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.cpu.incr == 0)
73 if (now < timer->it.cpu.expires)
76 incr = timer->it.cpu.incr;
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 if (WARN_ON_ONCE(!p))
382 * Protect against sighand release/switch in exit/exec and process/
383 * thread timer list entry concurrent read/writes.
385 sighand = lock_task_sighand(p, &flags);
386 if (unlikely(sighand == NULL)) {
388 * We raced with the reaping of the task.
389 * The deletion should have cleared us off the list.
391 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
393 if (timer->it.cpu.firing)
396 list_del(&timer->it.cpu.entry);
398 unlock_task_sighand(p, &flags);
407 static void cleanup_timers_list(struct list_head *head)
409 struct cpu_timer_list *timer, *next;
411 list_for_each_entry_safe(timer, next, head, entry)
412 list_del_init(&timer->entry);
416 * Clean out CPU timers still ticking when a thread exited. The task
417 * pointer is cleared, and the expiry time is replaced with the residual
418 * time for later timer_gettime calls to return.
419 * This must be called with the siglock held.
421 static void cleanup_timers(struct list_head *head)
423 cleanup_timers_list(head);
424 cleanup_timers_list(++head);
425 cleanup_timers_list(++head);
429 * These are both called with the siglock held, when the current thread
430 * is being reaped. When the final (leader) thread in the group is reaped,
431 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
433 void posix_cpu_timers_exit(struct task_struct *tsk)
435 cleanup_timers(tsk->cpu_timers);
437 void posix_cpu_timers_exit_group(struct task_struct *tsk)
439 cleanup_timers(tsk->signal->cpu_timers);
442 static inline int expires_gt(u64 expires, u64 new_exp)
444 return expires == 0 || expires > new_exp;
448 * Insert the timer on the appropriate list before any timers that
449 * expire later. This must be called with the sighand lock held.
451 static void arm_timer(struct k_itimer *timer)
453 struct task_struct *p = timer->it.cpu.task;
454 struct list_head *head, *listpos;
455 struct task_cputime *cputime_expires;
456 struct cpu_timer_list *const nt = &timer->it.cpu;
457 struct cpu_timer_list *next;
459 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
460 head = p->cpu_timers;
461 cputime_expires = &p->cputime_expires;
463 head = p->signal->cpu_timers;
464 cputime_expires = &p->signal->cputime_expires;
466 head += CPUCLOCK_WHICH(timer->it_clock);
469 list_for_each_entry(next, head, entry) {
470 if (nt->expires < next->expires)
472 listpos = &next->entry;
474 list_add(&nt->entry, listpos);
476 if (listpos == head) {
477 u64 exp = nt->expires;
480 * We are the new earliest-expiring POSIX 1.b timer, hence
481 * need to update expiration cache. Take into account that
482 * for process timers we share expiration cache with itimers
483 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
486 switch (CPUCLOCK_WHICH(timer->it_clock)) {
488 if (expires_gt(cputime_expires->prof_exp, exp))
489 cputime_expires->prof_exp = exp;
492 if (expires_gt(cputime_expires->virt_exp, exp))
493 cputime_expires->virt_exp = exp;
496 if (expires_gt(cputime_expires->sched_exp, exp))
497 cputime_expires->sched_exp = exp;
500 if (CPUCLOCK_PERTHREAD(timer->it_clock))
501 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
503 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
508 * The timer is locked, fire it and arrange for its reload.
510 static void cpu_timer_fire(struct k_itimer *timer)
512 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
514 * User don't want any signal.
516 timer->it.cpu.expires = 0;
517 } else if (unlikely(timer->sigq == NULL)) {
519 * This a special case for clock_nanosleep,
520 * not a normal timer from sys_timer_create.
522 wake_up_process(timer->it_process);
523 timer->it.cpu.expires = 0;
524 } else if (timer->it.cpu.incr == 0) {
526 * One-shot timer. Clear it as soon as it's fired.
528 posix_timer_event(timer, 0);
529 timer->it.cpu.expires = 0;
530 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
532 * The signal did not get queued because the signal
533 * was ignored, so we won't get any callback to
534 * reload the timer. But we need to keep it
535 * ticking in case the signal is deliverable next time.
537 posix_cpu_timer_rearm(timer);
538 ++timer->it_requeue_pending;
543 * Sample a process (thread group) timer for the given group_leader task.
544 * Must be called with task sighand lock held for safe while_each_thread()
547 static int cpu_timer_sample_group(const clockid_t which_clock,
548 struct task_struct *p, u64 *sample)
550 struct task_cputime cputime;
552 thread_group_cputimer(p, &cputime);
553 switch (CPUCLOCK_WHICH(which_clock)) {
557 *sample = cputime.utime + cputime.stime;
560 *sample = cputime.utime;
563 *sample = cputime.sum_exec_runtime;
570 * Guts of sys_timer_settime for CPU timers.
571 * This is called with the timer locked and interrupts disabled.
572 * If we return TIMER_RETRY, it's necessary to release the timer's lock
573 * and try again. (This happens when the timer is in the middle of firing.)
575 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
576 struct itimerspec64 *new, struct itimerspec64 *old)
579 struct sighand_struct *sighand;
580 struct task_struct *p = timer->it.cpu.task;
581 u64 old_expires, new_expires, old_incr, val;
584 if (WARN_ON_ONCE(!p))
588 * Use the to_ktime conversion because that clamps the maximum
589 * value to KTIME_MAX and avoid multiplication overflows.
591 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
594 * Protect against sighand release/switch in exit/exec and p->cpu_timers
595 * and p->signal->cpu_timers read/write in arm_timer()
597 sighand = lock_task_sighand(p, &flags);
599 * If p has just been reaped, we can no
600 * longer get any information about it at all.
602 if (unlikely(sighand == NULL)) {
607 * Disarm any old timer after extracting its expiry time.
611 old_incr = timer->it.cpu.incr;
612 old_expires = timer->it.cpu.expires;
613 if (unlikely(timer->it.cpu.firing)) {
614 timer->it.cpu.firing = -1;
617 list_del_init(&timer->it.cpu.entry);
620 * We need to sample the current value to convert the new
621 * value from to relative and absolute, and to convert the
622 * old value from absolute to relative. To set a process
623 * timer, we need a sample to balance the thread expiry
624 * times (in arm_timer). With an absolute time, we must
625 * check if it's already passed. In short, we need a sample.
627 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
628 cpu_clock_sample(timer->it_clock, p, &val);
630 cpu_timer_sample_group(timer->it_clock, p, &val);
634 if (old_expires == 0) {
635 old->it_value.tv_sec = 0;
636 old->it_value.tv_nsec = 0;
639 * Update the timer in case it has
640 * overrun already. If it has,
641 * we'll report it as having overrun
642 * and with the next reloaded timer
643 * already ticking, though we are
644 * swallowing that pending
645 * notification here to install the
648 bump_cpu_timer(timer, val);
649 if (val < timer->it.cpu.expires) {
650 old_expires = timer->it.cpu.expires - val;
651 old->it_value = ns_to_timespec64(old_expires);
653 old->it_value.tv_nsec = 1;
654 old->it_value.tv_sec = 0;
661 * We are colliding with the timer actually firing.
662 * Punt after filling in the timer's old value, and
663 * disable this firing since we are already reporting
664 * it as an overrun (thanks to bump_cpu_timer above).
666 unlock_task_sighand(p, &flags);
670 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
675 * Install the new expiry time (or zero).
676 * For a timer with no notification action, we don't actually
677 * arm the timer (we'll just fake it for timer_gettime).
679 timer->it.cpu.expires = new_expires;
680 if (new_expires != 0 && val < new_expires) {
684 unlock_task_sighand(p, &flags);
686 * Install the new reload setting, and
687 * set up the signal and overrun bookkeeping.
689 timer->it.cpu.incr = timespec64_to_ns(&new->it_interval);
690 timer->it_interval = ns_to_ktime(timer->it.cpu.incr);
693 * This acts as a modification timestamp for the timer,
694 * so any automatic reload attempt will punt on seeing
695 * that we have reset the timer manually.
697 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
699 timer->it_overrun_last = 0;
700 timer->it_overrun = -1;
702 if (new_expires != 0 && !(val < new_expires)) {
704 * The designated time already passed, so we notify
705 * immediately, even if the thread never runs to
706 * accumulate more time on this clock.
708 cpu_timer_fire(timer);
714 old->it_interval = ns_to_timespec64(old_incr);
719 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
721 struct task_struct *p = timer->it.cpu.task;
724 if (WARN_ON_ONCE(!p))
728 * Easy part: convert the reload time.
730 itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
732 if (!timer->it.cpu.expires)
736 * Sample the clock to take the difference with the expiry time.
738 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
739 cpu_clock_sample(timer->it_clock, p, &now);
741 struct sighand_struct *sighand;
745 * Protect against sighand release/switch in exit/exec and
746 * also make timer sampling safe if it ends up calling
747 * thread_group_cputime().
749 sighand = lock_task_sighand(p, &flags);
750 if (unlikely(sighand == NULL)) {
752 * The process has been reaped.
753 * We can't even collect a sample any more.
754 * Call the timer disarmed, nothing else to do.
756 timer->it.cpu.expires = 0;
759 cpu_timer_sample_group(timer->it_clock, p, &now);
760 unlock_task_sighand(p, &flags);
764 if (now < timer->it.cpu.expires) {
765 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
768 * The timer should have expired already, but the firing
769 * hasn't taken place yet. Say it's just about to expire.
771 itp->it_value.tv_nsec = 1;
772 itp->it_value.tv_sec = 0;
776 static unsigned long long
777 check_timers_list(struct list_head *timers,
778 struct list_head *firing,
779 unsigned long long curr)
783 while (!list_empty(timers)) {
784 struct cpu_timer_list *t;
786 t = list_first_entry(timers, struct cpu_timer_list, entry);
788 if (!--maxfire || curr < t->expires)
792 list_move_tail(&t->entry, firing);
798 static inline void check_dl_overrun(struct task_struct *tsk)
800 if (tsk->dl.dl_overrun) {
801 tsk->dl.dl_overrun = 0;
802 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
807 * Check for any per-thread CPU timers that have fired and move them off
808 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
809 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
811 static void check_thread_timers(struct task_struct *tsk,
812 struct list_head *firing)
814 struct list_head *timers = tsk->cpu_timers;
815 struct task_cputime *tsk_expires = &tsk->cputime_expires;
820 check_dl_overrun(tsk);
823 * If cputime_expires is zero, then there are no active
824 * per thread CPU timers.
826 if (task_cputime_zero(&tsk->cputime_expires))
829 expires = check_timers_list(timers, firing, prof_ticks(tsk));
830 tsk_expires->prof_exp = expires;
832 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
833 tsk_expires->virt_exp = expires;
835 tsk_expires->sched_exp = check_timers_list(++timers, firing,
836 tsk->se.sum_exec_runtime);
839 * Check for the special case thread timers.
841 soft = task_rlimit(tsk, RLIMIT_RTTIME);
842 if (soft != RLIM_INFINITY) {
843 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
845 if (hard != RLIM_INFINITY &&
846 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
848 * At the hard limit, we just die.
849 * No need to calculate anything else now.
851 if (print_fatal_signals) {
852 pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
853 tsk->comm, task_pid_nr(tsk));
855 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
858 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
860 * At the soft limit, send a SIGXCPU every second.
863 soft += USEC_PER_SEC;
864 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
867 if (print_fatal_signals) {
868 pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
869 tsk->comm, task_pid_nr(tsk));
871 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
874 if (task_cputime_zero(tsk_expires))
875 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
878 static inline void stop_process_timers(struct signal_struct *sig)
880 struct thread_group_cputimer *cputimer = &sig->cputimer;
882 /* Turn off cputimer->running. This is done without locking. */
883 WRITE_ONCE(cputimer->running, false);
884 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
887 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
888 u64 *expires, u64 cur_time, int signo)
893 if (cur_time >= it->expires) {
895 it->expires += it->incr;
899 trace_itimer_expire(signo == SIGPROF ?
900 ITIMER_PROF : ITIMER_VIRTUAL,
901 task_tgid(tsk), cur_time);
902 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
905 if (it->expires && (!*expires || it->expires < *expires))
906 *expires = it->expires;
910 * Check for any per-thread CPU timers that have fired and move them
911 * off the tsk->*_timers list onto the firing list. Per-thread timers
912 * have already been taken off.
914 static void check_process_timers(struct task_struct *tsk,
915 struct list_head *firing)
917 struct signal_struct *const sig = tsk->signal;
918 u64 utime, ptime, virt_expires, prof_expires;
919 u64 sum_sched_runtime, sched_expires;
920 struct list_head *timers = sig->cpu_timers;
921 struct task_cputime cputime;
925 check_dl_overrun(tsk);
928 * If cputimer is not running, then there are no active
929 * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
931 if (!READ_ONCE(tsk->signal->cputimer.running))
935 * Signify that a thread is checking for process timers.
936 * Write access to this field is protected by the sighand lock.
938 sig->cputimer.checking_timer = true;
941 * Collect the current process totals.
943 thread_group_cputimer(tsk, &cputime);
944 utime = cputime.utime;
945 ptime = utime + cputime.stime;
946 sum_sched_runtime = cputime.sum_exec_runtime;
948 prof_expires = check_timers_list(timers, firing, ptime);
949 virt_expires = check_timers_list(++timers, firing, utime);
950 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
953 * Check for the special case process timers.
955 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
957 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
959 soft = task_rlimit(tsk, RLIMIT_CPU);
960 if (soft != RLIM_INFINITY) {
961 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
962 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
966 * At the hard limit, we just die.
967 * No need to calculate anything else now.
969 if (print_fatal_signals) {
970 pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
971 tsk->comm, task_pid_nr(tsk));
973 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
978 * At the soft limit, send a SIGXCPU every second.
980 if (print_fatal_signals) {
981 pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
982 tsk->comm, task_pid_nr(tsk));
984 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
987 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
990 x = soft * NSEC_PER_SEC;
991 if (!prof_expires || x < prof_expires)
995 sig->cputime_expires.prof_exp = prof_expires;
996 sig->cputime_expires.virt_exp = virt_expires;
997 sig->cputime_expires.sched_exp = sched_expires;
998 if (task_cputime_zero(&sig->cputime_expires))
999 stop_process_timers(sig);
1001 sig->cputimer.checking_timer = false;
1005 * This is called from the signal code (via posixtimer_rearm)
1006 * when the last timer signal was delivered and we have to reload the timer.
1008 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1010 struct task_struct *p = timer->it.cpu.task;
1011 struct sighand_struct *sighand;
1012 unsigned long flags;
1015 if (WARN_ON_ONCE(!p))
1019 * Fetch the current sample and update the timer's expiry time.
1021 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1022 cpu_clock_sample(timer->it_clock, p, &now);
1023 bump_cpu_timer(timer, now);
1024 if (unlikely(p->exit_state))
1027 /* Protect timer list r/w in arm_timer() */
1028 sighand = lock_task_sighand(p, &flags);
1033 * Protect arm_timer() and timer sampling in case of call to
1034 * thread_group_cputime().
1036 sighand = lock_task_sighand(p, &flags);
1037 if (unlikely(sighand == NULL)) {
1039 * The process has been reaped.
1040 * We can't even collect a sample any more.
1042 timer->it.cpu.expires = 0;
1044 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1045 /* If the process is dying, no need to rearm */
1048 cpu_timer_sample_group(timer->it_clock, p, &now);
1049 bump_cpu_timer(timer, now);
1050 /* Leave the sighand locked for the call below. */
1054 * Now re-arm for the new expiry time.
1058 unlock_task_sighand(p, &flags);
1062 * task_cputime_expired - Compare two task_cputime entities.
1064 * @sample: The task_cputime structure to be checked for expiration.
1065 * @expires: Expiration times, against which @sample will be checked.
1067 * Checks @sample against @expires to see if any field of @sample has expired.
1068 * Returns true if any field of the former is greater than the corresponding
1069 * field of the latter if the latter field is set. Otherwise returns false.
1071 static inline int task_cputime_expired(const struct task_cputime *sample,
1072 const struct task_cputime *expires)
1074 if (expires->utime && sample->utime >= expires->utime)
1076 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1078 if (expires->sum_exec_runtime != 0 &&
1079 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1085 * fastpath_timer_check - POSIX CPU timers fast path.
1087 * @tsk: The task (thread) being checked.
1089 * Check the task and thread group timers. If both are zero (there are no
1090 * timers set) return false. Otherwise snapshot the task and thread group
1091 * timers and compare them with the corresponding expiration times. Return
1092 * true if a timer has expired, else return false.
1094 static inline int fastpath_timer_check(struct task_struct *tsk)
1096 struct signal_struct *sig;
1098 if (!task_cputime_zero(&tsk->cputime_expires)) {
1099 struct task_cputime task_sample;
1101 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1102 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1103 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1109 * Check if thread group timers expired when the cputimer is
1110 * running and no other thread in the group is already checking
1111 * for thread group cputimers. These fields are read without the
1112 * sighand lock. However, this is fine because this is meant to
1113 * be a fastpath heuristic to determine whether we should try to
1114 * acquire the sighand lock to check/handle timers.
1116 * In the worst case scenario, if 'running' or 'checking_timer' gets
1117 * set but the current thread doesn't see the change yet, we'll wait
1118 * until the next thread in the group gets a scheduler interrupt to
1119 * handle the timer. This isn't an issue in practice because these
1120 * types of delays with signals actually getting sent are expected.
1122 if (READ_ONCE(sig->cputimer.running) &&
1123 !READ_ONCE(sig->cputimer.checking_timer)) {
1124 struct task_cputime group_sample;
1126 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1128 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1132 if (dl_task(tsk) && tsk->dl.dl_overrun)
1139 * This is called from the timer interrupt handler. The irq handler has
1140 * already updated our counts. We need to check if any timers fire now.
1141 * Interrupts are disabled.
1143 void run_posix_cpu_timers(struct task_struct *tsk)
1146 struct k_itimer *timer, *next;
1147 unsigned long flags;
1149 lockdep_assert_irqs_disabled();
1152 * The fast path checks that there are no expired thread or thread
1153 * group timers. If that's so, just return.
1155 if (!fastpath_timer_check(tsk))
1158 if (!lock_task_sighand(tsk, &flags))
1161 * Here we take off tsk->signal->cpu_timers[N] and
1162 * tsk->cpu_timers[N] all the timers that are firing, and
1163 * put them on the firing list.
1165 check_thread_timers(tsk, &firing);
1167 check_process_timers(tsk, &firing);
1170 * We must release these locks before taking any timer's lock.
1171 * There is a potential race with timer deletion here, as the
1172 * siglock now protects our private firing list. We have set
1173 * the firing flag in each timer, so that a deletion attempt
1174 * that gets the timer lock before we do will give it up and
1175 * spin until we've taken care of that timer below.
1177 unlock_task_sighand(tsk, &flags);
1180 * Now that all the timers on our list have the firing flag,
1181 * no one will touch their list entries but us. We'll take
1182 * each timer's lock before clearing its firing flag, so no
1183 * timer call will interfere.
1185 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1188 spin_lock(&timer->it_lock);
1189 list_del_init(&timer->it.cpu.entry);
1190 cpu_firing = timer->it.cpu.firing;
1191 timer->it.cpu.firing = 0;
1193 * The firing flag is -1 if we collided with a reset
1194 * of the timer, which already reported this
1195 * almost-firing as an overrun. So don't generate an event.
1197 if (likely(cpu_firing >= 0))
1198 cpu_timer_fire(timer);
1199 spin_unlock(&timer->it_lock);
1204 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1205 * The tsk->sighand->siglock must be held by the caller.
1207 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1208 u64 *newval, u64 *oldval)
1213 if (WARN_ON_ONCE(clock_idx >= CPUCLOCK_SCHED))
1216 ret = cpu_timer_sample_group(clock_idx, tsk, &now);
1218 if (oldval && ret != -EINVAL) {
1220 * We are setting itimer. The *oldval is absolute and we update
1221 * it to be relative, *newval argument is relative and we update
1222 * it to be absolute.
1225 if (*oldval <= now) {
1226 /* Just about to fire. */
1227 *oldval = TICK_NSEC;
1239 * Update expiration cache if we are the earliest timer, or eventually
1240 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1242 switch (clock_idx) {
1244 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1245 tsk->signal->cputime_expires.prof_exp = *newval;
1248 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1249 tsk->signal->cputime_expires.virt_exp = *newval;
1253 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1256 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1257 const struct timespec64 *rqtp)
1259 struct itimerspec64 it;
1260 struct k_itimer timer;
1265 * Set up a temporary timer and then wait for it to go off.
1267 memset(&timer, 0, sizeof timer);
1268 spin_lock_init(&timer.it_lock);
1269 timer.it_clock = which_clock;
1270 timer.it_overrun = -1;
1271 error = posix_cpu_timer_create(&timer);
1272 timer.it_process = current;
1274 static struct itimerspec64 zero_it;
1275 struct restart_block *restart;
1277 memset(&it, 0, sizeof(it));
1278 it.it_value = *rqtp;
1280 spin_lock_irq(&timer.it_lock);
1281 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1283 spin_unlock_irq(&timer.it_lock);
1287 while (!signal_pending(current)) {
1288 if (timer.it.cpu.expires == 0) {
1290 * Our timer fired and was reset, below
1291 * deletion can not fail.
1293 posix_cpu_timer_del(&timer);
1294 spin_unlock_irq(&timer.it_lock);
1299 * Block until cpu_timer_fire (or a signal) wakes us.
1301 __set_current_state(TASK_INTERRUPTIBLE);
1302 spin_unlock_irq(&timer.it_lock);
1304 spin_lock_irq(&timer.it_lock);
1308 * We were interrupted by a signal.
1310 expires = timer.it.cpu.expires;
1311 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1314 * Timer is now unarmed, deletion can not fail.
1316 posix_cpu_timer_del(&timer);
1318 spin_unlock_irq(&timer.it_lock);
1320 while (error == TIMER_RETRY) {
1322 * We need to handle case when timer was or is in the
1323 * middle of firing. In other cases we already freed
1326 spin_lock_irq(&timer.it_lock);
1327 error = posix_cpu_timer_del(&timer);
1328 spin_unlock_irq(&timer.it_lock);
1331 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1333 * It actually did fire already.
1338 error = -ERESTART_RESTARTBLOCK;
1340 * Report back to the user the time still remaining.
1342 restart = ¤t->restart_block;
1343 restart->nanosleep.expires = expires;
1344 if (restart->nanosleep.type != TT_NONE)
1345 error = nanosleep_copyout(restart, &it.it_value);
1351 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1353 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1354 const struct timespec64 *rqtp)
1356 struct restart_block *restart_block = ¤t->restart_block;
1360 * Diagnose required errors first.
1362 if (CPUCLOCK_PERTHREAD(which_clock) &&
1363 (CPUCLOCK_PID(which_clock) == 0 ||
1364 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1367 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1369 if (error == -ERESTART_RESTARTBLOCK) {
1371 if (flags & TIMER_ABSTIME)
1372 return -ERESTARTNOHAND;
1374 restart_block->nanosleep.clockid = which_clock;
1375 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1380 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1382 clockid_t which_clock = restart_block->nanosleep.clockid;
1383 struct timespec64 t;
1385 t = ns_to_timespec64(restart_block->nanosleep.expires);
1387 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1390 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1391 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1393 static int process_cpu_clock_getres(const clockid_t which_clock,
1394 struct timespec64 *tp)
1396 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1398 static int process_cpu_clock_get(const clockid_t which_clock,
1399 struct timespec64 *tp)
1401 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1403 static int process_cpu_timer_create(struct k_itimer *timer)
1405 timer->it_clock = PROCESS_CLOCK;
1406 return posix_cpu_timer_create(timer);
1408 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1409 const struct timespec64 *rqtp)
1411 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1413 static int thread_cpu_clock_getres(const clockid_t which_clock,
1414 struct timespec64 *tp)
1416 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1418 static int thread_cpu_clock_get(const clockid_t which_clock,
1419 struct timespec64 *tp)
1421 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1423 static int thread_cpu_timer_create(struct k_itimer *timer)
1425 timer->it_clock = THREAD_CLOCK;
1426 return posix_cpu_timer_create(timer);
1429 const struct k_clock clock_posix_cpu = {
1430 .clock_getres = posix_cpu_clock_getres,
1431 .clock_set = posix_cpu_clock_set,
1432 .clock_get = posix_cpu_clock_get,
1433 .timer_create = posix_cpu_timer_create,
1434 .nsleep = posix_cpu_nsleep,
1435 .timer_set = posix_cpu_timer_set,
1436 .timer_del = posix_cpu_timer_del,
1437 .timer_get = posix_cpu_timer_get,
1438 .timer_rearm = posix_cpu_timer_rearm,
1441 const struct k_clock clock_process = {
1442 .clock_getres = process_cpu_clock_getres,
1443 .clock_get = process_cpu_clock_get,
1444 .timer_create = process_cpu_timer_create,
1445 .nsleep = process_cpu_nsleep,
1448 const struct k_clock clock_thread = {
1449 .clock_getres = thread_cpu_clock_getres,
1450 .clock_get = thread_cpu_clock_get,
1451 .timer_create = thread_cpu_timer_create,