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);
23 void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
25 posix_cputimers_init(pct);
26 if (cpu_limit != RLIM_INFINITY) {
27 pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
28 pct->timers_active = true;
33 * Called after updating RLIMIT_CPU to run cpu timer and update
34 * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
35 * necessary. Needs siglock protection since other code may update the
36 * expiration cache as well.
38 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
40 u64 nsecs = rlim_new * NSEC_PER_SEC;
42 spin_lock_irq(&task->sighand->siglock);
43 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
44 spin_unlock_irq(&task->sighand->siglock);
48 * Functions for validating access to tasks.
50 static struct task_struct *lookup_task(const pid_t pid, bool thread,
53 struct task_struct *p;
56 * If the encoded PID is 0, then the timer is targeted at current
57 * or the process to which current belongs.
60 return thread ? current : current->group_leader;
62 p = find_task_by_vpid(pid);
67 return same_thread_group(p, current) ? p : NULL;
70 * For clock_gettime(PROCESS) the task does not need to be
71 * the actual group leader. task->signal gives
72 * access to the group's clock.
74 if (gettime && (p == current))
78 * For processes require that p is group leader.
80 return thread_group_leader(p) ? p : NULL;
83 static struct task_struct *__get_task_for_clock(const clockid_t clock,
86 const bool thread = !!CPUCLOCK_PERTHREAD(clock);
87 const pid_t pid = CPUCLOCK_PID(clock);
89 if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
92 return lookup_task(pid, thread, gettime);
95 static inline struct task_struct *get_task_for_clock(const clockid_t clock)
97 return __get_task_for_clock(clock, false);
100 static inline struct task_struct *get_task_for_clock_get(const clockid_t clock)
102 return __get_task_for_clock(clock, true);
105 static inline int validate_clock_permissions(const clockid_t clock)
110 ret = __get_task_for_clock(clock, false) ? 0 : -EINVAL;
116 static inline enum pid_type clock_pid_type(const clockid_t clock)
118 return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
121 static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
123 return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
127 * Update expiry time from increment, and increase overrun count,
128 * given the current clock sample.
130 static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
132 u64 delta, incr, expires = timer->it.cpu.node.expires;
135 if (!timer->it_interval)
141 incr = timer->it_interval;
142 delta = now + incr - expires;
144 /* Don't use (incr*2 < delta), incr*2 might overflow. */
145 for (i = 0; incr < delta - incr; i++)
148 for (; i >= 0; incr >>= 1, i--) {
152 timer->it.cpu.node.expires += incr;
153 timer->it_overrun += 1LL << i;
156 return timer->it.cpu.node.expires;
159 /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
160 static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
162 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
163 ~pct->bases[CPUCLOCK_VIRT].nextevt |
164 ~pct->bases[CPUCLOCK_SCHED].nextevt);
168 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
170 int error = validate_clock_permissions(which_clock);
174 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
175 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
177 * If sched_clock is using a cycle counter, we
178 * don't have any idea of its true resolution
179 * exported, but it is much more than 1s/HZ.
188 posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
190 int error = validate_clock_permissions(clock);
193 * You can never reset a CPU clock, but we check for other errors
194 * in the call before failing with EPERM.
196 return error ? : -EPERM;
200 * Sample a per-thread clock for the given task. clkid is validated.
202 static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
206 if (clkid == CPUCLOCK_SCHED)
207 return task_sched_runtime(p);
209 task_cputime(p, &utime, &stime);
213 return utime + stime;
222 static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
224 samples[CPUCLOCK_PROF] = stime + utime;
225 samples[CPUCLOCK_VIRT] = utime;
226 samples[CPUCLOCK_SCHED] = rtime;
229 static void task_sample_cputime(struct task_struct *p, u64 *samples)
233 task_cputime(p, &utime, &stime);
234 store_samples(samples, stime, utime, p->se.sum_exec_runtime);
237 static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
240 u64 stime, utime, rtime;
242 utime = atomic64_read(&at->utime);
243 stime = atomic64_read(&at->stime);
244 rtime = atomic64_read(&at->sum_exec_runtime);
245 store_samples(samples, stime, utime, rtime);
249 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
250 * to avoid race conditions with concurrent updates to cputime.
252 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
256 curr_cputime = atomic64_read(cputime);
257 if (sum_cputime > curr_cputime) {
258 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
263 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
264 struct task_cputime *sum)
266 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
267 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
268 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
272 * thread_group_sample_cputime - Sample cputime for a given task
273 * @tsk: Task for which cputime needs to be started
274 * @samples: Storage for time samples
276 * Called from sys_getitimer() to calculate the expiry time of an active
277 * timer. That means group cputime accounting is already active. Called
278 * with task sighand lock held.
280 * Updates @times with an uptodate sample of the thread group cputimes.
282 void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
284 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
285 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
287 WARN_ON_ONCE(!pct->timers_active);
289 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
293 * thread_group_start_cputime - Start cputime and return a sample
294 * @tsk: Task for which cputime needs to be started
295 * @samples: Storage for time samples
297 * The thread group cputime accouting is avoided when there are no posix
298 * CPU timers armed. Before starting a timer it's required to check whether
299 * the time accounting is active. If not, a full update of the atomic
300 * accounting store needs to be done and the accounting enabled.
302 * Updates @times with an uptodate sample of the thread group cputimes.
304 static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
306 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
307 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
309 /* Check if cputimer isn't running. This is accessed without locking. */
310 if (!READ_ONCE(pct->timers_active)) {
311 struct task_cputime sum;
314 * The POSIX timer interface allows for absolute time expiry
315 * values through the TIMER_ABSTIME flag, therefore we have
316 * to synchronize the timer to the clock every time we start it.
318 thread_group_cputime(tsk, &sum);
319 update_gt_cputime(&cputimer->cputime_atomic, &sum);
322 * We're setting timers_active without a lock. Ensure this
323 * only gets written to in one operation. We set it after
324 * update_gt_cputime() as a small optimization, but
325 * barriers are not required because update_gt_cputime()
326 * can handle concurrent updates.
328 WRITE_ONCE(pct->timers_active, true);
330 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
333 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
335 struct task_cputime ct;
337 thread_group_cputime(tsk, &ct);
338 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
342 * Sample a process (thread group) clock for the given task clkid. If the
343 * group's cputime accounting is already enabled, read the atomic
344 * store. Otherwise a full update is required. clkid is already validated.
346 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
349 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
350 struct posix_cputimers *pct = &p->signal->posix_cputimers;
351 u64 samples[CPUCLOCK_MAX];
353 if (!READ_ONCE(pct->timers_active)) {
355 thread_group_start_cputime(p, samples);
357 __thread_group_cputime(p, samples);
359 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
362 return samples[clkid];
365 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
367 const clockid_t clkid = CPUCLOCK_WHICH(clock);
368 struct task_struct *tsk;
372 tsk = get_task_for_clock_get(clock);
378 if (CPUCLOCK_PERTHREAD(clock))
379 t = cpu_clock_sample(clkid, tsk);
381 t = cpu_clock_sample_group(clkid, tsk, false);
384 *tp = ns_to_timespec64(t);
389 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
390 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
391 * new timer already all-zeros initialized.
393 static int posix_cpu_timer_create(struct k_itimer *new_timer)
395 struct task_struct *p;
398 p = get_task_for_clock(new_timer->it_clock);
404 new_timer->kclock = &clock_posix_cpu;
405 timerqueue_init(&new_timer->it.cpu.node);
406 new_timer->it.cpu.pid = get_task_pid(p, clock_pid_type(new_timer->it_clock));
412 * Clean up a CPU-clock timer that is about to be destroyed.
413 * This is called from timer deletion with the timer already locked.
414 * If we return TIMER_RETRY, it's necessary to release the timer's lock
415 * and try again. (This happens when the timer is in the middle of firing.)
417 static int posix_cpu_timer_del(struct k_itimer *timer)
419 struct cpu_timer *ctmr = &timer->it.cpu;
420 struct sighand_struct *sighand;
421 struct task_struct *p;
426 p = cpu_timer_task_rcu(timer);
431 * Protect against sighand release/switch in exit/exec and process/
432 * thread timer list entry concurrent read/writes.
434 sighand = lock_task_sighand(p, &flags);
435 if (unlikely(sighand == NULL)) {
437 * This raced with the reaping of the task. The exit cleanup
438 * should have removed this timer from the timer queue.
440 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
442 if (timer->it.cpu.firing)
445 cpu_timer_dequeue(ctmr);
447 unlock_task_sighand(p, &flags);
458 static void cleanup_timerqueue(struct timerqueue_head *head)
460 struct timerqueue_node *node;
461 struct cpu_timer *ctmr;
463 while ((node = timerqueue_getnext(head))) {
464 timerqueue_del(head, node);
465 ctmr = container_of(node, struct cpu_timer, node);
471 * Clean out CPU timers which are still armed when a thread exits. The
472 * timers are only removed from the list. No other updates are done. The
473 * corresponding posix timers are still accessible, but cannot be rearmed.
475 * This must be called with the siglock held.
477 static void cleanup_timers(struct posix_cputimers *pct)
479 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
480 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
481 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
485 * These are both called with the siglock held, when the current thread
486 * is being reaped. When the final (leader) thread in the group is reaped,
487 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
489 void posix_cpu_timers_exit(struct task_struct *tsk)
491 cleanup_timers(&tsk->posix_cputimers);
493 void posix_cpu_timers_exit_group(struct task_struct *tsk)
495 cleanup_timers(&tsk->signal->posix_cputimers);
499 * Insert the timer on the appropriate list before any timers that
500 * expire later. This must be called with the sighand lock held.
502 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
504 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
505 struct cpu_timer *ctmr = &timer->it.cpu;
506 u64 newexp = cpu_timer_getexpires(ctmr);
507 struct posix_cputimer_base *base;
509 if (CPUCLOCK_PERTHREAD(timer->it_clock))
510 base = p->posix_cputimers.bases + clkidx;
512 base = p->signal->posix_cputimers.bases + clkidx;
514 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
518 * We are the new earliest-expiring POSIX 1.b timer, hence
519 * need to update expiration cache. Take into account that
520 * for process timers we share expiration cache with itimers
521 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
523 if (newexp < base->nextevt)
524 base->nextevt = newexp;
526 if (CPUCLOCK_PERTHREAD(timer->it_clock))
527 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
529 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
533 * The timer is locked, fire it and arrange for its reload.
535 static void cpu_timer_fire(struct k_itimer *timer)
537 struct cpu_timer *ctmr = &timer->it.cpu;
539 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
541 * User don't want any signal.
543 cpu_timer_setexpires(ctmr, 0);
544 } else if (unlikely(timer->sigq == NULL)) {
546 * This a special case for clock_nanosleep,
547 * not a normal timer from sys_timer_create.
549 wake_up_process(timer->it_process);
550 cpu_timer_setexpires(ctmr, 0);
551 } else if (!timer->it_interval) {
553 * One-shot timer. Clear it as soon as it's fired.
555 posix_timer_event(timer, 0);
556 cpu_timer_setexpires(ctmr, 0);
557 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
559 * The signal did not get queued because the signal
560 * was ignored, so we won't get any callback to
561 * reload the timer. But we need to keep it
562 * ticking in case the signal is deliverable next time.
564 posix_cpu_timer_rearm(timer);
565 ++timer->it_requeue_pending;
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)
578 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
579 u64 old_expires, new_expires, old_incr, val;
580 struct cpu_timer *ctmr = &timer->it.cpu;
581 struct sighand_struct *sighand;
582 struct task_struct *p;
587 p = cpu_timer_task_rcu(timer);
590 * If p has just been reaped, we can no
591 * longer get any information about it at all.
598 * Use the to_ktime conversion because that clamps the maximum
599 * value to KTIME_MAX and avoid multiplication overflows.
601 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
604 * Protect against sighand release/switch in exit/exec and p->cpu_timers
605 * and p->signal->cpu_timers read/write in arm_timer()
607 sighand = lock_task_sighand(p, &flags);
609 * If p has just been reaped, we can no
610 * longer get any information about it at all.
612 if (unlikely(sighand == NULL)) {
618 * Disarm any old timer after extracting its expiry time.
620 old_incr = timer->it_interval;
621 old_expires = cpu_timer_getexpires(ctmr);
623 if (unlikely(timer->it.cpu.firing)) {
624 timer->it.cpu.firing = -1;
627 cpu_timer_dequeue(ctmr);
631 * We need to sample the current value to convert the new
632 * value from to relative and absolute, and to convert the
633 * old value from absolute to relative. To set a process
634 * timer, we need a sample to balance the thread expiry
635 * times (in arm_timer). With an absolute time, we must
636 * check if it's already passed. In short, we need a sample.
638 if (CPUCLOCK_PERTHREAD(timer->it_clock))
639 val = cpu_clock_sample(clkid, p);
641 val = cpu_clock_sample_group(clkid, p, true);
644 if (old_expires == 0) {
645 old->it_value.tv_sec = 0;
646 old->it_value.tv_nsec = 0;
649 * Update the timer in case it has overrun already.
650 * If it has, we'll report it as having overrun and
651 * with the next reloaded timer already ticking,
652 * though we are swallowing that pending
653 * notification here to install the new setting.
655 u64 exp = bump_cpu_timer(timer, val);
658 old_expires = exp - val;
659 old->it_value = ns_to_timespec64(old_expires);
661 old->it_value.tv_nsec = 1;
662 old->it_value.tv_sec = 0;
669 * We are colliding with the timer actually firing.
670 * Punt after filling in the timer's old value, and
671 * disable this firing since we are already reporting
672 * it as an overrun (thanks to bump_cpu_timer above).
674 unlock_task_sighand(p, &flags);
678 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
683 * Install the new expiry time (or zero).
684 * For a timer with no notification action, we don't actually
685 * arm the timer (we'll just fake it for timer_gettime).
687 cpu_timer_setexpires(ctmr, new_expires);
688 if (new_expires != 0 && val < new_expires) {
692 unlock_task_sighand(p, &flags);
694 * Install the new reload setting, and
695 * set up the signal and overrun bookkeeping.
697 timer->it_interval = timespec64_to_ktime(new->it_interval);
700 * This acts as a modification timestamp for the timer,
701 * so any automatic reload attempt will punt on seeing
702 * that we have reset the timer manually.
704 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
706 timer->it_overrun_last = 0;
707 timer->it_overrun = -1;
709 if (new_expires != 0 && !(val < new_expires)) {
711 * The designated time already passed, so we notify
712 * immediately, even if the thread never runs to
713 * accumulate more time on this clock.
715 cpu_timer_fire(timer);
722 old->it_interval = ns_to_timespec64(old_incr);
727 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
729 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
730 struct cpu_timer *ctmr = &timer->it.cpu;
731 u64 now, expires = cpu_timer_getexpires(ctmr);
732 struct task_struct *p;
735 p = cpu_timer_task_rcu(timer);
740 * Easy part: convert the reload time.
742 itp->it_interval = ktime_to_timespec64(timer->it_interval);
748 * Sample the clock to take the difference with the expiry time.
750 if (CPUCLOCK_PERTHREAD(timer->it_clock))
751 now = cpu_clock_sample(clkid, p);
753 now = cpu_clock_sample_group(clkid, p, false);
756 itp->it_value = ns_to_timespec64(expires - now);
759 * The timer should have expired already, but the firing
760 * hasn't taken place yet. Say it's just about to expire.
762 itp->it_value.tv_nsec = 1;
763 itp->it_value.tv_sec = 0;
769 #define MAX_COLLECTED 20
771 static u64 collect_timerqueue(struct timerqueue_head *head,
772 struct list_head *firing, u64 now)
774 struct timerqueue_node *next;
777 while ((next = timerqueue_getnext(head))) {
778 struct cpu_timer *ctmr;
781 ctmr = container_of(next, struct cpu_timer, node);
782 expires = cpu_timer_getexpires(ctmr);
783 /* Limit the number of timers to expire at once */
784 if (++i == MAX_COLLECTED || now < expires)
788 cpu_timer_dequeue(ctmr);
789 list_add_tail(&ctmr->elist, firing);
795 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
796 struct list_head *firing)
798 struct posix_cputimer_base *base = pct->bases;
801 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
802 base->nextevt = collect_timerqueue(&base->tqhead, firing,
807 static inline void check_dl_overrun(struct task_struct *tsk)
809 if (tsk->dl.dl_overrun) {
810 tsk->dl.dl_overrun = 0;
811 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
815 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
820 if (print_fatal_signals) {
821 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
822 rt ? "RT" : "CPU", hard ? "hard" : "soft",
823 current->comm, task_pid_nr(current));
825 __group_send_sig_info(signo, SEND_SIG_PRIV, current);
830 * Check for any per-thread CPU timers that have fired and move them off
831 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
832 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
834 static void check_thread_timers(struct task_struct *tsk,
835 struct list_head *firing)
837 struct posix_cputimers *pct = &tsk->posix_cputimers;
838 u64 samples[CPUCLOCK_MAX];
842 check_dl_overrun(tsk);
844 if (expiry_cache_is_inactive(pct))
847 task_sample_cputime(tsk, samples);
848 collect_posix_cputimers(pct, samples, firing);
851 * Check for the special case thread timers.
853 soft = task_rlimit(tsk, RLIMIT_RTTIME);
854 if (soft != RLIM_INFINITY) {
855 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
856 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
857 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
859 /* At the hard limit, send SIGKILL. No further action. */
860 if (hard != RLIM_INFINITY &&
861 check_rlimit(rttime, hard, SIGKILL, true, true))
864 /* At the soft limit, send a SIGXCPU every second */
865 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
866 soft += USEC_PER_SEC;
867 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
871 if (expiry_cache_is_inactive(pct))
872 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
875 static inline void stop_process_timers(struct signal_struct *sig)
877 struct posix_cputimers *pct = &sig->posix_cputimers;
879 /* Turn off the active flag. This is done without locking. */
880 WRITE_ONCE(pct->timers_active, false);
881 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
884 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
885 u64 *expires, u64 cur_time, int signo)
890 if (cur_time >= it->expires) {
892 it->expires += it->incr;
896 trace_itimer_expire(signo == SIGPROF ?
897 ITIMER_PROF : ITIMER_VIRTUAL,
898 task_tgid(tsk), cur_time);
899 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
902 if (it->expires && it->expires < *expires)
903 *expires = it->expires;
907 * Check for any per-thread CPU timers that have fired and move them
908 * off the tsk->*_timers list onto the firing list. Per-thread timers
909 * have already been taken off.
911 static void check_process_timers(struct task_struct *tsk,
912 struct list_head *firing)
914 struct signal_struct *const sig = tsk->signal;
915 struct posix_cputimers *pct = &sig->posix_cputimers;
916 u64 samples[CPUCLOCK_MAX];
920 * If there are no active process wide timers (POSIX 1.b, itimers,
921 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
922 * processing when there is already another task handling them.
924 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
928 * Signify that a thread is checking for process timers.
929 * Write access to this field is protected by the sighand lock.
931 pct->expiry_active = true;
934 * Collect the current process totals. Group accounting is active
935 * so the sample can be taken directly.
937 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
938 collect_posix_cputimers(pct, samples, firing);
941 * Check for the special case process timers.
943 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
944 &pct->bases[CPUCLOCK_PROF].nextevt,
945 samples[CPUCLOCK_PROF], SIGPROF);
946 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
947 &pct->bases[CPUCLOCK_VIRT].nextevt,
948 samples[CPUCLOCK_VIRT], SIGVTALRM);
950 soft = task_rlimit(tsk, RLIMIT_CPU);
951 if (soft != RLIM_INFINITY) {
952 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
953 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
954 u64 ptime = samples[CPUCLOCK_PROF];
955 u64 softns = (u64)soft * NSEC_PER_SEC;
956 u64 hardns = (u64)hard * NSEC_PER_SEC;
958 /* At the hard limit, send SIGKILL. No further action. */
959 if (hard != RLIM_INFINITY &&
960 check_rlimit(ptime, hardns, SIGKILL, false, true))
963 /* At the soft limit, send a SIGXCPU every second */
964 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
965 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
966 softns += NSEC_PER_SEC;
969 /* Update the expiry cache */
970 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
971 pct->bases[CPUCLOCK_PROF].nextevt = softns;
974 if (expiry_cache_is_inactive(pct))
975 stop_process_timers(sig);
977 pct->expiry_active = false;
981 * This is called from the signal code (via posixtimer_rearm)
982 * when the last timer signal was delivered and we have to reload the timer.
984 static void posix_cpu_timer_rearm(struct k_itimer *timer)
986 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
987 struct task_struct *p;
988 struct sighand_struct *sighand;
993 p = cpu_timer_task_rcu(timer);
998 * Fetch the current sample and update the timer's expiry time.
1000 if (CPUCLOCK_PERTHREAD(timer->it_clock))
1001 now = cpu_clock_sample(clkid, p);
1003 now = cpu_clock_sample_group(clkid, p, true);
1005 bump_cpu_timer(timer, now);
1007 /* Protect timer list r/w in arm_timer() */
1008 sighand = lock_task_sighand(p, &flags);
1009 if (unlikely(sighand == NULL))
1013 * Now re-arm for the new expiry time.
1015 arm_timer(timer, p);
1016 unlock_task_sighand(p, &flags);
1022 * task_cputimers_expired - Check whether posix CPU timers are expired
1024 * @samples: Array of current samples for the CPUCLOCK clocks
1025 * @pct: Pointer to a posix_cputimers container
1027 * Returns true if any member of @samples is greater than the corresponding
1028 * member of @pct->bases[CLK].nextevt. False otherwise
1031 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1035 for (i = 0; i < CPUCLOCK_MAX; i++) {
1036 if (samples[i] >= pct->bases[i].nextevt)
1043 * fastpath_timer_check - POSIX CPU timers fast path.
1045 * @tsk: The task (thread) being checked.
1047 * Check the task and thread group timers. If both are zero (there are no
1048 * timers set) return false. Otherwise snapshot the task and thread group
1049 * timers and compare them with the corresponding expiration times. Return
1050 * true if a timer has expired, else return false.
1052 static inline bool fastpath_timer_check(struct task_struct *tsk)
1054 struct posix_cputimers *pct = &tsk->posix_cputimers;
1055 struct signal_struct *sig;
1057 if (!expiry_cache_is_inactive(pct)) {
1058 u64 samples[CPUCLOCK_MAX];
1060 task_sample_cputime(tsk, samples);
1061 if (task_cputimers_expired(samples, pct))
1066 pct = &sig->posix_cputimers;
1068 * Check if thread group timers expired when timers are active and
1069 * no other thread in the group is already handling expiry for
1070 * thread group cputimers. These fields are read without the
1071 * sighand lock. However, this is fine because this is meant to be
1072 * a fastpath heuristic to determine whether we should try to
1073 * acquire the sighand lock to handle timer expiry.
1075 * In the worst case scenario, if concurrently timers_active is set
1076 * or expiry_active is cleared, but the current thread doesn't see
1077 * the change yet, the timer checks are delayed until the next
1078 * thread in the group gets a scheduler interrupt to handle the
1079 * timer. This isn't an issue in practice because these types of
1080 * delays with signals actually getting sent are expected.
1082 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1083 u64 samples[CPUCLOCK_MAX];
1085 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1088 if (task_cputimers_expired(samples, pct))
1092 if (dl_task(tsk) && tsk->dl.dl_overrun)
1099 * This is called from the timer interrupt handler. The irq handler has
1100 * already updated our counts. We need to check if any timers fire now.
1101 * Interrupts are disabled.
1103 void run_posix_cpu_timers(void)
1105 struct task_struct *tsk = current;
1106 struct k_itimer *timer, *next;
1107 unsigned long flags;
1110 lockdep_assert_irqs_disabled();
1113 * The fast path checks that there are no expired thread or thread
1114 * group timers. If that's so, just return.
1116 if (!fastpath_timer_check(tsk))
1119 lockdep_posixtimer_enter();
1120 if (!lock_task_sighand(tsk, &flags)) {
1121 lockdep_posixtimer_exit();
1125 * Here we take off tsk->signal->cpu_timers[N] and
1126 * tsk->cpu_timers[N] all the timers that are firing, and
1127 * put them on the firing list.
1129 check_thread_timers(tsk, &firing);
1131 check_process_timers(tsk, &firing);
1134 * We must release these locks before taking any timer's lock.
1135 * There is a potential race with timer deletion here, as the
1136 * siglock now protects our private firing list. We have set
1137 * the firing flag in each timer, so that a deletion attempt
1138 * that gets the timer lock before we do will give it up and
1139 * spin until we've taken care of that timer below.
1141 unlock_task_sighand(tsk, &flags);
1144 * Now that all the timers on our list have the firing flag,
1145 * no one will touch their list entries but us. We'll take
1146 * each timer's lock before clearing its firing flag, so no
1147 * timer call will interfere.
1149 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1152 spin_lock(&timer->it_lock);
1153 list_del_init(&timer->it.cpu.elist);
1154 cpu_firing = timer->it.cpu.firing;
1155 timer->it.cpu.firing = 0;
1157 * The firing flag is -1 if we collided with a reset
1158 * of the timer, which already reported this
1159 * almost-firing as an overrun. So don't generate an event.
1161 if (likely(cpu_firing >= 0))
1162 cpu_timer_fire(timer);
1163 spin_unlock(&timer->it_lock);
1165 lockdep_posixtimer_exit();
1169 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1170 * The tsk->sighand->siglock must be held by the caller.
1172 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1173 u64 *newval, u64 *oldval)
1177 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1180 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1181 now = cpu_clock_sample_group(clkid, tsk, true);
1185 * We are setting itimer. The *oldval is absolute and we update
1186 * it to be relative, *newval argument is relative and we update
1187 * it to be absolute.
1190 if (*oldval <= now) {
1191 /* Just about to fire. */
1192 *oldval = TICK_NSEC;
1204 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1205 * expiry cache is also used by RLIMIT_CPU!.
1207 if (*newval < *nextevt)
1210 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1213 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1214 const struct timespec64 *rqtp)
1216 struct itimerspec64 it;
1217 struct k_itimer timer;
1222 * Set up a temporary timer and then wait for it to go off.
1224 memset(&timer, 0, sizeof timer);
1225 spin_lock_init(&timer.it_lock);
1226 timer.it_clock = which_clock;
1227 timer.it_overrun = -1;
1228 error = posix_cpu_timer_create(&timer);
1229 timer.it_process = current;
1232 static struct itimerspec64 zero_it;
1233 struct restart_block *restart;
1235 memset(&it, 0, sizeof(it));
1236 it.it_value = *rqtp;
1238 spin_lock_irq(&timer.it_lock);
1239 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1241 spin_unlock_irq(&timer.it_lock);
1245 while (!signal_pending(current)) {
1246 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1248 * Our timer fired and was reset, below
1249 * deletion can not fail.
1251 posix_cpu_timer_del(&timer);
1252 spin_unlock_irq(&timer.it_lock);
1257 * Block until cpu_timer_fire (or a signal) wakes us.
1259 __set_current_state(TASK_INTERRUPTIBLE);
1260 spin_unlock_irq(&timer.it_lock);
1262 spin_lock_irq(&timer.it_lock);
1266 * We were interrupted by a signal.
1268 expires = cpu_timer_getexpires(&timer.it.cpu);
1269 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1272 * Timer is now unarmed, deletion can not fail.
1274 posix_cpu_timer_del(&timer);
1276 spin_unlock_irq(&timer.it_lock);
1278 while (error == TIMER_RETRY) {
1280 * We need to handle case when timer was or is in the
1281 * middle of firing. In other cases we already freed
1284 spin_lock_irq(&timer.it_lock);
1285 error = posix_cpu_timer_del(&timer);
1286 spin_unlock_irq(&timer.it_lock);
1289 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1291 * It actually did fire already.
1296 error = -ERESTART_RESTARTBLOCK;
1298 * Report back to the user the time still remaining.
1300 restart = ¤t->restart_block;
1301 restart->nanosleep.expires = expires;
1302 if (restart->nanosleep.type != TT_NONE)
1303 error = nanosleep_copyout(restart, &it.it_value);
1309 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1311 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1312 const struct timespec64 *rqtp)
1314 struct restart_block *restart_block = ¤t->restart_block;
1318 * Diagnose required errors first.
1320 if (CPUCLOCK_PERTHREAD(which_clock) &&
1321 (CPUCLOCK_PID(which_clock) == 0 ||
1322 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1325 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1327 if (error == -ERESTART_RESTARTBLOCK) {
1329 if (flags & TIMER_ABSTIME)
1330 return -ERESTARTNOHAND;
1332 restart_block->fn = posix_cpu_nsleep_restart;
1333 restart_block->nanosleep.clockid = which_clock;
1338 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1340 clockid_t which_clock = restart_block->nanosleep.clockid;
1341 struct timespec64 t;
1343 t = ns_to_timespec64(restart_block->nanosleep.expires);
1345 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1348 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1349 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1351 static int process_cpu_clock_getres(const clockid_t which_clock,
1352 struct timespec64 *tp)
1354 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1356 static int process_cpu_clock_get(const clockid_t which_clock,
1357 struct timespec64 *tp)
1359 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1361 static int process_cpu_timer_create(struct k_itimer *timer)
1363 timer->it_clock = PROCESS_CLOCK;
1364 return posix_cpu_timer_create(timer);
1366 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1367 const struct timespec64 *rqtp)
1369 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1371 static int thread_cpu_clock_getres(const clockid_t which_clock,
1372 struct timespec64 *tp)
1374 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1376 static int thread_cpu_clock_get(const clockid_t which_clock,
1377 struct timespec64 *tp)
1379 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1381 static int thread_cpu_timer_create(struct k_itimer *timer)
1383 timer->it_clock = THREAD_CLOCK;
1384 return posix_cpu_timer_create(timer);
1387 const struct k_clock clock_posix_cpu = {
1388 .clock_getres = posix_cpu_clock_getres,
1389 .clock_set = posix_cpu_clock_set,
1390 .clock_get_timespec = posix_cpu_clock_get,
1391 .timer_create = posix_cpu_timer_create,
1392 .nsleep = posix_cpu_nsleep,
1393 .timer_set = posix_cpu_timer_set,
1394 .timer_del = posix_cpu_timer_del,
1395 .timer_get = posix_cpu_timer_get,
1396 .timer_rearm = posix_cpu_timer_rearm,
1399 const struct k_clock clock_process = {
1400 .clock_getres = process_cpu_clock_getres,
1401 .clock_get_timespec = process_cpu_clock_get,
1402 .timer_create = process_cpu_timer_create,
1403 .nsleep = process_cpu_nsleep,
1406 const struct k_clock clock_thread = {
1407 .clock_getres = thread_cpu_clock_getres,
1408 .clock_get_timespec = thread_cpu_clock_get,
1409 .timer_create = thread_cpu_timer_create,