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 pid *pid_for_clock(const clockid_t clock, bool gettime)
52 const bool thread = !!CPUCLOCK_PERTHREAD(clock);
53 const pid_t upid = CPUCLOCK_PID(clock);
56 if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
60 * If the encoded PID is 0, then the timer is targeted at current
61 * or the process to which current belongs.
64 return thread ? task_pid(current) : task_tgid(current);
66 pid = find_vpid(upid);
71 struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
72 return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
76 * For clock_gettime(PROCESS) allow finding the process by
77 * with the pid of the current task. The code needs the tgid
78 * of the process so that pid_task(pid, PIDTYPE_TGID) can be
79 * used to find the process.
81 if (gettime && (pid == task_pid(current)))
82 return task_tgid(current);
85 * For processes require that pid identifies a process.
87 return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
90 static inline int validate_clock_permissions(const clockid_t clock)
95 ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
101 static inline enum pid_type clock_pid_type(const clockid_t clock)
103 return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
106 static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
108 return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
112 * Update expiry time from increment, and increase overrun count,
113 * given the current clock sample.
115 static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
117 u64 delta, incr, expires = timer->it.cpu.node.expires;
120 if (!timer->it_interval)
126 incr = timer->it_interval;
127 delta = now + incr - expires;
129 /* Don't use (incr*2 < delta), incr*2 might overflow. */
130 for (i = 0; incr < delta - incr; i++)
133 for (; i >= 0; incr >>= 1, i--) {
137 timer->it.cpu.node.expires += incr;
138 timer->it_overrun += 1LL << i;
141 return timer->it.cpu.node.expires;
144 /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
145 static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
147 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
148 ~pct->bases[CPUCLOCK_VIRT].nextevt |
149 ~pct->bases[CPUCLOCK_SCHED].nextevt);
153 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
155 int error = validate_clock_permissions(which_clock);
159 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
160 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
162 * If sched_clock is using a cycle counter, we
163 * don't have any idea of its true resolution
164 * exported, but it is much more than 1s/HZ.
173 posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
175 int error = validate_clock_permissions(clock);
178 * You can never reset a CPU clock, but we check for other errors
179 * in the call before failing with EPERM.
181 return error ? : -EPERM;
185 * Sample a per-thread clock for the given task. clkid is validated.
187 static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
191 if (clkid == CPUCLOCK_SCHED)
192 return task_sched_runtime(p);
194 task_cputime(p, &utime, &stime);
198 return utime + stime;
207 static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
209 samples[CPUCLOCK_PROF] = stime + utime;
210 samples[CPUCLOCK_VIRT] = utime;
211 samples[CPUCLOCK_SCHED] = rtime;
214 static void task_sample_cputime(struct task_struct *p, u64 *samples)
218 task_cputime(p, &utime, &stime);
219 store_samples(samples, stime, utime, p->se.sum_exec_runtime);
222 static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
225 u64 stime, utime, rtime;
227 utime = atomic64_read(&at->utime);
228 stime = atomic64_read(&at->stime);
229 rtime = atomic64_read(&at->sum_exec_runtime);
230 store_samples(samples, stime, utime, rtime);
234 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
235 * to avoid race conditions with concurrent updates to cputime.
237 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
241 curr_cputime = atomic64_read(cputime);
242 if (sum_cputime > curr_cputime) {
243 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
248 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
249 struct task_cputime *sum)
251 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
252 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
253 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
257 * thread_group_sample_cputime - Sample cputime for a given task
258 * @tsk: Task for which cputime needs to be started
259 * @samples: Storage for time samples
261 * Called from sys_getitimer() to calculate the expiry time of an active
262 * timer. That means group cputime accounting is already active. Called
263 * with task sighand lock held.
265 * Updates @times with an uptodate sample of the thread group cputimes.
267 void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
269 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
270 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
272 WARN_ON_ONCE(!pct->timers_active);
274 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
278 * thread_group_start_cputime - Start cputime and return a sample
279 * @tsk: Task for which cputime needs to be started
280 * @samples: Storage for time samples
282 * The thread group cputime accounting is avoided when there are no posix
283 * CPU timers armed. Before starting a timer it's required to check whether
284 * the time accounting is active. If not, a full update of the atomic
285 * accounting store needs to be done and the accounting enabled.
287 * Updates @times with an uptodate sample of the thread group cputimes.
289 static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
291 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
292 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
294 lockdep_assert_task_sighand_held(tsk);
296 /* Check if cputimer isn't running. This is accessed without locking. */
297 if (!READ_ONCE(pct->timers_active)) {
298 struct task_cputime sum;
301 * The POSIX timer interface allows for absolute time expiry
302 * values through the TIMER_ABSTIME flag, therefore we have
303 * to synchronize the timer to the clock every time we start it.
305 thread_group_cputime(tsk, &sum);
306 update_gt_cputime(&cputimer->cputime_atomic, &sum);
309 * We're setting timers_active without a lock. Ensure this
310 * only gets written to in one operation. We set it after
311 * update_gt_cputime() as a small optimization, but
312 * barriers are not required because update_gt_cputime()
313 * can handle concurrent updates.
315 WRITE_ONCE(pct->timers_active, true);
317 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
320 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
322 struct task_cputime ct;
324 thread_group_cputime(tsk, &ct);
325 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
329 * Sample a process (thread group) clock for the given task clkid. If the
330 * group's cputime accounting is already enabled, read the atomic
331 * store. Otherwise a full update is required. clkid is already validated.
333 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
336 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
337 struct posix_cputimers *pct = &p->signal->posix_cputimers;
338 u64 samples[CPUCLOCK_MAX];
340 if (!READ_ONCE(pct->timers_active)) {
342 thread_group_start_cputime(p, samples);
344 __thread_group_cputime(p, samples);
346 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
349 return samples[clkid];
352 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
354 const clockid_t clkid = CPUCLOCK_WHICH(clock);
355 struct task_struct *tsk;
359 tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
365 if (CPUCLOCK_PERTHREAD(clock))
366 t = cpu_clock_sample(clkid, tsk);
368 t = cpu_clock_sample_group(clkid, tsk, false);
371 *tp = ns_to_timespec64(t);
376 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
377 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
378 * new timer already all-zeros initialized.
380 static int posix_cpu_timer_create(struct k_itimer *new_timer)
382 static struct lock_class_key posix_cpu_timers_key;
386 pid = pid_for_clock(new_timer->it_clock, false);
393 * If posix timer expiry is handled in task work context then
394 * timer::it_lock can be taken without disabling interrupts as all
395 * other locking happens in task context. This requires a separate
396 * lock class key otherwise regular posix timer expiry would record
397 * the lock class being taken in interrupt context and generate a
398 * false positive warning.
400 if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
401 lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
403 new_timer->kclock = &clock_posix_cpu;
404 timerqueue_init(&new_timer->it.cpu.node);
405 new_timer->it.cpu.pid = get_pid(pid);
410 static struct posix_cputimer_base *timer_base(struct k_itimer *timer,
411 struct task_struct *tsk)
413 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
415 if (CPUCLOCK_PERTHREAD(timer->it_clock))
416 return tsk->posix_cputimers.bases + clkidx;
418 return tsk->signal->posix_cputimers.bases + clkidx;
422 * Dequeue the timer and reset the base if it was its earliest expiration.
423 * It makes sure the next tick recalculates the base next expiration so we
424 * don't keep the costly process wide cputime counter around for a random
425 * amount of time, along with the tick dependency.
427 * If another timer gets queued between this and the next tick, its
428 * expiration will update the base next event if necessary on the next
431 static void disarm_timer(struct k_itimer *timer, struct task_struct *p)
433 struct cpu_timer *ctmr = &timer->it.cpu;
434 struct posix_cputimer_base *base;
436 if (!cpu_timer_dequeue(ctmr))
439 base = timer_base(timer, p);
440 if (cpu_timer_getexpires(ctmr) == base->nextevt)
446 * Clean up a CPU-clock timer that is about to be destroyed.
447 * This is called from timer deletion with the timer already locked.
448 * If we return TIMER_RETRY, it's necessary to release the timer's lock
449 * and try again. (This happens when the timer is in the middle of firing.)
451 static int posix_cpu_timer_del(struct k_itimer *timer)
453 struct cpu_timer *ctmr = &timer->it.cpu;
454 struct sighand_struct *sighand;
455 struct task_struct *p;
460 p = cpu_timer_task_rcu(timer);
465 * Protect against sighand release/switch in exit/exec and process/
466 * thread timer list entry concurrent read/writes.
468 sighand = lock_task_sighand(p, &flags);
469 if (unlikely(sighand == NULL)) {
471 * This raced with the reaping of the task. The exit cleanup
472 * should have removed this timer from the timer queue.
474 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
476 if (timer->it.cpu.firing)
479 disarm_timer(timer, p);
481 unlock_task_sighand(p, &flags);
492 static void cleanup_timerqueue(struct timerqueue_head *head)
494 struct timerqueue_node *node;
495 struct cpu_timer *ctmr;
497 while ((node = timerqueue_getnext(head))) {
498 timerqueue_del(head, node);
499 ctmr = container_of(node, struct cpu_timer, node);
505 * Clean out CPU timers which are still armed when a thread exits. The
506 * timers are only removed from the list. No other updates are done. The
507 * corresponding posix timers are still accessible, but cannot be rearmed.
509 * This must be called with the siglock held.
511 static void cleanup_timers(struct posix_cputimers *pct)
513 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
514 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
515 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
519 * These are both called with the siglock held, when the current thread
520 * is being reaped. When the final (leader) thread in the group is reaped,
521 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
523 void posix_cpu_timers_exit(struct task_struct *tsk)
525 cleanup_timers(&tsk->posix_cputimers);
527 void posix_cpu_timers_exit_group(struct task_struct *tsk)
529 cleanup_timers(&tsk->signal->posix_cputimers);
533 * Insert the timer on the appropriate list before any timers that
534 * expire later. This must be called with the sighand lock held.
536 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
538 struct posix_cputimer_base *base = timer_base(timer, p);
539 struct cpu_timer *ctmr = &timer->it.cpu;
540 u64 newexp = cpu_timer_getexpires(ctmr);
542 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
546 * We are the new earliest-expiring POSIX 1.b timer, hence
547 * need to update expiration cache. Take into account that
548 * for process timers we share expiration cache with itimers
549 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
551 if (newexp < base->nextevt)
552 base->nextevt = newexp;
554 if (CPUCLOCK_PERTHREAD(timer->it_clock))
555 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
557 tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER);
561 * The timer is locked, fire it and arrange for its reload.
563 static void cpu_timer_fire(struct k_itimer *timer)
565 struct cpu_timer *ctmr = &timer->it.cpu;
567 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
569 * User don't want any signal.
571 cpu_timer_setexpires(ctmr, 0);
572 } else if (unlikely(timer->sigq == NULL)) {
574 * This a special case for clock_nanosleep,
575 * not a normal timer from sys_timer_create.
577 wake_up_process(timer->it_process);
578 cpu_timer_setexpires(ctmr, 0);
579 } else if (!timer->it_interval) {
581 * One-shot timer. Clear it as soon as it's fired.
583 posix_timer_event(timer, 0);
584 cpu_timer_setexpires(ctmr, 0);
585 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
587 * The signal did not get queued because the signal
588 * was ignored, so we won't get any callback to
589 * reload the timer. But we need to keep it
590 * ticking in case the signal is deliverable next time.
592 posix_cpu_timer_rearm(timer);
593 ++timer->it_requeue_pending;
598 * Guts of sys_timer_settime for CPU timers.
599 * This is called with the timer locked and interrupts disabled.
600 * If we return TIMER_RETRY, it's necessary to release the timer's lock
601 * and try again. (This happens when the timer is in the middle of firing.)
603 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
604 struct itimerspec64 *new, struct itimerspec64 *old)
606 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
607 u64 old_expires, new_expires, old_incr, val;
608 struct cpu_timer *ctmr = &timer->it.cpu;
609 struct sighand_struct *sighand;
610 struct task_struct *p;
615 p = cpu_timer_task_rcu(timer);
618 * If p has just been reaped, we can no
619 * longer get any information about it at all.
626 * Use the to_ktime conversion because that clamps the maximum
627 * value to KTIME_MAX and avoid multiplication overflows.
629 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
632 * Protect against sighand release/switch in exit/exec and p->cpu_timers
633 * and p->signal->cpu_timers read/write in arm_timer()
635 sighand = lock_task_sighand(p, &flags);
637 * If p has just been reaped, we can no
638 * longer get any information about it at all.
640 if (unlikely(sighand == NULL)) {
646 * Disarm any old timer after extracting its expiry time.
648 old_incr = timer->it_interval;
649 old_expires = cpu_timer_getexpires(ctmr);
651 if (unlikely(timer->it.cpu.firing)) {
652 timer->it.cpu.firing = -1;
655 cpu_timer_dequeue(ctmr);
659 * We need to sample the current value to convert the new
660 * value from to relative and absolute, and to convert the
661 * old value from absolute to relative. To set a process
662 * timer, we need a sample to balance the thread expiry
663 * times (in arm_timer). With an absolute time, we must
664 * check if it's already passed. In short, we need a sample.
666 if (CPUCLOCK_PERTHREAD(timer->it_clock))
667 val = cpu_clock_sample(clkid, p);
669 val = cpu_clock_sample_group(clkid, p, true);
672 if (old_expires == 0) {
673 old->it_value.tv_sec = 0;
674 old->it_value.tv_nsec = 0;
677 * Update the timer in case it has overrun already.
678 * If it has, we'll report it as having overrun and
679 * with the next reloaded timer already ticking,
680 * though we are swallowing that pending
681 * notification here to install the new setting.
683 u64 exp = bump_cpu_timer(timer, val);
686 old_expires = exp - val;
687 old->it_value = ns_to_timespec64(old_expires);
689 old->it_value.tv_nsec = 1;
690 old->it_value.tv_sec = 0;
697 * We are colliding with the timer actually firing.
698 * Punt after filling in the timer's old value, and
699 * disable this firing since we are already reporting
700 * it as an overrun (thanks to bump_cpu_timer above).
702 unlock_task_sighand(p, &flags);
706 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
711 * Install the new expiry time (or zero).
712 * For a timer with no notification action, we don't actually
713 * arm the timer (we'll just fake it for timer_gettime).
715 cpu_timer_setexpires(ctmr, new_expires);
716 if (new_expires != 0 && val < new_expires) {
720 unlock_task_sighand(p, &flags);
722 * Install the new reload setting, and
723 * set up the signal and overrun bookkeeping.
725 timer->it_interval = timespec64_to_ktime(new->it_interval);
728 * This acts as a modification timestamp for the timer,
729 * so any automatic reload attempt will punt on seeing
730 * that we have reset the timer manually.
732 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
734 timer->it_overrun_last = 0;
735 timer->it_overrun = -1;
737 if (new_expires != 0 && !(val < new_expires)) {
739 * The designated time already passed, so we notify
740 * immediately, even if the thread never runs to
741 * accumulate more time on this clock.
743 cpu_timer_fire(timer);
748 old->it_interval = ns_to_timespec64(old_incr);
753 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
755 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
756 struct cpu_timer *ctmr = &timer->it.cpu;
757 u64 now, expires = cpu_timer_getexpires(ctmr);
758 struct task_struct *p;
761 p = cpu_timer_task_rcu(timer);
766 * Easy part: convert the reload time.
768 itp->it_interval = ktime_to_timespec64(timer->it_interval);
774 * Sample the clock to take the difference with the expiry time.
776 if (CPUCLOCK_PERTHREAD(timer->it_clock))
777 now = cpu_clock_sample(clkid, p);
779 now = cpu_clock_sample_group(clkid, p, false);
782 itp->it_value = ns_to_timespec64(expires - now);
785 * The timer should have expired already, but the firing
786 * hasn't taken place yet. Say it's just about to expire.
788 itp->it_value.tv_nsec = 1;
789 itp->it_value.tv_sec = 0;
795 #define MAX_COLLECTED 20
797 static u64 collect_timerqueue(struct timerqueue_head *head,
798 struct list_head *firing, u64 now)
800 struct timerqueue_node *next;
803 while ((next = timerqueue_getnext(head))) {
804 struct cpu_timer *ctmr;
807 ctmr = container_of(next, struct cpu_timer, node);
808 expires = cpu_timer_getexpires(ctmr);
809 /* Limit the number of timers to expire at once */
810 if (++i == MAX_COLLECTED || now < expires)
814 cpu_timer_dequeue(ctmr);
815 list_add_tail(&ctmr->elist, firing);
821 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
822 struct list_head *firing)
824 struct posix_cputimer_base *base = pct->bases;
827 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
828 base->nextevt = collect_timerqueue(&base->tqhead, firing,
833 static inline void check_dl_overrun(struct task_struct *tsk)
835 if (tsk->dl.dl_overrun) {
836 tsk->dl.dl_overrun = 0;
837 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
841 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
846 if (print_fatal_signals) {
847 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
848 rt ? "RT" : "CPU", hard ? "hard" : "soft",
849 current->comm, task_pid_nr(current));
851 __group_send_sig_info(signo, SEND_SIG_PRIV, current);
856 * Check for any per-thread CPU timers that have fired and move them off
857 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
858 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
860 static void check_thread_timers(struct task_struct *tsk,
861 struct list_head *firing)
863 struct posix_cputimers *pct = &tsk->posix_cputimers;
864 u64 samples[CPUCLOCK_MAX];
868 check_dl_overrun(tsk);
870 if (expiry_cache_is_inactive(pct))
873 task_sample_cputime(tsk, samples);
874 collect_posix_cputimers(pct, samples, firing);
877 * Check for the special case thread timers.
879 soft = task_rlimit(tsk, RLIMIT_RTTIME);
880 if (soft != RLIM_INFINITY) {
881 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
882 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
883 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
885 /* At the hard limit, send SIGKILL. No further action. */
886 if (hard != RLIM_INFINITY &&
887 check_rlimit(rttime, hard, SIGKILL, true, true))
890 /* At the soft limit, send a SIGXCPU every second */
891 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
892 soft += USEC_PER_SEC;
893 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
897 if (expiry_cache_is_inactive(pct))
898 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
901 static inline void stop_process_timers(struct signal_struct *sig)
903 struct posix_cputimers *pct = &sig->posix_cputimers;
905 /* Turn off the active flag. This is done without locking. */
906 WRITE_ONCE(pct->timers_active, false);
907 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
910 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
911 u64 *expires, u64 cur_time, int signo)
916 if (cur_time >= it->expires) {
918 it->expires += it->incr;
922 trace_itimer_expire(signo == SIGPROF ?
923 ITIMER_PROF : ITIMER_VIRTUAL,
924 task_tgid(tsk), cur_time);
925 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
928 if (it->expires && it->expires < *expires)
929 *expires = it->expires;
933 * Check for any per-thread CPU timers that have fired and move them
934 * off the tsk->*_timers list onto the firing list. Per-thread timers
935 * have already been taken off.
937 static void check_process_timers(struct task_struct *tsk,
938 struct list_head *firing)
940 struct signal_struct *const sig = tsk->signal;
941 struct posix_cputimers *pct = &sig->posix_cputimers;
942 u64 samples[CPUCLOCK_MAX];
946 * If there are no active process wide timers (POSIX 1.b, itimers,
947 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
948 * processing when there is already another task handling them.
950 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
954 * Signify that a thread is checking for process timers.
955 * Write access to this field is protected by the sighand lock.
957 pct->expiry_active = true;
960 * Collect the current process totals. Group accounting is active
961 * so the sample can be taken directly.
963 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
964 collect_posix_cputimers(pct, samples, firing);
967 * Check for the special case process timers.
969 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
970 &pct->bases[CPUCLOCK_PROF].nextevt,
971 samples[CPUCLOCK_PROF], SIGPROF);
972 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
973 &pct->bases[CPUCLOCK_VIRT].nextevt,
974 samples[CPUCLOCK_VIRT], SIGVTALRM);
976 soft = task_rlimit(tsk, RLIMIT_CPU);
977 if (soft != RLIM_INFINITY) {
978 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
979 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
980 u64 ptime = samples[CPUCLOCK_PROF];
981 u64 softns = (u64)soft * NSEC_PER_SEC;
982 u64 hardns = (u64)hard * NSEC_PER_SEC;
984 /* At the hard limit, send SIGKILL. No further action. */
985 if (hard != RLIM_INFINITY &&
986 check_rlimit(ptime, hardns, SIGKILL, false, true))
989 /* At the soft limit, send a SIGXCPU every second */
990 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
991 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
992 softns += NSEC_PER_SEC;
995 /* Update the expiry cache */
996 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
997 pct->bases[CPUCLOCK_PROF].nextevt = softns;
1000 if (expiry_cache_is_inactive(pct))
1001 stop_process_timers(sig);
1003 pct->expiry_active = false;
1007 * This is called from the signal code (via posixtimer_rearm)
1008 * when the last timer signal was delivered and we have to reload the timer.
1010 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1012 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
1013 struct task_struct *p;
1014 struct sighand_struct *sighand;
1015 unsigned long flags;
1019 p = cpu_timer_task_rcu(timer);
1023 /* Protect timer list r/w in arm_timer() */
1024 sighand = lock_task_sighand(p, &flags);
1025 if (unlikely(sighand == NULL))
1029 * Fetch the current sample and update the timer's expiry time.
1031 if (CPUCLOCK_PERTHREAD(timer->it_clock))
1032 now = cpu_clock_sample(clkid, p);
1034 now = cpu_clock_sample_group(clkid, p, true);
1036 bump_cpu_timer(timer, now);
1039 * Now re-arm for the new expiry time.
1041 arm_timer(timer, p);
1042 unlock_task_sighand(p, &flags);
1048 * task_cputimers_expired - Check whether posix CPU timers are expired
1050 * @samples: Array of current samples for the CPUCLOCK clocks
1051 * @pct: Pointer to a posix_cputimers container
1053 * Returns true if any member of @samples is greater than the corresponding
1054 * member of @pct->bases[CLK].nextevt. False otherwise
1057 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1061 for (i = 0; i < CPUCLOCK_MAX; i++) {
1062 if (samples[i] >= pct->bases[i].nextevt)
1069 * fastpath_timer_check - POSIX CPU timers fast path.
1071 * @tsk: The task (thread) being checked.
1073 * Check the task and thread group timers. If both are zero (there are no
1074 * timers set) return false. Otherwise snapshot the task and thread group
1075 * timers and compare them with the corresponding expiration times. Return
1076 * true if a timer has expired, else return false.
1078 static inline bool fastpath_timer_check(struct task_struct *tsk)
1080 struct posix_cputimers *pct = &tsk->posix_cputimers;
1081 struct signal_struct *sig;
1083 if (!expiry_cache_is_inactive(pct)) {
1084 u64 samples[CPUCLOCK_MAX];
1086 task_sample_cputime(tsk, samples);
1087 if (task_cputimers_expired(samples, pct))
1092 pct = &sig->posix_cputimers;
1094 * Check if thread group timers expired when timers are active and
1095 * no other thread in the group is already handling expiry for
1096 * thread group cputimers. These fields are read without the
1097 * sighand lock. However, this is fine because this is meant to be
1098 * a fastpath heuristic to determine whether we should try to
1099 * acquire the sighand lock to handle timer expiry.
1101 * In the worst case scenario, if concurrently timers_active is set
1102 * or expiry_active is cleared, but the current thread doesn't see
1103 * the change yet, the timer checks are delayed until the next
1104 * thread in the group gets a scheduler interrupt to handle the
1105 * timer. This isn't an issue in practice because these types of
1106 * delays with signals actually getting sent are expected.
1108 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1109 u64 samples[CPUCLOCK_MAX];
1111 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1114 if (task_cputimers_expired(samples, pct))
1118 if (dl_task(tsk) && tsk->dl.dl_overrun)
1124 static void handle_posix_cpu_timers(struct task_struct *tsk);
1126 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1127 static void posix_cpu_timers_work(struct callback_head *work)
1129 handle_posix_cpu_timers(current);
1133 * Initialize posix CPU timers task work in init task. Out of line to
1134 * keep the callback static and to avoid header recursion hell.
1136 void __init posix_cputimers_init_work(void)
1138 init_task_work(¤t->posix_cputimers_work.work,
1139 posix_cpu_timers_work);
1143 * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1144 * in hard interrupt context or in task context with interrupts
1145 * disabled. Aside of that the writer/reader interaction is always in the
1146 * context of the current task, which means they are strict per CPU.
1148 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1150 return tsk->posix_cputimers_work.scheduled;
1153 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1155 if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1158 /* Schedule task work to actually expire the timers */
1159 tsk->posix_cputimers_work.scheduled = true;
1160 task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1163 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1164 unsigned long start)
1169 * On !RT kernels interrupts are disabled while collecting expired
1170 * timers, so no tick can happen and the fast path check can be
1171 * reenabled without further checks.
1173 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1174 tsk->posix_cputimers_work.scheduled = false;
1179 * On RT enabled kernels ticks can happen while the expired timers
1180 * are collected under sighand lock. But any tick which observes
1181 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1182 * checks. So reenabling the tick work has do be done carefully:
1184 * Disable interrupts and run the fast path check if jiffies have
1185 * advanced since the collecting of expired timers started. If
1186 * jiffies have not advanced or the fast path check did not find
1187 * newly expired timers, reenable the fast path check in the timer
1188 * interrupt. If there are newly expired timers, return false and
1189 * let the collection loop repeat.
1191 local_irq_disable();
1192 if (start != jiffies && fastpath_timer_check(tsk))
1195 tsk->posix_cputimers_work.scheduled = false;
1200 #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1201 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1203 lockdep_posixtimer_enter();
1204 handle_posix_cpu_timers(tsk);
1205 lockdep_posixtimer_exit();
1208 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1213 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1214 unsigned long start)
1218 #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1220 static void handle_posix_cpu_timers(struct task_struct *tsk)
1222 struct k_itimer *timer, *next;
1223 unsigned long flags, start;
1226 if (!lock_task_sighand(tsk, &flags))
1231 * On RT locking sighand lock does not disable interrupts,
1232 * so this needs to be careful vs. ticks. Store the current
1235 start = READ_ONCE(jiffies);
1239 * Here we take off tsk->signal->cpu_timers[N] and
1240 * tsk->cpu_timers[N] all the timers that are firing, and
1241 * put them on the firing list.
1243 check_thread_timers(tsk, &firing);
1245 check_process_timers(tsk, &firing);
1248 * The above timer checks have updated the expiry cache and
1249 * because nothing can have queued or modified timers after
1250 * sighand lock was taken above it is guaranteed to be
1251 * consistent. So the next timer interrupt fastpath check
1252 * will find valid data.
1254 * If timer expiry runs in the timer interrupt context then
1255 * the loop is not relevant as timers will be directly
1256 * expired in interrupt context. The stub function below
1257 * returns always true which allows the compiler to
1258 * optimize the loop out.
1260 * If timer expiry is deferred to task work context then
1261 * the following rules apply:
1263 * - On !RT kernels no tick can have happened on this CPU
1264 * after sighand lock was acquired because interrupts are
1265 * disabled. So reenabling task work before dropping
1266 * sighand lock and reenabling interrupts is race free.
1268 * - On RT kernels ticks might have happened but the tick
1269 * work ignored posix CPU timer handling because the
1270 * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1271 * must be done very carefully including a check whether
1272 * ticks have happened since the start of the timer
1273 * expiry checks. posix_cpu_timers_enable_work() takes
1274 * care of that and eventually lets the expiry checks
1277 } while (!posix_cpu_timers_enable_work(tsk, start));
1280 * We must release sighand lock before taking any timer's lock.
1281 * There is a potential race with timer deletion here, as the
1282 * siglock now protects our private firing list. We have set
1283 * the firing flag in each timer, so that a deletion attempt
1284 * that gets the timer lock before we do will give it up and
1285 * spin until we've taken care of that timer below.
1287 unlock_task_sighand(tsk, &flags);
1290 * Now that all the timers on our list have the firing flag,
1291 * no one will touch their list entries but us. We'll take
1292 * each timer's lock before clearing its firing flag, so no
1293 * timer call will interfere.
1295 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1299 * spin_lock() is sufficient here even independent of the
1300 * expiry context. If expiry happens in hard interrupt
1301 * context it's obvious. For task work context it's safe
1302 * because all other operations on timer::it_lock happen in
1303 * task context (syscall or exit).
1305 spin_lock(&timer->it_lock);
1306 list_del_init(&timer->it.cpu.elist);
1307 cpu_firing = timer->it.cpu.firing;
1308 timer->it.cpu.firing = 0;
1310 * The firing flag is -1 if we collided with a reset
1311 * of the timer, which already reported this
1312 * almost-firing as an overrun. So don't generate an event.
1314 if (likely(cpu_firing >= 0))
1315 cpu_timer_fire(timer);
1316 spin_unlock(&timer->it_lock);
1321 * This is called from the timer interrupt handler. The irq handler has
1322 * already updated our counts. We need to check if any timers fire now.
1323 * Interrupts are disabled.
1325 void run_posix_cpu_timers(void)
1327 struct task_struct *tsk = current;
1329 lockdep_assert_irqs_disabled();
1332 * If the actual expiry is deferred to task work context and the
1333 * work is already scheduled there is no point to do anything here.
1335 if (posix_cpu_timers_work_scheduled(tsk))
1339 * The fast path checks that there are no expired thread or thread
1340 * group timers. If that's so, just return.
1342 if (!fastpath_timer_check(tsk))
1345 __run_posix_cpu_timers(tsk);
1349 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1350 * The tsk->sighand->siglock must be held by the caller.
1352 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1353 u64 *newval, u64 *oldval)
1357 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1360 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1361 now = cpu_clock_sample_group(clkid, tsk, true);
1365 * We are setting itimer. The *oldval is absolute and we update
1366 * it to be relative, *newval argument is relative and we update
1367 * it to be absolute.
1370 if (*oldval <= now) {
1371 /* Just about to fire. */
1372 *oldval = TICK_NSEC;
1382 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1383 * expiry cache is also used by RLIMIT_CPU!.
1385 if (*newval < *nextevt)
1388 tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER);
1391 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1392 const struct timespec64 *rqtp)
1394 struct itimerspec64 it;
1395 struct k_itimer timer;
1400 * Set up a temporary timer and then wait for it to go off.
1402 memset(&timer, 0, sizeof timer);
1403 spin_lock_init(&timer.it_lock);
1404 timer.it_clock = which_clock;
1405 timer.it_overrun = -1;
1406 error = posix_cpu_timer_create(&timer);
1407 timer.it_process = current;
1410 static struct itimerspec64 zero_it;
1411 struct restart_block *restart;
1413 memset(&it, 0, sizeof(it));
1414 it.it_value = *rqtp;
1416 spin_lock_irq(&timer.it_lock);
1417 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1419 spin_unlock_irq(&timer.it_lock);
1423 while (!signal_pending(current)) {
1424 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1426 * Our timer fired and was reset, below
1427 * deletion can not fail.
1429 posix_cpu_timer_del(&timer);
1430 spin_unlock_irq(&timer.it_lock);
1435 * Block until cpu_timer_fire (or a signal) wakes us.
1437 __set_current_state(TASK_INTERRUPTIBLE);
1438 spin_unlock_irq(&timer.it_lock);
1440 spin_lock_irq(&timer.it_lock);
1444 * We were interrupted by a signal.
1446 expires = cpu_timer_getexpires(&timer.it.cpu);
1447 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1450 * Timer is now unarmed, deletion can not fail.
1452 posix_cpu_timer_del(&timer);
1454 spin_unlock_irq(&timer.it_lock);
1456 while (error == TIMER_RETRY) {
1458 * We need to handle case when timer was or is in the
1459 * middle of firing. In other cases we already freed
1462 spin_lock_irq(&timer.it_lock);
1463 error = posix_cpu_timer_del(&timer);
1464 spin_unlock_irq(&timer.it_lock);
1467 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1469 * It actually did fire already.
1474 error = -ERESTART_RESTARTBLOCK;
1476 * Report back to the user the time still remaining.
1478 restart = ¤t->restart_block;
1479 restart->nanosleep.expires = expires;
1480 if (restart->nanosleep.type != TT_NONE)
1481 error = nanosleep_copyout(restart, &it.it_value);
1487 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1489 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1490 const struct timespec64 *rqtp)
1492 struct restart_block *restart_block = ¤t->restart_block;
1496 * Diagnose required errors first.
1498 if (CPUCLOCK_PERTHREAD(which_clock) &&
1499 (CPUCLOCK_PID(which_clock) == 0 ||
1500 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1503 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1505 if (error == -ERESTART_RESTARTBLOCK) {
1507 if (flags & TIMER_ABSTIME)
1508 return -ERESTARTNOHAND;
1510 restart_block->nanosleep.clockid = which_clock;
1511 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1516 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1518 clockid_t which_clock = restart_block->nanosleep.clockid;
1519 struct timespec64 t;
1521 t = ns_to_timespec64(restart_block->nanosleep.expires);
1523 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1526 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1527 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1529 static int process_cpu_clock_getres(const clockid_t which_clock,
1530 struct timespec64 *tp)
1532 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1534 static int process_cpu_clock_get(const clockid_t which_clock,
1535 struct timespec64 *tp)
1537 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1539 static int process_cpu_timer_create(struct k_itimer *timer)
1541 timer->it_clock = PROCESS_CLOCK;
1542 return posix_cpu_timer_create(timer);
1544 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1545 const struct timespec64 *rqtp)
1547 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1549 static int thread_cpu_clock_getres(const clockid_t which_clock,
1550 struct timespec64 *tp)
1552 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1554 static int thread_cpu_clock_get(const clockid_t which_clock,
1555 struct timespec64 *tp)
1557 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1559 static int thread_cpu_timer_create(struct k_itimer *timer)
1561 timer->it_clock = THREAD_CLOCK;
1562 return posix_cpu_timer_create(timer);
1565 const struct k_clock clock_posix_cpu = {
1566 .clock_getres = posix_cpu_clock_getres,
1567 .clock_set = posix_cpu_clock_set,
1568 .clock_get_timespec = posix_cpu_clock_get,
1569 .timer_create = posix_cpu_timer_create,
1570 .nsleep = posix_cpu_nsleep,
1571 .timer_set = posix_cpu_timer_set,
1572 .timer_del = posix_cpu_timer_del,
1573 .timer_get = posix_cpu_timer_get,
1574 .timer_rearm = posix_cpu_timer_rearm,
1577 const struct k_clock clock_process = {
1578 .clock_getres = process_cpu_clock_getres,
1579 .clock_get_timespec = process_cpu_clock_get,
1580 .timer_create = process_cpu_timer_create,
1581 .nsleep = process_cpu_nsleep,
1584 const struct k_clock clock_thread = {
1585 .clock_getres = thread_cpu_clock_getres,
1586 .clock_get_timespec = thread_cpu_clock_get,
1587 .timer_create = thread_cpu_timer_create,