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 accouting 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 /* Check if cputimer isn't running. This is accessed without locking. */
295 if (!READ_ONCE(pct->timers_active)) {
296 struct task_cputime sum;
299 * The POSIX timer interface allows for absolute time expiry
300 * values through the TIMER_ABSTIME flag, therefore we have
301 * to synchronize the timer to the clock every time we start it.
303 thread_group_cputime(tsk, &sum);
304 update_gt_cputime(&cputimer->cputime_atomic, &sum);
307 * We're setting timers_active without a lock. Ensure this
308 * only gets written to in one operation. We set it after
309 * update_gt_cputime() as a small optimization, but
310 * barriers are not required because update_gt_cputime()
311 * can handle concurrent updates.
313 WRITE_ONCE(pct->timers_active, true);
315 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
318 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
320 struct task_cputime ct;
322 thread_group_cputime(tsk, &ct);
323 store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
327 * Sample a process (thread group) clock for the given task clkid. If the
328 * group's cputime accounting is already enabled, read the atomic
329 * store. Otherwise a full update is required. clkid is already validated.
331 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
334 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
335 struct posix_cputimers *pct = &p->signal->posix_cputimers;
336 u64 samples[CPUCLOCK_MAX];
338 if (!READ_ONCE(pct->timers_active)) {
340 thread_group_start_cputime(p, samples);
342 __thread_group_cputime(p, samples);
344 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
347 return samples[clkid];
350 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
352 const clockid_t clkid = CPUCLOCK_WHICH(clock);
353 struct task_struct *tsk;
357 tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
363 if (CPUCLOCK_PERTHREAD(clock))
364 t = cpu_clock_sample(clkid, tsk);
366 t = cpu_clock_sample_group(clkid, tsk, false);
369 *tp = ns_to_timespec64(t);
374 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
375 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
376 * new timer already all-zeros initialized.
378 static int posix_cpu_timer_create(struct k_itimer *new_timer)
380 static struct lock_class_key posix_cpu_timers_key;
384 pid = pid_for_clock(new_timer->it_clock, false);
391 * If posix timer expiry is handled in task work context then
392 * timer::it_lock can be taken without disabling interrupts as all
393 * other locking happens in task context. This requires a seperate
394 * lock class key otherwise regular posix timer expiry would record
395 * the lock class being taken in interrupt context and generate a
396 * false positive warning.
398 if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
399 lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
401 new_timer->kclock = &clock_posix_cpu;
402 timerqueue_init(&new_timer->it.cpu.node);
403 new_timer->it.cpu.pid = get_pid(pid);
409 * Clean up a CPU-clock timer that is about to be destroyed.
410 * This is called from timer deletion with the timer already locked.
411 * If we return TIMER_RETRY, it's necessary to release the timer's lock
412 * and try again. (This happens when the timer is in the middle of firing.)
414 static int posix_cpu_timer_del(struct k_itimer *timer)
416 struct cpu_timer *ctmr = &timer->it.cpu;
417 struct sighand_struct *sighand;
418 struct task_struct *p;
423 p = cpu_timer_task_rcu(timer);
428 * Protect against sighand release/switch in exit/exec and process/
429 * thread timer list entry concurrent read/writes.
431 sighand = lock_task_sighand(p, &flags);
432 if (unlikely(sighand == NULL)) {
434 * This raced with the reaping of the task. The exit cleanup
435 * should have removed this timer from the timer queue.
437 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
439 if (timer->it.cpu.firing)
442 cpu_timer_dequeue(ctmr);
444 unlock_task_sighand(p, &flags);
455 static void cleanup_timerqueue(struct timerqueue_head *head)
457 struct timerqueue_node *node;
458 struct cpu_timer *ctmr;
460 while ((node = timerqueue_getnext(head))) {
461 timerqueue_del(head, node);
462 ctmr = container_of(node, struct cpu_timer, node);
468 * Clean out CPU timers which are still armed when a thread exits. The
469 * timers are only removed from the list. No other updates are done. The
470 * corresponding posix timers are still accessible, but cannot be rearmed.
472 * This must be called with the siglock held.
474 static void cleanup_timers(struct posix_cputimers *pct)
476 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
477 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
478 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
482 * These are both called with the siglock held, when the current thread
483 * is being reaped. When the final (leader) thread in the group is reaped,
484 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
486 void posix_cpu_timers_exit(struct task_struct *tsk)
488 cleanup_timers(&tsk->posix_cputimers);
490 void posix_cpu_timers_exit_group(struct task_struct *tsk)
492 cleanup_timers(&tsk->signal->posix_cputimers);
496 * Insert the timer on the appropriate list before any timers that
497 * expire later. This must be called with the sighand lock held.
499 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
501 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
502 struct cpu_timer *ctmr = &timer->it.cpu;
503 u64 newexp = cpu_timer_getexpires(ctmr);
504 struct posix_cputimer_base *base;
506 if (CPUCLOCK_PERTHREAD(timer->it_clock))
507 base = p->posix_cputimers.bases + clkidx;
509 base = p->signal->posix_cputimers.bases + clkidx;
511 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
515 * We are the new earliest-expiring POSIX 1.b timer, hence
516 * need to update expiration cache. Take into account that
517 * for process timers we share expiration cache with itimers
518 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
520 if (newexp < base->nextevt)
521 base->nextevt = newexp;
523 if (CPUCLOCK_PERTHREAD(timer->it_clock))
524 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
526 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
530 * The timer is locked, fire it and arrange for its reload.
532 static void cpu_timer_fire(struct k_itimer *timer)
534 struct cpu_timer *ctmr = &timer->it.cpu;
536 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
538 * User don't want any signal.
540 cpu_timer_setexpires(ctmr, 0);
541 } else if (unlikely(timer->sigq == NULL)) {
543 * This a special case for clock_nanosleep,
544 * not a normal timer from sys_timer_create.
546 wake_up_process(timer->it_process);
547 cpu_timer_setexpires(ctmr, 0);
548 } else if (!timer->it_interval) {
550 * One-shot timer. Clear it as soon as it's fired.
552 posix_timer_event(timer, 0);
553 cpu_timer_setexpires(ctmr, 0);
554 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
556 * The signal did not get queued because the signal
557 * was ignored, so we won't get any callback to
558 * reload the timer. But we need to keep it
559 * ticking in case the signal is deliverable next time.
561 posix_cpu_timer_rearm(timer);
562 ++timer->it_requeue_pending;
567 * Guts of sys_timer_settime for CPU timers.
568 * This is called with the timer locked and interrupts disabled.
569 * If we return TIMER_RETRY, it's necessary to release the timer's lock
570 * and try again. (This happens when the timer is in the middle of firing.)
572 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
573 struct itimerspec64 *new, struct itimerspec64 *old)
575 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
576 u64 old_expires, new_expires, old_incr, val;
577 struct cpu_timer *ctmr = &timer->it.cpu;
578 struct sighand_struct *sighand;
579 struct task_struct *p;
584 p = cpu_timer_task_rcu(timer);
587 * If p has just been reaped, we can no
588 * longer get any information about it at all.
595 * Use the to_ktime conversion because that clamps the maximum
596 * value to KTIME_MAX and avoid multiplication overflows.
598 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
601 * Protect against sighand release/switch in exit/exec and p->cpu_timers
602 * and p->signal->cpu_timers read/write in arm_timer()
604 sighand = lock_task_sighand(p, &flags);
606 * If p has just been reaped, we can no
607 * longer get any information about it at all.
609 if (unlikely(sighand == NULL)) {
615 * Disarm any old timer after extracting its expiry time.
617 old_incr = timer->it_interval;
618 old_expires = cpu_timer_getexpires(ctmr);
620 if (unlikely(timer->it.cpu.firing)) {
621 timer->it.cpu.firing = -1;
624 cpu_timer_dequeue(ctmr);
628 * We need to sample the current value to convert the new
629 * value from to relative and absolute, and to convert the
630 * old value from absolute to relative. To set a process
631 * timer, we need a sample to balance the thread expiry
632 * times (in arm_timer). With an absolute time, we must
633 * check if it's already passed. In short, we need a sample.
635 if (CPUCLOCK_PERTHREAD(timer->it_clock))
636 val = cpu_clock_sample(clkid, p);
638 val = cpu_clock_sample_group(clkid, p, true);
641 if (old_expires == 0) {
642 old->it_value.tv_sec = 0;
643 old->it_value.tv_nsec = 0;
646 * Update the timer in case it has overrun already.
647 * If it has, we'll report it as having overrun and
648 * with the next reloaded timer already ticking,
649 * though we are swallowing that pending
650 * notification here to install the new setting.
652 u64 exp = bump_cpu_timer(timer, val);
655 old_expires = exp - val;
656 old->it_value = ns_to_timespec64(old_expires);
658 old->it_value.tv_nsec = 1;
659 old->it_value.tv_sec = 0;
666 * We are colliding with the timer actually firing.
667 * Punt after filling in the timer's old value, and
668 * disable this firing since we are already reporting
669 * it as an overrun (thanks to bump_cpu_timer above).
671 unlock_task_sighand(p, &flags);
675 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
680 * Install the new expiry time (or zero).
681 * For a timer with no notification action, we don't actually
682 * arm the timer (we'll just fake it for timer_gettime).
684 cpu_timer_setexpires(ctmr, new_expires);
685 if (new_expires != 0 && val < new_expires) {
689 unlock_task_sighand(p, &flags);
691 * Install the new reload setting, and
692 * set up the signal and overrun bookkeeping.
694 timer->it_interval = timespec64_to_ktime(new->it_interval);
697 * This acts as a modification timestamp for the timer,
698 * so any automatic reload attempt will punt on seeing
699 * that we have reset the timer manually.
701 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
703 timer->it_overrun_last = 0;
704 timer->it_overrun = -1;
706 if (new_expires != 0 && !(val < new_expires)) {
708 * The designated time already passed, so we notify
709 * immediately, even if the thread never runs to
710 * accumulate more time on this clock.
712 cpu_timer_fire(timer);
719 old->it_interval = ns_to_timespec64(old_incr);
724 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
726 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
727 struct cpu_timer *ctmr = &timer->it.cpu;
728 u64 now, expires = cpu_timer_getexpires(ctmr);
729 struct task_struct *p;
732 p = cpu_timer_task_rcu(timer);
737 * Easy part: convert the reload time.
739 itp->it_interval = ktime_to_timespec64(timer->it_interval);
745 * Sample the clock to take the difference with the expiry time.
747 if (CPUCLOCK_PERTHREAD(timer->it_clock))
748 now = cpu_clock_sample(clkid, p);
750 now = cpu_clock_sample_group(clkid, p, false);
753 itp->it_value = ns_to_timespec64(expires - now);
756 * The timer should have expired already, but the firing
757 * hasn't taken place yet. Say it's just about to expire.
759 itp->it_value.tv_nsec = 1;
760 itp->it_value.tv_sec = 0;
766 #define MAX_COLLECTED 20
768 static u64 collect_timerqueue(struct timerqueue_head *head,
769 struct list_head *firing, u64 now)
771 struct timerqueue_node *next;
774 while ((next = timerqueue_getnext(head))) {
775 struct cpu_timer *ctmr;
778 ctmr = container_of(next, struct cpu_timer, node);
779 expires = cpu_timer_getexpires(ctmr);
780 /* Limit the number of timers to expire at once */
781 if (++i == MAX_COLLECTED || now < expires)
785 /* See posix_cpu_timer_wait_running() */
786 rcu_assign_pointer(ctmr->handling, current);
787 cpu_timer_dequeue(ctmr);
788 list_add_tail(&ctmr->elist, firing);
794 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
795 struct list_head *firing)
797 struct posix_cputimer_base *base = pct->bases;
800 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
801 base->nextevt = collect_timerqueue(&base->tqhead, firing,
806 static inline void check_dl_overrun(struct task_struct *tsk)
808 if (tsk->dl.dl_overrun) {
809 tsk->dl.dl_overrun = 0;
810 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
814 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
819 if (print_fatal_signals) {
820 pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
821 rt ? "RT" : "CPU", hard ? "hard" : "soft",
822 current->comm, task_pid_nr(current));
824 __group_send_sig_info(signo, SEND_SIG_PRIV, current);
829 * Check for any per-thread CPU timers that have fired and move them off
830 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
831 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
833 static void check_thread_timers(struct task_struct *tsk,
834 struct list_head *firing)
836 struct posix_cputimers *pct = &tsk->posix_cputimers;
837 u64 samples[CPUCLOCK_MAX];
841 check_dl_overrun(tsk);
843 if (expiry_cache_is_inactive(pct))
846 task_sample_cputime(tsk, samples);
847 collect_posix_cputimers(pct, samples, firing);
850 * Check for the special case thread timers.
852 soft = task_rlimit(tsk, RLIMIT_RTTIME);
853 if (soft != RLIM_INFINITY) {
854 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
855 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
856 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
858 /* At the hard limit, send SIGKILL. No further action. */
859 if (hard != RLIM_INFINITY &&
860 check_rlimit(rttime, hard, SIGKILL, true, true))
863 /* At the soft limit, send a SIGXCPU every second */
864 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
865 soft += USEC_PER_SEC;
866 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
870 if (expiry_cache_is_inactive(pct))
871 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
874 static inline void stop_process_timers(struct signal_struct *sig)
876 struct posix_cputimers *pct = &sig->posix_cputimers;
878 /* Turn off the active flag. This is done without locking. */
879 WRITE_ONCE(pct->timers_active, false);
880 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
883 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
884 u64 *expires, u64 cur_time, int signo)
889 if (cur_time >= it->expires) {
891 it->expires += it->incr;
895 trace_itimer_expire(signo == SIGPROF ?
896 ITIMER_PROF : ITIMER_VIRTUAL,
897 task_tgid(tsk), cur_time);
898 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
901 if (it->expires && it->expires < *expires)
902 *expires = it->expires;
906 * Check for any per-thread CPU timers that have fired and move them
907 * off the tsk->*_timers list onto the firing list. Per-thread timers
908 * have already been taken off.
910 static void check_process_timers(struct task_struct *tsk,
911 struct list_head *firing)
913 struct signal_struct *const sig = tsk->signal;
914 struct posix_cputimers *pct = &sig->posix_cputimers;
915 u64 samples[CPUCLOCK_MAX];
919 * If there are no active process wide timers (POSIX 1.b, itimers,
920 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
921 * processing when there is already another task handling them.
923 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
927 * Signify that a thread is checking for process timers.
928 * Write access to this field is protected by the sighand lock.
930 pct->expiry_active = true;
933 * Collect the current process totals. Group accounting is active
934 * so the sample can be taken directly.
936 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
937 collect_posix_cputimers(pct, samples, firing);
940 * Check for the special case process timers.
942 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
943 &pct->bases[CPUCLOCK_PROF].nextevt,
944 samples[CPUCLOCK_PROF], SIGPROF);
945 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
946 &pct->bases[CPUCLOCK_VIRT].nextevt,
947 samples[CPUCLOCK_VIRT], SIGVTALRM);
949 soft = task_rlimit(tsk, RLIMIT_CPU);
950 if (soft != RLIM_INFINITY) {
951 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
952 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
953 u64 ptime = samples[CPUCLOCK_PROF];
954 u64 softns = (u64)soft * NSEC_PER_SEC;
955 u64 hardns = (u64)hard * NSEC_PER_SEC;
957 /* At the hard limit, send SIGKILL. No further action. */
958 if (hard != RLIM_INFINITY &&
959 check_rlimit(ptime, hardns, SIGKILL, false, true))
962 /* At the soft limit, send a SIGXCPU every second */
963 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
964 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
965 softns += NSEC_PER_SEC;
968 /* Update the expiry cache */
969 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
970 pct->bases[CPUCLOCK_PROF].nextevt = softns;
973 if (expiry_cache_is_inactive(pct))
974 stop_process_timers(sig);
976 pct->expiry_active = false;
980 * This is called from the signal code (via posixtimer_rearm)
981 * when the last timer signal was delivered and we have to reload the timer.
983 static void posix_cpu_timer_rearm(struct k_itimer *timer)
985 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
986 struct task_struct *p;
987 struct sighand_struct *sighand;
992 p = cpu_timer_task_rcu(timer);
996 /* Protect timer list r/w in arm_timer() */
997 sighand = lock_task_sighand(p, &flags);
998 if (unlikely(sighand == NULL))
1002 * Fetch the current sample and update the timer's expiry time.
1004 if (CPUCLOCK_PERTHREAD(timer->it_clock))
1005 now = cpu_clock_sample(clkid, p);
1007 now = cpu_clock_sample_group(clkid, p, true);
1009 bump_cpu_timer(timer, now);
1012 * Now re-arm for the new expiry time.
1014 arm_timer(timer, p);
1015 unlock_task_sighand(p, &flags);
1021 * task_cputimers_expired - Check whether posix CPU timers are expired
1023 * @samples: Array of current samples for the CPUCLOCK clocks
1024 * @pct: Pointer to a posix_cputimers container
1026 * Returns true if any member of @samples is greater than the corresponding
1027 * member of @pct->bases[CLK].nextevt. False otherwise
1030 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1034 for (i = 0; i < CPUCLOCK_MAX; i++) {
1035 if (samples[i] >= pct->bases[i].nextevt)
1042 * fastpath_timer_check - POSIX CPU timers fast path.
1044 * @tsk: The task (thread) being checked.
1046 * Check the task and thread group timers. If both are zero (there are no
1047 * timers set) return false. Otherwise snapshot the task and thread group
1048 * timers and compare them with the corresponding expiration times. Return
1049 * true if a timer has expired, else return false.
1051 static inline bool fastpath_timer_check(struct task_struct *tsk)
1053 struct posix_cputimers *pct = &tsk->posix_cputimers;
1054 struct signal_struct *sig;
1056 if (!expiry_cache_is_inactive(pct)) {
1057 u64 samples[CPUCLOCK_MAX];
1059 task_sample_cputime(tsk, samples);
1060 if (task_cputimers_expired(samples, pct))
1065 pct = &sig->posix_cputimers;
1067 * Check if thread group timers expired when timers are active and
1068 * no other thread in the group is already handling expiry for
1069 * thread group cputimers. These fields are read without the
1070 * sighand lock. However, this is fine because this is meant to be
1071 * a fastpath heuristic to determine whether we should try to
1072 * acquire the sighand lock to handle timer expiry.
1074 * In the worst case scenario, if concurrently timers_active is set
1075 * or expiry_active is cleared, but the current thread doesn't see
1076 * the change yet, the timer checks are delayed until the next
1077 * thread in the group gets a scheduler interrupt to handle the
1078 * timer. This isn't an issue in practice because these types of
1079 * delays with signals actually getting sent are expected.
1081 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1082 u64 samples[CPUCLOCK_MAX];
1084 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1087 if (task_cputimers_expired(samples, pct))
1091 if (dl_task(tsk) && tsk->dl.dl_overrun)
1097 static void handle_posix_cpu_timers(struct task_struct *tsk);
1099 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1100 static void posix_cpu_timers_work(struct callback_head *work)
1102 struct posix_cputimers_work *cw = container_of(work, typeof(*cw), work);
1104 mutex_lock(&cw->mutex);
1105 handle_posix_cpu_timers(current);
1106 mutex_unlock(&cw->mutex);
1110 * Invoked from the posix-timer core when a cancel operation failed because
1111 * the timer is marked firing. The caller holds rcu_read_lock(), which
1112 * protects the timer and the task which is expiring it from being freed.
1114 static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1116 struct task_struct *tsk = rcu_dereference(timr->it.cpu.handling);
1118 /* Has the handling task completed expiry already? */
1122 /* Ensure that the task cannot go away */
1123 get_task_struct(tsk);
1124 /* Now drop the RCU protection so the mutex can be locked */
1126 /* Wait on the expiry mutex */
1127 mutex_lock(&tsk->posix_cputimers_work.mutex);
1128 /* Release it immediately again. */
1129 mutex_unlock(&tsk->posix_cputimers_work.mutex);
1130 /* Drop the task reference. */
1131 put_task_struct(tsk);
1132 /* Relock RCU so the callsite is balanced */
1136 static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1138 /* Ensure that timr->it.cpu.handling task cannot go away */
1140 spin_unlock_irq(&timr->it_lock);
1141 posix_cpu_timer_wait_running(timr);
1143 /* @timr is on stack and is valid */
1144 spin_lock_irq(&timr->it_lock);
1148 * Clear existing posix CPU timers task work.
1150 void clear_posix_cputimers_work(struct task_struct *p)
1153 * A copied work entry from the old task is not meaningful, clear it.
1154 * N.B. init_task_work will not do this.
1156 memset(&p->posix_cputimers_work.work, 0,
1157 sizeof(p->posix_cputimers_work.work));
1158 init_task_work(&p->posix_cputimers_work.work,
1159 posix_cpu_timers_work);
1160 mutex_init(&p->posix_cputimers_work.mutex);
1161 p->posix_cputimers_work.scheduled = false;
1165 * Initialize posix CPU timers task work in init task. Out of line to
1166 * keep the callback static and to avoid header recursion hell.
1168 void __init posix_cputimers_init_work(void)
1170 clear_posix_cputimers_work(current);
1174 * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1175 * in hard interrupt context or in task context with interrupts
1176 * disabled. Aside of that the writer/reader interaction is always in the
1177 * context of the current task, which means they are strict per CPU.
1179 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1181 return tsk->posix_cputimers_work.scheduled;
1184 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1186 if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1189 /* Schedule task work to actually expire the timers */
1190 tsk->posix_cputimers_work.scheduled = true;
1191 task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1194 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1195 unsigned long start)
1200 * On !RT kernels interrupts are disabled while collecting expired
1201 * timers, so no tick can happen and the fast path check can be
1202 * reenabled without further checks.
1204 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1205 tsk->posix_cputimers_work.scheduled = false;
1210 * On RT enabled kernels ticks can happen while the expired timers
1211 * are collected under sighand lock. But any tick which observes
1212 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1213 * checks. So reenabling the tick work has do be done carefully:
1215 * Disable interrupts and run the fast path check if jiffies have
1216 * advanced since the collecting of expired timers started. If
1217 * jiffies have not advanced or the fast path check did not find
1218 * newly expired timers, reenable the fast path check in the timer
1219 * interrupt. If there are newly expired timers, return false and
1220 * let the collection loop repeat.
1222 local_irq_disable();
1223 if (start != jiffies && fastpath_timer_check(tsk))
1226 tsk->posix_cputimers_work.scheduled = false;
1231 #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1232 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1234 lockdep_posixtimer_enter();
1235 handle_posix_cpu_timers(tsk);
1236 lockdep_posixtimer_exit();
1239 static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1244 static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
1246 spin_unlock_irq(&timr->it_lock);
1248 spin_lock_irq(&timr->it_lock);
1251 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1256 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1257 unsigned long start)
1261 #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1263 static void handle_posix_cpu_timers(struct task_struct *tsk)
1265 struct k_itimer *timer, *next;
1266 unsigned long flags, start;
1269 if (!lock_task_sighand(tsk, &flags))
1274 * On RT locking sighand lock does not disable interrupts,
1275 * so this needs to be careful vs. ticks. Store the current
1278 start = READ_ONCE(jiffies);
1282 * Here we take off tsk->signal->cpu_timers[N] and
1283 * tsk->cpu_timers[N] all the timers that are firing, and
1284 * put them on the firing list.
1286 check_thread_timers(tsk, &firing);
1288 check_process_timers(tsk, &firing);
1291 * The above timer checks have updated the exipry cache and
1292 * because nothing can have queued or modified timers after
1293 * sighand lock was taken above it is guaranteed to be
1294 * consistent. So the next timer interrupt fastpath check
1295 * will find valid data.
1297 * If timer expiry runs in the timer interrupt context then
1298 * the loop is not relevant as timers will be directly
1299 * expired in interrupt context. The stub function below
1300 * returns always true which allows the compiler to
1301 * optimize the loop out.
1303 * If timer expiry is deferred to task work context then
1304 * the following rules apply:
1306 * - On !RT kernels no tick can have happened on this CPU
1307 * after sighand lock was acquired because interrupts are
1308 * disabled. So reenabling task work before dropping
1309 * sighand lock and reenabling interrupts is race free.
1311 * - On RT kernels ticks might have happened but the tick
1312 * work ignored posix CPU timer handling because the
1313 * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1314 * must be done very carefully including a check whether
1315 * ticks have happened since the start of the timer
1316 * expiry checks. posix_cpu_timers_enable_work() takes
1317 * care of that and eventually lets the expiry checks
1320 } while (!posix_cpu_timers_enable_work(tsk, start));
1323 * We must release sighand lock before taking any timer's lock.
1324 * There is a potential race with timer deletion here, as the
1325 * siglock now protects our private firing list. We have set
1326 * the firing flag in each timer, so that a deletion attempt
1327 * that gets the timer lock before we do will give it up and
1328 * spin until we've taken care of that timer below.
1330 unlock_task_sighand(tsk, &flags);
1333 * Now that all the timers on our list have the firing flag,
1334 * no one will touch their list entries but us. We'll take
1335 * each timer's lock before clearing its firing flag, so no
1336 * timer call will interfere.
1338 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1342 * spin_lock() is sufficient here even independent of the
1343 * expiry context. If expiry happens in hard interrupt
1344 * context it's obvious. For task work context it's safe
1345 * because all other operations on timer::it_lock happen in
1346 * task context (syscall or exit).
1348 spin_lock(&timer->it_lock);
1349 list_del_init(&timer->it.cpu.elist);
1350 cpu_firing = timer->it.cpu.firing;
1351 timer->it.cpu.firing = 0;
1353 * The firing flag is -1 if we collided with a reset
1354 * of the timer, which already reported this
1355 * almost-firing as an overrun. So don't generate an event.
1357 if (likely(cpu_firing >= 0))
1358 cpu_timer_fire(timer);
1359 /* See posix_cpu_timer_wait_running() */
1360 rcu_assign_pointer(timer->it.cpu.handling, NULL);
1361 spin_unlock(&timer->it_lock);
1366 * This is called from the timer interrupt handler. The irq handler has
1367 * already updated our counts. We need to check if any timers fire now.
1368 * Interrupts are disabled.
1370 void run_posix_cpu_timers(void)
1372 struct task_struct *tsk = current;
1374 lockdep_assert_irqs_disabled();
1377 * If the actual expiry is deferred to task work context and the
1378 * work is already scheduled there is no point to do anything here.
1380 if (posix_cpu_timers_work_scheduled(tsk))
1384 * The fast path checks that there are no expired thread or thread
1385 * group timers. If that's so, just return.
1387 if (!fastpath_timer_check(tsk))
1390 __run_posix_cpu_timers(tsk);
1394 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1395 * The tsk->sighand->siglock must be held by the caller.
1397 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1398 u64 *newval, u64 *oldval)
1402 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1405 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1406 now = cpu_clock_sample_group(clkid, tsk, true);
1410 * We are setting itimer. The *oldval is absolute and we update
1411 * it to be relative, *newval argument is relative and we update
1412 * it to be absolute.
1415 if (*oldval <= now) {
1416 /* Just about to fire. */
1417 *oldval = TICK_NSEC;
1429 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1430 * expiry cache is also used by RLIMIT_CPU!.
1432 if (*newval < *nextevt)
1435 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1438 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1439 const struct timespec64 *rqtp)
1441 struct itimerspec64 it;
1442 struct k_itimer timer;
1447 * Set up a temporary timer and then wait for it to go off.
1449 memset(&timer, 0, sizeof timer);
1450 spin_lock_init(&timer.it_lock);
1451 timer.it_clock = which_clock;
1452 timer.it_overrun = -1;
1453 error = posix_cpu_timer_create(&timer);
1454 timer.it_process = current;
1457 static struct itimerspec64 zero_it;
1458 struct restart_block *restart;
1460 memset(&it, 0, sizeof(it));
1461 it.it_value = *rqtp;
1463 spin_lock_irq(&timer.it_lock);
1464 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1466 spin_unlock_irq(&timer.it_lock);
1470 while (!signal_pending(current)) {
1471 if (!cpu_timer_getexpires(&timer.it.cpu)) {
1473 * Our timer fired and was reset, below
1474 * deletion can not fail.
1476 posix_cpu_timer_del(&timer);
1477 spin_unlock_irq(&timer.it_lock);
1482 * Block until cpu_timer_fire (or a signal) wakes us.
1484 __set_current_state(TASK_INTERRUPTIBLE);
1485 spin_unlock_irq(&timer.it_lock);
1487 spin_lock_irq(&timer.it_lock);
1491 * We were interrupted by a signal.
1493 expires = cpu_timer_getexpires(&timer.it.cpu);
1494 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1496 /* Timer is now unarmed, deletion can not fail. */
1497 posix_cpu_timer_del(&timer);
1499 while (error == TIMER_RETRY) {
1500 posix_cpu_timer_wait_running_nsleep(&timer);
1501 error = posix_cpu_timer_del(&timer);
1505 spin_unlock_irq(&timer.it_lock);
1507 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1509 * It actually did fire already.
1514 error = -ERESTART_RESTARTBLOCK;
1516 * Report back to the user the time still remaining.
1518 restart = ¤t->restart_block;
1519 restart->nanosleep.expires = expires;
1520 if (restart->nanosleep.type != TT_NONE)
1521 error = nanosleep_copyout(restart, &it.it_value);
1527 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1529 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1530 const struct timespec64 *rqtp)
1532 struct restart_block *restart_block = ¤t->restart_block;
1536 * Diagnose required errors first.
1538 if (CPUCLOCK_PERTHREAD(which_clock) &&
1539 (CPUCLOCK_PID(which_clock) == 0 ||
1540 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1543 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1545 if (error == -ERESTART_RESTARTBLOCK) {
1547 if (flags & TIMER_ABSTIME)
1548 return -ERESTARTNOHAND;
1550 restart_block->nanosleep.clockid = which_clock;
1551 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1556 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1558 clockid_t which_clock = restart_block->nanosleep.clockid;
1559 struct timespec64 t;
1561 t = ns_to_timespec64(restart_block->nanosleep.expires);
1563 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1566 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1567 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1569 static int process_cpu_clock_getres(const clockid_t which_clock,
1570 struct timespec64 *tp)
1572 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1574 static int process_cpu_clock_get(const clockid_t which_clock,
1575 struct timespec64 *tp)
1577 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1579 static int process_cpu_timer_create(struct k_itimer *timer)
1581 timer->it_clock = PROCESS_CLOCK;
1582 return posix_cpu_timer_create(timer);
1584 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1585 const struct timespec64 *rqtp)
1587 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1589 static int thread_cpu_clock_getres(const clockid_t which_clock,
1590 struct timespec64 *tp)
1592 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1594 static int thread_cpu_clock_get(const clockid_t which_clock,
1595 struct timespec64 *tp)
1597 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1599 static int thread_cpu_timer_create(struct k_itimer *timer)
1601 timer->it_clock = THREAD_CLOCK;
1602 return posix_cpu_timer_create(timer);
1605 const struct k_clock clock_posix_cpu = {
1606 .clock_getres = posix_cpu_clock_getres,
1607 .clock_set = posix_cpu_clock_set,
1608 .clock_get_timespec = posix_cpu_clock_get,
1609 .timer_create = posix_cpu_timer_create,
1610 .nsleep = posix_cpu_nsleep,
1611 .timer_set = posix_cpu_timer_set,
1612 .timer_del = posix_cpu_timer_del,
1613 .timer_get = posix_cpu_timer_get,
1614 .timer_rearm = posix_cpu_timer_rearm,
1615 .timer_wait_running = posix_cpu_timer_wait_running,
1618 const struct k_clock clock_process = {
1619 .clock_getres = process_cpu_clock_getres,
1620 .clock_get_timespec = process_cpu_clock_get,
1621 .timer_create = process_cpu_timer_create,
1622 .nsleep = process_cpu_nsleep,
1625 const struct k_clock clock_thread = {
1626 .clock_getres = thread_cpu_clock_getres,
1627 .clock_get_timespec = thread_cpu_clock_get,
1628 .timer_create = thread_cpu_timer_create,