2 * Implement CPU time clocks for the POSIX clock interface.
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <linux/errno.h>
8 #include <linux/math64.h>
9 #include <asm/uaccess.h>
10 #include <linux/kernel_stat.h>
11 #include <trace/events/timer.h>
12 #include <linux/random.h>
13 #include <linux/tick.h>
14 #include <linux/workqueue.h>
17 * Called after updating RLIMIT_CPU to run cpu timer and update
18 * tsk->signal->cputime_expires expiration cache if necessary. Needs
19 * siglock protection since other code may update expiration cache as
22 void update_rlimit_cpu(struct task_struct
*task
, unsigned long rlim_new
)
24 cputime_t cputime
= secs_to_cputime(rlim_new
);
26 spin_lock_irq(&task
->sighand
->siglock
);
27 set_process_cpu_timer(task
, CPUCLOCK_PROF
, &cputime
, NULL
);
28 spin_unlock_irq(&task
->sighand
->siglock
);
31 static int check_clock(const clockid_t which_clock
)
34 struct task_struct
*p
;
35 const pid_t pid
= CPUCLOCK_PID(which_clock
);
37 if (CPUCLOCK_WHICH(which_clock
) >= CPUCLOCK_MAX
)
44 p
= find_task_by_vpid(pid
);
45 if (!p
|| !(CPUCLOCK_PERTHREAD(which_clock
) ?
46 same_thread_group(p
, current
) : has_group_leader_pid(p
))) {
54 static inline unsigned long long
55 timespec_to_sample(const clockid_t which_clock
, const struct timespec
*tp
)
57 unsigned long long ret
;
59 ret
= 0; /* high half always zero when .cpu used */
60 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
61 ret
= (unsigned long long)tp
->tv_sec
* NSEC_PER_SEC
+ tp
->tv_nsec
;
63 ret
= cputime_to_expires(timespec_to_cputime(tp
));
68 static void sample_to_timespec(const clockid_t which_clock
,
69 unsigned long long expires
,
72 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
)
73 *tp
= ns_to_timespec(expires
);
75 cputime_to_timespec((__force cputime_t
)expires
, tp
);
79 * Update expiry time from increment, and increase overrun count,
80 * given the current clock sample.
82 static void bump_cpu_timer(struct k_itimer
*timer
,
83 unsigned long long now
)
86 unsigned long long delta
, incr
;
88 if (timer
->it
.cpu
.incr
== 0)
91 if (now
< timer
->it
.cpu
.expires
)
94 incr
= timer
->it
.cpu
.incr
;
95 delta
= now
+ incr
- timer
->it
.cpu
.expires
;
97 /* Don't use (incr*2 < delta), incr*2 might overflow. */
98 for (i
= 0; incr
< delta
- incr
; i
++)
101 for (; i
>= 0; incr
>>= 1, i
--) {
105 timer
->it
.cpu
.expires
+= incr
;
106 timer
->it_overrun
+= 1LL << i
;
112 * task_cputime_zero - Check a task_cputime struct for all zero fields.
114 * @cputime: The struct to compare.
116 * Checks @cputime to see if all fields are zero. Returns true if all fields
117 * are zero, false if any field is nonzero.
119 static inline int task_cputime_zero(const struct task_cputime
*cputime
)
121 if (!cputime
->utime
&& !cputime
->stime
&& !cputime
->sum_exec_runtime
)
126 static inline unsigned long long prof_ticks(struct task_struct
*p
)
128 cputime_t utime
, stime
;
130 task_cputime(p
, &utime
, &stime
);
132 return cputime_to_expires(utime
+ stime
);
134 static inline unsigned long long virt_ticks(struct task_struct
*p
)
138 task_cputime(p
, &utime
, NULL
);
140 return cputime_to_expires(utime
);
144 posix_cpu_clock_getres(const clockid_t which_clock
, struct timespec
*tp
)
146 int error
= check_clock(which_clock
);
149 tp
->tv_nsec
= ((NSEC_PER_SEC
+ HZ
- 1) / HZ
);
150 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
152 * If sched_clock is using a cycle counter, we
153 * don't have any idea of its true resolution
154 * exported, but it is much more than 1s/HZ.
163 posix_cpu_clock_set(const clockid_t which_clock
, const struct timespec
*tp
)
166 * You can never reset a CPU clock, but we check for other errors
167 * in the call before failing with EPERM.
169 int error
= check_clock(which_clock
);
178 * Sample a per-thread clock for the given task.
180 static int cpu_clock_sample(const clockid_t which_clock
, struct task_struct
*p
,
181 unsigned long long *sample
)
183 switch (CPUCLOCK_WHICH(which_clock
)) {
187 *sample
= prof_ticks(p
);
190 *sample
= virt_ticks(p
);
193 *sample
= task_sched_runtime(p
);
199 static void update_gt_cputime(struct task_cputime
*a
, struct task_cputime
*b
)
201 if (b
->utime
> a
->utime
)
204 if (b
->stime
> a
->stime
)
207 if (b
->sum_exec_runtime
> a
->sum_exec_runtime
)
208 a
->sum_exec_runtime
= b
->sum_exec_runtime
;
211 void thread_group_cputimer(struct task_struct
*tsk
, struct task_cputime
*times
)
213 struct thread_group_cputimer
*cputimer
= &tsk
->signal
->cputimer
;
214 struct task_cputime sum
;
217 if (!cputimer
->running
) {
219 * The POSIX timer interface allows for absolute time expiry
220 * values through the TIMER_ABSTIME flag, therefore we have
221 * to synchronize the timer to the clock every time we start
224 thread_group_cputime(tsk
, &sum
);
225 raw_spin_lock_irqsave(&cputimer
->lock
, flags
);
226 cputimer
->running
= 1;
227 update_gt_cputime(&cputimer
->cputime
, &sum
);
229 raw_spin_lock_irqsave(&cputimer
->lock
, flags
);
230 *times
= cputimer
->cputime
;
231 raw_spin_unlock_irqrestore(&cputimer
->lock
, flags
);
235 * Sample a process (thread group) clock for the given group_leader task.
236 * Must be called with task sighand lock held for safe while_each_thread()
239 static int cpu_clock_sample_group(const clockid_t which_clock
,
240 struct task_struct
*p
,
241 unsigned long long *sample
)
243 struct task_cputime cputime
;
245 switch (CPUCLOCK_WHICH(which_clock
)) {
249 thread_group_cputime(p
, &cputime
);
250 *sample
= cputime_to_expires(cputime
.utime
+ cputime
.stime
);
253 thread_group_cputime(p
, &cputime
);
254 *sample
= cputime_to_expires(cputime
.utime
);
257 thread_group_cputime(p
, &cputime
);
258 *sample
= cputime
.sum_exec_runtime
;
264 static int posix_cpu_clock_get_task(struct task_struct
*tsk
,
265 const clockid_t which_clock
,
269 unsigned long long rtn
;
271 if (CPUCLOCK_PERTHREAD(which_clock
)) {
272 if (same_thread_group(tsk
, current
))
273 err
= cpu_clock_sample(which_clock
, tsk
, &rtn
);
276 struct sighand_struct
*sighand
;
279 * while_each_thread() is not yet entirely RCU safe,
280 * keep locking the group while sampling process
283 sighand
= lock_task_sighand(tsk
, &flags
);
287 if (tsk
== current
|| thread_group_leader(tsk
))
288 err
= cpu_clock_sample_group(which_clock
, tsk
, &rtn
);
290 unlock_task_sighand(tsk
, &flags
);
294 sample_to_timespec(which_clock
, rtn
, tp
);
300 static int posix_cpu_clock_get(const clockid_t which_clock
, struct timespec
*tp
)
302 const pid_t pid
= CPUCLOCK_PID(which_clock
);
307 * Special case constant value for our own clocks.
308 * We don't have to do any lookup to find ourselves.
310 err
= posix_cpu_clock_get_task(current
, which_clock
, tp
);
313 * Find the given PID, and validate that the caller
314 * should be able to see it.
316 struct task_struct
*p
;
318 p
= find_task_by_vpid(pid
);
320 err
= posix_cpu_clock_get_task(p
, which_clock
, tp
);
329 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
330 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
331 * new timer already all-zeros initialized.
333 static int posix_cpu_timer_create(struct k_itimer
*new_timer
)
336 const pid_t pid
= CPUCLOCK_PID(new_timer
->it_clock
);
337 struct task_struct
*p
;
339 if (CPUCLOCK_WHICH(new_timer
->it_clock
) >= CPUCLOCK_MAX
)
342 INIT_LIST_HEAD(&new_timer
->it
.cpu
.entry
);
345 if (CPUCLOCK_PERTHREAD(new_timer
->it_clock
)) {
349 p
= find_task_by_vpid(pid
);
350 if (p
&& !same_thread_group(p
, current
))
355 p
= current
->group_leader
;
357 p
= find_task_by_vpid(pid
);
358 if (p
&& !has_group_leader_pid(p
))
362 new_timer
->it
.cpu
.task
= p
;
374 * Clean up a CPU-clock timer that is about to be destroyed.
375 * This is called from timer deletion with the timer already locked.
376 * If we return TIMER_RETRY, it's necessary to release the timer's lock
377 * and try again. (This happens when the timer is in the middle of firing.)
379 static int posix_cpu_timer_del(struct k_itimer
*timer
)
383 struct sighand_struct
*sighand
;
384 struct task_struct
*p
= timer
->it
.cpu
.task
;
386 WARN_ON_ONCE(p
== NULL
);
389 * Protect against sighand release/switch in exit/exec and process/
390 * thread timer list entry concurrent read/writes.
392 sighand
= lock_task_sighand(p
, &flags
);
393 if (unlikely(sighand
== NULL
)) {
395 * We raced with the reaping of the task.
396 * The deletion should have cleared us off the list.
398 WARN_ON_ONCE(!list_empty(&timer
->it
.cpu
.entry
));
400 if (timer
->it
.cpu
.firing
)
403 list_del(&timer
->it
.cpu
.entry
);
405 unlock_task_sighand(p
, &flags
);
414 static void cleanup_timers_list(struct list_head
*head
)
416 struct cpu_timer_list
*timer
, *next
;
418 list_for_each_entry_safe(timer
, next
, head
, entry
)
419 list_del_init(&timer
->entry
);
423 * Clean out CPU timers still ticking when a thread exited. The task
424 * pointer is cleared, and the expiry time is replaced with the residual
425 * time for later timer_gettime calls to return.
426 * This must be called with the siglock held.
428 static void cleanup_timers(struct list_head
*head
)
430 cleanup_timers_list(head
);
431 cleanup_timers_list(++head
);
432 cleanup_timers_list(++head
);
436 * These are both called with the siglock held, when the current thread
437 * is being reaped. When the final (leader) thread in the group is reaped,
438 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
440 void posix_cpu_timers_exit(struct task_struct
*tsk
)
442 add_device_randomness((const void*) &tsk
->se
.sum_exec_runtime
,
443 sizeof(unsigned long long));
444 cleanup_timers(tsk
->cpu_timers
);
447 void posix_cpu_timers_exit_group(struct task_struct
*tsk
)
449 cleanup_timers(tsk
->signal
->cpu_timers
);
452 static inline int expires_gt(cputime_t expires
, cputime_t new_exp
)
454 return expires
== 0 || expires
> new_exp
;
458 * Insert the timer on the appropriate list before any timers that
459 * expire later. This must be called with the sighand lock held.
461 static void arm_timer(struct k_itimer
*timer
)
463 struct task_struct
*p
= timer
->it
.cpu
.task
;
464 struct list_head
*head
, *listpos
;
465 struct task_cputime
*cputime_expires
;
466 struct cpu_timer_list
*const nt
= &timer
->it
.cpu
;
467 struct cpu_timer_list
*next
;
469 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
470 head
= p
->cpu_timers
;
471 cputime_expires
= &p
->cputime_expires
;
473 head
= p
->signal
->cpu_timers
;
474 cputime_expires
= &p
->signal
->cputime_expires
;
476 head
+= CPUCLOCK_WHICH(timer
->it_clock
);
479 list_for_each_entry(next
, head
, entry
) {
480 if (nt
->expires
< next
->expires
)
482 listpos
= &next
->entry
;
484 list_add(&nt
->entry
, listpos
);
486 if (listpos
== head
) {
487 unsigned long long exp
= nt
->expires
;
490 * We are the new earliest-expiring POSIX 1.b timer, hence
491 * need to update expiration cache. Take into account that
492 * for process timers we share expiration cache with itimers
493 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
496 switch (CPUCLOCK_WHICH(timer
->it_clock
)) {
498 if (expires_gt(cputime_expires
->prof_exp
, expires_to_cputime(exp
)))
499 cputime_expires
->prof_exp
= expires_to_cputime(exp
);
502 if (expires_gt(cputime_expires
->virt_exp
, expires_to_cputime(exp
)))
503 cputime_expires
->virt_exp
= expires_to_cputime(exp
);
506 if (cputime_expires
->sched_exp
== 0 ||
507 cputime_expires
->sched_exp
> exp
)
508 cputime_expires
->sched_exp
= exp
;
515 * The timer is locked, fire it and arrange for its reload.
517 static void cpu_timer_fire(struct k_itimer
*timer
)
519 if ((timer
->it_sigev_notify
& ~SIGEV_THREAD_ID
) == SIGEV_NONE
) {
521 * User don't want any signal.
523 timer
->it
.cpu
.expires
= 0;
524 } else if (unlikely(timer
->sigq
== NULL
)) {
526 * This a special case for clock_nanosleep,
527 * not a normal timer from sys_timer_create.
529 wake_up_process(timer
->it_process
);
530 timer
->it
.cpu
.expires
= 0;
531 } else if (timer
->it
.cpu
.incr
== 0) {
533 * One-shot timer. Clear it as soon as it's fired.
535 posix_timer_event(timer
, 0);
536 timer
->it
.cpu
.expires
= 0;
537 } else if (posix_timer_event(timer
, ++timer
->it_requeue_pending
)) {
539 * The signal did not get queued because the signal
540 * was ignored, so we won't get any callback to
541 * reload the timer. But we need to keep it
542 * ticking in case the signal is deliverable next time.
544 posix_cpu_timer_schedule(timer
);
549 * Sample a process (thread group) timer for the given group_leader task.
550 * Must be called with task sighand lock held for safe while_each_thread()
553 static int cpu_timer_sample_group(const clockid_t which_clock
,
554 struct task_struct
*p
,
555 unsigned long long *sample
)
557 struct task_cputime cputime
;
559 thread_group_cputimer(p
, &cputime
);
560 switch (CPUCLOCK_WHICH(which_clock
)) {
564 *sample
= cputime_to_expires(cputime
.utime
+ cputime
.stime
);
567 *sample
= cputime_to_expires(cputime
.utime
);
570 *sample
= cputime
.sum_exec_runtime
+ task_delta_exec(p
);
576 #ifdef CONFIG_NO_HZ_FULL
577 static void nohz_kick_work_fn(struct work_struct
*work
)
579 tick_nohz_full_kick_all();
582 static DECLARE_WORK(nohz_kick_work
, nohz_kick_work_fn
);
585 * We need the IPIs to be sent from sane process context.
586 * The posix cpu timers are always set with irqs disabled.
588 static void posix_cpu_timer_kick_nohz(void)
590 if (context_tracking_is_enabled())
591 schedule_work(&nohz_kick_work
);
594 bool posix_cpu_timers_can_stop_tick(struct task_struct
*tsk
)
596 if (!task_cputime_zero(&tsk
->cputime_expires
))
599 if (tsk
->signal
->cputimer
.running
)
605 static inline void posix_cpu_timer_kick_nohz(void) { }
609 * Guts of sys_timer_settime for CPU timers.
610 * This is called with the timer locked and interrupts disabled.
611 * If we return TIMER_RETRY, it's necessary to release the timer's lock
612 * and try again. (This happens when the timer is in the middle of firing.)
614 static int posix_cpu_timer_set(struct k_itimer
*timer
, int timer_flags
,
615 struct itimerspec
*new, struct itimerspec
*old
)
618 struct sighand_struct
*sighand
;
619 struct task_struct
*p
= timer
->it
.cpu
.task
;
620 unsigned long long old_expires
, new_expires
, old_incr
, val
;
623 WARN_ON_ONCE(p
== NULL
);
625 new_expires
= timespec_to_sample(timer
->it_clock
, &new->it_value
);
628 * Protect against sighand release/switch in exit/exec and p->cpu_timers
629 * and p->signal->cpu_timers read/write in arm_timer()
631 sighand
= lock_task_sighand(p
, &flags
);
633 * If p has just been reaped, we can no
634 * longer get any information about it at all.
636 if (unlikely(sighand
== NULL
)) {
641 * Disarm any old timer after extracting its expiry time.
643 WARN_ON_ONCE(!irqs_disabled());
646 old_incr
= timer
->it
.cpu
.incr
;
647 old_expires
= timer
->it
.cpu
.expires
;
648 if (unlikely(timer
->it
.cpu
.firing
)) {
649 timer
->it
.cpu
.firing
= -1;
652 list_del_init(&timer
->it
.cpu
.entry
);
655 * We need to sample the current value to convert the new
656 * value from to relative and absolute, and to convert the
657 * old value from absolute to relative. To set a process
658 * timer, we need a sample to balance the thread expiry
659 * times (in arm_timer). With an absolute time, we must
660 * check if it's already passed. In short, we need a sample.
662 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
663 cpu_clock_sample(timer
->it_clock
, p
, &val
);
665 cpu_timer_sample_group(timer
->it_clock
, p
, &val
);
669 if (old_expires
== 0) {
670 old
->it_value
.tv_sec
= 0;
671 old
->it_value
.tv_nsec
= 0;
674 * Update the timer in case it has
675 * overrun already. If it has,
676 * we'll report it as having overrun
677 * and with the next reloaded timer
678 * already ticking, though we are
679 * swallowing that pending
680 * notification here to install the
683 bump_cpu_timer(timer
, val
);
684 if (val
< timer
->it
.cpu
.expires
) {
685 old_expires
= timer
->it
.cpu
.expires
- val
;
686 sample_to_timespec(timer
->it_clock
,
690 old
->it_value
.tv_nsec
= 1;
691 old
->it_value
.tv_sec
= 0;
698 * We are colliding with the timer actually firing.
699 * Punt after filling in the timer's old value, and
700 * disable this firing since we are already reporting
701 * it as an overrun (thanks to bump_cpu_timer above).
703 unlock_task_sighand(p
, &flags
);
707 if (new_expires
!= 0 && !(timer_flags
& TIMER_ABSTIME
)) {
712 * Install the new expiry time (or zero).
713 * For a timer with no notification action, we don't actually
714 * arm the timer (we'll just fake it for timer_gettime).
716 timer
->it
.cpu
.expires
= new_expires
;
717 if (new_expires
!= 0 && val
< new_expires
) {
721 unlock_task_sighand(p
, &flags
);
723 * Install the new reload setting, and
724 * set up the signal and overrun bookkeeping.
726 timer
->it
.cpu
.incr
= timespec_to_sample(timer
->it_clock
,
730 * This acts as a modification timestamp for the timer,
731 * so any automatic reload attempt will punt on seeing
732 * that we have reset the timer manually.
734 timer
->it_requeue_pending
= (timer
->it_requeue_pending
+ 2) &
736 timer
->it_overrun_last
= 0;
737 timer
->it_overrun
= -1;
739 if (new_expires
!= 0 && !(val
< new_expires
)) {
741 * The designated time already passed, so we notify
742 * immediately, even if the thread never runs to
743 * accumulate more time on this clock.
745 cpu_timer_fire(timer
);
751 sample_to_timespec(timer
->it_clock
,
752 old_incr
, &old
->it_interval
);
755 posix_cpu_timer_kick_nohz();
759 static void posix_cpu_timer_get(struct k_itimer
*timer
, struct itimerspec
*itp
)
761 unsigned long long now
;
762 struct task_struct
*p
= timer
->it
.cpu
.task
;
764 WARN_ON_ONCE(p
== NULL
);
767 * Easy part: convert the reload time.
769 sample_to_timespec(timer
->it_clock
,
770 timer
->it
.cpu
.incr
, &itp
->it_interval
);
772 if (timer
->it
.cpu
.expires
== 0) { /* Timer not armed at all. */
773 itp
->it_value
.tv_sec
= itp
->it_value
.tv_nsec
= 0;
778 * Sample the clock to take the difference with the expiry time.
780 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
781 cpu_clock_sample(timer
->it_clock
, p
, &now
);
783 struct sighand_struct
*sighand
;
787 * Protect against sighand release/switch in exit/exec and
788 * also make timer sampling safe if it ends up calling
789 * thread_group_cputime().
791 sighand
= lock_task_sighand(p
, &flags
);
792 if (unlikely(sighand
== NULL
)) {
794 * The process has been reaped.
795 * We can't even collect a sample any more.
796 * Call the timer disarmed, nothing else to do.
798 timer
->it
.cpu
.expires
= 0;
799 sample_to_timespec(timer
->it_clock
, timer
->it
.cpu
.expires
,
803 cpu_timer_sample_group(timer
->it_clock
, p
, &now
);
804 unlock_task_sighand(p
, &flags
);
808 if (now
< timer
->it
.cpu
.expires
) {
809 sample_to_timespec(timer
->it_clock
,
810 timer
->it
.cpu
.expires
- now
,
814 * The timer should have expired already, but the firing
815 * hasn't taken place yet. Say it's just about to expire.
817 itp
->it_value
.tv_nsec
= 1;
818 itp
->it_value
.tv_sec
= 0;
822 static unsigned long long
823 check_timers_list(struct list_head
*timers
,
824 struct list_head
*firing
,
825 unsigned long long curr
)
829 while (!list_empty(timers
)) {
830 struct cpu_timer_list
*t
;
832 t
= list_first_entry(timers
, struct cpu_timer_list
, entry
);
834 if (!--maxfire
|| curr
< t
->expires
)
838 list_move_tail(&t
->entry
, firing
);
845 * Check for any per-thread CPU timers that have fired and move them off
846 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
847 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
849 static void check_thread_timers(struct task_struct
*tsk
,
850 struct list_head
*firing
)
852 struct list_head
*timers
= tsk
->cpu_timers
;
853 struct signal_struct
*const sig
= tsk
->signal
;
854 struct task_cputime
*tsk_expires
= &tsk
->cputime_expires
;
855 unsigned long long expires
;
858 expires
= check_timers_list(timers
, firing
, prof_ticks(tsk
));
859 tsk_expires
->prof_exp
= expires_to_cputime(expires
);
861 expires
= check_timers_list(++timers
, firing
, virt_ticks(tsk
));
862 tsk_expires
->virt_exp
= expires_to_cputime(expires
);
864 tsk_expires
->sched_exp
= check_timers_list(++timers
, firing
,
865 tsk
->se
.sum_exec_runtime
);
868 * Check for the special case thread timers.
870 soft
= ACCESS_ONCE(sig
->rlim
[RLIMIT_RTTIME
].rlim_cur
);
871 if (soft
!= RLIM_INFINITY
) {
873 ACCESS_ONCE(sig
->rlim
[RLIMIT_RTTIME
].rlim_max
);
875 if (hard
!= RLIM_INFINITY
&&
876 tsk
->rt
.timeout
> DIV_ROUND_UP(hard
, USEC_PER_SEC
/HZ
)) {
878 * At the hard limit, we just die.
879 * No need to calculate anything else now.
881 __group_send_sig_info(SIGKILL
, SEND_SIG_PRIV
, tsk
);
884 if (tsk
->rt
.timeout
> DIV_ROUND_UP(soft
, USEC_PER_SEC
/HZ
)) {
886 * At the soft limit, send a SIGXCPU every second.
889 soft
+= USEC_PER_SEC
;
890 sig
->rlim
[RLIMIT_RTTIME
].rlim_cur
= soft
;
893 "RT Watchdog Timeout: %s[%d]\n",
894 tsk
->comm
, task_pid_nr(tsk
));
895 __group_send_sig_info(SIGXCPU
, SEND_SIG_PRIV
, tsk
);
900 static void stop_process_timers(struct signal_struct
*sig
)
902 struct thread_group_cputimer
*cputimer
= &sig
->cputimer
;
905 raw_spin_lock_irqsave(&cputimer
->lock
, flags
);
906 cputimer
->running
= 0;
907 raw_spin_unlock_irqrestore(&cputimer
->lock
, flags
);
910 static u32 onecputick
;
912 static void check_cpu_itimer(struct task_struct
*tsk
, struct cpu_itimer
*it
,
913 unsigned long long *expires
,
914 unsigned long long cur_time
, int signo
)
919 if (cur_time
>= it
->expires
) {
921 it
->expires
+= it
->incr
;
922 it
->error
+= it
->incr_error
;
923 if (it
->error
>= onecputick
) {
924 it
->expires
-= cputime_one_jiffy
;
925 it
->error
-= onecputick
;
931 trace_itimer_expire(signo
== SIGPROF
?
932 ITIMER_PROF
: ITIMER_VIRTUAL
,
933 tsk
->signal
->leader_pid
, cur_time
);
934 __group_send_sig_info(signo
, SEND_SIG_PRIV
, tsk
);
937 if (it
->expires
&& (!*expires
|| it
->expires
< *expires
)) {
938 *expires
= it
->expires
;
943 * Check for any per-thread CPU timers that have fired and move them
944 * off the tsk->*_timers list onto the firing list. Per-thread timers
945 * have already been taken off.
947 static void check_process_timers(struct task_struct
*tsk
,
948 struct list_head
*firing
)
950 struct signal_struct
*const sig
= tsk
->signal
;
951 unsigned long long utime
, ptime
, virt_expires
, prof_expires
;
952 unsigned long long sum_sched_runtime
, sched_expires
;
953 struct list_head
*timers
= sig
->cpu_timers
;
954 struct task_cputime cputime
;
958 * Collect the current process totals.
960 thread_group_cputimer(tsk
, &cputime
);
961 utime
= cputime_to_expires(cputime
.utime
);
962 ptime
= utime
+ cputime_to_expires(cputime
.stime
);
963 sum_sched_runtime
= cputime
.sum_exec_runtime
;
965 prof_expires
= check_timers_list(timers
, firing
, ptime
);
966 virt_expires
= check_timers_list(++timers
, firing
, utime
);
967 sched_expires
= check_timers_list(++timers
, firing
, sum_sched_runtime
);
970 * Check for the special case process timers.
972 check_cpu_itimer(tsk
, &sig
->it
[CPUCLOCK_PROF
], &prof_expires
, ptime
,
974 check_cpu_itimer(tsk
, &sig
->it
[CPUCLOCK_VIRT
], &virt_expires
, utime
,
976 soft
= ACCESS_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
977 if (soft
!= RLIM_INFINITY
) {
978 unsigned long psecs
= cputime_to_secs(ptime
);
980 ACCESS_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_max
);
984 * At the hard limit, we just die.
985 * No need to calculate anything else now.
987 __group_send_sig_info(SIGKILL
, SEND_SIG_PRIV
, tsk
);
992 * At the soft limit, send a SIGXCPU every second.
994 __group_send_sig_info(SIGXCPU
, SEND_SIG_PRIV
, tsk
);
997 sig
->rlim
[RLIMIT_CPU
].rlim_cur
= soft
;
1000 x
= secs_to_cputime(soft
);
1001 if (!prof_expires
|| x
< prof_expires
) {
1006 sig
->cputime_expires
.prof_exp
= expires_to_cputime(prof_expires
);
1007 sig
->cputime_expires
.virt_exp
= expires_to_cputime(virt_expires
);
1008 sig
->cputime_expires
.sched_exp
= sched_expires
;
1009 if (task_cputime_zero(&sig
->cputime_expires
))
1010 stop_process_timers(sig
);
1014 * This is called from the signal code (via do_schedule_next_timer)
1015 * when the last timer signal was delivered and we have to reload the timer.
1017 void posix_cpu_timer_schedule(struct k_itimer
*timer
)
1019 struct sighand_struct
*sighand
;
1020 unsigned long flags
;
1021 struct task_struct
*p
= timer
->it
.cpu
.task
;
1022 unsigned long long now
;
1024 WARN_ON_ONCE(p
== NULL
);
1027 * Fetch the current sample and update the timer's expiry time.
1029 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
1030 cpu_clock_sample(timer
->it_clock
, p
, &now
);
1031 bump_cpu_timer(timer
, now
);
1032 if (unlikely(p
->exit_state
))
1035 /* Protect timer list r/w in arm_timer() */
1036 sighand
= lock_task_sighand(p
, &flags
);
1041 * Protect arm_timer() and timer sampling in case of call to
1042 * thread_group_cputime().
1044 sighand
= lock_task_sighand(p
, &flags
);
1045 if (unlikely(sighand
== NULL
)) {
1047 * The process has been reaped.
1048 * We can't even collect a sample any more.
1050 timer
->it
.cpu
.expires
= 0;
1052 } else if (unlikely(p
->exit_state
) && thread_group_empty(p
)) {
1053 unlock_task_sighand(p
, &flags
);
1054 /* Optimizations: if the process is dying, no need to rearm */
1057 cpu_timer_sample_group(timer
->it_clock
, p
, &now
);
1058 bump_cpu_timer(timer
, now
);
1059 /* Leave the sighand locked for the call below. */
1063 * Now re-arm for the new expiry time.
1065 WARN_ON_ONCE(!irqs_disabled());
1067 unlock_task_sighand(p
, &flags
);
1069 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1070 posix_cpu_timer_kick_nohz();
1072 timer
->it_overrun_last
= timer
->it_overrun
;
1073 timer
->it_overrun
= -1;
1074 ++timer
->it_requeue_pending
;
1078 * task_cputime_expired - Compare two task_cputime entities.
1080 * @sample: The task_cputime structure to be checked for expiration.
1081 * @expires: Expiration times, against which @sample will be checked.
1083 * Checks @sample against @expires to see if any field of @sample has expired.
1084 * Returns true if any field of the former is greater than the corresponding
1085 * field of the latter if the latter field is set. Otherwise returns false.
1087 static inline int task_cputime_expired(const struct task_cputime
*sample
,
1088 const struct task_cputime
*expires
)
1090 if (expires
->utime
&& sample
->utime
>= expires
->utime
)
1092 if (expires
->stime
&& sample
->utime
+ sample
->stime
>= expires
->stime
)
1094 if (expires
->sum_exec_runtime
!= 0 &&
1095 sample
->sum_exec_runtime
>= expires
->sum_exec_runtime
)
1101 * fastpath_timer_check - POSIX CPU timers fast path.
1103 * @tsk: The task (thread) being checked.
1105 * Check the task and thread group timers. If both are zero (there are no
1106 * timers set) return false. Otherwise snapshot the task and thread group
1107 * timers and compare them with the corresponding expiration times. Return
1108 * true if a timer has expired, else return false.
1110 static inline int fastpath_timer_check(struct task_struct
*tsk
)
1112 struct signal_struct
*sig
;
1113 cputime_t utime
, stime
;
1115 task_cputime(tsk
, &utime
, &stime
);
1117 if (!task_cputime_zero(&tsk
->cputime_expires
)) {
1118 struct task_cputime task_sample
= {
1121 .sum_exec_runtime
= tsk
->se
.sum_exec_runtime
1124 if (task_cputime_expired(&task_sample
, &tsk
->cputime_expires
))
1129 if (sig
->cputimer
.running
) {
1130 struct task_cputime group_sample
;
1132 raw_spin_lock(&sig
->cputimer
.lock
);
1133 group_sample
= sig
->cputimer
.cputime
;
1134 raw_spin_unlock(&sig
->cputimer
.lock
);
1136 if (task_cputime_expired(&group_sample
, &sig
->cputime_expires
))
1144 * This is called from the timer interrupt handler. The irq handler has
1145 * already updated our counts. We need to check if any timers fire now.
1146 * Interrupts are disabled.
1148 void run_posix_cpu_timers(struct task_struct
*tsk
)
1151 struct k_itimer
*timer
, *next
;
1152 unsigned long flags
;
1154 WARN_ON_ONCE(!irqs_disabled());
1157 * The fast path checks that there are no expired thread or thread
1158 * group timers. If that's so, just return.
1160 if (!fastpath_timer_check(tsk
))
1163 if (!lock_task_sighand(tsk
, &flags
))
1166 * Here we take off tsk->signal->cpu_timers[N] and
1167 * tsk->cpu_timers[N] all the timers that are firing, and
1168 * put them on the firing list.
1170 check_thread_timers(tsk
, &firing
);
1172 * If there are any active process wide timers (POSIX 1.b, itimers,
1173 * RLIMIT_CPU) cputimer must be running.
1175 if (tsk
->signal
->cputimer
.running
)
1176 check_process_timers(tsk
, &firing
);
1179 * We must release these locks before taking any timer's lock.
1180 * There is a potential race with timer deletion here, as the
1181 * siglock now protects our private firing list. We have set
1182 * the firing flag in each timer, so that a deletion attempt
1183 * that gets the timer lock before we do will give it up and
1184 * spin until we've taken care of that timer below.
1186 unlock_task_sighand(tsk
, &flags
);
1189 * Now that all the timers on our list have the firing flag,
1190 * no one will touch their list entries but us. We'll take
1191 * each timer's lock before clearing its firing flag, so no
1192 * timer call will interfere.
1194 list_for_each_entry_safe(timer
, next
, &firing
, it
.cpu
.entry
) {
1197 spin_lock(&timer
->it_lock
);
1198 list_del_init(&timer
->it
.cpu
.entry
);
1199 cpu_firing
= timer
->it
.cpu
.firing
;
1200 timer
->it
.cpu
.firing
= 0;
1202 * The firing flag is -1 if we collided with a reset
1203 * of the timer, which already reported this
1204 * almost-firing as an overrun. So don't generate an event.
1206 if (likely(cpu_firing
>= 0))
1207 cpu_timer_fire(timer
);
1208 spin_unlock(&timer
->it_lock
);
1213 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1214 * The tsk->sighand->siglock must be held by the caller.
1216 void set_process_cpu_timer(struct task_struct
*tsk
, unsigned int clock_idx
,
1217 cputime_t
*newval
, cputime_t
*oldval
)
1219 unsigned long long now
;
1221 WARN_ON_ONCE(clock_idx
== CPUCLOCK_SCHED
);
1222 cpu_timer_sample_group(clock_idx
, tsk
, &now
);
1226 * We are setting itimer. The *oldval is absolute and we update
1227 * it to be relative, *newval argument is relative and we update
1228 * it to be absolute.
1231 if (*oldval
<= now
) {
1232 /* Just about to fire. */
1233 *oldval
= cputime_one_jiffy
;
1245 * Update expiration cache if we are the earliest timer, or eventually
1246 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1248 switch (clock_idx
) {
1250 if (expires_gt(tsk
->signal
->cputime_expires
.prof_exp
, *newval
))
1251 tsk
->signal
->cputime_expires
.prof_exp
= *newval
;
1254 if (expires_gt(tsk
->signal
->cputime_expires
.virt_exp
, *newval
))
1255 tsk
->signal
->cputime_expires
.virt_exp
= *newval
;
1259 posix_cpu_timer_kick_nohz();
1262 static int do_cpu_nanosleep(const clockid_t which_clock
, int flags
,
1263 struct timespec
*rqtp
, struct itimerspec
*it
)
1265 struct k_itimer timer
;
1269 * Set up a temporary timer and then wait for it to go off.
1271 memset(&timer
, 0, sizeof timer
);
1272 spin_lock_init(&timer
.it_lock
);
1273 timer
.it_clock
= which_clock
;
1274 timer
.it_overrun
= -1;
1275 error
= posix_cpu_timer_create(&timer
);
1276 timer
.it_process
= current
;
1278 static struct itimerspec zero_it
;
1280 memset(it
, 0, sizeof *it
);
1281 it
->it_value
= *rqtp
;
1283 spin_lock_irq(&timer
.it_lock
);
1284 error
= posix_cpu_timer_set(&timer
, flags
, it
, NULL
);
1286 spin_unlock_irq(&timer
.it_lock
);
1290 while (!signal_pending(current
)) {
1291 if (timer
.it
.cpu
.expires
== 0) {
1293 * Our timer fired and was reset, below
1294 * deletion can not fail.
1296 posix_cpu_timer_del(&timer
);
1297 spin_unlock_irq(&timer
.it_lock
);
1302 * Block until cpu_timer_fire (or a signal) wakes us.
1304 __set_current_state(TASK_INTERRUPTIBLE
);
1305 spin_unlock_irq(&timer
.it_lock
);
1307 spin_lock_irq(&timer
.it_lock
);
1311 * We were interrupted by a signal.
1313 sample_to_timespec(which_clock
, timer
.it
.cpu
.expires
, rqtp
);
1314 error
= posix_cpu_timer_set(&timer
, 0, &zero_it
, it
);
1317 * Timer is now unarmed, deletion can not fail.
1319 posix_cpu_timer_del(&timer
);
1321 spin_unlock_irq(&timer
.it_lock
);
1323 while (error
== TIMER_RETRY
) {
1325 * We need to handle case when timer was or is in the
1326 * middle of firing. In other cases we already freed
1329 spin_lock_irq(&timer
.it_lock
);
1330 error
= posix_cpu_timer_del(&timer
);
1331 spin_unlock_irq(&timer
.it_lock
);
1334 if ((it
->it_value
.tv_sec
| it
->it_value
.tv_nsec
) == 0) {
1336 * It actually did fire already.
1341 error
= -ERESTART_RESTARTBLOCK
;
1347 static long posix_cpu_nsleep_restart(struct restart_block
*restart_block
);
1349 static int posix_cpu_nsleep(const clockid_t which_clock
, int flags
,
1350 struct timespec
*rqtp
, struct timespec __user
*rmtp
)
1352 struct restart_block
*restart_block
=
1353 ¤t_thread_info()->restart_block
;
1354 struct itimerspec it
;
1358 * Diagnose required errors first.
1360 if (CPUCLOCK_PERTHREAD(which_clock
) &&
1361 (CPUCLOCK_PID(which_clock
) == 0 ||
1362 CPUCLOCK_PID(which_clock
) == current
->pid
))
1365 error
= do_cpu_nanosleep(which_clock
, flags
, rqtp
, &it
);
1367 if (error
== -ERESTART_RESTARTBLOCK
) {
1369 if (flags
& TIMER_ABSTIME
)
1370 return -ERESTARTNOHAND
;
1372 * Report back to the user the time still remaining.
1374 if (rmtp
&& copy_to_user(rmtp
, &it
.it_value
, sizeof *rmtp
))
1377 restart_block
->fn
= posix_cpu_nsleep_restart
;
1378 restart_block
->nanosleep
.clockid
= which_clock
;
1379 restart_block
->nanosleep
.rmtp
= rmtp
;
1380 restart_block
->nanosleep
.expires
= timespec_to_ns(rqtp
);
1385 static long posix_cpu_nsleep_restart(struct restart_block
*restart_block
)
1387 clockid_t which_clock
= restart_block
->nanosleep
.clockid
;
1389 struct itimerspec it
;
1392 t
= ns_to_timespec(restart_block
->nanosleep
.expires
);
1394 error
= do_cpu_nanosleep(which_clock
, TIMER_ABSTIME
, &t
, &it
);
1396 if (error
== -ERESTART_RESTARTBLOCK
) {
1397 struct timespec __user
*rmtp
= restart_block
->nanosleep
.rmtp
;
1399 * Report back to the user the time still remaining.
1401 if (rmtp
&& copy_to_user(rmtp
, &it
.it_value
, sizeof *rmtp
))
1404 restart_block
->nanosleep
.expires
= timespec_to_ns(&t
);
1410 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1411 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1413 static int process_cpu_clock_getres(const clockid_t which_clock
,
1414 struct timespec
*tp
)
1416 return posix_cpu_clock_getres(PROCESS_CLOCK
, tp
);
1418 static int process_cpu_clock_get(const clockid_t which_clock
,
1419 struct timespec
*tp
)
1421 return posix_cpu_clock_get(PROCESS_CLOCK
, tp
);
1423 static int process_cpu_timer_create(struct k_itimer
*timer
)
1425 timer
->it_clock
= PROCESS_CLOCK
;
1426 return posix_cpu_timer_create(timer
);
1428 static int process_cpu_nsleep(const clockid_t which_clock
, int flags
,
1429 struct timespec
*rqtp
,
1430 struct timespec __user
*rmtp
)
1432 return posix_cpu_nsleep(PROCESS_CLOCK
, flags
, rqtp
, rmtp
);
1434 static long process_cpu_nsleep_restart(struct restart_block
*restart_block
)
1438 static int thread_cpu_clock_getres(const clockid_t which_clock
,
1439 struct timespec
*tp
)
1441 return posix_cpu_clock_getres(THREAD_CLOCK
, tp
);
1443 static int thread_cpu_clock_get(const clockid_t which_clock
,
1444 struct timespec
*tp
)
1446 return posix_cpu_clock_get(THREAD_CLOCK
, tp
);
1448 static int thread_cpu_timer_create(struct k_itimer
*timer
)
1450 timer
->it_clock
= THREAD_CLOCK
;
1451 return posix_cpu_timer_create(timer
);
1454 struct k_clock clock_posix_cpu
= {
1455 .clock_getres
= posix_cpu_clock_getres
,
1456 .clock_set
= posix_cpu_clock_set
,
1457 .clock_get
= posix_cpu_clock_get
,
1458 .timer_create
= posix_cpu_timer_create
,
1459 .nsleep
= posix_cpu_nsleep
,
1460 .nsleep_restart
= posix_cpu_nsleep_restart
,
1461 .timer_set
= posix_cpu_timer_set
,
1462 .timer_del
= posix_cpu_timer_del
,
1463 .timer_get
= posix_cpu_timer_get
,
1466 static __init
int init_posix_cpu_timers(void)
1468 struct k_clock process
= {
1469 .clock_getres
= process_cpu_clock_getres
,
1470 .clock_get
= process_cpu_clock_get
,
1471 .timer_create
= process_cpu_timer_create
,
1472 .nsleep
= process_cpu_nsleep
,
1473 .nsleep_restart
= process_cpu_nsleep_restart
,
1475 struct k_clock thread
= {
1476 .clock_getres
= thread_cpu_clock_getres
,
1477 .clock_get
= thread_cpu_clock_get
,
1478 .timer_create
= thread_cpu_timer_create
,
1482 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID
, &process
);
1483 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID
, &thread
);
1485 cputime_to_timespec(cputime_one_jiffy
, &ts
);
1486 onecputick
= ts
.tv_nsec
;
1487 WARN_ON(ts
.tv_sec
!= 0);
1491 __initcall(init_posix_cpu_timers
);