2 * Implement CPU time clocks for the POSIX clock interface.
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <asm/uaccess.h>
8 #include <linux/errno.h>
10 static int check_clock(clockid_t which_clock
)
13 struct task_struct
*p
;
14 const pid_t pid
= CPUCLOCK_PID(which_clock
);
16 if (CPUCLOCK_WHICH(which_clock
) >= CPUCLOCK_MAX
)
22 read_lock(&tasklist_lock
);
23 p
= find_task_by_pid(pid
);
24 if (!p
|| (CPUCLOCK_PERTHREAD(which_clock
) ?
25 p
->tgid
!= current
->tgid
: p
->tgid
!= pid
)) {
28 read_unlock(&tasklist_lock
);
33 static inline union cpu_time_count
34 timespec_to_sample(clockid_t which_clock
, const struct timespec
*tp
)
36 union cpu_time_count ret
;
37 ret
.sched
= 0; /* high half always zero when .cpu used */
38 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
39 ret
.sched
= tp
->tv_sec
* NSEC_PER_SEC
+ tp
->tv_nsec
;
41 ret
.cpu
= timespec_to_cputime(tp
);
46 static void sample_to_timespec(clockid_t which_clock
,
47 union cpu_time_count cpu
,
50 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
51 tp
->tv_sec
= div_long_long_rem(cpu
.sched
,
52 NSEC_PER_SEC
, &tp
->tv_nsec
);
54 cputime_to_timespec(cpu
.cpu
, tp
);
58 static inline int cpu_time_before(clockid_t which_clock
,
59 union cpu_time_count now
,
60 union cpu_time_count then
)
62 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
63 return now
.sched
< then
.sched
;
65 return cputime_lt(now
.cpu
, then
.cpu
);
68 static inline void cpu_time_add(clockid_t which_clock
,
69 union cpu_time_count
*acc
,
70 union cpu_time_count val
)
72 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
73 acc
->sched
+= val
.sched
;
75 acc
->cpu
= cputime_add(acc
->cpu
, val
.cpu
);
78 static inline union cpu_time_count
cpu_time_sub(clockid_t which_clock
,
79 union cpu_time_count a
,
80 union cpu_time_count b
)
82 if (CPUCLOCK_WHICH(which_clock
) == CPUCLOCK_SCHED
) {
85 a
.cpu
= cputime_sub(a
.cpu
, b
.cpu
);
91 * Update expiry time from increment, and increase overrun count,
92 * given the current clock sample.
94 static inline void bump_cpu_timer(struct k_itimer
*timer
,
95 union cpu_time_count now
)
99 if (timer
->it
.cpu
.incr
.sched
== 0)
102 if (CPUCLOCK_WHICH(timer
->it_clock
) == CPUCLOCK_SCHED
) {
103 unsigned long long delta
, incr
;
105 if (now
.sched
< timer
->it
.cpu
.expires
.sched
)
107 incr
= timer
->it
.cpu
.incr
.sched
;
108 delta
= now
.sched
+ incr
- timer
->it
.cpu
.expires
.sched
;
109 /* Don't use (incr*2 < delta), incr*2 might overflow. */
110 for (i
= 0; incr
< delta
- incr
; i
++)
112 for (; i
>= 0; incr
>>= 1, i
--) {
115 timer
->it
.cpu
.expires
.sched
+= incr
;
116 timer
->it_overrun
+= 1 << i
;
120 cputime_t delta
, incr
;
122 if (cputime_lt(now
.cpu
, timer
->it
.cpu
.expires
.cpu
))
124 incr
= timer
->it
.cpu
.incr
.cpu
;
125 delta
= cputime_sub(cputime_add(now
.cpu
, incr
),
126 timer
->it
.cpu
.expires
.cpu
);
127 /* Don't use (incr*2 < delta), incr*2 might overflow. */
128 for (i
= 0; cputime_lt(incr
, cputime_sub(delta
, incr
)); i
++)
129 incr
= cputime_add(incr
, incr
);
130 for (; i
>= 0; incr
= cputime_halve(incr
), i
--) {
131 if (cputime_le(delta
, incr
))
133 timer
->it
.cpu
.expires
.cpu
=
134 cputime_add(timer
->it
.cpu
.expires
.cpu
, incr
);
135 timer
->it_overrun
+= 1 << i
;
136 delta
= cputime_sub(delta
, incr
);
141 static inline cputime_t
prof_ticks(struct task_struct
*p
)
143 return cputime_add(p
->utime
, p
->stime
);
145 static inline cputime_t
virt_ticks(struct task_struct
*p
)
149 static inline unsigned long long sched_ns(struct task_struct
*p
)
151 return (p
== current
) ? current_sched_time(p
) : p
->sched_time
;
154 int posix_cpu_clock_getres(clockid_t which_clock
, struct timespec
*tp
)
156 int error
= check_clock(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.
172 int posix_cpu_clock_set(clockid_t which_clock
, const struct timespec
*tp
)
175 * You can never reset a CPU clock, but we check for other errors
176 * in the call before failing with EPERM.
178 int error
= check_clock(which_clock
);
187 * Sample a per-thread clock for the given task.
189 static int cpu_clock_sample(clockid_t which_clock
, struct task_struct
*p
,
190 union cpu_time_count
*cpu
)
192 switch (CPUCLOCK_WHICH(which_clock
)) {
196 cpu
->cpu
= prof_ticks(p
);
199 cpu
->cpu
= virt_ticks(p
);
202 cpu
->sched
= sched_ns(p
);
209 * Sample a process (thread group) clock for the given group_leader task.
210 * Must be called with tasklist_lock held for reading.
211 * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
213 static int cpu_clock_sample_group_locked(unsigned int clock_idx
,
214 struct task_struct
*p
,
215 union cpu_time_count
*cpu
)
217 struct task_struct
*t
= p
;
222 cpu
->cpu
= cputime_add(p
->signal
->utime
, p
->signal
->stime
);
224 cpu
->cpu
= cputime_add(cpu
->cpu
, prof_ticks(t
));
229 cpu
->cpu
= p
->signal
->utime
;
231 cpu
->cpu
= cputime_add(cpu
->cpu
, virt_ticks(t
));
236 cpu
->sched
= p
->signal
->sched_time
;
237 /* Add in each other live thread. */
238 while ((t
= next_thread(t
)) != p
) {
239 cpu
->sched
+= t
->sched_time
;
241 if (p
->tgid
== current
->tgid
) {
243 * We're sampling ourselves, so include the
244 * cycles not yet banked. We still omit
245 * other threads running on other CPUs,
246 * so the total can always be behind as
247 * much as max(nthreads-1,ncpus) * (NSEC_PER_SEC/HZ).
249 cpu
->sched
+= current_sched_time(current
);
251 cpu
->sched
+= p
->sched_time
;
259 * Sample a process (thread group) clock for the given group_leader task.
260 * Must be called with tasklist_lock held for reading.
262 static int cpu_clock_sample_group(clockid_t which_clock
,
263 struct task_struct
*p
,
264 union cpu_time_count
*cpu
)
268 spin_lock_irqsave(&p
->sighand
->siglock
, flags
);
269 ret
= cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock
), p
,
271 spin_unlock_irqrestore(&p
->sighand
->siglock
, flags
);
276 int posix_cpu_clock_get(clockid_t which_clock
, struct timespec
*tp
)
278 const pid_t pid
= CPUCLOCK_PID(which_clock
);
280 union cpu_time_count rtn
;
284 * Special case constant value for our own clocks.
285 * We don't have to do any lookup to find ourselves.
287 if (CPUCLOCK_PERTHREAD(which_clock
)) {
289 * Sampling just ourselves we can do with no locking.
291 error
= cpu_clock_sample(which_clock
,
294 read_lock(&tasklist_lock
);
295 error
= cpu_clock_sample_group(which_clock
,
297 read_unlock(&tasklist_lock
);
301 * Find the given PID, and validate that the caller
302 * should be able to see it.
304 struct task_struct
*p
;
305 read_lock(&tasklist_lock
);
306 p
= find_task_by_pid(pid
);
308 if (CPUCLOCK_PERTHREAD(which_clock
)) {
309 if (p
->tgid
== current
->tgid
) {
310 error
= cpu_clock_sample(which_clock
,
313 } else if (p
->tgid
== pid
&& p
->signal
) {
314 error
= cpu_clock_sample_group(which_clock
,
318 read_unlock(&tasklist_lock
);
323 sample_to_timespec(which_clock
, rtn
, tp
);
329 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
330 * This is called from sys_timer_create with the new timer already locked.
332 int posix_cpu_timer_create(struct k_itimer
*new_timer
)
335 const pid_t pid
= CPUCLOCK_PID(new_timer
->it_clock
);
336 struct task_struct
*p
;
338 if (CPUCLOCK_WHICH(new_timer
->it_clock
) >= CPUCLOCK_MAX
)
341 INIT_LIST_HEAD(&new_timer
->it
.cpu
.entry
);
342 new_timer
->it
.cpu
.incr
.sched
= 0;
343 new_timer
->it
.cpu
.expires
.sched
= 0;
345 read_lock(&tasklist_lock
);
346 if (CPUCLOCK_PERTHREAD(new_timer
->it_clock
)) {
350 p
= find_task_by_pid(pid
);
351 if (p
&& p
->tgid
!= current
->tgid
)
356 p
= current
->group_leader
;
358 p
= find_task_by_pid(pid
);
359 if (p
&& p
->tgid
!= pid
)
363 new_timer
->it
.cpu
.task
= p
;
369 read_unlock(&tasklist_lock
);
375 * Clean up a CPU-clock timer that is about to be destroyed.
376 * This is called from timer deletion with the timer already locked.
377 * If we return TIMER_RETRY, it's necessary to release the timer's lock
378 * and try again. (This happens when the timer is in the middle of firing.)
380 int posix_cpu_timer_del(struct k_itimer
*timer
)
382 struct task_struct
*p
= timer
->it
.cpu
.task
;
385 if (likely(p
!= NULL
)) {
386 read_lock(&tasklist_lock
);
387 if (unlikely(p
->signal
== NULL
)) {
389 * We raced with the reaping of the task.
390 * The deletion should have cleared us off the list.
392 BUG_ON(!list_empty(&timer
->it
.cpu
.entry
));
394 spin_lock(&p
->sighand
->siglock
);
395 if (timer
->it
.cpu
.firing
)
398 list_del(&timer
->it
.cpu
.entry
);
399 spin_unlock(&p
->sighand
->siglock
);
401 read_unlock(&tasklist_lock
);
411 * Clean out CPU timers still ticking when a thread exited. The task
412 * pointer is cleared, and the expiry time is replaced with the residual
413 * time for later timer_gettime calls to return.
414 * This must be called with the siglock held.
416 static void cleanup_timers(struct list_head
*head
,
417 cputime_t utime
, cputime_t stime
,
418 unsigned long long sched_time
)
420 struct cpu_timer_list
*timer
, *next
;
421 cputime_t ptime
= cputime_add(utime
, stime
);
423 list_for_each_entry_safe(timer
, next
, head
, entry
) {
424 list_del_init(&timer
->entry
);
425 if (cputime_lt(timer
->expires
.cpu
, ptime
)) {
426 timer
->expires
.cpu
= cputime_zero
;
428 timer
->expires
.cpu
= cputime_sub(timer
->expires
.cpu
,
434 list_for_each_entry_safe(timer
, next
, head
, entry
) {
435 list_del_init(&timer
->entry
);
436 if (cputime_lt(timer
->expires
.cpu
, utime
)) {
437 timer
->expires
.cpu
= cputime_zero
;
439 timer
->expires
.cpu
= cputime_sub(timer
->expires
.cpu
,
445 list_for_each_entry_safe(timer
, next
, head
, entry
) {
446 list_del_init(&timer
->entry
);
447 if (timer
->expires
.sched
< sched_time
) {
448 timer
->expires
.sched
= 0;
450 timer
->expires
.sched
-= sched_time
;
456 * These are both called with the siglock held, when the current thread
457 * is being reaped. When the final (leader) thread in the group is reaped,
458 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
460 void posix_cpu_timers_exit(struct task_struct
*tsk
)
462 cleanup_timers(tsk
->cpu_timers
,
463 tsk
->utime
, tsk
->stime
, tsk
->sched_time
);
466 void posix_cpu_timers_exit_group(struct task_struct
*tsk
)
468 cleanup_timers(tsk
->signal
->cpu_timers
,
469 cputime_add(tsk
->utime
, tsk
->signal
->utime
),
470 cputime_add(tsk
->stime
, tsk
->signal
->stime
),
471 tsk
->sched_time
+ tsk
->signal
->sched_time
);
476 * Set the expiry times of all the threads in the process so one of them
477 * will go off before the process cumulative expiry total is reached.
479 static void process_timer_rebalance(struct task_struct
*p
,
480 unsigned int clock_idx
,
481 union cpu_time_count expires
,
482 union cpu_time_count val
)
484 cputime_t ticks
, left
;
485 unsigned long long ns
, nsleft
;
486 struct task_struct
*t
= p
;
487 unsigned int nthreads
= atomic_read(&p
->signal
->live
);
494 left
= cputime_div(cputime_sub(expires
.cpu
, val
.cpu
),
497 if (!unlikely(t
->exit_state
)) {
498 ticks
= cputime_add(prof_ticks(t
), left
);
499 if (cputime_eq(t
->it_prof_expires
,
501 cputime_gt(t
->it_prof_expires
, ticks
)) {
502 t
->it_prof_expires
= ticks
;
509 left
= cputime_div(cputime_sub(expires
.cpu
, val
.cpu
),
512 if (!unlikely(t
->exit_state
)) {
513 ticks
= cputime_add(virt_ticks(t
), left
);
514 if (cputime_eq(t
->it_virt_expires
,
516 cputime_gt(t
->it_virt_expires
, ticks
)) {
517 t
->it_virt_expires
= ticks
;
524 nsleft
= expires
.sched
- val
.sched
;
525 do_div(nsleft
, nthreads
);
527 if (!unlikely(t
->exit_state
)) {
528 ns
= t
->sched_time
+ nsleft
;
529 if (t
->it_sched_expires
== 0 ||
530 t
->it_sched_expires
> ns
) {
531 t
->it_sched_expires
= ns
;
540 static void clear_dead_task(struct k_itimer
*timer
, union cpu_time_count now
)
543 * That's all for this thread or process.
544 * We leave our residual in expires to be reported.
546 put_task_struct(timer
->it
.cpu
.task
);
547 timer
->it
.cpu
.task
= NULL
;
548 timer
->it
.cpu
.expires
= cpu_time_sub(timer
->it_clock
,
549 timer
->it
.cpu
.expires
,
554 * Insert the timer on the appropriate list before any timers that
555 * expire later. This must be called with the tasklist_lock held
556 * for reading, and interrupts disabled.
558 static void arm_timer(struct k_itimer
*timer
, union cpu_time_count now
)
560 struct task_struct
*p
= timer
->it
.cpu
.task
;
561 struct list_head
*head
, *listpos
;
562 struct cpu_timer_list
*const nt
= &timer
->it
.cpu
;
563 struct cpu_timer_list
*next
;
566 head
= (CPUCLOCK_PERTHREAD(timer
->it_clock
) ?
567 p
->cpu_timers
: p
->signal
->cpu_timers
);
568 head
+= CPUCLOCK_WHICH(timer
->it_clock
);
570 BUG_ON(!irqs_disabled());
571 spin_lock(&p
->sighand
->siglock
);
574 if (CPUCLOCK_WHICH(timer
->it_clock
) == CPUCLOCK_SCHED
) {
575 list_for_each_entry(next
, head
, entry
) {
576 if (next
->expires
.sched
> nt
->expires
.sched
) {
577 listpos
= &next
->entry
;
582 list_for_each_entry(next
, head
, entry
) {
583 if (cputime_gt(next
->expires
.cpu
, nt
->expires
.cpu
)) {
584 listpos
= &next
->entry
;
589 list_add(&nt
->entry
, listpos
);
591 if (listpos
== head
) {
593 * We are the new earliest-expiring timer.
594 * If we are a thread timer, there can always
595 * be a process timer telling us to stop earlier.
598 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
599 switch (CPUCLOCK_WHICH(timer
->it_clock
)) {
603 if (cputime_eq(p
->it_prof_expires
,
605 cputime_gt(p
->it_prof_expires
,
607 p
->it_prof_expires
= nt
->expires
.cpu
;
610 if (cputime_eq(p
->it_virt_expires
,
612 cputime_gt(p
->it_virt_expires
,
614 p
->it_virt_expires
= nt
->expires
.cpu
;
617 if (p
->it_sched_expires
== 0 ||
618 p
->it_sched_expires
> nt
->expires
.sched
)
619 p
->it_sched_expires
= nt
->expires
.sched
;
624 * For a process timer, we must balance
625 * all the live threads' expirations.
627 switch (CPUCLOCK_WHICH(timer
->it_clock
)) {
631 if (!cputime_eq(p
->signal
->it_virt_expires
,
633 cputime_lt(p
->signal
->it_virt_expires
,
634 timer
->it
.cpu
.expires
.cpu
))
638 if (!cputime_eq(p
->signal
->it_prof_expires
,
640 cputime_lt(p
->signal
->it_prof_expires
,
641 timer
->it
.cpu
.expires
.cpu
))
643 i
= p
->signal
->rlim
[RLIMIT_CPU
].rlim_cur
;
644 if (i
!= RLIM_INFINITY
&&
645 i
<= cputime_to_secs(timer
->it
.cpu
.expires
.cpu
))
650 process_timer_rebalance(
652 CPUCLOCK_WHICH(timer
->it_clock
),
653 timer
->it
.cpu
.expires
, now
);
659 spin_unlock(&p
->sighand
->siglock
);
663 * The timer is locked, fire it and arrange for its reload.
665 static void cpu_timer_fire(struct k_itimer
*timer
)
667 if (unlikely(timer
->sigq
== NULL
)) {
669 * This a special case for clock_nanosleep,
670 * not a normal timer from sys_timer_create.
672 wake_up_process(timer
->it_process
);
673 timer
->it
.cpu
.expires
.sched
= 0;
674 } else if (timer
->it
.cpu
.incr
.sched
== 0) {
676 * One-shot timer. Clear it as soon as it's fired.
678 posix_timer_event(timer
, 0);
679 timer
->it
.cpu
.expires
.sched
= 0;
680 } else if (posix_timer_event(timer
, ++timer
->it_requeue_pending
)) {
682 * The signal did not get queued because the signal
683 * was ignored, so we won't get any callback to
684 * reload the timer. But we need to keep it
685 * ticking in case the signal is deliverable next time.
687 posix_cpu_timer_schedule(timer
);
692 * Guts of sys_timer_settime for CPU timers.
693 * This is called with the timer locked and interrupts disabled.
694 * If we return TIMER_RETRY, it's necessary to release the timer's lock
695 * and try again. (This happens when the timer is in the middle of firing.)
697 int posix_cpu_timer_set(struct k_itimer
*timer
, int flags
,
698 struct itimerspec
*new, struct itimerspec
*old
)
700 struct task_struct
*p
= timer
->it
.cpu
.task
;
701 union cpu_time_count old_expires
, new_expires
, val
;
704 if (unlikely(p
== NULL
)) {
706 * Timer refers to a dead task's clock.
711 new_expires
= timespec_to_sample(timer
->it_clock
, &new->it_value
);
713 read_lock(&tasklist_lock
);
715 * We need the tasklist_lock to protect against reaping that
716 * clears p->signal. If p has just been reaped, we can no
717 * longer get any information about it at all.
719 if (unlikely(p
->signal
== NULL
)) {
720 read_unlock(&tasklist_lock
);
722 timer
->it
.cpu
.task
= NULL
;
727 * Disarm any old timer after extracting its expiry time.
729 BUG_ON(!irqs_disabled());
730 spin_lock(&p
->sighand
->siglock
);
731 old_expires
= timer
->it
.cpu
.expires
;
732 list_del_init(&timer
->it
.cpu
.entry
);
733 spin_unlock(&p
->sighand
->siglock
);
736 * We need to sample the current value to convert the new
737 * value from to relative and absolute, and to convert the
738 * old value from absolute to relative. To set a process
739 * timer, we need a sample to balance the thread expiry
740 * times (in arm_timer). With an absolute time, we must
741 * check if it's already passed. In short, we need a sample.
743 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
744 cpu_clock_sample(timer
->it_clock
, p
, &val
);
746 cpu_clock_sample_group(timer
->it_clock
, p
, &val
);
750 if (old_expires
.sched
== 0) {
751 old
->it_value
.tv_sec
= 0;
752 old
->it_value
.tv_nsec
= 0;
755 * Update the timer in case it has
756 * overrun already. If it has,
757 * we'll report it as having overrun
758 * and with the next reloaded timer
759 * already ticking, though we are
760 * swallowing that pending
761 * notification here to install the
764 bump_cpu_timer(timer
, val
);
765 if (cpu_time_before(timer
->it_clock
, val
,
766 timer
->it
.cpu
.expires
)) {
767 old_expires
= cpu_time_sub(
769 timer
->it
.cpu
.expires
, val
);
770 sample_to_timespec(timer
->it_clock
,
774 old
->it_value
.tv_nsec
= 1;
775 old
->it_value
.tv_sec
= 0;
780 if (unlikely(timer
->it
.cpu
.firing
)) {
782 * We are colliding with the timer actually firing.
783 * Punt after filling in the timer's old value, and
784 * disable this firing since we are already reporting
785 * it as an overrun (thanks to bump_cpu_timer above).
787 read_unlock(&tasklist_lock
);
788 timer
->it
.cpu
.firing
= -1;
793 if (new_expires
.sched
!= 0 && !(flags
& TIMER_ABSTIME
)) {
794 cpu_time_add(timer
->it_clock
, &new_expires
, val
);
798 * Install the new expiry time (or zero).
799 * For a timer with no notification action, we don't actually
800 * arm the timer (we'll just fake it for timer_gettime).
802 timer
->it
.cpu
.expires
= new_expires
;
803 if (new_expires
.sched
!= 0 &&
804 (timer
->it_sigev_notify
& ~SIGEV_THREAD_ID
) != SIGEV_NONE
&&
805 cpu_time_before(timer
->it_clock
, val
, new_expires
)) {
806 arm_timer(timer
, val
);
809 read_unlock(&tasklist_lock
);
812 * Install the new reload setting, and
813 * set up the signal and overrun bookkeeping.
815 timer
->it
.cpu
.incr
= timespec_to_sample(timer
->it_clock
,
819 * This acts as a modification timestamp for the timer,
820 * so any automatic reload attempt will punt on seeing
821 * that we have reset the timer manually.
823 timer
->it_requeue_pending
= (timer
->it_requeue_pending
+ 2) &
825 timer
->it_overrun_last
= 0;
826 timer
->it_overrun
= -1;
828 if (new_expires
.sched
!= 0 &&
829 (timer
->it_sigev_notify
& ~SIGEV_THREAD_ID
) != SIGEV_NONE
&&
830 !cpu_time_before(timer
->it_clock
, val
, new_expires
)) {
832 * The designated time already passed, so we notify
833 * immediately, even if the thread never runs to
834 * accumulate more time on this clock.
836 cpu_timer_fire(timer
);
842 sample_to_timespec(timer
->it_clock
,
843 timer
->it
.cpu
.incr
, &old
->it_interval
);
848 void posix_cpu_timer_get(struct k_itimer
*timer
, struct itimerspec
*itp
)
850 union cpu_time_count now
;
851 struct task_struct
*p
= timer
->it
.cpu
.task
;
855 * Easy part: convert the reload time.
857 sample_to_timespec(timer
->it_clock
,
858 timer
->it
.cpu
.incr
, &itp
->it_interval
);
860 if (timer
->it
.cpu
.expires
.sched
== 0) { /* Timer not armed at all. */
861 itp
->it_value
.tv_sec
= itp
->it_value
.tv_nsec
= 0;
865 if (unlikely(p
== NULL
)) {
867 * This task already died and the timer will never fire.
868 * In this case, expires is actually the dead value.
871 sample_to_timespec(timer
->it_clock
, timer
->it
.cpu
.expires
,
877 * Sample the clock to take the difference with the expiry time.
879 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
880 cpu_clock_sample(timer
->it_clock
, p
, &now
);
881 clear_dead
= p
->exit_state
;
883 read_lock(&tasklist_lock
);
884 if (unlikely(p
->signal
== NULL
)) {
886 * The process has been reaped.
887 * We can't even collect a sample any more.
888 * Call the timer disarmed, nothing else to do.
891 timer
->it
.cpu
.task
= NULL
;
892 timer
->it
.cpu
.expires
.sched
= 0;
893 read_unlock(&tasklist_lock
);
896 cpu_clock_sample_group(timer
->it_clock
, p
, &now
);
897 clear_dead
= (unlikely(p
->exit_state
) &&
898 thread_group_empty(p
));
900 read_unlock(&tasklist_lock
);
903 if ((timer
->it_sigev_notify
& ~SIGEV_THREAD_ID
) == SIGEV_NONE
) {
904 if (timer
->it
.cpu
.incr
.sched
== 0 &&
905 cpu_time_before(timer
->it_clock
,
906 timer
->it
.cpu
.expires
, now
)) {
908 * Do-nothing timer expired and has no reload,
909 * so it's as if it was never set.
911 timer
->it
.cpu
.expires
.sched
= 0;
912 itp
->it_value
.tv_sec
= itp
->it_value
.tv_nsec
= 0;
916 * Account for any expirations and reloads that should
919 bump_cpu_timer(timer
, now
);
922 if (unlikely(clear_dead
)) {
924 * We've noticed that the thread is dead, but
925 * not yet reaped. Take this opportunity to
928 clear_dead_task(timer
, now
);
932 if (cpu_time_before(timer
->it_clock
, now
, timer
->it
.cpu
.expires
)) {
933 sample_to_timespec(timer
->it_clock
,
934 cpu_time_sub(timer
->it_clock
,
935 timer
->it
.cpu
.expires
, now
),
939 * The timer should have expired already, but the firing
940 * hasn't taken place yet. Say it's just about to expire.
942 itp
->it_value
.tv_nsec
= 1;
943 itp
->it_value
.tv_sec
= 0;
948 * Check for any per-thread CPU timers that have fired and move them off
949 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
950 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
952 static void check_thread_timers(struct task_struct
*tsk
,
953 struct list_head
*firing
)
956 struct list_head
*timers
= tsk
->cpu_timers
;
959 tsk
->it_prof_expires
= cputime_zero
;
960 while (!list_empty(timers
)) {
961 struct cpu_timer_list
*t
= list_entry(timers
->next
,
962 struct cpu_timer_list
,
964 if (!--maxfire
|| cputime_lt(prof_ticks(tsk
), t
->expires
.cpu
)) {
965 tsk
->it_prof_expires
= t
->expires
.cpu
;
969 list_move_tail(&t
->entry
, firing
);
974 tsk
->it_virt_expires
= cputime_zero
;
975 while (!list_empty(timers
)) {
976 struct cpu_timer_list
*t
= list_entry(timers
->next
,
977 struct cpu_timer_list
,
979 if (!--maxfire
|| cputime_lt(virt_ticks(tsk
), t
->expires
.cpu
)) {
980 tsk
->it_virt_expires
= t
->expires
.cpu
;
984 list_move_tail(&t
->entry
, firing
);
989 tsk
->it_sched_expires
= 0;
990 while (!list_empty(timers
)) {
991 struct cpu_timer_list
*t
= list_entry(timers
->next
,
992 struct cpu_timer_list
,
994 if (!--maxfire
|| tsk
->sched_time
< t
->expires
.sched
) {
995 tsk
->it_sched_expires
= t
->expires
.sched
;
999 list_move_tail(&t
->entry
, firing
);
1004 * Check for any per-thread CPU timers that have fired and move them
1005 * off the tsk->*_timers list onto the firing list. Per-thread timers
1006 * have already been taken off.
1008 static void check_process_timers(struct task_struct
*tsk
,
1009 struct list_head
*firing
)
1012 struct signal_struct
*const sig
= tsk
->signal
;
1013 cputime_t utime
, stime
, ptime
, virt_expires
, prof_expires
;
1014 unsigned long long sched_time
, sched_expires
;
1015 struct task_struct
*t
;
1016 struct list_head
*timers
= sig
->cpu_timers
;
1019 * Don't sample the current process CPU clocks if there are no timers.
1021 if (list_empty(&timers
[CPUCLOCK_PROF
]) &&
1022 cputime_eq(sig
->it_prof_expires
, cputime_zero
) &&
1023 sig
->rlim
[RLIMIT_CPU
].rlim_cur
== RLIM_INFINITY
&&
1024 list_empty(&timers
[CPUCLOCK_VIRT
]) &&
1025 cputime_eq(sig
->it_virt_expires
, cputime_zero
) &&
1026 list_empty(&timers
[CPUCLOCK_SCHED
]))
1030 * Collect the current process totals.
1034 sched_time
= sig
->sched_time
;
1037 utime
= cputime_add(utime
, t
->utime
);
1038 stime
= cputime_add(stime
, t
->stime
);
1039 sched_time
+= t
->sched_time
;
1042 ptime
= cputime_add(utime
, stime
);
1045 prof_expires
= cputime_zero
;
1046 while (!list_empty(timers
)) {
1047 struct cpu_timer_list
*t
= list_entry(timers
->next
,
1048 struct cpu_timer_list
,
1050 if (!--maxfire
|| cputime_lt(ptime
, t
->expires
.cpu
)) {
1051 prof_expires
= t
->expires
.cpu
;
1055 list_move_tail(&t
->entry
, firing
);
1060 virt_expires
= cputime_zero
;
1061 while (!list_empty(timers
)) {
1062 struct cpu_timer_list
*t
= list_entry(timers
->next
,
1063 struct cpu_timer_list
,
1065 if (!--maxfire
|| cputime_lt(utime
, t
->expires
.cpu
)) {
1066 virt_expires
= t
->expires
.cpu
;
1070 list_move_tail(&t
->entry
, firing
);
1076 while (!list_empty(timers
)) {
1077 struct cpu_timer_list
*t
= list_entry(timers
->next
,
1078 struct cpu_timer_list
,
1080 if (!--maxfire
|| sched_time
< t
->expires
.sched
) {
1081 sched_expires
= t
->expires
.sched
;
1085 list_move_tail(&t
->entry
, firing
);
1089 * Check for the special case process timers.
1091 if (!cputime_eq(sig
->it_prof_expires
, cputime_zero
)) {
1092 if (cputime_ge(ptime
, sig
->it_prof_expires
)) {
1093 /* ITIMER_PROF fires and reloads. */
1094 sig
->it_prof_expires
= sig
->it_prof_incr
;
1095 if (!cputime_eq(sig
->it_prof_expires
, cputime_zero
)) {
1096 sig
->it_prof_expires
= cputime_add(
1097 sig
->it_prof_expires
, ptime
);
1099 __group_send_sig_info(SIGPROF
, SEND_SIG_PRIV
, tsk
);
1101 if (!cputime_eq(sig
->it_prof_expires
, cputime_zero
) &&
1102 (cputime_eq(prof_expires
, cputime_zero
) ||
1103 cputime_lt(sig
->it_prof_expires
, prof_expires
))) {
1104 prof_expires
= sig
->it_prof_expires
;
1107 if (!cputime_eq(sig
->it_virt_expires
, cputime_zero
)) {
1108 if (cputime_ge(utime
, sig
->it_virt_expires
)) {
1109 /* ITIMER_VIRTUAL fires and reloads. */
1110 sig
->it_virt_expires
= sig
->it_virt_incr
;
1111 if (!cputime_eq(sig
->it_virt_expires
, cputime_zero
)) {
1112 sig
->it_virt_expires
= cputime_add(
1113 sig
->it_virt_expires
, utime
);
1115 __group_send_sig_info(SIGVTALRM
, SEND_SIG_PRIV
, tsk
);
1117 if (!cputime_eq(sig
->it_virt_expires
, cputime_zero
) &&
1118 (cputime_eq(virt_expires
, cputime_zero
) ||
1119 cputime_lt(sig
->it_virt_expires
, virt_expires
))) {
1120 virt_expires
= sig
->it_virt_expires
;
1123 if (sig
->rlim
[RLIMIT_CPU
].rlim_cur
!= RLIM_INFINITY
) {
1124 unsigned long psecs
= cputime_to_secs(ptime
);
1126 if (psecs
>= sig
->rlim
[RLIMIT_CPU
].rlim_max
) {
1128 * At the hard limit, we just die.
1129 * No need to calculate anything else now.
1131 __group_send_sig_info(SIGKILL
, SEND_SIG_PRIV
, tsk
);
1134 if (psecs
>= sig
->rlim
[RLIMIT_CPU
].rlim_cur
) {
1136 * At the soft limit, send a SIGXCPU every second.
1138 __group_send_sig_info(SIGXCPU
, SEND_SIG_PRIV
, tsk
);
1139 if (sig
->rlim
[RLIMIT_CPU
].rlim_cur
1140 < sig
->rlim
[RLIMIT_CPU
].rlim_max
) {
1141 sig
->rlim
[RLIMIT_CPU
].rlim_cur
++;
1144 x
= secs_to_cputime(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1145 if (cputime_eq(prof_expires
, cputime_zero
) ||
1146 cputime_lt(x
, prof_expires
)) {
1151 if (!cputime_eq(prof_expires
, cputime_zero
) ||
1152 !cputime_eq(virt_expires
, cputime_zero
) ||
1153 sched_expires
!= 0) {
1155 * Rebalance the threads' expiry times for the remaining
1156 * process CPU timers.
1159 cputime_t prof_left
, virt_left
, ticks
;
1160 unsigned long long sched_left
, sched
;
1161 const unsigned int nthreads
= atomic_read(&sig
->live
);
1163 prof_left
= cputime_sub(prof_expires
, utime
);
1164 prof_left
= cputime_sub(prof_left
, stime
);
1165 prof_left
= cputime_div(prof_left
, nthreads
);
1166 virt_left
= cputime_sub(virt_expires
, utime
);
1167 virt_left
= cputime_div(virt_left
, nthreads
);
1168 if (sched_expires
) {
1169 sched_left
= sched_expires
- sched_time
;
1170 do_div(sched_left
, nthreads
);
1176 ticks
= cputime_add(cputime_add(t
->utime
, t
->stime
),
1178 if (!cputime_eq(prof_expires
, cputime_zero
) &&
1179 (cputime_eq(t
->it_prof_expires
, cputime_zero
) ||
1180 cputime_gt(t
->it_prof_expires
, ticks
))) {
1181 t
->it_prof_expires
= ticks
;
1184 ticks
= cputime_add(t
->utime
, virt_left
);
1185 if (!cputime_eq(virt_expires
, cputime_zero
) &&
1186 (cputime_eq(t
->it_virt_expires
, cputime_zero
) ||
1187 cputime_gt(t
->it_virt_expires
, ticks
))) {
1188 t
->it_virt_expires
= ticks
;
1191 sched
= t
->sched_time
+ sched_left
;
1192 if (sched_expires
&& (t
->it_sched_expires
== 0 ||
1193 t
->it_sched_expires
> sched
)) {
1194 t
->it_sched_expires
= sched
;
1199 } while (unlikely(t
->exit_state
));
1205 * This is called from the signal code (via do_schedule_next_timer)
1206 * when the last timer signal was delivered and we have to reload the timer.
1208 void posix_cpu_timer_schedule(struct k_itimer
*timer
)
1210 struct task_struct
*p
= timer
->it
.cpu
.task
;
1211 union cpu_time_count now
;
1213 if (unlikely(p
== NULL
))
1215 * The task was cleaned up already, no future firings.
1220 * Fetch the current sample and update the timer's expiry time.
1222 if (CPUCLOCK_PERTHREAD(timer
->it_clock
)) {
1223 cpu_clock_sample(timer
->it_clock
, p
, &now
);
1224 bump_cpu_timer(timer
, now
);
1225 if (unlikely(p
->exit_state
)) {
1226 clear_dead_task(timer
, now
);
1229 read_lock(&tasklist_lock
); /* arm_timer needs it. */
1231 read_lock(&tasklist_lock
);
1232 if (unlikely(p
->signal
== NULL
)) {
1234 * The process has been reaped.
1235 * We can't even collect a sample any more.
1238 timer
->it
.cpu
.task
= p
= NULL
;
1239 timer
->it
.cpu
.expires
.sched
= 0;
1240 read_unlock(&tasklist_lock
);
1242 } else if (unlikely(p
->exit_state
) && thread_group_empty(p
)) {
1244 * We've noticed that the thread is dead, but
1245 * not yet reaped. Take this opportunity to
1246 * drop our task ref.
1248 clear_dead_task(timer
, now
);
1249 read_unlock(&tasklist_lock
);
1252 cpu_clock_sample_group(timer
->it_clock
, p
, &now
);
1253 bump_cpu_timer(timer
, now
);
1254 /* Leave the tasklist_lock locked for the call below. */
1258 * Now re-arm for the new expiry time.
1260 arm_timer(timer
, now
);
1262 read_unlock(&tasklist_lock
);
1266 * This is called from the timer interrupt handler. The irq handler has
1267 * already updated our counts. We need to check if any timers fire now.
1268 * Interrupts are disabled.
1270 void run_posix_cpu_timers(struct task_struct
*tsk
)
1273 struct k_itimer
*timer
, *next
;
1275 BUG_ON(!irqs_disabled());
1277 #define UNEXPIRED(clock) \
1278 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1279 cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1281 if (UNEXPIRED(prof
) && UNEXPIRED(virt
) &&
1282 (tsk
->it_sched_expires
== 0 ||
1283 tsk
->sched_time
< tsk
->it_sched_expires
))
1288 BUG_ON(tsk
->exit_state
);
1291 * Double-check with locks held.
1293 read_lock(&tasklist_lock
);
1294 spin_lock(&tsk
->sighand
->siglock
);
1297 * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1298 * all the timers that are firing, and put them on the firing list.
1300 check_thread_timers(tsk
, &firing
);
1301 check_process_timers(tsk
, &firing
);
1304 * We must release these locks before taking any timer's lock.
1305 * There is a potential race with timer deletion here, as the
1306 * siglock now protects our private firing list. We have set
1307 * the firing flag in each timer, so that a deletion attempt
1308 * that gets the timer lock before we do will give it up and
1309 * spin until we've taken care of that timer below.
1311 spin_unlock(&tsk
->sighand
->siglock
);
1312 read_unlock(&tasklist_lock
);
1315 * Now that all the timers on our list have the firing flag,
1316 * noone will touch their list entries but us. We'll take
1317 * each timer's lock before clearing its firing flag, so no
1318 * timer call will interfere.
1320 list_for_each_entry_safe(timer
, next
, &firing
, it
.cpu
.entry
) {
1322 spin_lock(&timer
->it_lock
);
1323 list_del_init(&timer
->it
.cpu
.entry
);
1324 firing
= timer
->it
.cpu
.firing
;
1325 timer
->it
.cpu
.firing
= 0;
1327 * The firing flag is -1 if we collided with a reset
1328 * of the timer, which already reported this
1329 * almost-firing as an overrun. So don't generate an event.
1331 if (likely(firing
>= 0)) {
1332 cpu_timer_fire(timer
);
1334 spin_unlock(&timer
->it_lock
);
1339 * Set one of the process-wide special case CPU timers.
1340 * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1341 * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1342 * absolute; non-null for ITIMER_*, where *newval is relative and we update
1343 * it to be absolute, *oldval is absolute and we update it to be relative.
1345 void set_process_cpu_timer(struct task_struct
*tsk
, unsigned int clock_idx
,
1346 cputime_t
*newval
, cputime_t
*oldval
)
1348 union cpu_time_count now
;
1349 struct list_head
*head
;
1351 BUG_ON(clock_idx
== CPUCLOCK_SCHED
);
1352 cpu_clock_sample_group_locked(clock_idx
, tsk
, &now
);
1355 if (!cputime_eq(*oldval
, cputime_zero
)) {
1356 if (cputime_le(*oldval
, now
.cpu
)) {
1357 /* Just about to fire. */
1358 *oldval
= jiffies_to_cputime(1);
1360 *oldval
= cputime_sub(*oldval
, now
.cpu
);
1364 if (cputime_eq(*newval
, cputime_zero
))
1366 *newval
= cputime_add(*newval
, now
.cpu
);
1369 * If the RLIMIT_CPU timer will expire before the
1370 * ITIMER_PROF timer, we have nothing else to do.
1372 if (tsk
->signal
->rlim
[RLIMIT_CPU
].rlim_cur
1373 < cputime_to_secs(*newval
))
1378 * Check whether there are any process timers already set to fire
1379 * before this one. If so, we don't have anything more to do.
1381 head
= &tsk
->signal
->cpu_timers
[clock_idx
];
1382 if (list_empty(head
) ||
1383 cputime_ge(list_entry(head
->next
,
1384 struct cpu_timer_list
, entry
)->expires
.cpu
,
1387 * Rejigger each thread's expiry time so that one will
1388 * notice before we hit the process-cumulative expiry time.
1390 union cpu_time_count expires
= { .sched
= 0 };
1391 expires
.cpu
= *newval
;
1392 process_timer_rebalance(tsk
, clock_idx
, expires
, now
);
1396 static long posix_cpu_clock_nanosleep_restart(struct restart_block
*);
1398 int posix_cpu_nsleep(clockid_t which_clock
, int flags
,
1399 struct timespec
*rqtp
)
1401 struct restart_block
*restart_block
=
1402 ¤t_thread_info()->restart_block
;
1403 struct k_itimer timer
;
1407 * Diagnose required errors first.
1409 if (CPUCLOCK_PERTHREAD(which_clock
) &&
1410 (CPUCLOCK_PID(which_clock
) == 0 ||
1411 CPUCLOCK_PID(which_clock
) == current
->pid
))
1415 * Set up a temporary timer and then wait for it to go off.
1417 memset(&timer
, 0, sizeof timer
);
1418 spin_lock_init(&timer
.it_lock
);
1419 timer
.it_clock
= which_clock
;
1420 timer
.it_overrun
= -1;
1421 error
= posix_cpu_timer_create(&timer
);
1422 timer
.it_process
= current
;
1424 struct timespec __user
*rmtp
;
1425 static struct itimerspec zero_it
;
1426 struct itimerspec it
= { .it_value
= *rqtp
,
1427 .it_interval
= {} };
1429 spin_lock_irq(&timer
.it_lock
);
1430 error
= posix_cpu_timer_set(&timer
, flags
, &it
, NULL
);
1432 spin_unlock_irq(&timer
.it_lock
);
1436 while (!signal_pending(current
)) {
1437 if (timer
.it
.cpu
.expires
.sched
== 0) {
1439 * Our timer fired and was reset.
1441 spin_unlock_irq(&timer
.it_lock
);
1446 * Block until cpu_timer_fire (or a signal) wakes us.
1448 __set_current_state(TASK_INTERRUPTIBLE
);
1449 spin_unlock_irq(&timer
.it_lock
);
1451 spin_lock_irq(&timer
.it_lock
);
1455 * We were interrupted by a signal.
1457 sample_to_timespec(which_clock
, timer
.it
.cpu
.expires
, rqtp
);
1458 posix_cpu_timer_set(&timer
, 0, &zero_it
, &it
);
1459 spin_unlock_irq(&timer
.it_lock
);
1461 if ((it
.it_value
.tv_sec
| it
.it_value
.tv_nsec
) == 0) {
1463 * It actually did fire already.
1469 * Report back to the user the time still remaining.
1471 rmtp
= (struct timespec __user
*) restart_block
->arg1
;
1472 if (rmtp
!= NULL
&& !(flags
& TIMER_ABSTIME
) &&
1473 copy_to_user(rmtp
, &it
.it_value
, sizeof *rmtp
))
1476 restart_block
->fn
= posix_cpu_clock_nanosleep_restart
;
1477 /* Caller already set restart_block->arg1 */
1478 restart_block
->arg0
= which_clock
;
1479 restart_block
->arg2
= rqtp
->tv_sec
;
1480 restart_block
->arg3
= rqtp
->tv_nsec
;
1482 error
= -ERESTART_RESTARTBLOCK
;
1489 posix_cpu_clock_nanosleep_restart(struct restart_block
*restart_block
)
1491 clockid_t which_clock
= restart_block
->arg0
;
1492 struct timespec t
= { .tv_sec
= restart_block
->arg2
,
1493 .tv_nsec
= restart_block
->arg3
};
1494 restart_block
->fn
= do_no_restart_syscall
;
1495 return posix_cpu_nsleep(which_clock
, TIMER_ABSTIME
, &t
);
1499 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1500 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1502 static int process_cpu_clock_getres(clockid_t which_clock
, struct timespec
*tp
)
1504 return posix_cpu_clock_getres(PROCESS_CLOCK
, tp
);
1506 static int process_cpu_clock_get(clockid_t which_clock
, struct timespec
*tp
)
1508 return posix_cpu_clock_get(PROCESS_CLOCK
, tp
);
1510 static int process_cpu_timer_create(struct k_itimer
*timer
)
1512 timer
->it_clock
= PROCESS_CLOCK
;
1513 return posix_cpu_timer_create(timer
);
1515 static int process_cpu_nsleep(clockid_t which_clock
, int flags
,
1516 struct timespec
*rqtp
)
1518 return posix_cpu_nsleep(PROCESS_CLOCK
, flags
, rqtp
);
1520 static int thread_cpu_clock_getres(clockid_t which_clock
, struct timespec
*tp
)
1522 return posix_cpu_clock_getres(THREAD_CLOCK
, tp
);
1524 static int thread_cpu_clock_get(clockid_t which_clock
, struct timespec
*tp
)
1526 return posix_cpu_clock_get(THREAD_CLOCK
, tp
);
1528 static int thread_cpu_timer_create(struct k_itimer
*timer
)
1530 timer
->it_clock
= THREAD_CLOCK
;
1531 return posix_cpu_timer_create(timer
);
1533 static int thread_cpu_nsleep(clockid_t which_clock
, int flags
,
1534 struct timespec
*rqtp
)
1539 static __init
int init_posix_cpu_timers(void)
1541 struct k_clock process
= {
1542 .clock_getres
= process_cpu_clock_getres
,
1543 .clock_get
= process_cpu_clock_get
,
1544 .clock_set
= do_posix_clock_nosettime
,
1545 .timer_create
= process_cpu_timer_create
,
1546 .nsleep
= process_cpu_nsleep
,
1548 struct k_clock thread
= {
1549 .clock_getres
= thread_cpu_clock_getres
,
1550 .clock_get
= thread_cpu_clock_get
,
1551 .clock_set
= do_posix_clock_nosettime
,
1552 .timer_create
= thread_cpu_timer_create
,
1553 .nsleep
= thread_cpu_nsleep
,
1556 register_posix_clock(CLOCK_PROCESS_CPUTIME_ID
, &process
);
1557 register_posix_clock(CLOCK_THREAD_CPUTIME_ID
, &thread
);
1561 __initcall(init_posix_cpu_timers
);