2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/limits.h>
38 #include <sys/clock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/eventhandler.h>
43 #include <sys/resourcevar.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/sysctl.h>
48 #include <sys/sysent.h>
51 #include <sys/posix4.h>
53 #include <sys/timers.h>
54 #include <sys/timetc.h>
55 #include <sys/vnode.h>
58 #include <vm/vm_extern.h>
60 #define MAX_CLOCKS (CLOCK_MONOTONIC+1)
61 #define CPUCLOCK_BIT 0x80000000
62 #define CPUCLOCK_PROCESS_BIT 0x40000000
63 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
64 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid))
65 #define MAKE_PROCESS_CPUCLOCK(pid) \
66 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
68 static struct kclock posix_clocks[MAX_CLOCKS];
69 static uma_zone_t itimer_zone = NULL;
72 * Time of day and interval timer support.
74 * These routines provide the kernel entry points to get and set
75 * the time-of-day and per-process interval timers. Subroutines
76 * here provide support for adding and subtracting timeval structures
77 * and decrementing interval timers, optionally reloading the interval
78 * timers when they expire.
81 static int settime(struct thread *, struct timeval *);
82 static void timevalfix(struct timeval *);
84 static void itimer_start(void);
85 static int itimer_init(void *, int, int);
86 static void itimer_fini(void *, int);
87 static void itimer_enter(struct itimer *);
88 static void itimer_leave(struct itimer *);
89 static struct itimer *itimer_find(struct proc *, int);
90 static void itimers_alloc(struct proc *);
91 static void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
92 static void itimers_event_hook_exit(void *arg, struct proc *p);
93 static int realtimer_create(struct itimer *);
94 static int realtimer_gettime(struct itimer *, struct itimerspec *);
95 static int realtimer_settime(struct itimer *, int,
96 struct itimerspec *, struct itimerspec *);
97 static int realtimer_delete(struct itimer *);
98 static void realtimer_clocktime(clockid_t, struct timespec *);
99 static void realtimer_expire(void *);
100 static int kern_timer_create(struct thread *, clockid_t,
101 struct sigevent *, int *, int);
102 static int kern_timer_delete(struct thread *, int);
104 int register_posix_clock(int, struct kclock *);
105 void itimer_fire(struct itimer *it);
106 int itimespecfix(struct timespec *ts);
108 #define CLOCK_CALL(clock, call, arglist) \
109 ((*posix_clocks[clock].call) arglist)
111 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
115 settime(struct thread *td, struct timeval *tv)
117 struct timeval delta, tv1, tv2;
118 static struct timeval maxtime, laststep;
125 timevalsub(&delta, &tv1);
128 * If the system is secure, we do not allow the time to be
129 * set to a value earlier than 1 second less than the highest
130 * time we have yet seen. The worst a miscreant can do in
131 * this circumstance is "freeze" time. He couldn't go
134 * We similarly do not allow the clock to be stepped more
135 * than one second, nor more than once per second. This allows
136 * a miscreant to make the clock march double-time, but no worse.
138 if (securelevel_gt(td->td_ucred, 1) != 0) {
139 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
141 * Update maxtime to latest time we've seen.
143 if (tv1.tv_sec > maxtime.tv_sec)
146 timevalsub(&tv2, &maxtime);
147 if (tv2.tv_sec < -1) {
148 tv->tv_sec = maxtime.tv_sec - 1;
149 printf("Time adjustment clamped to -1 second\n");
152 if (tv1.tv_sec == laststep.tv_sec) {
156 if (delta.tv_sec > 1) {
157 tv->tv_sec = tv1.tv_sec + 1;
158 printf("Time adjustment clamped to +1 second\n");
164 ts.tv_sec = tv->tv_sec;
165 ts.tv_nsec = tv->tv_usec * 1000;
173 #ifndef _SYS_SYSPROTO_H_
174 struct clock_getcpuclockid2_args {
182 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
191 case CPUCLOCK_WHICH_PID:
196 error = p_cansee(td, p);
202 pid = td->td_proc->p_pid;
204 clk_id = MAKE_PROCESS_CPUCLOCK(pid);
206 case CPUCLOCK_WHICH_TID:
211 clk_id = MAKE_THREAD_CPUCLOCK(tid);
216 return (copyout(&clk_id, uap->clock_id, sizeof(clockid_t)));
219 #ifndef _SYS_SYSPROTO_H_
220 struct clock_gettime_args {
227 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
232 error = kern_clock_gettime(td, uap->clock_id, &ats);
234 error = copyout(&ats, uap->tp, sizeof(ats));
240 cputick2timespec(uint64_t runtime, struct timespec *ats)
242 runtime = cputick2usec(runtime);
243 ats->tv_sec = runtime / 1000000;
244 ats->tv_nsec = runtime % 1000000 * 1000;
248 get_thread_cputime(struct thread *targettd, struct timespec *ats)
250 uint64_t runtime, curtime, switchtime;
252 if (targettd == NULL) { /* current thread */
254 switchtime = PCPU_GET(switchtime);
255 curtime = cpu_ticks();
256 runtime = curthread->td_runtime;
258 runtime += curtime - switchtime;
260 thread_lock(targettd);
261 runtime = targettd->td_runtime;
262 thread_unlock(targettd);
264 cputick2timespec(runtime, ats);
268 get_process_cputime(struct proc *targetp, struct timespec *ats)
274 rufetch(targetp, &ru);
275 runtime = targetp->p_rux.rux_runtime;
276 PROC_SUNLOCK(targetp);
277 cputick2timespec(runtime, ats);
281 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
290 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
291 tid = clock_id & CPUCLOCK_ID_MASK;
292 td2 = tdfind(tid, p->p_pid);
295 get_thread_cputime(td2, ats);
296 PROC_UNLOCK(td2->td_proc);
298 pid = clock_id & CPUCLOCK_ID_MASK;
302 error = p_cansee(td, p2);
307 get_process_cputime(p2, ats);
314 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
316 struct timeval sys, user;
321 case CLOCK_REALTIME: /* Default to precise. */
322 case CLOCK_REALTIME_PRECISE:
325 case CLOCK_REALTIME_FAST:
331 calcru(p, &user, &sys);
334 TIMEVAL_TO_TIMESPEC(&user, ats);
339 calcru(p, &user, &sys);
342 timevaladd(&user, &sys);
343 TIMEVAL_TO_TIMESPEC(&user, ats);
345 case CLOCK_MONOTONIC: /* Default to precise. */
346 case CLOCK_MONOTONIC_PRECISE:
348 case CLOCK_UPTIME_PRECISE:
351 case CLOCK_UPTIME_FAST:
352 case CLOCK_MONOTONIC_FAST:
356 ats->tv_sec = time_second;
359 case CLOCK_THREAD_CPUTIME_ID:
360 get_thread_cputime(NULL, ats);
362 case CLOCK_PROCESS_CPUTIME_ID:
364 get_process_cputime(p, ats);
368 if ((int)clock_id >= 0)
370 return (get_cputime(td, clock_id, ats));
375 #ifndef _SYS_SYSPROTO_H_
376 struct clock_settime_args {
378 const struct timespec *tp;
383 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
388 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
390 return (kern_clock_settime(td, uap->clock_id, &ats));
394 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
399 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
401 if (clock_id != CLOCK_REALTIME)
403 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000)
405 /* XXX Don't convert nsec->usec and back */
406 TIMESPEC_TO_TIMEVAL(&atv, ats);
407 error = settime(td, &atv);
411 #ifndef _SYS_SYSPROTO_H_
412 struct clock_getres_args {
418 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
426 error = kern_clock_getres(td, uap->clock_id, &ts);
428 error = copyout(&ts, uap->tp, sizeof(ts));
433 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
439 case CLOCK_REALTIME_FAST:
440 case CLOCK_REALTIME_PRECISE:
441 case CLOCK_MONOTONIC:
442 case CLOCK_MONOTONIC_FAST:
443 case CLOCK_MONOTONIC_PRECISE:
445 case CLOCK_UPTIME_FAST:
446 case CLOCK_UPTIME_PRECISE:
448 * Round up the result of the division cheaply by adding 1.
449 * Rounding up is especially important if rounding down
450 * would give 0. Perfect rounding is unimportant.
452 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
456 /* Accurately round up here because we can do so cheaply. */
457 ts->tv_nsec = (1000000000 + hz - 1) / hz;
463 case CLOCK_THREAD_CPUTIME_ID:
464 case CLOCK_PROCESS_CPUTIME_ID:
466 /* sync with cputick2usec */
467 ts->tv_nsec = 1000000 / cpu_tickrate();
468 if (ts->tv_nsec == 0)
472 if ((int)clock_id < 0)
482 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
484 struct timespec ts, ts2, ts3;
488 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
490 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
493 timespecadd(&ts, rqt);
494 TIMESPEC_TO_TIMEVAL(&tv, rqt);
496 error = tsleep(&nanowait, PWAIT | PCATCH, "nanslp",
499 if (error != EWOULDBLOCK) {
500 if (error == ERESTART)
503 timespecsub(&ts, &ts2);
510 if (timespeccmp(&ts2, &ts, >=))
513 timespecsub(&ts3, &ts2);
514 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
518 #ifndef _SYS_SYSPROTO_H_
519 struct nanosleep_args {
520 struct timespec *rqtp;
521 struct timespec *rmtp;
526 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
528 struct timespec rmt, rqt;
531 error = copyin(uap->rqtp, &rqt, sizeof(rqt));
536 !useracc((caddr_t)uap->rmtp, sizeof(rmt), VM_PROT_WRITE))
538 error = kern_nanosleep(td, &rqt, &rmt);
539 if (error && uap->rmtp) {
542 error2 = copyout(&rmt, uap->rmtp, sizeof(rmt));
549 #ifndef _SYS_SYSPROTO_H_
550 struct gettimeofday_args {
552 struct timezone *tzp;
557 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
565 error = copyout(&atv, uap->tp, sizeof (atv));
567 if (error == 0 && uap->tzp != NULL) {
568 rtz.tz_minuteswest = tz_minuteswest;
569 rtz.tz_dsttime = tz_dsttime;
570 error = copyout(&rtz, uap->tzp, sizeof (rtz));
575 #ifndef _SYS_SYSPROTO_H_
576 struct settimeofday_args {
578 struct timezone *tzp;
583 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
585 struct timeval atv, *tvp;
586 struct timezone atz, *tzp;
590 error = copyin(uap->tv, &atv, sizeof(atv));
597 error = copyin(uap->tzp, &atz, sizeof(atz));
603 return (kern_settimeofday(td, tvp, tzp));
607 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
611 error = priv_check(td, PRIV_SETTIMEOFDAY);
614 /* Verify all parameters before changing time. */
616 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000)
618 error = settime(td, tv);
620 if (tzp && error == 0) {
621 tz_minuteswest = tzp->tz_minuteswest;
622 tz_dsttime = tzp->tz_dsttime;
628 * Get value of an interval timer. The process virtual and profiling virtual
629 * time timers are kept in the p_stats area, since they can be swapped out.
630 * These are kept internally in the way they are specified externally: in
631 * time until they expire.
633 * The real time interval timer is kept in the process table slot for the
634 * process, and its value (it_value) is kept as an absolute time rather than
635 * as a delta, so that it is easy to keep periodic real-time signals from
638 * Virtual time timers are processed in the hardclock() routine of
639 * kern_clock.c. The real time timer is processed by a timeout routine,
640 * called from the softclock() routine. Since a callout may be delayed in
641 * real time due to interrupt processing in the system, it is possible for
642 * the real time timeout routine (realitexpire, given below), to be delayed
643 * in real time past when it is supposed to occur. It does not suffice,
644 * therefore, to reload the real timer .it_value from the real time timers
645 * .it_interval. Rather, we compute the next time in absolute time the timer
648 #ifndef _SYS_SYSPROTO_H_
649 struct getitimer_args {
651 struct itimerval *itv;
655 sys_getitimer(struct thread *td, struct getitimer_args *uap)
657 struct itimerval aitv;
660 error = kern_getitimer(td, uap->which, &aitv);
663 return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
667 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
669 struct proc *p = td->td_proc;
672 if (which > ITIMER_PROF)
675 if (which == ITIMER_REAL) {
677 * Convert from absolute to relative time in .it_value
678 * part of real time timer. If time for real time timer
679 * has passed return 0, else return difference between
680 * current time and time for the timer to go off.
683 *aitv = p->p_realtimer;
685 if (timevalisset(&aitv->it_value)) {
686 getmicrouptime(&ctv);
687 if (timevalcmp(&aitv->it_value, &ctv, <))
688 timevalclear(&aitv->it_value);
690 timevalsub(&aitv->it_value, &ctv);
694 *aitv = p->p_stats->p_timer[which];
700 #ifndef _SYS_SYSPROTO_H_
701 struct setitimer_args {
703 struct itimerval *itv, *oitv;
707 sys_setitimer(struct thread *td, struct setitimer_args *uap)
709 struct itimerval aitv, oitv;
712 if (uap->itv == NULL) {
713 uap->itv = uap->oitv;
714 return (sys_getitimer(td, (struct getitimer_args *)uap));
717 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
719 error = kern_setitimer(td, uap->which, &aitv, &oitv);
720 if (error != 0 || uap->oitv == NULL)
722 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
726 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
727 struct itimerval *oitv)
729 struct proc *p = td->td_proc;
733 return (kern_getitimer(td, which, oitv));
735 if (which > ITIMER_PROF)
737 if (itimerfix(&aitv->it_value))
739 if (!timevalisset(&aitv->it_value))
740 timevalclear(&aitv->it_interval);
741 else if (itimerfix(&aitv->it_interval))
744 if (which == ITIMER_REAL) {
746 if (timevalisset(&p->p_realtimer.it_value))
747 callout_stop(&p->p_itcallout);
748 getmicrouptime(&ctv);
749 if (timevalisset(&aitv->it_value)) {
750 callout_reset(&p->p_itcallout, tvtohz(&aitv->it_value),
752 timevaladd(&aitv->it_value, &ctv);
754 *oitv = p->p_realtimer;
755 p->p_realtimer = *aitv;
757 if (timevalisset(&oitv->it_value)) {
758 if (timevalcmp(&oitv->it_value, &ctv, <))
759 timevalclear(&oitv->it_value);
761 timevalsub(&oitv->it_value, &ctv);
765 *oitv = p->p_stats->p_timer[which];
766 p->p_stats->p_timer[which] = *aitv;
773 * Real interval timer expired:
774 * send process whose timer expired an alarm signal.
775 * If time is not set up to reload, then just return.
776 * Else compute next time timer should go off which is > current time.
777 * This is where delay in processing this timeout causes multiple
778 * SIGALRM calls to be compressed into one.
779 * tvtohz() always adds 1 to allow for the time until the next clock
780 * interrupt being strictly less than 1 clock tick, but we don't want
781 * that here since we want to appear to be in sync with the clock
782 * interrupt even when we're delayed.
785 realitexpire(void *arg)
788 struct timeval ctv, ntv;
790 p = (struct proc *)arg;
792 kern_psignal(p, SIGALRM);
793 if (!timevalisset(&p->p_realtimer.it_interval)) {
794 timevalclear(&p->p_realtimer.it_value);
795 if (p->p_flag & P_WEXIT)
796 wakeup(&p->p_itcallout);
801 timevaladd(&p->p_realtimer.it_value,
802 &p->p_realtimer.it_interval);
803 getmicrouptime(&ctv);
804 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
805 ntv = p->p_realtimer.it_value;
806 timevalsub(&ntv, &ctv);
807 callout_reset(&p->p_itcallout, tvtohz(&ntv) - 1,
817 * Check that a proposed value to load into the .it_value or
818 * .it_interval part of an interval timer is acceptable, and
819 * fix it to have at least minimal value (i.e. if it is less
820 * than the resolution of the clock, round it up.)
823 itimerfix(struct timeval *tv)
826 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
828 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
834 * Decrement an interval timer by a specified number
835 * of microseconds, which must be less than a second,
836 * i.e. < 1000000. If the timer expires, then reload
837 * it. In this case, carry over (usec - old value) to
838 * reduce the value reloaded into the timer so that
839 * the timer does not drift. This routine assumes
840 * that it is called in a context where the timers
841 * on which it is operating cannot change in value.
844 itimerdecr(struct itimerval *itp, int usec)
847 if (itp->it_value.tv_usec < usec) {
848 if (itp->it_value.tv_sec == 0) {
849 /* expired, and already in next interval */
850 usec -= itp->it_value.tv_usec;
853 itp->it_value.tv_usec += 1000000;
854 itp->it_value.tv_sec--;
856 itp->it_value.tv_usec -= usec;
858 if (timevalisset(&itp->it_value))
860 /* expired, exactly at end of interval */
862 if (timevalisset(&itp->it_interval)) {
863 itp->it_value = itp->it_interval;
864 itp->it_value.tv_usec -= usec;
865 if (itp->it_value.tv_usec < 0) {
866 itp->it_value.tv_usec += 1000000;
867 itp->it_value.tv_sec--;
870 itp->it_value.tv_usec = 0; /* sec is already 0 */
875 * Add and subtract routines for timevals.
876 * N.B.: subtract routine doesn't deal with
877 * results which are before the beginning,
878 * it just gets very confused in this case.
882 timevaladd(struct timeval *t1, const struct timeval *t2)
885 t1->tv_sec += t2->tv_sec;
886 t1->tv_usec += t2->tv_usec;
891 timevalsub(struct timeval *t1, const struct timeval *t2)
894 t1->tv_sec -= t2->tv_sec;
895 t1->tv_usec -= t2->tv_usec;
900 timevalfix(struct timeval *t1)
903 if (t1->tv_usec < 0) {
905 t1->tv_usec += 1000000;
907 if (t1->tv_usec >= 1000000) {
909 t1->tv_usec -= 1000000;
914 * ratecheck(): simple time-based rate-limit checking.
917 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
919 struct timeval tv, delta;
922 getmicrouptime(&tv); /* NB: 10ms precision */
924 timevalsub(&delta, lasttime);
927 * check for 0,0 is so that the message will be seen at least once,
928 * even if interval is huge.
930 if (timevalcmp(&delta, mininterval, >=) ||
931 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
940 * ppsratecheck(): packets (or events) per second limitation.
942 * Return 0 if the limit is to be enforced (e.g. the caller
943 * should drop a packet because of the rate limitation).
945 * maxpps of 0 always causes zero to be returned. maxpps of -1
946 * always causes 1 to be returned; this effectively defeats rate
949 * Note that we maintain the struct timeval for compatibility
950 * with other bsd systems. We reuse the storage and just monitor
951 * clock ticks for minimal overhead.
954 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
959 * Reset the last time and counter if this is the first call
960 * or more than a second has passed since the last update of
964 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
965 lasttime->tv_sec = now;
967 return (maxpps != 0);
969 (*curpps)++; /* NB: ignore potential overflow */
970 return (maxpps < 0 || *curpps < maxpps);
977 struct kclock rt_clock = {
978 .timer_create = realtimer_create,
979 .timer_delete = realtimer_delete,
980 .timer_settime = realtimer_settime,
981 .timer_gettime = realtimer_gettime,
985 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
986 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
987 register_posix_clock(CLOCK_REALTIME, &rt_clock);
988 register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
989 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
990 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
991 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
992 EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
993 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
994 EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
995 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
999 register_posix_clock(int clockid, struct kclock *clk)
1001 if ((unsigned)clockid >= MAX_CLOCKS) {
1002 printf("%s: invalid clockid\n", __func__);
1005 posix_clocks[clockid] = *clk;
1010 itimer_init(void *mem, int size, int flags)
1014 it = (struct itimer *)mem;
1015 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1020 itimer_fini(void *mem, int size)
1024 it = (struct itimer *)mem;
1025 mtx_destroy(&it->it_mtx);
1029 itimer_enter(struct itimer *it)
1032 mtx_assert(&it->it_mtx, MA_OWNED);
1037 itimer_leave(struct itimer *it)
1040 mtx_assert(&it->it_mtx, MA_OWNED);
1041 KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1043 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1047 #ifndef _SYS_SYSPROTO_H_
1048 struct ktimer_create_args {
1050 struct sigevent * evp;
1055 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1057 struct sigevent *evp1, ev;
1061 if (uap->evp != NULL) {
1062 error = copyin(uap->evp, &ev, sizeof(ev));
1069 error = kern_timer_create(td, uap->clock_id, evp1, &id, -1);
1072 error = copyout(&id, uap->timerid, sizeof(int));
1074 kern_timer_delete(td, id);
1080 kern_timer_create(struct thread *td, clockid_t clock_id,
1081 struct sigevent *evp, int *timerid, int preset_id)
1083 struct proc *p = td->td_proc;
1088 if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1091 if (posix_clocks[clock_id].timer_create == NULL)
1095 if (evp->sigev_notify != SIGEV_NONE &&
1096 evp->sigev_notify != SIGEV_SIGNAL &&
1097 evp->sigev_notify != SIGEV_THREAD_ID)
1099 if ((evp->sigev_notify == SIGEV_SIGNAL ||
1100 evp->sigev_notify == SIGEV_THREAD_ID) &&
1101 !_SIG_VALID(evp->sigev_signo))
1105 if (p->p_itimers == NULL)
1108 it = uma_zalloc(itimer_zone, M_WAITOK);
1110 it->it_usecount = 0;
1112 timespecclear(&it->it_time.it_value);
1113 timespecclear(&it->it_time.it_interval);
1115 it->it_overrun_last = 0;
1116 it->it_clockid = clock_id;
1117 it->it_timerid = -1;
1119 ksiginfo_init(&it->it_ksi);
1120 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1121 error = CLOCK_CALL(clock_id, timer_create, (it));
1126 if (preset_id != -1) {
1127 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1129 if (p->p_itimers->its_timers[id] != NULL) {
1136 * Find a free timer slot, skipping those reserved
1139 for (id = 3; id < TIMER_MAX; id++)
1140 if (p->p_itimers->its_timers[id] == NULL)
1142 if (id == TIMER_MAX) {
1148 it->it_timerid = id;
1149 p->p_itimers->its_timers[id] = it;
1151 it->it_sigev = *evp;
1153 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1156 case CLOCK_REALTIME:
1157 it->it_sigev.sigev_signo = SIGALRM;
1160 it->it_sigev.sigev_signo = SIGVTALRM;
1163 it->it_sigev.sigev_signo = SIGPROF;
1166 it->it_sigev.sigev_value.sival_int = id;
1169 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1170 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1171 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1172 it->it_ksi.ksi_code = SI_TIMER;
1173 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1174 it->it_ksi.ksi_timerid = id;
1182 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1184 uma_zfree(itimer_zone, it);
1188 #ifndef _SYS_SYSPROTO_H_
1189 struct ktimer_delete_args {
1194 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1196 return (kern_timer_delete(td, uap->timerid));
1199 static struct itimer *
1200 itimer_find(struct proc *p, int timerid)
1204 PROC_LOCK_ASSERT(p, MA_OWNED);
1205 if ((p->p_itimers == NULL) ||
1206 (timerid < 0) || (timerid >= TIMER_MAX) ||
1207 (it = p->p_itimers->its_timers[timerid]) == NULL) {
1211 if ((it->it_flags & ITF_DELETING) != 0) {
1219 kern_timer_delete(struct thread *td, int timerid)
1221 struct proc *p = td->td_proc;
1225 it = itimer_find(p, timerid);
1232 it->it_flags |= ITF_DELETING;
1233 while (it->it_usecount > 0) {
1234 it->it_flags |= ITF_WANTED;
1235 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1237 it->it_flags &= ~ITF_WANTED;
1238 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1242 if (KSI_ONQ(&it->it_ksi))
1243 sigqueue_take(&it->it_ksi);
1244 p->p_itimers->its_timers[timerid] = NULL;
1246 uma_zfree(itimer_zone, it);
1250 #ifndef _SYS_SYSPROTO_H_
1251 struct ktimer_settime_args {
1254 const struct itimerspec * value;
1255 struct itimerspec * ovalue;
1259 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1261 struct proc *p = td->td_proc;
1263 struct itimerspec val, oval, *ovalp;
1266 error = copyin(uap->value, &val, sizeof(val));
1270 if (uap->ovalue != NULL)
1276 if (uap->timerid < 3 ||
1277 (it = itimer_find(p, uap->timerid)) == NULL) {
1283 error = CLOCK_CALL(it->it_clockid, timer_settime,
1284 (it, uap->flags, &val, ovalp));
1288 if (error == 0 && uap->ovalue != NULL)
1289 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1293 #ifndef _SYS_SYSPROTO_H_
1294 struct ktimer_gettime_args {
1296 struct itimerspec * value;
1300 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1302 struct proc *p = td->td_proc;
1304 struct itimerspec val;
1308 if (uap->timerid < 3 ||
1309 (it = itimer_find(p, uap->timerid)) == NULL) {
1315 error = CLOCK_CALL(it->it_clockid, timer_gettime,
1321 error = copyout(&val, uap->value, sizeof(val));
1325 #ifndef _SYS_SYSPROTO_H_
1326 struct timer_getoverrun_args {
1331 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1333 struct proc *p = td->td_proc;
1338 if (uap->timerid < 3 ||
1339 (it = itimer_find(p, uap->timerid)) == NULL) {
1343 td->td_retval[0] = it->it_overrun_last;
1352 realtimer_create(struct itimer *it)
1354 callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1359 realtimer_delete(struct itimer *it)
1361 mtx_assert(&it->it_mtx, MA_OWNED);
1364 * clear timer's value and interval to tell realtimer_expire
1365 * to not rearm the timer.
1367 timespecclear(&it->it_time.it_value);
1368 timespecclear(&it->it_time.it_interval);
1370 callout_drain(&it->it_callout);
1376 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1378 struct timespec cts;
1380 mtx_assert(&it->it_mtx, MA_OWNED);
1382 realtimer_clocktime(it->it_clockid, &cts);
1383 *ovalue = it->it_time;
1384 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1385 timespecsub(&ovalue->it_value, &cts);
1386 if (ovalue->it_value.tv_sec < 0 ||
1387 (ovalue->it_value.tv_sec == 0 &&
1388 ovalue->it_value.tv_nsec == 0)) {
1389 ovalue->it_value.tv_sec = 0;
1390 ovalue->it_value.tv_nsec = 1;
1397 realtimer_settime(struct itimer *it, int flags,
1398 struct itimerspec *value, struct itimerspec *ovalue)
1400 struct timespec cts, ts;
1402 struct itimerspec val;
1404 mtx_assert(&it->it_mtx, MA_OWNED);
1407 if (itimespecfix(&val.it_value))
1410 if (timespecisset(&val.it_value)) {
1411 if (itimespecfix(&val.it_interval))
1414 timespecclear(&val.it_interval);
1418 realtimer_gettime(it, ovalue);
1421 if (timespecisset(&val.it_value)) {
1422 realtimer_clocktime(it->it_clockid, &cts);
1424 if ((flags & TIMER_ABSTIME) == 0) {
1425 /* Convert to absolute time. */
1426 timespecadd(&it->it_time.it_value, &cts);
1428 timespecsub(&ts, &cts);
1430 * We don't care if ts is negative, tztohz will
1434 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1435 callout_reset(&it->it_callout, tvtohz(&tv),
1436 realtimer_expire, it);
1438 callout_stop(&it->it_callout);
1445 realtimer_clocktime(clockid_t id, struct timespec *ts)
1447 if (id == CLOCK_REALTIME)
1449 else /* CLOCK_MONOTONIC */
1454 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1458 PROC_LOCK_ASSERT(p, MA_OWNED);
1459 it = itimer_find(p, timerid);
1461 ksi->ksi_overrun = it->it_overrun;
1462 it->it_overrun_last = it->it_overrun;
1471 itimespecfix(struct timespec *ts)
1474 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1476 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1477 ts->tv_nsec = tick * 1000;
1481 /* Timeout callback for realtime timer */
1483 realtimer_expire(void *arg)
1485 struct timespec cts, ts;
1489 it = (struct itimer *)arg;
1491 realtimer_clocktime(it->it_clockid, &cts);
1492 /* Only fire if time is reached. */
1493 if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1494 if (timespecisset(&it->it_time.it_interval)) {
1495 timespecadd(&it->it_time.it_value,
1496 &it->it_time.it_interval);
1497 while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1498 if (it->it_overrun < INT_MAX)
1501 it->it_ksi.ksi_errno = ERANGE;
1502 timespecadd(&it->it_time.it_value,
1503 &it->it_time.it_interval);
1506 /* single shot timer ? */
1507 timespecclear(&it->it_time.it_value);
1509 if (timespecisset(&it->it_time.it_value)) {
1510 ts = it->it_time.it_value;
1511 timespecsub(&ts, &cts);
1512 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1513 callout_reset(&it->it_callout, tvtohz(&tv),
1514 realtimer_expire, it);
1521 } else if (timespecisset(&it->it_time.it_value)) {
1522 ts = it->it_time.it_value;
1523 timespecsub(&ts, &cts);
1524 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1525 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1531 itimer_fire(struct itimer *it)
1533 struct proc *p = it->it_proc;
1536 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1537 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1538 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1540 timespecclear(&it->it_time.it_value);
1541 timespecclear(&it->it_time.it_interval);
1542 callout_stop(&it->it_callout);
1546 if (!KSI_ONQ(&it->it_ksi)) {
1547 it->it_ksi.ksi_errno = 0;
1548 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1549 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1551 if (it->it_overrun < INT_MAX)
1554 it->it_ksi.ksi_errno = ERANGE;
1561 itimers_alloc(struct proc *p)
1563 struct itimers *its;
1566 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1567 LIST_INIT(&its->its_virtual);
1568 LIST_INIT(&its->its_prof);
1569 TAILQ_INIT(&its->its_worklist);
1570 for (i = 0; i < TIMER_MAX; i++)
1571 its->its_timers[i] = NULL;
1573 if (p->p_itimers == NULL) {
1579 free(its, M_SUBPROC);
1584 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1586 itimers_event_hook_exit(arg, p);
1589 /* Clean up timers when some process events are being triggered. */
1591 itimers_event_hook_exit(void *arg, struct proc *p)
1593 struct itimers *its;
1595 int event = (int)(intptr_t)arg;
1598 if (p->p_itimers != NULL) {
1600 for (i = 0; i < MAX_CLOCKS; ++i) {
1601 if (posix_clocks[i].event_hook != NULL)
1602 CLOCK_CALL(i, event_hook, (p, i, event));
1605 * According to susv3, XSI interval timers should be inherited
1608 if (event == ITIMER_EV_EXEC)
1610 else if (event == ITIMER_EV_EXIT)
1613 panic("unhandled event");
1614 for (; i < TIMER_MAX; ++i) {
1615 if ((it = its->its_timers[i]) != NULL)
1616 kern_timer_delete(curthread, i);
1618 if (its->its_timers[0] == NULL &&
1619 its->its_timers[1] == NULL &&
1620 its->its_timers[2] == NULL) {
1621 free(its, M_SUBPROC);
1622 p->p_itimers = NULL;