2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93
34 #include <sys/cdefs.h>
35 #include "opt_ktrace.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/limits.h>
40 #include <sys/clock.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/resourcevar.h>
45 #include <sys/signalvar.h>
46 #include <sys/kernel.h>
47 #include <sys/sleepqueue.h>
48 #include <sys/syscallsubr.h>
49 #include <sys/sysctl.h>
52 #include <sys/posix4.h>
54 #include <sys/timers.h>
55 #include <sys/timetc.h>
56 #include <sys/vnode.h>
58 #include <sys/ktrace.h>
62 #include <vm/vm_extern.h>
64 #define MAX_CLOCKS (CLOCK_MONOTONIC+1)
65 #define CPUCLOCK_BIT 0x80000000
66 #define CPUCLOCK_PROCESS_BIT 0x40000000
67 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
68 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid))
69 #define MAKE_PROCESS_CPUCLOCK(pid) \
70 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
72 #define NS_PER_SEC 1000000000
74 static struct kclock posix_clocks[MAX_CLOCKS];
75 static uma_zone_t itimer_zone = NULL;
78 * Time of day and interval timer support.
80 * These routines provide the kernel entry points to get and set
81 * the time-of-day and per-process interval timers. Subroutines
82 * here provide support for adding and subtracting timeval structures
83 * and decrementing interval timers, optionally reloading the interval
84 * timers when they expire.
87 static int settime(struct thread *, struct timeval *);
88 static void timevalfix(struct timeval *);
89 static int user_clock_nanosleep(struct thread *td, clockid_t clock_id,
90 int flags, const struct timespec *ua_rqtp,
91 struct timespec *ua_rmtp);
93 static void itimer_start(void);
94 static int itimer_init(void *, int, int);
95 static void itimer_fini(void *, int);
96 static void itimer_enter(struct itimer *);
97 static void itimer_leave(struct itimer *);
98 static struct itimer *itimer_find(struct proc *, int);
99 static void itimers_alloc(struct proc *);
100 static int realtimer_create(struct itimer *);
101 static int realtimer_gettime(struct itimer *, struct itimerspec *);
102 static int realtimer_settime(struct itimer *, int,
103 struct itimerspec *, struct itimerspec *);
104 static int realtimer_delete(struct itimer *);
105 static void realtimer_clocktime(clockid_t, struct timespec *);
106 static void realtimer_expire(void *);
107 static void realtimer_expire_l(struct itimer *it, bool proc_locked);
109 static void realitexpire(void *arg);
111 static int register_posix_clock(int, const struct kclock *);
112 static void itimer_fire(struct itimer *it);
113 static int itimespecfix(struct timespec *ts);
115 #define CLOCK_CALL(clock, call, arglist) \
116 ((*posix_clocks[clock].call) arglist)
118 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
121 settime(struct thread *td, struct timeval *tv)
123 struct timeval delta, tv1, tv2;
124 static struct timeval maxtime, laststep;
129 timevalsub(&delta, &tv1);
132 * If the system is secure, we do not allow the time to be
133 * set to a value earlier than 1 second less than the highest
134 * time we have yet seen. The worst a miscreant can do in
135 * this circumstance is "freeze" time. He couldn't go
138 * We similarly do not allow the clock to be stepped more
139 * than one second, nor more than once per second. This allows
140 * a miscreant to make the clock march double-time, but no worse.
142 if (securelevel_gt(td->td_ucred, 1) != 0) {
143 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
145 * Update maxtime to latest time we've seen.
147 if (tv1.tv_sec > maxtime.tv_sec)
150 timevalsub(&tv2, &maxtime);
151 if (tv2.tv_sec < -1) {
152 tv->tv_sec = maxtime.tv_sec - 1;
153 printf("Time adjustment clamped to -1 second\n");
156 if (tv1.tv_sec == laststep.tv_sec)
158 if (delta.tv_sec > 1) {
159 tv->tv_sec = tv1.tv_sec + 1;
160 printf("Time adjustment clamped to +1 second\n");
166 ts.tv_sec = tv->tv_sec;
167 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)
187 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
189 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
194 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
203 case CPUCLOCK_WHICH_PID:
205 error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
211 pid = td->td_proc->p_pid;
213 *clk_id = MAKE_PROCESS_CPUCLOCK(pid);
215 case CPUCLOCK_WHICH_TID:
216 tid = id == 0 ? td->td_tid : id;
217 *clk_id = MAKE_THREAD_CPUCLOCK(tid);
224 #ifndef _SYS_SYSPROTO_H_
225 struct clock_gettime_args {
232 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
237 error = kern_clock_gettime(td, uap->clock_id, &ats);
239 error = copyout(&ats, uap->tp, sizeof(ats));
245 cputick2timespec(uint64_t runtime, struct timespec *ats)
249 ats->tv_sec = runtime / tr;
250 ats->tv_nsec = ((runtime % tr) * 1000000000ULL) / tr;
254 kern_thread_cputime(struct thread *targettd, struct timespec *ats)
256 uint64_t runtime, curtime, switchtime;
258 if (targettd == NULL) { /* current thread */
260 switchtime = PCPU_GET(switchtime);
261 curtime = cpu_ticks();
262 runtime = curthread->td_runtime;
264 runtime += curtime - switchtime;
266 PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED);
267 thread_lock(targettd);
268 runtime = targettd->td_runtime;
269 thread_unlock(targettd);
271 cputick2timespec(runtime, ats);
275 kern_process_cputime(struct proc *targetp, struct timespec *ats)
280 PROC_LOCK_ASSERT(targetp, MA_OWNED);
281 PROC_STATLOCK(targetp);
282 rufetch(targetp, &ru);
283 runtime = targetp->p_rux.rux_runtime;
284 if (curthread->td_proc == targetp)
285 runtime += cpu_ticks() - PCPU_GET(switchtime);
286 PROC_STATUNLOCK(targetp);
287 cputick2timespec(runtime, ats);
291 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
300 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
301 tid = clock_id & CPUCLOCK_ID_MASK;
302 td2 = tdfind(tid, p->p_pid);
305 kern_thread_cputime(td2, ats);
306 PROC_UNLOCK(td2->td_proc);
308 pid = clock_id & CPUCLOCK_ID_MASK;
309 error = pget(pid, PGET_CANSEE, &p2);
312 kern_process_cputime(p2, ats);
319 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
321 struct timeval sys, user;
326 case CLOCK_REALTIME: /* Default to precise. */
327 case CLOCK_REALTIME_PRECISE:
330 case CLOCK_REALTIME_FAST:
336 calcru(p, &user, &sys);
339 TIMEVAL_TO_TIMESPEC(&user, ats);
344 calcru(p, &user, &sys);
347 timevaladd(&user, &sys);
348 TIMEVAL_TO_TIMESPEC(&user, ats);
350 case CLOCK_MONOTONIC: /* Default to precise. */
351 case CLOCK_MONOTONIC_PRECISE:
353 case CLOCK_UPTIME_PRECISE:
356 case CLOCK_UPTIME_FAST:
357 case CLOCK_MONOTONIC_FAST:
361 ats->tv_sec = time_second;
364 case CLOCK_THREAD_CPUTIME_ID:
365 kern_thread_cputime(NULL, ats);
367 case CLOCK_PROCESS_CPUTIME_ID:
369 kern_process_cputime(p, ats);
373 if ((int)clock_id >= 0)
375 return (get_cputime(td, clock_id, ats));
380 #ifndef _SYS_SYSPROTO_H_
381 struct clock_settime_args {
383 const struct timespec *tp;
388 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
393 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
395 return (kern_clock_settime(td, uap->clock_id, &ats));
398 static int allow_insane_settime = 0;
399 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
400 &allow_insane_settime, 0,
401 "do not perform possibly restrictive checks on settime(2) args");
404 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
409 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
411 if (clock_id != CLOCK_REALTIME)
413 if (!timespecvalid_interval(ats))
415 if (!allow_insane_settime &&
416 (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 ||
417 ats->tv_sec < utc_offset()))
419 /* XXX Don't convert nsec->usec and back */
420 TIMESPEC_TO_TIMEVAL(&atv, ats);
421 error = settime(td, &atv);
425 #ifndef _SYS_SYSPROTO_H_
426 struct clock_getres_args {
432 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
440 error = kern_clock_getres(td, uap->clock_id, &ts);
442 error = copyout(&ts, uap->tp, sizeof(ts));
447 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
453 case CLOCK_REALTIME_FAST:
454 case CLOCK_REALTIME_PRECISE:
455 case CLOCK_MONOTONIC:
456 case CLOCK_MONOTONIC_FAST:
457 case CLOCK_MONOTONIC_PRECISE:
459 case CLOCK_UPTIME_FAST:
460 case CLOCK_UPTIME_PRECISE:
462 * Round up the result of the division cheaply by adding 1.
463 * Rounding up is especially important if rounding down
464 * would give 0. Perfect rounding is unimportant.
466 ts->tv_nsec = NS_PER_SEC / tc_getfrequency() + 1;
470 /* Accurately round up here because we can do so cheaply. */
471 ts->tv_nsec = howmany(NS_PER_SEC, hz);
477 case CLOCK_THREAD_CPUTIME_ID:
478 case CLOCK_PROCESS_CPUTIME_ID:
480 ts->tv_nsec = 1000000000 / cpu_tickrate() + 1;
483 if ((int)clock_id < 0)
491 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
494 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
498 static uint8_t nanowait[MAXCPU];
501 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
502 const struct timespec *rqt, struct timespec *rmt)
504 struct timespec ts, now;
505 sbintime_t sbt, sbtt, prec, tmp;
510 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= NS_PER_SEC)
512 if ((flags & ~TIMER_ABSTIME) != 0)
516 case CLOCK_REALTIME_PRECISE:
517 case CLOCK_REALTIME_FAST:
519 is_abs_real = (flags & TIMER_ABSTIME) != 0;
521 case CLOCK_MONOTONIC:
522 case CLOCK_MONOTONIC_PRECISE:
523 case CLOCK_MONOTONIC_FAST:
525 case CLOCK_UPTIME_PRECISE:
526 case CLOCK_UPTIME_FAST:
531 case CLOCK_PROCESS_CPUTIME_ID:
533 case CLOCK_THREAD_CPUTIME_ID:
539 if ((flags & TIMER_ABSTIME) != 0) {
542 atomic_load_acq_int(&rtc_generation);
543 error = kern_clock_gettime(td, clock_id, &now);
544 KASSERT(error == 0, ("kern_clock_gettime: %d", error));
545 timespecsub(&ts, &now, &ts);
547 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
551 if (ts.tv_sec > INT32_MAX / 2) {
552 over = ts.tv_sec - INT32_MAX / 2;
559 if (TIMESEL(&sbt, tmp))
562 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
563 sbt, prec, C_ABSOLUTE);
564 } while (error == 0 && is_abs_real && td->td_rtcgen == 0);
566 if (error != EWOULDBLOCK) {
567 if (TIMESEL(&sbtt, tmp))
571 if (error == ERESTART)
573 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
574 ts = sbttots(sbt - sbtt);
585 #ifndef _SYS_SYSPROTO_H_
586 struct nanosleep_args {
587 struct timespec *rqtp;
588 struct timespec *rmtp;
593 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
596 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
597 uap->rqtp, uap->rmtp));
600 #ifndef _SYS_SYSPROTO_H_
601 struct clock_nanosleep_args {
604 struct timespec *rqtp;
605 struct timespec *rmtp;
610 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
614 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
616 return (kern_posix_error(td, error));
620 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
621 const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
623 struct timespec rmt, rqt;
626 error = copyin(ua_rqtp, &rqt, sizeof(rqt));
629 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
630 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
631 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
638 #ifndef _SYS_SYSPROTO_H_
639 struct gettimeofday_args {
641 struct timezone *tzp;
646 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
654 error = copyout(&atv, uap->tp, sizeof (atv));
656 if (error == 0 && uap->tzp != NULL) {
657 rtz.tz_minuteswest = 0;
659 error = copyout(&rtz, uap->tzp, sizeof (rtz));
664 #ifndef _SYS_SYSPROTO_H_
665 struct settimeofday_args {
667 struct timezone *tzp;
672 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
674 struct timeval atv, *tvp;
675 struct timezone atz, *tzp;
679 error = copyin(uap->tv, &atv, sizeof(atv));
686 error = copyin(uap->tzp, &atz, sizeof(atz));
692 return (kern_settimeofday(td, tvp, tzp));
696 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
700 error = priv_check(td, PRIV_SETTIMEOFDAY);
703 /* Verify all parameters before changing time. */
705 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
708 error = settime(td, tv);
714 * Get value of an interval timer. The process virtual and profiling virtual
715 * time timers are kept in the p_stats area, since they can be swapped out.
716 * These are kept internally in the way they are specified externally: in
717 * time until they expire.
719 * The real time interval timer is kept in the process table slot for the
720 * process, and its value (it_value) is kept as an absolute time rather than
721 * as a delta, so that it is easy to keep periodic real-time signals from
724 * Virtual time timers are processed in the hardclock() routine of
725 * kern_clock.c. The real time timer is processed by a timeout routine,
726 * called from the softclock() routine. Since a callout may be delayed in
727 * real time due to interrupt processing in the system, it is possible for
728 * the real time timeout routine (realitexpire, given below), to be delayed
729 * in real time past when it is supposed to occur. It does not suffice,
730 * therefore, to reload the real timer .it_value from the real time timers
731 * .it_interval. Rather, we compute the next time in absolute time the timer
734 #ifndef _SYS_SYSPROTO_H_
735 struct getitimer_args {
737 struct itimerval *itv;
741 sys_getitimer(struct thread *td, struct getitimer_args *uap)
743 struct itimerval aitv;
746 error = kern_getitimer(td, uap->which, &aitv);
749 return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
753 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
755 struct proc *p = td->td_proc;
758 if (which > ITIMER_PROF)
761 if (which == ITIMER_REAL) {
763 * Convert from absolute to relative time in .it_value
764 * part of real time timer. If time for real time timer
765 * has passed return 0, else return difference between
766 * current time and time for the timer to go off.
769 *aitv = p->p_realtimer;
771 if (timevalisset(&aitv->it_value)) {
773 if (timevalcmp(&aitv->it_value, &ctv, <))
774 timevalclear(&aitv->it_value);
776 timevalsub(&aitv->it_value, &ctv);
780 *aitv = p->p_stats->p_timer[which];
784 if (KTRPOINT(td, KTR_STRUCT))
790 #ifndef _SYS_SYSPROTO_H_
791 struct setitimer_args {
793 struct itimerval *itv, *oitv;
797 sys_setitimer(struct thread *td, struct setitimer_args *uap)
799 struct itimerval aitv, oitv;
802 if (uap->itv == NULL) {
803 uap->itv = uap->oitv;
804 return (sys_getitimer(td, (struct getitimer_args *)uap));
807 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
809 error = kern_setitimer(td, uap->which, &aitv, &oitv);
810 if (error != 0 || uap->oitv == NULL)
812 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
816 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
817 struct itimerval *oitv)
819 struct proc *p = td->td_proc;
824 return (kern_getitimer(td, which, oitv));
826 if (which > ITIMER_PROF)
829 if (KTRPOINT(td, KTR_STRUCT))
832 if (itimerfix(&aitv->it_value) ||
833 aitv->it_value.tv_sec > INT32_MAX / 2)
835 if (!timevalisset(&aitv->it_value))
836 timevalclear(&aitv->it_interval);
837 else if (itimerfix(&aitv->it_interval) ||
838 aitv->it_interval.tv_sec > INT32_MAX / 2)
841 if (which == ITIMER_REAL) {
843 if (timevalisset(&p->p_realtimer.it_value))
844 callout_stop(&p->p_itcallout);
846 if (timevalisset(&aitv->it_value)) {
847 pr = tvtosbt(aitv->it_value) >> tc_precexp;
848 timevaladd(&aitv->it_value, &ctv);
849 sbt = tvtosbt(aitv->it_value);
850 callout_reset_sbt(&p->p_itcallout, sbt, pr,
851 realitexpire, p, C_ABSOLUTE);
853 *oitv = p->p_realtimer;
854 p->p_realtimer = *aitv;
856 if (timevalisset(&oitv->it_value)) {
857 if (timevalcmp(&oitv->it_value, &ctv, <))
858 timevalclear(&oitv->it_value);
860 timevalsub(&oitv->it_value, &ctv);
863 if (aitv->it_interval.tv_sec == 0 &&
864 aitv->it_interval.tv_usec != 0 &&
865 aitv->it_interval.tv_usec < tick)
866 aitv->it_interval.tv_usec = tick;
867 if (aitv->it_value.tv_sec == 0 &&
868 aitv->it_value.tv_usec != 0 &&
869 aitv->it_value.tv_usec < tick)
870 aitv->it_value.tv_usec = tick;
872 *oitv = p->p_stats->p_timer[which];
873 p->p_stats->p_timer[which] = *aitv;
877 if (KTRPOINT(td, KTR_STRUCT))
884 realitexpire_reset_callout(struct proc *p, sbintime_t *isbtp)
888 prec = isbtp == NULL ? tvtosbt(p->p_realtimer.it_interval) : *isbtp;
889 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
890 prec >> tc_precexp, realitexpire, p, C_ABSOLUTE);
894 itimer_proc_continue(struct proc *p)
900 PROC_LOCK_ASSERT(p, MA_OWNED);
902 if ((p->p_flag2 & P2_ITSTOPPED) != 0) {
903 p->p_flag2 &= ~P2_ITSTOPPED;
905 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >=))
908 realitexpire_reset_callout(p, NULL);
911 if (p->p_itimers != NULL) {
912 for (id = 3; id < TIMER_MAX; id++) {
913 it = p->p_itimers->its_timers[id];
916 if ((it->it_flags & ITF_PSTOPPED) != 0) {
918 if ((it->it_flags & ITF_PSTOPPED) != 0) {
919 it->it_flags &= ~ITF_PSTOPPED;
920 if ((it->it_flags & ITF_DELETING) == 0)
921 realtimer_expire_l(it, true);
930 * Real interval timer expired:
931 * send process whose timer expired an alarm signal.
932 * If time is not set up to reload, then just return.
933 * Else compute next time timer should go off which is > current time.
934 * This is where delay in processing this timeout causes multiple
935 * SIGALRM calls to be compressed into one.
936 * tvtohz() always adds 1 to allow for the time until the next clock
937 * interrupt being strictly less than 1 clock tick, but we don't want
938 * that here since we want to appear to be in sync with the clock
939 * interrupt even when we're delayed.
942 realitexpire(void *arg)
948 p = (struct proc *)arg;
949 kern_psignal(p, SIGALRM);
950 if (!timevalisset(&p->p_realtimer.it_interval)) {
951 timevalclear(&p->p_realtimer.it_value);
955 isbt = tvtosbt(p->p_realtimer.it_interval);
956 if (isbt >= sbt_timethreshold)
957 getmicrouptime(&ctv);
961 timevaladd(&p->p_realtimer.it_value,
962 &p->p_realtimer.it_interval);
963 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
965 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
966 p->p_flag2 |= P2_ITSTOPPED;
970 p->p_flag2 &= ~P2_ITSTOPPED;
971 realitexpire_reset_callout(p, &isbt);
975 * Check that a proposed value to load into the .it_value or
976 * .it_interval part of an interval timer is acceptable, and
977 * fix it to have at least minimal value (i.e. if it is less
978 * than the resolution of the clock, round it up.)
981 itimerfix(struct timeval *tv)
984 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
986 if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
987 tv->tv_usec < (u_int)tick / 16)
988 tv->tv_usec = (u_int)tick / 16;
993 * Decrement an interval timer by a specified number
994 * of microseconds, which must be less than a second,
995 * i.e. < 1000000. If the timer expires, then reload
996 * it. In this case, carry over (usec - old value) to
997 * reduce the value reloaded into the timer so that
998 * the timer does not drift. This routine assumes
999 * that it is called in a context where the timers
1000 * on which it is operating cannot change in value.
1003 itimerdecr(struct itimerval *itp, int usec)
1006 if (itp->it_value.tv_usec < usec) {
1007 if (itp->it_value.tv_sec == 0) {
1008 /* expired, and already in next interval */
1009 usec -= itp->it_value.tv_usec;
1012 itp->it_value.tv_usec += 1000000;
1013 itp->it_value.tv_sec--;
1015 itp->it_value.tv_usec -= usec;
1017 if (timevalisset(&itp->it_value))
1019 /* expired, exactly at end of interval */
1021 if (timevalisset(&itp->it_interval)) {
1022 itp->it_value = itp->it_interval;
1023 itp->it_value.tv_usec -= usec;
1024 if (itp->it_value.tv_usec < 0) {
1025 itp->it_value.tv_usec += 1000000;
1026 itp->it_value.tv_sec--;
1029 itp->it_value.tv_usec = 0; /* sec is already 0 */
1034 * Add and subtract routines for timevals.
1035 * N.B.: subtract routine doesn't deal with
1036 * results which are before the beginning,
1037 * it just gets very confused in this case.
1041 timevaladd(struct timeval *t1, const struct timeval *t2)
1044 t1->tv_sec += t2->tv_sec;
1045 t1->tv_usec += t2->tv_usec;
1050 timevalsub(struct timeval *t1, const struct timeval *t2)
1053 t1->tv_sec -= t2->tv_sec;
1054 t1->tv_usec -= t2->tv_usec;
1059 timevalfix(struct timeval *t1)
1062 if (t1->tv_usec < 0) {
1064 t1->tv_usec += 1000000;
1066 if (t1->tv_usec >= 1000000) {
1068 t1->tv_usec -= 1000000;
1073 * ratecheck(): simple time-based rate-limit checking.
1076 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1078 struct timeval tv, delta;
1081 getmicrouptime(&tv); /* NB: 10ms precision */
1083 timevalsub(&delta, lasttime);
1086 * check for 0,0 is so that the message will be seen at least once,
1087 * even if interval is huge.
1089 if (timevalcmp(&delta, mininterval, >=) ||
1090 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1099 * eventratecheck(): events per second limitation.
1101 * Return 0 if the limit is to be enforced (e.g. the caller
1102 * should ignore the event because of the rate limitation).
1104 * maxeps of 0 always causes zero to be returned. maxeps of -1
1105 * always causes 1 to be returned; this effectively defeats rate
1108 * Note that we maintain the struct timeval for compatibility
1109 * with other bsd systems. We reuse the storage and just monitor
1110 * clock ticks for minimal overhead.
1113 eventratecheck(struct timeval *lasttime, int *cureps, int maxeps)
1118 * Reset the last time and counter if this is the first call
1119 * or more than a second has passed since the last update of
1123 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1124 lasttime->tv_sec = now;
1126 return (maxeps != 0);
1128 (*cureps)++; /* NB: ignore potential overflow */
1129 return (maxeps < 0 || *cureps <= maxeps);
1136 static const struct kclock rt_clock = {
1137 .timer_create = realtimer_create,
1138 .timer_delete = realtimer_delete,
1139 .timer_settime = realtimer_settime,
1140 .timer_gettime = realtimer_gettime,
1143 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1144 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1145 register_posix_clock(CLOCK_REALTIME, &rt_clock);
1146 register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1147 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1148 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1149 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1153 register_posix_clock(int clockid, const struct kclock *clk)
1155 if ((unsigned)clockid >= MAX_CLOCKS) {
1156 printf("%s: invalid clockid\n", __func__);
1159 posix_clocks[clockid] = *clk;
1164 itimer_init(void *mem, int size, int flags)
1168 it = (struct itimer *)mem;
1169 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1174 itimer_fini(void *mem, int size)
1178 it = (struct itimer *)mem;
1179 mtx_destroy(&it->it_mtx);
1183 itimer_enter(struct itimer *it)
1186 mtx_assert(&it->it_mtx, MA_OWNED);
1191 itimer_leave(struct itimer *it)
1194 mtx_assert(&it->it_mtx, MA_OWNED);
1195 KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1197 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1201 #ifndef _SYS_SYSPROTO_H_
1202 struct ktimer_create_args {
1204 struct sigevent * evp;
1209 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1211 struct sigevent *evp, ev;
1215 if (uap->evp == NULL) {
1218 error = copyin(uap->evp, &ev, sizeof(ev));
1223 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1225 error = copyout(&id, uap->timerid, sizeof(int));
1227 kern_ktimer_delete(td, id);
1233 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1234 int *timerid, int preset_id)
1236 struct proc *p = td->td_proc;
1241 if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1244 if (posix_clocks[clock_id].timer_create == NULL)
1248 if (evp->sigev_notify != SIGEV_NONE &&
1249 evp->sigev_notify != SIGEV_SIGNAL &&
1250 evp->sigev_notify != SIGEV_THREAD_ID)
1252 if ((evp->sigev_notify == SIGEV_SIGNAL ||
1253 evp->sigev_notify == SIGEV_THREAD_ID) &&
1254 !_SIG_VALID(evp->sigev_signo))
1258 if (p->p_itimers == NULL)
1261 it = uma_zalloc(itimer_zone, M_WAITOK);
1263 it->it_usecount = 0;
1264 timespecclear(&it->it_time.it_value);
1265 timespecclear(&it->it_time.it_interval);
1267 it->it_overrun_last = 0;
1268 it->it_clockid = clock_id;
1270 ksiginfo_init(&it->it_ksi);
1271 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1272 error = CLOCK_CALL(clock_id, timer_create, (it));
1277 if (preset_id != -1) {
1278 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1280 if (p->p_itimers->its_timers[id] != NULL) {
1287 * Find a free timer slot, skipping those reserved
1290 for (id = 3; id < TIMER_MAX; id++)
1291 if (p->p_itimers->its_timers[id] == NULL)
1293 if (id == TIMER_MAX) {
1299 p->p_itimers->its_timers[id] = it;
1301 it->it_sigev = *evp;
1303 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1306 case CLOCK_REALTIME:
1307 it->it_sigev.sigev_signo = SIGALRM;
1310 it->it_sigev.sigev_signo = SIGVTALRM;
1313 it->it_sigev.sigev_signo = SIGPROF;
1316 it->it_sigev.sigev_value.sival_int = id;
1319 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1320 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1321 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1322 it->it_ksi.ksi_code = SI_TIMER;
1323 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1324 it->it_ksi.ksi_timerid = id;
1332 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1334 uma_zfree(itimer_zone, it);
1338 #ifndef _SYS_SYSPROTO_H_
1339 struct ktimer_delete_args {
1344 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1347 return (kern_ktimer_delete(td, uap->timerid));
1350 static struct itimer *
1351 itimer_find(struct proc *p, int timerid)
1355 PROC_LOCK_ASSERT(p, MA_OWNED);
1356 if ((p->p_itimers == NULL) ||
1357 (timerid < 0) || (timerid >= TIMER_MAX) ||
1358 (it = p->p_itimers->its_timers[timerid]) == NULL) {
1362 if ((it->it_flags & ITF_DELETING) != 0) {
1370 kern_ktimer_delete(struct thread *td, int timerid)
1372 struct proc *p = td->td_proc;
1376 it = itimer_find(p, timerid);
1383 it->it_flags |= ITF_DELETING;
1384 while (it->it_usecount > 0) {
1385 it->it_flags |= ITF_WANTED;
1386 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1388 it->it_flags &= ~ITF_WANTED;
1389 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1393 if (KSI_ONQ(&it->it_ksi))
1394 sigqueue_take(&it->it_ksi);
1395 p->p_itimers->its_timers[timerid] = NULL;
1397 uma_zfree(itimer_zone, it);
1401 #ifndef _SYS_SYSPROTO_H_
1402 struct ktimer_settime_args {
1405 const struct itimerspec * value;
1406 struct itimerspec * ovalue;
1410 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1412 struct itimerspec val, oval, *ovalp;
1415 error = copyin(uap->value, &val, sizeof(val));
1418 ovalp = uap->ovalue != NULL ? &oval : NULL;
1419 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1420 if (error == 0 && uap->ovalue != NULL)
1421 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1426 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1427 struct itimerspec *val, struct itimerspec *oval)
1435 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1441 error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1449 #ifndef _SYS_SYSPROTO_H_
1450 struct ktimer_gettime_args {
1452 struct itimerspec * value;
1456 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1458 struct itimerspec val;
1461 error = kern_ktimer_gettime(td, uap->timerid, &val);
1463 error = copyout(&val, uap->value, sizeof(val));
1468 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1476 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1482 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1489 #ifndef _SYS_SYSPROTO_H_
1490 struct timer_getoverrun_args {
1495 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1498 return (kern_ktimer_getoverrun(td, uap->timerid));
1502 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1504 struct proc *p = td->td_proc;
1510 (it = itimer_find(p, timer_id)) == NULL) {
1514 td->td_retval[0] = it->it_overrun_last;
1523 realtimer_create(struct itimer *it)
1525 callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1530 realtimer_delete(struct itimer *it)
1532 mtx_assert(&it->it_mtx, MA_OWNED);
1535 * clear timer's value and interval to tell realtimer_expire
1536 * to not rearm the timer.
1538 timespecclear(&it->it_time.it_value);
1539 timespecclear(&it->it_time.it_interval);
1541 callout_drain(&it->it_callout);
1547 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1549 struct timespec cts;
1551 mtx_assert(&it->it_mtx, MA_OWNED);
1553 realtimer_clocktime(it->it_clockid, &cts);
1554 *ovalue = it->it_time;
1555 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1556 timespecsub(&ovalue->it_value, &cts, &ovalue->it_value);
1557 if (ovalue->it_value.tv_sec < 0 ||
1558 (ovalue->it_value.tv_sec == 0 &&
1559 ovalue->it_value.tv_nsec == 0)) {
1560 ovalue->it_value.tv_sec = 0;
1561 ovalue->it_value.tv_nsec = 1;
1568 realtimer_settime(struct itimer *it, int flags, struct itimerspec *value,
1569 struct itimerspec *ovalue)
1571 struct timespec cts, ts;
1573 struct itimerspec val;
1575 mtx_assert(&it->it_mtx, MA_OWNED);
1578 if (itimespecfix(&val.it_value))
1581 if (timespecisset(&val.it_value)) {
1582 if (itimespecfix(&val.it_interval))
1585 timespecclear(&val.it_interval);
1589 realtimer_gettime(it, ovalue);
1592 if (timespecisset(&val.it_value)) {
1593 realtimer_clocktime(it->it_clockid, &cts);
1595 if ((flags & TIMER_ABSTIME) == 0) {
1596 /* Convert to absolute time. */
1597 timespecadd(&it->it_time.it_value, &cts,
1598 &it->it_time.it_value);
1600 timespecsub(&ts, &cts, &ts);
1602 * We don't care if ts is negative, tztohz will
1606 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1607 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1610 callout_stop(&it->it_callout);
1617 realtimer_clocktime(clockid_t id, struct timespec *ts)
1619 if (id == CLOCK_REALTIME)
1621 else /* CLOCK_MONOTONIC */
1626 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1630 PROC_LOCK_ASSERT(p, MA_OWNED);
1631 it = itimer_find(p, timerid);
1633 ksi->ksi_overrun = it->it_overrun;
1634 it->it_overrun_last = it->it_overrun;
1643 itimespecfix(struct timespec *ts)
1646 if (!timespecvalid_interval(ts))
1648 if ((UINT64_MAX - ts->tv_nsec) / NS_PER_SEC < ts->tv_sec)
1650 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1651 ts->tv_nsec = tick * 1000;
1655 #define timespectons(tsp) \
1656 ((uint64_t)(tsp)->tv_sec * NS_PER_SEC + (tsp)->tv_nsec)
1657 #define timespecfromns(ns) (struct timespec){ \
1658 .tv_sec = (ns) / NS_PER_SEC, \
1659 .tv_nsec = (ns) % NS_PER_SEC \
1663 realtimer_expire_l(struct itimer *it, bool proc_locked)
1665 struct timespec cts, ts;
1668 uint64_t interval, now, overruns, value;
1670 realtimer_clocktime(it->it_clockid, &cts);
1671 /* Only fire if time is reached. */
1672 if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1673 if (timespecisset(&it->it_time.it_interval)) {
1674 timespecadd(&it->it_time.it_value,
1675 &it->it_time.it_interval,
1676 &it->it_time.it_value);
1678 interval = timespectons(&it->it_time.it_interval);
1679 value = timespectons(&it->it_time.it_value);
1680 now = timespectons(&cts);
1684 * We missed at least one period.
1686 overruns = howmany(now - value + 1, interval);
1687 if (it->it_overrun + overruns >=
1689 it->it_overrun + overruns <= INT_MAX) {
1690 it->it_overrun += (int)overruns;
1692 it->it_overrun = INT_MAX;
1693 it->it_ksi.ksi_errno = ERANGE;
1696 now + interval - (now - value) % interval;
1697 it->it_time.it_value = timespecfromns(value);
1700 /* single shot timer ? */
1701 timespecclear(&it->it_time.it_value);
1705 if (timespecisset(&it->it_time.it_value)) {
1706 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1707 it->it_flags |= ITF_PSTOPPED;
1709 timespecsub(&it->it_time.it_value, &cts, &ts);
1710 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1711 callout_reset(&it->it_callout, tvtohz(&tv),
1712 realtimer_expire, it);
1725 } else if (timespecisset(&it->it_time.it_value)) {
1727 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1728 it->it_flags |= ITF_PSTOPPED;
1730 ts = it->it_time.it_value;
1731 timespecsub(&ts, &cts, &ts);
1732 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1733 callout_reset(&it->it_callout, tvtohz(&tv),
1734 realtimer_expire, it);
1739 /* Timeout callback for realtime timer */
1741 realtimer_expire(void *arg)
1743 realtimer_expire_l(arg, false);
1747 itimer_fire(struct itimer *it)
1749 struct proc *p = it->it_proc;
1752 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1753 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1754 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1756 timespecclear(&it->it_time.it_value);
1757 timespecclear(&it->it_time.it_interval);
1758 callout_stop(&it->it_callout);
1762 if (!KSI_ONQ(&it->it_ksi)) {
1763 it->it_ksi.ksi_errno = 0;
1764 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1765 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1767 if (it->it_overrun < INT_MAX)
1770 it->it_ksi.ksi_errno = ERANGE;
1777 itimers_alloc(struct proc *p)
1779 struct itimers *its;
1781 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1783 if (p->p_itimers == NULL) {
1789 free(its, M_SUBPROC);
1793 /* Clean up timers when some process events are being triggered. */
1795 itimers_event_exit_exec(int start_idx, struct proc *p)
1797 struct itimers *its;
1805 for (i = start_idx; i < TIMER_MAX; ++i) {
1806 if ((it = its->its_timers[i]) != NULL)
1807 kern_ktimer_delete(curthread, i);
1809 if (its->its_timers[0] == NULL && its->its_timers[1] == NULL &&
1810 its->its_timers[2] == NULL) {
1811 /* Synchronize with itimer_proc_continue(). */
1813 p->p_itimers = NULL;
1815 free(its, M_SUBPROC);
1820 itimers_exec(struct proc *p)
1823 * According to susv3, XSI interval timers should be inherited
1826 itimers_event_exit_exec(3, p);
1830 itimers_exit(struct proc *p)
1832 itimers_event_exit_exec(0, p);