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 __FBSDID("$FreeBSD$");
37 #include "opt_ktrace.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/limits.h>
42 #include <sys/clock.h>
44 #include <sys/mutex.h>
45 #include <sys/sysproto.h>
46 #include <sys/resourcevar.h>
47 #include <sys/signalvar.h>
48 #include <sys/kernel.h>
49 #include <sys/sleepqueue.h>
50 #include <sys/syscallsubr.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysent.h>
55 #include <sys/posix4.h>
57 #include <sys/timers.h>
58 #include <sys/timetc.h>
59 #include <sys/vnode.h>
61 #include <sys/ktrace.h>
65 #include <vm/vm_extern.h>
67 #define MAX_CLOCKS (CLOCK_MONOTONIC+1)
68 #define CPUCLOCK_BIT 0x80000000
69 #define CPUCLOCK_PROCESS_BIT 0x40000000
70 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
71 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid))
72 #define MAKE_PROCESS_CPUCLOCK(pid) \
73 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
75 static struct kclock posix_clocks[MAX_CLOCKS];
76 static uma_zone_t itimer_zone = NULL;
79 * Time of day and interval timer support.
81 * These routines provide the kernel entry points to get and set
82 * the time-of-day and per-process interval timers. Subroutines
83 * here provide support for adding and subtracting timeval structures
84 * and decrementing interval timers, optionally reloading the interval
85 * timers when they expire.
88 static int settime(struct thread *, struct timeval *);
89 static void timevalfix(struct timeval *);
90 static int user_clock_nanosleep(struct thread *td, clockid_t clock_id,
91 int flags, const struct timespec *ua_rqtp,
92 struct timespec *ua_rmtp);
94 static void itimer_start(void);
95 static int itimer_init(void *, int, int);
96 static void itimer_fini(void *, int);
97 static void itimer_enter(struct itimer *);
98 static void itimer_leave(struct itimer *);
99 static struct itimer *itimer_find(struct proc *, int);
100 static void itimers_alloc(struct proc *);
101 static int realtimer_create(struct itimer *);
102 static int realtimer_gettime(struct itimer *, struct itimerspec *);
103 static int realtimer_settime(struct itimer *, int,
104 struct itimerspec *, struct itimerspec *);
105 static int realtimer_delete(struct itimer *);
106 static void realtimer_clocktime(clockid_t, struct timespec *);
107 static void realtimer_expire(void *);
108 static void realtimer_expire_l(struct itimer *it, bool proc_locked);
110 static int register_posix_clock(int, const struct kclock *);
111 static void itimer_fire(struct itimer *it);
112 static int itimespecfix(struct timespec *ts);
114 #define CLOCK_CALL(clock, call, arglist) \
115 ((*posix_clocks[clock].call) arglist)
117 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
120 settime(struct thread *td, struct timeval *tv)
122 struct timeval delta, tv1, tv2;
123 static struct timeval maxtime, laststep;
128 timevalsub(&delta, &tv1);
131 * If the system is secure, we do not allow the time to be
132 * set to a value earlier than 1 second less than the highest
133 * time we have yet seen. The worst a miscreant can do in
134 * this circumstance is "freeze" time. He couldn't go
137 * We similarly do not allow the clock to be stepped more
138 * than one second, nor more than once per second. This allows
139 * a miscreant to make the clock march double-time, but no worse.
141 if (securelevel_gt(td->td_ucred, 1) != 0) {
142 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
144 * Update maxtime to latest time we've seen.
146 if (tv1.tv_sec > maxtime.tv_sec)
149 timevalsub(&tv2, &maxtime);
150 if (tv2.tv_sec < -1) {
151 tv->tv_sec = maxtime.tv_sec - 1;
152 printf("Time adjustment clamped to -1 second\n");
155 if (tv1.tv_sec == laststep.tv_sec)
157 if (delta.tv_sec > 1) {
158 tv->tv_sec = tv1.tv_sec + 1;
159 printf("Time adjustment clamped to +1 second\n");
165 ts.tv_sec = tv->tv_sec;
166 ts.tv_nsec = tv->tv_usec * 1000;
172 #ifndef _SYS_SYSPROTO_H_
173 struct clock_getcpuclockid2_args {
181 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
186 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
188 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
193 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
202 case CPUCLOCK_WHICH_PID:
204 error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
210 pid = td->td_proc->p_pid;
212 *clk_id = MAKE_PROCESS_CPUCLOCK(pid);
214 case CPUCLOCK_WHICH_TID:
215 tid = id == 0 ? td->td_tid : id;
216 *clk_id = MAKE_THREAD_CPUCLOCK(tid);
223 #ifndef _SYS_SYSPROTO_H_
224 struct clock_gettime_args {
231 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
236 error = kern_clock_gettime(td, uap->clock_id, &ats);
238 error = copyout(&ats, uap->tp, sizeof(ats));
244 cputick2timespec(uint64_t runtime, struct timespec *ats)
246 runtime = cputick2usec(runtime);
247 ats->tv_sec = runtime / 1000000;
248 ats->tv_nsec = runtime % 1000000 * 1000;
252 kern_thread_cputime(struct thread *targettd, struct timespec *ats)
254 uint64_t runtime, curtime, switchtime;
256 if (targettd == NULL) { /* current thread */
258 switchtime = PCPU_GET(switchtime);
259 curtime = cpu_ticks();
260 runtime = curthread->td_runtime;
262 runtime += curtime - switchtime;
264 PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED);
265 thread_lock(targettd);
266 runtime = targettd->td_runtime;
267 thread_unlock(targettd);
269 cputick2timespec(runtime, ats);
273 kern_process_cputime(struct proc *targetp, struct timespec *ats)
278 PROC_LOCK_ASSERT(targetp, MA_OWNED);
279 PROC_STATLOCK(targetp);
280 rufetch(targetp, &ru);
281 runtime = targetp->p_rux.rux_runtime;
282 if (curthread->td_proc == targetp)
283 runtime += cpu_ticks() - PCPU_GET(switchtime);
284 PROC_STATUNLOCK(targetp);
285 cputick2timespec(runtime, ats);
289 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
298 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
299 tid = clock_id & CPUCLOCK_ID_MASK;
300 td2 = tdfind(tid, p->p_pid);
303 kern_thread_cputime(td2, ats);
304 PROC_UNLOCK(td2->td_proc);
306 pid = clock_id & CPUCLOCK_ID_MASK;
307 error = pget(pid, PGET_CANSEE, &p2);
310 kern_process_cputime(p2, ats);
317 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
319 struct timeval sys, user;
324 case CLOCK_REALTIME: /* Default to precise. */
325 case CLOCK_REALTIME_PRECISE:
328 case CLOCK_REALTIME_FAST:
334 calcru(p, &user, &sys);
337 TIMEVAL_TO_TIMESPEC(&user, ats);
342 calcru(p, &user, &sys);
345 timevaladd(&user, &sys);
346 TIMEVAL_TO_TIMESPEC(&user, ats);
348 case CLOCK_MONOTONIC: /* Default to precise. */
349 case CLOCK_MONOTONIC_PRECISE:
351 case CLOCK_UPTIME_PRECISE:
354 case CLOCK_UPTIME_FAST:
355 case CLOCK_MONOTONIC_FAST:
359 ats->tv_sec = time_second;
362 case CLOCK_THREAD_CPUTIME_ID:
363 kern_thread_cputime(NULL, ats);
365 case CLOCK_PROCESS_CPUTIME_ID:
367 kern_process_cputime(p, ats);
371 if ((int)clock_id >= 0)
373 return (get_cputime(td, clock_id, ats));
378 #ifndef _SYS_SYSPROTO_H_
379 struct clock_settime_args {
381 const struct timespec *tp;
386 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
391 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
393 return (kern_clock_settime(td, uap->clock_id, &ats));
396 static int allow_insane_settime = 0;
397 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
398 &allow_insane_settime, 0,
399 "do not perform possibly restrictive checks on settime(2) args");
402 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
407 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
409 if (clock_id != CLOCK_REALTIME)
411 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 ||
414 if (!allow_insane_settime &&
415 (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 ||
416 ats->tv_sec < utc_offset()))
418 /* XXX Don't convert nsec->usec and back */
419 TIMESPEC_TO_TIMEVAL(&atv, ats);
420 error = settime(td, &atv);
424 #ifndef _SYS_SYSPROTO_H_
425 struct clock_getres_args {
431 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
439 error = kern_clock_getres(td, uap->clock_id, &ts);
441 error = copyout(&ts, uap->tp, sizeof(ts));
446 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
452 case CLOCK_REALTIME_FAST:
453 case CLOCK_REALTIME_PRECISE:
454 case CLOCK_MONOTONIC:
455 case CLOCK_MONOTONIC_FAST:
456 case CLOCK_MONOTONIC_PRECISE:
458 case CLOCK_UPTIME_FAST:
459 case CLOCK_UPTIME_PRECISE:
461 * Round up the result of the division cheaply by adding 1.
462 * Rounding up is especially important if rounding down
463 * would give 0. Perfect rounding is unimportant.
465 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
469 /* Accurately round up here because we can do so cheaply. */
470 ts->tv_nsec = howmany(1000000000, hz);
476 case CLOCK_THREAD_CPUTIME_ID:
477 case CLOCK_PROCESS_CPUTIME_ID:
479 /* sync with cputick2usec */
480 ts->tv_nsec = 1000000 / cpu_tickrate();
481 if (ts->tv_nsec == 0)
485 if ((int)clock_id < 0)
493 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
496 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
500 static uint8_t nanowait[MAXCPU];
503 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
504 const struct timespec *rqt, struct timespec *rmt)
506 struct timespec ts, now;
507 sbintime_t sbt, sbtt, prec, tmp;
512 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
514 if ((flags & ~TIMER_ABSTIME) != 0)
518 case CLOCK_REALTIME_PRECISE:
519 case CLOCK_REALTIME_FAST:
521 is_abs_real = (flags & TIMER_ABSTIME) != 0;
523 case CLOCK_MONOTONIC:
524 case CLOCK_MONOTONIC_PRECISE:
525 case CLOCK_MONOTONIC_FAST:
527 case CLOCK_UPTIME_PRECISE:
528 case CLOCK_UPTIME_FAST:
533 case CLOCK_PROCESS_CPUTIME_ID:
535 case CLOCK_THREAD_CPUTIME_ID:
541 if ((flags & TIMER_ABSTIME) != 0) {
544 atomic_load_acq_int(&rtc_generation);
545 error = kern_clock_gettime(td, clock_id, &now);
546 KASSERT(error == 0, ("kern_clock_gettime: %d", error));
547 timespecsub(&ts, &now, &ts);
549 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
553 if (ts.tv_sec > INT32_MAX / 2) {
554 over = ts.tv_sec - INT32_MAX / 2;
561 if (TIMESEL(&sbt, tmp))
564 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
565 sbt, prec, C_ABSOLUTE);
566 } while (error == 0 && is_abs_real && td->td_rtcgen == 0);
568 if (error != EWOULDBLOCK) {
569 if (TIMESEL(&sbtt, tmp))
573 if (error == ERESTART)
575 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
576 ts = sbttots(sbt - sbtt);
587 #ifndef _SYS_SYSPROTO_H_
588 struct nanosleep_args {
589 struct timespec *rqtp;
590 struct timespec *rmtp;
595 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
598 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
599 uap->rqtp, uap->rmtp));
602 #ifndef _SYS_SYSPROTO_H_
603 struct clock_nanosleep_args {
606 struct timespec *rqtp;
607 struct timespec *rmtp;
612 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
616 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
618 return (kern_posix_error(td, error));
622 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
623 const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
625 struct timespec rmt, rqt;
628 error = copyin(ua_rqtp, &rqt, sizeof(rqt));
631 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
632 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
633 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
640 #ifndef _SYS_SYSPROTO_H_
641 struct gettimeofday_args {
643 struct timezone *tzp;
648 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
656 error = copyout(&atv, uap->tp, sizeof (atv));
658 if (error == 0 && uap->tzp != NULL) {
659 rtz.tz_minuteswest = 0;
661 error = copyout(&rtz, uap->tzp, sizeof (rtz));
666 #ifndef _SYS_SYSPROTO_H_
667 struct settimeofday_args {
669 struct timezone *tzp;
674 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
676 struct timeval atv, *tvp;
677 struct timezone atz, *tzp;
681 error = copyin(uap->tv, &atv, sizeof(atv));
688 error = copyin(uap->tzp, &atz, sizeof(atz));
694 return (kern_settimeofday(td, tvp, tzp));
698 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
702 error = priv_check(td, PRIV_SETTIMEOFDAY);
705 /* Verify all parameters before changing time. */
707 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
710 error = settime(td, tv);
716 * Get value of an interval timer. The process virtual and profiling virtual
717 * time timers are kept in the p_stats area, since they can be swapped out.
718 * These are kept internally in the way they are specified externally: in
719 * time until they expire.
721 * The real time interval timer is kept in the process table slot for the
722 * process, and its value (it_value) is kept as an absolute time rather than
723 * as a delta, so that it is easy to keep periodic real-time signals from
726 * Virtual time timers are processed in the hardclock() routine of
727 * kern_clock.c. The real time timer is processed by a timeout routine,
728 * called from the softclock() routine. Since a callout may be delayed in
729 * real time due to interrupt processing in the system, it is possible for
730 * the real time timeout routine (realitexpire, given below), to be delayed
731 * in real time past when it is supposed to occur. It does not suffice,
732 * therefore, to reload the real timer .it_value from the real time timers
733 * .it_interval. Rather, we compute the next time in absolute time the timer
736 #ifndef _SYS_SYSPROTO_H_
737 struct getitimer_args {
739 struct itimerval *itv;
743 sys_getitimer(struct thread *td, struct getitimer_args *uap)
745 struct itimerval aitv;
748 error = kern_getitimer(td, uap->which, &aitv);
751 return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
755 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
757 struct proc *p = td->td_proc;
760 if (which > ITIMER_PROF)
763 if (which == ITIMER_REAL) {
765 * Convert from absolute to relative time in .it_value
766 * part of real time timer. If time for real time timer
767 * has passed return 0, else return difference between
768 * current time and time for the timer to go off.
771 *aitv = p->p_realtimer;
773 if (timevalisset(&aitv->it_value)) {
775 if (timevalcmp(&aitv->it_value, &ctv, <))
776 timevalclear(&aitv->it_value);
778 timevalsub(&aitv->it_value, &ctv);
782 *aitv = p->p_stats->p_timer[which];
786 if (KTRPOINT(td, KTR_STRUCT))
792 #ifndef _SYS_SYSPROTO_H_
793 struct setitimer_args {
795 struct itimerval *itv, *oitv;
799 sys_setitimer(struct thread *td, struct setitimer_args *uap)
801 struct itimerval aitv, oitv;
804 if (uap->itv == NULL) {
805 uap->itv = uap->oitv;
806 return (sys_getitimer(td, (struct getitimer_args *)uap));
809 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
811 error = kern_setitimer(td, uap->which, &aitv, &oitv);
812 if (error != 0 || uap->oitv == NULL)
814 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
818 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
819 struct itimerval *oitv)
821 struct proc *p = td->td_proc;
826 return (kern_getitimer(td, which, oitv));
828 if (which > ITIMER_PROF)
831 if (KTRPOINT(td, KTR_STRUCT))
834 if (itimerfix(&aitv->it_value) ||
835 aitv->it_value.tv_sec > INT32_MAX / 2)
837 if (!timevalisset(&aitv->it_value))
838 timevalclear(&aitv->it_interval);
839 else if (itimerfix(&aitv->it_interval) ||
840 aitv->it_interval.tv_sec > INT32_MAX / 2)
843 if (which == ITIMER_REAL) {
845 if (timevalisset(&p->p_realtimer.it_value))
846 callout_stop(&p->p_itcallout);
848 if (timevalisset(&aitv->it_value)) {
849 pr = tvtosbt(aitv->it_value) >> tc_precexp;
850 timevaladd(&aitv->it_value, &ctv);
851 sbt = tvtosbt(aitv->it_value);
852 callout_reset_sbt(&p->p_itcallout, sbt, pr,
853 realitexpire, p, C_ABSOLUTE);
855 *oitv = p->p_realtimer;
856 p->p_realtimer = *aitv;
858 if (timevalisset(&oitv->it_value)) {
859 if (timevalcmp(&oitv->it_value, &ctv, <))
860 timevalclear(&oitv->it_value);
862 timevalsub(&oitv->it_value, &ctv);
865 if (aitv->it_interval.tv_sec == 0 &&
866 aitv->it_interval.tv_usec != 0 &&
867 aitv->it_interval.tv_usec < tick)
868 aitv->it_interval.tv_usec = tick;
869 if (aitv->it_value.tv_sec == 0 &&
870 aitv->it_value.tv_usec != 0 &&
871 aitv->it_value.tv_usec < tick)
872 aitv->it_value.tv_usec = tick;
874 *oitv = p->p_stats->p_timer[which];
875 p->p_stats->p_timer[which] = *aitv;
879 if (KTRPOINT(td, KTR_STRUCT))
886 realitexpire_reset_callout(struct proc *p, sbintime_t *isbtp)
890 prec = isbtp == NULL ? tvtosbt(p->p_realtimer.it_interval) : *isbtp;
891 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
892 prec >> tc_precexp, realitexpire, p, C_ABSOLUTE);
896 itimer_proc_continue(struct proc *p)
902 PROC_LOCK_ASSERT(p, MA_OWNED);
904 if ((p->p_flag2 & P2_ITSTOPPED) != 0) {
905 p->p_flag2 &= ~P2_ITSTOPPED;
907 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >=))
910 realitexpire_reset_callout(p, NULL);
913 if (p->p_itimers != NULL) {
914 for (id = 3; id < TIMER_MAX; id++) {
915 it = p->p_itimers->its_timers[id];
918 if ((it->it_flags & ITF_PSTOPPED) != 0) {
920 if ((it->it_flags & ITF_PSTOPPED) != 0) {
921 it->it_flags &= ~ITF_PSTOPPED;
922 if ((it->it_flags & ITF_DELETING) == 0)
923 realtimer_expire_l(it, true);
932 * Real interval timer expired:
933 * send process whose timer expired an alarm signal.
934 * If time is not set up to reload, then just return.
935 * Else compute next time timer should go off which is > current time.
936 * This is where delay in processing this timeout causes multiple
937 * SIGALRM calls to be compressed into one.
938 * tvtohz() always adds 1 to allow for the time until the next clock
939 * interrupt being strictly less than 1 clock tick, but we don't want
940 * that here since we want to appear to be in sync with the clock
941 * interrupt even when we're delayed.
944 realitexpire(void *arg)
950 p = (struct proc *)arg;
951 kern_psignal(p, SIGALRM);
952 if (!timevalisset(&p->p_realtimer.it_interval)) {
953 timevalclear(&p->p_realtimer.it_value);
954 if (p->p_flag & P_WEXIT)
955 wakeup(&p->p_itcallout);
959 isbt = tvtosbt(p->p_realtimer.it_interval);
960 if (isbt >= sbt_timethreshold)
961 getmicrouptime(&ctv);
965 timevaladd(&p->p_realtimer.it_value,
966 &p->p_realtimer.it_interval);
967 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
969 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
970 p->p_flag2 |= P2_ITSTOPPED;
974 p->p_flag2 &= ~P2_ITSTOPPED;
975 realitexpire_reset_callout(p, &isbt);
979 * Check that a proposed value to load into the .it_value or
980 * .it_interval part of an interval timer is acceptable, and
981 * fix it to have at least minimal value (i.e. if it is less
982 * than the resolution of the clock, round it up.)
985 itimerfix(struct timeval *tv)
988 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
990 if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
991 tv->tv_usec < (u_int)tick / 16)
992 tv->tv_usec = (u_int)tick / 16;
997 * Decrement an interval timer by a specified number
998 * of microseconds, which must be less than a second,
999 * i.e. < 1000000. If the timer expires, then reload
1000 * it. In this case, carry over (usec - old value) to
1001 * reduce the value reloaded into the timer so that
1002 * the timer does not drift. This routine assumes
1003 * that it is called in a context where the timers
1004 * on which it is operating cannot change in value.
1007 itimerdecr(struct itimerval *itp, int usec)
1010 if (itp->it_value.tv_usec < usec) {
1011 if (itp->it_value.tv_sec == 0) {
1012 /* expired, and already in next interval */
1013 usec -= itp->it_value.tv_usec;
1016 itp->it_value.tv_usec += 1000000;
1017 itp->it_value.tv_sec--;
1019 itp->it_value.tv_usec -= usec;
1021 if (timevalisset(&itp->it_value))
1023 /* expired, exactly at end of interval */
1025 if (timevalisset(&itp->it_interval)) {
1026 itp->it_value = itp->it_interval;
1027 itp->it_value.tv_usec -= usec;
1028 if (itp->it_value.tv_usec < 0) {
1029 itp->it_value.tv_usec += 1000000;
1030 itp->it_value.tv_sec--;
1033 itp->it_value.tv_usec = 0; /* sec is already 0 */
1038 * Add and subtract routines for timevals.
1039 * N.B.: subtract routine doesn't deal with
1040 * results which are before the beginning,
1041 * it just gets very confused in this case.
1045 timevaladd(struct timeval *t1, const struct timeval *t2)
1048 t1->tv_sec += t2->tv_sec;
1049 t1->tv_usec += t2->tv_usec;
1054 timevalsub(struct timeval *t1, const struct timeval *t2)
1057 t1->tv_sec -= t2->tv_sec;
1058 t1->tv_usec -= t2->tv_usec;
1063 timevalfix(struct timeval *t1)
1066 if (t1->tv_usec < 0) {
1068 t1->tv_usec += 1000000;
1070 if (t1->tv_usec >= 1000000) {
1072 t1->tv_usec -= 1000000;
1077 * ratecheck(): simple time-based rate-limit checking.
1080 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1082 struct timeval tv, delta;
1085 getmicrouptime(&tv); /* NB: 10ms precision */
1087 timevalsub(&delta, lasttime);
1090 * check for 0,0 is so that the message will be seen at least once,
1091 * even if interval is huge.
1093 if (timevalcmp(&delta, mininterval, >=) ||
1094 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1103 * ppsratecheck(): packets (or events) per second limitation.
1105 * Return 0 if the limit is to be enforced (e.g. the caller
1106 * should drop a packet because of the rate limitation).
1108 * maxpps of 0 always causes zero to be returned. maxpps of -1
1109 * always causes 1 to be returned; this effectively defeats rate
1112 * Note that we maintain the struct timeval for compatibility
1113 * with other bsd systems. We reuse the storage and just monitor
1114 * clock ticks for minimal overhead.
1117 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1122 * Reset the last time and counter if this is the first call
1123 * or more than a second has passed since the last update of
1127 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1128 lasttime->tv_sec = now;
1130 return (maxpps != 0);
1132 (*curpps)++; /* NB: ignore potential overflow */
1133 return (maxpps < 0 || *curpps <= maxpps);
1140 static const struct kclock rt_clock = {
1141 .timer_create = realtimer_create,
1142 .timer_delete = realtimer_delete,
1143 .timer_settime = realtimer_settime,
1144 .timer_gettime = realtimer_gettime,
1147 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1148 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1149 register_posix_clock(CLOCK_REALTIME, &rt_clock);
1150 register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1151 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1152 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1153 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1157 register_posix_clock(int clockid, const struct kclock *clk)
1159 if ((unsigned)clockid >= MAX_CLOCKS) {
1160 printf("%s: invalid clockid\n", __func__);
1163 posix_clocks[clockid] = *clk;
1168 itimer_init(void *mem, int size, int flags)
1172 it = (struct itimer *)mem;
1173 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1178 itimer_fini(void *mem, int size)
1182 it = (struct itimer *)mem;
1183 mtx_destroy(&it->it_mtx);
1187 itimer_enter(struct itimer *it)
1190 mtx_assert(&it->it_mtx, MA_OWNED);
1195 itimer_leave(struct itimer *it)
1198 mtx_assert(&it->it_mtx, MA_OWNED);
1199 KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1201 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1205 #ifndef _SYS_SYSPROTO_H_
1206 struct ktimer_create_args {
1208 struct sigevent * evp;
1213 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1215 struct sigevent *evp, ev;
1219 if (uap->evp == NULL) {
1222 error = copyin(uap->evp, &ev, sizeof(ev));
1227 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1229 error = copyout(&id, uap->timerid, sizeof(int));
1231 kern_ktimer_delete(td, id);
1237 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1238 int *timerid, int preset_id)
1240 struct proc *p = td->td_proc;
1245 if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1248 if (posix_clocks[clock_id].timer_create == NULL)
1252 if (evp->sigev_notify != SIGEV_NONE &&
1253 evp->sigev_notify != SIGEV_SIGNAL &&
1254 evp->sigev_notify != SIGEV_THREAD_ID)
1256 if ((evp->sigev_notify == SIGEV_SIGNAL ||
1257 evp->sigev_notify == SIGEV_THREAD_ID) &&
1258 !_SIG_VALID(evp->sigev_signo))
1262 if (p->p_itimers == NULL)
1265 it = uma_zalloc(itimer_zone, M_WAITOK);
1267 it->it_usecount = 0;
1269 timespecclear(&it->it_time.it_value);
1270 timespecclear(&it->it_time.it_interval);
1272 it->it_overrun_last = 0;
1273 it->it_clockid = clock_id;
1274 it->it_timerid = -1;
1276 ksiginfo_init(&it->it_ksi);
1277 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1278 error = CLOCK_CALL(clock_id, timer_create, (it));
1283 if (preset_id != -1) {
1284 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1286 if (p->p_itimers->its_timers[id] != NULL) {
1293 * Find a free timer slot, skipping those reserved
1296 for (id = 3; id < TIMER_MAX; id++)
1297 if (p->p_itimers->its_timers[id] == NULL)
1299 if (id == TIMER_MAX) {
1305 it->it_timerid = id;
1306 p->p_itimers->its_timers[id] = it;
1308 it->it_sigev = *evp;
1310 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1313 case CLOCK_REALTIME:
1314 it->it_sigev.sigev_signo = SIGALRM;
1317 it->it_sigev.sigev_signo = SIGVTALRM;
1320 it->it_sigev.sigev_signo = SIGPROF;
1323 it->it_sigev.sigev_value.sival_int = id;
1326 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1327 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1328 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1329 it->it_ksi.ksi_code = SI_TIMER;
1330 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1331 it->it_ksi.ksi_timerid = id;
1339 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1341 uma_zfree(itimer_zone, it);
1345 #ifndef _SYS_SYSPROTO_H_
1346 struct ktimer_delete_args {
1351 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1354 return (kern_ktimer_delete(td, uap->timerid));
1357 static struct itimer *
1358 itimer_find(struct proc *p, int timerid)
1362 PROC_LOCK_ASSERT(p, MA_OWNED);
1363 if ((p->p_itimers == NULL) ||
1364 (timerid < 0) || (timerid >= TIMER_MAX) ||
1365 (it = p->p_itimers->its_timers[timerid]) == NULL) {
1369 if ((it->it_flags & ITF_DELETING) != 0) {
1377 kern_ktimer_delete(struct thread *td, int timerid)
1379 struct proc *p = td->td_proc;
1383 it = itimer_find(p, timerid);
1390 it->it_flags |= ITF_DELETING;
1391 while (it->it_usecount > 0) {
1392 it->it_flags |= ITF_WANTED;
1393 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1395 it->it_flags &= ~ITF_WANTED;
1396 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1400 if (KSI_ONQ(&it->it_ksi))
1401 sigqueue_take(&it->it_ksi);
1402 p->p_itimers->its_timers[timerid] = NULL;
1404 uma_zfree(itimer_zone, it);
1408 #ifndef _SYS_SYSPROTO_H_
1409 struct ktimer_settime_args {
1412 const struct itimerspec * value;
1413 struct itimerspec * ovalue;
1417 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1419 struct itimerspec val, oval, *ovalp;
1422 error = copyin(uap->value, &val, sizeof(val));
1425 ovalp = uap->ovalue != NULL ? &oval : NULL;
1426 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1427 if (error == 0 && uap->ovalue != NULL)
1428 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1433 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1434 struct itimerspec *val, struct itimerspec *oval)
1442 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1448 error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1456 #ifndef _SYS_SYSPROTO_H_
1457 struct ktimer_gettime_args {
1459 struct itimerspec * value;
1463 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1465 struct itimerspec val;
1468 error = kern_ktimer_gettime(td, uap->timerid, &val);
1470 error = copyout(&val, uap->value, sizeof(val));
1475 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1483 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1489 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1496 #ifndef _SYS_SYSPROTO_H_
1497 struct timer_getoverrun_args {
1502 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1505 return (kern_ktimer_getoverrun(td, uap->timerid));
1509 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1511 struct proc *p = td->td_proc;
1517 (it = itimer_find(p, timer_id)) == NULL) {
1521 td->td_retval[0] = it->it_overrun_last;
1530 realtimer_create(struct itimer *it)
1532 callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1537 realtimer_delete(struct itimer *it)
1539 mtx_assert(&it->it_mtx, MA_OWNED);
1542 * clear timer's value and interval to tell realtimer_expire
1543 * to not rearm the timer.
1545 timespecclear(&it->it_time.it_value);
1546 timespecclear(&it->it_time.it_interval);
1548 callout_drain(&it->it_callout);
1554 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1556 struct timespec cts;
1558 mtx_assert(&it->it_mtx, MA_OWNED);
1560 realtimer_clocktime(it->it_clockid, &cts);
1561 *ovalue = it->it_time;
1562 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1563 timespecsub(&ovalue->it_value, &cts, &ovalue->it_value);
1564 if (ovalue->it_value.tv_sec < 0 ||
1565 (ovalue->it_value.tv_sec == 0 &&
1566 ovalue->it_value.tv_nsec == 0)) {
1567 ovalue->it_value.tv_sec = 0;
1568 ovalue->it_value.tv_nsec = 1;
1575 realtimer_settime(struct itimer *it, int flags, struct itimerspec *value,
1576 struct itimerspec *ovalue)
1578 struct timespec cts, ts;
1580 struct itimerspec val;
1582 mtx_assert(&it->it_mtx, MA_OWNED);
1585 if (itimespecfix(&val.it_value))
1588 if (timespecisset(&val.it_value)) {
1589 if (itimespecfix(&val.it_interval))
1592 timespecclear(&val.it_interval);
1596 realtimer_gettime(it, ovalue);
1599 if (timespecisset(&val.it_value)) {
1600 realtimer_clocktime(it->it_clockid, &cts);
1602 if ((flags & TIMER_ABSTIME) == 0) {
1603 /* Convert to absolute time. */
1604 timespecadd(&it->it_time.it_value, &cts,
1605 &it->it_time.it_value);
1607 timespecsub(&ts, &cts, &ts);
1609 * We don't care if ts is negative, tztohz will
1613 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1614 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1617 callout_stop(&it->it_callout);
1624 realtimer_clocktime(clockid_t id, struct timespec *ts)
1626 if (id == CLOCK_REALTIME)
1628 else /* CLOCK_MONOTONIC */
1633 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1637 PROC_LOCK_ASSERT(p, MA_OWNED);
1638 it = itimer_find(p, timerid);
1640 ksi->ksi_overrun = it->it_overrun;
1641 it->it_overrun_last = it->it_overrun;
1650 itimespecfix(struct timespec *ts)
1653 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1655 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1656 ts->tv_nsec = tick * 1000;
1660 #define timespectons(tsp) \
1661 ((uint64_t)(tsp)->tv_sec * 1000000000 + (tsp)->tv_nsec)
1662 #define timespecfromns(ns) (struct timespec){ \
1663 .tv_sec = (ns) / 1000000000, \
1664 .tv_nsec = (ns) % 1000000000 \
1668 realtimer_expire_l(struct itimer *it, bool proc_locked)
1670 struct timespec cts, ts;
1673 uint64_t interval, now, overruns, value;
1675 realtimer_clocktime(it->it_clockid, &cts);
1676 /* Only fire if time is reached. */
1677 if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1678 if (timespecisset(&it->it_time.it_interval)) {
1679 timespecadd(&it->it_time.it_value,
1680 &it->it_time.it_interval,
1681 &it->it_time.it_value);
1683 interval = timespectons(&it->it_time.it_interval);
1684 value = timespectons(&it->it_time.it_value);
1685 now = timespectons(&cts);
1689 * We missed at least one period.
1691 overruns = howmany(now - value + 1, interval);
1692 if (it->it_overrun + overruns >=
1694 it->it_overrun + overruns <= INT_MAX) {
1695 it->it_overrun += (int)overruns;
1697 it->it_overrun = INT_MAX;
1698 it->it_ksi.ksi_errno = ERANGE;
1701 now + interval - (now - value) % interval;
1702 it->it_time.it_value = timespecfromns(value);
1705 /* single shot timer ? */
1706 timespecclear(&it->it_time.it_value);
1710 if (timespecisset(&it->it_time.it_value)) {
1711 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1712 it->it_flags |= ITF_PSTOPPED;
1714 timespecsub(&it->it_time.it_value, &cts, &ts);
1715 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1716 callout_reset(&it->it_callout, tvtohz(&tv),
1717 realtimer_expire, it);
1730 } else if (timespecisset(&it->it_time.it_value)) {
1732 if (P_SHOULDSTOP(p) || P_KILLED(p)) {
1733 it->it_flags |= ITF_PSTOPPED;
1735 ts = it->it_time.it_value;
1736 timespecsub(&ts, &cts, &ts);
1737 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1738 callout_reset(&it->it_callout, tvtohz(&tv),
1739 realtimer_expire, it);
1744 /* Timeout callback for realtime timer */
1746 realtimer_expire(void *arg)
1748 realtimer_expire_l(arg, false);
1752 itimer_fire(struct itimer *it)
1754 struct proc *p = it->it_proc;
1757 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1758 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1759 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1761 timespecclear(&it->it_time.it_value);
1762 timespecclear(&it->it_time.it_interval);
1763 callout_stop(&it->it_callout);
1767 if (!KSI_ONQ(&it->it_ksi)) {
1768 it->it_ksi.ksi_errno = 0;
1769 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1770 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1772 if (it->it_overrun < INT_MAX)
1775 it->it_ksi.ksi_errno = ERANGE;
1782 itimers_alloc(struct proc *p)
1784 struct itimers *its;
1787 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1788 LIST_INIT(&its->its_virtual);
1789 LIST_INIT(&its->its_prof);
1790 TAILQ_INIT(&its->its_worklist);
1791 for (i = 0; i < TIMER_MAX; i++)
1792 its->its_timers[i] = NULL;
1794 if (p->p_itimers == NULL) {
1800 free(its, M_SUBPROC);
1804 /* Clean up timers when some process events are being triggered. */
1806 itimers_event_exit_exec(int start_idx, struct proc *p)
1808 struct itimers *its;
1816 for (i = start_idx; i < TIMER_MAX; ++i) {
1817 if ((it = its->its_timers[i]) != NULL)
1818 kern_ktimer_delete(curthread, i);
1820 if (its->its_timers[0] == NULL && its->its_timers[1] == NULL &&
1821 its->its_timers[2] == NULL) {
1822 free(its, M_SUBPROC);
1823 p->p_itimers = NULL;
1828 itimers_exec(struct proc *p)
1831 * According to susv3, XSI interval timers should be inherited
1834 itimers_event_exit_exec(3, p);
1838 itimers_exit(struct proc *p)
1840 itimers_event_exit_exec(0, p);