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 *);
109 static int register_posix_clock(int, const struct kclock *);
110 void itimer_fire(struct itimer *it);
111 int itimespecfix(struct timespec *ts);
113 #define CLOCK_CALL(clock, call, arglist) \
114 ((*posix_clocks[clock].call) arglist)
116 SYSINIT(posix_timer, SI_SUB_P1003_1B, SI_ORDER_FIRST+4, itimer_start, NULL);
119 settime(struct thread *td, struct timeval *tv)
121 struct timeval delta, tv1, tv2;
122 static struct timeval maxtime, laststep;
127 timevalsub(&delta, &tv1);
130 * If the system is secure, we do not allow the time to be
131 * set to a value earlier than 1 second less than the highest
132 * time we have yet seen. The worst a miscreant can do in
133 * this circumstance is "freeze" time. He couldn't go
136 * We similarly do not allow the clock to be stepped more
137 * than one second, nor more than once per second. This allows
138 * a miscreant to make the clock march double-time, but no worse.
140 if (securelevel_gt(td->td_ucred, 1) != 0) {
141 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
143 * Update maxtime to latest time we've seen.
145 if (tv1.tv_sec > maxtime.tv_sec)
148 timevalsub(&tv2, &maxtime);
149 if (tv2.tv_sec < -1) {
150 tv->tv_sec = maxtime.tv_sec - 1;
151 printf("Time adjustment clamped to -1 second\n");
154 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;
171 #ifndef _SYS_SYSPROTO_H_
172 struct clock_getcpuclockid2_args {
180 sys_clock_getcpuclockid2(struct thread *td, struct clock_getcpuclockid2_args *uap)
185 error = kern_clock_getcpuclockid2(td, uap->id, uap->which, &clk_id);
187 error = copyout(&clk_id, uap->clock_id, sizeof(clockid_t));
192 kern_clock_getcpuclockid2(struct thread *td, id_t id, int which,
201 case CPUCLOCK_WHICH_PID:
203 error = pget(id, PGET_CANSEE | PGET_NOTID, &p);
209 pid = td->td_proc->p_pid;
211 *clk_id = MAKE_PROCESS_CPUCLOCK(pid);
213 case CPUCLOCK_WHICH_TID:
214 tid = id == 0 ? td->td_tid : id;
215 *clk_id = MAKE_THREAD_CPUCLOCK(tid);
222 #ifndef _SYS_SYSPROTO_H_
223 struct clock_gettime_args {
230 sys_clock_gettime(struct thread *td, struct clock_gettime_args *uap)
235 error = kern_clock_gettime(td, uap->clock_id, &ats);
237 error = copyout(&ats, uap->tp, sizeof(ats));
243 cputick2timespec(uint64_t runtime, struct timespec *ats)
245 runtime = cputick2usec(runtime);
246 ats->tv_sec = runtime / 1000000;
247 ats->tv_nsec = runtime % 1000000 * 1000;
251 kern_thread_cputime(struct thread *targettd, struct timespec *ats)
253 uint64_t runtime, curtime, switchtime;
255 if (targettd == NULL) { /* current thread */
257 switchtime = PCPU_GET(switchtime);
258 curtime = cpu_ticks();
259 runtime = curthread->td_runtime;
261 runtime += curtime - switchtime;
263 PROC_LOCK_ASSERT(targettd->td_proc, MA_OWNED);
264 thread_lock(targettd);
265 runtime = targettd->td_runtime;
266 thread_unlock(targettd);
268 cputick2timespec(runtime, ats);
272 kern_process_cputime(struct proc *targetp, struct timespec *ats)
277 PROC_LOCK_ASSERT(targetp, MA_OWNED);
278 PROC_STATLOCK(targetp);
279 rufetch(targetp, &ru);
280 runtime = targetp->p_rux.rux_runtime;
281 if (curthread->td_proc == targetp)
282 runtime += cpu_ticks() - PCPU_GET(switchtime);
283 PROC_STATUNLOCK(targetp);
284 cputick2timespec(runtime, ats);
288 get_cputime(struct thread *td, clockid_t clock_id, struct timespec *ats)
297 if ((clock_id & CPUCLOCK_PROCESS_BIT) == 0) {
298 tid = clock_id & CPUCLOCK_ID_MASK;
299 td2 = tdfind(tid, p->p_pid);
302 kern_thread_cputime(td2, ats);
303 PROC_UNLOCK(td2->td_proc);
305 pid = clock_id & CPUCLOCK_ID_MASK;
306 error = pget(pid, PGET_CANSEE, &p2);
309 kern_process_cputime(p2, ats);
316 kern_clock_gettime(struct thread *td, clockid_t clock_id, struct timespec *ats)
318 struct timeval sys, user;
323 case CLOCK_REALTIME: /* Default to precise. */
324 case CLOCK_REALTIME_PRECISE:
327 case CLOCK_REALTIME_FAST:
333 calcru(p, &user, &sys);
336 TIMEVAL_TO_TIMESPEC(&user, ats);
341 calcru(p, &user, &sys);
344 timevaladd(&user, &sys);
345 TIMEVAL_TO_TIMESPEC(&user, ats);
347 case CLOCK_MONOTONIC: /* Default to precise. */
348 case CLOCK_MONOTONIC_PRECISE:
350 case CLOCK_UPTIME_PRECISE:
353 case CLOCK_UPTIME_FAST:
354 case CLOCK_MONOTONIC_FAST:
358 ats->tv_sec = time_second;
361 case CLOCK_THREAD_CPUTIME_ID:
362 kern_thread_cputime(NULL, ats);
364 case CLOCK_PROCESS_CPUTIME_ID:
366 kern_process_cputime(p, ats);
370 if ((int)clock_id >= 0)
372 return (get_cputime(td, clock_id, ats));
377 #ifndef _SYS_SYSPROTO_H_
378 struct clock_settime_args {
380 const struct timespec *tp;
385 sys_clock_settime(struct thread *td, struct clock_settime_args *uap)
390 if ((error = copyin(uap->tp, &ats, sizeof(ats))) != 0)
392 return (kern_clock_settime(td, uap->clock_id, &ats));
395 static int allow_insane_settime = 0;
396 SYSCTL_INT(_debug, OID_AUTO, allow_insane_settime, CTLFLAG_RWTUN,
397 &allow_insane_settime, 0,
398 "do not perform possibly restrictive checks on settime(2) args");
401 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
406 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
408 if (clock_id != CLOCK_REALTIME)
410 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 ||
413 if (!allow_insane_settime &&
414 (ats->tv_sec > 8000ULL * 365 * 24 * 60 * 60 ||
415 ats->tv_sec < utc_offset()))
417 /* XXX Don't convert nsec->usec and back */
418 TIMESPEC_TO_TIMEVAL(&atv, ats);
419 error = settime(td, &atv);
423 #ifndef _SYS_SYSPROTO_H_
424 struct clock_getres_args {
430 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
438 error = kern_clock_getres(td, uap->clock_id, &ts);
440 error = copyout(&ts, uap->tp, sizeof(ts));
445 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
451 case CLOCK_REALTIME_FAST:
452 case CLOCK_REALTIME_PRECISE:
453 case CLOCK_MONOTONIC:
454 case CLOCK_MONOTONIC_FAST:
455 case CLOCK_MONOTONIC_PRECISE:
457 case CLOCK_UPTIME_FAST:
458 case CLOCK_UPTIME_PRECISE:
460 * Round up the result of the division cheaply by adding 1.
461 * Rounding up is especially important if rounding down
462 * would give 0. Perfect rounding is unimportant.
464 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
468 /* Accurately round up here because we can do so cheaply. */
469 ts->tv_nsec = howmany(1000000000, hz);
475 case CLOCK_THREAD_CPUTIME_ID:
476 case CLOCK_PROCESS_CPUTIME_ID:
478 /* sync with cputick2usec */
479 ts->tv_nsec = 1000000 / cpu_tickrate();
480 if (ts->tv_nsec == 0)
484 if ((int)clock_id < 0)
492 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
495 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
499 static uint8_t nanowait[MAXCPU];
502 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
503 const struct timespec *rqt, struct timespec *rmt)
505 struct timespec ts, now;
506 sbintime_t sbt, sbtt, prec, tmp;
511 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
513 if ((flags & ~TIMER_ABSTIME) != 0)
517 case CLOCK_REALTIME_PRECISE:
518 case CLOCK_REALTIME_FAST:
520 is_abs_real = (flags & TIMER_ABSTIME) != 0;
522 case CLOCK_MONOTONIC:
523 case CLOCK_MONOTONIC_PRECISE:
524 case CLOCK_MONOTONIC_FAST:
526 case CLOCK_UPTIME_PRECISE:
527 case CLOCK_UPTIME_FAST:
532 case CLOCK_PROCESS_CPUTIME_ID:
534 case CLOCK_THREAD_CPUTIME_ID:
540 if ((flags & TIMER_ABSTIME) != 0) {
543 atomic_load_acq_int(&rtc_generation);
544 error = kern_clock_gettime(td, clock_id, &now);
545 KASSERT(error == 0, ("kern_clock_gettime: %d", error));
546 timespecsub(&ts, &now, &ts);
548 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
552 if (ts.tv_sec > INT32_MAX / 2) {
553 over = ts.tv_sec - INT32_MAX / 2;
560 if (TIMESEL(&sbt, tmp))
563 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
564 sbt, prec, C_ABSOLUTE);
565 } while (error == 0 && is_abs_real && td->td_rtcgen == 0);
567 if (error != EWOULDBLOCK) {
568 if (TIMESEL(&sbtt, tmp))
572 if (error == ERESTART)
574 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
575 ts = sbttots(sbt - sbtt);
586 #ifndef _SYS_SYSPROTO_H_
587 struct nanosleep_args {
588 struct timespec *rqtp;
589 struct timespec *rmtp;
594 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
597 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
598 uap->rqtp, uap->rmtp));
601 #ifndef _SYS_SYSPROTO_H_
602 struct clock_nanosleep_args {
605 struct timespec *rqtp;
606 struct timespec *rmtp;
611 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
615 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
617 return (kern_posix_error(td, error));
621 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
622 const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
624 struct timespec rmt, rqt;
627 error = copyin(ua_rqtp, &rqt, sizeof(rqt));
630 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
631 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
632 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
639 #ifndef _SYS_SYSPROTO_H_
640 struct gettimeofday_args {
642 struct timezone *tzp;
647 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
655 error = copyout(&atv, uap->tp, sizeof (atv));
657 if (error == 0 && uap->tzp != NULL) {
658 rtz.tz_minuteswest = 0;
660 error = copyout(&rtz, uap->tzp, sizeof (rtz));
665 #ifndef _SYS_SYSPROTO_H_
666 struct settimeofday_args {
668 struct timezone *tzp;
673 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
675 struct timeval atv, *tvp;
676 struct timezone atz, *tzp;
680 error = copyin(uap->tv, &atv, sizeof(atv));
687 error = copyin(uap->tzp, &atz, sizeof(atz));
693 return (kern_settimeofday(td, tvp, tzp));
697 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
701 error = priv_check(td, PRIV_SETTIMEOFDAY);
704 /* Verify all parameters before changing time. */
706 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
709 error = settime(td, tv);
715 * Get value of an interval timer. The process virtual and profiling virtual
716 * time timers are kept in the p_stats area, since they can be swapped out.
717 * These are kept internally in the way they are specified externally: in
718 * time until they expire.
720 * The real time interval timer is kept in the process table slot for the
721 * process, and its value (it_value) is kept as an absolute time rather than
722 * as a delta, so that it is easy to keep periodic real-time signals from
725 * Virtual time timers are processed in the hardclock() routine of
726 * kern_clock.c. The real time timer is processed by a timeout routine,
727 * called from the softclock() routine. Since a callout may be delayed in
728 * real time due to interrupt processing in the system, it is possible for
729 * the real time timeout routine (realitexpire, given below), to be delayed
730 * in real time past when it is supposed to occur. It does not suffice,
731 * therefore, to reload the real timer .it_value from the real time timers
732 * .it_interval. Rather, we compute the next time in absolute time the timer
735 #ifndef _SYS_SYSPROTO_H_
736 struct getitimer_args {
738 struct itimerval *itv;
742 sys_getitimer(struct thread *td, struct getitimer_args *uap)
744 struct itimerval aitv;
747 error = kern_getitimer(td, uap->which, &aitv);
750 return (copyout(&aitv, uap->itv, sizeof (struct itimerval)));
754 kern_getitimer(struct thread *td, u_int which, struct itimerval *aitv)
756 struct proc *p = td->td_proc;
759 if (which > ITIMER_PROF)
762 if (which == ITIMER_REAL) {
764 * Convert from absolute to relative time in .it_value
765 * part of real time timer. If time for real time timer
766 * has passed return 0, else return difference between
767 * current time and time for the timer to go off.
770 *aitv = p->p_realtimer;
772 if (timevalisset(&aitv->it_value)) {
774 if (timevalcmp(&aitv->it_value, &ctv, <))
775 timevalclear(&aitv->it_value);
777 timevalsub(&aitv->it_value, &ctv);
781 *aitv = p->p_stats->p_timer[which];
785 if (KTRPOINT(td, KTR_STRUCT))
791 #ifndef _SYS_SYSPROTO_H_
792 struct setitimer_args {
794 struct itimerval *itv, *oitv;
798 sys_setitimer(struct thread *td, struct setitimer_args *uap)
800 struct itimerval aitv, oitv;
803 if (uap->itv == NULL) {
804 uap->itv = uap->oitv;
805 return (sys_getitimer(td, (struct getitimer_args *)uap));
808 if ((error = copyin(uap->itv, &aitv, sizeof(struct itimerval))))
810 error = kern_setitimer(td, uap->which, &aitv, &oitv);
811 if (error != 0 || uap->oitv == NULL)
813 return (copyout(&oitv, uap->oitv, sizeof(struct itimerval)));
817 kern_setitimer(struct thread *td, u_int which, struct itimerval *aitv,
818 struct itimerval *oitv)
820 struct proc *p = td->td_proc;
825 return (kern_getitimer(td, which, oitv));
827 if (which > ITIMER_PROF)
830 if (KTRPOINT(td, KTR_STRUCT))
833 if (itimerfix(&aitv->it_value) ||
834 aitv->it_value.tv_sec > INT32_MAX / 2)
836 if (!timevalisset(&aitv->it_value))
837 timevalclear(&aitv->it_interval);
838 else if (itimerfix(&aitv->it_interval) ||
839 aitv->it_interval.tv_sec > INT32_MAX / 2)
842 if (which == ITIMER_REAL) {
844 if (timevalisset(&p->p_realtimer.it_value))
845 callout_stop(&p->p_itcallout);
847 if (timevalisset(&aitv->it_value)) {
848 pr = tvtosbt(aitv->it_value) >> tc_precexp;
849 timevaladd(&aitv->it_value, &ctv);
850 sbt = tvtosbt(aitv->it_value);
851 callout_reset_sbt(&p->p_itcallout, sbt, pr,
852 realitexpire, p, C_ABSOLUTE);
854 *oitv = p->p_realtimer;
855 p->p_realtimer = *aitv;
857 if (timevalisset(&oitv->it_value)) {
858 if (timevalcmp(&oitv->it_value, &ctv, <))
859 timevalclear(&oitv->it_value);
861 timevalsub(&oitv->it_value, &ctv);
864 if (aitv->it_interval.tv_sec == 0 &&
865 aitv->it_interval.tv_usec != 0 &&
866 aitv->it_interval.tv_usec < tick)
867 aitv->it_interval.tv_usec = tick;
868 if (aitv->it_value.tv_sec == 0 &&
869 aitv->it_value.tv_usec != 0 &&
870 aitv->it_value.tv_usec < tick)
871 aitv->it_value.tv_usec = tick;
873 *oitv = p->p_stats->p_timer[which];
874 p->p_stats->p_timer[which] = *aitv;
878 if (KTRPOINT(td, KTR_STRUCT))
885 * Real interval timer expired:
886 * send process whose timer expired an alarm signal.
887 * If time is not set up to reload, then just return.
888 * Else compute next time timer should go off which is > current time.
889 * This is where delay in processing this timeout causes multiple
890 * SIGALRM calls to be compressed into one.
891 * tvtohz() always adds 1 to allow for the time until the next clock
892 * interrupt being strictly less than 1 clock tick, but we don't want
893 * that here since we want to appear to be in sync with the clock
894 * interrupt even when we're delayed.
897 realitexpire(void *arg)
903 p = (struct proc *)arg;
904 kern_psignal(p, SIGALRM);
905 if (!timevalisset(&p->p_realtimer.it_interval)) {
906 timevalclear(&p->p_realtimer.it_value);
907 if (p->p_flag & P_WEXIT)
908 wakeup(&p->p_itcallout);
911 isbt = tvtosbt(p->p_realtimer.it_interval);
912 if (isbt >= sbt_timethreshold)
913 getmicrouptime(&ctv);
917 timevaladd(&p->p_realtimer.it_value,
918 &p->p_realtimer.it_interval);
919 } while (timevalcmp(&p->p_realtimer.it_value, &ctv, <=));
920 callout_reset_sbt(&p->p_itcallout, tvtosbt(p->p_realtimer.it_value),
921 isbt >> tc_precexp, realitexpire, p, C_ABSOLUTE);
925 * Check that a proposed value to load into the .it_value or
926 * .it_interval part of an interval timer is acceptable, and
927 * fix it to have at least minimal value (i.e. if it is less
928 * than the resolution of the clock, round it up.)
931 itimerfix(struct timeval *tv)
934 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000)
936 if (tv->tv_sec == 0 && tv->tv_usec != 0 &&
937 tv->tv_usec < (u_int)tick / 16)
938 tv->tv_usec = (u_int)tick / 16;
943 * Decrement an interval timer by a specified number
944 * of microseconds, which must be less than a second,
945 * i.e. < 1000000. If the timer expires, then reload
946 * it. In this case, carry over (usec - old value) to
947 * reduce the value reloaded into the timer so that
948 * the timer does not drift. This routine assumes
949 * that it is called in a context where the timers
950 * on which it is operating cannot change in value.
953 itimerdecr(struct itimerval *itp, int usec)
956 if (itp->it_value.tv_usec < usec) {
957 if (itp->it_value.tv_sec == 0) {
958 /* expired, and already in next interval */
959 usec -= itp->it_value.tv_usec;
962 itp->it_value.tv_usec += 1000000;
963 itp->it_value.tv_sec--;
965 itp->it_value.tv_usec -= usec;
967 if (timevalisset(&itp->it_value))
969 /* expired, exactly at end of interval */
971 if (timevalisset(&itp->it_interval)) {
972 itp->it_value = itp->it_interval;
973 itp->it_value.tv_usec -= usec;
974 if (itp->it_value.tv_usec < 0) {
975 itp->it_value.tv_usec += 1000000;
976 itp->it_value.tv_sec--;
979 itp->it_value.tv_usec = 0; /* sec is already 0 */
984 * Add and subtract routines for timevals.
985 * N.B.: subtract routine doesn't deal with
986 * results which are before the beginning,
987 * it just gets very confused in this case.
991 timevaladd(struct timeval *t1, const struct timeval *t2)
994 t1->tv_sec += t2->tv_sec;
995 t1->tv_usec += t2->tv_usec;
1000 timevalsub(struct timeval *t1, const struct timeval *t2)
1003 t1->tv_sec -= t2->tv_sec;
1004 t1->tv_usec -= t2->tv_usec;
1009 timevalfix(struct timeval *t1)
1012 if (t1->tv_usec < 0) {
1014 t1->tv_usec += 1000000;
1016 if (t1->tv_usec >= 1000000) {
1018 t1->tv_usec -= 1000000;
1023 * ratecheck(): simple time-based rate-limit checking.
1026 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
1028 struct timeval tv, delta;
1031 getmicrouptime(&tv); /* NB: 10ms precision */
1033 timevalsub(&delta, lasttime);
1036 * check for 0,0 is so that the message will be seen at least once,
1037 * even if interval is huge.
1039 if (timevalcmp(&delta, mininterval, >=) ||
1040 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
1049 * ppsratecheck(): packets (or events) per second limitation.
1051 * Return 0 if the limit is to be enforced (e.g. the caller
1052 * should drop a packet because of the rate limitation).
1054 * maxpps of 0 always causes zero to be returned. maxpps of -1
1055 * always causes 1 to be returned; this effectively defeats rate
1058 * Note that we maintain the struct timeval for compatibility
1059 * with other bsd systems. We reuse the storage and just monitor
1060 * clock ticks for minimal overhead.
1063 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
1068 * Reset the last time and counter if this is the first call
1069 * or more than a second has passed since the last update of
1073 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
1074 lasttime->tv_sec = now;
1076 return (maxpps != 0);
1078 (*curpps)++; /* NB: ignore potential overflow */
1079 return (maxpps < 0 || *curpps <= maxpps);
1086 static const struct kclock rt_clock = {
1087 .timer_create = realtimer_create,
1088 .timer_delete = realtimer_delete,
1089 .timer_settime = realtimer_settime,
1090 .timer_gettime = realtimer_gettime,
1093 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1094 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1095 register_posix_clock(CLOCK_REALTIME, &rt_clock);
1096 register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1097 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1098 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1099 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1103 register_posix_clock(int clockid, const struct kclock *clk)
1105 if ((unsigned)clockid >= MAX_CLOCKS) {
1106 printf("%s: invalid clockid\n", __func__);
1109 posix_clocks[clockid] = *clk;
1114 itimer_init(void *mem, int size, int flags)
1118 it = (struct itimer *)mem;
1119 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1124 itimer_fini(void *mem, int size)
1128 it = (struct itimer *)mem;
1129 mtx_destroy(&it->it_mtx);
1133 itimer_enter(struct itimer *it)
1136 mtx_assert(&it->it_mtx, MA_OWNED);
1141 itimer_leave(struct itimer *it)
1144 mtx_assert(&it->it_mtx, MA_OWNED);
1145 KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1147 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1151 #ifndef _SYS_SYSPROTO_H_
1152 struct ktimer_create_args {
1154 struct sigevent * evp;
1159 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1161 struct sigevent *evp, ev;
1165 if (uap->evp == NULL) {
1168 error = copyin(uap->evp, &ev, sizeof(ev));
1173 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1175 error = copyout(&id, uap->timerid, sizeof(int));
1177 kern_ktimer_delete(td, id);
1183 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1184 int *timerid, int preset_id)
1186 struct proc *p = td->td_proc;
1191 if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1194 if (posix_clocks[clock_id].timer_create == NULL)
1198 if (evp->sigev_notify != SIGEV_NONE &&
1199 evp->sigev_notify != SIGEV_SIGNAL &&
1200 evp->sigev_notify != SIGEV_THREAD_ID)
1202 if ((evp->sigev_notify == SIGEV_SIGNAL ||
1203 evp->sigev_notify == SIGEV_THREAD_ID) &&
1204 !_SIG_VALID(evp->sigev_signo))
1208 if (p->p_itimers == NULL)
1211 it = uma_zalloc(itimer_zone, M_WAITOK);
1213 it->it_usecount = 0;
1215 timespecclear(&it->it_time.it_value);
1216 timespecclear(&it->it_time.it_interval);
1218 it->it_overrun_last = 0;
1219 it->it_clockid = clock_id;
1220 it->it_timerid = -1;
1222 ksiginfo_init(&it->it_ksi);
1223 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1224 error = CLOCK_CALL(clock_id, timer_create, (it));
1229 if (preset_id != -1) {
1230 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1232 if (p->p_itimers->its_timers[id] != NULL) {
1239 * Find a free timer slot, skipping those reserved
1242 for (id = 3; id < TIMER_MAX; id++)
1243 if (p->p_itimers->its_timers[id] == NULL)
1245 if (id == TIMER_MAX) {
1251 it->it_timerid = id;
1252 p->p_itimers->its_timers[id] = it;
1254 it->it_sigev = *evp;
1256 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1259 case CLOCK_REALTIME:
1260 it->it_sigev.sigev_signo = SIGALRM;
1263 it->it_sigev.sigev_signo = SIGVTALRM;
1266 it->it_sigev.sigev_signo = SIGPROF;
1269 it->it_sigev.sigev_value.sival_int = id;
1272 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1273 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1274 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1275 it->it_ksi.ksi_code = SI_TIMER;
1276 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1277 it->it_ksi.ksi_timerid = id;
1285 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1287 uma_zfree(itimer_zone, it);
1291 #ifndef _SYS_SYSPROTO_H_
1292 struct ktimer_delete_args {
1297 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1300 return (kern_ktimer_delete(td, uap->timerid));
1303 static struct itimer *
1304 itimer_find(struct proc *p, int timerid)
1308 PROC_LOCK_ASSERT(p, MA_OWNED);
1309 if ((p->p_itimers == NULL) ||
1310 (timerid < 0) || (timerid >= TIMER_MAX) ||
1311 (it = p->p_itimers->its_timers[timerid]) == NULL) {
1315 if ((it->it_flags & ITF_DELETING) != 0) {
1323 kern_ktimer_delete(struct thread *td, int timerid)
1325 struct proc *p = td->td_proc;
1329 it = itimer_find(p, timerid);
1336 it->it_flags |= ITF_DELETING;
1337 while (it->it_usecount > 0) {
1338 it->it_flags |= ITF_WANTED;
1339 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1341 it->it_flags &= ~ITF_WANTED;
1342 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1346 if (KSI_ONQ(&it->it_ksi))
1347 sigqueue_take(&it->it_ksi);
1348 p->p_itimers->its_timers[timerid] = NULL;
1350 uma_zfree(itimer_zone, it);
1354 #ifndef _SYS_SYSPROTO_H_
1355 struct ktimer_settime_args {
1358 const struct itimerspec * value;
1359 struct itimerspec * ovalue;
1363 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1365 struct itimerspec val, oval, *ovalp;
1368 error = copyin(uap->value, &val, sizeof(val));
1371 ovalp = uap->ovalue != NULL ? &oval : NULL;
1372 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1373 if (error == 0 && uap->ovalue != NULL)
1374 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1379 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1380 struct itimerspec *val, struct itimerspec *oval)
1388 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1394 error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct ktimer_gettime_args {
1405 struct itimerspec * value;
1409 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1411 struct itimerspec val;
1414 error = kern_ktimer_gettime(td, uap->timerid, &val);
1416 error = copyout(&val, uap->value, sizeof(val));
1421 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1429 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1435 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1442 #ifndef _SYS_SYSPROTO_H_
1443 struct timer_getoverrun_args {
1448 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1451 return (kern_ktimer_getoverrun(td, uap->timerid));
1455 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1457 struct proc *p = td->td_proc;
1463 (it = itimer_find(p, timer_id)) == NULL) {
1467 td->td_retval[0] = it->it_overrun_last;
1476 realtimer_create(struct itimer *it)
1478 callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1483 realtimer_delete(struct itimer *it)
1485 mtx_assert(&it->it_mtx, MA_OWNED);
1488 * clear timer's value and interval to tell realtimer_expire
1489 * to not rearm the timer.
1491 timespecclear(&it->it_time.it_value);
1492 timespecclear(&it->it_time.it_interval);
1494 callout_drain(&it->it_callout);
1500 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1502 struct timespec cts;
1504 mtx_assert(&it->it_mtx, MA_OWNED);
1506 realtimer_clocktime(it->it_clockid, &cts);
1507 *ovalue = it->it_time;
1508 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1509 timespecsub(&ovalue->it_value, &cts, &ovalue->it_value);
1510 if (ovalue->it_value.tv_sec < 0 ||
1511 (ovalue->it_value.tv_sec == 0 &&
1512 ovalue->it_value.tv_nsec == 0)) {
1513 ovalue->it_value.tv_sec = 0;
1514 ovalue->it_value.tv_nsec = 1;
1521 realtimer_settime(struct itimer *it, int flags,
1522 struct itimerspec *value, struct itimerspec *ovalue)
1524 struct timespec cts, ts;
1526 struct itimerspec val;
1528 mtx_assert(&it->it_mtx, MA_OWNED);
1531 if (itimespecfix(&val.it_value))
1534 if (timespecisset(&val.it_value)) {
1535 if (itimespecfix(&val.it_interval))
1538 timespecclear(&val.it_interval);
1542 realtimer_gettime(it, ovalue);
1545 if (timespecisset(&val.it_value)) {
1546 realtimer_clocktime(it->it_clockid, &cts);
1548 if ((flags & TIMER_ABSTIME) == 0) {
1549 /* Convert to absolute time. */
1550 timespecadd(&it->it_time.it_value, &cts,
1551 &it->it_time.it_value);
1553 timespecsub(&ts, &cts, &ts);
1555 * We don't care if ts is negative, tztohz will
1559 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1560 callout_reset(&it->it_callout, tvtohz(&tv),
1561 realtimer_expire, it);
1563 callout_stop(&it->it_callout);
1570 realtimer_clocktime(clockid_t id, struct timespec *ts)
1572 if (id == CLOCK_REALTIME)
1574 else /* CLOCK_MONOTONIC */
1579 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1583 PROC_LOCK_ASSERT(p, MA_OWNED);
1584 it = itimer_find(p, timerid);
1586 ksi->ksi_overrun = it->it_overrun;
1587 it->it_overrun_last = it->it_overrun;
1596 itimespecfix(struct timespec *ts)
1599 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1601 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1602 ts->tv_nsec = tick * 1000;
1606 /* Timeout callback for realtime timer */
1608 realtimer_expire(void *arg)
1610 struct timespec cts, ts;
1614 it = (struct itimer *)arg;
1616 realtimer_clocktime(it->it_clockid, &cts);
1617 /* Only fire if time is reached. */
1618 if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1619 if (timespecisset(&it->it_time.it_interval)) {
1620 timespecadd(&it->it_time.it_value,
1621 &it->it_time.it_interval,
1622 &it->it_time.it_value);
1623 while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1624 if (it->it_overrun < INT_MAX)
1627 it->it_ksi.ksi_errno = ERANGE;
1628 timespecadd(&it->it_time.it_value,
1629 &it->it_time.it_interval,
1630 &it->it_time.it_value);
1633 /* single shot timer ? */
1634 timespecclear(&it->it_time.it_value);
1636 if (timespecisset(&it->it_time.it_value)) {
1637 timespecsub(&it->it_time.it_value, &cts, &ts);
1638 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1639 callout_reset(&it->it_callout, tvtohz(&tv),
1640 realtimer_expire, it);
1647 } else if (timespecisset(&it->it_time.it_value)) {
1648 ts = it->it_time.it_value;
1649 timespecsub(&ts, &cts, &ts);
1650 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1651 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1657 itimer_fire(struct itimer *it)
1659 struct proc *p = it->it_proc;
1662 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1663 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1664 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1666 timespecclear(&it->it_time.it_value);
1667 timespecclear(&it->it_time.it_interval);
1668 callout_stop(&it->it_callout);
1672 if (!KSI_ONQ(&it->it_ksi)) {
1673 it->it_ksi.ksi_errno = 0;
1674 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1675 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1677 if (it->it_overrun < INT_MAX)
1680 it->it_ksi.ksi_errno = ERANGE;
1687 itimers_alloc(struct proc *p)
1689 struct itimers *its;
1692 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1693 LIST_INIT(&its->its_virtual);
1694 LIST_INIT(&its->its_prof);
1695 TAILQ_INIT(&its->its_worklist);
1696 for (i = 0; i < TIMER_MAX; i++)
1697 its->its_timers[i] = NULL;
1699 if (p->p_itimers == NULL) {
1705 free(its, M_SUBPROC);
1709 /* Clean up timers when some process events are being triggered. */
1711 itimers_event_exit_exec(int start_idx, struct proc *p)
1713 struct itimers *its;
1721 for (i = start_idx; i < TIMER_MAX; ++i) {
1722 if ((it = its->its_timers[i]) != NULL)
1723 kern_ktimer_delete(curthread, i);
1725 if (its->its_timers[0] == NULL && its->its_timers[1] == NULL &&
1726 its->its_timers[2] == NULL) {
1727 free(its, M_SUBPROC);
1728 p->p_itimers = NULL;
1733 itimers_exec(struct proc *p)
1736 * According to susv3, XSI interval timers should be inherited
1739 itimers_event_exit_exec(3, p);
1743 itimers_exit(struct proc *p)
1745 itimers_event_exit_exec(0, p);