2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "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/eventhandler.h>
45 #include <sys/resourcevar.h>
46 #include <sys/signalvar.h>
47 #include <sys/kernel.h>
48 #include <sys/sleepqueue.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysctl.h>
51 #include <sys/sysent.h>
54 #include <sys/posix4.h>
56 #include <sys/timers.h>
57 #include <sys/timetc.h>
58 #include <sys/vnode.h>
60 #include <sys/ktrace.h>
64 #include <vm/vm_extern.h>
66 #define MAX_CLOCKS (CLOCK_MONOTONIC+1)
67 #define CPUCLOCK_BIT 0x80000000
68 #define CPUCLOCK_PROCESS_BIT 0x40000000
69 #define CPUCLOCK_ID_MASK (~(CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT))
70 #define MAKE_THREAD_CPUCLOCK(tid) (CPUCLOCK_BIT|(tid))
71 #define MAKE_PROCESS_CPUCLOCK(pid) \
72 (CPUCLOCK_BIT|CPUCLOCK_PROCESS_BIT|(pid))
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 void itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp);
101 static void itimers_event_hook_exit(void *arg, struct proc *p);
102 static int realtimer_create(struct itimer *);
103 static int realtimer_gettime(struct itimer *, struct itimerspec *);
104 static int realtimer_settime(struct itimer *, int,
105 struct itimerspec *, struct itimerspec *);
106 static int realtimer_delete(struct itimer *);
107 static void realtimer_clocktime(clockid_t, struct timespec *);
108 static void realtimer_expire(void *);
110 int register_posix_clock(int, struct kclock *);
111 void itimer_fire(struct itimer *it);
112 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);
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)
247 runtime = cputick2usec(runtime);
248 ats->tv_sec = runtime / 1000000;
249 ats->tv_nsec = runtime % 1000000 * 1000;
253 get_thread_cputime(struct thread *targettd, struct timespec *ats)
255 uint64_t runtime, curtime, switchtime;
257 if (targettd == NULL) { /* current thread */
259 switchtime = PCPU_GET(switchtime);
260 curtime = cpu_ticks();
261 runtime = curthread->td_runtime;
263 runtime += curtime - switchtime;
265 thread_lock(targettd);
266 runtime = targettd->td_runtime;
267 thread_unlock(targettd);
269 cputick2timespec(runtime, ats);
273 get_process_cputime(struct proc *targetp, struct timespec *ats)
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 get_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 get_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 get_thread_cputime(NULL, ats);
364 case CLOCK_PROCESS_CPUTIME_ID:
366 get_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));
396 kern_clock_settime(struct thread *td, clockid_t clock_id, struct timespec *ats)
401 if ((error = priv_check(td, PRIV_CLOCK_SETTIME)) != 0)
403 if (clock_id != CLOCK_REALTIME)
405 if (ats->tv_nsec < 0 || ats->tv_nsec >= 1000000000 ||
408 /* XXX Don't convert nsec->usec and back */
409 TIMESPEC_TO_TIMEVAL(&atv, ats);
410 error = settime(td, &atv);
414 #ifndef _SYS_SYSPROTO_H_
415 struct clock_getres_args {
421 sys_clock_getres(struct thread *td, struct clock_getres_args *uap)
429 error = kern_clock_getres(td, uap->clock_id, &ts);
431 error = copyout(&ts, uap->tp, sizeof(ts));
436 kern_clock_getres(struct thread *td, clockid_t clock_id, struct timespec *ts)
442 case CLOCK_REALTIME_FAST:
443 case CLOCK_REALTIME_PRECISE:
444 case CLOCK_MONOTONIC:
445 case CLOCK_MONOTONIC_FAST:
446 case CLOCK_MONOTONIC_PRECISE:
448 case CLOCK_UPTIME_FAST:
449 case CLOCK_UPTIME_PRECISE:
451 * Round up the result of the division cheaply by adding 1.
452 * Rounding up is especially important if rounding down
453 * would give 0. Perfect rounding is unimportant.
455 ts->tv_nsec = 1000000000 / tc_getfrequency() + 1;
459 /* Accurately round up here because we can do so cheaply. */
460 ts->tv_nsec = howmany(1000000000, hz);
466 case CLOCK_THREAD_CPUTIME_ID:
467 case CLOCK_PROCESS_CPUTIME_ID:
469 /* sync with cputick2usec */
470 ts->tv_nsec = 1000000 / cpu_tickrate();
471 if (ts->tv_nsec == 0)
475 if ((int)clock_id < 0)
483 kern_nanosleep(struct thread *td, struct timespec *rqt, struct timespec *rmt)
486 return (kern_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME, rqt,
490 static uint8_t nanowait[MAXCPU];
493 kern_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
494 const struct timespec *rqt, struct timespec *rmt)
496 struct timespec ts, now;
497 sbintime_t sbt, sbtt, prec, tmp;
502 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
504 if ((flags & ~TIMER_ABSTIME) != 0)
508 case CLOCK_REALTIME_PRECISE:
509 case CLOCK_REALTIME_FAST:
511 is_abs_real = (flags & TIMER_ABSTIME) != 0;
513 case CLOCK_MONOTONIC:
514 case CLOCK_MONOTONIC_PRECISE:
515 case CLOCK_MONOTONIC_FAST:
517 case CLOCK_UPTIME_PRECISE:
518 case CLOCK_UPTIME_FAST:
523 case CLOCK_PROCESS_CPUTIME_ID:
525 case CLOCK_THREAD_CPUTIME_ID:
531 if ((flags & TIMER_ABSTIME) != 0) {
534 atomic_load_acq_int(&rtc_generation);
535 error = kern_clock_gettime(td, clock_id, &now);
536 KASSERT(error == 0, ("kern_clock_gettime: %d", error));
537 timespecsub(&ts, &now);
539 if (ts.tv_sec < 0 || (ts.tv_sec == 0 && ts.tv_nsec == 0)) {
543 if (ts.tv_sec > INT32_MAX / 2) {
544 over = ts.tv_sec - INT32_MAX / 2;
551 if (TIMESEL(&sbt, tmp))
554 error = tsleep_sbt(&nanowait[curcpu], PWAIT | PCATCH, "nanslp",
555 sbt, prec, C_ABSOLUTE);
556 } while (error == 0 && is_abs_real && td->td_rtcgen == 0);
558 if (error != EWOULDBLOCK) {
559 if (TIMESEL(&sbtt, tmp))
563 if (error == ERESTART)
565 if ((flags & TIMER_ABSTIME) == 0 && rmt != NULL) {
566 ts = sbttots(sbt - sbtt);
577 #ifndef _SYS_SYSPROTO_H_
578 struct nanosleep_args {
579 struct timespec *rqtp;
580 struct timespec *rmtp;
585 sys_nanosleep(struct thread *td, struct nanosleep_args *uap)
588 return (user_clock_nanosleep(td, CLOCK_REALTIME, TIMER_RELTIME,
589 uap->rqtp, uap->rmtp));
592 #ifndef _SYS_SYSPROTO_H_
593 struct clock_nanosleep_args {
596 struct timespec *rqtp;
597 struct timespec *rmtp;
602 sys_clock_nanosleep(struct thread *td, struct clock_nanosleep_args *uap)
606 error = user_clock_nanosleep(td, uap->clock_id, uap->flags, uap->rqtp,
608 return (kern_posix_error(td, error));
612 user_clock_nanosleep(struct thread *td, clockid_t clock_id, int flags,
613 const struct timespec *ua_rqtp, struct timespec *ua_rmtp)
615 struct timespec rmt, rqt;
618 error = copyin(ua_rqtp, &rqt, sizeof(rqt));
621 if (ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0 &&
622 !useracc(ua_rmtp, sizeof(rmt), VM_PROT_WRITE))
624 error = kern_clock_nanosleep(td, clock_id, flags, &rqt, &rmt);
625 if (error == EINTR && ua_rmtp != NULL && (flags & TIMER_ABSTIME) == 0) {
628 error2 = copyout(&rmt, ua_rmtp, sizeof(rmt));
635 #ifndef _SYS_SYSPROTO_H_
636 struct gettimeofday_args {
638 struct timezone *tzp;
643 sys_gettimeofday(struct thread *td, struct gettimeofday_args *uap)
651 error = copyout(&atv, uap->tp, sizeof (atv));
653 if (error == 0 && uap->tzp != NULL) {
654 rtz.tz_minuteswest = tz_minuteswest;
655 rtz.tz_dsttime = tz_dsttime;
656 error = copyout(&rtz, uap->tzp, sizeof (rtz));
661 #ifndef _SYS_SYSPROTO_H_
662 struct settimeofday_args {
664 struct timezone *tzp;
669 sys_settimeofday(struct thread *td, struct settimeofday_args *uap)
671 struct timeval atv, *tvp;
672 struct timezone atz, *tzp;
676 error = copyin(uap->tv, &atv, sizeof(atv));
683 error = copyin(uap->tzp, &atz, sizeof(atz));
689 return (kern_settimeofday(td, tvp, tzp));
693 kern_settimeofday(struct thread *td, struct timeval *tv, struct timezone *tzp)
697 error = priv_check(td, PRIV_SETTIMEOFDAY);
700 /* Verify all parameters before changing time. */
702 if (tv->tv_usec < 0 || tv->tv_usec >= 1000000 ||
705 error = settime(td, tv);
707 if (tzp && error == 0) {
708 tz_minuteswest = tzp->tz_minuteswest;
709 tz_dsttime = tzp->tz_dsttime;
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 struct kclock rt_clock = {
1087 .timer_create = realtimer_create,
1088 .timer_delete = realtimer_delete,
1089 .timer_settime = realtimer_settime,
1090 .timer_gettime = realtimer_gettime,
1094 itimer_zone = uma_zcreate("itimer", sizeof(struct itimer),
1095 NULL, NULL, itimer_init, itimer_fini, UMA_ALIGN_PTR, 0);
1096 register_posix_clock(CLOCK_REALTIME, &rt_clock);
1097 register_posix_clock(CLOCK_MONOTONIC, &rt_clock);
1098 p31b_setcfg(CTL_P1003_1B_TIMERS, 200112L);
1099 p31b_setcfg(CTL_P1003_1B_DELAYTIMER_MAX, INT_MAX);
1100 p31b_setcfg(CTL_P1003_1B_TIMER_MAX, TIMER_MAX);
1101 EVENTHANDLER_REGISTER(process_exit, itimers_event_hook_exit,
1102 (void *)ITIMER_EV_EXIT, EVENTHANDLER_PRI_ANY);
1103 EVENTHANDLER_REGISTER(process_exec, itimers_event_hook_exec,
1104 (void *)ITIMER_EV_EXEC, EVENTHANDLER_PRI_ANY);
1108 register_posix_clock(int clockid, struct kclock *clk)
1110 if ((unsigned)clockid >= MAX_CLOCKS) {
1111 printf("%s: invalid clockid\n", __func__);
1114 posix_clocks[clockid] = *clk;
1119 itimer_init(void *mem, int size, int flags)
1123 it = (struct itimer *)mem;
1124 mtx_init(&it->it_mtx, "itimer lock", NULL, MTX_DEF);
1129 itimer_fini(void *mem, int size)
1133 it = (struct itimer *)mem;
1134 mtx_destroy(&it->it_mtx);
1138 itimer_enter(struct itimer *it)
1141 mtx_assert(&it->it_mtx, MA_OWNED);
1146 itimer_leave(struct itimer *it)
1149 mtx_assert(&it->it_mtx, MA_OWNED);
1150 KASSERT(it->it_usecount > 0, ("invalid it_usecount"));
1152 if (--it->it_usecount == 0 && (it->it_flags & ITF_WANTED) != 0)
1156 #ifndef _SYS_SYSPROTO_H_
1157 struct ktimer_create_args {
1159 struct sigevent * evp;
1164 sys_ktimer_create(struct thread *td, struct ktimer_create_args *uap)
1166 struct sigevent *evp, ev;
1170 if (uap->evp == NULL) {
1173 error = copyin(uap->evp, &ev, sizeof(ev));
1178 error = kern_ktimer_create(td, uap->clock_id, evp, &id, -1);
1180 error = copyout(&id, uap->timerid, sizeof(int));
1182 kern_ktimer_delete(td, id);
1188 kern_ktimer_create(struct thread *td, clockid_t clock_id, struct sigevent *evp,
1189 int *timerid, int preset_id)
1191 struct proc *p = td->td_proc;
1196 if (clock_id < 0 || clock_id >= MAX_CLOCKS)
1199 if (posix_clocks[clock_id].timer_create == NULL)
1203 if (evp->sigev_notify != SIGEV_NONE &&
1204 evp->sigev_notify != SIGEV_SIGNAL &&
1205 evp->sigev_notify != SIGEV_THREAD_ID)
1207 if ((evp->sigev_notify == SIGEV_SIGNAL ||
1208 evp->sigev_notify == SIGEV_THREAD_ID) &&
1209 !_SIG_VALID(evp->sigev_signo))
1213 if (p->p_itimers == NULL)
1216 it = uma_zalloc(itimer_zone, M_WAITOK);
1218 it->it_usecount = 0;
1220 timespecclear(&it->it_time.it_value);
1221 timespecclear(&it->it_time.it_interval);
1223 it->it_overrun_last = 0;
1224 it->it_clockid = clock_id;
1225 it->it_timerid = -1;
1227 ksiginfo_init(&it->it_ksi);
1228 it->it_ksi.ksi_flags |= KSI_INS | KSI_EXT;
1229 error = CLOCK_CALL(clock_id, timer_create, (it));
1234 if (preset_id != -1) {
1235 KASSERT(preset_id >= 0 && preset_id < 3, ("invalid preset_id"));
1237 if (p->p_itimers->its_timers[id] != NULL) {
1244 * Find a free timer slot, skipping those reserved
1247 for (id = 3; id < TIMER_MAX; id++)
1248 if (p->p_itimers->its_timers[id] == NULL)
1250 if (id == TIMER_MAX) {
1256 it->it_timerid = id;
1257 p->p_itimers->its_timers[id] = it;
1259 it->it_sigev = *evp;
1261 it->it_sigev.sigev_notify = SIGEV_SIGNAL;
1264 case CLOCK_REALTIME:
1265 it->it_sigev.sigev_signo = SIGALRM;
1268 it->it_sigev.sigev_signo = SIGVTALRM;
1271 it->it_sigev.sigev_signo = SIGPROF;
1274 it->it_sigev.sigev_value.sival_int = id;
1277 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1278 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1279 it->it_ksi.ksi_signo = it->it_sigev.sigev_signo;
1280 it->it_ksi.ksi_code = SI_TIMER;
1281 it->it_ksi.ksi_value = it->it_sigev.sigev_value;
1282 it->it_ksi.ksi_timerid = id;
1290 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1292 uma_zfree(itimer_zone, it);
1296 #ifndef _SYS_SYSPROTO_H_
1297 struct ktimer_delete_args {
1302 sys_ktimer_delete(struct thread *td, struct ktimer_delete_args *uap)
1305 return (kern_ktimer_delete(td, uap->timerid));
1308 static struct itimer *
1309 itimer_find(struct proc *p, int timerid)
1313 PROC_LOCK_ASSERT(p, MA_OWNED);
1314 if ((p->p_itimers == NULL) ||
1315 (timerid < 0) || (timerid >= TIMER_MAX) ||
1316 (it = p->p_itimers->its_timers[timerid]) == NULL) {
1320 if ((it->it_flags & ITF_DELETING) != 0) {
1328 kern_ktimer_delete(struct thread *td, int timerid)
1330 struct proc *p = td->td_proc;
1334 it = itimer_find(p, timerid);
1341 it->it_flags |= ITF_DELETING;
1342 while (it->it_usecount > 0) {
1343 it->it_flags |= ITF_WANTED;
1344 msleep(it, &it->it_mtx, PPAUSE, "itimer", 0);
1346 it->it_flags &= ~ITF_WANTED;
1347 CLOCK_CALL(it->it_clockid, timer_delete, (it));
1351 if (KSI_ONQ(&it->it_ksi))
1352 sigqueue_take(&it->it_ksi);
1353 p->p_itimers->its_timers[timerid] = NULL;
1355 uma_zfree(itimer_zone, it);
1359 #ifndef _SYS_SYSPROTO_H_
1360 struct ktimer_settime_args {
1363 const struct itimerspec * value;
1364 struct itimerspec * ovalue;
1368 sys_ktimer_settime(struct thread *td, struct ktimer_settime_args *uap)
1370 struct itimerspec val, oval, *ovalp;
1373 error = copyin(uap->value, &val, sizeof(val));
1376 ovalp = uap->ovalue != NULL ? &oval : NULL;
1377 error = kern_ktimer_settime(td, uap->timerid, uap->flags, &val, ovalp);
1378 if (error == 0 && uap->ovalue != NULL)
1379 error = copyout(ovalp, uap->ovalue, sizeof(*ovalp));
1384 kern_ktimer_settime(struct thread *td, int timer_id, int flags,
1385 struct itimerspec *val, struct itimerspec *oval)
1393 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1399 error = CLOCK_CALL(it->it_clockid, timer_settime, (it,
1407 #ifndef _SYS_SYSPROTO_H_
1408 struct ktimer_gettime_args {
1410 struct itimerspec * value;
1414 sys_ktimer_gettime(struct thread *td, struct ktimer_gettime_args *uap)
1416 struct itimerspec val;
1419 error = kern_ktimer_gettime(td, uap->timerid, &val);
1421 error = copyout(&val, uap->value, sizeof(val));
1426 kern_ktimer_gettime(struct thread *td, int timer_id, struct itimerspec *val)
1434 if (timer_id < 3 || (it = itimer_find(p, timer_id)) == NULL) {
1440 error = CLOCK_CALL(it->it_clockid, timer_gettime, (it, val));
1447 #ifndef _SYS_SYSPROTO_H_
1448 struct timer_getoverrun_args {
1453 sys_ktimer_getoverrun(struct thread *td, struct ktimer_getoverrun_args *uap)
1456 return (kern_ktimer_getoverrun(td, uap->timerid));
1460 kern_ktimer_getoverrun(struct thread *td, int timer_id)
1462 struct proc *p = td->td_proc;
1468 (it = itimer_find(p, timer_id)) == NULL) {
1472 td->td_retval[0] = it->it_overrun_last;
1481 realtimer_create(struct itimer *it)
1483 callout_init_mtx(&it->it_callout, &it->it_mtx, 0);
1488 realtimer_delete(struct itimer *it)
1490 mtx_assert(&it->it_mtx, MA_OWNED);
1493 * clear timer's value and interval to tell realtimer_expire
1494 * to not rearm the timer.
1496 timespecclear(&it->it_time.it_value);
1497 timespecclear(&it->it_time.it_interval);
1499 callout_drain(&it->it_callout);
1505 realtimer_gettime(struct itimer *it, struct itimerspec *ovalue)
1507 struct timespec cts;
1509 mtx_assert(&it->it_mtx, MA_OWNED);
1511 realtimer_clocktime(it->it_clockid, &cts);
1512 *ovalue = it->it_time;
1513 if (ovalue->it_value.tv_sec != 0 || ovalue->it_value.tv_nsec != 0) {
1514 timespecsub(&ovalue->it_value, &cts);
1515 if (ovalue->it_value.tv_sec < 0 ||
1516 (ovalue->it_value.tv_sec == 0 &&
1517 ovalue->it_value.tv_nsec == 0)) {
1518 ovalue->it_value.tv_sec = 0;
1519 ovalue->it_value.tv_nsec = 1;
1526 realtimer_settime(struct itimer *it, int flags,
1527 struct itimerspec *value, struct itimerspec *ovalue)
1529 struct timespec cts, ts;
1531 struct itimerspec val;
1533 mtx_assert(&it->it_mtx, MA_OWNED);
1536 if (itimespecfix(&val.it_value))
1539 if (timespecisset(&val.it_value)) {
1540 if (itimespecfix(&val.it_interval))
1543 timespecclear(&val.it_interval);
1547 realtimer_gettime(it, ovalue);
1550 if (timespecisset(&val.it_value)) {
1551 realtimer_clocktime(it->it_clockid, &cts);
1553 if ((flags & TIMER_ABSTIME) == 0) {
1554 /* Convert to absolute time. */
1555 timespecadd(&it->it_time.it_value, &cts);
1557 timespecsub(&ts, &cts);
1559 * We don't care if ts is negative, tztohz will
1563 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1564 callout_reset(&it->it_callout, tvtohz(&tv),
1565 realtimer_expire, it);
1567 callout_stop(&it->it_callout);
1574 realtimer_clocktime(clockid_t id, struct timespec *ts)
1576 if (id == CLOCK_REALTIME)
1578 else /* CLOCK_MONOTONIC */
1583 itimer_accept(struct proc *p, int timerid, ksiginfo_t *ksi)
1587 PROC_LOCK_ASSERT(p, MA_OWNED);
1588 it = itimer_find(p, timerid);
1590 ksi->ksi_overrun = it->it_overrun;
1591 it->it_overrun_last = it->it_overrun;
1600 itimespecfix(struct timespec *ts)
1603 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1605 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000)
1606 ts->tv_nsec = tick * 1000;
1610 /* Timeout callback for realtime timer */
1612 realtimer_expire(void *arg)
1614 struct timespec cts, ts;
1618 it = (struct itimer *)arg;
1620 realtimer_clocktime(it->it_clockid, &cts);
1621 /* Only fire if time is reached. */
1622 if (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1623 if (timespecisset(&it->it_time.it_interval)) {
1624 timespecadd(&it->it_time.it_value,
1625 &it->it_time.it_interval);
1626 while (timespeccmp(&cts, &it->it_time.it_value, >=)) {
1627 if (it->it_overrun < INT_MAX)
1630 it->it_ksi.ksi_errno = ERANGE;
1631 timespecadd(&it->it_time.it_value,
1632 &it->it_time.it_interval);
1635 /* single shot timer ? */
1636 timespecclear(&it->it_time.it_value);
1638 if (timespecisset(&it->it_time.it_value)) {
1639 ts = it->it_time.it_value;
1640 timespecsub(&ts, &cts);
1641 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1642 callout_reset(&it->it_callout, tvtohz(&tv),
1643 realtimer_expire, it);
1650 } else if (timespecisset(&it->it_time.it_value)) {
1651 ts = it->it_time.it_value;
1652 timespecsub(&ts, &cts);
1653 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1654 callout_reset(&it->it_callout, tvtohz(&tv), realtimer_expire,
1660 itimer_fire(struct itimer *it)
1662 struct proc *p = it->it_proc;
1665 if (it->it_sigev.sigev_notify == SIGEV_SIGNAL ||
1666 it->it_sigev.sigev_notify == SIGEV_THREAD_ID) {
1667 if (sigev_findtd(p, &it->it_sigev, &td) != 0) {
1669 timespecclear(&it->it_time.it_value);
1670 timespecclear(&it->it_time.it_interval);
1671 callout_stop(&it->it_callout);
1675 if (!KSI_ONQ(&it->it_ksi)) {
1676 it->it_ksi.ksi_errno = 0;
1677 ksiginfo_set_sigev(&it->it_ksi, &it->it_sigev);
1678 tdsendsignal(p, td, it->it_ksi.ksi_signo, &it->it_ksi);
1680 if (it->it_overrun < INT_MAX)
1683 it->it_ksi.ksi_errno = ERANGE;
1690 itimers_alloc(struct proc *p)
1692 struct itimers *its;
1695 its = malloc(sizeof (struct itimers), M_SUBPROC, M_WAITOK | M_ZERO);
1696 LIST_INIT(&its->its_virtual);
1697 LIST_INIT(&its->its_prof);
1698 TAILQ_INIT(&its->its_worklist);
1699 for (i = 0; i < TIMER_MAX; i++)
1700 its->its_timers[i] = NULL;
1702 if (p->p_itimers == NULL) {
1708 free(its, M_SUBPROC);
1713 itimers_event_hook_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
1715 itimers_event_hook_exit(arg, p);
1718 /* Clean up timers when some process events are being triggered. */
1720 itimers_event_hook_exit(void *arg, struct proc *p)
1722 struct itimers *its;
1724 int event = (int)(intptr_t)arg;
1727 if (p->p_itimers != NULL) {
1729 for (i = 0; i < MAX_CLOCKS; ++i) {
1730 if (posix_clocks[i].event_hook != NULL)
1731 CLOCK_CALL(i, event_hook, (p, i, event));
1734 * According to susv3, XSI interval timers should be inherited
1737 if (event == ITIMER_EV_EXEC)
1739 else if (event == ITIMER_EV_EXIT)
1742 panic("unhandled event");
1743 for (; i < TIMER_MAX; ++i) {
1744 if ((it = its->its_timers[i]) != NULL)
1745 kern_ktimer_delete(curthread, i);
1747 if (its->its_timers[0] == NULL &&
1748 its->its_timers[1] == NULL &&
1749 its->its_timers[2] == NULL) {
1750 free(its, M_SUBPROC);
1751 p->p_itimers = NULL;