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 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)kern_time.c 8.1 (Berkeley) 6/10/93
34 * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $
35 * $DragonFly: src/sys/kern/kern_time.c,v 1.17 2004/09/17 01:29:45 joerg Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
41 #include <sys/sysproto.h>
42 #include <sys/resourcevar.h>
43 #include <sys/signalvar.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/sysent.h>
47 #include <sys/sysunion.h>
50 #include <sys/vnode.h>
51 #include <sys/sysctl.h>
53 #include <vm/vm_extern.h>
54 #include <sys/msgport2.h>
55 #include <sys/thread2.h>
60 * Time of day and interval timer support.
62 * These routines provide the kernel entry points to get and set
63 * the time-of-day and per-process interval timers. Subroutines
64 * here provide support for adding and subtracting timeval structures
65 * and decrementing interval timers, optionally reloading the interval
66 * timers when they expire.
69 static int nanosleep1 (struct timespec *rqt,
70 struct timespec *rmt);
71 static int settime (struct timeval *);
72 static void timevalfix (struct timeval *);
73 static void no_lease_updatetime (int);
75 static int sleep_hard_us = 100;
76 SYSCTL_INT(_kern, OID_AUTO, sleep_hard_us, CTLFLAG_RW, &sleep_hard_us, 0, "")
79 no_lease_updatetime(deltat)
84 void (*lease_updatetime) (int) = no_lease_updatetime;
90 struct timeval delta, tv1, tv2;
91 static struct timeval maxtime, laststep;
97 timevalsub(&delta, &tv1);
100 * If the system is secure, we do not allow the time to be
101 * set to a value earlier than 1 second less than the highest
102 * time we have yet seen. The worst a miscreant can do in
103 * this circumstance is "freeze" time. He couldn't go
106 * We similarly do not allow the clock to be stepped more
107 * than one second, nor more than once per second. This allows
108 * a miscreant to make the clock march double-time, but no worse.
110 if (securelevel > 1) {
111 if (delta.tv_sec < 0 || delta.tv_usec < 0) {
113 * Update maxtime to latest time we've seen.
115 if (tv1.tv_sec > maxtime.tv_sec)
118 timevalsub(&tv2, &maxtime);
119 if (tv2.tv_sec < -1) {
120 tv->tv_sec = maxtime.tv_sec - 1;
121 printf("Time adjustment clamped to -1 second\n");
124 if (tv1.tv_sec == laststep.tv_sec) {
128 if (delta.tv_sec > 1) {
129 tv->tv_sec = tv1.tv_sec + 1;
130 printf("Time adjustment clamped to +1 second\n");
136 ts.tv_sec = tv->tv_sec;
137 ts.tv_nsec = tv->tv_usec * 1000;
139 lease_updatetime(delta.tv_sec);
147 clock_gettime(struct clock_gettime_args *uap)
151 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
154 return (copyout(&ats, SCARG(uap, tp), sizeof(ats)));
159 clock_settime(struct clock_settime_args *uap)
161 struct thread *td = curthread;
166 if ((error = suser(td)) != 0)
168 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
170 if ((error = copyin(SCARG(uap, tp), &ats, sizeof(ats))) != 0)
172 if (ats.tv_nsec < 0 || ats.tv_nsec >= 1000000000)
174 /* XXX Don't convert nsec->usec and back */
175 TIMESPEC_TO_TIMEVAL(&atv, &ats);
176 if ((error = settime(&atv)))
182 clock_getres(struct clock_getres_args *uap)
187 if (SCARG(uap, clock_id) != CLOCK_REALTIME)
190 if (SCARG(uap, tp)) {
193 * Round up the result of the division cheaply by adding 1.
194 * Rounding up is especially important if rounding down
195 * would give 0. Perfect rounding is unimportant.
197 ts.tv_nsec = 1000000000 / cputimer_freq + 1;
198 error = copyout(&ts, SCARG(uap, tp), sizeof(ts));
206 * This is a general helper function for nanosleep() (aka sleep() aka
209 * If there is less then one tick's worth of time left and
210 * we haven't done a yield, or the remaining microseconds is
211 * ridiculously low, do a yield. This avoids having
212 * to deal with systimer overheads when the system is under
213 * heavy loads. If we have done a yield already then use
214 * a systimer and an uninterruptable thread wait.
216 * If there is more then a tick's worth of time left,
217 * calculate the baseline ticks and use an interruptable
218 * tsleep, then handle the fine-grained delay on the next
219 * loop. This usually results in two sleeps occuring, a long one
223 ns1_systimer(systimer_t info)
225 lwkt_schedule(info->data);
229 nanosleep1(struct timespec *rqt, struct timespec *rmt)
232 struct timespec ts, ts2, ts3;
237 if (rqt->tv_nsec < 0 || rqt->tv_nsec >= 1000000000)
239 if (rqt->tv_sec < 0 || (rqt->tv_sec == 0 && rqt->tv_nsec == 0))
242 timespecadd(&ts, rqt); /* ts = target timestamp compare */
243 TIMESPEC_TO_TIMEVAL(&tv, rqt); /* tv = sleep interval */
248 struct systimer info;
250 ticks = tv.tv_usec / tick; /* approximate */
252 if (tv.tv_sec == 0 && ticks == 0) {
253 thread_t td = curthread;
254 if (tried_yield || tv.tv_usec < sleep_hard_us) {
258 crit_enter_quick(td);
259 systimer_init_oneshot(&info, ns1_systimer,
261 lwkt_deschedule_self(td);
264 systimer_del(&info); /* make sure it's gone */
266 error = iscaught(td->td_proc);
267 } else if (tv.tv_sec == 0) {
268 error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
270 ticks = tvtohz_low(&tv); /* also handles overflow */
271 error = tsleep(&nanowait, PCATCH, "nanslp", ticks);
274 if (error && error != EWOULDBLOCK) {
275 if (error == ERESTART)
278 timespecsub(&ts, &ts2);
285 if (timespeccmp(&ts2, &ts, >=))
288 timespecsub(&ts3, &ts2);
289 TIMESPEC_TO_TIMEVAL(&tv, &ts3);
293 static void nanosleep_done(void *arg);
294 static void nanosleep_copyout(union sysunion *sysun);
298 nanosleep(struct nanosleep_args *uap)
301 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
303 error = copyin(uap->rqtp, &smsleep->rqt, sizeof(smsleep->rqt));
307 * YYY clean this up to always use the callout, note that an abort
308 * implementation should record the residual in the async case.
310 if (uap->sysmsg.lmsg.ms_flags & MSGF_ASYNC) {
313 ticks = (quad_t)smsleep->rqt.tv_nsec * hz / 1000000000LL;
314 if (smsleep->rqt.tv_sec)
315 ticks += (quad_t)smsleep->rqt.tv_sec * hz;
322 uap->sysmsg.copyout = nanosleep_copyout;
323 uap->sysmsg.lmsg.ms_flags &= ~MSGF_DONE;
324 callout_init(&smsleep->timer);
325 callout_reset(&smsleep->timer, ticks, nanosleep_done, uap);
330 * Old synchronous sleep code, copyout the residual if
331 * nanosleep was interrupted.
333 error = nanosleep1(&smsleep->rqt, &smsleep->rmt);
334 if (error && SCARG(uap, rmtp))
335 error = copyout(&smsleep->rmt, SCARG(uap, rmtp), sizeof(smsleep->rmt));
341 * Asynch completion for the nanosleep() syscall. This function may be
342 * called from any context and cannot legally access the originating
343 * thread, proc, or its user space.
345 * YYY change the callout interface API so we can simply assign the replymsg
346 * function to it directly.
349 nanosleep_done(void *arg)
351 struct nanosleep_args *uap = arg;
352 lwkt_msg_t msg = &uap->sysmsg.lmsg;
354 lwkt_replymsg(msg, 0);
358 * Asynch return for the nanosleep() syscall, called in the context of the
359 * originating thread when it pulls the message off the reply port. This
360 * function is responsible for any copyouts to userland. Kernel threads
361 * which do their own internal system calls will not usually call the return
365 nanosleep_copyout(union sysunion *sysun)
367 struct nanosleep_args *uap = &sysun->nanosleep;
368 struct sysmsg_sleep *smsleep = &uap->sysmsg.sm.sleep;
370 if (sysun->lmsg.ms_error && uap->rmtp) {
371 sysun->lmsg.ms_error =
372 copyout(&smsleep->rmt, uap->rmtp, sizeof(smsleep->rmt));
378 gettimeofday(struct gettimeofday_args *uap)
385 if ((error = copyout((caddr_t)&atv, (caddr_t)uap->tp,
390 error = copyout((caddr_t)&tz, (caddr_t)uap->tzp,
397 settimeofday(struct settimeofday_args *uap)
399 struct thread *td = curthread;
404 if ((error = suser(td)))
406 /* Verify all parameters before changing time. */
408 if ((error = copyin((caddr_t)uap->tv, (caddr_t)&atv,
411 if (atv.tv_usec < 0 || atv.tv_usec >= 1000000)
415 (error = copyin((caddr_t)uap->tzp, (caddr_t)&atz, sizeof(atz))))
417 if (uap->tv && (error = settime(&atv)))
424 int tickdelta; /* current clock skew, us. per tick */
425 long timedelta; /* unapplied time correction, us. */
426 static long bigadj = 1000000; /* use 10x skew above bigadj us. */
430 adjtime(struct adjtime_args *uap)
432 struct thread *td = curthread;
434 long ndelta, ntickdelta, odelta;
437 if ((error = suser(td)))
440 copyin((caddr_t)uap->delta, (caddr_t)&atv, sizeof(struct timeval))))
444 * Compute the total correction and the rate at which to apply it.
445 * Round the adjustment down to a whole multiple of the per-tick
446 * delta, so that after some number of incremental changes in
447 * hardclock(), tickdelta will become zero, lest the correction
448 * overshoot and start taking us away from the desired final time.
450 ndelta = atv.tv_sec * 1000000 + atv.tv_usec;
451 if (ndelta > bigadj || ndelta < -bigadj)
452 ntickdelta = 10 * tickadj;
454 ntickdelta = tickadj;
455 if (ndelta % ntickdelta)
456 ndelta = ndelta / ntickdelta * ntickdelta;
459 * To make hardclock()'s job easier, make the per-tick delta negative
460 * if we want time to run slower; then hardclock can simply compute
461 * tick + tickdelta, and subtract tickdelta from timedelta.
464 ntickdelta = -ntickdelta;
466 * XXX not MP safe , but will probably work anyway.
471 tickdelta = ntickdelta;
475 atv.tv_sec = odelta / 1000000;
476 atv.tv_usec = odelta % 1000000;
477 (void) copyout((caddr_t)&atv, (caddr_t)uap->olddelta,
478 sizeof(struct timeval));
484 * Get value of an interval timer. The process virtual and
485 * profiling virtual time timers are kept in the p_stats area, since
486 * they can be swapped out. These are kept internally in the
487 * way they are specified externally: in time until they expire.
489 * The real time interval timer is kept in the process table slot
490 * for the process, and its value (it_value) is kept as an
491 * absolute time rather than as a delta, so that it is easy to keep
492 * periodic real-time signals from drifting.
494 * Virtual time timers are processed in the hardclock() routine of
495 * kern_clock.c. The real time timer is processed by a timeout
496 * routine, called from the softclock() routine. Since a callout
497 * may be delayed in real time due to interrupt processing in the system,
498 * it is possible for the real time timeout routine (realitexpire, given below),
499 * to be delayed in real time past when it is supposed to occur. It
500 * does not suffice, therefore, to reload the real timer .it_value from the
501 * real time timers .it_interval. Rather, we compute the next time in
502 * absolute time the timer should go off.
506 getitimer(struct getitimer_args *uap)
508 struct proc *p = curproc;
510 struct itimerval aitv;
512 if (uap->which > ITIMER_PROF)
515 if (uap->which == ITIMER_REAL) {
517 * Convert from absolute to relative time in .it_value
518 * part of real time timer. If time for real time timer
519 * has passed return 0, else return difference between
520 * current time and time for the timer to go off.
522 aitv = p->p_realtimer;
523 if (timevalisset(&aitv.it_value)) {
524 getmicrouptime(&ctv);
525 if (timevalcmp(&aitv.it_value, &ctv, <))
526 timevalclear(&aitv.it_value);
528 timevalsub(&aitv.it_value, &ctv);
531 aitv = p->p_stats->p_timer[uap->which];
534 return (copyout((caddr_t)&aitv, (caddr_t)uap->itv,
535 sizeof (struct itimerval)));
540 setitimer(struct setitimer_args *uap)
542 struct itimerval aitv;
544 struct itimerval *itvp;
545 struct proc *p = curproc;
548 if (uap->which > ITIMER_PROF)
551 if (itvp && (error = copyin((caddr_t)itvp, (caddr_t)&aitv,
552 sizeof(struct itimerval))))
554 if ((uap->itv = uap->oitv) &&
555 (error = getitimer((struct getitimer_args *)uap)))
559 if (itimerfix(&aitv.it_value))
561 if (!timevalisset(&aitv.it_value))
562 timevalclear(&aitv.it_interval);
563 else if (itimerfix(&aitv.it_interval))
566 if (uap->which == ITIMER_REAL) {
567 if (timevalisset(&p->p_realtimer.it_value))
568 callout_stop(&p->p_ithandle);
569 if (timevalisset(&aitv.it_value))
570 callout_reset(&p->p_ithandle,
571 tvtohz_high(&aitv.it_value), realitexpire, p);
572 getmicrouptime(&ctv);
573 timevaladd(&aitv.it_value, &ctv);
574 p->p_realtimer = aitv;
576 p->p_stats->p_timer[uap->which] = aitv;
583 * Real interval timer expired:
584 * send process whose timer expired an alarm signal.
585 * If time is not set up to reload, then just return.
586 * Else compute next time timer should go off which is > current time.
587 * This is where delay in processing this timeout causes multiple
588 * SIGALRM calls to be compressed into one.
589 * tvtohz_high() always adds 1 to allow for the time until the next clock
590 * interrupt being strictly less than 1 clock tick, but we don't want
591 * that here since we want to appear to be in sync with the clock
592 * interrupt even when we're delayed.
599 struct timeval ctv, ntv;
601 p = (struct proc *)arg;
603 if (!timevalisset(&p->p_realtimer.it_interval)) {
604 timevalclear(&p->p_realtimer.it_value);
609 timevaladd(&p->p_realtimer.it_value,
610 &p->p_realtimer.it_interval);
611 getmicrouptime(&ctv);
612 if (timevalcmp(&p->p_realtimer.it_value, &ctv, >)) {
613 ntv = p->p_realtimer.it_value;
614 timevalsub(&ntv, &ctv);
615 callout_reset(&p->p_ithandle, tvtohz_low(&ntv),
625 * Check that a proposed value to load into the .it_value or
626 * .it_interval part of an interval timer is acceptable, and
627 * fix it to have at least minimal value (i.e. if it is less
628 * than the resolution of the clock, round it up.)
635 if (tv->tv_sec < 0 || tv->tv_sec > 100000000 ||
636 tv->tv_usec < 0 || tv->tv_usec >= 1000000)
638 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick)
644 * Decrement an interval timer by a specified number
645 * of microseconds, which must be less than a second,
646 * i.e. < 1000000. If the timer expires, then reload
647 * it. In this case, carry over (usec - old value) to
648 * reduce the value reloaded into the timer so that
649 * the timer does not drift. This routine assumes
650 * that it is called in a context where the timers
651 * on which it is operating cannot change in value.
654 itimerdecr(itp, usec)
655 struct itimerval *itp;
659 if (itp->it_value.tv_usec < usec) {
660 if (itp->it_value.tv_sec == 0) {
661 /* expired, and already in next interval */
662 usec -= itp->it_value.tv_usec;
665 itp->it_value.tv_usec += 1000000;
666 itp->it_value.tv_sec--;
668 itp->it_value.tv_usec -= usec;
670 if (timevalisset(&itp->it_value))
672 /* expired, exactly at end of interval */
674 if (timevalisset(&itp->it_interval)) {
675 itp->it_value = itp->it_interval;
676 itp->it_value.tv_usec -= usec;
677 if (itp->it_value.tv_usec < 0) {
678 itp->it_value.tv_usec += 1000000;
679 itp->it_value.tv_sec--;
682 itp->it_value.tv_usec = 0; /* sec is already 0 */
687 * Add and subtract routines for timevals.
688 * N.B.: subtract routine doesn't deal with
689 * results which are before the beginning,
690 * it just gets very confused in this case.
695 struct timeval *t1, *t2;
698 t1->tv_sec += t2->tv_sec;
699 t1->tv_usec += t2->tv_usec;
705 struct timeval *t1, *t2;
708 t1->tv_sec -= t2->tv_sec;
709 t1->tv_usec -= t2->tv_usec;
718 if (t1->tv_usec < 0) {
720 t1->tv_usec += 1000000;
722 if (t1->tv_usec >= 1000000) {
724 t1->tv_usec -= 1000000;
729 * ratecheck(): simple time-based rate-limit checking.
732 ratecheck(struct timeval *lasttime, const struct timeval *mininterval)
734 struct timeval tv, delta;
737 getmicrouptime(&tv); /* NB: 10ms precision */
739 timevalsub(&delta, lasttime);
742 * check for 0,0 is so that the message will be seen at least once,
743 * even if interval is huge.
745 if (timevalcmp(&delta, mininterval, >=) ||
746 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) {
755 * ppsratecheck(): packets (or events) per second limitation.
757 * Return 0 if the limit is to be enforced (e.g. the caller
758 * should drop a packet because of the rate limitation).
760 * maxpps of 0 always causes zero to be returned. maxpps of -1
761 * always causes 1 to be returned; this effectively defeats rate
764 * Note that we maintain the struct timeval for compatibility
765 * with other bsd systems. We reuse the storage and just monitor
766 * clock ticks for minimal overhead.
769 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps)
774 * Reset the last time and counter if this is the first call
775 * or more than a second has passed since the last update of
779 if (lasttime->tv_sec == 0 || (u_int)(now - lasttime->tv_sec) >= hz) {
780 lasttime->tv_sec = now;
782 return (maxpps != 0);
784 (*curpps)++; /* NB: ignore potential overflow */
785 return (maxpps < 0 || *curpps < maxpps);