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. 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
30 * $FreeBSD: src/sys/kern/kern_time.c,v 1.68.2.1 2002/10/01 08:00:41 bde Exp $
33 #include <sys/param.h>
34 #include <sys/systm.h>
36 #include <sys/sysproto.h>
37 #include <sys/resourcevar.h>
38 #include <sys/signalvar.h>
39 #include <sys/kernel.h>
40 #include <sys/sysent.h>
41 #include <sys/sysunion.h>
45 #include <sys/vnode.h>
46 #include <sys/sysctl.h>
47 #include <sys/kern_syscall.h>
48 #include <sys/upmap.h>
50 #include <vm/vm_extern.h>
52 #include <sys/msgport2.h>
53 #include <sys/spinlock2.h>
54 #include <sys/thread2.h>
56 extern struct spinlock ntp_spin
;
58 #define CPUCLOCK_BIT 0x80000000
59 #define CPUCLOCK_ID_MASK ~CPUCLOCK_BIT
60 #define CPUCLOCK2LWPID(clock_id) (((clockid_t)(clock_id) >> 32) & CPUCLOCK_ID_MASK)
61 #define CPUCLOCK2PID(clock_id) ((clock_id) & CPUCLOCK_ID_MASK)
62 #define MAKE_CPUCLOCK(pid, lwp_id) ((clockid_t)(lwp_id) << 32 | (pid) | CPUCLOCK_BIT)
67 * Time of day and interval timer support.
69 * These routines provide the kernel entry points to get and set
70 * the time-of-day and per-process interval timers. Subroutines
71 * here provide support for adding and subtracting timeval structures
72 * and decrementing interval timers, optionally reloading the interval
73 * timers when they expire.
76 static int settime(struct timeval
*);
77 static void timevalfix(struct timeval
*);
78 static void realitexpire(void *arg
);
80 static int sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS
);
84 * Nanosleep tries very hard to sleep for a precisely requested time
85 * interval, down to 1uS. The administrator can impose a minimum delay
86 * and a delay below which we hard-loop instead of initiate a timer
87 * interrupt and sleep.
89 * For machines under high loads it might be beneficial to increase min_us
90 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully.
92 static int nanosleep_min_us
= 10;
93 static int nanosleep_hard_us
= 100;
94 static int gettimeofday_quick
= 0;
95 SYSCTL_INT(_kern
, OID_AUTO
, nanosleep_min_us
, CTLFLAG_RW
,
96 &nanosleep_min_us
, 0, "");
97 SYSCTL_INT(_kern
, OID_AUTO
, nanosleep_hard_us
, CTLFLAG_RW
,
98 &nanosleep_hard_us
, 0, "");
99 SYSCTL_PROC(_kern
, OID_AUTO
, gettimeofday_quick
, CTLTYPE_INT
| CTLFLAG_RW
,
100 0, 0, sysctl_gettimeofday_quick
, "I", "Quick mode gettimeofday");
102 static struct lock masterclock_lock
= LOCK_INITIALIZER("mstrclk", 0, 0);
105 settime(struct timeval
*tv
)
107 struct timeval delta
, tv1
, tv2
;
108 static struct timeval maxtime
, laststep
;
112 if ((origcpu
= mycpu
->gd_cpuid
) != 0)
113 lwkt_setcpu_self(globaldata_find(0));
118 timevalsub(&delta
, &tv1
);
121 * If the system is secure, we do not allow the time to be
122 * set to a value earlier than 1 second less than the highest
123 * time we have yet seen. The worst a miscreant can do in
124 * this circumstance is "freeze" time. He couldn't go
127 * We similarly do not allow the clock to be stepped more
128 * than one second, nor more than once per second. This allows
129 * a miscreant to make the clock march double-time, but no worse.
131 if (securelevel
> 1) {
132 if (delta
.tv_sec
< 0 || delta
.tv_usec
< 0) {
134 * Update maxtime to latest time we've seen.
136 if (tv1
.tv_sec
> maxtime
.tv_sec
)
139 timevalsub(&tv2
, &maxtime
);
140 if (tv2
.tv_sec
< -1) {
141 tv
->tv_sec
= maxtime
.tv_sec
- 1;
142 kprintf("Time adjustment clamped to -1 second\n");
145 if (tv1
.tv_sec
== laststep
.tv_sec
) {
149 if (delta
.tv_sec
> 1) {
150 tv
->tv_sec
= tv1
.tv_sec
+ 1;
151 kprintf("Time adjustment clamped to +1 second\n");
157 ts
.tv_sec
= tv
->tv_sec
;
158 ts
.tv_nsec
= tv
->tv_usec
* 1000;
163 lwkt_setcpu_self(globaldata_find(origcpu
));
170 get_process_cputime(struct proc
*p
, struct timespec
*ats
)
174 lwkt_gettoken(&p
->p_token
);
176 lwkt_reltoken(&p
->p_token
);
177 timevaladd(&ru
.ru_utime
, &ru
.ru_stime
);
178 TIMEVAL_TO_TIMESPEC(&ru
.ru_utime
, ats
);
182 get_process_usertime(struct proc
*p
, struct timespec
*ats
)
186 lwkt_gettoken(&p
->p_token
);
188 lwkt_reltoken(&p
->p_token
);
189 TIMEVAL_TO_TIMESPEC(&ru
.ru_utime
, ats
);
193 get_thread_cputime(struct thread
*td
, struct timespec
*ats
)
195 struct timeval sys
, user
;
197 calcru(td
->td_lwp
, &user
, &sys
);
198 timevaladd(&user
, &sys
);
199 TIMEVAL_TO_TIMESPEC(&user
, ats
);
206 kern_clock_gettime(clockid_t clock_id
, struct timespec
*ats
)
215 case CLOCK_REALTIME_PRECISE
:
218 case CLOCK_REALTIME_FAST
:
221 case CLOCK_MONOTONIC
:
222 case CLOCK_MONOTONIC_PRECISE
:
224 case CLOCK_UPTIME_PRECISE
:
227 case CLOCK_MONOTONIC_FAST
:
228 case CLOCK_UPTIME_FAST
:
232 get_process_usertime(p
, ats
);
235 case CLOCK_PROCESS_CPUTIME_ID
:
236 get_process_cputime(p
, ats
);
239 ats
->tv_sec
= time_second
;
242 case CLOCK_THREAD_CPUTIME_ID
:
243 get_thread_cputime(curthread
, ats
);
246 if ((clock_id
& CPUCLOCK_BIT
) == 0)
248 if ((p
= pfind(CPUCLOCK2PID(clock_id
))) == NULL
)
250 lwp_id
= CPUCLOCK2LWPID(clock_id
);
252 get_process_cputime(p
, ats
);
254 lwkt_gettoken(&p
->p_token
);
255 lp
= lwp_rb_tree_RB_LOOKUP(&p
->p_lwp_tree
, lwp_id
);
257 lwkt_reltoken(&p
->p_token
);
261 get_thread_cputime(lp
->lwp_thread
, ats
);
262 lwkt_reltoken(&p
->p_token
);
273 sys_clock_gettime(struct clock_gettime_args
*uap
)
278 error
= kern_clock_gettime(uap
->clock_id
, &ats
);
280 error
= copyout(&ats
, uap
->tp
, sizeof(ats
));
286 kern_clock_settime(clockid_t clock_id
, struct timespec
*ats
)
288 struct thread
*td
= curthread
;
292 if ((error
= priv_check(td
, PRIV_CLOCK_SETTIME
)) != 0)
294 if (clock_id
!= CLOCK_REALTIME
)
296 if (ats
->tv_nsec
< 0 || ats
->tv_nsec
>= 1000000000)
299 lockmgr(&masterclock_lock
, LK_EXCLUSIVE
);
300 TIMESPEC_TO_TIMEVAL(&atv
, ats
);
301 error
= settime(&atv
);
302 lockmgr(&masterclock_lock
, LK_RELEASE
);
311 sys_clock_settime(struct clock_settime_args
*uap
)
316 if ((error
= copyin(uap
->tp
, &ats
, sizeof(ats
))) != 0)
319 error
= kern_clock_settime(uap
->clock_id
, &ats
);
328 kern_clock_getres(clockid_t clock_id
, struct timespec
*ts
)
333 case CLOCK_REALTIME_FAST
:
334 case CLOCK_REALTIME_PRECISE
:
335 case CLOCK_MONOTONIC
:
336 case CLOCK_MONOTONIC_FAST
:
337 case CLOCK_MONOTONIC_PRECISE
:
339 case CLOCK_UPTIME_FAST
:
340 case CLOCK_UPTIME_PRECISE
:
342 * Round up the result of the division cheaply
343 * by adding 1. Rounding up is especially important
344 * if rounding down would give 0. Perfect rounding
347 ts
->tv_nsec
= 1000000000 / sys_cputimer
->freq
+ 1;
351 /* Accurately round up here because we can do so cheaply. */
352 ts
->tv_nsec
= (1000000000 + hz
- 1) / hz
;
358 case CLOCK_THREAD_CPUTIME_ID
:
359 case CLOCK_PROCESS_CPUTIME_ID
:
363 if ((clock_id
& CPUCLOCK_BIT
) != 0)
376 sys_clock_getres(struct clock_getres_args
*uap
)
381 error
= kern_clock_getres(uap
->clock_id
, &ts
);
383 error
= copyout(&ts
, uap
->tp
, sizeof(ts
));
389 kern_getcpuclockid(pid_t pid
, lwpid_t lwp_id
, clockid_t
*clock_id
)
403 /* lwp_id can be 0 when called by clock_getcpuclockid() */
408 lwkt_gettoken(&p
->p_token
);
410 lwp_rb_tree_RB_LOOKUP(&p
->p_lwp_tree
, lwp_id
) == NULL
) {
411 lwkt_reltoken(&p
->p_token
);
415 *clock_id
= MAKE_CPUCLOCK(pid
, lwp_id
);
416 lwkt_reltoken(&p
->p_token
);
423 sys_getcpuclockid(struct getcpuclockid_args
*uap
)
428 error
= kern_getcpuclockid(uap
->pid
, uap
->lwp_id
, &clk_id
);
430 error
= copyout(&clk_id
, uap
->clock_id
, sizeof(clockid_t
));
438 * This is a general helper function for nanosleep() (aka sleep() aka
441 * If there is less then one tick's worth of time left and
442 * we haven't done a yield, or the remaining microseconds is
443 * ridiculously low, do a yield. This avoids having
444 * to deal with systimer overheads when the system is under
445 * heavy loads. If we have done a yield already then use
446 * a systimer and an uninterruptable thread wait.
448 * If there is more then a tick's worth of time left,
449 * calculate the baseline ticks and use an interruptable
450 * tsleep, then handle the fine-grained delay on the next
451 * loop. This usually results in two sleeps occuring, a long one
457 ns1_systimer(systimer_t info
, int in_ipi __unused
,
458 struct intrframe
*frame __unused
)
460 lwkt_schedule(info
->data
);
464 nanosleep1(struct timespec
*rqt
, struct timespec
*rmt
)
467 struct timespec ts
, ts2
, ts3
;
471 if (rqt
->tv_nsec
< 0 || rqt
->tv_nsec
>= 1000000000)
473 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */
474 if (rqt
->tv_sec
< 0 || (rqt
->tv_sec
== 0 && rqt
->tv_nsec
== 0))
477 timespecadd(&ts
, rqt
); /* ts = target timestamp compare */
478 TIMESPEC_TO_TIMEVAL(&tv
, rqt
); /* tv = sleep interval */
482 struct systimer info
;
484 ticks
= tv
.tv_usec
/ ustick
; /* approximate */
486 if (tv
.tv_sec
== 0 && ticks
== 0) {
487 thread_t td
= curthread
;
488 if (tv
.tv_usec
> 0 && tv
.tv_usec
< nanosleep_min_us
)
489 tv
.tv_usec
= nanosleep_min_us
;
490 if (tv
.tv_usec
< nanosleep_hard_us
) {
494 crit_enter_quick(td
);
495 systimer_init_oneshot(&info
, ns1_systimer
,
497 lwkt_deschedule_self(td
);
500 systimer_del(&info
); /* make sure it's gone */
502 error
= iscaught(td
->td_lwp
);
503 } else if (tv
.tv_sec
== 0) {
504 error
= tsleep(&nanowait
, PCATCH
, "nanslp", ticks
);
506 ticks
= tvtohz_low(&tv
); /* also handles overflow */
507 error
= tsleep(&nanowait
, PCATCH
, "nanslp", ticks
);
510 if (error
&& error
!= EWOULDBLOCK
) {
511 if (error
== ERESTART
)
514 timespecsub(&ts
, &ts2
);
521 if (timespeccmp(&ts2
, &ts
, >=))
524 timespecsub(&ts3
, &ts2
);
525 TIMESPEC_TO_TIMEVAL(&tv
, &ts3
);
533 sys_nanosleep(struct nanosleep_args
*uap
)
539 error
= copyin(uap
->rqtp
, &rqt
, sizeof(rqt
));
543 error
= nanosleep1(&rqt
, &rmt
);
546 * copyout the residual if nanosleep was interrupted.
548 if (error
&& uap
->rmtp
) {
551 error2
= copyout(&rmt
, uap
->rmtp
, sizeof(rmt
));
559 * The gettimeofday() system call is supposed to return a fine-grained
560 * realtime stamp. However, acquiring a fine-grained stamp can create a
561 * bottleneck when multiple cpu cores are trying to accessing e.g. the
562 * HPET hardware timer all at the same time, so we have a sysctl that
563 * allows its behavior to be changed to a more coarse-grained timestamp
564 * which does not have to access a hardware timer.
567 sys_gettimeofday(struct gettimeofday_args
*uap
)
573 if (gettimeofday_quick
)
577 if ((error
= copyout((caddr_t
)&atv
, (caddr_t
)uap
->tp
,
582 error
= copyout((caddr_t
)&tz
, (caddr_t
)uap
->tzp
,
591 sys_settimeofday(struct settimeofday_args
*uap
)
593 struct thread
*td
= curthread
;
598 if ((error
= priv_check(td
, PRIV_SETTIMEOFDAY
)))
601 * Verify all parameters before changing time.
603 * XXX: We do not allow the time to be set to 0.0, which also by
604 * happy coincidence works around a pkgsrc bulk build bug.
607 if ((error
= copyin((caddr_t
)uap
->tv
, (caddr_t
)&atv
,
610 if (atv
.tv_usec
< 0 || atv
.tv_usec
>= 1000000)
612 if (atv
.tv_sec
== 0 && atv
.tv_usec
== 0)
616 (error
= copyin((caddr_t
)uap
->tzp
, (caddr_t
)&atz
, sizeof(atz
))))
619 lockmgr(&masterclock_lock
, LK_EXCLUSIVE
);
620 if (uap
->tv
&& (error
= settime(&atv
))) {
621 lockmgr(&masterclock_lock
, LK_RELEASE
);
624 lockmgr(&masterclock_lock
, LK_RELEASE
);
632 * WARNING! Run with ntp_spin held
635 kern_adjtime_common(void)
637 if ((ntp_delta
>= 0 && ntp_delta
< ntp_default_tick_delta
) ||
638 (ntp_delta
< 0 && ntp_delta
> -ntp_default_tick_delta
))
639 ntp_tick_delta
= ntp_delta
;
640 else if (ntp_delta
> ntp_big_delta
)
641 ntp_tick_delta
= 10 * ntp_default_tick_delta
;
642 else if (ntp_delta
< -ntp_big_delta
)
643 ntp_tick_delta
= -10 * ntp_default_tick_delta
;
644 else if (ntp_delta
> 0)
645 ntp_tick_delta
= ntp_default_tick_delta
;
647 ntp_tick_delta
= -ntp_default_tick_delta
;
651 kern_adjtime(int64_t delta
, int64_t *odelta
)
653 spin_lock(&ntp_spin
);
656 kern_adjtime_common();
657 spin_unlock(&ntp_spin
);
661 kern_get_ntp_delta(int64_t *delta
)
667 kern_reladjtime(int64_t delta
)
669 spin_lock(&ntp_spin
);
671 kern_adjtime_common();
672 spin_unlock(&ntp_spin
);
676 kern_adjfreq(int64_t rate
)
678 spin_lock(&ntp_spin
);
679 ntp_tick_permanent
= rate
;
680 spin_unlock(&ntp_spin
);
687 sys_adjtime(struct adjtime_args
*uap
)
689 struct thread
*td
= curthread
;
691 int64_t ndelta
, odelta
;
694 if ((error
= priv_check(td
, PRIV_ADJTIME
)))
696 error
= copyin(uap
->delta
, &atv
, sizeof(struct timeval
));
701 * Compute the total correction and the rate at which to apply it.
702 * Round the adjustment down to a whole multiple of the per-tick
703 * delta, so that after some number of incremental changes in
704 * hardclock(), tickdelta will become zero, lest the correction
705 * overshoot and start taking us away from the desired final time.
707 ndelta
= (int64_t)atv
.tv_sec
* 1000000000 + atv
.tv_usec
* 1000;
708 kern_adjtime(ndelta
, &odelta
);
711 atv
.tv_sec
= odelta
/ 1000000000;
712 atv
.tv_usec
= odelta
% 1000000000 / 1000;
713 copyout(&atv
, uap
->olddelta
, sizeof(struct timeval
));
719 sysctl_adjtime(SYSCTL_HANDLER_ARGS
)
724 if (req
->newptr
!= NULL
) {
725 if (priv_check(curthread
, PRIV_ROOT
))
727 error
= SYSCTL_IN(req
, &delta
, sizeof(delta
));
730 kern_reladjtime(delta
);
734 kern_get_ntp_delta(&delta
);
735 error
= SYSCTL_OUT(req
, &delta
, sizeof(delta
));
740 * delta is in nanoseconds.
743 sysctl_delta(SYSCTL_HANDLER_ARGS
)
745 int64_t delta
, old_delta
;
748 if (req
->newptr
!= NULL
) {
749 if (priv_check(curthread
, PRIV_ROOT
))
751 error
= SYSCTL_IN(req
, &delta
, sizeof(delta
));
754 kern_adjtime(delta
, &old_delta
);
757 if (req
->oldptr
!= NULL
)
758 kern_get_ntp_delta(&old_delta
);
759 error
= SYSCTL_OUT(req
, &old_delta
, sizeof(old_delta
));
764 * frequency is in nanoseconds per second shifted left 32.
765 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32.
768 sysctl_adjfreq(SYSCTL_HANDLER_ARGS
)
773 if (req
->newptr
!= NULL
) {
774 if (priv_check(curthread
, PRIV_ROOT
))
776 error
= SYSCTL_IN(req
, &freqdelta
, sizeof(freqdelta
));
781 kern_adjfreq(freqdelta
);
784 if (req
->oldptr
!= NULL
)
785 freqdelta
= ntp_tick_permanent
* hz
;
786 error
= SYSCTL_OUT(req
, &freqdelta
, sizeof(freqdelta
));
793 SYSCTL_NODE(_kern
, OID_AUTO
, ntp
, CTLFLAG_RW
, 0, "NTP related controls");
794 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, permanent
,
795 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
796 sysctl_adjfreq
, "Q", "permanent correction per second");
797 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, delta
,
798 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
799 sysctl_delta
, "Q", "one-time delta");
800 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, big_delta
, CTLFLAG_RD
,
801 &ntp_big_delta
, sizeof(ntp_big_delta
), "Q",
802 "threshold for fast adjustment");
803 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, tick_delta
, CTLFLAG_RD
,
804 &ntp_tick_delta
, sizeof(ntp_tick_delta
), "LU",
805 "per-tick adjustment");
806 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, default_tick_delta
, CTLFLAG_RD
,
807 &ntp_default_tick_delta
, sizeof(ntp_default_tick_delta
), "LU",
808 "default per-tick adjustment");
809 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, next_leap_second
, CTLFLAG_RW
,
810 &ntp_leap_second
, sizeof(ntp_leap_second
), "LU",
812 SYSCTL_INT(_kern_ntp
, OID_AUTO
, insert_leap_second
, CTLFLAG_RW
,
813 &ntp_leap_insert
, 0, "insert or remove leap second");
814 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, adjust
,
815 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
816 sysctl_adjtime
, "Q", "relative adjust for delta");
819 * Get value of an interval timer. The process virtual and
820 * profiling virtual time timers are kept in the p_stats area, since
821 * they can be swapped out. These are kept internally in the
822 * way they are specified externally: in time until they expire.
824 * The real time interval timer is kept in the process table slot
825 * for the process, and its value (it_value) is kept as an
826 * absolute time rather than as a delta, so that it is easy to keep
827 * periodic real-time signals from drifting.
829 * Virtual time timers are processed in the hardclock() routine of
830 * kern_clock.c. The real time timer is processed by a timeout
831 * routine, called from the softclock() routine. Since a callout
832 * may be delayed in real time due to interrupt processing in the system,
833 * it is possible for the real time timeout routine (realitexpire, given below),
834 * to be delayed in real time past when it is supposed to occur. It
835 * does not suffice, therefore, to reload the real timer .it_value from the
836 * real time timers .it_interval. Rather, we compute the next time in
837 * absolute time the timer should go off.
842 sys_getitimer(struct getitimer_args
*uap
)
844 struct proc
*p
= curproc
;
846 struct itimerval aitv
;
848 if (uap
->which
> ITIMER_PROF
)
850 lwkt_gettoken(&p
->p_token
);
851 if (uap
->which
== ITIMER_REAL
) {
853 * Convert from absolute to relative time in .it_value
854 * part of real time timer. If time for real time timer
855 * has passed return 0, else return difference between
856 * current time and time for the timer to go off.
858 aitv
= p
->p_realtimer
;
859 if (timevalisset(&aitv
.it_value
)) {
860 getmicrouptime(&ctv
);
861 if (timevalcmp(&aitv
.it_value
, &ctv
, <))
862 timevalclear(&aitv
.it_value
);
864 timevalsub(&aitv
.it_value
, &ctv
);
867 aitv
= p
->p_timer
[uap
->which
];
869 lwkt_reltoken(&p
->p_token
);
870 return (copyout(&aitv
, uap
->itv
, sizeof (struct itimerval
)));
877 sys_setitimer(struct setitimer_args
*uap
)
879 struct itimerval aitv
;
881 struct itimerval
*itvp
;
882 struct proc
*p
= curproc
;
885 if (uap
->which
> ITIMER_PROF
)
888 if (itvp
&& (error
= copyin((caddr_t
)itvp
, (caddr_t
)&aitv
,
889 sizeof(struct itimerval
))))
891 if ((uap
->itv
= uap
->oitv
) &&
892 (error
= sys_getitimer((struct getitimer_args
*)uap
)))
896 if (itimerfix(&aitv
.it_value
))
898 if (!timevalisset(&aitv
.it_value
))
899 timevalclear(&aitv
.it_interval
);
900 else if (itimerfix(&aitv
.it_interval
))
902 lwkt_gettoken(&p
->p_token
);
903 if (uap
->which
== ITIMER_REAL
) {
904 if (timevalisset(&p
->p_realtimer
.it_value
))
905 callout_stop_sync(&p
->p_ithandle
);
906 if (timevalisset(&aitv
.it_value
))
907 callout_reset(&p
->p_ithandle
,
908 tvtohz_high(&aitv
.it_value
), realitexpire
, p
);
909 getmicrouptime(&ctv
);
910 timevaladd(&aitv
.it_value
, &ctv
);
911 p
->p_realtimer
= aitv
;
913 p
->p_timer
[uap
->which
] = aitv
;
916 p
->p_flags
&= ~P_SIGVTALRM
;
919 p
->p_flags
&= ~P_SIGPROF
;
923 lwkt_reltoken(&p
->p_token
);
928 * Real interval timer expired:
929 * send process whose timer expired an alarm signal.
930 * If time is not set up to reload, then just return.
931 * Else compute next time timer should go off which is > current time.
932 * This is where delay in processing this timeout causes multiple
933 * SIGALRM calls to be compressed into one.
934 * tvtohz_high() always adds 1 to allow for the time until the next clock
935 * interrupt being strictly less than 1 clock tick, but we don't want
936 * that here since we want to appear to be in sync with the clock
937 * interrupt even when we're delayed.
941 realitexpire(void *arg
)
944 struct timeval ctv
, ntv
;
946 p
= (struct proc
*)arg
;
948 lwkt_gettoken(&p
->p_token
);
950 if (!timevalisset(&p
->p_realtimer
.it_interval
)) {
951 timevalclear(&p
->p_realtimer
.it_value
);
955 timevaladd(&p
->p_realtimer
.it_value
,
956 &p
->p_realtimer
.it_interval
);
957 getmicrouptime(&ctv
);
958 if (timevalcmp(&p
->p_realtimer
.it_value
, &ctv
, >)) {
959 ntv
= p
->p_realtimer
.it_value
;
960 timevalsub(&ntv
, &ctv
);
961 callout_reset(&p
->p_ithandle
, tvtohz_low(&ntv
),
967 lwkt_reltoken(&p
->p_token
);
972 * Used to validate itimer timeouts and utimes*() timespecs.
975 itimerfix(struct timeval
*tv
)
977 if (tv
->tv_sec
< 0 || tv
->tv_usec
< 0 || tv
->tv_usec
>= 1000000)
979 if (tv
->tv_sec
== 0 && tv
->tv_usec
!= 0 && tv
->tv_usec
< ustick
)
980 tv
->tv_usec
= ustick
;
985 * Used to validate timeouts and utimes*() timespecs.
988 itimespecfix(struct timespec
*ts
)
990 if (ts
->tv_sec
< 0 || ts
->tv_nsec
< 0 || ts
->tv_nsec
>= 1000000000ULL)
992 if (ts
->tv_sec
== 0 && ts
->tv_nsec
!= 0 && ts
->tv_nsec
< nstick
)
993 ts
->tv_nsec
= nstick
;
998 * Decrement an interval timer by a specified number
999 * of microseconds, which must be less than a second,
1000 * i.e. < 1000000. If the timer expires, then reload
1001 * it. In this case, carry over (usec - old value) to
1002 * reduce the value reloaded into the timer so that
1003 * the timer does not drift. This routine assumes
1004 * that it is called in a context where the timers
1005 * on which it is operating cannot change in value.
1008 itimerdecr(struct itimerval
*itp
, int usec
)
1011 if (itp
->it_value
.tv_usec
< usec
) {
1012 if (itp
->it_value
.tv_sec
== 0) {
1013 /* expired, and already in next interval */
1014 usec
-= itp
->it_value
.tv_usec
;
1017 itp
->it_value
.tv_usec
+= 1000000;
1018 itp
->it_value
.tv_sec
--;
1020 itp
->it_value
.tv_usec
-= usec
;
1022 if (timevalisset(&itp
->it_value
))
1024 /* expired, exactly at end of interval */
1026 if (timevalisset(&itp
->it_interval
)) {
1027 itp
->it_value
= itp
->it_interval
;
1028 itp
->it_value
.tv_usec
-= usec
;
1029 if (itp
->it_value
.tv_usec
< 0) {
1030 itp
->it_value
.tv_usec
+= 1000000;
1031 itp
->it_value
.tv_sec
--;
1034 itp
->it_value
.tv_usec
= 0; /* sec is already 0 */
1039 * Add and subtract routines for timevals.
1040 * N.B.: subtract routine doesn't deal with
1041 * results which are before the beginning,
1042 * it just gets very confused in this case.
1046 timevaladd(struct timeval
*t1
, const struct timeval
*t2
)
1049 t1
->tv_sec
+= t2
->tv_sec
;
1050 t1
->tv_usec
+= t2
->tv_usec
;
1055 timevalsub(struct timeval
*t1
, const struct timeval
*t2
)
1058 t1
->tv_sec
-= t2
->tv_sec
;
1059 t1
->tv_usec
-= t2
->tv_usec
;
1064 timevalfix(struct timeval
*t1
)
1067 if (t1
->tv_usec
< 0) {
1069 t1
->tv_usec
+= 1000000;
1071 if (t1
->tv_usec
>= 1000000) {
1073 t1
->tv_usec
-= 1000000;
1078 * ratecheck(): simple time-based rate-limit checking.
1081 ratecheck(struct timeval
*lasttime
, const struct timeval
*mininterval
)
1083 struct timeval tv
, delta
;
1086 getmicrouptime(&tv
); /* NB: 10ms precision */
1088 timevalsub(&delta
, lasttime
);
1091 * check for 0,0 is so that the message will be seen at least once,
1092 * even if interval is huge.
1094 if (timevalcmp(&delta
, mininterval
, >=) ||
1095 (lasttime
->tv_sec
== 0 && lasttime
->tv_usec
== 0)) {
1104 * ppsratecheck(): packets (or events) per second limitation.
1106 * Return 0 if the limit is to be enforced (e.g. the caller
1107 * should drop a packet because of the rate limitation).
1109 * maxpps of 0 always causes zero to be returned. maxpps of -1
1110 * always causes 1 to be returned; this effectively defeats rate
1113 * Note that we maintain the struct timeval for compatibility
1114 * with other bsd systems. We reuse the storage and just monitor
1115 * clock ticks for minimal overhead.
1118 ppsratecheck(struct timeval
*lasttime
, int *curpps
, int maxpps
)
1123 * Reset the last time and counter if this is the first call
1124 * or more than a second has passed since the last update of
1128 if (lasttime
->tv_sec
== 0 || (u_int
)(now
- lasttime
->tv_sec
) >= hz
) {
1129 lasttime
->tv_sec
= now
;
1131 return (maxpps
!= 0);
1133 (*curpps
)++; /* NB: ignore potential overflow */
1134 return (maxpps
< 0 || *curpps
< maxpps
);
1139 sysctl_gettimeofday_quick(SYSCTL_HANDLER_ARGS
)
1144 gtod
= gettimeofday_quick
;
1145 error
= sysctl_handle_int(oidp
, >od
, 0, req
);
1146 if (error
|| req
->newptr
== NULL
)
1148 gettimeofday_quick
= gtod
;
1150 kpmap
->fast_gtod
= gtod
;