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>
49 #include <vm/vm_extern.h>
51 #include <sys/msgport2.h>
52 #include <sys/spinlock2.h>
53 #include <sys/thread2.h>
55 extern struct spinlock ntp_spin
;
57 #define CPUCLOCK_BIT 0x80000000
58 #define CPUCLOCK_ID_MASK ~CPUCLOCK_BIT
59 #define CPUCLOCK2LWPID(clock_id) (((clockid_t)(clock_id) >> 32) & CPUCLOCK_ID_MASK)
60 #define CPUCLOCK2PID(clock_id) ((clock_id) & CPUCLOCK_ID_MASK)
61 #define MAKE_CPUCLOCK(pid, lwp_id) ((clockid_t)(lwp_id) << 32 | (pid) | CPUCLOCK_BIT)
66 * Time of day and interval timer support.
68 * These routines provide the kernel entry points to get and set
69 * the time-of-day and per-process interval timers. Subroutines
70 * here provide support for adding and subtracting timeval structures
71 * and decrementing interval timers, optionally reloading the interval
72 * timers when they expire.
75 static int settime(struct timeval
*);
76 static void timevalfix(struct timeval
*);
77 static void realitexpire(void *arg
);
80 * Nanosleep tries very hard to sleep for a precisely requested time
81 * interval, down to 1uS. The administrator can impose a minimum delay
82 * and a delay below which we hard-loop instead of initiate a timer
83 * interrupt and sleep.
85 * For machines under high loads it might be beneficial to increase min_us
86 * to e.g. 1000uS (1ms) so spining processes sleep meaningfully.
88 static int nanosleep_min_us
= 10;
89 static int nanosleep_hard_us
= 100;
90 static int gettimeofday_quick
= 0;
91 SYSCTL_INT(_kern
, OID_AUTO
, nanosleep_min_us
, CTLFLAG_RW
,
92 &nanosleep_min_us
, 0, "");
93 SYSCTL_INT(_kern
, OID_AUTO
, nanosleep_hard_us
, CTLFLAG_RW
,
94 &nanosleep_hard_us
, 0, "");
95 SYSCTL_INT(_kern
, OID_AUTO
, gettimeofday_quick
, CTLFLAG_RW
,
96 &gettimeofday_quick
, 0, "");
98 static struct lock masterclock_lock
= LOCK_INITIALIZER("mstrclk", 0, 0);
101 settime(struct timeval
*tv
)
103 struct timeval delta
, tv1
, tv2
;
104 static struct timeval maxtime
, laststep
;
108 if ((origcpu
= mycpu
->gd_cpuid
) != 0)
109 lwkt_setcpu_self(globaldata_find(0));
114 timevalsub(&delta
, &tv1
);
117 * If the system is secure, we do not allow the time to be
118 * set to a value earlier than 1 second less than the highest
119 * time we have yet seen. The worst a miscreant can do in
120 * this circumstance is "freeze" time. He couldn't go
123 * We similarly do not allow the clock to be stepped more
124 * than one second, nor more than once per second. This allows
125 * a miscreant to make the clock march double-time, but no worse.
127 if (securelevel
> 1) {
128 if (delta
.tv_sec
< 0 || delta
.tv_usec
< 0) {
130 * Update maxtime to latest time we've seen.
132 if (tv1
.tv_sec
> maxtime
.tv_sec
)
135 timevalsub(&tv2
, &maxtime
);
136 if (tv2
.tv_sec
< -1) {
137 tv
->tv_sec
= maxtime
.tv_sec
- 1;
138 kprintf("Time adjustment clamped to -1 second\n");
141 if (tv1
.tv_sec
== laststep
.tv_sec
) {
145 if (delta
.tv_sec
> 1) {
146 tv
->tv_sec
= tv1
.tv_sec
+ 1;
147 kprintf("Time adjustment clamped to +1 second\n");
153 ts
.tv_sec
= tv
->tv_sec
;
154 ts
.tv_nsec
= tv
->tv_usec
* 1000;
159 lwkt_setcpu_self(globaldata_find(origcpu
));
166 get_process_cputime(struct proc
*p
, struct timespec
*ats
)
170 lwkt_gettoken(&p
->p_token
);
172 lwkt_reltoken(&p
->p_token
);
173 timevaladd(&ru
.ru_utime
, &ru
.ru_stime
);
174 TIMEVAL_TO_TIMESPEC(&ru
.ru_utime
, ats
);
178 get_process_usertime(struct proc
*p
, struct timespec
*ats
)
182 lwkt_gettoken(&p
->p_token
);
184 lwkt_reltoken(&p
->p_token
);
185 TIMEVAL_TO_TIMESPEC(&ru
.ru_utime
, ats
);
189 get_thread_cputime(struct thread
*td
, struct timespec
*ats
)
191 struct timeval sys
, user
;
193 calcru(td
->td_lwp
, &user
, &sys
);
194 timevaladd(&user
, &sys
);
195 TIMEVAL_TO_TIMESPEC(&user
, ats
);
202 kern_clock_gettime(clockid_t clock_id
, struct timespec
*ats
)
211 case CLOCK_REALTIME_PRECISE
:
214 case CLOCK_REALTIME_FAST
:
217 case CLOCK_MONOTONIC
:
218 case CLOCK_MONOTONIC_PRECISE
:
220 case CLOCK_UPTIME_PRECISE
:
223 case CLOCK_MONOTONIC_FAST
:
224 case CLOCK_UPTIME_FAST
:
228 get_process_usertime(p
, ats
);
231 case CLOCK_PROCESS_CPUTIME_ID
:
232 get_process_cputime(p
, ats
);
235 ats
->tv_sec
= time_second
;
238 case CLOCK_THREAD_CPUTIME_ID
:
239 get_thread_cputime(curthread
, ats
);
242 if ((clock_id
& CPUCLOCK_BIT
) == 0)
244 if ((p
= pfind(CPUCLOCK2PID(clock_id
))) == NULL
)
246 lwp_id
= CPUCLOCK2LWPID(clock_id
);
248 get_process_cputime(p
, ats
);
250 lwkt_gettoken(&p
->p_token
);
251 lp
= lwp_rb_tree_RB_LOOKUP(&p
->p_lwp_tree
, lwp_id
);
253 lwkt_reltoken(&p
->p_token
);
257 get_thread_cputime(lp
->lwp_thread
, ats
);
258 lwkt_reltoken(&p
->p_token
);
269 sys_clock_gettime(struct clock_gettime_args
*uap
)
274 error
= kern_clock_gettime(uap
->clock_id
, &ats
);
276 error
= copyout(&ats
, uap
->tp
, sizeof(ats
));
282 kern_clock_settime(clockid_t clock_id
, struct timespec
*ats
)
284 struct thread
*td
= curthread
;
288 if ((error
= priv_check(td
, PRIV_CLOCK_SETTIME
)) != 0)
290 if (clock_id
!= CLOCK_REALTIME
)
292 if (ats
->tv_nsec
< 0 || ats
->tv_nsec
>= 1000000000)
295 lockmgr(&masterclock_lock
, LK_EXCLUSIVE
);
296 TIMESPEC_TO_TIMEVAL(&atv
, ats
);
297 error
= settime(&atv
);
298 lockmgr(&masterclock_lock
, LK_RELEASE
);
307 sys_clock_settime(struct clock_settime_args
*uap
)
312 if ((error
= copyin(uap
->tp
, &ats
, sizeof(ats
))) != 0)
315 error
= kern_clock_settime(uap
->clock_id
, &ats
);
324 kern_clock_getres(clockid_t clock_id
, struct timespec
*ts
)
329 case CLOCK_REALTIME_FAST
:
330 case CLOCK_REALTIME_PRECISE
:
331 case CLOCK_MONOTONIC
:
332 case CLOCK_MONOTONIC_FAST
:
333 case CLOCK_MONOTONIC_PRECISE
:
335 case CLOCK_UPTIME_FAST
:
336 case CLOCK_UPTIME_PRECISE
:
338 * Round up the result of the division cheaply
339 * by adding 1. Rounding up is especially important
340 * if rounding down would give 0. Perfect rounding
343 ts
->tv_nsec
= 1000000000 / sys_cputimer
->freq
+ 1;
347 /* Accurately round up here because we can do so cheaply. */
348 ts
->tv_nsec
= (1000000000 + hz
- 1) / hz
;
354 case CLOCK_THREAD_CPUTIME_ID
:
355 case CLOCK_PROCESS_CPUTIME_ID
:
359 if ((clock_id
& CPUCLOCK_BIT
) != 0)
372 sys_clock_getres(struct clock_getres_args
*uap
)
377 error
= kern_clock_getres(uap
->clock_id
, &ts
);
379 error
= copyout(&ts
, uap
->tp
, sizeof(ts
));
385 kern_getcpuclockid(pid_t pid
, lwpid_t lwp_id
, clockid_t
*clock_id
)
399 /* lwp_id can be 0 when called by clock_getcpuclockid() */
404 lwkt_gettoken(&p
->p_token
);
406 lwp_rb_tree_RB_LOOKUP(&p
->p_lwp_tree
, lwp_id
) == NULL
) {
407 lwkt_reltoken(&p
->p_token
);
411 *clock_id
= MAKE_CPUCLOCK(pid
, lwp_id
);
412 lwkt_reltoken(&p
->p_token
);
419 sys_getcpuclockid(struct getcpuclockid_args
*uap
)
424 error
= kern_getcpuclockid(uap
->pid
, uap
->lwp_id
, &clk_id
);
426 error
= copyout(&clk_id
, uap
->clock_id
, sizeof(clockid_t
));
434 * This is a general helper function for nanosleep() (aka sleep() aka
437 * If there is less then one tick's worth of time left and
438 * we haven't done a yield, or the remaining microseconds is
439 * ridiculously low, do a yield. This avoids having
440 * to deal with systimer overheads when the system is under
441 * heavy loads. If we have done a yield already then use
442 * a systimer and an uninterruptable thread wait.
444 * If there is more then a tick's worth of time left,
445 * calculate the baseline ticks and use an interruptable
446 * tsleep, then handle the fine-grained delay on the next
447 * loop. This usually results in two sleeps occuring, a long one
453 ns1_systimer(systimer_t info
, int in_ipi __unused
,
454 struct intrframe
*frame __unused
)
456 lwkt_schedule(info
->data
);
460 nanosleep1(struct timespec
*rqt
, struct timespec
*rmt
)
463 struct timespec ts
, ts2
, ts3
;
467 if (rqt
->tv_nsec
< 0 || rqt
->tv_nsec
>= 1000000000)
469 /* XXX: imho this should return EINVAL at least for tv_sec < 0 */
470 if (rqt
->tv_sec
< 0 || (rqt
->tv_sec
== 0 && rqt
->tv_nsec
== 0))
473 timespecadd(&ts
, rqt
); /* ts = target timestamp compare */
474 TIMESPEC_TO_TIMEVAL(&tv
, rqt
); /* tv = sleep interval */
478 struct systimer info
;
480 ticks
= tv
.tv_usec
/ ustick
; /* approximate */
482 if (tv
.tv_sec
== 0 && ticks
== 0) {
483 thread_t td
= curthread
;
484 if (tv
.tv_usec
> 0 && tv
.tv_usec
< nanosleep_min_us
)
485 tv
.tv_usec
= nanosleep_min_us
;
486 if (tv
.tv_usec
< nanosleep_hard_us
) {
490 crit_enter_quick(td
);
491 systimer_init_oneshot(&info
, ns1_systimer
,
493 lwkt_deschedule_self(td
);
496 systimer_del(&info
); /* make sure it's gone */
498 error
= iscaught(td
->td_lwp
);
499 } else if (tv
.tv_sec
== 0) {
500 error
= tsleep(&nanowait
, PCATCH
, "nanslp", ticks
);
502 ticks
= tvtohz_low(&tv
); /* also handles overflow */
503 error
= tsleep(&nanowait
, PCATCH
, "nanslp", ticks
);
506 if (error
&& error
!= EWOULDBLOCK
) {
507 if (error
== ERESTART
)
510 timespecsub(&ts
, &ts2
);
517 if (timespeccmp(&ts2
, &ts
, >=))
520 timespecsub(&ts3
, &ts2
);
521 TIMESPEC_TO_TIMEVAL(&tv
, &ts3
);
529 sys_nanosleep(struct nanosleep_args
*uap
)
535 error
= copyin(uap
->rqtp
, &rqt
, sizeof(rqt
));
539 error
= nanosleep1(&rqt
, &rmt
);
542 * copyout the residual if nanosleep was interrupted.
544 if (error
&& uap
->rmtp
) {
547 error2
= copyout(&rmt
, uap
->rmtp
, sizeof(rmt
));
555 * The gettimeofday() system call is supposed to return a fine-grained
556 * realtime stamp. However, acquiring a fine-grained stamp can create a
557 * bottleneck when multiple cpu cores are trying to accessing e.g. the
558 * HPET hardware timer all at the same time, so we have a sysctl that
559 * allows its behavior to be changed to a more coarse-grained timestamp
560 * which does not have to access a hardware timer.
563 sys_gettimeofday(struct gettimeofday_args
*uap
)
569 if (gettimeofday_quick
)
573 if ((error
= copyout((caddr_t
)&atv
, (caddr_t
)uap
->tp
,
578 error
= copyout((caddr_t
)&tz
, (caddr_t
)uap
->tzp
,
587 sys_settimeofday(struct settimeofday_args
*uap
)
589 struct thread
*td
= curthread
;
594 if ((error
= priv_check(td
, PRIV_SETTIMEOFDAY
)))
597 * Verify all parameters before changing time.
599 * XXX: We do not allow the time to be set to 0.0, which also by
600 * happy coincidence works around a pkgsrc bulk build bug.
603 if ((error
= copyin((caddr_t
)uap
->tv
, (caddr_t
)&atv
,
606 if (atv
.tv_usec
< 0 || atv
.tv_usec
>= 1000000)
608 if (atv
.tv_sec
== 0 && atv
.tv_usec
== 0)
612 (error
= copyin((caddr_t
)uap
->tzp
, (caddr_t
)&atz
, sizeof(atz
))))
615 lockmgr(&masterclock_lock
, LK_EXCLUSIVE
);
616 if (uap
->tv
&& (error
= settime(&atv
))) {
617 lockmgr(&masterclock_lock
, LK_RELEASE
);
620 lockmgr(&masterclock_lock
, LK_RELEASE
);
628 * WARNING! Run with ntp_spin held
631 kern_adjtime_common(void)
633 if ((ntp_delta
>= 0 && ntp_delta
< ntp_default_tick_delta
) ||
634 (ntp_delta
< 0 && ntp_delta
> -ntp_default_tick_delta
))
635 ntp_tick_delta
= ntp_delta
;
636 else if (ntp_delta
> ntp_big_delta
)
637 ntp_tick_delta
= 10 * ntp_default_tick_delta
;
638 else if (ntp_delta
< -ntp_big_delta
)
639 ntp_tick_delta
= -10 * ntp_default_tick_delta
;
640 else if (ntp_delta
> 0)
641 ntp_tick_delta
= ntp_default_tick_delta
;
643 ntp_tick_delta
= -ntp_default_tick_delta
;
647 kern_adjtime(int64_t delta
, int64_t *odelta
)
649 spin_lock(&ntp_spin
);
652 kern_adjtime_common();
653 spin_unlock(&ntp_spin
);
657 kern_get_ntp_delta(int64_t *delta
)
663 kern_reladjtime(int64_t delta
)
665 spin_lock(&ntp_spin
);
667 kern_adjtime_common();
668 spin_unlock(&ntp_spin
);
672 kern_adjfreq(int64_t rate
)
674 spin_lock(&ntp_spin
);
675 ntp_tick_permanent
= rate
;
676 spin_unlock(&ntp_spin
);
683 sys_adjtime(struct adjtime_args
*uap
)
685 struct thread
*td
= curthread
;
687 int64_t ndelta
, odelta
;
690 if ((error
= priv_check(td
, PRIV_ADJTIME
)))
692 error
= copyin(uap
->delta
, &atv
, sizeof(struct timeval
));
697 * Compute the total correction and the rate at which to apply it.
698 * Round the adjustment down to a whole multiple of the per-tick
699 * delta, so that after some number of incremental changes in
700 * hardclock(), tickdelta will become zero, lest the correction
701 * overshoot and start taking us away from the desired final time.
703 ndelta
= (int64_t)atv
.tv_sec
* 1000000000 + atv
.tv_usec
* 1000;
704 kern_adjtime(ndelta
, &odelta
);
707 atv
.tv_sec
= odelta
/ 1000000000;
708 atv
.tv_usec
= odelta
% 1000000000 / 1000;
709 copyout(&atv
, uap
->olddelta
, sizeof(struct timeval
));
715 sysctl_adjtime(SYSCTL_HANDLER_ARGS
)
720 if (req
->newptr
!= NULL
) {
721 if (priv_check(curthread
, PRIV_ROOT
))
723 error
= SYSCTL_IN(req
, &delta
, sizeof(delta
));
726 kern_reladjtime(delta
);
730 kern_get_ntp_delta(&delta
);
731 error
= SYSCTL_OUT(req
, &delta
, sizeof(delta
));
736 * delta is in nanoseconds.
739 sysctl_delta(SYSCTL_HANDLER_ARGS
)
741 int64_t delta
, old_delta
;
744 if (req
->newptr
!= NULL
) {
745 if (priv_check(curthread
, PRIV_ROOT
))
747 error
= SYSCTL_IN(req
, &delta
, sizeof(delta
));
750 kern_adjtime(delta
, &old_delta
);
753 if (req
->oldptr
!= NULL
)
754 kern_get_ntp_delta(&old_delta
);
755 error
= SYSCTL_OUT(req
, &old_delta
, sizeof(old_delta
));
760 * frequency is in nanoseconds per second shifted left 32.
761 * kern_adjfreq() needs it in nanoseconds per tick shifted left 32.
764 sysctl_adjfreq(SYSCTL_HANDLER_ARGS
)
769 if (req
->newptr
!= NULL
) {
770 if (priv_check(curthread
, PRIV_ROOT
))
772 error
= SYSCTL_IN(req
, &freqdelta
, sizeof(freqdelta
));
777 kern_adjfreq(freqdelta
);
780 if (req
->oldptr
!= NULL
)
781 freqdelta
= ntp_tick_permanent
* hz
;
782 error
= SYSCTL_OUT(req
, &freqdelta
, sizeof(freqdelta
));
789 SYSCTL_NODE(_kern
, OID_AUTO
, ntp
, CTLFLAG_RW
, 0, "NTP related controls");
790 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, permanent
,
791 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
792 sysctl_adjfreq
, "Q", "permanent correction per second");
793 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, delta
,
794 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
795 sysctl_delta
, "Q", "one-time delta");
796 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, big_delta
, CTLFLAG_RD
,
797 &ntp_big_delta
, sizeof(ntp_big_delta
), "Q",
798 "threshold for fast adjustment");
799 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, tick_delta
, CTLFLAG_RD
,
800 &ntp_tick_delta
, sizeof(ntp_tick_delta
), "LU",
801 "per-tick adjustment");
802 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, default_tick_delta
, CTLFLAG_RD
,
803 &ntp_default_tick_delta
, sizeof(ntp_default_tick_delta
), "LU",
804 "default per-tick adjustment");
805 SYSCTL_OPAQUE(_kern_ntp
, OID_AUTO
, next_leap_second
, CTLFLAG_RW
,
806 &ntp_leap_second
, sizeof(ntp_leap_second
), "LU",
808 SYSCTL_INT(_kern_ntp
, OID_AUTO
, insert_leap_second
, CTLFLAG_RW
,
809 &ntp_leap_insert
, 0, "insert or remove leap second");
810 SYSCTL_PROC(_kern_ntp
, OID_AUTO
, adjust
,
811 CTLTYPE_QUAD
|CTLFLAG_RW
, 0, 0,
812 sysctl_adjtime
, "Q", "relative adjust for delta");
815 * Get value of an interval timer. The process virtual and
816 * profiling virtual time timers are kept in the p_stats area, since
817 * they can be swapped out. These are kept internally in the
818 * way they are specified externally: in time until they expire.
820 * The real time interval timer is kept in the process table slot
821 * for the process, and its value (it_value) is kept as an
822 * absolute time rather than as a delta, so that it is easy to keep
823 * periodic real-time signals from drifting.
825 * Virtual time timers are processed in the hardclock() routine of
826 * kern_clock.c. The real time timer is processed by a timeout
827 * routine, called from the softclock() routine. Since a callout
828 * may be delayed in real time due to interrupt processing in the system,
829 * it is possible for the real time timeout routine (realitexpire, given below),
830 * to be delayed in real time past when it is supposed to occur. It
831 * does not suffice, therefore, to reload the real timer .it_value from the
832 * real time timers .it_interval. Rather, we compute the next time in
833 * absolute time the timer should go off.
838 sys_getitimer(struct getitimer_args
*uap
)
840 struct proc
*p
= curproc
;
842 struct itimerval aitv
;
844 if (uap
->which
> ITIMER_PROF
)
846 lwkt_gettoken(&p
->p_token
);
847 if (uap
->which
== ITIMER_REAL
) {
849 * Convert from absolute to relative time in .it_value
850 * part of real time timer. If time for real time timer
851 * has passed return 0, else return difference between
852 * current time and time for the timer to go off.
854 aitv
= p
->p_realtimer
;
855 if (timevalisset(&aitv
.it_value
)) {
856 getmicrouptime(&ctv
);
857 if (timevalcmp(&aitv
.it_value
, &ctv
, <))
858 timevalclear(&aitv
.it_value
);
860 timevalsub(&aitv
.it_value
, &ctv
);
863 aitv
= p
->p_timer
[uap
->which
];
865 lwkt_reltoken(&p
->p_token
);
866 return (copyout(&aitv
, uap
->itv
, sizeof (struct itimerval
)));
873 sys_setitimer(struct setitimer_args
*uap
)
875 struct itimerval aitv
;
877 struct itimerval
*itvp
;
878 struct proc
*p
= curproc
;
881 if (uap
->which
> ITIMER_PROF
)
884 if (itvp
&& (error
= copyin((caddr_t
)itvp
, (caddr_t
)&aitv
,
885 sizeof(struct itimerval
))))
887 if ((uap
->itv
= uap
->oitv
) &&
888 (error
= sys_getitimer((struct getitimer_args
*)uap
)))
892 if (itimerfix(&aitv
.it_value
))
894 if (!timevalisset(&aitv
.it_value
))
895 timevalclear(&aitv
.it_interval
);
896 else if (itimerfix(&aitv
.it_interval
))
898 lwkt_gettoken(&p
->p_token
);
899 if (uap
->which
== ITIMER_REAL
) {
900 if (timevalisset(&p
->p_realtimer
.it_value
))
901 callout_stop_sync(&p
->p_ithandle
);
902 if (timevalisset(&aitv
.it_value
))
903 callout_reset(&p
->p_ithandle
,
904 tvtohz_high(&aitv
.it_value
), realitexpire
, p
);
905 getmicrouptime(&ctv
);
906 timevaladd(&aitv
.it_value
, &ctv
);
907 p
->p_realtimer
= aitv
;
909 p
->p_timer
[uap
->which
] = aitv
;
912 p
->p_flags
&= ~P_SIGVTALRM
;
915 p
->p_flags
&= ~P_SIGPROF
;
919 lwkt_reltoken(&p
->p_token
);
924 * Real interval timer expired:
925 * send process whose timer expired an alarm signal.
926 * If time is not set up to reload, then just return.
927 * Else compute next time timer should go off which is > current time.
928 * This is where delay in processing this timeout causes multiple
929 * SIGALRM calls to be compressed into one.
930 * tvtohz_high() always adds 1 to allow for the time until the next clock
931 * interrupt being strictly less than 1 clock tick, but we don't want
932 * that here since we want to appear to be in sync with the clock
933 * interrupt even when we're delayed.
937 realitexpire(void *arg
)
940 struct timeval ctv
, ntv
;
942 p
= (struct proc
*)arg
;
944 lwkt_gettoken(&p
->p_token
);
946 if (!timevalisset(&p
->p_realtimer
.it_interval
)) {
947 timevalclear(&p
->p_realtimer
.it_value
);
951 timevaladd(&p
->p_realtimer
.it_value
,
952 &p
->p_realtimer
.it_interval
);
953 getmicrouptime(&ctv
);
954 if (timevalcmp(&p
->p_realtimer
.it_value
, &ctv
, >)) {
955 ntv
= p
->p_realtimer
.it_value
;
956 timevalsub(&ntv
, &ctv
);
957 callout_reset(&p
->p_ithandle
, tvtohz_low(&ntv
),
963 lwkt_reltoken(&p
->p_token
);
968 * Used to validate itimer timeouts and utimes*() timespecs.
971 itimerfix(struct timeval
*tv
)
973 if (tv
->tv_sec
< 0 || tv
->tv_usec
< 0 || tv
->tv_usec
>= 1000000)
975 if (tv
->tv_sec
== 0 && tv
->tv_usec
!= 0 && tv
->tv_usec
< ustick
)
976 tv
->tv_usec
= ustick
;
981 * Used to validate timeouts and utimes*() timespecs.
984 itimespecfix(struct timespec
*ts
)
986 if (ts
->tv_sec
< 0 || ts
->tv_nsec
< 0 || ts
->tv_nsec
>= 1000000000ULL)
988 if (ts
->tv_sec
== 0 && ts
->tv_nsec
!= 0 && ts
->tv_nsec
< nstick
)
989 ts
->tv_nsec
= nstick
;
994 * Decrement an interval timer by a specified number
995 * of microseconds, which must be less than a second,
996 * i.e. < 1000000. If the timer expires, then reload
997 * it. In this case, carry over (usec - old value) to
998 * reduce the value reloaded into the timer so that
999 * the timer does not drift. This routine assumes
1000 * that it is called in a context where the timers
1001 * on which it is operating cannot change in value.
1004 itimerdecr(struct itimerval
*itp
, int usec
)
1007 if (itp
->it_value
.tv_usec
< usec
) {
1008 if (itp
->it_value
.tv_sec
== 0) {
1009 /* expired, and already in next interval */
1010 usec
-= itp
->it_value
.tv_usec
;
1013 itp
->it_value
.tv_usec
+= 1000000;
1014 itp
->it_value
.tv_sec
--;
1016 itp
->it_value
.tv_usec
-= usec
;
1018 if (timevalisset(&itp
->it_value
))
1020 /* expired, exactly at end of interval */
1022 if (timevalisset(&itp
->it_interval
)) {
1023 itp
->it_value
= itp
->it_interval
;
1024 itp
->it_value
.tv_usec
-= usec
;
1025 if (itp
->it_value
.tv_usec
< 0) {
1026 itp
->it_value
.tv_usec
+= 1000000;
1027 itp
->it_value
.tv_sec
--;
1030 itp
->it_value
.tv_usec
= 0; /* sec is already 0 */
1035 * Add and subtract routines for timevals.
1036 * N.B.: subtract routine doesn't deal with
1037 * results which are before the beginning,
1038 * it just gets very confused in this case.
1042 timevaladd(struct timeval
*t1
, const struct timeval
*t2
)
1045 t1
->tv_sec
+= t2
->tv_sec
;
1046 t1
->tv_usec
+= t2
->tv_usec
;
1051 timevalsub(struct timeval
*t1
, const struct timeval
*t2
)
1054 t1
->tv_sec
-= t2
->tv_sec
;
1055 t1
->tv_usec
-= t2
->tv_usec
;
1060 timevalfix(struct timeval
*t1
)
1063 if (t1
->tv_usec
< 0) {
1065 t1
->tv_usec
+= 1000000;
1067 if (t1
->tv_usec
>= 1000000) {
1069 t1
->tv_usec
-= 1000000;
1074 * ratecheck(): simple time-based rate-limit checking.
1077 ratecheck(struct timeval
*lasttime
, const struct timeval
*mininterval
)
1079 struct timeval tv
, delta
;
1082 getmicrouptime(&tv
); /* NB: 10ms precision */
1084 timevalsub(&delta
, lasttime
);
1087 * check for 0,0 is so that the message will be seen at least once,
1088 * even if interval is huge.
1090 if (timevalcmp(&delta
, mininterval
, >=) ||
1091 (lasttime
->tv_sec
== 0 && lasttime
->tv_usec
== 0)) {
1100 * ppsratecheck(): packets (or events) per second limitation.
1102 * Return 0 if the limit is to be enforced (e.g. the caller
1103 * should drop a packet because of the rate limitation).
1105 * maxpps of 0 always causes zero to be returned. maxpps of -1
1106 * always causes 1 to be returned; this effectively defeats rate
1109 * Note that we maintain the struct timeval for compatibility
1110 * with other bsd systems. We reuse the storage and just monitor
1111 * clock ticks for minimal overhead.
1114 ppsratecheck(struct timeval
*lasttime
, int *curpps
, int maxpps
)
1119 * Reset the last time and counter if this is the first call
1120 * or more than a second has passed since the last update of
1124 if (lasttime
->tv_sec
== 0 || (u_int
)(now
- lasttime
->tv_sec
) >= hz
) {
1125 lasttime
->tv_sec
= now
;
1127 return (maxpps
!= 0);
1129 (*curpps
)++; /* NB: ignore potential overflow */
1130 return (maxpps
< 0 || *curpps
< maxpps
);