2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org>
35 * Copyright (c) 1982, 1986, 1991, 1993
36 * The Regents of the University of California. All rights reserved.
37 * (c) UNIX System Laboratories, Inc.
38 * All or some portions of this file are derived from material licensed
39 * to the University of California by American Telephone and Telegraph
40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
41 * the permission of UNIX System Laboratories, Inc.
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
72 * $FreeBSD: src/sys/kern/kern_clock.c,v 1.105.2.10 2002/10/17 13:19:40 maxim Exp $
73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.34 2005/04/14 11:15:52 joerg Exp $
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/dkstat.h>
81 #include <sys/callout.h>
82 #include <sys/kernel.h>
83 #include <sys/kinfo.h>
85 #include <sys/malloc.h>
86 #include <sys/resourcevar.h>
87 #include <sys/signalvar.h>
88 #include <sys/timex.h>
89 #include <sys/timepps.h>
93 #include <vm/vm_map.h>
94 #include <sys/sysctl.h>
95 #include <sys/thread2.h>
97 #include <machine/cpu.h>
98 #include <machine/limits.h>
99 #include <machine/smp.h>
102 #include <sys/gmon.h>
105 #ifdef DEVICE_POLLING
106 extern void init_device_poll(void);
107 extern void hardclock_device_poll(void);
108 #endif /* DEVICE_POLLING */
110 static void initclocks (void *dummy
);
111 SYSINIT(clocks
, SI_SUB_CLOCKS
, SI_ORDER_FIRST
, initclocks
, NULL
)
114 * Some of these don't belong here, but it's easiest to concentrate them.
115 * Note that cp_time counts in microseconds, but most userland programs
116 * just compare relative times against the total by delta.
118 struct cp_time cp_time
;
120 SYSCTL_OPAQUE(_kern
, OID_AUTO
, cp_time
, CTLFLAG_RD
, &cp_time
, sizeof(cp_time
),
121 "LU", "CPU time statistics");
124 * boottime is used to calculate the 'real' uptime. Do not confuse this with
125 * microuptime(). microtime() is not drift compensated. The real uptime
126 * with compensation is nanotime() - bootime. boottime is recalculated
127 * whenever the real time is set based on the compensated elapsed time
128 * in seconds (gd->gd_time_seconds).
130 * basetime is used to calculate the compensated real time of day. Chunky
131 * changes to the time, aka settimeofday(), are made by modifying basetime.
133 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
134 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
137 struct timespec boottime
; /* boot time (realtime) for reference only */
138 struct timespec basetime
; /* base time adjusts uptime -> realtime */
139 time_t time_second
; /* read-only 'passive' uptime in seconds */
141 SYSCTL_STRUCT(_kern
, KERN_BOOTTIME
, boottime
, CTLFLAG_RD
,
142 &boottime
, timeval
, "System boottime");
143 SYSCTL_STRUCT(_kern
, OID_AUTO
, basetime
, CTLFLAG_RD
,
144 &basetime
, timeval
, "System basetime");
146 static void hardclock(systimer_t info
, struct intrframe
*frame
);
147 static void statclock(systimer_t info
, struct intrframe
*frame
);
148 static void schedclock(systimer_t info
, struct intrframe
*frame
);
150 int ticks
; /* system master ticks at hz */
151 int clocks_running
; /* tsleep/timeout clocks operational */
152 int64_t nsec_adj
; /* ntpd per-tick adjustment in nsec << 32 */
153 int64_t nsec_acc
; /* accumulator */
155 /* NTPD time correction fields */
156 int64_t ntp_tick_permanent
; /* per-tick adjustment in nsec << 32 */
157 int64_t ntp_tick_acc
; /* accumulator for per-tick adjustment */
158 int64_t ntp_delta
; /* one-time correction in nsec */
159 int64_t ntp_big_delta
= 1000000000;
160 int32_t ntp_tick_delta
; /* current adjustment rate */
161 int32_t ntp_default_tick_delta
; /* adjustment rate for ntp_delta */
162 time_t ntp_leap_second
; /* time of next leap second */
163 int ntp_leap_insert
; /* whether to insert or remove a second */
166 * Finish initializing clock frequencies and start all clocks running.
170 initclocks(void *dummy
)
173 #ifdef DEVICE_POLLING
176 /*psratio = profhz / stathz;*/
182 * Called on a per-cpu basis
185 initclocks_pcpu(void)
187 struct globaldata
*gd
= mycpu
;
190 if (gd
->gd_cpuid
== 0) {
191 gd
->gd_time_seconds
= 1;
192 gd
->gd_cpuclock_base
= cputimer_count();
195 gd
->gd_time_seconds
= globaldata_find(0)->gd_time_seconds
;
196 gd
->gd_cpuclock_base
= globaldata_find(0)->gd_cpuclock_base
;
200 * Use a non-queued periodic systimer to prevent multiple ticks from
201 * building up if the sysclock jumps forward (8254 gets reset). The
202 * sysclock will never jump backwards. Our time sync is based on
203 * the actual sysclock, not the ticks count.
205 systimer_init_periodic_nq(&gd
->gd_hardclock
, hardclock
, NULL
, hz
);
206 systimer_init_periodic_nq(&gd
->gd_statclock
, statclock
, NULL
, stathz
);
207 /* XXX correct the frequency for scheduler / estcpu tests */
208 systimer_init_periodic_nq(&gd
->gd_schedclock
, schedclock
,
214 * This sets the current real time of day. Timespecs are in seconds and
215 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
216 * instead we adjust basetime so basetime + gd_* results in the current
217 * time of day. This way the gd_* fields are guarenteed to represent
218 * a monotonically increasing 'uptime' value.
221 set_timeofday(struct timespec
*ts
)
226 * XXX SMP / non-atomic basetime updates
230 basetime
.tv_sec
= ts
->tv_sec
- ts2
.tv_sec
;
231 basetime
.tv_nsec
= ts
->tv_nsec
- ts2
.tv_nsec
;
232 if (basetime
.tv_nsec
< 0) {
233 basetime
.tv_nsec
+= 1000000000;
238 * Note that basetime diverges from boottime as the clock drift is
239 * compensated for, so we cannot do away with boottime. When setting
240 * the absolute time of day the drift is 0 (for an instant) and we
241 * can simply assign boottime to basetime.
243 * Note that nanouptime() is based on gd_time_seconds which is drift
244 * compensated up to a point (it is guarenteed to remain monotonically
245 * increasing). gd_time_seconds is thus our best uptime guess and
246 * suitable for use in the boottime calculation. It is already taken
247 * into account in the basetime calculation above.
249 boottime
.tv_sec
= basetime
.tv_sec
;
255 * Each cpu has its own hardclock, but we only increments ticks and softticks
258 * NOTE! systimer! the MP lock might not be held here. We can only safely
259 * manipulate objects owned by the current cpu.
262 hardclock(systimer_t info
, struct intrframe
*frame
)
266 struct pstats
*pstats
;
267 struct globaldata
*gd
= mycpu
;
270 * Realtime updates are per-cpu. Note that timer corrections as
271 * returned by microtime() and friends make an additional adjustment
272 * using a system-wise 'basetime', but the running time is always
273 * taken from the per-cpu globaldata area. Since the same clock
274 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
277 * Note that we never allow info->time (aka gd->gd_hardclock.time)
278 * to reverse index gd_cpuclock_base, but that it is possible for
279 * it to temporarily get behind in the seconds if something in the
280 * system locks interrupts for a long period of time. Since periodic
281 * timers count events, though everything should resynch again
284 cputicks
= info
->time
- gd
->gd_cpuclock_base
;
285 if (cputicks
>= cputimer_freq
) {
286 ++gd
->gd_time_seconds
;
287 gd
->gd_cpuclock_base
+= cputimer_freq
;
291 * The system-wide ticks counter and NTP related timedelta/tickdelta
292 * adjustments only occur on cpu #0. NTP adjustments are accomplished
293 * by updating basetime.
295 if (gd
->gd_cpuid
== 0) {
301 #ifdef DEVICE_POLLING
302 hardclock_device_poll(); /* mpsafe, short and quick */
303 #endif /* DEVICE_POLLING */
306 if (tco
->tc_poll_pps
)
307 tco
->tc_poll_pps(tco
);
310 * Apply adjtime corrections. At the moment only do this if
311 * we can get the MP lock to interlock with adjtime's modification
312 * of these variables. Note that basetime adjustments are not
313 * MP safe either XXX.
315 if (ntp_delta
!= 0) {
316 basetime
.tv_nsec
+= ntp_tick_delta
;
317 ntp_delta
-= ntp_tick_delta
;
318 if ((ntp_delta
> 0 && ntp_delta
< ntp_tick_delta
) ||
319 (ntp_delta
< 0 && ntp_delta
> ntp_tick_delta
)) {
320 ntp_tick_delta
= ntp_delta
;
324 if (ntp_tick_permanent
!= 0) {
325 ntp_tick_acc
+= ntp_tick_permanent
;
326 if (ntp_tick_acc
>= (1LL << 32)) {
327 basetime
.tv_nsec
+= (-ntp_tick_acc
) >> 32;
328 ntp_tick_acc
&= (1LL << 32) - 1;
329 } else if (ntp_tick_acc
<= -(1LL << 32)) {
330 basetime
.tv_nsec
-= (-ntp_tick_acc
) >> 32;
331 ntp_tick_acc
= -((-ntp_tick_acc
) & ((1LL << 32) - 1));
335 if (basetime
.tv_nsec
>= 1000000000) {
337 basetime
.tv_nsec
-= 1000000000;
338 } else if (basetime
.tv_nsec
< 0) {
340 basetime
.tv_nsec
+= 1000000000;
343 if (ntp_leap_second
) {
347 if (ntp_leap_second
== tsp
.tv_sec
) {
357 * Apply per-tick compensation. ticks_adj adjusts for both
358 * offset and frequency, and could be negative.
360 if (nsec_adj
!= 0 && try_mplock()) {
361 nsec_acc
+= nsec_adj
;
362 if (nsec_acc
>= 0x100000000LL
) {
363 basetime
.tv_nsec
+= nsec_acc
>> 32;
364 nsec_acc
= (nsec_acc
& 0xFFFFFFFFLL
);
365 } else if (nsec_acc
<= -0x100000000LL
) {
366 basetime
.tv_nsec
-= -nsec_acc
>> 32;
367 nsec_acc
= -(-nsec_acc
& 0xFFFFFFFFLL
);
369 if (basetime
.tv_nsec
>= 1000000000) {
370 basetime
.tv_nsec
-= 1000000000;
372 } else if (basetime
.tv_nsec
< 0) {
373 basetime
.tv_nsec
+= 1000000000;
380 * If the realtime-adjusted seconds hand rolls over then tell
381 * ntp_update_second() what we did in the last second so it can
382 * calculate what to do in the next second. It may also add
383 * or subtract a leap second.
386 if (time_second
!= nts
.tv_sec
) {
387 leap
= ntp_update_second(time_second
, &nsec_adj
);
388 basetime
.tv_sec
+= leap
;
389 time_second
= nts
.tv_sec
+ leap
;
395 * softticks are handled for all cpus
397 hardclock_softtick(gd
);
400 * ITimer handling is per-tick, per-cpu. I don't think psignal()
401 * is mpsafe on curproc, so XXX get the mplock.
403 if ((p
= curproc
) != NULL
&& try_mplock()) {
405 if (frame
&& CLKF_USERMODE(frame
) &&
406 timevalisset(&pstats
->p_timer
[ITIMER_VIRTUAL
].it_value
) &&
407 itimerdecr(&pstats
->p_timer
[ITIMER_VIRTUAL
], tick
) == 0)
408 psignal(p
, SIGVTALRM
);
409 if (timevalisset(&pstats
->p_timer
[ITIMER_PROF
].it_value
) &&
410 itimerdecr(&pstats
->p_timer
[ITIMER_PROF
], tick
) == 0)
418 * The statistics clock typically runs at a 125Hz rate, and is intended
419 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
421 * NOTE! systimer! the MP lock might not be held here. We can only safely
422 * manipulate objects owned by the current cpu.
424 * The stats clock is responsible for grabbing a profiling sample.
425 * Most of the statistics are only used by user-level statistics programs.
426 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
429 * Like the other clocks, the stat clock is called from what is effectively
430 * a fast interrupt, so the context should be the thread/process that got
434 statclock(systimer_t info
, struct intrframe
*frame
)
447 * How big was our timeslice relative to the last time?
449 microuptime(&tv
); /* mpsafe */
450 stv
= &mycpu
->gd_stattv
;
451 if (stv
->tv_sec
== 0) {
454 bump
= tv
.tv_usec
- stv
->tv_usec
+
455 (tv
.tv_sec
- stv
->tv_sec
) * 1000000;
466 if (frame
&& CLKF_USERMODE(frame
)) {
468 * Came from userland, handle user time and deal with
471 if (p
&& (p
->p_flag
& P_PROFIL
))
472 addupc_intr(p
, CLKF_PC(frame
), 1);
473 td
->td_uticks
+= bump
;
476 * Charge the time as appropriate
478 if (p
&& p
->p_nice
> NZERO
)
479 cp_time
.cp_nice
+= bump
;
481 cp_time
.cp_user
+= bump
;
485 * Kernel statistics are just like addupc_intr, only easier.
488 if (g
->state
== GMON_PROF_ON
&& frame
) {
489 i
= CLKF_PC(frame
) - g
->lowpc
;
490 if (i
< g
->textsize
) {
491 i
/= HISTFRACTION
* sizeof(*g
->kcount
);
497 * Came from kernel mode, so we were:
498 * - handling an interrupt,
499 * - doing syscall or trap work on behalf of the current
501 * - spinning in the idle loop.
502 * Whichever it is, charge the time as appropriate.
503 * Note that we charge interrupts to the current process,
504 * regardless of whether they are ``for'' that process,
505 * so that we know how much of its real time was spent
506 * in ``non-process'' (i.e., interrupt) work.
508 * XXX assume system if frame is NULL. A NULL frame
509 * can occur if ipi processing is done from an splx().
511 if (frame
&& CLKF_INTR(frame
))
512 td
->td_iticks
+= bump
;
514 td
->td_sticks
+= bump
;
516 if (frame
&& CLKF_INTR(frame
)) {
517 cp_time
.cp_intr
+= bump
;
519 if (td
== &mycpu
->gd_idlethread
)
520 cp_time
.cp_idle
+= bump
;
522 cp_time
.cp_sys
+= bump
;
528 * The scheduler clock typically runs at a 20Hz rate. NOTE! systimer,
529 * the MP lock might not be held. We can safely manipulate parts of curproc
530 * but that's about it.
533 schedclock(systimer_t info
, struct intrframe
*frame
)
536 struct pstats
*pstats
;
541 schedulerclock(NULL
); /* mpsafe */
542 if ((p
= curproc
) != NULL
) {
543 /* Update resource usage integrals and maximums. */
544 if ((pstats
= p
->p_stats
) != NULL
&&
545 (ru
= &pstats
->p_ru
) != NULL
&&
546 (vm
= p
->p_vmspace
) != NULL
) {
547 ru
->ru_ixrss
+= pgtok(vm
->vm_tsize
);
548 ru
->ru_idrss
+= pgtok(vm
->vm_dsize
);
549 ru
->ru_isrss
+= pgtok(vm
->vm_ssize
);
550 rss
= pgtok(vmspace_resident_count(vm
));
551 if (ru
->ru_maxrss
< rss
)
558 * Compute number of ticks for the specified amount of time. The
559 * return value is intended to be used in a clock interrupt timed
560 * operation and guarenteed to meet or exceed the requested time.
561 * If the representation overflows, return INT_MAX. The minimum return
562 * value is 1 ticks and the function will average the calculation up.
563 * If any value greater then 0 microseconds is supplied, a value
564 * of at least 2 will be returned to ensure that a near-term clock
565 * interrupt does not cause the timeout to occur (degenerately) early.
567 * Note that limit checks must take into account microseconds, which is
568 * done simply by using the smaller signed long maximum instead of
569 * the unsigned long maximum.
571 * If ints have 32 bits, then the maximum value for any timeout in
572 * 10ms ticks is 248 days.
575 tvtohz_high(struct timeval
*tv
)
592 printf("tvotohz: negative time difference %ld sec %ld usec\n",
596 } else if (sec
<= INT_MAX
/ hz
) {
597 ticks
= (int)(sec
* hz
+
598 ((u_long
)usec
+ (tick
- 1)) / tick
) + 1;
606 * Compute number of ticks for the specified amount of time, erroring on
607 * the side of it being too low to ensure that sleeping the returned number
608 * of ticks will not result in a late return.
610 * The supplied timeval may not be negative and should be normalized. A
611 * return value of 0 is possible if the timeval converts to less then
614 * If ints have 32 bits, then the maximum value for any timeout in
615 * 10ms ticks is 248 days.
618 tvtohz_low(struct timeval
*tv
)
624 if (sec
<= INT_MAX
/ hz
)
625 ticks
= (int)(sec
* hz
+ (u_long
)tv
->tv_usec
/ tick
);
633 * Start profiling on a process.
635 * Kernel profiling passes proc0 which never exits and hence
636 * keeps the profile clock running constantly.
639 startprofclock(struct proc
*p
)
641 if ((p
->p_flag
& P_PROFIL
) == 0) {
642 p
->p_flag
|= P_PROFIL
;
644 if (++profprocs
== 1 && stathz
!= 0) {
647 setstatclockrate(profhz
);
655 * Stop profiling on a process.
658 stopprofclock(struct proc
*p
)
660 if (p
->p_flag
& P_PROFIL
) {
661 p
->p_flag
&= ~P_PROFIL
;
663 if (--profprocs
== 0 && stathz
!= 0) {
666 setstatclockrate(stathz
);
674 * Return information about system clocks.
677 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS
)
679 struct kinfo_clockinfo clkinfo
;
681 * Construct clockinfo structure.
684 clkinfo
.ci_tick
= tick
;
685 clkinfo
.ci_tickadj
= ntp_default_tick_delta
/ 1000;
686 clkinfo
.ci_profhz
= profhz
;
687 clkinfo
.ci_stathz
= stathz
? stathz
: hz
;
688 return (sysctl_handle_opaque(oidp
, &clkinfo
, sizeof clkinfo
, req
));
691 SYSCTL_PROC(_kern
, KERN_CLOCKRATE
, clockrate
, CTLTYPE_STRUCT
|CTLFLAG_RD
,
692 0, 0, sysctl_kern_clockrate
, "S,clockinfo","");
695 * We have eight functions for looking at the clock, four for
696 * microseconds and four for nanoseconds. For each there is fast
697 * but less precise version "get{nano|micro}[up]time" which will
698 * return a time which is up to 1/HZ previous to the call, whereas
699 * the raw version "{nano|micro}[up]time" will return a timestamp
700 * which is as precise as possible. The "up" variants return the
701 * time relative to system boot, these are well suited for time
702 * interval measurements.
704 * Each cpu independantly maintains the current time of day, so all
705 * we need to do to protect ourselves from changes is to do a loop
706 * check on the seconds field changing out from under us.
708 * The system timer maintains a 32 bit count and due to various issues
709 * it is possible for the calculated delta to occassionally exceed
710 * cputimer_freq. If this occurs the cputimer_freq64_nsec multiplication
711 * can easily overflow, so we deal with the case. For uniformity we deal
712 * with the case in the usec case too.
715 getmicrouptime(struct timeval
*tvp
)
717 struct globaldata
*gd
= mycpu
;
721 tvp
->tv_sec
= gd
->gd_time_seconds
;
722 delta
= gd
->gd_hardclock
.time
- gd
->gd_cpuclock_base
;
723 } while (tvp
->tv_sec
!= gd
->gd_time_seconds
);
725 if (delta
>= cputimer_freq
) {
726 tvp
->tv_sec
+= delta
/ cputimer_freq
;
727 delta
%= cputimer_freq
;
729 tvp
->tv_usec
= (cputimer_freq64_usec
* delta
) >> 32;
730 if (tvp
->tv_usec
>= 1000000) {
731 tvp
->tv_usec
-= 1000000;
737 getnanouptime(struct timespec
*tsp
)
739 struct globaldata
*gd
= mycpu
;
743 tsp
->tv_sec
= gd
->gd_time_seconds
;
744 delta
= gd
->gd_hardclock
.time
- gd
->gd_cpuclock_base
;
745 } while (tsp
->tv_sec
!= gd
->gd_time_seconds
);
747 if (delta
>= cputimer_freq
) {
748 tsp
->tv_sec
+= delta
/ cputimer_freq
;
749 delta
%= cputimer_freq
;
751 tsp
->tv_nsec
= (cputimer_freq64_nsec
* delta
) >> 32;
755 microuptime(struct timeval
*tvp
)
757 struct globaldata
*gd
= mycpu
;
761 tvp
->tv_sec
= gd
->gd_time_seconds
;
762 delta
= cputimer_count() - gd
->gd_cpuclock_base
;
763 } while (tvp
->tv_sec
!= gd
->gd_time_seconds
);
765 if (delta
>= cputimer_freq
) {
766 tvp
->tv_sec
+= delta
/ cputimer_freq
;
767 delta
%= cputimer_freq
;
769 tvp
->tv_usec
= (cputimer_freq64_usec
* delta
) >> 32;
773 nanouptime(struct timespec
*tsp
)
775 struct globaldata
*gd
= mycpu
;
779 tsp
->tv_sec
= gd
->gd_time_seconds
;
780 delta
= cputimer_count() - gd
->gd_cpuclock_base
;
781 } while (tsp
->tv_sec
!= gd
->gd_time_seconds
);
783 if (delta
>= cputimer_freq
) {
784 tsp
->tv_sec
+= delta
/ cputimer_freq
;
785 delta
%= cputimer_freq
;
787 tsp
->tv_nsec
= (cputimer_freq64_nsec
* delta
) >> 32;
795 getmicrotime(struct timeval
*tvp
)
797 struct globaldata
*gd
= mycpu
;
801 tvp
->tv_sec
= gd
->gd_time_seconds
;
802 delta
= gd
->gd_hardclock
.time
- gd
->gd_cpuclock_base
;
803 } while (tvp
->tv_sec
!= gd
->gd_time_seconds
);
805 if (delta
>= cputimer_freq
) {
806 tvp
->tv_sec
+= delta
/ cputimer_freq
;
807 delta
%= cputimer_freq
;
809 tvp
->tv_usec
= (cputimer_freq64_usec
* delta
) >> 32;
811 tvp
->tv_sec
+= basetime
.tv_sec
;
812 tvp
->tv_usec
+= basetime
.tv_nsec
/ 1000;
813 while (tvp
->tv_usec
>= 1000000) {
814 tvp
->tv_usec
-= 1000000;
820 getnanotime(struct timespec
*tsp
)
822 struct globaldata
*gd
= mycpu
;
826 tsp
->tv_sec
= gd
->gd_time_seconds
;
827 delta
= gd
->gd_hardclock
.time
- gd
->gd_cpuclock_base
;
828 } while (tsp
->tv_sec
!= gd
->gd_time_seconds
);
830 if (delta
>= cputimer_freq
) {
831 tsp
->tv_sec
+= delta
/ cputimer_freq
;
832 delta
%= cputimer_freq
;
834 tsp
->tv_nsec
= (cputimer_freq64_nsec
* delta
) >> 32;
836 tsp
->tv_sec
+= basetime
.tv_sec
;
837 tsp
->tv_nsec
+= basetime
.tv_nsec
;
838 while (tsp
->tv_nsec
>= 1000000000) {
839 tsp
->tv_nsec
-= 1000000000;
845 microtime(struct timeval
*tvp
)
847 struct globaldata
*gd
= mycpu
;
851 tvp
->tv_sec
= gd
->gd_time_seconds
;
852 delta
= cputimer_count() - gd
->gd_cpuclock_base
;
853 } while (tvp
->tv_sec
!= gd
->gd_time_seconds
);
855 if (delta
>= cputimer_freq
) {
856 tvp
->tv_sec
+= delta
/ cputimer_freq
;
857 delta
%= cputimer_freq
;
859 tvp
->tv_usec
= (cputimer_freq64_usec
* delta
) >> 32;
861 tvp
->tv_sec
+= basetime
.tv_sec
;
862 tvp
->tv_usec
+= basetime
.tv_nsec
/ 1000;
863 while (tvp
->tv_usec
>= 1000000) {
864 tvp
->tv_usec
-= 1000000;
870 nanotime(struct timespec
*tsp
)
872 struct globaldata
*gd
= mycpu
;
876 tsp
->tv_sec
= gd
->gd_time_seconds
;
877 delta
= cputimer_count() - gd
->gd_cpuclock_base
;
878 } while (tsp
->tv_sec
!= gd
->gd_time_seconds
);
880 if (delta
>= cputimer_freq
) {
881 tsp
->tv_sec
+= delta
/ cputimer_freq
;
882 delta
%= cputimer_freq
;
884 tsp
->tv_nsec
= (cputimer_freq64_nsec
* delta
) >> 32;
886 tsp
->tv_sec
+= basetime
.tv_sec
;
887 tsp
->tv_nsec
+= basetime
.tv_nsec
;
888 while (tsp
->tv_nsec
>= 1000000000) {
889 tsp
->tv_nsec
-= 1000000000;
895 * note: this is not exactly synchronized with real time. To do that we
896 * would have to do what microtime does and check for a nanoseconds overflow.
899 get_approximate_time_t(void)
901 struct globaldata
*gd
= mycpu
;
902 return(gd
->gd_time_seconds
+ basetime
.tv_sec
);
906 pps_ioctl(u_long cmd
, caddr_t data
, struct pps_state
*pps
)
909 struct pps_fetch_args
*fapi
;
911 struct pps_kcbind_args
*kapi
;
917 case PPS_IOC_DESTROY
:
919 case PPS_IOC_SETPARAMS
:
920 app
= (pps_params_t
*)data
;
921 if (app
->mode
& ~pps
->ppscap
)
923 pps
->ppsparam
= *app
;
925 case PPS_IOC_GETPARAMS
:
926 app
= (pps_params_t
*)data
;
927 *app
= pps
->ppsparam
;
928 app
->api_version
= PPS_API_VERS_1
;
931 *(int*)data
= pps
->ppscap
;
934 fapi
= (struct pps_fetch_args
*)data
;
935 if (fapi
->tsformat
&& fapi
->tsformat
!= PPS_TSFMT_TSPEC
)
937 if (fapi
->timeout
.tv_sec
|| fapi
->timeout
.tv_nsec
)
939 pps
->ppsinfo
.current_mode
= pps
->ppsparam
.mode
;
940 fapi
->pps_info_buf
= pps
->ppsinfo
;
944 kapi
= (struct pps_kcbind_args
*)data
;
945 /* XXX Only root should be able to do this */
946 if (kapi
->tsformat
&& kapi
->tsformat
!= PPS_TSFMT_TSPEC
)
948 if (kapi
->kernel_consumer
!= PPS_KC_HARDPPS
)
950 if (kapi
->edge
& ~pps
->ppscap
)
952 pps
->kcmode
= kapi
->edge
;
963 pps_init(struct pps_state
*pps
)
965 pps
->ppscap
|= PPS_TSFMT_TSPEC
;
966 if (pps
->ppscap
& PPS_CAPTUREASSERT
)
967 pps
->ppscap
|= PPS_OFFSETASSERT
;
968 if (pps
->ppscap
& PPS_CAPTURECLEAR
)
969 pps
->ppscap
|= PPS_OFFSETCLEAR
;
973 pps_event(struct pps_state
*pps
, sysclock_t count
, int event
)
975 struct globaldata
*gd
;
976 struct timespec
*tsp
;
977 struct timespec
*osp
;
990 /* Things would be easier with arrays... */
991 if (event
== PPS_CAPTUREASSERT
) {
992 tsp
= &pps
->ppsinfo
.assert_timestamp
;
993 osp
= &pps
->ppsparam
.assert_offset
;
994 foff
= pps
->ppsparam
.mode
& PPS_OFFSETASSERT
;
995 fhard
= pps
->kcmode
& PPS_CAPTUREASSERT
;
996 pcount
= &pps
->ppscount
[0];
997 pseq
= &pps
->ppsinfo
.assert_sequence
;
999 tsp
= &pps
->ppsinfo
.clear_timestamp
;
1000 osp
= &pps
->ppsparam
.clear_offset
;
1001 foff
= pps
->ppsparam
.mode
& PPS_OFFSETCLEAR
;
1002 fhard
= pps
->kcmode
& PPS_CAPTURECLEAR
;
1003 pcount
= &pps
->ppscount
[1];
1004 pseq
= &pps
->ppsinfo
.clear_sequence
;
1007 /* Nothing really happened */
1008 if (*pcount
== count
)
1014 ts
.tv_sec
= gd
->gd_time_seconds
;
1015 delta
= count
- gd
->gd_cpuclock_base
;
1016 } while (ts
.tv_sec
!= gd
->gd_time_seconds
);
1018 if (delta
>= cputimer_freq
) {
1019 ts
.tv_sec
+= delta
/ cputimer_freq
;
1020 delta
%= cputimer_freq
;
1022 ts
.tv_nsec
= (cputimer_freq64_nsec
* delta
) >> 32;
1023 ts
.tv_sec
+= basetime
.tv_sec
;
1024 ts
.tv_nsec
+= basetime
.tv_nsec
;
1025 while (ts
.tv_nsec
>= 1000000000) {
1026 ts
.tv_nsec
-= 1000000000;
1034 timespecadd(tsp
, osp
);
1035 if (tsp
->tv_nsec
< 0) {
1036 tsp
->tv_nsec
+= 1000000000;
1042 /* magic, at its best... */
1043 tcount
= count
- pps
->ppscount
[2];
1044 pps
->ppscount
[2] = count
;
1045 if (tcount
>= cputimer_freq
) {
1046 delta
= (1000000000 * (tcount
/ cputimer_freq
) +
1047 cputimer_freq64_nsec
*
1048 (tcount
% cputimer_freq
)) >> 32;
1050 delta
= (cputimer_freq64_nsec
* tcount
) >> 32;
1052 hardpps(tsp
, delta
);