Handle window updates inside header prediction to increase the hit rate.
[dragonfly.git] / sys / kern / kern_clock.c
CommitLineData
8c10bfcf
MD
1/*
2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
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
16 * distribution.
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.
20 *
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
32 * SUCH DAMAGE.
33 *
984263bc
MD
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.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
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.
58 *
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
69 * SUCH DAMAGE.
70 *
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 $
0d1dffdf 73 * $DragonFly: src/sys/kern/kern_clock.c,v 1.27 2004/11/20 20:25:09 dillon Exp $
984263bc
MD
74 */
75
76#include "opt_ntp.h"
77
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/proc.h>
84#include <sys/malloc.h>
85#include <sys/resourcevar.h>
86#include <sys/signalvar.h>
87#include <sys/timex.h>
88#include <sys/timepps.h>
89#include <vm/vm.h>
90#include <sys/lock.h>
91#include <vm/pmap.h>
92#include <vm/vm_map.h>
93#include <sys/sysctl.h>
2689779e 94#include <sys/thread2.h>
984263bc
MD
95
96#include <machine/cpu.h>
97#include <machine/limits.h>
98#include <machine/smp.h>
99
100#ifdef GPROF
101#include <sys/gmon.h>
102#endif
103
104#ifdef DEVICE_POLLING
105extern void init_device_poll(void);
106extern void hardclock_device_poll(void);
107#endif /* DEVICE_POLLING */
108
402ed7e1 109static void initclocks (void *dummy);
984263bc
MD
110SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
111
6ad39cae
MD
112/*
113 * Some of these don't belong here, but it's easiest to concentrate them.
114 * Note that cp_time[] counts in microseconds, but most userland programs
115 * just compare relative times against the total by delta.
116 */
984263bc
MD
117long cp_time[CPUSTATES];
118
119SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time),
120 "LU", "CPU time statistics");
121
122long tk_cancc;
123long tk_nin;
124long tk_nout;
125long tk_rawcc;
126
88c4d2f6
MD
127/*
128 * boottime is used to calculate the 'real' uptime. Do not confuse this with
129 * microuptime(). microtime() is not drift compensated. The real uptime
60b2809b
MD
130 * with compensation is nanotime() - bootime. boottime is recalculated
131 * whenever the real time is set based on the compensated elapsed time
132 * in seconds (gd->gd_time_seconds).
88c4d2f6
MD
133 *
134 * basetime is used to calculate the compensated real time of day. Chunky
135 * changes to the time, aka settimeofday(), are made by modifying basetime.
136 *
137 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
138 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
139 * the real time.
140 */
141struct timespec boottime; /* boot time (realtime) for reference only */
142struct timespec basetime; /* base time adjusts uptime -> realtime */
143time_t time_second; /* read-only 'passive' uptime in seconds */
984263bc 144
984263bc
MD
145SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
146 &boottime, timeval, "System boottime");
88c4d2f6
MD
147SYSCTL_STRUCT(_kern, OID_AUTO, basetime, CTLFLAG_RD,
148 &basetime, timeval, "System basetime");
984263bc 149
88c4d2f6
MD
150static void hardclock(systimer_t info, struct intrframe *frame);
151static void statclock(systimer_t info, struct intrframe *frame);
152static void schedclock(systimer_t info, struct intrframe *frame);
153
154int ticks; /* system master ticks at hz */
da3639ef 155int clocks_running; /* tsleep/timeout clocks operational */
88c4d2f6
MD
156int64_t nsec_adj; /* ntpd per-tick adjustment in nsec << 32 */
157int64_t nsec_acc; /* accumulator */
984263bc
MD
158
159/*
88c4d2f6 160 * Finish initializing clock frequencies and start all clocks running.
984263bc 161 */
88c4d2f6
MD
162/* ARGSUSED*/
163static void
164initclocks(void *dummy)
984263bc 165{
88c4d2f6
MD
166 cpu_initclocks();
167#ifdef DEVICE_POLLING
168 init_device_poll();
169#endif
170 /*psratio = profhz / stathz;*/
171 initclocks_pcpu();
da3639ef 172 clocks_running = 1;
984263bc
MD
173}
174
88c4d2f6
MD
175/*
176 * Called on a per-cpu basis
177 */
178void
179initclocks_pcpu(void)
180{
181 struct globaldata *gd = mycpu;
984263bc 182
88c4d2f6
MD
183 crit_enter();
184 if (gd->gd_cpuid == 0) {
185 gd->gd_time_seconds = 1;
186 gd->gd_cpuclock_base = cputimer_count();
187 } else {
188 /* XXX */
189 gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
190 gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
191 }
0d1dffdf
MD
192
193 /*
194 * Use a non-queued periodic systimer to prevent multiple ticks from
195 * building up if the sysclock jumps forward (8254 gets reset). The
196 * sysclock will never jump backwards. Our time sync is based on
197 * the actual sysclock, not the ticks count.
198 */
199 systimer_init_periodic_nq(&gd->gd_hardclock, hardclock, NULL, hz);
200 systimer_init_periodic_nq(&gd->gd_statclock, statclock, NULL, stathz);
88c4d2f6 201 /* XXX correct the frequency for scheduler / estcpu tests */
0d1dffdf 202 systimer_init_periodic_nq(&gd->gd_schedclock, schedclock,
8478264a 203 NULL, ESTCPUFREQ);
88c4d2f6
MD
204 crit_exit();
205}
984263bc
MD
206
207/*
88c4d2f6
MD
208 * This sets the current real time of day. Timespecs are in seconds and
209 * nanoseconds. We do not mess with gd_time_seconds and gd_cpuclock_base,
210 * instead we adjust basetime so basetime + gd_* results in the current
211 * time of day. This way the gd_* fields are guarenteed to represent
212 * a monotonically increasing 'uptime' value.
984263bc 213 */
88c4d2f6
MD
214void
215set_timeofday(struct timespec *ts)
216{
217 struct timespec ts2;
984263bc 218
88c4d2f6
MD
219 /*
220 * XXX SMP / non-atomic basetime updates
221 */
222 crit_enter();
223 nanouptime(&ts2);
224 basetime.tv_sec = ts->tv_sec - ts2.tv_sec;
225 basetime.tv_nsec = ts->tv_nsec - ts2.tv_nsec;
226 if (basetime.tv_nsec < 0) {
227 basetime.tv_nsec += 1000000000;
228 --basetime.tv_sec;
229 }
60b2809b 230 boottime.tv_sec = basetime.tv_sec - mycpu->gd_time_seconds;
88c4d2f6
MD
231 timedelta = 0;
232 crit_exit();
233}
234
984263bc 235/*
88c4d2f6
MD
236 * Each cpu has its own hardclock, but we only increments ticks and softticks
237 * on cpu #0.
238 *
239 * NOTE! systimer! the MP lock might not be held here. We can only safely
240 * manipulate objects owned by the current cpu.
984263bc 241 */
984263bc 242static void
88c4d2f6 243hardclock(systimer_t info, struct intrframe *frame)
984263bc 244{
88c4d2f6
MD
245 sysclock_t cputicks;
246 struct proc *p;
247 struct pstats *pstats;
248 struct globaldata *gd = mycpu;
984263bc
MD
249
250 /*
88c4d2f6
MD
251 * Realtime updates are per-cpu. Note that timer corrections as
252 * returned by microtime() and friends make an additional adjustment
253 * using a system-wise 'basetime', but the running time is always
254 * taken from the per-cpu globaldata area. Since the same clock
255 * is distributing (XXX SMP) to all cpus, the per-cpu timebases
256 * stay in synch.
257 *
258 * Note that we never allow info->time (aka gd->gd_hardclock.time)
fad57d0e
MD
259 * to reverse index gd_cpuclock_base, but that it is possible for
260 * it to temporarily get behind in the seconds if something in the
261 * system locks interrupts for a long period of time. Since periodic
262 * timers count events, though everything should resynch again
263 * immediately.
984263bc 264 */
88c4d2f6 265 cputicks = info->time - gd->gd_cpuclock_base;
fad57d0e 266 if (cputicks >= cputimer_freq) {
88c4d2f6
MD
267 ++gd->gd_time_seconds;
268 gd->gd_cpuclock_base += cputimer_freq;
269 }
984263bc
MD
270
271 /*
92b561b7
MD
272 * The system-wide ticks counter and NTP related timedelta/tickdelta
273 * adjustments only occur on cpu #0. NTP adjustments are accomplished
274 * by updating basetime.
984263bc 275 */
88c4d2f6
MD
276 if (gd->gd_cpuid == 0) {
277 struct timespec nts;
278 int leap;
984263bc 279
88c4d2f6 280 ++ticks;
984263bc 281
88c4d2f6
MD
282#ifdef DEVICE_POLLING
283 hardclock_device_poll(); /* mpsafe, short and quick */
284#endif /* DEVICE_POLLING */
984263bc 285
88c4d2f6
MD
286#if 0
287 if (tco->tc_poll_pps)
288 tco->tc_poll_pps(tco);
289#endif
290 /*
291 * Apply adjtime corrections. At the moment only do this if
292 * we can get the MP lock to interlock with adjtime's modification
293 * of these variables. Note that basetime adjustments are not
294 * MP safe either XXX.
295 */
296 if (timedelta != 0 && try_mplock()) {
297 basetime.tv_nsec += tickdelta * 1000;
298 if (basetime.tv_nsec >= 1000000000) {
299 basetime.tv_nsec -= 1000000000;
300 ++basetime.tv_sec;
301 } else if (basetime.tv_nsec < 0) {
302 basetime.tv_nsec += 1000000000;
303 --basetime.tv_sec;
304 }
305 timedelta -= tickdelta;
306 rel_mplock();
307 }
308
309 /*
310 * Apply per-tick compensation. ticks_adj adjusts for both
311 * offset and frequency, and could be negative.
312 */
313 if (nsec_adj != 0 && try_mplock()) {
314 nsec_acc += nsec_adj;
315 if (nsec_acc >= 0x100000000LL) {
316 basetime.tv_nsec += nsec_acc >> 32;
317 nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
318 } else if (nsec_acc <= -0x100000000LL) {
319 basetime.tv_nsec -= -nsec_acc >> 32;
320 nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
321 }
322 if (basetime.tv_nsec >= 1000000000) {
323 basetime.tv_nsec -= 1000000000;
324 ++basetime.tv_sec;
325 } else if (basetime.tv_nsec < 0) {
326 basetime.tv_nsec += 1000000000;
327 --basetime.tv_sec;
328 }
329 rel_mplock();
330 }
331
332 /*
333 * If the realtime-adjusted seconds hand rolls over then tell
334 * ntp_update_second() what we did in the last second so it can
335 * calculate what to do in the next second. It may also add
336 * or subtract a leap second.
337 */
338 getnanotime(&nts);
339 if (time_second != nts.tv_sec) {
340 leap = ntp_update_second(time_second, &nsec_adj);
341 basetime.tv_sec += leap;
342 time_second = nts.tv_sec + leap;
343 nsec_adj /= hz;
344 }
345 }
346
92b561b7
MD
347 /*
348 * softticks are handled for all cpus
349 */
350 hardclock_softtick(gd);
351
88c4d2f6
MD
352 /*
353 * ITimer handling is per-tick, per-cpu. I don't think psignal()
354 * is mpsafe on curproc, so XXX get the mplock.
355 */
356 if ((p = curproc) != NULL && try_mplock()) {
984263bc 357 pstats = p->p_stats;
88c4d2f6 358 if (frame && CLKF_USERMODE(frame) &&
984263bc
MD
359 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
360 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
361 psignal(p, SIGVTALRM);
362 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
363 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
364 psignal(p, SIGPROF);
88c4d2f6 365 rel_mplock();
984263bc 366 }
604e1e09 367 setdelayed();
88c4d2f6 368}
984263bc 369
88c4d2f6
MD
370/*
371 * The statistics clock typically runs at a 125Hz rate, and is intended
372 * to be frequency offset from the hardclock (typ 100Hz). It is per-cpu.
373 *
374 * NOTE! systimer! the MP lock might not be held here. We can only safely
375 * manipulate objects owned by the current cpu.
376 *
377 * The stats clock is responsible for grabbing a profiling sample.
378 * Most of the statistics are only used by user-level statistics programs.
379 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
380 * p->p_estcpu.
381 *
382 * Like the other clocks, the stat clock is called from what is effectively
383 * a fast interrupt, so the context should be the thread/process that got
384 * interrupted.
385 */
386static void
387statclock(systimer_t info, struct intrframe *frame)
388{
389#ifdef GPROF
390 struct gmonparam *g;
391 int i;
984263bc 392#endif
88c4d2f6
MD
393 thread_t td;
394 struct proc *p;
395 int bump;
396 struct timeval tv;
397 struct timeval *stv;
984263bc
MD
398
399 /*
88c4d2f6 400 * How big was our timeslice relative to the last time?
984263bc 401 */
88c4d2f6
MD
402 microuptime(&tv); /* mpsafe */
403 stv = &mycpu->gd_stattv;
404 if (stv->tv_sec == 0) {
405 bump = 1;
406 } else {
407 bump = tv.tv_usec - stv->tv_usec +
408 (tv.tv_sec - stv->tv_sec) * 1000000;
409 if (bump < 0)
410 bump = 0;
411 if (bump > 1000000)
412 bump = 1000000;
413 }
414 *stv = tv;
984263bc 415
88c4d2f6
MD
416 td = curthread;
417 p = td->td_proc;
984263bc 418
88c4d2f6
MD
419 if (frame && CLKF_USERMODE(frame)) {
420 /*
421 * Came from userland, handle user time and deal with
422 * possible process.
423 */
424 if (p && (p->p_flag & P_PROFIL))
425 addupc_intr(p, CLKF_PC(frame), 1);
426 td->td_uticks += bump;
984263bc 427
88c4d2f6
MD
428 /*
429 * Charge the time as appropriate
430 */
431 if (p && p->p_nice > NZERO)
432 cp_time[CP_NICE] += bump;
433 else
434 cp_time[CP_USER] += bump;
435 } else {
436#ifdef GPROF
437 /*
438 * Kernel statistics are just like addupc_intr, only easier.
439 */
440 g = &_gmonparam;
441 if (g->state == GMON_PROF_ON && frame) {
442 i = CLKF_PC(frame) - g->lowpc;
443 if (i < g->textsize) {
444 i /= HISTFRACTION * sizeof(*g->kcount);
445 g->kcount[i]++;
446 }
447 }
448#endif
449 /*
450 * Came from kernel mode, so we were:
451 * - handling an interrupt,
452 * - doing syscall or trap work on behalf of the current
453 * user process, or
454 * - spinning in the idle loop.
455 * Whichever it is, charge the time as appropriate.
456 * Note that we charge interrupts to the current process,
457 * regardless of whether they are ``for'' that process,
458 * so that we know how much of its real time was spent
459 * in ``non-process'' (i.e., interrupt) work.
460 *
461 * XXX assume system if frame is NULL. A NULL frame
462 * can occur if ipi processing is done from an splx().
463 */
464 if (frame && CLKF_INTR(frame))
465 td->td_iticks += bump;
466 else
467 td->td_sticks += bump;
468
469 if (frame && CLKF_INTR(frame)) {
470 cp_time[CP_INTR] += bump;
471 } else {
472 if (td == &mycpu->gd_idlethread)
473 cp_time[CP_IDLE] += bump;
474 else
475 cp_time[CP_SYS] += bump;
476 }
477 }
478}
479
480/*
0a3f9b47 481 * The scheduler clock typically runs at a 20Hz rate. NOTE! systimer,
88c4d2f6
MD
482 * the MP lock might not be held. We can safely manipulate parts of curproc
483 * but that's about it.
484 */
485static void
486schedclock(systimer_t info, struct intrframe *frame)
487{
488 struct proc *p;
489 struct pstats *pstats;
490 struct rusage *ru;
491 struct vmspace *vm;
492 long rss;
493
494 schedulerclock(NULL); /* mpsafe */
495 if ((p = curproc) != NULL) {
496 /* Update resource usage integrals and maximums. */
497 if ((pstats = p->p_stats) != NULL &&
498 (ru = &pstats->p_ru) != NULL &&
499 (vm = p->p_vmspace) != NULL) {
500 ru->ru_ixrss += pgtok(vm->vm_tsize);
501 ru->ru_idrss += pgtok(vm->vm_dsize);
502 ru->ru_isrss += pgtok(vm->vm_ssize);
503 rss = pgtok(vmspace_resident_count(vm));
504 if (ru->ru_maxrss < rss)
505 ru->ru_maxrss = rss;
506 }
b68b7282 507 }
984263bc
MD
508}
509
510/*
a94976ad
MD
511 * Compute number of ticks for the specified amount of time. The
512 * return value is intended to be used in a clock interrupt timed
513 * operation and guarenteed to meet or exceed the requested time.
514 * If the representation overflows, return INT_MAX. The minimum return
515 * value is 1 ticks and the function will average the calculation up.
516 * If any value greater then 0 microseconds is supplied, a value
517 * of at least 2 will be returned to ensure that a near-term clock
518 * interrupt does not cause the timeout to occur (degenerately) early.
519 *
520 * Note that limit checks must take into account microseconds, which is
521 * done simply by using the smaller signed long maximum instead of
522 * the unsigned long maximum.
523 *
524 * If ints have 32 bits, then the maximum value for any timeout in
525 * 10ms ticks is 248 days.
984263bc
MD
526 */
527int
a94976ad 528tvtohz_high(struct timeval *tv)
984263bc 529{
a94976ad 530 int ticks;
1fd87d54 531 long sec, usec;
984263bc 532
984263bc
MD
533 sec = tv->tv_sec;
534 usec = tv->tv_usec;
535 if (usec < 0) {
536 sec--;
537 usec += 1000000;
538 }
539 if (sec < 0) {
540#ifdef DIAGNOSTIC
541 if (usec > 0) {
542 sec++;
543 usec -= 1000000;
544 }
545 printf("tvotohz: negative time difference %ld sec %ld usec\n",
546 sec, usec);
547#endif
548 ticks = 1;
a94976ad
MD
549 } else if (sec <= INT_MAX / hz) {
550 ticks = (int)(sec * hz +
551 ((u_long)usec + (tick - 1)) / tick) + 1;
552 } else {
553 ticks = INT_MAX;
554 }
555 return (ticks);
556}
557
558/*
559 * Compute number of ticks for the specified amount of time, erroring on
560 * the side of it being too low to ensure that sleeping the returned number
561 * of ticks will not result in a late return.
562 *
563 * The supplied timeval may not be negative and should be normalized. A
564 * return value of 0 is possible if the timeval converts to less then
565 * 1 tick.
566 *
567 * If ints have 32 bits, then the maximum value for any timeout in
568 * 10ms ticks is 248 days.
569 */
570int
571tvtohz_low(struct timeval *tv)
572{
573 int ticks;
574 long sec;
575
576 sec = tv->tv_sec;
577 if (sec <= INT_MAX / hz)
578 ticks = (int)(sec * hz + (u_long)tv->tv_usec / tick);
984263bc 579 else
984263bc 580 ticks = INT_MAX;
a94976ad 581 return (ticks);
984263bc
MD
582}
583
a94976ad 584
984263bc
MD
585/*
586 * Start profiling on a process.
587 *
588 * Kernel profiling passes proc0 which never exits and hence
589 * keeps the profile clock running constantly.
590 */
591void
88c4d2f6 592startprofclock(struct proc *p)
984263bc 593{
984263bc
MD
594 if ((p->p_flag & P_PROFIL) == 0) {
595 p->p_flag |= P_PROFIL;
88c4d2f6 596#if 0 /* XXX */
984263bc
MD
597 if (++profprocs == 1 && stathz != 0) {
598 s = splstatclock();
6ad39cae 599 psdiv = psratio;
984263bc
MD
600 setstatclockrate(profhz);
601 splx(s);
602 }
88c4d2f6 603#endif
984263bc
MD
604 }
605}
606
607/*
608 * Stop profiling on a process.
609 */
610void
88c4d2f6 611stopprofclock(struct proc *p)
984263bc 612{
984263bc
MD
613 if (p->p_flag & P_PROFIL) {
614 p->p_flag &= ~P_PROFIL;
88c4d2f6 615#if 0 /* XXX */
984263bc
MD
616 if (--profprocs == 0 && stathz != 0) {
617 s = splstatclock();
6ad39cae 618 psdiv = 1;
984263bc
MD
619 setstatclockrate(stathz);
620 splx(s);
621 }
984263bc 622#endif
984263bc
MD
623 }
624}
625
626/*
627 * Return information about system clocks.
628 */
629static int
630sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
631{
632 struct clockinfo clkinfo;
633 /*
634 * Construct clockinfo structure.
635 */
636 clkinfo.hz = hz;
637 clkinfo.tick = tick;
638 clkinfo.tickadj = tickadj;
639 clkinfo.profhz = profhz;
640 clkinfo.stathz = stathz ? stathz : hz;
641 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
642}
643
644SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
645 0, 0, sysctl_kern_clockrate, "S,clockinfo","");
646
984263bc
MD
647/*
648 * We have eight functions for looking at the clock, four for
649 * microseconds and four for nanoseconds. For each there is fast
650 * but less precise version "get{nano|micro}[up]time" which will
651 * return a time which is up to 1/HZ previous to the call, whereas
652 * the raw version "{nano|micro}[up]time" will return a timestamp
653 * which is as precise as possible. The "up" variants return the
654 * time relative to system boot, these are well suited for time
655 * interval measurements.
88c4d2f6
MD
656 *
657 * Each cpu independantly maintains the current time of day, so all
658 * we need to do to protect ourselves from changes is to do a loop
659 * check on the seconds field changing out from under us.
fad57d0e
MD
660 *
661 * The system timer maintains a 32 bit count and due to various issues
662 * it is possible for the calculated delta to occassionally exceed
663 * cputimer_freq. If this occurs the cputimer_freq64_nsec multiplication
664 * can easily overflow, so we deal with the case. For uniformity we deal
665 * with the case in the usec case too.
984263bc 666 */
984263bc
MD
667void
668getmicrouptime(struct timeval *tvp)
669{
88c4d2f6
MD
670 struct globaldata *gd = mycpu;
671 sysclock_t delta;
672
673 do {
674 tvp->tv_sec = gd->gd_time_seconds;
675 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
676 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
677
678 if (delta >= cputimer_freq) {
679 tvp->tv_sec += delta / cputimer_freq;
680 delta %= cputimer_freq;
681 }
88c4d2f6
MD
682 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
683 if (tvp->tv_usec >= 1000000) {
684 tvp->tv_usec -= 1000000;
685 ++tvp->tv_sec;
984263bc
MD
686 }
687}
688
689void
690getnanouptime(struct timespec *tsp)
691{
88c4d2f6
MD
692 struct globaldata *gd = mycpu;
693 sysclock_t delta;
694
695 do {
696 tsp->tv_sec = gd->gd_time_seconds;
697 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
698 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
699
700 if (delta >= cputimer_freq) {
701 tsp->tv_sec += delta / cputimer_freq;
702 delta %= cputimer_freq;
984263bc 703 }
fad57d0e 704 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc
MD
705}
706
707void
88c4d2f6 708microuptime(struct timeval *tvp)
984263bc 709{
88c4d2f6
MD
710 struct globaldata *gd = mycpu;
711 sysclock_t delta;
712
713 do {
714 tvp->tv_sec = gd->gd_time_seconds;
715 delta = cputimer_count() - gd->gd_cpuclock_base;
716 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
717
718 if (delta >= cputimer_freq) {
719 tvp->tv_sec += delta / cputimer_freq;
720 delta %= cputimer_freq;
984263bc 721 }
fad57d0e 722 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
984263bc
MD
723}
724
725void
88c4d2f6 726nanouptime(struct timespec *tsp)
984263bc 727{
88c4d2f6
MD
728 struct globaldata *gd = mycpu;
729 sysclock_t delta;
730
731 do {
732 tsp->tv_sec = gd->gd_time_seconds;
733 delta = cputimer_count() - gd->gd_cpuclock_base;
734 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
735
736 if (delta >= cputimer_freq) {
737 tsp->tv_sec += delta / cputimer_freq;
738 delta %= cputimer_freq;
984263bc 739 }
fad57d0e 740 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc
MD
741}
742
88c4d2f6
MD
743/*
744 * realtime routines
745 */
984263bc
MD
746
747void
88c4d2f6 748getmicrotime(struct timeval *tvp)
984263bc 749{
88c4d2f6
MD
750 struct globaldata *gd = mycpu;
751 sysclock_t delta;
984263bc 752
88c4d2f6
MD
753 do {
754 tvp->tv_sec = gd->gd_time_seconds;
755 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
756 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
757
758 if (delta >= cputimer_freq) {
759 tvp->tv_sec += delta / cputimer_freq;
760 delta %= cputimer_freq;
761 }
88c4d2f6 762 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
984263bc 763
88c4d2f6
MD
764 tvp->tv_sec += basetime.tv_sec;
765 tvp->tv_usec += basetime.tv_nsec / 1000;
766 while (tvp->tv_usec >= 1000000) {
767 tvp->tv_usec -= 1000000;
768 ++tvp->tv_sec;
984263bc 769 }
984263bc
MD
770}
771
772void
88c4d2f6 773getnanotime(struct timespec *tsp)
984263bc 774{
88c4d2f6
MD
775 struct globaldata *gd = mycpu;
776 sysclock_t delta;
984263bc 777
88c4d2f6
MD
778 do {
779 tsp->tv_sec = gd->gd_time_seconds;
780 delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
781 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
782
783 if (delta >= cputimer_freq) {
784 tsp->tv_sec += delta / cputimer_freq;
785 delta %= cputimer_freq;
786 }
88c4d2f6 787 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc 788
88c4d2f6
MD
789 tsp->tv_sec += basetime.tv_sec;
790 tsp->tv_nsec += basetime.tv_nsec;
791 while (tsp->tv_nsec >= 1000000000) {
792 tsp->tv_nsec -= 1000000000;
793 ++tsp->tv_sec;
984263bc 794 }
984263bc
MD
795}
796
88c4d2f6
MD
797void
798microtime(struct timeval *tvp)
984263bc 799{
88c4d2f6
MD
800 struct globaldata *gd = mycpu;
801 sysclock_t delta;
984263bc 802
88c4d2f6
MD
803 do {
804 tvp->tv_sec = gd->gd_time_seconds;
805 delta = cputimer_count() - gd->gd_cpuclock_base;
806 } while (tvp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
807
808 if (delta >= cputimer_freq) {
809 tvp->tv_sec += delta / cputimer_freq;
810 delta %= cputimer_freq;
811 }
88c4d2f6 812 tvp->tv_usec = (cputimer_freq64_usec * delta) >> 32;
984263bc 813
88c4d2f6
MD
814 tvp->tv_sec += basetime.tv_sec;
815 tvp->tv_usec += basetime.tv_nsec / 1000;
816 while (tvp->tv_usec >= 1000000) {
817 tvp->tv_usec -= 1000000;
818 ++tvp->tv_sec;
984263bc 819 }
984263bc
MD
820}
821
88c4d2f6
MD
822void
823nanotime(struct timespec *tsp)
824{
825 struct globaldata *gd = mycpu;
826 sysclock_t delta;
984263bc 827
88c4d2f6
MD
828 do {
829 tsp->tv_sec = gd->gd_time_seconds;
830 delta = cputimer_count() - gd->gd_cpuclock_base;
831 } while (tsp->tv_sec != gd->gd_time_seconds);
fad57d0e
MD
832
833 if (delta >= cputimer_freq) {
834 tsp->tv_sec += delta / cputimer_freq;
835 delta %= cputimer_freq;
836 }
88c4d2f6 837 tsp->tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
984263bc 838
88c4d2f6
MD
839 tsp->tv_sec += basetime.tv_sec;
840 tsp->tv_nsec += basetime.tv_nsec;
841 while (tsp->tv_nsec >= 1000000000) {
842 tsp->tv_nsec -= 1000000000;
843 ++tsp->tv_sec;
984263bc 844 }
984263bc
MD
845}
846
984263bc
MD
847int
848pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
849{
850 pps_params_t *app;
851 struct pps_fetch_args *fapi;
852#ifdef PPS_SYNC
853 struct pps_kcbind_args *kapi;
854#endif
855
856 switch (cmd) {
857 case PPS_IOC_CREATE:
858 return (0);
859 case PPS_IOC_DESTROY:
860 return (0);
861 case PPS_IOC_SETPARAMS:
862 app = (pps_params_t *)data;
863 if (app->mode & ~pps->ppscap)
864 return (EINVAL);
865 pps->ppsparam = *app;
866 return (0);
867 case PPS_IOC_GETPARAMS:
868 app = (pps_params_t *)data;
869 *app = pps->ppsparam;
870 app->api_version = PPS_API_VERS_1;
871 return (0);
872 case PPS_IOC_GETCAP:
873 *(int*)data = pps->ppscap;
874 return (0);
875 case PPS_IOC_FETCH:
876 fapi = (struct pps_fetch_args *)data;
877 if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
878 return (EINVAL);
879 if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
880 return (EOPNOTSUPP);
881 pps->ppsinfo.current_mode = pps->ppsparam.mode;
882 fapi->pps_info_buf = pps->ppsinfo;
883 return (0);
884 case PPS_IOC_KCBIND:
885#ifdef PPS_SYNC
886 kapi = (struct pps_kcbind_args *)data;
887 /* XXX Only root should be able to do this */
888 if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
889 return (EINVAL);
890 if (kapi->kernel_consumer != PPS_KC_HARDPPS)
891 return (EINVAL);
892 if (kapi->edge & ~pps->ppscap)
893 return (EINVAL);
894 pps->kcmode = kapi->edge;
895 return (0);
896#else
897 return (EOPNOTSUPP);
898#endif
899 default:
900 return (ENOTTY);
901 }
902}
903
904void
905pps_init(struct pps_state *pps)
906{
907 pps->ppscap |= PPS_TSFMT_TSPEC;
908 if (pps->ppscap & PPS_CAPTUREASSERT)
909 pps->ppscap |= PPS_OFFSETASSERT;
910 if (pps->ppscap & PPS_CAPTURECLEAR)
911 pps->ppscap |= PPS_OFFSETCLEAR;
912}
913
914void
88c4d2f6 915pps_event(struct pps_state *pps, sysclock_t count, int event)
984263bc 916{
88c4d2f6
MD
917 struct globaldata *gd;
918 struct timespec *tsp;
919 struct timespec *osp;
920 struct timespec ts;
921 sysclock_t *pcount;
922#ifdef PPS_SYNC
923 sysclock_t tcount;
924#endif
925 sysclock_t delta;
926 pps_seq_t *pseq;
927 int foff;
928 int fhard;
929
930 gd = mycpu;
984263bc
MD
931
932 /* Things would be easier with arrays... */
933 if (event == PPS_CAPTUREASSERT) {
934 tsp = &pps->ppsinfo.assert_timestamp;
935 osp = &pps->ppsparam.assert_offset;
936 foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
937 fhard = pps->kcmode & PPS_CAPTUREASSERT;
938 pcount = &pps->ppscount[0];
939 pseq = &pps->ppsinfo.assert_sequence;
940 } else {
941 tsp = &pps->ppsinfo.clear_timestamp;
942 osp = &pps->ppsparam.clear_offset;
943 foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
944 fhard = pps->kcmode & PPS_CAPTURECLEAR;
945 pcount = &pps->ppscount[1];
946 pseq = &pps->ppsinfo.clear_sequence;
947 }
948
984263bc
MD
949 /* Nothing really happened */
950 if (*pcount == count)
951 return;
952
953 *pcount = count;
954
88c4d2f6
MD
955 do {
956 ts.tv_sec = gd->gd_time_seconds;
957 delta = count - gd->gd_cpuclock_base;
958 } while (ts.tv_sec != gd->gd_time_seconds);
fad57d0e
MD
959
960 if (delta >= cputimer_freq) {
88c4d2f6
MD
961 ts.tv_sec += delta / cputimer_freq;
962 delta %= cputimer_freq;
963 }
964 ts.tv_nsec = (cputimer_freq64_nsec * delta) >> 32;
965 ts.tv_sec += basetime.tv_sec;
966 ts.tv_nsec += basetime.tv_nsec;
967 while (ts.tv_nsec >= 1000000000) {
968 ts.tv_nsec -= 1000000000;
969 ++ts.tv_sec;
984263bc 970 }
984263bc
MD
971
972 (*pseq)++;
973 *tsp = ts;
974
975 if (foff) {
976 timespecadd(tsp, osp);
977 if (tsp->tv_nsec < 0) {
978 tsp->tv_nsec += 1000000000;
979 tsp->tv_sec -= 1;
980 }
981 }
982#ifdef PPS_SYNC
983 if (fhard) {
984 /* magic, at its best... */
985 tcount = count - pps->ppscount[2];
986 pps->ppscount[2] = count;
fad57d0e
MD
987 if (tcount >= cputimer_freq) {
988 delta = 1000000000 * (tcount / cputimer_freq) +
989 (cputimer_freq64_nsec *
990 (tcount % cputimer_freq)) >> 32;
991 } else {
992 delta = (cputimer_freq64_nsec * tcount) >> 32;
993 }
984263bc
MD
994 hardpps(tsp, delta);
995 }
996#endif
997}
88c4d2f6 998