FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_clock.c
1 /*-
2 * Copyright (c) 1982, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD: releng/5.2/sys/kern/kern_clock.c 121127 2003-10-16 08:39:15Z jeff $");
43
44 #include "opt_ntp.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/callout.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/ktr.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/signalvar.h>
60 #include <sys/smp.h>
61 #include <vm/vm.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_map.h>
64 #include <sys/sysctl.h>
65 #include <sys/bus.h>
66 #include <sys/interrupt.h>
67 #include <sys/limits.h>
68 #include <sys/timetc.h>
69
70 #include <machine/cpu.h>
71
72 #ifdef GPROF
73 #include <sys/gmon.h>
74 #endif
75
76 #ifdef DDB
77 #include <ddb/ddb.h>
78 #endif
79
80 #ifdef DEVICE_POLLING
81 extern void hardclock_device_poll(void);
82 #endif /* DEVICE_POLLING */
83
84 static void initclocks(void *dummy);
85 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
86
87 /* Some of these don't belong here, but it's easiest to concentrate them. */
88 long cp_time[CPUSTATES];
89
90 SYSCTL_OPAQUE(_kern, OID_AUTO, cp_time, CTLFLAG_RD, &cp_time, sizeof(cp_time),
91 "LU", "CPU time statistics");
92
93 #ifdef WATCHDOG
94 static int sysctl_watchdog_reset(SYSCTL_HANDLER_ARGS);
95 static void watchdog_fire(void);
96
97 static int watchdog_enabled;
98 static unsigned int watchdog_ticks;
99 static int watchdog_timeout = 20;
100
101 SYSCTL_NODE(_debug, OID_AUTO, watchdog, CTLFLAG_RW, 0, "System watchdog");
102 SYSCTL_INT(_debug_watchdog, OID_AUTO, enabled, CTLFLAG_RW, &watchdog_enabled,
103 0, "Enable the watchdog");
104 SYSCTL_INT(_debug_watchdog, OID_AUTO, timeout, CTLFLAG_RW, &watchdog_timeout,
105 0, "Timeout for watchdog checkins");
106
107 #endif /* WATCHDOG */
108
109 /*
110 * Clock handling routines.
111 *
112 * This code is written to operate with two timers that run independently of
113 * each other.
114 *
115 * The main timer, running hz times per second, is used to trigger interval
116 * timers, timeouts and rescheduling as needed.
117 *
118 * The second timer handles kernel and user profiling,
119 * and does resource use estimation. If the second timer is programmable,
120 * it is randomized to avoid aliasing between the two clocks. For example,
121 * the randomization prevents an adversary from always giving up the cpu
122 * just before its quantum expires. Otherwise, it would never accumulate
123 * cpu ticks. The mean frequency of the second timer is stathz.
124 *
125 * If no second timer exists, stathz will be zero; in this case we drive
126 * profiling and statistics off the main clock. This WILL NOT be accurate;
127 * do not do it unless absolutely necessary.
128 *
129 * The statistics clock may (or may not) be run at a higher rate while
130 * profiling. This profile clock runs at profhz. We require that profhz
131 * be an integral multiple of stathz.
132 *
133 * If the statistics clock is running fast, it must be divided by the ratio
134 * profhz/stathz for statistics. (For profiling, every tick counts.)
135 *
136 * Time-of-day is maintained using a "timecounter", which may or may
137 * not be related to the hardware generating the above mentioned
138 * interrupts.
139 */
140
141 int stathz;
142 int profhz;
143 int profprocs;
144 int ticks;
145 int psratio;
146
147 /*
148 * Initialize clock frequencies and start both clocks running.
149 */
150 /* ARGSUSED*/
151 static void
152 initclocks(dummy)
153 void *dummy;
154 {
155 register int i;
156
157 /*
158 * Set divisors to 1 (normal case) and let the machine-specific
159 * code do its bit.
160 */
161 cpu_initclocks();
162
163 /*
164 * Compute profhz/stathz, and fix profhz if needed.
165 */
166 i = stathz ? stathz : hz;
167 if (profhz == 0)
168 profhz = i;
169 psratio = profhz / i;
170 }
171
172 /*
173 * Each time the real-time timer fires, this function is called on all CPUs.
174 * Note that hardclock() calls hardclock_process() for the boot CPU, so only
175 * the other CPUs in the system need to call this function.
176 */
177 void
178 hardclock_process(frame)
179 register struct clockframe *frame;
180 {
181 struct pstats *pstats;
182 struct thread *td = curthread;
183 struct proc *p = td->td_proc;
184
185 /*
186 * Run current process's virtual and profile time, as needed.
187 */
188 mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
189 if (p->p_flag & P_SA) {
190 /* XXXKSE What to do? */
191 } else {
192 pstats = p->p_stats;
193 if (CLKF_USERMODE(frame) &&
194 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
195 itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
196 p->p_sflag |= PS_ALRMPEND;
197 td->td_flags |= TDF_ASTPENDING;
198 }
199 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
200 itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
201 p->p_sflag |= PS_PROFPEND;
202 td->td_flags |= TDF_ASTPENDING;
203 }
204 }
205 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
206 }
207
208 /*
209 * The real-time timer, interrupting hz times per second.
210 */
211 void
212 hardclock(frame)
213 register struct clockframe *frame;
214 {
215 int need_softclock = 0;
216
217 CTR0(KTR_CLK, "hardclock fired");
218 hardclock_process(frame);
219
220 tc_ticktock();
221 /*
222 * If no separate statistics clock is available, run it from here.
223 *
224 * XXX: this only works for UP
225 */
226 if (stathz == 0) {
227 profclock(frame);
228 statclock(frame);
229 }
230
231 #ifdef DEVICE_POLLING
232 hardclock_device_poll(); /* this is very short and quick */
233 #endif /* DEVICE_POLLING */
234
235 /*
236 * Process callouts at a very low cpu priority, so we don't keep the
237 * relatively high clock interrupt priority any longer than necessary.
238 */
239 mtx_lock_spin_flags(&callout_lock, MTX_QUIET);
240 ticks++;
241 if (TAILQ_FIRST(&callwheel[ticks & callwheelmask]) != NULL) {
242 need_softclock = 1;
243 } else if (softticks + 1 == ticks)
244 ++softticks;
245 mtx_unlock_spin_flags(&callout_lock, MTX_QUIET);
246
247 /*
248 * swi_sched acquires sched_lock, so we don't want to call it with
249 * callout_lock held; incorrect locking order.
250 */
251 if (need_softclock)
252 swi_sched(softclock_ih, 0);
253
254 #ifdef WATCHDOG
255 if (watchdog_enabled > 0 &&
256 (int)(ticks - watchdog_ticks) >= (hz * watchdog_timeout))
257 watchdog_fire();
258 #endif /* WATCHDOG */
259 }
260
261 /*
262 * Compute number of ticks in the specified amount of time.
263 */
264 int
265 tvtohz(tv)
266 struct timeval *tv;
267 {
268 register unsigned long ticks;
269 register long sec, usec;
270
271 /*
272 * If the number of usecs in the whole seconds part of the time
273 * difference fits in a long, then the total number of usecs will
274 * fit in an unsigned long. Compute the total and convert it to
275 * ticks, rounding up and adding 1 to allow for the current tick
276 * to expire. Rounding also depends on unsigned long arithmetic
277 * to avoid overflow.
278 *
279 * Otherwise, if the number of ticks in the whole seconds part of
280 * the time difference fits in a long, then convert the parts to
281 * ticks separately and add, using similar rounding methods and
282 * overflow avoidance. This method would work in the previous
283 * case but it is slightly slower and assumes that hz is integral.
284 *
285 * Otherwise, round the time difference down to the maximum
286 * representable value.
287 *
288 * If ints have 32 bits, then the maximum value for any timeout in
289 * 10ms ticks is 248 days.
290 */
291 sec = tv->tv_sec;
292 usec = tv->tv_usec;
293 if (usec < 0) {
294 sec--;
295 usec += 1000000;
296 }
297 if (sec < 0) {
298 #ifdef DIAGNOSTIC
299 if (usec > 0) {
300 sec++;
301 usec -= 1000000;
302 }
303 printf("tvotohz: negative time difference %ld sec %ld usec\n",
304 sec, usec);
305 #endif
306 ticks = 1;
307 } else if (sec <= LONG_MAX / 1000000)
308 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
309 / tick + 1;
310 else if (sec <= LONG_MAX / hz)
311 ticks = sec * hz
312 + ((unsigned long)usec + (tick - 1)) / tick + 1;
313 else
314 ticks = LONG_MAX;
315 if (ticks > INT_MAX)
316 ticks = INT_MAX;
317 return ((int)ticks);
318 }
319
320 /*
321 * Start profiling on a process.
322 *
323 * Kernel profiling passes proc0 which never exits and hence
324 * keeps the profile clock running constantly.
325 */
326 void
327 startprofclock(p)
328 register struct proc *p;
329 {
330
331 /*
332 * XXX; Right now sched_lock protects statclock(), but perhaps
333 * it should be protected later on by a time_lock, which would
334 * cover psdiv, etc. as well.
335 */
336 PROC_LOCK_ASSERT(p, MA_OWNED);
337 if (p->p_flag & P_STOPPROF)
338 return;
339 if ((p->p_flag & P_PROFIL) == 0) {
340 mtx_lock_spin(&sched_lock);
341 p->p_flag |= P_PROFIL;
342 if (++profprocs == 1)
343 cpu_startprofclock();
344 mtx_unlock_spin(&sched_lock);
345 }
346 }
347
348 /*
349 * Stop profiling on a process.
350 */
351 void
352 stopprofclock(p)
353 register struct proc *p;
354 {
355
356 PROC_LOCK_ASSERT(p, MA_OWNED);
357 if (p->p_flag & P_PROFIL) {
358 if (p->p_profthreads != 0) {
359 p->p_flag |= P_STOPPROF;
360 while (p->p_profthreads != 0)
361 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
362 "stopprof", NULL);
363 p->p_flag &= ~P_STOPPROF;
364 }
365 mtx_lock_spin(&sched_lock);
366 p->p_flag &= ~P_PROFIL;
367 if (--profprocs == 0)
368 cpu_stopprofclock();
369 mtx_unlock_spin(&sched_lock);
370 }
371 }
372
373 /*
374 * Statistics clock. Grab profile sample, and if divider reaches 0,
375 * do process and kernel statistics. Most of the statistics are only
376 * used by user-level statistics programs. The main exceptions are
377 * ke->ke_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu.
378 * This should be called by all active processors.
379 */
380 void
381 statclock(frame)
382 register struct clockframe *frame;
383 {
384 struct pstats *pstats;
385 struct rusage *ru;
386 struct vmspace *vm;
387 struct thread *td;
388 struct proc *p;
389 long rss;
390
391 td = curthread;
392 p = td->td_proc;
393
394 mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
395 if (CLKF_USERMODE(frame)) {
396 /*
397 * Charge the time as appropriate.
398 */
399 if (p->p_flag & P_SA)
400 thread_statclock(1);
401 p->p_uticks++;
402 if (td->td_ksegrp->kg_nice > NZERO)
403 cp_time[CP_NICE]++;
404 else
405 cp_time[CP_USER]++;
406 } else {
407 /*
408 * Came from kernel mode, so we were:
409 * - handling an interrupt,
410 * - doing syscall or trap work on behalf of the current
411 * user process, or
412 * - spinning in the idle loop.
413 * Whichever it is, charge the time as appropriate.
414 * Note that we charge interrupts to the current process,
415 * regardless of whether they are ``for'' that process,
416 * so that we know how much of its real time was spent
417 * in ``non-process'' (i.e., interrupt) work.
418 */
419 if ((td->td_ithd != NULL) || td->td_intr_nesting_level >= 2) {
420 p->p_iticks++;
421 cp_time[CP_INTR]++;
422 } else {
423 if (p->p_flag & P_SA)
424 thread_statclock(0);
425 td->td_sticks++;
426 p->p_sticks++;
427 if (p != PCPU_GET(idlethread)->td_proc)
428 cp_time[CP_SYS]++;
429 else
430 cp_time[CP_IDLE]++;
431 }
432 }
433
434 sched_clock(td);
435
436 /* Update resource usage integrals and maximums. */
437 if ((pstats = p->p_stats) != NULL &&
438 (ru = &pstats->p_ru) != NULL &&
439 (vm = p->p_vmspace) != NULL) {
440 ru->ru_ixrss += pgtok(vm->vm_tsize);
441 ru->ru_idrss += pgtok(vm->vm_dsize);
442 ru->ru_isrss += pgtok(vm->vm_ssize);
443 rss = pgtok(vmspace_resident_count(vm));
444 if (ru->ru_maxrss < rss)
445 ru->ru_maxrss = rss;
446 }
447 mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
448 }
449
450 void
451 profclock(frame)
452 register struct clockframe *frame;
453 {
454 struct thread *td;
455 #ifdef GPROF
456 struct gmonparam *g;
457 int i;
458 #endif
459
460 td = curthread;
461 if (CLKF_USERMODE(frame)) {
462 /*
463 * Came from user mode; CPU was in user state.
464 * If this process is being profiled, record the tick.
465 * if there is no related user location yet, don't
466 * bother trying to count it.
467 */
468 td = curthread;
469 if (td->td_proc->p_flag & P_PROFIL)
470 addupc_intr(td, CLKF_PC(frame), 1);
471 }
472 #ifdef GPROF
473 else {
474 /*
475 * Kernel statistics are just like addupc_intr, only easier.
476 */
477 g = &_gmonparam;
478 if (g->state == GMON_PROF_ON) {
479 i = CLKF_PC(frame) - g->lowpc;
480 if (i < g->textsize) {
481 i /= HISTFRACTION * sizeof(*g->kcount);
482 g->kcount[i]++;
483 }
484 }
485 }
486 #endif
487 }
488
489 /*
490 * Return information about system clocks.
491 */
492 static int
493 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
494 {
495 struct clockinfo clkinfo;
496 /*
497 * Construct clockinfo structure.
498 */
499 bzero(&clkinfo, sizeof(clkinfo));
500 clkinfo.hz = hz;
501 clkinfo.tick = tick;
502 clkinfo.profhz = profhz;
503 clkinfo.stathz = stathz ? stathz : hz;
504 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
505 }
506
507 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
508 0, 0, sysctl_kern_clockrate, "S,clockinfo",
509 "Rate and period of various kernel clocks");
510
511 #ifdef WATCHDOG
512 /*
513 * Reset the watchdog timer to ticks, thus preventing the watchdog
514 * from firing for another watchdog timeout period.
515 */
516 static int
517 sysctl_watchdog_reset(SYSCTL_HANDLER_ARGS)
518 {
519 int ret;
520
521 ret = 0;
522 watchdog_ticks = ticks;
523 return sysctl_handle_int(oidp, &ret, 0, req);
524 }
525
526 SYSCTL_PROC(_debug_watchdog, OID_AUTO, reset, CTLFLAG_RW, 0, 0,
527 sysctl_watchdog_reset, "I", "Reset the watchdog");
528
529 /*
530 * Handle a watchdog timeout by dumping interrupt information and
531 * then either dropping to DDB or panicing.
532 */
533 static void
534 watchdog_fire(void)
535 {
536 int nintr;
537 u_int64_t inttotal;
538 u_long *curintr;
539 char *curname;
540
541 curintr = intrcnt;
542 curname = intrnames;
543 inttotal = 0;
544 nintr = eintrcnt - intrcnt;
545
546 printf("interrupt total\n");
547 while (--nintr >= 0) {
548 if (*curintr)
549 printf("%-12s %20lu\n", curname, *curintr);
550 curname += strlen(curname) + 1;
551 inttotal += *curintr++;
552 }
553 printf("Total %20ju\n", (uintmax_t)inttotal);
554
555 #ifdef DDB
556 db_print_backtrace();
557 Debugger("watchdog timeout");
558 #else /* !DDB */
559 panic("watchdog timeout");
560 #endif /* DDB */
561 }
562
563 #endif /* WATCHDOG */
Cache object: 67ec95efceeb30ed29ed7be5853802e9
|