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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/11.1/sys/kern/kern_clock.c 310436 2016-12-23 00:38:43Z jhb $");
39
40 #include "opt_kdb.h"
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_ntp.h"
44 #include "opt_watchdog.h"
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/callout.h>
49 #include <sys/kdb.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/ktr.h>
53 #include <sys/lock.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/sdt.h>
60 #include <sys/signalvar.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/smp.h>
63 #include <vm/vm.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_map.h>
66 #include <sys/sysctl.h>
67 #include <sys/bus.h>
68 #include <sys/interrupt.h>
69 #include <sys/limits.h>
70 #include <sys/timetc.h>
71
72 #ifdef GPROF
73 #include <sys/gmon.h>
74 #endif
75
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 PMC_SOFT_DEFINE( , , clock, hard);
79 PMC_SOFT_DEFINE( , , clock, stat);
80 PMC_SOFT_DEFINE_EX( , , clock, prof, \
81 cpu_startprofclock, cpu_stopprofclock);
82 #endif
83
84 #ifdef DEVICE_POLLING
85 extern void hardclock_device_poll(void);
86 #endif /* DEVICE_POLLING */
87
88 static void initclocks(void *dummy);
89 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
90
91 /* Spin-lock protecting profiling statistics. */
92 static struct mtx time_lock;
93
94 SDT_PROVIDER_DECLARE(sched);
95 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
96
97 static int
98 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
99 {
100 int error;
101 long cp_time[CPUSTATES];
102 #ifdef SCTL_MASK32
103 int i;
104 unsigned int cp_time32[CPUSTATES];
105 #endif
106
107 read_cpu_time(cp_time);
108 #ifdef SCTL_MASK32
109 if (req->flags & SCTL_MASK32) {
110 if (!req->oldptr)
111 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
112 for (i = 0; i < CPUSTATES; i++)
113 cp_time32[i] = (unsigned int)cp_time[i];
114 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
115 } else
116 #endif
117 {
118 if (!req->oldptr)
119 return SYSCTL_OUT(req, 0, sizeof(cp_time));
120 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
121 }
122 return error;
123 }
124
125 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
126 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
127
128 static long empty[CPUSTATES];
129
130 static int
131 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
132 {
133 struct pcpu *pcpu;
134 int error;
135 int c;
136 long *cp_time;
137 #ifdef SCTL_MASK32
138 unsigned int cp_time32[CPUSTATES];
139 int i;
140 #endif
141
142 if (!req->oldptr) {
143 #ifdef SCTL_MASK32
144 if (req->flags & SCTL_MASK32)
145 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
146 else
147 #endif
148 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
149 }
150 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
151 if (!CPU_ABSENT(c)) {
152 pcpu = pcpu_find(c);
153 cp_time = pcpu->pc_cp_time;
154 } else {
155 cp_time = empty;
156 }
157 #ifdef SCTL_MASK32
158 if (req->flags & SCTL_MASK32) {
159 for (i = 0; i < CPUSTATES; i++)
160 cp_time32[i] = (unsigned int)cp_time[i];
161 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
162 } else
163 #endif
164 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
165 }
166 return error;
167 }
168
169 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
170 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
171
172 #ifdef DEADLKRES
173 static const char *blessed[] = {
174 "getblk",
175 "so_snd_sx",
176 "so_rcv_sx",
177 NULL
178 };
179 static int slptime_threshold = 1800;
180 static int blktime_threshold = 900;
181 static int sleepfreq = 3;
182
183 static void
184 deadlkres(void)
185 {
186 struct proc *p;
187 struct thread *td;
188 void *wchan;
189 int blkticks, i, slpticks, slptype, tryl, tticks;
190
191 tryl = 0;
192 for (;;) {
193 blkticks = blktime_threshold * hz;
194 slpticks = slptime_threshold * hz;
195
196 /*
197 * Avoid to sleep on the sx_lock in order to avoid a possible
198 * priority inversion problem leading to starvation.
199 * If the lock can't be held after 100 tries, panic.
200 */
201 if (!sx_try_slock(&allproc_lock)) {
202 if (tryl > 100)
203 panic("%s: possible deadlock detected on allproc_lock\n",
204 __func__);
205 tryl++;
206 pause("allproc", sleepfreq * hz);
207 continue;
208 }
209 tryl = 0;
210 FOREACH_PROC_IN_SYSTEM(p) {
211 PROC_LOCK(p);
212 if (p->p_state == PRS_NEW) {
213 PROC_UNLOCK(p);
214 continue;
215 }
216 FOREACH_THREAD_IN_PROC(p, td) {
217
218 thread_lock(td);
219 if (TD_ON_LOCK(td)) {
220
221 /*
222 * The thread should be blocked on a
223 * turnstile, simply check if the
224 * turnstile channel is in good state.
225 */
226 MPASS(td->td_blocked != NULL);
227
228 tticks = ticks - td->td_blktick;
229 thread_unlock(td);
230 if (tticks > blkticks) {
231
232 /*
233 * Accordingly with provided
234 * thresholds, this thread is
235 * stuck for too long on a
236 * turnstile.
237 */
238 PROC_UNLOCK(p);
239 sx_sunlock(&allproc_lock);
240 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
241 __func__, td, tticks);
242 }
243 } else if (TD_IS_SLEEPING(td) &&
244 TD_ON_SLEEPQ(td)) {
245
246 /*
247 * Check if the thread is sleeping on a
248 * lock, otherwise skip the check.
249 * Drop the thread lock in order to
250 * avoid a LOR with the sleepqueue
251 * spinlock.
252 */
253 wchan = td->td_wchan;
254 tticks = ticks - td->td_slptick;
255 thread_unlock(td);
256 slptype = sleepq_type(wchan);
257 if ((slptype == SLEEPQ_SX ||
258 slptype == SLEEPQ_LK) &&
259 tticks > slpticks) {
260
261 /*
262 * Accordingly with provided
263 * thresholds, this thread is
264 * stuck for too long on a
265 * sleepqueue.
266 * However, being on a
267 * sleepqueue, we might still
268 * check for the blessed
269 * list.
270 */
271 tryl = 0;
272 for (i = 0; blessed[i] != NULL;
273 i++) {
274 if (!strcmp(blessed[i],
275 td->td_wmesg)) {
276 tryl = 1;
277 break;
278 }
279 }
280 if (tryl != 0) {
281 tryl = 0;
282 continue;
283 }
284 PROC_UNLOCK(p);
285 sx_sunlock(&allproc_lock);
286 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
287 __func__, td, tticks);
288 }
289 } else
290 thread_unlock(td);
291 }
292 PROC_UNLOCK(p);
293 }
294 sx_sunlock(&allproc_lock);
295
296 /* Sleep for sleepfreq seconds. */
297 pause("-", sleepfreq * hz);
298 }
299 }
300
301 static struct kthread_desc deadlkres_kd = {
302 "deadlkres",
303 deadlkres,
304 (struct thread **)NULL
305 };
306
307 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
308
309 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
310 "Deadlock resolver");
311 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
312 &slptime_threshold, 0,
313 "Number of seconds within is valid to sleep on a sleepqueue");
314 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
315 &blktime_threshold, 0,
316 "Number of seconds within is valid to block on a turnstile");
317 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
318 "Number of seconds between any deadlock resolver thread run");
319 #endif /* DEADLKRES */
320
321 void
322 read_cpu_time(long *cp_time)
323 {
324 struct pcpu *pc;
325 int i, j;
326
327 /* Sum up global cp_time[]. */
328 bzero(cp_time, sizeof(long) * CPUSTATES);
329 CPU_FOREACH(i) {
330 pc = pcpu_find(i);
331 for (j = 0; j < CPUSTATES; j++)
332 cp_time[j] += pc->pc_cp_time[j];
333 }
334 }
335
336 #ifdef SW_WATCHDOG
337 #include <sys/watchdog.h>
338
339 static int watchdog_ticks;
340 static int watchdog_enabled;
341 static void watchdog_fire(void);
342 static void watchdog_config(void *, u_int, int *);
343 #endif /* SW_WATCHDOG */
344
345 /*
346 * Clock handling routines.
347 *
348 * This code is written to operate with two timers that run independently of
349 * each other.
350 *
351 * The main timer, running hz times per second, is used to trigger interval
352 * timers, timeouts and rescheduling as needed.
353 *
354 * The second timer handles kernel and user profiling,
355 * and does resource use estimation. If the second timer is programmable,
356 * it is randomized to avoid aliasing between the two clocks. For example,
357 * the randomization prevents an adversary from always giving up the cpu
358 * just before its quantum expires. Otherwise, it would never accumulate
359 * cpu ticks. The mean frequency of the second timer is stathz.
360 *
361 * If no second timer exists, stathz will be zero; in this case we drive
362 * profiling and statistics off the main clock. This WILL NOT be accurate;
363 * do not do it unless absolutely necessary.
364 *
365 * The statistics clock may (or may not) be run at a higher rate while
366 * profiling. This profile clock runs at profhz. We require that profhz
367 * be an integral multiple of stathz.
368 *
369 * If the statistics clock is running fast, it must be divided by the ratio
370 * profhz/stathz for statistics. (For profiling, every tick counts.)
371 *
372 * Time-of-day is maintained using a "timecounter", which may or may
373 * not be related to the hardware generating the above mentioned
374 * interrupts.
375 */
376
377 int stathz;
378 int profhz;
379 int profprocs;
380 volatile int ticks;
381 int psratio;
382
383 static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */
384 #ifdef DEVICE_POLLING
385 static int devpoll_run = 0;
386 #endif
387
388 /*
389 * Initialize clock frequencies and start both clocks running.
390 */
391 /* ARGSUSED*/
392 static void
393 initclocks(dummy)
394 void *dummy;
395 {
396 register int i;
397
398 /*
399 * Set divisors to 1 (normal case) and let the machine-specific
400 * code do its bit.
401 */
402 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
403 cpu_initclocks();
404
405 /*
406 * Compute profhz/stathz, and fix profhz if needed.
407 */
408 i = stathz ? stathz : hz;
409 if (profhz == 0)
410 profhz = i;
411 psratio = profhz / i;
412 #ifdef SW_WATCHDOG
413 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
414 #endif
415 }
416
417 /*
418 * Each time the real-time timer fires, this function is called on all CPUs.
419 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
420 * the other CPUs in the system need to call this function.
421 */
422 void
423 hardclock_cpu(int usermode)
424 {
425 struct pstats *pstats;
426 struct thread *td = curthread;
427 struct proc *p = td->td_proc;
428 int flags;
429
430 /*
431 * Run current process's virtual and profile time, as needed.
432 */
433 pstats = p->p_stats;
434 flags = 0;
435 if (usermode &&
436 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
437 PROC_ITIMLOCK(p);
438 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
439 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
440 PROC_ITIMUNLOCK(p);
441 }
442 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
443 PROC_ITIMLOCK(p);
444 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
445 flags |= TDF_PROFPEND | TDF_ASTPENDING;
446 PROC_ITIMUNLOCK(p);
447 }
448 thread_lock(td);
449 td->td_flags |= flags;
450 thread_unlock(td);
451
452 #ifdef HWPMC_HOOKS
453 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
454 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
455 if (td->td_intr_frame != NULL)
456 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
457 #endif
458 callout_process(sbinuptime());
459 }
460
461 /*
462 * The real-time timer, interrupting hz times per second.
463 */
464 void
465 hardclock(int usermode, uintfptr_t pc)
466 {
467
468 atomic_add_int(&ticks, 1);
469 hardclock_cpu(usermode);
470 tc_ticktock(1);
471 cpu_tick_calibration();
472 /*
473 * If no separate statistics clock is available, run it from here.
474 *
475 * XXX: this only works for UP
476 */
477 if (stathz == 0) {
478 profclock(usermode, pc);
479 statclock(usermode);
480 }
481 #ifdef DEVICE_POLLING
482 hardclock_device_poll(); /* this is very short and quick */
483 #endif /* DEVICE_POLLING */
484 #ifdef SW_WATCHDOG
485 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
486 watchdog_fire();
487 #endif /* SW_WATCHDOG */
488 }
489
490 void
491 hardclock_cnt(int cnt, int usermode)
492 {
493 struct pstats *pstats;
494 struct thread *td = curthread;
495 struct proc *p = td->td_proc;
496 int *t = DPCPU_PTR(pcputicks);
497 int flags, global, newticks;
498 #ifdef SW_WATCHDOG
499 int i;
500 #endif /* SW_WATCHDOG */
501
502 /*
503 * Update per-CPU and possibly global ticks values.
504 */
505 *t += cnt;
506 do {
507 global = ticks;
508 newticks = *t - global;
509 if (newticks <= 0) {
510 if (newticks < -1)
511 *t = global - 1;
512 newticks = 0;
513 break;
514 }
515 } while (!atomic_cmpset_int(&ticks, global, *t));
516
517 /*
518 * Run current process's virtual and profile time, as needed.
519 */
520 pstats = p->p_stats;
521 flags = 0;
522 if (usermode &&
523 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
524 PROC_ITIMLOCK(p);
525 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
526 tick * cnt) == 0)
527 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
528 PROC_ITIMUNLOCK(p);
529 }
530 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
531 PROC_ITIMLOCK(p);
532 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
533 tick * cnt) == 0)
534 flags |= TDF_PROFPEND | TDF_ASTPENDING;
535 PROC_ITIMUNLOCK(p);
536 }
537 if (flags != 0) {
538 thread_lock(td);
539 td->td_flags |= flags;
540 thread_unlock(td);
541 }
542
543 #ifdef HWPMC_HOOKS
544 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
545 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
546 if (td->td_intr_frame != NULL)
547 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
548 #endif
549 /* We are in charge to handle this tick duty. */
550 if (newticks > 0) {
551 tc_ticktock(newticks);
552 #ifdef DEVICE_POLLING
553 /* Dangerous and no need to call these things concurrently. */
554 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
555 /* This is very short and quick. */
556 hardclock_device_poll();
557 atomic_store_rel_int(&devpoll_run, 0);
558 }
559 #endif /* DEVICE_POLLING */
560 #ifdef SW_WATCHDOG
561 if (watchdog_enabled > 0) {
562 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
563 if (i > 0 && i <= newticks)
564 watchdog_fire();
565 }
566 #endif /* SW_WATCHDOG */
567 }
568 if (curcpu == CPU_FIRST())
569 cpu_tick_calibration();
570 }
571
572 void
573 hardclock_sync(int cpu)
574 {
575 int *t = DPCPU_ID_PTR(cpu, pcputicks);
576
577 *t = ticks;
578 }
579
580 /*
581 * Compute number of ticks in the specified amount of time.
582 */
583 int
584 tvtohz(tv)
585 struct timeval *tv;
586 {
587 register unsigned long ticks;
588 register long sec, usec;
589
590 /*
591 * If the number of usecs in the whole seconds part of the time
592 * difference fits in a long, then the total number of usecs will
593 * fit in an unsigned long. Compute the total and convert it to
594 * ticks, rounding up and adding 1 to allow for the current tick
595 * to expire. Rounding also depends on unsigned long arithmetic
596 * to avoid overflow.
597 *
598 * Otherwise, if the number of ticks in the whole seconds part of
599 * the time difference fits in a long, then convert the parts to
600 * ticks separately and add, using similar rounding methods and
601 * overflow avoidance. This method would work in the previous
602 * case but it is slightly slower and assumes that hz is integral.
603 *
604 * Otherwise, round the time difference down to the maximum
605 * representable value.
606 *
607 * If ints have 32 bits, then the maximum value for any timeout in
608 * 10ms ticks is 248 days.
609 */
610 sec = tv->tv_sec;
611 usec = tv->tv_usec;
612 if (usec < 0) {
613 sec--;
614 usec += 1000000;
615 }
616 if (sec < 0) {
617 #ifdef DIAGNOSTIC
618 if (usec > 0) {
619 sec++;
620 usec -= 1000000;
621 }
622 printf("tvotohz: negative time difference %ld sec %ld usec\n",
623 sec, usec);
624 #endif
625 ticks = 1;
626 } else if (sec <= LONG_MAX / 1000000)
627 ticks = howmany(sec * 1000000 + (unsigned long)usec, tick) + 1;
628 else if (sec <= LONG_MAX / hz)
629 ticks = sec * hz
630 + howmany((unsigned long)usec, tick) + 1;
631 else
632 ticks = LONG_MAX;
633 if (ticks > INT_MAX)
634 ticks = INT_MAX;
635 return ((int)ticks);
636 }
637
638 /*
639 * Start profiling on a process.
640 *
641 * Kernel profiling passes proc0 which never exits and hence
642 * keeps the profile clock running constantly.
643 */
644 void
645 startprofclock(p)
646 register struct proc *p;
647 {
648
649 PROC_LOCK_ASSERT(p, MA_OWNED);
650 if (p->p_flag & P_STOPPROF)
651 return;
652 if ((p->p_flag & P_PROFIL) == 0) {
653 p->p_flag |= P_PROFIL;
654 mtx_lock(&time_lock);
655 if (++profprocs == 1)
656 cpu_startprofclock();
657 mtx_unlock(&time_lock);
658 }
659 }
660
661 /*
662 * Stop profiling on a process.
663 */
664 void
665 stopprofclock(p)
666 register struct proc *p;
667 {
668
669 PROC_LOCK_ASSERT(p, MA_OWNED);
670 if (p->p_flag & P_PROFIL) {
671 if (p->p_profthreads != 0) {
672 while (p->p_profthreads != 0) {
673 p->p_flag |= P_STOPPROF;
674 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
675 "stopprof", 0);
676 }
677 }
678 if ((p->p_flag & P_PROFIL) == 0)
679 return;
680 p->p_flag &= ~P_PROFIL;
681 mtx_lock(&time_lock);
682 if (--profprocs == 0)
683 cpu_stopprofclock();
684 mtx_unlock(&time_lock);
685 }
686 }
687
688 /*
689 * Statistics clock. Updates rusage information and calls the scheduler
690 * to adjust priorities of the active thread.
691 *
692 * This should be called by all active processors.
693 */
694 void
695 statclock(int usermode)
696 {
697
698 statclock_cnt(1, usermode);
699 }
700
701 void
702 statclock_cnt(int cnt, int usermode)
703 {
704 struct rusage *ru;
705 struct vmspace *vm;
706 struct thread *td;
707 struct proc *p;
708 long rss;
709 long *cp_time;
710
711 td = curthread;
712 p = td->td_proc;
713
714 cp_time = (long *)PCPU_PTR(cp_time);
715 if (usermode) {
716 /*
717 * Charge the time as appropriate.
718 */
719 td->td_uticks += cnt;
720 if (p->p_nice > NZERO)
721 cp_time[CP_NICE] += cnt;
722 else
723 cp_time[CP_USER] += cnt;
724 } else {
725 /*
726 * Came from kernel mode, so we were:
727 * - handling an interrupt,
728 * - doing syscall or trap work on behalf of the current
729 * user process, or
730 * - spinning in the idle loop.
731 * Whichever it is, charge the time as appropriate.
732 * Note that we charge interrupts to the current process,
733 * regardless of whether they are ``for'' that process,
734 * so that we know how much of its real time was spent
735 * in ``non-process'' (i.e., interrupt) work.
736 */
737 if ((td->td_pflags & TDP_ITHREAD) ||
738 td->td_intr_nesting_level >= 2) {
739 td->td_iticks += cnt;
740 cp_time[CP_INTR] += cnt;
741 } else {
742 td->td_pticks += cnt;
743 td->td_sticks += cnt;
744 if (!TD_IS_IDLETHREAD(td))
745 cp_time[CP_SYS] += cnt;
746 else
747 cp_time[CP_IDLE] += cnt;
748 }
749 }
750
751 /* Update resource usage integrals and maximums. */
752 MPASS(p->p_vmspace != NULL);
753 vm = p->p_vmspace;
754 ru = &td->td_ru;
755 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
756 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
757 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
758 rss = pgtok(vmspace_resident_count(vm));
759 if (ru->ru_maxrss < rss)
760 ru->ru_maxrss = rss;
761 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
762 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
763 SDT_PROBE2(sched, , , tick, td, td->td_proc);
764 thread_lock_flags(td, MTX_QUIET);
765 for ( ; cnt > 0; cnt--)
766 sched_clock(td);
767 thread_unlock(td);
768 #ifdef HWPMC_HOOKS
769 if (td->td_intr_frame != NULL)
770 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
771 #endif
772 }
773
774 void
775 profclock(int usermode, uintfptr_t pc)
776 {
777
778 profclock_cnt(1, usermode, pc);
779 }
780
781 void
782 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
783 {
784 struct thread *td;
785 #ifdef GPROF
786 struct gmonparam *g;
787 uintfptr_t i;
788 #endif
789
790 td = curthread;
791 if (usermode) {
792 /*
793 * Came from user mode; CPU was in user state.
794 * If this process is being profiled, record the tick.
795 * if there is no related user location yet, don't
796 * bother trying to count it.
797 */
798 if (td->td_proc->p_flag & P_PROFIL)
799 addupc_intr(td, pc, cnt);
800 }
801 #ifdef GPROF
802 else {
803 /*
804 * Kernel statistics are just like addupc_intr, only easier.
805 */
806 g = &_gmonparam;
807 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
808 i = PC_TO_I(g, pc);
809 if (i < g->textsize) {
810 KCOUNT(g, i) += cnt;
811 }
812 }
813 }
814 #endif
815 #ifdef HWPMC_HOOKS
816 if (td->td_intr_frame != NULL)
817 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
818 #endif
819 }
820
821 /*
822 * Return information about system clocks.
823 */
824 static int
825 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
826 {
827 struct clockinfo clkinfo;
828 /*
829 * Construct clockinfo structure.
830 */
831 bzero(&clkinfo, sizeof(clkinfo));
832 clkinfo.hz = hz;
833 clkinfo.tick = tick;
834 clkinfo.profhz = profhz;
835 clkinfo.stathz = stathz ? stathz : hz;
836 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
837 }
838
839 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
840 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
841 0, 0, sysctl_kern_clockrate, "S,clockinfo",
842 "Rate and period of various kernel clocks");
843
844 #ifdef SW_WATCHDOG
845
846 static void
847 watchdog_config(void *unused __unused, u_int cmd, int *error)
848 {
849 u_int u;
850
851 u = cmd & WD_INTERVAL;
852 if (u >= WD_TO_1SEC) {
853 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
854 watchdog_enabled = 1;
855 *error = 0;
856 } else {
857 watchdog_enabled = 0;
858 }
859 }
860
861 /*
862 * Handle a watchdog timeout by dumping interrupt information and
863 * then either dropping to DDB or panicking.
864 */
865 static void
866 watchdog_fire(void)
867 {
868 int nintr;
869 uint64_t inttotal;
870 u_long *curintr;
871 char *curname;
872
873 curintr = intrcnt;
874 curname = intrnames;
875 inttotal = 0;
876 nintr = sintrcnt / sizeof(u_long);
877
878 printf("interrupt total\n");
879 while (--nintr >= 0) {
880 if (*curintr)
881 printf("%-12s %20lu\n", curname, *curintr);
882 curname += strlen(curname) + 1;
883 inttotal += *curintr++;
884 }
885 printf("Total %20ju\n", (uintmax_t)inttotal);
886
887 #if defined(KDB) && !defined(KDB_UNATTENDED)
888 kdb_backtrace();
889 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
890 #else
891 panic("watchdog timeout");
892 #endif
893 }
894
895 #endif /* SW_WATCHDOG */
Cache object: dd72a6ea132c54c69c1d6d8ce086c0e9
|