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/9.1/sys/kern/kern_clock.c 236344 2012-05-30 23:22:52Z rstone $");
39
40 #include "opt_kdb.h"
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_kdtrace.h"
44 #include "opt_ntp.h"
45 #include "opt_watchdog.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/callout.h>
50 #include <sys/kdb.h>
51 #include <sys/kernel.h>
52 #include <sys/kthread.h>
53 #include <sys/ktr.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/resource.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/signalvar.h>
62 #include <sys/sleepqueue.h>
63 #include <sys/smp.h>
64 #include <vm/vm.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <sys/sysctl.h>
68 #include <sys/bus.h>
69 #include <sys/interrupt.h>
70 #include <sys/limits.h>
71 #include <sys/timetc.h>
72
73 #ifdef GPROF
74 #include <sys/gmon.h>
75 #endif
76
77 #ifdef HWPMC_HOOKS
78 #include <sys/pmckern.h>
79 PMC_SOFT_DEFINE( , , clock, hard);
80 PMC_SOFT_DEFINE( , , clock, stat);
81 #endif
82
83 #ifdef DEVICE_POLLING
84 extern void hardclock_device_poll(void);
85 #endif /* DEVICE_POLLING */
86
87 static void initclocks(void *dummy);
88 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
89
90 /* Spin-lock protecting profiling statistics. */
91 static struct mtx time_lock;
92
93 SDT_PROVIDER_DECLARE(sched);
94 SDT_PROBE_DEFINE2(sched, , , tick, tick, "struct thread *", "struct proc *");
95
96 static int
97 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
98 {
99 int error;
100 long cp_time[CPUSTATES];
101 #ifdef SCTL_MASK32
102 int i;
103 unsigned int cp_time32[CPUSTATES];
104 #endif
105
106 read_cpu_time(cp_time);
107 #ifdef SCTL_MASK32
108 if (req->flags & SCTL_MASK32) {
109 if (!req->oldptr)
110 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
111 for (i = 0; i < CPUSTATES; i++)
112 cp_time32[i] = (unsigned int)cp_time[i];
113 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
114 } else
115 #endif
116 {
117 if (!req->oldptr)
118 return SYSCTL_OUT(req, 0, sizeof(cp_time));
119 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
120 }
121 return error;
122 }
123
124 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
125 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
126
127 static long empty[CPUSTATES];
128
129 static int
130 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
131 {
132 struct pcpu *pcpu;
133 int error;
134 int c;
135 long *cp_time;
136 #ifdef SCTL_MASK32
137 unsigned int cp_time32[CPUSTATES];
138 int i;
139 #endif
140
141 if (!req->oldptr) {
142 #ifdef SCTL_MASK32
143 if (req->flags & SCTL_MASK32)
144 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
145 else
146 #endif
147 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
148 }
149 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
150 if (!CPU_ABSENT(c)) {
151 pcpu = pcpu_find(c);
152 cp_time = pcpu->pc_cp_time;
153 } else {
154 cp_time = empty;
155 }
156 #ifdef SCTL_MASK32
157 if (req->flags & SCTL_MASK32) {
158 for (i = 0; i < CPUSTATES; i++)
159 cp_time32[i] = (unsigned int)cp_time[i];
160 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
161 } else
162 #endif
163 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
164 }
165 return error;
166 }
167
168 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
169 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
170
171 #ifdef DEADLKRES
172 static const char *blessed[] = {
173 "getblk",
174 "so_snd_sx",
175 "so_rcv_sx",
176 NULL
177 };
178 static int slptime_threshold = 1800;
179 static int blktime_threshold = 900;
180 static int sleepfreq = 3;
181
182 static void
183 deadlkres(void)
184 {
185 struct proc *p;
186 struct thread *td;
187 void *wchan;
188 int blkticks, i, slpticks, slptype, tryl, tticks;
189
190 tryl = 0;
191 for (;;) {
192 blkticks = blktime_threshold * hz;
193 slpticks = slptime_threshold * hz;
194
195 /*
196 * Avoid to sleep on the sx_lock in order to avoid a possible
197 * priority inversion problem leading to starvation.
198 * If the lock can't be held after 100 tries, panic.
199 */
200 if (!sx_try_slock(&allproc_lock)) {
201 if (tryl > 100)
202 panic("%s: possible deadlock detected on allproc_lock\n",
203 __func__);
204 tryl++;
205 pause("allproc", sleepfreq * hz);
206 continue;
207 }
208 tryl = 0;
209 FOREACH_PROC_IN_SYSTEM(p) {
210 PROC_LOCK(p);
211 if (p->p_state == PRS_NEW) {
212 PROC_UNLOCK(p);
213 continue;
214 }
215 FOREACH_THREAD_IN_PROC(p, td) {
216
217 /*
218 * Once a thread is found in "interesting"
219 * state a possible ticks wrap-up needs to be
220 * checked.
221 */
222 thread_lock(td);
223 if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
224
225 /*
226 * The thread should be blocked on a
227 * turnstile, simply check if the
228 * turnstile channel is in good state.
229 */
230 MPASS(td->td_blocked != NULL);
231
232 tticks = ticks - td->td_blktick;
233 thread_unlock(td);
234 if (tticks > blkticks) {
235
236 /*
237 * Accordingly with provided
238 * thresholds, this thread is
239 * stuck for too long on a
240 * turnstile.
241 */
242 PROC_UNLOCK(p);
243 sx_sunlock(&allproc_lock);
244 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
245 __func__, td, tticks);
246 }
247 } else if (TD_IS_SLEEPING(td) &&
248 TD_ON_SLEEPQ(td) &&
249 ticks < td->td_blktick) {
250
251 /*
252 * Check if the thread is sleeping on a
253 * lock, otherwise skip the check.
254 * Drop the thread lock in order to
255 * avoid a LOR with the sleepqueue
256 * spinlock.
257 */
258 wchan = td->td_wchan;
259 tticks = ticks - td->td_slptick;
260 thread_unlock(td);
261 slptype = sleepq_type(wchan);
262 if ((slptype == SLEEPQ_SX ||
263 slptype == SLEEPQ_LK) &&
264 tticks > slpticks) {
265
266 /*
267 * Accordingly with provided
268 * thresholds, this thread is
269 * stuck for too long on a
270 * sleepqueue.
271 * However, being on a
272 * sleepqueue, we might still
273 * check for the blessed
274 * list.
275 */
276 tryl = 0;
277 for (i = 0; blessed[i] != NULL;
278 i++) {
279 if (!strcmp(blessed[i],
280 td->td_wmesg)) {
281 tryl = 1;
282 break;
283 }
284 }
285 if (tryl != 0) {
286 tryl = 0;
287 continue;
288 }
289 PROC_UNLOCK(p);
290 sx_sunlock(&allproc_lock);
291 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
292 __func__, td, tticks);
293 }
294 } else
295 thread_unlock(td);
296 }
297 PROC_UNLOCK(p);
298 }
299 sx_sunlock(&allproc_lock);
300
301 /* Sleep for sleepfreq seconds. */
302 pause("-", sleepfreq * hz);
303 }
304 }
305
306 static struct kthread_desc deadlkres_kd = {
307 "deadlkres",
308 deadlkres,
309 (struct thread **)NULL
310 };
311
312 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
313
314 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
315 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
316 &slptime_threshold, 0,
317 "Number of seconds within is valid to sleep on a sleepqueue");
318 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
319 &blktime_threshold, 0,
320 "Number of seconds within is valid to block on a turnstile");
321 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
322 "Number of seconds between any deadlock resolver thread run");
323 #endif /* DEADLKRES */
324
325 void
326 read_cpu_time(long *cp_time)
327 {
328 struct pcpu *pc;
329 int i, j;
330
331 /* Sum up global cp_time[]. */
332 bzero(cp_time, sizeof(long) * CPUSTATES);
333 CPU_FOREACH(i) {
334 pc = pcpu_find(i);
335 for (j = 0; j < CPUSTATES; j++)
336 cp_time[j] += pc->pc_cp_time[j];
337 }
338 }
339
340 #ifdef SW_WATCHDOG
341 #include <sys/watchdog.h>
342
343 static int watchdog_ticks;
344 static int watchdog_enabled;
345 static void watchdog_fire(void);
346 static void watchdog_config(void *, u_int, int *);
347 #endif /* SW_WATCHDOG */
348
349 /*
350 * Clock handling routines.
351 *
352 * This code is written to operate with two timers that run independently of
353 * each other.
354 *
355 * The main timer, running hz times per second, is used to trigger interval
356 * timers, timeouts and rescheduling as needed.
357 *
358 * The second timer handles kernel and user profiling,
359 * and does resource use estimation. If the second timer is programmable,
360 * it is randomized to avoid aliasing between the two clocks. For example,
361 * the randomization prevents an adversary from always giving up the cpu
362 * just before its quantum expires. Otherwise, it would never accumulate
363 * cpu ticks. The mean frequency of the second timer is stathz.
364 *
365 * If no second timer exists, stathz will be zero; in this case we drive
366 * profiling and statistics off the main clock. This WILL NOT be accurate;
367 * do not do it unless absolutely necessary.
368 *
369 * The statistics clock may (or may not) be run at a higher rate while
370 * profiling. This profile clock runs at profhz. We require that profhz
371 * be an integral multiple of stathz.
372 *
373 * If the statistics clock is running fast, it must be divided by the ratio
374 * profhz/stathz for statistics. (For profiling, every tick counts.)
375 *
376 * Time-of-day is maintained using a "timecounter", which may or may
377 * not be related to the hardware generating the above mentioned
378 * interrupts.
379 */
380
381 int stathz;
382 int profhz;
383 int profprocs;
384 int ticks;
385 int psratio;
386
387 static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */
388 static int global_hardclock_run = 0;
389
390 /*
391 * Initialize clock frequencies and start both clocks running.
392 */
393 /* ARGSUSED*/
394 static void
395 initclocks(dummy)
396 void *dummy;
397 {
398 register int i;
399
400 /*
401 * Set divisors to 1 (normal case) and let the machine-specific
402 * code do its bit.
403 */
404 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
405 cpu_initclocks();
406
407 /*
408 * Compute profhz/stathz, and fix profhz if needed.
409 */
410 i = stathz ? stathz : hz;
411 if (profhz == 0)
412 profhz = i;
413 psratio = profhz / i;
414 #ifdef SW_WATCHDOG
415 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
416 #endif
417 }
418
419 /*
420 * Each time the real-time timer fires, this function is called on all CPUs.
421 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
422 * the other CPUs in the system need to call this function.
423 */
424 void
425 hardclock_cpu(int usermode)
426 {
427 struct pstats *pstats;
428 struct thread *td = curthread;
429 struct proc *p = td->td_proc;
430 int flags;
431
432 /*
433 * Run current process's virtual and profile time, as needed.
434 */
435 pstats = p->p_stats;
436 flags = 0;
437 if (usermode &&
438 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
439 PROC_SLOCK(p);
440 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
441 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
442 PROC_SUNLOCK(p);
443 }
444 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
445 PROC_SLOCK(p);
446 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
447 flags |= TDF_PROFPEND | TDF_ASTPENDING;
448 PROC_SUNLOCK(p);
449 }
450 thread_lock(td);
451 sched_tick(1);
452 td->td_flags |= flags;
453 thread_unlock(td);
454
455 #ifdef HWPMC_HOOKS
456 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
457 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
458 if (td->td_intr_frame != NULL)
459 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
460 #endif
461 callout_tick();
462 }
463
464 /*
465 * The real-time timer, interrupting hz times per second.
466 */
467 void
468 hardclock(int usermode, uintfptr_t pc)
469 {
470
471 atomic_add_int((volatile int *)&ticks, 1);
472 hardclock_cpu(usermode);
473 tc_ticktock(1);
474 cpu_tick_calibration();
475 /*
476 * If no separate statistics clock is available, run it from here.
477 *
478 * XXX: this only works for UP
479 */
480 if (stathz == 0) {
481 profclock(usermode, pc);
482 statclock(usermode);
483 }
484 #ifdef DEVICE_POLLING
485 hardclock_device_poll(); /* this is very short and quick */
486 #endif /* DEVICE_POLLING */
487 #ifdef SW_WATCHDOG
488 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
489 watchdog_fire();
490 #endif /* SW_WATCHDOG */
491 }
492
493 void
494 hardclock_cnt(int cnt, int usermode)
495 {
496 struct pstats *pstats;
497 struct thread *td = curthread;
498 struct proc *p = td->td_proc;
499 int *t = DPCPU_PTR(pcputicks);
500 int flags, global, newticks;
501 #ifdef SW_WATCHDOG
502 int i;
503 #endif /* SW_WATCHDOG */
504
505 /*
506 * Update per-CPU and possibly global ticks values.
507 */
508 *t += cnt;
509 do {
510 global = ticks;
511 newticks = *t - global;
512 if (newticks <= 0) {
513 if (newticks < -1)
514 *t = global - 1;
515 newticks = 0;
516 break;
517 }
518 } while (!atomic_cmpset_int(&ticks, global, *t));
519
520 /*
521 * Run current process's virtual and profile time, as needed.
522 */
523 pstats = p->p_stats;
524 flags = 0;
525 if (usermode &&
526 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
527 PROC_SLOCK(p);
528 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
529 tick * cnt) == 0)
530 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
531 PROC_SUNLOCK(p);
532 }
533 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
534 PROC_SLOCK(p);
535 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
536 tick * cnt) == 0)
537 flags |= TDF_PROFPEND | TDF_ASTPENDING;
538 PROC_SUNLOCK(p);
539 }
540 thread_lock(td);
541 sched_tick(cnt);
542 td->td_flags |= flags;
543 thread_unlock(td);
544
545 #ifdef HWPMC_HOOKS
546 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
547 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
548 if (td->td_intr_frame != NULL)
549 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
550 #endif
551 callout_tick();
552 /* We are in charge to handle this tick duty. */
553 if (newticks > 0) {
554 /* Dangerous and no need to call these things concurrently. */
555 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
556 tc_ticktock(newticks);
557 #ifdef DEVICE_POLLING
558 /* This is very short and quick. */
559 hardclock_device_poll();
560 #endif /* DEVICE_POLLING */
561 atomic_store_rel_int(&global_hardclock_run, 0);
562 }
563 #ifdef SW_WATCHDOG
564 if (watchdog_enabled > 0) {
565 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
566 if (i > 0 && i <= newticks)
567 watchdog_fire();
568 }
569 #endif /* SW_WATCHDOG */
570 }
571 if (curcpu == CPU_FIRST())
572 cpu_tick_calibration();
573 }
574
575 void
576 hardclock_sync(int cpu)
577 {
578 int *t = DPCPU_ID_PTR(cpu, pcputicks);
579
580 *t = ticks;
581 }
582
583 /*
584 * Compute number of ticks in the specified amount of time.
585 */
586 int
587 tvtohz(tv)
588 struct timeval *tv;
589 {
590 register unsigned long ticks;
591 register long sec, usec;
592
593 /*
594 * If the number of usecs in the whole seconds part of the time
595 * difference fits in a long, then the total number of usecs will
596 * fit in an unsigned long. Compute the total and convert it to
597 * ticks, rounding up and adding 1 to allow for the current tick
598 * to expire. Rounding also depends on unsigned long arithmetic
599 * to avoid overflow.
600 *
601 * Otherwise, if the number of ticks in the whole seconds part of
602 * the time difference fits in a long, then convert the parts to
603 * ticks separately and add, using similar rounding methods and
604 * overflow avoidance. This method would work in the previous
605 * case but it is slightly slower and assumes that hz is integral.
606 *
607 * Otherwise, round the time difference down to the maximum
608 * representable value.
609 *
610 * If ints have 32 bits, then the maximum value for any timeout in
611 * 10ms ticks is 248 days.
612 */
613 sec = tv->tv_sec;
614 usec = tv->tv_usec;
615 if (usec < 0) {
616 sec--;
617 usec += 1000000;
618 }
619 if (sec < 0) {
620 #ifdef DIAGNOSTIC
621 if (usec > 0) {
622 sec++;
623 usec -= 1000000;
624 }
625 printf("tvotohz: negative time difference %ld sec %ld usec\n",
626 sec, usec);
627 #endif
628 ticks = 1;
629 } else if (sec <= LONG_MAX / 1000000)
630 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
631 / tick + 1;
632 else if (sec <= LONG_MAX / hz)
633 ticks = sec * hz
634 + ((unsigned long)usec + (tick - 1)) / tick + 1;
635 else
636 ticks = LONG_MAX;
637 if (ticks > INT_MAX)
638 ticks = INT_MAX;
639 return ((int)ticks);
640 }
641
642 /*
643 * Start profiling on a process.
644 *
645 * Kernel profiling passes proc0 which never exits and hence
646 * keeps the profile clock running constantly.
647 */
648 void
649 startprofclock(p)
650 register struct proc *p;
651 {
652
653 PROC_LOCK_ASSERT(p, MA_OWNED);
654 if (p->p_flag & P_STOPPROF)
655 return;
656 if ((p->p_flag & P_PROFIL) == 0) {
657 p->p_flag |= P_PROFIL;
658 mtx_lock(&time_lock);
659 if (++profprocs == 1)
660 cpu_startprofclock();
661 mtx_unlock(&time_lock);
662 }
663 }
664
665 /*
666 * Stop profiling on a process.
667 */
668 void
669 stopprofclock(p)
670 register struct proc *p;
671 {
672
673 PROC_LOCK_ASSERT(p, MA_OWNED);
674 if (p->p_flag & P_PROFIL) {
675 if (p->p_profthreads != 0) {
676 p->p_flag |= P_STOPPROF;
677 while (p->p_profthreads != 0)
678 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
679 "stopprof", 0);
680 p->p_flag &= ~P_STOPPROF;
681 }
682 if ((p->p_flag & P_PROFIL) == 0)
683 return;
684 p->p_flag &= ~P_PROFIL;
685 mtx_lock(&time_lock);
686 if (--profprocs == 0)
687 cpu_stopprofclock();
688 mtx_unlock(&time_lock);
689 }
690 }
691
692 /*
693 * Statistics clock. Updates rusage information and calls the scheduler
694 * to adjust priorities of the active thread.
695 *
696 * This should be called by all active processors.
697 */
698 void
699 statclock(int usermode)
700 {
701
702 statclock_cnt(1, usermode);
703 }
704
705 void
706 statclock_cnt(int cnt, int usermode)
707 {
708 struct rusage *ru;
709 struct vmspace *vm;
710 struct thread *td;
711 struct proc *p;
712 long rss;
713 long *cp_time;
714
715 td = curthread;
716 p = td->td_proc;
717
718 cp_time = (long *)PCPU_PTR(cp_time);
719 if (usermode) {
720 /*
721 * Charge the time as appropriate.
722 */
723 td->td_uticks += cnt;
724 if (p->p_nice > NZERO)
725 cp_time[CP_NICE] += cnt;
726 else
727 cp_time[CP_USER] += cnt;
728 } else {
729 /*
730 * Came from kernel mode, so we were:
731 * - handling an interrupt,
732 * - doing syscall or trap work on behalf of the current
733 * user process, or
734 * - spinning in the idle loop.
735 * Whichever it is, charge the time as appropriate.
736 * Note that we charge interrupts to the current process,
737 * regardless of whether they are ``for'' that process,
738 * so that we know how much of its real time was spent
739 * in ``non-process'' (i.e., interrupt) work.
740 */
741 if ((td->td_pflags & TDP_ITHREAD) ||
742 td->td_intr_nesting_level >= 2) {
743 td->td_iticks += cnt;
744 cp_time[CP_INTR] += cnt;
745 } else {
746 td->td_pticks += cnt;
747 td->td_sticks += cnt;
748 if (!TD_IS_IDLETHREAD(td))
749 cp_time[CP_SYS] += cnt;
750 else
751 cp_time[CP_IDLE] += cnt;
752 }
753 }
754
755 /* Update resource usage integrals and maximums. */
756 MPASS(p->p_vmspace != NULL);
757 vm = p->p_vmspace;
758 ru = &td->td_ru;
759 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
760 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
761 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
762 rss = pgtok(vmspace_resident_count(vm));
763 if (ru->ru_maxrss < rss)
764 ru->ru_maxrss = rss;
765 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
766 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
767 SDT_PROBE2(sched, , , tick, td, td->td_proc);
768 thread_lock_flags(td, MTX_QUIET);
769 for ( ; cnt > 0; cnt--)
770 sched_clock(td);
771 thread_unlock(td);
772 #ifdef HWPMC_HOOKS
773 if (td->td_intr_frame != NULL)
774 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
775 #endif
776 }
777
778 void
779 profclock(int usermode, uintfptr_t pc)
780 {
781
782 profclock_cnt(1, usermode, pc);
783 }
784
785 void
786 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
787 {
788 struct thread *td;
789 #ifdef GPROF
790 struct gmonparam *g;
791 uintfptr_t i;
792 #endif
793
794 td = curthread;
795 if (usermode) {
796 /*
797 * Came from user mode; CPU was in user state.
798 * If this process is being profiled, record the tick.
799 * if there is no related user location yet, don't
800 * bother trying to count it.
801 */
802 if (td->td_proc->p_flag & P_PROFIL)
803 addupc_intr(td, pc, cnt);
804 }
805 #ifdef GPROF
806 else {
807 /*
808 * Kernel statistics are just like addupc_intr, only easier.
809 */
810 g = &_gmonparam;
811 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
812 i = PC_TO_I(g, pc);
813 if (i < g->textsize) {
814 KCOUNT(g, i) += cnt;
815 }
816 }
817 }
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: 85c2efa0ae56c07eb3cd17b38149716b
|