1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
37 */
38
39 #include <sys/cdefs.h>
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/epoch.h>
50 #include <sys/eventhandler.h>
51 #include <sys/gtaskqueue.h>
52 #include <sys/kdb.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/ktr.h>
56 #include <sys/lock.h>
57 #include <sys/mutex.h>
58 #include <sys/proc.h>
59 #include <sys/resource.h>
60 #include <sys/resourcevar.h>
61 #include <sys/sched.h>
62 #include <sys/sdt.h>
63 #include <sys/signalvar.h>
64 #include <sys/sleepqueue.h>
65 #include <sys/smp.h>
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_map.h>
69 #include <sys/sysctl.h>
70 #include <sys/bus.h>
71 #include <sys/interrupt.h>
72 #include <sys/limits.h>
73 #include <sys/timetc.h>
74
75 #ifdef HWPMC_HOOKS
76 #include <sys/pmckern.h>
77 PMC_SOFT_DEFINE( , , clock, hard);
78 PMC_SOFT_DEFINE( , , clock, stat);
79 PMC_SOFT_DEFINE_EX( , , clock, prof, \
80 cpu_startprofclock, cpu_stopprofclock);
81 #endif
82
83 #ifdef DEVICE_POLLING
84 extern void hardclock_device_poll(void);
85 #endif /* DEVICE_POLLING */
86
87 /* Spin-lock protecting profiling statistics. */
88 static struct mtx time_lock;
89
90 SDT_PROVIDER_DECLARE(sched);
91 SDT_PROBE_DEFINE2(sched, , , tick, "struct thread *", "struct proc *");
92
93 static int
sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)94 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
95 {
96 int error;
97 long cp_time[CPUSTATES];
98 #ifdef SCTL_MASK32
99 int i;
100 unsigned int cp_time32[CPUSTATES];
101 #endif
102
103 read_cpu_time(cp_time);
104 #ifdef SCTL_MASK32
105 if (req->flags & SCTL_MASK32) {
106 if (!req->oldptr)
107 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
108 for (i = 0; i < CPUSTATES; i++)
109 cp_time32[i] = (unsigned int)cp_time[i];
110 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
111 } else
112 #endif
113 {
114 if (!req->oldptr)
115 return SYSCTL_OUT(req, 0, sizeof(cp_time));
116 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
117 }
118 return error;
119 }
120
121 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
122 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
123
124 static long empty[CPUSTATES];
125
126 static int
sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)127 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
128 {
129 struct pcpu *pcpu;
130 int error;
131 int c;
132 long *cp_time;
133 #ifdef SCTL_MASK32
134 unsigned int cp_time32[CPUSTATES];
135 int i;
136 #endif
137
138 if (!req->oldptr) {
139 #ifdef SCTL_MASK32
140 if (req->flags & SCTL_MASK32)
141 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
142 else
143 #endif
144 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
145 }
146 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
147 if (!CPU_ABSENT(c)) {
148 pcpu = pcpu_find(c);
149 cp_time = pcpu->pc_cp_time;
150 } else {
151 cp_time = empty;
152 }
153 #ifdef SCTL_MASK32
154 if (req->flags & SCTL_MASK32) {
155 for (i = 0; i < CPUSTATES; i++)
156 cp_time32[i] = (unsigned int)cp_time[i];
157 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
158 } else
159 #endif
160 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
161 }
162 return error;
163 }
164
165 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
166 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
167
168 #ifdef DEADLKRES
169 static const char *blessed[] = {
170 "getblk",
171 "so_snd_sx",
172 "so_rcv_sx",
173 NULL
174 };
175 static int slptime_threshold = 1800;
176 static int blktime_threshold = 900;
177 static int sleepfreq = 3;
178
179 static void
deadlres_td_on_lock(struct proc * p,struct thread * td,int blkticks)180 deadlres_td_on_lock(struct proc *p, struct thread *td, int blkticks)
181 {
182 int tticks;
183
184 sx_assert(&allproc_lock, SX_LOCKED);
185 PROC_LOCK_ASSERT(p, MA_OWNED);
186 THREAD_LOCK_ASSERT(td, MA_OWNED);
187 /*
188 * The thread should be blocked on a turnstile, simply check
189 * if the turnstile channel is in good state.
190 */
191 MPASS(td->td_blocked != NULL);
192
193 tticks = ticks - td->td_blktick;
194 if (tticks > blkticks)
195 /*
196 * Accordingly with provided thresholds, this thread is stuck
197 * for too long on a turnstile.
198 */
199 panic("%s: possible deadlock detected for %p (%s), "
200 "blocked for %d ticks\n", __func__,
201 td, sched_tdname(td), tticks);
202 }
203
204 static void
deadlres_td_sleep_q(struct proc * p,struct thread * td,int slpticks)205 deadlres_td_sleep_q(struct proc *p, struct thread *td, int slpticks)
206 {
207 const void *wchan;
208 int i, slptype, tticks;
209
210 sx_assert(&allproc_lock, SX_LOCKED);
211 PROC_LOCK_ASSERT(p, MA_OWNED);
212 THREAD_LOCK_ASSERT(td, MA_OWNED);
213 /*
214 * Check if the thread is sleeping on a lock, otherwise skip the check.
215 * Drop the thread lock in order to avoid a LOR with the sleepqueue
216 * spinlock.
217 */
218 wchan = td->td_wchan;
219 tticks = ticks - td->td_slptick;
220 slptype = sleepq_type(wchan);
221 if ((slptype == SLEEPQ_SX || slptype == SLEEPQ_LK) &&
222 tticks > slpticks) {
223 /*
224 * Accordingly with provided thresholds, this thread is stuck
225 * for too long on a sleepqueue.
226 * However, being on a sleepqueue, we might still check for the
227 * blessed list.
228 */
229 for (i = 0; blessed[i] != NULL; i++)
230 if (!strcmp(blessed[i], td->td_wmesg))
231 return;
232
233 panic("%s: possible deadlock detected for %p (%s), "
234 "blocked for %d ticks\n", __func__,
235 td, sched_tdname(td), tticks);
236 }
237 }
238
239 static void
deadlkres(void)240 deadlkres(void)
241 {
242 struct proc *p;
243 struct thread *td;
244 int blkticks, slpticks, tryl;
245
246 tryl = 0;
247 for (;;) {
248 blkticks = blktime_threshold * hz;
249 slpticks = slptime_threshold * hz;
250
251 /*
252 * Avoid to sleep on the sx_lock in order to avoid a
253 * possible priority inversion problem leading to
254 * starvation.
255 * If the lock can't be held after 100 tries, panic.
256 */
257 if (!sx_try_slock(&allproc_lock)) {
258 if (tryl > 100)
259 panic("%s: possible deadlock detected "
260 "on allproc_lock\n", __func__);
261 tryl++;
262 pause("allproc", sleepfreq * hz);
263 continue;
264 }
265 tryl = 0;
266 FOREACH_PROC_IN_SYSTEM(p) {
267 PROC_LOCK(p);
268 if (p->p_state == PRS_NEW) {
269 PROC_UNLOCK(p);
270 continue;
271 }
272 FOREACH_THREAD_IN_PROC(p, td) {
273 thread_lock(td);
274 if (TD_ON_LOCK(td))
275 deadlres_td_on_lock(p, td,
276 blkticks);
277 else if (TD_IS_SLEEPING(td))
278 deadlres_td_sleep_q(p, td,
279 slpticks);
280 thread_unlock(td);
281 }
282 PROC_UNLOCK(p);
283 }
284 sx_sunlock(&allproc_lock);
285
286 /* Sleep for sleepfreq seconds. */
287 pause("-", sleepfreq * hz);
288 }
289 }
290
291 static struct kthread_desc deadlkres_kd = {
292 "deadlkres",
293 deadlkres,
294 (struct thread **)NULL
295 };
296
297 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
298
299 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
300 "Deadlock resolver");
301 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RWTUN,
302 &slptime_threshold, 0,
303 "Number of seconds within is valid to sleep on a sleepqueue");
304 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RWTUN,
305 &blktime_threshold, 0,
306 "Number of seconds within is valid to block on a turnstile");
307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RWTUN, &sleepfreq, 0,
308 "Number of seconds between any deadlock resolver thread run");
309 #endif /* DEADLKRES */
310
311 void
read_cpu_time(long * cp_time)312 read_cpu_time(long *cp_time)
313 {
314 struct pcpu *pc;
315 int i, j;
316
317 /* Sum up global cp_time[]. */
318 bzero(cp_time, sizeof(long) * CPUSTATES);
319 CPU_FOREACH(i) {
320 pc = pcpu_find(i);
321 for (j = 0; j < CPUSTATES; j++)
322 cp_time[j] += pc->pc_cp_time[j];
323 }
324 }
325
326 #include <sys/watchdog.h>
327
328 static long watchdog_ticks;
329 static int watchdog_enabled;
330 static void watchdog_fire(void);
331 static void watchdog_config(void *, u_int, int *);
332
333 static void
watchdog_attach(void)334 watchdog_attach(void)
335 {
336 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
337 }
338
339 /*
340 * Clock handling routines.
341 *
342 * This code is written to operate with two timers that run independently of
343 * each other.
344 *
345 * The main timer, running hz times per second, is used to trigger interval
346 * timers, timeouts and rescheduling as needed.
347 *
348 * The second timer handles kernel and user profiling,
349 * and does resource use estimation. If the second timer is programmable,
350 * it is randomized to avoid aliasing between the two clocks. For example,
351 * the randomization prevents an adversary from always giving up the cpu
352 * just before its quantum expires. Otherwise, it would never accumulate
353 * cpu ticks. The mean frequency of the second timer is stathz.
354 *
355 * If no second timer exists, stathz will be zero; in this case we drive
356 * profiling and statistics off the main clock. This WILL NOT be accurate;
357 * do not do it unless absolutely necessary.
358 *
359 * The statistics clock may (or may not) be run at a higher rate while
360 * profiling. This profile clock runs at profhz. We require that profhz
361 * be an integral multiple of stathz.
362 *
363 * If the statistics clock is running fast, it must be divided by the ratio
364 * profhz/stathz for statistics. (For profiling, every tick counts.)
365 *
366 * Time-of-day is maintained using a "timecounter", which may or may
367 * not be related to the hardware generating the above mentioned
368 * interrupts.
369 */
370
371 int stathz;
372 int profhz;
373 int profprocs;
374 int psratio;
375
376 DPCPU_DEFINE_STATIC(long, pcputicks); /* Per-CPU version of ticks. */
377 #ifdef DEVICE_POLLING
378 static int devpoll_run = 0;
379 #endif
380
381 static void
ast_oweupc(struct thread * td,int tda __unused)382 ast_oweupc(struct thread *td, int tda __unused)
383 {
384 if ((td->td_proc->p_flag & P_PROFIL) == 0)
385 return;
386 addupc_task(td, td->td_profil_addr, td->td_profil_ticks);
387 td->td_profil_ticks = 0;
388 td->td_pflags &= ~TDP_OWEUPC;
389 }
390
391 static void
ast_alrm(struct thread * td,int tda __unused)392 ast_alrm(struct thread *td, int tda __unused)
393 {
394 struct proc *p;
395
396 p = td->td_proc;
397 PROC_LOCK(p);
398 kern_psignal(p, SIGVTALRM);
399 PROC_UNLOCK(p);
400 }
401
402 static void
ast_prof(struct thread * td,int tda __unused)403 ast_prof(struct thread *td, int tda __unused)
404 {
405 struct proc *p;
406
407 p = td->td_proc;
408 PROC_LOCK(p);
409 kern_psignal(p, SIGPROF);
410 PROC_UNLOCK(p);
411 }
412
413 /*
414 * Initialize clock frequencies and start both clocks running.
415 */
416 static void
initclocks(void * dummy __unused)417 initclocks(void *dummy __unused)
418 {
419 int i;
420
421 /*
422 * Set divisors to 1 (normal case) and let the machine-specific
423 * code do its bit.
424 */
425 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
426 cpu_initclocks();
427
428 /*
429 * Compute profhz/stathz, and fix profhz if needed.
430 */
431 i = stathz ? stathz : hz;
432 if (profhz == 0)
433 profhz = i;
434 psratio = profhz / i;
435
436 ast_register(TDA_OWEUPC, ASTR_ASTF_REQUIRED, 0, ast_oweupc);
437 ast_register(TDA_ALRM, ASTR_ASTF_REQUIRED, 0, ast_alrm);
438 ast_register(TDA_PROF, ASTR_ASTF_REQUIRED, 0, ast_prof);
439
440 #ifdef SW_WATCHDOG
441 /* Enable hardclock watchdog now, even if a hardware watchdog exists. */
442 watchdog_attach();
443 #else
444 /* Volunteer to run a software watchdog. */
445 if (wdog_software_attach == NULL)
446 wdog_software_attach = watchdog_attach;
447 #endif
448 }
449 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
450
451 static __noinline void
hardclock_itimer(struct thread * td,struct pstats * pstats,int cnt,int usermode)452 hardclock_itimer(struct thread *td, struct pstats *pstats, int cnt, int usermode)
453 {
454 struct proc *p;
455 int ast;
456
457 ast = 0;
458 p = td->td_proc;
459 if (usermode &&
460 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
461 PROC_ITIMLOCK(p);
462 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
463 tick * cnt) == 0)
464 ast |= TDAI(TDA_ALRM);
465 PROC_ITIMUNLOCK(p);
466 }
467 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
468 PROC_ITIMLOCK(p);
469 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
470 tick * cnt) == 0)
471 ast |= TDAI(TDA_PROF);
472 PROC_ITIMUNLOCK(p);
473 }
474 if (ast != 0)
475 ast_sched_mask(td, ast);
476 }
477
478 void
hardclock(int cnt,int usermode)479 hardclock(int cnt, int usermode)
480 {
481 struct pstats *pstats;
482 struct thread *td = curthread;
483 struct proc *p = td->td_proc;
484 long global, newticks, *t;
485
486 /*
487 * Update per-CPU and possibly global ticks values.
488 */
489 t = DPCPU_PTR(pcputicks);
490 *t += cnt;
491 global = atomic_load_long(&ticksl);
492 do {
493 newticks = *t - global;
494 if (newticks <= 0) {
495 if (newticks < -1)
496 *t = global - 1;
497 newticks = 0;
498 break;
499 }
500 } while (!atomic_fcmpset_long(&ticksl, &global, *t));
501
502 /*
503 * Run current process's virtual and profile time, as needed.
504 */
505 pstats = p->p_stats;
506 if (__predict_false(
507 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) ||
508 timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)))
509 hardclock_itimer(td, pstats, cnt, usermode);
510
511 #ifdef HWPMC_HOOKS
512 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
513 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
514 if (td->td_intr_frame != NULL)
515 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
516 #endif
517 /* We are in charge to handle this tick duty. */
518 if (newticks > 0) {
519 tc_ticktock(newticks);
520 #ifdef DEVICE_POLLING
521 /* Dangerous and no need to call these things concurrently. */
522 if (atomic_cmpset_acq_int(&devpoll_run, 0, 1)) {
523 /* This is very short and quick. */
524 hardclock_device_poll();
525 atomic_store_rel_int(&devpoll_run, 0);
526 }
527 #endif /* DEVICE_POLLING */
528 if (watchdog_enabled > 0) {
529 long left;
530
531 left = atomic_fetchadd_long(&watchdog_ticks, -newticks);
532 if (left > 0 && left <= newticks)
533 watchdog_fire();
534 }
535 intr_event_handle(clk_intr_event, NULL);
536 }
537 if (curcpu == CPU_FIRST())
538 cpu_tick_calibration();
539 if (__predict_false(DPCPU_GET(epoch_cb_count)))
540 GROUPTASK_ENQUEUE(DPCPU_PTR(epoch_cb_task));
541 }
542
543 void
hardclock_sync(int cpu)544 hardclock_sync(int cpu)
545 {
546 long *t;
547
548 KASSERT(!CPU_ABSENT(cpu), ("Absent CPU %d", cpu));
549
550 t = DPCPU_ID_PTR(cpu, pcputicks);
551 *t = ticksl;
552 }
553
554 /*
555 * Regular integer scaling formula without losing precision:
556 */
557 #define TIME_INT_SCALE(value, mul, div) \
558 (((value) / (div)) * (mul) + (((value) % (div)) * (mul)) / (div))
559
560 /*
561 * Macro for converting seconds and microseconds into actual ticks,
562 * based on the given hz value:
563 */
564 #define TIME_TO_TICKS(sec, usec, hz) \
565 ((sec) * (hz) + TIME_INT_SCALE(usec, hz, 1 << 6) / (1000000 >> 6))
566
567 #define TIME_ASSERT_VALID_HZ(hz) \
568 _Static_assert(TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) >= 0 && \
569 TIME_TO_TICKS(INT_MAX / (hz) - 1, 999999, hz) < INT_MAX, \
570 "tvtohz() can overflow the regular integer type")
571
572 /*
573 * Compile time assert the maximum and minimum values to fit into a
574 * regular integer when computing TIME_TO_TICKS():
575 */
576 TIME_ASSERT_VALID_HZ(HZ_MAXIMUM);
577 TIME_ASSERT_VALID_HZ(HZ_MINIMUM);
578
579 /*
580 * The formula is mostly linear, but test some more common values just
581 * in case:
582 */
583 TIME_ASSERT_VALID_HZ(1024);
584 TIME_ASSERT_VALID_HZ(1000);
585 TIME_ASSERT_VALID_HZ(128);
586 TIME_ASSERT_VALID_HZ(100);
587
588 /*
589 * Compute number of ticks representing the specified amount of time.
590 * If the specified time is negative, a value of 1 is returned. This
591 * function returns a value from 1 up to and including INT_MAX.
592 */
593 int
tvtohz(struct timeval * tv)594 tvtohz(struct timeval *tv)
595 {
596 int retval;
597
598 /*
599 * The values passed here may come from user-space and these
600 * checks ensure "tv_usec" is within its allowed range:
601 */
602
603 /* check for tv_usec underflow */
604 if (__predict_false(tv->tv_usec < 0)) {
605 tv->tv_sec += tv->tv_usec / 1000000;
606 tv->tv_usec = tv->tv_usec % 1000000;
607 /* convert tv_usec to a positive value */
608 if (__predict_true(tv->tv_usec < 0)) {
609 tv->tv_usec += 1000000;
610 tv->tv_sec -= 1;
611 }
612 /* check for tv_usec overflow */
613 } else if (__predict_false(tv->tv_usec >= 1000000)) {
614 tv->tv_sec += tv->tv_usec / 1000000;
615 tv->tv_usec = tv->tv_usec % 1000000;
616 }
617
618 /* check for tv_sec underflow */
619 if (__predict_false(tv->tv_sec < 0))
620 return (1);
621 /* check for tv_sec overflow (including room for the tv_usec part) */
622 else if (__predict_false(tv->tv_sec >= tick_seconds_max))
623 return (INT_MAX);
624
625 /* cast to "int" to avoid platform differences */
626 retval = TIME_TO_TICKS((int)tv->tv_sec, (int)tv->tv_usec, hz);
627
628 /* add one additional tick */
629 return (retval + 1);
630 }
631
632 /*
633 * Start profiling on a process.
634 *
635 * Kernel profiling passes proc0 which never exits and hence
636 * keeps the profile clock running constantly.
637 */
638 void
startprofclock(struct proc * p)639 startprofclock(struct proc *p)
640 {
641
642 PROC_LOCK_ASSERT(p, MA_OWNED);
643 if (p->p_flag & P_STOPPROF)
644 return;
645 if ((p->p_flag & P_PROFIL) == 0) {
646 p->p_flag |= P_PROFIL;
647 mtx_lock(&time_lock);
648 if (++profprocs == 1)
649 cpu_startprofclock();
650 mtx_unlock(&time_lock);
651 }
652 }
653
654 /*
655 * Stop profiling on a process.
656 */
657 void
stopprofclock(struct proc * p)658 stopprofclock(struct proc *p)
659 {
660
661 PROC_LOCK_ASSERT(p, MA_OWNED);
662 if (p->p_flag & P_PROFIL) {
663 if (p->p_profthreads != 0) {
664 while (p->p_profthreads != 0) {
665 p->p_flag |= P_STOPPROF;
666 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
667 "stopprof", 0);
668 }
669 }
670 if ((p->p_flag & P_PROFIL) == 0)
671 return;
672 p->p_flag &= ~P_PROFIL;
673 mtx_lock(&time_lock);
674 if (--profprocs == 0)
675 cpu_stopprofclock();
676 mtx_unlock(&time_lock);
677 }
678 }
679
680 /*
681 * Statistics clock. Updates rusage information and calls the scheduler
682 * to adjust priorities of the active thread.
683 *
684 * This should be called by all active processors.
685 */
686 void
statclock(int cnt,int usermode)687 statclock(int cnt, int usermode)
688 {
689 struct rusage *ru;
690 struct vmspace *vm;
691 struct thread *td;
692 struct proc *p;
693 long rss;
694 long *cp_time;
695 uint64_t runtime, new_switchtime;
696
697 td = curthread;
698 p = td->td_proc;
699
700 cp_time = (long *)PCPU_PTR(cp_time);
701 if (usermode) {
702 /*
703 * Charge the time as appropriate.
704 */
705 td->td_uticks += cnt;
706 if (p->p_nice > NZERO)
707 cp_time[CP_NICE] += cnt;
708 else
709 cp_time[CP_USER] += cnt;
710 } else {
711 /*
712 * Came from kernel mode, so we were:
713 * - handling an interrupt,
714 * - doing syscall or trap work on behalf of the current
715 * user process, or
716 * - spinning in the idle loop.
717 * Whichever it is, charge the time as appropriate.
718 * Note that we charge interrupts to the current process,
719 * regardless of whether they are ``for'' that process,
720 * so that we know how much of its real time was spent
721 * in ``non-process'' (i.e., interrupt) work.
722 */
723 if ((td->td_pflags & TDP_ITHREAD) ||
724 td->td_intr_nesting_level >= 2) {
725 td->td_iticks += cnt;
726 cp_time[CP_INTR] += cnt;
727 } else {
728 td->td_pticks += cnt;
729 td->td_sticks += cnt;
730 if (!TD_IS_IDLETHREAD(td))
731 cp_time[CP_SYS] += cnt;
732 else
733 cp_time[CP_IDLE] += cnt;
734 }
735 }
736
737 /* Update resource usage integrals and maximums. */
738 MPASS(p->p_vmspace != NULL);
739 vm = p->p_vmspace;
740 ru = &td->td_ru;
741 ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
742 ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
743 ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
744 rss = pgtok(vmspace_resident_count(vm));
745 if (ru->ru_maxrss < rss)
746 ru->ru_maxrss = rss;
747 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
748 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
749 SDT_PROBE2(sched, , , tick, td, td->td_proc);
750 thread_lock_flags(td, MTX_QUIET);
751
752 /*
753 * Compute the amount of time during which the current
754 * thread was running, and add that to its total so far.
755 */
756 new_switchtime = cpu_ticks();
757 runtime = new_switchtime - PCPU_GET(switchtime);
758 td->td_runtime += runtime;
759 td->td_incruntime += runtime;
760 PCPU_SET(switchtime, new_switchtime);
761
762 sched_clock(td, cnt);
763 thread_unlock(td);
764 #ifdef HWPMC_HOOKS
765 if (td->td_intr_frame != NULL)
766 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
767 #endif
768 }
769
770 void
profclock(int cnt,int usermode,uintfptr_t pc)771 profclock(int cnt, int usermode, uintfptr_t pc)
772 {
773 struct thread *td;
774
775 td = curthread;
776 if (usermode) {
777 /*
778 * Came from user mode; CPU was in user state.
779 * If this process is being profiled, record the tick.
780 * if there is no related user location yet, don't
781 * bother trying to count it.
782 */
783 if (td->td_proc->p_flag & P_PROFIL)
784 addupc_intr(td, pc, cnt);
785 }
786 #ifdef HWPMC_HOOKS
787 if (td->td_intr_frame != NULL)
788 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
789 #endif
790 }
791
792 /*
793 * Return information about system clocks.
794 */
795 static int
sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)796 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
797 {
798 struct clockinfo clkinfo;
799 /*
800 * Construct clockinfo structure.
801 */
802 bzero(&clkinfo, sizeof(clkinfo));
803 clkinfo.hz = hz;
804 clkinfo.tick = tick;
805 clkinfo.profhz = profhz;
806 clkinfo.stathz = stathz ? stathz : hz;
807 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
808 }
809
810 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
811 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
812 0, 0, sysctl_kern_clockrate, "S,clockinfo",
813 "Rate and period of various kernel clocks");
814
815 static void
watchdog_config(void * unused __unused,u_int cmd,int * error)816 watchdog_config(void *unused __unused, u_int cmd, int *error)
817 {
818 u_int u;
819
820 u = cmd & WD_INTERVAL;
821 if (u >= WD_TO_1SEC) {
822 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
823 watchdog_enabled = 1;
824 *error = 0;
825 } else {
826 watchdog_enabled = 0;
827 }
828 }
829
830 /*
831 * Handle a watchdog timeout by dropping to DDB or panicking.
832 */
833 static void
watchdog_fire(void)834 watchdog_fire(void)
835 {
836
837 #if defined(KDB) && !defined(KDB_UNATTENDED)
838 kdb_backtrace();
839 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
840 #else
841 panic("watchdog timeout");
842 #endif
843 }
844