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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
34 * @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
41 #include "opt_device_polling.h"
42 #include "opt_hwpmc_hooks.h"
44 #include "opt_watchdog.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/callout.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
54 #include <sys/mutex.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/signalvar.h>
60 #include <sys/sleepqueue.h>
64 #include <vm/vm_map.h>
65 #include <sys/sysctl.h>
67 #include <sys/interrupt.h>
68 #include <sys/limits.h>
69 #include <sys/timetc.h>
76 #include <sys/pmckern.h>
80 extern void hardclock_device_poll(void);
81 #endif /* DEVICE_POLLING */
83 static void initclocks(void *dummy);
84 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
86 /* Spin-lock protecting profiling statistics. */
87 static struct mtx time_lock;
90 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
93 long cp_time[CPUSTATES];
96 unsigned int cp_time32[CPUSTATES];
99 read_cpu_time(cp_time);
101 if (req->flags & SCTL_MASK32) {
103 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
104 for (i = 0; i < CPUSTATES; i++)
105 cp_time32[i] = (unsigned int)cp_time[i];
106 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
111 return SYSCTL_OUT(req, 0, sizeof(cp_time));
112 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
117 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
118 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
120 static long empty[CPUSTATES];
123 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
130 unsigned int cp_time32[CPUSTATES];
136 if (req->flags & SCTL_MASK32)
137 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
140 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
142 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
143 if (!CPU_ABSENT(c)) {
145 cp_time = pcpu->pc_cp_time;
150 if (req->flags & SCTL_MASK32) {
151 for (i = 0; i < CPUSTATES; i++)
152 cp_time32[i] = (unsigned int)cp_time[i];
153 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
156 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
161 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
162 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
165 static const char *blessed[] = {
170 static int slptime_threshold = 1800;
171 static int blktime_threshold = 900;
172 static int sleepfreq = 3;
180 int blkticks, i, slpticks, slptype, tryl, tticks;
184 blkticks = blktime_threshold * hz;
185 slpticks = slptime_threshold * hz;
188 * Avoid to sleep on the sx_lock in order to avoid a possible
189 * priority inversion problem leading to starvation.
190 * If the lock can't be held after 100 tries, panic.
192 if (!sx_try_slock(&allproc_lock)) {
194 panic("%s: possible deadlock detected on allproc_lock\n",
197 pause("allproc_lock deadlkres", sleepfreq * hz);
201 FOREACH_PROC_IN_SYSTEM(p) {
203 FOREACH_THREAD_IN_PROC(p, td) {
205 if (TD_ON_LOCK(td)) {
208 * The thread should be blocked on a
209 * turnstile, simply check if the
210 * turnstile channel is in good state.
212 MPASS(td->td_blocked != NULL);
214 /* Handle ticks wrap-up. */
215 if (ticks < td->td_blktick)
217 tticks = ticks - td->td_blktick;
219 if (tticks > blkticks) {
222 * Accordingly with provided
223 * thresholds, this thread is
224 * stuck for too long on a
228 sx_sunlock(&allproc_lock);
229 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
230 __func__, td, tticks);
232 } else if (TD_IS_SLEEPING(td)) {
234 /* Handle ticks wrap-up. */
235 if (ticks < td->td_blktick)
239 * Check if the thread is sleeping on a
240 * lock, otherwise skip the check.
241 * Drop the thread lock in order to
242 * avoid a LOR with the sleepqueue
245 wchan = td->td_wchan;
246 tticks = ticks - td->td_slptick;
248 slptype = sleepq_type(wchan);
249 if ((slptype == SLEEPQ_SX ||
250 slptype == SLEEPQ_LK) &&
254 * Accordingly with provided
255 * thresholds, this thread is
256 * stuck for too long on a
258 * However, being on a
259 * sleepqueue, we might still
260 * check for the blessed
264 for (i = 0; blessed[i] != NULL;
266 if (!strcmp(blessed[i],
277 sx_sunlock(&allproc_lock);
278 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
279 __func__, td, tticks);
286 sx_sunlock(&allproc_lock);
288 /* Sleep for sleepfreq seconds. */
289 pause("deadlkres", sleepfreq * hz);
293 static struct kthread_desc deadlkres_kd = {
296 (struct thread **)NULL
299 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
301 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
302 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
303 &slptime_threshold, 0,
304 "Number of seconds within is valid to sleep on a sleepqueue");
305 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
306 &blktime_threshold, 0,
307 "Number of seconds within is valid to block on a turnstile");
308 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
309 "Number of seconds between any deadlock resolver thread run");
310 #endif /* DEADLKRES */
313 read_cpu_time(long *cp_time)
318 /* Sum up global cp_time[]. */
319 bzero(cp_time, sizeof(long) * CPUSTATES);
320 for (i = 0; i <= mp_maxid; i++) {
324 for (j = 0; j < CPUSTATES; j++)
325 cp_time[j] += pc->pc_cp_time[j];
330 #include <sys/watchdog.h>
332 static int watchdog_ticks;
333 static int watchdog_enabled;
334 static void watchdog_fire(void);
335 static void watchdog_config(void *, u_int, int *);
336 #endif /* SW_WATCHDOG */
339 * Clock handling routines.
341 * This code is written to operate with two timers that run independently of
344 * The main timer, running hz times per second, is used to trigger interval
345 * timers, timeouts and rescheduling as needed.
347 * The second timer handles kernel and user profiling,
348 * and does resource use estimation. If the second timer is programmable,
349 * it is randomized to avoid aliasing between the two clocks. For example,
350 * the randomization prevents an adversary from always giving up the cpu
351 * just before its quantum expires. Otherwise, it would never accumulate
352 * cpu ticks. The mean frequency of the second timer is stathz.
354 * If no second timer exists, stathz will be zero; in this case we drive
355 * profiling and statistics off the main clock. This WILL NOT be accurate;
356 * do not do it unless absolutely necessary.
358 * The statistics clock may (or may not) be run at a higher rate while
359 * profiling. This profile clock runs at profhz. We require that profhz
360 * be an integral multiple of stathz.
362 * If the statistics clock is running fast, it must be divided by the ratio
363 * profhz/stathz for statistics. (For profiling, every tick counts.)
365 * Time-of-day is maintained using a "timecounter", which may or may
366 * not be related to the hardware generating the above mentioned
377 * Initialize clock frequencies and start both clocks running.
387 * Set divisors to 1 (normal case) and let the machine-specific
390 mtx_init(&time_lock, "time lock", NULL, MTX_SPIN);
394 * Compute profhz/stathz, and fix profhz if needed.
396 i = stathz ? stathz : hz;
399 psratio = profhz / i;
401 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
406 * Each time the real-time timer fires, this function is called on all CPUs.
407 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
408 * the other CPUs in the system need to call this function.
411 hardclock_cpu(int usermode)
413 struct pstats *pstats;
414 struct thread *td = curthread;
415 struct proc *p = td->td_proc;
419 * Run current process's virtual and profile time, as needed.
424 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
426 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
427 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
430 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
432 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
433 flags |= TDF_PROFPEND | TDF_ASTPENDING;
438 td->td_flags |= flags;
442 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
443 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
449 * The real-time timer, interrupting hz times per second.
452 hardclock(int usermode, uintfptr_t pc)
455 atomic_add_int((volatile int *)&ticks, 1);
456 hardclock_cpu(usermode);
459 * If no separate statistics clock is available, run it from here.
461 * XXX: this only works for UP
464 profclock(usermode, pc);
467 #ifdef DEVICE_POLLING
468 hardclock_device_poll(); /* this is very short and quick */
469 #endif /* DEVICE_POLLING */
471 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
473 #endif /* SW_WATCHDOG */
477 * Compute number of ticks in the specified amount of time.
483 register unsigned long ticks;
484 register long sec, usec;
487 * If the number of usecs in the whole seconds part of the time
488 * difference fits in a long, then the total number of usecs will
489 * fit in an unsigned long. Compute the total and convert it to
490 * ticks, rounding up and adding 1 to allow for the current tick
491 * to expire. Rounding also depends on unsigned long arithmetic
494 * Otherwise, if the number of ticks in the whole seconds part of
495 * the time difference fits in a long, then convert the parts to
496 * ticks separately and add, using similar rounding methods and
497 * overflow avoidance. This method would work in the previous
498 * case but it is slightly slower and assumes that hz is integral.
500 * Otherwise, round the time difference down to the maximum
501 * representable value.
503 * If ints have 32 bits, then the maximum value for any timeout in
504 * 10ms ticks is 248 days.
518 printf("tvotohz: negative time difference %ld sec %ld usec\n",
522 } else if (sec <= LONG_MAX / 1000000)
523 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
525 else if (sec <= LONG_MAX / hz)
527 + ((unsigned long)usec + (tick - 1)) / tick + 1;
536 * Start profiling on a process.
538 * Kernel profiling passes proc0 which never exits and hence
539 * keeps the profile clock running constantly.
543 register struct proc *p;
546 PROC_LOCK_ASSERT(p, MA_OWNED);
547 if (p->p_flag & P_STOPPROF)
549 if ((p->p_flag & P_PROFIL) == 0) {
550 p->p_flag |= P_PROFIL;
551 mtx_lock_spin(&time_lock);
552 if (++profprocs == 1)
553 cpu_startprofclock();
554 mtx_unlock_spin(&time_lock);
559 * Stop profiling on a process.
563 register struct proc *p;
566 PROC_LOCK_ASSERT(p, MA_OWNED);
567 if (p->p_flag & P_PROFIL) {
568 if (p->p_profthreads != 0) {
569 p->p_flag |= P_STOPPROF;
570 while (p->p_profthreads != 0)
571 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
573 p->p_flag &= ~P_STOPPROF;
575 if ((p->p_flag & P_PROFIL) == 0)
577 p->p_flag &= ~P_PROFIL;
578 mtx_lock_spin(&time_lock);
579 if (--profprocs == 0)
581 mtx_unlock_spin(&time_lock);
586 * Statistics clock. Updates rusage information and calls the scheduler
587 * to adjust priorities of the active thread.
589 * This should be called by all active processors.
592 statclock(int usermode)
604 cp_time = (long *)PCPU_PTR(cp_time);
607 * Charge the time as appropriate.
610 if (p->p_nice > NZERO)
616 * Came from kernel mode, so we were:
617 * - handling an interrupt,
618 * - doing syscall or trap work on behalf of the current
620 * - spinning in the idle loop.
621 * Whichever it is, charge the time as appropriate.
622 * Note that we charge interrupts to the current process,
623 * regardless of whether they are ``for'' that process,
624 * so that we know how much of its real time was spent
625 * in ``non-process'' (i.e., interrupt) work.
627 if ((td->td_pflags & TDP_ITHREAD) ||
628 td->td_intr_nesting_level >= 2) {
634 if (!TD_IS_IDLETHREAD(td))
641 /* Update resource usage integrals and maximums. */
642 MPASS(p->p_vmspace != NULL);
645 ru->ru_ixrss += pgtok(vm->vm_tsize);
646 ru->ru_idrss += pgtok(vm->vm_dsize);
647 ru->ru_isrss += pgtok(vm->vm_ssize);
648 rss = pgtok(vmspace_resident_count(vm));
649 if (ru->ru_maxrss < rss)
651 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
652 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
653 thread_lock_flags(td, MTX_QUIET);
659 profclock(int usermode, uintfptr_t pc)
670 * Came from user mode; CPU was in user state.
671 * If this process is being profiled, record the tick.
672 * if there is no related user location yet, don't
673 * bother trying to count it.
675 if (td->td_proc->p_flag & P_PROFIL)
676 addupc_intr(td, pc, 1);
681 * Kernel statistics are just like addupc_intr, only easier.
684 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
686 if (i < g->textsize) {
695 * Return information about system clocks.
698 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
700 struct clockinfo clkinfo;
702 * Construct clockinfo structure.
704 bzero(&clkinfo, sizeof(clkinfo));
707 clkinfo.profhz = profhz;
708 clkinfo.stathz = stathz ? stathz : hz;
709 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
712 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
713 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
714 0, 0, sysctl_kern_clockrate, "S,clockinfo",
715 "Rate and period of various kernel clocks");
720 watchdog_config(void *unused __unused, u_int cmd, int *error)
724 u = cmd & WD_INTERVAL;
725 if (u >= WD_TO_1SEC) {
726 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
727 watchdog_enabled = 1;
730 watchdog_enabled = 0;
735 * Handle a watchdog timeout by dumping interrupt information and
736 * then either dropping to DDB or panicking.
749 nintr = eintrcnt - intrcnt;
751 printf("interrupt total\n");
752 while (--nintr >= 0) {
754 printf("%-12s %20lu\n", curname, *curintr);
755 curname += strlen(curname) + 1;
756 inttotal += *curintr++;
758 printf("Total %20ju\n", (uintmax_t)inttotal);
760 #if defined(KDB) && !defined(KDB_UNATTENDED)
762 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
764 panic("watchdog timeout");
768 #endif /* SW_WATCHDOG */