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>
53 #include <sys/mutex.h>
55 #include <sys/resource.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sched.h>
58 #include <sys/signalvar.h>
62 #include <vm/vm_map.h>
63 #include <sys/sysctl.h>
65 #include <sys/interrupt.h>
66 #include <sys/limits.h>
67 #include <sys/timetc.h>
74 #include <sys/pmckern.h>
78 extern void hardclock_device_poll(void);
79 #endif /* DEVICE_POLLING */
81 static void initclocks(void *dummy);
82 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL)
84 /* Some of these don't belong here, but it's easiest to concentrate them. */
85 long cp_time[CPUSTATES];
87 /* Spin-lock protecting profiling statistics. */
88 static struct mtx time_lock;
91 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
96 unsigned int cp_time32[CPUSTATES];
98 if (req->flags & SCTL_MASK32) {
100 return SYSCTL_OUT(req, 0, sizeof(cp_time32));
101 for (i = 0; i < CPUSTATES; i++)
102 cp_time32[i] = (unsigned int)cp_time[i];
103 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
108 return SYSCTL_OUT(req, 0, sizeof(cp_time));
109 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
114 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD,
115 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
118 #include <sys/watchdog.h>
120 static int watchdog_ticks;
121 static int watchdog_enabled;
122 static void watchdog_fire(void);
123 static void watchdog_config(void *, u_int, int *);
124 #endif /* SW_WATCHDOG */
127 * Clock handling routines.
129 * This code is written to operate with two timers that run independently of
132 * The main timer, running hz times per second, is used to trigger interval
133 * timers, timeouts and rescheduling as needed.
135 * The second timer handles kernel and user profiling,
136 * and does resource use estimation. If the second timer is programmable,
137 * it is randomized to avoid aliasing between the two clocks. For example,
138 * the randomization prevents an adversary from always giving up the cpu
139 * just before its quantum expires. Otherwise, it would never accumulate
140 * cpu ticks. The mean frequency of the second timer is stathz.
142 * If no second timer exists, stathz will be zero; in this case we drive
143 * profiling and statistics off the main clock. This WILL NOT be accurate;
144 * do not do it unless absolutely necessary.
146 * The statistics clock may (or may not) be run at a higher rate while
147 * profiling. This profile clock runs at profhz. We require that profhz
148 * be an integral multiple of stathz.
150 * If the statistics clock is running fast, it must be divided by the ratio
151 * profhz/stathz for statistics. (For profiling, every tick counts.)
153 * Time-of-day is maintained using a "timecounter", which may or may
154 * not be related to the hardware generating the above mentioned
165 * Initialize clock frequencies and start both clocks running.
175 * Set divisors to 1 (normal case) and let the machine-specific
178 mtx_init(&time_lock, "time lock", NULL, MTX_SPIN);
182 * Compute profhz/stathz, and fix profhz if needed.
184 i = stathz ? stathz : hz;
187 psratio = profhz / i;
189 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
194 * Each time the real-time timer fires, this function is called on all CPUs.
195 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
196 * the other CPUs in the system need to call this function.
199 hardclock_cpu(int usermode)
201 struct pstats *pstats;
202 struct thread *td = curthread;
203 struct proc *p = td->td_proc;
207 * Run current process's virtual and profile time, as needed.
212 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
214 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
215 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
218 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
220 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
221 flags |= TDF_PROFPEND | TDF_ASTPENDING;
226 td->td_flags |= flags;
230 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
231 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
236 * The real-time timer, interrupting hz times per second.
239 hardclock(int usermode, uintfptr_t pc)
241 int need_softclock = 0;
243 hardclock_cpu(usermode);
247 * If no separate statistics clock is available, run it from here.
249 * XXX: this only works for UP
252 profclock(usermode, pc);
256 #ifdef DEVICE_POLLING
257 hardclock_device_poll(); /* this is very short and quick */
258 #endif /* DEVICE_POLLING */
261 * Process callouts at a very low cpu priority, so we don't keep the
262 * relatively high clock interrupt priority any longer than necessary.
264 mtx_lock_spin_flags(&callout_lock, MTX_QUIET);
266 if (!TAILQ_EMPTY(&callwheel[ticks & callwheelmask])) {
268 } else if (softticks + 1 == ticks)
270 mtx_unlock_spin_flags(&callout_lock, MTX_QUIET);
273 * swi_sched acquires the thread lock, so we don't want to call it
274 * with callout_lock held; incorrect locking order.
277 swi_sched(softclock_ih, 0);
280 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
282 #endif /* SW_WATCHDOG */
286 * Compute number of ticks in the specified amount of time.
292 register unsigned long ticks;
293 register long sec, usec;
296 * If the number of usecs in the whole seconds part of the time
297 * difference fits in a long, then the total number of usecs will
298 * fit in an unsigned long. Compute the total and convert it to
299 * ticks, rounding up and adding 1 to allow for the current tick
300 * to expire. Rounding also depends on unsigned long arithmetic
303 * Otherwise, if the number of ticks in the whole seconds part of
304 * the time difference fits in a long, then convert the parts to
305 * ticks separately and add, using similar rounding methods and
306 * overflow avoidance. This method would work in the previous
307 * case but it is slightly slower and assumes that hz is integral.
309 * Otherwise, round the time difference down to the maximum
310 * representable value.
312 * If ints have 32 bits, then the maximum value for any timeout in
313 * 10ms ticks is 248 days.
327 printf("tvotohz: negative time difference %ld sec %ld usec\n",
331 } else if (sec <= LONG_MAX / 1000000)
332 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
334 else if (sec <= LONG_MAX / hz)
336 + ((unsigned long)usec + (tick - 1)) / tick + 1;
345 * Start profiling on a process.
347 * Kernel profiling passes proc0 which never exits and hence
348 * keeps the profile clock running constantly.
352 register struct proc *p;
355 PROC_LOCK_ASSERT(p, MA_OWNED);
356 if (p->p_flag & P_STOPPROF)
358 if ((p->p_flag & P_PROFIL) == 0) {
359 p->p_flag |= P_PROFIL;
360 mtx_lock_spin(&time_lock);
361 if (++profprocs == 1)
362 cpu_startprofclock();
363 mtx_unlock_spin(&time_lock);
368 * Stop profiling on a process.
372 register struct proc *p;
375 PROC_LOCK_ASSERT(p, MA_OWNED);
376 if (p->p_flag & P_PROFIL) {
377 if (p->p_profthreads != 0) {
378 p->p_flag |= P_STOPPROF;
379 while (p->p_profthreads != 0)
380 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
382 p->p_flag &= ~P_STOPPROF;
384 if ((p->p_flag & P_PROFIL) == 0)
386 p->p_flag &= ~P_PROFIL;
387 mtx_lock_spin(&time_lock);
388 if (--profprocs == 0)
390 mtx_unlock_spin(&time_lock);
395 * Statistics clock. Updates rusage information and calls the scheduler
396 * to adjust priorities of the active thread.
398 * This should be called by all active processors.
401 statclock(int usermode)
412 thread_lock_flags(td, MTX_QUIET);
415 * Charge the time as appropriate.
418 if (p->p_flag & P_SA)
422 if (p->p_nice > NZERO)
423 atomic_add_long(&cp_time[CP_NICE], 1);
425 atomic_add_long(&cp_time[CP_USER], 1);
428 * Came from kernel mode, so we were:
429 * - handling an interrupt,
430 * - doing syscall or trap work on behalf of the current
432 * - spinning in the idle loop.
433 * Whichever it is, charge the time as appropriate.
434 * Note that we charge interrupts to the current process,
435 * regardless of whether they are ``for'' that process,
436 * so that we know how much of its real time was spent
437 * in ``non-process'' (i.e., interrupt) work.
439 if ((td->td_pflags & TDP_ITHREAD) ||
440 td->td_intr_nesting_level >= 2) {
442 atomic_add_long(&cp_time[CP_INTR], 1);
445 if (p->p_flag & P_SA)
450 if (!TD_IS_IDLETHREAD(td))
451 atomic_add_long(&cp_time[CP_SYS], 1);
453 atomic_add_long(&cp_time[CP_IDLE], 1);
457 /* Update resource usage integrals and maximums. */
458 MPASS(p->p_vmspace != NULL);
461 ru->ru_ixrss += pgtok(vm->vm_tsize);
462 ru->ru_idrss += pgtok(vm->vm_dsize);
463 ru->ru_isrss += pgtok(vm->vm_ssize);
464 rss = pgtok(vmspace_resident_count(vm));
465 if (ru->ru_maxrss < rss)
467 CTR4(KTR_SCHED, "statclock: %p(%s) prio %d stathz %d",
468 td, td->td_proc->p_comm, td->td_priority, (stathz)?stathz:hz);
474 profclock(int usermode, uintfptr_t pc)
485 * Came from user mode; CPU was in user state.
486 * If this process is being profiled, record the tick.
487 * if there is no related user location yet, don't
488 * bother trying to count it.
490 if (td->td_proc->p_flag & P_PROFIL)
491 addupc_intr(td, pc, 1);
496 * Kernel statistics are just like addupc_intr, only easier.
499 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
501 if (i < g->textsize) {
510 * Return information about system clocks.
513 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
515 struct clockinfo clkinfo;
517 * Construct clockinfo structure.
519 bzero(&clkinfo, sizeof(clkinfo));
522 clkinfo.profhz = profhz;
523 clkinfo.stathz = stathz ? stathz : hz;
524 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
527 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
528 0, 0, sysctl_kern_clockrate, "S,clockinfo",
529 "Rate and period of various kernel clocks");
534 watchdog_config(void *unused __unused, u_int cmd, int *error)
538 u = cmd & WD_INTERVAL;
539 if (u >= WD_TO_1SEC) {
540 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
541 watchdog_enabled = 1;
544 watchdog_enabled = 0;
549 * Handle a watchdog timeout by dumping interrupt information and
550 * then either dropping to DDB or panicking.
563 nintr = eintrcnt - intrcnt;
565 printf("interrupt total\n");
566 while (--nintr >= 0) {
568 printf("%-12s %20lu\n", curname, *curintr);
569 curname += strlen(curname) + 1;
570 inttotal += *curintr++;
572 printf("Total %20ju\n", (uintmax_t)inttotal);
574 #if defined(KDB) && !defined(KDB_UNATTENDED)
576 kdb_enter("watchdog timeout");
578 panic("watchdog timeout");
582 #endif /* SW_WATCHDOG */