xref: /openbsd/sys/kern/kern_clock.c (revision 241d6723)
1 /*	$OpenBSD: kern_clock.c,v 1.124 2024/07/08 13:17:11 claudio Exp $	*/
2 /*	$NetBSD: kern_clock.c,v 1.34 1996/06/09 04:51:03 briggs Exp $	*/
3 
4 /*-
5  * Copyright (c) 1982, 1986, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)kern_clock.c	8.5 (Berkeley) 1/21/94
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/clockintr.h>
43 #include <sys/timeout.h>
44 #include <sys/kernel.h>
45 #include <sys/limits.h>
46 #include <sys/proc.h>
47 #include <sys/user.h>
48 #include <sys/resourcevar.h>
49 #include <sys/sysctl.h>
50 #include <sys/sched.h>
51 #include <sys/timetc.h>
52 
53 /*
54  * Clock handling routines.
55  *
56  * This code is written to operate with two timers that run independently of
57  * each other.  The main clock, running hz times per second, is used to keep
58  * track of real time.  The second timer handles kernel and user profiling,
59  * and does resource use estimation.  If the second timer is programmable,
60  * it is randomized to avoid aliasing between the two clocks.  For example,
61  * the randomization prevents an adversary from always giving up the cpu
62  * just before its quantum expires.  Otherwise, it would never accumulate
63  * cpu ticks.  The mean frequency of the second timer is stathz.
64  *
65  * If no second timer exists, stathz will be zero; in this case we drive
66  * profiling and statistics off the main clock.  This WILL NOT be accurate;
67  * do not do it unless absolutely necessary.
68  *
69  * The statistics clock may (or may not) be run at a higher rate while
70  * profiling.  This profile clock runs at profhz.  We require that profhz
71  * be an integral multiple of stathz.
72  *
73  * If the statistics clock is running fast, it must be divided by the ratio
74  * profhz/stathz for statistics.  (For profiling, every tick counts.)
75  */
76 
77 int	stathz;
78 int	profhz;
79 int	profprocs;
80 int	ticks = INT_MAX - (15 * 60 * HZ);
81 
82 /* Don't force early wrap around, triggers bug in inteldrm */
83 volatile unsigned long jiffies;
84 
85 uint64_t hardclock_period;	/* [I] hardclock period (ns) */
86 uint64_t statclock_avg;		/* [I] average statclock period (ns) */
87 uint64_t statclock_min;		/* [I] minimum statclock period (ns) */
88 uint32_t statclock_mask;	/* [I] set of allowed offsets */
89 int statclock_is_randomized;	/* [I] fixed or pseudorandom period? */
90 
91 /*
92  * Initialize clock frequencies and start both clocks running.
93  */
94 void
initclocks(void)95 initclocks(void)
96 {
97 	uint64_t half_avg;
98 	uint32_t var;
99 
100 	/*
101 	 * Let the machine-specific code do its bit.
102 	 */
103 	cpu_initclocks();
104 
105 	KASSERT(hz > 0 && hz <= 1000000000);
106 	hardclock_period = 1000000000 / hz;
107 	roundrobin_period = hardclock_period * 10;
108 
109 	KASSERT(stathz >= 1 && stathz <= 1000000000);
110 
111 	/*
112 	 * Compute the average statclock() period.  Then find var, the
113 	 * largest 32-bit power of two such that var <= statclock_avg / 2.
114 	 */
115 	statclock_avg = 1000000000 / stathz;
116 	half_avg = statclock_avg / 2;
117 	for (var = 1U << 31; var > half_avg; var /= 2)
118 		continue;
119 
120 	/*
121 	 * Set a lower bound for the range using statclock_avg and var.
122 	 * The mask for that range is just (var - 1).
123 	 */
124 	statclock_min = statclock_avg - (var / 2);
125 	statclock_mask = var - 1;
126 
127 	KASSERT(profhz >= stathz && profhz <= 1000000000);
128 	KASSERT(profhz % stathz == 0);
129 	profclock_period = 1000000000 / profhz;
130 
131 	inittimecounter();
132 
133 	/* Start dispatching clock interrupts on the primary CPU. */
134 	cpu_startclock();
135 }
136 
137 /*
138  * The real-time timer, interrupting hz times per second.
139  */
140 void
hardclock(struct clockframe * frame)141 hardclock(struct clockframe *frame)
142 {
143 	tc_ticktock();
144 	ticks++;
145 	jiffies++;
146 
147 	/*
148 	 * Update the timeout wheel.
149 	 */
150 	timeout_hardclock_update();
151 }
152 
153 /*
154  * Compute number of hz in the specified amount of time.
155  */
156 int
tvtohz(const struct timeval * tv)157 tvtohz(const struct timeval *tv)
158 {
159 	unsigned long nticks;
160 	time_t sec;
161 	long usec;
162 
163 	/*
164 	 * If the number of usecs in the whole seconds part of the time
165 	 * fits in a long, then the total number of usecs will
166 	 * fit in an unsigned long.  Compute the total and convert it to
167 	 * ticks, rounding up and adding 1 to allow for the current tick
168 	 * to expire.  Rounding also depends on unsigned long arithmetic
169 	 * to avoid overflow.
170 	 *
171 	 * Otherwise, if the number of ticks in the whole seconds part of
172 	 * the time fits in a long, then convert the parts to
173 	 * ticks separately and add, using similar rounding methods and
174 	 * overflow avoidance.  This method would work in the previous
175 	 * case but it is slightly slower and assumes that hz is integral.
176 	 *
177 	 * Otherwise, round the time down to the maximum
178 	 * representable value.
179 	 *
180 	 * If ints have 32 bits, then the maximum value for any timeout in
181 	 * 10ms ticks is 248 days.
182 	 */
183 	sec = tv->tv_sec;
184 	usec = tv->tv_usec;
185 	if (sec < 0 || (sec == 0 && usec <= 0))
186 		nticks = 0;
187 	else if (sec <= LONG_MAX / 1000000)
188 		nticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
189 		    / tick + 1;
190 	else if (sec <= LONG_MAX / hz)
191 		nticks = sec * hz
192 		    + ((unsigned long)usec + (tick - 1)) / tick + 1;
193 	else
194 		nticks = LONG_MAX;
195 	if (nticks > INT_MAX)
196 		nticks = INT_MAX;
197 	return ((int)nticks);
198 }
199 
200 int
tstohz(const struct timespec * ts)201 tstohz(const struct timespec *ts)
202 {
203 	struct timeval tv;
204 	TIMESPEC_TO_TIMEVAL(&tv, ts);
205 
206 	/* Round up. */
207 	if ((ts->tv_nsec % 1000) != 0) {
208 		tv.tv_usec += 1;
209 		if (tv.tv_usec >= 1000000) {
210 			tv.tv_usec -= 1000000;
211 			tv.tv_sec += 1;
212 		}
213 	}
214 
215 	return (tvtohz(&tv));
216 }
217 
218 /*
219  * Start profiling on a process.
220  *
221  * Kernel profiling passes proc0 which never exits and hence
222  * keeps the profile clock running constantly.
223  */
224 void
startprofclock(struct process * pr)225 startprofclock(struct process *pr)
226 {
227 	int s;
228 
229 	if ((pr->ps_flags & PS_PROFIL) == 0) {
230 		atomic_setbits_int(&pr->ps_flags, PS_PROFIL);
231 		if (++profprocs == 1) {
232 			s = splstatclock();
233 			setstatclockrate(profhz);
234 			splx(s);
235 		}
236 	}
237 }
238 
239 /*
240  * Stop profiling on a process.
241  */
242 void
stopprofclock(struct process * pr)243 stopprofclock(struct process *pr)
244 {
245 	int s;
246 
247 	if (pr->ps_flags & PS_PROFIL) {
248 		atomic_clearbits_int(&pr->ps_flags, PS_PROFIL);
249 		if (--profprocs == 0) {
250 			s = splstatclock();
251 			setstatclockrate(stathz);
252 			splx(s);
253 		}
254 	}
255 }
256 
257 /*
258  * Statistics clock.  Grab profile sample, and if divider reaches 0,
259  * do process and kernel statistics.
260  */
261 void
statclock(struct clockrequest * cr,void * cf,void * arg)262 statclock(struct clockrequest *cr, void *cf, void *arg)
263 {
264 	uint64_t count, i;
265 	struct clockframe *frame = cf;
266 	struct cpu_info *ci = curcpu();
267 	struct schedstate_percpu *spc = &ci->ci_schedstate;
268 	struct proc *p = curproc;
269 	struct process *pr;
270 
271 	if (statclock_is_randomized) {
272 		count = clockrequest_advance_random(cr, statclock_min,
273 		    statclock_mask);
274 	} else {
275 		count = clockrequest_advance(cr, statclock_avg);
276 	}
277 
278 	if (CLKF_USERMODE(frame)) {
279 		pr = p->p_p;
280 		/*
281 		 * Came from user mode; CPU was in user state.
282 		 * If this process is being profiled record the tick.
283 		 */
284 		tu_enter(&p->p_tu);
285 		p->p_tu.tu_uticks += count;
286 		tu_leave(&p->p_tu);
287 		if (pr->ps_nice > NZERO)
288 			spc->spc_cp_time[CP_NICE] += count;
289 		else
290 			spc->spc_cp_time[CP_USER] += count;
291 	} else {
292 		/*
293 		 * Came from kernel mode, so we were:
294 		 * - spinning on a lock
295 		 * - handling an interrupt,
296 		 * - doing syscall or trap work on behalf of the current
297 		 *   user process, or
298 		 * - spinning in the idle loop.
299 		 * Whichever it is, charge the time as appropriate.
300 		 * Note that we charge interrupts to the current process,
301 		 * regardless of whether they are ``for'' that process,
302 		 * so that we know how much of its real time was spent
303 		 * in ``non-process'' (i.e., interrupt) work.
304 		 */
305 		if (CLKF_INTR(frame)) {
306 			if (p != NULL) {
307 				tu_enter(&p->p_tu);
308 				p->p_tu.tu_iticks += count;
309 				tu_leave(&p->p_tu);
310 			}
311 			spc->spc_cp_time[spc->spc_spinning ?
312 			    CP_SPIN : CP_INTR] += count;
313 		} else if (p != NULL && p != spc->spc_idleproc) {
314 			tu_enter(&p->p_tu);
315 			p->p_tu.tu_sticks += count;
316 			tu_leave(&p->p_tu);
317 			spc->spc_cp_time[spc->spc_spinning ?
318 			    CP_SPIN : CP_SYS] += count;
319 		} else
320 			spc->spc_cp_time[spc->spc_spinning ?
321 			    CP_SPIN : CP_IDLE] += count;
322 	}
323 
324 	if (p != NULL) {
325 		p->p_cpticks += count;
326 		/*
327 		 * schedclock() runs every fourth statclock().
328 		 */
329 		for (i = 0; i < count; i++) {
330 			if ((++spc->spc_schedticks & 3) == 0)
331 				schedclock(p);
332 		}
333 	}
334 }
335 
336 /*
337  * Return information about system clocks.
338  */
339 int
sysctl_clockrate(char * where,size_t * sizep,void * newp)340 sysctl_clockrate(char *where, size_t *sizep, void *newp)
341 {
342 	struct clockinfo clkinfo;
343 
344 	/*
345 	 * Construct clockinfo structure.
346 	 */
347 	memset(&clkinfo, 0, sizeof clkinfo);
348 	clkinfo.tick = tick;
349 	clkinfo.hz = hz;
350 	clkinfo.profhz = profhz;
351 	clkinfo.stathz = stathz;
352 	return (sysctl_rdstruct(where, sizep, newp, &clkinfo, sizeof(clkinfo)));
353 }
354