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>
| 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: head/sys/kern/kern_clock.c 215317 2010-11-14 20:38:11Z dim $");
| 38__FBSDID("$FreeBSD: head/sys/kern/kern_clock.c 215701 2010-11-22 19:32:54Z dim $");
|
39
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/kdb.h>
50#include <sys/kernel.h>
51#include <sys/kthread.h>
52#include <sys/ktr.h>
53#include <sys/lock.h>
54#include <sys/mutex.h>
55#include <sys/proc.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>
61#include <sys/smp.h>
62#include <vm/vm.h>
63#include <vm/pmap.h>
64#include <vm/vm_map.h>
65#include <sys/sysctl.h>
66#include <sys/bus.h>
67#include <sys/interrupt.h>
68#include <sys/limits.h>
69#include <sys/timetc.h>
70
71#ifdef GPROF
72#include <sys/gmon.h>
73#endif
74
75#ifdef HWPMC_HOOKS
76#include <sys/pmckern.h>
77#endif
78
79#ifdef DEVICE_POLLING
80extern void hardclock_device_poll(void);
81#endif /* DEVICE_POLLING */
82
83static void initclocks(void *dummy);
84SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
85
86/* Spin-lock protecting profiling statistics. */
87static struct mtx time_lock;
88
89static int
90sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
91{
92 int error;
93 long cp_time[CPUSTATES];
94#ifdef SCTL_MASK32
95 int i;
96 unsigned int cp_time32[CPUSTATES];
97#endif
98
99 read_cpu_time(cp_time);
100#ifdef SCTL_MASK32
101 if (req->flags & SCTL_MASK32) {
102 if (!req->oldptr)
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));
107 } else
108#endif
109 {
110 if (!req->oldptr)
111 return SYSCTL_OUT(req, 0, sizeof(cp_time));
112 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
113 }
114 return error;
115}
116
117SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
118 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
119
120static long empty[CPUSTATES];
121
122static int
123sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
124{
125 struct pcpu *pcpu;
126 int error;
127 int c;
128 long *cp_time;
129#ifdef SCTL_MASK32
130 unsigned int cp_time32[CPUSTATES];
131 int i;
132#endif
133
134 if (!req->oldptr) {
135#ifdef SCTL_MASK32
136 if (req->flags & SCTL_MASK32)
137 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
138 else
139#endif
140 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
141 }
142 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
143 if (!CPU_ABSENT(c)) {
144 pcpu = pcpu_find(c);
145 cp_time = pcpu->pc_cp_time;
146 } else {
147 cp_time = empty;
148 }
149#ifdef SCTL_MASK32
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));
154 } else
155#endif
156 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
157 }
158 return error;
159}
160
161SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
162 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
163
164#ifdef DEADLKRES
165static const char *blessed[] = {
166 "getblk",
167 "so_snd_sx",
168 "so_rcv_sx",
169 NULL
170};
171static int slptime_threshold = 1800;
172static int blktime_threshold = 900;
173static int sleepfreq = 3;
174
175static void
176deadlkres(void)
177{
178 struct proc *p;
179 struct thread *td;
180 void *wchan;
181 int blkticks, i, slpticks, slptype, tryl, tticks;
182
183 tryl = 0;
184 for (;;) {
185 blkticks = blktime_threshold * hz;
186 slpticks = slptime_threshold * hz;
187
188 /*
189 * Avoid to sleep on the sx_lock in order to avoid a possible
190 * priority inversion problem leading to starvation.
191 * If the lock can't be held after 100 tries, panic.
192 */
193 if (!sx_try_slock(&allproc_lock)) {
194 if (tryl > 100)
195 panic("%s: possible deadlock detected on allproc_lock\n",
196 __func__);
197 tryl++;
198 pause("allproc", sleepfreq * hz);
199 continue;
200 }
201 tryl = 0;
202 FOREACH_PROC_IN_SYSTEM(p) {
203 PROC_LOCK(p);
204 FOREACH_THREAD_IN_PROC(p, td) {
205
206 /*
207 * Once a thread is found in "interesting"
208 * state a possible ticks wrap-up needs to be
209 * checked.
210 */
211 thread_lock(td);
212 if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
213
214 /*
215 * The thread should be blocked on a
216 * turnstile, simply check if the
217 * turnstile channel is in good state.
218 */
219 MPASS(td->td_blocked != NULL);
220
221 tticks = ticks - td->td_blktick;
222 thread_unlock(td);
223 if (tticks > blkticks) {
224
225 /*
226 * Accordingly with provided
227 * thresholds, this thread is
228 * stuck for too long on a
229 * turnstile.
230 */
231 PROC_UNLOCK(p);
232 sx_sunlock(&allproc_lock);
233 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
234 __func__, td, tticks);
235 }
236 } else if (TD_IS_SLEEPING(td) &&
237 TD_ON_SLEEPQ(td) &&
238 ticks < td->td_blktick) {
239
240 /*
241 * Check if the thread is sleeping on a
242 * lock, otherwise skip the check.
243 * Drop the thread lock in order to
244 * avoid a LOR with the sleepqueue
245 * spinlock.
246 */
247 wchan = td->td_wchan;
248 tticks = ticks - td->td_slptick;
249 thread_unlock(td);
250 slptype = sleepq_type(wchan);
251 if ((slptype == SLEEPQ_SX ||
252 slptype == SLEEPQ_LK) &&
253 tticks > slpticks) {
254
255 /*
256 * Accordingly with provided
257 * thresholds, this thread is
258 * stuck for too long on a
259 * sleepqueue.
260 * However, being on a
261 * sleepqueue, we might still
262 * check for the blessed
263 * list.
264 */
265 tryl = 0;
266 for (i = 0; blessed[i] != NULL;
267 i++) {
268 if (!strcmp(blessed[i],
269 td->td_wmesg)) {
270 tryl = 1;
271 break;
272 }
273 }
274 if (tryl != 0) {
275 tryl = 0;
276 continue;
277 }
278 PROC_UNLOCK(p);
279 sx_sunlock(&allproc_lock);
280 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
281 __func__, td, tticks);
282 }
283 } else
284 thread_unlock(td);
285 }
286 PROC_UNLOCK(p);
287 }
288 sx_sunlock(&allproc_lock);
289
290 /* Sleep for sleepfreq seconds. */
291 pause("-", sleepfreq * hz);
292 }
293}
294
295static struct kthread_desc deadlkres_kd = {
296 "deadlkres",
297 deadlkres,
298 (struct thread **)NULL
299};
300
301SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
302
303SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
304SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
305 &slptime_threshold, 0,
306 "Number of seconds within is valid to sleep on a sleepqueue");
307SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
308 &blktime_threshold, 0,
309 "Number of seconds within is valid to block on a turnstile");
310SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
311 "Number of seconds between any deadlock resolver thread run");
312#endif /* DEADLKRES */
313
314void
315read_cpu_time(long *cp_time)
316{
317 struct pcpu *pc;
318 int i, j;
319
320 /* Sum up global cp_time[]. */
321 bzero(cp_time, sizeof(long) * CPUSTATES);
322 CPU_FOREACH(i) {
323 pc = pcpu_find(i);
324 for (j = 0; j < CPUSTATES; j++)
325 cp_time[j] += pc->pc_cp_time[j];
326 }
327}
328
329#ifdef SW_WATCHDOG
330#include <sys/watchdog.h>
331
332static int watchdog_ticks;
333static int watchdog_enabled;
334static void watchdog_fire(void);
335static void watchdog_config(void *, u_int, int *);
336#endif /* SW_WATCHDOG */
337
338/*
339 * Clock handling routines.
340 *
341 * This code is written to operate with two timers that run independently of
342 * each other.
343 *
344 * The main timer, running hz times per second, is used to trigger interval
345 * timers, timeouts and rescheduling as needed.
346 *
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.
353 *
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.
357 *
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.
361 *
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.)
364 *
365 * Time-of-day is maintained using a "timecounter", which may or may
366 * not be related to the hardware generating the above mentioned
367 * interrupts.
368 */
369
370int stathz;
371int profhz;
372int profprocs;
373int ticks;
374int psratio;
375
| 39
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/kdb.h>
50#include <sys/kernel.h>
51#include <sys/kthread.h>
52#include <sys/ktr.h>
53#include <sys/lock.h>
54#include <sys/mutex.h>
55#include <sys/proc.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>
61#include <sys/smp.h>
62#include <vm/vm.h>
63#include <vm/pmap.h>
64#include <vm/vm_map.h>
65#include <sys/sysctl.h>
66#include <sys/bus.h>
67#include <sys/interrupt.h>
68#include <sys/limits.h>
69#include <sys/timetc.h>
70
71#ifdef GPROF
72#include <sys/gmon.h>
73#endif
74
75#ifdef HWPMC_HOOKS
76#include <sys/pmckern.h>
77#endif
78
79#ifdef DEVICE_POLLING
80extern void hardclock_device_poll(void);
81#endif /* DEVICE_POLLING */
82
83static void initclocks(void *dummy);
84SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
85
86/* Spin-lock protecting profiling statistics. */
87static struct mtx time_lock;
88
89static int
90sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
91{
92 int error;
93 long cp_time[CPUSTATES];
94#ifdef SCTL_MASK32
95 int i;
96 unsigned int cp_time32[CPUSTATES];
97#endif
98
99 read_cpu_time(cp_time);
100#ifdef SCTL_MASK32
101 if (req->flags & SCTL_MASK32) {
102 if (!req->oldptr)
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));
107 } else
108#endif
109 {
110 if (!req->oldptr)
111 return SYSCTL_OUT(req, 0, sizeof(cp_time));
112 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
113 }
114 return error;
115}
116
117SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
118 0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
119
120static long empty[CPUSTATES];
121
122static int
123sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
124{
125 struct pcpu *pcpu;
126 int error;
127 int c;
128 long *cp_time;
129#ifdef SCTL_MASK32
130 unsigned int cp_time32[CPUSTATES];
131 int i;
132#endif
133
134 if (!req->oldptr) {
135#ifdef SCTL_MASK32
136 if (req->flags & SCTL_MASK32)
137 return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
138 else
139#endif
140 return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
141 }
142 for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
143 if (!CPU_ABSENT(c)) {
144 pcpu = pcpu_find(c);
145 cp_time = pcpu->pc_cp_time;
146 } else {
147 cp_time = empty;
148 }
149#ifdef SCTL_MASK32
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));
154 } else
155#endif
156 error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
157 }
158 return error;
159}
160
161SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
162 0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
163
164#ifdef DEADLKRES
165static const char *blessed[] = {
166 "getblk",
167 "so_snd_sx",
168 "so_rcv_sx",
169 NULL
170};
171static int slptime_threshold = 1800;
172static int blktime_threshold = 900;
173static int sleepfreq = 3;
174
175static void
176deadlkres(void)
177{
178 struct proc *p;
179 struct thread *td;
180 void *wchan;
181 int blkticks, i, slpticks, slptype, tryl, tticks;
182
183 tryl = 0;
184 for (;;) {
185 blkticks = blktime_threshold * hz;
186 slpticks = slptime_threshold * hz;
187
188 /*
189 * Avoid to sleep on the sx_lock in order to avoid a possible
190 * priority inversion problem leading to starvation.
191 * If the lock can't be held after 100 tries, panic.
192 */
193 if (!sx_try_slock(&allproc_lock)) {
194 if (tryl > 100)
195 panic("%s: possible deadlock detected on allproc_lock\n",
196 __func__);
197 tryl++;
198 pause("allproc", sleepfreq * hz);
199 continue;
200 }
201 tryl = 0;
202 FOREACH_PROC_IN_SYSTEM(p) {
203 PROC_LOCK(p);
204 FOREACH_THREAD_IN_PROC(p, td) {
205
206 /*
207 * Once a thread is found in "interesting"
208 * state a possible ticks wrap-up needs to be
209 * checked.
210 */
211 thread_lock(td);
212 if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
213
214 /*
215 * The thread should be blocked on a
216 * turnstile, simply check if the
217 * turnstile channel is in good state.
218 */
219 MPASS(td->td_blocked != NULL);
220
221 tticks = ticks - td->td_blktick;
222 thread_unlock(td);
223 if (tticks > blkticks) {
224
225 /*
226 * Accordingly with provided
227 * thresholds, this thread is
228 * stuck for too long on a
229 * turnstile.
230 */
231 PROC_UNLOCK(p);
232 sx_sunlock(&allproc_lock);
233 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
234 __func__, td, tticks);
235 }
236 } else if (TD_IS_SLEEPING(td) &&
237 TD_ON_SLEEPQ(td) &&
238 ticks < td->td_blktick) {
239
240 /*
241 * Check if the thread is sleeping on a
242 * lock, otherwise skip the check.
243 * Drop the thread lock in order to
244 * avoid a LOR with the sleepqueue
245 * spinlock.
246 */
247 wchan = td->td_wchan;
248 tticks = ticks - td->td_slptick;
249 thread_unlock(td);
250 slptype = sleepq_type(wchan);
251 if ((slptype == SLEEPQ_SX ||
252 slptype == SLEEPQ_LK) &&
253 tticks > slpticks) {
254
255 /*
256 * Accordingly with provided
257 * thresholds, this thread is
258 * stuck for too long on a
259 * sleepqueue.
260 * However, being on a
261 * sleepqueue, we might still
262 * check for the blessed
263 * list.
264 */
265 tryl = 0;
266 for (i = 0; blessed[i] != NULL;
267 i++) {
268 if (!strcmp(blessed[i],
269 td->td_wmesg)) {
270 tryl = 1;
271 break;
272 }
273 }
274 if (tryl != 0) {
275 tryl = 0;
276 continue;
277 }
278 PROC_UNLOCK(p);
279 sx_sunlock(&allproc_lock);
280 panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
281 __func__, td, tticks);
282 }
283 } else
284 thread_unlock(td);
285 }
286 PROC_UNLOCK(p);
287 }
288 sx_sunlock(&allproc_lock);
289
290 /* Sleep for sleepfreq seconds. */
291 pause("-", sleepfreq * hz);
292 }
293}
294
295static struct kthread_desc deadlkres_kd = {
296 "deadlkres",
297 deadlkres,
298 (struct thread **)NULL
299};
300
301SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
302
303SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
304SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
305 &slptime_threshold, 0,
306 "Number of seconds within is valid to sleep on a sleepqueue");
307SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
308 &blktime_threshold, 0,
309 "Number of seconds within is valid to block on a turnstile");
310SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
311 "Number of seconds between any deadlock resolver thread run");
312#endif /* DEADLKRES */
313
314void
315read_cpu_time(long *cp_time)
316{
317 struct pcpu *pc;
318 int i, j;
319
320 /* Sum up global cp_time[]. */
321 bzero(cp_time, sizeof(long) * CPUSTATES);
322 CPU_FOREACH(i) {
323 pc = pcpu_find(i);
324 for (j = 0; j < CPUSTATES; j++)
325 cp_time[j] += pc->pc_cp_time[j];
326 }
327}
328
329#ifdef SW_WATCHDOG
330#include <sys/watchdog.h>
331
332static int watchdog_ticks;
333static int watchdog_enabled;
334static void watchdog_fire(void);
335static void watchdog_config(void *, u_int, int *);
336#endif /* SW_WATCHDOG */
337
338/*
339 * Clock handling routines.
340 *
341 * This code is written to operate with two timers that run independently of
342 * each other.
343 *
344 * The main timer, running hz times per second, is used to trigger interval
345 * timers, timeouts and rescheduling as needed.
346 *
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.
353 *
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.
357 *
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.
361 *
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.)
364 *
365 * Time-of-day is maintained using a "timecounter", which may or may
366 * not be related to the hardware generating the above mentioned
367 * interrupts.
368 */
369
370int stathz;
371int profhz;
372int profprocs;
373int ticks;
374int psratio;
375
|
376STATIC_DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */
| 376static DPCPU_DEFINE(int, pcputicks); /* Per-CPU version of ticks. */
|
377static int global_hardclock_run = 0;
378
379/*
380 * Initialize clock frequencies and start both clocks running.
381 */
382/* ARGSUSED*/
383static void
384initclocks(dummy)
385 void *dummy;
386{
387 register int i;
388
389 /*
390 * Set divisors to 1 (normal case) and let the machine-specific
391 * code do its bit.
392 */
393 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
394 cpu_initclocks();
395
396 /*
397 * Compute profhz/stathz, and fix profhz if needed.
398 */
399 i = stathz ? stathz : hz;
400 if (profhz == 0)
401 profhz = i;
402 psratio = profhz / i;
403#ifdef SW_WATCHDOG
404 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
405#endif
406}
407
408/*
409 * Each time the real-time timer fires, this function is called on all CPUs.
410 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
411 * the other CPUs in the system need to call this function.
412 */
413void
414hardclock_cpu(int usermode)
415{
416 struct pstats *pstats;
417 struct thread *td = curthread;
418 struct proc *p = td->td_proc;
419 int flags;
420
421 /*
422 * Run current process's virtual and profile time, as needed.
423 */
424 pstats = p->p_stats;
425 flags = 0;
426 if (usermode &&
427 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
428 PROC_SLOCK(p);
429 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
430 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
431 PROC_SUNLOCK(p);
432 }
433 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
434 PROC_SLOCK(p);
435 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
436 flags |= TDF_PROFPEND | TDF_ASTPENDING;
437 PROC_SUNLOCK(p);
438 }
439 thread_lock(td);
440 sched_tick(1);
441 td->td_flags |= flags;
442 thread_unlock(td);
443
444#ifdef HWPMC_HOOKS
445 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
446 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
447#endif
448 callout_tick();
449}
450
451/*
452 * The real-time timer, interrupting hz times per second.
453 */
454void
455hardclock(int usermode, uintfptr_t pc)
456{
457
458 atomic_add_int((volatile int *)&ticks, 1);
459 hardclock_cpu(usermode);
460 tc_ticktock(1);
461 cpu_tick_calibration();
462 /*
463 * If no separate statistics clock is available, run it from here.
464 *
465 * XXX: this only works for UP
466 */
467 if (stathz == 0) {
468 profclock(usermode, pc);
469 statclock(usermode);
470 }
471#ifdef DEVICE_POLLING
472 hardclock_device_poll(); /* this is very short and quick */
473#endif /* DEVICE_POLLING */
474#ifdef SW_WATCHDOG
475 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
476 watchdog_fire();
477#endif /* SW_WATCHDOG */
478}
479
480void
481hardclock_anycpu(int cnt, int usermode)
482{
483 struct pstats *pstats;
484 struct thread *td = curthread;
485 struct proc *p = td->td_proc;
486 int *t = DPCPU_PTR(pcputicks);
487 int flags, global, newticks;
488#ifdef SW_WATCHDOG
489 int i;
490#endif /* SW_WATCHDOG */
491
492 /*
493 * Update per-CPU and possibly global ticks values.
494 */
495 *t += cnt;
496 do {
497 global = ticks;
498 newticks = *t - global;
499 if (newticks <= 0) {
500 if (newticks < -1)
501 *t = global - 1;
502 newticks = 0;
503 break;
504 }
505 } while (!atomic_cmpset_int(&ticks, global, *t));
506
507 /*
508 * Run current process's virtual and profile time, as needed.
509 */
510 pstats = p->p_stats;
511 flags = 0;
512 if (usermode &&
513 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
514 PROC_SLOCK(p);
515 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
516 tick * cnt) == 0)
517 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
518 PROC_SUNLOCK(p);
519 }
520 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
521 PROC_SLOCK(p);
522 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
523 tick * cnt) == 0)
524 flags |= TDF_PROFPEND | TDF_ASTPENDING;
525 PROC_SUNLOCK(p);
526 }
527 thread_lock(td);
528 sched_tick(cnt);
529 td->td_flags |= flags;
530 thread_unlock(td);
531
532#ifdef HWPMC_HOOKS
533 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
534 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
535#endif
536 callout_tick();
537 /* We are in charge to handle this tick duty. */
538 if (newticks > 0) {
539 /* Dangerous and no need to call these things concurrently. */
540 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
541 tc_ticktock(newticks);
542#ifdef DEVICE_POLLING
543 /* This is very short and quick. */
544 hardclock_device_poll();
545#endif /* DEVICE_POLLING */
546 atomic_store_rel_int(&global_hardclock_run, 0);
547 }
548#ifdef SW_WATCHDOG
549 if (watchdog_enabled > 0) {
550 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
551 if (i > 0 && i <= newticks)
552 watchdog_fire();
553 }
554#endif /* SW_WATCHDOG */
555 }
556 if (curcpu == CPU_FIRST())
557 cpu_tick_calibration();
558}
559
560void
561hardclock_sync(int cpu)
562{
563 int *t = DPCPU_ID_PTR(cpu, pcputicks);
564
565 *t = ticks;
566}
567
568/*
569 * Compute number of ticks in the specified amount of time.
570 */
571int
572tvtohz(tv)
573 struct timeval *tv;
574{
575 register unsigned long ticks;
576 register long sec, usec;
577
578 /*
579 * If the number of usecs in the whole seconds part of the time
580 * difference fits in a long, then the total number of usecs will
581 * fit in an unsigned long. Compute the total and convert it to
582 * ticks, rounding up and adding 1 to allow for the current tick
583 * to expire. Rounding also depends on unsigned long arithmetic
584 * to avoid overflow.
585 *
586 * Otherwise, if the number of ticks in the whole seconds part of
587 * the time difference fits in a long, then convert the parts to
588 * ticks separately and add, using similar rounding methods and
589 * overflow avoidance. This method would work in the previous
590 * case but it is slightly slower and assumes that hz is integral.
591 *
592 * Otherwise, round the time difference down to the maximum
593 * representable value.
594 *
595 * If ints have 32 bits, then the maximum value for any timeout in
596 * 10ms ticks is 248 days.
597 */
598 sec = tv->tv_sec;
599 usec = tv->tv_usec;
600 if (usec < 0) {
601 sec--;
602 usec += 1000000;
603 }
604 if (sec < 0) {
605#ifdef DIAGNOSTIC
606 if (usec > 0) {
607 sec++;
608 usec -= 1000000;
609 }
610 printf("tvotohz: negative time difference %ld sec %ld usec\n",
611 sec, usec);
612#endif
613 ticks = 1;
614 } else if (sec <= LONG_MAX / 1000000)
615 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
616 / tick + 1;
617 else if (sec <= LONG_MAX / hz)
618 ticks = sec * hz
619 + ((unsigned long)usec + (tick - 1)) / tick + 1;
620 else
621 ticks = LONG_MAX;
622 if (ticks > INT_MAX)
623 ticks = INT_MAX;
624 return ((int)ticks);
625}
626
627/*
628 * Start profiling on a process.
629 *
630 * Kernel profiling passes proc0 which never exits and hence
631 * keeps the profile clock running constantly.
632 */
633void
634startprofclock(p)
635 register struct proc *p;
636{
637
638 PROC_LOCK_ASSERT(p, MA_OWNED);
639 if (p->p_flag & P_STOPPROF)
640 return;
641 if ((p->p_flag & P_PROFIL) == 0) {
642 p->p_flag |= P_PROFIL;
643 mtx_lock(&time_lock);
644 if (++profprocs == 1)
645 cpu_startprofclock();
646 mtx_unlock(&time_lock);
647 }
648}
649
650/*
651 * Stop profiling on a process.
652 */
653void
654stopprofclock(p)
655 register struct proc *p;
656{
657
658 PROC_LOCK_ASSERT(p, MA_OWNED);
659 if (p->p_flag & P_PROFIL) {
660 if (p->p_profthreads != 0) {
661 p->p_flag |= P_STOPPROF;
662 while (p->p_profthreads != 0)
663 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
664 "stopprof", 0);
665 p->p_flag &= ~P_STOPPROF;
666 }
667 if ((p->p_flag & P_PROFIL) == 0)
668 return;
669 p->p_flag &= ~P_PROFIL;
670 mtx_lock(&time_lock);
671 if (--profprocs == 0)
672 cpu_stopprofclock();
673 mtx_unlock(&time_lock);
674 }
675}
676
677/*
678 * Statistics clock. Updates rusage information and calls the scheduler
679 * to adjust priorities of the active thread.
680 *
681 * This should be called by all active processors.
682 */
683void
684statclock(int usermode)
685{
686 struct rusage *ru;
687 struct vmspace *vm;
688 struct thread *td;
689 struct proc *p;
690 long rss;
691 long *cp_time;
692
693 td = curthread;
694 p = td->td_proc;
695
696 cp_time = (long *)PCPU_PTR(cp_time);
697 if (usermode) {
698 /*
699 * Charge the time as appropriate.
700 */
701 td->td_uticks++;
702 if (p->p_nice > NZERO)
703 cp_time[CP_NICE]++;
704 else
705 cp_time[CP_USER]++;
706 } else {
707 /*
708 * Came from kernel mode, so we were:
709 * - handling an interrupt,
710 * - doing syscall or trap work on behalf of the current
711 * user process, or
712 * - spinning in the idle loop.
713 * Whichever it is, charge the time as appropriate.
714 * Note that we charge interrupts to the current process,
715 * regardless of whether they are ``for'' that process,
716 * so that we know how much of its real time was spent
717 * in ``non-process'' (i.e., interrupt) work.
718 */
719 if ((td->td_pflags & TDP_ITHREAD) ||
720 td->td_intr_nesting_level >= 2) {
721 td->td_iticks++;
722 cp_time[CP_INTR]++;
723 } else {
724 td->td_pticks++;
725 td->td_sticks++;
726 if (!TD_IS_IDLETHREAD(td))
727 cp_time[CP_SYS]++;
728 else
729 cp_time[CP_IDLE]++;
730 }
731 }
732
733 /* Update resource usage integrals and maximums. */
734 MPASS(p->p_vmspace != NULL);
735 vm = p->p_vmspace;
736 ru = &td->td_ru;
737 ru->ru_ixrss += pgtok(vm->vm_tsize);
738 ru->ru_idrss += pgtok(vm->vm_dsize);
739 ru->ru_isrss += pgtok(vm->vm_ssize);
740 rss = pgtok(vmspace_resident_count(vm));
741 if (ru->ru_maxrss < rss)
742 ru->ru_maxrss = rss;
743 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
744 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
745 thread_lock_flags(td, MTX_QUIET);
746 sched_clock(td);
747 thread_unlock(td);
748}
749
750void
751profclock(int usermode, uintfptr_t pc)
752{
753 struct thread *td;
754#ifdef GPROF
755 struct gmonparam *g;
756 uintfptr_t i;
757#endif
758
759 td = curthread;
760 if (usermode) {
761 /*
762 * Came from user mode; CPU was in user state.
763 * If this process is being profiled, record the tick.
764 * if there is no related user location yet, don't
765 * bother trying to count it.
766 */
767 if (td->td_proc->p_flag & P_PROFIL)
768 addupc_intr(td, pc, 1);
769 }
770#ifdef GPROF
771 else {
772 /*
773 * Kernel statistics are just like addupc_intr, only easier.
774 */
775 g = &_gmonparam;
776 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
777 i = PC_TO_I(g, pc);
778 if (i < g->textsize) {
779 KCOUNT(g, i)++;
780 }
781 }
782 }
783#endif
784}
785
786/*
787 * Return information about system clocks.
788 */
789static int
790sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
791{
792 struct clockinfo clkinfo;
793 /*
794 * Construct clockinfo structure.
795 */
796 bzero(&clkinfo, sizeof(clkinfo));
797 clkinfo.hz = hz;
798 clkinfo.tick = tick;
799 clkinfo.profhz = profhz;
800 clkinfo.stathz = stathz ? stathz : hz;
801 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
802}
803
804SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
805 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
806 0, 0, sysctl_kern_clockrate, "S,clockinfo",
807 "Rate and period of various kernel clocks");
808
809#ifdef SW_WATCHDOG
810
811static void
812watchdog_config(void *unused __unused, u_int cmd, int *error)
813{
814 u_int u;
815
816 u = cmd & WD_INTERVAL;
817 if (u >= WD_TO_1SEC) {
818 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
819 watchdog_enabled = 1;
820 *error = 0;
821 } else {
822 watchdog_enabled = 0;
823 }
824}
825
826/*
827 * Handle a watchdog timeout by dumping interrupt information and
828 * then either dropping to DDB or panicking.
829 */
830static void
831watchdog_fire(void)
832{
833 int nintr;
834 uint64_t inttotal;
835 u_long *curintr;
836 char *curname;
837
838 curintr = intrcnt;
839 curname = intrnames;
840 inttotal = 0;
841 nintr = eintrcnt - intrcnt;
842
843 printf("interrupt total\n");
844 while (--nintr >= 0) {
845 if (*curintr)
846 printf("%-12s %20lu\n", curname, *curintr);
847 curname += strlen(curname) + 1;
848 inttotal += *curintr++;
849 }
850 printf("Total %20ju\n", (uintmax_t)inttotal);
851
852#if defined(KDB) && !defined(KDB_UNATTENDED)
853 kdb_backtrace();
854 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
855#else
856 panic("watchdog timeout");
857#endif
858}
859
860#endif /* SW_WATCHDOG */
| 377static int global_hardclock_run = 0;
378
379/*
380 * Initialize clock frequencies and start both clocks running.
381 */
382/* ARGSUSED*/
383static void
384initclocks(dummy)
385 void *dummy;
386{
387 register int i;
388
389 /*
390 * Set divisors to 1 (normal case) and let the machine-specific
391 * code do its bit.
392 */
393 mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
394 cpu_initclocks();
395
396 /*
397 * Compute profhz/stathz, and fix profhz if needed.
398 */
399 i = stathz ? stathz : hz;
400 if (profhz == 0)
401 profhz = i;
402 psratio = profhz / i;
403#ifdef SW_WATCHDOG
404 EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
405#endif
406}
407
408/*
409 * Each time the real-time timer fires, this function is called on all CPUs.
410 * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
411 * the other CPUs in the system need to call this function.
412 */
413void
414hardclock_cpu(int usermode)
415{
416 struct pstats *pstats;
417 struct thread *td = curthread;
418 struct proc *p = td->td_proc;
419 int flags;
420
421 /*
422 * Run current process's virtual and profile time, as needed.
423 */
424 pstats = p->p_stats;
425 flags = 0;
426 if (usermode &&
427 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
428 PROC_SLOCK(p);
429 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
430 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
431 PROC_SUNLOCK(p);
432 }
433 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
434 PROC_SLOCK(p);
435 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
436 flags |= TDF_PROFPEND | TDF_ASTPENDING;
437 PROC_SUNLOCK(p);
438 }
439 thread_lock(td);
440 sched_tick(1);
441 td->td_flags |= flags;
442 thread_unlock(td);
443
444#ifdef HWPMC_HOOKS
445 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
446 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
447#endif
448 callout_tick();
449}
450
451/*
452 * The real-time timer, interrupting hz times per second.
453 */
454void
455hardclock(int usermode, uintfptr_t pc)
456{
457
458 atomic_add_int((volatile int *)&ticks, 1);
459 hardclock_cpu(usermode);
460 tc_ticktock(1);
461 cpu_tick_calibration();
462 /*
463 * If no separate statistics clock is available, run it from here.
464 *
465 * XXX: this only works for UP
466 */
467 if (stathz == 0) {
468 profclock(usermode, pc);
469 statclock(usermode);
470 }
471#ifdef DEVICE_POLLING
472 hardclock_device_poll(); /* this is very short and quick */
473#endif /* DEVICE_POLLING */
474#ifdef SW_WATCHDOG
475 if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
476 watchdog_fire();
477#endif /* SW_WATCHDOG */
478}
479
480void
481hardclock_anycpu(int cnt, int usermode)
482{
483 struct pstats *pstats;
484 struct thread *td = curthread;
485 struct proc *p = td->td_proc;
486 int *t = DPCPU_PTR(pcputicks);
487 int flags, global, newticks;
488#ifdef SW_WATCHDOG
489 int i;
490#endif /* SW_WATCHDOG */
491
492 /*
493 * Update per-CPU and possibly global ticks values.
494 */
495 *t += cnt;
496 do {
497 global = ticks;
498 newticks = *t - global;
499 if (newticks <= 0) {
500 if (newticks < -1)
501 *t = global - 1;
502 newticks = 0;
503 break;
504 }
505 } while (!atomic_cmpset_int(&ticks, global, *t));
506
507 /*
508 * Run current process's virtual and profile time, as needed.
509 */
510 pstats = p->p_stats;
511 flags = 0;
512 if (usermode &&
513 timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
514 PROC_SLOCK(p);
515 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
516 tick * cnt) == 0)
517 flags |= TDF_ALRMPEND | TDF_ASTPENDING;
518 PROC_SUNLOCK(p);
519 }
520 if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
521 PROC_SLOCK(p);
522 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
523 tick * cnt) == 0)
524 flags |= TDF_PROFPEND | TDF_ASTPENDING;
525 PROC_SUNLOCK(p);
526 }
527 thread_lock(td);
528 sched_tick(cnt);
529 td->td_flags |= flags;
530 thread_unlock(td);
531
532#ifdef HWPMC_HOOKS
533 if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
534 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
535#endif
536 callout_tick();
537 /* We are in charge to handle this tick duty. */
538 if (newticks > 0) {
539 /* Dangerous and no need to call these things concurrently. */
540 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
541 tc_ticktock(newticks);
542#ifdef DEVICE_POLLING
543 /* This is very short and quick. */
544 hardclock_device_poll();
545#endif /* DEVICE_POLLING */
546 atomic_store_rel_int(&global_hardclock_run, 0);
547 }
548#ifdef SW_WATCHDOG
549 if (watchdog_enabled > 0) {
550 i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
551 if (i > 0 && i <= newticks)
552 watchdog_fire();
553 }
554#endif /* SW_WATCHDOG */
555 }
556 if (curcpu == CPU_FIRST())
557 cpu_tick_calibration();
558}
559
560void
561hardclock_sync(int cpu)
562{
563 int *t = DPCPU_ID_PTR(cpu, pcputicks);
564
565 *t = ticks;
566}
567
568/*
569 * Compute number of ticks in the specified amount of time.
570 */
571int
572tvtohz(tv)
573 struct timeval *tv;
574{
575 register unsigned long ticks;
576 register long sec, usec;
577
578 /*
579 * If the number of usecs in the whole seconds part of the time
580 * difference fits in a long, then the total number of usecs will
581 * fit in an unsigned long. Compute the total and convert it to
582 * ticks, rounding up and adding 1 to allow for the current tick
583 * to expire. Rounding also depends on unsigned long arithmetic
584 * to avoid overflow.
585 *
586 * Otherwise, if the number of ticks in the whole seconds part of
587 * the time difference fits in a long, then convert the parts to
588 * ticks separately and add, using similar rounding methods and
589 * overflow avoidance. This method would work in the previous
590 * case but it is slightly slower and assumes that hz is integral.
591 *
592 * Otherwise, round the time difference down to the maximum
593 * representable value.
594 *
595 * If ints have 32 bits, then the maximum value for any timeout in
596 * 10ms ticks is 248 days.
597 */
598 sec = tv->tv_sec;
599 usec = tv->tv_usec;
600 if (usec < 0) {
601 sec--;
602 usec += 1000000;
603 }
604 if (sec < 0) {
605#ifdef DIAGNOSTIC
606 if (usec > 0) {
607 sec++;
608 usec -= 1000000;
609 }
610 printf("tvotohz: negative time difference %ld sec %ld usec\n",
611 sec, usec);
612#endif
613 ticks = 1;
614 } else if (sec <= LONG_MAX / 1000000)
615 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
616 / tick + 1;
617 else if (sec <= LONG_MAX / hz)
618 ticks = sec * hz
619 + ((unsigned long)usec + (tick - 1)) / tick + 1;
620 else
621 ticks = LONG_MAX;
622 if (ticks > INT_MAX)
623 ticks = INT_MAX;
624 return ((int)ticks);
625}
626
627/*
628 * Start profiling on a process.
629 *
630 * Kernel profiling passes proc0 which never exits and hence
631 * keeps the profile clock running constantly.
632 */
633void
634startprofclock(p)
635 register struct proc *p;
636{
637
638 PROC_LOCK_ASSERT(p, MA_OWNED);
639 if (p->p_flag & P_STOPPROF)
640 return;
641 if ((p->p_flag & P_PROFIL) == 0) {
642 p->p_flag |= P_PROFIL;
643 mtx_lock(&time_lock);
644 if (++profprocs == 1)
645 cpu_startprofclock();
646 mtx_unlock(&time_lock);
647 }
648}
649
650/*
651 * Stop profiling on a process.
652 */
653void
654stopprofclock(p)
655 register struct proc *p;
656{
657
658 PROC_LOCK_ASSERT(p, MA_OWNED);
659 if (p->p_flag & P_PROFIL) {
660 if (p->p_profthreads != 0) {
661 p->p_flag |= P_STOPPROF;
662 while (p->p_profthreads != 0)
663 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
664 "stopprof", 0);
665 p->p_flag &= ~P_STOPPROF;
666 }
667 if ((p->p_flag & P_PROFIL) == 0)
668 return;
669 p->p_flag &= ~P_PROFIL;
670 mtx_lock(&time_lock);
671 if (--profprocs == 0)
672 cpu_stopprofclock();
673 mtx_unlock(&time_lock);
674 }
675}
676
677/*
678 * Statistics clock. Updates rusage information and calls the scheduler
679 * to adjust priorities of the active thread.
680 *
681 * This should be called by all active processors.
682 */
683void
684statclock(int usermode)
685{
686 struct rusage *ru;
687 struct vmspace *vm;
688 struct thread *td;
689 struct proc *p;
690 long rss;
691 long *cp_time;
692
693 td = curthread;
694 p = td->td_proc;
695
696 cp_time = (long *)PCPU_PTR(cp_time);
697 if (usermode) {
698 /*
699 * Charge the time as appropriate.
700 */
701 td->td_uticks++;
702 if (p->p_nice > NZERO)
703 cp_time[CP_NICE]++;
704 else
705 cp_time[CP_USER]++;
706 } else {
707 /*
708 * Came from kernel mode, so we were:
709 * - handling an interrupt,
710 * - doing syscall or trap work on behalf of the current
711 * user process, or
712 * - spinning in the idle loop.
713 * Whichever it is, charge the time as appropriate.
714 * Note that we charge interrupts to the current process,
715 * regardless of whether they are ``for'' that process,
716 * so that we know how much of its real time was spent
717 * in ``non-process'' (i.e., interrupt) work.
718 */
719 if ((td->td_pflags & TDP_ITHREAD) ||
720 td->td_intr_nesting_level >= 2) {
721 td->td_iticks++;
722 cp_time[CP_INTR]++;
723 } else {
724 td->td_pticks++;
725 td->td_sticks++;
726 if (!TD_IS_IDLETHREAD(td))
727 cp_time[CP_SYS]++;
728 else
729 cp_time[CP_IDLE]++;
730 }
731 }
732
733 /* Update resource usage integrals and maximums. */
734 MPASS(p->p_vmspace != NULL);
735 vm = p->p_vmspace;
736 ru = &td->td_ru;
737 ru->ru_ixrss += pgtok(vm->vm_tsize);
738 ru->ru_idrss += pgtok(vm->vm_dsize);
739 ru->ru_isrss += pgtok(vm->vm_ssize);
740 rss = pgtok(vmspace_resident_count(vm));
741 if (ru->ru_maxrss < rss)
742 ru->ru_maxrss = rss;
743 KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
744 "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
745 thread_lock_flags(td, MTX_QUIET);
746 sched_clock(td);
747 thread_unlock(td);
748}
749
750void
751profclock(int usermode, uintfptr_t pc)
752{
753 struct thread *td;
754#ifdef GPROF
755 struct gmonparam *g;
756 uintfptr_t i;
757#endif
758
759 td = curthread;
760 if (usermode) {
761 /*
762 * Came from user mode; CPU was in user state.
763 * If this process is being profiled, record the tick.
764 * if there is no related user location yet, don't
765 * bother trying to count it.
766 */
767 if (td->td_proc->p_flag & P_PROFIL)
768 addupc_intr(td, pc, 1);
769 }
770#ifdef GPROF
771 else {
772 /*
773 * Kernel statistics are just like addupc_intr, only easier.
774 */
775 g = &_gmonparam;
776 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
777 i = PC_TO_I(g, pc);
778 if (i < g->textsize) {
779 KCOUNT(g, i)++;
780 }
781 }
782 }
783#endif
784}
785
786/*
787 * Return information about system clocks.
788 */
789static int
790sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
791{
792 struct clockinfo clkinfo;
793 /*
794 * Construct clockinfo structure.
795 */
796 bzero(&clkinfo, sizeof(clkinfo));
797 clkinfo.hz = hz;
798 clkinfo.tick = tick;
799 clkinfo.profhz = profhz;
800 clkinfo.stathz = stathz ? stathz : hz;
801 return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
802}
803
804SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
805 CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
806 0, 0, sysctl_kern_clockrate, "S,clockinfo",
807 "Rate and period of various kernel clocks");
808
809#ifdef SW_WATCHDOG
810
811static void
812watchdog_config(void *unused __unused, u_int cmd, int *error)
813{
814 u_int u;
815
816 u = cmd & WD_INTERVAL;
817 if (u >= WD_TO_1SEC) {
818 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
819 watchdog_enabled = 1;
820 *error = 0;
821 } else {
822 watchdog_enabled = 0;
823 }
824}
825
826/*
827 * Handle a watchdog timeout by dumping interrupt information and
828 * then either dropping to DDB or panicking.
829 */
830static void
831watchdog_fire(void)
832{
833 int nintr;
834 uint64_t inttotal;
835 u_long *curintr;
836 char *curname;
837
838 curintr = intrcnt;
839 curname = intrnames;
840 inttotal = 0;
841 nintr = eintrcnt - intrcnt;
842
843 printf("interrupt total\n");
844 while (--nintr >= 0) {
845 if (*curintr)
846 printf("%-12s %20lu\n", curname, *curintr);
847 curname += strlen(curname) + 1;
848 inttotal += *curintr++;
849 }
850 printf("Total %20ju\n", (uintmax_t)inttotal);
851
852#if defined(KDB) && !defined(KDB_UNATTENDED)
853 kdb_backtrace();
854 kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
855#else
856 panic("watchdog timeout");
857#endif
858}
859
860#endif /* SW_WATCHDOG */
|