FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_synch.c
1 /*-
2 * Copyright (c) 1982, 1986, 1990, 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_synch.c 8.9 (Berkeley) 5/19/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/7.3/sys/kern/kern_synch.c 188730 2009-02-17 21:02:35Z jhb $");
39
40 #include "opt_ktrace.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/condvar.h>
45 #include <sys/kdb.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/proc.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sched.h>
53 #include <sys/signalvar.h>
54 #include <sys/sleepqueue.h>
55 #include <sys/smp.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysproto.h>
59 #include <sys/vmmeter.h>
60 #ifdef KTRACE
61 #include <sys/uio.h>
62 #include <sys/ktrace.h>
63 #endif
64
65 #include <machine/cpu.h>
66
67 static void synch_setup(void *dummy);
68 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
69 NULL);
70
71 int hogticks;
72 int lbolt;
73 static int pause_wchan;
74
75 static struct callout loadav_callout;
76 static struct callout lbolt_callout;
77
78 struct loadavg averunnable =
79 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
80 /*
81 * Constants for averages over 1, 5, and 15 minutes
82 * when sampling at 5 second intervals.
83 */
84 static fixpt_t cexp[3] = {
85 0.9200444146293232 * FSCALE, /* exp(-1/12) */
86 0.9834714538216174 * FSCALE, /* exp(-1/60) */
87 0.9944598480048967 * FSCALE, /* exp(-1/180) */
88 };
89
90 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
91 static int fscale __unused = FSCALE;
92 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
93
94 static void loadav(void *arg);
95 static void lboltcb(void *arg);
96
97 void
98 sleepinit(void)
99 {
100
101 hogticks = (hz / 10) * 2; /* Default only. */
102 init_sleepqueues();
103 }
104
105 /*
106 * General sleep call. Suspends the current thread until a wakeup is
107 * performed on the specified identifier. The thread will then be made
108 * runnable with the specified priority. Sleeps at most timo/hz seconds
109 * (0 means no timeout). If pri includes PCATCH flag, signals are checked
110 * before and after sleeping, else signals are not checked. Returns 0 if
111 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
112 * signal needs to be delivered, ERESTART is returned if the current system
113 * call should be restarted if possible, and EINTR is returned if the system
114 * call should be interrupted by the signal (return EINTR).
115 *
116 * The lock argument is unlocked before the caller is suspended, and
117 * re-locked before _sleep() returns. If priority includes the PDROP
118 * flag the lock is not re-locked before returning.
119 */
120 int
121 _sleep(void *ident, struct lock_object *lock, int priority,
122 const char *wmesg, int timo)
123 {
124 struct thread *td;
125 struct proc *p;
126 struct lock_class *class;
127 int catch, flags, lock_state, pri, rval;
128 WITNESS_SAVE_DECL(lock_witness);
129
130 td = curthread;
131 p = td->td_proc;
132 #ifdef KTRACE
133 if (KTRPOINT(td, KTR_CSW))
134 ktrcsw(1, 0);
135 #endif
136 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
137 "Sleeping on \"%s\"", wmesg);
138 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL ||
139 ident == &lbolt, ("sleeping without a lock"));
140 KASSERT(p != NULL, ("msleep1"));
141 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
142 if (priority & PDROP)
143 KASSERT(lock != NULL && lock != &Giant.lock_object,
144 ("PDROP requires a non-Giant lock"));
145 if (lock != NULL)
146 class = LOCK_CLASS(lock);
147 else
148 class = NULL;
149
150 if (cold) {
151 /*
152 * During autoconfiguration, just return;
153 * don't run any other threads or panic below,
154 * in case this is the idle thread and already asleep.
155 * XXX: this used to do "s = splhigh(); splx(safepri);
156 * splx(s);" to give interrupts a chance, but there is
157 * no way to give interrupts a chance now.
158 */
159 if (lock != NULL && priority & PDROP)
160 class->lc_unlock(lock);
161 return (0);
162 }
163 catch = priority & PCATCH;
164 rval = 0;
165
166 /*
167 * If we are already on a sleep queue, then remove us from that
168 * sleep queue first. We have to do this to handle recursive
169 * sleeps.
170 */
171 if (TD_ON_SLEEPQ(td))
172 sleepq_remove(td, td->td_wchan);
173
174 if (ident == &pause_wchan)
175 flags = SLEEPQ_PAUSE;
176 else
177 flags = SLEEPQ_SLEEP;
178 if (catch)
179 flags |= SLEEPQ_INTERRUPTIBLE;
180
181 sleepq_lock(ident);
182 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
183 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
184
185 DROP_GIANT();
186 if (lock != NULL && lock != &Giant.lock_object &&
187 !(class->lc_flags & LC_SLEEPABLE)) {
188 WITNESS_SAVE(lock, lock_witness);
189 lock_state = class->lc_unlock(lock);
190 } else
191 /* GCC needs to follow the Yellow Brick Road */
192 lock_state = -1;
193
194 /*
195 * We put ourselves on the sleep queue and start our timeout
196 * before calling thread_suspend_check, as we could stop there,
197 * and a wakeup or a SIGCONT (or both) could occur while we were
198 * stopped without resuming us. Thus, we must be ready for sleep
199 * when cursig() is called. If the wakeup happens while we're
200 * stopped, then td will no longer be on a sleep queue upon
201 * return from cursig().
202 */
203 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0);
204 if (timo)
205 sleepq_set_timeout(ident, timo);
206 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
207 sleepq_release(ident);
208 WITNESS_SAVE(lock, lock_witness);
209 lock_state = class->lc_unlock(lock);
210 sleepq_lock(ident);
211 }
212
213 /*
214 * Adjust this thread's priority, if necessary.
215 */
216 pri = priority & PRIMASK;
217 if (pri != 0 && pri != td->td_priority) {
218 thread_lock(td);
219 sched_prio(td, pri);
220 thread_unlock(td);
221 }
222
223 if (timo && catch)
224 rval = sleepq_timedwait_sig(ident);
225 else if (timo)
226 rval = sleepq_timedwait(ident);
227 else if (catch)
228 rval = sleepq_wait_sig(ident);
229 else {
230 sleepq_wait(ident);
231 rval = 0;
232 }
233 #ifdef KTRACE
234 if (KTRPOINT(td, KTR_CSW))
235 ktrcsw(0, 0);
236 #endif
237 PICKUP_GIANT();
238 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
239 class->lc_lock(lock, lock_state);
240 WITNESS_RESTORE(lock, lock_witness);
241 }
242 return (rval);
243 }
244
245 int
246 msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, int timo)
247 {
248 struct thread *td;
249 struct proc *p;
250 int rval;
251 WITNESS_SAVE_DECL(mtx);
252
253 td = curthread;
254 p = td->td_proc;
255 KASSERT(mtx != NULL, ("sleeping without a mutex"));
256 KASSERT(p != NULL, ("msleep1"));
257 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
258
259 if (cold) {
260 /*
261 * During autoconfiguration, just return;
262 * don't run any other threads or panic below,
263 * in case this is the idle thread and already asleep.
264 * XXX: this used to do "s = splhigh(); splx(safepri);
265 * splx(s);" to give interrupts a chance, but there is
266 * no way to give interrupts a chance now.
267 */
268 return (0);
269 }
270
271 sleepq_lock(ident);
272 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
273 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
274
275 DROP_GIANT();
276 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
277 WITNESS_SAVE(&mtx->lock_object, mtx);
278 mtx_unlock_spin(mtx);
279
280 /*
281 * We put ourselves on the sleep queue and start our timeout.
282 */
283 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
284 if (timo)
285 sleepq_set_timeout(ident, timo);
286
287 /*
288 * Can't call ktrace with any spin locks held so it can lock the
289 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
290 * any spin lock. Thus, we have to drop the sleepq spin lock while
291 * we handle those requests. This is safe since we have placed our
292 * thread on the sleep queue already.
293 */
294 #ifdef KTRACE
295 if (KTRPOINT(td, KTR_CSW)) {
296 sleepq_release(ident);
297 ktrcsw(1, 0);
298 sleepq_lock(ident);
299 }
300 #endif
301 #ifdef WITNESS
302 sleepq_release(ident);
303 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
304 wmesg);
305 sleepq_lock(ident);
306 #endif
307 if (timo)
308 rval = sleepq_timedwait(ident);
309 else {
310 sleepq_wait(ident);
311 rval = 0;
312 }
313 #ifdef KTRACE
314 if (KTRPOINT(td, KTR_CSW))
315 ktrcsw(0, 0);
316 #endif
317 PICKUP_GIANT();
318 mtx_lock_spin(mtx);
319 WITNESS_RESTORE(&mtx->lock_object, mtx);
320 return (rval);
321 }
322
323 /*
324 * pause() is like tsleep() except that the intention is to not be
325 * explicitly woken up by another thread. Instead, the current thread
326 * simply wishes to sleep until the timeout expires. It is
327 * implemented using a dummy wait channel.
328 */
329 int
330 pause(const char *wmesg, int timo)
331 {
332
333 KASSERT(timo != 0, ("pause: timeout required"));
334 return (tsleep(&pause_wchan, 0, wmesg, timo));
335 }
336
337 /*
338 * Make all threads sleeping on the specified identifier runnable.
339 */
340 void
341 wakeup(void *ident)
342 {
343 int wakeup_swapper;
344
345 sleepq_lock(ident);
346 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, -1, 0);
347 if (wakeup_swapper)
348 kick_proc0();
349 }
350
351 /*
352 * Make a thread sleeping on the specified identifier runnable.
353 * May wake more than one thread if a target thread is currently
354 * swapped out.
355 */
356 void
357 wakeup_one(void *ident)
358 {
359 int wakeup_swapper;
360
361 sleepq_lock(ident);
362 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, -1, 0);
363 sleepq_release(ident);
364 if (wakeup_swapper)
365 kick_proc0();
366 }
367
368 /*
369 * The machine independent parts of context switching.
370 */
371 void
372 mi_switch(int flags, struct thread *newtd)
373 {
374 uint64_t runtime, new_switchtime;
375 struct thread *td;
376 struct proc *p;
377
378 td = curthread; /* XXX */
379 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
380 p = td->td_proc; /* XXX */
381 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
382 #ifdef INVARIANTS
383 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
384 mtx_assert(&Giant, MA_NOTOWNED);
385 #endif
386 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
387 (td->td_owepreempt) && (flags & SW_INVOL) != 0 &&
388 newtd == NULL) || panicstr,
389 ("mi_switch: switch in a critical section"));
390 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
391 ("mi_switch: switch must be voluntary or involuntary"));
392 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
393
394 /*
395 * Don't perform context switches from the debugger.
396 */
397 if (kdb_active) {
398 thread_unlock(td);
399 kdb_backtrace();
400 kdb_reenter();
401 panic("%s: did not reenter debugger", __func__);
402 }
403 if (flags & SW_VOL)
404 td->td_ru.ru_nvcsw++;
405 else
406 td->td_ru.ru_nivcsw++;
407 /*
408 * Compute the amount of time during which the current
409 * thread was running, and add that to its total so far.
410 */
411 new_switchtime = cpu_ticks();
412 runtime = new_switchtime - PCPU_GET(switchtime);
413 td->td_runtime += runtime;
414 td->td_incruntime += runtime;
415 PCPU_SET(switchtime, new_switchtime);
416 td->td_generation++; /* bump preempt-detect counter */
417 PCPU_INC(cnt.v_swtch);
418 PCPU_SET(switchticks, ticks);
419 CTR4(KTR_PROC, "mi_switch: old thread %ld (kse %p, pid %ld, %s)",
420 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
421 #if (KTR_COMPILE & KTR_SCHED) != 0
422 if (TD_IS_IDLETHREAD(td))
423 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",
424 td, td->td_proc->p_comm, td->td_priority);
425 else if (newtd != NULL)
426 CTR5(KTR_SCHED,
427 "mi_switch: %p(%s) prio %d preempted by %p(%s)",
428 td, td->td_proc->p_comm, td->td_priority, newtd,
429 newtd->td_proc->p_comm);
430 else
431 CTR6(KTR_SCHED,
432 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s",
433 td, td->td_proc->p_comm, td->td_priority,
434 td->td_inhibitors, td->td_wmesg, td->td_lockname);
435 #endif
436 /*
437 * We call thread_switchout after the KTR_SCHED prints above so kse
438 * selecting a new thread to run does not show up as a preemption.
439 */
440 #ifdef KSE
441 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
442 newtd = thread_switchout(td, flags, newtd);
443 #endif
444 sched_switch(td, newtd, flags);
445 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d",
446 td, td->td_proc->p_comm, td->td_priority);
447
448 CTR4(KTR_PROC, "mi_switch: new thread %ld (kse %p, pid %ld, %s)",
449 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
450
451 /*
452 * If the last thread was exiting, finish cleaning it up.
453 */
454 if ((td = PCPU_GET(deadthread))) {
455 PCPU_SET(deadthread, NULL);
456 thread_stash(td);
457 }
458 }
459
460 /*
461 * Change thread state to be runnable, placing it on the run queue if
462 * it is in memory. If it is swapped out, return true so our caller
463 * will know to awaken the swapper.
464 */
465 int
466 setrunnable(struct thread *td)
467 {
468
469 THREAD_LOCK_ASSERT(td, MA_OWNED);
470 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
471 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
472 switch (td->td_state) {
473 case TDS_RUNNING:
474 case TDS_RUNQ:
475 return (0);
476 case TDS_INHIBITED:
477 /*
478 * If we are only inhibited because we are swapped out
479 * then arange to swap in this process. Otherwise just return.
480 */
481 if (td->td_inhibitors != TDI_SWAPPED)
482 return (0);
483 /* FALLTHROUGH */
484 case TDS_CAN_RUN:
485 break;
486 default:
487 printf("state is 0x%x", td->td_state);
488 panic("setrunnable(2)");
489 }
490 if ((td->td_flags & TDF_INMEM) == 0) {
491 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
492 td->td_flags |= TDF_SWAPINREQ;
493 return (1);
494 }
495 } else
496 sched_wakeup(td);
497 return (0);
498 }
499
500 /*
501 * Compute a tenex style load average of a quantity on
502 * 1, 5 and 15 minute intervals.
503 * XXXKSE Needs complete rewrite when correct info is available.
504 * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
505 */
506 static void
507 loadav(void *arg)
508 {
509 int i, nrun;
510 struct loadavg *avg;
511
512 nrun = sched_load();
513 avg = &averunnable;
514
515 for (i = 0; i < 3; i++)
516 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
517 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
518
519 /*
520 * Schedule the next update to occur after 5 seconds, but add a
521 * random variation to avoid synchronisation with processes that
522 * run at regular intervals.
523 */
524 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
525 loadav, NULL);
526 }
527
528 static void
529 lboltcb(void *arg)
530 {
531 wakeup(&lbolt);
532 callout_reset(&lbolt_callout, hz, lboltcb, NULL);
533 }
534
535 /* ARGSUSED */
536 static void
537 synch_setup(void *dummy)
538 {
539 callout_init(&loadav_callout, CALLOUT_MPSAFE);
540 callout_init(&lbolt_callout, CALLOUT_MPSAFE);
541
542 /* Kick off timeout driven events by calling first time. */
543 loadav(NULL);
544 lboltcb(NULL);
545 }
546
547 /*
548 * General purpose yield system call.
549 */
550 int
551 yield(struct thread *td, struct yield_args *uap)
552 {
553
554 thread_lock(td);
555 sched_prio(td, PRI_MAX_TIMESHARE);
556 mi_switch(SW_VOL, NULL);
557 thread_unlock(td);
558 td->td_retval[0] = 0;
559 return (0);
560 }
Cache object: bd7beb532e0d7adcc2a8c48e15a1834a
|