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("$FreeBSD$");
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 (lock != NULL)
143 class = LOCK_CLASS(lock);
144 else
145 class = NULL;
146
147 if (cold) {
148 /*
149 * During autoconfiguration, just return;
150 * don't run any other threads or panic below,
151 * in case this is the idle thread and already asleep.
152 * XXX: this used to do "s = splhigh(); splx(safepri);
153 * splx(s);" to give interrupts a chance, but there is
154 * no way to give interrupts a chance now.
155 */
156 if (lock != NULL && priority & PDROP)
157 class->lc_unlock(lock);
158 return (0);
159 }
160 catch = priority & PCATCH;
161 rval = 0;
162
163 /*
164 * If we are already on a sleep queue, then remove us from that
165 * sleep queue first. We have to do this to handle recursive
166 * sleeps.
167 */
168 if (TD_ON_SLEEPQ(td))
169 sleepq_remove(td, td->td_wchan);
170
171 if (ident == &pause_wchan)
172 flags = SLEEPQ_PAUSE;
173 else
174 flags = SLEEPQ_SLEEP;
175 if (catch)
176 flags |= SLEEPQ_INTERRUPTIBLE;
177
178 sleepq_lock(ident);
179 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
180 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
181
182 DROP_GIANT();
183 if (lock != NULL && !(class->lc_flags & LC_SLEEPABLE)) {
184 WITNESS_SAVE(lock, lock_witness);
185 lock_state = class->lc_unlock(lock);
186 } else
187 /* GCC needs to follow the Yellow Brick Road */
188 lock_state = -1;
189
190 /*
191 * We put ourselves on the sleep queue and start our timeout
192 * before calling thread_suspend_check, as we could stop there,
193 * and a wakeup or a SIGCONT (or both) could occur while we were
194 * stopped without resuming us. Thus, we must be ready for sleep
195 * when cursig() is called. If the wakeup happens while we're
196 * stopped, then td will no longer be on a sleep queue upon
197 * return from cursig().
198 */
199 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0);
200 if (timo)
201 sleepq_set_timeout(ident, timo);
202 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
203 sleepq_release(ident);
204 WITNESS_SAVE(lock, lock_witness);
205 lock_state = class->lc_unlock(lock);
206 sleepq_lock(ident);
207 }
208
209 /*
210 * Adjust this thread's priority, if necessary.
211 */
212 pri = priority & PRIMASK;
213 if (pri != 0 && pri != td->td_priority) {
214 thread_lock(td);
215 sched_prio(td, pri);
216 thread_unlock(td);
217 }
218
219 if (timo && catch)
220 rval = sleepq_timedwait_sig(ident);
221 else if (timo)
222 rval = sleepq_timedwait(ident);
223 else if (catch)
224 rval = sleepq_wait_sig(ident);
225 else {
226 sleepq_wait(ident);
227 rval = 0;
228 }
229 #ifdef KTRACE
230 if (KTRPOINT(td, KTR_CSW))
231 ktrcsw(0, 0);
232 #endif
233 PICKUP_GIANT();
234 if (lock != NULL && !(priority & PDROP)) {
235 class->lc_lock(lock, lock_state);
236 WITNESS_RESTORE(lock, lock_witness);
237 }
238 return (rval);
239 }
240
241 int
242 msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, int timo)
243 {
244 struct thread *td;
245 struct proc *p;
246 int rval;
247 WITNESS_SAVE_DECL(mtx);
248
249 td = curthread;
250 p = td->td_proc;
251 KASSERT(mtx != NULL, ("sleeping without a mutex"));
252 KASSERT(p != NULL, ("msleep1"));
253 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
254
255 if (cold) {
256 /*
257 * During autoconfiguration, just return;
258 * don't run any other threads or panic below,
259 * in case this is the idle thread and already asleep.
260 * XXX: this used to do "s = splhigh(); splx(safepri);
261 * splx(s);" to give interrupts a chance, but there is
262 * no way to give interrupts a chance now.
263 */
264 return (0);
265 }
266
267 sleepq_lock(ident);
268 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
269 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
270
271 DROP_GIANT();
272 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
273 WITNESS_SAVE(&mtx->lock_object, mtx);
274 mtx_unlock_spin(mtx);
275
276 /*
277 * We put ourselves on the sleep queue and start our timeout.
278 */
279 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
280 if (timo)
281 sleepq_set_timeout(ident, timo);
282
283 /*
284 * Can't call ktrace with any spin locks held so it can lock the
285 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
286 * any spin lock. Thus, we have to drop the sleepq spin lock while
287 * we handle those requests. This is safe since we have placed our
288 * thread on the sleep queue already.
289 */
290 #ifdef KTRACE
291 if (KTRPOINT(td, KTR_CSW)) {
292 sleepq_release(ident);
293 ktrcsw(1, 0);
294 sleepq_lock(ident);
295 }
296 #endif
297 #ifdef WITNESS
298 sleepq_release(ident);
299 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
300 wmesg);
301 sleepq_lock(ident);
302 #endif
303 if (timo)
304 rval = sleepq_timedwait(ident);
305 else {
306 sleepq_wait(ident);
307 rval = 0;
308 }
309 #ifdef KTRACE
310 if (KTRPOINT(td, KTR_CSW))
311 ktrcsw(0, 0);
312 #endif
313 PICKUP_GIANT();
314 mtx_lock_spin(mtx);
315 WITNESS_RESTORE(&mtx->lock_object, mtx);
316 return (rval);
317 }
318
319 /*
320 * pause() is like tsleep() except that the intention is to not be
321 * explicitly woken up by another thread. Instead, the current thread
322 * simply wishes to sleep until the timeout expires. It is
323 * implemented using a dummy wait channel.
324 */
325 int
326 pause(const char *wmesg, int timo)
327 {
328
329 KASSERT(timo != 0, ("pause: timeout required"));
330 return (tsleep(&pause_wchan, 0, wmesg, timo));
331 }
332
333 /*
334 * Make all threads sleeping on the specified identifier runnable.
335 */
336 void
337 wakeup(void *ident)
338 {
339 int wakeup_swapper;
340
341 sleepq_lock(ident);
342 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, -1, 0);
343 if (wakeup_swapper)
344 kick_proc0();
345 }
346
347 /*
348 * Make a thread sleeping on the specified identifier runnable.
349 * May wake more than one thread if a target thread is currently
350 * swapped out.
351 */
352 void
353 wakeup_one(void *ident)
354 {
355 int wakeup_swapper;
356
357 sleepq_lock(ident);
358 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, -1, 0);
359 sleepq_release(ident);
360 if (wakeup_swapper)
361 kick_proc0();
362 }
363
364 /*
365 * The machine independent parts of context switching.
366 */
367 void
368 mi_switch(int flags, struct thread *newtd)
369 {
370 uint64_t runtime, new_switchtime;
371 struct thread *td;
372 struct proc *p;
373
374 td = curthread; /* XXX */
375 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
376 p = td->td_proc; /* XXX */
377 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
378 #ifdef INVARIANTS
379 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
380 mtx_assert(&Giant, MA_NOTOWNED);
381 #endif
382 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
383 (td->td_owepreempt) && (flags & SW_INVOL) != 0 &&
384 newtd == NULL) || panicstr,
385 ("mi_switch: switch in a critical section"));
386 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
387 ("mi_switch: switch must be voluntary or involuntary"));
388 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
389
390 /*
391 * Don't perform context switches from the debugger.
392 */
393 if (kdb_active) {
394 thread_unlock(td);
395 kdb_backtrace();
396 kdb_reenter();
397 panic("%s: did not reenter debugger", __func__);
398 }
399 if (flags & SW_VOL)
400 td->td_ru.ru_nvcsw++;
401 else
402 td->td_ru.ru_nivcsw++;
403 /*
404 * Compute the amount of time during which the current
405 * thread was running, and add that to its total so far.
406 */
407 new_switchtime = cpu_ticks();
408 runtime = new_switchtime - PCPU_GET(switchtime);
409 td->td_runtime += runtime;
410 td->td_incruntime += runtime;
411 PCPU_SET(switchtime, new_switchtime);
412 td->td_generation++; /* bump preempt-detect counter */
413 PCPU_INC(cnt.v_swtch);
414 PCPU_SET(switchticks, ticks);
415 CTR4(KTR_PROC, "mi_switch: old thread %ld (kse %p, pid %ld, %s)",
416 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
417 #if (KTR_COMPILE & KTR_SCHED) != 0
418 if (TD_IS_IDLETHREAD(td))
419 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",
420 td, td->td_proc->p_comm, td->td_priority);
421 else if (newtd != NULL)
422 CTR5(KTR_SCHED,
423 "mi_switch: %p(%s) prio %d preempted by %p(%s)",
424 td, td->td_proc->p_comm, td->td_priority, newtd,
425 newtd->td_proc->p_comm);
426 else
427 CTR6(KTR_SCHED,
428 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s",
429 td, td->td_proc->p_comm, td->td_priority,
430 td->td_inhibitors, td->td_wmesg, td->td_lockname);
431 #endif
432 /*
433 * We call thread_switchout after the KTR_SCHED prints above so kse
434 * selecting a new thread to run does not show up as a preemption.
435 */
436 #ifdef KSE
437 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
438 newtd = thread_switchout(td, flags, newtd);
439 #endif
440 sched_switch(td, newtd, flags);
441 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d",
442 td, td->td_proc->p_comm, td->td_priority);
443
444 CTR4(KTR_PROC, "mi_switch: new thread %ld (kse %p, pid %ld, %s)",
445 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
446
447 /*
448 * If the last thread was exiting, finish cleaning it up.
449 */
450 if ((td = PCPU_GET(deadthread))) {
451 PCPU_SET(deadthread, NULL);
452 thread_stash(td);
453 }
454 }
455
456 /*
457 * Change thread state to be runnable, placing it on the run queue if
458 * it is in memory. If it is swapped out, return true so our caller
459 * will know to awaken the swapper.
460 */
461 int
462 setrunnable(struct thread *td)
463 {
464
465 THREAD_LOCK_ASSERT(td, MA_OWNED);
466 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
467 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
468 switch (td->td_state) {
469 case TDS_RUNNING:
470 case TDS_RUNQ:
471 return (0);
472 case TDS_INHIBITED:
473 /*
474 * If we are only inhibited because we are swapped out
475 * then arange to swap in this process. Otherwise just return.
476 */
477 if (td->td_inhibitors != TDI_SWAPPED)
478 return (0);
479 /* FALLTHROUGH */
480 case TDS_CAN_RUN:
481 break;
482 default:
483 printf("state is 0x%x", td->td_state);
484 panic("setrunnable(2)");
485 }
486 if ((td->td_flags & TDF_INMEM) == 0) {
487 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
488 td->td_flags |= TDF_SWAPINREQ;
489 return (1);
490 }
491 } else
492 sched_wakeup(td);
493 return (0);
494 }
495
496 /*
497 * Compute a tenex style load average of a quantity on
498 * 1, 5 and 15 minute intervals.
499 * XXXKSE Needs complete rewrite when correct info is available.
500 * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
501 */
502 static void
503 loadav(void *arg)
504 {
505 int i, nrun;
506 struct loadavg *avg;
507
508 nrun = sched_load();
509 avg = &averunnable;
510
511 for (i = 0; i < 3; i++)
512 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
513 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
514
515 /*
516 * Schedule the next update to occur after 5 seconds, but add a
517 * random variation to avoid synchronisation with processes that
518 * run at regular intervals.
519 */
520 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
521 loadav, NULL);
522 }
523
524 static void
525 lboltcb(void *arg)
526 {
527 wakeup(&lbolt);
528 callout_reset(&lbolt_callout, hz, lboltcb, NULL);
529 }
530
531 /* ARGSUSED */
532 static void
533 synch_setup(void *dummy)
534 {
535 callout_init(&loadav_callout, CALLOUT_MPSAFE);
536 callout_init(&lbolt_callout, CALLOUT_MPSAFE);
537
538 /* Kick off timeout driven events by calling first time. */
539 loadav(NULL);
540 lboltcb(NULL);
541 }
542
543 /*
544 * General purpose yield system call.
545 */
546 int
547 yield(struct thread *td, struct yield_args *uap)
548 {
549
550 thread_lock(td);
551 sched_prio(td, PRI_MAX_TIMESHARE);
552 mi_switch(SW_VOL, NULL);
553 thread_unlock(td);
554 td->td_retval[0] = 0;
555 return (0);
556 }
Cache object: b456eb8f49148b35cec0f7a53b90f249
|