xref: /freebsd/sys/kern/kern_synch.c (revision 2e5b60079b7d8c3ca68f1390cd90f305e651f8d3)
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 #include "opt_sched.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/condvar.h>
46 #include <sys/kdb.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
54 #include <sys/sdt.h>
55 #include <sys/signalvar.h>
56 #include <sys/sleepqueue.h>
57 #include <sys/smp.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysproto.h>
61 #include <sys/vmmeter.h>
62 #ifdef KTRACE
63 #include <sys/uio.h>
64 #include <sys/ktrace.h>
65 #endif
66 
67 #include <machine/cpu.h>
68 
69 #ifdef XEN
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/pmap.h>
73 #endif
74 
75 #define	KTDSTATE(td)							\
76 	(((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep"  :		\
77 	((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" :	\
78 	((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" :		\
79 	((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" :		\
80 	((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding")
81 
82 static void synch_setup(void *dummy);
83 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
84     NULL);
85 
86 int	hogticks;
87 static uint8_t pause_wchan[MAXCPU];
88 
89 static struct callout loadav_callout;
90 
91 struct loadavg averunnable =
92 	{ {0, 0, 0}, FSCALE };	/* load average, of runnable procs */
93 /*
94  * Constants for averages over 1, 5, and 15 minutes
95  * when sampling at 5 second intervals.
96  */
97 static fixpt_t cexp[3] = {
98 	0.9200444146293232 * FSCALE,	/* exp(-1/12) */
99 	0.9834714538216174 * FSCALE,	/* exp(-1/60) */
100 	0.9944598480048967 * FSCALE,	/* exp(-1/180) */
101 };
102 
103 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
104 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
105 
106 static void	loadav(void *arg);
107 
108 SDT_PROVIDER_DECLARE(sched);
109 SDT_PROBE_DEFINE(sched, , , preempt);
110 
111 /*
112  * These probes reference Solaris features that are not implemented in FreeBSD.
113  * Create the probes anyway for compatibility with existing D scripts; they'll
114  * just never fire.
115  */
116 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep);
117 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
118 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
119 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
120 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
121 
122 static void
123 sleepinit(void *unused)
124 {
125 
126 	hogticks = (hz / 10) * 2;	/* Default only. */
127 	init_sleepqueues();
128 }
129 
130 /*
131  * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
132  * it is available.
133  */
134 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
135 
136 /*
137  * General sleep call.  Suspends the current thread until a wakeup is
138  * performed on the specified identifier.  The thread will then be made
139  * runnable with the specified priority.  Sleeps at most sbt units of time
140  * (0 means no timeout).  If pri includes the PCATCH flag, let signals
141  * interrupt the sleep, otherwise ignore them while sleeping.  Returns 0 if
142  * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
143  * signal becomes pending, ERESTART is returned if the current system
144  * call should be restarted if possible, and EINTR is returned if the system
145  * call should be interrupted by the signal (return EINTR).
146  *
147  * The lock argument is unlocked before the caller is suspended, and
148  * re-locked before _sleep() returns.  If priority includes the PDROP
149  * flag the lock is not re-locked before returning.
150  */
151 int
152 _sleep(void *ident, struct lock_object *lock, int priority,
153     const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
154 {
155 	struct thread *td;
156 	struct proc *p;
157 	struct lock_class *class;
158 	uintptr_t lock_state;
159 	int catch, pri, rval, sleepq_flags;
160 	WITNESS_SAVE_DECL(lock_witness);
161 
162 	td = curthread;
163 	p = td->td_proc;
164 #ifdef KTRACE
165 	if (KTRPOINT(td, KTR_CSW))
166 		ktrcsw(1, 0, wmesg);
167 #endif
168 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
169 	    "Sleeping on \"%s\"", wmesg);
170 	KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
171 	    ("sleeping without a lock"));
172 	KASSERT(p != NULL, ("msleep1"));
173 	KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
174 	if (priority & PDROP)
175 		KASSERT(lock != NULL && lock != &Giant.lock_object,
176 		    ("PDROP requires a non-Giant lock"));
177 	if (lock != NULL)
178 		class = LOCK_CLASS(lock);
179 	else
180 		class = NULL;
181 
182 	if (cold || SCHEDULER_STOPPED()) {
183 		/*
184 		 * During autoconfiguration, just return;
185 		 * don't run any other threads or panic below,
186 		 * in case this is the idle thread and already asleep.
187 		 * XXX: this used to do "s = splhigh(); splx(safepri);
188 		 * splx(s);" to give interrupts a chance, but there is
189 		 * no way to give interrupts a chance now.
190 		 */
191 		if (lock != NULL && priority & PDROP)
192 			class->lc_unlock(lock);
193 		return (0);
194 	}
195 	catch = priority & PCATCH;
196 	pri = priority & PRIMASK;
197 
198 	/*
199 	 * If we are already on a sleep queue, then remove us from that
200 	 * sleep queue first.  We have to do this to handle recursive
201 	 * sleeps.
202 	 */
203 	if (TD_ON_SLEEPQ(td))
204 		sleepq_remove(td, td->td_wchan);
205 
206 	if ((uint8_t *)ident >= &pause_wchan[0] &&
207 	    (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
208 		sleepq_flags = SLEEPQ_PAUSE;
209 	else
210 		sleepq_flags = SLEEPQ_SLEEP;
211 	if (catch)
212 		sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
213 
214 	sleepq_lock(ident);
215 	CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
216 	    td->td_tid, p->p_pid, td->td_name, wmesg, ident);
217 
218 	if (lock == &Giant.lock_object)
219 		mtx_assert(&Giant, MA_OWNED);
220 	DROP_GIANT();
221 	if (lock != NULL && lock != &Giant.lock_object &&
222 	    !(class->lc_flags & LC_SLEEPABLE)) {
223 		WITNESS_SAVE(lock, lock_witness);
224 		lock_state = class->lc_unlock(lock);
225 	} else
226 		/* GCC needs to follow the Yellow Brick Road */
227 		lock_state = -1;
228 
229 	/*
230 	 * We put ourselves on the sleep queue and start our timeout
231 	 * before calling thread_suspend_check, as we could stop there,
232 	 * and a wakeup or a SIGCONT (or both) could occur while we were
233 	 * stopped without resuming us.  Thus, we must be ready for sleep
234 	 * when cursig() is called.  If the wakeup happens while we're
235 	 * stopped, then td will no longer be on a sleep queue upon
236 	 * return from cursig().
237 	 */
238 	sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
239 	if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
240 		sleepq_release(ident);
241 		WITNESS_SAVE(lock, lock_witness);
242 		lock_state = class->lc_unlock(lock);
243 		if (sbt != 0)
244 			sleepq_set_timeout_sbt(ident, sbt, pr, flags);
245 		sleepq_lock(ident);
246 	} else if (sbt != 0) {
247 		sleepq_release(ident);
248 		sleepq_set_timeout_sbt(ident, sbt, pr, flags);
249 		sleepq_lock(ident);
250 	}
251 	if (sbt != 0 && catch)
252 		rval = sleepq_timedwait_sig(ident, pri);
253 	else if (sbt != 0)
254 		rval = sleepq_timedwait(ident, pri);
255 	else if (catch)
256 		rval = sleepq_wait_sig(ident, pri);
257 	else {
258 		sleepq_wait(ident, pri);
259 		rval = 0;
260 	}
261 #ifdef KTRACE
262 	if (KTRPOINT(td, KTR_CSW))
263 		ktrcsw(0, 0, wmesg);
264 #endif
265 	PICKUP_GIANT();
266 	if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
267 		class->lc_lock(lock, lock_state);
268 		WITNESS_RESTORE(lock, lock_witness);
269 	}
270 	return (rval);
271 }
272 
273 int
274 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
275     sbintime_t sbt, sbintime_t pr, int flags)
276 {
277 	struct thread *td;
278 	struct proc *p;
279 	int rval;
280 	WITNESS_SAVE_DECL(mtx);
281 
282 	td = curthread;
283 	p = td->td_proc;
284 	KASSERT(mtx != NULL, ("sleeping without a mutex"));
285 	KASSERT(p != NULL, ("msleep1"));
286 	KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
287 
288 	if (cold || SCHEDULER_STOPPED()) {
289 		/*
290 		 * During autoconfiguration, just return;
291 		 * don't run any other threads or panic below,
292 		 * in case this is the idle thread and already asleep.
293 		 * XXX: this used to do "s = splhigh(); splx(safepri);
294 		 * splx(s);" to give interrupts a chance, but there is
295 		 * no way to give interrupts a chance now.
296 		 */
297 		return (0);
298 	}
299 
300 	sleepq_lock(ident);
301 	CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
302 	    td->td_tid, p->p_pid, td->td_name, wmesg, ident);
303 
304 	DROP_GIANT();
305 	mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
306 	WITNESS_SAVE(&mtx->lock_object, mtx);
307 	mtx_unlock_spin(mtx);
308 
309 	/*
310 	 * We put ourselves on the sleep queue and start our timeout.
311 	 */
312 	sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
313 	if (sbt != 0) {
314 		sleepq_release(ident);
315 		sleepq_set_timeout_sbt(ident, sbt, pr, flags);
316 		sleepq_lock(ident);
317 	}
318 
319 	/*
320 	 * Can't call ktrace with any spin locks held so it can lock the
321 	 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
322 	 * any spin lock.  Thus, we have to drop the sleepq spin lock while
323 	 * we handle those requests.  This is safe since we have placed our
324 	 * thread on the sleep queue already.
325 	 */
326 #ifdef KTRACE
327 	if (KTRPOINT(td, KTR_CSW)) {
328 		sleepq_release(ident);
329 		ktrcsw(1, 0, wmesg);
330 		sleepq_lock(ident);
331 	}
332 #endif
333 #ifdef WITNESS
334 	sleepq_release(ident);
335 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
336 	    wmesg);
337 	sleepq_lock(ident);
338 #endif
339 	if (sbt != 0)
340 		rval = sleepq_timedwait(ident, 0);
341 	else {
342 		sleepq_wait(ident, 0);
343 		rval = 0;
344 	}
345 #ifdef KTRACE
346 	if (KTRPOINT(td, KTR_CSW))
347 		ktrcsw(0, 0, wmesg);
348 #endif
349 	PICKUP_GIANT();
350 	mtx_lock_spin(mtx);
351 	WITNESS_RESTORE(&mtx->lock_object, mtx);
352 	return (rval);
353 }
354 
355 /*
356  * pause() delays the calling thread by the given number of system ticks.
357  * During cold bootup, pause() uses the DELAY() function instead of
358  * the tsleep() function to do the waiting. The "timo" argument must be
359  * greater than or equal to zero. A "timo" value of zero is equivalent
360  * to a "timo" value of one.
361  */
362 int
363 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
364 {
365 	KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
366 
367 	/* silently convert invalid timeouts */
368 	if (sbt == 0)
369 		sbt = tick_sbt;
370 
371 	if (cold || kdb_active) {
372 		/*
373 		 * We delay one second at a time to avoid overflowing the
374 		 * system specific DELAY() function(s):
375 		 */
376 		while (sbt >= SBT_1S) {
377 			DELAY(1000000);
378 			sbt -= SBT_1S;
379 		}
380 		/* Do the delay remainder, if any */
381 		sbt = (sbt + SBT_1US - 1) / SBT_1US;
382 		if (sbt > 0)
383 			DELAY(sbt);
384 		return (0);
385 	}
386 	return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
387 }
388 
389 /*
390  * Make all threads sleeping on the specified identifier runnable.
391  */
392 void
393 wakeup(void *ident)
394 {
395 	int wakeup_swapper;
396 
397 	sleepq_lock(ident);
398 	wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
399 	sleepq_release(ident);
400 	if (wakeup_swapper) {
401 		KASSERT(ident != &proc0,
402 		    ("wakeup and wakeup_swapper and proc0"));
403 		kick_proc0();
404 	}
405 }
406 
407 /*
408  * Make a thread sleeping on the specified identifier runnable.
409  * May wake more than one thread if a target thread is currently
410  * swapped out.
411  */
412 void
413 wakeup_one(void *ident)
414 {
415 	int wakeup_swapper;
416 
417 	sleepq_lock(ident);
418 	wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
419 	sleepq_release(ident);
420 	if (wakeup_swapper)
421 		kick_proc0();
422 }
423 
424 static void
425 kdb_switch(void)
426 {
427 	thread_unlock(curthread);
428 	kdb_backtrace();
429 	kdb_reenter();
430 	panic("%s: did not reenter debugger", __func__);
431 }
432 
433 /*
434  * The machine independent parts of context switching.
435  */
436 void
437 mi_switch(int flags, struct thread *newtd)
438 {
439 	uint64_t runtime, new_switchtime;
440 	struct thread *td;
441 	struct proc *p;
442 
443 	td = curthread;			/* XXX */
444 	THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
445 	p = td->td_proc;		/* XXX */
446 	KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
447 #ifdef INVARIANTS
448 	if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
449 		mtx_assert(&Giant, MA_NOTOWNED);
450 #endif
451 	KASSERT(td->td_critnest == 1 || panicstr,
452 	    ("mi_switch: switch in a critical section"));
453 	KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
454 	    ("mi_switch: switch must be voluntary or involuntary"));
455 	KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
456 
457 	/*
458 	 * Don't perform context switches from the debugger.
459 	 */
460 	if (kdb_active)
461 		kdb_switch();
462 	if (SCHEDULER_STOPPED())
463 		return;
464 	if (flags & SW_VOL) {
465 		td->td_ru.ru_nvcsw++;
466 		td->td_swvoltick = ticks;
467 	} else
468 		td->td_ru.ru_nivcsw++;
469 #ifdef SCHED_STATS
470 	SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
471 #endif
472 	/*
473 	 * Compute the amount of time during which the current
474 	 * thread was running, and add that to its total so far.
475 	 */
476 	new_switchtime = cpu_ticks();
477 	runtime = new_switchtime - PCPU_GET(switchtime);
478 	td->td_runtime += runtime;
479 	td->td_incruntime += runtime;
480 	PCPU_SET(switchtime, new_switchtime);
481 	td->td_generation++;	/* bump preempt-detect counter */
482 	PCPU_INC(cnt.v_swtch);
483 	PCPU_SET(switchticks, ticks);
484 	CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
485 	    td->td_tid, td->td_sched, p->p_pid, td->td_name);
486 #if (KTR_COMPILE & KTR_SCHED) != 0
487 	if (TD_IS_IDLETHREAD(td))
488 		KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
489 		    "prio:%d", td->td_priority);
490 	else
491 		KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
492 		    "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
493 		    "lockname:\"%s\"", td->td_lockname);
494 #endif
495 	SDT_PROBE0(sched, , , preempt);
496 #ifdef XEN
497 	PT_UPDATES_FLUSH();
498 #endif
499 	sched_switch(td, newtd, flags);
500 	KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
501 	    "prio:%d", td->td_priority);
502 
503 	CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
504 	    td->td_tid, td->td_sched, p->p_pid, td->td_name);
505 
506 	/*
507 	 * If the last thread was exiting, finish cleaning it up.
508 	 */
509 	if ((td = PCPU_GET(deadthread))) {
510 		PCPU_SET(deadthread, NULL);
511 		thread_stash(td);
512 	}
513 }
514 
515 /*
516  * Change thread state to be runnable, placing it on the run queue if
517  * it is in memory.  If it is swapped out, return true so our caller
518  * will know to awaken the swapper.
519  */
520 int
521 setrunnable(struct thread *td)
522 {
523 
524 	THREAD_LOCK_ASSERT(td, MA_OWNED);
525 	KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
526 	    ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
527 	switch (td->td_state) {
528 	case TDS_RUNNING:
529 	case TDS_RUNQ:
530 		return (0);
531 	case TDS_INHIBITED:
532 		/*
533 		 * If we are only inhibited because we are swapped out
534 		 * then arange to swap in this process. Otherwise just return.
535 		 */
536 		if (td->td_inhibitors != TDI_SWAPPED)
537 			return (0);
538 		/* FALLTHROUGH */
539 	case TDS_CAN_RUN:
540 		break;
541 	default:
542 		printf("state is 0x%x", td->td_state);
543 		panic("setrunnable(2)");
544 	}
545 	if ((td->td_flags & TDF_INMEM) == 0) {
546 		if ((td->td_flags & TDF_SWAPINREQ) == 0) {
547 			td->td_flags |= TDF_SWAPINREQ;
548 			return (1);
549 		}
550 	} else
551 		sched_wakeup(td);
552 	return (0);
553 }
554 
555 /*
556  * Compute a tenex style load average of a quantity on
557  * 1, 5 and 15 minute intervals.
558  */
559 static void
560 loadav(void *arg)
561 {
562 	int i, nrun;
563 	struct loadavg *avg;
564 
565 	nrun = sched_load();
566 	avg = &averunnable;
567 
568 	for (i = 0; i < 3; i++)
569 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
570 		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
571 
572 	/*
573 	 * Schedule the next update to occur after 5 seconds, but add a
574 	 * random variation to avoid synchronisation with processes that
575 	 * run at regular intervals.
576 	 */
577 	callout_reset_sbt(&loadav_callout,
578 	    SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
579 	    loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
580 }
581 
582 /* ARGSUSED */
583 static void
584 synch_setup(void *dummy)
585 {
586 	callout_init(&loadav_callout, CALLOUT_MPSAFE);
587 
588 	/* Kick off timeout driven events by calling first time. */
589 	loadav(NULL);
590 }
591 
592 int
593 should_yield(void)
594 {
595 
596 	return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
597 }
598 
599 void
600 maybe_yield(void)
601 {
602 
603 	if (should_yield())
604 		kern_yield(PRI_USER);
605 }
606 
607 void
608 kern_yield(int prio)
609 {
610 	struct thread *td;
611 
612 	td = curthread;
613 	DROP_GIANT();
614 	thread_lock(td);
615 	if (prio == PRI_USER)
616 		prio = td->td_user_pri;
617 	if (prio >= 0)
618 		sched_prio(td, prio);
619 	mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
620 	thread_unlock(td);
621 	PICKUP_GIANT();
622 }
623 
624 /*
625  * General purpose yield system call.
626  */
627 int
628 sys_yield(struct thread *td, struct yield_args *uap)
629 {
630 
631 	thread_lock(td);
632 	if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
633 		sched_prio(td, PRI_MAX_TIMESHARE);
634 	mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
635 	thread_unlock(td);
636 	td->td_retval[0] = 0;
637 	return (0);
638 }
639