kern_synch.c revision 107719
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 * 3. All advertising materials mentioning features or use of this software
19 *    must display the following acknowledgement:
20 *	This product includes software developed by the University of
21 *	California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 *    may be used to endorse or promote products derived from this software
24 *    without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 *	@(#)kern_synch.c	8.9 (Berkeley) 5/19/95
39 * $FreeBSD: head/sys/kern/kern_synch.c 107719 2002-12-10 02:33:45Z julian $
40 */
41
42#include "opt_ddb.h"
43#include "opt_ktrace.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/condvar.h>
48#include <sys/kernel.h>
49#include <sys/ktr.h>
50#include <sys/lock.h>
51#include <sys/mutex.h>
52#include <sys/proc.h>
53#include <sys/resourcevar.h>
54#include <sys/sched.h>
55#include <sys/signalvar.h>
56#include <sys/smp.h>
57#include <sys/sx.h>
58#include <sys/sysctl.h>
59#include <sys/sysproto.h>
60#include <sys/vmmeter.h>
61#ifdef DDB
62#include <ddb/ddb.h>
63#endif
64#ifdef KTRACE
65#include <sys/uio.h>
66#include <sys/ktrace.h>
67#endif
68
69#include <machine/cpu.h>
70
71static void sched_setup(void *dummy);
72SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
73
74int	hogticks;
75int	lbolt;
76
77static struct callout loadav_callout;
78static struct callout lbolt_callout;
79
80struct loadavg averunnable =
81	{ {0, 0, 0}, FSCALE };	/* load average, of runnable procs */
82/*
83 * Constants for averages over 1, 5, and 15 minutes
84 * when sampling at 5 second intervals.
85 */
86static fixpt_t cexp[3] = {
87	0.9200444146293232 * FSCALE,	/* exp(-1/12) */
88	0.9834714538216174 * FSCALE,	/* exp(-1/60) */
89	0.9944598480048967 * FSCALE,	/* exp(-1/180) */
90};
91
92/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
93static int      fscale __unused = FSCALE;
94SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
95
96static void	endtsleep(void *);
97static void	loadav(void *arg);
98static void	lboltcb(void *arg);
99
100/*
101 * We're only looking at 7 bits of the address; everything is
102 * aligned to 4, lots of things are aligned to greater powers
103 * of 2.  Shift right by 8, i.e. drop the bottom 256 worth.
104 */
105#define TABLESIZE	128
106static TAILQ_HEAD(slpquehead, thread) slpque[TABLESIZE];
107#define LOOKUP(x)	(((intptr_t)(x) >> 8) & (TABLESIZE - 1))
108
109void
110sleepinit(void)
111{
112	int i;
113
114	hogticks = (hz / 10) * 2;	/* Default only. */
115	for (i = 0; i < TABLESIZE; i++)
116		TAILQ_INIT(&slpque[i]);
117}
118
119/*
120 * General sleep call.  Suspends the current process until a wakeup is
121 * performed on the specified identifier.  The process will then be made
122 * runnable with the specified priority.  Sleeps at most timo/hz seconds
123 * (0 means no timeout).  If pri includes PCATCH flag, signals are checked
124 * before and after sleeping, else signals are not checked.  Returns 0 if
125 * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
126 * signal needs to be delivered, ERESTART is returned if the current system
127 * call should be restarted if possible, and EINTR is returned if the system
128 * call should be interrupted by the signal (return EINTR).
129 *
130 * The mutex argument is exited before the caller is suspended, and
131 * entered before msleep returns.  If priority includes the PDROP
132 * flag the mutex is not entered before returning.
133 */
134
135int
136msleep(ident, mtx, priority, wmesg, timo)
137	void *ident;
138	struct mtx *mtx;
139	int priority, timo;
140	const char *wmesg;
141{
142	struct thread *td = curthread;
143	struct proc *p = td->td_proc;
144	int sig, catch = priority & PCATCH;
145	int rval = 0;
146	WITNESS_SAVE_DECL(mtx);
147
148#ifdef KTRACE
149	if (KTRPOINT(td, KTR_CSW))
150		ktrcsw(1, 0);
151#endif
152	WITNESS_SLEEP(0, &mtx->mtx_object);
153	KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL,
154	    ("sleeping without a mutex"));
155	/*
156	 * If we are capable of async syscalls and there isn't already
157	 * another one ready to return, start a new thread
158	 * and queue it as ready to run. Note that there is danger here
159	 * because we need to make sure that we don't sleep allocating
160	 * the thread (recursion here might be bad).
161	 * Hence the TDF_INMSLEEP flag.
162	 */
163	if (p->p_flag & P_KSES) {
164		/*
165		 * Just don't bother if we are exiting
166		 * and not the exiting thread or thread was marked as
167		 * interrupted.
168		 */
169		if (catch &&
170		    (((p->p_flag & P_WEXIT) && (p->p_singlethread != td)) ||
171		     (td->td_flags & TDF_INTERRUPT))) {
172			td->td_flags &= ~TDF_INTERRUPT;
173			return (EINTR);
174		}
175		mtx_lock_spin(&sched_lock);
176		if ((td->td_flags & (TDF_UNBOUND|TDF_INMSLEEP)) ==
177		    TDF_UNBOUND) {
178			/*
179			 * Arrange for an upcall to be readied.
180			 * it will not actually happen until all
181			 * pending in-kernel work for this KSEGRP
182			 * has been done.
183			 */
184			/* Don't recurse here! */
185			td->td_flags |= TDF_INMSLEEP;
186			thread_schedule_upcall(td, td->td_kse);
187			td->td_flags &= ~TDF_INMSLEEP;
188		}
189	} else {
190		mtx_lock_spin(&sched_lock);
191	}
192	if (cold ) {
193		/*
194		 * During autoconfiguration, just give interrupts
195		 * a chance, then just return.
196		 * Don't run any other procs or panic below,
197		 * in case this is the idle process and already asleep.
198		 */
199		if (mtx != NULL && priority & PDROP)
200			mtx_unlock(mtx);
201		mtx_unlock_spin(&sched_lock);
202		return (0);
203	}
204
205	DROP_GIANT();
206
207	if (mtx != NULL) {
208		mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
209		WITNESS_SAVE(&mtx->mtx_object, mtx);
210		mtx_unlock(mtx);
211		if (priority & PDROP)
212			mtx = NULL;
213	}
214
215	KASSERT(p != NULL, ("msleep1"));
216	KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
217
218	CTR5(KTR_PROC, "msleep: thread %p (pid %d, %s) on %s (%p)",
219	    td, p->p_pid, p->p_comm, wmesg, ident);
220
221	td->td_wchan = ident;
222	td->td_wmesg = wmesg;
223	TAILQ_INSERT_TAIL(&slpque[LOOKUP(ident)], td, td_slpq);
224	TD_SET_ON_SLEEPQ(td);
225	if (timo)
226		callout_reset(&td->td_slpcallout, timo, endtsleep, td);
227	/*
228	 * We put ourselves on the sleep queue and start our timeout
229	 * before calling thread_suspend_check, as we could stop there, and
230	 * a wakeup or a SIGCONT (or both) could occur while we were stopped.
231	 * without resuming us, thus we must be ready for sleep
232	 * when cursig is called.  If the wakeup happens while we're
233	 * stopped, td->td_wchan will be 0 upon return from cursig.
234	 */
235	if (catch) {
236		CTR3(KTR_PROC, "msleep caught: thread %p (pid %d, %s)", td,
237		    p->p_pid, p->p_comm);
238		td->td_flags |= TDF_SINTR;
239		mtx_unlock_spin(&sched_lock);
240		PROC_LOCK(p);
241		sig = cursig(td);
242		if (sig == 0 && thread_suspend_check(1))
243			sig = SIGSTOP;
244		mtx_lock_spin(&sched_lock);
245		PROC_UNLOCK(p);
246		if (sig != 0) {
247			if (TD_ON_SLEEPQ(td))
248				unsleep(td);
249		} else if (!TD_ON_SLEEPQ(td))
250			catch = 0;
251	} else
252		sig = 0;
253
254	/*
255	 * Let the scheduler know we're about to voluntarily go to sleep.
256	 */
257	sched_sleep(td, priority & PRIMASK);
258
259	if (TD_ON_SLEEPQ(td)) {
260		p->p_stats->p_ru.ru_nvcsw++;
261		TD_SET_SLEEPING(td);
262		mi_switch();
263	}
264	/*
265	 * We're awake from voluntary sleep.
266	 */
267	CTR3(KTR_PROC, "msleep resume: thread %p (pid %d, %s)", td, p->p_pid,
268	    p->p_comm);
269	KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING"));
270	td->td_flags &= ~TDF_SINTR;
271	if (td->td_flags & TDF_TIMEOUT) {
272		td->td_flags &= ~TDF_TIMEOUT;
273		if (sig == 0)
274			rval = EWOULDBLOCK;
275	} else if (td->td_flags & TDF_TIMOFAIL) {
276		td->td_flags &= ~TDF_TIMOFAIL;
277	} else if (timo && callout_stop(&td->td_slpcallout) == 0) {
278		/*
279		 * This isn't supposed to be pretty.  If we are here, then
280		 * the endtsleep() callout is currently executing on another
281		 * CPU and is either spinning on the sched_lock or will be
282		 * soon.  If we don't synchronize here, there is a chance
283		 * that this process may msleep() again before the callout
284		 * has a chance to run and the callout may end up waking up
285		 * the wrong msleep().  Yuck.
286		 */
287		TD_SET_SLEEPING(td);
288		p->p_stats->p_ru.ru_nivcsw++;
289		mi_switch();
290		td->td_flags &= ~TDF_TIMOFAIL;
291	}
292	if ((td->td_flags & TDF_INTERRUPT) && (priority & PCATCH) &&
293	    (rval == 0)) {
294		td->td_flags &= ~TDF_INTERRUPT;
295		rval = EINTR;
296	}
297	mtx_unlock_spin(&sched_lock);
298
299	if (rval == 0 && catch) {
300		PROC_LOCK(p);
301		/* XXX: shouldn't we always be calling cursig() */
302		if (sig != 0 || (sig = cursig(td))) {
303			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
304				rval = EINTR;
305			else
306				rval = ERESTART;
307		}
308		PROC_UNLOCK(p);
309	}
310#ifdef KTRACE
311	if (KTRPOINT(td, KTR_CSW))
312		ktrcsw(0, 0);
313#endif
314	PICKUP_GIANT();
315	if (mtx != NULL) {
316		mtx_lock(mtx);
317		WITNESS_RESTORE(&mtx->mtx_object, mtx);
318	}
319	return (rval);
320}
321
322/*
323 * Implement timeout for msleep()
324 *
325 * If process hasn't been awakened (wchan non-zero),
326 * set timeout flag and undo the sleep.  If proc
327 * is stopped, just unsleep so it will remain stopped.
328 * MP-safe, called without the Giant mutex.
329 */
330static void
331endtsleep(arg)
332	void *arg;
333{
334	register struct thread *td = arg;
335
336	CTR3(KTR_PROC, "endtsleep: thread %p (pid %d, %s)",
337	    td, td->td_proc->p_pid, td->td_proc->p_comm);
338	mtx_lock_spin(&sched_lock);
339	/*
340	 * This is the other half of the synchronization with msleep()
341	 * described above.  If the TDS_TIMEOUT flag is set, we lost the
342	 * race and just need to put the process back on the runqueue.
343	 */
344	if (TD_ON_SLEEPQ(td)) {
345		TAILQ_REMOVE(&slpque[LOOKUP(td->td_wchan)], td, td_slpq);
346		TD_CLR_ON_SLEEPQ(td);
347		td->td_flags |= TDF_TIMEOUT;
348	} else {
349		td->td_flags |= TDF_TIMOFAIL;
350	}
351	TD_CLR_SLEEPING(td);
352	setrunnable(td);
353	mtx_unlock_spin(&sched_lock);
354}
355
356/*
357 * Abort a thread, as if an interrupt had occured.  Only abort
358 * interruptable waits (unfortunatly it isn't only safe to abort others).
359 * This is about identical to cv_abort().
360 * Think about merging them?
361 * Also, whatever the signal code does...
362 */
363void
364abortsleep(struct thread *td)
365{
366
367	mtx_assert(&sched_lock, MA_OWNED);
368	/*
369	 * If the TDF_TIMEOUT flag is set, just leave. A
370	 * timeout is scheduled anyhow.
371	 */
372	if ((td->td_flags & (TDF_TIMEOUT | TDF_SINTR)) == TDF_SINTR) {
373		if (TD_ON_SLEEPQ(td)) {
374			unsleep(td);
375			TD_CLR_SLEEPING(td);
376			setrunnable(td);
377		}
378	}
379}
380
381/*
382 * Remove a process from its wait queue
383 */
384void
385unsleep(struct thread *td)
386{
387
388	mtx_lock_spin(&sched_lock);
389	if (TD_ON_SLEEPQ(td)) {
390		TAILQ_REMOVE(&slpque[LOOKUP(td->td_wchan)], td, td_slpq);
391		TD_CLR_ON_SLEEPQ(td);
392	}
393	mtx_unlock_spin(&sched_lock);
394}
395
396/*
397 * Make all processes sleeping on the specified identifier runnable.
398 */
399void
400wakeup(ident)
401	register void *ident;
402{
403	register struct slpquehead *qp;
404	register struct thread *td;
405	struct thread *ntd;
406	struct proc *p;
407
408	mtx_lock_spin(&sched_lock);
409	qp = &slpque[LOOKUP(ident)];
410restart:
411	for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
412		ntd = TAILQ_NEXT(td, td_slpq);
413		if (td->td_wchan == ident) {
414			unsleep(td);
415			TD_CLR_SLEEPING(td);
416			setrunnable(td);
417			p = td->td_proc;
418			CTR3(KTR_PROC,"wakeup: thread %p (pid %d, %s)",
419			    td, p->p_pid, p->p_comm);
420			goto restart;
421		}
422	}
423	mtx_unlock_spin(&sched_lock);
424}
425
426/*
427 * Make a process sleeping on the specified identifier runnable.
428 * May wake more than one process if a target process is currently
429 * swapped out.
430 */
431void
432wakeup_one(ident)
433	register void *ident;
434{
435	register struct slpquehead *qp;
436	register struct thread *td;
437	register struct proc *p;
438	struct thread *ntd;
439
440	mtx_lock_spin(&sched_lock);
441	qp = &slpque[LOOKUP(ident)];
442	for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
443		ntd = TAILQ_NEXT(td, td_slpq);
444		if (td->td_wchan == ident) {
445			unsleep(td);
446			TD_CLR_SLEEPING(td);
447			setrunnable(td);
448			p = td->td_proc;
449			CTR3(KTR_PROC,"wakeup1: thread %p (pid %d, %s)",
450			    td, p->p_pid, p->p_comm);
451			break;
452		}
453	}
454	mtx_unlock_spin(&sched_lock);
455}
456
457/*
458 * The machine independent parts of mi_switch().
459 */
460void
461mi_switch(void)
462{
463	struct bintime new_switchtime;
464	struct thread *td = curthread;	/* XXX */
465	struct proc *p = td->td_proc;	/* XXX */
466	struct kse *ke = td->td_kse;
467	u_int sched_nest;
468
469	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
470
471	KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
472#ifdef INVARIANTS
473	if (!TD_ON_LOCK(td) &&
474	    !TD_ON_RUNQ(td) &&
475	    !TD_IS_RUNNING(td))
476		mtx_assert(&Giant, MA_NOTOWNED);
477#endif
478	KASSERT(td->td_critnest == 1,
479	    ("mi_switch: switch in a critical section"));
480
481	/*
482	 * Compute the amount of time during which the current
483	 * process was running, and add that to its total so far.
484	 */
485	binuptime(&new_switchtime);
486	bintime_add(&p->p_runtime, &new_switchtime);
487	bintime_sub(&p->p_runtime, PCPU_PTR(switchtime));
488
489#ifdef DDB
490	/*
491	 * Don't perform context switches from the debugger.
492	 */
493	if (db_active) {
494		mtx_unlock_spin(&sched_lock);
495		db_error("Context switches not allowed in the debugger.");
496	}
497#endif
498
499	/*
500	 * Check if the process exceeds its cpu resource allocation.  If
501	 * over max, arrange to kill the process in ast().
502	 */
503	if (p->p_cpulimit != RLIM_INFINITY &&
504	    p->p_runtime.sec > p->p_cpulimit) {
505		p->p_sflag |= PS_XCPU;
506		ke->ke_flags |= KEF_ASTPENDING;
507	}
508
509	/*
510	 * Finish up stats for outgoing thread.
511	 */
512	cnt.v_swtch++;
513	PCPU_SET(switchtime, new_switchtime);
514	CTR3(KTR_PROC, "mi_switch: old thread %p (pid %d, %s)", td, p->p_pid,
515	    p->p_comm);
516
517	sched_nest = sched_lock.mtx_recurse;
518	sched_switchout(td);
519
520	cpu_switch();		/* SHAZAM!!*/
521
522	sched_lock.mtx_recurse = sched_nest;
523	sched_lock.mtx_lock = (uintptr_t)td;
524	sched_switchin(td);
525
526	/*
527	 * Start setting up stats etc. for the incoming thread.
528	 * Similar code in fork_exit() is returned to by cpu_switch()
529	 * in the case of a new thread/process.
530	 */
531	CTR3(KTR_PROC, "mi_switch: new thread %p (pid %d, %s)", td, p->p_pid,
532	    p->p_comm);
533	if (PCPU_GET(switchtime.sec) == 0)
534		binuptime(PCPU_PTR(switchtime));
535	PCPU_SET(switchticks, ticks);
536
537	/*
538	 * Call the switchin function while still holding the scheduler lock
539	 * (used by the idlezero code and the general page-zeroing code)
540	 */
541	if (td->td_switchin)
542		td->td_switchin();
543
544	/*
545	 * If the last thread was exiting, finish cleaning it up.
546	 */
547	if ((td = PCPU_GET(deadthread))) {
548		PCPU_SET(deadthread, NULL);
549		thread_stash(td);
550	}
551}
552
553/*
554 * Change process state to be runnable,
555 * placing it on the run queue if it is in memory,
556 * and awakening the swapper if it isn't in memory.
557 */
558void
559setrunnable(struct thread *td)
560{
561	struct proc *p = td->td_proc;
562
563	mtx_assert(&sched_lock, MA_OWNED);
564	switch (p->p_state) {
565	case PRS_ZOMBIE:
566		panic("setrunnable(1)");
567	default:
568		break;
569	}
570	switch (td->td_state) {
571	case TDS_RUNNING:
572	case TDS_RUNQ:
573		return;
574	case TDS_INHIBITED:
575		/*
576		 * If we are only inhibited because we are swapped out
577		 * then arange to swap in this process. Otherwise just return.
578		 */
579		if (td->td_inhibitors != TDI_SWAPPED)
580			return;
581	case TDS_CAN_RUN:
582		break;
583	default:
584		printf("state is 0x%x", td->td_state);
585		panic("setrunnable(2)");
586	}
587	if ((p->p_sflag & PS_INMEM) == 0) {
588		if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
589			p->p_sflag |= PS_SWAPINREQ;
590			wakeup(&proc0);
591		}
592	} else
593		sched_wakeup(td);
594}
595
596/*
597 * Compute a tenex style load average of a quantity on
598 * 1, 5 and 15 minute intervals.
599 * XXXKSE   Needs complete rewrite when correct info is available.
600 * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
601 */
602static void
603loadav(void *arg)
604{
605	int i, nrun;
606	struct loadavg *avg;
607	struct proc *p;
608	struct thread *td;
609
610	avg = &averunnable;
611	sx_slock(&allproc_lock);
612	nrun = 0;
613	FOREACH_PROC_IN_SYSTEM(p) {
614		FOREACH_THREAD_IN_PROC(p, td) {
615			switch (td->td_state) {
616			case TDS_RUNQ:
617			case TDS_RUNNING:
618				if ((p->p_flag & P_NOLOAD) != 0)
619					goto nextproc;
620				nrun++; /* XXXKSE */
621			default:
622				break;
623			}
624nextproc:
625			continue;
626		}
627	}
628	sx_sunlock(&allproc_lock);
629	for (i = 0; i < 3; i++)
630		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
631		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
632
633	/*
634	 * Schedule the next update to occur after 5 seconds, but add a
635	 * random variation to avoid synchronisation with processes that
636	 * run at regular intervals.
637	 */
638	callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
639	    loadav, NULL);
640}
641
642static void
643lboltcb(void *arg)
644{
645	wakeup(&lbolt);
646	callout_reset(&lbolt_callout, hz, lboltcb, NULL);
647}
648
649/* ARGSUSED */
650static void
651sched_setup(dummy)
652	void *dummy;
653{
654	callout_init(&loadav_callout, 0);
655	callout_init(&lbolt_callout, 1);
656
657	/* Kick off timeout driven events by calling first time. */
658	loadav(NULL);
659	lboltcb(NULL);
660}
661
662/*
663 * General purpose yield system call
664 */
665int
666yield(struct thread *td, struct yield_args *uap)
667{
668	struct ksegrp *kg = td->td_ksegrp;
669
670	mtx_assert(&Giant, MA_NOTOWNED);
671	mtx_lock_spin(&sched_lock);
672	kg->kg_proc->p_stats->p_ru.ru_nvcsw++;
673	sched_prio(td, PRI_MAX_TIMESHARE);
674	mi_switch();
675	mtx_unlock_spin(&sched_lock);
676	td->td_retval[0] = 0;
677
678	return (0);
679}
680
681