xref: /freebsd/sys/kern/kern_synch.c (revision b52f49a9a0f22207ad5130ad8faba08de3ed23d8)
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  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_ktrace.h"
46 #ifdef __i386__
47 #include "opt_swtch.h"
48 #endif
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/condvar.h>
53 #include <sys/kernel.h>
54 #include <sys/ktr.h>
55 #include <sys/lock.h>
56 #include <sys/mutex.h>
57 #include <sys/proc.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/signalvar.h>
61 #include <sys/smp.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysproto.h>
65 #include <sys/vmmeter.h>
66 #ifdef DDB
67 #include <ddb/ddb.h>
68 #endif
69 #ifdef KTRACE
70 #include <sys/uio.h>
71 #include <sys/ktrace.h>
72 #endif
73 
74 #include <machine/cpu.h>
75 #ifdef SWTCH_OPTIM_STATS
76 #include <machine/md_var.h>
77 #endif
78 
79 static void sched_setup(void *dummy);
80 SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
81 
82 int	hogticks;
83 int	lbolt;
84 
85 static struct callout loadav_callout;
86 static struct callout lbolt_callout;
87 
88 struct loadavg averunnable =
89 	{ {0, 0, 0}, FSCALE };	/* load average, of runnable procs */
90 /*
91  * Constants for averages over 1, 5, and 15 minutes
92  * when sampling at 5 second intervals.
93  */
94 static fixpt_t cexp[3] = {
95 	0.9200444146293232 * FSCALE,	/* exp(-1/12) */
96 	0.9834714538216174 * FSCALE,	/* exp(-1/60) */
97 	0.9944598480048967 * FSCALE,	/* exp(-1/180) */
98 };
99 
100 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
101 static int      fscale __unused = FSCALE;
102 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
103 
104 static void	endtsleep(void *);
105 static void	loadav(void *arg);
106 static void	lboltcb(void *arg);
107 
108 /*
109  * We're only looking at 7 bits of the address; everything is
110  * aligned to 4, lots of things are aligned to greater powers
111  * of 2.  Shift right by 8, i.e. drop the bottom 256 worth.
112  */
113 #define TABLESIZE	128
114 static TAILQ_HEAD(slpquehead, thread) slpque[TABLESIZE];
115 #define LOOKUP(x)	(((intptr_t)(x) >> 8) & (TABLESIZE - 1))
116 
117 void
118 sleepinit(void)
119 {
120 	int i;
121 
122 	hogticks = (hz / 10) * 2;	/* Default only. */
123 	for (i = 0; i < TABLESIZE; i++)
124 		TAILQ_INIT(&slpque[i]);
125 }
126 
127 /*
128  * General sleep call.  Suspends the current process until a wakeup is
129  * performed on the specified identifier.  The process will then be made
130  * runnable with the specified priority.  Sleeps at most timo/hz seconds
131  * (0 means no timeout).  If pri includes PCATCH flag, signals are checked
132  * before and after sleeping, else signals are not checked.  Returns 0 if
133  * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
134  * signal needs to be delivered, ERESTART is returned if the current system
135  * call should be restarted if possible, and EINTR is returned if the system
136  * call should be interrupted by the signal (return EINTR).
137  *
138  * The mutex argument is exited before the caller is suspended, and
139  * entered before msleep returns.  If priority includes the PDROP
140  * flag the mutex is not entered before returning.
141  */
142 
143 int
144 msleep(ident, mtx, priority, wmesg, timo)
145 	void *ident;
146 	struct mtx *mtx;
147 	int priority, timo;
148 	const char *wmesg;
149 {
150 	struct thread *td = curthread;
151 	struct proc *p = td->td_proc;
152 	int sig, catch = priority & PCATCH;
153 	int rval = 0;
154 	WITNESS_SAVE_DECL(mtx);
155 
156 #ifdef KTRACE
157 	if (KTRPOINT(td, KTR_CSW))
158 		ktrcsw(1, 0);
159 #endif
160 	/* XXX: mtx == NULL ?? */
161 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, &mtx->mtx_object,
162 	    "Sleeping on \"%s\"", wmesg);
163 	KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL,
164 	    ("sleeping without a mutex"));
165 	/*
166 	 * If we are capable of async syscalls and there isn't already
167 	 * another one ready to return, start a new thread
168 	 * and queue it as ready to run. Note that there is danger here
169 	 * because we need to make sure that we don't sleep allocating
170 	 * the thread (recursion here might be bad).
171 	 */
172 	mtx_lock_spin(&sched_lock);
173 	if (p->p_flag & P_SA || p->p_numthreads > 1) {
174 		/*
175 		 * Just don't bother if we are exiting
176 		 * and not the exiting thread or thread was marked as
177 		 * interrupted.
178 		 */
179 		if (catch &&
180 		    (((p->p_flag & P_WEXIT) && (p->p_singlethread != td)) ||
181 		     (td->td_flags & TDF_INTERRUPT))) {
182 			td->td_flags &= ~TDF_INTERRUPT;
183 			mtx_unlock_spin(&sched_lock);
184 			return (EINTR);
185 		}
186 	}
187 	if (cold ) {
188 		/*
189 		 * During autoconfiguration, just give interrupts
190 		 * a chance, then just return.
191 		 * Don't run any other procs or panic below,
192 		 * in case this is the idle process and already asleep.
193 		 */
194 		if (mtx != NULL && priority & PDROP)
195 			mtx_unlock(mtx);
196 		mtx_unlock_spin(&sched_lock);
197 		return (0);
198 	}
199 
200 	DROP_GIANT();
201 
202 	if (mtx != NULL) {
203 		mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
204 		WITNESS_SAVE(&mtx->mtx_object, mtx);
205 		mtx_unlock(mtx);
206 		if (priority & PDROP)
207 			mtx = NULL;
208 	}
209 
210 	KASSERT(p != NULL, ("msleep1"));
211 	KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
212 
213 	CTR5(KTR_PROC, "msleep: thread %p (pid %d, %s) on %s (%p)",
214 	    td, p->p_pid, p->p_comm, wmesg, ident);
215 
216 	td->td_wchan = ident;
217 	td->td_wmesg = wmesg;
218 	TAILQ_INSERT_TAIL(&slpque[LOOKUP(ident)], td, td_slpq);
219 	TD_SET_ON_SLEEPQ(td);
220 	if (timo)
221 		callout_reset(&td->td_slpcallout, timo, endtsleep, td);
222 	/*
223 	 * We put ourselves on the sleep queue and start our timeout
224 	 * before calling thread_suspend_check, as we could stop there, and
225 	 * a wakeup or a SIGCONT (or both) could occur while we were stopped.
226 	 * without resuming us, thus we must be ready for sleep
227 	 * when cursig is called.  If the wakeup happens while we're
228 	 * stopped, td->td_wchan will be 0 upon return from cursig.
229 	 */
230 	if (catch) {
231 		CTR3(KTR_PROC, "msleep caught: thread %p (pid %d, %s)", td,
232 		    p->p_pid, p->p_comm);
233 		td->td_flags |= TDF_SINTR;
234 		mtx_unlock_spin(&sched_lock);
235 		PROC_LOCK(p);
236 		mtx_lock(&p->p_sigacts->ps_mtx);
237 		sig = cursig(td);
238 		mtx_unlock(&p->p_sigacts->ps_mtx);
239 		if (sig == 0 && thread_suspend_check(1))
240 			sig = SIGSTOP;
241 		mtx_lock_spin(&sched_lock);
242 		PROC_UNLOCK(p);
243 		if (sig != 0) {
244 			if (TD_ON_SLEEPQ(td))
245 				unsleep(td);
246 		} else if (!TD_ON_SLEEPQ(td))
247 			catch = 0;
248 	} else
249 		sig = 0;
250 
251 	/*
252 	 * Let the scheduler know we're about to voluntarily go to sleep.
253 	 */
254 	sched_sleep(td, priority & PRIMASK);
255 
256 	if (TD_ON_SLEEPQ(td)) {
257 		p->p_stats->p_ru.ru_nvcsw++;
258 		TD_SET_SLEEPING(td);
259 		mi_switch();
260 	}
261 	/*
262 	 * We're awake from voluntary sleep.
263 	 */
264 	CTR3(KTR_PROC, "msleep resume: thread %p (pid %d, %s)", td, p->p_pid,
265 	    p->p_comm);
266 	KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING"));
267 	td->td_flags &= ~TDF_SINTR;
268 	if (td->td_flags & TDF_TIMEOUT) {
269 		td->td_flags &= ~TDF_TIMEOUT;
270 		if (sig == 0)
271 			rval = EWOULDBLOCK;
272 	} else if (td->td_flags & TDF_TIMOFAIL) {
273 		td->td_flags &= ~TDF_TIMOFAIL;
274 	} else if (timo && callout_stop(&td->td_slpcallout) == 0) {
275 		/*
276 		 * This isn't supposed to be pretty.  If we are here, then
277 		 * the endtsleep() callout is currently executing on another
278 		 * CPU and is either spinning on the sched_lock or will be
279 		 * soon.  If we don't synchronize here, there is a chance
280 		 * that this process may msleep() again before the callout
281 		 * has a chance to run and the callout may end up waking up
282 		 * the wrong msleep().  Yuck.
283 		 */
284 		TD_SET_SLEEPING(td);
285 		p->p_stats->p_ru.ru_nivcsw++;
286 		mi_switch();
287 		td->td_flags &= ~TDF_TIMOFAIL;
288 	}
289 	if ((td->td_flags & TDF_INTERRUPT) && (priority & PCATCH) &&
290 	    (rval == 0)) {
291 		td->td_flags &= ~TDF_INTERRUPT;
292 		rval = EINTR;
293 	}
294 	mtx_unlock_spin(&sched_lock);
295 
296 	if (rval == 0 && catch) {
297 		PROC_LOCK(p);
298 		/* XXX: shouldn't we always be calling cursig() */
299 		mtx_lock(&p->p_sigacts->ps_mtx);
300 		if (sig != 0 || (sig = cursig(td))) {
301 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
302 				rval = EINTR;
303 			else
304 				rval = ERESTART;
305 		}
306 		mtx_unlock(&p->p_sigacts->ps_mtx);
307 		PROC_UNLOCK(p);
308 	}
309 #ifdef KTRACE
310 	if (KTRPOINT(td, KTR_CSW))
311 		ktrcsw(0, 0);
312 #endif
313 	PICKUP_GIANT();
314 	if (mtx != NULL) {
315 		mtx_lock(mtx);
316 		WITNESS_RESTORE(&mtx->mtx_object, mtx);
317 	}
318 	return (rval);
319 }
320 
321 /*
322  * Implement timeout for msleep()
323  *
324  * If process hasn't been awakened (wchan non-zero),
325  * set timeout flag and undo the sleep.  If proc
326  * is stopped, just unsleep so it will remain stopped.
327  * MP-safe, called without the Giant mutex.
328  */
329 static void
330 endtsleep(arg)
331 	void *arg;
332 {
333 	register struct thread *td = arg;
334 
335 	CTR3(KTR_PROC, "endtsleep: thread %p (pid %d, %s)",
336 	    td, td->td_proc->p_pid, td->td_proc->p_comm);
337 	mtx_lock_spin(&sched_lock);
338 	/*
339 	 * This is the other half of the synchronization with msleep()
340 	 * described above.  If the TDS_TIMEOUT flag is set, we lost the
341 	 * race and just need to put the process back on the runqueue.
342 	 */
343 	if (TD_ON_SLEEPQ(td)) {
344 		TAILQ_REMOVE(&slpque[LOOKUP(td->td_wchan)], td, td_slpq);
345 		TD_CLR_ON_SLEEPQ(td);
346 		td->td_flags |= TDF_TIMEOUT;
347 		td->td_wmesg = NULL;
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  */
363 void
364 abortsleep(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  */
384 void
385 unsleep(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 		td->td_wmesg = NULL;
393 	}
394 	mtx_unlock_spin(&sched_lock);
395 }
396 
397 /*
398  * Make all processes sleeping on the specified identifier runnable.
399  */
400 void
401 wakeup(ident)
402 	register void *ident;
403 {
404 	register struct slpquehead *qp;
405 	register struct thread *td;
406 	struct thread *ntd;
407 	struct proc *p;
408 
409 	mtx_lock_spin(&sched_lock);
410 	qp = &slpque[LOOKUP(ident)];
411 restart:
412 	for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
413 		ntd = TAILQ_NEXT(td, td_slpq);
414 		if (td->td_wchan == ident) {
415 			unsleep(td);
416 			TD_CLR_SLEEPING(td);
417 			setrunnable(td);
418 			p = td->td_proc;
419 			CTR3(KTR_PROC,"wakeup: thread %p (pid %d, %s)",
420 			    td, p->p_pid, p->p_comm);
421 			goto restart;
422 		}
423 	}
424 	mtx_unlock_spin(&sched_lock);
425 }
426 
427 /*
428  * Make a process sleeping on the specified identifier runnable.
429  * May wake more than one process if a target process is currently
430  * swapped out.
431  */
432 void
433 wakeup_one(ident)
434 	register void *ident;
435 {
436 	register struct slpquehead *qp;
437 	register struct thread *td;
438 	register struct proc *p;
439 	struct thread *ntd;
440 
441 	mtx_lock_spin(&sched_lock);
442 	qp = &slpque[LOOKUP(ident)];
443 	for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
444 		ntd = TAILQ_NEXT(td, td_slpq);
445 		if (td->td_wchan == ident) {
446 			unsleep(td);
447 			TD_CLR_SLEEPING(td);
448 			setrunnable(td);
449 			p = td->td_proc;
450 			CTR3(KTR_PROC,"wakeup1: thread %p (pid %d, %s)",
451 			    td, p->p_pid, p->p_comm);
452 			break;
453 		}
454 	}
455 	mtx_unlock_spin(&sched_lock);
456 }
457 
458 /*
459  * The machine independent parts of mi_switch().
460  */
461 void
462 mi_switch(void)
463 {
464 	struct bintime new_switchtime;
465 	struct thread *td;
466 #if !defined(__alpha__) && !defined(__powerpc__)
467 	struct thread *newtd;
468 #endif
469 	struct proc *p;
470 	u_int sched_nest;
471 
472 	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
473 	td = curthread;			/* XXX */
474 	p = td->td_proc;		/* XXX */
475 	KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
476 #ifdef INVARIANTS
477 	if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
478 		mtx_assert(&Giant, MA_NOTOWNED);
479 #endif
480 	KASSERT(td->td_critnest == 1,
481 	    ("mi_switch: switch in a critical section"));
482 
483 	/*
484 	 * Compute the amount of time during which the current
485 	 * process was running, and add that to its total so far.
486 	 */
487 	binuptime(&new_switchtime);
488 	bintime_add(&p->p_runtime, &new_switchtime);
489 	bintime_sub(&p->p_runtime, PCPU_PTR(switchtime));
490 
491 #ifdef DDB
492 	/*
493 	 * Don't perform context switches from the debugger.
494 	 */
495 	if (db_active) {
496 		mtx_unlock_spin(&sched_lock);
497 		db_print_backtrace();
498 		db_error("Context switches not allowed in the debugger.");
499 	}
500 #endif
501 
502 	/*
503 	 * Check if the process exceeds its cpu resource allocation.  If
504 	 * over max, arrange to kill the process in ast().
505 	 */
506 	if (p->p_cpulimit != RLIM_INFINITY &&
507 	    p->p_runtime.sec > p->p_cpulimit) {
508 		p->p_sflag |= PS_XCPU;
509 		td->td_flags |= TDF_ASTPENDING;
510 	}
511 
512 	/*
513 	 * Finish up stats for outgoing thread.
514 	 */
515 	cnt.v_swtch++;
516 	PCPU_SET(switchtime, new_switchtime);
517 	CTR3(KTR_PROC, "mi_switch: old thread %p (pid %d, %s)", td, p->p_pid,
518 	    p->p_comm);
519 	sched_nest = sched_lock.mtx_recurse;
520 	if (td->td_proc->p_flag & P_SA)
521 		thread_switchout(td);
522 	sched_switchout(td);
523 
524 #if !defined(__alpha__) && !defined(__powerpc__)
525 	newtd = choosethread();
526 	if (td != newtd)
527 		cpu_switch(td, newtd);	/* SHAZAM!! */
528 #ifdef SWTCH_OPTIM_STATS
529 	else
530 		stupid_switch++;
531 #endif
532 #else
533 	cpu_switch();		/* SHAZAM!!*/
534 #endif
535 
536 	sched_lock.mtx_recurse = sched_nest;
537 	sched_lock.mtx_lock = (uintptr_t)td;
538 	sched_switchin(td);
539 
540 	/*
541 	 * Start setting up stats etc. for the incoming thread.
542 	 * Similar code in fork_exit() is returned to by cpu_switch()
543 	 * in the case of a new thread/process.
544 	 */
545 	CTR3(KTR_PROC, "mi_switch: new thread %p (pid %d, %s)", td, p->p_pid,
546 	    p->p_comm);
547 	if (PCPU_GET(switchtime.sec) == 0)
548 		binuptime(PCPU_PTR(switchtime));
549 	PCPU_SET(switchticks, ticks);
550 
551 	/*
552 	 * Call the switchin function while still holding the scheduler lock
553 	 * (used by the idlezero code and the general page-zeroing code)
554 	 */
555 	if (td->td_switchin)
556 		td->td_switchin();
557 
558 	/*
559 	 * If the last thread was exiting, finish cleaning it up.
560 	 */
561 	if ((td = PCPU_GET(deadthread))) {
562 		PCPU_SET(deadthread, NULL);
563 		thread_stash(td);
564 	}
565 }
566 
567 /*
568  * Change process state to be runnable,
569  * placing it on the run queue if it is in memory,
570  * and awakening the swapper if it isn't in memory.
571  */
572 void
573 setrunnable(struct thread *td)
574 {
575 	struct proc *p = td->td_proc;
576 
577 	mtx_assert(&sched_lock, MA_OWNED);
578 	switch (p->p_state) {
579 	case PRS_ZOMBIE:
580 		panic("setrunnable(1)");
581 	default:
582 		break;
583 	}
584 	switch (td->td_state) {
585 	case TDS_RUNNING:
586 	case TDS_RUNQ:
587 		return;
588 	case TDS_INHIBITED:
589 		/*
590 		 * If we are only inhibited because we are swapped out
591 		 * then arange to swap in this process. Otherwise just return.
592 		 */
593 		if (td->td_inhibitors != TDI_SWAPPED)
594 			return;
595 		/* XXX: intentional fall-through ? */
596 	case TDS_CAN_RUN:
597 		break;
598 	default:
599 		printf("state is 0x%x", td->td_state);
600 		panic("setrunnable(2)");
601 	}
602 	if ((p->p_sflag & PS_INMEM) == 0) {
603 		if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
604 			p->p_sflag |= PS_SWAPINREQ;
605 			wakeup(&proc0);
606 		}
607 	} else
608 		sched_wakeup(td);
609 }
610 
611 /*
612  * Compute a tenex style load average of a quantity on
613  * 1, 5 and 15 minute intervals.
614  * XXXKSE   Needs complete rewrite when correct info is available.
615  * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
616  */
617 static void
618 loadav(void *arg)
619 {
620 	int i, nrun;
621 	struct loadavg *avg;
622 	struct proc *p;
623 	struct thread *td;
624 
625 	avg = &averunnable;
626 	sx_slock(&allproc_lock);
627 	nrun = 0;
628 	FOREACH_PROC_IN_SYSTEM(p) {
629 		FOREACH_THREAD_IN_PROC(p, td) {
630 			switch (td->td_state) {
631 			case TDS_RUNQ:
632 			case TDS_RUNNING:
633 				if ((p->p_flag & P_NOLOAD) != 0)
634 					goto nextproc;
635 				nrun++; /* XXXKSE */
636 			default:
637 				break;
638 			}
639 nextproc:
640 			continue;
641 		}
642 	}
643 	sx_sunlock(&allproc_lock);
644 	for (i = 0; i < 3; i++)
645 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
646 		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
647 
648 	/*
649 	 * Schedule the next update to occur after 5 seconds, but add a
650 	 * random variation to avoid synchronisation with processes that
651 	 * run at regular intervals.
652 	 */
653 	callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
654 	    loadav, NULL);
655 }
656 
657 static void
658 lboltcb(void *arg)
659 {
660 	wakeup(&lbolt);
661 	callout_reset(&lbolt_callout, hz, lboltcb, NULL);
662 }
663 
664 /* ARGSUSED */
665 static void
666 sched_setup(dummy)
667 	void *dummy;
668 {
669 	callout_init(&loadav_callout, 0);
670 	callout_init(&lbolt_callout, 1);
671 
672 	/* Kick off timeout driven events by calling first time. */
673 	loadav(NULL);
674 	lboltcb(NULL);
675 }
676 
677 /*
678  * General purpose yield system call
679  */
680 int
681 yield(struct thread *td, struct yield_args *uap)
682 {
683 	struct ksegrp *kg = td->td_ksegrp;
684 
685 	mtx_assert(&Giant, MA_NOTOWNED);
686 	mtx_lock_spin(&sched_lock);
687 	kg->kg_proc->p_stats->p_ru.ru_nvcsw++;
688 	sched_prio(td, PRI_MAX_TIMESHARE);
689 	mi_switch();
690 	mtx_unlock_spin(&sched_lock);
691 	td->td_retval[0] = 0;
692 
693 	return (0);
694 }
695 
696