xref: /netbsd-src/sys/kern/kern_synch.c (revision ce666bb8ce77792a3948ca697c2fdad578a542a7)
1 /*	$NetBSD: kern_synch.c,v 1.158 2005/12/24 19:12:23 perry Exp $	*/
2 
3 /*-
4  * Copyright (c) 1999, 2000, 2004 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  * This code is derived from software contributed to The NetBSD Foundation
11  * by Charles M. Hannum.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *	This product includes software developed by the NetBSD
24  *	Foundation, Inc. and its contributors.
25  * 4. Neither the name of The NetBSD Foundation nor the names of its
26  *    contributors may be used to endorse or promote products derived
27  *    from this software without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
30  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
31  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
32  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
33  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
34  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
35  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
36  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
37  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
38  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
39  * POSSIBILITY OF SUCH DAMAGE.
40  */
41 
42 /*-
43  * Copyright (c) 1982, 1986, 1990, 1991, 1993
44  *	The Regents of the University of California.  All rights reserved.
45  * (c) UNIX System Laboratories, Inc.
46  * All or some portions of this file are derived from material licensed
47  * to the University of California by American Telephone and Telegraph
48  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
49  * the permission of UNIX System Laboratories, Inc.
50  *
51  * Redistribution and use in source and binary forms, with or without
52  * modification, are permitted provided that the following conditions
53  * are met:
54  * 1. Redistributions of source code must retain the above copyright
55  *    notice, this list of conditions and the following disclaimer.
56  * 2. Redistributions in binary form must reproduce the above copyright
57  *    notice, this list of conditions and the following disclaimer in the
58  *    documentation and/or other materials provided with the distribution.
59  * 3. Neither the name of the University nor the names of its contributors
60  *    may be used to endorse or promote products derived from this software
61  *    without specific prior written permission.
62  *
63  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
65  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
66  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
67  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
68  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
69  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
70  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
71  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
72  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
73  * SUCH DAMAGE.
74  *
75  *	@(#)kern_synch.c	8.9 (Berkeley) 5/19/95
76  */
77 
78 #include <sys/cdefs.h>
79 __KERNEL_RCSID(0, "$NetBSD: kern_synch.c,v 1.158 2005/12/24 19:12:23 perry Exp $");
80 
81 #include "opt_ddb.h"
82 #include "opt_ktrace.h"
83 #include "opt_kstack.h"
84 #include "opt_lockdebug.h"
85 #include "opt_multiprocessor.h"
86 #include "opt_perfctrs.h"
87 
88 #include <sys/param.h>
89 #include <sys/systm.h>
90 #include <sys/callout.h>
91 #include <sys/proc.h>
92 #include <sys/kernel.h>
93 #include <sys/buf.h>
94 #if defined(PERFCTRS)
95 #include <sys/pmc.h>
96 #endif
97 #include <sys/signalvar.h>
98 #include <sys/resourcevar.h>
99 #include <sys/sched.h>
100 #include <sys/sa.h>
101 #include <sys/savar.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 #ifdef KTRACE
106 #include <sys/ktrace.h>
107 #endif
108 
109 #include <machine/cpu.h>
110 
111 int	lbolt;			/* once a second sleep address */
112 int	rrticks;		/* number of hardclock ticks per roundrobin() */
113 
114 /*
115  * Sleep queues.
116  *
117  * We're only looking at 7 bits of the address; everything is
118  * aligned to 4, lots of things are aligned to greater powers
119  * of 2.  Shift right by 8, i.e. drop the bottom 256 worth.
120  */
121 #define	SLPQUE_TABLESIZE	128
122 #define	SLPQUE_LOOKUP(x)	(((u_long)(x) >> 8) & (SLPQUE_TABLESIZE - 1))
123 
124 #define	SLPQUE(ident)	(&sched_slpque[SLPQUE_LOOKUP(ident)])
125 
126 /*
127  * The global scheduler state.
128  */
129 struct prochd sched_qs[RUNQUE_NQS];	/* run queues */
130 volatile u_int32_t sched_whichqs;	/* bitmap of non-empty queues */
131 struct slpque sched_slpque[SLPQUE_TABLESIZE]; /* sleep queues */
132 
133 struct simplelock sched_lock = SIMPLELOCK_INITIALIZER;
134 
135 void schedcpu(void *);
136 void updatepri(struct lwp *);
137 void endtsleep(void *);
138 
139 inline void sa_awaken(struct lwp *);
140 inline void awaken(struct lwp *);
141 
142 struct callout schedcpu_ch = CALLOUT_INITIALIZER_SETFUNC(schedcpu, NULL);
143 static unsigned int schedcpu_ticks;
144 
145 
146 /*
147  * Force switch among equal priority processes every 100ms.
148  * Called from hardclock every hz/10 == rrticks hardclock ticks.
149  */
150 /* ARGSUSED */
151 void
152 roundrobin(struct cpu_info *ci)
153 {
154 	struct schedstate_percpu *spc = &ci->ci_schedstate;
155 
156 	spc->spc_rrticks = rrticks;
157 
158 	if (curlwp != NULL) {
159 		if (spc->spc_flags & SPCF_SEENRR) {
160 			/*
161 			 * The process has already been through a roundrobin
162 			 * without switching and may be hogging the CPU.
163 			 * Indicate that the process should yield.
164 			 */
165 			spc->spc_flags |= SPCF_SHOULDYIELD;
166 		} else
167 			spc->spc_flags |= SPCF_SEENRR;
168 	}
169 	need_resched(curcpu());
170 }
171 
172 #define	PPQ	(128 / RUNQUE_NQS)	/* priorities per queue */
173 #define	NICE_WEIGHT 2			/* priorities per nice level */
174 
175 #define	ESTCPU_SHIFT	11
176 #define	ESTCPU_MAX	((NICE_WEIGHT * PRIO_MAX - PPQ) << ESTCPU_SHIFT)
177 #define	ESTCPULIM(e)	min((e), ESTCPU_MAX)
178 
179 /*
180  * Constants for digital decay and forget:
181  *	90% of (p_estcpu) usage in 5 * loadav time
182  *	95% of (p_pctcpu) usage in 60 seconds (load insensitive)
183  *          Note that, as ps(1) mentions, this can let percentages
184  *          total over 100% (I've seen 137.9% for 3 processes).
185  *
186  * Note that hardclock updates p_estcpu and p_cpticks independently.
187  *
188  * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds.
189  * That is, the system wants to compute a value of decay such
190  * that the following for loop:
191  * 	for (i = 0; i < (5 * loadavg); i++)
192  * 		p_estcpu *= decay;
193  * will compute
194  * 	p_estcpu *= 0.1;
195  * for all values of loadavg:
196  *
197  * Mathematically this loop can be expressed by saying:
198  * 	decay ** (5 * loadavg) ~= .1
199  *
200  * The system computes decay as:
201  * 	decay = (2 * loadavg) / (2 * loadavg + 1)
202  *
203  * We wish to prove that the system's computation of decay
204  * will always fulfill the equation:
205  * 	decay ** (5 * loadavg) ~= .1
206  *
207  * If we compute b as:
208  * 	b = 2 * loadavg
209  * then
210  * 	decay = b / (b + 1)
211  *
212  * We now need to prove two things:
213  *	1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1)
214  *	2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg)
215  *
216  * Facts:
217  *         For x close to zero, exp(x) =~ 1 + x, since
218  *              exp(x) = 0! + x**1/1! + x**2/2! + ... .
219  *              therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b.
220  *         For x close to zero, ln(1+x) =~ x, since
221  *              ln(1+x) = x - x**2/2 + x**3/3 - ...     -1 < x < 1
222  *              therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1).
223  *         ln(.1) =~ -2.30
224  *
225  * Proof of (1):
226  *    Solve (factor)**(power) =~ .1 given power (5*loadav):
227  *	solving for factor,
228  *      ln(factor) =~ (-2.30/5*loadav), or
229  *      factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) =
230  *          exp(-1/b) =~ (b-1)/b =~ b/(b+1).                    QED
231  *
232  * Proof of (2):
233  *    Solve (factor)**(power) =~ .1 given factor == (b/(b+1)):
234  *	solving for power,
235  *      power*ln(b/(b+1)) =~ -2.30, or
236  *      power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav.  QED
237  *
238  * Actual power values for the implemented algorithm are as follows:
239  *      loadav: 1       2       3       4
240  *      power:  5.68    10.32   14.94   19.55
241  */
242 
243 /* calculations for digital decay to forget 90% of usage in 5*loadav sec */
244 #define	loadfactor(loadav)	(2 * (loadav))
245 
246 static fixpt_t
247 decay_cpu(fixpt_t loadfac, fixpt_t estcpu)
248 {
249 
250 	if (estcpu == 0) {
251 		return 0;
252 	}
253 
254 #if !defined(_LP64)
255 	/* avoid 64bit arithmetics. */
256 #define	FIXPT_MAX ((fixpt_t)((UINTMAX_C(1) << sizeof(fixpt_t) * CHAR_BIT) - 1))
257 	if (__predict_true(loadfac <= FIXPT_MAX / ESTCPU_MAX)) {
258 		return estcpu * loadfac / (loadfac + FSCALE);
259 	}
260 #endif /* !defined(_LP64) */
261 
262 	return (uint64_t)estcpu * loadfac / (loadfac + FSCALE);
263 }
264 
265 /*
266  * For all load averages >= 1 and max p_estcpu of (255 << ESTCPU_SHIFT),
267  * sleeping for at least seven times the loadfactor will decay p_estcpu to
268  * less than (1 << ESTCPU_SHIFT).
269  *
270  * note that our ESTCPU_MAX is actually much smaller than (255 << ESTCPU_SHIFT).
271  */
272 static fixpt_t
273 decay_cpu_batch(fixpt_t loadfac, fixpt_t estcpu, unsigned int n)
274 {
275 
276 	if ((n << FSHIFT) >= 7 * loadfac) {
277 		return 0;
278 	}
279 
280 	while (estcpu != 0 && n > 1) {
281 		estcpu = decay_cpu(loadfac, estcpu);
282 		n--;
283 	}
284 
285 	return estcpu;
286 }
287 
288 /* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
289 fixpt_t	ccpu = 0.95122942450071400909 * FSCALE;		/* exp(-1/20) */
290 
291 /*
292  * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
293  * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
294  * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
295  *
296  * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
297  *	1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
298  *
299  * If you dont want to bother with the faster/more-accurate formula, you
300  * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
301  * (more general) method of calculating the %age of CPU used by a process.
302  */
303 #define	CCPU_SHIFT	11
304 
305 /*
306  * Recompute process priorities, every hz ticks.
307  */
308 /* ARGSUSED */
309 void
310 schedcpu(void *arg)
311 {
312 	fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
313 	struct lwp *l;
314 	struct proc *p;
315 	int s, minslp;
316 	int clkhz;
317 
318 	schedcpu_ticks++;
319 
320 	proclist_lock_read();
321 	PROCLIST_FOREACH(p, &allproc) {
322 		/*
323 		 * Increment time in/out of memory and sleep time
324 		 * (if sleeping).  We ignore overflow; with 16-bit int's
325 		 * (remember them?) overflow takes 45 days.
326 		 */
327 		minslp = 2;
328 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
329 			l->l_swtime++;
330 			if (l->l_stat == LSSLEEP || l->l_stat == LSSTOP ||
331 			    l->l_stat == LSSUSPENDED) {
332 				l->l_slptime++;
333 				minslp = min(minslp, l->l_slptime);
334 			} else
335 				minslp = 0;
336 		}
337 		p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
338 		/*
339 		 * If the process has slept the entire second,
340 		 * stop recalculating its priority until it wakes up.
341 		 */
342 		if (minslp > 1)
343 			continue;
344 		s = splstatclock();	/* prevent state changes */
345 		/*
346 		 * p_pctcpu is only for ps.
347 		 */
348 		clkhz = stathz != 0 ? stathz : hz;
349 #if	(FSHIFT >= CCPU_SHIFT)
350 		p->p_pctcpu += (clkhz == 100)?
351 			((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT):
352                 	100 * (((fixpt_t) p->p_cpticks)
353 				<< (FSHIFT - CCPU_SHIFT)) / clkhz;
354 #else
355 		p->p_pctcpu += ((FSCALE - ccpu) *
356 			(p->p_cpticks * FSCALE / clkhz)) >> FSHIFT;
357 #endif
358 		p->p_cpticks = 0;
359 		p->p_estcpu = decay_cpu(loadfac, p->p_estcpu);
360 		splx(s);	/* Done with the process CPU ticks update */
361 		SCHED_LOCK(s);
362 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
363 			if (l->l_slptime > 1)
364 				continue;
365 			resetpriority(l);
366 			if (l->l_priority >= PUSER) {
367 				if (l->l_stat == LSRUN &&
368 				    (l->l_flag & L_INMEM) &&
369 				    (l->l_priority / PPQ) != (l->l_usrpri / PPQ)) {
370 					remrunqueue(l);
371 					l->l_priority = l->l_usrpri;
372 					setrunqueue(l);
373 				} else
374 					l->l_priority = l->l_usrpri;
375 			}
376 		}
377 		SCHED_UNLOCK(s);
378 	}
379 	proclist_unlock_read();
380 	uvm_meter();
381 	wakeup((caddr_t)&lbolt);
382 	callout_schedule(&schedcpu_ch, hz);
383 }
384 
385 /*
386  * Recalculate the priority of a process after it has slept for a while.
387  */
388 void
389 updatepri(struct lwp *l)
390 {
391 	struct proc *p = l->l_proc;
392 	fixpt_t loadfac;
393 
394 	SCHED_ASSERT_LOCKED();
395 	KASSERT(l->l_slptime > 1);
396 
397 	loadfac = loadfactor(averunnable.ldavg[0]);
398 
399 	l->l_slptime--; /* the first time was done in schedcpu */
400 	/* XXX NJWLWP */
401 	p->p_estcpu = decay_cpu_batch(loadfac, p->p_estcpu, l->l_slptime);
402 	resetpriority(l);
403 }
404 
405 /*
406  * During autoconfiguration or after a panic, a sleep will simply
407  * lower the priority briefly to allow interrupts, then return.
408  * The priority to be used (safepri) is machine-dependent, thus this
409  * value is initialized and maintained in the machine-dependent layers.
410  * This priority will typically be 0, or the lowest priority
411  * that is safe for use on the interrupt stack; it can be made
412  * higher to block network software interrupts after panics.
413  */
414 int safepri;
415 
416 /*
417  * General sleep call.  Suspends the current process until a wakeup is
418  * performed on the specified identifier.  The process will then be made
419  * runnable with the specified priority.  Sleeps at most timo/hz seconds
420  * (0 means no timeout).  If pri includes PCATCH flag, signals are checked
421  * before and after sleeping, else signals are not checked.  Returns 0 if
422  * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
423  * signal needs to be delivered, ERESTART is returned if the current system
424  * call should be restarted if possible, and EINTR is returned if the system
425  * call should be interrupted by the signal (return EINTR).
426  *
427  * The interlock is held until the scheduler_slock is acquired.  The
428  * interlock will be locked before returning back to the caller
429  * unless the PNORELOCK flag is specified, in which case the
430  * interlock will always be unlocked upon return.
431  */
432 int
433 ltsleep(volatile const void *ident, int priority, const char *wmesg, int timo,
434     volatile struct simplelock *interlock)
435 {
436 	struct lwp *l = curlwp;
437 	struct proc *p = l ? l->l_proc : NULL;
438 	struct slpque *qp;
439 	struct sadata_upcall *sau;
440 	int sig, s;
441 	int catch = priority & PCATCH;
442 	int relock = (priority & PNORELOCK) == 0;
443 	int exiterr = (priority & PNOEXITERR) == 0;
444 
445 	/*
446 	 * XXXSMP
447 	 * This is probably bogus.  Figure out what the right
448 	 * thing to do here really is.
449 	 * Note that not sleeping if ltsleep is called with curlwp == NULL
450 	 * in the shutdown case is disgusting but partly necessary given
451 	 * how shutdown (barely) works.
452 	 */
453 	if (cold || (doing_shutdown && (panicstr || (l == NULL)))) {
454 		/*
455 		 * After a panic, or during autoconfiguration,
456 		 * just give interrupts a chance, then just return;
457 		 * don't run any other procs or panic below,
458 		 * in case this is the idle process and already asleep.
459 		 */
460 		s = splhigh();
461 		splx(safepri);
462 		splx(s);
463 		if (interlock != NULL && relock == 0)
464 			simple_unlock(interlock);
465 		return (0);
466 	}
467 
468 	KASSERT(p != NULL);
469 	LOCK_ASSERT(interlock == NULL || simple_lock_held(interlock));
470 
471 #ifdef KTRACE
472 	if (KTRPOINT(p, KTR_CSW))
473 		ktrcsw(l, 1, 0);
474 #endif
475 
476 	/*
477 	 * XXX We need to allocate the sadata_upcall structure here,
478 	 * XXX since we can't sleep while waiting for memory inside
479 	 * XXX sa_upcall().  It would be nice if we could safely
480 	 * XXX allocate the sadata_upcall structure on the stack, here.
481 	 */
482 	if (l->l_flag & L_SA) {
483 		sau = sadata_upcall_alloc(0);
484 	} else {
485 		sau = NULL;
486 	}
487 
488 	SCHED_LOCK(s);
489 
490 #ifdef DIAGNOSTIC
491 	if (ident == NULL)
492 		panic("ltsleep: ident == NULL");
493 	if (l->l_stat != LSONPROC)
494 		panic("ltsleep: l_stat %d != LSONPROC", l->l_stat);
495 	if (l->l_back != NULL)
496 		panic("ltsleep: p_back != NULL");
497 #endif
498 
499 	l->l_wchan = ident;
500 	l->l_wmesg = wmesg;
501 	l->l_slptime = 0;
502 	l->l_priority = priority & PRIMASK;
503 
504 	qp = SLPQUE(ident);
505 	if (qp->sq_head == 0)
506 		qp->sq_head = l;
507 	else {
508 		*qp->sq_tailp = l;
509 	}
510 	*(qp->sq_tailp = &l->l_forw) = 0;
511 
512 	if (timo)
513 		callout_reset(&l->l_tsleep_ch, timo, endtsleep, l);
514 
515 	/*
516 	 * We can now release the interlock; the scheduler_slock
517 	 * is held, so a thread can't get in to do wakeup() before
518 	 * we do the switch.
519 	 *
520 	 * XXX We leave the code block here, after inserting ourselves
521 	 * on the sleep queue, because we might want a more clever
522 	 * data structure for the sleep queues at some point.
523 	 */
524 	if (interlock != NULL)
525 		simple_unlock(interlock);
526 
527 	/*
528 	 * We put ourselves on the sleep queue and start our timeout
529 	 * before calling CURSIG, as we could stop there, and a wakeup
530 	 * or a SIGCONT (or both) could occur while we were stopped.
531 	 * A SIGCONT would cause us to be marked as SSLEEP
532 	 * without resuming us, thus we must be ready for sleep
533 	 * when CURSIG is called.  If the wakeup happens while we're
534 	 * stopped, p->p_wchan will be 0 upon return from CURSIG.
535 	 */
536 	if (catch) {
537 		l->l_flag |= L_SINTR;
538 		if (((sig = CURSIG(l)) != 0) ||
539 		    ((p->p_flag & P_WEXIT) && p->p_nlwps > 1)) {
540 			if (l->l_wchan != NULL)
541 				unsleep(l);
542 			l->l_stat = LSONPROC;
543 			SCHED_UNLOCK(s);
544 			goto resume;
545 		}
546 		if (l->l_wchan == NULL) {
547 			catch = 0;
548 			SCHED_UNLOCK(s);
549 			goto resume;
550 		}
551 	} else
552 		sig = 0;
553 	l->l_stat = LSSLEEP;
554 	p->p_nrlwps--;
555 	p->p_stats->p_ru.ru_nvcsw++;
556 	SCHED_ASSERT_LOCKED();
557 	if (l->l_flag & L_SA)
558 		sa_switch(l, sau, SA_UPCALL_BLOCKED);
559 	else
560 		mi_switch(l, NULL);
561 
562 #if	defined(DDB) && !defined(GPROF)
563 	/* handy breakpoint location after process "wakes" */
564 	__asm(".globl bpendtsleep\nbpendtsleep:");
565 #endif
566 	/*
567 	 * p->p_nrlwps is incremented by whoever made us runnable again,
568 	 * either setrunnable() or awaken().
569 	 */
570 
571 	SCHED_ASSERT_UNLOCKED();
572 	splx(s);
573 
574  resume:
575 	KDASSERT(l->l_cpu != NULL);
576 	KDASSERT(l->l_cpu == curcpu());
577 	l->l_cpu->ci_schedstate.spc_curpriority = l->l_usrpri;
578 
579 	l->l_flag &= ~L_SINTR;
580 	if (l->l_flag & L_TIMEOUT) {
581 		l->l_flag &= ~(L_TIMEOUT|L_CANCELLED);
582 		if (sig == 0) {
583 #ifdef KTRACE
584 			if (KTRPOINT(p, KTR_CSW))
585 				ktrcsw(l, 0, 0);
586 #endif
587 			if (relock && interlock != NULL)
588 				simple_lock(interlock);
589 			return (EWOULDBLOCK);
590 		}
591 	} else if (timo)
592 		callout_stop(&l->l_tsleep_ch);
593 
594 	if (catch) {
595 		const int cancelled = l->l_flag & L_CANCELLED;
596 		l->l_flag &= ~L_CANCELLED;
597 		if (sig != 0 || (sig = CURSIG(l)) != 0 || cancelled) {
598 #ifdef KTRACE
599 			if (KTRPOINT(p, KTR_CSW))
600 				ktrcsw(l, 0, 0);
601 #endif
602 			if (relock && interlock != NULL)
603 				simple_lock(interlock);
604 			/*
605 			 * If this sleep was canceled, don't let the syscall
606 			 * restart.
607 			 */
608 			if (cancelled ||
609 			    (SIGACTION(p, sig).sa_flags & SA_RESTART) == 0)
610 				return (EINTR);
611 			return (ERESTART);
612 		}
613 	}
614 
615 #ifdef KTRACE
616 	if (KTRPOINT(p, KTR_CSW))
617 		ktrcsw(l, 0, 0);
618 #endif
619 	if (relock && interlock != NULL)
620 		simple_lock(interlock);
621 
622 	/* XXXNJW this is very much a kluge.
623 	 * revisit. a better way of preventing looping/hanging syscalls like
624 	 * wait4() and _lwp_wait() from wedging an exiting process
625 	 * would be preferred.
626 	 */
627 	if (catch && ((p->p_flag & P_WEXIT) && p->p_nlwps > 1 && exiterr))
628 		return (EINTR);
629 	return (0);
630 }
631 
632 /*
633  * Implement timeout for tsleep.
634  * If process hasn't been awakened (wchan non-zero),
635  * set timeout flag and undo the sleep.  If proc
636  * is stopped, just unsleep so it will remain stopped.
637  */
638 void
639 endtsleep(void *arg)
640 {
641 	struct lwp *l;
642 	int s;
643 
644 	l = (struct lwp *)arg;
645 	SCHED_LOCK(s);
646 	if (l->l_wchan) {
647 		if (l->l_stat == LSSLEEP)
648 			setrunnable(l);
649 		else
650 			unsleep(l);
651 		l->l_flag |= L_TIMEOUT;
652 	}
653 	SCHED_UNLOCK(s);
654 }
655 
656 /*
657  * Remove a process from its wait queue
658  */
659 void
660 unsleep(struct lwp *l)
661 {
662 	struct slpque *qp;
663 	struct lwp **hp;
664 
665 	SCHED_ASSERT_LOCKED();
666 
667 	if (l->l_wchan) {
668 		hp = &(qp = SLPQUE(l->l_wchan))->sq_head;
669 		while (*hp != l)
670 			hp = &(*hp)->l_forw;
671 		*hp = l->l_forw;
672 		if (qp->sq_tailp == &l->l_forw)
673 			qp->sq_tailp = hp;
674 		l->l_wchan = 0;
675 	}
676 }
677 
678 inline void
679 sa_awaken(struct lwp *l)
680 {
681 
682 	SCHED_ASSERT_LOCKED();
683 
684 	if (l == l->l_savp->savp_lwp && l->l_flag & L_SA_YIELD)
685 		l->l_flag &= ~L_SA_IDLE;
686 }
687 
688 /*
689  * Optimized-for-wakeup() version of setrunnable().
690  */
691 inline void
692 awaken(struct lwp *l)
693 {
694 
695 	SCHED_ASSERT_LOCKED();
696 
697 	if (l->l_proc->p_sa)
698 		sa_awaken(l);
699 
700 	if (l->l_slptime > 1)
701 		updatepri(l);
702 	l->l_slptime = 0;
703 	l->l_stat = LSRUN;
704 	l->l_proc->p_nrlwps++;
705 	/*
706 	 * Since curpriority is a user priority, p->p_priority
707 	 * is always better than curpriority on the last CPU on
708 	 * which it ran.
709 	 *
710 	 * XXXSMP See affinity comment in resched_proc().
711 	 */
712 	if (l->l_flag & L_INMEM) {
713 		setrunqueue(l);
714 		KASSERT(l->l_cpu != NULL);
715 		need_resched(l->l_cpu);
716 	} else
717 		sched_wakeup(&proc0);
718 }
719 
720 #if defined(MULTIPROCESSOR) || defined(LOCKDEBUG)
721 void
722 sched_unlock_idle(void)
723 {
724 
725 	simple_unlock(&sched_lock);
726 }
727 
728 void
729 sched_lock_idle(void)
730 {
731 
732 	simple_lock(&sched_lock);
733 }
734 #endif /* MULTIPROCESSOR || LOCKDEBUG */
735 
736 /*
737  * Make all processes sleeping on the specified identifier runnable.
738  */
739 
740 void
741 wakeup(volatile const void *ident)
742 {
743 	int s;
744 
745 	SCHED_ASSERT_UNLOCKED();
746 
747 	SCHED_LOCK(s);
748 	sched_wakeup(ident);
749 	SCHED_UNLOCK(s);
750 }
751 
752 void
753 sched_wakeup(volatile const void *ident)
754 {
755 	struct slpque *qp;
756 	struct lwp *l, **q;
757 
758 	SCHED_ASSERT_LOCKED();
759 
760 	qp = SLPQUE(ident);
761  restart:
762 	for (q = &qp->sq_head; (l = *q) != NULL; ) {
763 #ifdef DIAGNOSTIC
764 		if (l->l_back || (l->l_stat != LSSLEEP &&
765 		    l->l_stat != LSSTOP && l->l_stat != LSSUSPENDED))
766 			panic("wakeup");
767 #endif
768 		if (l->l_wchan == ident) {
769 			l->l_wchan = 0;
770 			*q = l->l_forw;
771 			if (qp->sq_tailp == &l->l_forw)
772 				qp->sq_tailp = q;
773 			if (l->l_stat == LSSLEEP) {
774 				awaken(l);
775 				goto restart;
776 			}
777 		} else
778 			q = &l->l_forw;
779 	}
780 }
781 
782 /*
783  * Make the highest priority process first in line on the specified
784  * identifier runnable.
785  */
786 void
787 wakeup_one(volatile const void *ident)
788 {
789 	struct slpque *qp;
790 	struct lwp *l, **q;
791 	struct lwp *best_sleepp, **best_sleepq;
792 	struct lwp *best_stopp, **best_stopq;
793 	int s;
794 
795 	best_sleepp = best_stopp = NULL;
796 	best_sleepq = best_stopq = NULL;
797 
798 	SCHED_LOCK(s);
799 
800 	qp = SLPQUE(ident);
801 
802 	for (q = &qp->sq_head; (l = *q) != NULL; q = &l->l_forw) {
803 #ifdef DIAGNOSTIC
804 		if (l->l_back || (l->l_stat != LSSLEEP &&
805 		    l->l_stat != LSSTOP && l->l_stat != LSSUSPENDED))
806 			panic("wakeup_one");
807 #endif
808 		if (l->l_wchan == ident) {
809 			if (l->l_stat == LSSLEEP) {
810 				if (best_sleepp == NULL ||
811 				    l->l_priority < best_sleepp->l_priority) {
812 					best_sleepp = l;
813 					best_sleepq = q;
814 				}
815 			} else {
816 				if (best_stopp == NULL ||
817 				    l->l_priority < best_stopp->l_priority) {
818 				    	best_stopp = l;
819 					best_stopq = q;
820 				}
821 			}
822 		}
823 	}
824 
825 	/*
826 	 * Consider any SSLEEP process higher than the highest priority SSTOP
827 	 * process.
828 	 */
829 	if (best_sleepp != NULL) {
830 		l = best_sleepp;
831 		q = best_sleepq;
832 	} else {
833 		l = best_stopp;
834 		q = best_stopq;
835 	}
836 
837 	if (l != NULL) {
838 		l->l_wchan = NULL;
839 		*q = l->l_forw;
840 		if (qp->sq_tailp == &l->l_forw)
841 			qp->sq_tailp = q;
842 		if (l->l_stat == LSSLEEP)
843 			awaken(l);
844 	}
845 	SCHED_UNLOCK(s);
846 }
847 
848 /*
849  * General yield call.  Puts the current process back on its run queue and
850  * performs a voluntary context switch.  Should only be called when the
851  * current process explicitly requests it (eg sched_yield(2) in compat code).
852  */
853 void
854 yield(void)
855 {
856 	struct lwp *l = curlwp;
857 	int s;
858 
859 	SCHED_LOCK(s);
860 	l->l_priority = l->l_usrpri;
861 	l->l_stat = LSRUN;
862 	setrunqueue(l);
863 	l->l_proc->p_stats->p_ru.ru_nvcsw++;
864 	mi_switch(l, NULL);
865 	SCHED_ASSERT_UNLOCKED();
866 	splx(s);
867 }
868 
869 /*
870  * General preemption call.  Puts the current process back on its run queue
871  * and performs an involuntary context switch.
872  * The 'more' ("more work to do") argument is boolean. Returning to userspace
873  * preempt() calls pass 0. "Voluntary" preemptions in e.g. uiomove() pass 1.
874  * This will be used to indicate to the SA subsystem that the LWP is
875  * not yet finished in the kernel.
876  */
877 
878 void
879 preempt(int more)
880 {
881 	struct lwp *l = curlwp;
882 	int r, s;
883 
884 	SCHED_LOCK(s);
885 	l->l_priority = l->l_usrpri;
886 	l->l_stat = LSRUN;
887 	setrunqueue(l);
888 	l->l_proc->p_stats->p_ru.ru_nivcsw++;
889 	r = mi_switch(l, NULL);
890 	SCHED_ASSERT_UNLOCKED();
891 	splx(s);
892 	if ((l->l_flag & L_SA) != 0 && r != 0 && more == 0)
893 		sa_preempt(l);
894 }
895 
896 /*
897  * The machine independent parts of context switch.
898  * Must be called at splsched() (no higher!) and with
899  * the sched_lock held.
900  * Switch to "new" if non-NULL, otherwise let cpu_switch choose
901  * the next lwp.
902  *
903  * Returns 1 if another process was actually run.
904  */
905 int
906 mi_switch(struct lwp *l, struct lwp *newl)
907 {
908 	struct schedstate_percpu *spc;
909 	struct rlimit *rlim;
910 	long s, u;
911 	struct timeval tv;
912 	int hold_count;
913 	struct proc *p = l->l_proc;
914 	int retval;
915 
916 	SCHED_ASSERT_LOCKED();
917 
918 	/*
919 	 * Release the kernel_lock, as we are about to yield the CPU.
920 	 * The scheduler lock is still held until cpu_switch()
921 	 * selects a new process and removes it from the run queue.
922 	 */
923 	hold_count = KERNEL_LOCK_RELEASE_ALL();
924 
925 	KDASSERT(l->l_cpu != NULL);
926 	KDASSERT(l->l_cpu == curcpu());
927 
928 	spc = &l->l_cpu->ci_schedstate;
929 
930 #if defined(LOCKDEBUG) || defined(DIAGNOSTIC)
931 	spinlock_switchcheck();
932 #endif
933 #ifdef LOCKDEBUG
934 	simple_lock_switchcheck();
935 #endif
936 
937 	/*
938 	 * Compute the amount of time during which the current
939 	 * process was running.
940 	 */
941 	microtime(&tv);
942 	u = p->p_rtime.tv_usec +
943 	    (tv.tv_usec - spc->spc_runtime.tv_usec);
944 	s = p->p_rtime.tv_sec + (tv.tv_sec - spc->spc_runtime.tv_sec);
945 	if (u < 0) {
946 		u += 1000000;
947 		s--;
948 	} else if (u >= 1000000) {
949 		u -= 1000000;
950 		s++;
951 	}
952 	p->p_rtime.tv_usec = u;
953 	p->p_rtime.tv_sec = s;
954 
955 	/*
956 	 * Check if the process exceeds its CPU resource allocation.
957 	 * If over max, kill it.  In any case, if it has run for more
958 	 * than 10 minutes, reduce priority to give others a chance.
959 	 */
960 	rlim = &p->p_rlimit[RLIMIT_CPU];
961 	if (s >= rlim->rlim_cur) {
962 		/*
963 		 * XXXSMP: we're inside the scheduler lock perimeter;
964 		 * use sched_psignal.
965 		 */
966 		if (s >= rlim->rlim_max)
967 			sched_psignal(p, SIGKILL);
968 		else {
969 			sched_psignal(p, SIGXCPU);
970 			if (rlim->rlim_cur < rlim->rlim_max)
971 				rlim->rlim_cur += 5;
972 		}
973 	}
974 	if (autonicetime && s > autonicetime && p->p_ucred->cr_uid &&
975 	    p->p_nice == NZERO) {
976 		p->p_nice = autoniceval + NZERO;
977 		resetpriority(l);
978 	}
979 
980 	/*
981 	 * Process is about to yield the CPU; clear the appropriate
982 	 * scheduling flags.
983 	 */
984 	spc->spc_flags &= ~SPCF_SWITCHCLEAR;
985 
986 #ifdef KSTACK_CHECK_MAGIC
987 	kstack_check_magic(l);
988 #endif
989 
990 	/*
991 	 * If we are using h/w performance counters, save context.
992 	 */
993 #if PERFCTRS
994 	if (PMC_ENABLED(p))
995 		pmc_save_context(p);
996 #endif
997 
998 	/*
999 	 * Switch to the new current process.  When we
1000 	 * run again, we'll return back here.
1001 	 */
1002 	uvmexp.swtch++;
1003 	if (newl == NULL) {
1004 		retval = cpu_switch(l, NULL);
1005 	} else {
1006 		remrunqueue(newl);
1007 		cpu_switchto(l, newl);
1008 		retval = 0;
1009 	}
1010 
1011 	/*
1012 	 * If we are using h/w performance counters, restore context.
1013 	 */
1014 #if PERFCTRS
1015 	if (PMC_ENABLED(p))
1016 		pmc_restore_context(p);
1017 #endif
1018 
1019 	/*
1020 	 * Make sure that MD code released the scheduler lock before
1021 	 * resuming us.
1022 	 */
1023 	SCHED_ASSERT_UNLOCKED();
1024 
1025 	/*
1026 	 * We're running again; record our new start time.  We might
1027 	 * be running on a new CPU now, so don't use the cache'd
1028 	 * schedstate_percpu pointer.
1029 	 */
1030 	KDASSERT(l->l_cpu != NULL);
1031 	KDASSERT(l->l_cpu == curcpu());
1032 	microtime(&l->l_cpu->ci_schedstate.spc_runtime);
1033 
1034 	/*
1035 	 * Reacquire the kernel_lock now.  We do this after we've
1036 	 * released the scheduler lock to avoid deadlock, and before
1037 	 * we reacquire the interlock.
1038 	 */
1039 	KERNEL_LOCK_ACQUIRE_COUNT(hold_count);
1040 
1041 	return retval;
1042 }
1043 
1044 /*
1045  * Initialize the (doubly-linked) run queues
1046  * to be empty.
1047  */
1048 void
1049 rqinit()
1050 {
1051 	int i;
1052 
1053 	for (i = 0; i < RUNQUE_NQS; i++)
1054 		sched_qs[i].ph_link = sched_qs[i].ph_rlink =
1055 		    (struct lwp *)&sched_qs[i];
1056 }
1057 
1058 static inline void
1059 resched_proc(struct lwp *l, u_char pri)
1060 {
1061 	struct cpu_info *ci;
1062 
1063 	/*
1064 	 * XXXSMP
1065 	 * Since l->l_cpu persists across a context switch,
1066 	 * this gives us *very weak* processor affinity, in
1067 	 * that we notify the CPU on which the process last
1068 	 * ran that it should try to switch.
1069 	 *
1070 	 * This does not guarantee that the process will run on
1071 	 * that processor next, because another processor might
1072 	 * grab it the next time it performs a context switch.
1073 	 *
1074 	 * This also does not handle the case where its last
1075 	 * CPU is running a higher-priority process, but every
1076 	 * other CPU is running a lower-priority process.  There
1077 	 * are ways to handle this situation, but they're not
1078 	 * currently very pretty, and we also need to weigh the
1079 	 * cost of moving a process from one CPU to another.
1080 	 *
1081 	 * XXXSMP
1082 	 * There is also the issue of locking the other CPU's
1083 	 * sched state, which we currently do not do.
1084 	 */
1085 	ci = (l->l_cpu != NULL) ? l->l_cpu : curcpu();
1086 	if (pri < ci->ci_schedstate.spc_curpriority)
1087 		need_resched(ci);
1088 }
1089 
1090 /*
1091  * Change process state to be runnable,
1092  * placing it on the run queue if it is in memory,
1093  * and awakening the swapper if it isn't in memory.
1094  */
1095 void
1096 setrunnable(struct lwp *l)
1097 {
1098 	struct proc *p = l->l_proc;
1099 
1100 	SCHED_ASSERT_LOCKED();
1101 
1102 	switch (l->l_stat) {
1103 	case 0:
1104 	case LSRUN:
1105 	case LSONPROC:
1106 	case LSZOMB:
1107 	case LSDEAD:
1108 	default:
1109 		panic("setrunnable: lwp %p state was %d", l, l->l_stat);
1110 	case LSSTOP:
1111 		/*
1112 		 * If we're being traced (possibly because someone attached us
1113 		 * while we were stopped), check for a signal from the debugger.
1114 		 */
1115 		if ((p->p_flag & P_TRACED) != 0 && p->p_xstat != 0) {
1116 			sigaddset(&p->p_sigctx.ps_siglist, p->p_xstat);
1117 			CHECKSIGS(p);
1118 		}
1119 	case LSSLEEP:
1120 		unsleep(l);		/* e.g. when sending signals */
1121 		break;
1122 
1123 	case LSIDL:
1124 		break;
1125 	case LSSUSPENDED:
1126 		break;
1127 	}
1128 
1129 	if (l->l_proc->p_sa)
1130 		sa_awaken(l);
1131 
1132 	l->l_stat = LSRUN;
1133 	p->p_nrlwps++;
1134 
1135 	if (l->l_flag & L_INMEM)
1136 		setrunqueue(l);
1137 
1138 	if (l->l_slptime > 1)
1139 		updatepri(l);
1140 	l->l_slptime = 0;
1141 	if ((l->l_flag & L_INMEM) == 0)
1142 		sched_wakeup((caddr_t)&proc0);
1143 	else
1144 		resched_proc(l, l->l_priority);
1145 }
1146 
1147 /*
1148  * Compute the priority of a process when running in user mode.
1149  * Arrange to reschedule if the resulting priority is better
1150  * than that of the current process.
1151  */
1152 void
1153 resetpriority(struct lwp *l)
1154 {
1155 	unsigned int newpriority;
1156 	struct proc *p = l->l_proc;
1157 
1158 	SCHED_ASSERT_LOCKED();
1159 
1160 	newpriority = PUSER + (p->p_estcpu >> ESTCPU_SHIFT) +
1161 			NICE_WEIGHT * (p->p_nice - NZERO);
1162 	newpriority = min(newpriority, MAXPRI);
1163 	l->l_usrpri = newpriority;
1164 	resched_proc(l, l->l_usrpri);
1165 }
1166 
1167 /*
1168  * Recompute priority for all LWPs in a process.
1169  */
1170 void
1171 resetprocpriority(struct proc *p)
1172 {
1173 	struct lwp *l;
1174 
1175 	LIST_FOREACH(l, &p->p_lwps, l_sibling)
1176 	    resetpriority(l);
1177 }
1178 
1179 /*
1180  * We adjust the priority of the current process.  The priority of a process
1181  * gets worse as it accumulates CPU time.  The CPU usage estimator (p_estcpu)
1182  * is increased here.  The formula for computing priorities (in kern_synch.c)
1183  * will compute a different value each time p_estcpu increases. This can
1184  * cause a switch, but unless the priority crosses a PPQ boundary the actual
1185  * queue will not change.  The CPU usage estimator ramps up quite quickly
1186  * when the process is running (linearly), and decays away exponentially, at
1187  * a rate which is proportionally slower when the system is busy.  The basic
1188  * principle is that the system will 90% forget that the process used a lot
1189  * of CPU time in 5 * loadav seconds.  This causes the system to favor
1190  * processes which haven't run much recently, and to round-robin among other
1191  * processes.
1192  */
1193 
1194 void
1195 schedclock(struct lwp *l)
1196 {
1197 	struct proc *p = l->l_proc;
1198 	int s;
1199 
1200 	p->p_estcpu = ESTCPULIM(p->p_estcpu + (1 << ESTCPU_SHIFT));
1201 	SCHED_LOCK(s);
1202 	resetpriority(l);
1203 	SCHED_UNLOCK(s);
1204 
1205 	if (l->l_priority >= PUSER)
1206 		l->l_priority = l->l_usrpri;
1207 }
1208 
1209 void
1210 suspendsched()
1211 {
1212 	struct lwp *l;
1213 	int s;
1214 
1215 	/*
1216 	 * Convert all non-P_SYSTEM LSSLEEP or LSRUN processes to
1217 	 * LSSUSPENDED.
1218 	 */
1219 	proclist_lock_read();
1220 	SCHED_LOCK(s);
1221 	LIST_FOREACH(l, &alllwp, l_list) {
1222 		if ((l->l_proc->p_flag & P_SYSTEM) != 0)
1223 			continue;
1224 
1225 		switch (l->l_stat) {
1226 		case LSRUN:
1227 			l->l_proc->p_nrlwps--;
1228 			if ((l->l_flag & L_INMEM) != 0)
1229 				remrunqueue(l);
1230 			/* FALLTHROUGH */
1231 		case LSSLEEP:
1232 			l->l_stat = LSSUSPENDED;
1233 			break;
1234 		case LSONPROC:
1235 			/*
1236 			 * XXX SMP: we need to deal with processes on
1237 			 * others CPU !
1238 			 */
1239 			break;
1240 		default:
1241 			break;
1242 		}
1243 	}
1244 	SCHED_UNLOCK(s);
1245 	proclist_unlock_read();
1246 }
1247 
1248 /*
1249  * scheduler_fork_hook:
1250  *
1251  *	Inherit the parent's scheduler history.
1252  */
1253 void
1254 scheduler_fork_hook(struct proc *parent, struct proc *child)
1255 {
1256 
1257 	child->p_estcpu = child->p_estcpu_inherited = parent->p_estcpu;
1258 	child->p_forktime = schedcpu_ticks;
1259 }
1260 
1261 /*
1262  * scheduler_wait_hook:
1263  *
1264  *	Chargeback parents for the sins of their children.
1265  */
1266 void
1267 scheduler_wait_hook(struct proc *parent, struct proc *child)
1268 {
1269 	fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
1270 	fixpt_t estcpu;
1271 
1272 	/* XXX Only if parent != init?? */
1273 
1274 	estcpu = decay_cpu_batch(loadfac, child->p_estcpu_inherited,
1275 	    schedcpu_ticks - child->p_forktime);
1276 	if (child->p_estcpu > estcpu) {
1277 		parent->p_estcpu =
1278 		    ESTCPULIM(parent->p_estcpu + child->p_estcpu - estcpu);
1279 	}
1280 }
1281 
1282 /*
1283  * Low-level routines to access the run queue.  Optimised assembler
1284  * routines can override these.
1285  */
1286 
1287 #ifndef __HAVE_MD_RUNQUEUE
1288 
1289 /*
1290  * On some architectures, it's faster to use a MSB ordering for the priorites
1291  * than the traditional LSB ordering.
1292  */
1293 #ifdef __HAVE_BIGENDIAN_BITOPS
1294 #define	RQMASK(n) (0x80000000 >> (n))
1295 #else
1296 #define	RQMASK(n) (0x00000001 << (n))
1297 #endif
1298 
1299 /*
1300  * The primitives that manipulate the run queues.  whichqs tells which
1301  * of the 32 queues qs have processes in them.  Setrunqueue puts processes
1302  * into queues, remrunqueue removes them from queues.  The running process is
1303  * on no queue, other processes are on a queue related to p->p_priority,
1304  * divided by 4 actually to shrink the 0-127 range of priorities into the 32
1305  * available queues.
1306  */
1307 
1308 #ifdef RQDEBUG
1309 static void
1310 checkrunqueue(int whichq, struct lwp *l)
1311 {
1312 	const struct prochd * const rq = &sched_qs[whichq];
1313 	struct lwp *l2;
1314 	int found = 0;
1315 	int die = 0;
1316 	int empty = 1;
1317 	for (l2 = rq->ph_link; l2 != (void*) rq; l2 = l2->l_forw) {
1318 		if (l2->l_stat != LSRUN) {
1319 			printf("checkrunqueue[%d]: lwp %p state (%d) "
1320 			    " != LSRUN\n", whichq, l2, l2->l_stat);
1321 		}
1322 		if (l2->l_back->l_forw != l2) {
1323 			printf("checkrunqueue[%d]: lwp %p back-qptr (%p) "
1324 			    "corrupt %p\n", whichq, l2, l2->l_back,
1325 			    l2->l_back->l_forw);
1326 			die = 1;
1327 		}
1328 		if (l2->l_forw->l_back != l2) {
1329 			printf("checkrunqueue[%d]: lwp %p forw-qptr (%p) "
1330 			    "corrupt %p\n", whichq, l2, l2->l_forw,
1331 			    l2->l_forw->l_back);
1332 			die = 1;
1333 		}
1334 		if (l2 == l)
1335 			found = 1;
1336 		empty = 0;
1337 	}
1338 	if (empty && (sched_whichqs & RQMASK(whichq)) != 0) {
1339 		printf("checkrunqueue[%d]: bit set for empty run-queue %p\n",
1340 		    whichq, rq);
1341 		die = 1;
1342 	} else if (!empty && (sched_whichqs & RQMASK(whichq)) == 0) {
1343 		printf("checkrunqueue[%d]: bit clear for non-empty "
1344 		    "run-queue %p\n", whichq, rq);
1345 		die = 1;
1346 	}
1347 	if (l != NULL && (sched_whichqs & RQMASK(whichq)) == 0) {
1348 		printf("checkrunqueue[%d]: bit clear for active lwp %p\n",
1349 		    whichq, l);
1350 		die = 1;
1351 	}
1352 	if (l != NULL && empty) {
1353 		printf("checkrunqueue[%d]: empty run-queue %p with "
1354 		    "active lwp %p\n", whichq, rq, l);
1355 		die = 1;
1356 	}
1357 	if (l != NULL && !found) {
1358 		printf("checkrunqueue[%d]: lwp %p not in runqueue %p!",
1359 		    whichq, l, rq);
1360 		die = 1;
1361 	}
1362 	if (die)
1363 		panic("checkrunqueue: inconsistency found");
1364 }
1365 #endif /* RQDEBUG */
1366 
1367 void
1368 setrunqueue(struct lwp *l)
1369 {
1370 	struct prochd *rq;
1371 	struct lwp *prev;
1372 	const int whichq = l->l_priority / PPQ;
1373 
1374 #ifdef RQDEBUG
1375 	checkrunqueue(whichq, NULL);
1376 #endif
1377 #ifdef DIAGNOSTIC
1378 	if (l->l_back != NULL || l->l_wchan != NULL || l->l_stat != LSRUN)
1379 		panic("setrunqueue");
1380 #endif
1381 	sched_whichqs |= RQMASK(whichq);
1382 	rq = &sched_qs[whichq];
1383 	prev = rq->ph_rlink;
1384 	l->l_forw = (struct lwp *)rq;
1385 	rq->ph_rlink = l;
1386 	prev->l_forw = l;
1387 	l->l_back = prev;
1388 #ifdef RQDEBUG
1389 	checkrunqueue(whichq, l);
1390 #endif
1391 }
1392 
1393 void
1394 remrunqueue(struct lwp *l)
1395 {
1396 	struct lwp *prev, *next;
1397 	const int whichq = l->l_priority / PPQ;
1398 #ifdef RQDEBUG
1399 	checkrunqueue(whichq, l);
1400 #endif
1401 #ifdef DIAGNOSTIC
1402 	if (((sched_whichqs & RQMASK(whichq)) == 0))
1403 		panic("remrunqueue: bit %d not set", whichq);
1404 #endif
1405 	prev = l->l_back;
1406 	l->l_back = NULL;
1407 	next = l->l_forw;
1408 	prev->l_forw = next;
1409 	next->l_back = prev;
1410 	if (prev == next)
1411 		sched_whichqs &= ~RQMASK(whichq);
1412 #ifdef RQDEBUG
1413 	checkrunqueue(whichq, NULL);
1414 #endif
1415 }
1416 
1417 #undef RQMASK
1418 #endif /* !defined(__HAVE_MD_RUNQUEUE) */
1419