sched_4bsd.c revision 1.1.2.24 1 /* $NetBSD: sched_4bsd.c,v 1.1.2.24 2007/04/03 15:23:26 matt Exp $ */
2
3 /*-
4 * Copyright (c) 1999, 2000, 2004, 2006, 2007 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, by Charles M. Hannum, Andrew Doran, and
10 * Daniel Sieger.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the NetBSD
23 * Foundation, Inc. and its contributors.
24 * 4. Neither the name of The NetBSD Foundation nor the names of its
25 * contributors may be used to endorse or promote products derived
26 * from this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
29 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
30 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
31 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
32 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
34 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
35 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
36 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
37 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
38 * POSSIBILITY OF SUCH DAMAGE.
39 */
40
41 /*-
42 * Copyright (c) 1982, 1986, 1990, 1991, 1993
43 * The Regents of the University of California. All rights reserved.
44 * (c) UNIX System Laboratories, Inc.
45 * All or some portions of this file are derived from material licensed
46 * to the University of California by American Telephone and Telegraph
47 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
48 * the permission of UNIX System Laboratories, Inc.
49 *
50 * Redistribution and use in source and binary forms, with or without
51 * modification, are permitted provided that the following conditions
52 * are met:
53 * 1. Redistributions of source code must retain the above copyright
54 * notice, this list of conditions and the following disclaimer.
55 * 2. Redistributions in binary form must reproduce the above copyright
56 * notice, this list of conditions and the following disclaimer in the
57 * documentation and/or other materials provided with the distribution.
58 * 3. Neither the name of the University nor the names of its contributors
59 * may be used to endorse or promote products derived from this software
60 * without specific prior written permission.
61 *
62 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
63 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
64 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
65 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
66 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
67 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
68 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
69 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
70 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
71 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
72 * SUCH DAMAGE.
73 *
74 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
75 */
76
77 #include <sys/cdefs.h>
78 __KERNEL_RCSID(0, "$NetBSD: sched_4bsd.c,v 1.1.2.24 2007/04/03 15:23:26 matt Exp $");
79
80 #include "opt_ddb.h"
81 #include "opt_lockdebug.h"
82 #include "opt_perfctrs.h"
83
84 #define __MUTEX_PRIVATE
85
86 #include <sys/param.h>
87 #include <sys/systm.h>
88 #include <sys/callout.h>
89 #include <sys/cpu.h>
90 #include <sys/proc.h>
91 #include <sys/kernel.h>
92 #include <sys/signalvar.h>
93 #include <sys/resourcevar.h>
94 #include <sys/sched.h>
95 #include <sys/sysctl.h>
96 #include <sys/kauth.h>
97 #include <sys/lockdebug.h>
98
99 #include <uvm/uvm_extern.h>
100
101 /*
102 * Run queues.
103 *
104 * We have 32 run queues in descending priority of 0..31. We maintain
105 * a bitmask of non-empty queues in order speed up finding the first
106 * runnable process. The bitmask is maintained only by machine-dependent
107 * code, allowing the most efficient instructions to be used to find the
108 * first non-empty queue.
109 */
110
111 #define RUNQUE_NQS 32 /* number of runqueues */
112 #define PPQ (128 / RUNQUE_NQS) /* priorities per queue */
113
114 typedef struct subqueue {
115 TAILQ_HEAD(, lwp) sq_queue;
116 } subqueue_t;
117 typedef struct runqueue {
118 subqueue_t rq_subqueues[RUNQUE_NQS]; /* run queues */
119 uint32_t rq_bitmap; /* bitmap of non-empty queues */
120 } runqueue_t;
121 static runqueue_t global_queue;
122
123 static void updatepri(struct lwp *);
124 static void resetpriority(struct lwp *);
125 static void resetprocpriority(struct proc *);
126
127 extern unsigned int sched_pstats_ticks; /* defined in kern_synch.c */
128
129 /* The global scheduler state */
130 kmutex_t sched_mutex;
131
132 /* Number of hardclock ticks per sched_tick() */
133 int rrticks;
134
135 /*
136 * Force switch among equal priority processes every 100ms.
137 * Called from hardclock every hz/10 == rrticks hardclock ticks.
138 */
139 /* ARGSUSED */
140 void
141 sched_tick(struct cpu_info *ci)
142 {
143 struct schedstate_percpu *spc = &ci->ci_schedstate;
144
145 spc->spc_ticks = rrticks;
146
147 if (!CURCPU_IDLE_P()) {
148 if (spc->spc_flags & SPCF_SEENRR) {
149 /*
150 * The process has already been through a roundrobin
151 * without switching and may be hogging the CPU.
152 * Indicate that the process should yield.
153 */
154 spc->spc_flags |= SPCF_SHOULDYIELD;
155 } else
156 spc->spc_flags |= SPCF_SEENRR;
157 }
158 cpu_need_resched(curcpu(), 0);
159 }
160
161 #define NICE_WEIGHT 2 /* priorities per nice level */
162
163 #define ESTCPU_SHIFT 11
164 #define ESTCPU_MAX ((NICE_WEIGHT * PRIO_MAX - PPQ) << ESTCPU_SHIFT)
165 #define ESTCPULIM(e) min((e), ESTCPU_MAX)
166
167 /*
168 * Constants for digital decay and forget:
169 * 90% of (p_estcpu) usage in 5 * loadav time
170 * 95% of (p_pctcpu) usage in 60 seconds (load insensitive)
171 * Note that, as ps(1) mentions, this can let percentages
172 * total over 100% (I've seen 137.9% for 3 processes).
173 *
174 * Note that hardclock updates p_estcpu and p_cpticks independently.
175 *
176 * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds.
177 * That is, the system wants to compute a value of decay such
178 * that the following for loop:
179 * for (i = 0; i < (5 * loadavg); i++)
180 * p_estcpu *= decay;
181 * will compute
182 * p_estcpu *= 0.1;
183 * for all values of loadavg:
184 *
185 * Mathematically this loop can be expressed by saying:
186 * decay ** (5 * loadavg) ~= .1
187 *
188 * The system computes decay as:
189 * decay = (2 * loadavg) / (2 * loadavg + 1)
190 *
191 * We wish to prove that the system's computation of decay
192 * will always fulfill the equation:
193 * decay ** (5 * loadavg) ~= .1
194 *
195 * If we compute b as:
196 * b = 2 * loadavg
197 * then
198 * decay = b / (b + 1)
199 *
200 * We now need to prove two things:
201 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1)
202 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg)
203 *
204 * Facts:
205 * For x close to zero, exp(x) =~ 1 + x, since
206 * exp(x) = 0! + x**1/1! + x**2/2! + ... .
207 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b.
208 * For x close to zero, ln(1+x) =~ x, since
209 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1
210 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1).
211 * ln(.1) =~ -2.30
212 *
213 * Proof of (1):
214 * Solve (factor)**(power) =~ .1 given power (5*loadav):
215 * solving for factor,
216 * ln(factor) =~ (-2.30/5*loadav), or
217 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) =
218 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED
219 *
220 * Proof of (2):
221 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)):
222 * solving for power,
223 * power*ln(b/(b+1)) =~ -2.30, or
224 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED
225 *
226 * Actual power values for the implemented algorithm are as follows:
227 * loadav: 1 2 3 4
228 * power: 5.68 10.32 14.94 19.55
229 */
230
231 /* calculations for digital decay to forget 90% of usage in 5*loadav sec */
232 #define loadfactor(loadav) (2 * (loadav))
233
234 static fixpt_t
235 decay_cpu(fixpt_t loadfac, fixpt_t estcpu)
236 {
237
238 if (estcpu == 0) {
239 return 0;
240 }
241
242 #if !defined(_LP64)
243 /* avoid 64bit arithmetics. */
244 #define FIXPT_MAX ((fixpt_t)((UINTMAX_C(1) << sizeof(fixpt_t) * CHAR_BIT) - 1))
245 if (__predict_true(loadfac <= FIXPT_MAX / ESTCPU_MAX)) {
246 return estcpu * loadfac / (loadfac + FSCALE);
247 }
248 #endif /* !defined(_LP64) */
249
250 return (uint64_t)estcpu * loadfac / (loadfac + FSCALE);
251 }
252
253 /*
254 * For all load averages >= 1 and max p_estcpu of (255 << ESTCPU_SHIFT),
255 * sleeping for at least seven times the loadfactor will decay p_estcpu to
256 * less than (1 << ESTCPU_SHIFT).
257 *
258 * note that our ESTCPU_MAX is actually much smaller than (255 << ESTCPU_SHIFT).
259 */
260 static fixpt_t
261 decay_cpu_batch(fixpt_t loadfac, fixpt_t estcpu, unsigned int n)
262 {
263
264 if ((n << FSHIFT) >= 7 * loadfac) {
265 return 0;
266 }
267
268 while (estcpu != 0 && n > 1) {
269 estcpu = decay_cpu(loadfac, estcpu);
270 n--;
271 }
272
273 return estcpu;
274 }
275
276 /*
277 * sched_pstats_hook:
278 *
279 * Periodically called from sched_pstats(); used to recalculate priorities.
280 */
281 inline void
282 sched_pstats_hook(struct proc *p, int minslp)
283 {
284 struct lwp *l;
285 fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
286
287 /*
288 * If the process has slept the entire second,
289 * stop recalculating its priority until it wakes up.
290 */
291 if (minslp <= 1) {
292 p->p_estcpu = decay_cpu(loadfac, p->p_estcpu);
293
294 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
295 if ((l->l_flag & LW_IDLE) != 0)
296 continue;
297 lwp_lock(l);
298 if (l->l_slptime <= 1 &&
299 l->l_priority >= PUSER)
300 resetpriority(l);
301 lwp_unlock(l);
302 }
303 }
304 }
305
306 /*
307 * Recalculate the priority of a process after it has slept for a while.
308 */
309 static void
310 updatepri(struct lwp *l)
311 {
312 struct proc *p = l->l_proc;
313 fixpt_t loadfac;
314
315 LOCK_ASSERT(lwp_locked(l, NULL));
316 KASSERT(l->l_slptime > 1);
317
318 loadfac = loadfactor(averunnable.ldavg[0]);
319
320 l->l_slptime--; /* the first time was done in sched_pstats */
321 /* XXX NJWLWP */
322 /* XXXSMP occasionally unlocked, should be per-LWP */
323 p->p_estcpu = decay_cpu_batch(loadfac, p->p_estcpu, l->l_slptime);
324 resetpriority(l);
325 }
326
327 /*
328 * On some architectures, it's faster to use a MSB ordering for the priorites
329 * than the traditional LSB ordering.
330 */
331 #define RQMASK(n) (0x00000001 << (n))
332
333 /*
334 * The primitives that manipulate the run queues. whichqs tells which
335 * of the 32 queues qs have processes in them. sched_enqueue() puts processes
336 * into queues, sched_dequeue removes them from queues. The running process is
337 * on no queue, other processes are on a queue related to p->p_priority,
338 * divided by 4 actually to shrink the 0-127 range of priorities into the 32
339 * available queues.
340 */
341 #ifdef RQDEBUG
342 static void
343 runqueue_check(const runqueue_t *rq, int whichq, struct lwp *l)
344 {
345 const subqueue_t * const sq = &rq->rq_subqueues[whichq];
346 const uint32_t bitmap = rq->rq_bitmap;
347 struct lwp *l2;
348 int found = 0;
349 int die = 0;
350 int empty = 1;
351
352 TAILQ_FOREACH(l2, &sq->sq_queue, l_runq) {
353 if (l2->l_stat != LSRUN) {
354 printf("runqueue_check[%d]: lwp %p state (%d) "
355 " != LSRUN\n", whichq, l2, l2->l_stat);
356 }
357 if (l2 == l)
358 found = 1;
359 empty = 0;
360 }
361 if (empty && (bitmap & RQMASK(whichq)) != 0) {
362 printf("runqueue_check[%d]: bit set for empty run-queue %p\n",
363 whichq, rq);
364 die = 1;
365 } else if (!empty && (bitmap & RQMASK(whichq)) == 0) {
366 printf("runqueue_check[%d]: bit clear for non-empty "
367 "run-queue %p\n", whichq, rq);
368 die = 1;
369 }
370 if (l != NULL && (bitmap & RQMASK(whichq)) == 0) {
371 printf("runqueue_check[%d]: bit clear for active lwp %p\n",
372 whichq, l);
373 die = 1;
374 }
375 if (l != NULL && empty) {
376 printf("runqueue_check[%d]: empty run-queue %p with "
377 "active lwp %p\n", whichq, rq, l);
378 die = 1;
379 }
380 if (l != NULL && !found) {
381 printf("runqueue_check[%d]: lwp %p not in runqueue %p!",
382 whichq, l, rq);
383 die = 1;
384 }
385 if (die)
386 panic("runqueue_check: inconsistency found");
387 }
388 #else /* RQDEBUG */
389 #define runqueue_check(a, b, c) /* nothing */
390 #endif /* RQDEBUG */
391
392 static void
393 runqueue_init(runqueue_t *rq)
394 {
395 int i;
396
397 for (i = 0; i < RUNQUE_NQS; i++)
398 TAILQ_INIT(&rq->rq_subqueues[i].sq_queue);
399 }
400
401 static void
402 runqueue_enqueue(runqueue_t *rq, struct lwp *l)
403 {
404 subqueue_t *sq;
405 const int whichq = lwp_eprio(l) / PPQ;
406
407 LOCK_ASSERT(lwp_locked(l, l->l_cpu->ci_schedstate.spc_mutex));
408
409 runqueue_check(rq, whichq, NULL);
410 rq->rq_bitmap |= RQMASK(whichq);
411 sq = &rq->rq_subqueues[whichq];
412 TAILQ_INSERT_TAIL(&sq->sq_queue, l, l_runq);
413 runqueue_check(rq, whichq, l);
414 }
415
416 static void
417 runqueue_dequeue(runqueue_t *rq, struct lwp *l)
418 {
419 subqueue_t *sq;
420 const int whichq = lwp_eprio(l) / PPQ;
421
422 LOCK_ASSERT(lwp_locked(l, l->l_cpu->ci_schedstate.spc_mutex));
423
424 runqueue_check(rq, whichq, l);
425 KASSERT((rq->rq_bitmap & RQMASK(whichq)) != 0);
426 sq = &rq->rq_subqueues[whichq];
427 TAILQ_REMOVE(&sq->sq_queue, l, l_runq);
428 if (TAILQ_EMPTY(&sq->sq_queue))
429 rq->rq_bitmap &= ~RQMASK(whichq);
430 runqueue_check(rq, whichq, NULL);
431 }
432
433 static struct lwp *
434 runqueue_nextlwp(runqueue_t *rq)
435 {
436 const uint32_t bitmap = rq->rq_bitmap;
437 int whichq;
438
439 if (bitmap == 0) {
440 return NULL;
441 }
442 whichq = ffs(bitmap) - 1;
443 return TAILQ_FIRST(&rq->rq_subqueues[whichq].sq_queue);
444 }
445
446 #if defined(DDB)
447 static void
448 runqueue_print(const runqueue_t *rq, void (*pr)(const char *, ...))
449 {
450 const uint32_t bitmap = rq->rq_bitmap;
451 struct lwp *l;
452 int i, first;
453
454 for (i = 0; i < RUNQUE_NQS; i++) {
455 const subqueue_t *sq;
456 first = 1;
457 sq = &rq->rq_subqueues[i];
458 TAILQ_FOREACH(l, &sq->sq_queue, l_runq) {
459 if (first) {
460 (*pr)("%c%d",
461 (bitmap & RQMASK(i)) ? ' ' : '!', i);
462 first = 0;
463 }
464 (*pr)("\t%d.%d (%s) pri=%d usrpri=%d\n",
465 l->l_proc->p_pid,
466 l->l_lid, l->l_proc->p_comm,
467 (int)l->l_priority, (int)l->l_usrpri);
468 }
469 }
470 }
471 #endif /* defined(DDB) */
472 #undef RQMASK
473
474 /*
475 * Initialize the (doubly-linked) run queues
476 * to be empty.
477 */
478 void
479 sched_rqinit()
480 {
481
482 runqueue_init(&global_queue);
483 mutex_init(&sched_mutex, MUTEX_SPIN, IPL_SCHED);
484 /* Initialize the lock pointer for lwp0 */
485 lwp0.l_mutex = &sched_mutex;
486 }
487
488 void
489 sched_cpuattach(struct cpu_info *ci)
490 {
491
492 ci->ci_schedstate.spc_mutex = &sched_mutex;
493 }
494
495 void
496 sched_setup()
497 {
498
499 rrticks = hz / 10;
500 sched_pstats(NULL);
501 }
502
503 void
504 sched_setrunnable(struct lwp *l)
505 {
506
507 if (l->l_slptime > 1)
508 updatepri(l);
509 }
510
511 bool
512 sched_curcpu_runnable_p(void)
513 {
514
515 return global_queue.rq_bitmap != 0;
516 }
517
518 void
519 sched_nice(struct proc *chgp, int n)
520 {
521
522 chgp->p_nice = n;
523 (void)resetprocpriority(chgp);
524 }
525
526 /*
527 * Compute the priority of a process when running in user mode.
528 * Arrange to reschedule if the resulting priority is better
529 * than that of the current process.
530 */
531 static void
532 resetpriority(struct lwp *l)
533 {
534 unsigned int newpriority;
535 struct proc *p = l->l_proc;
536
537 /* XXXSMP LOCK_ASSERT(mutex_owned(&p->p_stmutex)); */
538 LOCK_ASSERT(lwp_locked(l, NULL));
539
540 if ((l->l_flag & LW_SYSTEM) != 0)
541 return;
542
543 newpriority = PUSER + (p->p_estcpu >> ESTCPU_SHIFT) +
544 NICE_WEIGHT * (p->p_nice - NZERO);
545 newpriority = min(newpriority, MAXPRI);
546 lwp_changepri(l, newpriority);
547 }
548
549 /*
550 * Recompute priority for all LWPs in a process.
551 */
552 static void
553 resetprocpriority(struct proc *p)
554 {
555 struct lwp *l;
556
557 LOCK_ASSERT(mutex_owned(&p->p_stmutex));
558
559 LIST_FOREACH(l, &p->p_lwps, l_sibling) {
560 lwp_lock(l);
561 resetpriority(l);
562 lwp_unlock(l);
563 }
564 }
565
566 /*
567 * We adjust the priority of the current process. The priority of a process
568 * gets worse as it accumulates CPU time. The CPU usage estimator (p_estcpu)
569 * is increased here. The formula for computing priorities (in kern_synch.c)
570 * will compute a different value each time p_estcpu increases. This can
571 * cause a switch, but unless the priority crosses a PPQ boundary the actual
572 * queue will not change. The CPU usage estimator ramps up quite quickly
573 * when the process is running (linearly), and decays away exponentially, at
574 * a rate which is proportionally slower when the system is busy. The basic
575 * principle is that the system will 90% forget that the process used a lot
576 * of CPU time in 5 * loadav seconds. This causes the system to favor
577 * processes which haven't run much recently, and to round-robin among other
578 * processes.
579 */
580
581 void
582 sched_schedclock(struct lwp *l)
583 {
584 struct proc *p = l->l_proc;
585
586 KASSERT(!CURCPU_IDLE_P());
587 mutex_spin_enter(&p->p_stmutex);
588 p->p_estcpu = ESTCPULIM(p->p_estcpu + (1 << ESTCPU_SHIFT));
589 lwp_lock(l);
590 resetpriority(l);
591 mutex_spin_exit(&p->p_stmutex);
592 if ((l->l_flag & LW_SYSTEM) == 0 && l->l_priority >= PUSER)
593 l->l_priority = l->l_usrpri;
594 lwp_unlock(l);
595 }
596
597 /*
598 * scheduler_fork_hook:
599 *
600 * Inherit the parent's scheduler history.
601 */
602 void
603 sched_proc_fork(struct proc *parent, struct proc *child)
604 {
605
606 LOCK_ASSERT(mutex_owned(&parent->p_smutex));
607
608 child->p_estcpu = child->p_estcpu_inherited = parent->p_estcpu;
609 child->p_forktime = sched_pstats_ticks;
610 }
611
612 /*
613 * scheduler_wait_hook:
614 *
615 * Chargeback parents for the sins of their children.
616 */
617 void
618 sched_proc_exit(struct proc *parent, struct proc *child)
619 {
620 fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
621 fixpt_t estcpu;
622
623 /* XXX Only if parent != init?? */
624
625 mutex_spin_enter(&parent->p_stmutex);
626 estcpu = decay_cpu_batch(loadfac, child->p_estcpu_inherited,
627 sched_pstats_ticks - child->p_forktime);
628 if (child->p_estcpu > estcpu)
629 parent->p_estcpu =
630 ESTCPULIM(parent->p_estcpu + child->p_estcpu - estcpu);
631 mutex_spin_exit(&parent->p_stmutex);
632 }
633
634 void
635 sched_enqueue(struct lwp *l, bool ctxswitch)
636 {
637
638 runqueue_enqueue(&global_queue, l);
639 }
640
641 /*
642 * XXXSMP When LWP dispatch (cpu_switch()) is changed to use sched_dequeue(),
643 * drop of the effective priority level from kernel to user needs to be
644 * moved here from userret(). The assignment in userret() is currently
645 * done unlocked.
646 */
647 void
648 sched_dequeue(struct lwp *l)
649 {
650
651 runqueue_dequeue(&global_queue, l);
652 }
653
654 struct lwp *
655 sched_nextlwp()
656 {
657
658 return runqueue_nextlwp(&global_queue);
659 }
660
661 /* Dummy */
662 void
663 sched_lwp_fork(struct lwp *l)
664 {
665
666 }
667
668 void
669 sched_lwp_exit(struct lwp *l)
670 {
671
672 }
673
674 void
675 sched_slept(struct lwp *l)
676 {
677
678 }
679
680 /* SysCtl */
681
682 SYSCTL_SETUP(sysctl_sched_setup, "sysctl kern.sched subtree setup")
683 {
684 sysctl_createv(clog, 0, NULL, NULL,
685 CTLFLAG_PERMANENT,
686 CTLTYPE_NODE, "kern", NULL,
687 NULL, 0, NULL, 0,
688 CTL_KERN, CTL_EOL);
689 sysctl_createv(clog, 0, NULL, NULL,
690 CTLFLAG_PERMANENT,
691 CTLTYPE_NODE, "sched",
692 SYSCTL_DESCR("Scheduler options"),
693 NULL, 0, NULL, 0,
694 CTL_KERN, KERN_SCHED, CTL_EOL);
695 sysctl_createv(clog, 0, NULL, NULL,
696 CTLFLAG_PERMANENT,
697 CTLTYPE_STRING, "name", NULL,
698 NULL, 0, __UNCONST("4.4BSD"), 0,
699 CTL_KERN, KERN_SCHED, CTL_CREATE, CTL_EOL);
700 }
701
702 #if defined(DDB)
703 void
704 sched_print_runqueue(void (*pr)(const char *, ...))
705 {
706
707 runqueue_print(&global_queue, pr);
708 }
709 #endif /* defined(DDB) */
710