xref: /freebsd/sys/kern/sched_ule.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
1 /*-
2  * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/kernel.h>
32 #include <sys/ktr.h>
33 #include <sys/lock.h>
34 #include <sys/mutex.h>
35 #include <sys/proc.h>
36 #include <sys/sched.h>
37 #include <sys/smp.h>
38 #include <sys/sx.h>
39 #include <sys/sysctl.h>
40 #include <sys/sysproto.h>
41 #include <sys/vmmeter.h>
42 #ifdef DDB
43 #include <ddb/ddb.h>
44 #endif
45 #ifdef KTRACE
46 #include <sys/uio.h>
47 #include <sys/ktrace.h>
48 #endif
49 
50 #include <machine/cpu.h>
51 
52 /* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
53 /* XXX This is bogus compatability crap for ps */
54 static fixpt_t  ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
55 SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
56 
57 static void sched_setup(void *dummy);
58 SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL)
59 
60 #define	SCHED_STRICT_RESCHED 1
61 
62 /*
63  * These datastructures are allocated within their parent datastructure but
64  * are scheduler specific.
65  */
66 
67 struct ke_sched {
68 	int		ske_slice;
69 	struct runq	*ske_runq;
70 	/* The following variables are only used for pctcpu calculation */
71 	int		ske_ltick;	/* Last tick that we were running on */
72 	int		ske_ftick;	/* First tick that we were running on */
73 	int		ske_ticks;	/* Tick count */
74 	u_char		ske_cpu;
75 };
76 #define	ke_slice	ke_sched->ske_slice
77 #define	ke_runq		ke_sched->ske_runq
78 #define	ke_ltick	ke_sched->ske_ltick
79 #define	ke_ftick	ke_sched->ske_ftick
80 #define	ke_ticks	ke_sched->ske_ticks
81 #define	ke_cpu		ke_sched->ske_cpu
82 
83 struct kg_sched {
84 	int	skg_slptime;		/* Number of ticks we vol. slept */
85 	int	skg_runtime;		/* Number of ticks we were running */
86 };
87 #define	kg_slptime	kg_sched->skg_slptime
88 #define	kg_runtime	kg_sched->skg_runtime
89 
90 struct td_sched {
91 	int	std_slptime;
92 	int	std_schedflag;
93 };
94 #define	td_slptime	td_sched->std_slptime
95 #define	td_schedflag	td_sched->std_schedflag
96 
97 #define	TD_SCHED_BLOAD	0x0001		/*
98 					 * thread was counted as being in short
99 					 * term sleep.
100 					 */
101 struct td_sched td_sched;
102 struct ke_sched ke_sched;
103 struct kg_sched kg_sched;
104 
105 struct ke_sched *kse0_sched = &ke_sched;
106 struct kg_sched *ksegrp0_sched = &kg_sched;
107 struct p_sched *proc0_sched = NULL;
108 struct td_sched *thread0_sched = &td_sched;
109 
110 /*
111  * This priority range has 20 priorities on either end that are reachable
112  * only through nice values.
113  */
114 #define	SCHED_PRI_RANGE	(PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE + 1)
115 #define	SCHED_PRI_NRESV	40
116 #define	SCHED_PRI_BASE	(SCHED_PRI_NRESV / 2)
117 #define	SCHED_PRI_DYN	(SCHED_PRI_RANGE - SCHED_PRI_NRESV)
118 #define	SCHED_PRI_DYN_HALF	(SCHED_PRI_DYN / 2)
119 
120 /*
121  * These determine how sleep time effects the priority of a process.
122  *
123  * SLP_RUN_MAX:	Maximum amount of sleep time + run time we'll accumulate
124  *		before throttling back.
125  * SLP_RUN_THORTTLE:	Divisor for reducing slp/run time.
126  * SLP_RATIO:	Compute a bounded ratio of slp time vs run time.
127  * SLP_TOPRI:	Convert a number of ticks slept and ticks ran into a priority
128  */
129 #define	SCHED_SLP_RUN_MAX	((hz * 30) * 1024)
130 #define	SCHED_SLP_RUN_THROTTLE	(10)
131 static __inline int
132 sched_slp_ratio(int b, int s)
133 {
134 	b /= SCHED_PRI_DYN_HALF;
135 	if (b == 0)
136 		return (0);
137 	s /= b;
138 	return (s);
139 }
140 #define	SCHED_SLP_TOPRI(slp, run)					\
141     ((((slp) > (run))?							\
142     sched_slp_ratio((slp), (run)):					\
143     SCHED_PRI_DYN_HALF + (SCHED_PRI_DYN_HALF - sched_slp_ratio((run), (slp))))+ \
144     SCHED_PRI_NRESV / 2)
145 /*
146  * These parameters and macros determine the size of the time slice that is
147  * granted to each thread.
148  *
149  * SLICE_MIN:	Minimum time slice granted, in units of ticks.
150  * SLICE_MAX:	Maximum time slice granted.
151  * SLICE_RANGE:	Range of available time slices scaled by hz.
152  * SLICE_SCALE:	The number slices granted per unit of pri or slp.
153  * PRI_TOSLICE:	Compute a slice size that is proportional to the priority.
154  * SLP_TOSLICE:	Compute a slice size that is inversely proportional to the
155  *		amount of time slept. (smaller slices for interactive ksegs)
156  * PRI_COMP:	This determines what fraction of the actual slice comes from
157  *		the slice size computed from the priority.
158  * SLP_COMP:	This determines what component of the actual slice comes from
159  *		the slize size computed from the sleep time.
160  */
161 #define	SCHED_SLICE_MIN		(hz / 100)
162 #define	SCHED_SLICE_MAX		(hz / 4)
163 #define	SCHED_SLICE_RANGE	(SCHED_SLICE_MAX - SCHED_SLICE_MIN + 1)
164 #define	SCHED_SLICE_SCALE(val, max)	(((val) * SCHED_SLICE_RANGE) / (max))
165 #define	SCHED_PRI_TOSLICE(pri)						\
166     (SCHED_SLICE_MAX - SCHED_SLICE_SCALE((pri), SCHED_PRI_RANGE))
167 #define	SCHED_SLP_TOSLICE(slp)						\
168     (SCHED_SLICE_MAX - SCHED_SLICE_SCALE((slp), SCHED_PRI_DYN))
169 #define	SCHED_SLP_COMP(slice)	(((slice) / 5) * 3)	/* 60% */
170 #define	SCHED_PRI_COMP(slice)	(((slice) / 5) * 2)	/* 40% */
171 
172 /*
173  * This macro determines whether or not the kse belongs on the current or
174  * next run queue.
175  *
176  * XXX nice value should effect how interactive a kg is.
177  */
178 #define	SCHED_CURR(kg)	(((kg)->kg_slptime > (kg)->kg_runtime &&	\
179 	sched_slp_ratio((kg)->kg_slptime, (kg)->kg_runtime) > 4))
180 
181 /*
182  * Cpu percentage computation macros and defines.
183  *
184  * SCHED_CPU_TIME:	Number of seconds to average the cpu usage across.
185  * SCHED_CPU_TICKS:	Number of hz ticks to average the cpu usage across.
186  */
187 
188 #define	SCHED_CPU_TIME	60
189 #define	SCHED_CPU_TICKS	(hz * SCHED_CPU_TIME)
190 
191 /*
192  * kseq - pair of runqs per processor
193  */
194 
195 struct kseq {
196 	struct runq	ksq_runqs[2];
197 	struct runq	*ksq_curr;
198 	struct runq	*ksq_next;
199 	int		ksq_load;	/* Total runnable */
200 #ifdef SMP
201 	unsigned int	ksq_rslices;	/* Slices on run queue */
202 	unsigned int	ksq_bload;	/* Threads waiting on IO */
203 #endif
204 };
205 
206 /*
207  * One kse queue per processor.
208  */
209 #ifdef SMP
210 struct kseq	kseq_cpu[MAXCPU];
211 #define	KSEQ_SELF()	(&kseq_cpu[PCPU_GET(cpuid)])
212 #define	KSEQ_CPU(x)	(&kseq_cpu[(x)])
213 #else
214 struct kseq	kseq_cpu;
215 #define	KSEQ_SELF()	(&kseq_cpu)
216 #define	KSEQ_CPU(x)	(&kseq_cpu)
217 #endif
218 
219 static int sched_slice(struct ksegrp *kg);
220 static int sched_priority(struct ksegrp *kg);
221 void sched_pctcpu_update(struct kse *ke);
222 int sched_pickcpu(void);
223 
224 /* Operations on per processor queues */
225 static struct kse * kseq_choose(struct kseq *kseq);
226 static void kseq_setup(struct kseq *kseq);
227 static __inline void kseq_add(struct kseq *kseq, struct kse *ke);
228 static __inline void kseq_rem(struct kseq *kseq, struct kse *ke);
229 #ifdef SMP
230 static __inline void kseq_sleep(struct kseq *kseq, struct kse *ke);
231 static __inline void kseq_wakeup(struct kseq *kseq, struct kse *ke);
232 struct kseq * kseq_load_highest(void);
233 #endif
234 
235 static __inline void
236 kseq_add(struct kseq *kseq, struct kse *ke)
237 {
238 	runq_add(ke->ke_runq, ke);
239 	kseq->ksq_load++;
240 #ifdef SMP
241 	kseq->ksq_rslices += ke->ke_slice;
242 #endif
243 }
244 static __inline void
245 kseq_rem(struct kseq *kseq, struct kse *ke)
246 {
247 	kseq->ksq_load--;
248 	runq_remove(ke->ke_runq, ke);
249 #ifdef SMP
250 	kseq->ksq_rslices -= ke->ke_slice;
251 #endif
252 }
253 
254 #ifdef SMP
255 static __inline void
256 kseq_sleep(struct kseq *kseq, struct kse *ke)
257 {
258 	kseq->ksq_bload++;
259 }
260 
261 static __inline void
262 kseq_wakeup(struct kseq *kseq, struct kse *ke)
263 {
264 	kseq->ksq_bload--;
265 }
266 
267 struct kseq *
268 kseq_load_highest(void)
269 {
270 	struct kseq *kseq;
271 	int load;
272 	int cpu;
273 	int i;
274 
275 	cpu = 0;
276 	load = 0;
277 
278 	for (i = 0; i < mp_maxid; i++) {
279 		if (CPU_ABSENT(i))
280 			continue;
281 		kseq = KSEQ_CPU(i);
282 		if (kseq->ksq_load > load) {
283 			load = kseq->ksq_load;
284 			cpu = i;
285 		}
286 	}
287 	if (load)
288 		return (KSEQ_CPU(cpu));
289 
290 	return (NULL);
291 }
292 #endif
293 
294 struct kse *
295 kseq_choose(struct kseq *kseq)
296 {
297 	struct kse *ke;
298 	struct runq *swap;
299 
300 	if ((ke = runq_choose(kseq->ksq_curr)) == NULL) {
301 		swap = kseq->ksq_curr;
302 		kseq->ksq_curr = kseq->ksq_next;
303 		kseq->ksq_next = swap;
304 		ke = runq_choose(kseq->ksq_curr);
305 	}
306 
307 	return (ke);
308 }
309 
310 
311 static void
312 kseq_setup(struct kseq *kseq)
313 {
314 	kseq->ksq_curr = &kseq->ksq_runqs[0];
315 	kseq->ksq_next = &kseq->ksq_runqs[1];
316 	runq_init(kseq->ksq_curr);
317 	runq_init(kseq->ksq_next);
318 	kseq->ksq_load = 0;
319 #ifdef SMP
320 	kseq->ksq_rslices = 0;
321 	kseq->ksq_bload = 0;
322 #endif
323 }
324 
325 static void
326 sched_setup(void *dummy)
327 {
328 	int i;
329 
330 	mtx_lock_spin(&sched_lock);
331 	/* init kseqs */
332 	for (i = 0; i < MAXCPU; i++)
333 		kseq_setup(KSEQ_CPU(i));
334 	mtx_unlock_spin(&sched_lock);
335 }
336 
337 /*
338  * Scale the scheduling priority according to the "interactivity" of this
339  * process.
340  */
341 static int
342 sched_priority(struct ksegrp *kg)
343 {
344 	int pri;
345 
346 	if (kg->kg_pri_class != PRI_TIMESHARE)
347 		return (kg->kg_user_pri);
348 
349 	pri = SCHED_SLP_TOPRI(kg->kg_slptime, kg->kg_runtime);
350 	CTR2(KTR_RUNQ, "sched_priority: slptime: %d\tpri: %d",
351 	    kg->kg_slptime, pri);
352 
353 	pri += PRI_MIN_TIMESHARE;
354 	pri += kg->kg_nice;
355 
356 	if (pri > PRI_MAX_TIMESHARE)
357 		pri = PRI_MAX_TIMESHARE;
358 	else if (pri < PRI_MIN_TIMESHARE)
359 		pri = PRI_MIN_TIMESHARE;
360 
361 	kg->kg_user_pri = pri;
362 
363 	return (kg->kg_user_pri);
364 }
365 
366 /*
367  * Calculate a time slice based on the process priority.
368  */
369 static int
370 sched_slice(struct ksegrp *kg)
371 {
372 	int pslice;
373 	int sslice;
374 	int slice;
375 	int pri;
376 
377 	pri = kg->kg_user_pri;
378 	pri -= PRI_MIN_TIMESHARE;
379 	pslice = SCHED_PRI_TOSLICE(pri);
380 	sslice = SCHED_PRI_TOSLICE(SCHED_SLP_TOPRI(kg->kg_slptime, kg->kg_runtime));
381 /*
382 SCHED_SLP_TOSLICE(SCHED_SLP_RATIO(
383 	    kg->kg_slptime, kg->kg_runtime));
384 */
385 	slice = SCHED_SLP_COMP(sslice) + SCHED_PRI_COMP(pslice);
386 
387 	CTR4(KTR_RUNQ,
388 	    "sched_slice: pri: %d\tsslice: %d\tpslice: %d\tslice: %d",
389 	    pri, sslice, pslice, slice);
390 
391 	if (slice < SCHED_SLICE_MIN)
392 		slice = SCHED_SLICE_MIN;
393 	else if (slice > SCHED_SLICE_MAX)
394 		slice = SCHED_SLICE_MAX;
395 
396 	/*
397 	 * Every time we grant a new slice check to see if we need to scale
398 	 * back the slp and run time in the kg.  This will cause us to forget
399 	 * old interactivity while maintaining the current ratio.
400 	 */
401 	if ((kg->kg_runtime + kg->kg_slptime) >  SCHED_SLP_RUN_MAX) {
402 		kg->kg_runtime /= SCHED_SLP_RUN_THROTTLE;
403 		kg->kg_slptime /= SCHED_SLP_RUN_THROTTLE;
404 	}
405 
406 	return (slice);
407 }
408 
409 int
410 sched_rr_interval(void)
411 {
412 	return (SCHED_SLICE_MAX);
413 }
414 
415 void
416 sched_pctcpu_update(struct kse *ke)
417 {
418 	/*
419 	 * Adjust counters and watermark for pctcpu calc.
420 	 */
421 	/*
422 	 * Shift the tick count out so that the divide doesn't round away
423 	 * our results.
424 	 */
425 	ke->ke_ticks <<= 10;
426 	ke->ke_ticks = (ke->ke_ticks / (ke->ke_ltick - ke->ke_ftick)) *
427 		    SCHED_CPU_TICKS;
428 	ke->ke_ticks >>= 10;
429 	ke->ke_ltick = ticks;
430 	ke->ke_ftick = ke->ke_ltick - SCHED_CPU_TICKS;
431 }
432 
433 #ifdef SMP
434 /* XXX Should be changed to kseq_load_lowest() */
435 int
436 sched_pickcpu(void)
437 {
438 	struct kseq *kseq;
439 	int load;
440 	int cpu;
441 	int i;
442 
443 	if (!smp_started)
444 		return (0);
445 
446 	load = 0;
447 	cpu = 0;
448 
449 	for (i = 0; i < mp_maxid; i++) {
450 		if (CPU_ABSENT(i))
451 			continue;
452 		kseq = KSEQ_CPU(i);
453 		if (kseq->ksq_load < load) {
454 			cpu = i;
455 			load = kseq->ksq_load;
456 		}
457 	}
458 
459 	CTR1(KTR_RUNQ, "sched_pickcpu: %d", cpu);
460 	return (cpu);
461 }
462 #else
463 int
464 sched_pickcpu(void)
465 {
466 	return (0);
467 }
468 #endif
469 
470 void
471 sched_prio(struct thread *td, u_char prio)
472 {
473 	struct kse *ke;
474 	struct runq *rq;
475 
476 	mtx_assert(&sched_lock, MA_OWNED);
477 	ke = td->td_kse;
478 	td->td_priority = prio;
479 
480 	if (TD_ON_RUNQ(td)) {
481 		rq = ke->ke_runq;
482 
483 		runq_remove(rq, ke);
484 		runq_add(rq, ke);
485 	}
486 }
487 
488 void
489 sched_switchout(struct thread *td)
490 {
491 	struct kse *ke;
492 
493 	mtx_assert(&sched_lock, MA_OWNED);
494 
495 	ke = td->td_kse;
496 
497 	td->td_last_kse = ke;
498         td->td_lastcpu = ke->ke_oncpu;
499 	ke->ke_oncpu = NOCPU;
500         td->td_flags &= ~TDF_NEEDRESCHED;
501 
502 	if (TD_IS_RUNNING(td)) {
503 		setrunqueue(td);
504 		return;
505 	} else
506 		td->td_kse->ke_runq = NULL;
507 
508 	/*
509 	 * We will not be on the run queue. So we must be
510 	 * sleeping or similar.
511 	 */
512 	if (td->td_proc->p_flag & P_THREADED)
513 		kse_reassign(ke);
514 }
515 
516 void
517 sched_switchin(struct thread *td)
518 {
519 	/* struct kse *ke = td->td_kse; */
520 	mtx_assert(&sched_lock, MA_OWNED);
521 
522 	td->td_kse->ke_oncpu = PCPU_GET(cpuid);
523 #if SCHED_STRICT_RESCHED
524 	if (td->td_ksegrp->kg_pri_class == PRI_TIMESHARE &&
525 	    td->td_priority != td->td_ksegrp->kg_user_pri)
526 		curthread->td_flags |= TDF_NEEDRESCHED;
527 #endif
528 }
529 
530 void
531 sched_nice(struct ksegrp *kg, int nice)
532 {
533 	struct thread *td;
534 
535 	kg->kg_nice = nice;
536 	sched_priority(kg);
537 	FOREACH_THREAD_IN_GROUP(kg, td) {
538 		td->td_flags |= TDF_NEEDRESCHED;
539 	}
540 }
541 
542 void
543 sched_sleep(struct thread *td, u_char prio)
544 {
545 	mtx_assert(&sched_lock, MA_OWNED);
546 
547 	td->td_slptime = ticks;
548 	td->td_priority = prio;
549 
550 	/*
551 	 * If this is an interactive task clear its queue so it moves back
552 	 * on to curr when it wakes up.  Otherwise let it stay on the queue
553 	 * that it was assigned to.
554 	 */
555 	if (SCHED_CURR(td->td_kse->ke_ksegrp))
556 		td->td_kse->ke_runq = NULL;
557 #ifdef SMP
558 	if (td->td_priority < PZERO) {
559 		kseq_sleep(KSEQ_CPU(td->td_kse->ke_cpu), td->td_kse);
560 		td->td_schedflag |= TD_SCHED_BLOAD;
561 	}
562 #endif
563 }
564 
565 void
566 sched_wakeup(struct thread *td)
567 {
568 	mtx_assert(&sched_lock, MA_OWNED);
569 
570 	/*
571 	 * Let the kseg know how long we slept for.  This is because process
572 	 * interactivity behavior is modeled in the kseg.
573 	 */
574 	if (td->td_slptime) {
575 		struct ksegrp *kg;
576 
577 		kg = td->td_ksegrp;
578 		kg->kg_slptime += (ticks - td->td_slptime) * 1024;
579 		sched_priority(kg);
580 		td->td_slptime = 0;
581 	}
582 #ifdef SMP
583 	if (td->td_priority < PZERO && td->td_schedflag & TD_SCHED_BLOAD) {
584 		kseq_wakeup(KSEQ_CPU(td->td_kse->ke_cpu), td->td_kse);
585 		td->td_schedflag &= ~TD_SCHED_BLOAD;
586 	}
587 #endif
588 	setrunqueue(td);
589 #if SCHED_STRICT_RESCHED
590         if (td->td_priority < curthread->td_priority)
591                 curthread->td_flags |= TDF_NEEDRESCHED;
592 #endif
593 }
594 
595 /*
596  * Penalize the parent for creating a new child and initialize the child's
597  * priority.
598  */
599 void
600 sched_fork(struct ksegrp *kg, struct ksegrp *child)
601 {
602 	struct kse *ckse;
603 	struct kse *pkse;
604 
605 	mtx_assert(&sched_lock, MA_OWNED);
606 	ckse = FIRST_KSE_IN_KSEGRP(child);
607 	pkse = FIRST_KSE_IN_KSEGRP(kg);
608 
609 	/* XXX Need something better here */
610 	if (kg->kg_slptime > kg->kg_runtime) {
611 		child->kg_slptime = SCHED_PRI_DYN;
612 		child->kg_runtime = kg->kg_slptime / SCHED_PRI_DYN;
613 	} else {
614 		child->kg_runtime = SCHED_PRI_DYN;
615 		child->kg_slptime = kg->kg_runtime / SCHED_PRI_DYN;
616 	}
617 #if 0
618 	child->kg_slptime = kg->kg_slptime;
619 	child->kg_runtime = kg->kg_runtime;
620 #endif
621 	child->kg_user_pri = kg->kg_user_pri;
622 
623 #if 0
624 	if (pkse->ke_cpu != PCPU_GET(cpuid)) {
625 		printf("pkse->ke_cpu = %d\n", pkse->ke_cpu);
626 		printf("cpuid = %d", PCPU_GET(cpuid));
627 		Debugger("stop");
628 	}
629 #endif
630 
631 	ckse->ke_slice = pkse->ke_slice;
632 	ckse->ke_cpu = pkse->ke_cpu; /* sched_pickcpu(); */
633 	ckse->ke_runq = NULL;
634 	/*
635 	 * Claim that we've been running for one second for statistical
636 	 * purposes.
637 	 */
638 	ckse->ke_ticks = 0;
639 	ckse->ke_ltick = ticks;
640 	ckse->ke_ftick = ticks - hz;
641 }
642 
643 /*
644  * Return some of the child's priority and interactivity to the parent.
645  */
646 void
647 sched_exit(struct ksegrp *kg, struct ksegrp *child)
648 {
649 	/* XXX Need something better here */
650 	mtx_assert(&sched_lock, MA_OWNED);
651 	kg->kg_slptime = child->kg_slptime;
652 	kg->kg_runtime = child->kg_runtime;
653 	sched_priority(kg);
654 }
655 
656 void
657 sched_clock(struct thread *td)
658 {
659 	struct kse *ke;
660 #if SCHED_STRICT_RESCHED
661 	struct kse *nke;
662 	struct kseq *kseq;
663 #endif
664 	struct ksegrp *kg;
665 
666 
667 	ke = td->td_kse;
668 	kg = td->td_ksegrp;
669 
670 	mtx_assert(&sched_lock, MA_OWNED);
671 	KASSERT((td != NULL), ("schedclock: null thread pointer"));
672 
673 	/* Adjust ticks for pctcpu */
674 	ke->ke_ticks++;
675 	ke->ke_ltick = ticks;
676 	/* Go up to one second beyond our max and then trim back down */
677 	if (ke->ke_ftick + SCHED_CPU_TICKS + hz < ke->ke_ltick)
678 		sched_pctcpu_update(ke);
679 
680 	if (td->td_kse->ke_flags & KEF_IDLEKSE)
681 		return;
682 
683 	/*
684 	 * Check for a higher priority task on the run queue.  This can happen
685 	 * on SMP if another processor woke up a process on our runq.
686 	 */
687 #if SCHED_STRICT_RESCHED
688 	kseq = KSEQ_SELF();
689 	nke = runq_choose(kseq->ksq_curr);
690 
691 	if (nke && nke->ke_thread &&
692 	    nke->ke_thread->td_priority < td->td_priority)
693 		td->td_flags |= TDF_NEEDRESCHED;
694 #endif
695 	/*
696 	 * We used a tick charge it to the ksegrp so that we can compute our
697 	 * "interactivity".
698 	 */
699 	kg->kg_runtime += 1024;
700 
701 	/*
702 	 * We used up one time slice.
703 	 */
704 	ke->ke_slice--;
705 	/*
706 	 * We're out of time, recompute priorities and requeue
707 	 */
708 	if (ke->ke_slice == 0) {
709 		td->td_priority = sched_priority(kg);
710 		ke->ke_slice = sched_slice(kg);
711 		td->td_flags |= TDF_NEEDRESCHED;
712 		ke->ke_runq = NULL;
713 	}
714 }
715 
716 int
717 sched_runnable(void)
718 {
719 	struct kseq *kseq;
720 
721 	kseq = KSEQ_SELF();
722 
723 	if (kseq->ksq_load)
724 		return (1);
725 #ifdef SMP
726 	/*
727 	 * For SMP we may steal other processor's KSEs.  Just search until we
728 	 * verify that at least on other cpu has a runnable task.
729 	 */
730 	if (smp_started) {
731 		int i;
732 
733 #if 0
734 		if (kseq->ksq_bload)
735 			return (0);
736 #endif
737 
738 		for (i = 0; i < mp_maxid; i++) {
739 			if (CPU_ABSENT(i))
740 				continue;
741 			kseq = KSEQ_CPU(i);
742 			if (kseq->ksq_load)
743 				return (1);
744 		}
745 	}
746 #endif
747 	return (0);
748 }
749 
750 void
751 sched_userret(struct thread *td)
752 {
753 	struct ksegrp *kg;
754 
755 	kg = td->td_ksegrp;
756 
757 	if (td->td_priority != kg->kg_user_pri) {
758 		mtx_lock_spin(&sched_lock);
759 		td->td_priority = kg->kg_user_pri;
760 		mtx_unlock_spin(&sched_lock);
761 	}
762 }
763 
764 struct kse *
765 sched_choose(void)
766 {
767 	struct kseq *kseq;
768 	struct kse *ke;
769 
770 	kseq = KSEQ_SELF();
771 	ke = kseq_choose(kseq);
772 
773 	if (ke) {
774 		ke->ke_state = KES_THREAD;
775 		kseq_rem(kseq, ke);
776 	}
777 
778 #ifdef SMP
779 	if (ke == NULL && smp_started) {
780 #if 0
781 		if (kseq->ksq_bload)
782 			return (NULL);
783 #endif
784 		/*
785 		 * Find the cpu with the highest load and steal one proc.
786 		 */
787 		kseq = kseq_load_highest();
788 		if (kseq == NULL)
789 			return (NULL);
790 		ke = kseq_choose(kseq);
791 		kseq_rem(kseq, ke);
792 
793 		ke->ke_state = KES_THREAD;
794 		ke->ke_runq = NULL;
795 		ke->ke_cpu = PCPU_GET(cpuid);
796 	}
797 #endif
798 	return (ke);
799 }
800 
801 void
802 sched_add(struct kse *ke)
803 {
804 	struct kseq *kseq;
805 
806 	mtx_assert(&sched_lock, MA_OWNED);
807 	KASSERT((ke->ke_thread != NULL), ("sched_add: No thread on KSE"));
808 	KASSERT((ke->ke_thread->td_kse != NULL),
809 	    ("sched_add: No KSE on thread"));
810 	KASSERT(ke->ke_state != KES_ONRUNQ,
811 	    ("sched_add: kse %p (%s) already in run queue", ke,
812 	    ke->ke_proc->p_comm));
813 	KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
814 	    ("sched_add: process swapped out"));
815 
816 	/*
817 	 * Timeshare threads get placed on the appropriate queue on their
818 	 * bound cpu.
819 	 */
820 	if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) {
821 		kseq = KSEQ_CPU(ke->ke_cpu);
822 
823 		if (ke->ke_runq == NULL) {
824 			if (SCHED_CURR(ke->ke_ksegrp))
825 				ke->ke_runq = kseq->ksq_curr;
826 			else
827 				ke->ke_runq = kseq->ksq_next;
828 		}
829 	/*
830 	 * If we're a real-time or interrupt thread place us on the curr
831 	 * queue for the current processor.  Hopefully this will yield the
832 	 * lowest latency response.
833 	 */
834 	} else {
835 		kseq = KSEQ_SELF();
836 		ke->ke_runq = kseq->ksq_curr;
837 	}
838 	ke->ke_ksegrp->kg_runq_kses++;
839 	ke->ke_state = KES_ONRUNQ;
840 
841 	kseq_add(kseq, ke);
842 }
843 
844 void
845 sched_rem(struct kse *ke)
846 {
847 	mtx_assert(&sched_lock, MA_OWNED);
848 	/* KASSERT((ke->ke_state == KES_ONRUNQ), ("KSE not on run queue")); */
849 
850 	ke->ke_runq = NULL;
851 	ke->ke_state = KES_THREAD;
852 	ke->ke_ksegrp->kg_runq_kses--;
853 
854 	kseq_rem(KSEQ_CPU(ke->ke_cpu), ke);
855 }
856 
857 fixpt_t
858 sched_pctcpu(struct kse *ke)
859 {
860 	fixpt_t pctcpu;
861 	int realstathz;
862 
863 	pctcpu = 0;
864 	realstathz = stathz ? stathz : hz;
865 
866 	if (ke->ke_ticks) {
867 		int rtick;
868 
869 		/* Update to account for time potentially spent sleeping */
870 		ke->ke_ltick = ticks;
871 		sched_pctcpu_update(ke);
872 
873 		/* How many rtick per second ? */
874 		rtick = ke->ke_ticks / SCHED_CPU_TIME;
875 		pctcpu = (FSCALE * ((FSCALE * rtick)/realstathz)) >> FSHIFT;
876 	}
877 
878 	ke->ke_proc->p_swtime = ke->ke_ltick - ke->ke_ftick;
879 
880 	return (pctcpu);
881 }
882 
883 int
884 sched_sizeof_kse(void)
885 {
886 	return (sizeof(struct kse) + sizeof(struct ke_sched));
887 }
888 
889 int
890 sched_sizeof_ksegrp(void)
891 {
892 	return (sizeof(struct ksegrp) + sizeof(struct kg_sched));
893 }
894 
895 int
896 sched_sizeof_proc(void)
897 {
898 	return (sizeof(struct proc));
899 }
900 
901 int
902 sched_sizeof_thread(void)
903 {
904 	return (sizeof(struct thread) + sizeof(struct td_sched));
905 }
906