sched_ule.c revision 109864 
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: head/sys/kern/sched_ule.c 109864 2003-01-26 05:23:15Z jeff $
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 */
54static fixpt_t  ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
55SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
56
57static void sched_setup(void *dummy);
58SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL)
59
60/*
61 * These datastructures are allocated within their parent datastructure but
62 * are scheduler specific.
63 */
64
65struct ke_sched {
66	int		ske_slice;
67	struct runq	*ske_runq;
68	/* The following variables are only used for pctcpu calculation */
69	int		ske_ltick;	/* Last tick that we were running on */
70	int		ske_ftick;	/* First tick that we were running on */
71	int		ske_ticks;	/* Tick count */
72};
73#define	ke_slice	ke_sched->ske_slice
74#define	ke_runq		ke_sched->ske_runq
75#define	ke_ltick	ke_sched->ske_ltick
76#define	ke_ftick	ke_sched->ske_ftick
77#define	ke_ticks	ke_sched->ske_ticks
78
79struct kg_sched {
80	int	skg_slptime;
81};
82#define	kg_slptime	kg_sched->skg_slptime
83
84struct td_sched {
85	int	std_slptime;
86};
87#define	td_slptime	td_sched->std_slptime
88
89struct ke_sched ke_sched;
90struct kg_sched kg_sched;
91struct td_sched td_sched;
92
93struct ke_sched *kse0_sched = &ke_sched;
94struct kg_sched *ksegrp0_sched = &kg_sched;
95struct p_sched *proc0_sched = NULL;
96struct td_sched *thread0_sched = &td_sched;
97
98/*
99 * This priority range has 20 priorities on either end that are reachable
100 * only through nice values.
101 */
102#define	SCHED_PRI_NRESV	40
103#define	SCHED_PRI_RANGE	((PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE + 1) - \
104    SCHED_PRI_NRESV)
105
106/*
107 * These determine how sleep time effects the priority of a process.
108 *
109 * SLP_MAX:	Maximum amount of accrued sleep time.
110 * SLP_SCALE:	Scale the number of ticks slept across the dynamic priority
111 *		range.
112 * SLP_TOPRI:	Convert a number of ticks slept into a priority value.
113 * SLP_DECAY:	Reduce the sleep time to 50% for every granted slice.
114 */
115#define	SCHED_SLP_MAX	(hz * 2)
116#define	SCHED_SLP_SCALE(slp)	(((slp) * SCHED_PRI_RANGE) / SCHED_SLP_MAX)
117#define	SCHED_SLP_TOPRI(slp)	(SCHED_PRI_RANGE - SCHED_SLP_SCALE((slp)) + \
118    SCHED_PRI_NRESV / 2)
119#define	SCHED_SLP_DECAY(slp)	((slp) / 2)	/* XXX Multiple kses break */
120
121/*
122 * These parameters and macros determine the size of the time slice that is
123 * granted to each thread.
124 *
125 * SLICE_MIN:	Minimum time slice granted, in units of ticks.
126 * SLICE_MAX:	Maximum time slice granted.
127 * SLICE_RANGE:	Range of available time slices scaled by hz.
128 * SLICE_SCALE:	The number slices granted per unit of pri or slp.
129 * PRI_TOSLICE:	Compute a slice size that is proportional to the priority.
130 * SLP_TOSLICE:	Compute a slice size that is inversely proportional to the
131 *		amount of time slept. (smaller slices for interactive ksegs)
132 * PRI_COMP:	This determines what fraction of the actual slice comes from
133 *		the slice size computed from the priority.
134 * SLP_COMP:	This determines what component of the actual slice comes from
135 *		the slize size computed from the sleep time.
136 */
137#define	SCHED_SLICE_MIN		(hz / 100)
138#define	SCHED_SLICE_MAX		(hz / 10)
139#define	SCHED_SLICE_RANGE	(SCHED_SLICE_MAX - SCHED_SLICE_MIN + 1)
140#define	SCHED_SLICE_SCALE(val, max)	(((val) * SCHED_SLICE_RANGE) / (max))
141#define	SCHED_PRI_TOSLICE(pri)						\
142    (SCHED_SLICE_MAX - SCHED_SLICE_SCALE((pri), SCHED_PRI_RANGE))
143#define	SCHED_SLP_TOSLICE(slp)						\
144    (SCHED_SLICE_MAX - SCHED_SLICE_SCALE((slp), SCHED_SLP_MAX))
145#define	SCHED_SLP_COMP(slice)	(((slice) / 5) * 3)	/* 60% */
146#define	SCHED_PRI_COMP(slice)	(((slice) / 5) * 2)	/* 40% */
147
148/*
149 * This macro determines whether or not the kse belongs on the current or
150 * next run queue.
151 */
152#define	SCHED_CURR(kg)	((kg)->kg_slptime > (hz / 4) || \
153    (kg)->kg_pri_class != PRI_TIMESHARE)
154
155/*
156 * Cpu percentage computation macros and defines.
157 *
158 * SCHED_CPU_TIME:	Number of seconds to average the cpu usage across.
159 * SCHED_CPU_TICKS:	Number of hz ticks to average the cpu usage across.
160 */
161
162#define	SCHED_CPU_TIME	60
163#define	SCHED_CPU_TICKS	(hz * SCHED_CPU_TIME)
164
165/*
166 * kseq - pair of runqs per processor
167 */
168
169struct kseq {
170	struct runq	ksq_runqs[2];
171	struct runq	*ksq_curr;
172	struct runq	*ksq_next;
173	int		ksq_load;	/* Total runnable */
174};
175
176/*
177 * One kse queue per processor.
178 */
179struct kseq	kseq_cpu[MAXCPU];
180
181static int sched_slice(struct ksegrp *kg);
182static int sched_priority(struct ksegrp *kg);
183void sched_pctcpu_update(struct kse *ke);
184int sched_pickcpu(void);
185
186static void
187sched_setup(void *dummy)
188{
189	int i;
190
191	mtx_lock_spin(&sched_lock);
192	/* init kseqs */
193	for (i = 0; i < MAXCPU; i++) {
194		kseq_cpu[i].ksq_load = 0;
195		kseq_cpu[i].ksq_curr = &kseq_cpu[i].ksq_runqs[0];
196		kseq_cpu[i].ksq_next = &kseq_cpu[i].ksq_runqs[1];
197		runq_init(kseq_cpu[i].ksq_curr);
198		runq_init(kseq_cpu[i].ksq_next);
199	}
200	/* CPU0 has proc0 */
201	kseq_cpu[0].ksq_load++;
202	mtx_unlock_spin(&sched_lock);
203}
204
205/*
206 * Scale the scheduling priority according to the "interactivity" of this
207 * process.
208 */
209static int
210sched_priority(struct ksegrp *kg)
211{
212	int pri;
213
214	if (kg->kg_pri_class != PRI_TIMESHARE)
215		return (kg->kg_user_pri);
216
217	pri = SCHED_SLP_TOPRI(kg->kg_slptime);
218	CTR2(KTR_RUNQ, "sched_priority: slptime: %d\tpri: %d",
219	    kg->kg_slptime, pri);
220
221	pri += PRI_MIN_TIMESHARE;
222	pri += kg->kg_nice;
223
224	if (pri > PRI_MAX_TIMESHARE)
225		pri = PRI_MAX_TIMESHARE;
226	else if (pri < PRI_MIN_TIMESHARE)
227		pri = PRI_MIN_TIMESHARE;
228
229	kg->kg_user_pri = pri;
230
231	return (kg->kg_user_pri);
232}
233
234/*
235 * Calculate a time slice based on the process priority.
236 */
237static int
238sched_slice(struct ksegrp *kg)
239{
240	int pslice;
241	int sslice;
242	int slice;
243	int pri;
244
245	pri = kg->kg_user_pri;
246	pri -= PRI_MIN_TIMESHARE;
247	pslice = SCHED_PRI_TOSLICE(pri);
248	sslice = SCHED_SLP_TOSLICE(kg->kg_slptime);
249	slice = SCHED_SLP_COMP(sslice) + SCHED_PRI_COMP(pslice);
250	kg->kg_slptime = SCHED_SLP_DECAY(kg->kg_slptime);
251
252	CTR4(KTR_RUNQ,
253	    "sched_slice: pri: %d\tsslice: %d\tpslice: %d\tslice: %d",
254	    pri, sslice, pslice, slice);
255
256	if (slice < SCHED_SLICE_MIN)
257		slice = SCHED_SLICE_MIN;
258	else if (slice > SCHED_SLICE_MAX)
259		slice = SCHED_SLICE_MAX;
260
261	return (slice);
262}
263
264int
265sched_rr_interval(void)
266{
267	return (SCHED_SLICE_MAX);
268}
269
270void
271sched_pctcpu_update(struct kse *ke)
272{
273	/*
274	 * Adjust counters and watermark for pctcpu calc.
275	 */
276	ke->ke_ticks = (ke->ke_ticks / (ke->ke_ltick - ke->ke_ftick)) *
277		    SCHED_CPU_TICKS;
278	ke->ke_ltick = ticks;
279	ke->ke_ftick = ke->ke_ltick - SCHED_CPU_TICKS;
280}
281
282#ifdef SMP
283int
284sched_pickcpu(void)
285{
286	int cpu;
287	int load;
288	int i;
289
290	if (!smp_started)
291		return (0);
292
293	cpu = PCPU_GET(cpuid);
294	load = kseq_cpu[cpu].ksq_load;
295
296	for (i = 0; i < mp_maxid; i++) {
297		if (CPU_ABSENT(i))
298			continue;
299		if (kseq_cpu[i].ksq_load < load) {
300			cpu = i;
301			load = kseq_cpu[i].ksq_load;
302		}
303	}
304
305	CTR1(KTR_RUNQ, "sched_pickcpu: %d", cpu);
306	return (cpu);
307}
308#else
309int
310sched_pickcpu(void)
311{
312	return (0);
313}
314#endif
315
316void
317sched_prio(struct thread *td, u_char prio)
318{
319	struct kse *ke;
320	struct runq *rq;
321
322	mtx_assert(&sched_lock, MA_OWNED);
323	ke = td->td_kse;
324	td->td_priority = prio;
325
326	if (TD_ON_RUNQ(td)) {
327		rq = ke->ke_runq;
328
329		runq_remove(rq, ke);
330		runq_add(rq, ke);
331	}
332}
333
334void
335sched_switchout(struct thread *td)
336{
337	struct kse *ke;
338
339	mtx_assert(&sched_lock, MA_OWNED);
340
341	ke = td->td_kse;
342
343	td->td_last_kse = ke;
344        td->td_lastcpu = ke->ke_oncpu;
345        ke->ke_flags &= ~KEF_NEEDRESCHED;
346
347	if (TD_IS_RUNNING(td)) {
348		setrunqueue(td);
349		return;
350	} else
351		td->td_kse->ke_runq = NULL;
352
353	/*
354	 * We will not be on the run queue. So we must be
355	 * sleeping or similar.
356	 */
357	if (td->td_proc->p_flag & P_KSES)
358		kse_reassign(ke);
359}
360
361void
362sched_switchin(struct thread *td)
363{
364	/* struct kse *ke = td->td_kse; */
365	mtx_assert(&sched_lock, MA_OWNED);
366
367	td->td_kse->ke_oncpu = PCPU_GET(cpuid); /* XXX */
368	if (td->td_ksegrp->kg_pri_class == PRI_TIMESHARE &&
369	    td->td_priority != td->td_ksegrp->kg_user_pri)
370		curthread->td_kse->ke_flags |= KEF_NEEDRESCHED;
371
372}
373
374void
375sched_nice(struct ksegrp *kg, int nice)
376{
377	struct thread *td;
378
379	kg->kg_nice = nice;
380	sched_priority(kg);
381	FOREACH_THREAD_IN_GROUP(kg, td) {
382		td->td_kse->ke_flags |= KEF_NEEDRESCHED;
383	}
384}
385
386void
387sched_sleep(struct thread *td, u_char prio)
388{
389	mtx_assert(&sched_lock, MA_OWNED);
390
391	td->td_slptime = ticks;
392	td->td_priority = prio;
393
394	/*
395	 * If this is an interactive task clear its queue so it moves back
396	 * on to curr when it wakes up.  Otherwise let it stay on the queue
397	 * that it was assigned to.
398	 */
399	if (SCHED_CURR(td->td_kse->ke_ksegrp))
400		td->td_kse->ke_runq = NULL;
401}
402
403void
404sched_wakeup(struct thread *td)
405{
406	struct ksegrp *kg;
407
408	mtx_assert(&sched_lock, MA_OWNED);
409
410	/*
411	 * Let the kseg know how long we slept for.  This is because process
412	 * interactivity behavior is modeled in the kseg.
413	 */
414	kg = td->td_ksegrp;
415
416	if (td->td_slptime) {
417		kg->kg_slptime += ticks - td->td_slptime;
418		if (kg->kg_slptime > SCHED_SLP_MAX)
419			kg->kg_slptime = SCHED_SLP_MAX;
420		td->td_priority = sched_priority(kg);
421	}
422	td->td_slptime = 0;
423	setrunqueue(td);
424        if (td->td_priority < curthread->td_priority)
425                curthread->td_kse->ke_flags |= KEF_NEEDRESCHED;
426}
427
428/*
429 * Penalize the parent for creating a new child and initialize the child's
430 * priority.
431 */
432void
433sched_fork(struct ksegrp *kg, struct ksegrp *child)
434{
435	struct kse *ckse;
436	struct kse *pkse;
437
438	mtx_assert(&sched_lock, MA_OWNED);
439	ckse = FIRST_KSE_IN_KSEGRP(child);
440	pkse = FIRST_KSE_IN_KSEGRP(kg);
441
442	/* XXX Need something better here */
443	child->kg_slptime = kg->kg_slptime;
444	child->kg_user_pri = kg->kg_user_pri;
445
446	ckse->ke_slice = pkse->ke_slice;
447	ckse->ke_oncpu = sched_pickcpu();
448	ckse->ke_runq = NULL;
449	/*
450	 * Claim that we've been running for one second for statistical
451	 * purposes.
452	 */
453	ckse->ke_ticks = 0;
454	ckse->ke_ltick = ticks;
455	ckse->ke_ftick = ticks - hz;
456}
457
458/*
459 * Return some of the child's priority and interactivity to the parent.
460 */
461void
462sched_exit(struct ksegrp *kg, struct ksegrp *child)
463{
464	struct kseq *kseq;
465	struct kse *ke;
466
467	/* XXX Need something better here */
468	mtx_assert(&sched_lock, MA_OWNED);
469	kg->kg_slptime = child->kg_slptime;
470	sched_priority(kg);
471
472	/*
473	 * We drop the load here so that the running process leaves us with a
474	 * load of at least one.
475	 */
476	ke = FIRST_KSE_IN_KSEGRP(kg);
477	kseq = &kseq_cpu[ke->ke_oncpu];
478	kseq->ksq_load--;
479}
480
481int sched_clock_switches;
482
483void
484sched_clock(struct thread *td)
485{
486	struct kse *ke;
487	struct kse *nke;
488	struct ksegrp *kg;
489	struct kseq *kseq;
490	int cpu;
491
492	cpu = PCPU_GET(cpuid);
493	kseq = &kseq_cpu[cpu];
494
495	mtx_assert(&sched_lock, MA_OWNED);
496	KASSERT((td != NULL), ("schedclock: null thread pointer"));
497	ke = td->td_kse;
498	kg = td->td_ksegrp;
499
500	nke = runq_choose(kseq->ksq_curr);
501
502	if (td->td_kse->ke_flags & KEF_IDLEKSE) {
503#if 0
504		if (nke && nke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) {
505			printf("Idle running with %s on the runq!\n",
506			    nke->ke_proc->p_comm);
507			Debugger("stop");
508		}
509#endif
510		return;
511	}
512	if (nke && nke->ke_thread &&
513	    nke->ke_thread->td_priority < td->td_priority) {
514		sched_clock_switches++;
515		ke->ke_flags |= KEF_NEEDRESCHED;
516	}
517
518	/*
519	 * We used a tick, decrease our total sleep time.  This decreases our
520	 * "interactivity".
521	 */
522	if (kg->kg_slptime)
523		kg->kg_slptime--;
524	/*
525	 * We used up one time slice.
526	 */
527	ke->ke_slice--;
528	/*
529	 * We're out of time, recompute priorities and requeue
530	 */
531	if (ke->ke_slice == 0) {
532		struct kseq *kseq;
533
534		kseq = &kseq_cpu[ke->ke_oncpu];
535
536		td->td_priority = sched_priority(kg);
537		ke->ke_slice = sched_slice(kg);
538		ke->ke_flags |= KEF_NEEDRESCHED;
539		ke->ke_runq = NULL;
540	}
541	ke->ke_ticks += 10000;
542	ke->ke_ltick = ticks;
543	/* Go up to one second beyond our max and then trim back down */
544	if (ke->ke_ftick + SCHED_CPU_TICKS + hz < ke->ke_ltick)
545		sched_pctcpu_update(ke);
546}
547
548int
549sched_runnable(void)
550{
551	struct kseq *kseq;
552	int cpu;
553
554	cpu = PCPU_GET(cpuid);
555	kseq = &kseq_cpu[cpu];
556
557	if (runq_check(kseq->ksq_curr) == 0)
558		return (runq_check(kseq->ksq_next));
559	return (1);
560}
561
562void
563sched_userret(struct thread *td)
564{
565	struct ksegrp *kg;
566
567	kg = td->td_ksegrp;
568
569	if (td->td_priority != kg->kg_user_pri) {
570		mtx_lock_spin(&sched_lock);
571		td->td_priority = kg->kg_user_pri;
572		mtx_unlock_spin(&sched_lock);
573	}
574}
575
576struct kse *
577sched_choose(void)
578{
579	struct kseq *kseq;
580	struct kse *ke;
581	struct runq *swap;
582	int cpu;
583
584	cpu = PCPU_GET(cpuid);
585	kseq = &kseq_cpu[cpu];
586
587	if ((ke = runq_choose(kseq->ksq_curr)) == NULL) {
588		swap = kseq->ksq_curr;
589		kseq->ksq_curr = kseq->ksq_next;
590		kseq->ksq_next = swap;
591		ke = runq_choose(kseq->ksq_curr);
592	}
593	if (ke) {
594		runq_remove(ke->ke_runq, ke);
595		ke->ke_state = KES_THREAD;
596	}
597
598	return (ke);
599}
600
601void
602sched_add(struct kse *ke)
603{
604	struct kseq *kseq;
605	int cpu;
606
607	mtx_assert(&sched_lock, MA_OWNED);
608	KASSERT((ke->ke_thread != NULL), ("runq_add: No thread on KSE"));
609	KASSERT((ke->ke_thread->td_kse != NULL),
610	    ("runq_add: No KSE on thread"));
611	KASSERT(ke->ke_state != KES_ONRUNQ,
612	    ("runq_add: kse %p (%s) already in run queue", ke,
613	    ke->ke_proc->p_comm));
614	KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
615	    ("runq_add: process swapped out"));
616
617	/* cpu = PCPU_GET(cpuid); */
618	cpu = ke->ke_oncpu;
619	kseq = &kseq_cpu[cpu];
620	kseq->ksq_load++;
621
622	if (ke->ke_runq == NULL) {
623		if (SCHED_CURR(ke->ke_ksegrp))
624			ke->ke_runq = kseq->ksq_curr;
625		else
626			ke->ke_runq = kseq->ksq_next;
627	}
628	ke->ke_ksegrp->kg_runq_kses++;
629	ke->ke_state = KES_ONRUNQ;
630
631	runq_add(ke->ke_runq, ke);
632}
633
634void
635sched_rem(struct kse *ke)
636{
637	struct kseq *kseq;
638
639	mtx_assert(&sched_lock, MA_OWNED);
640	/* KASSERT((ke->ke_state == KES_ONRUNQ), ("KSE not on run queue")); */
641
642	kseq = &kseq_cpu[ke->ke_oncpu];
643	kseq->ksq_load--;
644
645	runq_remove(ke->ke_runq, ke);
646	ke->ke_runq = NULL;
647	ke->ke_state = KES_THREAD;
648	ke->ke_ksegrp->kg_runq_kses--;
649}
650
651fixpt_t
652sched_pctcpu(struct kse *ke)
653{
654	fixpt_t pctcpu;
655
656	pctcpu = 0;
657
658	if (ke->ke_ticks) {
659		int rtick;
660
661		/* Update to account for time potentially spent sleeping */
662		ke->ke_ltick = ticks;
663		sched_pctcpu_update(ke);
664
665		/* How many rtick per second ? */
666		rtick = ke->ke_ticks / (SCHED_CPU_TIME * 10000);
667		pctcpu = (FSCALE * ((FSCALE * rtick)/stathz)) >> FSHIFT;
668	}
669
670	ke->ke_proc->p_swtime = ke->ke_ltick - ke->ke_ftick;
671
672	return (pctcpu);
673}
674
675int
676sched_sizeof_kse(void)
677{
678	return (sizeof(struct kse) + sizeof(struct ke_sched));
679}
680
681int
682sched_sizeof_ksegrp(void)
683{
684	return (sizeof(struct ksegrp) + sizeof(struct kg_sched));
685}
686
687int
688sched_sizeof_proc(void)
689{
690	return (sizeof(struct proc));
691}
692
693int
694sched_sizeof_thread(void)
695{
696	return (sizeof(struct thread) + sizeof(struct td_sched));
697}
698