kern_switch.c revision 132543
1/*
2 * Copyright (c) 2001 Jake Burkholder <jake@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, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27/***
28Here is the logic..
29
30If there are N processors, then there are at most N KSEs (kernel
31schedulable entities) working to process threads that belong to a
32KSEGROUP (kg). If there are X of these KSEs actually running at the
33moment in question, then there are at most M (N-X) of these KSEs on
34the run queue, as running KSEs are not on the queue.
35
36Runnable threads are queued off the KSEGROUP in priority order.
37If there are M or more threads runnable, the top M threads
38(by priority) are 'preassigned' to the M KSEs not running. The KSEs take
39their priority from those threads and are put on the run queue.
40
41The last thread that had a priority high enough to have a KSE associated
42with it, AND IS ON THE RUN QUEUE is pointed to by
43kg->kg_last_assigned. If no threads queued off the KSEGROUP have KSEs
44assigned as all the available KSEs are activly running, or because there
45are no threads queued, that pointer is NULL.
46
47When a KSE is removed from the run queue to become runnable, we know
48it was associated with the highest priority thread in the queue (at the head
49of the queue). If it is also the last assigned we know M was 1 and must
50now be 0. Since the thread is no longer queued that pointer must be
51removed from it. Since we know there were no more KSEs available,
52(M was 1 and is now 0) and since we are not FREEING our KSE
53but using it, we know there are STILL no more KSEs available, we can prove
54that the next thread in the ksegrp list will not have a KSE to assign to
55it, so we can show that the pointer must be made 'invalid' (NULL).
56
57The pointer exists so that when a new thread is made runnable, it can
58have its priority compared with the last assigned thread to see if
59it should 'steal' its KSE or not.. i.e. is it 'earlier'
60on the list than that thread or later.. If it's earlier, then the KSE is
61removed from the last assigned (which is now not assigned a KSE)
62and reassigned to the new thread, which is placed earlier in the list.
63The pointer is then backed up to the previous thread (which may or may not
64be the new thread).
65
66When a thread sleeps or is removed, the KSE becomes available and if there
67are queued threads that are not assigned KSEs, the highest priority one of
68them is assigned the KSE, which is then placed back on the run queue at
69the approipriate place, and the kg->kg_last_assigned pointer is adjusted down
70to point to it.
71
72The following diagram shows 2 KSEs and 3 threads from a single process.
73
74 RUNQ: --->KSE---KSE--...    (KSEs queued at priorities from threads)
75              \    \____
76               \        \
77    KSEGROUP---thread--thread--thread    (queued in priority order)
78        \                 /
79         \_______________/
80          (last_assigned)
81
82The result of this scheme is that the M available KSEs are always
83queued at the priorities they have inherrited from the M highest priority
84threads for that KSEGROUP. If this situation changes, the KSEs are
85reassigned to keep this true.
86***/
87
88#include <sys/cdefs.h>
89__FBSDID("$FreeBSD: head/sys/kern/kern_switch.c 132543 2004-07-22 14:32:48Z scottl $");
90
91#include "opt_full_preemption.h"
92
93#include <sys/param.h>
94#include <sys/systm.h>
95#include <sys/kdb.h>
96#include <sys/kernel.h>
97#include <sys/ktr.h>
98#include <sys/lock.h>
99#include <sys/mutex.h>
100#include <sys/proc.h>
101#include <sys/queue.h>
102#include <sys/sched.h>
103#if defined(SMP) && (defined(__i386__) || defined(__amd64__))
104#include <sys/smp.h>
105#endif
106#include <machine/critical.h>
107
108CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS);
109
110void panc(char *string1, char *string2);
111
112#if 0
113static void runq_readjust(struct runq *rq, struct kse *ke);
114#endif
115/************************************************************************
116 * Functions that manipulate runnability from a thread perspective.	*
117 ************************************************************************/
118/*
119 * Select the KSE that will be run next.  From that find the thread, and
120 * remove it from the KSEGRP's run queue.  If there is thread clustering,
121 * this will be what does it.
122 */
123struct thread *
124choosethread(void)
125{
126	struct kse *ke;
127	struct thread *td;
128	struct ksegrp *kg;
129
130#if defined(SMP) && (defined(__i386__) || defined(__amd64__))
131	if (smp_active == 0 && PCPU_GET(cpuid) != 0) {
132		/* Shutting down, run idlethread on AP's */
133		td = PCPU_GET(idlethread);
134		ke = td->td_kse;
135		CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
136		ke->ke_flags |= KEF_DIDRUN;
137		TD_SET_RUNNING(td);
138		return (td);
139	}
140#endif
141
142retry:
143	ke = sched_choose();
144	if (ke) {
145		td = ke->ke_thread;
146		KASSERT((td->td_kse == ke), ("kse/thread mismatch"));
147		kg = ke->ke_ksegrp;
148		if (td->td_proc->p_flag & P_SA) {
149			if (kg->kg_last_assigned == td) {
150				kg->kg_last_assigned = TAILQ_PREV(td,
151				    threadqueue, td_runq);
152			}
153			TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
154		}
155		kg->kg_runnable--;
156		CTR2(KTR_RUNQ, "choosethread: td=%p pri=%d",
157		    td, td->td_priority);
158	} else {
159		/* Simulate runq_choose() having returned the idle thread */
160		td = PCPU_GET(idlethread);
161		ke = td->td_kse;
162		CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
163	}
164	ke->ke_flags |= KEF_DIDRUN;
165
166	/*
167	 * If we are in panic, only allow system threads,
168	 * plus the one we are running in, to be run.
169	 */
170	if (panicstr && ((td->td_proc->p_flag & P_SYSTEM) == 0 &&
171	    (td->td_flags & TDF_INPANIC) == 0)) {
172		/* note that it is no longer on the run queue */
173		TD_SET_CAN_RUN(td);
174		goto retry;
175	}
176
177	TD_SET_RUNNING(td);
178	return (td);
179}
180
181/*
182 * Given a surplus KSE, either assign a new runable thread to it
183 * (and put it in the run queue) or put it in the ksegrp's idle KSE list.
184 * Assumes that the original thread is not runnable.
185 */
186void
187kse_reassign(struct kse *ke)
188{
189	struct ksegrp *kg;
190	struct thread *td;
191	struct thread *original;
192
193	mtx_assert(&sched_lock, MA_OWNED);
194	original = ke->ke_thread;
195	KASSERT(original == NULL || TD_IS_INHIBITED(original),
196    	    ("reassigning KSE with runnable thread"));
197	kg = ke->ke_ksegrp;
198	if (original)
199		original->td_kse = NULL;
200
201	/*
202	 * Find the first unassigned thread
203	 */
204	if ((td = kg->kg_last_assigned) != NULL)
205		td = TAILQ_NEXT(td, td_runq);
206	else
207		td = TAILQ_FIRST(&kg->kg_runq);
208
209	/*
210	 * If we found one, assign it the kse, otherwise idle the kse.
211	 */
212	if (td) {
213		kg->kg_last_assigned = td;
214		td->td_kse = ke;
215		ke->ke_thread = td;
216		sched_add(td);
217		CTR2(KTR_RUNQ, "kse_reassign: ke%p -> td%p", ke, td);
218		return;
219	}
220
221	ke->ke_state = KES_IDLE;
222	ke->ke_thread = NULL;
223	TAILQ_INSERT_TAIL(&kg->kg_iq, ke, ke_kgrlist);
224	kg->kg_idle_kses++;
225	CTR1(KTR_RUNQ, "kse_reassign: ke%p on idle queue", ke);
226	return;
227}
228
229#if 0
230/*
231 * Remove a thread from its KSEGRP's run queue.
232 * This in turn may remove it from a KSE if it was already assigned
233 * to one, possibly causing a new thread to be assigned to the KSE
234 * and the KSE getting a new priority.
235 */
236static void
237remrunqueue(struct thread *td)
238{
239	struct thread *td2, *td3;
240	struct ksegrp *kg;
241	struct kse *ke;
242
243	mtx_assert(&sched_lock, MA_OWNED);
244	KASSERT((TD_ON_RUNQ(td)), ("remrunqueue: Bad state on run queue"));
245	kg = td->td_ksegrp;
246	ke = td->td_kse;
247	CTR1(KTR_RUNQ, "remrunqueue: td%p", td);
248	kg->kg_runnable--;
249	TD_SET_CAN_RUN(td);
250	/*
251	 * If it is not a threaded process, take the shortcut.
252	 */
253	if ((td->td_proc->p_flag & P_SA) == 0) {
254		/* Bring its kse with it, leave the thread attached */
255		sched_rem(td);
256		ke->ke_state = KES_THREAD;
257		return;
258	}
259   	td3 = TAILQ_PREV(td, threadqueue, td_runq);
260	TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
261	if (ke) {
262		/*
263		 * This thread has been assigned to a KSE.
264		 * We need to dissociate it and try assign the
265		 * KSE to the next available thread. Then, we should
266		 * see if we need to move the KSE in the run queues.
267		 */
268		sched_rem(td);
269		ke->ke_state = KES_THREAD;
270		td2 = kg->kg_last_assigned;
271		KASSERT((td2 != NULL), ("last assigned has wrong value"));
272		if (td2 == td)
273			kg->kg_last_assigned = td3;
274		kse_reassign(ke);
275	}
276}
277#endif
278
279/*
280 * Change the priority of a thread that is on the run queue.
281 */
282void
283adjustrunqueue( struct thread *td, int newpri)
284{
285	struct ksegrp *kg;
286	struct kse *ke;
287
288	mtx_assert(&sched_lock, MA_OWNED);
289	KASSERT((TD_ON_RUNQ(td)), ("adjustrunqueue: Bad state on run queue"));
290
291	ke = td->td_kse;
292	CTR1(KTR_RUNQ, "adjustrunqueue: td%p", td);
293	/*
294	 * If it is not a threaded process, take the shortcut.
295	 */
296	if ((td->td_proc->p_flag & P_SA) == 0) {
297		/* We only care about the kse in the run queue. */
298		td->td_priority = newpri;
299		if (ke->ke_rqindex != (newpri / RQ_PPQ)) {
300			sched_rem(td);
301			sched_add(td);
302		}
303		return;
304	}
305
306	/* It is a threaded process */
307	kg = td->td_ksegrp;
308	kg->kg_runnable--;
309	TD_SET_CAN_RUN(td);
310	if (ke) {
311		if (kg->kg_last_assigned == td) {
312			kg->kg_last_assigned =
313			    TAILQ_PREV(td, threadqueue, td_runq);
314		}
315		sched_rem(td);
316	}
317	TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
318	td->td_priority = newpri;
319	setrunqueue(td);
320}
321
322void
323setrunqueue(struct thread *td)
324{
325	struct kse *ke;
326	struct ksegrp *kg;
327	struct thread *td2;
328	struct thread *tda;
329
330	CTR1(KTR_RUNQ, "setrunqueue: td%p", td);
331	mtx_assert(&sched_lock, MA_OWNED);
332	KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
333	    ("setrunqueue: bad thread state"));
334	TD_SET_RUNQ(td);
335	kg = td->td_ksegrp;
336	kg->kg_runnable++;
337	if ((td->td_proc->p_flag & P_SA) == 0) {
338		/*
339		 * Common path optimisation: Only one of everything
340		 * and the KSE is always already attached.
341		 * Totally ignore the ksegrp run queue.
342		 */
343		sched_add(td);
344		return;
345	}
346
347	tda = kg->kg_last_assigned;
348	if ((ke = td->td_kse) == NULL) {
349		if (kg->kg_idle_kses) {
350			/*
351			 * There is a free one so it's ours for the asking..
352			 */
353			ke = TAILQ_FIRST(&kg->kg_iq);
354			TAILQ_REMOVE(&kg->kg_iq, ke, ke_kgrlist);
355			ke->ke_state = KES_THREAD;
356			kg->kg_idle_kses--;
357		} else if (tda && (tda->td_priority > td->td_priority)) {
358			/*
359			 * None free, but there is one we can commandeer.
360			 */
361			ke = tda->td_kse;
362			sched_rem(tda);
363			tda->td_kse = NULL;
364			ke->ke_thread = NULL;
365			tda = kg->kg_last_assigned =
366		    	    TAILQ_PREV(tda, threadqueue, td_runq);
367		}
368	} else {
369		/*
370		 * Temporarily disassociate so it looks like the other cases.
371		 */
372		ke->ke_thread = NULL;
373		td->td_kse = NULL;
374	}
375
376	/*
377	 * Add the thread to the ksegrp's run queue at
378	 * the appropriate place.
379	 */
380	TAILQ_FOREACH(td2, &kg->kg_runq, td_runq) {
381		if (td2->td_priority > td->td_priority) {
382			TAILQ_INSERT_BEFORE(td2, td, td_runq);
383			break;
384		}
385	}
386	if (td2 == NULL) {
387		/* We ran off the end of the TAILQ or it was empty. */
388		TAILQ_INSERT_TAIL(&kg->kg_runq, td, td_runq);
389	}
390
391	/*
392	 * If we have a ke to use, then put it on the run queue and
393	 * If needed, readjust the last_assigned pointer.
394	 */
395	if (ke) {
396		if (tda == NULL) {
397			/*
398			 * No pre-existing last assigned so whoever is first
399			 * gets the KSE we brought in.. (maybe us)
400			 */
401			td2 = TAILQ_FIRST(&kg->kg_runq);
402			KASSERT((td2->td_kse == NULL),
403			    ("unexpected ke present"));
404			td2->td_kse = ke;
405			ke->ke_thread = td2;
406			kg->kg_last_assigned = td2;
407		} else if (tda->td_priority > td->td_priority) {
408			/*
409			 * It's ours, grab it, but last_assigned is past us
410			 * so don't change it.
411			 */
412			td->td_kse = ke;
413			ke->ke_thread = td;
414		} else {
415			/*
416			 * We are past last_assigned, so
417			 * put the new kse on whatever is next,
418			 * which may or may not be us.
419			 */
420			td2 = TAILQ_NEXT(tda, td_runq);
421			kg->kg_last_assigned = td2;
422			td2->td_kse = ke;
423			ke->ke_thread = td2;
424		}
425		sched_add(ke->ke_thread);
426	}
427}
428
429/*
430 * Kernel thread preemption implementation.  Critical sections mark
431 * regions of code in which preemptions are not allowed.
432 */
433void
434critical_enter(void)
435{
436	struct thread *td;
437
438	td = curthread;
439	if (td->td_critnest == 0)
440		cpu_critical_enter();
441	td->td_critnest++;
442}
443
444void
445critical_exit(void)
446{
447	struct thread *td;
448
449	td = curthread;
450	KASSERT(td->td_critnest != 0,
451	    ("critical_exit: td_critnest == 0"));
452	if (td->td_critnest == 1) {
453#if 0
454#ifdef PREEMPTION
455		mtx_assert(&sched_lock, MA_NOTOWNED);
456		if (td->td_pflags & TDP_OWEPREEMPT) {
457			mtx_lock_spin(&sched_lock);
458			mi_switch(SW_INVOL, NULL);
459			mtx_unlock_spin(&sched_lock);
460		}
461#endif
462#endif
463		td->td_critnest = 0;
464		cpu_critical_exit();
465	} else {
466		td->td_critnest--;
467	}
468}
469
470/*
471 * This function is called when a thread is about to be put on run queue
472 * because it has been made runnable or its priority has been adjusted.  It
473 * determines if the new thread should be immediately preempted to.  If so,
474 * it switches to it and eventually returns true.  If not, it returns false
475 * so that the caller may place the thread on an appropriate run queue.
476 */
477int
478maybe_preempt(struct thread *td)
479{
480#ifdef PREEMPTION
481	struct thread *ctd;
482	int cpri, pri;
483#endif
484
485	mtx_assert(&sched_lock, MA_OWNED);
486#ifdef PREEMPTION
487	/*
488	 * The new thread should not preempt the current thread if any of the
489	 * following conditions are true:
490	 *
491	 *  - The current thread has a higher (numerically lower) or
492	 *    equivalent priority.  Note that this prevents curthread from
493	 *    trying to preempt to itself.
494	 *  - It is too early in the boot for context switches (cold is set).
495	 *  - The current thread has an inhibitor set or is in the process of
496	 *    exiting.  In this case, the current thread is about to switch
497	 *    out anyways, so there's no point in preempting.  If we did,
498	 *    the current thread would not be properly resumed as well, so
499	 *    just avoid that whole landmine.
500	 *  - If the new thread's priority is not a realtime priority and
501	 *    the current thread's priority is not an idle priority and
502	 *    FULL_PREEMPTION is disabled.
503	 *
504	 * If all of these conditions are false, but the current thread is in
505	 * a nested critical section, then we have to defer the preemption
506	 * until we exit the critical section.  Otherwise, switch immediately
507	 * to the new thread.
508	 */
509	ctd = curthread;
510	pri = td->td_priority;
511	cpri = ctd->td_priority;
512	if (pri >= cpri || cold /* || dumping */ || TD_IS_INHIBITED(ctd) ||
513	    td->td_kse->ke_state != KES_THREAD)
514		return (0);
515#ifndef FULL_PREEMPTION
516	if (!(pri >= PRI_MIN_ITHD && pri <= PRI_MAX_ITHD) &&
517	    !(cpri >= PRI_MIN_IDLE))
518		return (0);
519#endif
520	if (ctd->td_critnest > 1) {
521		CTR1(KTR_PROC, "maybe_preempt: in critical section %d",
522		    ctd->td_critnest);
523		ctd->td_pflags |= TDP_OWEPREEMPT;
524		return (0);
525	}
526
527	/*
528	 * Our thread state says that we are already on a run queue, so
529	 * update our state as if we had been dequeued by choosethread().
530	 */
531	MPASS(TD_ON_RUNQ(td));
532	TD_SET_RUNNING(td);
533	CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td,
534	    td->td_proc->p_pid, td->td_proc->p_comm);
535	mi_switch(SW_INVOL, td);
536	return (1);
537#else
538	return (0);
539#endif
540}
541
542#ifndef PREEMPTION
543/* XXX: There should be a non-static version of this. */
544static void
545printf_caddr_t(void *data)
546{
547	printf("%s", (char *)data);
548}
549static char preempt_warning[] =
550    "WARNING: Kernel preemption is disabled, expect reduced performance.\n";
551SYSINIT(preempt_warning, SI_SUB_COPYRIGHT, SI_ORDER_ANY, printf_caddr_t,
552    preempt_warning)
553#endif
554
555/************************************************************************
556 * SYSTEM RUN QUEUE manipulations and tests				*
557 ************************************************************************/
558/*
559 * Initialize a run structure.
560 */
561void
562runq_init(struct runq *rq)
563{
564	int i;
565
566	bzero(rq, sizeof *rq);
567	for (i = 0; i < RQ_NQS; i++)
568		TAILQ_INIT(&rq->rq_queues[i]);
569}
570
571/*
572 * Clear the status bit of the queue corresponding to priority level pri,
573 * indicating that it is empty.
574 */
575static __inline void
576runq_clrbit(struct runq *rq, int pri)
577{
578	struct rqbits *rqb;
579
580	rqb = &rq->rq_status;
581	CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d",
582	    rqb->rqb_bits[RQB_WORD(pri)],
583	    rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri),
584	    RQB_BIT(pri), RQB_WORD(pri));
585	rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri);
586}
587
588/*
589 * Find the index of the first non-empty run queue.  This is done by
590 * scanning the status bits, a set bit indicates a non-empty queue.
591 */
592static __inline int
593runq_findbit(struct runq *rq)
594{
595	struct rqbits *rqb;
596	int pri;
597	int i;
598
599	rqb = &rq->rq_status;
600	for (i = 0; i < RQB_LEN; i++)
601		if (rqb->rqb_bits[i]) {
602			pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW);
603			CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d",
604			    rqb->rqb_bits[i], i, pri);
605			return (pri);
606		}
607
608	return (-1);
609}
610
611/*
612 * Set the status bit of the queue corresponding to priority level pri,
613 * indicating that it is non-empty.
614 */
615static __inline void
616runq_setbit(struct runq *rq, int pri)
617{
618	struct rqbits *rqb;
619
620	rqb = &rq->rq_status;
621	CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d",
622	    rqb->rqb_bits[RQB_WORD(pri)],
623	    rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri),
624	    RQB_BIT(pri), RQB_WORD(pri));
625	rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri);
626}
627
628/*
629 * Add the KSE to the queue specified by its priority, and set the
630 * corresponding status bit.
631 */
632void
633runq_add(struct runq *rq, struct kse *ke)
634{
635	struct rqhead *rqh;
636	int pri;
637
638	pri = ke->ke_thread->td_priority / RQ_PPQ;
639	ke->ke_rqindex = pri;
640	runq_setbit(rq, pri);
641	rqh = &rq->rq_queues[pri];
642	CTR4(KTR_RUNQ, "runq_add: p=%p pri=%d %d rqh=%p",
643	    ke->ke_proc, ke->ke_thread->td_priority, pri, rqh);
644	TAILQ_INSERT_TAIL(rqh, ke, ke_procq);
645}
646
647/*
648 * Return true if there are runnable processes of any priority on the run
649 * queue, false otherwise.  Has no side effects, does not modify the run
650 * queue structure.
651 */
652int
653runq_check(struct runq *rq)
654{
655	struct rqbits *rqb;
656	int i;
657
658	rqb = &rq->rq_status;
659	for (i = 0; i < RQB_LEN; i++)
660		if (rqb->rqb_bits[i]) {
661			CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d",
662			    rqb->rqb_bits[i], i);
663			return (1);
664		}
665	CTR0(KTR_RUNQ, "runq_check: empty");
666
667	return (0);
668}
669
670/*
671 * Find the highest priority process on the run queue.
672 */
673struct kse *
674runq_choose(struct runq *rq)
675{
676	struct rqhead *rqh;
677	struct kse *ke;
678	int pri;
679
680	mtx_assert(&sched_lock, MA_OWNED);
681	while ((pri = runq_findbit(rq)) != -1) {
682		rqh = &rq->rq_queues[pri];
683		ke = TAILQ_FIRST(rqh);
684		KASSERT(ke != NULL, ("runq_choose: no proc on busy queue"));
685		CTR3(KTR_RUNQ,
686		    "runq_choose: pri=%d kse=%p rqh=%p", pri, ke, rqh);
687		return (ke);
688	}
689	CTR1(KTR_RUNQ, "runq_choose: idleproc pri=%d", pri);
690
691	return (NULL);
692}
693
694/*
695 * Remove the KSE from the queue specified by its priority, and clear the
696 * corresponding status bit if the queue becomes empty.
697 * Caller must set ke->ke_state afterwards.
698 */
699void
700runq_remove(struct runq *rq, struct kse *ke)
701{
702	struct rqhead *rqh;
703	int pri;
704
705	KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
706		("runq_remove: process swapped out"));
707	pri = ke->ke_rqindex;
708	rqh = &rq->rq_queues[pri];
709	CTR4(KTR_RUNQ, "runq_remove: p=%p pri=%d %d rqh=%p",
710	    ke, ke->ke_thread->td_priority, pri, rqh);
711	KASSERT(ke != NULL, ("runq_remove: no proc on busy queue"));
712	TAILQ_REMOVE(rqh, ke, ke_procq);
713	if (TAILQ_EMPTY(rqh)) {
714		CTR0(KTR_RUNQ, "runq_remove: empty");
715		runq_clrbit(rq, pri);
716	}
717}
718
719#if 0
720void
721panc(char *string1, char *string2)
722{
723	printf("%s", string1);
724	kdb_enter(string2);
725}
726
727void
728thread_sanity_check(struct thread *td, char *string)
729{
730	struct proc *p;
731	struct ksegrp *kg;
732	struct kse *ke;
733	struct thread *td2 = NULL;
734	unsigned int prevpri;
735	int	saw_lastassigned = 0;
736	int unassigned = 0;
737	int assigned = 0;
738
739	p = td->td_proc;
740	kg = td->td_ksegrp;
741	ke = td->td_kse;
742
743
744	if (ke) {
745		if (p != ke->ke_proc) {
746			panc(string, "wrong proc");
747		}
748		if (ke->ke_thread != td) {
749			panc(string, "wrong thread");
750		}
751	}
752
753	if ((p->p_flag & P_SA) == 0) {
754		if (ke == NULL) {
755			panc(string, "non KSE thread lost kse");
756		}
757	} else {
758		prevpri = 0;
759		saw_lastassigned = 0;
760		unassigned = 0;
761		assigned = 0;
762		TAILQ_FOREACH(td2, &kg->kg_runq, td_runq) {
763			if (td2->td_priority < prevpri) {
764				panc(string, "thread runqueue unosorted");
765			}
766			if ((td2->td_state == TDS_RUNQ) &&
767			    td2->td_kse &&
768			    (td2->td_kse->ke_state != KES_ONRUNQ)) {
769				panc(string, "KSE wrong state");
770			}
771			prevpri = td2->td_priority;
772			if (td2->td_kse) {
773				assigned++;
774				if (unassigned) {
775					panc(string, "unassigned before assigned");
776				}
777 				if  (kg->kg_last_assigned == NULL) {
778					panc(string, "lastassigned corrupt");
779				}
780				if (saw_lastassigned) {
781					panc(string, "last assigned not last");
782				}
783				if (td2->td_kse->ke_thread != td2) {
784					panc(string, "mismatched kse/thread");
785				}
786			} else {
787				unassigned++;
788			}
789			if (td2 == kg->kg_last_assigned) {
790				saw_lastassigned = 1;
791				if (td2->td_kse == NULL) {
792					panc(string, "last assigned not assigned");
793				}
794			}
795		}
796		if (kg->kg_last_assigned && (saw_lastassigned == 0)) {
797			panc(string, "where on earth does lastassigned point?");
798		}
799#if 0
800		FOREACH_THREAD_IN_GROUP(kg, td2) {
801			if (((td2->td_flags & TDF_UNBOUND) == 0) &&
802			    (TD_ON_RUNQ(td2))) {
803				assigned++;
804				if (td2->td_kse == NULL) {
805					panc(string, "BOUND thread with no KSE");
806				}
807			}
808		}
809#endif
810#if 0
811		if ((unassigned + assigned) != kg->kg_runnable) {
812			panc(string, "wrong number in runnable");
813		}
814#endif
815	}
816	if (assigned == 12345) {
817		printf("%p %p %p %p %p %d, %d",
818		    td, td2, ke, kg, p, assigned, saw_lastassigned);
819	}
820}
821#endif
822
823