2 * Copyright (c) 2001 Jake Burkholder <jake@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
30 If there are N processors, then there are at most N KSEs (kernel
31 schedulable entities) working to process threads that belong to a
32 KSEGROUP (kg). If there are X of these KSEs actually running at the
33 moment in question, then there are at most M (N-X) of these KSEs on
34 the run queue, as running KSEs are not on the queue.
36 Runnable threads are queued off the KSEGROUP in priority order.
37 If 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
39 their priority from those threads and are put on the run queue.
41 The last thread that had a priority high enough to have a KSE associated
42 with it, AND IS ON THE RUN QUEUE is pointed to by
43 kg->kg_last_assigned. If no threads queued off the KSEGROUP have KSEs
44 assigned as all the available KSEs are activly running, or because there
45 are no threads queued, that pointer is NULL.
47 When a KSE is removed from the run queue to become runnable, we know
48 it was associated with the highest priority thread in the queue (at the head
49 of the queue). If it is also the last assigned we know M was 1 and must
50 now be 0. Since the thread is no longer queued that pointer must be
51 removed 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
53 but using it, we know there are STILL no more KSEs available, we can prove
54 that the next thread in the ksegrp list will not have a KSE to assign to
55 it, so we can show that the pointer must be made 'invalid' (NULL).
57 The pointer exists so that when a new thread is made runnable, it can
58 have its priority compared with the last assigned thread to see if
59 it should 'steal' its KSE or not.. i.e. is it 'earlier'
60 on the list than that thread or later.. If it's earlier, then the KSE is
61 removed from the last assigned (which is now not assigned a KSE)
62 and reassigned to the new thread, which is placed earlier in the list.
63 The pointer is then backed up to the previous thread (which may or may not
66 When a thread sleeps or is removed, the KSE becomes available and if there
67 are queued threads that are not assigned KSEs, the highest priority one of
68 them is assigned the KSE, which is then placed back on the run queue at
69 the approipriate place, and the kg->kg_last_assigned pointer is adjusted down
72 The following diagram shows 2 KSEs and 3 threads from a single process.
74 RUNQ: --->KSE---KSE--... (KSEs queued at priorities from threads)
77 KSEGROUP---thread--thread--thread (queued in priority order)
82 The result of this scheme is that the M available KSEs are always
83 queued at the priorities they have inherrited from the M highest priority
84 threads for that KSEGROUP. If this situation changes, the KSEs are
85 reassigned to keep this true.
88 #include <sys/cdefs.h>
89 __FBSDID("$FreeBSD$");
91 #include "opt_sched.h"
93 #ifndef KERN_SWITCH_INCLUDE
94 #include <sys/param.h>
95 #include <sys/systm.h>
97 #include <sys/kernel.h>
100 #include <sys/mutex.h>
101 #include <sys/proc.h>
102 #include <sys/queue.h>
103 #include <sys/sched.h>
104 #else /* KERN_SWITCH_INCLUDE */
105 #if defined(SMP) && (defined(__i386__) || defined(__amd64__))
108 #if defined(SMP) && defined(SCHED_4BSD)
109 #include <sys/sysctl.h>
112 #ifdef FULL_PREEMPTION
114 #error "The FULL_PREEMPTION option requires the PREEMPTION option"
118 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS);
120 #define td_kse td_sched
123 * kern.sched.preemption allows user space to determine if preemption support
124 * is compiled in or not. It is not currently a boot or runtime flag that
128 static int kern_sched_preemption = 1;
130 static int kern_sched_preemption = 0;
132 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD,
133 &kern_sched_preemption, 0, "Kernel preemption enabled");
135 /************************************************************************
136 * Functions that manipulate runnability from a thread perspective. *
137 ************************************************************************/
139 * Select the KSE that will be run next. From that find the thread, and
140 * remove it from the KSEGRP's run queue. If there is thread clustering,
141 * this will be what does it.
150 #if defined(SMP) && (defined(__i386__) || defined(__amd64__))
151 if (smp_active == 0 && PCPU_GET(cpuid) != 0) {
152 /* Shutting down, run idlethread on AP's */
153 td = PCPU_GET(idlethread);
155 CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
156 ke->ke_flags |= KEF_DIDRUN;
166 KASSERT((td->td_kse == ke), ("kse/thread mismatch"));
168 if (td->td_proc->p_flag & P_HADTHREADS) {
169 if (kg->kg_last_assigned == td) {
170 kg->kg_last_assigned = TAILQ_PREV(td,
171 threadqueue, td_runq);
173 TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
175 CTR2(KTR_RUNQ, "choosethread: td=%p pri=%d",
176 td, td->td_priority);
178 /* Simulate runq_choose() having returned the idle thread */
179 td = PCPU_GET(idlethread);
181 CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
183 ke->ke_flags |= KEF_DIDRUN;
186 * If we are in panic, only allow system threads,
187 * plus the one we are running in, to be run.
189 if (panicstr && ((td->td_proc->p_flag & P_SYSTEM) == 0 &&
190 (td->td_flags & TDF_INPANIC) == 0)) {
191 /* note that it is no longer on the run queue */
201 * Given a surplus system slot, try assign a new runnable thread to it.
203 * sched_thread_exit() (local)
204 * sched_switch() (local)
205 * sched_thread_exit() (local)
206 * remrunqueue() (local) (not at the moment)
209 slot_fill(struct ksegrp *kg)
213 mtx_assert(&sched_lock, MA_OWNED);
214 while (kg->kg_avail_opennings > 0) {
216 * Find the first unassigned thread
218 if ((td = kg->kg_last_assigned) != NULL)
219 td = TAILQ_NEXT(td, td_runq);
221 td = TAILQ_FIRST(&kg->kg_runq);
224 * If we found one, send it to the system scheduler.
227 kg->kg_last_assigned = td;
228 sched_add(td, SRQ_YIELDING);
229 CTR2(KTR_RUNQ, "slot_fill: td%p -> kg%p", td, kg);
231 /* no threads to use up the slots. quit now */
239 * Remove a thread from its KSEGRP's run queue.
240 * This in turn may remove it from a KSE if it was already assigned
241 * to one, possibly causing a new thread to be assigned to the KSE
242 * and the KSE getting a new priority.
245 remrunqueue(struct thread *td)
247 struct thread *td2, *td3;
251 mtx_assert(&sched_lock, MA_OWNED);
252 KASSERT((TD_ON_RUNQ(td)), ("remrunqueue: Bad state on run queue"));
255 CTR1(KTR_RUNQ, "remrunqueue: td%p", td);
258 * If it is not a threaded process, take the shortcut.
260 if ((td->td_proc->p_flag & P_HADTHREADS) == 0) {
261 /* remve from sys run queue and free up a slot */
263 ke->ke_state = KES_THREAD;
266 td3 = TAILQ_PREV(td, threadqueue, td_runq);
267 TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
268 if (ke->ke_state == KES_ONRUNQ) {
270 * This thread has been assigned to the system run queue.
271 * We need to dissociate it and try assign the
272 * KSE to the next available thread. Then, we should
273 * see if we need to move the KSE in the run queues.
276 ke->ke_state = KES_THREAD;
277 td2 = kg->kg_last_assigned;
278 KASSERT((td2 != NULL), ("last assigned has wrong value"));
280 kg->kg_last_assigned = td3;
281 /* slot_fill(kg); */ /* will replace it with another */
287 * Change the priority of a thread that is on the run queue.
290 adjustrunqueue( struct thread *td, int newpri)
295 mtx_assert(&sched_lock, MA_OWNED);
296 KASSERT((TD_ON_RUNQ(td)), ("adjustrunqueue: Bad state on run queue"));
299 CTR1(KTR_RUNQ, "adjustrunqueue: td%p", td);
301 * If it is not a threaded process, take the shortcut.
303 if ((td->td_proc->p_flag & P_HADTHREADS) == 0) {
304 /* We only care about the kse in the run queue. */
305 td->td_priority = newpri;
306 if (ke->ke_rqindex != (newpri / RQ_PPQ)) {
308 sched_add(td, SRQ_BORING);
313 /* It is a threaded process */
315 if (ke->ke_state == KES_ONRUNQ) {
316 if (kg->kg_last_assigned == td) {
317 kg->kg_last_assigned =
318 TAILQ_PREV(td, threadqueue, td_runq);
322 TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
324 td->td_priority = newpri;
325 setrunqueue(td, SRQ_BORING);
329 * This function is called when a thread is about to be put on a
330 * ksegrp run queue because it has been made runnable or its
331 * priority has been adjusted and the ksegrp does not have a
332 * free kse slot. It determines if a thread from the same ksegrp
333 * should be preempted. If so, it tries to switch threads
334 * if the thread is on the same cpu or notifies another cpu that
335 * it should switch threads.
339 maybe_preempt_in_ksegrp(struct thread *td)
342 struct thread *running_thread;
344 #ifndef FULL_PREEMPTION
346 pri = td->td_priority;
347 if (!(pri >= PRI_MIN_ITHD && pri <= PRI_MAX_ITHD))
350 mtx_assert(&sched_lock, MA_OWNED);
351 running_thread = curthread;
353 if (running_thread->td_ksegrp != td->td_ksegrp)
356 if (td->td_priority > running_thread->td_priority)
359 if (running_thread->td_critnest > 1)
360 running_thread->td_owepreempt = 1;
362 mi_switch(SW_INVOL, NULL);
365 running_thread->td_flags |= TDF_NEEDRESCHED;
372 struct thread *running_thread;
375 cpumask_t cpumask,dontuse;
377 struct pcpu *best_pcpu;
378 struct thread *cputhread;
380 #ifndef FULL_PREEMPTION
382 pri = td->td_priority;
383 if (!(pri >= PRI_MIN_ITHD && pri <= PRI_MAX_ITHD))
387 mtx_assert(&sched_lock, MA_OWNED);
389 running_thread = curthread;
391 #if !defined(KSEG_PEEMPT_BEST_CPU)
392 if (running_thread->td_ksegrp != td->td_ksegrp) {
396 /* if someone is ahead of this thread, wait our turn */
397 if (td != TAILQ_FIRST(&kg->kg_runq))
400 worst_pri = td->td_priority;
402 dontuse = stopped_cpus | idle_cpus_mask;
405 * Find a cpu with the worst priority that runs at thread from
406 * the same ksegrp - if multiple exist give first the last run
407 * cpu and then the current cpu priority
410 SLIST_FOREACH(pc, &cpuhead, pc_allcpu) {
411 cpumask = pc->pc_cpumask;
412 cputhread = pc->pc_curthread;
414 if ((cpumask & dontuse) ||
415 cputhread->td_ksegrp != kg)
418 if (cputhread->td_priority > worst_pri) {
419 worst_pri = cputhread->td_priority;
424 if (cputhread->td_priority == worst_pri &&
426 (td->td_lastcpu == pc->pc_cpuid ||
427 (PCPU_GET(cpumask) == cpumask &&
428 td->td_lastcpu != best_pcpu->pc_cpuid)))
432 /* Check if we need to preempt someone */
433 if (best_pcpu == NULL)
436 if (PCPU_GET(cpuid) != best_pcpu->pc_cpuid) {
437 best_pcpu->pc_curthread->td_flags |= TDF_NEEDRESCHED;
438 ipi_selected(best_pcpu->pc_cpumask, IPI_AST);
441 #if !defined(KSEG_PEEMPT_BEST_CPU)
445 if (td->td_priority > running_thread->td_priority)
448 if (running_thread->td_critnest > 1)
449 running_thread->td_owepreempt = 1;
451 mi_switch(SW_INVOL, NULL);
454 running_thread->td_flags |= TDF_NEEDRESCHED;
463 setrunqueue(struct thread *td, int flags)
469 CTR3(KTR_RUNQ, "setrunqueue: td:%p kg:%p pid:%d",
470 td, td->td_ksegrp, td->td_proc->p_pid);
471 CTR5(KTR_SCHED, "setrunqueue: %p(%s) prio %d by %p(%s)",
472 td, td->td_proc->p_comm, td->td_priority, curthread,
473 curthread->td_proc->p_comm);
474 mtx_assert(&sched_lock, MA_OWNED);
475 KASSERT((td->td_inhibitors == 0),
476 ("setrunqueue: trying to run inhibitted thread"));
477 KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
478 ("setrunqueue: bad thread state"));
481 if ((td->td_proc->p_flag & P_HADTHREADS) == 0) {
483 * Common path optimisation: Only one of everything
484 * and the KSE is always already attached.
485 * Totally ignore the ksegrp run queue.
487 if (kg->kg_avail_opennings != 1) {
488 if (limitcount < 1) {
490 printf("pid %d: corrected slot count (%d->1)\n",
491 td->td_proc->p_pid, kg->kg_avail_opennings);
494 kg->kg_avail_opennings = 1;
496 sched_add(td, flags);
501 * If the concurrency has reduced, and we would go in the
502 * assigned section, then keep removing entries from the
503 * system run queue, until we are not in that section
504 * or there is room for us to be put in that section.
505 * What we MUST avoid is the case where there are threads of less
506 * priority than the new one scheduled, but it can not
507 * be scheduled itself. That would lead to a non contiguous set
508 * of scheduled threads, and everything would break.
510 tda = kg->kg_last_assigned;
511 while ((kg->kg_avail_opennings <= 0) &&
512 (tda && (tda->td_priority > td->td_priority))) {
514 * None free, but there is one we can commandeer.
517 "setrunqueue: kg:%p: take slot from td: %p", kg, tda);
519 tda = kg->kg_last_assigned =
520 TAILQ_PREV(tda, threadqueue, td_runq);
524 * Add the thread to the ksegrp's run queue at
525 * the appropriate place.
527 TAILQ_FOREACH(td2, &kg->kg_runq, td_runq) {
528 if (td2->td_priority > td->td_priority) {
529 TAILQ_INSERT_BEFORE(td2, td, td_runq);
534 /* We ran off the end of the TAILQ or it was empty. */
535 TAILQ_INSERT_TAIL(&kg->kg_runq, td, td_runq);
539 * If we have a slot to use, then put the thread on the system
540 * run queue and if needed, readjust the last_assigned pointer.
541 * it may be that we need to schedule something anyhow
542 * even if the availabel slots are -ve so that
543 * all the items < last_assigned are scheduled.
545 if (kg->kg_avail_opennings > 0) {
548 * No pre-existing last assigned so whoever is first
549 * gets the slot.. (maybe us)
551 td2 = TAILQ_FIRST(&kg->kg_runq);
552 kg->kg_last_assigned = td2;
553 } else if (tda->td_priority > td->td_priority) {
557 * We are past last_assigned, so
558 * give the next slot to whatever is next,
559 * which may or may not be us.
561 td2 = TAILQ_NEXT(tda, td_runq);
562 kg->kg_last_assigned = td2;
564 sched_add(td2, flags);
566 CTR3(KTR_RUNQ, "setrunqueue: held: td%p kg%p pid%d",
567 td, td->td_ksegrp, td->td_proc->p_pid);
568 if ((flags & SRQ_YIELDING) == 0)
569 maybe_preempt_in_ksegrp(td);
574 * Kernel thread preemption implementation. Critical sections mark
575 * regions of code in which preemptions are not allowed.
584 CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td,
585 (long)td->td_proc->p_pid, td->td_proc->p_comm, td->td_critnest);
594 KASSERT(td->td_critnest != 0,
595 ("critical_exit: td_critnest == 0"));
596 if (td->td_critnest == 1) {
597 if (td->td_pflags & TDP_WAKEPROC0) {
598 td->td_pflags &= ~TDP_WAKEPROC0;
605 mtx_assert(&sched_lock, MA_NOTOWNED);
606 if (td->td_owepreempt) {
608 mtx_lock_spin(&sched_lock);
610 mi_switch(SW_INVOL, NULL);
611 mtx_unlock_spin(&sched_lock);
619 CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td,
620 (long)td->td_proc->p_pid, td->td_proc->p_comm, td->td_critnest);
624 * This function is called when a thread is about to be put on run queue
625 * because it has been made runnable or its priority has been adjusted. It
626 * determines if the new thread should be immediately preempted to. If so,
627 * it switches to it and eventually returns true. If not, it returns false
628 * so that the caller may place the thread on an appropriate run queue.
631 maybe_preempt(struct thread *td)
638 mtx_assert(&sched_lock, MA_OWNED);
641 * The new thread should not preempt the current thread if any of the
642 * following conditions are true:
644 * - The kernel is in the throes of crashing (panicstr).
645 * - The current thread has a higher (numerically lower) or
646 * equivalent priority. Note that this prevents curthread from
647 * trying to preempt to itself.
648 * - It is too early in the boot for context switches (cold is set).
649 * - The current thread has an inhibitor set or is in the process of
650 * exiting. In this case, the current thread is about to switch
651 * out anyways, so there's no point in preempting. If we did,
652 * the current thread would not be properly resumed as well, so
653 * just avoid that whole landmine.
654 * - If the new thread's priority is not a realtime priority and
655 * the current thread's priority is not an idle priority and
656 * FULL_PREEMPTION is disabled.
658 * If all of these conditions are false, but the current thread is in
659 * a nested critical section, then we have to defer the preemption
660 * until we exit the critical section. Otherwise, switch immediately
664 KASSERT ((ctd->td_kse != NULL && ctd->td_kse->ke_thread == ctd),
665 ("thread has no (or wrong) sched-private part."));
666 KASSERT((td->td_inhibitors == 0),
667 ("maybe_preempt: trying to run inhibitted thread"));
668 pri = td->td_priority;
669 cpri = ctd->td_priority;
670 if (panicstr != NULL || pri >= cpri || cold /* || dumping */ ||
671 TD_IS_INHIBITED(ctd) || td->td_kse->ke_state != KES_THREAD)
673 #ifndef FULL_PREEMPTION
674 if (!(pri >= PRI_MIN_ITHD && pri <= PRI_MAX_ITHD) &&
675 !(cpri >= PRI_MIN_IDLE))
678 if (ctd->td_critnest > 1) {
679 CTR1(KTR_PROC, "maybe_preempt: in critical section %d",
681 ctd->td_owepreempt = 1;
686 * Thread is runnable but not yet put on system run queue.
688 MPASS(TD_ON_RUNQ(td));
689 MPASS(td->td_sched->ke_state != KES_ONRUNQ);
690 if (td->td_proc->p_flag & P_HADTHREADS) {
692 * If this is a threaded process we actually ARE on the
693 * ksegrp run queue so take it off that first.
694 * Also undo any damage done to the last_assigned pointer.
695 * XXX Fix setrunqueue so this isn't needed
700 if (kg->kg_last_assigned == td)
701 kg->kg_last_assigned =
702 TAILQ_PREV(td, threadqueue, td_runq);
703 TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
707 CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td,
708 td->td_proc->p_pid, td->td_proc->p_comm);
709 mi_switch(SW_INVOL|SW_PREEMPT, td);
718 /* XXX: There should be a non-static version of this. */
720 printf_caddr_t(void *data)
722 printf("%s", (char *)data);
724 static char preempt_warning[] =
725 "WARNING: Kernel preemption is disabled, expect reduced performance.\n";
726 SYSINIT(preempt_warning, SI_SUB_COPYRIGHT, SI_ORDER_ANY, printf_caddr_t,
731 /************************************************************************
732 * SYSTEM RUN QUEUE manipulations and tests *
733 ************************************************************************/
735 * Initialize a run structure.
738 runq_init(struct runq *rq)
742 bzero(rq, sizeof *rq);
743 for (i = 0; i < RQ_NQS; i++)
744 TAILQ_INIT(&rq->rq_queues[i]);
748 * Clear the status bit of the queue corresponding to priority level pri,
749 * indicating that it is empty.
752 runq_clrbit(struct runq *rq, int pri)
756 rqb = &rq->rq_status;
757 CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d",
758 rqb->rqb_bits[RQB_WORD(pri)],
759 rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri),
760 RQB_BIT(pri), RQB_WORD(pri));
761 rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri);
765 * Find the index of the first non-empty run queue. This is done by
766 * scanning the status bits, a set bit indicates a non-empty queue.
769 runq_findbit(struct runq *rq)
775 rqb = &rq->rq_status;
776 for (i = 0; i < RQB_LEN; i++)
777 if (rqb->rqb_bits[i]) {
778 pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW);
779 CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d",
780 rqb->rqb_bits[i], i, pri);
788 * Set the status bit of the queue corresponding to priority level pri,
789 * indicating that it is non-empty.
792 runq_setbit(struct runq *rq, int pri)
796 rqb = &rq->rq_status;
797 CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d",
798 rqb->rqb_bits[RQB_WORD(pri)],
799 rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri),
800 RQB_BIT(pri), RQB_WORD(pri));
801 rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri);
805 * Add the KSE to the queue specified by its priority, and set the
806 * corresponding status bit.
809 runq_add(struct runq *rq, struct kse *ke, int flags)
814 pri = ke->ke_thread->td_priority / RQ_PPQ;
815 ke->ke_rqindex = pri;
816 runq_setbit(rq, pri);
817 rqh = &rq->rq_queues[pri];
818 CTR5(KTR_RUNQ, "runq_add: td=%p ke=%p pri=%d %d rqh=%p",
819 ke->ke_thread, ke, ke->ke_thread->td_priority, pri, rqh);
820 if (flags & SRQ_PREEMPTED) {
821 TAILQ_INSERT_HEAD(rqh, ke, ke_procq);
823 TAILQ_INSERT_TAIL(rqh, ke, ke_procq);
828 * Return true if there are runnable processes of any priority on the run
829 * queue, false otherwise. Has no side effects, does not modify the run
833 runq_check(struct runq *rq)
838 rqb = &rq->rq_status;
839 for (i = 0; i < RQB_LEN; i++)
840 if (rqb->rqb_bits[i]) {
841 CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d",
842 rqb->rqb_bits[i], i);
845 CTR0(KTR_RUNQ, "runq_check: empty");
850 #if defined(SMP) && defined(SCHED_4BSD)
852 SYSCTL_INT(_kern_sched, OID_AUTO, runq_fuzz, CTLFLAG_RW, &runq_fuzz, 0, "");
856 * Find the highest priority process on the run queue.
859 runq_choose(struct runq *rq)
865 mtx_assert(&sched_lock, MA_OWNED);
866 while ((pri = runq_findbit(rq)) != -1) {
867 rqh = &rq->rq_queues[pri];
868 #if defined(SMP) && defined(SCHED_4BSD)
869 /* fuzz == 1 is normal.. 0 or less are ignored */
872 * In the first couple of entries, check if
873 * there is one for our CPU as a preference.
875 int count = runq_fuzz;
876 int cpu = PCPU_GET(cpuid);
878 ke2 = ke = TAILQ_FIRST(rqh);
880 while (count-- && ke2) {
881 if (ke->ke_thread->td_lastcpu == cpu) {
885 ke2 = TAILQ_NEXT(ke2, ke_procq);
889 ke = TAILQ_FIRST(rqh);
890 KASSERT(ke != NULL, ("runq_choose: no proc on busy queue"));
892 "runq_choose: pri=%d kse=%p rqh=%p", pri, ke, rqh);
895 CTR1(KTR_RUNQ, "runq_choose: idleproc pri=%d", pri);
901 * Remove the KSE from the queue specified by its priority, and clear the
902 * corresponding status bit if the queue becomes empty.
903 * Caller must set ke->ke_state afterwards.
906 runq_remove(struct runq *rq, struct kse *ke)
911 KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
912 ("runq_remove: process swapped out"));
913 pri = ke->ke_rqindex;
914 rqh = &rq->rq_queues[pri];
915 CTR5(KTR_RUNQ, "runq_remove: td=%p, ke=%p pri=%d %d rqh=%p",
916 ke->ke_thread, ke, ke->ke_thread->td_priority, pri, rqh);
917 KASSERT(ke != NULL, ("runq_remove: no proc on busy queue"));
918 TAILQ_REMOVE(rqh, ke, ke_procq);
919 if (TAILQ_EMPTY(rqh)) {
920 CTR0(KTR_RUNQ, "runq_remove: empty");
921 runq_clrbit(rq, pri);
925 /****** functions that are temporarily here ***********/
927 extern struct mtx kse_zombie_lock;
930 * Allocate scheduler specific per-process resources.
931 * The thread and ksegrp have already been linked in.
932 * In this case just set the default concurrency value.
935 * proc_init() (UMA init method)
938 sched_newproc(struct proc *p, struct ksegrp *kg, struct thread *td)
941 /* This can go in sched_fork */
942 sched_init_concurrency(kg);
946 * thread is being either created or recycled.
947 * Fix up the per-scheduler resources associated with it.
949 * sched_fork_thread()
950 * thread_dtor() (*may go away)
951 * thread_init() (*may go away)
954 sched_newthread(struct thread *td)
958 ke = (struct td_sched *) (td + 1);
959 bzero(ke, sizeof(*ke));
962 ke->ke_state = KES_THREAD;
966 * Set up an initial concurrency of 1
967 * and set the given thread (if given) to be using that
969 * May be used "offline"..before the ksegrp is attached to the world
970 * and thus wouldn't need schedlock in that case.
973 * proc_init() (UMA) via sched_newproc()
976 sched_init_concurrency(struct ksegrp *kg)
979 CTR1(KTR_RUNQ,"kg %p init slots and concurrency to 1", kg);
980 kg->kg_concurrency = 1;
981 kg->kg_avail_opennings = 1;
985 * Change the concurrency of an existing ksegrp to N
993 sched_set_concurrency(struct ksegrp *kg, int concurrency)
996 CTR4(KTR_RUNQ,"kg %p set concurrency to %d, slots %d -> %d",
999 kg->kg_avail_opennings,
1000 kg->kg_avail_opennings + (concurrency - kg->kg_concurrency));
1001 kg->kg_avail_opennings += (concurrency - kg->kg_concurrency);
1002 kg->kg_concurrency = concurrency;
1006 * Called from thread_exit() for all exiting thread
1008 * Not to be confused with sched_exit_thread()
1009 * that is only called from thread_exit() for threads exiting
1010 * without the rest of the process exiting because it is also called from
1011 * sched_exit() and we wouldn't want to call it twice.
1012 * XXX This can probably be fixed.
1015 sched_thread_exit(struct thread *td)
1018 SLOT_RELEASE(td->td_ksegrp);
1019 slot_fill(td->td_ksegrp);
1022 #endif /* KERN_SWITCH_INCLUDE */