2 * Copyright (c) 1982, 1986, 1990, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
39 * $FreeBSD: src/sys/kern/kern_synch.c,v 1.87.2.6 2002/10/13 07:29:53 kbyanc Exp $
40 * $DragonFly: src/sys/kern/kern_synch.c,v 1.70 2007/01/12 03:05:49 dillon Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/signalvar.h>
50 #include <sys/resourcevar.h>
51 #include <sys/vmmeter.h>
52 #include <sys/sysctl.h>
56 #include <sys/ktrace.h>
58 #include <sys/xwait.h>
61 #include <sys/thread2.h>
62 #include <sys/spinlock2.h>
64 #include <machine/cpu.h>
65 #include <machine/smp.h>
67 TAILQ_HEAD(tslpque, thread);
69 static void sched_setup (void *dummy);
70 SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
75 int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
77 int ncpus2, ncpus2_shift, ncpus2_mask;
80 static struct callout loadav_callout;
81 static struct callout schedcpu_callout;
82 MALLOC_DEFINE(M_TSLEEP, "tslpque", "tsleep queues");
84 #if !defined(KTR_TSLEEP)
85 #define KTR_TSLEEP KTR_ALL
87 KTR_INFO_MASTER(tsleep);
88 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_beg, 0, "tsleep enter", 0);
89 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_end, 0, "tsleep exit", 0);
90 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_beg, 0, "wakeup enter", 0);
91 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_end, 0, "wakeup exit", 0);
92 #define logtsleep(name) KTR_LOG(tsleep_ ## name)
94 struct loadavg averunnable =
95 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
97 * Constants for averages over 1, 5, and 15 minutes
98 * when sampling at 5 second intervals.
100 static fixpt_t cexp[3] = {
101 0.9200444146293232 * FSCALE, /* exp(-1/12) */
102 0.9834714538216174 * FSCALE, /* exp(-1/60) */
103 0.9944598480048967 * FSCALE, /* exp(-1/180) */
106 static void endtsleep (void *);
107 static void unsleep_and_wakeup_thread(struct thread *td);
108 static void loadav (void *arg);
109 static void schedcpu (void *arg);
112 * Adjust the scheduler quantum. The quantum is specified in microseconds.
113 * Note that 'tick' is in microseconds per tick.
116 sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
120 new_val = sched_quantum * tick;
121 error = sysctl_handle_int(oidp, &new_val, 0, req);
122 if (error != 0 || req->newptr == NULL)
126 sched_quantum = new_val / tick;
127 hogticks = 2 * sched_quantum;
131 SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW,
132 0, sizeof sched_quantum, sysctl_kern_quantum, "I", "");
135 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
136 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
137 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
139 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
140 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
142 * If you don't want to bother with the faster/more-accurate formula, you
143 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
144 * (more general) method of calculating the %age of CPU used by a process.
146 * decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing
148 #define CCPU_SHIFT 11
150 static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
151 SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
154 * kernel uses `FSCALE', userland (SHOULD) use kern.fscale
156 int fscale __unused = FSCALE; /* exported to systat */
157 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
160 * Recompute process priorities, once a second.
162 * Since the userland schedulers are typically event oriented, if the
163 * estcpu calculation at wakeup() time is not sufficient to make a
164 * process runnable relative to other processes in the system we have
165 * a 1-second recalc to help out.
167 * This code also allows us to store sysclock_t data in the process structure
168 * without fear of an overrun, since sysclock_t are guarenteed to hold
169 * several seconds worth of count.
171 * WARNING! callouts can preempt normal threads. However, they will not
172 * preempt a thread holding a spinlock so we *can* safely use spinlocks.
174 static int schedcpu_stats(struct proc *p, void *data __unused);
175 static int schedcpu_resource(struct proc *p, void *data __unused);
180 allproc_scan(schedcpu_stats, NULL);
181 allproc_scan(schedcpu_resource, NULL);
182 wakeup((caddr_t)&lbolt);
183 wakeup((caddr_t)&lbolt_syncer);
184 callout_reset(&schedcpu_callout, hz, schedcpu, NULL);
188 * General process statistics once a second
191 schedcpu_stats(struct proc *p, void *data __unused)
195 if (p->p_stat == SSLEEP)
199 * Only recalculate processes that are active or have slept
200 * less then 2 seconds. The schedulers understand this.
202 if (p->p_slptime <= 1) {
203 p->p_usched->recalculate(&p->p_lwp);
205 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
212 * Resource checks. XXX break out since ksignal/killproc can block,
213 * limiting us to one process killed per second. There is probably
217 schedcpu_resource(struct proc *p, void *data __unused)
222 if (p->p_stat == SIDL ||
223 (p->p_flag & P_ZOMBIE) ||
224 p->p_limit == NULL ||
231 ttime = p->p_thread->td_sticks + p->p_thread->td_uticks;
233 switch(plimit_testcpulimit(p->p_limit, ttime)) {
234 case PLIMIT_TESTCPU_KILL:
235 killproc(p, "exceeded maximum CPU limit");
237 case PLIMIT_TESTCPU_XCPU:
238 if ((p->p_flag & P_XCPU) == 0) {
251 * This is only used by ps. Generate a cpu percentage use over
252 * a period of one second.
257 updatepcpu(struct lwp *lp, int cpticks, int ttlticks)
262 acc = (cpticks << FSHIFT) / ttlticks;
263 if (ttlticks >= ESTCPUFREQ) {
264 lp->lwp_pctcpu = acc;
266 remticks = ESTCPUFREQ - ttlticks;
267 lp->lwp_pctcpu = (acc * ttlticks + lp->lwp_pctcpu * remticks) /
273 * We're only looking at 7 bits of the address; everything is
274 * aligned to 4, lots of things are aligned to greater powers
275 * of 2. Shift right by 8, i.e. drop the bottom 256 worth.
277 #define TABLESIZE 128
278 #define LOOKUP(x) (((intptr_t)(x) >> 8) & (TABLESIZE - 1))
280 static cpumask_t slpque_cpumasks[TABLESIZE];
283 * General scheduler initialization. We force a reschedule 25 times
284 * a second by default. Note that cpu0 is initialized in early boot and
285 * cannot make any high level calls.
287 * Each cpu has its own sleep queue.
290 sleep_gdinit(globaldata_t gd)
292 static struct tslpque slpque_cpu0[TABLESIZE];
295 if (gd->gd_cpuid == 0) {
296 sched_quantum = (hz + 24) / 25;
297 hogticks = 2 * sched_quantum;
299 gd->gd_tsleep_hash = slpque_cpu0;
301 gd->gd_tsleep_hash = kmalloc(sizeof(slpque_cpu0),
302 M_TSLEEP, M_WAITOK | M_ZERO);
304 for (i = 0; i < TABLESIZE; ++i)
305 TAILQ_INIT(&gd->gd_tsleep_hash[i]);
309 * General sleep call. Suspends the current process until a wakeup is
310 * performed on the specified identifier. The process will then be made
311 * runnable with the specified priority. Sleeps at most timo/hz seconds
312 * (0 means no timeout). If flags includes PCATCH flag, signals are checked
313 * before and after sleeping, else signals are not checked. Returns 0 if
314 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
315 * signal needs to be delivered, ERESTART is returned if the current system
316 * call should be restarted if possible, and EINTR is returned if the system
317 * call should be interrupted by the signal (return EINTR).
319 * Note that if we are a process, we release_curproc() before messing with
320 * the LWKT scheduler.
322 * During autoconfiguration or after a panic, a sleep will simply
323 * lower the priority briefly to allow interrupts, then return.
326 tsleep(void *ident, int flags, const char *wmesg, int timo)
328 struct thread *td = curthread;
329 struct proc *p = td->td_proc; /* may be NULL */
336 struct callout thandle;
339 * NOTE: removed KTRPOINT, it could cause races due to blocking
340 * even in stable. Just scrap it for now.
342 if (cold || panicstr) {
344 * After a panic, or during autoconfiguration,
345 * just give interrupts a chance, then just return;
346 * don't run any other procs or panic below,
347 * in case this is the idle process and already asleep.
350 oldpri = td->td_pri & TDPRI_MASK;
351 lwkt_setpri_self(safepri);
353 lwkt_setpri_self(oldpri);
356 logtsleep(tsleep_beg);
358 KKASSERT(td != &gd->gd_idlethread); /* you must be kidding! */
361 * NOTE: all of this occurs on the current cpu, including any
362 * callout-based wakeups, so a critical section is a sufficient
365 * The entire sequence through to where we actually sleep must
366 * run without breaking the critical section.
369 catch = flags & PCATCH;
373 crit_enter_quick(td);
375 KASSERT(ident != NULL, ("tsleep: no ident"));
376 KASSERT(p == NULL || p->p_stat == SRUN, ("tsleep %p %s %d",
377 ident, wmesg, p->p_stat));
380 * Setup for the current process (if this is a process).
385 * Early termination if PCATCH was set and a
386 * signal is pending, interlocked with the
389 * Early termination only occurs when tsleep() is
390 * entered while in a normal SRUN state.
392 if ((sig = CURSIG(p)) != 0)
396 * Causes ksignal to wake us up when.
398 p->p_flag |= P_SINTR;
402 * Make sure the current process has been untangled from
403 * the userland scheduler and initialize slptime to start
406 if (flags & PNORESCHED)
407 td->td_flags |= TDF_NORESCHED;
408 p->p_usched->release_curproc(&p->p_lwp);
413 * Move our thread to the correct queue and setup our wchan, etc.
415 lwkt_deschedule_self(td);
416 td->td_flags |= TDF_TSLEEPQ;
417 TAILQ_INSERT_TAIL(&gd->gd_tsleep_hash[id], td, td_threadq);
418 atomic_set_int(&slpque_cpumasks[id], gd->gd_cpumask);
420 td->td_wchan = ident;
421 td->td_wmesg = wmesg;
422 td->td_wdomain = flags & PDOMAIN_MASK;
425 * Setup the timeout, if any
428 callout_init(&thandle);
429 callout_reset(&thandle, timo, endtsleep, td);
437 * Ok, we are sleeping. Place us in the SSLEEP state.
439 KKASSERT((p->p_flag & P_ONRUNQ) == 0);
441 p->p_lwp.lwp_ru.ru_nvcsw++;
445 * And when we are woken up, put us back in SRUN. If we
446 * slept for over a second, recalculate our estcpu.
450 p->p_usched->recalculate(&p->p_lwp);
457 * Make sure we haven't switched cpus while we were asleep. It's
458 * not supposed to happen. Cleanup our temporary flags.
460 KKASSERT(gd == td->td_gd);
461 td->td_flags &= ~TDF_NORESCHED;
464 * Cleanup the timeout.
467 if (td->td_flags & TDF_TIMEOUT) {
468 td->td_flags &= ~TDF_TIMEOUT;
472 callout_stop(&thandle);
477 * Since td_threadq is used both for our run queue AND for the
478 * tsleep hash queue, we can't still be on it at this point because
479 * we've gotten cpu back.
481 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0, ("tsleep: impossible thread flags %08x", td->td_flags));
487 * Figure out the correct error return
491 p->p_flag &= ~(P_BREAKTSLEEP | P_SINTR);
492 if (catch && error == 0 && (sig != 0 || (sig = CURSIG(p)))) {
493 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
499 logtsleep(tsleep_end);
505 * This is a dandy function that allows us to interlock tsleep/wakeup
506 * operations with unspecified upper level locks, such as lockmgr locks,
507 * simply by holding a critical section. The sequence is:
509 * (enter critical section)
510 * (acquire upper level lock)
511 * tsleep_interlock(blah)
512 * (release upper level lock)
514 * (exit critical section)
516 * Basically this function sets our cpumask for the ident which informs
517 * other cpus that our cpu 'might' be waiting (or about to wait on) the
518 * hash index related to the ident. The critical section prevents another
519 * cpu's wakeup() from being processed on our cpu until we are actually
520 * able to enter the tsleep(). Thus, no race occurs between our attempt
521 * to release a resource and sleep, and another cpu's attempt to acquire
522 * a resource and call wakeup.
524 * There isn't much of a point to this function unless you call it while
525 * holding a critical section.
528 _tsleep_interlock(globaldata_t gd, void *ident)
530 int id = LOOKUP(ident);
532 atomic_set_int(&slpque_cpumasks[id], gd->gd_cpumask);
536 tsleep_interlock(void *ident)
538 _tsleep_interlock(mycpu, ident);
542 * Interlocked spinlock sleep. An exclusively held spinlock must
543 * be passed to msleep(). The function will atomically release the
544 * spinlock and tsleep on the ident, then reacquire the spinlock and
547 * This routine is fairly important along the critical path, so optimize it
551 msleep(void *ident, struct spinlock *spin, int flags,
552 const char *wmesg, int timo)
554 globaldata_t gd = mycpu;
558 _tsleep_interlock(gd, ident);
559 spin_unlock_wr_quick(gd, spin);
560 error = tsleep(ident, flags, wmesg, timo);
561 spin_lock_wr_quick(gd, spin);
568 * Implement the timeout for tsleep.
570 * We set P_BREAKTSLEEP to indicate that an event has occured, but
571 * we only call setrunnable if the process is not stopped.
573 * This type of callout timeout is scheduled on the same cpu the process
574 * is sleeping on. Also, at the moment, the MP lock is held.
582 ASSERT_MP_LOCK_HELD(curthread);
586 * cpu interlock. Thread flags are only manipulated on
587 * the cpu owning the thread. proc flags are only manipulated
588 * by the older of the MP lock. We have both.
590 if (td->td_flags & TDF_TSLEEPQ) {
591 td->td_flags |= TDF_TIMEOUT;
593 if ((p = td->td_proc) != NULL) {
594 p->p_flag |= P_BREAKTSLEEP;
595 if ((p->p_flag & P_STOPPED) == 0)
598 unsleep_and_wakeup_thread(td);
605 * Unsleep and wakeup a thread. This function runs without the MP lock
606 * which means that it can only manipulate thread state on the owning cpu,
607 * and cannot touch the process state at all.
611 unsleep_and_wakeup_thread(struct thread *td)
613 globaldata_t gd = mycpu;
617 if (td->td_gd != gd) {
618 lwkt_send_ipiq(td->td_gd, (ipifunc1_t)unsleep_and_wakeup_thread, td);
623 if (td->td_flags & TDF_TSLEEPQ) {
624 td->td_flags &= ~TDF_TSLEEPQ;
625 id = LOOKUP(td->td_wchan);
626 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_threadq);
627 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL)
628 atomic_clear_int(&slpque_cpumasks[id], gd->gd_cpumask);
635 * Make all processes sleeping on the specified identifier runnable.
636 * count may be zero or one only.
638 * The domain encodes the sleep/wakeup domain AND the first cpu to check
639 * (which is always the current cpu). As we iterate across cpus
641 * This call may run without the MP lock held. We can only manipulate thread
642 * state on the cpu owning the thread. We CANNOT manipulate process state
646 _wakeup(void *ident, int domain)
661 logtsleep(wakeup_beg);
664 qp = &gd->gd_tsleep_hash[id];
666 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
667 ntd = TAILQ_NEXT(td, td_threadq);
668 if (td->td_wchan == ident &&
669 td->td_wdomain == (domain & PDOMAIN_MASK)
671 KKASSERT(td->td_flags & TDF_TSLEEPQ);
672 td->td_flags &= ~TDF_TSLEEPQ;
673 TAILQ_REMOVE(qp, td, td_threadq);
674 if (TAILQ_FIRST(qp) == NULL) {
675 atomic_clear_int(&slpque_cpumasks[id],
679 if (domain & PWAKEUP_ONE)
687 * We finished checking the current cpu but there still may be
688 * more work to do. Either wakeup_one was requested and no matching
689 * thread was found, or a normal wakeup was requested and we have
690 * to continue checking cpus.
692 * The cpu that started the wakeup sequence is encoded in the domain.
693 * We use this information to determine which cpus still need to be
694 * checked, locate a candidate cpu, and chain the wakeup
695 * asynchronously with an IPI message.
697 * It should be noted that this scheme is actually less expensive then
698 * the old scheme when waking up multiple threads, since we send
699 * only one IPI message per target candidate which may then schedule
700 * multiple threads. Before we could have wound up sending an IPI
701 * message for each thread on the target cpu (!= current cpu) that
702 * needed to be woken up.
704 * NOTE: Wakeups occuring on remote cpus are asynchronous. This
705 * should be ok since we are passing idents in the IPI rather then
708 if ((domain & PWAKEUP_MYCPU) == 0 &&
709 (mask = slpque_cpumasks[id]) != 0
712 * Look for a cpu that might have work to do. Mask out cpus
713 * which have already been processed.
715 * 31xxxxxxxxxxxxxxxxxxxxxxxxxxxxx0
717 * start currentcpu start
720 * 11111111111111110000000000000111 case1
721 * 00000000111111110000000000000000 case2
723 * case1: We started at start_case1 and processed through
724 * to the current cpu. We have to check any bits
725 * after the current cpu, then check bits before
728 * case2: We have already checked all the bits from
729 * start_case2 to the end, and from 0 to the current
730 * cpu. We just have the bits from the current cpu
731 * to start_case2 left to check.
733 startcpu = PWAKEUP_DECODE(domain);
734 if (gd->gd_cpuid >= startcpu) {
738 tmask = mask & ~((gd->gd_cpumask << 1) - 1);
740 nextcpu = bsfl(mask & tmask);
741 lwkt_send_ipiq2(globaldata_find(nextcpu),
742 _wakeup, ident, domain);
744 tmask = (1 << startcpu) - 1;
746 nextcpu = bsfl(mask & tmask);
748 globaldata_find(nextcpu),
749 _wakeup, ident, domain);
756 tmask = ~((gd->gd_cpumask << 1) - 1) &
757 ((1 << startcpu) - 1);
759 nextcpu = bsfl(mask & tmask);
760 lwkt_send_ipiq2(globaldata_find(nextcpu),
761 _wakeup, ident, domain);
767 logtsleep(wakeup_end);
772 * Wakeup all threads tsleep()ing on the specified ident, on all cpus
777 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid));
781 * Wakeup one thread tsleep()ing on the specified ident, on any cpu.
784 wakeup_one(void *ident)
786 /* XXX potentially round-robin the first responding cpu */
787 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid) | PWAKEUP_ONE);
791 * Wakeup threads tsleep()ing on the specified ident on the current cpu
795 wakeup_mycpu(void *ident)
797 _wakeup(ident, PWAKEUP_MYCPU);
801 * Wakeup one thread tsleep()ing on the specified ident on the current cpu
805 wakeup_mycpu_one(void *ident)
807 /* XXX potentially round-robin the first responding cpu */
808 _wakeup(ident, PWAKEUP_MYCPU|PWAKEUP_ONE);
812 * Wakeup all thread tsleep()ing on the specified ident on the specified cpu
816 wakeup_oncpu(globaldata_t gd, void *ident)
820 _wakeup(ident, PWAKEUP_MYCPU);
822 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU);
825 _wakeup(ident, PWAKEUP_MYCPU);
830 * Wakeup one thread tsleep()ing on the specified ident on the specified cpu
834 wakeup_oncpu_one(globaldata_t gd, void *ident)
838 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
840 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
843 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
848 * Wakeup all threads waiting on the specified ident that slept using
849 * the specified domain, on all cpus.
852 wakeup_domain(void *ident, int domain)
854 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid));
858 * Wakeup one thread waiting on the specified ident that slept using
859 * the specified domain, on any cpu.
862 wakeup_domain_one(void *ident, int domain)
864 /* XXX potentially round-robin the first responding cpu */
865 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid) | PWAKEUP_ONE);
871 * Make a process runnable. The MP lock must be held on call. This only
872 * has an effect if we are in SSLEEP. We only break out of the
873 * tsleep if P_BREAKTSLEEP is set, otherwise we just fix-up the state.
875 * NOTE: With the MP lock held we can only safely manipulate the process
876 * structure. We cannot safely manipulate the thread structure.
879 setrunnable(struct proc *p)
882 ASSERT_MP_LOCK_HELD(curthread);
883 p->p_flag &= ~P_STOPPED;
884 if (p->p_stat == SSLEEP && (p->p_flag & P_BREAKTSLEEP)) {
885 unsleep_and_wakeup_thread(p->p_thread);
891 * The process is stopped due to some condition, usually because P_STOPPED
892 * is set but also possibly due to being traced.
894 * NOTE! If the caller sets P_STOPPED, the caller must also clear P_WAITED
895 * because the parent may check the child's status before the child actually
896 * gets to this routine.
898 * This routine is called with the current process only, typically just
899 * before returning to userland.
901 * Setting P_BREAKTSLEEP before entering the tsleep will cause a passive
902 * SIGCONT to break out of the tsleep.
905 tstop(struct proc *p)
907 wakeup((caddr_t)p->p_pptr);
908 p->p_flag |= P_BREAKTSLEEP;
909 tsleep(p, 0, "stop", 0);
913 * Yield / synchronous reschedule. This is a bit tricky because the trap
914 * code might have set a lazy release on the switch function. Setting
915 * P_PASSIVE_ACQ will ensure that the lazy release executes when we call
916 * switch, and that we are given a greater chance of affinity with our
919 * We call lwkt_setpri_self() to rotate our thread to the end of the lwkt
920 * run queue. lwkt_switch() will also execute any assigned passive release
921 * (which usually calls release_curproc()), allowing a same/higher priority
922 * process to be designated as the current process.
924 * While it is possible for a lower priority process to be designated,
925 * it's call to lwkt_maybe_switch() in acquire_curproc() will likely
926 * round-robin back to us and we will be able to re-acquire the current
927 * process designation.
932 struct thread *td = curthread;
933 struct proc *p = td->td_proc;
935 lwkt_setpri_self(td->td_pri & TDPRI_MASK);
937 p->p_flag |= P_PASSIVE_ACQ;
939 p->p_flag &= ~P_PASSIVE_ACQ;
946 * Compute a tenex style load average of a quantity on
947 * 1, 5 and 15 minute intervals.
949 static int loadav_count_runnable(struct proc *p, void *data);
958 allproc_scan(loadav_count_runnable, &nrun);
960 for (i = 0; i < 3; i++) {
961 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
962 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
966 * Schedule the next update to occur after 5 seconds, but add a
967 * random variation to avoid synchronisation with processes that
968 * run at regular intervals.
970 callout_reset(&loadav_callout, hz * 4 + (int)(krandom() % (hz * 2 + 1)),
975 loadav_count_runnable(struct proc *p, void *data)
982 if ((td = p->p_thread) == NULL)
984 if (td->td_flags & TDF_BLOCKED)
998 sched_setup(void *dummy)
1000 callout_init(&loadav_callout);
1001 callout_init(&schedcpu_callout);
1003 /* Kick off timeout driven events by calling first time. */