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 1.66 2006/09/03 18:52:28 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/ipl.h>
66 #include <machine/smp.h>
68 TAILQ_HEAD(tslpque, thread);
70 static void sched_setup (void *dummy);
71 SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
76 int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
78 int ncpus2, ncpus2_shift, ncpus2_mask;
81 static struct callout loadav_callout;
82 static struct callout schedcpu_callout;
83 MALLOC_DEFINE(M_TSLEEP, "tslpque", "tsleep queues");
85 #if !defined(KTR_TSLEEP)
86 #define KTR_TSLEEP KTR_ALL
88 KTR_INFO_MASTER(tsleep);
89 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_beg, 0, "tsleep enter", 0);
90 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_end, 0, "tsleep exit", 0);
91 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_beg, 0, "wakeup enter", 0);
92 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_end, 0, "wakeup exit", 0);
93 #define logtsleep(name) KTR_LOG(tsleep_ ## name)
95 struct loadavg averunnable =
96 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
98 * Constants for averages over 1, 5, and 15 minutes
99 * when sampling at 5 second intervals.
101 static fixpt_t cexp[3] = {
102 0.9200444146293232 * FSCALE, /* exp(-1/12) */
103 0.9834714538216174 * FSCALE, /* exp(-1/60) */
104 0.9944598480048967 * FSCALE, /* exp(-1/180) */
107 static void endtsleep (void *);
108 static void unsleep_and_wakeup_thread(struct thread *td);
109 static void loadav (void *arg);
110 static void schedcpu (void *arg);
113 * Adjust the scheduler quantum. The quantum is specified in microseconds.
114 * Note that 'tick' is in microseconds per tick.
117 sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
121 new_val = sched_quantum * tick;
122 error = sysctl_handle_int(oidp, &new_val, 0, req);
123 if (error != 0 || req->newptr == NULL)
127 sched_quantum = new_val / tick;
128 hogticks = 2 * sched_quantum;
132 SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW,
133 0, sizeof sched_quantum, sysctl_kern_quantum, "I", "");
136 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
137 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
138 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
140 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
141 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
143 * If you don't want to bother with the faster/more-accurate formula, you
144 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
145 * (more general) method of calculating the %age of CPU used by a process.
147 * decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing
149 #define CCPU_SHIFT 11
151 static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
152 SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
155 * kernel uses `FSCALE', userland (SHOULD) use kern.fscale
157 static int fscale __unused = FSCALE;
158 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
161 * Recompute process priorities, once a second.
163 * Since the userland schedulers are typically event oriented, if the
164 * estcpu calculation at wakeup() time is not sufficient to make a
165 * process runnable relative to other processes in the system we have
166 * a 1-second recalc to help out.
168 * This code also allows us to store sysclock_t data in the process structure
169 * without fear of an overrun, since sysclock_t are guarenteed to hold
170 * several seconds worth of count.
172 * WARNING! callouts can preempt normal threads. However, they will not
173 * preempt a thread holding a spinlock so we *can* safely use spinlocks.
175 static int schedcpu_stats(struct proc *p, void *data __unused);
176 static int schedcpu_resource(struct proc *p, void *data __unused);
181 allproc_scan(schedcpu_stats, NULL);
182 allproc_scan(schedcpu_resource, NULL);
183 wakeup((caddr_t)&lbolt);
184 wakeup((caddr_t)&lbolt_syncer);
185 callout_reset(&schedcpu_callout, hz, schedcpu, NULL);
189 * General process statistics once a second
192 schedcpu_stats(struct proc *p, void *data __unused)
196 if (p->p_stat == SSLEEP)
200 * Only recalculate processes that are active or have slept
201 * less then 2 seconds. The schedulers understand this.
203 if (p->p_slptime <= 1) {
204 p->p_usched->recalculate(&p->p_lwp);
206 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
213 * Resource checks. XXX break out since ksignal/killproc can block,
214 * limiting us to one process killed per second. There is probably
218 schedcpu_resource(struct proc *p, void *data __unused)
223 if (p->p_stat == SIDL ||
224 (p->p_flag & P_ZOMBIE) ||
225 p->p_limit == NULL ||
232 ttime = p->p_thread->td_sticks + p->p_thread->td_uticks;
234 switch(plimit_testcpulimit(p->p_limit, ttime)) {
235 case PLIMIT_TESTCPU_KILL:
236 killproc(p, "exceeded maximum CPU limit");
238 case PLIMIT_TESTCPU_XCPU:
239 if ((p->p_flag & P_XCPU) == 0) {
252 * This is only used by ps. Generate a cpu percentage use over
253 * a period of one second.
258 updatepcpu(struct lwp *lp, int cpticks, int ttlticks)
263 acc = (cpticks << FSHIFT) / ttlticks;
264 if (ttlticks >= ESTCPUFREQ) {
265 lp->lwp_pctcpu = acc;
267 remticks = ESTCPUFREQ - ttlticks;
268 lp->lwp_pctcpu = (acc * ttlticks + lp->lwp_pctcpu * remticks) /
274 * We're only looking at 7 bits of the address; everything is
275 * aligned to 4, lots of things are aligned to greater powers
276 * of 2. Shift right by 8, i.e. drop the bottom 256 worth.
278 #define TABLESIZE 128
279 #define LOOKUP(x) (((intptr_t)(x) >> 8) & (TABLESIZE - 1))
281 static cpumask_t slpque_cpumasks[TABLESIZE];
284 * General scheduler initialization. We force a reschedule 25 times
285 * a second by default. Note that cpu0 is initialized in early boot and
286 * cannot make any high level calls.
288 * Each cpu has its own sleep queue.
291 sleep_gdinit(globaldata_t gd)
293 static struct tslpque slpque_cpu0[TABLESIZE];
296 if (gd->gd_cpuid == 0) {
297 sched_quantum = (hz + 24) / 25;
298 hogticks = 2 * sched_quantum;
300 gd->gd_tsleep_hash = slpque_cpu0;
302 gd->gd_tsleep_hash = malloc(sizeof(slpque_cpu0),
303 M_TSLEEP, M_WAITOK | M_ZERO);
305 for (i = 0; i < TABLESIZE; ++i)
306 TAILQ_INIT(&gd->gd_tsleep_hash[i]);
310 * General sleep call. Suspends the current process until a wakeup is
311 * performed on the specified identifier. The process will then be made
312 * runnable with the specified priority. Sleeps at most timo/hz seconds
313 * (0 means no timeout). If flags includes PCATCH flag, signals are checked
314 * before and after sleeping, else signals are not checked. Returns 0 if
315 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
316 * signal needs to be delivered, ERESTART is returned if the current system
317 * call should be restarted if possible, and EINTR is returned if the system
318 * call should be interrupted by the signal (return EINTR).
320 * Note that if we are a process, we release_curproc() before messing with
321 * the LWKT scheduler.
323 * During autoconfiguration or after a panic, a sleep will simply
324 * lower the priority briefly to allow interrupts, then return.
327 tsleep(void *ident, int flags, const char *wmesg, int timo)
329 struct thread *td = curthread;
330 struct proc *p = td->td_proc; /* may be NULL */
337 struct callout thandle;
340 * NOTE: removed KTRPOINT, it could cause races due to blocking
341 * even in stable. Just scrap it for now.
343 if (cold || panicstr) {
345 * After a panic, or during autoconfiguration,
346 * just give interrupts a chance, then just return;
347 * don't run any other procs or panic below,
348 * in case this is the idle process and already asleep.
351 oldpri = td->td_pri & TDPRI_MASK;
352 lwkt_setpri_self(safepri);
354 lwkt_setpri_self(oldpri);
357 logtsleep(tsleep_beg);
359 KKASSERT(td != &gd->gd_idlethread); /* you must be kidding! */
362 * NOTE: all of this occurs on the current cpu, including any
363 * callout-based wakeups, so a critical section is a sufficient
366 * The entire sequence through to where we actually sleep must
367 * run without breaking the critical section.
370 catch = flags & PCATCH;
374 crit_enter_quick(td);
376 KASSERT(ident != NULL, ("tsleep: no ident"));
377 KASSERT(p == NULL || p->p_stat == SRUN, ("tsleep %p %s %d",
378 ident, wmesg, p->p_stat));
381 * Setup for the current process (if this is a process).
386 * Early termination if PCATCH was set and a
387 * signal is pending, interlocked with the
390 * Early termination only occurs when tsleep() is
391 * entered while in a normal SRUN state.
393 if ((sig = CURSIG(p)) != 0)
397 * Causes ksignal to wake us up when.
399 p->p_flag |= P_SINTR;
403 * Make sure the current process has been untangled from
404 * the userland scheduler and initialize slptime to start
407 if (flags & PNORESCHED)
408 td->td_flags |= TDF_NORESCHED;
409 p->p_usched->release_curproc(&p->p_lwp);
414 * Move our thread to the correct queue and setup our wchan, etc.
416 lwkt_deschedule_self(td);
417 td->td_flags |= TDF_TSLEEPQ;
418 TAILQ_INSERT_TAIL(&gd->gd_tsleep_hash[id], td, td_threadq);
419 atomic_set_int(&slpque_cpumasks[id], gd->gd_cpumask);
421 td->td_wchan = ident;
422 td->td_wmesg = wmesg;
423 td->td_wdomain = flags & PDOMAIN_MASK;
426 * Setup the timeout, if any
429 callout_init(&thandle);
430 callout_reset(&thandle, timo, endtsleep, td);
438 * Ok, we are sleeping. Place us in the SSLEEP state.
440 KKASSERT((p->p_flag & P_ONRUNQ) == 0);
442 p->p_stats->p_ru.ru_nvcsw++;
446 * And when we are woken up, put us back in SRUN. If we
447 * slept for over a second, recalculate our estcpu.
451 p->p_usched->recalculate(&p->p_lwp);
458 * Make sure we haven't switched cpus while we were asleep. It's
459 * not supposed to happen. Cleanup our temporary flags.
461 KKASSERT(gd == td->td_gd);
462 td->td_flags &= ~TDF_NORESCHED;
465 * Cleanup the timeout.
468 if (td->td_flags & TDF_TIMEOUT) {
469 td->td_flags &= ~TDF_TIMEOUT;
473 callout_stop(&thandle);
478 * Since td_threadq is used both for our run queue AND for the
479 * tsleep hash queue, we can't still be on it at this point because
480 * we've gotten cpu back.
482 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0, ("tsleep: impossible thread flags %08x", td->td_flags));
488 * Figure out the correct error return
492 p->p_flag &= ~(P_BREAKTSLEEP | P_SINTR);
493 if (catch && error == 0 && (sig != 0 || (sig = CURSIG(p)))) {
494 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
500 logtsleep(tsleep_end);
506 * This is a dandy function that allows us to interlock tsleep/wakeup
507 * operations with unspecified upper level locks, such as lockmgr locks,
508 * simply by holding a critical section. The sequence is:
510 * (enter critical section)
511 * (acquire upper level lock)
512 * tsleep_interlock(blah)
513 * (release upper level lock)
515 * (exit critical section)
517 * Basically this function sets our cpumask for the ident which informs
518 * other cpus that our cpu 'might' be waiting (or about to wait on) the
519 * hash index related to the ident. The critical section prevents another
520 * cpu's wakeup() from being processed on our cpu until we are actually
521 * able to enter the tsleep(). Thus, no race occurs between our attempt
522 * to release a resource and sleep, and another cpu's attempt to acquire
523 * a resource and call wakeup.
525 * There isn't much of a point to this function unless you call it while
526 * holding a critical section.
529 _tsleep_interlock(globaldata_t gd, void *ident)
531 int id = LOOKUP(ident);
533 atomic_set_int(&slpque_cpumasks[id], gd->gd_cpumask);
537 tsleep_interlock(void *ident)
539 _tsleep_interlock(mycpu, ident);
543 * Interlocked spinlock sleep. An exclusively held spinlock must
544 * be passed to msleep(). The function will atomically release the
545 * spinlock and tsleep on the ident, then reacquire the spinlock and
548 * This routine is fairly important along the critical path, so optimize it
552 msleep(void *ident, struct spinlock *spin, int flags,
553 const char *wmesg, int timo)
555 globaldata_t gd = mycpu;
559 _tsleep_interlock(gd, ident);
560 spin_unlock_wr_quick(gd, spin);
561 error = tsleep(ident, flags, wmesg, timo);
562 spin_lock_wr_quick(gd, spin);
569 * Implement the timeout for tsleep.
571 * We set P_BREAKTSLEEP to indicate that an event has occured, but
572 * we only call setrunnable if the process is not stopped.
574 * This type of callout timeout is scheduled on the same cpu the process
575 * is sleeping on. Also, at the moment, the MP lock is held.
583 ASSERT_MP_LOCK_HELD(curthread);
587 * cpu interlock. Thread flags are only manipulated on
588 * the cpu owning the thread. proc flags are only manipulated
589 * by the older of the MP lock. We have both.
591 if (td->td_flags & TDF_TSLEEPQ) {
592 td->td_flags |= TDF_TIMEOUT;
594 if ((p = td->td_proc) != NULL) {
595 p->p_flag |= P_BREAKTSLEEP;
596 if ((p->p_flag & P_STOPPED) == 0)
599 unsleep_and_wakeup_thread(td);
606 * Unsleep and wakeup a thread. This function runs without the MP lock
607 * which means that it can only manipulate thread state on the owning cpu,
608 * and cannot touch the process state at all.
612 unsleep_and_wakeup_thread(struct thread *td)
614 globaldata_t gd = mycpu;
618 if (td->td_gd != gd) {
619 lwkt_send_ipiq(td->td_gd, (ipifunc1_t)unsleep_and_wakeup_thread, td);
624 if (td->td_flags & TDF_TSLEEPQ) {
625 td->td_flags &= ~TDF_TSLEEPQ;
626 id = LOOKUP(td->td_wchan);
627 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_threadq);
628 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL)
629 atomic_clear_int(&slpque_cpumasks[id], gd->gd_cpumask);
636 * Make all processes sleeping on the specified identifier runnable.
637 * count may be zero or one only.
639 * The domain encodes the sleep/wakeup domain AND the first cpu to check
640 * (which is always the current cpu). As we iterate across cpus
642 * This call may run without the MP lock held. We can only manipulate thread
643 * state on the cpu owning the thread. We CANNOT manipulate process state
647 _wakeup(void *ident, int domain)
662 logtsleep(wakeup_beg);
665 qp = &gd->gd_tsleep_hash[id];
667 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
668 ntd = TAILQ_NEXT(td, td_threadq);
669 if (td->td_wchan == ident &&
670 td->td_wdomain == (domain & PDOMAIN_MASK)
672 KKASSERT(td->td_flags & TDF_TSLEEPQ);
673 td->td_flags &= ~TDF_TSLEEPQ;
674 TAILQ_REMOVE(qp, td, td_threadq);
675 if (TAILQ_FIRST(qp) == NULL) {
676 atomic_clear_int(&slpque_cpumasks[id],
680 if (domain & PWAKEUP_ONE)
688 * We finished checking the current cpu but there still may be
689 * more work to do. Either wakeup_one was requested and no matching
690 * thread was found, or a normal wakeup was requested and we have
691 * to continue checking cpus.
693 * The cpu that started the wakeup sequence is encoded in the domain.
694 * We use this information to determine which cpus still need to be
695 * checked, locate a candidate cpu, and chain the wakeup
696 * asynchronously with an IPI message.
698 * It should be noted that this scheme is actually less expensive then
699 * the old scheme when waking up multiple threads, since we send
700 * only one IPI message per target candidate which may then schedule
701 * multiple threads. Before we could have wound up sending an IPI
702 * message for each thread on the target cpu (!= current cpu) that
703 * needed to be woken up.
705 * NOTE: Wakeups occuring on remote cpus are asynchronous. This
706 * should be ok since we are passing idents in the IPI rather then
709 if ((domain & PWAKEUP_MYCPU) == 0 &&
710 (mask = slpque_cpumasks[id]) != 0
713 * Look for a cpu that might have work to do. Mask out cpus
714 * which have already been processed.
716 * 31xxxxxxxxxxxxxxxxxxxxxxxxxxxxx0
718 * start currentcpu start
721 * 11111111111111110000000000000111 case1
722 * 00000000111111110000000000000000 case2
724 * case1: We started at start_case1 and processed through
725 * to the current cpu. We have to check any bits
726 * after the current cpu, then check bits before
729 * case2: We have already checked all the bits from
730 * start_case2 to the end, and from 0 to the current
731 * cpu. We just have the bits from the current cpu
732 * to start_case2 left to check.
734 startcpu = PWAKEUP_DECODE(domain);
735 if (gd->gd_cpuid >= startcpu) {
739 tmask = mask & ~((gd->gd_cpumask << 1) - 1);
741 nextcpu = bsfl(mask & tmask);
742 lwkt_send_ipiq2(globaldata_find(nextcpu),
743 _wakeup, ident, domain);
745 tmask = (1 << startcpu) - 1;
747 nextcpu = bsfl(mask & tmask);
749 globaldata_find(nextcpu),
750 _wakeup, ident, domain);
757 tmask = ~((gd->gd_cpumask << 1) - 1) &
758 ((1 << startcpu) - 1);
760 nextcpu = bsfl(mask & tmask);
761 lwkt_send_ipiq2(globaldata_find(nextcpu),
762 _wakeup, ident, domain);
768 logtsleep(wakeup_end);
773 * Wakeup all threads tsleep()ing on the specified ident, on all cpus
778 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid));
782 * Wakeup one thread tsleep()ing on the specified ident, on any cpu.
785 wakeup_one(void *ident)
787 /* XXX potentially round-robin the first responding cpu */
788 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid) | PWAKEUP_ONE);
792 * Wakeup threads tsleep()ing on the specified ident on the current cpu
796 wakeup_mycpu(void *ident)
798 _wakeup(ident, PWAKEUP_MYCPU);
802 * Wakeup one thread tsleep()ing on the specified ident on the current cpu
806 wakeup_mycpu_one(void *ident)
808 /* XXX potentially round-robin the first responding cpu */
809 _wakeup(ident, PWAKEUP_MYCPU|PWAKEUP_ONE);
813 * Wakeup all thread tsleep()ing on the specified ident on the specified cpu
817 wakeup_oncpu(globaldata_t gd, void *ident)
821 _wakeup(ident, PWAKEUP_MYCPU);
823 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU);
826 _wakeup(ident, PWAKEUP_MYCPU);
831 * Wakeup one thread tsleep()ing on the specified ident on the specified cpu
835 wakeup_oncpu_one(globaldata_t gd, void *ident)
839 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
841 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
844 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
849 * Wakeup all threads waiting on the specified ident that slept using
850 * the specified domain, on all cpus.
853 wakeup_domain(void *ident, int domain)
855 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid));
859 * Wakeup one thread waiting on the specified ident that slept using
860 * the specified domain, on any cpu.
863 wakeup_domain_one(void *ident, int domain)
865 /* XXX potentially round-robin the first responding cpu */
866 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid) | PWAKEUP_ONE);
872 * Make a process runnable. The MP lock must be held on call. This only
873 * has an effect if we are in SSLEEP. We only break out of the
874 * tsleep if P_BREAKTSLEEP is set, otherwise we just fix-up the state.
876 * NOTE: With the MP lock held we can only safely manipulate the process
877 * structure. We cannot safely manipulate the thread structure.
880 setrunnable(struct proc *p)
883 ASSERT_MP_LOCK_HELD(curthread);
884 p->p_flag &= ~P_STOPPED;
885 if (p->p_stat == SSLEEP && (p->p_flag & P_BREAKTSLEEP)) {
886 unsleep_and_wakeup_thread(p->p_thread);
892 * The process is stopped due to some condition, usually because P_STOPPED
893 * is set but also possibly due to being traced.
895 * NOTE! If the caller sets P_STOPPED, the caller must also clear P_WAITED
896 * because the parent may check the child's status before the child actually
897 * gets to this routine.
899 * This routine is called with the current process only, typically just
900 * before returning to userland.
902 * Setting P_BREAKTSLEEP before entering the tsleep will cause a passive
903 * SIGCONT to break out of the tsleep.
906 tstop(struct proc *p)
908 wakeup((caddr_t)p->p_pptr);
909 p->p_flag |= P_BREAKTSLEEP;
910 tsleep(p, 0, "stop", 0);
914 * Yield / synchronous reschedule. This is a bit tricky because the trap
915 * code might have set a lazy release on the switch function. Setting
916 * P_PASSIVE_ACQ will ensure that the lazy release executes when we call
917 * switch, and that we are given a greater chance of affinity with our
920 * We call lwkt_setpri_self() to rotate our thread to the end of the lwkt
921 * run queue. lwkt_switch() will also execute any assigned passive release
922 * (which usually calls release_curproc()), allowing a same/higher priority
923 * process to be designated as the current process.
925 * While it is possible for a lower priority process to be designated,
926 * it's call to lwkt_maybe_switch() in acquire_curproc() will likely
927 * round-robin back to us and we will be able to re-acquire the current
928 * process designation.
933 struct thread *td = curthread;
934 struct proc *p = td->td_proc;
936 lwkt_setpri_self(td->td_pri & TDPRI_MASK);
938 p->p_flag |= P_PASSIVE_ACQ;
940 p->p_flag &= ~P_PASSIVE_ACQ;
947 * Compute a tenex style load average of a quantity on
948 * 1, 5 and 15 minute intervals.
950 static int loadav_count_runnable(struct proc *p, void *data);
959 allproc_scan(loadav_count_runnable, &nrun);
961 for (i = 0; i < 3; i++) {
962 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
963 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
967 * Schedule the next update to occur after 5 seconds, but add a
968 * random variation to avoid synchronisation with processes that
969 * run at regular intervals.
971 callout_reset(&loadav_callout, hz * 4 + (int)(krandom() % (hz * 2 + 1)),
976 loadav_count_runnable(struct proc *p, void *data)
983 if ((td = p->p_thread) == NULL)
985 if (td->td_flags & TDF_BLOCKED)
999 sched_setup(void *dummy)
1001 callout_init(&loadav_callout);
1002 callout_init(&schedcpu_callout);
1004 /* Kick off timeout driven events by calling first time. */