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 1.61 2006/05/25 07:36:34 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>
53 #include <sys/thread2.h>
57 #include <sys/ktrace.h>
59 #include <sys/xwait.h>
62 #include <machine/cpu.h>
63 #include <machine/ipl.h>
64 #include <machine/smp.h>
66 TAILQ_HEAD(tslpque, thread);
68 static void sched_setup (void *dummy);
69 SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
74 int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
76 int ncpus2, ncpus2_shift, ncpus2_mask;
79 static struct callout loadav_callout;
80 static struct callout schedcpu_callout;
81 MALLOC_DEFINE(M_TSLEEP, "tslpque", "tsleep queues");
83 #if !defined(KTR_TSLEEP)
84 #define KTR_TSLEEP KTR_ALL
86 KTR_INFO_MASTER(tsleep);
87 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_beg, 0, "tsleep enter", 0);
88 KTR_INFO(KTR_TSLEEP, tsleep, tsleep_end, 0, "tsleep exit", 0);
89 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_beg, 0, "wakeup enter", 0);
90 KTR_INFO(KTR_TSLEEP, tsleep, wakeup_end, 0, "wakeup exit", 0);
91 #define logtsleep(name) KTR_LOG(tsleep_ ## name)
93 struct loadavg averunnable =
94 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
96 * Constants for averages over 1, 5, and 15 minutes
97 * when sampling at 5 second intervals.
99 static fixpt_t cexp[3] = {
100 0.9200444146293232 * FSCALE, /* exp(-1/12) */
101 0.9834714538216174 * FSCALE, /* exp(-1/60) */
102 0.9944598480048967 * FSCALE, /* exp(-1/180) */
105 static void endtsleep (void *);
106 static void unsleep_and_wakeup_thread(struct thread *td);
107 static void loadav (void *arg);
108 static void schedcpu (void *arg);
111 * Adjust the scheduler quantum. The quantum is specified in microseconds.
112 * Note that 'tick' is in microseconds per tick.
115 sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
119 new_val = sched_quantum * tick;
120 error = sysctl_handle_int(oidp, &new_val, 0, req);
121 if (error != 0 || req->newptr == NULL)
125 sched_quantum = new_val / tick;
126 hogticks = 2 * sched_quantum;
130 SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW,
131 0, sizeof sched_quantum, sysctl_kern_quantum, "I", "");
134 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
135 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
136 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
138 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
139 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
141 * If you don't want to bother with the faster/more-accurate formula, you
142 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
143 * (more general) method of calculating the %age of CPU used by a process.
145 * decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing
147 #define CCPU_SHIFT 11
149 static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
150 SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
153 * kernel uses `FSCALE', userland (SHOULD) use kern.fscale
155 static int fscale __unused = FSCALE;
156 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
159 * Recompute process priorities, once a second.
161 * Since the userland schedulers are typically event oriented, if the
162 * estcpu calculation at wakeup() time is not sufficient to make a
163 * process runnable relative to other processes in the system we have
164 * a 1-second recalc to help out.
166 * This code also allows us to store sysclock_t data in the process structure
167 * without fear of an overrun, since sysclock_t are guarenteed to hold
168 * several seconds worth of count.
170 * WARNING! callouts can preempt normal threads. However, they will not
171 * preempt a thread holding a spinlock so we *can* safely use spinlocks.
173 static int schedcpu_stats(struct proc *p, void *data __unused);
174 static int schedcpu_resource(struct proc *p, void *data __unused);
179 allproc_scan(schedcpu_stats, NULL);
180 allproc_scan(schedcpu_resource, NULL);
181 wakeup((caddr_t)&lbolt);
182 wakeup((caddr_t)&lbolt_syncer);
183 callout_reset(&schedcpu_callout, hz, schedcpu, NULL);
187 * General process statistics once a second
190 schedcpu_stats(struct proc *p, void *data __unused)
194 if (p->p_stat == SSLEEP)
198 * Only recalculate processes that are active or have slept
199 * less then 2 seconds. The schedulers understand this.
201 if (p->p_slptime <= 1) {
202 p->p_usched->recalculate(&p->p_lwp);
204 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
211 * Resource checks. XXX break out since psignal/killproc can block,
212 * limiting us to one process killed per second. There is probably
216 schedcpu_resource(struct proc *p, void *data __unused)
221 if (p->p_stat == SIDL ||
222 (p->p_flag & P_ZOMBIE) ||
223 p->p_limit == NULL ||
230 ttime = p->p_thread->td_sticks + p->p_thread->td_uticks;
232 switch(plimit_testcpulimit(p->p_limit, ttime)) {
233 case PLIMIT_TESTCPU_KILL:
234 killproc(p, "exceeded maximum CPU limit");
236 case PLIMIT_TESTCPU_XCPU:
237 if ((p->p_flag & P_XCPU) == 0) {
250 * This is only used by ps. Generate a cpu percentage use over
251 * a period of one second.
254 updatepcpu(struct lwp *lp, int cpticks, int ttlticks)
259 acc = (cpticks << FSHIFT) / ttlticks;
260 if (ttlticks >= ESTCPUFREQ) {
261 lp->lwp_pctcpu = acc;
263 remticks = ESTCPUFREQ - ttlticks;
264 lp->lwp_pctcpu = (acc * ttlticks + lp->lwp_pctcpu * remticks) /
270 * We're only looking at 7 bits of the address; everything is
271 * aligned to 4, lots of things are aligned to greater powers
272 * of 2. Shift right by 8, i.e. drop the bottom 256 worth.
274 #define TABLESIZE 128
275 #define LOOKUP(x) (((intptr_t)(x) >> 8) & (TABLESIZE - 1))
277 static cpumask_t slpque_cpumasks[TABLESIZE];
280 * General scheduler initialization. We force a reschedule 25 times
281 * a second by default. Note that cpu0 is initialized in early boot and
282 * cannot make any high level calls.
284 * Each cpu has its own sleep queue.
287 sleep_gdinit(globaldata_t gd)
289 static struct tslpque slpque_cpu0[TABLESIZE];
292 if (gd->gd_cpuid == 0) {
293 sched_quantum = (hz + 24) / 25;
294 hogticks = 2 * sched_quantum;
296 gd->gd_tsleep_hash = slpque_cpu0;
298 gd->gd_tsleep_hash = malloc(sizeof(slpque_cpu0),
299 M_TSLEEP, M_WAITOK | M_ZERO);
301 for (i = 0; i < TABLESIZE; ++i)
302 TAILQ_INIT(&gd->gd_tsleep_hash[i]);
306 * General sleep call. Suspends the current process until a wakeup is
307 * performed on the specified identifier. The process will then be made
308 * runnable with the specified priority. Sleeps at most timo/hz seconds
309 * (0 means no timeout). If flags includes PCATCH flag, signals are checked
310 * before and after sleeping, else signals are not checked. Returns 0 if
311 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
312 * signal needs to be delivered, ERESTART is returned if the current system
313 * call should be restarted if possible, and EINTR is returned if the system
314 * call should be interrupted by the signal (return EINTR).
316 * Note that if we are a process, we release_curproc() before messing with
317 * the LWKT scheduler.
319 * During autoconfiguration or after a panic, a sleep will simply
320 * lower the priority briefly to allow interrupts, then return.
323 tsleep(void *ident, int flags, const char *wmesg, int timo)
325 struct thread *td = curthread;
326 struct proc *p = td->td_proc; /* may be NULL */
333 struct callout thandle;
336 * NOTE: removed KTRPOINT, it could cause races due to blocking
337 * even in stable. Just scrap it for now.
339 if (cold || panicstr) {
341 * After a panic, or during autoconfiguration,
342 * just give interrupts a chance, then just return;
343 * don't run any other procs or panic below,
344 * in case this is the idle process and already asleep.
347 oldpri = td->td_pri & TDPRI_MASK;
348 lwkt_setpri_self(safepri);
350 lwkt_setpri_self(oldpri);
353 logtsleep(tsleep_beg);
355 KKASSERT(td != &gd->gd_idlethread); /* you must be kidding! */
358 * NOTE: all of this occurs on the current cpu, including any
359 * callout-based wakeups, so a critical section is a sufficient
362 * The entire sequence through to where we actually sleep must
363 * run without breaking the critical section.
366 catch = flags & PCATCH;
370 crit_enter_quick(td);
372 KASSERT(ident != NULL, ("tsleep: no ident"));
373 KASSERT(p == NULL || p->p_stat == SRUN, ("tsleep %p %s %d",
374 ident, wmesg, p->p_stat));
377 * Setup for the current process (if this is a process).
382 * Early termination if PCATCH was set and a
383 * signal is pending, interlocked with the
386 * Early termination only occurs when tsleep() is
387 * entered while in a normal SRUN state.
389 if ((sig = CURSIG(p)) != 0)
393 * Causes psignal to wake us up when.
395 p->p_flag |= P_SINTR;
399 * Make sure the current process has been untangled from
400 * the userland scheduler and initialize slptime to start
403 if (flags & PNORESCHED)
404 td->td_flags |= TDF_NORESCHED;
405 p->p_usched->release_curproc(&p->p_lwp);
410 * Move our thread to the correct queue and setup our wchan, etc.
412 lwkt_deschedule_self(td);
413 td->td_flags |= TDF_TSLEEPQ;
414 TAILQ_INSERT_TAIL(&gd->gd_tsleep_hash[id], td, td_threadq);
415 atomic_set_int(&slpque_cpumasks[id], gd->gd_cpumask);
417 td->td_wchan = ident;
418 td->td_wmesg = wmesg;
419 td->td_wdomain = flags & PDOMAIN_MASK;
422 * Setup the timeout, if any
425 callout_init(&thandle);
426 callout_reset(&thandle, timo, endtsleep, td);
434 * Ok, we are sleeping. Remove us from the userland runq
435 * and place us in the SSLEEP state.
437 if (p->p_flag & P_ONRUNQ)
438 p->p_usched->remrunqueue(&p->p_lwp);
440 p->p_stats->p_ru.ru_nvcsw++;
448 * Make sure we haven't switched cpus while we were asleep. It's
449 * not supposed to happen. Cleanup our temporary flags.
451 KKASSERT(gd == td->td_gd);
452 td->td_flags &= ~TDF_NORESCHED;
455 * Cleanup the timeout.
458 if (td->td_flags & TDF_TIMEOUT) {
459 td->td_flags &= ~TDF_TIMEOUT;
463 callout_stop(&thandle);
468 * Since td_threadq is used both for our run queue AND for the
469 * tsleep hash queue, we can't still be on it at this point because
470 * we've gotten cpu back.
472 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0, ("tsleep: impossible thread flags %08x", td->td_flags));
478 * Figure out the correct error return
482 p->p_flag &= ~(P_BREAKTSLEEP | P_SINTR);
483 if (catch && error == 0 && (sig != 0 || (sig = CURSIG(p)))) {
484 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
490 logtsleep(tsleep_end);
496 * This is a dandy function that allows us to interlock tsleep/wakeup
497 * operations with unspecified upper level locks, such as lockmgr locks,
498 * simply by holding a critical section. The sequence is:
500 * (enter critical section)
501 * (acquire upper level lock)
502 * tsleep_interlock(blah)
503 * (release upper level lock)
505 * (exit critical section)
507 * Basically this function sets our cpumask for the ident which informs
508 * other cpus that our cpu 'might' be waiting (or about to wait on) the
509 * hash index related to the ident. The critical section prevents another
510 * cpu's wakeup() from being processed on our cpu until we are actually
511 * able to enter the tsleep(). Thus, no race occurs between our attempt
512 * to release a resource and sleep, and another cpu's attempt to acquire
513 * a resource and call wakeup.
515 * There isn't much of a point to this function unless you call it while
516 * holding a critical section.
519 tsleep_interlock(void *ident)
521 int id = LOOKUP(ident);
523 atomic_set_int(&slpque_cpumasks[id], mycpu->gd_cpumask);
527 * Implement the timeout for tsleep.
529 * We set P_BREAKTSLEEP to indicate that an event has occured, but
530 * we only call setrunnable if the process is not stopped.
532 * This type of callout timeout is scheduled on the same cpu the process
533 * is sleeping on. Also, at the moment, the MP lock is held.
541 ASSERT_MP_LOCK_HELD(curthread);
545 * cpu interlock. Thread flags are only manipulated on
546 * the cpu owning the thread. proc flags are only manipulated
547 * by the older of the MP lock. We have both.
549 if (td->td_flags & TDF_TSLEEPQ) {
550 td->td_flags |= TDF_TIMEOUT;
552 if ((p = td->td_proc) != NULL) {
553 p->p_flag |= P_BREAKTSLEEP;
554 if ((p->p_flag & P_STOPPED) == 0)
557 unsleep_and_wakeup_thread(td);
564 * Unsleep and wakeup a thread. This function runs without the MP lock
565 * which means that it can only manipulate thread state on the owning cpu,
566 * and cannot touch the process state at all.
570 unsleep_and_wakeup_thread(struct thread *td)
572 globaldata_t gd = mycpu;
576 if (td->td_gd != gd) {
577 lwkt_send_ipiq(td->td_gd, (ipifunc1_t)unsleep_and_wakeup_thread, td);
582 if (td->td_flags & TDF_TSLEEPQ) {
583 td->td_flags &= ~TDF_TSLEEPQ;
584 id = LOOKUP(td->td_wchan);
585 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_threadq);
586 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL)
587 atomic_clear_int(&slpque_cpumasks[id], gd->gd_cpumask);
594 * Make all processes sleeping on the specified identifier runnable.
595 * count may be zero or one only.
597 * The domain encodes the sleep/wakeup domain AND the first cpu to check
598 * (which is always the current cpu). As we iterate across cpus
600 * This call may run without the MP lock held. We can only manipulate thread
601 * state on the cpu owning the thread. We CANNOT manipulate process state
605 _wakeup(void *ident, int domain)
620 logtsleep(wakeup_beg);
623 qp = &gd->gd_tsleep_hash[id];
625 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
626 ntd = TAILQ_NEXT(td, td_threadq);
627 if (td->td_wchan == ident &&
628 td->td_wdomain == (domain & PDOMAIN_MASK)
630 KKASSERT(td->td_flags & TDF_TSLEEPQ);
631 td->td_flags &= ~TDF_TSLEEPQ;
632 TAILQ_REMOVE(qp, td, td_threadq);
633 if (TAILQ_FIRST(qp) == NULL) {
634 atomic_clear_int(&slpque_cpumasks[id],
638 if (domain & PWAKEUP_ONE)
646 * We finished checking the current cpu but there still may be
647 * more work to do. Either wakeup_one was requested and no matching
648 * thread was found, or a normal wakeup was requested and we have
649 * to continue checking cpus.
651 * The cpu that started the wakeup sequence is encoded in the domain.
652 * We use this information to determine which cpus still need to be
653 * checked, locate a candidate cpu, and chain the wakeup
654 * asynchronously with an IPI message.
656 * It should be noted that this scheme is actually less expensive then
657 * the old scheme when waking up multiple threads, since we send
658 * only one IPI message per target candidate which may then schedule
659 * multiple threads. Before we could have wound up sending an IPI
660 * message for each thread on the target cpu (!= current cpu) that
661 * needed to be woken up.
663 * NOTE: Wakeups occuring on remote cpus are asynchronous. This
664 * should be ok since we are passing idents in the IPI rather then
667 if ((domain & PWAKEUP_MYCPU) == 0 &&
668 (mask = slpque_cpumasks[id]) != 0
671 * Look for a cpu that might have work to do. Mask out cpus
672 * which have already been processed.
674 * 31xxxxxxxxxxxxxxxxxxxxxxxxxxxxx0
676 * start currentcpu start
679 * 11111111111111110000000000000111 case1
680 * 00000000111111110000000000000000 case2
682 * case1: We started at start_case1 and processed through
683 * to the current cpu. We have to check any bits
684 * after the current cpu, then check bits before
687 * case2: We have already checked all the bits from
688 * start_case2 to the end, and from 0 to the current
689 * cpu. We just have the bits from the current cpu
690 * to start_case2 left to check.
692 startcpu = PWAKEUP_DECODE(domain);
693 if (gd->gd_cpuid >= startcpu) {
697 tmask = mask & ~((gd->gd_cpumask << 1) - 1);
699 nextcpu = bsfl(mask & tmask);
700 lwkt_send_ipiq2(globaldata_find(nextcpu),
701 _wakeup, ident, domain);
703 tmask = (1 << startcpu) - 1;
705 nextcpu = bsfl(mask & tmask);
707 globaldata_find(nextcpu),
708 _wakeup, ident, domain);
715 tmask = ~((gd->gd_cpumask << 1) - 1) &
716 ((1 << startcpu) - 1);
718 nextcpu = bsfl(mask & tmask);
719 lwkt_send_ipiq2(globaldata_find(nextcpu),
720 _wakeup, ident, domain);
726 logtsleep(wakeup_end);
731 * Wakeup all threads tsleep()ing on the specified ident, on all cpus
736 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid));
740 * Wakeup one thread tsleep()ing on the specified ident, on any cpu.
743 wakeup_one(void *ident)
745 /* XXX potentially round-robin the first responding cpu */
746 _wakeup(ident, PWAKEUP_ENCODE(0, mycpu->gd_cpuid) | PWAKEUP_ONE);
750 * Wakeup threads tsleep()ing on the specified ident on the current cpu
754 wakeup_mycpu(void *ident)
756 _wakeup(ident, PWAKEUP_MYCPU);
760 * Wakeup one thread tsleep()ing on the specified ident on the current cpu
764 wakeup_mycpu_one(void *ident)
766 /* XXX potentially round-robin the first responding cpu */
767 _wakeup(ident, PWAKEUP_MYCPU|PWAKEUP_ONE);
771 * Wakeup all thread tsleep()ing on the specified ident on the specified cpu
775 wakeup_oncpu(globaldata_t gd, void *ident)
779 _wakeup(ident, PWAKEUP_MYCPU);
781 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU);
784 _wakeup(ident, PWAKEUP_MYCPU);
789 * Wakeup one thread tsleep()ing on the specified ident on the specified cpu
793 wakeup_oncpu_one(globaldata_t gd, void *ident)
797 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
799 lwkt_send_ipiq2(gd, _wakeup, ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
802 _wakeup(ident, PWAKEUP_MYCPU | PWAKEUP_ONE);
807 * Wakeup all threads waiting on the specified ident that slept using
808 * the specified domain, on all cpus.
811 wakeup_domain(void *ident, int domain)
813 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid));
817 * Wakeup one thread waiting on the specified ident that slept using
818 * the specified domain, on any cpu.
821 wakeup_domain_one(void *ident, int domain)
823 /* XXX potentially round-robin the first responding cpu */
824 _wakeup(ident, PWAKEUP_ENCODE(domain, mycpu->gd_cpuid) | PWAKEUP_ONE);
830 * Make a process runnable. The MP lock must be held on call. This only
831 * has an effect if we are in SSLEEP. We only break out of the
832 * tsleep if P_BREAKTSLEEP is set, otherwise we just fix-up the state.
834 * NOTE: With the MP lock held we can only safely manipulate the process
835 * structure. We cannot safely manipulate the thread structure.
838 setrunnable(struct proc *p)
841 ASSERT_MP_LOCK_HELD(curthread);
842 p->p_flag &= ~P_STOPPED;
843 if (p->p_stat == SSLEEP && (p->p_flag & P_BREAKTSLEEP)) {
844 unsleep_and_wakeup_thread(p->p_thread);
850 * The process is stopped due to some condition, usually because P_STOPPED
851 * is set but also possibly due to being traced.
853 * NOTE! If the caller sets P_STOPPED, the caller must also clear P_WAITED
854 * because the parent may check the child's status before the child actually
855 * gets to this routine.
857 * This routine is called with the current process only, typically just
858 * before returning to userland.
860 * Setting P_BREAKTSLEEP before entering the tsleep will cause a passive
861 * SIGCONT to break out of the tsleep.
864 tstop(struct proc *p)
866 wakeup((caddr_t)p->p_pptr);
867 p->p_flag |= P_BREAKTSLEEP;
868 tsleep(p, 0, "stop", 0);
872 * Yield / synchronous reschedule. This is a bit tricky because the trap
873 * code might have set a lazy release on the switch function. Setting
874 * P_PASSIVE_ACQ will ensure that the lazy release executes when we call
875 * switch, and that we are given a greater chance of affinity with our
878 * We call lwkt_setpri_self() to rotate our thread to the end of the lwkt
879 * run queue. lwkt_switch() will also execute any assigned passive release
880 * (which usually calls release_curproc()), allowing a same/higher priority
881 * process to be designated as the current process.
883 * While it is possible for a lower priority process to be designated,
884 * it's call to lwkt_maybe_switch() in acquire_curproc() will likely
885 * round-robin back to us and we will be able to re-acquire the current
886 * process designation.
891 struct thread *td = curthread;
892 struct proc *p = td->td_proc;
894 lwkt_setpri_self(td->td_pri & TDPRI_MASK);
896 p->p_flag |= P_PASSIVE_ACQ;
898 p->p_flag &= ~P_PASSIVE_ACQ;
905 * Compute a tenex style load average of a quantity on
906 * 1, 5 and 15 minute intervals.
908 static int loadav_count_runnable(struct proc *p, void *data);
917 allproc_scan(loadav_count_runnable, &nrun);
919 for (i = 0; i < 3; i++) {
920 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
921 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
925 * Schedule the next update to occur after 5 seconds, but add a
926 * random variation to avoid synchronisation with processes that
927 * run at regular intervals.
929 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
934 loadav_count_runnable(struct proc *p, void *data)
941 if ((td = p->p_thread) == NULL)
943 if (td->td_flags & TDF_BLOCKED)
957 sched_setup(void *dummy)
959 callout_init(&loadav_callout);
960 callout_init(&schedcpu_callout);
962 /* Kick off timeout driven events by calling first time. */