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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_ktrace.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/condvar.h>
46 #include <sys/kernel.h>
49 #include <sys/mutex.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sched.h>
53 #include <sys/signalvar.h>
54 #include <sys/sleepqueue.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysproto.h>
59 #include <sys/vmmeter.h>
62 #include <sys/ktrace.h>
65 #include <machine/cpu.h>
67 static void synch_setup(void *dummy);
68 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL)
72 static int pause_wchan;
74 static struct callout loadav_callout;
75 static struct callout lbolt_callout;
77 struct loadavg averunnable =
78 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
80 * Constants for averages over 1, 5, and 15 minutes
81 * when sampling at 5 second intervals.
83 static fixpt_t cexp[3] = {
84 0.9200444146293232 * FSCALE, /* exp(-1/12) */
85 0.9834714538216174 * FSCALE, /* exp(-1/60) */
86 0.9944598480048967 * FSCALE, /* exp(-1/180) */
89 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
90 static int fscale __unused = FSCALE;
91 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
93 static void loadav(void *arg);
94 static void lboltcb(void *arg);
100 hogticks = (hz / 10) * 2; /* Default only. */
105 * General sleep call. Suspends the current thread until a wakeup is
106 * performed on the specified identifier. The thread will then be made
107 * runnable with the specified priority. Sleeps at most timo/hz seconds
108 * (0 means no timeout). If pri includes PCATCH flag, signals are checked
109 * before and after sleeping, else signals are not checked. Returns 0 if
110 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
111 * signal needs to be delivered, ERESTART is returned if the current system
112 * call should be restarted if possible, and EINTR is returned if the system
113 * call should be interrupted by the signal (return EINTR).
115 * The lock argument is unlocked before the caller is suspended, and
116 * re-locked before _sleep() returns. If priority includes the PDROP
117 * flag the lock is not re-locked before returning.
120 _sleep(ident, lock, priority, wmesg, timo)
122 struct lock_object *lock;
128 struct lock_class *class;
129 int catch, flags, lock_state, pri, rval;
130 WITNESS_SAVE_DECL(lock_witness);
135 if (KTRPOINT(td, KTR_CSW))
138 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
139 "Sleeping on \"%s\"", wmesg);
140 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL ||
141 ident == &lbolt, ("sleeping without a lock"));
142 KASSERT(p != NULL, ("msleep1"));
143 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
145 class = LOCK_CLASS(lock);
151 * During autoconfiguration, just return;
152 * don't run any other threads or panic below,
153 * in case this is the idle thread and already asleep.
154 * XXX: this used to do "s = splhigh(); splx(safepri);
155 * splx(s);" to give interrupts a chance, but there is
156 * no way to give interrupts a chance now.
158 if (lock != NULL && priority & PDROP)
159 class->lc_unlock(lock);
162 catch = priority & PCATCH;
166 * If we are already on a sleep queue, then remove us from that
167 * sleep queue first. We have to do this to handle recursive
170 if (TD_ON_SLEEPQ(td))
171 sleepq_remove(td, td->td_wchan);
173 if (ident == &pause_wchan)
174 flags = SLEEPQ_PAUSE;
176 flags = SLEEPQ_SLEEP;
178 flags |= SLEEPQ_INTERRUPTIBLE;
181 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
182 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
185 if (lock != NULL && !(class->lc_flags & LC_SLEEPABLE)) {
186 WITNESS_SAVE(lock, lock_witness);
187 lock_state = class->lc_unlock(lock);
189 /* GCC needs to follow the Yellow Brick Road */
193 * We put ourselves on the sleep queue and start our timeout
194 * before calling thread_suspend_check, as we could stop there,
195 * and a wakeup or a SIGCONT (or both) could occur while we were
196 * stopped without resuming us. Thus, we must be ready for sleep
197 * when cursig() is called. If the wakeup happens while we're
198 * stopped, then td will no longer be on a sleep queue upon
199 * return from cursig().
201 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0);
203 sleepq_set_timeout(ident, timo);
204 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
205 sleepq_release(ident);
206 WITNESS_SAVE(lock, lock_witness);
207 lock_state = class->lc_unlock(lock);
212 * Adjust this thread's priority, if necessary.
214 pri = priority & PRIMASK;
215 if (pri != 0 && pri != td->td_priority) {
216 mtx_lock_spin(&sched_lock);
218 mtx_unlock_spin(&sched_lock);
222 rval = sleepq_timedwait_sig(ident);
224 rval = sleepq_timedwait(ident);
226 rval = sleepq_wait_sig(ident);
232 if (KTRPOINT(td, KTR_CSW))
236 if (lock != NULL && !(priority & PDROP)) {
237 class->lc_lock(lock, lock_state);
238 WITNESS_RESTORE(lock, lock_witness);
244 msleep_spin(ident, mtx, wmesg, timo)
253 WITNESS_SAVE_DECL(mtx);
257 KASSERT(mtx != NULL, ("sleeping without a mutex"));
258 KASSERT(p != NULL, ("msleep1"));
259 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
263 * During autoconfiguration, just return;
264 * don't run any other threads or panic below,
265 * in case this is the idle thread and already asleep.
266 * XXX: this used to do "s = splhigh(); splx(safepri);
267 * splx(s);" to give interrupts a chance, but there is
268 * no way to give interrupts a chance now.
274 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
275 td->td_tid, p->p_pid, p->p_comm, wmesg, ident);
278 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
279 WITNESS_SAVE(&mtx->lock_object, mtx);
280 mtx_unlock_spin(mtx);
283 * We put ourselves on the sleep queue and start our timeout.
285 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
287 sleepq_set_timeout(ident, timo);
290 * Can't call ktrace with any spin locks held so it can lock the
291 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
292 * any spin lock. Thus, we have to drop the sleepq spin lock while
293 * we handle those requests. This is safe since we have placed our
294 * thread on the sleep queue already.
297 if (KTRPOINT(td, KTR_CSW)) {
298 sleepq_release(ident);
304 sleepq_release(ident);
305 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
310 rval = sleepq_timedwait(ident);
316 if (KTRPOINT(td, KTR_CSW))
321 WITNESS_RESTORE(&mtx->lock_object, mtx);
326 * pause() is like tsleep() except that the intention is to not be
327 * explicitly woken up by another thread. Instead, the current thread
328 * simply wishes to sleep until the timeout expires. It is
329 * implemented using a dummy wait channel.
337 KASSERT(timo != 0, ("pause: timeout required"));
338 return (tsleep(&pause_wchan, 0, wmesg, timo));
342 * Make all threads sleeping on the specified identifier runnable.
346 register void *ident;
350 sleepq_broadcast(ident, SLEEPQ_SLEEP, -1, 0);
354 * Make a thread sleeping on the specified identifier runnable.
355 * May wake more than one thread if a target thread is currently
360 register void *ident;
364 sleepq_signal(ident, SLEEPQ_SLEEP, -1, 0);
368 * The machine independent parts of context switching.
371 mi_switch(int flags, struct thread *newtd)
373 uint64_t new_switchtime;
377 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
378 td = curthread; /* XXX */
379 p = td->td_proc; /* XXX */
380 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
382 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
383 mtx_assert(&Giant, MA_NOTOWNED);
385 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
386 (td->td_owepreempt) && (flags & SW_INVOL) != 0 &&
387 newtd == NULL) || panicstr,
388 ("mi_switch: switch in a critical section"));
389 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
390 ("mi_switch: switch must be voluntary or involuntary"));
391 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
394 * Don't perform context switches from the debugger.
397 mtx_unlock_spin(&sched_lock);
400 panic("%s: did not reenter debugger", __func__);
404 p->p_stats->p_ru.ru_nvcsw++;
406 p->p_stats->p_ru.ru_nivcsw++;
409 * Compute the amount of time during which the current
410 * process was running, and add that to its total so far.
412 new_switchtime = cpu_ticks();
413 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime));
414 p->p_rux.rux_uticks += td->td_uticks;
416 p->p_rux.rux_iticks += td->td_iticks;
418 p->p_rux.rux_sticks += td->td_sticks;
421 td->td_generation++; /* bump preempt-detect counter */
424 * Check if the process exceeds its cpu resource allocation. If
425 * it reaches the max, arrange to kill the process in ast().
427 if (p->p_cpulimit != RLIM_INFINITY &&
428 p->p_rux.rux_runtime >= p->p_cpulimit * cpu_tickrate()) {
429 p->p_sflag |= PS_XCPU;
430 td->td_flags |= TDF_ASTPENDING;
434 * Finish up stats for outgoing thread.
437 PCPU_SET(switchtime, new_switchtime);
438 PCPU_SET(switchticks, ticks);
439 CTR4(KTR_PROC, "mi_switch: old thread %ld (kse %p, pid %ld, %s)",
440 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
441 #if (KTR_COMPILE & KTR_SCHED) != 0
442 if (TD_IS_IDLETHREAD(td))
443 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",
444 td, td->td_proc->p_comm, td->td_priority);
445 else if (newtd != NULL)
447 "mi_switch: %p(%s) prio %d preempted by %p(%s)",
448 td, td->td_proc->p_comm, td->td_priority, newtd,
449 newtd->td_proc->p_comm);
452 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s",
453 td, td->td_proc->p_comm, td->td_priority,
454 td->td_inhibitors, td->td_wmesg, td->td_lockname);
457 * We call thread_switchout after the KTR_SCHED prints above so kse
458 * selecting a new thread to run does not show up as a preemption.
461 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
462 newtd = thread_switchout(td, flags, newtd);
464 sched_switch(td, newtd, flags);
465 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d",
466 td, td->td_proc->p_comm, td->td_priority);
468 CTR4(KTR_PROC, "mi_switch: new thread %ld (kse %p, pid %ld, %s)",
469 td->td_tid, td->td_sched, p->p_pid, p->p_comm);
472 * If the last thread was exiting, finish cleaning it up.
474 if ((td = PCPU_GET(deadthread))) {
475 PCPU_SET(deadthread, NULL);
481 * Change process state to be runnable,
482 * placing it on the run queue if it is in memory,
483 * and awakening the swapper if it isn't in memory.
486 setrunnable(struct thread *td)
491 mtx_assert(&sched_lock, MA_OWNED);
492 switch (p->p_state) {
494 panic("setrunnable(1)");
498 switch (td->td_state) {
504 * If we are only inhibited because we are swapped out
505 * then arange to swap in this process. Otherwise just return.
507 if (td->td_inhibitors != TDI_SWAPPED)
509 /* XXX: intentional fall-through ? */
513 printf("state is 0x%x", td->td_state);
514 panic("setrunnable(2)");
516 if ((p->p_sflag & PS_INMEM) == 0) {
517 if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
518 p->p_sflag |= PS_SWAPINREQ;
520 * due to a LOR between sched_lock and
521 * the sleepqueue chain locks, use
522 * lower level scheduling functions.
531 * Compute a tenex style load average of a quantity on
532 * 1, 5 and 15 minute intervals.
533 * XXXKSE Needs complete rewrite when correct info is available.
534 * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
545 for (i = 0; i < 3; i++)
546 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
547 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
550 * Schedule the next update to occur after 5 seconds, but add a
551 * random variation to avoid synchronisation with processes that
552 * run at regular intervals.
554 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
562 callout_reset(&lbolt_callout, hz, lboltcb, NULL);
570 callout_init(&loadav_callout, CALLOUT_MPSAFE);
571 callout_init(&lbolt_callout, CALLOUT_MPSAFE);
573 /* Kick off timeout driven events by calling first time. */
579 * General purpose yield system call.
582 yield(struct thread *td, struct yield_args *uap)
584 mtx_assert(&Giant, MA_NOTOWNED);
586 sched_relinquish(td);