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_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_sched.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/condvar.h>
48 #include <sys/kernel.h>
51 #include <sys/mutex.h>
53 #include <sys/resourcevar.h>
54 #include <sys/sched.h>
56 #include <sys/signalvar.h>
57 #include <sys/sleepqueue.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysproto.h>
62 #include <sys/vmmeter.h>
65 #include <sys/ktrace.h>
68 #include <machine/cpu.h>
72 #include <vm/vm_param.h>
76 static void synch_setup(void *dummy);
77 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
81 static uint8_t pause_wchan[MAXCPU];
83 static struct callout loadav_callout;
85 struct loadavg averunnable =
86 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
88 * Constants for averages over 1, 5, and 15 minutes
89 * when sampling at 5 second intervals.
91 static fixpt_t cexp[3] = {
92 0.9200444146293232 * FSCALE, /* exp(-1/12) */
93 0.9834714538216174 * FSCALE, /* exp(-1/60) */
94 0.9944598480048967 * FSCALE, /* exp(-1/180) */
97 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
98 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
100 static void loadav(void *arg);
102 SDT_PROVIDER_DECLARE(sched);
103 SDT_PROBE_DEFINE(sched, , , preempt);
106 * These probes reference Solaris features that are not implemented in FreeBSD.
107 * Create the probes anyway for compatibility with existing D scripts; they'll
110 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep);
111 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
112 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
113 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
114 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
117 sleepinit(void *unused)
120 hogticks = (hz / 10) * 2; /* Default only. */
125 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
128 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
131 * General sleep call. Suspends the current thread until a wakeup is
132 * performed on the specified identifier. The thread will then be made
133 * runnable with the specified priority. Sleeps at most sbt units of time
134 * (0 means no timeout). If pri includes the PCATCH flag, let signals
135 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
136 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
137 * signal becomes pending, ERESTART is returned if the current system
138 * call should be restarted if possible, and EINTR is returned if the system
139 * call should be interrupted by the signal (return EINTR).
141 * The lock argument is unlocked before the caller is suspended, and
142 * re-locked before _sleep() returns. If priority includes the PDROP
143 * flag the lock is not re-locked before returning.
146 _sleep(void *ident, struct lock_object *lock, int priority,
147 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
151 struct lock_class *class;
152 uintptr_t lock_state;
153 int catch, pri, rval, sleepq_flags;
154 WITNESS_SAVE_DECL(lock_witness);
159 if (KTRPOINT(td, KTR_CSW))
162 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
163 "Sleeping on \"%s\"", wmesg);
164 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
165 ("sleeping without a lock"));
166 KASSERT(p != NULL, ("msleep1"));
167 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
168 if (priority & PDROP)
169 KASSERT(lock != NULL && lock != &Giant.lock_object,
170 ("PDROP requires a non-Giant lock"));
172 class = LOCK_CLASS(lock);
176 if (cold || SCHEDULER_STOPPED()) {
178 * During autoconfiguration, just return;
179 * don't run any other threads or panic below,
180 * in case this is the idle thread and already asleep.
181 * XXX: this used to do "s = splhigh(); splx(safepri);
182 * splx(s);" to give interrupts a chance, but there is
183 * no way to give interrupts a chance now.
185 if (lock != NULL && priority & PDROP)
186 class->lc_unlock(lock);
189 catch = priority & PCATCH;
190 pri = priority & PRIMASK;
192 KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep"));
194 if ((uint8_t *)ident >= &pause_wchan[0] &&
195 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
196 sleepq_flags = SLEEPQ_PAUSE;
198 sleepq_flags = SLEEPQ_SLEEP;
200 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
203 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
204 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
206 if (lock == &Giant.lock_object)
207 mtx_assert(&Giant, MA_OWNED);
209 if (lock != NULL && lock != &Giant.lock_object &&
210 !(class->lc_flags & LC_SLEEPABLE)) {
211 WITNESS_SAVE(lock, lock_witness);
212 lock_state = class->lc_unlock(lock);
214 /* GCC needs to follow the Yellow Brick Road */
218 * We put ourselves on the sleep queue and start our timeout
219 * before calling thread_suspend_check, as we could stop there,
220 * and a wakeup or a SIGCONT (or both) could occur while we were
221 * stopped without resuming us. Thus, we must be ready for sleep
222 * when cursig() is called. If the wakeup happens while we're
223 * stopped, then td will no longer be on a sleep queue upon
224 * return from cursig().
226 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
228 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
229 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
230 sleepq_release(ident);
231 WITNESS_SAVE(lock, lock_witness);
232 lock_state = class->lc_unlock(lock);
235 if (sbt != 0 && catch)
236 rval = sleepq_timedwait_sig(ident, pri);
238 rval = sleepq_timedwait(ident, pri);
240 rval = sleepq_wait_sig(ident, pri);
242 sleepq_wait(ident, pri);
246 if (KTRPOINT(td, KTR_CSW))
250 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
251 class->lc_lock(lock, lock_state);
252 WITNESS_RESTORE(lock, lock_witness);
258 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
259 sbintime_t sbt, sbintime_t pr, int flags)
264 WITNESS_SAVE_DECL(mtx);
268 KASSERT(mtx != NULL, ("sleeping without a mutex"));
269 KASSERT(p != NULL, ("msleep1"));
270 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
272 if (cold || SCHEDULER_STOPPED()) {
274 * During autoconfiguration, just return;
275 * don't run any other threads or panic below,
276 * in case this is the idle thread and already asleep.
277 * XXX: this used to do "s = splhigh(); splx(safepri);
278 * splx(s);" to give interrupts a chance, but there is
279 * no way to give interrupts a chance now.
285 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
286 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
289 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
290 WITNESS_SAVE(&mtx->lock_object, mtx);
291 mtx_unlock_spin(mtx);
294 * We put ourselves on the sleep queue and start our timeout.
296 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
298 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
301 * Can't call ktrace with any spin locks held so it can lock the
302 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
303 * any spin lock. Thus, we have to drop the sleepq spin lock while
304 * we handle those requests. This is safe since we have placed our
305 * thread on the sleep queue already.
308 if (KTRPOINT(td, KTR_CSW)) {
309 sleepq_release(ident);
315 sleepq_release(ident);
316 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
321 rval = sleepq_timedwait(ident, 0);
323 sleepq_wait(ident, 0);
327 if (KTRPOINT(td, KTR_CSW))
332 WITNESS_RESTORE(&mtx->lock_object, mtx);
337 * pause_sbt() delays the calling thread by the given signed binary
338 * time. During cold bootup, pause_sbt() uses the DELAY() function
339 * instead of the _sleep() function to do the waiting. The "sbt"
340 * argument must be greater than or equal to zero. A "sbt" value of
341 * zero is equivalent to a "sbt" value of one tick.
344 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
346 KASSERT(sbt >= 0, ("pause_sbt: timeout must be >= 0"));
348 /* silently convert invalid timeouts */
352 if (cold || kdb_active || SCHEDULER_STOPPED()) {
354 * We delay one second at a time to avoid overflowing the
355 * system specific DELAY() function(s):
357 while (sbt >= SBT_1S) {
361 /* Do the delay remainder, if any */
362 sbt = (sbt + SBT_1US - 1) / SBT_1US;
365 return (EWOULDBLOCK);
367 return (_sleep(&pause_wchan[curcpu], NULL,
368 (flags & C_CATCH) ? PCATCH : 0, wmesg, sbt, pr, flags));
372 * Make all threads sleeping on the specified identifier runnable.
380 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
381 sleepq_release(ident);
382 if (wakeup_swapper) {
383 KASSERT(ident != &proc0,
384 ("wakeup and wakeup_swapper and proc0"));
390 * Make a thread sleeping on the specified identifier runnable.
391 * May wake more than one thread if a target thread is currently
395 wakeup_one(void *ident)
400 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
401 sleepq_release(ident);
409 thread_unlock(curthread);
412 panic("%s: did not reenter debugger", __func__);
416 * The machine independent parts of context switching.
419 mi_switch(int flags, struct thread *newtd)
421 uint64_t runtime, new_switchtime;
424 td = curthread; /* XXX */
425 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
426 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
428 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
429 mtx_assert(&Giant, MA_NOTOWNED);
431 KASSERT(td->td_critnest == 1 || panicstr,
432 ("mi_switch: switch in a critical section"));
433 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
434 ("mi_switch: switch must be voluntary or involuntary"));
435 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
438 * Don't perform context switches from the debugger.
442 if (SCHEDULER_STOPPED())
444 if (flags & SW_VOL) {
445 td->td_ru.ru_nvcsw++;
446 td->td_swvoltick = ticks;
448 td->td_ru.ru_nivcsw++;
450 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
453 * Compute the amount of time during which the current
454 * thread was running, and add that to its total so far.
456 new_switchtime = cpu_ticks();
457 runtime = new_switchtime - PCPU_GET(switchtime);
458 td->td_runtime += runtime;
459 td->td_incruntime += runtime;
460 PCPU_SET(switchtime, new_switchtime);
461 td->td_generation++; /* bump preempt-detect counter */
462 PCPU_INC(cnt.v_swtch);
463 PCPU_SET(switchticks, ticks);
464 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
465 td->td_tid, td->td_sched, td->td_proc->p_pid, td->td_name);
467 if ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 &&
468 (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED))
469 SDT_PROBE0(sched, , , preempt);
474 sched_switch(td, newtd, flags);
475 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
476 td->td_tid, td->td_sched, td->td_proc->p_pid, td->td_name);
479 * If the last thread was exiting, finish cleaning it up.
481 if ((td = PCPU_GET(deadthread))) {
482 PCPU_SET(deadthread, NULL);
488 * Change thread state to be runnable, placing it on the run queue if
489 * it is in memory. If it is swapped out, return true so our caller
490 * will know to awaken the swapper.
493 setrunnable(struct thread *td)
496 THREAD_LOCK_ASSERT(td, MA_OWNED);
497 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
498 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
499 switch (td->td_state) {
505 * If we are only inhibited because we are swapped out
506 * then arange to swap in this process. Otherwise just return.
508 if (td->td_inhibitors != TDI_SWAPPED)
514 printf("state is 0x%x", td->td_state);
515 panic("setrunnable(2)");
517 if ((td->td_flags & TDF_INMEM) == 0) {
518 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
519 td->td_flags |= TDF_SWAPINREQ;
528 * Compute a tenex style load average of a quantity on
529 * 1, 5 and 15 minute intervals.
540 for (i = 0; i < 3; i++)
541 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
542 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
545 * Schedule the next update to occur after 5 seconds, but add a
546 * random variation to avoid synchronisation with processes that
547 * run at regular intervals.
549 callout_reset_sbt(&loadav_callout,
550 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
551 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
556 synch_setup(void *dummy)
558 callout_init(&loadav_callout, 1);
560 /* Kick off timeout driven events by calling first time. */
568 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
576 kern_yield(PRI_USER);
587 if (prio == PRI_USER)
588 prio = td->td_user_pri;
590 sched_prio(td, prio);
591 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
597 * General purpose yield system call.
600 sys_yield(struct thread *td, struct yield_args *uap)
604 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
605 sched_prio(td, PRI_MAX_TIMESHARE);
606 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
608 td->td_retval[0] = 0;