2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1990, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_ktrace.h"
43 #include "opt_sched.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/condvar.h>
49 #include <sys/kernel.h>
52 #include <sys/mutex.h>
54 #include <sys/resourcevar.h>
55 #include <sys/sched.h>
57 #include <sys/signalvar.h>
58 #include <sys/sleepqueue.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/vmmeter.h>
66 #include <sys/ktrace.h>
69 #include <machine/cpu.h>
71 static void synch_setup(void *dummy);
72 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
76 static uint8_t pause_wchan[MAXCPU];
78 static struct callout loadav_callout;
80 struct loadavg averunnable =
81 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
83 * Constants for averages over 1, 5, and 15 minutes
84 * when sampling at 5 second intervals.
86 static fixpt_t cexp[3] = {
87 0.9200444146293232 * FSCALE, /* exp(-1/12) */
88 0.9834714538216174 * FSCALE, /* exp(-1/60) */
89 0.9944598480048967 * FSCALE, /* exp(-1/180) */
92 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
93 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
95 static void loadav(void *arg);
97 SDT_PROVIDER_DECLARE(sched);
98 SDT_PROBE_DEFINE(sched, , , preempt);
101 sleepinit(void *unused)
104 hogticks = (hz / 10) * 2; /* Default only. */
109 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
112 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, NULL);
115 * General sleep call. Suspends the current thread until a wakeup is
116 * performed on the specified identifier. The thread will then be made
117 * runnable with the specified priority. Sleeps at most sbt units of time
118 * (0 means no timeout). If pri includes the PCATCH flag, let signals
119 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
120 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
121 * signal becomes pending, ERESTART is returned if the current system
122 * call should be restarted if possible, and EINTR is returned if the system
123 * call should be interrupted by the signal (return EINTR).
125 * The lock argument is unlocked before the caller is suspended, and
126 * re-locked before _sleep() returns. If priority includes the PDROP
127 * flag the lock is not re-locked before returning.
130 _sleep(void *ident, struct lock_object *lock, int priority,
131 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
134 struct lock_class *class;
135 uintptr_t lock_state;
136 int catch, pri, rval, sleepq_flags;
137 WITNESS_SAVE_DECL(lock_witness);
141 if (KTRPOINT(td, KTR_CSW))
144 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
145 "Sleeping on \"%s\"", wmesg);
146 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
147 ("sleeping without a lock"));
148 KASSERT(ident != NULL, ("_sleep: NULL ident"));
149 KASSERT(TD_IS_RUNNING(td), ("_sleep: curthread not running"));
150 KASSERT(td->td_epochnest == 0, ("sleeping in an epoch section"));
151 if (priority & PDROP)
152 KASSERT(lock != NULL && lock != &Giant.lock_object,
153 ("PDROP requires a non-Giant lock"));
155 class = LOCK_CLASS(lock);
159 if (SCHEDULER_STOPPED_TD(td)) {
160 if (lock != NULL && priority & PDROP)
161 class->lc_unlock(lock);
164 catch = priority & PCATCH;
165 pri = priority & PRIMASK;
167 KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep"));
169 if ((uint8_t *)ident >= &pause_wchan[0] &&
170 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
171 sleepq_flags = SLEEPQ_PAUSE;
173 sleepq_flags = SLEEPQ_SLEEP;
175 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
178 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
179 td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident);
181 if (lock == &Giant.lock_object)
182 mtx_assert(&Giant, MA_OWNED);
184 if (lock != NULL && lock != &Giant.lock_object &&
185 !(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, lock, wmesg, sleepq_flags, 0);
203 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
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);
210 if (sbt != 0 && catch)
211 rval = sleepq_timedwait_sig(ident, pri);
213 rval = sleepq_timedwait(ident, pri);
215 rval = sleepq_wait_sig(ident, pri);
217 sleepq_wait(ident, pri);
221 if (KTRPOINT(td, KTR_CSW))
225 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
226 class->lc_lock(lock, lock_state);
227 WITNESS_RESTORE(lock, lock_witness);
233 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
234 sbintime_t sbt, sbintime_t pr, int flags)
238 WITNESS_SAVE_DECL(mtx);
241 KASSERT(mtx != NULL, ("sleeping without a mutex"));
242 KASSERT(ident != NULL, ("msleep_spin_sbt: NULL ident"));
243 KASSERT(TD_IS_RUNNING(td), ("msleep_spin_sbt: curthread not running"));
245 if (SCHEDULER_STOPPED_TD(td))
249 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
250 td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident);
253 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
254 WITNESS_SAVE(&mtx->lock_object, mtx);
255 mtx_unlock_spin(mtx);
258 * We put ourselves on the sleep queue and start our timeout.
260 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
262 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
265 * Can't call ktrace with any spin locks held so it can lock the
266 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
267 * any spin lock. Thus, we have to drop the sleepq spin lock while
268 * we handle those requests. This is safe since we have placed our
269 * thread on the sleep queue already.
272 if (KTRPOINT(td, KTR_CSW)) {
273 sleepq_release(ident);
279 sleepq_release(ident);
280 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
285 rval = sleepq_timedwait(ident, 0);
287 sleepq_wait(ident, 0);
291 if (KTRPOINT(td, KTR_CSW))
296 WITNESS_RESTORE(&mtx->lock_object, mtx);
301 * pause_sbt() delays the calling thread by the given signed binary
302 * time. During cold bootup, pause_sbt() uses the DELAY() function
303 * instead of the _sleep() function to do the waiting. The "sbt"
304 * argument must be greater than or equal to zero. A "sbt" value of
305 * zero is equivalent to a "sbt" value of one tick.
308 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
310 KASSERT(sbt >= 0, ("pause_sbt: timeout must be >= 0"));
312 /* silently convert invalid timeouts */
316 if ((cold && curthread == &thread0) || kdb_active ||
317 SCHEDULER_STOPPED()) {
319 * We delay one second at a time to avoid overflowing the
320 * system specific DELAY() function(s):
322 while (sbt >= SBT_1S) {
326 /* Do the delay remainder, if any */
327 sbt = howmany(sbt, SBT_1US);
330 return (EWOULDBLOCK);
332 return (_sleep(&pause_wchan[curcpu], NULL,
333 (flags & C_CATCH) ? PCATCH : 0, wmesg, sbt, pr, flags));
337 * Make all threads sleeping on the specified identifier runnable.
345 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
346 sleepq_release(ident);
347 if (wakeup_swapper) {
348 KASSERT(ident != &proc0,
349 ("wakeup and wakeup_swapper and proc0"));
355 * Make a thread sleeping on the specified identifier runnable.
356 * May wake more than one thread if a target thread is currently
360 wakeup_one(void *ident)
365 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
366 sleepq_release(ident);
372 wakeup_any(void *ident)
377 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_UNFAIR,
379 sleepq_release(ident);
387 thread_unlock(curthread);
390 panic("%s: did not reenter debugger", __func__);
394 * The machine independent parts of context switching.
397 mi_switch(int flags, struct thread *newtd)
399 uint64_t runtime, new_switchtime;
402 td = curthread; /* XXX */
403 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
404 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
406 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
407 mtx_assert(&Giant, MA_NOTOWNED);
409 KASSERT(td->td_critnest == 1 || panicstr,
410 ("mi_switch: switch in a critical section"));
411 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
412 ("mi_switch: switch must be voluntary or involuntary"));
413 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
416 * Don't perform context switches from the debugger.
420 if (SCHEDULER_STOPPED_TD(td))
422 if (flags & SW_VOL) {
423 td->td_ru.ru_nvcsw++;
424 td->td_swvoltick = ticks;
426 td->td_ru.ru_nivcsw++;
427 td->td_swinvoltick = ticks;
430 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
433 * Compute the amount of time during which the current
434 * thread was running, and add that to its total so far.
436 new_switchtime = cpu_ticks();
437 runtime = new_switchtime - PCPU_GET(switchtime);
438 td->td_runtime += runtime;
439 td->td_incruntime += runtime;
440 PCPU_SET(switchtime, new_switchtime);
441 td->td_generation++; /* bump preempt-detect counter */
443 PCPU_SET(switchticks, ticks);
444 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
445 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
447 if (SDT_PROBES_ENABLED() &&
448 ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 &&
449 (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED)))
450 SDT_PROBE0(sched, , , preempt);
452 sched_switch(td, newtd, flags);
453 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
454 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
457 * If the last thread was exiting, finish cleaning it up.
459 if ((td = PCPU_GET(deadthread))) {
460 PCPU_SET(deadthread, NULL);
466 * Change thread state to be runnable, placing it on the run queue if
467 * it is in memory. If it is swapped out, return true so our caller
468 * will know to awaken the swapper.
471 setrunnable(struct thread *td)
474 THREAD_LOCK_ASSERT(td, MA_OWNED);
475 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
476 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
477 switch (td->td_state) {
483 * If we are only inhibited because we are swapped out
484 * then arange to swap in this process. Otherwise just return.
486 if (td->td_inhibitors != TDI_SWAPPED)
492 printf("state is 0x%x", td->td_state);
493 panic("setrunnable(2)");
495 if ((td->td_flags & TDF_INMEM) == 0) {
496 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
497 td->td_flags |= TDF_SWAPINREQ;
506 * Compute a tenex style load average of a quantity on
507 * 1, 5 and 15 minute intervals.
518 for (i = 0; i < 3; i++)
519 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
520 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
523 * Schedule the next update to occur after 5 seconds, but add a
524 * random variation to avoid synchronisation with processes that
525 * run at regular intervals.
527 callout_reset_sbt(&loadav_callout,
528 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
529 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
534 synch_setup(void *dummy)
536 callout_init(&loadav_callout, 1);
538 /* Kick off timeout driven events by calling first time. */
546 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
554 kern_yield(PRI_USER);
565 if (prio == PRI_USER)
566 prio = td->td_user_pri;
568 sched_prio(td, prio);
569 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
575 * General purpose yield system call.
578 sys_yield(struct thread *td, struct yield_args *uap)
582 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
583 sched_prio(td, PRI_MAX_TIMESHARE);
584 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
586 td->td_retval[0] = 0;