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 #define KTDSTATE(td) \
77 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \
78 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \
79 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \
80 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \
81 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding")
83 static void synch_setup(void *dummy);
84 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
88 static uint8_t pause_wchan[MAXCPU];
90 static struct callout loadav_callout;
92 struct loadavg averunnable =
93 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
95 * Constants for averages over 1, 5, and 15 minutes
96 * when sampling at 5 second intervals.
98 static fixpt_t cexp[3] = {
99 0.9200444146293232 * FSCALE, /* exp(-1/12) */
100 0.9834714538216174 * FSCALE, /* exp(-1/60) */
101 0.9944598480048967 * FSCALE, /* exp(-1/180) */
104 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
105 static int fscale __unused = FSCALE;
106 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
108 static void loadav(void *arg);
110 SDT_PROVIDER_DECLARE(sched);
111 SDT_PROBE_DEFINE(sched, , , preempt, preempt);
114 * These probes reference Solaris features that are not implemented in FreeBSD.
115 * Create the probes anyway for compatibility with existing D scripts; they'll
118 SDT_PROBE_DEFINE(sched, , , cpucaps_sleep, cpucaps-sleep);
119 SDT_PROBE_DEFINE(sched, , , cpucaps_wakeup, cpucaps-wakeup);
120 SDT_PROBE_DEFINE(sched, , , schedctl_nopreempt, schedctl-nopreempt);
121 SDT_PROBE_DEFINE(sched, , , schedctl_preempt, schedctl-preempt);
122 SDT_PROBE_DEFINE(sched, , , schedctl_yield, schedctl-yield);
128 hogticks = (hz / 10) * 2; /* Default only. */
133 * General sleep call. Suspends the current thread until a wakeup is
134 * performed on the specified identifier. The thread will then be made
135 * runnable with the specified priority. Sleeps at most timo/hz seconds
136 * (0 means no timeout). If pri includes the PCATCH flag, let signals
137 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
138 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
139 * signal becomes pending, ERESTART is returned if the current system
140 * call should be restarted if possible, and EINTR is returned if the system
141 * call should be interrupted by the signal (return EINTR).
143 * The lock argument is unlocked before the caller is suspended, and
144 * re-locked before _sleep() returns. If priority includes the PDROP
145 * flag the lock is not re-locked before returning.
148 _sleep(void *ident, struct lock_object *lock, int priority,
149 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
153 struct lock_class *class;
154 int catch, lock_state, pri, rval, sleepq_flags;
155 WITNESS_SAVE_DECL(lock_witness);
160 if (KTRPOINT(td, KTR_CSW))
163 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
164 "Sleeping on \"%s\"", wmesg);
165 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
166 ("sleeping without a lock"));
167 KASSERT(p != NULL, ("msleep1"));
168 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
169 if (priority & PDROP)
170 KASSERT(lock != NULL && lock != &Giant.lock_object,
171 ("PDROP requires a non-Giant lock"));
173 class = LOCK_CLASS(lock);
177 if (cold || SCHEDULER_STOPPED()) {
179 * During autoconfiguration, just return;
180 * don't run any other threads or panic below,
181 * in case this is the idle thread and already asleep.
182 * XXX: this used to do "s = splhigh(); splx(safepri);
183 * splx(s);" to give interrupts a chance, but there is
184 * no way to give interrupts a chance now.
186 if (lock != NULL && priority & PDROP)
187 class->lc_unlock(lock);
190 catch = priority & PCATCH;
191 pri = priority & PRIMASK;
194 * If we are already on a sleep queue, then remove us from that
195 * sleep queue first. We have to do this to handle recursive
198 if (TD_ON_SLEEPQ(td))
199 sleepq_remove(td, td->td_wchan);
201 if ((uint8_t *)ident >= &pause_wchan[0] &&
202 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
203 sleepq_flags = SLEEPQ_PAUSE;
205 sleepq_flags = SLEEPQ_SLEEP;
207 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
208 if (priority & PBDRY)
209 sleepq_flags |= SLEEPQ_STOP_ON_BDRY;
212 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
213 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
215 if (lock == &Giant.lock_object)
216 mtx_assert(&Giant, MA_OWNED);
218 if (lock != NULL && lock != &Giant.lock_object &&
219 !(class->lc_flags & LC_SLEEPABLE)) {
220 WITNESS_SAVE(lock, lock_witness);
221 lock_state = class->lc_unlock(lock);
223 /* GCC needs to follow the Yellow Brick Road */
227 * We put ourselves on the sleep queue and start our timeout
228 * before calling thread_suspend_check, as we could stop there,
229 * and a wakeup or a SIGCONT (or both) could occur while we were
230 * stopped without resuming us. Thus, we must be ready for sleep
231 * when cursig() is called. If the wakeup happens while we're
232 * stopped, then td will no longer be on a sleep queue upon
233 * return from cursig().
235 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
237 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
238 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
239 sleepq_release(ident);
240 WITNESS_SAVE(lock, lock_witness);
241 lock_state = class->lc_unlock(lock);
244 if (sbt != 0 && catch)
245 rval = sleepq_timedwait_sig(ident, pri);
247 rval = sleepq_timedwait(ident, pri);
249 rval = sleepq_wait_sig(ident, pri);
251 sleepq_wait(ident, pri);
255 if (KTRPOINT(td, KTR_CSW))
259 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
260 class->lc_lock(lock, lock_state);
261 WITNESS_RESTORE(lock, lock_witness);
267 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
268 sbintime_t sbt, sbintime_t pr, int flags)
273 WITNESS_SAVE_DECL(mtx);
277 KASSERT(mtx != NULL, ("sleeping without a mutex"));
278 KASSERT(p != NULL, ("msleep1"));
279 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
281 if (cold || SCHEDULER_STOPPED()) {
283 * During autoconfiguration, just return;
284 * don't run any other threads or panic below,
285 * in case this is the idle thread and already asleep.
286 * XXX: this used to do "s = splhigh(); splx(safepri);
287 * splx(s);" to give interrupts a chance, but there is
288 * no way to give interrupts a chance now.
294 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
295 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
298 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
299 WITNESS_SAVE(&mtx->lock_object, mtx);
300 mtx_unlock_spin(mtx);
303 * We put ourselves on the sleep queue and start our timeout.
305 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
307 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
310 * Can't call ktrace with any spin locks held so it can lock the
311 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
312 * any spin lock. Thus, we have to drop the sleepq spin lock while
313 * we handle those requests. This is safe since we have placed our
314 * thread on the sleep queue already.
317 if (KTRPOINT(td, KTR_CSW)) {
318 sleepq_release(ident);
324 sleepq_release(ident);
325 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
330 rval = sleepq_timedwait(ident, 0);
332 sleepq_wait(ident, 0);
336 if (KTRPOINT(td, KTR_CSW))
341 WITNESS_RESTORE(&mtx->lock_object, mtx);
346 * pause() delays the calling thread by the given number of system ticks.
347 * During cold bootup, pause() uses the DELAY() function instead of
348 * the tsleep() function to do the waiting. The "timo" argument must be
349 * greater than or equal to zero. A "timo" value of zero is equivalent
350 * to a "timo" value of one.
353 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
357 sbt_sec = sbintime_getsec(sbt);
358 KASSERT(sbt_sec >= 0, ("pause: timo must be >= 0"));
360 /* silently convert invalid timeouts */
366 * We delay one second at a time to avoid overflowing the
367 * system specific DELAY() function(s):
369 while (sbt_sec > 0) {
373 DELAY((sbt & 0xffffffff) / SBT_1US);
376 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
380 * Make all threads sleeping on the specified identifier runnable.
388 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
389 sleepq_release(ident);
390 if (wakeup_swapper) {
391 KASSERT(ident != &proc0,
392 ("wakeup and wakeup_swapper and proc0"));
398 * Make a thread sleeping on the specified identifier runnable.
399 * May wake more than one thread if a target thread is currently
403 wakeup_one(void *ident)
408 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
409 sleepq_release(ident);
417 thread_unlock(curthread);
420 panic("%s: did not reenter debugger", __func__);
424 * The machine independent parts of context switching.
427 mi_switch(int flags, struct thread *newtd)
429 uint64_t runtime, new_switchtime;
433 td = curthread; /* XXX */
434 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
435 p = td->td_proc; /* XXX */
436 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
438 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
439 mtx_assert(&Giant, MA_NOTOWNED);
441 KASSERT(td->td_critnest == 1 || panicstr,
442 ("mi_switch: switch in a critical section"));
443 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
444 ("mi_switch: switch must be voluntary or involuntary"));
445 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
448 * Don't perform context switches from the debugger.
452 if (SCHEDULER_STOPPED())
454 if (flags & SW_VOL) {
455 td->td_ru.ru_nvcsw++;
456 td->td_swvoltick = ticks;
458 td->td_ru.ru_nivcsw++;
460 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
463 * Compute the amount of time during which the current
464 * thread was running, and add that to its total so far.
466 new_switchtime = cpu_ticks();
467 runtime = new_switchtime - PCPU_GET(switchtime);
468 td->td_runtime += runtime;
469 td->td_incruntime += runtime;
470 PCPU_SET(switchtime, new_switchtime);
471 td->td_generation++; /* bump preempt-detect counter */
472 PCPU_INC(cnt.v_swtch);
473 PCPU_SET(switchticks, ticks);
474 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
475 td->td_tid, td->td_sched, p->p_pid, td->td_name);
476 #if (KTR_COMPILE & KTR_SCHED) != 0
477 if (TD_IS_IDLETHREAD(td))
478 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
479 "prio:%d", td->td_priority);
481 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
482 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
483 "lockname:\"%s\"", td->td_lockname);
485 SDT_PROBE0(sched, , , preempt);
489 sched_switch(td, newtd, flags);
490 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
491 "prio:%d", td->td_priority);
493 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
494 td->td_tid, td->td_sched, p->p_pid, td->td_name);
497 * If the last thread was exiting, finish cleaning it up.
499 if ((td = PCPU_GET(deadthread))) {
500 PCPU_SET(deadthread, NULL);
506 * Change thread state to be runnable, placing it on the run queue if
507 * it is in memory. If it is swapped out, return true so our caller
508 * will know to awaken the swapper.
511 setrunnable(struct thread *td)
514 THREAD_LOCK_ASSERT(td, MA_OWNED);
515 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
516 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
517 switch (td->td_state) {
523 * If we are only inhibited because we are swapped out
524 * then arange to swap in this process. Otherwise just return.
526 if (td->td_inhibitors != TDI_SWAPPED)
532 printf("state is 0x%x", td->td_state);
533 panic("setrunnable(2)");
535 if ((td->td_flags & TDF_INMEM) == 0) {
536 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
537 td->td_flags |= TDF_SWAPINREQ;
546 * Compute a tenex style load average of a quantity on
547 * 1, 5 and 15 minute intervals.
558 for (i = 0; i < 3; i++)
559 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
560 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
563 * Schedule the next update to occur after 5 seconds, but add a
564 * random variation to avoid synchronisation with processes that
565 * run at regular intervals.
567 callout_reset_sbt(&loadav_callout,
568 tick_sbt * (hz * 4 + (int)(random() % (hz * 2 + 1))), 0,
569 loadav, NULL, C_DIRECT_EXEC | C_HARDCLOCK);
574 synch_setup(void *dummy)
576 callout_init(&loadav_callout, CALLOUT_MPSAFE);
578 /* Kick off timeout driven events by calling first time. */
586 return (ticks - curthread->td_swvoltick >= hogticks);
594 kern_yield(PRI_USER);
605 if (prio == PRI_USER)
606 prio = td->td_user_pri;
608 sched_prio(td, prio);
609 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
615 * General purpose yield system call.
618 sys_yield(struct thread *td, struct yield_args *uap)
622 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
623 sched_prio(td, PRI_MAX_TIMESHARE);
624 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
626 td->td_retval[0] = 0;