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"
41 #include "opt_sched.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/condvar.h>
47 #include <sys/kernel.h>
50 #include <sys/mutex.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
55 #include <sys/signalvar.h>
56 #include <sys/sleepqueue.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysproto.h>
61 #include <sys/vmmeter.h>
64 #include <sys/ktrace.h>
67 #include <machine/cpu.h>
69 #define KTDSTATE(td) \
70 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \
71 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \
72 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \
73 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \
74 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding")
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 sleepinit(void *unused)
109 hogticks = (hz / 10) * 2; /* Default only. */
114 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
117 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
120 * General sleep call. Suspends the current thread until a wakeup is
121 * performed on the specified identifier. The thread will then be made
122 * runnable with the specified priority. Sleeps at most sbt units of time
123 * (0 means no timeout). If pri includes the PCATCH flag, let signals
124 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
125 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
126 * signal becomes pending, ERESTART is returned if the current system
127 * call should be restarted if possible, and EINTR is returned if the system
128 * call should be interrupted by the signal (return EINTR).
130 * The lock argument is unlocked before the caller is suspended, and
131 * re-locked before _sleep() returns. If priority includes the PDROP
132 * flag the lock is not re-locked before returning.
135 _sleep(void *ident, struct lock_object *lock, int priority,
136 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
140 struct lock_class *class;
141 uintptr_t lock_state;
142 int catch, pri, rval, sleepq_flags;
143 WITNESS_SAVE_DECL(lock_witness);
148 if (KTRPOINT(td, KTR_CSW))
151 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
152 "Sleeping on \"%s\"", wmesg);
153 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
154 ("sleeping without a lock"));
155 KASSERT(p != NULL, ("msleep1"));
156 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
157 if (priority & PDROP)
158 KASSERT(lock != NULL && lock != &Giant.lock_object,
159 ("PDROP requires a non-Giant lock"));
161 class = LOCK_CLASS(lock);
165 if (SCHEDULER_STOPPED()) {
166 if (lock != NULL && priority & PDROP)
167 class->lc_unlock(lock);
170 catch = priority & PCATCH;
171 pri = priority & PRIMASK;
174 * If we are already on a sleep queue, then remove us from that
175 * sleep queue first. We have to do this to handle recursive
178 if (TD_ON_SLEEPQ(td))
179 sleepq_remove(td, td->td_wchan);
181 if ((uint8_t *)ident >= &pause_wchan[0] &&
182 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
183 sleepq_flags = SLEEPQ_PAUSE;
185 sleepq_flags = SLEEPQ_SLEEP;
187 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
190 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
191 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
193 if (lock == &Giant.lock_object)
194 mtx_assert(&Giant, MA_OWNED);
196 if (lock != NULL && lock != &Giant.lock_object &&
197 !(class->lc_flags & LC_SLEEPABLE)) {
198 WITNESS_SAVE(lock, lock_witness);
199 lock_state = class->lc_unlock(lock);
201 /* GCC needs to follow the Yellow Brick Road */
205 * We put ourselves on the sleep queue and start our timeout
206 * before calling thread_suspend_check, as we could stop there,
207 * and a wakeup or a SIGCONT (or both) could occur while we were
208 * stopped without resuming us. Thus, we must be ready for sleep
209 * when cursig() is called. If the wakeup happens while we're
210 * stopped, then td will no longer be on a sleep queue upon
211 * return from cursig().
213 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
215 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
216 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
217 sleepq_release(ident);
218 WITNESS_SAVE(lock, lock_witness);
219 lock_state = class->lc_unlock(lock);
222 if (sbt != 0 && catch)
223 rval = sleepq_timedwait_sig(ident, pri);
225 rval = sleepq_timedwait(ident, pri);
227 rval = sleepq_wait_sig(ident, pri);
229 sleepq_wait(ident, pri);
233 if (KTRPOINT(td, KTR_CSW))
237 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
238 class->lc_lock(lock, lock_state);
239 WITNESS_RESTORE(lock, lock_witness);
245 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
246 sbintime_t sbt, sbintime_t pr, int flags)
251 WITNESS_SAVE_DECL(mtx);
255 KASSERT(mtx != NULL, ("sleeping without a mutex"));
256 KASSERT(p != NULL, ("msleep1"));
257 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
259 if (SCHEDULER_STOPPED())
263 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
264 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
267 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
268 WITNESS_SAVE(&mtx->lock_object, mtx);
269 mtx_unlock_spin(mtx);
272 * We put ourselves on the sleep queue and start our timeout.
274 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
276 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
279 * Can't call ktrace with any spin locks held so it can lock the
280 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
281 * any spin lock. Thus, we have to drop the sleepq spin lock while
282 * we handle those requests. This is safe since we have placed our
283 * thread on the sleep queue already.
286 if (KTRPOINT(td, KTR_CSW)) {
287 sleepq_release(ident);
293 sleepq_release(ident);
294 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
299 rval = sleepq_timedwait(ident, 0);
301 sleepq_wait(ident, 0);
305 if (KTRPOINT(td, KTR_CSW))
310 WITNESS_RESTORE(&mtx->lock_object, mtx);
315 * pause() delays the calling thread by the given number of system ticks.
316 * During cold bootup, pause() uses the DELAY() function instead of
317 * the tsleep() function to do the waiting. The "timo" argument must be
318 * greater than or equal to zero. A "timo" value of zero is equivalent
319 * to a "timo" value of one.
322 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
324 KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
326 /* silently convert invalid timeouts */
330 if (cold || kdb_active || SCHEDULER_STOPPED()) {
332 * We delay one second at a time to avoid overflowing the
333 * system specific DELAY() function(s):
335 while (sbt >= SBT_1S) {
339 /* Do the delay remainder, if any */
340 sbt = howmany(sbt, SBT_1US);
345 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
349 * Make all threads sleeping on the specified identifier runnable.
357 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
358 sleepq_release(ident);
359 if (wakeup_swapper) {
360 KASSERT(ident != &proc0,
361 ("wakeup and wakeup_swapper and proc0"));
367 * Make a thread sleeping on the specified identifier runnable.
368 * May wake more than one thread if a target thread is currently
372 wakeup_one(void *ident)
377 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
378 sleepq_release(ident);
386 thread_unlock(curthread);
389 panic("%s: did not reenter debugger", __func__);
393 * The machine independent parts of context switching.
396 mi_switch(int flags, struct thread *newtd)
398 uint64_t runtime, new_switchtime;
401 td = curthread; /* XXX */
402 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
403 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
405 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
406 mtx_assert(&Giant, MA_NOTOWNED);
408 KASSERT(td->td_critnest == 1 || panicstr,
409 ("mi_switch: switch in a critical section"));
410 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
411 ("mi_switch: switch must be voluntary or involuntary"));
412 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
415 * Don't perform context switches from the debugger.
419 if (SCHEDULER_STOPPED())
421 if (flags & SW_VOL) {
422 td->td_ru.ru_nvcsw++;
423 td->td_swvoltick = ticks;
425 td->td_ru.ru_nivcsw++;
426 td->td_swinvoltick = ticks;
429 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
432 * Compute the amount of time during which the current
433 * thread was running, and add that to its total so far.
435 new_switchtime = cpu_ticks();
436 runtime = new_switchtime - PCPU_GET(switchtime);
437 td->td_runtime += runtime;
438 td->td_incruntime += runtime;
439 PCPU_SET(switchtime, new_switchtime);
440 td->td_generation++; /* bump preempt-detect counter */
441 PCPU_INC(cnt.v_swtch);
442 PCPU_SET(switchticks, ticks);
443 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
444 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
445 #if (KTR_COMPILE & KTR_SCHED) != 0
446 if (TD_IS_IDLETHREAD(td))
447 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
448 "prio:%d", td->td_priority);
450 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
451 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
452 "lockname:\"%s\"", td->td_lockname);
454 SDT_PROBE0(sched, , , preempt);
455 sched_switch(td, newtd, flags);
456 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
457 "prio:%d", td->td_priority);
459 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
460 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
463 * If the last thread was exiting, finish cleaning it up.
465 if ((td = PCPU_GET(deadthread))) {
466 PCPU_SET(deadthread, NULL);
472 * Change thread state to be runnable, placing it on the run queue if
473 * it is in memory. If it is swapped out, return true so our caller
474 * will know to awaken the swapper.
477 setrunnable(struct thread *td)
480 THREAD_LOCK_ASSERT(td, MA_OWNED);
481 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
482 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
483 switch (td->td_state) {
489 * If we are only inhibited because we are swapped out
490 * then arange to swap in this process. Otherwise just return.
492 if (td->td_inhibitors != TDI_SWAPPED)
498 printf("state is 0x%x", td->td_state);
499 panic("setrunnable(2)");
501 if ((td->td_flags & TDF_INMEM) == 0) {
502 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
503 td->td_flags |= TDF_SWAPINREQ;
512 * Compute a tenex style load average of a quantity on
513 * 1, 5 and 15 minute intervals.
524 for (i = 0; i < 3; i++)
525 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
526 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
529 * Schedule the next update to occur after 5 seconds, but add a
530 * random variation to avoid synchronisation with processes that
531 * run at regular intervals.
533 callout_reset_sbt(&loadav_callout,
534 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
535 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
540 synch_setup(void *dummy)
542 callout_init(&loadav_callout, 1);
544 /* Kick off timeout driven events by calling first time. */
552 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
560 kern_yield(PRI_USER);
571 if (prio == PRI_USER)
572 prio = td->td_user_pri;
574 sched_prio(td, prio);
575 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
581 * General purpose yield system call.
584 sys_yield(struct thread *td, struct yield_args *uap)
588 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
589 sched_prio(td, PRI_MAX_TIMESHARE);
590 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
592 td->td_retval[0] = 0;