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);
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);
119 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
120 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
121 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
122 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
125 sleepinit(void *unused)
128 hogticks = (hz / 10) * 2; /* Default only. */
133 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
136 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
139 * General sleep call. Suspends the current thread until a wakeup is
140 * performed on the specified identifier. The thread will then be made
141 * runnable with the specified priority. Sleeps at most sbt units of time
142 * (0 means no timeout). If pri includes the PCATCH flag, let signals
143 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
144 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
145 * signal becomes pending, ERESTART is returned if the current system
146 * call should be restarted if possible, and EINTR is returned if the system
147 * call should be interrupted by the signal (return EINTR).
149 * The lock argument is unlocked before the caller is suspended, and
150 * re-locked before _sleep() returns. If priority includes the PDROP
151 * flag the lock is not re-locked before returning.
154 _sleep(void *ident, struct lock_object *lock, int priority,
155 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
159 struct lock_class *class;
160 uintptr_t lock_state;
161 int catch, pri, rval, sleepq_flags;
162 WITNESS_SAVE_DECL(lock_witness);
167 if (KTRPOINT(td, KTR_CSW))
170 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
171 "Sleeping on \"%s\"", wmesg);
172 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
173 ("sleeping without a lock"));
174 KASSERT(p != NULL, ("msleep1"));
175 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
176 if (priority & PDROP)
177 KASSERT(lock != NULL && lock != &Giant.lock_object,
178 ("PDROP requires a non-Giant lock"));
180 class = LOCK_CLASS(lock);
184 if (cold || SCHEDULER_STOPPED()) {
186 * During autoconfiguration, just return;
187 * don't run any other threads or panic below,
188 * in case this is the idle thread and already asleep.
189 * XXX: this used to do "s = splhigh(); splx(safepri);
190 * splx(s);" to give interrupts a chance, but there is
191 * no way to give interrupts a chance now.
193 if (lock != NULL && priority & PDROP)
194 class->lc_unlock(lock);
197 catch = priority & PCATCH;
198 pri = priority & PRIMASK;
201 * If we are already on a sleep queue, then remove us from that
202 * sleep queue first. We have to do this to handle recursive
205 if (TD_ON_SLEEPQ(td))
206 sleepq_remove(td, td->td_wchan);
208 if ((uint8_t *)ident >= &pause_wchan[0] &&
209 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
210 sleepq_flags = SLEEPQ_PAUSE;
212 sleepq_flags = SLEEPQ_SLEEP;
214 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
217 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
218 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
220 if (lock == &Giant.lock_object)
221 mtx_assert(&Giant, MA_OWNED);
223 if (lock != NULL && lock != &Giant.lock_object &&
224 !(class->lc_flags & LC_SLEEPABLE)) {
225 WITNESS_SAVE(lock, lock_witness);
226 lock_state = class->lc_unlock(lock);
228 /* GCC needs to follow the Yellow Brick Road */
232 * We put ourselves on the sleep queue and start our timeout
233 * before calling thread_suspend_check, as we could stop there,
234 * and a wakeup or a SIGCONT (or both) could occur while we were
235 * stopped without resuming us. Thus, we must be ready for sleep
236 * when cursig() is called. If the wakeup happens while we're
237 * stopped, then td will no longer be on a sleep queue upon
238 * return from cursig().
240 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
242 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
243 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
244 sleepq_release(ident);
245 WITNESS_SAVE(lock, lock_witness);
246 lock_state = class->lc_unlock(lock);
249 if (sbt != 0 && catch)
250 rval = sleepq_timedwait_sig(ident, pri);
252 rval = sleepq_timedwait(ident, pri);
254 rval = sleepq_wait_sig(ident, pri);
256 sleepq_wait(ident, pri);
260 if (KTRPOINT(td, KTR_CSW))
264 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
265 class->lc_lock(lock, lock_state);
266 WITNESS_RESTORE(lock, lock_witness);
272 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
273 sbintime_t sbt, sbintime_t pr, int flags)
278 WITNESS_SAVE_DECL(mtx);
282 KASSERT(mtx != NULL, ("sleeping without a mutex"));
283 KASSERT(p != NULL, ("msleep1"));
284 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
286 if (cold || SCHEDULER_STOPPED()) {
288 * During autoconfiguration, just return;
289 * don't run any other threads or panic below,
290 * in case this is the idle thread and already asleep.
291 * XXX: this used to do "s = splhigh(); splx(safepri);
292 * splx(s);" to give interrupts a chance, but there is
293 * no way to give interrupts a chance now.
299 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
300 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
303 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
304 WITNESS_SAVE(&mtx->lock_object, mtx);
305 mtx_unlock_spin(mtx);
308 * We put ourselves on the sleep queue and start our timeout.
310 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
312 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
315 * Can't call ktrace with any spin locks held so it can lock the
316 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
317 * any spin lock. Thus, we have to drop the sleepq spin lock while
318 * we handle those requests. This is safe since we have placed our
319 * thread on the sleep queue already.
322 if (KTRPOINT(td, KTR_CSW)) {
323 sleepq_release(ident);
329 sleepq_release(ident);
330 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
335 rval = sleepq_timedwait(ident, 0);
337 sleepq_wait(ident, 0);
341 if (KTRPOINT(td, KTR_CSW))
346 WITNESS_RESTORE(&mtx->lock_object, mtx);
351 * pause() delays the calling thread by the given number of system ticks.
352 * During cold bootup, pause() uses the DELAY() function instead of
353 * the tsleep() function to do the waiting. The "timo" argument must be
354 * greater than or equal to zero. A "timo" value of zero is equivalent
355 * to a "timo" value of one.
358 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
360 KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
362 /* silently convert invalid timeouts */
366 if (cold || kdb_active) {
368 * We delay one second at a time to avoid overflowing the
369 * system specific DELAY() function(s):
371 while (sbt >= SBT_1S) {
375 /* Do the delay remainder, if any */
376 sbt = (sbt + SBT_1US - 1) / SBT_1US;
381 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
385 * Make all threads sleeping on the specified identifier runnable.
393 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
394 sleepq_release(ident);
395 if (wakeup_swapper) {
396 KASSERT(ident != &proc0,
397 ("wakeup and wakeup_swapper and proc0"));
403 * Make a thread sleeping on the specified identifier runnable.
404 * May wake more than one thread if a target thread is currently
408 wakeup_one(void *ident)
413 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
414 sleepq_release(ident);
422 thread_unlock(curthread);
425 panic("%s: did not reenter debugger", __func__);
429 * The machine independent parts of context switching.
432 mi_switch(int flags, struct thread *newtd)
434 uint64_t runtime, new_switchtime;
438 td = curthread; /* XXX */
439 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
440 p = td->td_proc; /* XXX */
441 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
443 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
444 mtx_assert(&Giant, MA_NOTOWNED);
446 KASSERT(td->td_critnest == 1 || panicstr,
447 ("mi_switch: switch in a critical section"));
448 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
449 ("mi_switch: switch must be voluntary or involuntary"));
450 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
453 * Don't perform context switches from the debugger.
457 if (SCHEDULER_STOPPED())
459 if (flags & SW_VOL) {
460 td->td_ru.ru_nvcsw++;
461 td->td_swvoltick = ticks;
463 td->td_ru.ru_nivcsw++;
465 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
468 * Compute the amount of time during which the current
469 * thread was running, and add that to its total so far.
471 new_switchtime = cpu_ticks();
472 runtime = new_switchtime - PCPU_GET(switchtime);
473 td->td_runtime += runtime;
474 td->td_incruntime += runtime;
475 PCPU_SET(switchtime, new_switchtime);
476 td->td_generation++; /* bump preempt-detect counter */
477 PCPU_INC(cnt.v_swtch);
478 PCPU_SET(switchticks, ticks);
479 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
480 td->td_tid, td->td_sched, p->p_pid, td->td_name);
481 #if (KTR_COMPILE & KTR_SCHED) != 0
482 if (TD_IS_IDLETHREAD(td))
483 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
484 "prio:%d", td->td_priority);
486 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
487 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
488 "lockname:\"%s\"", td->td_lockname);
490 SDT_PROBE0(sched, , , preempt);
494 sched_switch(td, newtd, flags);
495 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
496 "prio:%d", td->td_priority);
498 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
499 td->td_tid, td->td_sched, p->p_pid, td->td_name);
502 * If the last thread was exiting, finish cleaning it up.
504 if ((td = PCPU_GET(deadthread))) {
505 PCPU_SET(deadthread, NULL);
511 * Change thread state to be runnable, placing it on the run queue if
512 * it is in memory. If it is swapped out, return true so our caller
513 * will know to awaken the swapper.
516 setrunnable(struct thread *td)
519 THREAD_LOCK_ASSERT(td, MA_OWNED);
520 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
521 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
522 switch (td->td_state) {
528 * If we are only inhibited because we are swapped out
529 * then arange to swap in this process. Otherwise just return.
531 if (td->td_inhibitors != TDI_SWAPPED)
537 printf("state is 0x%x", td->td_state);
538 panic("setrunnable(2)");
540 if ((td->td_flags & TDF_INMEM) == 0) {
541 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
542 td->td_flags |= TDF_SWAPINREQ;
551 * Compute a tenex style load average of a quantity on
552 * 1, 5 and 15 minute intervals.
563 for (i = 0; i < 3; i++)
564 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
565 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
568 * Schedule the next update to occur after 5 seconds, but add a
569 * random variation to avoid synchronisation with processes that
570 * run at regular intervals.
572 callout_reset_sbt(&loadav_callout,
573 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
574 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
579 synch_setup(void *dummy)
581 callout_init(&loadav_callout, CALLOUT_MPSAFE);
583 /* Kick off timeout driven events by calling first time. */
591 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
599 kern_yield(PRI_USER);
610 if (prio == PRI_USER)
611 prio = td->td_user_pri;
613 sched_prio(td, prio);
614 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
620 * General purpose yield system call.
623 sys_yield(struct thread *td, struct yield_args *uap)
627 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
628 sched_prio(td, PRI_MAX_TIMESHARE);
629 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
631 td->td_retval[0] = 0;