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);
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 int catch, lock_state, pri, rval, sleepq_flags;
161 WITNESS_SAVE_DECL(lock_witness);
166 if (KTRPOINT(td, KTR_CSW))
169 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
170 "Sleeping on \"%s\"", wmesg);
171 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
172 ("sleeping without a lock"));
173 KASSERT(p != NULL, ("msleep1"));
174 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
175 if (priority & PDROP)
176 KASSERT(lock != NULL && lock != &Giant.lock_object,
177 ("PDROP requires a non-Giant lock"));
179 class = LOCK_CLASS(lock);
183 if (cold || SCHEDULER_STOPPED()) {
185 * During autoconfiguration, just return;
186 * don't run any other threads or panic below,
187 * in case this is the idle thread and already asleep.
188 * XXX: this used to do "s = splhigh(); splx(safepri);
189 * splx(s);" to give interrupts a chance, but there is
190 * no way to give interrupts a chance now.
192 if (lock != NULL && priority & PDROP)
193 class->lc_unlock(lock);
196 catch = priority & PCATCH;
197 pri = priority & PRIMASK;
200 * If we are already on a sleep queue, then remove us from that
201 * sleep queue first. We have to do this to handle recursive
204 if (TD_ON_SLEEPQ(td))
205 sleepq_remove(td, td->td_wchan);
207 if ((uint8_t *)ident >= &pause_wchan[0] &&
208 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
209 sleepq_flags = SLEEPQ_PAUSE;
211 sleepq_flags = SLEEPQ_SLEEP;
213 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
216 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
217 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
219 if (lock == &Giant.lock_object)
220 mtx_assert(&Giant, MA_OWNED);
222 if (lock != NULL && lock != &Giant.lock_object &&
223 !(class->lc_flags & LC_SLEEPABLE)) {
224 WITNESS_SAVE(lock, lock_witness);
225 lock_state = class->lc_unlock(lock);
227 /* GCC needs to follow the Yellow Brick Road */
231 * We put ourselves on the sleep queue and start our timeout
232 * before calling thread_suspend_check, as we could stop there,
233 * and a wakeup or a SIGCONT (or both) could occur while we were
234 * stopped without resuming us. Thus, we must be ready for sleep
235 * when cursig() is called. If the wakeup happens while we're
236 * stopped, then td will no longer be on a sleep queue upon
237 * return from cursig().
239 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
241 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
242 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
243 sleepq_release(ident);
244 WITNESS_SAVE(lock, lock_witness);
245 lock_state = class->lc_unlock(lock);
248 if (sbt != 0 && catch)
249 rval = sleepq_timedwait_sig(ident, pri);
251 rval = sleepq_timedwait(ident, pri);
253 rval = sleepq_wait_sig(ident, pri);
255 sleepq_wait(ident, pri);
259 if (KTRPOINT(td, KTR_CSW))
263 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
264 class->lc_lock(lock, lock_state);
265 WITNESS_RESTORE(lock, lock_witness);
271 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
272 sbintime_t sbt, sbintime_t pr, int flags)
277 WITNESS_SAVE_DECL(mtx);
281 KASSERT(mtx != NULL, ("sleeping without a mutex"));
282 KASSERT(p != NULL, ("msleep1"));
283 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
285 if (cold || SCHEDULER_STOPPED()) {
287 * During autoconfiguration, just return;
288 * don't run any other threads or panic below,
289 * in case this is the idle thread and already asleep.
290 * XXX: this used to do "s = splhigh(); splx(safepri);
291 * splx(s);" to give interrupts a chance, but there is
292 * no way to give interrupts a chance now.
298 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
299 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
302 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
303 WITNESS_SAVE(&mtx->lock_object, mtx);
304 mtx_unlock_spin(mtx);
307 * We put ourselves on the sleep queue and start our timeout.
309 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
311 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
314 * Can't call ktrace with any spin locks held so it can lock the
315 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
316 * any spin lock. Thus, we have to drop the sleepq spin lock while
317 * we handle those requests. This is safe since we have placed our
318 * thread on the sleep queue already.
321 if (KTRPOINT(td, KTR_CSW)) {
322 sleepq_release(ident);
328 sleepq_release(ident);
329 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
334 rval = sleepq_timedwait(ident, 0);
336 sleepq_wait(ident, 0);
340 if (KTRPOINT(td, KTR_CSW))
345 WITNESS_RESTORE(&mtx->lock_object, mtx);
350 * pause() delays the calling thread by the given number of system ticks.
351 * During cold bootup, pause() uses the DELAY() function instead of
352 * the tsleep() function to do the waiting. The "timo" argument must be
353 * greater than or equal to zero. A "timo" value of zero is equivalent
354 * to a "timo" value of one.
357 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
361 sbt_sec = sbintime_getsec(sbt);
362 KASSERT(sbt_sec >= 0, ("pause: timo must be >= 0"));
364 /* silently convert invalid timeouts */
370 * We delay one second at a time to avoid overflowing the
371 * system specific DELAY() function(s):
373 while (sbt_sec > 0) {
377 DELAY((sbt & 0xffffffff) / SBT_1US);
380 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
384 * Make all threads sleeping on the specified identifier runnable.
392 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
393 sleepq_release(ident);
394 if (wakeup_swapper) {
395 KASSERT(ident != &proc0,
396 ("wakeup and wakeup_swapper and proc0"));
402 * Make a thread sleeping on the specified identifier runnable.
403 * May wake more than one thread if a target thread is currently
407 wakeup_one(void *ident)
412 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
413 sleepq_release(ident);
421 thread_unlock(curthread);
424 panic("%s: did not reenter debugger", __func__);
428 * The machine independent parts of context switching.
431 mi_switch(int flags, struct thread *newtd)
433 uint64_t runtime, new_switchtime;
437 td = curthread; /* XXX */
438 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
439 p = td->td_proc; /* XXX */
440 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
442 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
443 mtx_assert(&Giant, MA_NOTOWNED);
445 KASSERT(td->td_critnest == 1 || panicstr,
446 ("mi_switch: switch in a critical section"));
447 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
448 ("mi_switch: switch must be voluntary or involuntary"));
449 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
452 * Don't perform context switches from the debugger.
456 if (SCHEDULER_STOPPED())
458 if (flags & SW_VOL) {
459 td->td_ru.ru_nvcsw++;
460 td->td_swvoltick = ticks;
462 td->td_ru.ru_nivcsw++;
464 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
467 * Compute the amount of time during which the current
468 * thread was running, and add that to its total so far.
470 new_switchtime = cpu_ticks();
471 runtime = new_switchtime - PCPU_GET(switchtime);
472 td->td_runtime += runtime;
473 td->td_incruntime += runtime;
474 PCPU_SET(switchtime, new_switchtime);
475 td->td_generation++; /* bump preempt-detect counter */
476 PCPU_INC(cnt.v_swtch);
477 PCPU_SET(switchticks, ticks);
478 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
479 td->td_tid, td->td_sched, p->p_pid, td->td_name);
480 #if (KTR_COMPILE & KTR_SCHED) != 0
481 if (TD_IS_IDLETHREAD(td))
482 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
483 "prio:%d", td->td_priority);
485 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
486 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
487 "lockname:\"%s\"", td->td_lockname);
489 SDT_PROBE0(sched, , , preempt);
493 sched_switch(td, newtd, flags);
494 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
495 "prio:%d", td->td_priority);
497 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
498 td->td_tid, td->td_sched, p->p_pid, td->td_name);
501 * If the last thread was exiting, finish cleaning it up.
503 if ((td = PCPU_GET(deadthread))) {
504 PCPU_SET(deadthread, NULL);
510 * Change thread state to be runnable, placing it on the run queue if
511 * it is in memory. If it is swapped out, return true so our caller
512 * will know to awaken the swapper.
515 setrunnable(struct thread *td)
518 THREAD_LOCK_ASSERT(td, MA_OWNED);
519 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
520 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
521 switch (td->td_state) {
527 * If we are only inhibited because we are swapped out
528 * then arange to swap in this process. Otherwise just return.
530 if (td->td_inhibitors != TDI_SWAPPED)
536 printf("state is 0x%x", td->td_state);
537 panic("setrunnable(2)");
539 if ((td->td_flags & TDF_INMEM) == 0) {
540 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
541 td->td_flags |= TDF_SWAPINREQ;
550 * Compute a tenex style load average of a quantity on
551 * 1, 5 and 15 minute intervals.
562 for (i = 0; i < 3; i++)
563 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
564 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
567 * Schedule the next update to occur after 5 seconds, but add a
568 * random variation to avoid synchronisation with processes that
569 * run at regular intervals.
571 callout_reset_sbt(&loadav_callout,
572 tick_sbt * (hz * 4 + (int)(random() % (hz * 2 + 1))), 0,
573 loadav, NULL, C_DIRECT_EXEC | C_HARDCLOCK);
578 synch_setup(void *dummy)
580 callout_init(&loadav_callout, CALLOUT_MPSAFE);
582 /* Kick off timeout driven events by calling first time. */
590 return ((unsigned int)(ticks - curthread->td_swvoltick) >= hogticks);
598 kern_yield(PRI_USER);
609 if (prio == PRI_USER)
610 prio = td->td_user_pri;
612 sched_prio(td, prio);
613 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
619 * General purpose yield system call.
622 sys_yield(struct thread *td, struct yield_args *uap)
626 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
627 sched_prio(td, PRI_MAX_TIMESHARE);
628 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
630 td->td_retval[0] = 0;