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>
71 #include <vm/vm_param.h>
75 #define KTDSTATE(td) \
76 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \
77 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \
78 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \
79 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \
80 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding")
82 static void synch_setup(void *dummy);
83 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
87 static uint8_t pause_wchan[MAXCPU];
89 static struct callout loadav_callout;
91 struct loadavg averunnable =
92 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
94 * Constants for averages over 1, 5, and 15 minutes
95 * when sampling at 5 second intervals.
97 static fixpt_t cexp[3] = {
98 0.9200444146293232 * FSCALE, /* exp(-1/12) */
99 0.9834714538216174 * FSCALE, /* exp(-1/60) */
100 0.9944598480048967 * FSCALE, /* exp(-1/180) */
103 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
104 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
106 static void loadav(void *arg);
108 SDT_PROVIDER_DECLARE(sched);
109 SDT_PROBE_DEFINE(sched, , , preempt);
112 * These probes reference Solaris features that are not implemented in FreeBSD.
113 * Create the probes anyway for compatibility with existing D scripts; they'll
116 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep);
117 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
118 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
119 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
120 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
123 sleepinit(void *unused)
126 hogticks = (hz / 10) * 2; /* Default only. */
131 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
134 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
137 * General sleep call. Suspends the current thread until a wakeup is
138 * performed on the specified identifier. The thread will then be made
139 * runnable with the specified priority. Sleeps at most sbt units of time
140 * (0 means no timeout). If pri includes the PCATCH flag, let signals
141 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
142 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
143 * signal becomes pending, ERESTART is returned if the current system
144 * call should be restarted if possible, and EINTR is returned if the system
145 * call should be interrupted by the signal (return EINTR).
147 * The lock argument is unlocked before the caller is suspended, and
148 * re-locked before _sleep() returns. If priority includes the PDROP
149 * flag the lock is not re-locked before returning.
152 _sleep(void *ident, struct lock_object *lock, int priority,
153 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
157 struct lock_class *class;
158 uintptr_t lock_state;
159 int catch, pri, rval, sleepq_flags;
160 WITNESS_SAVE_DECL(lock_witness);
165 if (KTRPOINT(td, KTR_CSW))
168 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
169 "Sleeping on \"%s\"", wmesg);
170 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
171 ("sleeping without a lock"));
172 KASSERT(p != NULL, ("msleep1"));
173 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
174 if (priority & PDROP)
175 KASSERT(lock != NULL && lock != &Giant.lock_object,
176 ("PDROP requires a non-Giant lock"));
178 class = LOCK_CLASS(lock);
182 if (cold || SCHEDULER_STOPPED()) {
184 * During autoconfiguration, just return;
185 * don't run any other threads or panic below,
186 * in case this is the idle thread and already asleep.
187 * XXX: this used to do "s = splhigh(); splx(safepri);
188 * splx(s);" to give interrupts a chance, but there is
189 * no way to give interrupts a chance now.
191 if (lock != NULL && priority & PDROP)
192 class->lc_unlock(lock);
195 catch = priority & PCATCH;
196 pri = priority & PRIMASK;
199 * If we are already on a sleep queue, then remove us from that
200 * sleep queue first. We have to do this to handle recursive
203 if (TD_ON_SLEEPQ(td))
204 sleepq_remove(td, td->td_wchan);
206 if ((uint8_t *)ident >= &pause_wchan[0] &&
207 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
208 sleepq_flags = SLEEPQ_PAUSE;
210 sleepq_flags = SLEEPQ_SLEEP;
212 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
215 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
216 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
218 if (lock == &Giant.lock_object)
219 mtx_assert(&Giant, MA_OWNED);
221 if (lock != NULL && lock != &Giant.lock_object &&
222 !(class->lc_flags & LC_SLEEPABLE)) {
223 WITNESS_SAVE(lock, lock_witness);
224 lock_state = class->lc_unlock(lock);
226 /* GCC needs to follow the Yellow Brick Road */
230 * We put ourselves on the sleep queue and start our timeout
231 * before calling thread_suspend_check, as we could stop there,
232 * and a wakeup or a SIGCONT (or both) could occur while we were
233 * stopped without resuming us. Thus, we must be ready for sleep
234 * when cursig() is called. If the wakeup happens while we're
235 * stopped, then td will no longer be on a sleep queue upon
236 * return from cursig().
238 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
240 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
241 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
242 sleepq_release(ident);
243 WITNESS_SAVE(lock, lock_witness);
244 lock_state = class->lc_unlock(lock);
247 if (sbt != 0 && catch)
248 rval = sleepq_timedwait_sig(ident, pri);
250 rval = sleepq_timedwait(ident, pri);
252 rval = sleepq_wait_sig(ident, pri);
254 sleepq_wait(ident, pri);
258 if (KTRPOINT(td, KTR_CSW))
262 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
263 class->lc_lock(lock, lock_state);
264 WITNESS_RESTORE(lock, lock_witness);
270 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
271 sbintime_t sbt, sbintime_t pr, int flags)
276 WITNESS_SAVE_DECL(mtx);
280 KASSERT(mtx != NULL, ("sleeping without a mutex"));
281 KASSERT(p != NULL, ("msleep1"));
282 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
284 if (cold || SCHEDULER_STOPPED()) {
286 * During autoconfiguration, just return;
287 * don't run any other threads or panic below,
288 * in case this is the idle thread and already asleep.
289 * XXX: this used to do "s = splhigh(); splx(safepri);
290 * splx(s);" to give interrupts a chance, but there is
291 * no way to give interrupts a chance now.
297 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
298 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
301 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
302 WITNESS_SAVE(&mtx->lock_object, mtx);
303 mtx_unlock_spin(mtx);
306 * We put ourselves on the sleep queue and start our timeout.
308 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
310 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
313 * Can't call ktrace with any spin locks held so it can lock the
314 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
315 * any spin lock. Thus, we have to drop the sleepq spin lock while
316 * we handle those requests. This is safe since we have placed our
317 * thread on the sleep queue already.
320 if (KTRPOINT(td, KTR_CSW)) {
321 sleepq_release(ident);
327 sleepq_release(ident);
328 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
333 rval = sleepq_timedwait(ident, 0);
335 sleepq_wait(ident, 0);
339 if (KTRPOINT(td, KTR_CSW))
344 WITNESS_RESTORE(&mtx->lock_object, mtx);
349 * pause() delays the calling thread by the given number of system ticks.
350 * During cold bootup, pause() uses the DELAY() function instead of
351 * the tsleep() function to do the waiting. The "timo" argument must be
352 * greater than or equal to zero. A "timo" value of zero is equivalent
353 * to a "timo" value of one.
356 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
358 KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
360 /* silently convert invalid timeouts */
364 if (cold || kdb_active) {
366 * We delay one second at a time to avoid overflowing the
367 * system specific DELAY() function(s):
369 while (sbt >= SBT_1S) {
373 /* Do the delay remainder, if any */
374 sbt = (sbt + SBT_1US - 1) / SBT_1US;
379 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
383 * Make all threads sleeping on the specified identifier runnable.
391 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
392 sleepq_release(ident);
393 if (wakeup_swapper) {
394 KASSERT(ident != &proc0,
395 ("wakeup and wakeup_swapper and proc0"));
401 * Make a thread sleeping on the specified identifier runnable.
402 * May wake more than one thread if a target thread is currently
406 wakeup_one(void *ident)
411 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
412 sleepq_release(ident);
420 thread_unlock(curthread);
423 panic("%s: did not reenter debugger", __func__);
427 * The machine independent parts of context switching.
430 mi_switch(int flags, struct thread *newtd)
432 uint64_t runtime, new_switchtime;
436 td = curthread; /* XXX */
437 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
438 p = td->td_proc; /* XXX */
439 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
441 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
442 mtx_assert(&Giant, MA_NOTOWNED);
444 KASSERT(td->td_critnest == 1 || panicstr,
445 ("mi_switch: switch in a critical section"));
446 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
447 ("mi_switch: switch must be voluntary or involuntary"));
448 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
451 * Don't perform context switches from the debugger.
455 if (SCHEDULER_STOPPED())
457 if (flags & SW_VOL) {
458 td->td_ru.ru_nvcsw++;
459 td->td_swvoltick = ticks;
461 td->td_ru.ru_nivcsw++;
463 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
466 * Compute the amount of time during which the current
467 * thread was running, and add that to its total so far.
469 new_switchtime = cpu_ticks();
470 runtime = new_switchtime - PCPU_GET(switchtime);
471 td->td_runtime += runtime;
472 td->td_incruntime += runtime;
473 PCPU_SET(switchtime, new_switchtime);
474 td->td_generation++; /* bump preempt-detect counter */
475 PCPU_INC(cnt.v_swtch);
476 PCPU_SET(switchticks, ticks);
477 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
478 td->td_tid, td->td_sched, p->p_pid, td->td_name);
479 #if (KTR_COMPILE & KTR_SCHED) != 0
480 if (TD_IS_IDLETHREAD(td))
481 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
482 "prio:%d", td->td_priority);
484 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
485 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
486 "lockname:\"%s\"", td->td_lockname);
488 SDT_PROBE0(sched, , , preempt);
492 sched_switch(td, newtd, flags);
493 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
494 "prio:%d", td->td_priority);
496 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
497 td->td_tid, td->td_sched, p->p_pid, td->td_name);
500 * If the last thread was exiting, finish cleaning it up.
502 if ((td = PCPU_GET(deadthread))) {
503 PCPU_SET(deadthread, NULL);
509 * Change thread state to be runnable, placing it on the run queue if
510 * it is in memory. If it is swapped out, return true so our caller
511 * will know to awaken the swapper.
514 setrunnable(struct thread *td)
517 THREAD_LOCK_ASSERT(td, MA_OWNED);
518 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
519 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
520 switch (td->td_state) {
526 * If we are only inhibited because we are swapped out
527 * then arange to swap in this process. Otherwise just return.
529 if (td->td_inhibitors != TDI_SWAPPED)
535 printf("state is 0x%x", td->td_state);
536 panic("setrunnable(2)");
538 if ((td->td_flags & TDF_INMEM) == 0) {
539 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
540 td->td_flags |= TDF_SWAPINREQ;
549 * Compute a tenex style load average of a quantity on
550 * 1, 5 and 15 minute intervals.
561 for (i = 0; i < 3; i++)
562 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
563 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
566 * Schedule the next update to occur after 5 seconds, but add a
567 * random variation to avoid synchronisation with processes that
568 * run at regular intervals.
570 callout_reset_sbt(&loadav_callout,
571 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
572 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
577 synch_setup(void *dummy)
579 callout_init(&loadav_callout, CALLOUT_MPSAFE);
581 /* Kick off timeout driven events by calling first time. */
589 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
597 kern_yield(PRI_USER);
608 if (prio == PRI_USER)
609 prio = td->td_user_pri;
611 sched_prio(td, prio);
612 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
618 * General purpose yield system call.
621 sys_yield(struct thread *td, struct yield_args *uap)
625 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
626 sched_prio(td, PRI_MAX_TIMESHARE);
627 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
629 td->td_retval[0] = 0;