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 int pause_wchan;
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 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
107 static void loadav(void *arg);
109 SDT_PROVIDER_DECLARE(sched);
110 SDT_PROBE_DEFINE(sched, , , preempt);
113 * These probes reference Solaris features that are not implemented in FreeBSD.
114 * Create the probes anyway for compatibility with existing D scripts; they'll
117 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep);
118 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
119 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
120 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
121 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
127 hogticks = (hz / 10) * 2; /* Default only. */
132 * General sleep call. Suspends the current thread until a wakeup is
133 * performed on the specified identifier. The thread will then be made
134 * runnable with the specified priority. Sleeps at most timo/hz seconds
135 * (0 means no timeout). If pri includes PCATCH flag, signals are checked
136 * before and after sleeping, else signals are not checked. Returns 0 if
137 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
138 * signal needs to be delivered, ERESTART is returned if the current system
139 * call should be restarted if possible, and EINTR is returned if the system
140 * call should be interrupted by the signal (return EINTR).
142 * The lock argument is unlocked before the caller is suspended, and
143 * re-locked before _sleep() returns. If priority includes the PDROP
144 * flag the lock is not re-locked before returning.
147 _sleep(void *ident, struct lock_object *lock, int priority,
148 const char *wmesg, int timo)
152 struct lock_class *class;
153 int catch, flags, lock_state, pri, rval;
154 WITNESS_SAVE_DECL(lock_witness);
159 if (KTRPOINT(td, KTR_CSW))
162 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
163 "Sleeping on \"%s\"", wmesg);
164 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL,
165 ("sleeping without a lock"));
166 KASSERT(p != NULL, ("msleep1"));
167 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
168 if (priority & PDROP)
169 KASSERT(lock != NULL && lock != &Giant.lock_object,
170 ("PDROP requires a non-Giant lock"));
172 class = LOCK_CLASS(lock);
176 if (cold || SCHEDULER_STOPPED()) {
178 * During autoconfiguration, just return;
179 * don't run any other threads or panic below,
180 * in case this is the idle thread and already asleep.
181 * XXX: this used to do "s = splhigh(); splx(safepri);
182 * splx(s);" to give interrupts a chance, but there is
183 * no way to give interrupts a chance now.
185 if (lock != NULL && priority & PDROP)
186 class->lc_unlock(lock);
189 catch = priority & PCATCH;
190 pri = priority & PRIMASK;
193 * If we are already on a sleep queue, then remove us from that
194 * sleep queue first. We have to do this to handle recursive
197 if (TD_ON_SLEEPQ(td))
198 sleepq_remove(td, td->td_wchan);
200 if (ident == &pause_wchan)
201 flags = SLEEPQ_PAUSE;
203 flags = SLEEPQ_SLEEP;
205 flags |= SLEEPQ_INTERRUPTIBLE;
206 if (priority & PBDRY)
207 flags |= SLEEPQ_STOP_ON_BDRY;
210 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
211 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
213 if (lock == &Giant.lock_object)
214 mtx_assert(&Giant, MA_OWNED);
216 if (lock != NULL && lock != &Giant.lock_object &&
217 !(class->lc_flags & LC_SLEEPABLE)) {
218 WITNESS_SAVE(lock, lock_witness);
219 lock_state = class->lc_unlock(lock);
221 /* GCC needs to follow the Yellow Brick Road */
225 * We put ourselves on the sleep queue and start our timeout
226 * before calling thread_suspend_check, as we could stop there,
227 * and a wakeup or a SIGCONT (or both) could occur while we were
228 * stopped without resuming us. Thus, we must be ready for sleep
229 * when cursig() is called. If the wakeup happens while we're
230 * stopped, then td will no longer be on a sleep queue upon
231 * return from cursig().
233 sleepq_add(ident, lock, wmesg, flags, 0);
235 sleepq_set_timeout(ident, timo);
236 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
237 sleepq_release(ident);
238 WITNESS_SAVE(lock, lock_witness);
239 lock_state = class->lc_unlock(lock);
243 rval = sleepq_timedwait_sig(ident, pri);
245 rval = sleepq_timedwait(ident, pri);
247 rval = sleepq_wait_sig(ident, pri);
249 sleepq_wait(ident, pri);
253 if (KTRPOINT(td, KTR_CSW))
257 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
258 class->lc_lock(lock, lock_state);
259 WITNESS_RESTORE(lock, lock_witness);
265 msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, int timo)
270 WITNESS_SAVE_DECL(mtx);
274 KASSERT(mtx != NULL, ("sleeping without a mutex"));
275 KASSERT(p != NULL, ("msleep1"));
276 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
278 if (cold || SCHEDULER_STOPPED()) {
280 * During autoconfiguration, just return;
281 * don't run any other threads or panic below,
282 * in case this is the idle thread and already asleep.
283 * XXX: this used to do "s = splhigh(); splx(safepri);
284 * splx(s);" to give interrupts a chance, but there is
285 * no way to give interrupts a chance now.
291 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
292 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
295 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
296 WITNESS_SAVE(&mtx->lock_object, mtx);
297 mtx_unlock_spin(mtx);
300 * We put ourselves on the sleep queue and start our timeout.
302 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
304 sleepq_set_timeout(ident, timo);
307 * Can't call ktrace with any spin locks held so it can lock the
308 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
309 * any spin lock. Thus, we have to drop the sleepq spin lock while
310 * we handle those requests. This is safe since we have placed our
311 * thread on the sleep queue already.
314 if (KTRPOINT(td, KTR_CSW)) {
315 sleepq_release(ident);
321 sleepq_release(ident);
322 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
327 rval = sleepq_timedwait(ident, 0);
329 sleepq_wait(ident, 0);
333 if (KTRPOINT(td, KTR_CSW))
338 WITNESS_RESTORE(&mtx->lock_object, mtx);
343 * pause() delays the calling thread by the given number of system ticks.
344 * During cold bootup, pause() uses the DELAY() function instead of
345 * the tsleep() function to do the waiting. The "timo" argument must be
346 * greater than or equal to zero. A "timo" value of zero is equivalent
347 * to a "timo" value of one.
350 pause(const char *wmesg, int timo)
352 KASSERT(timo >= 0, ("pause: timo must be >= 0"));
354 /* silently convert invalid timeouts */
358 if (cold || kdb_active || SCHEDULER_STOPPED()) {
360 * We delay one HZ at a time to avoid overflowing the
361 * system specific DELAY() function(s):
371 return (tsleep(&pause_wchan, 0, wmesg, timo));
375 * Make all threads sleeping on the specified identifier runnable.
383 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
384 sleepq_release(ident);
385 if (wakeup_swapper) {
386 KASSERT(ident != &proc0,
387 ("wakeup and wakeup_swapper and proc0"));
393 * Make a thread sleeping on the specified identifier runnable.
394 * May wake more than one thread if a target thread is currently
398 wakeup_one(void *ident)
403 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
404 sleepq_release(ident);
412 thread_unlock(curthread);
415 panic("%s: did not reenter debugger", __func__);
419 * The machine independent parts of context switching.
422 mi_switch(int flags, struct thread *newtd)
424 uint64_t runtime, new_switchtime;
428 td = curthread; /* XXX */
429 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
430 p = td->td_proc; /* XXX */
431 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
433 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
434 mtx_assert(&Giant, MA_NOTOWNED);
436 KASSERT(td->td_critnest == 1 || panicstr,
437 ("mi_switch: switch in a critical section"));
438 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
439 ("mi_switch: switch must be voluntary or involuntary"));
440 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
443 * Don't perform context switches from the debugger.
447 if (SCHEDULER_STOPPED())
449 if (flags & SW_VOL) {
450 td->td_ru.ru_nvcsw++;
451 td->td_swvoltick = ticks;
453 td->td_ru.ru_nivcsw++;
455 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
458 * Compute the amount of time during which the current
459 * thread was running, and add that to its total so far.
461 new_switchtime = cpu_ticks();
462 runtime = new_switchtime - PCPU_GET(switchtime);
463 td->td_runtime += runtime;
464 td->td_incruntime += runtime;
465 PCPU_SET(switchtime, new_switchtime);
466 td->td_generation++; /* bump preempt-detect counter */
467 PCPU_INC(cnt.v_swtch);
468 PCPU_SET(switchticks, ticks);
469 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
470 td->td_tid, td->td_sched, p->p_pid, td->td_name);
471 #if (KTR_COMPILE & KTR_SCHED) != 0
472 if (TD_IS_IDLETHREAD(td))
473 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
474 "prio:%d", td->td_priority);
476 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
477 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
478 "lockname:\"%s\"", td->td_lockname);
480 SDT_PROBE0(sched, , , preempt);
484 sched_switch(td, newtd, flags);
485 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
486 "prio:%d", td->td_priority);
488 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
489 td->td_tid, td->td_sched, p->p_pid, td->td_name);
492 * If the last thread was exiting, finish cleaning it up.
494 if ((td = PCPU_GET(deadthread))) {
495 PCPU_SET(deadthread, NULL);
501 * Change thread state to be runnable, placing it on the run queue if
502 * it is in memory. If it is swapped out, return true so our caller
503 * will know to awaken the swapper.
506 setrunnable(struct thread *td)
509 THREAD_LOCK_ASSERT(td, MA_OWNED);
510 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
511 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
512 switch (td->td_state) {
518 * If we are only inhibited because we are swapped out
519 * then arange to swap in this process. Otherwise just return.
521 if (td->td_inhibitors != TDI_SWAPPED)
527 printf("state is 0x%x", td->td_state);
528 panic("setrunnable(2)");
530 if ((td->td_flags & TDF_INMEM) == 0) {
531 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
532 td->td_flags |= TDF_SWAPINREQ;
541 * Compute a tenex style load average of a quantity on
542 * 1, 5 and 15 minute intervals.
553 for (i = 0; i < 3; i++)
554 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
555 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
558 * Schedule the next update to occur after 5 seconds, but add a
559 * random variation to avoid synchronisation with processes that
560 * run at regular intervals.
562 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
568 synch_setup(void *dummy)
570 callout_init(&loadav_callout, CALLOUT_MPSAFE);
572 /* Kick off timeout driven events by calling first time. */
580 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
588 kern_yield(PRI_USER);
599 if (prio == PRI_USER)
600 prio = td->td_user_pri;
602 sched_prio(td, prio);
603 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
609 * General purpose yield system call.
612 sys_yield(struct thread *td, struct yield_args *uap)
616 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
617 sched_prio(td, PRI_MAX_TIMESHARE);
618 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
620 td->td_retval[0] = 0;