1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1990, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
37 */
38
39 #include <sys/cdefs.h>
40 #include "opt_ktrace.h"
41 #include "opt_sched.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/blockcount.h>
46 #include <sys/condvar.h>
47 #include <sys/kdb.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/sched.h>
55 #include <sys/sdt.h>
56 #include <sys/signalvar.h>
57 #include <sys/sleepqueue.h>
58 #include <sys/smp.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysproto.h>
62 #include <sys/vmmeter.h>
63 #ifdef KTRACE
64 #include <sys/uio.h>
65 #include <sys/ktrace.h>
66 #endif
67 #ifdef EPOCH_TRACE
68 #include <sys/epoch.h>
69 #endif
70
71 #include <machine/cpu.h>
72
73 static void synch_setup(void *dummy);
74 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
75 NULL);
76
77 int hogticks;
78 static const char pause_wchan[MAXCPU];
79
80 static struct callout loadav_callout;
81
82 struct loadavg averunnable =
83 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
84 /*
85 * Constants for averages over 1, 5, and 15 minutes
86 * when sampling at 5 second intervals.
87 */
88 static uint64_t cexp[3] = {
89 0.9200444146293232 * FSCALE, /* exp(-1/12) */
90 0.9834714538216174 * FSCALE, /* exp(-1/60) */
91 0.9944598480048967 * FSCALE, /* exp(-1/180) */
92 };
93
94 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
95 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE,
96 "Fixed-point scale factor used for calculating load average values");
97
98 static void loadav(void *arg);
99
100 SDT_PROVIDER_DECLARE(sched);
101 SDT_PROBE_DEFINE(sched, , , preempt);
102
103 static void
sleepinit(void * unused)104 sleepinit(void *unused)
105 {
106
107 hogticks = (hz / 10) * 2; /* Default only. */
108 init_sleepqueues();
109 }
110
111 /*
112 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
113 * it is available.
114 */
115 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, NULL);
116
117 /*
118 * General sleep call. Suspends the current thread until a wakeup is
119 * performed on the specified identifier. The thread will then be made
120 * runnable with the specified priority. Sleeps at most sbt units of time
121 * (0 means no timeout). If pri includes the PCATCH flag, let signals
122 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
123 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
124 * signal becomes pending, ERESTART is returned if the current system
125 * call should be restarted if possible, and EINTR is returned if the system
126 * call should be interrupted by the signal (return EINTR).
127 *
128 * The lock argument is unlocked before the caller is suspended, and
129 * re-locked before _sleep() returns. If priority includes the PDROP
130 * flag the lock is not re-locked before returning.
131 */
132 int
_sleep(const void * ident,struct lock_object * lock,int priority,const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)133 _sleep(const void *ident, struct lock_object *lock, int priority,
134 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
135 {
136 struct thread *td;
137 struct lock_class *class;
138 uintptr_t lock_state;
139 int catch, pri, rval, sleepq_flags;
140 WITNESS_SAVE_DECL(lock_witness);
141
142 TSENTER();
143 td = curthread;
144 #ifdef KTRACE
145 if (KTRPOINT(td, KTR_CSW))
146 ktrcsw(1, 0, wmesg);
147 #endif
148 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
149 "Sleeping on \"%s\"", wmesg);
150 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
151 ("sleeping without a lock"));
152 KASSERT(ident != NULL, ("_sleep: NULL ident"));
153 KASSERT(TD_IS_RUNNING(td), ("_sleep: curthread not running"));
154 if (priority & PDROP)
155 KASSERT(lock != NULL && lock != &Giant.lock_object,
156 ("PDROP requires a non-Giant lock"));
157 if (lock != NULL)
158 class = LOCK_CLASS(lock);
159 else
160 class = NULL;
161
162 if (SCHEDULER_STOPPED_TD(td)) {
163 if (lock != NULL && priority & PDROP)
164 class->lc_unlock(lock);
165 return (0);
166 }
167 catch = priority & PCATCH;
168 pri = priority & PRIMASK;
169
170 KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep"));
171
172 if ((uintptr_t)ident >= (uintptr_t)&pause_wchan[0] &&
173 (uintptr_t)ident <= (uintptr_t)&pause_wchan[MAXCPU - 1])
174 sleepq_flags = SLEEPQ_PAUSE;
175 else
176 sleepq_flags = SLEEPQ_SLEEP;
177 if (catch)
178 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
179
180 sleepq_lock(ident);
181 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
182 td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident);
183
184 if (lock == &Giant.lock_object)
185 mtx_assert(&Giant, MA_OWNED);
186 DROP_GIANT();
187 if (lock != NULL && lock != &Giant.lock_object &&
188 !(class->lc_flags & LC_SLEEPABLE)) {
189 KASSERT(!(class->lc_flags & LC_SPINLOCK),
190 ("spin locks can only use msleep_spin"));
191 WITNESS_SAVE(lock, lock_witness);
192 lock_state = class->lc_unlock(lock);
193 } else
194 /* GCC needs to follow the Yellow Brick Road */
195 lock_state = -1;
196
197 /*
198 * We put ourselves on the sleep queue and start our timeout
199 * before calling thread_suspend_check, as we could stop there,
200 * and a wakeup or a SIGCONT (or both) could occur while we were
201 * stopped without resuming us. Thus, we must be ready for sleep
202 * when cursig() is called. If the wakeup happens while we're
203 * stopped, then td will no longer be on a sleep queue upon
204 * return from cursig().
205 */
206 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
207 if (sbt != 0)
208 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
209 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
210 sleepq_release(ident);
211 WITNESS_SAVE(lock, lock_witness);
212 lock_state = class->lc_unlock(lock);
213 sleepq_lock(ident);
214 }
215 if (sbt != 0 && catch)
216 rval = sleepq_timedwait_sig(ident, pri);
217 else if (sbt != 0)
218 rval = sleepq_timedwait(ident, pri);
219 else if (catch)
220 rval = sleepq_wait_sig(ident, pri);
221 else {
222 sleepq_wait(ident, pri);
223 rval = 0;
224 }
225 #ifdef KTRACE
226 if (KTRPOINT(td, KTR_CSW))
227 ktrcsw(0, 0, wmesg);
228 #endif
229 PICKUP_GIANT();
230 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
231 class->lc_lock(lock, lock_state);
232 WITNESS_RESTORE(lock, lock_witness);
233 }
234 TSEXIT();
235 return (rval);
236 }
237
238 int
msleep_spin_sbt(const void * ident,struct mtx * mtx,const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)239 msleep_spin_sbt(const void *ident, struct mtx *mtx, const char *wmesg,
240 sbintime_t sbt, sbintime_t pr, int flags)
241 {
242 struct thread *td;
243 int rval;
244 WITNESS_SAVE_DECL(mtx);
245
246 td = curthread;
247 KASSERT(mtx != NULL, ("sleeping without a mutex"));
248 KASSERT(ident != NULL, ("msleep_spin_sbt: NULL ident"));
249 KASSERT(TD_IS_RUNNING(td), ("msleep_spin_sbt: curthread not running"));
250
251 if (SCHEDULER_STOPPED_TD(td))
252 return (0);
253
254 sleepq_lock(ident);
255 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
256 td->td_tid, td->td_proc->p_pid, td->td_name, wmesg, ident);
257
258 DROP_GIANT();
259 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
260 WITNESS_SAVE(&mtx->lock_object, mtx);
261 mtx_unlock_spin(mtx);
262
263 /*
264 * We put ourselves on the sleep queue and start our timeout.
265 */
266 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
267 if (sbt != 0)
268 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
269
270 /*
271 * Can't call ktrace with any spin locks held so it can lock the
272 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
273 * any spin lock. Thus, we have to drop the sleepq spin lock while
274 * we handle those requests. This is safe since we have placed our
275 * thread on the sleep queue already.
276 */
277 #ifdef KTRACE
278 if (KTRPOINT(td, KTR_CSW)) {
279 sleepq_release(ident);
280 ktrcsw(1, 0, wmesg);
281 sleepq_lock(ident);
282 }
283 #endif
284 #ifdef WITNESS
285 sleepq_release(ident);
286 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
287 wmesg);
288 sleepq_lock(ident);
289 #endif
290 if (sbt != 0)
291 rval = sleepq_timedwait(ident, 0);
292 else {
293 sleepq_wait(ident, 0);
294 rval = 0;
295 }
296 #ifdef KTRACE
297 if (KTRPOINT(td, KTR_CSW))
298 ktrcsw(0, 0, wmesg);
299 #endif
300 PICKUP_GIANT();
301 mtx_lock_spin(mtx);
302 WITNESS_RESTORE(&mtx->lock_object, mtx);
303 return (rval);
304 }
305
306 /*
307 * pause_sbt() delays the calling thread by the given signed binary
308 * time. During cold bootup, pause_sbt() uses the DELAY() function
309 * instead of the _sleep() function to do the waiting. The "sbt"
310 * argument must be greater than or equal to zero. A "sbt" value of
311 * zero is equivalent to a "sbt" value of one tick.
312 */
313 int
pause_sbt(const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)314 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
315 {
316 KASSERT(sbt >= 0, ("pause_sbt: timeout must be >= 0"));
317
318 /* silently convert invalid timeouts */
319 if (sbt == 0)
320 sbt = tick_sbt;
321
322 if ((cold && curthread == &thread0) || kdb_active ||
323 SCHEDULER_STOPPED()) {
324 /*
325 * We delay one second at a time to avoid overflowing the
326 * system specific DELAY() function(s):
327 */
328 while (sbt >= SBT_1S) {
329 DELAY(1000000);
330 sbt -= SBT_1S;
331 }
332 /* Do the delay remainder, if any */
333 sbt = howmany(sbt, SBT_1US);
334 if (sbt > 0)
335 DELAY(sbt);
336 return (EWOULDBLOCK);
337 }
338 return (_sleep(&pause_wchan[curcpu], NULL,
339 (flags & C_CATCH) ? PCATCH : 0, wmesg, sbt, pr, flags));
340 }
341
342 /*
343 * Make all threads sleeping on the specified identifier runnable.
344 */
345 void
wakeup(const void * ident)346 wakeup(const void *ident)
347 {
348 int wakeup_swapper;
349
350 sleepq_lock(ident);
351 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
352 sleepq_release(ident);
353 if (wakeup_swapper) {
354 KASSERT(ident != &proc0,
355 ("wakeup and wakeup_swapper and proc0"));
356 kick_proc0();
357 }
358 }
359
360 /*
361 * Make a thread sleeping on the specified identifier runnable.
362 * May wake more than one thread if a target thread is currently
363 * swapped out.
364 */
365 void
wakeup_one(const void * ident)366 wakeup_one(const void *ident)
367 {
368 int wakeup_swapper;
369
370 sleepq_lock(ident);
371 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_DROP, 0, 0);
372 if (wakeup_swapper)
373 kick_proc0();
374 }
375
376 void
wakeup_any(const void * ident)377 wakeup_any(const void *ident)
378 {
379 int wakeup_swapper;
380
381 sleepq_lock(ident);
382 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP | SLEEPQ_UNFAIR |
383 SLEEPQ_DROP, 0, 0);
384 if (wakeup_swapper)
385 kick_proc0();
386 }
387
388 /*
389 * Signal sleeping waiters after the counter has reached zero.
390 */
391 void
_blockcount_wakeup(blockcount_t * bc,u_int old)392 _blockcount_wakeup(blockcount_t *bc, u_int old)
393 {
394
395 KASSERT(_BLOCKCOUNT_WAITERS(old),
396 ("%s: no waiters on %p", __func__, bc));
397
398 if (atomic_cmpset_int(&bc->__count, _BLOCKCOUNT_WAITERS_FLAG, 0))
399 wakeup(bc);
400 }
401
402 /*
403 * Wait for a wakeup or a signal. This does not guarantee that the count is
404 * still zero on return. Callers wanting a precise answer should use
405 * blockcount_wait() with an interlock.
406 *
407 * If there is no work to wait for, return 0. If the sleep was interrupted by a
408 * signal, return EINTR or ERESTART, and return EAGAIN otherwise.
409 */
410 int
_blockcount_sleep(blockcount_t * bc,struct lock_object * lock,const char * wmesg,int prio)411 _blockcount_sleep(blockcount_t *bc, struct lock_object *lock, const char *wmesg,
412 int prio)
413 {
414 void *wchan;
415 uintptr_t lock_state;
416 u_int old;
417 int ret;
418 bool catch, drop;
419
420 KASSERT(lock != &Giant.lock_object,
421 ("%s: cannot use Giant as the interlock", __func__));
422
423 catch = (prio & PCATCH) != 0;
424 drop = (prio & PDROP) != 0;
425 prio &= PRIMASK;
426
427 /*
428 * Synchronize with the fence in blockcount_release(). If we end up
429 * waiting, the sleepqueue lock acquisition will provide the required
430 * side effects.
431 *
432 * If there is no work to wait for, but waiters are present, try to put
433 * ourselves to sleep to avoid jumping ahead.
434 */
435 if (atomic_load_acq_int(&bc->__count) == 0) {
436 if (lock != NULL && drop)
437 LOCK_CLASS(lock)->lc_unlock(lock);
438 return (0);
439 }
440 lock_state = 0;
441 wchan = bc;
442 sleepq_lock(wchan);
443 DROP_GIANT();
444 if (lock != NULL)
445 lock_state = LOCK_CLASS(lock)->lc_unlock(lock);
446 old = blockcount_read(bc);
447 ret = 0;
448 do {
449 if (_BLOCKCOUNT_COUNT(old) == 0) {
450 sleepq_release(wchan);
451 goto out;
452 }
453 if (_BLOCKCOUNT_WAITERS(old))
454 break;
455 } while (!atomic_fcmpset_int(&bc->__count, &old,
456 old | _BLOCKCOUNT_WAITERS_FLAG));
457 sleepq_add(wchan, NULL, wmesg, catch ? SLEEPQ_INTERRUPTIBLE : 0, 0);
458 if (catch)
459 ret = sleepq_wait_sig(wchan, prio);
460 else
461 sleepq_wait(wchan, prio);
462 if (ret == 0)
463 ret = EAGAIN;
464
465 out:
466 PICKUP_GIANT();
467 if (lock != NULL && !drop)
468 LOCK_CLASS(lock)->lc_lock(lock, lock_state);
469
470 return (ret);
471 }
472
473 static void
kdb_switch(void)474 kdb_switch(void)
475 {
476 thread_unlock(curthread);
477 kdb_backtrace();
478 kdb_reenter();
479 panic("%s: did not reenter debugger", __func__);
480 }
481
482 /*
483 * The machine independent parts of context switching.
484 *
485 * The thread lock is required on entry and is no longer held on return.
486 */
487 void
mi_switch(int flags)488 mi_switch(int flags)
489 {
490 uint64_t runtime, new_switchtime;
491 struct thread *td;
492
493 td = curthread; /* XXX */
494 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
495 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
496 #ifdef INVARIANTS
497 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
498 mtx_assert(&Giant, MA_NOTOWNED);
499 #endif
500 KASSERT(td->td_critnest == 1 || KERNEL_PANICKED(),
501 ("mi_switch: switch in a critical section"));
502 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
503 ("mi_switch: switch must be voluntary or involuntary"));
504
505 /*
506 * Don't perform context switches from the debugger.
507 */
508 if (kdb_active)
509 kdb_switch();
510 if (SCHEDULER_STOPPED_TD(td))
511 return;
512 if (flags & SW_VOL) {
513 td->td_ru.ru_nvcsw++;
514 td->td_swvoltick = ticks;
515 } else {
516 td->td_ru.ru_nivcsw++;
517 td->td_swinvoltick = ticks;
518 }
519 #ifdef SCHED_STATS
520 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
521 #endif
522 /*
523 * Compute the amount of time during which the current
524 * thread was running, and add that to its total so far.
525 */
526 new_switchtime = cpu_ticks();
527 runtime = new_switchtime - PCPU_GET(switchtime);
528 td->td_runtime += runtime;
529 td->td_incruntime += runtime;
530 PCPU_SET(switchtime, new_switchtime);
531 td->td_generation++; /* bump preempt-detect counter */
532 VM_CNT_INC(v_swtch);
533 PCPU_SET(switchticks, ticks);
534 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
535 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
536 #ifdef KDTRACE_HOOKS
537 if (SDT_PROBES_ENABLED() &&
538 ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 &&
539 (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED)))
540 SDT_PROBE0(sched, , , preempt);
541 #endif
542 sched_switch(td, flags);
543 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
544 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name);
545
546 /*
547 * If the last thread was exiting, finish cleaning it up.
548 */
549 if ((td = PCPU_GET(deadthread))) {
550 PCPU_SET(deadthread, NULL);
551 thread_stash(td);
552 }
553 spinlock_exit();
554 }
555
556 /*
557 * Change thread state to be runnable, placing it on the run queue if
558 * it is in memory. If it is swapped out, return true so our caller
559 * will know to awaken the swapper.
560 *
561 * Requires the thread lock on entry, drops on exit.
562 */
563 int
setrunnable(struct thread * td,int srqflags)564 setrunnable(struct thread *td, int srqflags)
565 {
566 int swapin;
567
568 THREAD_LOCK_ASSERT(td, MA_OWNED);
569 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
570 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
571
572 swapin = 0;
573 switch (td->td_state) {
574 case TDS_RUNNING:
575 case TDS_RUNQ:
576 break;
577 case TDS_CAN_RUN:
578 KASSERT((td->td_flags & TDF_INMEM) != 0,
579 ("setrunnable: td %p not in mem, flags 0x%X inhibit 0x%X",
580 td, td->td_flags, td->td_inhibitors));
581 /* unlocks thread lock according to flags */
582 sched_wakeup(td, srqflags);
583 return (0);
584 case TDS_INHIBITED:
585 /*
586 * If we are only inhibited because we are swapped out
587 * arrange to swap in this process.
588 */
589 if (td->td_inhibitors == TDI_SWAPPED &&
590 (td->td_flags & TDF_SWAPINREQ) == 0) {
591 td->td_flags |= TDF_SWAPINREQ;
592 swapin = 1;
593 }
594 break;
595 default:
596 panic("setrunnable: state 0x%x", td->td_state);
597 }
598 if ((srqflags & (SRQ_HOLD | SRQ_HOLDTD)) == 0)
599 thread_unlock(td);
600
601 return (swapin);
602 }
603
604 /*
605 * Compute a tenex style load average of a quantity on
606 * 1, 5 and 15 minute intervals.
607 */
608 static void
loadav(void * arg)609 loadav(void *arg)
610 {
611 int i;
612 uint64_t nrun;
613 struct loadavg *avg;
614
615 nrun = (uint64_t)sched_load();
616 avg = &averunnable;
617
618 for (i = 0; i < 3; i++)
619 avg->ldavg[i] = (cexp[i] * (uint64_t)avg->ldavg[i] +
620 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
621
622 /*
623 * Schedule the next update to occur after 5 seconds, but add a
624 * random variation to avoid synchronisation with processes that
625 * run at regular intervals.
626 */
627 callout_reset_sbt(&loadav_callout,
628 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
629 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
630 }
631
632 /* ARGSUSED */
633 static void
synch_setup(void * dummy)634 synch_setup(void *dummy)
635 {
636 callout_init(&loadav_callout, 1);
637
638 /* Kick off timeout driven events by calling first time. */
639 loadav(NULL);
640 }
641
642 bool
should_yield(void)643 should_yield(void)
644 {
645
646 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
647 }
648
649 void
maybe_yield(void)650 maybe_yield(void)
651 {
652
653 if (should_yield())
654 kern_yield(PRI_USER);
655 }
656
657 void
kern_yield(int prio)658 kern_yield(int prio)
659 {
660 struct thread *td;
661
662 td = curthread;
663 DROP_GIANT();
664 thread_lock(td);
665 if (prio == PRI_USER)
666 prio = td->td_user_pri;
667 if (prio >= 0)
668 sched_prio(td, prio);
669 mi_switch(SW_VOL | SWT_RELINQUISH);
670 PICKUP_GIANT();
671 }
672
673 /*
674 * General purpose yield system call.
675 */
676 int
sys_yield(struct thread * td,struct yield_args * uap)677 sys_yield(struct thread *td, struct yield_args *uap)
678 {
679
680 thread_lock(td);
681 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
682 sched_prio(td, PRI_MAX_TIMESHARE);
683 mi_switch(SW_VOL | SWT_RELINQUISH);
684 td->td_retval[0] = 0;
685 return (0);
686 }
687
688 int
sys_sched_getcpu(struct thread * td,struct sched_getcpu_args * uap)689 sys_sched_getcpu(struct thread *td, struct sched_getcpu_args *uap)
690 {
691 td->td_retval[0] = td->td_oncpu;
692 return (0);
693 }
694