FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_synch.c
1 /*-
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
21 *
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
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/10.4/sys/kern/kern_synch.c 316843 2017-04-14 14:45:44Z avg $");
39
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_sched.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.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
68 #include <machine/cpu.h>
69
70 #ifdef XEN
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <vm/pmap.h>
74 #endif
75
76 static void synch_setup(void *dummy);
77 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup,
78 NULL);
79
80 int hogticks;
81 static uint8_t pause_wchan[MAXCPU];
82
83 static struct callout loadav_callout;
84
85 struct loadavg averunnable =
86 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
87 /*
88 * Constants for averages over 1, 5, and 15 minutes
89 * when sampling at 5 second intervals.
90 */
91 static fixpt_t cexp[3] = {
92 0.9200444146293232 * FSCALE, /* exp(-1/12) */
93 0.9834714538216174 * FSCALE, /* exp(-1/60) */
94 0.9944598480048967 * FSCALE, /* exp(-1/180) */
95 };
96
97 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
98 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, "");
99
100 static void loadav(void *arg);
101
102 SDT_PROVIDER_DECLARE(sched);
103 SDT_PROBE_DEFINE(sched, , , preempt);
104
105 /*
106 * These probes reference Solaris features that are not implemented in FreeBSD.
107 * Create the probes anyway for compatibility with existing D scripts; they'll
108 * just never fire.
109 */
110 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep);
111 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup);
112 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt);
113 SDT_PROBE_DEFINE(sched, , , schedctl__preempt);
114 SDT_PROBE_DEFINE(sched, , , schedctl__yield);
115
116 static void
117 sleepinit(void *unused)
118 {
119
120 hogticks = (hz / 10) * 2; /* Default only. */
121 init_sleepqueues();
122 }
123
124 /*
125 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
126 * it is available.
127 */
128 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
129
130 /*
131 * General sleep call. Suspends the current thread until a wakeup is
132 * performed on the specified identifier. The thread will then be made
133 * runnable with the specified priority. Sleeps at most sbt units of time
134 * (0 means no timeout). If pri includes the PCATCH flag, let signals
135 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
136 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
137 * signal becomes pending, ERESTART is returned if the current system
138 * call should be restarted if possible, and EINTR is returned if the system
139 * call should be interrupted by the signal (return EINTR).
140 *
141 * The lock argument is unlocked before the caller is suspended, and
142 * re-locked before _sleep() returns. If priority includes the PDROP
143 * flag the lock is not re-locked before returning.
144 */
145 int
146 _sleep(void *ident, struct lock_object *lock, int priority,
147 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
148 {
149 struct thread *td;
150 struct proc *p;
151 struct lock_class *class;
152 uintptr_t lock_state;
153 int catch, pri, rval, sleepq_flags;
154 WITNESS_SAVE_DECL(lock_witness);
155
156 td = curthread;
157 p = td->td_proc;
158 #ifdef KTRACE
159 if (KTRPOINT(td, KTR_CSW))
160 ktrcsw(1, 0, wmesg);
161 #endif
162 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
163 "Sleeping on \"%s\"", wmesg);
164 KASSERT(sbt != 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"));
171 if (lock != NULL)
172 class = LOCK_CLASS(lock);
173 else
174 class = NULL;
175
176 if (cold || SCHEDULER_STOPPED()) {
177 /*
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.
184 */
185 if (lock != NULL && priority & PDROP)
186 class->lc_unlock(lock);
187 return (0);
188 }
189 catch = priority & PCATCH;
190 pri = priority & PRIMASK;
191
192 KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep"));
193
194 if ((uint8_t *)ident >= &pause_wchan[0] &&
195 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
196 sleepq_flags = SLEEPQ_PAUSE;
197 else
198 sleepq_flags = SLEEPQ_SLEEP;
199 if (catch)
200 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
201
202 sleepq_lock(ident);
203 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
204 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
205
206 if (lock == &Giant.lock_object)
207 mtx_assert(&Giant, MA_OWNED);
208 DROP_GIANT();
209 if (lock != NULL && lock != &Giant.lock_object &&
210 !(class->lc_flags & LC_SLEEPABLE)) {
211 WITNESS_SAVE(lock, lock_witness);
212 lock_state = class->lc_unlock(lock);
213 } else
214 /* GCC needs to follow the Yellow Brick Road */
215 lock_state = -1;
216
217 /*
218 * We put ourselves on the sleep queue and start our timeout
219 * before calling thread_suspend_check, as we could stop there,
220 * and a wakeup or a SIGCONT (or both) could occur while we were
221 * stopped without resuming us. Thus, we must be ready for sleep
222 * when cursig() is called. If the wakeup happens while we're
223 * stopped, then td will no longer be on a sleep queue upon
224 * return from cursig().
225 */
226 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
227 if (sbt != 0)
228 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
229 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
230 sleepq_release(ident);
231 WITNESS_SAVE(lock, lock_witness);
232 lock_state = class->lc_unlock(lock);
233 sleepq_lock(ident);
234 }
235 if (sbt != 0 && catch)
236 rval = sleepq_timedwait_sig(ident, pri);
237 else if (sbt != 0)
238 rval = sleepq_timedwait(ident, pri);
239 else if (catch)
240 rval = sleepq_wait_sig(ident, pri);
241 else {
242 sleepq_wait(ident, pri);
243 rval = 0;
244 }
245 #ifdef KTRACE
246 if (KTRPOINT(td, KTR_CSW))
247 ktrcsw(0, 0, wmesg);
248 #endif
249 PICKUP_GIANT();
250 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
251 class->lc_lock(lock, lock_state);
252 WITNESS_RESTORE(lock, lock_witness);
253 }
254 return (rval);
255 }
256
257 int
258 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
259 sbintime_t sbt, sbintime_t pr, int flags)
260 {
261 struct thread *td;
262 struct proc *p;
263 int rval;
264 WITNESS_SAVE_DECL(mtx);
265
266 td = curthread;
267 p = td->td_proc;
268 KASSERT(mtx != NULL, ("sleeping without a mutex"));
269 KASSERT(p != NULL, ("msleep1"));
270 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
271
272 if (cold || SCHEDULER_STOPPED()) {
273 /*
274 * During autoconfiguration, just return;
275 * don't run any other threads or panic below,
276 * in case this is the idle thread and already asleep.
277 * XXX: this used to do "s = splhigh(); splx(safepri);
278 * splx(s);" to give interrupts a chance, but there is
279 * no way to give interrupts a chance now.
280 */
281 return (0);
282 }
283
284 sleepq_lock(ident);
285 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
286 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
287
288 DROP_GIANT();
289 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
290 WITNESS_SAVE(&mtx->lock_object, mtx);
291 mtx_unlock_spin(mtx);
292
293 /*
294 * We put ourselves on the sleep queue and start our timeout.
295 */
296 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
297 if (sbt != 0)
298 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
299
300 /*
301 * Can't call ktrace with any spin locks held so it can lock the
302 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
303 * any spin lock. Thus, we have to drop the sleepq spin lock while
304 * we handle those requests. This is safe since we have placed our
305 * thread on the sleep queue already.
306 */
307 #ifdef KTRACE
308 if (KTRPOINT(td, KTR_CSW)) {
309 sleepq_release(ident);
310 ktrcsw(1, 0, wmesg);
311 sleepq_lock(ident);
312 }
313 #endif
314 #ifdef WITNESS
315 sleepq_release(ident);
316 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
317 wmesg);
318 sleepq_lock(ident);
319 #endif
320 if (sbt != 0)
321 rval = sleepq_timedwait(ident, 0);
322 else {
323 sleepq_wait(ident, 0);
324 rval = 0;
325 }
326 #ifdef KTRACE
327 if (KTRPOINT(td, KTR_CSW))
328 ktrcsw(0, 0, wmesg);
329 #endif
330 PICKUP_GIANT();
331 mtx_lock_spin(mtx);
332 WITNESS_RESTORE(&mtx->lock_object, mtx);
333 return (rval);
334 }
335
336 /*
337 * pause() delays the calling thread by the given number of system ticks.
338 * During cold bootup, pause() uses the DELAY() function instead of
339 * the tsleep() function to do the waiting. The "timo" argument must be
340 * greater than or equal to zero. A "timo" value of zero is equivalent
341 * to a "timo" value of one.
342 */
343 int
344 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
345 {
346 KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
347
348 /* silently convert invalid timeouts */
349 if (sbt == 0)
350 sbt = tick_sbt;
351
352 if (cold || kdb_active || SCHEDULER_STOPPED()) {
353 /*
354 * We delay one second at a time to avoid overflowing the
355 * system specific DELAY() function(s):
356 */
357 while (sbt >= SBT_1S) {
358 DELAY(1000000);
359 sbt -= SBT_1S;
360 }
361 /* Do the delay remainder, if any */
362 sbt = (sbt + SBT_1US - 1) / SBT_1US;
363 if (sbt > 0)
364 DELAY(sbt);
365 return (0);
366 }
367 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
368 }
369
370 /*
371 * Make all threads sleeping on the specified identifier runnable.
372 */
373 void
374 wakeup(void *ident)
375 {
376 int wakeup_swapper;
377
378 sleepq_lock(ident);
379 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
380 sleepq_release(ident);
381 if (wakeup_swapper) {
382 KASSERT(ident != &proc0,
383 ("wakeup and wakeup_swapper and proc0"));
384 kick_proc0();
385 }
386 }
387
388 /*
389 * Make a thread sleeping on the specified identifier runnable.
390 * May wake more than one thread if a target thread is currently
391 * swapped out.
392 */
393 void
394 wakeup_one(void *ident)
395 {
396 int wakeup_swapper;
397
398 sleepq_lock(ident);
399 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
400 sleepq_release(ident);
401 if (wakeup_swapper)
402 kick_proc0();
403 }
404
405 static void
406 kdb_switch(void)
407 {
408 thread_unlock(curthread);
409 kdb_backtrace();
410 kdb_reenter();
411 panic("%s: did not reenter debugger", __func__);
412 }
413
414 /*
415 * The machine independent parts of context switching.
416 */
417 void
418 mi_switch(int flags, struct thread *newtd)
419 {
420 uint64_t runtime, new_switchtime;
421 struct thread *td;
422
423 td = curthread; /* XXX */
424 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
425 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
426 #ifdef INVARIANTS
427 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
428 mtx_assert(&Giant, MA_NOTOWNED);
429 #endif
430 KASSERT(td->td_critnest == 1 || panicstr,
431 ("mi_switch: switch in a critical section"));
432 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
433 ("mi_switch: switch must be voluntary or involuntary"));
434 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
435
436 /*
437 * Don't perform context switches from the debugger.
438 */
439 if (kdb_active)
440 kdb_switch();
441 if (SCHEDULER_STOPPED())
442 return;
443 if (flags & SW_VOL) {
444 td->td_ru.ru_nvcsw++;
445 td->td_swvoltick = ticks;
446 } else
447 td->td_ru.ru_nivcsw++;
448 #ifdef SCHED_STATS
449 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
450 #endif
451 /*
452 * Compute the amount of time during which the current
453 * thread was running, and add that to its total so far.
454 */
455 new_switchtime = cpu_ticks();
456 runtime = new_switchtime - PCPU_GET(switchtime);
457 td->td_runtime += runtime;
458 td->td_incruntime += runtime;
459 PCPU_SET(switchtime, new_switchtime);
460 td->td_generation++; /* bump preempt-detect counter */
461 PCPU_INC(cnt.v_swtch);
462 PCPU_SET(switchticks, ticks);
463 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)",
464 td->td_tid, td->td_sched, td->td_proc->p_pid, td->td_name);
465 #ifdef KDTRACE_HOOKS
466 if ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 &&
467 (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED))
468 SDT_PROBE0(sched, , , preempt);
469 #endif
470 #ifdef XEN
471 PT_UPDATES_FLUSH();
472 #endif
473 sched_switch(td, newtd, flags);
474 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
475 td->td_tid, td->td_sched, td->td_proc->p_pid, td->td_name);
476
477 /*
478 * If the last thread was exiting, finish cleaning it up.
479 */
480 if ((td = PCPU_GET(deadthread))) {
481 PCPU_SET(deadthread, NULL);
482 thread_stash(td);
483 }
484 }
485
486 /*
487 * Change thread state to be runnable, placing it on the run queue if
488 * it is in memory. If it is swapped out, return true so our caller
489 * will know to awaken the swapper.
490 */
491 int
492 setrunnable(struct thread *td)
493 {
494
495 THREAD_LOCK_ASSERT(td, MA_OWNED);
496 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
497 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
498 switch (td->td_state) {
499 case TDS_RUNNING:
500 case TDS_RUNQ:
501 return (0);
502 case TDS_INHIBITED:
503 /*
504 * If we are only inhibited because we are swapped out
505 * then arange to swap in this process. Otherwise just return.
506 */
507 if (td->td_inhibitors != TDI_SWAPPED)
508 return (0);
509 /* FALLTHROUGH */
510 case TDS_CAN_RUN:
511 break;
512 default:
513 printf("state is 0x%x", td->td_state);
514 panic("setrunnable(2)");
515 }
516 if ((td->td_flags & TDF_INMEM) == 0) {
517 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
518 td->td_flags |= TDF_SWAPINREQ;
519 return (1);
520 }
521 } else
522 sched_wakeup(td);
523 return (0);
524 }
525
526 /*
527 * Compute a tenex style load average of a quantity on
528 * 1, 5 and 15 minute intervals.
529 */
530 static void
531 loadav(void *arg)
532 {
533 int i, nrun;
534 struct loadavg *avg;
535
536 nrun = sched_load();
537 avg = &averunnable;
538
539 for (i = 0; i < 3; i++)
540 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
541 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
542
543 /*
544 * Schedule the next update to occur after 5 seconds, but add a
545 * random variation to avoid synchronisation with processes that
546 * run at regular intervals.
547 */
548 callout_reset_sbt(&loadav_callout,
549 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
550 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
551 }
552
553 /* ARGSUSED */
554 static void
555 synch_setup(void *dummy)
556 {
557 callout_init(&loadav_callout, 1);
558
559 /* Kick off timeout driven events by calling first time. */
560 loadav(NULL);
561 }
562
563 int
564 should_yield(void)
565 {
566
567 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
568 }
569
570 void
571 maybe_yield(void)
572 {
573
574 if (should_yield())
575 kern_yield(PRI_USER);
576 }
577
578 void
579 kern_yield(int prio)
580 {
581 struct thread *td;
582
583 td = curthread;
584 DROP_GIANT();
585 thread_lock(td);
586 if (prio == PRI_USER)
587 prio = td->td_user_pri;
588 if (prio >= 0)
589 sched_prio(td, prio);
590 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
591 thread_unlock(td);
592 PICKUP_GIANT();
593 }
594
595 /*
596 * General purpose yield system call.
597 */
598 int
599 sys_yield(struct thread *td, struct yield_args *uap)
600 {
601
602 thread_lock(td);
603 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
604 sched_prio(td, PRI_MAX_TIMESHARE);
605 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
606 thread_unlock(td);
607 td->td_retval[0] = 0;
608 return (0);
609 }
Cache object: 579354298de7230d9aae70ea8514f333
|