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("$FreeBSD$");
39
40 #include "opt_ktrace.h"
41 #include "opt_sched.h"
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/condvar.h>
46 #include <sys/kdb.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/sched.h>
54 #include <sys/sdt.h>
55 #include <sys/signalvar.h>
56 #include <sys/sleepqueue.h>
57 #include <sys/smp.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysproto.h>
61 #include <sys/vmmeter.h>
62 #ifdef KTRACE
63 #include <sys/uio.h>
64 #include <sys/ktrace.h>
65 #endif
66
67 #include <machine/cpu.h>
68
69 #define KTDSTATE(td) \
70 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \
71 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \
72 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \
73 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \
74 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding")
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 static void
sleepinit(void * unused)106 sleepinit(void *unused)
107 {
108
109 hogticks = (hz / 10) * 2; /* Default only. */
110 init_sleepqueues();
111 }
112
113 /*
114 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure
115 * it is available.
116 */
117 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0);
118
119 /*
120 * General sleep call. Suspends the current thread until a wakeup is
121 * performed on the specified identifier. The thread will then be made
122 * runnable with the specified priority. Sleeps at most sbt units of time
123 * (0 means no timeout). If pri includes the PCATCH flag, let signals
124 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if
125 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
126 * signal becomes pending, ERESTART is returned if the current system
127 * call should be restarted if possible, and EINTR is returned if the system
128 * call should be interrupted by the signal (return EINTR).
129 *
130 * The lock argument is unlocked before the caller is suspended, and
131 * re-locked before _sleep() returns. If priority includes the PDROP
132 * flag the lock is not re-locked before returning.
133 */
134 int
_sleep(void * ident,struct lock_object * lock,int priority,const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)135 _sleep(void *ident, struct lock_object *lock, int priority,
136 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
137 {
138 struct thread *td;
139 struct proc *p;
140 struct lock_class *class;
141 uintptr_t lock_state;
142 int catch, pri, rval, sleepq_flags;
143 WITNESS_SAVE_DECL(lock_witness);
144
145 td = curthread;
146 p = td->td_proc;
147 #ifdef KTRACE
148 if (KTRPOINT(td, KTR_CSW))
149 ktrcsw(1, 0, wmesg);
150 #endif
151 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
152 "Sleeping on \"%s\"", wmesg);
153 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL,
154 ("sleeping without a lock"));
155 KASSERT(p != NULL, ("msleep1"));
156 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
157 if (priority & PDROP)
158 KASSERT(lock != NULL && lock != &Giant.lock_object,
159 ("PDROP requires a non-Giant lock"));
160 if (lock != NULL)
161 class = LOCK_CLASS(lock);
162 else
163 class = NULL;
164
165 if (cold || SCHEDULER_STOPPED()) {
166 /*
167 * During autoconfiguration, just return;
168 * don't run any other threads or panic below,
169 * in case this is the idle thread and already asleep.
170 * XXX: this used to do "s = splhigh(); splx(safepri);
171 * splx(s);" to give interrupts a chance, but there is
172 * no way to give interrupts a chance now.
173 */
174 if (lock != NULL && priority & PDROP)
175 class->lc_unlock(lock);
176 return (0);
177 }
178 catch = priority & PCATCH;
179 pri = priority & PRIMASK;
180
181 /*
182 * If we are already on a sleep queue, then remove us from that
183 * sleep queue first. We have to do this to handle recursive
184 * sleeps.
185 */
186 if (TD_ON_SLEEPQ(td))
187 sleepq_remove(td, td->td_wchan);
188
189 if ((uint8_t *)ident >= &pause_wchan[0] &&
190 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1])
191 sleepq_flags = SLEEPQ_PAUSE;
192 else
193 sleepq_flags = SLEEPQ_SLEEP;
194 if (catch)
195 sleepq_flags |= SLEEPQ_INTERRUPTIBLE;
196
197 sleepq_lock(ident);
198 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)",
199 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
200
201 if (lock == &Giant.lock_object)
202 mtx_assert(&Giant, MA_OWNED);
203 DROP_GIANT();
204 if (lock != NULL && lock != &Giant.lock_object &&
205 !(class->lc_flags & LC_SLEEPABLE)) {
206 WITNESS_SAVE(lock, lock_witness);
207 lock_state = class->lc_unlock(lock);
208 } else
209 /* GCC needs to follow the Yellow Brick Road */
210 lock_state = -1;
211
212 /*
213 * We put ourselves on the sleep queue and start our timeout
214 * before calling thread_suspend_check, as we could stop there,
215 * and a wakeup or a SIGCONT (or both) could occur while we were
216 * stopped without resuming us. Thus, we must be ready for sleep
217 * when cursig() is called. If the wakeup happens while we're
218 * stopped, then td will no longer be on a sleep queue upon
219 * return from cursig().
220 */
221 sleepq_add(ident, lock, wmesg, sleepq_flags, 0);
222 if (sbt != 0)
223 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
224 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) {
225 sleepq_release(ident);
226 WITNESS_SAVE(lock, lock_witness);
227 lock_state = class->lc_unlock(lock);
228 sleepq_lock(ident);
229 }
230 if (sbt != 0 && catch)
231 rval = sleepq_timedwait_sig(ident, pri);
232 else if (sbt != 0)
233 rval = sleepq_timedwait(ident, pri);
234 else if (catch)
235 rval = sleepq_wait_sig(ident, pri);
236 else {
237 sleepq_wait(ident, pri);
238 rval = 0;
239 }
240 #ifdef KTRACE
241 if (KTRPOINT(td, KTR_CSW))
242 ktrcsw(0, 0, wmesg);
243 #endif
244 PICKUP_GIANT();
245 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) {
246 class->lc_lock(lock, lock_state);
247 WITNESS_RESTORE(lock, lock_witness);
248 }
249 return (rval);
250 }
251
252 int
msleep_spin_sbt(void * ident,struct mtx * mtx,const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)253 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg,
254 sbintime_t sbt, sbintime_t pr, int flags)
255 {
256 struct thread *td;
257 struct proc *p;
258 int rval;
259 WITNESS_SAVE_DECL(mtx);
260
261 td = curthread;
262 p = td->td_proc;
263 KASSERT(mtx != NULL, ("sleeping without a mutex"));
264 KASSERT(p != NULL, ("msleep1"));
265 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
266
267 if (cold || SCHEDULER_STOPPED()) {
268 /*
269 * During autoconfiguration, just return;
270 * don't run any other threads or panic below,
271 * in case this is the idle thread and already asleep.
272 * XXX: this used to do "s = splhigh(); splx(safepri);
273 * splx(s);" to give interrupts a chance, but there is
274 * no way to give interrupts a chance now.
275 */
276 return (0);
277 }
278
279 sleepq_lock(ident);
280 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)",
281 td->td_tid, p->p_pid, td->td_name, wmesg, ident);
282
283 DROP_GIANT();
284 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
285 WITNESS_SAVE(&mtx->lock_object, mtx);
286 mtx_unlock_spin(mtx);
287
288 /*
289 * We put ourselves on the sleep queue and start our timeout.
290 */
291 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0);
292 if (sbt != 0)
293 sleepq_set_timeout_sbt(ident, sbt, pr, flags);
294
295 /*
296 * Can't call ktrace with any spin locks held so it can lock the
297 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
298 * any spin lock. Thus, we have to drop the sleepq spin lock while
299 * we handle those requests. This is safe since we have placed our
300 * thread on the sleep queue already.
301 */
302 #ifdef KTRACE
303 if (KTRPOINT(td, KTR_CSW)) {
304 sleepq_release(ident);
305 ktrcsw(1, 0, wmesg);
306 sleepq_lock(ident);
307 }
308 #endif
309 #ifdef WITNESS
310 sleepq_release(ident);
311 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
312 wmesg);
313 sleepq_lock(ident);
314 #endif
315 if (sbt != 0)
316 rval = sleepq_timedwait(ident, 0);
317 else {
318 sleepq_wait(ident, 0);
319 rval = 0;
320 }
321 #ifdef KTRACE
322 if (KTRPOINT(td, KTR_CSW))
323 ktrcsw(0, 0, wmesg);
324 #endif
325 PICKUP_GIANT();
326 mtx_lock_spin(mtx);
327 WITNESS_RESTORE(&mtx->lock_object, mtx);
328 return (rval);
329 }
330
331 /*
332 * pause() delays the calling thread by the given number of system ticks.
333 * During cold bootup, pause() uses the DELAY() function instead of
334 * the tsleep() function to do the waiting. The "timo" argument must be
335 * greater than or equal to zero. A "timo" value of zero is equivalent
336 * to a "timo" value of one.
337 */
338 int
pause_sbt(const char * wmesg,sbintime_t sbt,sbintime_t pr,int flags)339 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags)
340 {
341 KASSERT(sbt >= 0, ("pause: timeout must be >= 0"));
342
343 /* silently convert invalid timeouts */
344 if (sbt == 0)
345 sbt = tick_sbt;
346
347 if (cold || kdb_active) {
348 /*
349 * We delay one second at a time to avoid overflowing the
350 * system specific DELAY() function(s):
351 */
352 while (sbt >= SBT_1S) {
353 DELAY(1000000);
354 sbt -= SBT_1S;
355 }
356 /* Do the delay remainder, if any */
357 sbt = (sbt + SBT_1US - 1) / SBT_1US;
358 if (sbt > 0)
359 DELAY(sbt);
360 return (0);
361 }
362 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags));
363 }
364
365 /*
366 * Make all threads sleeping on the specified identifier runnable.
367 */
368 void
wakeup(void * ident)369 wakeup(void *ident)
370 {
371 int wakeup_swapper;
372
373 sleepq_lock(ident);
374 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0);
375 sleepq_release(ident);
376 if (wakeup_swapper) {
377 KASSERT(ident != &proc0,
378 ("wakeup and wakeup_swapper and proc0"));
379 kick_proc0();
380 }
381 }
382
383 /*
384 * Make a thread sleeping on the specified identifier runnable.
385 * May wake more than one thread if a target thread is currently
386 * swapped out.
387 */
388 void
wakeup_one(void * ident)389 wakeup_one(void *ident)
390 {
391 int wakeup_swapper;
392
393 sleepq_lock(ident);
394 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0);
395 sleepq_release(ident);
396 if (wakeup_swapper)
397 kick_proc0();
398 }
399
400 static void
kdb_switch(void)401 kdb_switch(void)
402 {
403 thread_unlock(curthread);
404 kdb_backtrace();
405 kdb_reenter();
406 panic("%s: did not reenter debugger", __func__);
407 }
408
409 /*
410 * The machine independent parts of context switching.
411 */
412 void
mi_switch(int flags,struct thread * newtd)413 mi_switch(int flags, struct thread *newtd)
414 {
415 uint64_t runtime, new_switchtime;
416 struct thread *td;
417
418 td = curthread; /* XXX */
419 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED);
420 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
421 #ifdef INVARIANTS
422 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
423 mtx_assert(&Giant, MA_NOTOWNED);
424 #endif
425 KASSERT(td->td_critnest == 1 || panicstr,
426 ("mi_switch: switch in a critical section"));
427 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
428 ("mi_switch: switch must be voluntary or involuntary"));
429 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
430
431 /*
432 * Don't perform context switches from the debugger.
433 */
434 if (kdb_active)
435 kdb_switch();
436 if (SCHEDULER_STOPPED())
437 return;
438 if (flags & SW_VOL) {
439 td->td_ru.ru_nvcsw++;
440 td->td_swvoltick = ticks;
441 } else
442 td->td_ru.ru_nivcsw++;
443 #ifdef SCHED_STATS
444 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]);
445 #endif
446 /*
447 * Do the context switch callback before blocking.
448 */
449 if (td->td_cswitchcb != NULL)
450 (*td->td_cswitchcb)(SWCB_BLOCK, td);
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 #if (KTR_COMPILE & KTR_SCHED) != 0
466 if (TD_IS_IDLETHREAD(td))
467 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle",
468 "prio:%d", td->td_priority);
469 else
470 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td),
471 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg,
472 "lockname:\"%s\"", td->td_lockname);
473 #endif
474 SDT_PROBE0(sched, , , preempt);
475 sched_switch(td, newtd, flags);
476 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running",
477 "prio:%d", td->td_priority);
478
479 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)",
480 td->td_tid, td->td_sched, td->td_proc->p_pid, td->td_name);
481
482 /*
483 * Do the context switch callback for unblocking.
484 */
485 if (td->td_cswitchcb != NULL)
486 (*td->td_cswitchcb)(SWCB_UNBLOCK, td);
487 /*
488 * If the last thread was exiting, finish cleaning it up.
489 */
490 if ((td = PCPU_GET(deadthread))) {
491 PCPU_SET(deadthread, NULL);
492 thread_stash(td);
493 }
494 }
495
496 /*
497 * Change thread state to be runnable, placing it on the run queue if
498 * it is in memory. If it is swapped out, return true so our caller
499 * will know to awaken the swapper.
500 */
501 int
setrunnable(struct thread * td)502 setrunnable(struct thread *td)
503 {
504
505 THREAD_LOCK_ASSERT(td, MA_OWNED);
506 KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
507 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
508 switch (td->td_state) {
509 case TDS_RUNNING:
510 case TDS_RUNQ:
511 return (0);
512 case TDS_INHIBITED:
513 /*
514 * If we are only inhibited because we are swapped out
515 * then arange to swap in this process. Otherwise just return.
516 */
517 if (td->td_inhibitors != TDI_SWAPPED)
518 return (0);
519 /* FALLTHROUGH */
520 case TDS_CAN_RUN:
521 break;
522 default:
523 printf("state is 0x%x", td->td_state);
524 panic("setrunnable(2)");
525 }
526 if ((td->td_flags & TDF_INMEM) == 0) {
527 if ((td->td_flags & TDF_SWAPINREQ) == 0) {
528 td->td_flags |= TDF_SWAPINREQ;
529 return (1);
530 }
531 } else
532 sched_wakeup(td);
533 return (0);
534 }
535
536 /*
537 * Compute a tenex style load average of a quantity on
538 * 1, 5 and 15 minute intervals.
539 */
540 static void
loadav(void * arg)541 loadav(void *arg)
542 {
543 int i, nrun;
544 struct loadavg *avg;
545
546 nrun = sched_load();
547 avg = &averunnable;
548
549 for (i = 0; i < 3; i++)
550 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
551 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
552
553 /*
554 * Schedule the next update to occur after 5 seconds, but add a
555 * random variation to avoid synchronisation with processes that
556 * run at regular intervals.
557 */
558 callout_reset_sbt(&loadav_callout,
559 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US,
560 loadav, NULL, C_DIRECT_EXEC | C_PREL(32));
561 }
562
563 /* ARGSUSED */
564 static void
synch_setup(void * dummy)565 synch_setup(void *dummy)
566 {
567 callout_init(&loadav_callout, 1);
568
569 /* Kick off timeout driven events by calling first time. */
570 loadav(NULL);
571 }
572
573 int
should_yield(void)574 should_yield(void)
575 {
576
577 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks);
578 }
579
580 void
maybe_yield(void)581 maybe_yield(void)
582 {
583
584 if (should_yield())
585 kern_yield(PRI_USER);
586 }
587
588 void
kern_yield(int prio)589 kern_yield(int prio)
590 {
591 struct thread *td;
592
593 td = curthread;
594 DROP_GIANT();
595 thread_lock(td);
596 if (prio == PRI_USER)
597 prio = td->td_user_pri;
598 if (prio >= 0)
599 sched_prio(td, prio);
600 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
601 thread_unlock(td);
602 PICKUP_GIANT();
603 }
604
605 /*
606 * General purpose yield system call.
607 */
608 int
sys_yield(struct thread * td,struct yield_args * uap)609 sys_yield(struct thread *td, struct yield_args *uap)
610 {
611
612 thread_lock(td);
613 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE)
614 sched_prio(td, PRI_MAX_TIMESHARE);
615 mi_switch(SW_VOL | SWT_RELINQUISH, NULL);
616 thread_unlock(td);
617 td->td_retval[0] = 0;
618 return (0);
619 }
620