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("$FreeBSD$");
39
40 #include "opt_ktrace.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/condvar.h>
45 #include <sys/kdb.h>
46 #include <sys/kernel.h>
47 #include <sys/ktr.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/proc.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sched.h>
53 #include <sys/signalvar.h>
54 #include <sys/sleepqueue.h>
55 #include <sys/smp.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysproto.h>
59 #include <sys/vmmeter.h>
60 #ifdef KTRACE
61 #include <sys/uio.h>
62 #include <sys/ktrace.h>
63 #endif
64
65 #include <machine/cpu.h>
66
67 static void synch_setup(void *dummy);
68 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL)
69
70 int hogticks;
71 int lbolt;
72
73 static struct callout loadav_callout;
74 static struct callout lbolt_callout;
75
76 struct loadavg averunnable =
77 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
78 /*
79 * Constants for averages over 1, 5, and 15 minutes
80 * when sampling at 5 second intervals.
81 */
82 static fixpt_t cexp[3] = {
83 0.9200444146293232 * FSCALE, /* exp(-1/12) */
84 0.9834714538216174 * FSCALE, /* exp(-1/60) */
85 0.9944598480048967 * FSCALE, /* exp(-1/180) */
86 };
87
88 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */
89 static int fscale __unused = FSCALE;
90 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
91
92 static void loadav(void *arg);
93 static void lboltcb(void *arg);
94
95 void
96 sleepinit(void)
97 {
98
99 hogticks = (hz / 10) * 2; /* Default only. */
100 init_sleepqueues();
101 }
102
103 /*
104 * General sleep call. Suspends the current process until a wakeup is
105 * performed on the specified identifier. The process will then be made
106 * runnable with the specified priority. Sleeps at most timo/hz seconds
107 * (0 means no timeout). If pri includes PCATCH flag, signals are checked
108 * before and after sleeping, else signals are not checked. Returns 0 if
109 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
110 * signal needs to be delivered, ERESTART is returned if the current system
111 * call should be restarted if possible, and EINTR is returned if the system
112 * call should be interrupted by the signal (return EINTR).
113 *
114 * The mutex argument is exited before the caller is suspended, and
115 * entered before msleep returns. If priority includes the PDROP
116 * flag the mutex is not entered before returning.
117 */
118 int
119 msleep(ident, mtx, priority, wmesg, timo)
120 void *ident;
121 struct mtx *mtx;
122 int priority, timo;
123 const char *wmesg;
124 {
125 struct thread *td;
126 struct proc *p;
127 int catch, rval, flags;
128 WITNESS_SAVE_DECL(mtx);
129
130 td = curthread;
131 p = td->td_proc;
132 #ifdef KTRACE
133 if (KTRPOINT(td, KTR_CSW))
134 ktrcsw(1, 0);
135 #endif
136 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL :
137 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg);
138 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL,
139 ("sleeping without a mutex"));
140 KASSERT(p != NULL, ("msleep1"));
141 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
142
143 if (cold) {
144 /*
145 * During autoconfiguration, just return;
146 * don't run any other threads or panic below,
147 * in case this is the idle thread and already asleep.
148 * XXX: this used to do "s = splhigh(); splx(safepri);
149 * splx(s);" to give interrupts a chance, but there is
150 * no way to give interrupts a chance now.
151 */
152 if (mtx != NULL && priority & PDROP)
153 mtx_unlock(mtx);
154 return (0);
155 }
156 catch = priority & PCATCH;
157 rval = 0;
158
159 /*
160 * If we are already on a sleep queue, then remove us from that
161 * sleep queue first. We have to do this to handle recursive
162 * sleeps.
163 */
164 if (TD_ON_SLEEPQ(td))
165 sleepq_remove(td, td->td_wchan);
166
167 flags = SLEEPQ_MSLEEP;
168 if (catch)
169 flags |= SLEEPQ_INTERRUPTIBLE;
170
171 sleepq_lock(ident);
172 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)",
173 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident);
174
175 DROP_GIANT();
176 if (mtx != NULL) {
177 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
178 WITNESS_SAVE(&mtx->mtx_object, mtx);
179 mtx_unlock(mtx);
180 }
181
182 /*
183 * We put ourselves on the sleep queue and start our timeout
184 * before calling thread_suspend_check, as we could stop there,
185 * and a wakeup or a SIGCONT (or both) could occur while we were
186 * stopped without resuming us. Thus, we must be ready for sleep
187 * when cursig() is called. If the wakeup happens while we're
188 * stopped, then td will no longer be on a sleep queue upon
189 * return from cursig().
190 */
191 sleepq_add(ident, &mtx->mtx_object, wmesg, flags, 0);
192 if (timo)
193 sleepq_set_timeout(ident, timo);
194
195 /*
196 * Adjust this thread's priority.
197 */
198 if ((priority & PRIMASK) != 0) {
199 mtx_lock_spin(&sched_lock);
200 sched_prio(td, priority & PRIMASK);
201 mtx_unlock_spin(&sched_lock);
202 }
203
204 if (timo && catch)
205 rval = sleepq_timedwait_sig(ident);
206 else if (timo)
207 rval = sleepq_timedwait(ident);
208 else if (catch)
209 rval = sleepq_wait_sig(ident);
210 else {
211 sleepq_wait(ident);
212 rval = 0;
213 }
214 #ifdef KTRACE
215 if (KTRPOINT(td, KTR_CSW))
216 ktrcsw(0, 0);
217 #endif
218 PICKUP_GIANT();
219 if (mtx != NULL && !(priority & PDROP)) {
220 mtx_lock(mtx);
221 WITNESS_RESTORE(&mtx->mtx_object, mtx);
222 }
223 return (rval);
224 }
225
226 int
227 msleep_spin(ident, mtx, wmesg, timo)
228 void *ident;
229 struct mtx *mtx;
230 const char *wmesg;
231 int timo;
232 {
233 struct thread *td;
234 struct proc *p;
235 int rval;
236 WITNESS_SAVE_DECL(mtx);
237
238 td = curthread;
239 p = td->td_proc;
240 KASSERT(mtx != NULL, ("sleeping without a mutex"));
241 KASSERT(p != NULL, ("msleep1"));
242 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep"));
243
244 if (cold) {
245 /*
246 * During autoconfiguration, just return;
247 * don't run any other threads or panic below,
248 * in case this is the idle thread and already asleep.
249 * XXX: this used to do "s = splhigh(); splx(safepri);
250 * splx(s);" to give interrupts a chance, but there is
251 * no way to give interrupts a chance now.
252 */
253 return (0);
254 }
255
256 sleepq_lock(ident);
257 CTR5(KTR_PROC, "msleep_spin: thread %p (pid %ld, %s) on %s (%p)",
258 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident);
259
260 DROP_GIANT();
261 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
262 WITNESS_SAVE(&mtx->mtx_object, mtx);
263 mtx_unlock_spin(mtx);
264
265 /*
266 * We put ourselves on the sleep queue and start our timeout.
267 */
268 sleepq_add(ident, &mtx->mtx_object, wmesg, SLEEPQ_MSLEEP, 0);
269 if (timo)
270 sleepq_set_timeout(ident, timo);
271
272 /*
273 * Can't call ktrace with any spin locks held so it can lock the
274 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold
275 * any spin lock. Thus, we have to drop the sleepq spin lock while
276 * we handle those requests. This is safe since we have placed our
277 * thread on the sleep queue already.
278 */
279 #ifdef KTRACE
280 if (KTRPOINT(td, KTR_CSW)) {
281 sleepq_release(ident);
282 ktrcsw(1, 0);
283 sleepq_lock(ident);
284 }
285 #endif
286 #ifdef WITNESS
287 sleepq_release(ident);
288 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"",
289 wmesg);
290 sleepq_lock(ident);
291 #endif
292 if (timo)
293 rval = sleepq_timedwait(ident);
294 else {
295 sleepq_wait(ident);
296 rval = 0;
297 }
298 #ifdef KTRACE
299 if (KTRPOINT(td, KTR_CSW))
300 ktrcsw(0, 0);
301 #endif
302 PICKUP_GIANT();
303 mtx_lock_spin(mtx);
304 WITNESS_RESTORE(&mtx->mtx_object, mtx);
305 return (rval);
306 }
307
308 /*
309 * Make all threads sleeping on the specified identifier runnable.
310 */
311 void
312 wakeup(ident)
313 register void *ident;
314 {
315
316 sleepq_lock(ident);
317 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1, 0);
318 }
319
320 /*
321 * Make a thread sleeping on the specified identifier runnable.
322 * May wake more than one thread if a target thread is currently
323 * swapped out.
324 */
325 void
326 wakeup_one(ident)
327 register void *ident;
328 {
329
330 sleepq_lock(ident);
331 sleepq_signal(ident, SLEEPQ_MSLEEP, -1, 0);
332 }
333
334 /*
335 * The machine independent parts of context switching.
336 */
337 void
338 mi_switch(int flags, struct thread *newtd)
339 {
340 struct bintime new_switchtime;
341 struct thread *td;
342 struct proc *p;
343
344 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
345 td = curthread; /* XXX */
346 p = td->td_proc; /* XXX */
347 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
348 #ifdef INVARIANTS
349 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
350 mtx_assert(&Giant, MA_NOTOWNED);
351 #endif
352 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
353 (td->td_owepreempt) && (flags & SW_INVOL) != 0 &&
354 newtd == NULL) || panicstr,
355 ("mi_switch: switch in a critical section"));
356 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
357 ("mi_switch: switch must be voluntary or involuntary"));
358 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
359
360 if (flags & SW_VOL)
361 p->p_stats->p_ru.ru_nvcsw++;
362 else
363 p->p_stats->p_ru.ru_nivcsw++;
364
365 /*
366 * Compute the amount of time during which the current
367 * process was running, and add that to its total so far.
368 */
369 binuptime(&new_switchtime);
370 bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
371 bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
372
373 td->td_generation++; /* bump preempt-detect counter */
374
375 /*
376 * Don't perform context switches from the debugger.
377 */
378 if (kdb_active) {
379 mtx_unlock_spin(&sched_lock);
380 kdb_backtrace();
381 kdb_reenter();
382 panic("%s: did not reenter debugger", __func__);
383 }
384
385 /*
386 * Check if the process exceeds its cpu resource allocation. If
387 * it reaches the max, arrange to kill the process in ast().
388 */
389 if (p->p_cpulimit != RLIM_INFINITY &&
390 p->p_rux.rux_runtime.sec >= p->p_cpulimit) {
391 p->p_sflag |= PS_XCPU;
392 td->td_flags |= TDF_ASTPENDING;
393 }
394
395 /*
396 * Finish up stats for outgoing thread.
397 */
398 cnt.v_swtch++;
399 PCPU_SET(switchtime, new_switchtime);
400 PCPU_SET(switchticks, ticks);
401 CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)",
402 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
403 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
404 newtd = thread_switchout(td, flags, newtd);
405 #if (KTR_COMPILE & KTR_SCHED) != 0
406 if (td == PCPU_GET(idlethread))
407 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",
408 td, td->td_proc->p_comm, td->td_priority);
409 else if (newtd != NULL)
410 CTR5(KTR_SCHED,
411 "mi_switch: %p(%s) prio %d preempted by %p(%s)",
412 td, td->td_proc->p_comm, td->td_priority, newtd,
413 newtd->td_proc->p_comm);
414 else
415 CTR6(KTR_SCHED,
416 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s",
417 td, td->td_proc->p_comm, td->td_priority,
418 td->td_inhibitors, td->td_wmesg, td->td_lockname);
419 #endif
420 sched_switch(td, newtd, flags);
421 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d",
422 td, td->td_proc->p_comm, td->td_priority);
423
424 CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)",
425 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
426
427 /*
428 * If the last thread was exiting, finish cleaning it up.
429 */
430 if ((td = PCPU_GET(deadthread))) {
431 PCPU_SET(deadthread, NULL);
432 thread_stash(td);
433 }
434 }
435
436 /*
437 * Change process state to be runnable,
438 * placing it on the run queue if it is in memory,
439 * and awakening the swapper if it isn't in memory.
440 */
441 void
442 setrunnable(struct thread *td)
443 {
444 struct proc *p;
445
446 p = td->td_proc;
447 mtx_assert(&sched_lock, MA_OWNED);
448 switch (p->p_state) {
449 case PRS_ZOMBIE:
450 panic("setrunnable(1)");
451 default:
452 break;
453 }
454 switch (td->td_state) {
455 case TDS_RUNNING:
456 case TDS_RUNQ:
457 return;
458 case TDS_INHIBITED:
459 /*
460 * If we are only inhibited because we are swapped out
461 * then arange to swap in this process. Otherwise just return.
462 */
463 if (td->td_inhibitors != TDI_SWAPPED)
464 return;
465 /* XXX: intentional fall-through ? */
466 case TDS_CAN_RUN:
467 break;
468 default:
469 printf("state is 0x%x", td->td_state);
470 panic("setrunnable(2)");
471 }
472 if ((p->p_sflag & PS_INMEM) == 0) {
473 if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
474 p->p_sflag |= PS_SWAPINREQ;
475 /*
476 * due to a LOR between sched_lock and
477 * the sleepqueue chain locks, use
478 * lower level scheduling functions.
479 */
480 kick_proc0();
481 }
482 } else
483 sched_wakeup(td);
484 }
485
486 /*
487 * Compute a tenex style load average of a quantity on
488 * 1, 5 and 15 minute intervals.
489 * XXXKSE Needs complete rewrite when correct info is available.
490 * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
491 */
492 static void
493 loadav(void *arg)
494 {
495 int i, nrun;
496 struct loadavg *avg;
497
498 nrun = sched_load();
499 avg = &averunnable;
500
501 for (i = 0; i < 3; i++)
502 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
503 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
504
505 /*
506 * Schedule the next update to occur after 5 seconds, but add a
507 * random variation to avoid synchronisation with processes that
508 * run at regular intervals.
509 */
510 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
511 loadav, NULL);
512 }
513
514 static void
515 lboltcb(void *arg)
516 {
517 wakeup(&lbolt);
518 callout_reset(&lbolt_callout, hz, lboltcb, NULL);
519 }
520
521 /* ARGSUSED */
522 static void
523 synch_setup(dummy)
524 void *dummy;
525 {
526 callout_init(&loadav_callout, CALLOUT_MPSAFE);
527 callout_init(&lbolt_callout, CALLOUT_MPSAFE);
528
529 /* Kick off timeout driven events by calling first time. */
530 loadav(NULL);
531 lboltcb(NULL);
532 }
533
534 /*
535 * General purpose yield system call
536 */
537 int
538 yield(struct thread *td, struct yield_args *uap)
539 {
540 struct ksegrp *kg;
541
542 kg = td->td_ksegrp;
543 mtx_assert(&Giant, MA_NOTOWNED);
544 mtx_lock_spin(&sched_lock);
545 sched_prio(td, PRI_MAX_TIMESHARE);
546 mi_switch(SW_VOL, NULL);
547 mtx_unlock_spin(&sched_lock);
548 td->td_retval[0] = 0;
549 return (0);
550 }
Cache object: cbf33e5b76b28309b6259f3f922b884a
|