xref: /freebsd/sys/kern/kern_synch.c (revision 6af83ee0d2941d18880b6aaa2b4facd1d30c6106)
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, sig, 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 	sleepq_lock(ident);
168 	if (catch) {
169 		/*
170 		 * Don't bother sleeping if we are exiting and not the exiting
171 		 * thread or if our thread is marked as interrupted.
172 		 */
173 		mtx_lock_spin(&sched_lock);
174 		rval = thread_sleep_check(td);
175 		mtx_unlock_spin(&sched_lock);
176 		if (rval != 0) {
177 			sleepq_release(ident);
178 			if (mtx != NULL && priority & PDROP)
179 				mtx_unlock(mtx);
180 			return (rval);
181 		}
182 	}
183 	CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)",
184 	    (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident);
185 
186 	DROP_GIANT();
187 	if (mtx != NULL) {
188 		mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED);
189 		WITNESS_SAVE(&mtx->mtx_object, mtx);
190 		mtx_unlock(mtx);
191 	}
192 
193 	/*
194 	 * We put ourselves on the sleep queue and start our timeout
195 	 * before calling thread_suspend_check, as we could stop there,
196 	 * and a wakeup or a SIGCONT (or both) could occur while we were
197 	 * stopped without resuming us.  Thus, we must be ready for sleep
198 	 * when cursig() is called.  If the wakeup happens while we're
199 	 * stopped, then td will no longer be on a sleep queue upon
200 	 * return from cursig().
201 	 */
202 	flags = SLEEPQ_MSLEEP;
203 	if (catch)
204 		flags |= SLEEPQ_INTERRUPTIBLE;
205 	sleepq_add(ident, mtx, wmesg, flags);
206 	if (timo)
207 		sleepq_set_timeout(ident, timo);
208 	if (catch) {
209 		sig = sleepq_catch_signals(ident);
210 	} else
211 		sig = 0;
212 
213 	/*
214 	 * Adjust this thread's priority.
215 	 */
216 	mtx_lock_spin(&sched_lock);
217 	sched_prio(td, priority & PRIMASK);
218 	mtx_unlock_spin(&sched_lock);
219 
220 	if (timo && catch)
221 		rval = sleepq_timedwait_sig(ident, sig != 0);
222 	else if (timo)
223 		rval = sleepq_timedwait(ident);
224 	else if (catch)
225 		rval = sleepq_wait_sig(ident);
226 	else {
227 		sleepq_wait(ident);
228 		rval = 0;
229 	}
230 	if (rval == 0 && catch)
231 		rval = sleepq_calc_signal_retval(sig);
232 #ifdef KTRACE
233 	if (KTRPOINT(td, KTR_CSW))
234 		ktrcsw(0, 0);
235 #endif
236 	PICKUP_GIANT();
237 	if (mtx != NULL && !(priority & PDROP)) {
238 		mtx_lock(mtx);
239 		WITNESS_RESTORE(&mtx->mtx_object, mtx);
240 	}
241 	return (rval);
242 }
243 
244 /*
245  * Make all threads sleeping on the specified identifier runnable.
246  */
247 void
248 wakeup(ident)
249 	register void *ident;
250 {
251 
252 	sleepq_lock(ident);
253 	sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1);
254 }
255 
256 /*
257  * Make a thread sleeping on the specified identifier runnable.
258  * May wake more than one thread if a target thread is currently
259  * swapped out.
260  */
261 void
262 wakeup_one(ident)
263 	register void *ident;
264 {
265 
266 	sleepq_lock(ident);
267 	sleepq_signal(ident, SLEEPQ_MSLEEP, -1);
268 }
269 
270 /*
271  * The machine independent parts of context switching.
272  */
273 void
274 mi_switch(int flags, struct thread *newtd)
275 {
276 	struct bintime new_switchtime;
277 	struct thread *td;
278 	struct proc *p;
279 
280 	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
281 	td = curthread;			/* XXX */
282 	p = td->td_proc;		/* XXX */
283 	KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
284 #ifdef INVARIANTS
285 	if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
286 		mtx_assert(&Giant, MA_NOTOWNED);
287 #endif
288 	KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
289 	    (td->td_pflags & TDP_OWEPREEMPT) != 0 && (flags & SW_INVOL) != 0 &&
290 	    newtd == NULL),
291 	    ("mi_switch: switch in a critical section"));
292 	KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
293 	    ("mi_switch: switch must be voluntary or involuntary"));
294 	KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
295 
296 	if (flags & SW_VOL)
297 		p->p_stats->p_ru.ru_nvcsw++;
298 	else
299 		p->p_stats->p_ru.ru_nivcsw++;
300 
301 	/*
302 	 * Compute the amount of time during which the current
303 	 * process was running, and add that to its total so far.
304 	 */
305 	binuptime(&new_switchtime);
306 	bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
307 	bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
308 
309 	td->td_generation++;	/* bump preempt-detect counter */
310 
311 	/*
312 	 * Don't perform context switches from the debugger.
313 	 */
314 	if (kdb_active) {
315 		mtx_unlock_spin(&sched_lock);
316 		kdb_backtrace();
317 		kdb_reenter();
318 		panic("%s: did not reenter debugger", __func__);
319 	}
320 
321 	/*
322 	 * Check if the process exceeds its cpu resource allocation.  If
323 	 * over max, arrange to kill the process in ast().
324 	 */
325 	if (p->p_cpulimit != RLIM_INFINITY &&
326 	    p->p_rux.rux_runtime.sec > p->p_cpulimit) {
327 		p->p_sflag |= PS_XCPU;
328 		td->td_flags |= TDF_ASTPENDING;
329 	}
330 
331 	/*
332 	 * Finish up stats for outgoing thread.
333 	 */
334 	cnt.v_swtch++;
335 	PCPU_SET(switchtime, new_switchtime);
336 	PCPU_SET(switchticks, ticks);
337 	CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)",
338 	    (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
339 	if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
340 		newtd = thread_switchout(td, flags, newtd);
341 #if (KTR_COMPILE & KTR_SCHED) != 0
342 	if (td == PCPU_GET(idlethread))
343 		CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",
344 		    td, td->td_proc->p_comm, td->td_priority);
345 	else if (newtd != NULL)
346 		CTR5(KTR_SCHED,
347 		    "mi_switch: %p(%s) prio %d preempted by %p(%s)",
348 		    td, td->td_proc->p_comm, td->td_priority, newtd,
349 		    newtd->td_proc->p_comm);
350 	else
351 		CTR6(KTR_SCHED,
352 		    "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s",
353 		    td, td->td_proc->p_comm, td->td_priority,
354 		    td->td_inhibitors, td->td_wmesg, td->td_lockname);
355 #endif
356 	sched_switch(td, newtd, flags);
357 	CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d",
358 	    td, td->td_proc->p_comm, td->td_priority);
359 
360 	CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)",
361 	    (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
362 
363 	/*
364 	 * If the last thread was exiting, finish cleaning it up.
365 	 */
366 	if ((td = PCPU_GET(deadthread))) {
367 		PCPU_SET(deadthread, NULL);
368 		thread_stash(td);
369 	}
370 }
371 
372 /*
373  * Change process state to be runnable,
374  * placing it on the run queue if it is in memory,
375  * and awakening the swapper if it isn't in memory.
376  */
377 void
378 setrunnable(struct thread *td)
379 {
380 	struct proc *p;
381 
382 	p = td->td_proc;
383 	mtx_assert(&sched_lock, MA_OWNED);
384 	switch (p->p_state) {
385 	case PRS_ZOMBIE:
386 		panic("setrunnable(1)");
387 	default:
388 		break;
389 	}
390 	switch (td->td_state) {
391 	case TDS_RUNNING:
392 	case TDS_RUNQ:
393 		return;
394 	case TDS_INHIBITED:
395 		/*
396 		 * If we are only inhibited because we are swapped out
397 		 * then arange to swap in this process. Otherwise just return.
398 		 */
399 		if (td->td_inhibitors != TDI_SWAPPED)
400 			return;
401 		/* XXX: intentional fall-through ? */
402 	case TDS_CAN_RUN:
403 		break;
404 	default:
405 		printf("state is 0x%x", td->td_state);
406 		panic("setrunnable(2)");
407 	}
408 	if ((p->p_sflag & PS_INMEM) == 0) {
409 		if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
410 			p->p_sflag |= PS_SWAPINREQ;
411 			/*
412 			 * due to a LOR between sched_lock and
413 			 * the sleepqueue chain locks, delay
414 			 * wakeup proc0 until thread leaves
415 			 * critical region.
416 			 */
417 			curthread->td_pflags |= TDP_WAKEPROC0;
418 		}
419 	} else
420 		sched_wakeup(td);
421 }
422 
423 /*
424  * Compute a tenex style load average of a quantity on
425  * 1, 5 and 15 minute intervals.
426  * XXXKSE   Needs complete rewrite when correct info is available.
427  * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
428  */
429 static void
430 loadav(void *arg)
431 {
432 	int i, nrun;
433 	struct loadavg *avg;
434 
435 	nrun = sched_load();
436 	avg = &averunnable;
437 
438 	for (i = 0; i < 3; i++)
439 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
440 		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
441 
442 	/*
443 	 * Schedule the next update to occur after 5 seconds, but add a
444 	 * random variation to avoid synchronisation with processes that
445 	 * run at regular intervals.
446 	 */
447 	callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
448 	    loadav, NULL);
449 }
450 
451 static void
452 lboltcb(void *arg)
453 {
454 	wakeup(&lbolt);
455 	callout_reset(&lbolt_callout, hz, lboltcb, NULL);
456 }
457 
458 /* ARGSUSED */
459 static void
460 synch_setup(dummy)
461 	void *dummy;
462 {
463 	callout_init(&loadav_callout, CALLOUT_MPSAFE);
464 	callout_init(&lbolt_callout, CALLOUT_MPSAFE);
465 
466 	/* Kick off timeout driven events by calling first time. */
467 	loadav(NULL);
468 	lboltcb(NULL);
469 }
470 
471 /*
472  * General purpose yield system call
473  */
474 int
475 yield(struct thread *td, struct yield_args *uap)
476 {
477 	struct ksegrp *kg;
478 
479 	kg = td->td_ksegrp;
480 	mtx_assert(&Giant, MA_NOTOWNED);
481 	mtx_lock_spin(&sched_lock);
482 	sched_prio(td, PRI_MAX_TIMESHARE);
483 	mi_switch(SW_VOL, NULL);
484 	mtx_unlock_spin(&sched_lock);
485 	td->td_retval[0] = 0;
486 	return (0);
487 }
488