xref: /freebsd/sys/kern/kern_synch.c (revision 63f9a4cb2684a303e3eb2ffed39c03a2e2b28ae0)
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 	 * XXX: do we need to save priority in td_base_pri?
217 	 */
218 	mtx_lock_spin(&sched_lock);
219 	sched_prio(td, priority & PRIMASK);
220 	mtx_unlock_spin(&sched_lock);
221 
222 	if (timo && catch)
223 		rval = sleepq_timedwait_sig(ident, sig != 0);
224 	else if (timo)
225 		rval = sleepq_timedwait(ident);
226 	else if (catch)
227 		rval = sleepq_wait_sig(ident);
228 	else {
229 		sleepq_wait(ident);
230 		rval = 0;
231 	}
232 	if (rval == 0 && catch)
233 		rval = sleepq_calc_signal_retval(sig);
234 #ifdef KTRACE
235 	if (KTRPOINT(td, KTR_CSW))
236 		ktrcsw(0, 0);
237 #endif
238 	PICKUP_GIANT();
239 	if (mtx != NULL && !(priority & PDROP)) {
240 		mtx_lock(mtx);
241 		WITNESS_RESTORE(&mtx->mtx_object, mtx);
242 	}
243 	return (rval);
244 }
245 
246 /*
247  * Make all threads sleeping on the specified identifier runnable.
248  */
249 void
250 wakeup(ident)
251 	register void *ident;
252 {
253 
254 	sleepq_lock(ident);
255 	sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1);
256 }
257 
258 /*
259  * Make a thread sleeping on the specified identifier runnable.
260  * May wake more than one thread if a target thread is currently
261  * swapped out.
262  */
263 void
264 wakeup_one(ident)
265 	register void *ident;
266 {
267 
268 	sleepq_lock(ident);
269 	sleepq_signal(ident, SLEEPQ_MSLEEP, -1);
270 }
271 
272 /*
273  * The machine independent parts of context switching.
274  */
275 void
276 mi_switch(int flags, struct thread *newtd)
277 {
278 	struct bintime new_switchtime;
279 	struct thread *td;
280 	struct proc *p;
281 
282 	mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
283 	td = curthread;			/* XXX */
284 	p = td->td_proc;		/* XXX */
285 	KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code"));
286 #ifdef INVARIANTS
287 	if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td))
288 		mtx_assert(&Giant, MA_NOTOWNED);
289 #endif
290 	KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 &&
291 	    (td->td_pflags & TDP_OWEPREEMPT) != 0 && (flags & SW_INVOL) != 0 &&
292 	    newtd == NULL),
293 	    ("mi_switch: switch in a critical section"));
294 	KASSERT((flags & (SW_INVOL | SW_VOL)) != 0,
295 	    ("mi_switch: switch must be voluntary or involuntary"));
296 	KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself"));
297 
298 	if (flags & SW_VOL)
299 		p->p_stats->p_ru.ru_nvcsw++;
300 	else
301 		p->p_stats->p_ru.ru_nivcsw++;
302 
303 	/*
304 	 * Compute the amount of time during which the current
305 	 * process was running, and add that to its total so far.
306 	 */
307 	binuptime(&new_switchtime);
308 	bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
309 	bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
310 
311 	td->td_generation++;	/* bump preempt-detect counter */
312 
313 	/*
314 	 * Don't perform context switches from the debugger.
315 	 */
316 	if (kdb_active) {
317 		mtx_unlock_spin(&sched_lock);
318 		kdb_backtrace();
319 		kdb_reenter();
320 		panic("%s: did not reenter debugger", __func__);
321 	}
322 
323 	/*
324 	 * Check if the process exceeds its cpu resource allocation.  If
325 	 * over max, arrange to kill the process in ast().
326 	 */
327 	if (p->p_cpulimit != RLIM_INFINITY &&
328 	    p->p_rux.rux_runtime.sec > p->p_cpulimit) {
329 		p->p_sflag |= PS_XCPU;
330 		td->td_flags |= TDF_ASTPENDING;
331 	}
332 
333 	/*
334 	 * Finish up stats for outgoing thread.
335 	 */
336 	cnt.v_swtch++;
337 	PCPU_SET(switchtime, new_switchtime);
338 	PCPU_SET(switchticks, ticks);
339 	CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)",
340 	    (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
341 	if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
342 		newtd = thread_switchout(td, flags, newtd);
343 	sched_switch(td, newtd, flags);
344 
345 	CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)",
346 	    (void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
347 
348 	/*
349 	 * If the last thread was exiting, finish cleaning it up.
350 	 */
351 	if ((td = PCPU_GET(deadthread))) {
352 		PCPU_SET(deadthread, NULL);
353 		thread_stash(td);
354 	}
355 }
356 
357 /*
358  * Change process state to be runnable,
359  * placing it on the run queue if it is in memory,
360  * and awakening the swapper if it isn't in memory.
361  */
362 void
363 setrunnable(struct thread *td)
364 {
365 	struct proc *p;
366 
367 	p = td->td_proc;
368 	mtx_assert(&sched_lock, MA_OWNED);
369 	switch (p->p_state) {
370 	case PRS_ZOMBIE:
371 		panic("setrunnable(1)");
372 	default:
373 		break;
374 	}
375 	switch (td->td_state) {
376 	case TDS_RUNNING:
377 	case TDS_RUNQ:
378 		return;
379 	case TDS_INHIBITED:
380 		/*
381 		 * If we are only inhibited because we are swapped out
382 		 * then arange to swap in this process. Otherwise just return.
383 		 */
384 		if (td->td_inhibitors != TDI_SWAPPED)
385 			return;
386 		/* XXX: intentional fall-through ? */
387 	case TDS_CAN_RUN:
388 		break;
389 	default:
390 		printf("state is 0x%x", td->td_state);
391 		panic("setrunnable(2)");
392 	}
393 	if ((p->p_sflag & PS_INMEM) == 0) {
394 		if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
395 			p->p_sflag |= PS_SWAPINREQ;
396 			/*
397 			 * due to a LOR between sched_lock and
398 			 * the sleepqueue chain locks, delay
399 			 * wakeup proc0 until thread leaves
400 			 * critical region.
401 			 */
402 			curthread->td_pflags |= TDP_WAKEPROC0;
403 		}
404 	} else
405 		sched_wakeup(td);
406 }
407 
408 /*
409  * Compute a tenex style load average of a quantity on
410  * 1, 5 and 15 minute intervals.
411  * XXXKSE   Needs complete rewrite when correct info is available.
412  * Completely Bogus.. only works with 1:1 (but compiles ok now :-)
413  */
414 static void
415 loadav(void *arg)
416 {
417 	int i, nrun;
418 	struct loadavg *avg;
419 
420 	nrun = sched_load();
421 	avg = &averunnable;
422 
423 	for (i = 0; i < 3; i++)
424 		avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
425 		    nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
426 
427 	/*
428 	 * Schedule the next update to occur after 5 seconds, but add a
429 	 * random variation to avoid synchronisation with processes that
430 	 * run at regular intervals.
431 	 */
432 	callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)),
433 	    loadav, NULL);
434 }
435 
436 static void
437 lboltcb(void *arg)
438 {
439 	wakeup(&lbolt);
440 	callout_reset(&lbolt_callout, hz, lboltcb, NULL);
441 }
442 
443 /* ARGSUSED */
444 static void
445 synch_setup(dummy)
446 	void *dummy;
447 {
448 	callout_init(&loadav_callout, CALLOUT_MPSAFE);
449 	callout_init(&lbolt_callout, CALLOUT_MPSAFE);
450 
451 	/* Kick off timeout driven events by calling first time. */
452 	loadav(NULL);
453 	lboltcb(NULL);
454 }
455 
456 /*
457  * General purpose yield system call
458  */
459 int
460 yield(struct thread *td, struct yield_args *uap)
461 {
462 	struct ksegrp *kg;
463 
464 	kg = td->td_ksegrp;
465 	mtx_assert(&Giant, MA_NOTOWNED);
466 	mtx_lock_spin(&sched_lock);
467 	sched_prio(td, PRI_MAX_TIMESHARE);
468 	mi_switch(SW_VOL, NULL);
469 	mtx_unlock_spin(&sched_lock);
470 	td->td_retval[0] = 0;
471 	return (0);
472 }
473