kernel - Add per-process token, adjust signal code to use it.
[dragonfly.git] / sys / kern / kern_synch.c
CommitLineData
984263bc
MD
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 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95
39 * $FreeBSD: src/sys/kern/kern_synch.c,v 1.87.2.6 2002/10/13 07:29:53 kbyanc Exp $
c730be20 40 * $DragonFly: src/sys/kern/kern_synch.c,v 1.91 2008/09/09 04:06:13 dillon Exp $
984263bc
MD
41 */
42
43#include "opt_ktrace.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/proc.h>
48#include <sys/kernel.h>
49#include <sys/signalvar.h>
50#include <sys/resourcevar.h>
51#include <sys/vmmeter.h>
52#include <sys/sysctl.h>
344ad853 53#include <sys/lock.h>
984263bc 54#include <sys/uio.h>
fc9ae81d 55#ifdef KTRACE
984263bc
MD
56#include <sys/ktrace.h>
57#endif
f1d1c3fa 58#include <sys/xwait.h>
9afb0ffd 59#include <sys/ktr.h>
684a93c4 60#include <sys/serialize.h>
984263bc 61
684a93c4 62#include <sys/signal2.h>
bf765287
MD
63#include <sys/thread2.h>
64#include <sys/spinlock2.h>
7f6220a9 65#include <sys/mutex2.h>
684a93c4 66#include <sys/mplock2.h>
bf765287 67
984263bc 68#include <machine/cpu.h>
984263bc
MD
69#include <machine/smp.h>
70
fc17ad60
MD
71TAILQ_HEAD(tslpque, thread);
72
402ed7e1 73static void sched_setup (void *dummy);
984263bc
MD
74SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
75
984263bc
MD
76int hogticks;
77int lbolt;
344ad853 78int lbolt_syncer;
984263bc 79int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
17a9f566 80int ncpus;
da23a592
MD
81int ncpus2, ncpus2_shift, ncpus2_mask; /* note: mask not cpumask_t */
82int ncpus_fit, ncpus_fit_mask; /* note: mask not cpumask_t */
e43a034f 83int safepri;
dbcd0c9b 84int tsleep_now_works;
5ea440eb 85int tsleep_crypto_dump = 0;
984263bc
MD
86
87static struct callout loadav_callout;
35f9d051 88static struct callout schedcpu_callout;
fc17ad60 89MALLOC_DEFINE(M_TSLEEP, "tslpque", "tsleep queues");
984263bc 90
5decebc7
MD
91#define __DEALL(ident) __DEQUALIFY(void *, ident)
92
9afb0ffd
MD
93#if !defined(KTR_TSLEEP)
94#define KTR_TSLEEP KTR_ALL
95#endif
96KTR_INFO_MASTER(tsleep);
8aa3430c
MD
97KTR_INFO(KTR_TSLEEP, tsleep, tsleep_beg, 0, "tsleep enter %p", sizeof(void *));
98KTR_INFO(KTR_TSLEEP, tsleep, tsleep_end, 1, "tsleep exit", 0);
99KTR_INFO(KTR_TSLEEP, tsleep, wakeup_beg, 2, "wakeup enter %p", sizeof(void *));
100KTR_INFO(KTR_TSLEEP, tsleep, wakeup_end, 3, "wakeup exit", 0);
d9345d3a 101KTR_INFO(KTR_TSLEEP, tsleep, ilockfail, 4, "interlock failed %p", sizeof(void *));
8aa3430c
MD
102
103#define logtsleep1(name) KTR_LOG(tsleep_ ## name)
104#define logtsleep2(name, val) KTR_LOG(tsleep_ ## name, val)
9afb0ffd 105
984263bc
MD
106struct loadavg averunnable =
107 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
108/*
109 * Constants for averages over 1, 5, and 15 minutes
110 * when sampling at 5 second intervals.
111 */
112static fixpt_t cexp[3] = {
113 0.9200444146293232 * FSCALE, /* exp(-1/12) */
114 0.9834714538216174 * FSCALE, /* exp(-1/60) */
115 0.9944598480048967 * FSCALE, /* exp(-1/180) */
116};
117
402ed7e1
RG
118static void endtsleep (void *);
119static void loadav (void *arg);
402ed7e1 120static void schedcpu (void *arg);
ce4b5045 121#ifdef SMP
74c9628e 122static void tsleep_wakeup_remote(struct thread *td);
ce4b5045 123#endif
984263bc 124
a46fac56
MD
125/*
126 * Adjust the scheduler quantum. The quantum is specified in microseconds.
127 * Note that 'tick' is in microseconds per tick.
128 */
984263bc
MD
129static int
130sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
131{
132 int error, new_val;
133
a591f597 134 new_val = sched_quantum * ustick;
984263bc
MD
135 error = sysctl_handle_int(oidp, &new_val, 0, req);
136 if (error != 0 || req->newptr == NULL)
137 return (error);
a591f597 138 if (new_val < ustick)
984263bc 139 return (EINVAL);
a591f597 140 sched_quantum = new_val / ustick;
984263bc
MD
141 hogticks = 2 * sched_quantum;
142 return (0);
143}
144
145SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW,
146 0, sizeof sched_quantum, sysctl_kern_quantum, "I", "");
147
984263bc
MD
148/*
149 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
150 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
151 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
152 *
153 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
dcc99b62 154 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
984263bc
MD
155 *
156 * If you don't want to bother with the faster/more-accurate formula, you
157 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
158 * (more general) method of calculating the %age of CPU used by a process.
dcc99b62 159 *
08f2f1bb 160 * decay 95% of `lwp_pctcpu' in 60 seconds; see CCPU_SHIFT before changing
dcc99b62
MD
161 */
162#define CCPU_SHIFT 11
163
164static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
165SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
166
167/*
168 * kernel uses `FSCALE', userland (SHOULD) use kern.fscale
984263bc 169 */
460426e6 170int fscale __unused = FSCALE; /* exported to systat */
dcc99b62 171SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
984263bc
MD
172
173/*
0a3f9b47 174 * Recompute process priorities, once a second.
dcc99b62
MD
175 *
176 * Since the userland schedulers are typically event oriented, if the
177 * estcpu calculation at wakeup() time is not sufficient to make a
178 * process runnable relative to other processes in the system we have
179 * a 1-second recalc to help out.
180 *
181 * This code also allows us to store sysclock_t data in the process structure
182 * without fear of an overrun, since sysclock_t are guarenteed to hold
183 * several seconds worth of count.
8fa76237
MD
184 *
185 * WARNING! callouts can preempt normal threads. However, they will not
186 * preempt a thread holding a spinlock so we *can* safely use spinlocks.
984263bc 187 */
8fa76237
MD
188static int schedcpu_stats(struct proc *p, void *data __unused);
189static int schedcpu_resource(struct proc *p, void *data __unused);
190
984263bc 191static void
26a0694b 192schedcpu(void *arg)
984263bc 193{
8fa76237
MD
194 allproc_scan(schedcpu_stats, NULL);
195 allproc_scan(schedcpu_resource, NULL);
196 wakeup((caddr_t)&lbolt);
197 wakeup((caddr_t)&lbolt_syncer);
198 callout_reset(&schedcpu_callout, hz, schedcpu, NULL);
199}
200
201/*
202 * General process statistics once a second
203 */
204static int
205schedcpu_stats(struct proc *p, void *data __unused)
206{
08f2f1bb
SS
207 struct lwp *lp;
208
8fa76237
MD
209 crit_enter();
210 p->p_swtime++;
c7e98b2f
SS
211 FOREACH_LWP_IN_PROC(lp, p) {
212 if (lp->lwp_stat == LSSLEEP)
213 lp->lwp_slptime++;
4b5f931b 214
c7e98b2f
SS
215 /*
216 * Only recalculate processes that are active or have slept
217 * less then 2 seconds. The schedulers understand this.
218 */
219 if (lp->lwp_slptime <= 1) {
220 p->p_usched->recalculate(lp);
221 } else {
222 lp->lwp_pctcpu = (lp->lwp_pctcpu * ccpu) >> FSHIFT;
223 }
8fa76237
MD
224 }
225 crit_exit();
226 return(0);
227}
a46fac56 228
8fa76237 229/*
84204577 230 * Resource checks. XXX break out since ksignal/killproc can block,
8fa76237
MD
231 * limiting us to one process killed per second. There is probably
232 * a better way.
233 */
234static int
235schedcpu_resource(struct proc *p, void *data __unused)
236{
237 u_int64_t ttime;
08f2f1bb 238 struct lwp *lp;
8fa76237
MD
239
240 crit_enter();
241 if (p->p_stat == SIDL ||
416d05d7 242 p->p_stat == SZOMB ||
c7e98b2f 243 p->p_limit == NULL
8fa76237 244 ) {
e43a034f 245 crit_exit();
8fa76237 246 return(0);
984263bc 247 }
344ad853 248
c7e98b2f
SS
249 ttime = 0;
250 FOREACH_LWP_IN_PROC(lp, p) {
e595c6cd
MD
251 /*
252 * We may have caught an lp in the middle of being
253 * created, lwp_thread can be NULL.
254 */
255 if (lp->lwp_thread) {
256 ttime += lp->lwp_thread->td_sticks;
257 ttime += lp->lwp_thread->td_uticks;
258 }
c7e98b2f 259 }
8fa76237
MD
260
261 switch(plimit_testcpulimit(p->p_limit, ttime)) {
262 case PLIMIT_TESTCPU_KILL:
263 killproc(p, "exceeded maximum CPU limit");
264 break;
265 case PLIMIT_TESTCPU_XCPU:
266 if ((p->p_flag & P_XCPU) == 0) {
267 p->p_flag |= P_XCPU;
84204577 268 ksignal(p, SIGXCPU);
344ad853 269 }
8fa76237
MD
270 break;
271 default:
c0b8a06d 272 break;
344ad853 273 }
8fa76237
MD
274 crit_exit();
275 return(0);
984263bc
MD
276}
277
278/*
dcc99b62
MD
279 * This is only used by ps. Generate a cpu percentage use over
280 * a period of one second.
52eedfb5
MD
281 *
282 * MPSAFE
984263bc 283 */
dcc99b62 284void
553ea3c8 285updatepcpu(struct lwp *lp, int cpticks, int ttlticks)
984263bc 286{
dcc99b62
MD
287 fixpt_t acc;
288 int remticks;
289
290 acc = (cpticks << FSHIFT) / ttlticks;
291 if (ttlticks >= ESTCPUFREQ) {
553ea3c8 292 lp->lwp_pctcpu = acc;
dcc99b62
MD
293 } else {
294 remticks = ESTCPUFREQ - ttlticks;
553ea3c8 295 lp->lwp_pctcpu = (acc * ttlticks + lp->lwp_pctcpu * remticks) /
dcc99b62 296 ESTCPUFREQ;
a46fac56 297 }
984263bc
MD
298}
299
300/*
8aa3430c
MD
301 * tsleep/wakeup hash table parameters. Try to find the sweet spot for
302 * like addresses being slept on.
984263bc 303 */
8aa3430c
MD
304#define TABLESIZE 1024
305#define LOOKUP(x) (((intptr_t)(x) >> 6) & (TABLESIZE - 1))
984263bc 306
fc17ad60
MD
307static cpumask_t slpque_cpumasks[TABLESIZE];
308
984263bc 309/*
a46fac56 310 * General scheduler initialization. We force a reschedule 25 times
fc17ad60
MD
311 * a second by default. Note that cpu0 is initialized in early boot and
312 * cannot make any high level calls.
313 *
314 * Each cpu has its own sleep queue.
984263bc 315 */
984263bc 316void
fc17ad60 317sleep_gdinit(globaldata_t gd)
984263bc 318{
fc17ad60 319 static struct tslpque slpque_cpu0[TABLESIZE];
9c1fad94 320 int i;
984263bc 321
fc17ad60
MD
322 if (gd->gd_cpuid == 0) {
323 sched_quantum = (hz + 24) / 25;
324 hogticks = 2 * sched_quantum;
325
326 gd->gd_tsleep_hash = slpque_cpu0;
327 } else {
77652cad 328 gd->gd_tsleep_hash = kmalloc(sizeof(slpque_cpu0),
fc17ad60
MD
329 M_TSLEEP, M_WAITOK | M_ZERO);
330 }
331 for (i = 0; i < TABLESIZE; ++i)
332 TAILQ_INIT(&gd->gd_tsleep_hash[i]);
984263bc
MD
333}
334
335/*
ae8e83e6
MD
336 * This is a dandy function that allows us to interlock tsleep/wakeup
337 * operations with unspecified upper level locks, such as lockmgr locks,
338 * simply by holding a critical section. The sequence is:
339 *
340 * (acquire upper level lock)
341 * tsleep_interlock(blah)
342 * (release upper level lock)
343 * tsleep(blah, ...)
344 *
345 * Basically this functions queues us on the tsleep queue without actually
346 * descheduling us. When tsleep() is later called with PINTERLOCK it
347 * assumes the thread was already queued, otherwise it queues it there.
348 *
349 * Thus it is possible to receive the wakeup prior to going to sleep and
350 * the race conditions are covered.
351 */
352static __inline void
5decebc7 353_tsleep_interlock(globaldata_t gd, const volatile void *ident, int flags)
ae8e83e6
MD
354{
355 thread_t td = gd->gd_curthread;
356 int id;
357
358 crit_enter_quick(td);
359 if (td->td_flags & TDF_TSLEEPQ) {
360 id = LOOKUP(td->td_wchan);
361 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_sleepq);
362 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL)
da23a592 363 atomic_clear_cpumask(&slpque_cpumasks[id], gd->gd_cpumask);
ae8e83e6
MD
364 } else {
365 td->td_flags |= TDF_TSLEEPQ;
366 }
367 id = LOOKUP(ident);
368 TAILQ_INSERT_TAIL(&gd->gd_tsleep_hash[id], td, td_sleepq);
da23a592 369 atomic_set_cpumask(&slpque_cpumasks[id], gd->gd_cpumask);
ae8e83e6
MD
370 td->td_wchan = ident;
371 td->td_wdomain = flags & PDOMAIN_MASK;
ae8e83e6
MD
372 crit_exit_quick(td);
373}
374
375void
5decebc7 376tsleep_interlock(const volatile void *ident, int flags)
ae8e83e6
MD
377{
378 _tsleep_interlock(mycpu, ident, flags);
379}
380
381/*
382 * Remove thread from sleepq. Must be called with a critical section held.
383 */
384static __inline void
385_tsleep_remove(thread_t td)
386{
387 globaldata_t gd = mycpu;
388 int id;
389
390 KKASSERT(td->td_gd == gd);
391 if (td->td_flags & TDF_TSLEEPQ) {
392 td->td_flags &= ~TDF_TSLEEPQ;
393 id = LOOKUP(td->td_wchan);
394 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_sleepq);
395 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL)
da23a592 396 atomic_clear_cpumask(&slpque_cpumasks[id], gd->gd_cpumask);
ae8e83e6
MD
397 td->td_wchan = NULL;
398 td->td_wdomain = 0;
399 }
400}
401
402void
403tsleep_remove(thread_t td)
404{
405 _tsleep_remove(td);
406}
407
408/*
409 * This function removes a thread from the tsleep queue and schedules
410 * it. This function may act asynchronously. The target thread may be
411 * sleeping on a different cpu.
412 *
413 * This function mus be called while in a critical section but if the
414 * target thread is sleeping on a different cpu we cannot safely probe
415 * td_flags.
74c9628e
MD
416 *
417 * This function is only called from a different cpu via setrunnable()
418 * when the thread is in a known sleep. However, multiple wakeups are
419 * possible and we must hold the td to prevent a race against the thread
420 * exiting.
ae8e83e6
MD
421 */
422static __inline
423void
424_tsleep_wakeup(struct thread *td)
425{
c4e34360 426#ifdef SMP
ae8e83e6
MD
427 globaldata_t gd = mycpu;
428
ae8e83e6 429 if (td->td_gd != gd) {
74c9628e
MD
430 lwkt_hold(td);
431 lwkt_send_ipiq(td->td_gd, (ipifunc1_t)tsleep_wakeup_remote, td);
ae8e83e6
MD
432 return;
433 }
434#endif
435 _tsleep_remove(td);
436 if (td->td_flags & TDF_TSLEEP_DESCHEDULED) {
437 td->td_flags &= ~TDF_TSLEEP_DESCHEDULED;
438 lwkt_schedule(td);
439 }
440}
441
ce4b5045 442#ifdef SMP
ae8e83e6
MD
443static
444void
74c9628e 445tsleep_wakeup_remote(struct thread *td)
ae8e83e6
MD
446{
447 _tsleep_wakeup(td);
74c9628e 448 lwkt_rele(td);
ae8e83e6 449}
ce4b5045 450#endif
ae8e83e6
MD
451
452
453/*
984263bc
MD
454 * General sleep call. Suspends the current process until a wakeup is
455 * performed on the specified identifier. The process will then be made
456 * runnable with the specified priority. Sleeps at most timo/hz seconds
377d4740 457 * (0 means no timeout). If flags includes PCATCH flag, signals are checked
984263bc
MD
458 * before and after sleeping, else signals are not checked. Returns 0 if
459 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
460 * signal needs to be delivered, ERESTART is returned if the current system
461 * call should be restarted if possible, and EINTR is returned if the system
462 * call should be interrupted by the signal (return EINTR).
26a0694b 463 *
0a3f9b47
MD
464 * Note that if we are a process, we release_curproc() before messing with
465 * the LWKT scheduler.
a46fac56
MD
466 *
467 * During autoconfiguration or after a panic, a sleep will simply
468 * lower the priority briefly to allow interrupts, then return.
984263bc
MD
469 */
470int
5decebc7 471tsleep(const volatile void *ident, int flags, const char *wmesg, int timo)
984263bc 472{
dadab5e9 473 struct thread *td = curthread;
08f2f1bb 474 struct lwp *lp = td->td_lwp;
0cfcada1 475 struct proc *p = td->td_proc; /* may be NULL */
fc17ad60 476 globaldata_t gd;
344ad853
MD
477 int sig;
478 int catch;
479 int id;
480 int error;
e43a034f 481 int oldpri;
076fecef 482 struct callout thandle;
984263bc 483
0cfcada1
MD
484 /*
485 * NOTE: removed KTRPOINT, it could cause races due to blocking
486 * even in stable. Just scrap it for now.
487 */
5ea440eb 488 if (!tsleep_crypto_dump && (tsleep_now_works == 0 || panicstr)) {
984263bc 489 /*
dbcd0c9b
MD
490 * After a panic, or before we actually have an operational
491 * softclock, just give interrupts a chance, then just return;
492 *
984263bc
MD
493 * don't run any other procs or panic below,
494 * in case this is the idle process and already asleep.
495 */
e43a034f 496 splz();
f9235b6d 497 oldpri = td->td_pri;
e43a034f
MD
498 lwkt_setpri_self(safepri);
499 lwkt_switch();
500 lwkt_setpri_self(oldpri);
984263bc
MD
501 return (0);
502 }
8aa3430c 503 logtsleep2(tsleep_beg, ident);
fc17ad60
MD
504 gd = td->td_gd;
505 KKASSERT(td != &gd->gd_idlethread); /* you must be kidding! */
344ad853
MD
506
507 /*
508 * NOTE: all of this occurs on the current cpu, including any
509 * callout-based wakeups, so a critical section is a sufficient
510 * interlock.
511 *
512 * The entire sequence through to where we actually sleep must
513 * run without breaking the critical section.
514 */
344ad853
MD
515 catch = flags & PCATCH;
516 error = 0;
517 sig = 0;
518
37af14fe 519 crit_enter_quick(td);
344ad853 520
0cfcada1 521 KASSERT(ident != NULL, ("tsleep: no ident"));
7278a846
SS
522 KASSERT(lp == NULL ||
523 lp->lwp_stat == LSRUN || /* Obvious */
524 lp->lwp_stat == LSSTOP, /* Set in tstop */
525 ("tsleep %p %s %d",
526 ident, wmesg, lp->lwp_stat));
0cfcada1 527
344ad853 528 /*
5686ec5a
MD
529 * We interlock the sleep queue if the caller has not already done
530 * it for us. This must be done before we potentially acquire any
531 * tokens or we can loose the wakeup.
532 */
533 if ((flags & PINTERLOCKED) == 0) {
534 id = LOOKUP(ident);
535 _tsleep_interlock(gd, ident, flags);
536 }
537
538 /*
344ad853 539 * Setup for the current process (if this is a process).
5686ec5a
MD
540 *
541 * We hold the process token if lp && catch. The resume
542 * code will release it.
344ad853 543 */
08f2f1bb 544 if (lp) {
344ad853
MD
545 if (catch) {
546 /*
547 * Early termination if PCATCH was set and a
548 * signal is pending, interlocked with the
549 * critical section.
550 *
551 * Early termination only occurs when tsleep() is
164b8401 552 * entered while in a normal LSRUN state.
344ad853 553 */
5686ec5a 554 lwkt_gettoken(&p->p_token);
08f2f1bb 555 if ((sig = CURSIG(lp)) != 0)
344ad853
MD
556 goto resume;
557
558 /*
7c1212ec
MD
559 * Early termination if PCATCH was set and a
560 * mailbox signal was possibly delivered prior to
561 * the system call even being made, in order to
562 * allow the user to interlock without having to
563 * make additional system calls.
564 */
565 if (p->p_flag & P_MAILBOX)
566 goto resume;
567
568 /*
5686ec5a
MD
569 * Causes ksignal to wake us up if a signal is
570 * received (interlocked with p->p_token).
344ad853 571 */
9a379a4a 572 lp->lwp_flag |= LWP_SINTR;
344ad853 573 }
5686ec5a
MD
574 } else {
575 KKASSERT(p == NULL);
4ecd8190 576 }
344ad853 577
4ecd8190 578 /*
4ecd8190
MD
579 * Make sure the current process has been untangled from
580 * the userland scheduler and initialize slptime to start
5686ec5a 581 * counting.
4ecd8190
MD
582 */
583 if (lp) {
08f2f1bb
SS
584 p->p_usched->release_curproc(lp);
585 lp->lwp_slptime = 0;
0a3f9b47 586 }
fc17ad60
MD
587
588 /*
d9345d3a
MD
589 * If the interlocked flag is set but our cpu bit in the slpqueue
590 * is no longer set, then a wakeup was processed inbetween the
4ecd8190
MD
591 * tsleep_interlock() (ours or the callers), and here. This can
592 * occur under numerous circumstances including when we release the
593 * current process.
d9345d3a 594 *
4ecd8190
MD
595 * Extreme loads can cause the sending of an IPI (e.g. wakeup()'s)
596 * to process incoming IPIs, thus draining incoming wakeups.
d9345d3a 597 */
4ecd8190
MD
598 if ((td->td_flags & TDF_TSLEEPQ) == 0) {
599 logtsleep2(ilockfail, ident);
600 goto resume;
d9345d3a 601 }
4ecd8190
MD
602
603 /*
604 * scheduling is blocked while in a critical section. Coincide
605 * the descheduled-by-tsleep flag with the descheduling of the
606 * lwkt.
607 */
37af14fe 608 lwkt_deschedule_self(td);
ae8e83e6 609 td->td_flags |= TDF_TSLEEP_DESCHEDULED;
344ad853 610 td->td_wmesg = wmesg;
344ad853
MD
611
612 /*
613 * Setup the timeout, if any
614 */
076fecef
MD
615 if (timo) {
616 callout_init(&thandle);
617 callout_reset(&thandle, timo, endtsleep, td);
618 }
344ad853 619
984263bc 620 /*
344ad853 621 * Beddy bye bye.
984263bc 622 */
08f2f1bb 623 if (lp) {
26a0694b 624 /*
52eedfb5 625 * Ok, we are sleeping. Place us in the SSLEEP state.
26a0694b 626 */
9388413d 627 KKASSERT((lp->lwp_flag & LWP_ONRUNQ) == 0);
7278a846
SS
628 /*
629 * tstop() sets LSSTOP, so don't fiddle with that.
630 */
631 if (lp->lwp_stat != LSSTOP)
632 lp->lwp_stat = LSSLEEP;
08f2f1bb 633 lp->lwp_ru.ru_nvcsw++;
344ad853 634 lwkt_switch();
ab554892
MD
635
636 /*
164b8401 637 * And when we are woken up, put us back in LSRUN. If we
ab554892
MD
638 * slept for over a second, recalculate our estcpu.
639 */
164b8401 640 lp->lwp_stat = LSRUN;
08f2f1bb
SS
641 if (lp->lwp_slptime)
642 p->p_usched->recalculate(lp);
643 lp->lwp_slptime = 0;
0cfcada1
MD
644 } else {
645 lwkt_switch();
646 }
344ad853 647
fc17ad60
MD
648 /*
649 * Make sure we haven't switched cpus while we were asleep. It's
344ad853 650 * not supposed to happen. Cleanup our temporary flags.
fc17ad60
MD
651 */
652 KKASSERT(gd == td->td_gd);
344ad853
MD
653
654 /*
655 * Cleanup the timeout.
656 */
657 if (timo) {
658 if (td->td_flags & TDF_TIMEOUT) {
659 td->td_flags &= ~TDF_TIMEOUT;
a40da8f0 660 error = EWOULDBLOCK;
344ad853
MD
661 } else {
662 callout_stop(&thandle);
663 }
0cfcada1 664 }
344ad853
MD
665
666 /*
ae8e83e6
MD
667 * Make sure we have been removed from the sleepq. This should
668 * have been done for us already.
f022a370
MD
669 *
670 * However, it is possible for a scheduling IPI to be in flight
671 * from a previous tsleep/tsleep_interlock or due to a straight-out
672 * call to lwkt_schedule() (in the case of an interrupt thread).
673 * So don't complain if DESCHEDULED is still set.
344ad853 674 */
ae8e83e6 675 _tsleep_remove(td);
344ad853 676 td->td_wmesg = NULL;
ae8e83e6
MD
677 if (td->td_flags & TDF_TSLEEP_DESCHEDULED) {
678 td->td_flags &= ~TDF_TSLEEP_DESCHEDULED;
ae8e83e6 679 }
344ad853
MD
680
681 /*
7c1212ec
MD
682 * Figure out the correct error return. If interrupted by a
683 * signal we want to return EINTR or ERESTART.
684 *
685 * If P_MAILBOX is set no automatic system call restart occurs
686 * and we return EINTR. P_MAILBOX is meant to be used as an
687 * interlock, the user must poll it prior to any system call
688 * that it wishes to interlock a mailbox signal against since
689 * the flag is cleared on *any* system call that sleeps.
5686ec5a
MD
690 *
691 * p->p_token is held in the p && catch case.
344ad853
MD
692 */
693resume:
0cfcada1 694 if (p) {
7c1212ec
MD
695 if (catch && error == 0) {
696 if ((p->p_flag & P_MAILBOX) && sig == 0) {
344ad853 697 error = EINTR;
08f2f1bb 698 } else if (sig != 0 || (sig = CURSIG(lp))) {
7c1212ec
MD
699 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
700 error = EINTR;
701 else
702 error = ERESTART;
703 }
984263bc 704 }
5686ec5a
MD
705 if (catch)
706 lwkt_reltoken(&p->p_token);
9a379a4a
SS
707 lp->lwp_flag &= ~(LWP_BREAKTSLEEP | LWP_SINTR);
708 p->p_flag &= ~P_MAILBOX;
984263bc 709 }
8aa3430c 710 logtsleep1(tsleep_end);
344ad853
MD
711 crit_exit_quick(td);
712 return (error);
984263bc
MD
713}
714
984263bc 715/*
bf765287 716 * Interlocked spinlock sleep. An exclusively held spinlock must
e590ee86 717 * be passed to ssleep(). The function will atomically release the
bf765287
MD
718 * spinlock and tsleep on the ident, then reacquire the spinlock and
719 * return.
720 *
721 * This routine is fairly important along the critical path, so optimize it
722 * heavily.
723 */
724int
5decebc7 725ssleep(const volatile void *ident, struct spinlock *spin, int flags,
bf765287
MD
726 const char *wmesg, int timo)
727{
728 globaldata_t gd = mycpu;
729 int error;
16523a43 730
ae8e83e6 731 _tsleep_interlock(gd, ident, flags);
7cfe2b28 732 spin_unlock_quick(gd, spin);
ef48be0d 733 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
7cfe2b28 734 spin_lock_quick(gd, spin);
bf765287
MD
735
736 return (error);
16523a43
MD
737}
738
bed060de 739int
5decebc7
MD
740lksleep(const volatile void *ident, struct lock *lock, int flags,
741 const char *wmesg, int timo)
bed060de
AH
742{
743 globaldata_t gd = mycpu;
744 int error;
745
746 _tsleep_interlock(gd, ident, flags);
747 lockmgr(lock, LK_RELEASE);
748 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
749 lockmgr(lock, LK_EXCLUSIVE);
750
751 return (error);
752}
753
16523a43 754/*
7f6220a9
MD
755 * Interlocked mutex sleep. An exclusively held mutex must be passed
756 * to mtxsleep(). The function will atomically release the mutex
757 * and tsleep on the ident, then reacquire the mutex and return.
758 */
759int
5decebc7 760mtxsleep(const volatile void *ident, struct mtx *mtx, int flags,
7f6220a9
MD
761 const char *wmesg, int timo)
762{
763 globaldata_t gd = mycpu;
764 int error;
765
766 _tsleep_interlock(gd, ident, flags);
767 mtx_unlock(mtx);
768 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
769 mtx_lock_ex_quick(mtx, wmesg);
770
771 return (error);
772}
773
774/*
362e59be 775 * Interlocked serializer sleep. An exclusively held serializer must
ed3f6624 776 * be passed to zsleep(). The function will atomically release
362e59be
SZ
777 * the serializer and tsleep on the ident, then reacquire the serializer
778 * and return.
779 */
780int
5decebc7 781zsleep(const volatile void *ident, struct lwkt_serialize *slz, int flags,
ed3f6624 782 const char *wmesg, int timo)
362e59be 783{
ae8e83e6 784 globaldata_t gd = mycpu;
362e59be
SZ
785 int ret;
786
787 ASSERT_SERIALIZED(slz);
788
ae8e83e6 789 _tsleep_interlock(gd, ident, flags);
362e59be 790 lwkt_serialize_exit(slz);
ef48be0d 791 ret = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
362e59be 792 lwkt_serialize_enter(slz);
362e59be
SZ
793
794 return ret;
795}
796
797/*
a22c590e
MD
798 * Directly block on the LWKT thread by descheduling it. This
799 * is much faster then tsleep(), but the only legal way to wake
800 * us up is to directly schedule the thread.
801 *
802 * Setting TDF_SINTR will cause new signals to directly schedule us.
803 *
ae8e83e6 804 * This routine must be called while in a critical section.
a22c590e
MD
805 */
806int
807lwkt_sleep(const char *wmesg, int flags)
808{
809 thread_t td = curthread;
810 int sig;
811
812 if ((flags & PCATCH) == 0 || td->td_lwp == NULL) {
813 td->td_flags |= TDF_BLOCKED;
814 td->td_wmesg = wmesg;
815 lwkt_deschedule_self(td);
816 lwkt_switch();
817 td->td_wmesg = NULL;
818 td->td_flags &= ~TDF_BLOCKED;
819 return(0);
820 }
821 if ((sig = CURSIG(td->td_lwp)) != 0) {
822 if (SIGISMEMBER(td->td_proc->p_sigacts->ps_sigintr, sig))
823 return(EINTR);
824 else
825 return(ERESTART);
826
827 }
828 td->td_flags |= TDF_BLOCKED | TDF_SINTR;
829 td->td_wmesg = wmesg;
830 lwkt_deschedule_self(td);
831 lwkt_switch();
832 td->td_flags &= ~(TDF_BLOCKED | TDF_SINTR);
833 td->td_wmesg = NULL;
834 return(0);
835}
836
837/*
344ad853 838 * Implement the timeout for tsleep.
fc17ad60 839 *
9a379a4a 840 * We set LWP_BREAKTSLEEP to indicate that an event has occured, but
344ad853
MD
841 * we only call setrunnable if the process is not stopped.
842 *
843 * This type of callout timeout is scheduled on the same cpu the process
844 * is sleeping on. Also, at the moment, the MP lock is held.
984263bc
MD
845 */
846static void
0cfcada1 847endtsleep(void *arg)
984263bc 848{
0cfcada1 849 thread_t td = arg;
9a379a4a 850 struct lwp *lp;
984263bc 851
37af14fe 852 crit_enter();
5686ec5a
MD
853 lp = td->td_lwp;
854
855 if (lp)
856 lwkt_gettoken(&lp->lwp_proc->p_token);
344ad853
MD
857
858 /*
859 * cpu interlock. Thread flags are only manipulated on
860 * the cpu owning the thread. proc flags are only manipulated
5686ec5a 861 * by the holder of p->p_token. We have both.
344ad853 862 */
ae8e83e6 863 if (td->td_flags & TDF_TSLEEP_DESCHEDULED) {
0cfcada1 864 td->td_flags |= TDF_TIMEOUT;
344ad853 865
5686ec5a 866 if (lp) {
9a379a4a
SS
867 lp->lwp_flag |= LWP_BREAKTSLEEP;
868 if (lp->lwp_proc->p_stat != SSTOP)
869 setrunnable(lp);
0cfcada1 870 } else {
ae8e83e6 871 _tsleep_wakeup(td);
0cfcada1 872 }
984263bc 873 }
5686ec5a
MD
874 if (lp)
875 lwkt_reltoken(&lp->lwp_proc->p_token);
37af14fe 876 crit_exit();
984263bc
MD
877}
878
984263bc 879/*
8fb8bca6 880 * Make all processes sleeping on the specified identifier runnable.
fc17ad60
MD
881 * count may be zero or one only.
882 *
883 * The domain encodes the sleep/wakeup domain AND the first cpu to check
884 * (which is always the current cpu). As we iterate across cpus
344ad853
MD
885 *
886 * This call may run without the MP lock held. We can only manipulate thread
887 * state on the cpu owning the thread. We CANNOT manipulate process state
888 * at all.
5decebc7
MD
889 *
890 * _wakeup() can be passed to an IPI so we can't use (const volatile
891 * void *ident).
8fb8bca6
EN
892 */
893static void
fc17ad60 894_wakeup(void *ident, int domain)
8fb8bca6 895{
fc17ad60 896 struct tslpque *qp;
0cfcada1
MD
897 struct thread *td;
898 struct thread *ntd;
fc17ad60 899 globaldata_t gd;
fc17ad60
MD
900#ifdef SMP
901 cpumask_t mask;
fc17ad60
MD
902#endif
903 int id;
984263bc 904
37af14fe 905 crit_enter();
8aa3430c 906 logtsleep2(wakeup_beg, ident);
fc17ad60
MD
907 gd = mycpu;
908 id = LOOKUP(ident);
909 qp = &gd->gd_tsleep_hash[id];
984263bc 910restart:
0cfcada1 911 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
ae8e83e6 912 ntd = TAILQ_NEXT(td, td_sleepq);
fc17ad60
MD
913 if (td->td_wchan == ident &&
914 td->td_wdomain == (domain & PDOMAIN_MASK)
915 ) {
ae8e83e6
MD
916 KKASSERT(td->td_gd == gd);
917 _tsleep_remove(td);
918 if (td->td_flags & TDF_TSLEEP_DESCHEDULED) {
919 td->td_flags &= ~TDF_TSLEEP_DESCHEDULED;
920 lwkt_schedule(td);
921 if (domain & PWAKEUP_ONE)
922 goto done;
fc17ad60 923 }
0cfcada1 924 goto restart;
984263bc
MD
925 }
926 }
fc17ad60
MD
927
928#ifdef SMP
929 /*
930 * We finished checking the current cpu but there still may be
931 * more work to do. Either wakeup_one was requested and no matching
932 * thread was found, or a normal wakeup was requested and we have
933 * to continue checking cpus.
934 *
fc17ad60
MD
935 * It should be noted that this scheme is actually less expensive then
936 * the old scheme when waking up multiple threads, since we send
937 * only one IPI message per target candidate which may then schedule
938 * multiple threads. Before we could have wound up sending an IPI
939 * message for each thread on the target cpu (!= current cpu) that
940 * needed to be woken up.
941 *
942 * NOTE: Wakeups occuring on remote cpus are asynchronous. This
943 * should be ok since we are passing idents in the IPI rather then
944 * thread pointers.
945 */
1f4f6e0b
MD
946 if ((domain & PWAKEUP_MYCPU) == 0 &&
947 (mask = slpque_cpumasks[id] & gd->gd_other_cpus) != 0) {
948 lwkt_send_ipiq2_mask(mask, _wakeup, ident,
949 domain | PWAKEUP_MYCPU);
fc17ad60
MD
950 }
951#endif
952done:
8aa3430c 953 logtsleep1(wakeup_end);
37af14fe 954 crit_exit();
984263bc
MD
955}
956
b336a9b1
MD
957/*
958 * Wakeup all threads tsleep()ing on the specified ident, on all cpus
959 */
984263bc 960void
5decebc7 961wakeup(const volatile void *ident)
984263bc 962{
5decebc7 963 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mycpu->gd_cpuid));
0cfcada1 964}
984263bc 965
b336a9b1
MD
966/*
967 * Wakeup one thread tsleep()ing on the specified ident, on any cpu.
968 */
0cfcada1 969void
5decebc7 970wakeup_one(const volatile void *ident)
0cfcada1 971{
fc17ad60 972 /* XXX potentially round-robin the first responding cpu */
5decebc7 973 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mycpu->gd_cpuid) | PWAKEUP_ONE);
da5fb9ef
MD
974}
975
b336a9b1
MD
976/*
977 * Wakeup threads tsleep()ing on the specified ident on the current cpu
978 * only.
979 */
980void
5decebc7 981wakeup_mycpu(const volatile void *ident)
b336a9b1 982{
5decebc7 983 _wakeup(__DEALL(ident), PWAKEUP_MYCPU);
b336a9b1
MD
984}
985
986/*
987 * Wakeup one thread tsleep()ing on the specified ident on the current cpu
988 * only.
989 */
990void
5decebc7 991wakeup_mycpu_one(const volatile void *ident)
b336a9b1
MD
992{
993 /* XXX potentially round-robin the first responding cpu */
5decebc7 994 _wakeup(__DEALL(ident), PWAKEUP_MYCPU|PWAKEUP_ONE);
b336a9b1
MD
995}
996
997/*
998 * Wakeup all thread tsleep()ing on the specified ident on the specified cpu
999 * only.
1000 */
1001void
5decebc7 1002wakeup_oncpu(globaldata_t gd, const volatile void *ident)
b336a9b1 1003{
1699d292 1004#ifdef SMP
b336a9b1 1005 if (gd == mycpu) {
5decebc7 1006 _wakeup(__DEALL(ident), PWAKEUP_MYCPU);
b336a9b1 1007 } else {
5decebc7 1008 lwkt_send_ipiq2(gd, _wakeup, __DEALL(ident), PWAKEUP_MYCPU);
b336a9b1 1009 }
1699d292 1010#else
8e44b950 1011 _wakeup(__DEALL(ident), PWAKEUP_MYCPU);
1699d292 1012#endif
b336a9b1
MD
1013}
1014
1015/*
1016 * Wakeup one thread tsleep()ing on the specified ident on the specified cpu
1017 * only.
1018 */
1019void
5decebc7 1020wakeup_oncpu_one(globaldata_t gd, const volatile void *ident)
b336a9b1 1021{
1699d292 1022#ifdef SMP
b336a9b1 1023 if (gd == mycpu) {
5decebc7 1024 _wakeup(__DEALL(ident), PWAKEUP_MYCPU | PWAKEUP_ONE);
b336a9b1 1025 } else {
5decebc7
MD
1026 lwkt_send_ipiq2(gd, _wakeup, __DEALL(ident),
1027 PWAKEUP_MYCPU | PWAKEUP_ONE);
b336a9b1 1028 }
1699d292 1029#else
8e44b950 1030 _wakeup(__DEALL(ident), PWAKEUP_MYCPU | PWAKEUP_ONE);
1699d292 1031#endif
b336a9b1
MD
1032}
1033
1034/*
1035 * Wakeup all threads waiting on the specified ident that slept using
1036 * the specified domain, on all cpus.
1037 */
da5fb9ef 1038void
5decebc7 1039wakeup_domain(const volatile void *ident, int domain)
da5fb9ef 1040{
5decebc7 1041 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(domain, mycpu->gd_cpuid));
da5fb9ef
MD
1042}
1043
b336a9b1
MD
1044/*
1045 * Wakeup one thread waiting on the specified ident that slept using
1046 * the specified domain, on any cpu.
1047 */
da5fb9ef 1048void
5decebc7 1049wakeup_domain_one(const volatile void *ident, int domain)
da5fb9ef 1050{
fc17ad60 1051 /* XXX potentially round-robin the first responding cpu */
5decebc7
MD
1052 _wakeup(__DEALL(ident),
1053 PWAKEUP_ENCODE(domain, mycpu->gd_cpuid) | PWAKEUP_ONE);
984263bc
MD
1054}
1055
1056/*
344ad853
MD
1057 * setrunnable()
1058 *
5686ec5a
MD
1059 * Make a process runnable. lp->lwp_proc->p_token must be held on call.
1060 * This only has an effect if we are in SSLEEP. We only break out of the
9a379a4a 1061 * tsleep if LWP_BREAKTSLEEP is set, otherwise we just fix-up the state.
37af14fe 1062 *
5686ec5a 1063 * NOTE: With p_token held we can only safely manipulate the process
74c9628e 1064 * structure and the lp's lwp_stat.
984263bc
MD
1065 */
1066void
9a379a4a 1067setrunnable(struct lwp *lp)
984263bc 1068{
5686ec5a 1069 ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
344ad853 1070 crit_enter();
2daf83b0
SS
1071 if (lp->lwp_stat == LSSTOP)
1072 lp->lwp_stat = LSSLEEP;
1073 if (lp->lwp_stat == LSSLEEP && (lp->lwp_flag & LWP_BREAKTSLEEP))
ae8e83e6 1074 _tsleep_wakeup(lp->lwp_thread);
344ad853 1075 crit_exit();
984263bc
MD
1076}
1077
1078/*
164b8401
SS
1079 * The process is stopped due to some condition, usually because p_stat is
1080 * set to SSTOP, but also possibly due to being traced.
fc17ad60 1081 *
164b8401 1082 * NOTE! If the caller sets SSTOP, the caller must also clear P_WAITED
344ad853
MD
1083 * because the parent may check the child's status before the child actually
1084 * gets to this routine.
1085 *
9a379a4a 1086 * This routine is called with the current lwp only, typically just
344ad853
MD
1087 * before returning to userland.
1088 *
9a379a4a 1089 * Setting LWP_BREAKTSLEEP before entering the tsleep will cause a passive
344ad853 1090 * SIGCONT to break out of the tsleep.
984263bc
MD
1091 */
1092void
9a379a4a 1093tstop(void)
984263bc 1094{
9a379a4a 1095 struct lwp *lp = curthread->td_lwp;
7278a846 1096 struct proc *p = lp->lwp_proc;
9a379a4a 1097
7278a846 1098 crit_enter();
f33e8653
SS
1099 /*
1100 * If LWP_WSTOP is set, we were sleeping
1101 * while our process was stopped. At this point
1102 * we were already counted as stopped.
1103 */
1104 if ((lp->lwp_flag & LWP_WSTOP) == 0) {
1105 /*
1106 * If we're the last thread to stop, signal
1107 * our parent.
1108 */
1109 p->p_nstopped++;
1110 lp->lwp_flag |= LWP_WSTOP;
ea59a697 1111 wakeup(&p->p_nstopped);
f33e8653
SS
1112 if (p->p_nstopped == p->p_nthreads) {
1113 p->p_flag &= ~P_WAITED;
1114 wakeup(p->p_pptr);
1115 if ((p->p_pptr->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1116 ksignal(p->p_pptr, SIGCHLD);
1117 }
1118 }
ea59a697
SS
1119 while (p->p_stat == SSTOP) {
1120 lp->lwp_flag |= LWP_BREAKTSLEEP;
1121 lp->lwp_stat = LSSTOP;
1122 tsleep(p, 0, "stop", 0);
1123 }
7278a846 1124 p->p_nstopped--;
a5ff9d37 1125 lp->lwp_flag &= ~LWP_WSTOP;
7278a846 1126 crit_exit();
26a0694b
MD
1127}
1128
1129/*
984263bc
MD
1130 * Compute a tenex style load average of a quantity on
1131 * 1, 5 and 15 minute intervals.
1132 */
c7e98b2f 1133static int loadav_count_runnable(struct lwp *p, void *data);
8fa76237 1134
984263bc
MD
1135static void
1136loadav(void *arg)
1137{
984263bc 1138 struct loadavg *avg;
8fa76237 1139 int i, nrun;
984263bc 1140
984263bc 1141 nrun = 0;
c7e98b2f 1142 alllwp_scan(loadav_count_runnable, &nrun);
8fa76237
MD
1143 avg = &averunnable;
1144 for (i = 0; i < 3; i++) {
984263bc
MD
1145 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
1146 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
8fa76237 1147 }
984263bc
MD
1148
1149 /*
1150 * Schedule the next update to occur after 5 seconds, but add a
1151 * random variation to avoid synchronisation with processes that
1152 * run at regular intervals.
1153 */
cddfb7bb 1154 callout_reset(&loadav_callout, hz * 4 + (int)(krandom() % (hz * 2 + 1)),
8fa76237
MD
1155 loadav, NULL);
1156}
1157
1158static int
c7e98b2f 1159loadav_count_runnable(struct lwp *lp, void *data)
8fa76237
MD
1160{
1161 int *nrunp = data;
1162 thread_t td;
1163
164b8401
SS
1164 switch (lp->lwp_stat) {
1165 case LSRUN:
08f2f1bb 1166 if ((td = lp->lwp_thread) == NULL)
8fa76237
MD
1167 break;
1168 if (td->td_flags & TDF_BLOCKED)
1169 break;
8fa76237
MD
1170 ++*nrunp;
1171 break;
1172 default:
1173 break;
1174 }
1175 return(0);
984263bc
MD
1176}
1177
1178/* ARGSUSED */
1179static void
6656cd91 1180sched_setup(void *dummy)
984263bc 1181{
984263bc 1182 callout_init(&loadav_callout);
35f9d051 1183 callout_init(&schedcpu_callout);
984263bc
MD
1184
1185 /* Kick off timeout driven events by calling first time. */
984263bc
MD
1186 schedcpu(NULL);
1187 loadav(NULL);
1188}
1189