kernel: Remove unused hogticks global from kern_synch.c.
[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.
dc71b7ab 18 * 3. Neither the name of the University nor the names of its contributors
984263bc
MD
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 * $FreeBSD: src/sys/kern/kern_synch.c,v 1.87.2.6 2002/10/13 07:29:53 kbyanc Exp $
36 */
37
38#include "opt_ktrace.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/proc.h>
43#include <sys/kernel.h>
44#include <sys/signalvar.h>
45#include <sys/resourcevar.h>
46#include <sys/vmmeter.h>
47#include <sys/sysctl.h>
344ad853 48#include <sys/lock.h>
984263bc 49#include <sys/uio.h>
fc9ae81d 50#ifdef KTRACE
984263bc
MD
51#include <sys/ktrace.h>
52#endif
f1d1c3fa 53#include <sys/xwait.h>
9afb0ffd 54#include <sys/ktr.h>
684a93c4 55#include <sys/serialize.h>
984263bc 56
684a93c4 57#include <sys/signal2.h>
bf765287
MD
58#include <sys/thread2.h>
59#include <sys/spinlock2.h>
7f6220a9 60#include <sys/mutex2.h>
bf765287 61
984263bc 62#include <machine/cpu.h>
984263bc
MD
63#include <machine/smp.h>
64
fc17ad60
MD
65TAILQ_HEAD(tslpque, thread);
66
402ed7e1 67static void sched_setup (void *dummy);
984263bc
MD
68SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL)
69
984263bc 70int lbolt;
50e4012a 71void *lbolt_syncer;
984263bc 72int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
17a9f566 73int ncpus;
da23a592
MD
74int ncpus2, ncpus2_shift, ncpus2_mask; /* note: mask not cpumask_t */
75int ncpus_fit, ncpus_fit_mask; /* note: mask not cpumask_t */
e43a034f 76int safepri;
dbcd0c9b 77int tsleep_now_works;
5ea440eb 78int tsleep_crypto_dump = 0;
984263bc
MD
79
80static struct callout loadav_callout;
35f9d051 81static struct callout schedcpu_callout;
fc17ad60 82MALLOC_DEFINE(M_TSLEEP, "tslpque", "tsleep queues");
984263bc 83
5decebc7
MD
84#define __DEALL(ident) __DEQUALIFY(void *, ident)
85
9afb0ffd
MD
86#if !defined(KTR_TSLEEP)
87#define KTR_TSLEEP KTR_ALL
88#endif
89KTR_INFO_MASTER(tsleep);
5bf48697
AE
90KTR_INFO(KTR_TSLEEP, tsleep, tsleep_beg, 0, "tsleep enter %p", const volatile void *ident);
91KTR_INFO(KTR_TSLEEP, tsleep, tsleep_end, 1, "tsleep exit");
92KTR_INFO(KTR_TSLEEP, tsleep, wakeup_beg, 2, "wakeup enter %p", const volatile void *ident);
93KTR_INFO(KTR_TSLEEP, tsleep, wakeup_end, 3, "wakeup exit");
94KTR_INFO(KTR_TSLEEP, tsleep, ilockfail, 4, "interlock failed %p", const volatile void *ident);
8aa3430c
MD
95
96#define logtsleep1(name) KTR_LOG(tsleep_ ## name)
97#define logtsleep2(name, val) KTR_LOG(tsleep_ ## name, val)
9afb0ffd 98
984263bc
MD
99struct loadavg averunnable =
100 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */
101/*
102 * Constants for averages over 1, 5, and 15 minutes
103 * when sampling at 5 second intervals.
104 */
105static fixpt_t cexp[3] = {
106 0.9200444146293232 * FSCALE, /* exp(-1/12) */
107 0.9834714538216174 * FSCALE, /* exp(-1/60) */
108 0.9944598480048967 * FSCALE, /* exp(-1/180) */
109};
110
402ed7e1
RG
111static void endtsleep (void *);
112static void loadav (void *arg);
402ed7e1 113static void schedcpu (void *arg);
984263bc 114
a46fac56
MD
115/*
116 * Adjust the scheduler quantum. The quantum is specified in microseconds.
117 * Note that 'tick' is in microseconds per tick.
118 */
984263bc
MD
119static int
120sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
121{
122 int error, new_val;
123
a591f597 124 new_val = sched_quantum * ustick;
984263bc
MD
125 error = sysctl_handle_int(oidp, &new_val, 0, req);
126 if (error != 0 || req->newptr == NULL)
127 return (error);
a591f597 128 if (new_val < ustick)
984263bc 129 return (EINVAL);
a591f597 130 sched_quantum = new_val / ustick;
984263bc
MD
131 return (0);
132}
133
134SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW,
135 0, sizeof sched_quantum, sysctl_kern_quantum, "I", "");
136
bc55d64f
MD
137static int pctcpu_decay = 10;
138SYSCTL_INT(_kern, OID_AUTO, pctcpu_decay, CTLFLAG_RW, &pctcpu_decay, 0, "");
dcc99b62
MD
139
140/*
141 * kernel uses `FSCALE', userland (SHOULD) use kern.fscale
984263bc 142 */
460426e6 143int fscale __unused = FSCALE; /* exported to systat */
dcc99b62 144SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, "");
984263bc
MD
145
146/*
0a3f9b47 147 * Recompute process priorities, once a second.
dcc99b62
MD
148 *
149 * Since the userland schedulers are typically event oriented, if the
150 * estcpu calculation at wakeup() time is not sufficient to make a
151 * process runnable relative to other processes in the system we have
152 * a 1-second recalc to help out.
153 *
154 * This code also allows us to store sysclock_t data in the process structure
155 * without fear of an overrun, since sysclock_t are guarenteed to hold
156 * several seconds worth of count.
8fa76237
MD
157 *
158 * WARNING! callouts can preempt normal threads. However, they will not
159 * preempt a thread holding a spinlock so we *can* safely use spinlocks.
984263bc 160 */
8fa76237
MD
161static int schedcpu_stats(struct proc *p, void *data __unused);
162static int schedcpu_resource(struct proc *p, void *data __unused);
163
984263bc 164static void
26a0694b 165schedcpu(void *arg)
984263bc 166{
8fa76237
MD
167 allproc_scan(schedcpu_stats, NULL);
168 allproc_scan(schedcpu_resource, NULL);
169 wakeup((caddr_t)&lbolt);
50e4012a 170 wakeup(lbolt_syncer);
8fa76237
MD
171 callout_reset(&schedcpu_callout, hz, schedcpu, NULL);
172}
173
174/*
175 * General process statistics once a second
176 */
177static int
178schedcpu_stats(struct proc *p, void *data __unused)
179{
08f2f1bb
SS
180 struct lwp *lp;
181
7bea4e64
MD
182 /*
183 * Threads may not be completely set up if process in SIDL state.
184 */
185 if (p->p_stat == SIDL)
186 return(0);
187
0d78b86e 188 PHOLD(p);
85946b6c
MD
189 if (lwkt_trytoken(&p->p_token) == FALSE) {
190 PRELE(p);
191 return(0);
192 }
0d78b86e 193
8fa76237 194 p->p_swtime++;
c7e98b2f 195 FOREACH_LWP_IN_PROC(lp, p) {
de4d4cb0
MD
196 if (lp->lwp_stat == LSSLEEP) {
197 ++lp->lwp_slptime;
198 if (lp->lwp_slptime == 1)
199 p->p_usched->uload_update(lp);
200 }
4b5f931b 201
c7e98b2f
SS
202 /*
203 * Only recalculate processes that are active or have slept
204 * less then 2 seconds. The schedulers understand this.
bc55d64f 205 * Otherwise decay by 50% per second.
c7e98b2f
SS
206 */
207 if (lp->lwp_slptime <= 1) {
208 p->p_usched->recalculate(lp);
209 } else {
bc55d64f
MD
210 int decay;
211
212 decay = pctcpu_decay;
213 cpu_ccfence();
214 if (decay <= 1)
215 decay = 1;
216 if (decay > 100)
217 decay = 100;
218 lp->lwp_pctcpu = (lp->lwp_pctcpu * (decay - 1)) / decay;
c7e98b2f 219 }
8fa76237 220 }
0d78b86e 221 lwkt_reltoken(&p->p_token);
d2d8515b 222 lwkt_yield();
0d78b86e 223 PRELE(p);
8fa76237
MD
224 return(0);
225}
a46fac56 226
8fa76237 227/*
84204577 228 * Resource checks. XXX break out since ksignal/killproc can block,
8fa76237
MD
229 * limiting us to one process killed per second. There is probably
230 * a better way.
231 */
232static int
233schedcpu_resource(struct proc *p, void *data __unused)
234{
235 u_int64_t ttime;
08f2f1bb 236 struct lwp *lp;
8fa76237 237
0d78b86e
MD
238 if (p->p_stat == SIDL)
239 return(0);
240
241 PHOLD(p);
85946b6c
MD
242 if (lwkt_trytoken(&p->p_token) == FALSE) {
243 PRELE(p);
244 return(0);
245 }
0d78b86e
MD
246
247 if (p->p_stat == SZOMB || p->p_limit == NULL) {
248 lwkt_reltoken(&p->p_token);
249 PRELE(p);
8fa76237 250 return(0);
984263bc 251 }
344ad853 252
c7e98b2f
SS
253 ttime = 0;
254 FOREACH_LWP_IN_PROC(lp, p) {
e595c6cd
MD
255 /*
256 * We may have caught an lp in the middle of being
257 * created, lwp_thread can be NULL.
258 */
259 if (lp->lwp_thread) {
260 ttime += lp->lwp_thread->td_sticks;
261 ttime += lp->lwp_thread->td_uticks;
262 }
c7e98b2f 263 }
8fa76237
MD
264
265 switch(plimit_testcpulimit(p->p_limit, ttime)) {
266 case PLIMIT_TESTCPU_KILL:
267 killproc(p, "exceeded maximum CPU limit");
268 break;
269 case PLIMIT_TESTCPU_XCPU:
4643740a
MD
270 if ((p->p_flags & P_XCPU) == 0) {
271 p->p_flags |= P_XCPU;
84204577 272 ksignal(p, SIGXCPU);
344ad853 273 }
8fa76237
MD
274 break;
275 default:
c0b8a06d 276 break;
344ad853 277 }
0d78b86e 278 lwkt_reltoken(&p->p_token);
d2d8515b 279 lwkt_yield();
0d78b86e 280 PRELE(p);
8fa76237 281 return(0);
984263bc
MD
282}
283
284/*
dcc99b62
MD
285 * This is only used by ps. Generate a cpu percentage use over
286 * a period of one second.
984263bc 287 */
dcc99b62 288void
553ea3c8 289updatepcpu(struct lwp *lp, int cpticks, int ttlticks)
984263bc 290{
dcc99b62
MD
291 fixpt_t acc;
292 int remticks;
293
294 acc = (cpticks << FSHIFT) / ttlticks;
295 if (ttlticks >= ESTCPUFREQ) {
553ea3c8 296 lp->lwp_pctcpu = acc;
dcc99b62
MD
297 } else {
298 remticks = ESTCPUFREQ - ttlticks;
553ea3c8 299 lp->lwp_pctcpu = (acc * ttlticks + lp->lwp_pctcpu * remticks) /
dcc99b62 300 ESTCPUFREQ;
a46fac56 301 }
984263bc
MD
302}
303
304/*
8aa3430c
MD
305 * tsleep/wakeup hash table parameters. Try to find the sweet spot for
306 * like addresses being slept on.
984263bc 307 */
b12defdc
MD
308#define TABLESIZE 4001
309#define LOOKUP(x) (((u_int)(uintptr_t)(x)) % TABLESIZE)
984263bc 310
fc17ad60
MD
311static cpumask_t slpque_cpumasks[TABLESIZE];
312
984263bc 313/*
a46fac56 314 * General scheduler initialization. We force a reschedule 25 times
fc17ad60
MD
315 * a second by default. Note that cpu0 is initialized in early boot and
316 * cannot make any high level calls.
317 *
318 * Each cpu has its own sleep queue.
984263bc 319 */
984263bc 320void
fc17ad60 321sleep_gdinit(globaldata_t gd)
984263bc 322{
fc17ad60 323 static struct tslpque slpque_cpu0[TABLESIZE];
9c1fad94 324 int i;
984263bc 325
fc17ad60
MD
326 if (gd->gd_cpuid == 0) {
327 sched_quantum = (hz + 24) / 25;
fc17ad60
MD
328 gd->gd_tsleep_hash = slpque_cpu0;
329 } else {
77652cad 330 gd->gd_tsleep_hash = kmalloc(sizeof(slpque_cpu0),
fc17ad60
MD
331 M_TSLEEP, M_WAITOK | M_ZERO);
332 }
333 for (i = 0; i < TABLESIZE; ++i)
334 TAILQ_INIT(&gd->gd_tsleep_hash[i]);
984263bc
MD
335}
336
ae8e83e6
MD
337/*
338 * This is a dandy function that allows us to interlock tsleep/wakeup
339 * operations with unspecified upper level locks, such as lockmgr locks,
340 * simply by holding a critical section. The sequence is:
341 *
342 * (acquire upper level lock)
343 * tsleep_interlock(blah)
344 * (release upper level lock)
345 * tsleep(blah, ...)
346 *
347 * Basically this functions queues us on the tsleep queue without actually
348 * descheduling us. When tsleep() is later called with PINTERLOCK it
349 * assumes the thread was already queued, otherwise it queues it there.
350 *
351 * Thus it is possible to receive the wakeup prior to going to sleep and
352 * the race conditions are covered.
353 */
354static __inline void
5decebc7 355_tsleep_interlock(globaldata_t gd, const volatile void *ident, int flags)
ae8e83e6
MD
356{
357 thread_t td = gd->gd_curthread;
358 int id;
359
360 crit_enter_quick(td);
361 if (td->td_flags & TDF_TSLEEPQ) {
362 id = LOOKUP(td->td_wchan);
363 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_sleepq);
b12defdc 364 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL) {
c07315c4
MD
365 ATOMIC_CPUMASK_NANDBIT(slpque_cpumasks[id],
366 gd->gd_cpuid);
b12defdc 367 }
ae8e83e6
MD
368 } else {
369 td->td_flags |= TDF_TSLEEPQ;
370 }
371 id = LOOKUP(ident);
372 TAILQ_INSERT_TAIL(&gd->gd_tsleep_hash[id], td, td_sleepq);
c07315c4 373 ATOMIC_CPUMASK_ORBIT(slpque_cpumasks[id], gd->gd_cpuid);
ae8e83e6
MD
374 td->td_wchan = ident;
375 td->td_wdomain = flags & PDOMAIN_MASK;
ae8e83e6
MD
376 crit_exit_quick(td);
377}
378
379void
5decebc7 380tsleep_interlock(const volatile void *ident, int flags)
ae8e83e6
MD
381{
382 _tsleep_interlock(mycpu, ident, flags);
383}
384
385/*
386 * Remove thread from sleepq. Must be called with a critical section held.
4643740a 387 * The thread must not be migrating.
ae8e83e6
MD
388 */
389static __inline void
390_tsleep_remove(thread_t td)
391{
392 globaldata_t gd = mycpu;
393 int id;
394
957625b2 395 KKASSERT(td->td_gd == gd && IN_CRITICAL_SECT(td));
4643740a 396 KKASSERT((td->td_flags & TDF_MIGRATING) == 0);
ae8e83e6
MD
397 if (td->td_flags & TDF_TSLEEPQ) {
398 td->td_flags &= ~TDF_TSLEEPQ;
399 id = LOOKUP(td->td_wchan);
400 TAILQ_REMOVE(&gd->gd_tsleep_hash[id], td, td_sleepq);
c07315c4
MD
401 if (TAILQ_FIRST(&gd->gd_tsleep_hash[id]) == NULL) {
402 ATOMIC_CPUMASK_NANDBIT(slpque_cpumasks[id],
403 gd->gd_cpuid);
404 }
ae8e83e6
MD
405 td->td_wchan = NULL;
406 td->td_wdomain = 0;
407 }
408}
409
410void
411tsleep_remove(thread_t td)
412{
413 _tsleep_remove(td);
414}
415
984263bc
MD
416/*
417 * General sleep call. Suspends the current process until a wakeup is
418 * performed on the specified identifier. The process will then be made
419 * runnable with the specified priority. Sleeps at most timo/hz seconds
377d4740 420 * (0 means no timeout). If flags includes PCATCH flag, signals are checked
984263bc
MD
421 * before and after sleeping, else signals are not checked. Returns 0 if
422 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a
423 * signal needs to be delivered, ERESTART is returned if the current system
424 * call should be restarted if possible, and EINTR is returned if the system
425 * call should be interrupted by the signal (return EINTR).
26a0694b 426 *
0a3f9b47
MD
427 * Note that if we are a process, we release_curproc() before messing with
428 * the LWKT scheduler.
a46fac56
MD
429 *
430 * During autoconfiguration or after a panic, a sleep will simply
431 * lower the priority briefly to allow interrupts, then return.
94f98873
MD
432 *
433 * WARNING! This code can't block (short of switching away), or bad things
434 * will happen. No getting tokens, no blocking locks, etc.
984263bc
MD
435 */
436int
5decebc7 437tsleep(const volatile void *ident, int flags, const char *wmesg, int timo)
984263bc 438{
dadab5e9 439 struct thread *td = curthread;
08f2f1bb 440 struct lwp *lp = td->td_lwp;
0cfcada1 441 struct proc *p = td->td_proc; /* may be NULL */
fc17ad60 442 globaldata_t gd;
344ad853
MD
443 int sig;
444 int catch;
344ad853 445 int error;
e43a034f 446 int oldpri;
076fecef 447 struct callout thandle;
984263bc 448
b0da0c88
MD
449 /*
450 * Currently a severe hack. Make sure any delayed wakeups
451 * are flushed before we sleep or we might deadlock on whatever
452 * event we are sleeping on.
453 */
454 if (td->td_flags & TDF_DELAYED_WAKEUP)
455 wakeup_end_delayed();
456
0cfcada1
MD
457 /*
458 * NOTE: removed KTRPOINT, it could cause races due to blocking
459 * even in stable. Just scrap it for now.
460 */
5ea440eb 461 if (!tsleep_crypto_dump && (tsleep_now_works == 0 || panicstr)) {
984263bc 462 /*
dbcd0c9b
MD
463 * After a panic, or before we actually have an operational
464 * softclock, just give interrupts a chance, then just return;
465 *
984263bc
MD
466 * don't run any other procs or panic below,
467 * in case this is the idle process and already asleep.
468 */
e43a034f 469 splz();
f9235b6d 470 oldpri = td->td_pri;
e43a034f
MD
471 lwkt_setpri_self(safepri);
472 lwkt_switch();
473 lwkt_setpri_self(oldpri);
984263bc
MD
474 return (0);
475 }
8aa3430c 476 logtsleep2(tsleep_beg, ident);
fc17ad60
MD
477 gd = td->td_gd;
478 KKASSERT(td != &gd->gd_idlethread); /* you must be kidding! */
de4d4cb0 479 td->td_wakefromcpu = -1; /* overwritten by _wakeup */
344ad853
MD
480
481 /*
482 * NOTE: all of this occurs on the current cpu, including any
483 * callout-based wakeups, so a critical section is a sufficient
484 * interlock.
485 *
486 * The entire sequence through to where we actually sleep must
487 * run without breaking the critical section.
488 */
344ad853
MD
489 catch = flags & PCATCH;
490 error = 0;
491 sig = 0;
492
37af14fe 493 crit_enter_quick(td);
344ad853 494
0cfcada1 495 KASSERT(ident != NULL, ("tsleep: no ident"));
7278a846
SS
496 KASSERT(lp == NULL ||
497 lp->lwp_stat == LSRUN || /* Obvious */
498 lp->lwp_stat == LSSTOP, /* Set in tstop */
499 ("tsleep %p %s %d",
500 ident, wmesg, lp->lwp_stat));
0cfcada1 501
5686ec5a
MD
502 /*
503 * We interlock the sleep queue if the caller has not already done
504 * it for us. This must be done before we potentially acquire any
505 * tokens or we can loose the wakeup.
506 */
507 if ((flags & PINTERLOCKED) == 0) {
5686ec5a
MD
508 _tsleep_interlock(gd, ident, flags);
509 }
510
344ad853 511 /*
4643740a
MD
512 * Setup for the current process (if this is a process). We must
513 * interlock with lwp_token to avoid remote wakeup races via
514 * setrunnable()
344ad853 515 */
08f2f1bb 516 if (lp) {
4643740a 517 lwkt_gettoken(&lp->lwp_token);
344ad853
MD
518 if (catch) {
519 /*
520 * Early termination if PCATCH was set and a
521 * signal is pending, interlocked with the
522 * critical section.
523 *
524 * Early termination only occurs when tsleep() is
164b8401 525 * entered while in a normal LSRUN state.
344ad853 526 */
08f2f1bb 527 if ((sig = CURSIG(lp)) != 0)
344ad853
MD
528 goto resume;
529
530 /*
5686ec5a
MD
531 * Causes ksignal to wake us up if a signal is
532 * received (interlocked with p->p_token).
344ad853 533 */
4643740a 534 lp->lwp_flags |= LWP_SINTR;
344ad853 535 }
5686ec5a
MD
536 } else {
537 KKASSERT(p == NULL);
4ecd8190 538 }
344ad853 539
4ecd8190 540 /*
4ecd8190
MD
541 * Make sure the current process has been untangled from
542 * the userland scheduler and initialize slptime to start
5686ec5a 543 * counting.
c75e41b7
MD
544 *
545 * NOTE: td->td_wakefromcpu is pre-set by the release function
546 * for the dfly scheduler, and then adjusted by _wakeup()
4ecd8190
MD
547 */
548 if (lp) {
08f2f1bb
SS
549 p->p_usched->release_curproc(lp);
550 lp->lwp_slptime = 0;
0a3f9b47 551 }
fc17ad60 552
d9345d3a
MD
553 /*
554 * If the interlocked flag is set but our cpu bit in the slpqueue
555 * is no longer set, then a wakeup was processed inbetween the
4ecd8190
MD
556 * tsleep_interlock() (ours or the callers), and here. This can
557 * occur under numerous circumstances including when we release the
558 * current process.
d9345d3a 559 *
4ecd8190
MD
560 * Extreme loads can cause the sending of an IPI (e.g. wakeup()'s)
561 * to process incoming IPIs, thus draining incoming wakeups.
d9345d3a 562 */
4ecd8190
MD
563 if ((td->td_flags & TDF_TSLEEPQ) == 0) {
564 logtsleep2(ilockfail, ident);
565 goto resume;
d9345d3a 566 }
4ecd8190
MD
567
568 /*
569 * scheduling is blocked while in a critical section. Coincide
570 * the descheduled-by-tsleep flag with the descheduling of the
571 * lwkt.
8d446850
MD
572 *
573 * The timer callout is localized on our cpu and interlocked by
574 * our critical section.
4ecd8190 575 */
37af14fe 576 lwkt_deschedule_self(td);
ae8e83e6 577 td->td_flags |= TDF_TSLEEP_DESCHEDULED;
344ad853 578 td->td_wmesg = wmesg;
344ad853
MD
579
580 /*
8d446850
MD
581 * Setup the timeout, if any. The timeout is only operable while
582 * the thread is flagged descheduled.
344ad853 583 */
8d446850 584 KKASSERT((td->td_flags & TDF_TIMEOUT) == 0);
076fecef 585 if (timo) {
8d446850 586 callout_init_mp(&thandle);
076fecef
MD
587 callout_reset(&thandle, timo, endtsleep, td);
588 }
344ad853 589
984263bc 590 /*
344ad853 591 * Beddy bye bye.
984263bc 592 */
08f2f1bb 593 if (lp) {
26a0694b 594 /*
52eedfb5 595 * Ok, we are sleeping. Place us in the SSLEEP state.
26a0694b 596 */
4643740a 597 KKASSERT((lp->lwp_mpflags & LWP_MP_ONRUNQ) == 0);
6b4d33c2 598
7278a846
SS
599 /*
600 * tstop() sets LSSTOP, so don't fiddle with that.
601 */
602 if (lp->lwp_stat != LSSTOP)
603 lp->lwp_stat = LSSLEEP;
08f2f1bb 604 lp->lwp_ru.ru_nvcsw++;
e28d8b15 605 p->p_usched->uload_update(lp);
de4d4cb0 606 lwkt_switch();
ab554892
MD
607
608 /*
164b8401 609 * And when we are woken up, put us back in LSRUN. If we
ab554892
MD
610 * slept for over a second, recalculate our estcpu.
611 */
164b8401 612 lp->lwp_stat = LSRUN;
de4d4cb0
MD
613 if (lp->lwp_slptime) {
614 p->p_usched->uload_update(lp);
08f2f1bb 615 p->p_usched->recalculate(lp);
de4d4cb0 616 }
08f2f1bb 617 lp->lwp_slptime = 0;
0cfcada1
MD
618 } else {
619 lwkt_switch();
620 }
344ad853 621
fc17ad60
MD
622 /*
623 * Make sure we haven't switched cpus while we were asleep. It's
344ad853 624 * not supposed to happen. Cleanup our temporary flags.
fc17ad60
MD
625 */
626 KKASSERT(gd == td->td_gd);
344ad853
MD
627
628 /*
8d446850 629 * Cleanup the timeout. If the timeout has already occured thandle
4643740a
MD
630 * has already been stopped, otherwise stop thandle. If the timeout
631 * is running (the callout thread must be blocked trying to get
632 * lwp_token) then wait for us to get scheduled.
344ad853
MD
633 */
634 if (timo) {
4643740a
MD
635 while (td->td_flags & TDF_TIMEOUT_RUNNING) {
636 lwkt_deschedule_self(td);
637 td->td_wmesg = "tsrace";
638 lwkt_switch();
639 kprintf("td %p %s: timeout race\n", td, td->td_comm);
640 }
344ad853
MD
641 if (td->td_flags & TDF_TIMEOUT) {
642 td->td_flags &= ~TDF_TIMEOUT;
a40da8f0 643 error = EWOULDBLOCK;
344ad853 644 } else {
8d446850 645 /* does not block when on same cpu */
344ad853
MD
646 callout_stop(&thandle);
647 }
0cfcada1 648 }
4643740a 649 td->td_flags &= ~TDF_TSLEEP_DESCHEDULED;
344ad853
MD
650
651 /*
8d446850
MD
652 * Make sure we have been removed from the sleepq. In most
653 * cases this will have been done for us already but it is
654 * possible for a scheduling IPI to be in-flight from a
655 * previous tsleep/tsleep_interlock() or due to a straight-out
656 * call to lwkt_schedule() (in the case of an interrupt thread),
657 * causing a spurious wakeup.
344ad853 658 */
ae8e83e6 659 _tsleep_remove(td);
344ad853 660 td->td_wmesg = NULL;
344ad853
MD
661
662 /*
7c1212ec
MD
663 * Figure out the correct error return. If interrupted by a
664 * signal we want to return EINTR or ERESTART.
344ad853
MD
665 */
666resume:
4643740a 667 if (lp) {
7c1212ec 668 if (catch && error == 0) {
94f98873 669 if (sig != 0 || (sig = CURSIG(lp))) {
7c1212ec
MD
670 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
671 error = EINTR;
672 else
673 error = ERESTART;
674 }
984263bc 675 }
4643740a
MD
676 lp->lwp_flags &= ~LWP_SINTR;
677 lwkt_reltoken(&lp->lwp_token);
984263bc 678 }
8aa3430c 679 logtsleep1(tsleep_end);
344ad853
MD
680 crit_exit_quick(td);
681 return (error);
984263bc
MD
682}
683
bf765287
MD
684/*
685 * Interlocked spinlock sleep. An exclusively held spinlock must
e590ee86 686 * be passed to ssleep(). The function will atomically release the
bf765287
MD
687 * spinlock and tsleep on the ident, then reacquire the spinlock and
688 * return.
689 *
690 * This routine is fairly important along the critical path, so optimize it
691 * heavily.
692 */
693int
5decebc7 694ssleep(const volatile void *ident, struct spinlock *spin, int flags,
bf765287
MD
695 const char *wmesg, int timo)
696{
697 globaldata_t gd = mycpu;
698 int error;
16523a43 699
ae8e83e6 700 _tsleep_interlock(gd, ident, flags);
7cfe2b28 701 spin_unlock_quick(gd, spin);
ef48be0d 702 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
050032ec 703 _spin_lock_quick(gd, spin, wmesg);
bf765287
MD
704
705 return (error);
16523a43
MD
706}
707
bed060de 708int
5decebc7
MD
709lksleep(const volatile void *ident, struct lock *lock, int flags,
710 const char *wmesg, int timo)
bed060de
AH
711{
712 globaldata_t gd = mycpu;
713 int error;
714
715 _tsleep_interlock(gd, ident, flags);
716 lockmgr(lock, LK_RELEASE);
717 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
718 lockmgr(lock, LK_EXCLUSIVE);
719
720 return (error);
721}
722
7f6220a9
MD
723/*
724 * Interlocked mutex sleep. An exclusively held mutex must be passed
725 * to mtxsleep(). The function will atomically release the mutex
726 * and tsleep on the ident, then reacquire the mutex and return.
727 */
728int
5decebc7 729mtxsleep(const volatile void *ident, struct mtx *mtx, int flags,
7f6220a9
MD
730 const char *wmesg, int timo)
731{
732 globaldata_t gd = mycpu;
733 int error;
734
735 _tsleep_interlock(gd, ident, flags);
736 mtx_unlock(mtx);
737 error = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
738 mtx_lock_ex_quick(mtx, wmesg);
739
740 return (error);
741}
742
362e59be
SZ
743/*
744 * Interlocked serializer sleep. An exclusively held serializer must
ed3f6624 745 * be passed to zsleep(). The function will atomically release
362e59be
SZ
746 * the serializer and tsleep on the ident, then reacquire the serializer
747 * and return.
748 */
749int
5decebc7 750zsleep(const volatile void *ident, struct lwkt_serialize *slz, int flags,
ed3f6624 751 const char *wmesg, int timo)
362e59be 752{
ae8e83e6 753 globaldata_t gd = mycpu;
362e59be
SZ
754 int ret;
755
756 ASSERT_SERIALIZED(slz);
757
ae8e83e6 758 _tsleep_interlock(gd, ident, flags);
362e59be 759 lwkt_serialize_exit(slz);
ef48be0d 760 ret = tsleep(ident, flags | PINTERLOCKED, wmesg, timo);
362e59be 761 lwkt_serialize_enter(slz);
362e59be
SZ
762
763 return ret;
764}
765
a22c590e
MD
766/*
767 * Directly block on the LWKT thread by descheduling it. This
768 * is much faster then tsleep(), but the only legal way to wake
769 * us up is to directly schedule the thread.
770 *
771 * Setting TDF_SINTR will cause new signals to directly schedule us.
772 *
ae8e83e6 773 * This routine must be called while in a critical section.
a22c590e
MD
774 */
775int
776lwkt_sleep(const char *wmesg, int flags)
777{
778 thread_t td = curthread;
779 int sig;
780
781 if ((flags & PCATCH) == 0 || td->td_lwp == NULL) {
782 td->td_flags |= TDF_BLOCKED;
783 td->td_wmesg = wmesg;
784 lwkt_deschedule_self(td);
785 lwkt_switch();
786 td->td_wmesg = NULL;
787 td->td_flags &= ~TDF_BLOCKED;
788 return(0);
789 }
790 if ((sig = CURSIG(td->td_lwp)) != 0) {
791 if (SIGISMEMBER(td->td_proc->p_sigacts->ps_sigintr, sig))
792 return(EINTR);
793 else
794 return(ERESTART);
795
796 }
797 td->td_flags |= TDF_BLOCKED | TDF_SINTR;
798 td->td_wmesg = wmesg;
799 lwkt_deschedule_self(td);
800 lwkt_switch();
801 td->td_flags &= ~(TDF_BLOCKED | TDF_SINTR);
802 td->td_wmesg = NULL;
803 return(0);
804}
805
984263bc 806/*
344ad853 807 * Implement the timeout for tsleep.
fc17ad60 808 *
344ad853
MD
809 * This type of callout timeout is scheduled on the same cpu the process
810 * is sleeping on. Also, at the moment, the MP lock is held.
984263bc
MD
811 */
812static void
0cfcada1 813endtsleep(void *arg)
984263bc 814{
0cfcada1 815 thread_t td = arg;
9a379a4a 816 struct lwp *lp;
984263bc 817
8d446850 818 /*
4643740a
MD
819 * We are going to have to get the lwp_token, which means we might
820 * block. This can race a tsleep getting woken up by other means
821 * so set TDF_TIMEOUT_RUNNING to force the tsleep to wait for our
822 * processing to complete (sorry tsleep!).
823 *
824 * We can safely set td_flags because td MUST be on the same cpu
825 * as we are.
8d446850 826 */
4643740a
MD
827 KKASSERT(td->td_gd == mycpu);
828 crit_enter();
829 td->td_flags |= TDF_TIMEOUT_RUNNING | TDF_TIMEOUT;
344ad853
MD
830
831 /*
4643740a
MD
832 * This can block but TDF_TIMEOUT_RUNNING will prevent the thread
833 * from exiting the tsleep on us. The flag is interlocked by virtue
834 * of lp being on the same cpu as we are.
344ad853 835 */
8d446850 836 if ((lp = td->td_lwp) != NULL)
e2b148c6 837 lwkt_gettoken(&lp->lwp_token);
344ad853 838
4643740a
MD
839 KKASSERT(td->td_flags & TDF_TSLEEP_DESCHEDULED);
840
841 if (lp) {
d4688a1e
MD
842 /*
843 * callout timer should never be set in tstop() because
844 * it passes a timeout of 0.
845 */
846 KKASSERT(lp->lwp_stat != LSSTOP);
847 setrunnable(lp);
e2b148c6 848 lwkt_reltoken(&lp->lwp_token);
4643740a
MD
849 } else {
850 _tsleep_remove(td);
851 lwkt_schedule(td);
852 }
853 KKASSERT(td->td_gd == mycpu);
854 td->td_flags &= ~TDF_TIMEOUT_RUNNING;
37af14fe 855 crit_exit();
984263bc
MD
856}
857
8fb8bca6
EN
858/*
859 * Make all processes sleeping on the specified identifier runnable.
fc17ad60
MD
860 * count may be zero or one only.
861 *
c75e41b7
MD
862 * The domain encodes the sleep/wakeup domain, flags, plus the originating
863 * cpu.
344ad853
MD
864 *
865 * This call may run without the MP lock held. We can only manipulate thread
866 * state on the cpu owning the thread. We CANNOT manipulate process state
867 * at all.
5decebc7
MD
868 *
869 * _wakeup() can be passed to an IPI so we can't use (const volatile
870 * void *ident).
8fb8bca6
EN
871 */
872static void
fc17ad60 873_wakeup(void *ident, int domain)
984263bc 874{
fc17ad60 875 struct tslpque *qp;
0cfcada1
MD
876 struct thread *td;
877 struct thread *ntd;
fc17ad60 878 globaldata_t gd;
fc17ad60 879 cpumask_t mask;
fc17ad60 880 int id;
984263bc 881
37af14fe 882 crit_enter();
8aa3430c 883 logtsleep2(wakeup_beg, ident);
fc17ad60
MD
884 gd = mycpu;
885 id = LOOKUP(ident);
886 qp = &gd->gd_tsleep_hash[id];
984263bc 887restart:
0cfcada1 888 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) {
ae8e83e6 889 ntd = TAILQ_NEXT(td, td_sleepq);
fc17ad60
MD
890 if (td->td_wchan == ident &&
891 td->td_wdomain == (domain & PDOMAIN_MASK)
892 ) {
ae8e83e6
MD
893 KKASSERT(td->td_gd == gd);
894 _tsleep_remove(td);
c75e41b7 895 td->td_wakefromcpu = PWAKEUP_DECODE(domain);
ae8e83e6 896 if (td->td_flags & TDF_TSLEEP_DESCHEDULED) {
ae8e83e6
MD
897 lwkt_schedule(td);
898 if (domain & PWAKEUP_ONE)
899 goto done;
fc17ad60 900 }
0cfcada1 901 goto restart;
984263bc
MD
902 }
903 }
fc17ad60 904
fc17ad60
MD
905 /*
906 * We finished checking the current cpu but there still may be
907 * more work to do. Either wakeup_one was requested and no matching
908 * thread was found, or a normal wakeup was requested and we have
909 * to continue checking cpus.
910 *
fc17ad60
MD
911 * It should be noted that this scheme is actually less expensive then
912 * the old scheme when waking up multiple threads, since we send
913 * only one IPI message per target candidate which may then schedule
914 * multiple threads. Before we could have wound up sending an IPI
915 * message for each thread on the target cpu (!= current cpu) that
916 * needed to be woken up.
917 *
918 * NOTE: Wakeups occuring on remote cpus are asynchronous. This
919 * should be ok since we are passing idents in the IPI rather then
920 * thread pointers.
921 */
c07315c4
MD
922 if ((domain & PWAKEUP_MYCPU) == 0) {
923 mask = slpque_cpumasks[id];
924 CPUMASK_ANDMASK(mask, gd->gd_other_cpus);
925 if (CPUMASK_TESTNZERO(mask)) {
926 lwkt_send_ipiq2_mask(mask, _wakeup, ident,
927 domain | PWAKEUP_MYCPU);
928 }
fc17ad60 929 }
fc17ad60 930done:
8aa3430c 931 logtsleep1(wakeup_end);
37af14fe 932 crit_exit();
984263bc
MD
933}
934
b336a9b1
MD
935/*
936 * Wakeup all threads tsleep()ing on the specified ident, on all cpus
937 */
984263bc 938void
5decebc7 939wakeup(const volatile void *ident)
984263bc 940{
b0da0c88
MD
941 globaldata_t gd = mycpu;
942 thread_t td = gd->gd_curthread;
943
944 if (td && (td->td_flags & TDF_DELAYED_WAKEUP)) {
2e90abac
MD
945 /*
946 * If we are in a delayed wakeup section, record up to two wakeups in
947 * a per-CPU queue and issue them when we block or exit the delayed
948 * wakeup section.
949 */
950 if (atomic_cmpset_ptr(&gd->gd_delayed_wakeup[0], NULL, ident))
951 return;
952 if (atomic_cmpset_ptr(&gd->gd_delayed_wakeup[1], NULL, ident))
953 return;
954
955 ident = atomic_swap_ptr(__DEQUALIFY(volatile void **, &gd->gd_delayed_wakeup[1]),
956 __DEALL(ident));
957 ident = atomic_swap_ptr(__DEQUALIFY(volatile void **, &gd->gd_delayed_wakeup[0]),
958 __DEALL(ident));
b0da0c88 959 }
2e90abac 960
b0da0c88 961 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, gd->gd_cpuid));
0cfcada1 962}
984263bc 963
b336a9b1
MD
964/*
965 * Wakeup one thread tsleep()ing on the specified ident, on any cpu.
966 */
0cfcada1 967void
5decebc7 968wakeup_one(const volatile void *ident)
0cfcada1 969{
fc17ad60 970 /* XXX potentially round-robin the first responding cpu */
c75e41b7
MD
971 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mycpu->gd_cpuid) |
972 PWAKEUP_ONE);
da5fb9ef
MD
973}
974
b336a9b1
MD
975/*
976 * Wakeup threads tsleep()ing on the specified ident on the current cpu
977 * only.
978 */
979void
5decebc7 980wakeup_mycpu(const volatile void *ident)
b336a9b1 981{
c75e41b7
MD
982 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mycpu->gd_cpuid) |
983 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 */
c75e41b7
MD
994 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mycpu->gd_cpuid) |
995 PWAKEUP_MYCPU | PWAKEUP_ONE);
b336a9b1
MD
996}
997
998/*
999 * Wakeup all thread tsleep()ing on the specified ident on the specified cpu
1000 * only.
1001 */
1002void
5decebc7 1003wakeup_oncpu(globaldata_t gd, const volatile void *ident)
b336a9b1 1004{
c75e41b7 1005 globaldata_t mygd = mycpu;
b336a9b1 1006 if (gd == mycpu) {
c75e41b7
MD
1007 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mygd->gd_cpuid) |
1008 PWAKEUP_MYCPU);
b336a9b1 1009 } else {
c75e41b7
MD
1010 lwkt_send_ipiq2(gd, _wakeup, __DEALL(ident),
1011 PWAKEUP_ENCODE(0, mygd->gd_cpuid) |
1012 PWAKEUP_MYCPU);
b336a9b1
MD
1013 }
1014}
1015
1016/*
1017 * Wakeup one thread tsleep()ing on the specified ident on the specified cpu
1018 * only.
1019 */
1020void
5decebc7 1021wakeup_oncpu_one(globaldata_t gd, const volatile void *ident)
b336a9b1 1022{
c75e41b7
MD
1023 globaldata_t mygd = mycpu;
1024 if (gd == mygd) {
1025 _wakeup(__DEALL(ident), PWAKEUP_ENCODE(0, mygd->gd_cpuid) |
1026 PWAKEUP_MYCPU | PWAKEUP_ONE);
b336a9b1 1027 } else {
5decebc7 1028 lwkt_send_ipiq2(gd, _wakeup, __DEALL(ident),
c75e41b7 1029 PWAKEUP_ENCODE(0, mygd->gd_cpuid) |
5decebc7 1030 PWAKEUP_MYCPU | PWAKEUP_ONE);
b336a9b1
MD
1031 }
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
b0da0c88
MD
1056void
1057wakeup_start_delayed(void)
1058{
1059 globaldata_t gd = mycpu;
1060
1061 crit_enter();
1062 gd->gd_curthread->td_flags |= TDF_DELAYED_WAKEUP;
1063 crit_exit();
1064}
1065
1066void
1067wakeup_end_delayed(void)
1068{
1069 globaldata_t gd = mycpu;
1070
1071 if (gd->gd_curthread->td_flags & TDF_DELAYED_WAKEUP) {
1072 crit_enter();
1073 gd->gd_curthread->td_flags &= ~TDF_DELAYED_WAKEUP;
1074 if (gd->gd_delayed_wakeup[0] || gd->gd_delayed_wakeup[1]) {
1075 if (gd->gd_delayed_wakeup[0]) {
1076 wakeup(gd->gd_delayed_wakeup[0]);
1077 gd->gd_delayed_wakeup[0] = NULL;
1078 }
1079 if (gd->gd_delayed_wakeup[1]) {
1080 wakeup(gd->gd_delayed_wakeup[1]);
1081 gd->gd_delayed_wakeup[1] = NULL;
1082 }
1083 }
1084 crit_exit();
1085 }
1086}
1087
984263bc 1088/*
344ad853
MD
1089 * setrunnable()
1090 *
4643740a
MD
1091 * Make a process runnable. lp->lwp_token must be held on call and this
1092 * function must be called from the cpu owning lp.
37af14fe 1093 *
4643740a 1094 * This only has an effect if we are in LSSTOP or LSSLEEP.
984263bc
MD
1095 */
1096void
9a379a4a 1097setrunnable(struct lwp *lp)
984263bc 1098{
4643740a
MD
1099 thread_t td = lp->lwp_thread;
1100
e2b148c6 1101 ASSERT_LWKT_TOKEN_HELD(&lp->lwp_token);
4643740a 1102 KKASSERT(td->td_gd == mycpu);
344ad853 1103 crit_enter();
2daf83b0
SS
1104 if (lp->lwp_stat == LSSTOP)
1105 lp->lwp_stat = LSSLEEP;
4643740a
MD
1106 if (lp->lwp_stat == LSSLEEP) {
1107 _tsleep_remove(td);
1108 lwkt_schedule(td);
1109 } else if (td->td_flags & TDF_SINTR) {
1110 lwkt_schedule(td);
1111 }
344ad853 1112 crit_exit();
984263bc
MD
1113}
1114
1115/*
164b8401
SS
1116 * The process is stopped due to some condition, usually because p_stat is
1117 * set to SSTOP, but also possibly due to being traced.
fc17ad60 1118 *
4643740a
MD
1119 * Caller must hold p->p_token
1120 *
164b8401 1121 * NOTE! If the caller sets SSTOP, the caller must also clear P_WAITED
344ad853
MD
1122 * because the parent may check the child's status before the child actually
1123 * gets to this routine.
1124 *
9a379a4a 1125 * This routine is called with the current lwp only, typically just
4643740a
MD
1126 * before returning to userland if the process state is detected as
1127 * possibly being in a stopped state.
984263bc
MD
1128 */
1129void
9a379a4a 1130tstop(void)
984263bc 1131{
9a379a4a 1132 struct lwp *lp = curthread->td_lwp;
7278a846 1133 struct proc *p = lp->lwp_proc;
8c986a82 1134 struct proc *q;
9a379a4a 1135
4643740a 1136 lwkt_gettoken(&lp->lwp_token);
7278a846 1137 crit_enter();
4643740a 1138
f33e8653 1139 /*
4643740a 1140 * If LWP_MP_WSTOP is set, we were sleeping
f33e8653
SS
1141 * while our process was stopped. At this point
1142 * we were already counted as stopped.
1143 */
4643740a 1144 if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
f33e8653
SS
1145 /*
1146 * If we're the last thread to stop, signal
1147 * our parent.
1148 */
1149 p->p_nstopped++;
4643740a 1150 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
ea59a697 1151 wakeup(&p->p_nstopped);
f33e8653 1152 if (p->p_nstopped == p->p_nthreads) {
8c986a82
MD
1153 /*
1154 * Token required to interlock kern_wait()
1155 */
1156 q = p->p_pptr;
1157 PHOLD(q);
1158 lwkt_gettoken(&q->p_token);
4643740a 1159 p->p_flags &= ~P_WAITED;
f33e8653 1160 wakeup(p->p_pptr);
8c986a82
MD
1161 if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1162 ksignal(q, SIGCHLD);
1163 lwkt_reltoken(&q->p_token);
1164 PRELE(q);
f33e8653
SS
1165 }
1166 }
ea59a697 1167 while (p->p_stat == SSTOP) {
ea59a697
SS
1168 lp->lwp_stat = LSSTOP;
1169 tsleep(p, 0, "stop", 0);
1170 }
7278a846 1171 p->p_nstopped--;
4643740a 1172 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
7278a846 1173 crit_exit();
4643740a 1174 lwkt_reltoken(&lp->lwp_token);
26a0694b
MD
1175}
1176
984263bc
MD
1177/*
1178 * Compute a tenex style load average of a quantity on
1179 * 1, 5 and 15 minute intervals.
1180 */
c7e98b2f 1181static int loadav_count_runnable(struct lwp *p, void *data);
8fa76237 1182
984263bc
MD
1183static void
1184loadav(void *arg)
1185{
984263bc 1186 struct loadavg *avg;
8fa76237 1187 int i, nrun;
984263bc 1188
984263bc 1189 nrun = 0;
c7e98b2f 1190 alllwp_scan(loadav_count_runnable, &nrun);
8fa76237
MD
1191 avg = &averunnable;
1192 for (i = 0; i < 3; i++) {
984263bc
MD
1193 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
1194 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
8fa76237 1195 }
984263bc
MD
1196
1197 /*
1198 * Schedule the next update to occur after 5 seconds, but add a
1199 * random variation to avoid synchronisation with processes that
1200 * run at regular intervals.
1201 */
cddfb7bb 1202 callout_reset(&loadav_callout, hz * 4 + (int)(krandom() % (hz * 2 + 1)),
8fa76237
MD
1203 loadav, NULL);
1204}
1205
1206static int
c7e98b2f 1207loadav_count_runnable(struct lwp *lp, void *data)
8fa76237
MD
1208{
1209 int *nrunp = data;
1210 thread_t td;
1211
164b8401
SS
1212 switch (lp->lwp_stat) {
1213 case LSRUN:
08f2f1bb 1214 if ((td = lp->lwp_thread) == NULL)
8fa76237
MD
1215 break;
1216 if (td->td_flags & TDF_BLOCKED)
1217 break;
8fa76237
MD
1218 ++*nrunp;
1219 break;
1220 default:
1221 break;
1222 }
d2d8515b 1223 lwkt_yield();
8fa76237 1224 return(0);
984263bc
MD
1225}
1226
1227/* ARGSUSED */
1228static void
6656cd91 1229sched_setup(void *dummy)
984263bc 1230{
8d446850
MD
1231 callout_init_mp(&loadav_callout);
1232 callout_init_mp(&schedcpu_callout);
984263bc
MD
1233
1234 /* Kick off timeout driven events by calling first time. */
984263bc
MD
1235 schedcpu(NULL);
1236 loadav(NULL);
1237}
1238