kern_synch.c revision 167387
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: head/sys/kern/kern_synch.c 167387 2007-03-09 22:41:01Z jhb $"); 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 67static void synch_setup(void *dummy); 68SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL) 69 70int hogticks; 71int lbolt; 72static int pause_wchan; 73 74static struct callout loadav_callout; 75static struct callout lbolt_callout; 76 77struct loadavg averunnable = 78 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 79/* 80 * Constants for averages over 1, 5, and 15 minutes 81 * when sampling at 5 second intervals. 82 */ 83static fixpt_t cexp[3] = { 84 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 85 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 86 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 87}; 88 89/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 90static int fscale __unused = FSCALE; 91SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 92 93static void loadav(void *arg); 94static void lboltcb(void *arg); 95 96void 97sleepinit(void) 98{ 99 100 hogticks = (hz / 10) * 2; /* Default only. */ 101 init_sleepqueues(); 102} 103 104/* 105 * General sleep call. Suspends the current thread until a wakeup is 106 * performed on the specified identifier. The thread will then be made 107 * runnable with the specified priority. Sleeps at most timo/hz seconds 108 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 109 * before and after sleeping, else signals are not checked. Returns 0 if 110 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 111 * signal needs to be delivered, ERESTART is returned if the current system 112 * call should be restarted if possible, and EINTR is returned if the system 113 * call should be interrupted by the signal (return EINTR). 114 * 115 * The lock argument is unlocked before the caller is suspended, and 116 * re-locked before _sleep() returns. If priority includes the PDROP 117 * flag the lock is not re-locked before returning. 118 */ 119int 120_sleep(ident, lock, priority, wmesg, timo) 121 void *ident; 122 struct lock_object *lock; 123 int priority, timo; 124 const char *wmesg; 125{ 126 struct thread *td; 127 struct proc *p; 128 struct lock_class *class; 129 int catch, flags, lock_state, pri, rval; 130 WITNESS_SAVE_DECL(lock_witness); 131 132 td = curthread; 133 p = td->td_proc; 134#ifdef KTRACE 135 if (KTRPOINT(td, KTR_CSW)) 136 ktrcsw(1, 0); 137#endif 138 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 139 "Sleeping on \"%s\"", wmesg); 140 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL || 141 ident == &lbolt, ("sleeping without a lock")); 142 KASSERT(p != NULL, ("msleep1")); 143 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 144 if (lock != NULL) 145 class = LOCK_CLASS(lock); 146 else 147 class = NULL; 148 149 if (cold) { 150 /* 151 * During autoconfiguration, just return; 152 * don't run any other threads or panic below, 153 * in case this is the idle thread and already asleep. 154 * XXX: this used to do "s = splhigh(); splx(safepri); 155 * splx(s);" to give interrupts a chance, but there is 156 * no way to give interrupts a chance now. 157 */ 158 if (lock != NULL && priority & PDROP) 159 class->lc_unlock(lock); 160 return (0); 161 } 162 catch = priority & PCATCH; 163 rval = 0; 164 165 /* 166 * If we are already on a sleep queue, then remove us from that 167 * sleep queue first. We have to do this to handle recursive 168 * sleeps. 169 */ 170 if (TD_ON_SLEEPQ(td)) 171 sleepq_remove(td, td->td_wchan); 172 173 if (ident == &pause_wchan) 174 flags = SLEEPQ_PAUSE; 175 else 176 flags = SLEEPQ_SLEEP; 177 if (catch) 178 flags |= SLEEPQ_INTERRUPTIBLE; 179 180 sleepq_lock(ident); 181 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 182 td->td_tid, p->p_pid, p->p_comm, wmesg, ident); 183 184 DROP_GIANT(); 185 if (lock != NULL) { 186 WITNESS_SAVE(lock, lock_witness); 187 lock_state = class->lc_unlock(lock); 188 } else 189 /* GCC needs to follow the Yellow Brick Road */ 190 lock_state = -1; 191 192 /* 193 * We put ourselves on the sleep queue and start our timeout 194 * before calling thread_suspend_check, as we could stop there, 195 * and a wakeup or a SIGCONT (or both) could occur while we were 196 * stopped without resuming us. Thus, we must be ready for sleep 197 * when cursig() is called. If the wakeup happens while we're 198 * stopped, then td will no longer be on a sleep queue upon 199 * return from cursig(). 200 */ 201 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0); 202 if (timo) 203 sleepq_set_timeout(ident, timo); 204 205 /* 206 * Adjust this thread's priority, if necessary. 207 */ 208 pri = priority & PRIMASK; 209 if (pri != 0 && pri != td->td_priority) { 210 mtx_lock_spin(&sched_lock); 211 sched_prio(td, pri); 212 mtx_unlock_spin(&sched_lock); 213 } 214 215 if (timo && catch) 216 rval = sleepq_timedwait_sig(ident); 217 else if (timo) 218 rval = sleepq_timedwait(ident); 219 else if (catch) 220 rval = sleepq_wait_sig(ident); 221 else { 222 sleepq_wait(ident); 223 rval = 0; 224 } 225#ifdef KTRACE 226 if (KTRPOINT(td, KTR_CSW)) 227 ktrcsw(0, 0); 228#endif 229 PICKUP_GIANT(); 230 if (lock != NULL && !(priority & PDROP)) { 231 class->lc_lock(lock, lock_state); 232 WITNESS_RESTORE(lock, lock_witness); 233 } 234 return (rval); 235} 236 237int 238msleep_spin(ident, mtx, wmesg, timo) 239 void *ident; 240 struct mtx *mtx; 241 const char *wmesg; 242 int timo; 243{ 244 struct thread *td; 245 struct proc *p; 246 int rval; 247 WITNESS_SAVE_DECL(mtx); 248 249 td = curthread; 250 p = td->td_proc; 251 KASSERT(mtx != NULL, ("sleeping without a mutex")); 252 KASSERT(p != NULL, ("msleep1")); 253 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 254 255 if (cold) { 256 /* 257 * During autoconfiguration, just return; 258 * don't run any other threads or panic below, 259 * in case this is the idle thread and already asleep. 260 * XXX: this used to do "s = splhigh(); splx(safepri); 261 * splx(s);" to give interrupts a chance, but there is 262 * no way to give interrupts a chance now. 263 */ 264 return (0); 265 } 266 267 sleepq_lock(ident); 268 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 269 td->td_tid, p->p_pid, p->p_comm, wmesg, ident); 270 271 DROP_GIANT(); 272 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 273 WITNESS_SAVE(&mtx->mtx_object, mtx); 274 mtx_unlock_spin(mtx); 275 276 /* 277 * We put ourselves on the sleep queue and start our timeout. 278 */ 279 sleepq_add(ident, &mtx->mtx_object, wmesg, SLEEPQ_SLEEP, 0); 280 if (timo) 281 sleepq_set_timeout(ident, timo); 282 283 /* 284 * Can't call ktrace with any spin locks held so it can lock the 285 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 286 * any spin lock. Thus, we have to drop the sleepq spin lock while 287 * we handle those requests. This is safe since we have placed our 288 * thread on the sleep queue already. 289 */ 290#ifdef KTRACE 291 if (KTRPOINT(td, KTR_CSW)) { 292 sleepq_release(ident); 293 ktrcsw(1, 0); 294 sleepq_lock(ident); 295 } 296#endif 297#ifdef WITNESS 298 sleepq_release(ident); 299 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 300 wmesg); 301 sleepq_lock(ident); 302#endif 303 if (timo) 304 rval = sleepq_timedwait(ident); 305 else { 306 sleepq_wait(ident); 307 rval = 0; 308 } 309#ifdef KTRACE 310 if (KTRPOINT(td, KTR_CSW)) 311 ktrcsw(0, 0); 312#endif 313 PICKUP_GIANT(); 314 mtx_lock_spin(mtx); 315 WITNESS_RESTORE(&mtx->mtx_object, mtx); 316 return (rval); 317} 318 319/* 320 * pause() is like tsleep() except that the intention is to not be 321 * explicitly woken up by another thread. Instead, the current thread 322 * simply wishes to sleep until the timeout expires. It is 323 * implemented using a dummy wait channel. 324 */ 325int 326pause(wmesg, timo) 327 const char *wmesg; 328 int timo; 329{ 330 331 KASSERT(timo != 0, ("pause: timeout required")); 332 return (tsleep(&pause_wchan, 0, wmesg, timo)); 333} 334 335/* 336 * Make all threads sleeping on the specified identifier runnable. 337 */ 338void 339wakeup(ident) 340 register void *ident; 341{ 342 343 sleepq_lock(ident); 344 sleepq_broadcast(ident, SLEEPQ_SLEEP, -1, 0); 345} 346 347/* 348 * Make a thread sleeping on the specified identifier runnable. 349 * May wake more than one thread if a target thread is currently 350 * swapped out. 351 */ 352void 353wakeup_one(ident) 354 register void *ident; 355{ 356 357 sleepq_lock(ident); 358 sleepq_signal(ident, SLEEPQ_SLEEP, -1, 0); 359} 360 361/* 362 * The machine independent parts of context switching. 363 */ 364void 365mi_switch(int flags, struct thread *newtd) 366{ 367 uint64_t new_switchtime; 368 struct thread *td; 369 struct proc *p; 370 371 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 372 td = curthread; /* XXX */ 373 p = td->td_proc; /* XXX */ 374 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 375#ifdef INVARIANTS 376 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 377 mtx_assert(&Giant, MA_NOTOWNED); 378#endif 379 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 380 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 381 newtd == NULL) || panicstr, 382 ("mi_switch: switch in a critical section")); 383 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 384 ("mi_switch: switch must be voluntary or involuntary")); 385 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 386 387 /* 388 * Don't perform context switches from the debugger. 389 */ 390 if (kdb_active) { 391 mtx_unlock_spin(&sched_lock); 392 kdb_backtrace(); 393 kdb_reenter(); 394 panic("%s: did not reenter debugger", __func__); 395 } 396 397 if (flags & SW_VOL) 398 p->p_stats->p_ru.ru_nvcsw++; 399 else 400 p->p_stats->p_ru.ru_nivcsw++; 401 402 /* 403 * Compute the amount of time during which the current 404 * process was running, and add that to its total so far. 405 */ 406 new_switchtime = cpu_ticks(); 407 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime)); 408 p->p_rux.rux_uticks += td->td_uticks; 409 td->td_uticks = 0; 410 p->p_rux.rux_iticks += td->td_iticks; 411 td->td_iticks = 0; 412 p->p_rux.rux_sticks += td->td_sticks; 413 td->td_sticks = 0; 414 415 td->td_generation++; /* bump preempt-detect counter */ 416 417 /* 418 * Check if the process exceeds its cpu resource allocation. If 419 * it reaches the max, arrange to kill the process in ast(). 420 */ 421 if (p->p_cpulimit != RLIM_INFINITY && 422 p->p_rux.rux_runtime >= p->p_cpulimit * cpu_tickrate()) { 423 p->p_sflag |= PS_XCPU; 424 td->td_flags |= TDF_ASTPENDING; 425 } 426 427 /* 428 * Finish up stats for outgoing thread. 429 */ 430 cnt.v_swtch++; 431 PCPU_SET(switchtime, new_switchtime); 432 PCPU_SET(switchticks, ticks); 433 CTR4(KTR_PROC, "mi_switch: old thread %ld (kse %p, pid %ld, %s)", 434 td->td_tid, td->td_sched, p->p_pid, p->p_comm); 435#if (KTR_COMPILE & KTR_SCHED) != 0 436 if (TD_IS_IDLETHREAD(td)) 437 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 438 td, td->td_proc->p_comm, td->td_priority); 439 else if (newtd != NULL) 440 CTR5(KTR_SCHED, 441 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 442 td, td->td_proc->p_comm, td->td_priority, newtd, 443 newtd->td_proc->p_comm); 444 else 445 CTR6(KTR_SCHED, 446 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 447 td, td->td_proc->p_comm, td->td_priority, 448 td->td_inhibitors, td->td_wmesg, td->td_lockname); 449#endif 450 /* 451 * We call thread_switchout after the KTR_SCHED prints above so kse 452 * selecting a new thread to run does not show up as a preemption. 453 */ 454#ifdef KSE 455 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA)) 456 newtd = thread_switchout(td, flags, newtd); 457#endif 458 sched_switch(td, newtd, flags); 459 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 460 td, td->td_proc->p_comm, td->td_priority); 461 462 CTR4(KTR_PROC, "mi_switch: new thread %ld (kse %p, pid %ld, %s)", 463 td->td_tid, td->td_sched, p->p_pid, p->p_comm); 464 465 /* 466 * If the last thread was exiting, finish cleaning it up. 467 */ 468 if ((td = PCPU_GET(deadthread))) { 469 PCPU_SET(deadthread, NULL); 470 thread_stash(td); 471 } 472} 473 474/* 475 * Change process state to be runnable, 476 * placing it on the run queue if it is in memory, 477 * and awakening the swapper if it isn't in memory. 478 */ 479void 480setrunnable(struct thread *td) 481{ 482 struct proc *p; 483 484 p = td->td_proc; 485 mtx_assert(&sched_lock, MA_OWNED); 486 switch (p->p_state) { 487 case PRS_ZOMBIE: 488 panic("setrunnable(1)"); 489 default: 490 break; 491 } 492 switch (td->td_state) { 493 case TDS_RUNNING: 494 case TDS_RUNQ: 495 return; 496 case TDS_INHIBITED: 497 /* 498 * If we are only inhibited because we are swapped out 499 * then arange to swap in this process. Otherwise just return. 500 */ 501 if (td->td_inhibitors != TDI_SWAPPED) 502 return; 503 /* XXX: intentional fall-through ? */ 504 case TDS_CAN_RUN: 505 break; 506 default: 507 printf("state is 0x%x", td->td_state); 508 panic("setrunnable(2)"); 509 } 510 if ((p->p_sflag & PS_INMEM) == 0) { 511 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 512 p->p_sflag |= PS_SWAPINREQ; 513 /* 514 * due to a LOR between sched_lock and 515 * the sleepqueue chain locks, use 516 * lower level scheduling functions. 517 */ 518 kick_proc0(); 519 } 520 } else 521 sched_wakeup(td); 522} 523 524/* 525 * Compute a tenex style load average of a quantity on 526 * 1, 5 and 15 minute intervals. 527 * XXXKSE Needs complete rewrite when correct info is available. 528 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 529 */ 530static void 531loadav(void *arg) 532{ 533 int i, nrun; 534 struct loadavg *avg; 535 536 nrun = sched_load(); 537 avg = &averunnable; 538 539 for (i = 0; i < 3; i++) 540 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 541 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 542 543 /* 544 * Schedule the next update to occur after 5 seconds, but add a 545 * random variation to avoid synchronisation with processes that 546 * run at regular intervals. 547 */ 548 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 549 loadav, NULL); 550} 551 552static void 553lboltcb(void *arg) 554{ 555 wakeup(&lbolt); 556 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 557} 558 559/* ARGSUSED */ 560static void 561synch_setup(dummy) 562 void *dummy; 563{ 564 callout_init(&loadav_callout, CALLOUT_MPSAFE); 565 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 566 567 /* Kick off timeout driven events by calling first time. */ 568 loadav(NULL); 569 lboltcb(NULL); 570} 571 572/* 573 * General purpose yield system call. 574 */ 575int 576yield(struct thread *td, struct yield_args *uap) 577{ 578 mtx_assert(&Giant, MA_NOTOWNED); 579 (void)uap; 580 sched_relinquish(td); 581 return (0); 582} 583