kern_synch.c revision 167327
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 167327 2007-03-08 06:44:34Z julian $"); 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 mutex argument is unlocked before the caller is suspended, and 116 * re-locked before msleep returns. If priority includes the PDROP 117 * flag the mutex is not re-locked before returning. 118 */ 119int 120msleep(ident, mtx, priority, wmesg, timo) 121 void *ident; 122 struct mtx *mtx; 123 int priority, timo; 124 const char *wmesg; 125{ 126 struct thread *td; 127 struct proc *p; 128 int catch, rval, flags, pri; 129 WITNESS_SAVE_DECL(mtx); 130 131 td = curthread; 132 p = td->td_proc; 133#ifdef KTRACE 134 if (KTRPOINT(td, KTR_CSW)) 135 ktrcsw(1, 0); 136#endif 137 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL : 138 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg); 139 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL || 140 ident == &lbolt, ("sleeping without a mutex")); 141 KASSERT(p != NULL, ("msleep1")); 142 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 143 144 if (cold) { 145 /* 146 * During autoconfiguration, just return; 147 * don't run any other threads or panic below, 148 * in case this is the idle thread and already asleep. 149 * XXX: this used to do "s = splhigh(); splx(safepri); 150 * splx(s);" to give interrupts a chance, but there is 151 * no way to give interrupts a chance now. 152 */ 153 if (mtx != NULL && priority & PDROP) 154 mtx_unlock(mtx); 155 return (0); 156 } 157 catch = priority & PCATCH; 158 rval = 0; 159 160 /* 161 * If we are already on a sleep queue, then remove us from that 162 * sleep queue first. We have to do this to handle recursive 163 * sleeps. 164 */ 165 if (TD_ON_SLEEPQ(td)) 166 sleepq_remove(td, td->td_wchan); 167 168 if (ident == &pause_wchan) 169 flags = SLEEPQ_PAUSE; 170 else 171 flags = SLEEPQ_MSLEEP; 172 if (catch) 173 flags |= SLEEPQ_INTERRUPTIBLE; 174 175 sleepq_lock(ident); 176 CTR5(KTR_PROC, "msleep: thread %ld (pid %ld, %s) on %s (%p)", 177 td->td_tid, p->p_pid, p->p_comm, wmesg, ident); 178 179 DROP_GIANT(); 180 if (mtx != NULL) { 181 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 182 WITNESS_SAVE(&mtx->mtx_object, mtx); 183 mtx_unlock(mtx); 184 } 185 186 /* 187 * We put ourselves on the sleep queue and start our timeout 188 * before calling thread_suspend_check, as we could stop there, 189 * and a wakeup or a SIGCONT (or both) could occur while we were 190 * stopped without resuming us. Thus, we must be ready for sleep 191 * when cursig() is called. If the wakeup happens while we're 192 * stopped, then td will no longer be on a sleep queue upon 193 * return from cursig(). 194 */ 195 sleepq_add(ident, ident == &lbolt ? NULL : &mtx->mtx_object, wmesg, 196 flags, 0); 197 if (timo) 198 sleepq_set_timeout(ident, timo); 199 200 /* 201 * Adjust this thread's priority, if necessary. 202 */ 203 pri = priority & PRIMASK; 204 if (pri != 0 && pri != td->td_priority) { 205 mtx_lock_spin(&sched_lock); 206 sched_prio(td, pri); 207 mtx_unlock_spin(&sched_lock); 208 } 209 210 if (timo && catch) 211 rval = sleepq_timedwait_sig(ident); 212 else if (timo) 213 rval = sleepq_timedwait(ident); 214 else if (catch) 215 rval = sleepq_wait_sig(ident); 216 else { 217 sleepq_wait(ident); 218 rval = 0; 219 } 220#ifdef KTRACE 221 if (KTRPOINT(td, KTR_CSW)) 222 ktrcsw(0, 0); 223#endif 224 PICKUP_GIANT(); 225 if (mtx != NULL && !(priority & PDROP)) { 226 mtx_lock(mtx); 227 WITNESS_RESTORE(&mtx->mtx_object, mtx); 228 } 229 return (rval); 230} 231 232int 233msleep_spin(ident, mtx, wmesg, timo) 234 void *ident; 235 struct mtx *mtx; 236 const char *wmesg; 237 int timo; 238{ 239 struct thread *td; 240 struct proc *p; 241 int rval; 242 WITNESS_SAVE_DECL(mtx); 243 244 td = curthread; 245 p = td->td_proc; 246 KASSERT(mtx != NULL, ("sleeping without a mutex")); 247 KASSERT(p != NULL, ("msleep1")); 248 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 249 250 if (cold) { 251 /* 252 * During autoconfiguration, just return; 253 * don't run any other threads or panic below, 254 * in case this is the idle thread and already asleep. 255 * XXX: this used to do "s = splhigh(); splx(safepri); 256 * splx(s);" to give interrupts a chance, but there is 257 * no way to give interrupts a chance now. 258 */ 259 return (0); 260 } 261 262 sleepq_lock(ident); 263 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 264 td->td_tid, p->p_pid, p->p_comm, wmesg, ident); 265 266 DROP_GIANT(); 267 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 268 WITNESS_SAVE(&mtx->mtx_object, mtx); 269 mtx_unlock_spin(mtx); 270 271 /* 272 * We put ourselves on the sleep queue and start our timeout. 273 */ 274 sleepq_add(ident, &mtx->mtx_object, wmesg, SLEEPQ_MSLEEP, 0); 275 if (timo) 276 sleepq_set_timeout(ident, timo); 277 278 /* 279 * Can't call ktrace with any spin locks held so it can lock the 280 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 281 * any spin lock. Thus, we have to drop the sleepq spin lock while 282 * we handle those requests. This is safe since we have placed our 283 * thread on the sleep queue already. 284 */ 285#ifdef KTRACE 286 if (KTRPOINT(td, KTR_CSW)) { 287 sleepq_release(ident); 288 ktrcsw(1, 0); 289 sleepq_lock(ident); 290 } 291#endif 292#ifdef WITNESS 293 sleepq_release(ident); 294 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 295 wmesg); 296 sleepq_lock(ident); 297#endif 298 if (timo) 299 rval = sleepq_timedwait(ident); 300 else { 301 sleepq_wait(ident); 302 rval = 0; 303 } 304#ifdef KTRACE 305 if (KTRPOINT(td, KTR_CSW)) 306 ktrcsw(0, 0); 307#endif 308 PICKUP_GIANT(); 309 mtx_lock_spin(mtx); 310 WITNESS_RESTORE(&mtx->mtx_object, mtx); 311 return (rval); 312} 313 314/* 315 * pause() is like tsleep() except that the intention is to not be 316 * explicitly woken up by another thread. Instead, the current thread 317 * simply wishes to sleep until the timeout expires. It is 318 * implemented using a dummy wait channel. 319 */ 320int 321pause(wmesg, timo) 322 const char *wmesg; 323 int timo; 324{ 325 326 KASSERT(timo != 0, ("pause: timeout required")); 327 return (tsleep(&pause_wchan, 0, wmesg, timo)); 328} 329 330/* 331 * Make all threads sleeping on the specified identifier runnable. 332 */ 333void 334wakeup(ident) 335 register void *ident; 336{ 337 338 sleepq_lock(ident); 339 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1, 0); 340} 341 342/* 343 * Make a thread sleeping on the specified identifier runnable. 344 * May wake more than one thread if a target thread is currently 345 * swapped out. 346 */ 347void 348wakeup_one(ident) 349 register void *ident; 350{ 351 352 sleepq_lock(ident); 353 sleepq_signal(ident, SLEEPQ_MSLEEP, -1, 0); 354} 355 356/* 357 * The machine independent parts of context switching. 358 */ 359void 360mi_switch(int flags, struct thread *newtd) 361{ 362 uint64_t new_switchtime; 363 struct thread *td; 364 struct proc *p; 365 366 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 367 td = curthread; /* XXX */ 368 p = td->td_proc; /* XXX */ 369 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 370#ifdef INVARIANTS 371 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 372 mtx_assert(&Giant, MA_NOTOWNED); 373#endif 374 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 375 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 376 newtd == NULL) || panicstr, 377 ("mi_switch: switch in a critical section")); 378 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 379 ("mi_switch: switch must be voluntary or involuntary")); 380 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 381 382 /* 383 * Don't perform context switches from the debugger. 384 */ 385 if (kdb_active) { 386 mtx_unlock_spin(&sched_lock); 387 kdb_backtrace(); 388 kdb_reenter(); 389 panic("%s: did not reenter debugger", __func__); 390 } 391 392 if (flags & SW_VOL) 393 p->p_stats->p_ru.ru_nvcsw++; 394 else 395 p->p_stats->p_ru.ru_nivcsw++; 396 397 /* 398 * Compute the amount of time during which the current 399 * process was running, and add that to its total so far. 400 */ 401 new_switchtime = cpu_ticks(); 402 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime)); 403 p->p_rux.rux_uticks += td->td_uticks; 404 td->td_uticks = 0; 405 p->p_rux.rux_iticks += td->td_iticks; 406 td->td_iticks = 0; 407 p->p_rux.rux_sticks += td->td_sticks; 408 td->td_sticks = 0; 409 410 td->td_generation++; /* bump preempt-detect counter */ 411 412 /* 413 * Check if the process exceeds its cpu resource allocation. If 414 * it reaches the max, arrange to kill the process in ast(). 415 */ 416 if (p->p_cpulimit != RLIM_INFINITY && 417 p->p_rux.rux_runtime >= p->p_cpulimit * cpu_tickrate()) { 418 p->p_sflag |= PS_XCPU; 419 td->td_flags |= TDF_ASTPENDING; 420 } 421 422 /* 423 * Finish up stats for outgoing thread. 424 */ 425 cnt.v_swtch++; 426 PCPU_SET(switchtime, new_switchtime); 427 PCPU_SET(switchticks, ticks); 428 CTR4(KTR_PROC, "mi_switch: old thread %ld (kse %p, pid %ld, %s)", 429 td->td_tid, td->td_sched, p->p_pid, p->p_comm); 430#if (KTR_COMPILE & KTR_SCHED) != 0 431 if (TD_IS_IDLETHREAD(td)) 432 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 433 td, td->td_proc->p_comm, td->td_priority); 434 else if (newtd != NULL) 435 CTR5(KTR_SCHED, 436 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 437 td, td->td_proc->p_comm, td->td_priority, newtd, 438 newtd->td_proc->p_comm); 439 else 440 CTR6(KTR_SCHED, 441 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 442 td, td->td_proc->p_comm, td->td_priority, 443 td->td_inhibitors, td->td_wmesg, td->td_lockname); 444#endif 445 /* 446 * We call thread_switchout after the KTR_SCHED prints above so kse 447 * selecting a new thread to run does not show up as a preemption. 448 */ 449#ifdef KSE 450 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA)) 451 newtd = thread_switchout(td, flags, newtd); 452#endif 453 sched_switch(td, newtd, flags); 454 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 455 td, td->td_proc->p_comm, td->td_priority); 456 457 CTR4(KTR_PROC, "mi_switch: new thread %ld (kse %p, pid %ld, %s)", 458 td->td_tid, td->td_sched, p->p_pid, p->p_comm); 459 460 /* 461 * If the last thread was exiting, finish cleaning it up. 462 */ 463 if ((td = PCPU_GET(deadthread))) { 464 PCPU_SET(deadthread, NULL); 465 thread_stash(td); 466 } 467} 468 469/* 470 * Change process state to be runnable, 471 * placing it on the run queue if it is in memory, 472 * and awakening the swapper if it isn't in memory. 473 */ 474void 475setrunnable(struct thread *td) 476{ 477 struct proc *p; 478 479 p = td->td_proc; 480 mtx_assert(&sched_lock, MA_OWNED); 481 switch (p->p_state) { 482 case PRS_ZOMBIE: 483 panic("setrunnable(1)"); 484 default: 485 break; 486 } 487 switch (td->td_state) { 488 case TDS_RUNNING: 489 case TDS_RUNQ: 490 return; 491 case TDS_INHIBITED: 492 /* 493 * If we are only inhibited because we are swapped out 494 * then arange to swap in this process. Otherwise just return. 495 */ 496 if (td->td_inhibitors != TDI_SWAPPED) 497 return; 498 /* XXX: intentional fall-through ? */ 499 case TDS_CAN_RUN: 500 break; 501 default: 502 printf("state is 0x%x", td->td_state); 503 panic("setrunnable(2)"); 504 } 505 if ((p->p_sflag & PS_INMEM) == 0) { 506 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 507 p->p_sflag |= PS_SWAPINREQ; 508 /* 509 * due to a LOR between sched_lock and 510 * the sleepqueue chain locks, use 511 * lower level scheduling functions. 512 */ 513 kick_proc0(); 514 } 515 } else 516 sched_wakeup(td); 517} 518 519/* 520 * Compute a tenex style load average of a quantity on 521 * 1, 5 and 15 minute intervals. 522 * XXXKSE Needs complete rewrite when correct info is available. 523 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 524 */ 525static void 526loadav(void *arg) 527{ 528 int i, nrun; 529 struct loadavg *avg; 530 531 nrun = sched_load(); 532 avg = &averunnable; 533 534 for (i = 0; i < 3; i++) 535 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 536 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 537 538 /* 539 * Schedule the next update to occur after 5 seconds, but add a 540 * random variation to avoid synchronisation with processes that 541 * run at regular intervals. 542 */ 543 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 544 loadav, NULL); 545} 546 547static void 548lboltcb(void *arg) 549{ 550 wakeup(&lbolt); 551 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 552} 553 554/* ARGSUSED */ 555static void 556synch_setup(dummy) 557 void *dummy; 558{ 559 callout_init(&loadav_callout, CALLOUT_MPSAFE); 560 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 561 562 /* Kick off timeout driven events by calling first time. */ 563 loadav(NULL); 564 lboltcb(NULL); 565} 566 567/* 568 * General purpose yield system call. 569 */ 570int 571yield(struct thread *td, struct yield_args *uap) 572{ 573 mtx_assert(&Giant, MA_NOTOWNED); 574 (void)uap; 575 sched_relinquish(td); 576 return (0); 577} 578