kern_synch.c revision 153855
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 153855 2005-12-29 20:57:45Z 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; 72 73static struct callout loadav_callout; 74static struct callout lbolt_callout; 75 76struct loadavg averunnable = 77 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 78/* 79 * Constants for averages over 1, 5, and 15 minutes 80 * when sampling at 5 second intervals. 81 */ 82static fixpt_t cexp[3] = { 83 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 84 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 85 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 86}; 87 88/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 89static int fscale __unused = FSCALE; 90SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 91 92static void loadav(void *arg); 93static void lboltcb(void *arg); 94 95void 96sleepinit(void) 97{ 98 99 hogticks = (hz / 10) * 2; /* Default only. */ 100 init_sleepqueues(); 101} 102 103/* 104 * General sleep call. Suspends the current process until a wakeup is 105 * performed on the specified identifier. The process will then be made 106 * runnable with the specified priority. Sleeps at most timo/hz seconds 107 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 108 * before and after sleeping, else signals are not checked. Returns 0 if 109 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 110 * signal needs to be delivered, ERESTART is returned if the current system 111 * call should be restarted if possible, and EINTR is returned if the system 112 * call should be interrupted by the signal (return EINTR). 113 * 114 * The mutex argument is exited before the caller is suspended, and 115 * entered before msleep returns. If priority includes the PDROP 116 * flag the mutex is not entered before returning. 117 */ 118int 119msleep(ident, mtx, priority, wmesg, timo) 120 void *ident; 121 struct mtx *mtx; 122 int priority, timo; 123 const char *wmesg; 124{ 125 struct thread *td; 126 struct proc *p; 127 int catch, rval, sig, flags; 128 WITNESS_SAVE_DECL(mtx); 129 130 td = curthread; 131 p = td->td_proc; 132#ifdef KTRACE 133 if (KTRPOINT(td, KTR_CSW)) 134 ktrcsw(1, 0); 135#endif 136 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL : 137 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg); 138 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL, 139 ("sleeping without a mutex")); 140 KASSERT(p != NULL, ("msleep1")); 141 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 142 143 if (cold) { 144 /* 145 * During autoconfiguration, just return; 146 * don't run any other threads or panic below, 147 * in case this is the idle thread and already asleep. 148 * XXX: this used to do "s = splhigh(); splx(safepri); 149 * splx(s);" to give interrupts a chance, but there is 150 * no way to give interrupts a chance now. 151 */ 152 if (mtx != NULL && priority & PDROP) 153 mtx_unlock(mtx); 154 return (0); 155 } 156 catch = priority & PCATCH; 157 rval = 0; 158 159 /* 160 * If we are already on a sleep queue, then remove us from that 161 * sleep queue first. We have to do this to handle recursive 162 * sleeps. 163 */ 164 if (TD_ON_SLEEPQ(td)) 165 sleepq_remove(td, td->td_wchan); 166 167 sleepq_lock(ident); 168 if (catch) { 169 /* 170 * Don't bother sleeping if we are exiting and not the exiting 171 * thread or if our thread is marked as interrupted. 172 */ 173 mtx_lock_spin(&sched_lock); 174 rval = thread_sleep_check(td); 175 mtx_unlock_spin(&sched_lock); 176 if (rval != 0) { 177 sleepq_release(ident); 178 if (mtx != NULL && priority & PDROP) 179 mtx_unlock(mtx); 180 return (rval); 181 } 182 } 183 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)", 184 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 185 186 DROP_GIANT(); 187 if (mtx != NULL) { 188 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 189 WITNESS_SAVE(&mtx->mtx_object, mtx); 190 mtx_unlock(mtx); 191 } 192 193 /* 194 * We put ourselves on the sleep queue and start our timeout 195 * before calling thread_suspend_check, as we could stop there, 196 * and a wakeup or a SIGCONT (or both) could occur while we were 197 * stopped without resuming us. Thus, we must be ready for sleep 198 * when cursig() is called. If the wakeup happens while we're 199 * stopped, then td will no longer be on a sleep queue upon 200 * return from cursig(). 201 */ 202 flags = SLEEPQ_MSLEEP; 203 if (catch) 204 flags |= SLEEPQ_INTERRUPTIBLE; 205 sleepq_add(ident, mtx, wmesg, flags); 206 if (timo) 207 sleepq_set_timeout(ident, timo); 208 if (catch) { 209 sig = sleepq_catch_signals(ident); 210 } else 211 sig = 0; 212 213 /* 214 * Adjust this thread's priority. 215 */ 216 mtx_lock_spin(&sched_lock); 217 sched_prio(td, priority & PRIMASK); 218 mtx_unlock_spin(&sched_lock); 219 220 if (timo && catch) 221 rval = sleepq_timedwait_sig(ident, sig != 0); 222 else if (timo) 223 rval = sleepq_timedwait(ident); 224 else if (catch) 225 rval = sleepq_wait_sig(ident); 226 else { 227 sleepq_wait(ident); 228 rval = 0; 229 } 230 if (rval == 0 && catch) 231 rval = sleepq_calc_signal_retval(sig); 232#ifdef KTRACE 233 if (KTRPOINT(td, KTR_CSW)) 234 ktrcsw(0, 0); 235#endif 236 PICKUP_GIANT(); 237 if (mtx != NULL && !(priority & PDROP)) { 238 mtx_lock(mtx); 239 WITNESS_RESTORE(&mtx->mtx_object, mtx); 240 } 241 return (rval); 242} 243 244int 245msleep_spin(ident, mtx, wmesg, timo) 246 void *ident; 247 struct mtx *mtx; 248 const char *wmesg; 249 int timo; 250{ 251 struct thread *td; 252 struct proc *p; 253 int rval; 254 WITNESS_SAVE_DECL(mtx); 255 256 td = curthread; 257 p = td->td_proc; 258 KASSERT(mtx != NULL, ("sleeping without a mutex")); 259 KASSERT(p != NULL, ("msleep1")); 260 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 261 262 if (cold) { 263 /* 264 * During autoconfiguration, just return; 265 * don't run any other threads or panic below, 266 * in case this is the idle thread and already asleep. 267 * XXX: this used to do "s = splhigh(); splx(safepri); 268 * splx(s);" to give interrupts a chance, but there is 269 * no way to give interrupts a chance now. 270 */ 271 return (0); 272 } 273 274 sleepq_lock(ident); 275 CTR5(KTR_PROC, "msleep_spin: thread %p (pid %ld, %s) on %s (%p)", 276 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 277 278 DROP_GIANT(); 279 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 280 WITNESS_SAVE(&mtx->mtx_object, mtx); 281 mtx_unlock_spin(mtx); 282 283 /* 284 * We put ourselves on the sleep queue and start our timeout. 285 */ 286 sleepq_add(ident, mtx, wmesg, SLEEPQ_MSLEEP); 287 if (timo) 288 sleepq_set_timeout(ident, timo); 289 290 /* 291 * Can't call ktrace with any spin locks held so it can lock the 292 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 293 * any spin lock. Thus, we have to drop the sleepq spin lock while 294 * we handle those requests. This is safe since we have placed our 295 * thread on the sleep queue already. 296 */ 297#ifdef KTRACE 298 if (KTRPOINT(td, KTR_CSW)) { 299 sleepq_release(ident); 300 ktrcsw(1, 0); 301 sleepq_lock(ident); 302 } 303#endif 304#ifdef WITNESS 305 sleepq_release(ident); 306 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 307 wmesg); 308 sleepq_lock(ident); 309#endif 310 if (timo) 311 rval = sleepq_timedwait(ident); 312 else { 313 sleepq_wait(ident); 314 rval = 0; 315 } 316#ifdef KTRACE 317 if (KTRPOINT(td, KTR_CSW)) 318 ktrcsw(0, 0); 319#endif 320 PICKUP_GIANT(); 321 mtx_lock_spin(mtx); 322 WITNESS_RESTORE(&mtx->mtx_object, mtx); 323 return (rval); 324} 325 326int 327msleep_spin(ident, mtx, wmesg, timo) 328 void *ident; 329 struct mtx *mtx; 330 const char *wmesg; 331 int timo; 332{ 333 struct thread *td; 334 struct proc *p; 335 int rval; 336 WITNESS_SAVE_DECL(mtx); 337 338 td = curthread; 339 p = td->td_proc; 340 KASSERT(mtx != NULL, ("sleeping without a mutex")); 341 KASSERT(p != NULL, ("msleep1")); 342 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 343 344 if (cold) { 345 /* 346 * During autoconfiguration, just return; 347 * don't run any other threads or panic below, 348 * in case this is the idle thread and already asleep. 349 * XXX: this used to do "s = splhigh(); splx(safepri); 350 * splx(s);" to give interrupts a chance, but there is 351 * no way to give interrupts a chance now. 352 */ 353 return (0); 354 } 355 356 sleepq_lock(ident); 357 CTR5(KTR_PROC, "msleep_spin: thread %p (pid %ld, %s) on %s (%p)", 358 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 359 360 DROP_GIANT(); 361 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 362 WITNESS_SAVE(&mtx->mtx_object, mtx); 363 mtx_unlock_spin(mtx); 364 365 /* 366 * We put ourselves on the sleep queue and start our timeout. 367 */ 368 sleepq_add(ident, mtx, wmesg, SLEEPQ_MSLEEP); 369 if (timo) 370 sleepq_set_timeout(ident, timo); 371 372 /* 373 * Can't call ktrace with any spin locks held so it can lock the 374 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 375 * any spin lock. Thus, we have to drop the sleepq spin lock while 376 * we handle those requests. This is safe since we have placed our 377 * thread on the sleep queue already. 378 */ 379#ifdef KTRACE 380 if (KTRPOINT(td, KTR_CSW)) { 381 sleepq_release(ident); 382 ktrcsw(1, 0); 383 sleepq_lock(ident); 384 } 385#endif 386#ifdef WITNESS 387 sleepq_release(ident); 388 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 389 wmesg); 390 sleepq_lock(ident); 391#endif 392 if (timo) 393 rval = sleepq_timedwait(ident); 394 else { 395 sleepq_wait(ident); 396 rval = 0; 397 } 398#ifdef KTRACE 399 if (KTRPOINT(td, KTR_CSW)) 400 ktrcsw(0, 0); 401#endif 402 PICKUP_GIANT(); 403 mtx_lock_spin(mtx); 404 WITNESS_RESTORE(&mtx->mtx_object, mtx); 405 return (rval); 406} 407 408/* 409 * Make all threads sleeping on the specified identifier runnable. 410 */ 411void 412wakeup(ident) 413 register void *ident; 414{ 415 416 sleepq_lock(ident); 417 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1); 418} 419 420/* 421 * Make a thread sleeping on the specified identifier runnable. 422 * May wake more than one thread if a target thread is currently 423 * swapped out. 424 */ 425void 426wakeup_one(ident) 427 register void *ident; 428{ 429 430 sleepq_lock(ident); 431 sleepq_signal(ident, SLEEPQ_MSLEEP, -1); 432} 433 434/* 435 * The machine independent parts of context switching. 436 */ 437void 438mi_switch(int flags, struct thread *newtd) 439{ 440 struct bintime new_switchtime; 441 struct thread *td; 442 struct proc *p; 443 444 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 445 td = curthread; /* XXX */ 446 p = td->td_proc; /* XXX */ 447 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 448#ifdef INVARIANTS 449 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 450 mtx_assert(&Giant, MA_NOTOWNED); 451#endif 452 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 453 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 454 newtd == NULL) || panicstr, 455 ("mi_switch: switch in a critical section")); 456 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 457 ("mi_switch: switch must be voluntary or involuntary")); 458 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 459 460 if (flags & SW_VOL) 461 p->p_stats->p_ru.ru_nvcsw++; 462 else 463 p->p_stats->p_ru.ru_nivcsw++; 464 465 /* 466 * Compute the amount of time during which the current 467 * process was running, and add that to its total so far. 468 */ 469 binuptime(&new_switchtime); 470 bintime_add(&p->p_rux.rux_runtime, &new_switchtime); 471 bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime)); 472 473 td->td_generation++; /* bump preempt-detect counter */ 474 475 /* 476 * Don't perform context switches from the debugger. 477 */ 478 if (kdb_active) { 479 mtx_unlock_spin(&sched_lock); 480 kdb_backtrace(); 481 kdb_reenter(); 482 panic("%s: did not reenter debugger", __func__); 483 } 484 485 /* 486 * Check if the process exceeds its cpu resource allocation. If 487 * it reaches the max, arrange to kill the process in ast(). 488 */ 489 if (p->p_cpulimit != RLIM_INFINITY && 490 p->p_rux.rux_runtime.sec >= p->p_cpulimit) { 491 p->p_sflag |= PS_XCPU; 492 td->td_flags |= TDF_ASTPENDING; 493 } 494 495 /* 496 * Finish up stats for outgoing thread. 497 */ 498 cnt.v_swtch++; 499 PCPU_SET(switchtime, new_switchtime); 500 PCPU_SET(switchticks, ticks); 501 CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)", 502 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 503 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA)) 504 newtd = thread_switchout(td, flags, newtd); 505#if (KTR_COMPILE & KTR_SCHED) != 0 506 if (td == PCPU_GET(idlethread)) 507 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 508 td, td->td_proc->p_comm, td->td_priority); 509 else if (newtd != NULL) 510 CTR5(KTR_SCHED, 511 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 512 td, td->td_proc->p_comm, td->td_priority, newtd, 513 newtd->td_proc->p_comm); 514 else 515 CTR6(KTR_SCHED, 516 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 517 td, td->td_proc->p_comm, td->td_priority, 518 td->td_inhibitors, td->td_wmesg, td->td_lockname); 519#endif 520 sched_switch(td, newtd, flags); 521 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 522 td, td->td_proc->p_comm, td->td_priority); 523 524 CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)", 525 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 526 527 /* 528 * If the last thread was exiting, finish cleaning it up. 529 */ 530 if ((td = PCPU_GET(deadthread))) { 531 PCPU_SET(deadthread, NULL); 532 thread_stash(td); 533 } 534} 535 536/* 537 * Change process state to be runnable, 538 * placing it on the run queue if it is in memory, 539 * and awakening the swapper if it isn't in memory. 540 */ 541void 542setrunnable(struct thread *td) 543{ 544 struct proc *p; 545 546 p = td->td_proc; 547 mtx_assert(&sched_lock, MA_OWNED); 548 switch (p->p_state) { 549 case PRS_ZOMBIE: 550 panic("setrunnable(1)"); 551 default: 552 break; 553 } 554 switch (td->td_state) { 555 case TDS_RUNNING: 556 case TDS_RUNQ: 557 return; 558 case TDS_INHIBITED: 559 /* 560 * If we are only inhibited because we are swapped out 561 * then arange to swap in this process. Otherwise just return. 562 */ 563 if (td->td_inhibitors != TDI_SWAPPED) 564 return; 565 /* XXX: intentional fall-through ? */ 566 case TDS_CAN_RUN: 567 break; 568 default: 569 printf("state is 0x%x", td->td_state); 570 panic("setrunnable(2)"); 571 } 572 if ((p->p_sflag & PS_INMEM) == 0) { 573 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 574 p->p_sflag |= PS_SWAPINREQ; 575 /* 576 * due to a LOR between sched_lock and 577 * the sleepqueue chain locks, use 578 * lower level scheduling functions. 579 */ 580 kick_proc0(); 581 } 582 } else 583 sched_wakeup(td); 584} 585 586/* 587 * Compute a tenex style load average of a quantity on 588 * 1, 5 and 15 minute intervals. 589 * XXXKSE Needs complete rewrite when correct info is available. 590 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 591 */ 592static void 593loadav(void *arg) 594{ 595 int i, nrun; 596 struct loadavg *avg; 597 598 nrun = sched_load(); 599 avg = &averunnable; 600 601 for (i = 0; i < 3; i++) 602 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 603 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 604 605 /* 606 * Schedule the next update to occur after 5 seconds, but add a 607 * random variation to avoid synchronisation with processes that 608 * run at regular intervals. 609 */ 610 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 611 loadav, NULL); 612} 613 614static void 615lboltcb(void *arg) 616{ 617 wakeup(&lbolt); 618 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 619} 620 621/* ARGSUSED */ 622static void 623synch_setup(dummy) 624 void *dummy; 625{ 626 callout_init(&loadav_callout, CALLOUT_MPSAFE); 627 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 628 629 /* Kick off timeout driven events by calling first time. */ 630 loadav(NULL); 631 lboltcb(NULL); 632} 633 634/* 635 * General purpose yield system call 636 */ 637int 638yield(struct thread *td, struct yield_args *uap) 639{ 640 struct ksegrp *kg; 641 642 kg = td->td_ksegrp; 643 mtx_assert(&Giant, MA_NOTOWNED); 644 mtx_lock_spin(&sched_lock); 645 sched_prio(td, PRI_MAX_TIMESHARE); 646 mi_switch(SW_VOL, NULL); 647 mtx_unlock_spin(&sched_lock); 648 td->td_retval[0] = 0; 649 return (0); 650} 651