kern_synch.c revision 104964
1/*- 2 * Copyright (c) 1982, 1986, 1990, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95 39 * $FreeBSD: head/sys/kern/kern_synch.c 104964 2002-10-12 05:32:24Z jeff $ 40 */ 41 42#include "opt_ddb.h" 43#include "opt_ktrace.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/condvar.h> 48#include <sys/kernel.h> 49#include <sys/ktr.h> 50#include <sys/lock.h> 51#include <sys/mutex.h> 52#include <sys/proc.h> 53#include <sys/resourcevar.h> 54#include <sys/sched.h> 55#include <sys/signalvar.h> 56#include <sys/smp.h> 57#include <sys/sx.h> 58#include <sys/sysctl.h> 59#include <sys/sysproto.h> 60#include <sys/vmmeter.h> 61#ifdef DDB 62#include <ddb/ddb.h> 63#endif 64#ifdef KTRACE 65#include <sys/uio.h> 66#include <sys/ktrace.h> 67#endif 68 69#include <machine/cpu.h> 70 71static void sched_setup(void *dummy); 72SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL) 73 74int hogticks; 75int lbolt; 76 77static struct callout loadav_callout; 78 79struct loadavg averunnable = 80 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 81/* 82 * Constants for averages over 1, 5, and 15 minutes 83 * when sampling at 5 second intervals. 84 */ 85static fixpt_t cexp[3] = { 86 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 87 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 88 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 89}; 90 91static void endtsleep(void *); 92static void loadav(void *arg); 93 94/* 95 * We're only looking at 7 bits of the address; everything is 96 * aligned to 4, lots of things are aligned to greater powers 97 * of 2. Shift right by 8, i.e. drop the bottom 256 worth. 98 */ 99#define TABLESIZE 128 100static TAILQ_HEAD(slpquehead, thread) slpque[TABLESIZE]; 101#define LOOKUP(x) (((intptr_t)(x) >> 8) & (TABLESIZE - 1)) 102 103void 104sleepinit(void) 105{ 106 int i; 107 108 hogticks = (hz / 10) * 2; /* Default only. */ 109 for (i = 0; i < TABLESIZE; i++) 110 TAILQ_INIT(&slpque[i]); 111} 112 113/* 114 * General sleep call. Suspends the current process until a wakeup is 115 * performed on the specified identifier. The process will then be made 116 * runnable with the specified priority. Sleeps at most timo/hz seconds 117 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 118 * before and after sleeping, else signals are not checked. Returns 0 if 119 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 120 * signal needs to be delivered, ERESTART is returned if the current system 121 * call should be restarted if possible, and EINTR is returned if the system 122 * call should be interrupted by the signal (return EINTR). 123 * 124 * The mutex argument is exited before the caller is suspended, and 125 * entered before msleep returns. If priority includes the PDROP 126 * flag the mutex is not entered before returning. 127 */ 128 129int 130msleep(ident, mtx, priority, wmesg, timo) 131 void *ident; 132 struct mtx *mtx; 133 int priority, timo; 134 const char *wmesg; 135{ 136 struct thread *td = curthread; 137 struct proc *p = td->td_proc; 138 int sig, catch = priority & PCATCH; 139 int rval = 0; 140 WITNESS_SAVE_DECL(mtx); 141 142#ifdef KTRACE 143 if (KTRPOINT(td, KTR_CSW)) 144 ktrcsw(1, 0); 145#endif 146 WITNESS_SLEEP(0, &mtx->mtx_object); 147 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL, 148 ("sleeping without a mutex")); 149 /* 150 * If we are capable of async syscalls and there isn't already 151 * another one ready to return, start a new thread 152 * and queue it as ready to run. Note that there is danger here 153 * because we need to make sure that we don't sleep allocating 154 * the thread (recursion here might be bad). 155 * Hence the TDF_INMSLEEP flag. 156 */ 157 if (p->p_flag & P_KSES) { 158 /* Just don't bother if we are exiting 159 and not the exiting thread. */ 160 if ((p->p_flag & P_WEXIT) && catch && p->p_singlethread != td) 161 return (EINTR); 162 if (td->td_mailbox && (!(td->td_flags & TDF_INMSLEEP))) { 163 /* 164 * Arrange for an upcall to be readied. 165 * it will not actually happen until all 166 * pending in-kernel work for this KSEGRP 167 * has been done. 168 */ 169 mtx_lock_spin(&sched_lock); 170 /* Don't recurse here! */ 171 td->td_flags |= TDF_INMSLEEP; 172 thread_schedule_upcall(td, td->td_kse); 173 td->td_flags &= ~TDF_INMSLEEP; 174 mtx_unlock_spin(&sched_lock); 175 } 176 } 177 mtx_lock_spin(&sched_lock); 178 if (cold ) { 179 /* 180 * During autoconfiguration, just give interrupts 181 * a chance, then just return. 182 * Don't run any other procs or panic below, 183 * in case this is the idle process and already asleep. 184 */ 185 if (mtx != NULL && priority & PDROP) 186 mtx_unlock(mtx); 187 mtx_unlock_spin(&sched_lock); 188 return (0); 189 } 190 191 DROP_GIANT(); 192 193 if (mtx != NULL) { 194 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 195 WITNESS_SAVE(&mtx->mtx_object, mtx); 196 mtx_unlock(mtx); 197 if (priority & PDROP) 198 mtx = NULL; 199 } 200 201 KASSERT(p != NULL, ("msleep1")); 202 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 203 204 CTR5(KTR_PROC, "msleep: thread %p (pid %d, %s) on %s (%p)", 205 td, p->p_pid, p->p_comm, wmesg, ident); 206 207 td->td_wchan = ident; 208 td->td_wmesg = wmesg; 209 TAILQ_INSERT_TAIL(&slpque[LOOKUP(ident)], td, td_slpq); 210 TD_SET_ON_SLEEPQ(td); 211 if (timo) 212 callout_reset(&td->td_slpcallout, timo, endtsleep, td); 213 /* 214 * We put ourselves on the sleep queue and start our timeout 215 * before calling thread_suspend_check, as we could stop there, and 216 * a wakeup or a SIGCONT (or both) could occur while we were stopped. 217 * without resuming us, thus we must be ready for sleep 218 * when cursig is called. If the wakeup happens while we're 219 * stopped, td->td_wchan will be 0 upon return from cursig. 220 */ 221 if (catch) { 222 CTR3(KTR_PROC, "msleep caught: thread %p (pid %d, %s)", td, 223 p->p_pid, p->p_comm); 224 td->td_flags |= TDF_SINTR; 225 mtx_unlock_spin(&sched_lock); 226 PROC_LOCK(p); 227 sig = cursig(td); 228 if (sig == 0 && thread_suspend_check(1)) 229 sig = SIGSTOP; 230 mtx_lock_spin(&sched_lock); 231 PROC_UNLOCK(p); 232 if (sig != 0) { 233 if (TD_ON_SLEEPQ(td)) 234 unsleep(td); 235 } else if (!TD_ON_SLEEPQ(td)) 236 catch = 0; 237 } else 238 sig = 0; 239 240 /* 241 * Let the scheduler know we're about to voluntarily go to sleep. 242 */ 243 sched_sleep(td, priority & PRIMASK); 244 245 if (TD_ON_SLEEPQ(td)) { 246 p->p_stats->p_ru.ru_nvcsw++; 247 TD_SET_SLEEPING(td); 248 mi_switch(); 249 } 250 /* 251 * We're awake from voluntary sleep. 252 */ 253 CTR3(KTR_PROC, "msleep resume: thread %p (pid %d, %s)", td, p->p_pid, 254 p->p_comm); 255 KASSERT(TD_IS_RUNNING(td), ("running but not TDS_RUNNING")); 256 td->td_flags &= ~TDF_SINTR; 257 if (td->td_flags & TDF_TIMEOUT) { 258 td->td_flags &= ~TDF_TIMEOUT; 259 if (sig == 0) 260 rval = EWOULDBLOCK; 261 } else if (td->td_flags & TDF_TIMOFAIL) { 262 td->td_flags &= ~TDF_TIMOFAIL; 263 } else if (timo && callout_stop(&td->td_slpcallout) == 0) { 264 /* 265 * This isn't supposed to be pretty. If we are here, then 266 * the endtsleep() callout is currently executing on another 267 * CPU and is either spinning on the sched_lock or will be 268 * soon. If we don't synchronize here, there is a chance 269 * that this process may msleep() again before the callout 270 * has a chance to run and the callout may end up waking up 271 * the wrong msleep(). Yuck. 272 */ 273 TD_SET_SLEEPING(td); 274 p->p_stats->p_ru.ru_nivcsw++; 275 mi_switch(); 276 td->td_flags &= ~TDF_TIMOFAIL; 277 } 278 mtx_unlock_spin(&sched_lock); 279 280 if (rval == 0 && catch) { 281 PROC_LOCK(p); 282 /* XXX: shouldn't we always be calling cursig() */ 283 if (sig != 0 || (sig = cursig(td))) { 284 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig)) 285 rval = EINTR; 286 else 287 rval = ERESTART; 288 } 289 PROC_UNLOCK(p); 290 } 291#ifdef KTRACE 292 if (KTRPOINT(td, KTR_CSW)) 293 ktrcsw(0, 0); 294#endif 295 PICKUP_GIANT(); 296 if (mtx != NULL) { 297 mtx_lock(mtx); 298 WITNESS_RESTORE(&mtx->mtx_object, mtx); 299 } 300 return (rval); 301} 302 303/* 304 * Implement timeout for msleep() 305 * 306 * If process hasn't been awakened (wchan non-zero), 307 * set timeout flag and undo the sleep. If proc 308 * is stopped, just unsleep so it will remain stopped. 309 * MP-safe, called without the Giant mutex. 310 */ 311static void 312endtsleep(arg) 313 void *arg; 314{ 315 register struct thread *td = arg; 316 317 CTR3(KTR_PROC, "endtsleep: thread %p (pid %d, %s)", 318 td, td->td_proc->p_pid, td->td_proc->p_comm); 319 mtx_lock_spin(&sched_lock); 320 /* 321 * This is the other half of the synchronization with msleep() 322 * described above. If the TDS_TIMEOUT flag is set, we lost the 323 * race and just need to put the process back on the runqueue. 324 */ 325 if (TD_ON_SLEEPQ(td)) { 326 TAILQ_REMOVE(&slpque[LOOKUP(td->td_wchan)], td, td_slpq); 327 TD_CLR_ON_SLEEPQ(td); 328 td->td_flags |= TDF_TIMEOUT; 329 } else { 330 td->td_flags |= TDF_TIMOFAIL; 331 } 332 TD_CLR_SLEEPING(td); 333 setrunnable(td); 334 mtx_unlock_spin(&sched_lock); 335} 336 337/* 338 * Abort a thread, as if an interrupt had occured. Only abort 339 * interruptable waits (unfortunatly it isn't only safe to abort others). 340 * This is about identical to cv_abort(). 341 * Think about merging them? 342 * Also, whatever the signal code does... 343 */ 344void 345abortsleep(struct thread *td) 346{ 347 348 mtx_assert(&sched_lock, MA_OWNED); 349 /* 350 * If the TDF_TIMEOUT flag is set, just leave. A 351 * timeout is scheduled anyhow. 352 */ 353 if ((td->td_flags & (TDF_TIMEOUT | TDF_SINTR)) == TDF_SINTR) { 354 if (TD_ON_SLEEPQ(td)) { 355 unsleep(td); 356 TD_CLR_SLEEPING(td); 357 setrunnable(td); 358 } 359 } 360} 361 362/* 363 * Remove a process from its wait queue 364 */ 365void 366unsleep(struct thread *td) 367{ 368 369 mtx_lock_spin(&sched_lock); 370 if (TD_ON_SLEEPQ(td)) { 371 TAILQ_REMOVE(&slpque[LOOKUP(td->td_wchan)], td, td_slpq); 372 TD_CLR_ON_SLEEPQ(td); 373 } 374 mtx_unlock_spin(&sched_lock); 375} 376 377/* 378 * Make all processes sleeping on the specified identifier runnable. 379 */ 380void 381wakeup(ident) 382 register void *ident; 383{ 384 register struct slpquehead *qp; 385 register struct thread *td; 386 struct thread *ntd; 387 struct proc *p; 388 389 mtx_lock_spin(&sched_lock); 390 qp = &slpque[LOOKUP(ident)]; 391restart: 392 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) { 393 ntd = TAILQ_NEXT(td, td_slpq); 394 if (td->td_wchan == ident) { 395 unsleep(td); 396 TD_CLR_SLEEPING(td); 397 setrunnable(td); 398 p = td->td_proc; 399 CTR3(KTR_PROC,"wakeup: thread %p (pid %d, %s)", 400 td, p->p_pid, p->p_comm); 401 goto restart; 402 } 403 } 404 mtx_unlock_spin(&sched_lock); 405} 406 407/* 408 * Make a process sleeping on the specified identifier runnable. 409 * May wake more than one process if a target process is currently 410 * swapped out. 411 */ 412void 413wakeup_one(ident) 414 register void *ident; 415{ 416 register struct slpquehead *qp; 417 register struct thread *td; 418 register struct proc *p; 419 struct thread *ntd; 420 421 mtx_lock_spin(&sched_lock); 422 qp = &slpque[LOOKUP(ident)]; 423 for (td = TAILQ_FIRST(qp); td != NULL; td = ntd) { 424 ntd = TAILQ_NEXT(td, td_slpq); 425 if (td->td_wchan == ident) { 426 unsleep(td); 427 TD_CLR_SLEEPING(td); 428 setrunnable(td); 429 p = td->td_proc; 430 CTR3(KTR_PROC,"wakeup1: thread %p (pid %d, %s)", 431 td, p->p_pid, p->p_comm); 432 break; 433 } 434 } 435 mtx_unlock_spin(&sched_lock); 436} 437 438/* 439 * The machine independent parts of mi_switch(). 440 */ 441void 442mi_switch(void) 443{ 444 struct bintime new_switchtime; 445 struct thread *td = curthread; /* XXX */ 446 struct proc *p = td->td_proc; /* XXX */ 447 struct kse *ke = td->td_kse; 448 u_int sched_nest; 449 450 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 451 452 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 453#ifdef INVARIANTS 454 if (!TD_ON_LOCK(td) && 455 !TD_ON_RUNQ(td) && 456 !TD_IS_RUNNING(td)) 457 mtx_assert(&Giant, MA_NOTOWNED); 458#endif 459 KASSERT(td->td_critnest == 1, 460 ("mi_switch: switch in a critical section")); 461 462 /* 463 * Compute the amount of time during which the current 464 * process was running, and add that to its total so far. 465 */ 466 binuptime(&new_switchtime); 467 bintime_add(&p->p_runtime, &new_switchtime); 468 bintime_sub(&p->p_runtime, PCPU_PTR(switchtime)); 469 470#ifdef DDB 471 /* 472 * Don't perform context switches from the debugger. 473 */ 474 if (db_active) { 475 mtx_unlock_spin(&sched_lock); 476 db_error("Context switches not allowed in the debugger."); 477 } 478#endif 479 480 /* 481 * Check if the process exceeds its cpu resource allocation. If 482 * over max, arrange to kill the process in ast(). 483 */ 484 if (p->p_cpulimit != RLIM_INFINITY && 485 p->p_runtime.sec > p->p_cpulimit) { 486 p->p_sflag |= PS_XCPU; 487 ke->ke_flags |= KEF_ASTPENDING; 488 } 489 490 /* 491 * Finish up stats for outgoing thread. 492 */ 493 cnt.v_swtch++; 494 PCPU_SET(switchtime, new_switchtime); 495 CTR3(KTR_PROC, "mi_switch: old thread %p (pid %d, %s)", td, p->p_pid, 496 p->p_comm); 497 498 sched_nest = sched_lock.mtx_recurse; 499 sched_switchout(td); 500 501 cpu_switch(); /* SHAZAM!!*/ 502 503 sched_lock.mtx_recurse = sched_nest; 504 sched_lock.mtx_lock = (uintptr_t)td; 505 sched_switchin(td); 506 507 /* 508 * Start setting up stats etc. for the incoming thread. 509 * Similar code in fork_exit() is returned to by cpu_switch() 510 * in the case of a new thread/process. 511 */ 512 CTR3(KTR_PROC, "mi_switch: new thread %p (pid %d, %s)", td, p->p_pid, 513 p->p_comm); 514 if (PCPU_GET(switchtime.sec) == 0) 515 binuptime(PCPU_PTR(switchtime)); 516 PCPU_SET(switchticks, ticks); 517 518 /* 519 * Call the switchin function while still holding the scheduler lock 520 * (used by the idlezero code and the general page-zeroing code) 521 */ 522 if (td->td_switchin) 523 td->td_switchin(); 524} 525 526/* 527 * Change process state to be runnable, 528 * placing it on the run queue if it is in memory, 529 * and awakening the swapper if it isn't in memory. 530 */ 531void 532setrunnable(struct thread *td) 533{ 534 struct proc *p = td->td_proc; 535 536 mtx_assert(&sched_lock, MA_OWNED); 537 switch (p->p_state) { 538 case PRS_ZOMBIE: 539 panic("setrunnable(1)"); 540 default: 541 break; 542 } 543 switch (td->td_state) { 544 case TDS_RUNNING: 545 case TDS_RUNQ: 546 return; 547 case TDS_INHIBITED: 548 /* 549 * If we are only inhibited because we are swapped out 550 * then arange to swap in this process. Otherwise just return. 551 */ 552 if (td->td_inhibitors != TDI_SWAPPED) 553 return; 554 case TDS_CAN_RUN: 555 break; 556 default: 557 printf("state is 0x%x", td->td_state); 558 panic("setrunnable(2)"); 559 } 560 if ((p->p_sflag & PS_INMEM) == 0) { 561 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 562 p->p_sflag |= PS_SWAPINREQ; 563 wakeup(&proc0); 564 } 565 } else 566 sched_wakeup(td); 567} 568 569/* 570 * Compute a tenex style load average of a quantity on 571 * 1, 5 and 15 minute intervals. 572 * XXXKSE Needs complete rewrite when correct info is available. 573 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 574 */ 575static void 576loadav(void *arg) 577{ 578 int i, nrun; 579 struct loadavg *avg; 580 struct proc *p; 581 struct thread *td; 582 583 avg = &averunnable; 584 sx_slock(&allproc_lock); 585 nrun = 0; 586 FOREACH_PROC_IN_SYSTEM(p) { 587 FOREACH_THREAD_IN_PROC(p, td) { 588 switch (td->td_state) { 589 case TDS_RUNQ: 590 case TDS_RUNNING: 591 if ((p->p_flag & P_NOLOAD) != 0) 592 goto nextproc; 593 nrun++; /* XXXKSE */ 594 default: 595 break; 596 } 597nextproc: 598 continue; 599 } 600 } 601 sx_sunlock(&allproc_lock); 602 for (i = 0; i < 3; i++) 603 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 604 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 605 606 /* 607 * Schedule the next update to occur after 5 seconds, but add a 608 * random variation to avoid synchronisation with processes that 609 * run at regular intervals. 610 */ 611 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 612 loadav, NULL); 613} 614 615/* ARGSUSED */ 616static void 617sched_setup(dummy) 618 void *dummy; 619{ 620 callout_init(&loadav_callout, 0); 621 622 /* Kick off timeout driven events by calling first time. */ 623 loadav(NULL); 624} 625 626/* 627 * General purpose yield system call 628 */ 629int 630yield(struct thread *td, struct yield_args *uap) 631{ 632 struct ksegrp *kg = td->td_ksegrp; 633 634 mtx_assert(&Giant, MA_NOTOWNED); 635 mtx_lock_spin(&sched_lock); 636 kg->kg_proc->p_stats->p_ru.ru_nvcsw++; 637 sched_prio(td, PRI_MAX_TIMESHARE); 638 mi_switch(); 639 mtx_unlock_spin(&sched_lock); 640 td->td_retval[0] = 0; 641 642 return (0); 643} 644 645