kern_synch.c revision 164325
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 164325 2006-11-16 01:02:00Z pjd $"); 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 thread until a wakeup is 105 * performed on the specified identifier. The thread 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 unlocked before the caller is suspended, and 115 * re-locked before msleep returns. If priority includes the PDROP 116 * flag the mutex is not re-locked 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, 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 ident == &lbolt, ("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 flags = SLEEPQ_MSLEEP; 168 if (catch) 169 flags |= SLEEPQ_INTERRUPTIBLE; 170 171 sleepq_lock(ident); 172 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)", 173 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 174 175 DROP_GIANT(); 176 if (mtx != NULL) { 177 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 178 WITNESS_SAVE(&mtx->mtx_object, mtx); 179 mtx_unlock(mtx); 180 } 181 182 /* 183 * We put ourselves on the sleep queue and start our timeout 184 * before calling thread_suspend_check, as we could stop there, 185 * and a wakeup or a SIGCONT (or both) could occur while we were 186 * stopped without resuming us. Thus, we must be ready for sleep 187 * when cursig() is called. If the wakeup happens while we're 188 * stopped, then td will no longer be on a sleep queue upon 189 * return from cursig(). 190 */ 191 sleepq_add(ident, ident == &lbolt ? NULL : &mtx->mtx_object, wmesg, 192 flags); 193 if (timo) 194 sleepq_set_timeout(ident, timo); 195 196 /* 197 * Adjust this thread's priority. 198 */ 199 if ((priority & PRIMASK) != 0) { 200 mtx_lock_spin(&sched_lock); 201 sched_prio(td, priority & PRIMASK); 202 mtx_unlock_spin(&sched_lock); 203 } 204 205 if (timo && catch) 206 rval = sleepq_timedwait_sig(ident); 207 else if (timo) 208 rval = sleepq_timedwait(ident); 209 else if (catch) 210 rval = sleepq_wait_sig(ident); 211 else { 212 sleepq_wait(ident); 213 rval = 0; 214 } 215#ifdef KTRACE 216 if (KTRPOINT(td, KTR_CSW)) 217 ktrcsw(0, 0); 218#endif 219 PICKUP_GIANT(); 220 if (mtx != NULL && !(priority & PDROP)) { 221 mtx_lock(mtx); 222 WITNESS_RESTORE(&mtx->mtx_object, mtx); 223 } 224 return (rval); 225} 226 227int 228msleep_spin(ident, mtx, wmesg, timo) 229 void *ident; 230 struct mtx *mtx; 231 const char *wmesg; 232 int timo; 233{ 234 struct thread *td; 235 struct proc *p; 236 int rval; 237 WITNESS_SAVE_DECL(mtx); 238 239 td = curthread; 240 p = td->td_proc; 241 KASSERT(mtx != NULL, ("sleeping without a mutex")); 242 KASSERT(p != NULL, ("msleep1")); 243 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 244 245 if (cold) { 246 /* 247 * During autoconfiguration, just return; 248 * don't run any other threads or panic below, 249 * in case this is the idle thread and already asleep. 250 * XXX: this used to do "s = splhigh(); splx(safepri); 251 * splx(s);" to give interrupts a chance, but there is 252 * no way to give interrupts a chance now. 253 */ 254 return (0); 255 } 256 257 sleepq_lock(ident); 258 CTR5(KTR_PROC, "msleep_spin: thread %p (pid %ld, %s) on %s (%p)", 259 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 260 261 DROP_GIANT(); 262 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 263 WITNESS_SAVE(&mtx->mtx_object, mtx); 264 mtx_unlock_spin(mtx); 265 266 /* 267 * We put ourselves on the sleep queue and start our timeout. 268 */ 269 sleepq_add(ident, &mtx->mtx_object, wmesg, SLEEPQ_MSLEEP); 270 if (timo) 271 sleepq_set_timeout(ident, timo); 272 273 /* 274 * Can't call ktrace with any spin locks held so it can lock the 275 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 276 * any spin lock. Thus, we have to drop the sleepq spin lock while 277 * we handle those requests. This is safe since we have placed our 278 * thread on the sleep queue already. 279 */ 280#ifdef KTRACE 281 if (KTRPOINT(td, KTR_CSW)) { 282 sleepq_release(ident); 283 ktrcsw(1, 0); 284 sleepq_lock(ident); 285 } 286#endif 287#ifdef WITNESS 288 sleepq_release(ident); 289 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 290 wmesg); 291 sleepq_lock(ident); 292#endif 293 if (timo) 294 rval = sleepq_timedwait(ident); 295 else { 296 sleepq_wait(ident); 297 rval = 0; 298 } 299#ifdef KTRACE 300 if (KTRPOINT(td, KTR_CSW)) 301 ktrcsw(0, 0); 302#endif 303 PICKUP_GIANT(); 304 mtx_lock_spin(mtx); 305 WITNESS_RESTORE(&mtx->mtx_object, mtx); 306 return (rval); 307} 308 309/* 310 * Make all threads sleeping on the specified identifier runnable. 311 */ 312void 313wakeup(ident) 314 register void *ident; 315{ 316 317 sleepq_lock(ident); 318 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1); 319} 320 321/* 322 * Make a thread sleeping on the specified identifier runnable. 323 * May wake more than one thread if a target thread is currently 324 * swapped out. 325 */ 326void 327wakeup_one(ident) 328 register void *ident; 329{ 330 331 sleepq_lock(ident); 332 sleepq_signal(ident, SLEEPQ_MSLEEP, -1); 333} 334 335/* 336 * The machine independent parts of context switching. 337 */ 338void 339mi_switch(int flags, struct thread *newtd) 340{ 341 uint64_t new_switchtime; 342 struct thread *td; 343 struct proc *p; 344 345 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 346 td = curthread; /* XXX */ 347 p = td->td_proc; /* XXX */ 348 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 349#ifdef INVARIANTS 350 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 351 mtx_assert(&Giant, MA_NOTOWNED); 352#endif 353 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 354 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 355 newtd == NULL) || panicstr, 356 ("mi_switch: switch in a critical section")); 357 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 358 ("mi_switch: switch must be voluntary or involuntary")); 359 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 360 361 /* 362 * Don't perform context switches from the debugger. 363 */ 364 if (kdb_active) { 365 mtx_unlock_spin(&sched_lock); 366 kdb_backtrace(); 367 kdb_reenter(); 368 panic("%s: did not reenter debugger", __func__); 369 } 370 371 if (flags & SW_VOL) 372 p->p_stats->p_ru.ru_nvcsw++; 373 else 374 p->p_stats->p_ru.ru_nivcsw++; 375 376 /* 377 * Compute the amount of time during which the current 378 * process was running, and add that to its total so far. 379 */ 380 new_switchtime = cpu_ticks(); 381 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime)); 382 p->p_rux.rux_uticks += td->td_uticks; 383 td->td_uticks = 0; 384 p->p_rux.rux_iticks += td->td_iticks; 385 td->td_iticks = 0; 386 p->p_rux.rux_sticks += td->td_sticks; 387 td->td_sticks = 0; 388 389 td->td_generation++; /* bump preempt-detect counter */ 390 391 /* 392 * Check if the process exceeds its cpu resource allocation. If 393 * it reaches the max, arrange to kill the process in ast(). 394 */ 395 if (p->p_cpulimit != RLIM_INFINITY && 396 p->p_rux.rux_runtime >= p->p_cpulimit * cpu_tickrate()) { 397 p->p_sflag |= PS_XCPU; 398 td->td_flags |= TDF_ASTPENDING; 399 } 400 401 /* 402 * Finish up stats for outgoing thread. 403 */ 404 cnt.v_swtch++; 405 PCPU_SET(switchtime, new_switchtime); 406 PCPU_SET(switchticks, ticks); 407 CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)", 408 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 409#ifdef KSE 410 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA)) 411 newtd = thread_switchout(td, flags, newtd); 412#endif 413#if (KTR_COMPILE & KTR_SCHED) != 0 414 if (td == PCPU_GET(idlethread)) 415 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 416 td, td->td_proc->p_comm, td->td_priority); 417 else if (newtd != NULL) 418 CTR5(KTR_SCHED, 419 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 420 td, td->td_proc->p_comm, td->td_priority, newtd, 421 newtd->td_proc->p_comm); 422 else 423 CTR6(KTR_SCHED, 424 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 425 td, td->td_proc->p_comm, td->td_priority, 426 td->td_inhibitors, td->td_wmesg, td->td_lockname); 427#endif 428 sched_switch(td, newtd, flags); 429 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 430 td, td->td_proc->p_comm, td->td_priority); 431 432 CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)", 433 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 434 435 /* 436 * If the last thread was exiting, finish cleaning it up. 437 */ 438 if ((td = PCPU_GET(deadthread))) { 439 PCPU_SET(deadthread, NULL); 440 thread_stash(td); 441 } 442} 443 444/* 445 * Change process state to be runnable, 446 * placing it on the run queue if it is in memory, 447 * and awakening the swapper if it isn't in memory. 448 */ 449void 450setrunnable(struct thread *td) 451{ 452 struct proc *p; 453 454 p = td->td_proc; 455 mtx_assert(&sched_lock, MA_OWNED); 456 switch (p->p_state) { 457 case PRS_ZOMBIE: 458 panic("setrunnable(1)"); 459 default: 460 break; 461 } 462 switch (td->td_state) { 463 case TDS_RUNNING: 464 case TDS_RUNQ: 465 return; 466 case TDS_INHIBITED: 467 /* 468 * If we are only inhibited because we are swapped out 469 * then arange to swap in this process. Otherwise just return. 470 */ 471 if (td->td_inhibitors != TDI_SWAPPED) 472 return; 473 /* XXX: intentional fall-through ? */ 474 case TDS_CAN_RUN: 475 break; 476 default: 477 printf("state is 0x%x", td->td_state); 478 panic("setrunnable(2)"); 479 } 480 if ((p->p_sflag & PS_INMEM) == 0) { 481 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 482 p->p_sflag |= PS_SWAPINREQ; 483 /* 484 * due to a LOR between sched_lock and 485 * the sleepqueue chain locks, use 486 * lower level scheduling functions. 487 */ 488 kick_proc0(); 489 } 490 } else 491 sched_wakeup(td); 492} 493 494/* 495 * Compute a tenex style load average of a quantity on 496 * 1, 5 and 15 minute intervals. 497 * XXXKSE Needs complete rewrite when correct info is available. 498 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 499 */ 500static void 501loadav(void *arg) 502{ 503 int i, nrun; 504 struct loadavg *avg; 505 506 nrun = sched_load(); 507 avg = &averunnable; 508 509 for (i = 0; i < 3; i++) 510 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 511 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 512 513 /* 514 * Schedule the next update to occur after 5 seconds, but add a 515 * random variation to avoid synchronisation with processes that 516 * run at regular intervals. 517 */ 518 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 519 loadav, NULL); 520} 521 522static void 523lboltcb(void *arg) 524{ 525 wakeup(&lbolt); 526 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 527} 528 529/* ARGSUSED */ 530static void 531synch_setup(dummy) 532 void *dummy; 533{ 534 callout_init(&loadav_callout, CALLOUT_MPSAFE); 535 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 536 537 /* Kick off timeout driven events by calling first time. */ 538 loadav(NULL); 539 lboltcb(NULL); 540} 541 542/* 543 * General purpose yield system call 544 */ 545int 546yield(struct thread *td, struct yield_args *uap) 547{ 548 mtx_assert(&Giant, MA_NOTOWNED); 549 (void)uap; 550 sched_relinquish(td); 551 return (0); 552} 553