kern_synch.c revision 177091
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 177091 2008-03-12 10:12:01Z jeff $"); 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 pri = priority & PRIMASK; 164 rval = 0; 165 166 /* 167 * If we are already on a sleep queue, then remove us from that 168 * sleep queue first. We have to do this to handle recursive 169 * sleeps. 170 */ 171 if (TD_ON_SLEEPQ(td)) 172 sleepq_remove(td, td->td_wchan); 173 174 if (ident == &pause_wchan) 175 flags = SLEEPQ_PAUSE; 176 else 177 flags = SLEEPQ_SLEEP; 178 if (catch) 179 flags |= SLEEPQ_INTERRUPTIBLE; 180 181 sleepq_lock(ident); 182 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 183 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 184 185 DROP_GIANT(); 186 if (lock != NULL && !(class->lc_flags & LC_SLEEPABLE)) { 187 WITNESS_SAVE(lock, lock_witness); 188 lock_state = class->lc_unlock(lock); 189 } else 190 /* GCC needs to follow the Yellow Brick Road */ 191 lock_state = -1; 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 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0); 203 if (timo) 204 sleepq_set_timeout(ident, timo); 205 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 206 sleepq_release(ident); 207 WITNESS_SAVE(lock, lock_witness); 208 lock_state = class->lc_unlock(lock); 209 sleepq_lock(ident); 210 } 211 if (timo && catch) 212 rval = sleepq_timedwait_sig(ident, pri); 213 else if (timo) 214 rval = sleepq_timedwait(ident, pri); 215 else if (catch) 216 rval = sleepq_wait_sig(ident, pri); 217 else { 218 sleepq_wait(ident, pri); 219 rval = 0; 220 } 221#ifdef KTRACE 222 if (KTRPOINT(td, KTR_CSW)) 223 ktrcsw(0, 0); 224#endif 225 PICKUP_GIANT(); 226 if (lock != NULL && !(priority & PDROP)) { 227 class->lc_lock(lock, lock_state); 228 WITNESS_RESTORE(lock, lock_witness); 229 } 230 return (rval); 231} 232 233int 234msleep_spin(ident, mtx, wmesg, timo) 235 void *ident; 236 struct mtx *mtx; 237 const char *wmesg; 238 int timo; 239{ 240 struct thread *td; 241 struct proc *p; 242 int rval; 243 WITNESS_SAVE_DECL(mtx); 244 245 td = curthread; 246 p = td->td_proc; 247 KASSERT(mtx != NULL, ("sleeping without a mutex")); 248 KASSERT(p != NULL, ("msleep1")); 249 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 250 251 if (cold) { 252 /* 253 * During autoconfiguration, just return; 254 * don't run any other threads or panic below, 255 * in case this is the idle thread and already asleep. 256 * XXX: this used to do "s = splhigh(); splx(safepri); 257 * splx(s);" to give interrupts a chance, but there is 258 * no way to give interrupts a chance now. 259 */ 260 return (0); 261 } 262 263 sleepq_lock(ident); 264 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 265 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 266 267 DROP_GIANT(); 268 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 269 WITNESS_SAVE(&mtx->lock_object, mtx); 270 mtx_unlock_spin(mtx); 271 272 /* 273 * We put ourselves on the sleep queue and start our timeout. 274 */ 275 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 276 if (timo) 277 sleepq_set_timeout(ident, timo); 278 279 /* 280 * Can't call ktrace with any spin locks held so it can lock the 281 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 282 * any spin lock. Thus, we have to drop the sleepq spin lock while 283 * we handle those requests. This is safe since we have placed our 284 * thread on the sleep queue already. 285 */ 286#ifdef KTRACE 287 if (KTRPOINT(td, KTR_CSW)) { 288 sleepq_release(ident); 289 ktrcsw(1, 0); 290 sleepq_lock(ident); 291 } 292#endif 293#ifdef WITNESS 294 sleepq_release(ident); 295 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 296 wmesg); 297 sleepq_lock(ident); 298#endif 299 if (timo) 300 rval = sleepq_timedwait(ident, 0); 301 else { 302 sleepq_wait(ident, 0); 303 rval = 0; 304 } 305#ifdef KTRACE 306 if (KTRPOINT(td, KTR_CSW)) 307 ktrcsw(0, 0); 308#endif 309 PICKUP_GIANT(); 310 mtx_lock_spin(mtx); 311 WITNESS_RESTORE(&mtx->lock_object, mtx); 312 return (rval); 313} 314 315/* 316 * pause() is like tsleep() except that the intention is to not be 317 * explicitly woken up by another thread. Instead, the current thread 318 * simply wishes to sleep until the timeout expires. It is 319 * implemented using a dummy wait channel. 320 */ 321int 322pause(wmesg, timo) 323 const char *wmesg; 324 int timo; 325{ 326 327 KASSERT(timo != 0, ("pause: timeout required")); 328 return (tsleep(&pause_wchan, 0, wmesg, timo)); 329} 330 331/* 332 * Make all threads sleeping on the specified identifier runnable. 333 */ 334void 335wakeup(ident) 336 register void *ident; 337{ 338 339 sleepq_lock(ident); 340 sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 341 sleepq_release(ident); 342} 343 344/* 345 * Make a thread sleeping on the specified identifier runnable. 346 * May wake more than one thread if a target thread is currently 347 * swapped out. 348 */ 349void 350wakeup_one(ident) 351 register void *ident; 352{ 353 354 sleepq_lock(ident); 355 sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 356 sleepq_release(ident); 357} 358 359static void 360kdb_switch(void) 361{ 362 thread_unlock(curthread); 363 kdb_backtrace(); 364 kdb_reenter(); 365 panic("%s: did not reenter debugger", __func__); 366} 367 368/* 369 * The machine independent parts of context switching. 370 */ 371void 372mi_switch(int flags, struct thread *newtd) 373{ 374 uint64_t runtime, new_switchtime; 375 struct thread *td; 376 struct proc *p; 377 378 td = curthread; /* XXX */ 379 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 380 p = td->td_proc; /* XXX */ 381 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 382#ifdef INVARIANTS 383 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 384 mtx_assert(&Giant, MA_NOTOWNED); 385#endif 386 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 387 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 388 newtd == NULL) || panicstr, 389 ("mi_switch: switch in a critical section")); 390 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 391 ("mi_switch: switch must be voluntary or involuntary")); 392 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 393 394 /* 395 * Don't perform context switches from the debugger. 396 */ 397 if (kdb_active) 398 kdb_switch(); 399 if (flags & SW_VOL) 400 td->td_ru.ru_nvcsw++; 401 else 402 td->td_ru.ru_nivcsw++; 403 /* 404 * Compute the amount of time during which the current 405 * thread was running, and add that to its total so far. 406 */ 407 new_switchtime = cpu_ticks(); 408 runtime = new_switchtime - PCPU_GET(switchtime); 409 td->td_runtime += runtime; 410 td->td_incruntime += runtime; 411 PCPU_SET(switchtime, new_switchtime); 412 td->td_generation++; /* bump preempt-detect counter */ 413 PCPU_INC(cnt.v_swtch); 414 PCPU_SET(switchticks, ticks); 415 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 416 td->td_tid, td->td_sched, p->p_pid, td->td_name); 417#if (KTR_COMPILE & KTR_SCHED) != 0 418 if (TD_IS_IDLETHREAD(td)) 419 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 420 td, td->td_name, td->td_priority); 421 else if (newtd != NULL) 422 CTR5(KTR_SCHED, 423 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 424 td, td->td_name, td->td_priority, newtd, 425 newtd->td_name); 426 else 427 CTR6(KTR_SCHED, 428 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 429 td, td->td_name, td->td_priority, 430 td->td_inhibitors, td->td_wmesg, td->td_lockname); 431#endif 432 sched_switch(td, newtd, flags); 433 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 434 td, td->td_name, td->td_priority); 435 436 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 437 td->td_tid, td->td_sched, p->p_pid, td->td_name); 438 439 /* 440 * If the last thread was exiting, finish cleaning it up. 441 */ 442 if ((td = PCPU_GET(deadthread))) { 443 PCPU_SET(deadthread, NULL); 444 thread_stash(td); 445 } 446} 447 448/* 449 * Change process state to be runnable, 450 * placing it on the run queue if it is in memory, 451 * and awakening the swapper if it isn't in memory. 452 */ 453void 454setrunnable(struct thread *td) 455{ 456 457 THREAD_LOCK_ASSERT(td, MA_OWNED); 458 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 459 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 460 switch (td->td_state) { 461 case TDS_RUNNING: 462 case TDS_RUNQ: 463 return; 464 case TDS_INHIBITED: 465 /* 466 * If we are only inhibited because we are swapped out 467 * then arange to swap in this process. Otherwise just return. 468 */ 469 if (td->td_inhibitors != TDI_SWAPPED) 470 return; 471 /* XXX: intentional fall-through ? */ 472 case TDS_CAN_RUN: 473 break; 474 default: 475 printf("state is 0x%x", td->td_state); 476 panic("setrunnable(2)"); 477 } 478 if ((td->td_flags & TDF_INMEM) == 0) { 479 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 480 td->td_flags |= TDF_SWAPINREQ; 481 /* 482 * due to a LOR between the thread lock and 483 * the sleepqueue chain locks, use 484 * lower level scheduling functions. 485 */ 486 kick_proc0(); 487 } 488 } else 489 sched_wakeup(td); 490} 491 492/* 493 * Compute a tenex style load average of a quantity on 494 * 1, 5 and 15 minute intervals. 495 */ 496static void 497loadav(void *arg) 498{ 499 int i, nrun; 500 struct loadavg *avg; 501 502 nrun = sched_load(); 503 avg = &averunnable; 504 505 for (i = 0; i < 3; i++) 506 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 507 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 508 509 /* 510 * Schedule the next update to occur after 5 seconds, but add a 511 * random variation to avoid synchronisation with processes that 512 * run at regular intervals. 513 */ 514 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 515 loadav, NULL); 516} 517 518static void 519lboltcb(void *arg) 520{ 521 wakeup(&lbolt); 522 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 523} 524 525/* ARGSUSED */ 526static void 527synch_setup(dummy) 528 void *dummy; 529{ 530 callout_init(&loadav_callout, CALLOUT_MPSAFE); 531 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 532 533 /* Kick off timeout driven events by calling first time. */ 534 loadav(NULL); 535 lboltcb(NULL); 536} 537 538/* 539 * General purpose yield system call. 540 */ 541int 542yield(struct thread *td, struct yield_args *uap) 543{ 544 545 thread_lock(td); 546 sched_prio(td, PRI_MAX_TIMESHARE); 547 mi_switch(SW_VOL, NULL); 548 thread_unlock(td); 549 td->td_retval[0] = 0; 550 return (0); 551} 552