kern_synch.c revision 181394
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 181394 2008-08-07 21:00:13Z jhb $"); 39 40#include "opt_ktrace.h" 41#include "opt_sched.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/condvar.h> 46#include <sys/kdb.h> 47#include <sys/kernel.h> 48#include <sys/ktr.h> 49#include <sys/lock.h> 50#include <sys/mutex.h> 51#include <sys/proc.h> 52#include <sys/resourcevar.h> 53#include <sys/sched.h> 54#include <sys/signalvar.h> 55#include <sys/sleepqueue.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 KTRACE 62#include <sys/uio.h> 63#include <sys/ktrace.h> 64#endif 65 66#include <machine/cpu.h> 67 68static void synch_setup(void *dummy); 69SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 70 NULL); 71 72int hogticks; 73int lbolt; 74static int pause_wchan; 75 76static struct callout loadav_callout; 77static struct callout lbolt_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 91/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 92static int fscale __unused = FSCALE; 93SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 94 95static void loadav(void *arg); 96static void lboltcb(void *arg); 97 98void 99sleepinit(void) 100{ 101 102 hogticks = (hz / 10) * 2; /* Default only. */ 103 init_sleepqueues(); 104} 105 106/* 107 * General sleep call. Suspends the current thread until a wakeup is 108 * performed on the specified identifier. The thread will then be made 109 * runnable with the specified priority. Sleeps at most timo/hz seconds 110 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 111 * before and after sleeping, else signals are not checked. Returns 0 if 112 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 113 * signal needs to be delivered, ERESTART is returned if the current system 114 * call should be restarted if possible, and EINTR is returned if the system 115 * call should be interrupted by the signal (return EINTR). 116 * 117 * The lock argument is unlocked before the caller is suspended, and 118 * re-locked before _sleep() returns. If priority includes the PDROP 119 * flag the lock is not re-locked before returning. 120 */ 121int 122_sleep(void *ident, struct lock_object *lock, int priority, 123 const char *wmesg, int timo) 124{ 125 struct thread *td; 126 struct proc *p; 127 struct lock_class *class; 128 int catch, flags, lock_state, pri, rval; 129 WITNESS_SAVE_DECL(lock_witness); 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, lock, 138 "Sleeping on \"%s\"", wmesg); 139 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL || 140 ident == &lbolt, ("sleeping without a lock")); 141 KASSERT(p != NULL, ("msleep1")); 142 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 143 if (priority & PDROP) 144 KASSERT(lock != NULL && lock != &Giant.lock_object, 145 ("PDROP requires a non-Giant lock")); 146 if (lock != NULL) 147 class = LOCK_CLASS(lock); 148 else 149 class = NULL; 150 151 if (cold) { 152 /* 153 * During autoconfiguration, just return; 154 * don't run any other threads or panic below, 155 * in case this is the idle thread and already asleep. 156 * XXX: this used to do "s = splhigh(); splx(safepri); 157 * splx(s);" to give interrupts a chance, but there is 158 * no way to give interrupts a chance now. 159 */ 160 if (lock != NULL && priority & PDROP) 161 class->lc_unlock(lock); 162 return (0); 163 } 164 catch = priority & PCATCH; 165 pri = priority & PRIMASK; 166 rval = 0; 167 168 /* 169 * If we are already on a sleep queue, then remove us from that 170 * sleep queue first. We have to do this to handle recursive 171 * sleeps. 172 */ 173 if (TD_ON_SLEEPQ(td)) 174 sleepq_remove(td, td->td_wchan); 175 176 if (ident == &pause_wchan) 177 flags = SLEEPQ_PAUSE; 178 else 179 flags = SLEEPQ_SLEEP; 180 if (catch) 181 flags |= SLEEPQ_INTERRUPTIBLE; 182 183 sleepq_lock(ident); 184 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 185 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 186 187 DROP_GIANT(); 188 if (lock != NULL && lock != &Giant.lock_object && 189 !(class->lc_flags & LC_SLEEPABLE)) { 190 WITNESS_SAVE(lock, lock_witness); 191 lock_state = class->lc_unlock(lock); 192 } else 193 /* GCC needs to follow the Yellow Brick Road */ 194 lock_state = -1; 195 196 /* 197 * We put ourselves on the sleep queue and start our timeout 198 * before calling thread_suspend_check, as we could stop there, 199 * and a wakeup or a SIGCONT (or both) could occur while we were 200 * stopped without resuming us. Thus, we must be ready for sleep 201 * when cursig() is called. If the wakeup happens while we're 202 * stopped, then td will no longer be on a sleep queue upon 203 * return from cursig(). 204 */ 205 sleepq_add(ident, ident == &lbolt ? NULL : lock, wmesg, flags, 0); 206 if (timo) 207 sleepq_set_timeout(ident, timo); 208 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 209 sleepq_release(ident); 210 WITNESS_SAVE(lock, lock_witness); 211 lock_state = class->lc_unlock(lock); 212 sleepq_lock(ident); 213 } 214 if (timo && catch) 215 rval = sleepq_timedwait_sig(ident, pri); 216 else if (timo) 217 rval = sleepq_timedwait(ident, pri); 218 else if (catch) 219 rval = sleepq_wait_sig(ident, pri); 220 else { 221 sleepq_wait(ident, pri); 222 rval = 0; 223 } 224#ifdef KTRACE 225 if (KTRPOINT(td, KTR_CSW)) 226 ktrcsw(0, 0); 227#endif 228 PICKUP_GIANT(); 229 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 230 class->lc_lock(lock, lock_state); 231 WITNESS_RESTORE(lock, lock_witness); 232 } 233 return (rval); 234} 235 236int 237msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, 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, td->td_name, wmesg, ident); 265 266 DROP_GIANT(); 267 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 268 WITNESS_SAVE(&mtx->lock_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->lock_object, wmesg, SLEEPQ_SLEEP, 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, 0); 300 else { 301 sleepq_wait(ident, 0); 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->lock_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(const char *wmesg, int timo) 322{ 323 324 KASSERT(timo != 0, ("pause: timeout required")); 325 return (tsleep(&pause_wchan, 0, wmesg, timo)); 326} 327 328/* 329 * Make all threads sleeping on the specified identifier runnable. 330 */ 331void 332wakeup(void *ident) 333{ 334 int wakeup_swapper; 335 336 sleepq_lock(ident); 337 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 338 sleepq_release(ident); 339 if (wakeup_swapper) 340 kick_proc0(); 341} 342 343/* 344 * Make a thread sleeping on the specified identifier runnable. 345 * May wake more than one thread if a target thread is currently 346 * swapped out. 347 */ 348void 349wakeup_one(void *ident) 350{ 351 int wakeup_swapper; 352 353 sleepq_lock(ident); 354 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 355 sleepq_release(ident); 356 if (wakeup_swapper) 357 kick_proc0(); 358} 359 360static void 361kdb_switch(void) 362{ 363 thread_unlock(curthread); 364 kdb_backtrace(); 365 kdb_reenter(); 366 panic("%s: did not reenter debugger", __func__); 367} 368 369/* 370 * The machine independent parts of context switching. 371 */ 372void 373mi_switch(int flags, struct thread *newtd) 374{ 375 uint64_t runtime, new_switchtime; 376 struct thread *td; 377 struct proc *p; 378 379 td = curthread; /* XXX */ 380 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 381 p = td->td_proc; /* XXX */ 382 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 383#ifdef INVARIANTS 384 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 385 mtx_assert(&Giant, MA_NOTOWNED); 386#endif 387 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 388 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 389 newtd == NULL) || panicstr, 390 ("mi_switch: switch in a critical section")); 391 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 392 ("mi_switch: switch must be voluntary or involuntary")); 393 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 394 395 /* 396 * Don't perform context switches from the debugger. 397 */ 398 if (kdb_active) 399 kdb_switch(); 400 if (flags & SW_VOL) 401 td->td_ru.ru_nvcsw++; 402 else 403 td->td_ru.ru_nivcsw++; 404#ifdef SCHED_STATS 405 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 406#endif 407 /* 408 * Compute the amount of time during which the current 409 * thread was running, and add that to its total so far. 410 */ 411 new_switchtime = cpu_ticks(); 412 runtime = new_switchtime - PCPU_GET(switchtime); 413 td->td_runtime += runtime; 414 td->td_incruntime += runtime; 415 PCPU_SET(switchtime, new_switchtime); 416 td->td_generation++; /* bump preempt-detect counter */ 417 PCPU_INC(cnt.v_swtch); 418 PCPU_SET(switchticks, ticks); 419 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 420 td->td_tid, td->td_sched, p->p_pid, td->td_name); 421#if (KTR_COMPILE & KTR_SCHED) != 0 422 if (TD_IS_IDLETHREAD(td)) 423 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 424 td, td->td_name, td->td_priority); 425 else if (newtd != NULL) 426 CTR5(KTR_SCHED, 427 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 428 td, td->td_name, td->td_priority, newtd, 429 newtd->td_name); 430 else 431 CTR6(KTR_SCHED, 432 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 433 td, td->td_name, td->td_priority, 434 td->td_inhibitors, td->td_wmesg, td->td_lockname); 435#endif 436 sched_switch(td, newtd, flags); 437 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 438 td, td->td_name, td->td_priority); 439 440 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 441 td->td_tid, td->td_sched, p->p_pid, td->td_name); 442 443 /* 444 * If the last thread was exiting, finish cleaning it up. 445 */ 446 if ((td = PCPU_GET(deadthread))) { 447 PCPU_SET(deadthread, NULL); 448 thread_stash(td); 449 } 450} 451 452/* 453 * Change thread state to be runnable, placing it on the run queue if 454 * it is in memory. If it is swapped out, return true so our caller 455 * will know to awaken the swapper. 456 */ 457int 458setrunnable(struct thread *td) 459{ 460 461 THREAD_LOCK_ASSERT(td, MA_OWNED); 462 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 463 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 464 switch (td->td_state) { 465 case TDS_RUNNING: 466 case TDS_RUNQ: 467 return (0); 468 case TDS_INHIBITED: 469 /* 470 * If we are only inhibited because we are swapped out 471 * then arange to swap in this process. Otherwise just return. 472 */ 473 if (td->td_inhibitors != TDI_SWAPPED) 474 return (0); 475 /* FALLTHROUGH */ 476 case TDS_CAN_RUN: 477 break; 478 default: 479 printf("state is 0x%x", td->td_state); 480 panic("setrunnable(2)"); 481 } 482 if ((td->td_flags & TDF_INMEM) == 0) { 483 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 484 td->td_flags |= TDF_SWAPINREQ; 485 return (1); 486 } 487 } else 488 sched_wakeup(td); 489 return (0); 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(void *dummy) 528{ 529 callout_init(&loadav_callout, CALLOUT_MPSAFE); 530 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 531 532 /* Kick off timeout driven events by calling first time. */ 533 loadav(NULL); 534 lboltcb(NULL); 535} 536 537/* 538 * General purpose yield system call. 539 */ 540int 541yield(struct thread *td, struct yield_args *uap) 542{ 543 544 thread_lock(td); 545 sched_prio(td, PRI_MAX_TIMESHARE); 546 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 547 thread_unlock(td); 548 td->td_retval[0] = 0; 549 return (0); 550} 551