kern_synch.c revision 183352
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 183352 2008-09-25 13:42:19Z 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; 73static int pause_wchan; 74 75static struct callout loadav_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); 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 lock argument is unlocked before the caller is suspended, and 115 * re-locked before _sleep() returns. If priority includes the PDROP 116 * flag the lock is not re-locked before returning. 117 */ 118int 119_sleep(void *ident, struct lock_object *lock, int priority, 120 const char *wmesg, int timo) 121{ 122 struct thread *td; 123 struct proc *p; 124 struct lock_class *class; 125 int catch, flags, lock_state, pri, rval; 126 WITNESS_SAVE_DECL(lock_witness); 127 128 td = curthread; 129 p = td->td_proc; 130#ifdef KTRACE 131 if (KTRPOINT(td, KTR_CSW)) 132 ktrcsw(1, 0); 133#endif 134 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 135 "Sleeping on \"%s\"", wmesg); 136 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL, 137 ("sleeping without a lock")); 138 KASSERT(p != NULL, ("msleep1")); 139 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 140 if (priority & PDROP) 141 KASSERT(lock != NULL && lock != &Giant.lock_object, 142 ("PDROP requires a non-Giant lock")); 143 if (lock != NULL) 144 class = LOCK_CLASS(lock); 145 else 146 class = NULL; 147 148 if (cold) { 149 /* 150 * During autoconfiguration, just return; 151 * don't run any other threads or panic below, 152 * in case this is the idle thread and already asleep. 153 * XXX: this used to do "s = splhigh(); splx(safepri); 154 * splx(s);" to give interrupts a chance, but there is 155 * no way to give interrupts a chance now. 156 */ 157 if (lock != NULL && priority & PDROP) 158 class->lc_unlock(lock); 159 return (0); 160 } 161 catch = priority & PCATCH; 162 pri = priority & PRIMASK; 163 rval = 0; 164 165 /* 166 * If we are already on a sleep queue, then remove us from that 167 * sleep queue first. We have to do this to handle recursive 168 * sleeps. 169 */ 170 if (TD_ON_SLEEPQ(td)) 171 sleepq_remove(td, td->td_wchan); 172 173 if (ident == &pause_wchan) 174 flags = SLEEPQ_PAUSE; 175 else 176 flags = SLEEPQ_SLEEP; 177 if (catch) 178 flags |= SLEEPQ_INTERRUPTIBLE; 179 180 sleepq_lock(ident); 181 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 182 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 183 184 if (lock == &Giant.lock_object) 185 mtx_assert(&Giant, MA_OWNED); 186 DROP_GIANT(); 187 if (lock != NULL && lock != &Giant.lock_object && 188 !(class->lc_flags & LC_SLEEPABLE)) { 189 WITNESS_SAVE(lock, lock_witness); 190 lock_state = class->lc_unlock(lock); 191 } else 192 /* GCC needs to follow the Yellow Brick Road */ 193 lock_state = -1; 194 195 /* 196 * We put ourselves on the sleep queue and start our timeout 197 * before calling thread_suspend_check, as we could stop there, 198 * and a wakeup or a SIGCONT (or both) could occur while we were 199 * stopped without resuming us. Thus, we must be ready for sleep 200 * when cursig() is called. If the wakeup happens while we're 201 * stopped, then td will no longer be on a sleep queue upon 202 * return from cursig(). 203 */ 204 sleepq_add(ident, lock, wmesg, flags, 0); 205 if (timo) 206 sleepq_set_timeout(ident, timo); 207 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 208 sleepq_release(ident); 209 WITNESS_SAVE(lock, lock_witness); 210 lock_state = class->lc_unlock(lock); 211 sleepq_lock(ident); 212 } 213 if (timo && catch) 214 rval = sleepq_timedwait_sig(ident, pri); 215 else if (timo) 216 rval = sleepq_timedwait(ident, pri); 217 else if (catch) 218 rval = sleepq_wait_sig(ident, pri); 219 else { 220 sleepq_wait(ident, pri); 221 rval = 0; 222 } 223#ifdef KTRACE 224 if (KTRPOINT(td, KTR_CSW)) 225 ktrcsw(0, 0); 226#endif 227 PICKUP_GIANT(); 228 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 229 class->lc_lock(lock, lock_state); 230 WITNESS_RESTORE(lock, lock_witness); 231 } 232 return (rval); 233} 234 235int 236msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, int timo) 237{ 238 struct thread *td; 239 struct proc *p; 240 int rval; 241 WITNESS_SAVE_DECL(mtx); 242 243 td = curthread; 244 p = td->td_proc; 245 KASSERT(mtx != NULL, ("sleeping without a mutex")); 246 KASSERT(p != NULL, ("msleep1")); 247 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 248 249 if (cold) { 250 /* 251 * During autoconfiguration, just return; 252 * don't run any other threads or panic below, 253 * in case this is the idle thread and already asleep. 254 * XXX: this used to do "s = splhigh(); splx(safepri); 255 * splx(s);" to give interrupts a chance, but there is 256 * no way to give interrupts a chance now. 257 */ 258 return (0); 259 } 260 261 sleepq_lock(ident); 262 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 263 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 264 265 DROP_GIANT(); 266 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 267 WITNESS_SAVE(&mtx->lock_object, mtx); 268 mtx_unlock_spin(mtx); 269 270 /* 271 * We put ourselves on the sleep queue and start our timeout. 272 */ 273 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 274 if (timo) 275 sleepq_set_timeout(ident, timo); 276 277 /* 278 * Can't call ktrace with any spin locks held so it can lock the 279 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 280 * any spin lock. Thus, we have to drop the sleepq spin lock while 281 * we handle those requests. This is safe since we have placed our 282 * thread on the sleep queue already. 283 */ 284#ifdef KTRACE 285 if (KTRPOINT(td, KTR_CSW)) { 286 sleepq_release(ident); 287 ktrcsw(1, 0); 288 sleepq_lock(ident); 289 } 290#endif 291#ifdef WITNESS 292 sleepq_release(ident); 293 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 294 wmesg); 295 sleepq_lock(ident); 296#endif 297 if (timo) 298 rval = sleepq_timedwait(ident, 0); 299 else { 300 sleepq_wait(ident, 0); 301 rval = 0; 302 } 303#ifdef KTRACE 304 if (KTRPOINT(td, KTR_CSW)) 305 ktrcsw(0, 0); 306#endif 307 PICKUP_GIANT(); 308 mtx_lock_spin(mtx); 309 WITNESS_RESTORE(&mtx->lock_object, mtx); 310 return (rval); 311} 312 313/* 314 * pause() is like tsleep() except that the intention is to not be 315 * explicitly woken up by another thread. Instead, the current thread 316 * simply wishes to sleep until the timeout expires. It is 317 * implemented using a dummy wait channel. 318 */ 319int 320pause(const char *wmesg, int timo) 321{ 322 323 KASSERT(timo != 0, ("pause: timeout required")); 324 return (tsleep(&pause_wchan, 0, wmesg, timo)); 325} 326 327/* 328 * Make all threads sleeping on the specified identifier runnable. 329 */ 330void 331wakeup(void *ident) 332{ 333 int wakeup_swapper; 334 335 sleepq_lock(ident); 336 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 337 sleepq_release(ident); 338 if (wakeup_swapper) 339 kick_proc0(); 340} 341 342/* 343 * Make a thread sleeping on the specified identifier runnable. 344 * May wake more than one thread if a target thread is currently 345 * swapped out. 346 */ 347void 348wakeup_one(void *ident) 349{ 350 int wakeup_swapper; 351 352 sleepq_lock(ident); 353 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 354 sleepq_release(ident); 355 if (wakeup_swapper) 356 kick_proc0(); 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#ifdef SCHED_STATS 404 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 405#endif 406 /* 407 * Compute the amount of time during which the current 408 * thread was running, and add that to its total so far. 409 */ 410 new_switchtime = cpu_ticks(); 411 runtime = new_switchtime - PCPU_GET(switchtime); 412 td->td_runtime += runtime; 413 td->td_incruntime += runtime; 414 PCPU_SET(switchtime, new_switchtime); 415 td->td_generation++; /* bump preempt-detect counter */ 416 PCPU_INC(cnt.v_swtch); 417 PCPU_SET(switchticks, ticks); 418 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 419 td->td_tid, td->td_sched, p->p_pid, td->td_name); 420#if (KTR_COMPILE & KTR_SCHED) != 0 421 if (TD_IS_IDLETHREAD(td)) 422 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 423 td, td->td_name, td->td_priority); 424 else if (newtd != NULL) 425 CTR5(KTR_SCHED, 426 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 427 td, td->td_name, td->td_priority, newtd, 428 newtd->td_name); 429 else 430 CTR6(KTR_SCHED, 431 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 432 td, td->td_name, td->td_priority, 433 td->td_inhibitors, td->td_wmesg, td->td_lockname); 434#endif 435 sched_switch(td, newtd, flags); 436 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 437 td, td->td_name, td->td_priority); 438 439 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 440 td->td_tid, td->td_sched, p->p_pid, td->td_name); 441 442 /* 443 * If the last thread was exiting, finish cleaning it up. 444 */ 445 if ((td = PCPU_GET(deadthread))) { 446 PCPU_SET(deadthread, NULL); 447 thread_stash(td); 448 } 449} 450 451/* 452 * Change thread state to be runnable, placing it on the run queue if 453 * it is in memory. If it is swapped out, return true so our caller 454 * will know to awaken the swapper. 455 */ 456int 457setrunnable(struct thread *td) 458{ 459 460 THREAD_LOCK_ASSERT(td, MA_OWNED); 461 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 462 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 463 switch (td->td_state) { 464 case TDS_RUNNING: 465 case TDS_RUNQ: 466 return (0); 467 case TDS_INHIBITED: 468 /* 469 * If we are only inhibited because we are swapped out 470 * then arange to swap in this process. Otherwise just return. 471 */ 472 if (td->td_inhibitors != TDI_SWAPPED) 473 return (0); 474 /* FALLTHROUGH */ 475 case TDS_CAN_RUN: 476 break; 477 default: 478 printf("state is 0x%x", td->td_state); 479 panic("setrunnable(2)"); 480 } 481 if ((td->td_flags & TDF_INMEM) == 0) { 482 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 483 td->td_flags |= TDF_SWAPINREQ; 484 return (1); 485 } 486 } else 487 sched_wakeup(td); 488 return (0); 489} 490 491/* 492 * Compute a tenex style load average of a quantity on 493 * 1, 5 and 15 minute intervals. 494 */ 495static void 496loadav(void *arg) 497{ 498 int i, nrun; 499 struct loadavg *avg; 500 501 nrun = sched_load(); 502 avg = &averunnable; 503 504 for (i = 0; i < 3; i++) 505 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 506 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 507 508 /* 509 * Schedule the next update to occur after 5 seconds, but add a 510 * random variation to avoid synchronisation with processes that 511 * run at regular intervals. 512 */ 513 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 514 loadav, NULL); 515} 516 517/* ARGSUSED */ 518static void 519synch_setup(void *dummy) 520{ 521 callout_init(&loadav_callout, CALLOUT_MPSAFE); 522 523 /* Kick off timeout driven events by calling first time. */ 524 loadav(NULL); 525} 526 527/* 528 * General purpose yield system call. 529 */ 530int 531yield(struct thread *td, struct yield_args *uap) 532{ 533 534 thread_lock(td); 535 sched_prio(td, PRI_MAX_TIMESHARE); 536 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 537 thread_unlock(td); 538 td->td_retval[0] = 0; 539 return (0); 540} 541