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("$FreeBSD$"); 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 68 #ifdef XEN 69 #include <vm/vm.h> 70 #include <vm/vm_param.h> 71 #include <vm/pmap.h> 72 #endif 73 74 #define KTDSTATE(td) \ 75 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \ 76 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \ 77 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \ 78 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \ 79 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding") 80 81 static void synch_setup(void *dummy); 82 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 83 NULL); 84 85 int hogticks; 86 static int pause_wchan; 87 88 static struct callout loadav_callout; 89 90 struct loadavg averunnable = 91 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 92 /* 93 * Constants for averages over 1, 5, and 15 minutes 94 * when sampling at 5 second intervals. 95 */ 96 static fixpt_t cexp[3] = { 97 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 98 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 99 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 100 }; 101 102 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 103 static int fscale __unused = FSCALE; 104 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 105 106 static void loadav(void *arg); 107 108 void 109 sleepinit(void) 110 { 111 112 hogticks = (hz / 10) * 2; /* Default only. */ 113 init_sleepqueues(); 114 } 115 116 /* 117 * General sleep call. Suspends the current thread until a wakeup is 118 * performed on the specified identifier. The thread will then be made 119 * runnable with the specified priority. Sleeps at most timo/hz seconds 120 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 121 * before and after sleeping, else signals are not checked. Returns 0 if 122 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 123 * signal needs to be delivered, ERESTART is returned if the current system 124 * call should be restarted if possible, and EINTR is returned if the system 125 * call should be interrupted by the signal (return EINTR). 126 * 127 * The lock argument is unlocked before the caller is suspended, and 128 * re-locked before _sleep() returns. If priority includes the PDROP 129 * flag the lock is not re-locked before returning. 130 */ 131 int 132 _sleep(void *ident, struct lock_object *lock, int priority, 133 const char *wmesg, int timo) 134 { 135 struct thread *td; 136 struct proc *p; 137 struct lock_class *class; 138 int catch, flags, lock_state, pri, rval; 139 WITNESS_SAVE_DECL(lock_witness); 140 141 td = curthread; 142 p = td->td_proc; 143 #ifdef KTRACE 144 if (KTRPOINT(td, KTR_CSW)) 145 ktrcsw(1, 0); 146 #endif 147 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 148 "Sleeping on \"%s\"", wmesg); 149 KASSERT(timo != 0 || mtx_owned(&Giant) || lock != NULL, 150 ("sleeping without a lock")); 151 KASSERT(p != NULL, ("msleep1")); 152 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 153 if (priority & PDROP) 154 KASSERT(lock != NULL && lock != &Giant.lock_object, 155 ("PDROP requires a non-Giant lock")); 156 if (lock != NULL) 157 class = LOCK_CLASS(lock); 158 else 159 class = NULL; 160 161 if (cold || SCHEDULER_STOPPED()) { 162 /* 163 * During autoconfiguration, just return; 164 * don't run any other threads or panic below, 165 * in case this is the idle thread and already asleep. 166 * XXX: this used to do "s = splhigh(); splx(safepri); 167 * splx(s);" to give interrupts a chance, but there is 168 * no way to give interrupts a chance now. 169 */ 170 if (lock != NULL && priority & PDROP) 171 class->lc_unlock(lock); 172 return (0); 173 } 174 catch = priority & PCATCH; 175 pri = priority & PRIMASK; 176 177 /* 178 * If we are already on a sleep queue, then remove us from that 179 * sleep queue first. We have to do this to handle recursive 180 * sleeps. 181 */ 182 if (TD_ON_SLEEPQ(td)) 183 sleepq_remove(td, td->td_wchan); 184 185 if (ident == &pause_wchan) 186 flags = SLEEPQ_PAUSE; 187 else 188 flags = SLEEPQ_SLEEP; 189 if (catch) 190 flags |= SLEEPQ_INTERRUPTIBLE; 191 if (priority & PBDRY) 192 flags |= SLEEPQ_STOP_ON_BDRY; 193 194 sleepq_lock(ident); 195 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 196 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 197 198 if (lock == &Giant.lock_object) 199 mtx_assert(&Giant, MA_OWNED); 200 DROP_GIANT(); 201 if (lock != NULL && lock != &Giant.lock_object && 202 !(class->lc_flags & LC_SLEEPABLE)) { 203 WITNESS_SAVE(lock, lock_witness); 204 lock_state = class->lc_unlock(lock); 205 } else 206 /* GCC needs to follow the Yellow Brick Road */ 207 lock_state = -1; 208 209 /* 210 * We put ourselves on the sleep queue and start our timeout 211 * before calling thread_suspend_check, as we could stop there, 212 * and a wakeup or a SIGCONT (or both) could occur while we were 213 * stopped without resuming us. Thus, we must be ready for sleep 214 * when cursig() is called. If the wakeup happens while we're 215 * stopped, then td will no longer be on a sleep queue upon 216 * return from cursig(). 217 */ 218 sleepq_add(ident, lock, wmesg, flags, 0); 219 if (timo) 220 sleepq_set_timeout(ident, timo); 221 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 222 sleepq_release(ident); 223 WITNESS_SAVE(lock, lock_witness); 224 lock_state = class->lc_unlock(lock); 225 sleepq_lock(ident); 226 } 227 if (timo && catch) 228 rval = sleepq_timedwait_sig(ident, pri); 229 else if (timo) 230 rval = sleepq_timedwait(ident, pri); 231 else if (catch) 232 rval = sleepq_wait_sig(ident, pri); 233 else { 234 sleepq_wait(ident, pri); 235 rval = 0; 236 } 237 #ifdef KTRACE 238 if (KTRPOINT(td, KTR_CSW)) 239 ktrcsw(0, 0); 240 #endif 241 PICKUP_GIANT(); 242 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 243 class->lc_lock(lock, lock_state); 244 WITNESS_RESTORE(lock, lock_witness); 245 } 246 return (rval); 247 } 248 249 int 250 msleep_spin(void *ident, struct mtx *mtx, const char *wmesg, int timo) 251 { 252 struct thread *td; 253 struct proc *p; 254 int rval; 255 WITNESS_SAVE_DECL(mtx); 256 257 td = curthread; 258 p = td->td_proc; 259 KASSERT(mtx != NULL, ("sleeping without a mutex")); 260 KASSERT(p != NULL, ("msleep1")); 261 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 262 263 if (cold || SCHEDULER_STOPPED()) { 264 /* 265 * During autoconfiguration, just return; 266 * don't run any other threads or panic below, 267 * in case this is the idle thread and already asleep. 268 * XXX: this used to do "s = splhigh(); splx(safepri); 269 * splx(s);" to give interrupts a chance, but there is 270 * no way to give interrupts a chance now. 271 */ 272 return (0); 273 } 274 275 sleepq_lock(ident); 276 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 277 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 278 279 DROP_GIANT(); 280 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 281 WITNESS_SAVE(&mtx->lock_object, mtx); 282 mtx_unlock_spin(mtx); 283 284 /* 285 * We put ourselves on the sleep queue and start our timeout. 286 */ 287 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 288 if (timo) 289 sleepq_set_timeout(ident, timo); 290 291 /* 292 * Can't call ktrace with any spin locks held so it can lock the 293 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 294 * any spin lock. Thus, we have to drop the sleepq spin lock while 295 * we handle those requests. This is safe since we have placed our 296 * thread on the sleep queue already. 297 */ 298 #ifdef KTRACE 299 if (KTRPOINT(td, KTR_CSW)) { 300 sleepq_release(ident); 301 ktrcsw(1, 0); 302 sleepq_lock(ident); 303 } 304 #endif 305 #ifdef WITNESS 306 sleepq_release(ident); 307 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 308 wmesg); 309 sleepq_lock(ident); 310 #endif 311 if (timo) 312 rval = sleepq_timedwait(ident, 0); 313 else { 314 sleepq_wait(ident, 0); 315 rval = 0; 316 } 317 #ifdef KTRACE 318 if (KTRPOINT(td, KTR_CSW)) 319 ktrcsw(0, 0); 320 #endif 321 PICKUP_GIANT(); 322 mtx_lock_spin(mtx); 323 WITNESS_RESTORE(&mtx->lock_object, mtx); 324 return (rval); 325 } 326 327 /* 328 * pause() delays the calling thread by the given number of system ticks. 329 * During cold bootup, pause() uses the DELAY() function instead of 330 * the tsleep() function to do the waiting. The "timo" argument must be 331 * greater than or equal to zero. A "timo" value of zero is equivalent 332 * to a "timo" value of one. 333 */ 334 int 335 pause(const char *wmesg, int timo) 336 { 337 KASSERT(timo >= 0, ("pause: timo must be >= 0")); 338 339 /* silently convert invalid timeouts */ 340 if (timo < 1) 341 timo = 1; 342 343 if (cold) { 344 /* 345 * We delay one HZ at a time to avoid overflowing the 346 * system specific DELAY() function(s): 347 */ 348 while (timo >= hz) { 349 DELAY(1000000); 350 timo -= hz; 351 } 352 if (timo > 0) 353 DELAY(timo * tick); 354 return (0); 355 } 356 return (tsleep(&pause_wchan, 0, wmesg, timo)); 357 } 358 359 /* 360 * Make all threads sleeping on the specified identifier runnable. 361 */ 362 void 363 wakeup(void *ident) 364 { 365 int wakeup_swapper; 366 367 sleepq_lock(ident); 368 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 369 sleepq_release(ident); 370 if (wakeup_swapper) { 371 KASSERT(ident != &proc0, 372 ("wakeup and wakeup_swapper and proc0")); 373 kick_proc0(); 374 } 375 } 376 377 /* 378 * Make a thread sleeping on the specified identifier runnable. 379 * May wake more than one thread if a target thread is currently 380 * swapped out. 381 */ 382 void 383 wakeup_one(void *ident) 384 { 385 int wakeup_swapper; 386 387 sleepq_lock(ident); 388 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 389 sleepq_release(ident); 390 if (wakeup_swapper) 391 kick_proc0(); 392 } 393 394 static void 395 kdb_switch(void) 396 { 397 thread_unlock(curthread); 398 kdb_backtrace(); 399 kdb_reenter(); 400 panic("%s: did not reenter debugger", __func__); 401 } 402 403 /* 404 * The machine independent parts of context switching. 405 */ 406 void 407 mi_switch(int flags, struct thread *newtd) 408 { 409 uint64_t runtime, new_switchtime; 410 struct thread *td; 411 struct proc *p; 412 413 td = curthread; /* XXX */ 414 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 415 p = td->td_proc; /* XXX */ 416 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 417 #ifdef INVARIANTS 418 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 419 mtx_assert(&Giant, MA_NOTOWNED); 420 #endif 421 KASSERT(td->td_critnest == 1 || panicstr, 422 ("mi_switch: switch in a critical section")); 423 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 424 ("mi_switch: switch must be voluntary or involuntary")); 425 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 426 427 /* 428 * Don't perform context switches from the debugger. 429 */ 430 if (kdb_active) 431 kdb_switch(); 432 if (SCHEDULER_STOPPED()) 433 return; 434 if (flags & SW_VOL) { 435 td->td_ru.ru_nvcsw++; 436 td->td_swvoltick = ticks; 437 } else 438 td->td_ru.ru_nivcsw++; 439 #ifdef SCHED_STATS 440 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 441 #endif 442 /* 443 * Compute the amount of time during which the current 444 * thread was running, and add that to its total so far. 445 */ 446 new_switchtime = cpu_ticks(); 447 runtime = new_switchtime - PCPU_GET(switchtime); 448 td->td_runtime += runtime; 449 td->td_incruntime += runtime; 450 PCPU_SET(switchtime, new_switchtime); 451 td->td_generation++; /* bump preempt-detect counter */ 452 PCPU_INC(cnt.v_swtch); 453 PCPU_SET(switchticks, ticks); 454 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 455 td->td_tid, td->td_sched, p->p_pid, td->td_name); 456 #if (KTR_COMPILE & KTR_SCHED) != 0 457 if (TD_IS_IDLETHREAD(td)) 458 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", 459 "prio:%d", td->td_priority); 460 else 461 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), 462 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, 463 "lockname:\"%s\"", td->td_lockname); 464 #endif 465 #ifdef XEN 466 PT_UPDATES_FLUSH(); 467 #endif 468 sched_switch(td, newtd, flags); 469 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", 470 "prio:%d", td->td_priority); 471 472 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 473 td->td_tid, td->td_sched, p->p_pid, td->td_name); 474 475 /* 476 * If the last thread was exiting, finish cleaning it up. 477 */ 478 if ((td = PCPU_GET(deadthread))) { 479 PCPU_SET(deadthread, NULL); 480 thread_stash(td); 481 } 482 } 483 484 /* 485 * Change thread state to be runnable, placing it on the run queue if 486 * it is in memory. If it is swapped out, return true so our caller 487 * will know to awaken the swapper. 488 */ 489 int 490 setrunnable(struct thread *td) 491 { 492 493 THREAD_LOCK_ASSERT(td, MA_OWNED); 494 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 495 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 496 switch (td->td_state) { 497 case TDS_RUNNING: 498 case TDS_RUNQ: 499 return (0); 500 case TDS_INHIBITED: 501 /* 502 * If we are only inhibited because we are swapped out 503 * then arange to swap in this process. Otherwise just return. 504 */ 505 if (td->td_inhibitors != TDI_SWAPPED) 506 return (0); 507 /* FALLTHROUGH */ 508 case TDS_CAN_RUN: 509 break; 510 default: 511 printf("state is 0x%x", td->td_state); 512 panic("setrunnable(2)"); 513 } 514 if ((td->td_flags & TDF_INMEM) == 0) { 515 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 516 td->td_flags |= TDF_SWAPINREQ; 517 return (1); 518 } 519 } else 520 sched_wakeup(td); 521 return (0); 522 } 523 524 /* 525 * Compute a tenex style load average of a quantity on 526 * 1, 5 and 15 minute intervals. 527 */ 528 static void 529 loadav(void *arg) 530 { 531 int i, nrun; 532 struct loadavg *avg; 533 534 nrun = sched_load(); 535 avg = &averunnable; 536 537 for (i = 0; i < 3; i++) 538 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 539 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 540 541 /* 542 * Schedule the next update to occur after 5 seconds, but add a 543 * random variation to avoid synchronisation with processes that 544 * run at regular intervals. 545 */ 546 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 547 loadav, NULL); 548 } 549 550 /* ARGSUSED */ 551 static void 552 synch_setup(void *dummy) 553 { 554 callout_init(&loadav_callout, CALLOUT_MPSAFE); 555 556 /* Kick off timeout driven events by calling first time. */ 557 loadav(NULL); 558 } 559 560 int 561 should_yield(void) 562 { 563 564 return (ticks - curthread->td_swvoltick >= hogticks); 565 } 566 567 void 568 maybe_yield(void) 569 { 570 571 if (should_yield()) 572 kern_yield(PRI_USER); 573 } 574 575 void 576 kern_yield(int prio) 577 { 578 struct thread *td; 579 580 td = curthread; 581 DROP_GIANT(); 582 thread_lock(td); 583 if (prio == PRI_USER) 584 prio = td->td_user_pri; 585 if (prio >= 0) 586 sched_prio(td, prio); 587 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 588 thread_unlock(td); 589 PICKUP_GIANT(); 590 } 591 592 /* 593 * General purpose yield system call. 594 */ 595 int 596 sys_yield(struct thread *td, struct yield_args *uap) 597 { 598 599 thread_lock(td); 600 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 601 sched_prio(td, PRI_MAX_TIMESHARE); 602 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 603 thread_unlock(td); 604 td->td_retval[0] = 0; 605 return (0); 606 } 607