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 * 3. 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/sdt.h> 55 #include <sys/signalvar.h> 56 #include <sys/sleepqueue.h> 57 #include <sys/smp.h> 58 #include <sys/sx.h> 59 #include <sys/sysctl.h> 60 #include <sys/sysproto.h> 61 #include <sys/vmmeter.h> 62 #ifdef KTRACE 63 #include <sys/uio.h> 64 #include <sys/ktrace.h> 65 #endif 66 67 #include <machine/cpu.h> 68 69 static void synch_setup(void *dummy); 70 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 71 NULL); 72 73 int hogticks; 74 static uint8_t pause_wchan[MAXCPU]; 75 76 static struct callout loadav_callout; 77 78 struct loadavg averunnable = 79 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 80 /* 81 * Constants for averages over 1, 5, and 15 minutes 82 * when sampling at 5 second intervals. 83 */ 84 static fixpt_t cexp[3] = { 85 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 86 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 87 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 88 }; 89 90 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 91 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, ""); 92 93 static void loadav(void *arg); 94 95 SDT_PROVIDER_DECLARE(sched); 96 SDT_PROBE_DEFINE(sched, , , preempt); 97 98 static void 99 sleepinit(void *unused) 100 { 101 102 hogticks = (hz / 10) * 2; /* Default only. */ 103 init_sleepqueues(); 104 } 105 106 /* 107 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure 108 * it is available. 109 */ 110 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0); 111 112 /* 113 * General sleep call. Suspends the current thread until a wakeup is 114 * performed on the specified identifier. The thread will then be made 115 * runnable with the specified priority. Sleeps at most sbt units of time 116 * (0 means no timeout). If pri includes the PCATCH flag, let signals 117 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if 118 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 119 * signal becomes pending, ERESTART is returned if the current system 120 * call should be restarted if possible, and EINTR is returned if the system 121 * call should be interrupted by the signal (return EINTR). 122 * 123 * The lock argument is unlocked before the caller is suspended, and 124 * re-locked before _sleep() returns. If priority includes the PDROP 125 * flag the lock is not re-locked before returning. 126 */ 127 int 128 _sleep(void *ident, struct lock_object *lock, int priority, 129 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 130 { 131 struct thread *td; 132 struct proc *p; 133 struct lock_class *class; 134 uintptr_t lock_state; 135 int catch, pri, rval, sleepq_flags; 136 WITNESS_SAVE_DECL(lock_witness); 137 138 td = curthread; 139 p = td->td_proc; 140 #ifdef KTRACE 141 if (KTRPOINT(td, KTR_CSW)) 142 ktrcsw(1, 0, wmesg); 143 #endif 144 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 145 "Sleeping on \"%s\"", wmesg); 146 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, 147 ("sleeping without a lock")); 148 KASSERT(ident != NULL, ("_sleep: NULL ident")); 149 KASSERT(TD_IS_RUNNING(td), ("_sleep: curthread not running")); 150 if (priority & PDROP) 151 KASSERT(lock != NULL && lock != &Giant.lock_object, 152 ("PDROP requires a non-Giant lock")); 153 if (lock != NULL) 154 class = LOCK_CLASS(lock); 155 else 156 class = NULL; 157 158 if (SCHEDULER_STOPPED_TD(td)) { 159 if (lock != NULL && priority & PDROP) 160 class->lc_unlock(lock); 161 return (0); 162 } 163 catch = priority & PCATCH; 164 pri = priority & PRIMASK; 165 166 KASSERT(!TD_ON_SLEEPQ(td), ("recursive sleep")); 167 168 if ((uint8_t *)ident >= &pause_wchan[0] && 169 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) 170 sleepq_flags = SLEEPQ_PAUSE; 171 else 172 sleepq_flags = SLEEPQ_SLEEP; 173 if (catch) 174 sleepq_flags |= SLEEPQ_INTERRUPTIBLE; 175 176 sleepq_lock(ident); 177 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 178 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 179 180 if (lock == &Giant.lock_object) 181 mtx_assert(&Giant, MA_OWNED); 182 DROP_GIANT(); 183 if (lock != NULL && lock != &Giant.lock_object && 184 !(class->lc_flags & LC_SLEEPABLE)) { 185 WITNESS_SAVE(lock, lock_witness); 186 lock_state = class->lc_unlock(lock); 187 } else 188 /* GCC needs to follow the Yellow Brick Road */ 189 lock_state = -1; 190 191 /* 192 * We put ourselves on the sleep queue and start our timeout 193 * before calling thread_suspend_check, as we could stop there, 194 * and a wakeup or a SIGCONT (or both) could occur while we were 195 * stopped without resuming us. Thus, we must be ready for sleep 196 * when cursig() is called. If the wakeup happens while we're 197 * stopped, then td will no longer be on a sleep queue upon 198 * return from cursig(). 199 */ 200 sleepq_add(ident, lock, wmesg, sleepq_flags, 0); 201 if (sbt != 0) 202 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 203 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 204 sleepq_release(ident); 205 WITNESS_SAVE(lock, lock_witness); 206 lock_state = class->lc_unlock(lock); 207 sleepq_lock(ident); 208 } 209 if (sbt != 0 && catch) 210 rval = sleepq_timedwait_sig(ident, pri); 211 else if (sbt != 0) 212 rval = sleepq_timedwait(ident, pri); 213 else if (catch) 214 rval = sleepq_wait_sig(ident, pri); 215 else { 216 sleepq_wait(ident, pri); 217 rval = 0; 218 } 219 #ifdef KTRACE 220 if (KTRPOINT(td, KTR_CSW)) 221 ktrcsw(0, 0, wmesg); 222 #endif 223 PICKUP_GIANT(); 224 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 225 class->lc_lock(lock, lock_state); 226 WITNESS_RESTORE(lock, lock_witness); 227 } 228 return (rval); 229 } 230 231 int 232 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, 233 sbintime_t sbt, sbintime_t pr, int flags) 234 { 235 struct thread *td; 236 struct proc *p; 237 int rval; 238 WITNESS_SAVE_DECL(mtx); 239 240 td = curthread; 241 p = td->td_proc; 242 KASSERT(mtx != NULL, ("sleeping without a mutex")); 243 KASSERT(ident != NULL, ("msleep_spin_sbt: NULL ident")); 244 KASSERT(TD_IS_RUNNING(td), ("msleep_spin_sbt: curthread not running")); 245 246 if (SCHEDULER_STOPPED_TD(td)) 247 return (0); 248 249 sleepq_lock(ident); 250 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 251 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 252 253 DROP_GIANT(); 254 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 255 WITNESS_SAVE(&mtx->lock_object, mtx); 256 mtx_unlock_spin(mtx); 257 258 /* 259 * We put ourselves on the sleep queue and start our timeout. 260 */ 261 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 262 if (sbt != 0) 263 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 264 265 /* 266 * Can't call ktrace with any spin locks held so it can lock the 267 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 268 * any spin lock. Thus, we have to drop the sleepq spin lock while 269 * we handle those requests. This is safe since we have placed our 270 * thread on the sleep queue already. 271 */ 272 #ifdef KTRACE 273 if (KTRPOINT(td, KTR_CSW)) { 274 sleepq_release(ident); 275 ktrcsw(1, 0, wmesg); 276 sleepq_lock(ident); 277 } 278 #endif 279 #ifdef WITNESS 280 sleepq_release(ident); 281 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 282 wmesg); 283 sleepq_lock(ident); 284 #endif 285 if (sbt != 0) 286 rval = sleepq_timedwait(ident, 0); 287 else { 288 sleepq_wait(ident, 0); 289 rval = 0; 290 } 291 #ifdef KTRACE 292 if (KTRPOINT(td, KTR_CSW)) 293 ktrcsw(0, 0, wmesg); 294 #endif 295 PICKUP_GIANT(); 296 mtx_lock_spin(mtx); 297 WITNESS_RESTORE(&mtx->lock_object, mtx); 298 return (rval); 299 } 300 301 /* 302 * pause() delays the calling thread by the given number of system ticks. 303 * During cold bootup, pause() uses the DELAY() function instead of 304 * the tsleep() function to do the waiting. The "timo" argument must be 305 * greater than or equal to zero. A "timo" value of zero is equivalent 306 * to a "timo" value of one. 307 */ 308 int 309 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 310 { 311 KASSERT(sbt >= 0, ("pause: timeout must be >= 0")); 312 313 /* silently convert invalid timeouts */ 314 if (sbt == 0) 315 sbt = tick_sbt; 316 317 if ((cold && curthread == &thread0) || kdb_active || 318 SCHEDULER_STOPPED()) { 319 /* 320 * We delay one second at a time to avoid overflowing the 321 * system specific DELAY() function(s): 322 */ 323 while (sbt >= SBT_1S) { 324 DELAY(1000000); 325 sbt -= SBT_1S; 326 } 327 /* Do the delay remainder, if any */ 328 sbt = howmany(sbt, SBT_1US); 329 if (sbt > 0) 330 DELAY(sbt); 331 return (0); 332 } 333 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); 334 } 335 336 /* 337 * Make all threads sleeping on the specified identifier runnable. 338 */ 339 void 340 wakeup(void *ident) 341 { 342 int wakeup_swapper; 343 344 sleepq_lock(ident); 345 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 346 sleepq_release(ident); 347 if (wakeup_swapper) { 348 KASSERT(ident != &proc0, 349 ("wakeup and wakeup_swapper and proc0")); 350 kick_proc0(); 351 } 352 } 353 354 /* 355 * Make a thread sleeping on the specified identifier runnable. 356 * May wake more than one thread if a target thread is currently 357 * swapped out. 358 */ 359 void 360 wakeup_one(void *ident) 361 { 362 int wakeup_swapper; 363 364 sleepq_lock(ident); 365 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 366 sleepq_release(ident); 367 if (wakeup_swapper) 368 kick_proc0(); 369 } 370 371 static void 372 kdb_switch(void) 373 { 374 thread_unlock(curthread); 375 kdb_backtrace(); 376 kdb_reenter(); 377 panic("%s: did not reenter debugger", __func__); 378 } 379 380 /* 381 * The machine independent parts of context switching. 382 */ 383 void 384 mi_switch(int flags, struct thread *newtd) 385 { 386 uint64_t runtime, new_switchtime; 387 struct thread *td; 388 389 td = curthread; /* XXX */ 390 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 391 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 392 #ifdef INVARIANTS 393 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 394 mtx_assert(&Giant, MA_NOTOWNED); 395 #endif 396 KASSERT(td->td_critnest == 1 || panicstr, 397 ("mi_switch: switch in a critical section")); 398 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 399 ("mi_switch: switch must be voluntary or involuntary")); 400 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 401 402 /* 403 * Don't perform context switches from the debugger. 404 */ 405 if (kdb_active) 406 kdb_switch(); 407 if (SCHEDULER_STOPPED_TD(td)) 408 return; 409 if (flags & SW_VOL) { 410 td->td_ru.ru_nvcsw++; 411 td->td_swvoltick = ticks; 412 } else { 413 td->td_ru.ru_nivcsw++; 414 td->td_swinvoltick = ticks; 415 } 416 #ifdef SCHED_STATS 417 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 418 #endif 419 /* 420 * Compute the amount of time during which the current 421 * thread was running, and add that to its total so far. 422 */ 423 new_switchtime = cpu_ticks(); 424 runtime = new_switchtime - PCPU_GET(switchtime); 425 td->td_runtime += runtime; 426 td->td_incruntime += runtime; 427 PCPU_SET(switchtime, new_switchtime); 428 td->td_generation++; /* bump preempt-detect counter */ 429 PCPU_INC(cnt.v_swtch); 430 PCPU_SET(switchticks, ticks); 431 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 432 td->td_tid, td_get_sched(td), td->td_proc->p_pid, td->td_name); 433 #ifdef KDTRACE_HOOKS 434 if ((flags & SW_PREEMPT) != 0 || ((flags & SW_INVOL) != 0 && 435 (flags & SW_TYPE_MASK) == SWT_NEEDRESCHED)) 436 SDT_PROBE0(sched, , , preempt); 437 #endif 438 sched_switch(td, newtd, flags); 439 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 440 td->td_tid, td_get_sched(td), td->td_proc->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 */ 456 int 457 setrunnable(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 */ 495 static void 496 loadav(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_sbt(&loadav_callout, 514 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, 515 loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); 516 } 517 518 /* ARGSUSED */ 519 static void 520 synch_setup(void *dummy) 521 { 522 callout_init(&loadav_callout, 1); 523 524 /* Kick off timeout driven events by calling first time. */ 525 loadav(NULL); 526 } 527 528 int 529 should_yield(void) 530 { 531 532 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); 533 } 534 535 void 536 maybe_yield(void) 537 { 538 539 if (should_yield()) 540 kern_yield(PRI_USER); 541 } 542 543 void 544 kern_yield(int prio) 545 { 546 struct thread *td; 547 548 td = curthread; 549 DROP_GIANT(); 550 thread_lock(td); 551 if (prio == PRI_USER) 552 prio = td->td_user_pri; 553 if (prio >= 0) 554 sched_prio(td, prio); 555 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 556 thread_unlock(td); 557 PICKUP_GIANT(); 558 } 559 560 /* 561 * General purpose yield system call. 562 */ 563 int 564 sys_yield(struct thread *td, struct yield_args *uap) 565 { 566 567 thread_lock(td); 568 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 569 sched_prio(td, PRI_MAX_TIMESHARE); 570 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 571 thread_unlock(td); 572 td->td_retval[0] = 0; 573 return (0); 574 } 575