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/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 #ifdef XEN 70 #include <vm/vm.h> 71 #include <vm/vm_param.h> 72 #include <vm/pmap.h> 73 #endif 74 75 #define KTDSTATE(td) \ 76 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \ 77 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \ 78 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \ 79 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \ 80 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding") 81 82 static void synch_setup(void *dummy); 83 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 84 NULL); 85 86 int hogticks; 87 static uint8_t pause_wchan[MAXCPU]; 88 89 static struct callout loadav_callout; 90 91 struct loadavg averunnable = 92 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 93 /* 94 * Constants for averages over 1, 5, and 15 minutes 95 * when sampling at 5 second intervals. 96 */ 97 static fixpt_t cexp[3] = { 98 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 99 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 100 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 101 }; 102 103 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 104 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, ""); 105 106 static void loadav(void *arg); 107 108 SDT_PROVIDER_DECLARE(sched); 109 SDT_PROBE_DEFINE(sched, , , preempt); 110 111 /* 112 * These probes reference Solaris features that are not implemented in FreeBSD. 113 * Create the probes anyway for compatibility with existing D scripts; they'll 114 * just never fire. 115 */ 116 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep); 117 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup); 118 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt); 119 SDT_PROBE_DEFINE(sched, , , schedctl__preempt); 120 SDT_PROBE_DEFINE(sched, , , schedctl__yield); 121 122 static void 123 sleepinit(void *unused) 124 { 125 126 hogticks = (hz / 10) * 2; /* Default only. */ 127 init_sleepqueues(); 128 } 129 130 /* 131 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure 132 * it is available. 133 */ 134 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0); 135 136 /* 137 * General sleep call. Suspends the current thread until a wakeup is 138 * performed on the specified identifier. The thread will then be made 139 * runnable with the specified priority. Sleeps at most sbt units of time 140 * (0 means no timeout). If pri includes the PCATCH flag, let signals 141 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if 142 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 143 * signal becomes pending, ERESTART is returned if the current system 144 * call should be restarted if possible, and EINTR is returned if the system 145 * call should be interrupted by the signal (return EINTR). 146 * 147 * The lock argument is unlocked before the caller is suspended, and 148 * re-locked before _sleep() returns. If priority includes the PDROP 149 * flag the lock is not re-locked before returning. 150 */ 151 int 152 _sleep(void *ident, struct lock_object *lock, int priority, 153 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 154 { 155 struct thread *td; 156 struct proc *p; 157 struct lock_class *class; 158 uintptr_t lock_state; 159 int catch, pri, rval, sleepq_flags; 160 WITNESS_SAVE_DECL(lock_witness); 161 162 td = curthread; 163 p = td->td_proc; 164 #ifdef KTRACE 165 if (KTRPOINT(td, KTR_CSW)) 166 ktrcsw(1, 0, wmesg); 167 #endif 168 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 169 "Sleeping on \"%s\"", wmesg); 170 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, 171 ("sleeping without a lock")); 172 KASSERT(p != NULL, ("msleep1")); 173 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 174 if (priority & PDROP) 175 KASSERT(lock != NULL && lock != &Giant.lock_object, 176 ("PDROP requires a non-Giant lock")); 177 if (lock != NULL) 178 class = LOCK_CLASS(lock); 179 else 180 class = NULL; 181 182 if (cold || SCHEDULER_STOPPED()) { 183 /* 184 * During autoconfiguration, just return; 185 * don't run any other threads or panic below, 186 * in case this is the idle thread and already asleep. 187 * XXX: this used to do "s = splhigh(); splx(safepri); 188 * splx(s);" to give interrupts a chance, but there is 189 * no way to give interrupts a chance now. 190 */ 191 if (lock != NULL && priority & PDROP) 192 class->lc_unlock(lock); 193 return (0); 194 } 195 catch = priority & PCATCH; 196 pri = priority & PRIMASK; 197 198 /* 199 * If we are already on a sleep queue, then remove us from that 200 * sleep queue first. We have to do this to handle recursive 201 * sleeps. 202 */ 203 if (TD_ON_SLEEPQ(td)) 204 sleepq_remove(td, td->td_wchan); 205 206 if ((uint8_t *)ident >= &pause_wchan[0] && 207 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) 208 sleepq_flags = SLEEPQ_PAUSE; 209 else 210 sleepq_flags = SLEEPQ_SLEEP; 211 if (catch) 212 sleepq_flags |= SLEEPQ_INTERRUPTIBLE; 213 214 sleepq_lock(ident); 215 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 216 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 217 218 if (lock == &Giant.lock_object) 219 mtx_assert(&Giant, MA_OWNED); 220 DROP_GIANT(); 221 if (lock != NULL && lock != &Giant.lock_object && 222 !(class->lc_flags & LC_SLEEPABLE)) { 223 WITNESS_SAVE(lock, lock_witness); 224 lock_state = class->lc_unlock(lock); 225 } else 226 /* GCC needs to follow the Yellow Brick Road */ 227 lock_state = -1; 228 229 /* 230 * We put ourselves on the sleep queue and start our timeout 231 * before calling thread_suspend_check, as we could stop there, 232 * and a wakeup or a SIGCONT (or both) could occur while we were 233 * stopped without resuming us. Thus, we must be ready for sleep 234 * when cursig() is called. If the wakeup happens while we're 235 * stopped, then td will no longer be on a sleep queue upon 236 * return from cursig(). 237 */ 238 sleepq_add(ident, lock, wmesg, sleepq_flags, 0); 239 if (sbt != 0) 240 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 241 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 242 sleepq_release(ident); 243 WITNESS_SAVE(lock, lock_witness); 244 lock_state = class->lc_unlock(lock); 245 sleepq_lock(ident); 246 } 247 if (sbt != 0 && catch) 248 rval = sleepq_timedwait_sig(ident, pri); 249 else if (sbt != 0) 250 rval = sleepq_timedwait(ident, pri); 251 else if (catch) 252 rval = sleepq_wait_sig(ident, pri); 253 else { 254 sleepq_wait(ident, pri); 255 rval = 0; 256 } 257 #ifdef KTRACE 258 if (KTRPOINT(td, KTR_CSW)) 259 ktrcsw(0, 0, wmesg); 260 #endif 261 PICKUP_GIANT(); 262 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 263 class->lc_lock(lock, lock_state); 264 WITNESS_RESTORE(lock, lock_witness); 265 } 266 return (rval); 267 } 268 269 int 270 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, 271 sbintime_t sbt, sbintime_t pr, int flags) 272 { 273 struct thread *td; 274 struct proc *p; 275 int rval; 276 WITNESS_SAVE_DECL(mtx); 277 278 td = curthread; 279 p = td->td_proc; 280 KASSERT(mtx != NULL, ("sleeping without a mutex")); 281 KASSERT(p != NULL, ("msleep1")); 282 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 283 284 if (cold || SCHEDULER_STOPPED()) { 285 /* 286 * During autoconfiguration, just return; 287 * don't run any other threads or panic below, 288 * in case this is the idle thread and already asleep. 289 * XXX: this used to do "s = splhigh(); splx(safepri); 290 * splx(s);" to give interrupts a chance, but there is 291 * no way to give interrupts a chance now. 292 */ 293 return (0); 294 } 295 296 sleepq_lock(ident); 297 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 298 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 299 300 DROP_GIANT(); 301 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 302 WITNESS_SAVE(&mtx->lock_object, mtx); 303 mtx_unlock_spin(mtx); 304 305 /* 306 * We put ourselves on the sleep queue and start our timeout. 307 */ 308 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 309 if (sbt != 0) 310 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 311 312 /* 313 * Can't call ktrace with any spin locks held so it can lock the 314 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 315 * any spin lock. Thus, we have to drop the sleepq spin lock while 316 * we handle those requests. This is safe since we have placed our 317 * thread on the sleep queue already. 318 */ 319 #ifdef KTRACE 320 if (KTRPOINT(td, KTR_CSW)) { 321 sleepq_release(ident); 322 ktrcsw(1, 0, wmesg); 323 sleepq_lock(ident); 324 } 325 #endif 326 #ifdef WITNESS 327 sleepq_release(ident); 328 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 329 wmesg); 330 sleepq_lock(ident); 331 #endif 332 if (sbt != 0) 333 rval = sleepq_timedwait(ident, 0); 334 else { 335 sleepq_wait(ident, 0); 336 rval = 0; 337 } 338 #ifdef KTRACE 339 if (KTRPOINT(td, KTR_CSW)) 340 ktrcsw(0, 0, wmesg); 341 #endif 342 PICKUP_GIANT(); 343 mtx_lock_spin(mtx); 344 WITNESS_RESTORE(&mtx->lock_object, mtx); 345 return (rval); 346 } 347 348 /* 349 * pause() delays the calling thread by the given number of system ticks. 350 * During cold bootup, pause() uses the DELAY() function instead of 351 * the tsleep() function to do the waiting. The "timo" argument must be 352 * greater than or equal to zero. A "timo" value of zero is equivalent 353 * to a "timo" value of one. 354 */ 355 int 356 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 357 { 358 KASSERT(sbt >= 0, ("pause: timeout must be >= 0")); 359 360 /* silently convert invalid timeouts */ 361 if (sbt == 0) 362 sbt = tick_sbt; 363 364 if (cold || kdb_active) { 365 /* 366 * We delay one second at a time to avoid overflowing the 367 * system specific DELAY() function(s): 368 */ 369 while (sbt >= SBT_1S) { 370 DELAY(1000000); 371 sbt -= SBT_1S; 372 } 373 /* Do the delay remainder, if any */ 374 sbt = (sbt + SBT_1US - 1) / SBT_1US; 375 if (sbt > 0) 376 DELAY(sbt); 377 return (0); 378 } 379 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); 380 } 381 382 /* 383 * Make all threads sleeping on the specified identifier runnable. 384 */ 385 void 386 wakeup(void *ident) 387 { 388 int wakeup_swapper; 389 390 sleepq_lock(ident); 391 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 392 sleepq_release(ident); 393 if (wakeup_swapper) { 394 KASSERT(ident != &proc0, 395 ("wakeup and wakeup_swapper and proc0")); 396 kick_proc0(); 397 } 398 } 399 400 /* 401 * Make a thread sleeping on the specified identifier runnable. 402 * May wake more than one thread if a target thread is currently 403 * swapped out. 404 */ 405 void 406 wakeup_one(void *ident) 407 { 408 int wakeup_swapper; 409 410 sleepq_lock(ident); 411 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 412 sleepq_release(ident); 413 if (wakeup_swapper) 414 kick_proc0(); 415 } 416 417 static void 418 kdb_switch(void) 419 { 420 thread_unlock(curthread); 421 kdb_backtrace(); 422 kdb_reenter(); 423 panic("%s: did not reenter debugger", __func__); 424 } 425 426 /* 427 * The machine independent parts of context switching. 428 */ 429 void 430 mi_switch(int flags, struct thread *newtd) 431 { 432 uint64_t runtime, new_switchtime; 433 struct thread *td; 434 struct proc *p; 435 436 td = curthread; /* XXX */ 437 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 438 p = td->td_proc; /* XXX */ 439 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 440 #ifdef INVARIANTS 441 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 442 mtx_assert(&Giant, MA_NOTOWNED); 443 #endif 444 KASSERT(td->td_critnest == 1 || panicstr, 445 ("mi_switch: switch in a critical section")); 446 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 447 ("mi_switch: switch must be voluntary or involuntary")); 448 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 449 450 /* 451 * Don't perform context switches from the debugger. 452 */ 453 if (kdb_active) 454 kdb_switch(); 455 if (SCHEDULER_STOPPED()) 456 return; 457 if (flags & SW_VOL) { 458 td->td_ru.ru_nvcsw++; 459 td->td_swvoltick = ticks; 460 } else 461 td->td_ru.ru_nivcsw++; 462 #ifdef SCHED_STATS 463 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 464 #endif 465 /* 466 * Compute the amount of time during which the current 467 * thread was running, and add that to its total so far. 468 */ 469 new_switchtime = cpu_ticks(); 470 runtime = new_switchtime - PCPU_GET(switchtime); 471 td->td_runtime += runtime; 472 td->td_incruntime += runtime; 473 PCPU_SET(switchtime, new_switchtime); 474 td->td_generation++; /* bump preempt-detect counter */ 475 PCPU_INC(cnt.v_swtch); 476 PCPU_SET(switchticks, ticks); 477 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 478 td->td_tid, td->td_sched, p->p_pid, td->td_name); 479 #if (KTR_COMPILE & KTR_SCHED) != 0 480 if (TD_IS_IDLETHREAD(td)) 481 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", 482 "prio:%d", td->td_priority); 483 else 484 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), 485 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, 486 "lockname:\"%s\"", td->td_lockname); 487 #endif 488 SDT_PROBE0(sched, , , preempt); 489 #ifdef XEN 490 PT_UPDATES_FLUSH(); 491 #endif 492 sched_switch(td, newtd, flags); 493 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", 494 "prio:%d", td->td_priority); 495 496 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 497 td->td_tid, td->td_sched, p->p_pid, td->td_name); 498 499 /* 500 * If the last thread was exiting, finish cleaning it up. 501 */ 502 if ((td = PCPU_GET(deadthread))) { 503 PCPU_SET(deadthread, NULL); 504 thread_stash(td); 505 } 506 } 507 508 /* 509 * Change thread state to be runnable, placing it on the run queue if 510 * it is in memory. If it is swapped out, return true so our caller 511 * will know to awaken the swapper. 512 */ 513 int 514 setrunnable(struct thread *td) 515 { 516 517 THREAD_LOCK_ASSERT(td, MA_OWNED); 518 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 519 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 520 switch (td->td_state) { 521 case TDS_RUNNING: 522 case TDS_RUNQ: 523 return (0); 524 case TDS_INHIBITED: 525 /* 526 * If we are only inhibited because we are swapped out 527 * then arange to swap in this process. Otherwise just return. 528 */ 529 if (td->td_inhibitors != TDI_SWAPPED) 530 return (0); 531 /* FALLTHROUGH */ 532 case TDS_CAN_RUN: 533 break; 534 default: 535 printf("state is 0x%x", td->td_state); 536 panic("setrunnable(2)"); 537 } 538 if ((td->td_flags & TDF_INMEM) == 0) { 539 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 540 td->td_flags |= TDF_SWAPINREQ; 541 return (1); 542 } 543 } else 544 sched_wakeup(td); 545 return (0); 546 } 547 548 /* 549 * Compute a tenex style load average of a quantity on 550 * 1, 5 and 15 minute intervals. 551 */ 552 static void 553 loadav(void *arg) 554 { 555 int i, nrun; 556 struct loadavg *avg; 557 558 nrun = sched_load(); 559 avg = &averunnable; 560 561 for (i = 0; i < 3; i++) 562 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 563 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 564 565 /* 566 * Schedule the next update to occur after 5 seconds, but add a 567 * random variation to avoid synchronisation with processes that 568 * run at regular intervals. 569 */ 570 callout_reset_sbt(&loadav_callout, 571 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, 572 loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); 573 } 574 575 /* ARGSUSED */ 576 static void 577 synch_setup(void *dummy) 578 { 579 callout_init(&loadav_callout, CALLOUT_MPSAFE); 580 581 /* Kick off timeout driven events by calling first time. */ 582 loadav(NULL); 583 } 584 585 int 586 should_yield(void) 587 { 588 589 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); 590 } 591 592 void 593 maybe_yield(void) 594 { 595 596 if (should_yield()) 597 kern_yield(PRI_USER); 598 } 599 600 void 601 kern_yield(int prio) 602 { 603 struct thread *td; 604 605 td = curthread; 606 DROP_GIANT(); 607 thread_lock(td); 608 if (prio == PRI_USER) 609 prio = td->td_user_pri; 610 if (prio >= 0) 611 sched_prio(td, prio); 612 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 613 thread_unlock(td); 614 PICKUP_GIANT(); 615 } 616 617 /* 618 * General purpose yield system call. 619 */ 620 int 621 sys_yield(struct thread *td, struct yield_args *uap) 622 { 623 624 thread_lock(td); 625 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 626 sched_prio(td, PRI_MAX_TIMESHARE); 627 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 628 thread_unlock(td); 629 td->td_retval[0] = 0; 630 return (0); 631 } 632