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 static int fscale __unused = FSCALE; 105 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 106 107 static void loadav(void *arg); 108 109 SDT_PROVIDER_DECLARE(sched); 110 SDT_PROBE_DEFINE(sched, , , preempt); 111 112 /* 113 * These probes reference Solaris features that are not implemented in FreeBSD. 114 * Create the probes anyway for compatibility with existing D scripts; they'll 115 * just never fire. 116 */ 117 SDT_PROBE_DEFINE(sched, , , cpucaps__sleep); 118 SDT_PROBE_DEFINE(sched, , , cpucaps__wakeup); 119 SDT_PROBE_DEFINE(sched, , , schedctl__nopreempt); 120 SDT_PROBE_DEFINE(sched, , , schedctl__preempt); 121 SDT_PROBE_DEFINE(sched, , , schedctl__yield); 122 123 static void 124 sleepinit(void *unused) 125 { 126 127 hogticks = (hz / 10) * 2; /* Default only. */ 128 init_sleepqueues(); 129 } 130 131 /* 132 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure 133 * it is available. 134 */ 135 SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0); 136 137 /* 138 * General sleep call. Suspends the current thread until a wakeup is 139 * performed on the specified identifier. The thread will then be made 140 * runnable with the specified priority. Sleeps at most sbt units of time 141 * (0 means no timeout). If pri includes the PCATCH flag, let signals 142 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if 143 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 144 * signal becomes pending, ERESTART is returned if the current system 145 * call should be restarted if possible, and EINTR is returned if the system 146 * call should be interrupted by the signal (return EINTR). 147 * 148 * The lock argument is unlocked before the caller is suspended, and 149 * re-locked before _sleep() returns. If priority includes the PDROP 150 * flag the lock is not re-locked before returning. 151 */ 152 int 153 _sleep(void *ident, struct lock_object *lock, int priority, 154 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 155 { 156 struct thread *td; 157 struct proc *p; 158 struct lock_class *class; 159 uintptr_t lock_state; 160 int catch, pri, rval, sleepq_flags; 161 WITNESS_SAVE_DECL(lock_witness); 162 163 td = curthread; 164 p = td->td_proc; 165 #ifdef KTRACE 166 if (KTRPOINT(td, KTR_CSW)) 167 ktrcsw(1, 0, wmesg); 168 #endif 169 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 170 "Sleeping on \"%s\"", wmesg); 171 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, 172 ("sleeping without a lock")); 173 KASSERT(p != NULL, ("msleep1")); 174 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 175 if (priority & PDROP) 176 KASSERT(lock != NULL && lock != &Giant.lock_object, 177 ("PDROP requires a non-Giant lock")); 178 if (lock != NULL) 179 class = LOCK_CLASS(lock); 180 else 181 class = NULL; 182 183 if (cold || SCHEDULER_STOPPED()) { 184 /* 185 * During autoconfiguration, just return; 186 * don't run any other threads or panic below, 187 * in case this is the idle thread and already asleep. 188 * XXX: this used to do "s = splhigh(); splx(safepri); 189 * splx(s);" to give interrupts a chance, but there is 190 * no way to give interrupts a chance now. 191 */ 192 if (lock != NULL && priority & PDROP) 193 class->lc_unlock(lock); 194 return (0); 195 } 196 catch = priority & PCATCH; 197 pri = priority & PRIMASK; 198 199 /* 200 * If we are already on a sleep queue, then remove us from that 201 * sleep queue first. We have to do this to handle recursive 202 * sleeps. 203 */ 204 if (TD_ON_SLEEPQ(td)) 205 sleepq_remove(td, td->td_wchan); 206 207 if ((uint8_t *)ident >= &pause_wchan[0] && 208 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) 209 sleepq_flags = SLEEPQ_PAUSE; 210 else 211 sleepq_flags = SLEEPQ_SLEEP; 212 if (catch) 213 sleepq_flags |= SLEEPQ_INTERRUPTIBLE; 214 215 sleepq_lock(ident); 216 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 217 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 218 219 if (lock == &Giant.lock_object) 220 mtx_assert(&Giant, MA_OWNED); 221 DROP_GIANT(); 222 if (lock != NULL && lock != &Giant.lock_object && 223 !(class->lc_flags & LC_SLEEPABLE)) { 224 WITNESS_SAVE(lock, lock_witness); 225 lock_state = class->lc_unlock(lock); 226 } else 227 /* GCC needs to follow the Yellow Brick Road */ 228 lock_state = -1; 229 230 /* 231 * We put ourselves on the sleep queue and start our timeout 232 * before calling thread_suspend_check, as we could stop there, 233 * and a wakeup or a SIGCONT (or both) could occur while we were 234 * stopped without resuming us. Thus, we must be ready for sleep 235 * when cursig() is called. If the wakeup happens while we're 236 * stopped, then td will no longer be on a sleep queue upon 237 * return from cursig(). 238 */ 239 sleepq_add(ident, lock, wmesg, sleepq_flags, 0); 240 if (sbt != 0) 241 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 242 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 243 sleepq_release(ident); 244 WITNESS_SAVE(lock, lock_witness); 245 lock_state = class->lc_unlock(lock); 246 sleepq_lock(ident); 247 } 248 if (sbt != 0 && catch) 249 rval = sleepq_timedwait_sig(ident, pri); 250 else if (sbt != 0) 251 rval = sleepq_timedwait(ident, pri); 252 else if (catch) 253 rval = sleepq_wait_sig(ident, pri); 254 else { 255 sleepq_wait(ident, pri); 256 rval = 0; 257 } 258 #ifdef KTRACE 259 if (KTRPOINT(td, KTR_CSW)) 260 ktrcsw(0, 0, wmesg); 261 #endif 262 PICKUP_GIANT(); 263 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 264 class->lc_lock(lock, lock_state); 265 WITNESS_RESTORE(lock, lock_witness); 266 } 267 return (rval); 268 } 269 270 int 271 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, 272 sbintime_t sbt, sbintime_t pr, int flags) 273 { 274 struct thread *td; 275 struct proc *p; 276 int rval; 277 WITNESS_SAVE_DECL(mtx); 278 279 td = curthread; 280 p = td->td_proc; 281 KASSERT(mtx != NULL, ("sleeping without a mutex")); 282 KASSERT(p != NULL, ("msleep1")); 283 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 284 285 if (cold || SCHEDULER_STOPPED()) { 286 /* 287 * During autoconfiguration, just return; 288 * don't run any other threads or panic below, 289 * in case this is the idle thread and already asleep. 290 * XXX: this used to do "s = splhigh(); splx(safepri); 291 * splx(s);" to give interrupts a chance, but there is 292 * no way to give interrupts a chance now. 293 */ 294 return (0); 295 } 296 297 sleepq_lock(ident); 298 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 299 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 300 301 DROP_GIANT(); 302 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 303 WITNESS_SAVE(&mtx->lock_object, mtx); 304 mtx_unlock_spin(mtx); 305 306 /* 307 * We put ourselves on the sleep queue and start our timeout. 308 */ 309 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 310 if (sbt != 0) 311 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 312 313 /* 314 * Can't call ktrace with any spin locks held so it can lock the 315 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 316 * any spin lock. Thus, we have to drop the sleepq spin lock while 317 * we handle those requests. This is safe since we have placed our 318 * thread on the sleep queue already. 319 */ 320 #ifdef KTRACE 321 if (KTRPOINT(td, KTR_CSW)) { 322 sleepq_release(ident); 323 ktrcsw(1, 0, wmesg); 324 sleepq_lock(ident); 325 } 326 #endif 327 #ifdef WITNESS 328 sleepq_release(ident); 329 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 330 wmesg); 331 sleepq_lock(ident); 332 #endif 333 if (sbt != 0) 334 rval = sleepq_timedwait(ident, 0); 335 else { 336 sleepq_wait(ident, 0); 337 rval = 0; 338 } 339 #ifdef KTRACE 340 if (KTRPOINT(td, KTR_CSW)) 341 ktrcsw(0, 0, wmesg); 342 #endif 343 PICKUP_GIANT(); 344 mtx_lock_spin(mtx); 345 WITNESS_RESTORE(&mtx->lock_object, mtx); 346 return (rval); 347 } 348 349 /* 350 * pause() delays the calling thread by the given number of system ticks. 351 * During cold bootup, pause() uses the DELAY() function instead of 352 * the tsleep() function to do the waiting. The "timo" argument must be 353 * greater than or equal to zero. A "timo" value of zero is equivalent 354 * to a "timo" value of one. 355 */ 356 int 357 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 358 { 359 KASSERT(sbt >= 0, ("pause: timeout must be >= 0")); 360 361 /* silently convert invalid timeouts */ 362 if (sbt == 0) 363 sbt = tick_sbt; 364 365 if (cold) { 366 /* 367 * We delay one second at a time to avoid overflowing the 368 * system specific DELAY() function(s): 369 */ 370 while (sbt >= SBT_1S) { 371 DELAY(1000000); 372 sbt -= SBT_1S; 373 } 374 /* Do the delay remainder, if any */ 375 sbt = (sbt + SBT_1US - 1) / SBT_1US; 376 if (sbt > 0) 377 DELAY(sbt); 378 return (0); 379 } 380 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); 381 } 382 383 /* 384 * Make all threads sleeping on the specified identifier runnable. 385 */ 386 void 387 wakeup(void *ident) 388 { 389 int wakeup_swapper; 390 391 sleepq_lock(ident); 392 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 393 sleepq_release(ident); 394 if (wakeup_swapper) { 395 KASSERT(ident != &proc0, 396 ("wakeup and wakeup_swapper and proc0")); 397 kick_proc0(); 398 } 399 } 400 401 /* 402 * Make a thread sleeping on the specified identifier runnable. 403 * May wake more than one thread if a target thread is currently 404 * swapped out. 405 */ 406 void 407 wakeup_one(void *ident) 408 { 409 int wakeup_swapper; 410 411 sleepq_lock(ident); 412 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 413 sleepq_release(ident); 414 if (wakeup_swapper) 415 kick_proc0(); 416 } 417 418 static void 419 kdb_switch(void) 420 { 421 thread_unlock(curthread); 422 kdb_backtrace(); 423 kdb_reenter(); 424 panic("%s: did not reenter debugger", __func__); 425 } 426 427 /* 428 * The machine independent parts of context switching. 429 */ 430 void 431 mi_switch(int flags, struct thread *newtd) 432 { 433 uint64_t runtime, new_switchtime; 434 struct thread *td; 435 struct proc *p; 436 437 td = curthread; /* XXX */ 438 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 439 p = td->td_proc; /* XXX */ 440 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 441 #ifdef INVARIANTS 442 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 443 mtx_assert(&Giant, MA_NOTOWNED); 444 #endif 445 KASSERT(td->td_critnest == 1 || panicstr, 446 ("mi_switch: switch in a critical section")); 447 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 448 ("mi_switch: switch must be voluntary or involuntary")); 449 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 450 451 /* 452 * Don't perform context switches from the debugger. 453 */ 454 if (kdb_active) 455 kdb_switch(); 456 if (SCHEDULER_STOPPED()) 457 return; 458 if (flags & SW_VOL) { 459 td->td_ru.ru_nvcsw++; 460 td->td_swvoltick = ticks; 461 } else 462 td->td_ru.ru_nivcsw++; 463 #ifdef SCHED_STATS 464 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 465 #endif 466 /* 467 * Compute the amount of time during which the current 468 * thread was running, and add that to its total so far. 469 */ 470 new_switchtime = cpu_ticks(); 471 runtime = new_switchtime - PCPU_GET(switchtime); 472 td->td_runtime += runtime; 473 td->td_incruntime += runtime; 474 PCPU_SET(switchtime, new_switchtime); 475 td->td_generation++; /* bump preempt-detect counter */ 476 PCPU_INC(cnt.v_swtch); 477 PCPU_SET(switchticks, ticks); 478 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 479 td->td_tid, td->td_sched, p->p_pid, td->td_name); 480 #if (KTR_COMPILE & KTR_SCHED) != 0 481 if (TD_IS_IDLETHREAD(td)) 482 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", 483 "prio:%d", td->td_priority); 484 else 485 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), 486 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, 487 "lockname:\"%s\"", td->td_lockname); 488 #endif 489 SDT_PROBE0(sched, , , preempt); 490 #ifdef XEN 491 PT_UPDATES_FLUSH(); 492 #endif 493 sched_switch(td, newtd, flags); 494 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", 495 "prio:%d", td->td_priority); 496 497 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 498 td->td_tid, td->td_sched, p->p_pid, td->td_name); 499 500 /* 501 * If the last thread was exiting, finish cleaning it up. 502 */ 503 if ((td = PCPU_GET(deadthread))) { 504 PCPU_SET(deadthread, NULL); 505 thread_stash(td); 506 } 507 } 508 509 /* 510 * Change thread state to be runnable, placing it on the run queue if 511 * it is in memory. If it is swapped out, return true so our caller 512 * will know to awaken the swapper. 513 */ 514 int 515 setrunnable(struct thread *td) 516 { 517 518 THREAD_LOCK_ASSERT(td, MA_OWNED); 519 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 520 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 521 switch (td->td_state) { 522 case TDS_RUNNING: 523 case TDS_RUNQ: 524 return (0); 525 case TDS_INHIBITED: 526 /* 527 * If we are only inhibited because we are swapped out 528 * then arange to swap in this process. Otherwise just return. 529 */ 530 if (td->td_inhibitors != TDI_SWAPPED) 531 return (0); 532 /* FALLTHROUGH */ 533 case TDS_CAN_RUN: 534 break; 535 default: 536 printf("state is 0x%x", td->td_state); 537 panic("setrunnable(2)"); 538 } 539 if ((td->td_flags & TDF_INMEM) == 0) { 540 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 541 td->td_flags |= TDF_SWAPINREQ; 542 return (1); 543 } 544 } else 545 sched_wakeup(td); 546 return (0); 547 } 548 549 /* 550 * Compute a tenex style load average of a quantity on 551 * 1, 5 and 15 minute intervals. 552 */ 553 static void 554 loadav(void *arg) 555 { 556 int i, nrun; 557 struct loadavg *avg; 558 559 nrun = sched_load(); 560 avg = &averunnable; 561 562 for (i = 0; i < 3; i++) 563 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 564 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 565 566 /* 567 * Schedule the next update to occur after 5 seconds, but add a 568 * random variation to avoid synchronisation with processes that 569 * run at regular intervals. 570 */ 571 callout_reset_sbt(&loadav_callout, 572 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, 573 loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); 574 } 575 576 /* ARGSUSED */ 577 static void 578 synch_setup(void *dummy) 579 { 580 callout_init(&loadav_callout, CALLOUT_MPSAFE); 581 582 /* Kick off timeout driven events by calling first time. */ 583 loadav(NULL); 584 } 585 586 int 587 should_yield(void) 588 { 589 590 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); 591 } 592 593 void 594 maybe_yield(void) 595 { 596 597 if (should_yield()) 598 kern_yield(PRI_USER); 599 } 600 601 void 602 kern_yield(int prio) 603 { 604 struct thread *td; 605 606 td = curthread; 607 DROP_GIANT(); 608 thread_lock(td); 609 if (prio == PRI_USER) 610 prio = td->td_user_pri; 611 if (prio >= 0) 612 sched_prio(td, prio); 613 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 614 thread_unlock(td); 615 PICKUP_GIANT(); 616 } 617 618 /* 619 * General purpose yield system call. 620 */ 621 int 622 sys_yield(struct thread *td, struct yield_args *uap) 623 { 624 625 thread_lock(td); 626 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 627 sched_prio(td, PRI_MAX_TIMESHARE); 628 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 629 thread_unlock(td); 630 td->td_retval[0] = 0; 631 return (0); 632 } 633