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_kdtrace.h" 41 #include "opt_ktrace.h" 42 #include "opt_sched.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/condvar.h> 47 #include <sys/kdb.h> 48 #include <sys/kernel.h> 49 #include <sys/ktr.h> 50 #include <sys/lock.h> 51 #include <sys/mutex.h> 52 #include <sys/proc.h> 53 #include <sys/resourcevar.h> 54 #include <sys/sched.h> 55 #include <sys/sdt.h> 56 #include <sys/signalvar.h> 57 #include <sys/sleepqueue.h> 58 #include <sys/smp.h> 59 #include <sys/sx.h> 60 #include <sys/sysctl.h> 61 #include <sys/sysproto.h> 62 #include <sys/vmmeter.h> 63 #ifdef KTRACE 64 #include <sys/uio.h> 65 #include <sys/ktrace.h> 66 #endif 67 68 #include <machine/cpu.h> 69 70 #ifdef XEN 71 #include <vm/vm.h> 72 #include <vm/vm_param.h> 73 #include <vm/pmap.h> 74 #endif 75 76 #define KTDSTATE(td) \ 77 (((td)->td_inhibitors & TDI_SLEEPING) != 0 ? "sleep" : \ 78 ((td)->td_inhibitors & TDI_SUSPENDED) != 0 ? "suspended" : \ 79 ((td)->td_inhibitors & TDI_SWAPPED) != 0 ? "swapped" : \ 80 ((td)->td_inhibitors & TDI_LOCK) != 0 ? "blocked" : \ 81 ((td)->td_inhibitors & TDI_IWAIT) != 0 ? "iwait" : "yielding") 82 83 static void synch_setup(void *dummy); 84 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 85 NULL); 86 87 int hogticks; 88 static uint8_t pause_wchan[MAXCPU]; 89 90 static struct callout loadav_callout; 91 92 struct loadavg averunnable = 93 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 94 /* 95 * Constants for averages over 1, 5, and 15 minutes 96 * when sampling at 5 second intervals. 97 */ 98 static fixpt_t cexp[3] = { 99 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 100 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 101 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 102 }; 103 104 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 105 static int fscale __unused = FSCALE; 106 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 107 108 static void loadav(void *arg); 109 110 SDT_PROVIDER_DECLARE(sched); 111 SDT_PROBE_DEFINE(sched, , , preempt, preempt); 112 113 /* 114 * These probes reference Solaris features that are not implemented in FreeBSD. 115 * Create the probes anyway for compatibility with existing D scripts; they'll 116 * just never fire. 117 */ 118 SDT_PROBE_DEFINE(sched, , , cpucaps_sleep, cpucaps-sleep); 119 SDT_PROBE_DEFINE(sched, , , cpucaps_wakeup, cpucaps-wakeup); 120 SDT_PROBE_DEFINE(sched, , , schedctl_nopreempt, schedctl-nopreempt); 121 SDT_PROBE_DEFINE(sched, , , schedctl_preempt, schedctl-preempt); 122 SDT_PROBE_DEFINE(sched, , , schedctl_yield, schedctl-yield); 123 124 void 125 sleepinit(void) 126 { 127 128 hogticks = (hz / 10) * 2; /* Default only. */ 129 init_sleepqueues(); 130 } 131 132 /* 133 * General sleep call. Suspends the current thread until a wakeup is 134 * performed on the specified identifier. The thread will then be made 135 * runnable with the specified priority. Sleeps at most sbt units of time 136 * (0 means no timeout). If pri includes the PCATCH flag, let signals 137 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if 138 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 139 * signal becomes pending, ERESTART is returned if the current system 140 * call should be restarted if possible, and EINTR is returned if the system 141 * call should be interrupted by the signal (return EINTR). 142 * 143 * The lock argument is unlocked before the caller is suspended, and 144 * re-locked before _sleep() returns. If priority includes the PDROP 145 * flag the lock is not re-locked before returning. 146 */ 147 int 148 _sleep(void *ident, struct lock_object *lock, int priority, 149 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 150 { 151 struct thread *td; 152 struct proc *p; 153 struct lock_class *class; 154 int catch, lock_state, pri, rval, sleepq_flags; 155 WITNESS_SAVE_DECL(lock_witness); 156 157 td = curthread; 158 p = td->td_proc; 159 #ifdef KTRACE 160 if (KTRPOINT(td, KTR_CSW)) 161 ktrcsw(1, 0, wmesg); 162 #endif 163 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 164 "Sleeping on \"%s\"", wmesg); 165 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, 166 ("sleeping without a lock")); 167 KASSERT(p != NULL, ("msleep1")); 168 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 169 if (priority & PDROP) 170 KASSERT(lock != NULL && lock != &Giant.lock_object, 171 ("PDROP requires a non-Giant lock")); 172 if (lock != NULL) 173 class = LOCK_CLASS(lock); 174 else 175 class = NULL; 176 177 if (cold || SCHEDULER_STOPPED()) { 178 /* 179 * During autoconfiguration, just return; 180 * don't run any other threads or panic below, 181 * in case this is the idle thread and already asleep. 182 * XXX: this used to do "s = splhigh(); splx(safepri); 183 * splx(s);" to give interrupts a chance, but there is 184 * no way to give interrupts a chance now. 185 */ 186 if (lock != NULL && priority & PDROP) 187 class->lc_unlock(lock); 188 return (0); 189 } 190 catch = priority & PCATCH; 191 pri = priority & PRIMASK; 192 193 /* 194 * If we are already on a sleep queue, then remove us from that 195 * sleep queue first. We have to do this to handle recursive 196 * sleeps. 197 */ 198 if (TD_ON_SLEEPQ(td)) 199 sleepq_remove(td, td->td_wchan); 200 201 if ((uint8_t *)ident >= &pause_wchan[0] && 202 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) 203 sleepq_flags = SLEEPQ_PAUSE; 204 else 205 sleepq_flags = SLEEPQ_SLEEP; 206 if (catch) 207 sleepq_flags |= SLEEPQ_INTERRUPTIBLE; 208 209 sleepq_lock(ident); 210 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 211 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 212 213 if (lock == &Giant.lock_object) 214 mtx_assert(&Giant, MA_OWNED); 215 DROP_GIANT(); 216 if (lock != NULL && lock != &Giant.lock_object && 217 !(class->lc_flags & LC_SLEEPABLE)) { 218 WITNESS_SAVE(lock, lock_witness); 219 lock_state = class->lc_unlock(lock); 220 } else 221 /* GCC needs to follow the Yellow Brick Road */ 222 lock_state = -1; 223 224 /* 225 * We put ourselves on the sleep queue and start our timeout 226 * before calling thread_suspend_check, as we could stop there, 227 * and a wakeup or a SIGCONT (or both) could occur while we were 228 * stopped without resuming us. Thus, we must be ready for sleep 229 * when cursig() is called. If the wakeup happens while we're 230 * stopped, then td will no longer be on a sleep queue upon 231 * return from cursig(). 232 */ 233 sleepq_add(ident, lock, wmesg, sleepq_flags, 0); 234 if (sbt != 0) 235 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 236 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 237 sleepq_release(ident); 238 WITNESS_SAVE(lock, lock_witness); 239 lock_state = class->lc_unlock(lock); 240 sleepq_lock(ident); 241 } 242 if (sbt != 0 && catch) 243 rval = sleepq_timedwait_sig(ident, pri); 244 else if (sbt != 0) 245 rval = sleepq_timedwait(ident, pri); 246 else if (catch) 247 rval = sleepq_wait_sig(ident, pri); 248 else { 249 sleepq_wait(ident, pri); 250 rval = 0; 251 } 252 #ifdef KTRACE 253 if (KTRPOINT(td, KTR_CSW)) 254 ktrcsw(0, 0, wmesg); 255 #endif 256 PICKUP_GIANT(); 257 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 258 class->lc_lock(lock, lock_state); 259 WITNESS_RESTORE(lock, lock_witness); 260 } 261 return (rval); 262 } 263 264 int 265 msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, 266 sbintime_t sbt, sbintime_t pr, int flags) 267 { 268 struct thread *td; 269 struct proc *p; 270 int rval; 271 WITNESS_SAVE_DECL(mtx); 272 273 td = curthread; 274 p = td->td_proc; 275 KASSERT(mtx != NULL, ("sleeping without a mutex")); 276 KASSERT(p != NULL, ("msleep1")); 277 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 278 279 if (cold || SCHEDULER_STOPPED()) { 280 /* 281 * During autoconfiguration, just return; 282 * don't run any other threads or panic below, 283 * in case this is the idle thread and already asleep. 284 * XXX: this used to do "s = splhigh(); splx(safepri); 285 * splx(s);" to give interrupts a chance, but there is 286 * no way to give interrupts a chance now. 287 */ 288 return (0); 289 } 290 291 sleepq_lock(ident); 292 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 293 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 294 295 DROP_GIANT(); 296 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 297 WITNESS_SAVE(&mtx->lock_object, mtx); 298 mtx_unlock_spin(mtx); 299 300 /* 301 * We put ourselves on the sleep queue and start our timeout. 302 */ 303 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 304 if (sbt != 0) 305 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 306 307 /* 308 * Can't call ktrace with any spin locks held so it can lock the 309 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 310 * any spin lock. Thus, we have to drop the sleepq spin lock while 311 * we handle those requests. This is safe since we have placed our 312 * thread on the sleep queue already. 313 */ 314 #ifdef KTRACE 315 if (KTRPOINT(td, KTR_CSW)) { 316 sleepq_release(ident); 317 ktrcsw(1, 0, wmesg); 318 sleepq_lock(ident); 319 } 320 #endif 321 #ifdef WITNESS 322 sleepq_release(ident); 323 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 324 wmesg); 325 sleepq_lock(ident); 326 #endif 327 if (sbt != 0) 328 rval = sleepq_timedwait(ident, 0); 329 else { 330 sleepq_wait(ident, 0); 331 rval = 0; 332 } 333 #ifdef KTRACE 334 if (KTRPOINT(td, KTR_CSW)) 335 ktrcsw(0, 0, wmesg); 336 #endif 337 PICKUP_GIANT(); 338 mtx_lock_spin(mtx); 339 WITNESS_RESTORE(&mtx->lock_object, mtx); 340 return (rval); 341 } 342 343 /* 344 * pause() delays the calling thread by the given number of system ticks. 345 * During cold bootup, pause() uses the DELAY() function instead of 346 * the tsleep() function to do the waiting. The "timo" argument must be 347 * greater than or equal to zero. A "timo" value of zero is equivalent 348 * to a "timo" value of one. 349 */ 350 int 351 pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 352 { 353 int sbt_sec; 354 355 sbt_sec = sbintime_getsec(sbt); 356 KASSERT(sbt_sec >= 0, ("pause: timo must be >= 0")); 357 358 /* silently convert invalid timeouts */ 359 if (sbt == 0) 360 sbt = tick_sbt; 361 362 if (cold) { 363 /* 364 * We delay one second at a time to avoid overflowing the 365 * system specific DELAY() function(s): 366 */ 367 while (sbt_sec > 0) { 368 DELAY(1000000); 369 sbt_sec--; 370 } 371 DELAY((sbt & 0xffffffff) / SBT_1US); 372 return (0); 373 } 374 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); 375 } 376 377 /* 378 * Make all threads sleeping on the specified identifier runnable. 379 */ 380 void 381 wakeup(void *ident) 382 { 383 int wakeup_swapper; 384 385 sleepq_lock(ident); 386 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 387 sleepq_release(ident); 388 if (wakeup_swapper) { 389 KASSERT(ident != &proc0, 390 ("wakeup and wakeup_swapper and proc0")); 391 kick_proc0(); 392 } 393 } 394 395 /* 396 * Make a thread sleeping on the specified identifier runnable. 397 * May wake more than one thread if a target thread is currently 398 * swapped out. 399 */ 400 void 401 wakeup_one(void *ident) 402 { 403 int wakeup_swapper; 404 405 sleepq_lock(ident); 406 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 407 sleepq_release(ident); 408 if (wakeup_swapper) 409 kick_proc0(); 410 } 411 412 static void 413 kdb_switch(void) 414 { 415 thread_unlock(curthread); 416 kdb_backtrace(); 417 kdb_reenter(); 418 panic("%s: did not reenter debugger", __func__); 419 } 420 421 /* 422 * The machine independent parts of context switching. 423 */ 424 void 425 mi_switch(int flags, struct thread *newtd) 426 { 427 uint64_t runtime, new_switchtime; 428 struct thread *td; 429 struct proc *p; 430 431 td = curthread; /* XXX */ 432 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 433 p = td->td_proc; /* XXX */ 434 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 435 #ifdef INVARIANTS 436 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 437 mtx_assert(&Giant, MA_NOTOWNED); 438 #endif 439 KASSERT(td->td_critnest == 1 || panicstr, 440 ("mi_switch: switch in a critical section")); 441 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 442 ("mi_switch: switch must be voluntary or involuntary")); 443 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 444 445 /* 446 * Don't perform context switches from the debugger. 447 */ 448 if (kdb_active) 449 kdb_switch(); 450 if (SCHEDULER_STOPPED()) 451 return; 452 if (flags & SW_VOL) { 453 td->td_ru.ru_nvcsw++; 454 td->td_swvoltick = ticks; 455 } else 456 td->td_ru.ru_nivcsw++; 457 #ifdef SCHED_STATS 458 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 459 #endif 460 /* 461 * Compute the amount of time during which the current 462 * thread was running, and add that to its total so far. 463 */ 464 new_switchtime = cpu_ticks(); 465 runtime = new_switchtime - PCPU_GET(switchtime); 466 td->td_runtime += runtime; 467 td->td_incruntime += runtime; 468 PCPU_SET(switchtime, new_switchtime); 469 td->td_generation++; /* bump preempt-detect counter */ 470 PCPU_INC(cnt.v_swtch); 471 PCPU_SET(switchticks, ticks); 472 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 473 td->td_tid, td->td_sched, p->p_pid, td->td_name); 474 #if (KTR_COMPILE & KTR_SCHED) != 0 475 if (TD_IS_IDLETHREAD(td)) 476 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", 477 "prio:%d", td->td_priority); 478 else 479 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), 480 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, 481 "lockname:\"%s\"", td->td_lockname); 482 #endif 483 SDT_PROBE0(sched, , , preempt); 484 #ifdef XEN 485 PT_UPDATES_FLUSH(); 486 #endif 487 sched_switch(td, newtd, flags); 488 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", 489 "prio:%d", td->td_priority); 490 491 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 492 td->td_tid, td->td_sched, p->p_pid, td->td_name); 493 494 /* 495 * If the last thread was exiting, finish cleaning it up. 496 */ 497 if ((td = PCPU_GET(deadthread))) { 498 PCPU_SET(deadthread, NULL); 499 thread_stash(td); 500 } 501 } 502 503 /* 504 * Change thread state to be runnable, placing it on the run queue if 505 * it is in memory. If it is swapped out, return true so our caller 506 * will know to awaken the swapper. 507 */ 508 int 509 setrunnable(struct thread *td) 510 { 511 512 THREAD_LOCK_ASSERT(td, MA_OWNED); 513 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 514 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 515 switch (td->td_state) { 516 case TDS_RUNNING: 517 case TDS_RUNQ: 518 return (0); 519 case TDS_INHIBITED: 520 /* 521 * If we are only inhibited because we are swapped out 522 * then arange to swap in this process. Otherwise just return. 523 */ 524 if (td->td_inhibitors != TDI_SWAPPED) 525 return (0); 526 /* FALLTHROUGH */ 527 case TDS_CAN_RUN: 528 break; 529 default: 530 printf("state is 0x%x", td->td_state); 531 panic("setrunnable(2)"); 532 } 533 if ((td->td_flags & TDF_INMEM) == 0) { 534 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 535 td->td_flags |= TDF_SWAPINREQ; 536 return (1); 537 } 538 } else 539 sched_wakeup(td); 540 return (0); 541 } 542 543 /* 544 * Compute a tenex style load average of a quantity on 545 * 1, 5 and 15 minute intervals. 546 */ 547 static void 548 loadav(void *arg) 549 { 550 int i, nrun; 551 struct loadavg *avg; 552 553 nrun = sched_load(); 554 avg = &averunnable; 555 556 for (i = 0; i < 3; i++) 557 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 558 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 559 560 /* 561 * Schedule the next update to occur after 5 seconds, but add a 562 * random variation to avoid synchronisation with processes that 563 * run at regular intervals. 564 */ 565 callout_reset_sbt(&loadav_callout, 566 tick_sbt * (hz * 4 + (int)(random() % (hz * 2 + 1))), 0, 567 loadav, NULL, C_DIRECT_EXEC | C_HARDCLOCK); 568 } 569 570 /* ARGSUSED */ 571 static void 572 synch_setup(void *dummy) 573 { 574 callout_init(&loadav_callout, CALLOUT_MPSAFE); 575 576 /* Kick off timeout driven events by calling first time. */ 577 loadav(NULL); 578 } 579 580 int 581 should_yield(void) 582 { 583 584 return (ticks - curthread->td_swvoltick >= hogticks); 585 } 586 587 void 588 maybe_yield(void) 589 { 590 591 if (should_yield()) 592 kern_yield(PRI_USER); 593 } 594 595 void 596 kern_yield(int prio) 597 { 598 struct thread *td; 599 600 td = curthread; 601 DROP_GIANT(); 602 thread_lock(td); 603 if (prio == PRI_USER) 604 prio = td->td_user_pri; 605 if (prio >= 0) 606 sched_prio(td, prio); 607 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 608 thread_unlock(td); 609 PICKUP_GIANT(); 610 } 611 612 /* 613 * General purpose yield system call. 614 */ 615 int 616 sys_yield(struct thread *td, struct yield_args *uap) 617 { 618 619 thread_lock(td); 620 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 621 sched_prio(td, PRI_MAX_TIMESHARE); 622 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 623 thread_unlock(td); 624 td->td_retval[0] = 0; 625 return (0); 626 } 627