kern_synch.c revision 281702
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: head/sys/kern/kern_synch.c 281702 2015-04-18 20:36:58Z markj $"); 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 82static void synch_setup(void *dummy); 83SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, 84 NULL); 85 86int hogticks; 87static uint8_t pause_wchan[MAXCPU]; 88 89static struct callout loadav_callout; 90 91struct 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 */ 97static 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 */ 104SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, FSCALE, ""); 105 106static void loadav(void *arg); 107 108SDT_PROVIDER_DECLARE(sched); 109SDT_PROBE_DEFINE(sched, , , preempt); 110 111static void 112sleepinit(void *unused) 113{ 114 115 hogticks = (hz / 10) * 2; /* Default only. */ 116 init_sleepqueues(); 117} 118 119/* 120 * vmem tries to lock the sleepq mutexes when free'ing kva, so make sure 121 * it is available. 122 */ 123SYSINIT(sleepinit, SI_SUB_KMEM, SI_ORDER_ANY, sleepinit, 0); 124 125/* 126 * General sleep call. Suspends the current thread until a wakeup is 127 * performed on the specified identifier. The thread will then be made 128 * runnable with the specified priority. Sleeps at most sbt units of time 129 * (0 means no timeout). If pri includes the PCATCH flag, let signals 130 * interrupt the sleep, otherwise ignore them while sleeping. Returns 0 if 131 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 132 * signal becomes pending, ERESTART is returned if the current system 133 * call should be restarted if possible, and EINTR is returned if the system 134 * call should be interrupted by the signal (return EINTR). 135 * 136 * The lock argument is unlocked before the caller is suspended, and 137 * re-locked before _sleep() returns. If priority includes the PDROP 138 * flag the lock is not re-locked before returning. 139 */ 140int 141_sleep(void *ident, struct lock_object *lock, int priority, 142 const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 143{ 144 struct thread *td; 145 struct proc *p; 146 struct lock_class *class; 147 uintptr_t lock_state; 148 int catch, pri, rval, sleepq_flags; 149 WITNESS_SAVE_DECL(lock_witness); 150 151 td = curthread; 152 p = td->td_proc; 153#ifdef KTRACE 154 if (KTRPOINT(td, KTR_CSW)) 155 ktrcsw(1, 0, wmesg); 156#endif 157 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock, 158 "Sleeping on \"%s\"", wmesg); 159 KASSERT(sbt != 0 || mtx_owned(&Giant) || lock != NULL, 160 ("sleeping without a lock")); 161 KASSERT(p != NULL, ("msleep1")); 162 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 163 if (priority & PDROP) 164 KASSERT(lock != NULL && lock != &Giant.lock_object, 165 ("PDROP requires a non-Giant lock")); 166 if (lock != NULL) 167 class = LOCK_CLASS(lock); 168 else 169 class = NULL; 170 171 if (cold || SCHEDULER_STOPPED()) { 172 /* 173 * During autoconfiguration, just return; 174 * don't run any other threads or panic below, 175 * in case this is the idle thread and already asleep. 176 * XXX: this used to do "s = splhigh(); splx(safepri); 177 * splx(s);" to give interrupts a chance, but there is 178 * no way to give interrupts a chance now. 179 */ 180 if (lock != NULL && priority & PDROP) 181 class->lc_unlock(lock); 182 return (0); 183 } 184 catch = priority & PCATCH; 185 pri = priority & PRIMASK; 186 187 /* 188 * If we are already on a sleep queue, then remove us from that 189 * sleep queue first. We have to do this to handle recursive 190 * sleeps. 191 */ 192 if (TD_ON_SLEEPQ(td)) 193 sleepq_remove(td, td->td_wchan); 194 195 if ((uint8_t *)ident >= &pause_wchan[0] && 196 (uint8_t *)ident <= &pause_wchan[MAXCPU - 1]) 197 sleepq_flags = SLEEPQ_PAUSE; 198 else 199 sleepq_flags = SLEEPQ_SLEEP; 200 if (catch) 201 sleepq_flags |= SLEEPQ_INTERRUPTIBLE; 202 203 sleepq_lock(ident); 204 CTR5(KTR_PROC, "sleep: thread %ld (pid %ld, %s) on %s (%p)", 205 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 206 207 if (lock == &Giant.lock_object) 208 mtx_assert(&Giant, MA_OWNED); 209 DROP_GIANT(); 210 if (lock != NULL && lock != &Giant.lock_object && 211 !(class->lc_flags & LC_SLEEPABLE)) { 212 WITNESS_SAVE(lock, lock_witness); 213 lock_state = class->lc_unlock(lock); 214 } else 215 /* GCC needs to follow the Yellow Brick Road */ 216 lock_state = -1; 217 218 /* 219 * We put ourselves on the sleep queue and start our timeout 220 * before calling thread_suspend_check, as we could stop there, 221 * and a wakeup or a SIGCONT (or both) could occur while we were 222 * stopped without resuming us. Thus, we must be ready for sleep 223 * when cursig() is called. If the wakeup happens while we're 224 * stopped, then td will no longer be on a sleep queue upon 225 * return from cursig(). 226 */ 227 sleepq_add(ident, lock, wmesg, sleepq_flags, 0); 228 if (sbt != 0) 229 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 230 if (lock != NULL && class->lc_flags & LC_SLEEPABLE) { 231 sleepq_release(ident); 232 WITNESS_SAVE(lock, lock_witness); 233 lock_state = class->lc_unlock(lock); 234 sleepq_lock(ident); 235 } 236 if (sbt != 0 && catch) 237 rval = sleepq_timedwait_sig(ident, pri); 238 else if (sbt != 0) 239 rval = sleepq_timedwait(ident, pri); 240 else if (catch) 241 rval = sleepq_wait_sig(ident, pri); 242 else { 243 sleepq_wait(ident, pri); 244 rval = 0; 245 } 246#ifdef KTRACE 247 if (KTRPOINT(td, KTR_CSW)) 248 ktrcsw(0, 0, wmesg); 249#endif 250 PICKUP_GIANT(); 251 if (lock != NULL && lock != &Giant.lock_object && !(priority & PDROP)) { 252 class->lc_lock(lock, lock_state); 253 WITNESS_RESTORE(lock, lock_witness); 254 } 255 return (rval); 256} 257 258int 259msleep_spin_sbt(void *ident, struct mtx *mtx, const char *wmesg, 260 sbintime_t sbt, sbintime_t pr, int flags) 261{ 262 struct thread *td; 263 struct proc *p; 264 int rval; 265 WITNESS_SAVE_DECL(mtx); 266 267 td = curthread; 268 p = td->td_proc; 269 KASSERT(mtx != NULL, ("sleeping without a mutex")); 270 KASSERT(p != NULL, ("msleep1")); 271 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 272 273 if (cold || SCHEDULER_STOPPED()) { 274 /* 275 * During autoconfiguration, just return; 276 * don't run any other threads or panic below, 277 * in case this is the idle thread and already asleep. 278 * XXX: this used to do "s = splhigh(); splx(safepri); 279 * splx(s);" to give interrupts a chance, but there is 280 * no way to give interrupts a chance now. 281 */ 282 return (0); 283 } 284 285 sleepq_lock(ident); 286 CTR5(KTR_PROC, "msleep_spin: thread %ld (pid %ld, %s) on %s (%p)", 287 td->td_tid, p->p_pid, td->td_name, wmesg, ident); 288 289 DROP_GIANT(); 290 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 291 WITNESS_SAVE(&mtx->lock_object, mtx); 292 mtx_unlock_spin(mtx); 293 294 /* 295 * We put ourselves on the sleep queue and start our timeout. 296 */ 297 sleepq_add(ident, &mtx->lock_object, wmesg, SLEEPQ_SLEEP, 0); 298 if (sbt != 0) 299 sleepq_set_timeout_sbt(ident, sbt, pr, flags); 300 301 /* 302 * Can't call ktrace with any spin locks held so it can lock the 303 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 304 * any spin lock. Thus, we have to drop the sleepq spin lock while 305 * we handle those requests. This is safe since we have placed our 306 * thread on the sleep queue already. 307 */ 308#ifdef KTRACE 309 if (KTRPOINT(td, KTR_CSW)) { 310 sleepq_release(ident); 311 ktrcsw(1, 0, wmesg); 312 sleepq_lock(ident); 313 } 314#endif 315#ifdef WITNESS 316 sleepq_release(ident); 317 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 318 wmesg); 319 sleepq_lock(ident); 320#endif 321 if (sbt != 0) 322 rval = sleepq_timedwait(ident, 0); 323 else { 324 sleepq_wait(ident, 0); 325 rval = 0; 326 } 327#ifdef KTRACE 328 if (KTRPOINT(td, KTR_CSW)) 329 ktrcsw(0, 0, wmesg); 330#endif 331 PICKUP_GIANT(); 332 mtx_lock_spin(mtx); 333 WITNESS_RESTORE(&mtx->lock_object, mtx); 334 return (rval); 335} 336 337/* 338 * pause() delays the calling thread by the given number of system ticks. 339 * During cold bootup, pause() uses the DELAY() function instead of 340 * the tsleep() function to do the waiting. The "timo" argument must be 341 * greater than or equal to zero. A "timo" value of zero is equivalent 342 * to a "timo" value of one. 343 */ 344int 345pause_sbt(const char *wmesg, sbintime_t sbt, sbintime_t pr, int flags) 346{ 347 KASSERT(sbt >= 0, ("pause: timeout must be >= 0")); 348 349 /* silently convert invalid timeouts */ 350 if (sbt == 0) 351 sbt = tick_sbt; 352 353 if (cold || kdb_active) { 354 /* 355 * We delay one second at a time to avoid overflowing the 356 * system specific DELAY() function(s): 357 */ 358 while (sbt >= SBT_1S) { 359 DELAY(1000000); 360 sbt -= SBT_1S; 361 } 362 /* Do the delay remainder, if any */ 363 sbt = (sbt + SBT_1US - 1) / SBT_1US; 364 if (sbt > 0) 365 DELAY(sbt); 366 return (0); 367 } 368 return (_sleep(&pause_wchan[curcpu], NULL, 0, wmesg, sbt, pr, flags)); 369} 370 371/* 372 * Make all threads sleeping on the specified identifier runnable. 373 */ 374void 375wakeup(void *ident) 376{ 377 int wakeup_swapper; 378 379 sleepq_lock(ident); 380 wakeup_swapper = sleepq_broadcast(ident, SLEEPQ_SLEEP, 0, 0); 381 sleepq_release(ident); 382 if (wakeup_swapper) { 383 KASSERT(ident != &proc0, 384 ("wakeup and wakeup_swapper and proc0")); 385 kick_proc0(); 386 } 387} 388 389/* 390 * Make a thread sleeping on the specified identifier runnable. 391 * May wake more than one thread if a target thread is currently 392 * swapped out. 393 */ 394void 395wakeup_one(void *ident) 396{ 397 int wakeup_swapper; 398 399 sleepq_lock(ident); 400 wakeup_swapper = sleepq_signal(ident, SLEEPQ_SLEEP, 0, 0); 401 sleepq_release(ident); 402 if (wakeup_swapper) 403 kick_proc0(); 404} 405 406static void 407kdb_switch(void) 408{ 409 thread_unlock(curthread); 410 kdb_backtrace(); 411 kdb_reenter(); 412 panic("%s: did not reenter debugger", __func__); 413} 414 415/* 416 * The machine independent parts of context switching. 417 */ 418void 419mi_switch(int flags, struct thread *newtd) 420{ 421 uint64_t runtime, new_switchtime; 422 struct thread *td; 423 struct proc *p; 424 425 td = curthread; /* XXX */ 426 THREAD_LOCK_ASSERT(td, MA_OWNED | MA_NOTRECURSED); 427 p = td->td_proc; /* XXX */ 428 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 429#ifdef INVARIANTS 430 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 431 mtx_assert(&Giant, MA_NOTOWNED); 432#endif 433 KASSERT(td->td_critnest == 1 || panicstr, 434 ("mi_switch: switch in a critical section")); 435 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 436 ("mi_switch: switch must be voluntary or involuntary")); 437 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 438 439 /* 440 * Don't perform context switches from the debugger. 441 */ 442 if (kdb_active) 443 kdb_switch(); 444 if (SCHEDULER_STOPPED()) 445 return; 446 if (flags & SW_VOL) { 447 td->td_ru.ru_nvcsw++; 448 td->td_swvoltick = ticks; 449 } else 450 td->td_ru.ru_nivcsw++; 451#ifdef SCHED_STATS 452 SCHED_STAT_INC(sched_switch_stats[flags & SW_TYPE_MASK]); 453#endif 454 /* 455 * Compute the amount of time during which the current 456 * thread was running, and add that to its total so far. 457 */ 458 new_switchtime = cpu_ticks(); 459 runtime = new_switchtime - PCPU_GET(switchtime); 460 td->td_runtime += runtime; 461 td->td_incruntime += runtime; 462 PCPU_SET(switchtime, new_switchtime); 463 td->td_generation++; /* bump preempt-detect counter */ 464 PCPU_INC(cnt.v_swtch); 465 PCPU_SET(switchticks, ticks); 466 CTR4(KTR_PROC, "mi_switch: old thread %ld (td_sched %p, pid %ld, %s)", 467 td->td_tid, td->td_sched, p->p_pid, td->td_name); 468#if (KTR_COMPILE & KTR_SCHED) != 0 469 if (TD_IS_IDLETHREAD(td)) 470 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "idle", 471 "prio:%d", td->td_priority); 472 else 473 KTR_STATE3(KTR_SCHED, "thread", sched_tdname(td), KTDSTATE(td), 474 "prio:%d", td->td_priority, "wmesg:\"%s\"", td->td_wmesg, 475 "lockname:\"%s\"", td->td_lockname); 476#endif 477 SDT_PROBE0(sched, , , preempt); 478#ifdef XEN 479 PT_UPDATES_FLUSH(); 480#endif 481 sched_switch(td, newtd, flags); 482 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(td), "running", 483 "prio:%d", td->td_priority); 484 485 CTR4(KTR_PROC, "mi_switch: new thread %ld (td_sched %p, pid %ld, %s)", 486 td->td_tid, td->td_sched, p->p_pid, td->td_name); 487 488 /* 489 * If the last thread was exiting, finish cleaning it up. 490 */ 491 if ((td = PCPU_GET(deadthread))) { 492 PCPU_SET(deadthread, NULL); 493 thread_stash(td); 494 } 495} 496 497/* 498 * Change thread state to be runnable, placing it on the run queue if 499 * it is in memory. If it is swapped out, return true so our caller 500 * will know to awaken the swapper. 501 */ 502int 503setrunnable(struct thread *td) 504{ 505 506 THREAD_LOCK_ASSERT(td, MA_OWNED); 507 KASSERT(td->td_proc->p_state != PRS_ZOMBIE, 508 ("setrunnable: pid %d is a zombie", td->td_proc->p_pid)); 509 switch (td->td_state) { 510 case TDS_RUNNING: 511 case TDS_RUNQ: 512 return (0); 513 case TDS_INHIBITED: 514 /* 515 * If we are only inhibited because we are swapped out 516 * then arange to swap in this process. Otherwise just return. 517 */ 518 if (td->td_inhibitors != TDI_SWAPPED) 519 return (0); 520 /* FALLTHROUGH */ 521 case TDS_CAN_RUN: 522 break; 523 default: 524 printf("state is 0x%x", td->td_state); 525 panic("setrunnable(2)"); 526 } 527 if ((td->td_flags & TDF_INMEM) == 0) { 528 if ((td->td_flags & TDF_SWAPINREQ) == 0) { 529 td->td_flags |= TDF_SWAPINREQ; 530 return (1); 531 } 532 } else 533 sched_wakeup(td); 534 return (0); 535} 536 537/* 538 * Compute a tenex style load average of a quantity on 539 * 1, 5 and 15 minute intervals. 540 */ 541static void 542loadav(void *arg) 543{ 544 int i, nrun; 545 struct loadavg *avg; 546 547 nrun = sched_load(); 548 avg = &averunnable; 549 550 for (i = 0; i < 3; i++) 551 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 552 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 553 554 /* 555 * Schedule the next update to occur after 5 seconds, but add a 556 * random variation to avoid synchronisation with processes that 557 * run at regular intervals. 558 */ 559 callout_reset_sbt(&loadav_callout, 560 SBT_1US * (4000000 + (int)(random() % 2000001)), SBT_1US, 561 loadav, NULL, C_DIRECT_EXEC | C_PREL(32)); 562} 563 564/* ARGSUSED */ 565static void 566synch_setup(void *dummy) 567{ 568 callout_init(&loadav_callout, CALLOUT_MPSAFE); 569 570 /* Kick off timeout driven events by calling first time. */ 571 loadav(NULL); 572} 573 574int 575should_yield(void) 576{ 577 578 return ((u_int)ticks - (u_int)curthread->td_swvoltick >= hogticks); 579} 580 581void 582maybe_yield(void) 583{ 584 585 if (should_yield()) 586 kern_yield(PRI_USER); 587} 588 589void 590kern_yield(int prio) 591{ 592 struct thread *td; 593 594 td = curthread; 595 DROP_GIANT(); 596 thread_lock(td); 597 if (prio == PRI_USER) 598 prio = td->td_user_pri; 599 if (prio >= 0) 600 sched_prio(td, prio); 601 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 602 thread_unlock(td); 603 PICKUP_GIANT(); 604} 605 606/* 607 * General purpose yield system call. 608 */ 609int 610sys_yield(struct thread *td, struct yield_args *uap) 611{ 612 613 thread_lock(td); 614 if (PRI_BASE(td->td_pri_class) == PRI_TIMESHARE) 615 sched_prio(td, PRI_MAX_TIMESHARE); 616 mi_switch(SW_VOL | SWT_RELINQUISH, NULL); 617 thread_unlock(td); 618 td->td_retval[0] = 0; 619 return (0); 620} 621