kern_synch.c revision 155932
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 155932 2006-02-23 00:13:58Z davidxu $"); 39 40#include "opt_ktrace.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/condvar.h> 45#include <sys/kdb.h> 46#include <sys/kernel.h> 47#include <sys/ktr.h> 48#include <sys/lock.h> 49#include <sys/mutex.h> 50#include <sys/proc.h> 51#include <sys/resourcevar.h> 52#include <sys/sched.h> 53#include <sys/signalvar.h> 54#include <sys/sleepqueue.h> 55#include <sys/smp.h> 56#include <sys/sx.h> 57#include <sys/sysctl.h> 58#include <sys/sysproto.h> 59#include <sys/vmmeter.h> 60#ifdef KTRACE 61#include <sys/uio.h> 62#include <sys/ktrace.h> 63#endif 64 65#include <machine/cpu.h> 66 67static void synch_setup(void *dummy); 68SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL) 69 70int hogticks; 71int lbolt; 72 73static struct callout loadav_callout; 74static struct callout lbolt_callout; 75 76struct loadavg averunnable = 77 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 78/* 79 * Constants for averages over 1, 5, and 15 minutes 80 * when sampling at 5 second intervals. 81 */ 82static fixpt_t cexp[3] = { 83 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 84 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 85 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 86}; 87 88/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 89static int fscale __unused = FSCALE; 90SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 91 92static void loadav(void *arg); 93static void lboltcb(void *arg); 94 95void 96sleepinit(void) 97{ 98 99 hogticks = (hz / 10) * 2; /* Default only. */ 100 init_sleepqueues(); 101} 102 103/* 104 * General sleep call. Suspends the current thread until a wakeup is 105 * performed on the specified identifier. The thread will then be made 106 * runnable with the specified priority. Sleeps at most timo/hz seconds 107 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 108 * before and after sleeping, else signals are not checked. Returns 0 if 109 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 110 * signal needs to be delivered, ERESTART is returned if the current system 111 * call should be restarted if possible, and EINTR is returned if the system 112 * call should be interrupted by the signal (return EINTR). 113 * 114 * The mutex argument is unlocked before the caller is suspended, and 115 * re-locked before msleep returns. If priority includes the PDROP 116 * flag the mutex is not re-locked before returning. 117 */ 118int 119msleep(ident, mtx, priority, wmesg, timo) 120 void *ident; 121 struct mtx *mtx; 122 int priority, timo; 123 const char *wmesg; 124{ 125 struct thread *td; 126 struct proc *p; 127 int catch, rval, flags; 128 WITNESS_SAVE_DECL(mtx); 129 130 td = curthread; 131 p = td->td_proc; 132#ifdef KTRACE 133 if (KTRPOINT(td, KTR_CSW)) 134 ktrcsw(1, 0); 135#endif 136 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL : 137 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg); 138 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL, 139 ("sleeping without a mutex")); 140 KASSERT(p != NULL, ("msleep1")); 141 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 142 143 if (cold) { 144 /* 145 * During autoconfiguration, just return; 146 * don't run any other threads or panic below, 147 * in case this is the idle thread and already asleep. 148 * XXX: this used to do "s = splhigh(); splx(safepri); 149 * splx(s);" to give interrupts a chance, but there is 150 * no way to give interrupts a chance now. 151 */ 152 if (mtx != NULL && priority & PDROP) 153 mtx_unlock(mtx); 154 return (0); 155 } 156 catch = priority & PCATCH; 157 rval = 0; 158 159 /* 160 * If we are already on a sleep queue, then remove us from that 161 * sleep queue first. We have to do this to handle recursive 162 * sleeps. 163 */ 164 if (TD_ON_SLEEPQ(td)) 165 sleepq_remove(td, td->td_wchan); 166 167 flags = SLEEPQ_MSLEEP; 168 if (catch) 169 flags |= SLEEPQ_INTERRUPTIBLE; 170 171 sleepq_lock(ident); 172 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)", 173 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 174 175 DROP_GIANT(); 176 if (mtx != NULL) { 177 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 178 WITNESS_SAVE(&mtx->mtx_object, mtx); 179 mtx_unlock(mtx); 180 } 181 182 /* 183 * We put ourselves on the sleep queue and start our timeout 184 * before calling thread_suspend_check, as we could stop there, 185 * and a wakeup or a SIGCONT (or both) could occur while we were 186 * stopped without resuming us. Thus, we must be ready for sleep 187 * when cursig() is called. If the wakeup happens while we're 188 * stopped, then td will no longer be on a sleep queue upon 189 * return from cursig(). 190 */ 191 sleepq_add(ident, mtx, wmesg, flags); 192 if (timo) 193 sleepq_set_timeout(ident, timo); 194 195 /* 196 * Adjust this thread's priority. 197 */ 198 mtx_lock_spin(&sched_lock); 199 sched_prio(td, priority & PRIMASK); 200 mtx_unlock_spin(&sched_lock); 201 202 if (timo && catch) 203 rval = sleepq_timedwait_sig(ident); 204 else if (timo) 205 rval = sleepq_timedwait(ident); 206 else if (catch) 207 rval = sleepq_wait_sig(ident); 208 else { 209 sleepq_wait(ident); 210 rval = 0; 211 } 212#ifdef KTRACE 213 if (KTRPOINT(td, KTR_CSW)) 214 ktrcsw(0, 0); 215#endif 216 PICKUP_GIANT(); 217 if (mtx != NULL && !(priority & PDROP)) { 218 mtx_lock(mtx); 219 WITNESS_RESTORE(&mtx->mtx_object, mtx); 220 } 221 return (rval); 222} 223 224int 225msleep_spin(ident, mtx, wmesg, timo) 226 void *ident; 227 struct mtx *mtx; 228 const char *wmesg; 229 int timo; 230{ 231 struct thread *td; 232 struct proc *p; 233 int rval; 234 WITNESS_SAVE_DECL(mtx); 235 236 td = curthread; 237 p = td->td_proc; 238 KASSERT(mtx != NULL, ("sleeping without a mutex")); 239 KASSERT(p != NULL, ("msleep1")); 240 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 241 242 if (cold) { 243 /* 244 * During autoconfiguration, just return; 245 * don't run any other threads or panic below, 246 * in case this is the idle thread and already asleep. 247 * XXX: this used to do "s = splhigh(); splx(safepri); 248 * splx(s);" to give interrupts a chance, but there is 249 * no way to give interrupts a chance now. 250 */ 251 return (0); 252 } 253 254 sleepq_lock(ident); 255 CTR5(KTR_PROC, "msleep_spin: thread %p (pid %ld, %s) on %s (%p)", 256 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 257 258 DROP_GIANT(); 259 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 260 WITNESS_SAVE(&mtx->mtx_object, mtx); 261 mtx_unlock_spin(mtx); 262 263 /* 264 * We put ourselves on the sleep queue and start our timeout. 265 */ 266 sleepq_add(ident, mtx, wmesg, SLEEPQ_MSLEEP); 267 if (timo) 268 sleepq_set_timeout(ident, timo); 269 270 /* 271 * Can't call ktrace with any spin locks held so it can lock the 272 * ktrace_mtx lock, and WITNESS_WARN considers it an error to hold 273 * any spin lock. Thus, we have to drop the sleepq spin lock while 274 * we handle those requests. This is safe since we have placed our 275 * thread on the sleep queue already. 276 */ 277#ifdef KTRACE 278 if (KTRPOINT(td, KTR_CSW)) { 279 sleepq_release(ident); 280 ktrcsw(1, 0); 281 sleepq_lock(ident); 282 } 283#endif 284#ifdef WITNESS 285 sleepq_release(ident); 286 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "Sleeping on \"%s\"", 287 wmesg); 288 sleepq_lock(ident); 289#endif 290 if (timo) 291 rval = sleepq_timedwait(ident); 292 else { 293 sleepq_wait(ident); 294 rval = 0; 295 } 296#ifdef KTRACE 297 if (KTRPOINT(td, KTR_CSW)) 298 ktrcsw(0, 0); 299#endif 300 PICKUP_GIANT(); 301 mtx_lock_spin(mtx); 302 WITNESS_RESTORE(&mtx->mtx_object, mtx); 303 return (rval); 304} 305 306/* 307 * Make all threads sleeping on the specified identifier runnable. 308 */ 309void 310wakeup(ident) 311 register void *ident; 312{ 313 314 sleepq_lock(ident); 315 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1); 316} 317 318/* 319 * Make a thread sleeping on the specified identifier runnable. 320 * May wake more than one thread if a target thread is currently 321 * swapped out. 322 */ 323void 324wakeup_one(ident) 325 register void *ident; 326{ 327 328 sleepq_lock(ident); 329 sleepq_signal(ident, SLEEPQ_MSLEEP, -1); 330} 331 332/* 333 * The machine independent parts of context switching. 334 */ 335void 336mi_switch(int flags, struct thread *newtd) 337{ 338 uint64_t new_switchtime; 339 struct thread *td; 340 struct proc *p; 341 342 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 343 td = curthread; /* XXX */ 344 p = td->td_proc; /* XXX */ 345 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 346#ifdef INVARIANTS 347 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 348 mtx_assert(&Giant, MA_NOTOWNED); 349#endif 350 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 351 (td->td_owepreempt) && (flags & SW_INVOL) != 0 && 352 newtd == NULL) || panicstr, 353 ("mi_switch: switch in a critical section")); 354 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 355 ("mi_switch: switch must be voluntary or involuntary")); 356 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 357 358 if (flags & SW_VOL) 359 p->p_stats->p_ru.ru_nvcsw++; 360 else 361 p->p_stats->p_ru.ru_nivcsw++; 362 363 /* 364 * Compute the amount of time during which the current 365 * process was running, and add that to its total so far. 366 */ 367 new_switchtime = cpu_ticks(); 368 p->p_rux.rux_runtime += (new_switchtime - PCPU_GET(switchtime)); 369 p->p_rux.rux_uticks += td->td_uticks; 370 td->td_uticks = 0; 371 p->p_rux.rux_iticks += td->td_iticks; 372 td->td_iticks = 0; 373 p->p_rux.rux_sticks += td->td_sticks; 374 td->td_sticks = 0; 375 376 td->td_generation++; /* bump preempt-detect counter */ 377 378 /* 379 * Don't perform context switches from the debugger. 380 */ 381 if (kdb_active) { 382 mtx_unlock_spin(&sched_lock); 383 kdb_backtrace(); 384 kdb_reenter(); 385 panic("%s: did not reenter debugger", __func__); 386 } 387 388 /* 389 * Check if the process exceeds its cpu resource allocation. If 390 * it reaches the max, arrange to kill the process in ast(). 391 */ 392 if (p->p_cpulimit != RLIM_INFINITY && 393 p->p_rux.rux_runtime >= p->p_cpulimit * cpu_tickrate()) { 394 p->p_sflag |= PS_XCPU; 395 td->td_flags |= TDF_ASTPENDING; 396 } 397 398 /* 399 * Finish up stats for outgoing thread. 400 */ 401 cnt.v_swtch++; 402 PCPU_SET(switchtime, new_switchtime); 403 PCPU_SET(switchticks, ticks); 404 CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)", 405 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 406 if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA)) 407 newtd = thread_switchout(td, flags, newtd); 408#if (KTR_COMPILE & KTR_SCHED) != 0 409 if (td == PCPU_GET(idlethread)) 410 CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle", 411 td, td->td_proc->p_comm, td->td_priority); 412 else if (newtd != NULL) 413 CTR5(KTR_SCHED, 414 "mi_switch: %p(%s) prio %d preempted by %p(%s)", 415 td, td->td_proc->p_comm, td->td_priority, newtd, 416 newtd->td_proc->p_comm); 417 else 418 CTR6(KTR_SCHED, 419 "mi_switch: %p(%s) prio %d inhibit %d wmesg %s lock %s", 420 td, td->td_proc->p_comm, td->td_priority, 421 td->td_inhibitors, td->td_wmesg, td->td_lockname); 422#endif 423 sched_switch(td, newtd, flags); 424 CTR3(KTR_SCHED, "mi_switch: running %p(%s) prio %d", 425 td, td->td_proc->p_comm, td->td_priority); 426 427 CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)", 428 (void *)td, td->td_sched, (long)p->p_pid, p->p_comm); 429 430 /* 431 * If the last thread was exiting, finish cleaning it up. 432 */ 433 if ((td = PCPU_GET(deadthread))) { 434 PCPU_SET(deadthread, NULL); 435 thread_stash(td); 436 } 437} 438 439/* 440 * Change process state to be runnable, 441 * placing it on the run queue if it is in memory, 442 * and awakening the swapper if it isn't in memory. 443 */ 444void 445setrunnable(struct thread *td) 446{ 447 struct proc *p; 448 449 p = td->td_proc; 450 mtx_assert(&sched_lock, MA_OWNED); 451 switch (p->p_state) { 452 case PRS_ZOMBIE: 453 panic("setrunnable(1)"); 454 default: 455 break; 456 } 457 switch (td->td_state) { 458 case TDS_RUNNING: 459 case TDS_RUNQ: 460 return; 461 case TDS_INHIBITED: 462 /* 463 * If we are only inhibited because we are swapped out 464 * then arange to swap in this process. Otherwise just return. 465 */ 466 if (td->td_inhibitors != TDI_SWAPPED) 467 return; 468 /* XXX: intentional fall-through ? */ 469 case TDS_CAN_RUN: 470 break; 471 default: 472 printf("state is 0x%x", td->td_state); 473 panic("setrunnable(2)"); 474 } 475 if ((p->p_sflag & PS_INMEM) == 0) { 476 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 477 p->p_sflag |= PS_SWAPINREQ; 478 /* 479 * due to a LOR between sched_lock and 480 * the sleepqueue chain locks, use 481 * lower level scheduling functions. 482 */ 483 kick_proc0(); 484 } 485 } else 486 sched_wakeup(td); 487} 488 489/* 490 * Compute a tenex style load average of a quantity on 491 * 1, 5 and 15 minute intervals. 492 * XXXKSE Needs complete rewrite when correct info is available. 493 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 494 */ 495static void 496loadav(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(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 514 loadav, NULL); 515} 516 517static void 518lboltcb(void *arg) 519{ 520 wakeup(&lbolt); 521 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 522} 523 524/* ARGSUSED */ 525static void 526synch_setup(dummy) 527 void *dummy; 528{ 529 callout_init(&loadav_callout, CALLOUT_MPSAFE); 530 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 531 532 /* Kick off timeout driven events by calling first time. */ 533 loadav(NULL); 534 lboltcb(NULL); 535} 536 537/* 538 * General purpose yield system call 539 */ 540int 541yield(struct thread *td, struct yield_args *uap) 542{ 543 struct ksegrp *kg; 544 545 kg = td->td_ksegrp; 546 mtx_assert(&Giant, MA_NOTOWNED); 547 mtx_lock_spin(&sched_lock); 548 sched_prio(td, PRI_MAX_TIMESHARE); 549 mi_switch(SW_VOL, NULL); 550 mtx_unlock_spin(&sched_lock); 551 td->td_retval[0] = 0; 552 return (0); 553} 554