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_ddb.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/condvar.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 DDB 61 #include <ddb/ddb.h> 62 #endif 63 #ifdef KTRACE 64 #include <sys/uio.h> 65 #include <sys/ktrace.h> 66 #endif 67 68 #include <machine/cpu.h> 69 70 static void synch_setup(void *dummy); 71 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL) 72 73 int hogticks; 74 int lbolt; 75 76 static struct callout loadav_callout; 77 static struct callout lbolt_callout; 78 79 struct loadavg averunnable = 80 { {0, 0, 0}, FSCALE }; /* load average, of runnable procs */ 81 /* 82 * Constants for averages over 1, 5, and 15 minutes 83 * when sampling at 5 second intervals. 84 */ 85 static fixpt_t cexp[3] = { 86 0.9200444146293232 * FSCALE, /* exp(-1/12) */ 87 0.9834714538216174 * FSCALE, /* exp(-1/60) */ 88 0.9944598480048967 * FSCALE, /* exp(-1/180) */ 89 }; 90 91 /* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 92 static int fscale __unused = FSCALE; 93 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 94 95 static void loadav(void *arg); 96 static void lboltcb(void *arg); 97 98 void 99 sleepinit(void) 100 { 101 102 hogticks = (hz / 10) * 2; /* Default only. */ 103 init_sleepqueues(); 104 } 105 106 /* 107 * General sleep call. Suspends the current process until a wakeup is 108 * performed on the specified identifier. The process will then be made 109 * runnable with the specified priority. Sleeps at most timo/hz seconds 110 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 111 * before and after sleeping, else signals are not checked. Returns 0 if 112 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 113 * signal needs to be delivered, ERESTART is returned if the current system 114 * call should be restarted if possible, and EINTR is returned if the system 115 * call should be interrupted by the signal (return EINTR). 116 * 117 * The mutex argument is exited before the caller is suspended, and 118 * entered before msleep returns. If priority includes the PDROP 119 * flag the mutex is not entered before returning. 120 */ 121 int 122 msleep(ident, mtx, priority, wmesg, timo) 123 void *ident; 124 struct mtx *mtx; 125 int priority, timo; 126 const char *wmesg; 127 { 128 struct sleepqueue *sq; 129 struct thread *td; 130 struct proc *p; 131 int catch, rval, sig; 132 WITNESS_SAVE_DECL(mtx); 133 134 td = curthread; 135 p = td->td_proc; 136 #ifdef KTRACE 137 if (KTRPOINT(td, KTR_CSW)) 138 ktrcsw(1, 0); 139 #endif 140 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL : 141 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg); 142 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL, 143 ("sleeping without a mutex")); 144 KASSERT(p != NULL, ("msleep1")); 145 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 146 147 if (cold) { 148 /* 149 * During autoconfiguration, just return; 150 * don't run any other threads or panic below, 151 * in case this is the idle thread and already asleep. 152 * XXX: this used to do "s = splhigh(); splx(safepri); 153 * splx(s);" to give interrupts a chance, but there is 154 * no way to give interrupts a chance now. 155 */ 156 if (mtx != NULL && priority & PDROP) 157 mtx_unlock(mtx); 158 return (0); 159 } 160 catch = priority & PCATCH; 161 rval = 0; 162 163 /* 164 * If we are already on a sleep queue, then remove us from that 165 * sleep queue first. We have to do this to handle recursive 166 * sleeps. 167 */ 168 if (TD_ON_SLEEPQ(td)) 169 sleepq_remove(td, td->td_wchan); 170 171 sq = sleepq_lookup(ident); 172 mtx_lock_spin(&sched_lock); 173 174 /* 175 * If we are capable of async syscalls and there isn't already 176 * another one ready to return, start a new thread 177 * and queue it as ready to run. Note that there is danger here 178 * because we need to make sure that we don't sleep allocating 179 * the thread (recursion here might be bad). 180 */ 181 if (p->p_flag & P_SA || p->p_numthreads > 1) { 182 /* 183 * Just don't bother if we are exiting 184 * and not the exiting thread or thread was marked as 185 * interrupted. 186 */ 187 if (catch) { 188 if ((p->p_flag & P_WEXIT) && p->p_singlethread != td) { 189 mtx_unlock_spin(&sched_lock); 190 sleepq_release(ident); 191 return (EINTR); 192 } 193 if (td->td_flags & TDF_INTERRUPT) { 194 mtx_unlock_spin(&sched_lock); 195 sleepq_release(ident); 196 return (td->td_intrval); 197 } 198 } 199 } 200 mtx_unlock_spin(&sched_lock); 201 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)", 202 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 203 204 DROP_GIANT(); 205 if (mtx != NULL) { 206 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 207 WITNESS_SAVE(&mtx->mtx_object, mtx); 208 mtx_unlock(mtx); 209 } 210 211 /* 212 * We put ourselves on the sleep queue and start our timeout 213 * before calling thread_suspend_check, as we could stop there, 214 * and a wakeup or a SIGCONT (or both) could occur while we were 215 * stopped without resuming us. Thus, we must be ready for sleep 216 * when cursig() is called. If the wakeup happens while we're 217 * stopped, then td will no longer be on a sleep queue upon 218 * return from cursig(). 219 */ 220 sleepq_add(sq, ident, mtx, wmesg, 0); 221 if (timo) 222 sleepq_set_timeout(ident, timo); 223 if (catch) { 224 sig = sleepq_catch_signals(ident); 225 if (sig == 0 && !TD_ON_SLEEPQ(td)) { 226 mtx_lock_spin(&sched_lock); 227 td->td_flags &= ~TDF_SINTR; 228 mtx_unlock_spin(&sched_lock); 229 catch = 0; 230 } 231 } else 232 sig = 0; 233 234 /* 235 * Adjust this thread's priority. 236 * 237 * XXX: do we need to save priority in td_base_pri? 238 */ 239 mtx_lock_spin(&sched_lock); 240 sched_prio(td, priority & PRIMASK); 241 mtx_unlock_spin(&sched_lock); 242 243 if (timo && catch) 244 rval = sleepq_timedwait_sig(ident, sig != 0); 245 else if (timo) 246 rval = sleepq_timedwait(ident, sig != 0); 247 else if (catch) 248 rval = sleepq_wait_sig(ident); 249 else { 250 sleepq_wait(ident); 251 rval = 0; 252 } 253 if (rval == 0 && catch) 254 rval = sleepq_calc_signal_retval(sig); 255 #ifdef KTRACE 256 if (KTRPOINT(td, KTR_CSW)) 257 ktrcsw(0, 0); 258 #endif 259 PICKUP_GIANT(); 260 if (mtx != NULL && !(priority & PDROP)) { 261 mtx_lock(mtx); 262 WITNESS_RESTORE(&mtx->mtx_object, mtx); 263 } 264 return (rval); 265 } 266 267 /* 268 * Make all threads sleeping on the specified identifier runnable. 269 */ 270 void 271 wakeup(ident) 272 register void *ident; 273 { 274 275 sleepq_broadcast(ident, 0, -1); 276 } 277 278 /* 279 * Make a thread sleeping on the specified identifier runnable. 280 * May wake more than one thread if a target thread is currently 281 * swapped out. 282 */ 283 void 284 wakeup_one(ident) 285 register void *ident; 286 { 287 288 sleepq_signal(ident, 0, -1); 289 } 290 291 /* 292 * The machine independent parts of context switching. 293 */ 294 void 295 mi_switch(int flags) 296 { 297 struct bintime new_switchtime; 298 struct thread *td; 299 struct proc *p; 300 301 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 302 td = curthread; /* XXX */ 303 p = td->td_proc; /* XXX */ 304 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 305 #ifdef INVARIANTS 306 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 307 mtx_assert(&Giant, MA_NOTOWNED); 308 #endif 309 KASSERT(td->td_critnest == 1, 310 ("mi_switch: switch in a critical section")); 311 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 312 ("mi_switch: switch must be voluntary or involuntary")); 313 314 if (flags & SW_VOL) 315 p->p_stats->p_ru.ru_nvcsw++; 316 else 317 p->p_stats->p_ru.ru_nivcsw++; 318 /* 319 * Compute the amount of time during which the current 320 * process was running, and add that to its total so far. 321 */ 322 binuptime(&new_switchtime); 323 bintime_add(&p->p_runtime, &new_switchtime); 324 bintime_sub(&p->p_runtime, PCPU_PTR(switchtime)); 325 326 td->td_generation++; /* bump preempt-detect counter */ 327 328 #ifdef DDB 329 /* 330 * Don't perform context switches from the debugger. 331 */ 332 if (db_active) { 333 mtx_unlock_spin(&sched_lock); 334 db_print_backtrace(); 335 db_error("Context switches not allowed in the debugger"); 336 } 337 #endif 338 339 /* 340 * Check if the process exceeds its cpu resource allocation. If 341 * over max, arrange to kill the process in ast(). 342 */ 343 if (p->p_cpulimit != RLIM_INFINITY && 344 p->p_runtime.sec > p->p_cpulimit) { 345 p->p_sflag |= PS_XCPU; 346 td->td_flags |= TDF_ASTPENDING; 347 } 348 349 /* 350 * Finish up stats for outgoing thread. 351 */ 352 cnt.v_swtch++; 353 PCPU_SET(switchtime, new_switchtime); 354 PCPU_SET(switchticks, ticks); 355 CTR3(KTR_PROC, "mi_switch: old thread %p (pid %ld, %s)", 356 (void *)td, (long)p->p_pid, p->p_comm); 357 if (td->td_proc->p_flag & P_SA) 358 thread_switchout(td); 359 sched_switch(td); 360 361 CTR3(KTR_PROC, "mi_switch: new thread %p (pid %ld, %s)", 362 (void *)td, (long)p->p_pid, p->p_comm); 363 364 /* 365 * If the last thread was exiting, finish cleaning it up. 366 */ 367 if ((td = PCPU_GET(deadthread))) { 368 PCPU_SET(deadthread, NULL); 369 thread_stash(td); 370 } 371 } 372 373 /* 374 * Change process state to be runnable, 375 * placing it on the run queue if it is in memory, 376 * and awakening the swapper if it isn't in memory. 377 */ 378 void 379 setrunnable(struct thread *td) 380 { 381 struct proc *p; 382 383 p = td->td_proc; 384 mtx_assert(&sched_lock, MA_OWNED); 385 switch (p->p_state) { 386 case PRS_ZOMBIE: 387 panic("setrunnable(1)"); 388 default: 389 break; 390 } 391 switch (td->td_state) { 392 case TDS_RUNNING: 393 case TDS_RUNQ: 394 return; 395 case TDS_INHIBITED: 396 /* 397 * If we are only inhibited because we are swapped out 398 * then arange to swap in this process. Otherwise just return. 399 */ 400 if (td->td_inhibitors != TDI_SWAPPED) 401 return; 402 /* XXX: intentional fall-through ? */ 403 case TDS_CAN_RUN: 404 break; 405 default: 406 printf("state is 0x%x", td->td_state); 407 panic("setrunnable(2)"); 408 } 409 if ((p->p_sflag & PS_INMEM) == 0) { 410 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 411 p->p_sflag |= PS_SWAPINREQ; 412 wakeup(&proc0); 413 } 414 } else 415 sched_wakeup(td); 416 } 417 418 /* 419 * Compute a tenex style load average of a quantity on 420 * 1, 5 and 15 minute intervals. 421 * XXXKSE Needs complete rewrite when correct info is available. 422 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 423 */ 424 static void 425 loadav(void *arg) 426 { 427 int i, nrun; 428 struct loadavg *avg; 429 430 nrun = sched_load(); 431 avg = &averunnable; 432 433 for (i = 0; i < 3; i++) 434 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 435 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 436 437 /* 438 * Schedule the next update to occur after 5 seconds, but add a 439 * random variation to avoid synchronisation with processes that 440 * run at regular intervals. 441 */ 442 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 443 loadav, NULL); 444 } 445 446 static void 447 lboltcb(void *arg) 448 { 449 wakeup(&lbolt); 450 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 451 } 452 453 /* ARGSUSED */ 454 static void 455 synch_setup(dummy) 456 void *dummy; 457 { 458 callout_init(&loadav_callout, CALLOUT_MPSAFE); 459 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 460 461 /* Kick off timeout driven events by calling first time. */ 462 loadav(NULL); 463 lboltcb(NULL); 464 } 465 466 /* 467 * General purpose yield system call 468 */ 469 int 470 yield(struct thread *td, struct yield_args *uap) 471 { 472 struct ksegrp *kg; 473 474 kg = td->td_ksegrp; 475 mtx_assert(&Giant, MA_NOTOWNED); 476 mtx_lock_spin(&sched_lock); 477 sched_prio(td, PRI_MAX_TIMESHARE); 478 mi_switch(SW_VOL); 479 mtx_unlock_spin(&sched_lock); 480 td->td_retval[0] = 0; 481 return (0); 482 } 483