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 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 67 static void synch_setup(void *dummy); 68 SYSINIT(synch_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, synch_setup, NULL) 69 70 int hogticks; 71 int lbolt; 72 73 static struct callout loadav_callout; 74 static struct callout lbolt_callout; 75 76 struct 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 */ 82 static 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 */ 89 static int fscale __unused = FSCALE; 90 SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 91 92 static void loadav(void *arg); 93 static void lboltcb(void *arg); 94 95 void 96 sleepinit(void) 97 { 98 99 hogticks = (hz / 10) * 2; /* Default only. */ 100 init_sleepqueues(); 101 } 102 103 /* 104 * General sleep call. Suspends the current process until a wakeup is 105 * performed on the specified identifier. The process 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 exited before the caller is suspended, and 115 * entered before msleep returns. If priority includes the PDROP 116 * flag the mutex is not entered before returning. 117 */ 118 int 119 msleep(ident, mtx, priority, wmesg, timo) 120 void *ident; 121 struct mtx *mtx; 122 int priority, timo; 123 const char *wmesg; 124 { 125 struct sleepqueue *sq; 126 struct thread *td; 127 struct proc *p; 128 int catch, rval, sig, flags; 129 WITNESS_SAVE_DECL(mtx); 130 131 td = curthread; 132 p = td->td_proc; 133 #ifdef KTRACE 134 if (KTRPOINT(td, KTR_CSW)) 135 ktrcsw(1, 0); 136 #endif 137 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, mtx == NULL ? NULL : 138 &mtx->mtx_object, "Sleeping on \"%s\"", wmesg); 139 KASSERT(timo != 0 || mtx_owned(&Giant) || mtx != NULL, 140 ("sleeping without a mutex")); 141 KASSERT(p != NULL, ("msleep1")); 142 KASSERT(ident != NULL && TD_IS_RUNNING(td), ("msleep")); 143 144 if (cold) { 145 /* 146 * During autoconfiguration, just return; 147 * don't run any other threads or panic below, 148 * in case this is the idle thread and already asleep. 149 * XXX: this used to do "s = splhigh(); splx(safepri); 150 * splx(s);" to give interrupts a chance, but there is 151 * no way to give interrupts a chance now. 152 */ 153 if (mtx != NULL && priority & PDROP) 154 mtx_unlock(mtx); 155 return (0); 156 } 157 catch = priority & PCATCH; 158 rval = 0; 159 160 /* 161 * If we are already on a sleep queue, then remove us from that 162 * sleep queue first. We have to do this to handle recursive 163 * sleeps. 164 */ 165 if (TD_ON_SLEEPQ(td)) 166 sleepq_remove(td, td->td_wchan); 167 168 sq = sleepq_lookup(ident); 169 if (catch) { 170 /* 171 * Don't bother sleeping if we are exiting and not the exiting 172 * thread or if our thread is marked as interrupted. 173 */ 174 mtx_lock_spin(&sched_lock); 175 rval = thread_sleep_check(td); 176 mtx_unlock_spin(&sched_lock); 177 if (rval != 0) { 178 sleepq_release(ident); 179 return (rval); 180 } 181 } 182 CTR5(KTR_PROC, "msleep: thread %p (pid %ld, %s) on %s (%p)", 183 (void *)td, (long)p->p_pid, p->p_comm, wmesg, ident); 184 185 DROP_GIANT(); 186 if (mtx != NULL) { 187 mtx_assert(mtx, MA_OWNED | MA_NOTRECURSED); 188 WITNESS_SAVE(&mtx->mtx_object, mtx); 189 mtx_unlock(mtx); 190 } 191 192 /* 193 * We put ourselves on the sleep queue and start our timeout 194 * before calling thread_suspend_check, as we could stop there, 195 * and a wakeup or a SIGCONT (or both) could occur while we were 196 * stopped without resuming us. Thus, we must be ready for sleep 197 * when cursig() is called. If the wakeup happens while we're 198 * stopped, then td will no longer be on a sleep queue upon 199 * return from cursig(). 200 */ 201 flags = SLEEPQ_MSLEEP; 202 if (catch) 203 flags |= SLEEPQ_INTERRUPTIBLE; 204 sleepq_add(sq, ident, mtx, wmesg, flags); 205 if (timo) 206 sleepq_set_timeout(ident, timo); 207 if (catch) { 208 sig = sleepq_catch_signals(ident); 209 } else 210 sig = 0; 211 212 /* 213 * Adjust this thread's priority. 214 * 215 * XXX: do we need to save priority in td_base_pri? 216 */ 217 mtx_lock_spin(&sched_lock); 218 sched_prio(td, priority & PRIMASK); 219 mtx_unlock_spin(&sched_lock); 220 221 if (timo && catch) 222 rval = sleepq_timedwait_sig(ident, sig != 0); 223 else if (timo) 224 rval = sleepq_timedwait(ident); 225 else if (catch) 226 rval = sleepq_wait_sig(ident); 227 else { 228 sleepq_wait(ident); 229 rval = 0; 230 } 231 if (rval == 0 && catch) 232 rval = sleepq_calc_signal_retval(sig); 233 #ifdef KTRACE 234 if (KTRPOINT(td, KTR_CSW)) 235 ktrcsw(0, 0); 236 #endif 237 PICKUP_GIANT(); 238 if (mtx != NULL && !(priority & PDROP)) { 239 mtx_lock(mtx); 240 WITNESS_RESTORE(&mtx->mtx_object, mtx); 241 } 242 return (rval); 243 } 244 245 /* 246 * Make all threads sleeping on the specified identifier runnable. 247 */ 248 void 249 wakeup(ident) 250 register void *ident; 251 { 252 253 sleepq_broadcast(ident, SLEEPQ_MSLEEP, -1); 254 } 255 256 /* 257 * Make a thread sleeping on the specified identifier runnable. 258 * May wake more than one thread if a target thread is currently 259 * swapped out. 260 */ 261 void 262 wakeup_one(ident) 263 register void *ident; 264 { 265 266 sleepq_signal(ident, SLEEPQ_MSLEEP, -1); 267 } 268 269 /* 270 * The machine independent parts of context switching. 271 */ 272 void 273 mi_switch(int flags, struct thread *newtd) 274 { 275 struct bintime new_switchtime; 276 struct thread *td; 277 struct proc *p; 278 279 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED); 280 td = curthread; /* XXX */ 281 p = td->td_proc; /* XXX */ 282 KASSERT(!TD_ON_RUNQ(td), ("mi_switch: called by old code")); 283 #ifdef INVARIANTS 284 if (!TD_ON_LOCK(td) && !TD_IS_RUNNING(td)) 285 mtx_assert(&Giant, MA_NOTOWNED); 286 #endif 287 KASSERT(td->td_critnest == 1 || (td->td_critnest == 2 && 288 (td->td_pflags & TDP_OWEPREEMPT) != 0 && (flags & SW_INVOL) != 0 && 289 newtd == NULL), 290 ("mi_switch: switch in a critical section")); 291 KASSERT((flags & (SW_INVOL | SW_VOL)) != 0, 292 ("mi_switch: switch must be voluntary or involuntary")); 293 KASSERT(newtd != curthread, ("mi_switch: preempting back to ourself")); 294 295 if (flags & SW_VOL) 296 p->p_stats->p_ru.ru_nvcsw++; 297 else 298 p->p_stats->p_ru.ru_nivcsw++; 299 300 /* 301 * Compute the amount of time during which the current 302 * process was running, and add that to its total so far. 303 */ 304 binuptime(&new_switchtime); 305 bintime_add(&p->p_runtime, &new_switchtime); 306 bintime_sub(&p->p_runtime, PCPU_PTR(switchtime)); 307 308 td->td_generation++; /* bump preempt-detect counter */ 309 310 /* 311 * Don't perform context switches from the debugger. 312 */ 313 if (kdb_active) { 314 mtx_unlock_spin(&sched_lock); 315 kdb_backtrace(); 316 kdb_reenter(); 317 panic("%s: did not reenter debugger", __func__); 318 } 319 320 /* 321 * Check if the process exceeds its cpu resource allocation. If 322 * over max, arrange to kill the process in ast(). 323 */ 324 if (p->p_cpulimit != RLIM_INFINITY && 325 p->p_runtime.sec > p->p_cpulimit) { 326 p->p_sflag |= PS_XCPU; 327 td->td_flags |= TDF_ASTPENDING; 328 } 329 330 /* 331 * Finish up stats for outgoing thread. 332 */ 333 cnt.v_swtch++; 334 PCPU_SET(switchtime, new_switchtime); 335 PCPU_SET(switchticks, ticks); 336 CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)", 337 (void *)td, td->td_kse, (long)p->p_pid, p->p_comm); 338 if (td->td_proc->p_flag & P_SA) 339 thread_switchout(td); 340 sched_switch(td, newtd); 341 342 CTR4(KTR_PROC, "mi_switch: new thread %p (kse %p, pid %ld, %s)", 343 (void *)td, td->td_kse, (long)p->p_pid, p->p_comm); 344 345 /* 346 * If the last thread was exiting, finish cleaning it up. 347 */ 348 if ((td = PCPU_GET(deadthread))) { 349 PCPU_SET(deadthread, NULL); 350 thread_stash(td); 351 } 352 } 353 354 /* 355 * Change process state to be runnable, 356 * placing it on the run queue if it is in memory, 357 * and awakening the swapper if it isn't in memory. 358 */ 359 void 360 setrunnable(struct thread *td) 361 { 362 struct proc *p; 363 364 p = td->td_proc; 365 mtx_assert(&sched_lock, MA_OWNED); 366 switch (p->p_state) { 367 case PRS_ZOMBIE: 368 panic("setrunnable(1)"); 369 default: 370 break; 371 } 372 switch (td->td_state) { 373 case TDS_RUNNING: 374 case TDS_RUNQ: 375 return; 376 case TDS_INHIBITED: 377 /* 378 * If we are only inhibited because we are swapped out 379 * then arange to swap in this process. Otherwise just return. 380 */ 381 if (td->td_inhibitors != TDI_SWAPPED) 382 return; 383 /* XXX: intentional fall-through ? */ 384 case TDS_CAN_RUN: 385 break; 386 default: 387 printf("state is 0x%x", td->td_state); 388 panic("setrunnable(2)"); 389 } 390 if ((p->p_sflag & PS_INMEM) == 0) { 391 if ((p->p_sflag & PS_SWAPPINGIN) == 0) { 392 p->p_sflag |= PS_SWAPINREQ; 393 #ifndef SMP 394 /* 395 * XXX: Disabled on SMP due to a LOR between 396 * sched_lock and the sleepqueue chain locks. 397 */ 398 wakeup(&proc0); 399 #endif 400 } 401 } else 402 sched_wakeup(td); 403 } 404 405 /* 406 * Compute a tenex style load average of a quantity on 407 * 1, 5 and 15 minute intervals. 408 * XXXKSE Needs complete rewrite when correct info is available. 409 * Completely Bogus.. only works with 1:1 (but compiles ok now :-) 410 */ 411 static void 412 loadav(void *arg) 413 { 414 int i, nrun; 415 struct loadavg *avg; 416 417 nrun = sched_load(); 418 avg = &averunnable; 419 420 for (i = 0; i < 3; i++) 421 avg->ldavg[i] = (cexp[i] * avg->ldavg[i] + 422 nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT; 423 424 /* 425 * Schedule the next update to occur after 5 seconds, but add a 426 * random variation to avoid synchronisation with processes that 427 * run at regular intervals. 428 */ 429 callout_reset(&loadav_callout, hz * 4 + (int)(random() % (hz * 2 + 1)), 430 loadav, NULL); 431 } 432 433 static void 434 lboltcb(void *arg) 435 { 436 wakeup(&lbolt); 437 callout_reset(&lbolt_callout, hz, lboltcb, NULL); 438 } 439 440 /* ARGSUSED */ 441 static void 442 synch_setup(dummy) 443 void *dummy; 444 { 445 callout_init(&loadav_callout, CALLOUT_MPSAFE); 446 callout_init(&lbolt_callout, CALLOUT_MPSAFE); 447 448 /* Kick off timeout driven events by calling first time. */ 449 loadav(NULL); 450 lboltcb(NULL); 451 } 452 453 /* 454 * General purpose yield system call 455 */ 456 int 457 yield(struct thread *td, struct yield_args *uap) 458 { 459 struct ksegrp *kg; 460 461 kg = td->td_ksegrp; 462 mtx_assert(&Giant, MA_NOTOWNED); 463 mtx_lock_spin(&sched_lock); 464 sched_prio(td, PRI_MAX_TIMESHARE); 465 mi_switch(SW_VOL, NULL); 466 mtx_unlock_spin(&sched_lock); 467 td->td_retval[0] = 0; 468 return (0); 469 } 470