kern_synch.c revision 41971
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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95 39 * $Id: kern_synch.c,v 1.69 1998/11/27 11:44:22 dg Exp $ 40 */ 41 42#include "opt_ktrace.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/proc.h> 47#include <sys/kernel.h> 48#include <sys/signalvar.h> 49#include <sys/resourcevar.h> 50#include <sys/vmmeter.h> 51#include <sys/sysctl.h> 52#include <vm/vm.h> 53#include <vm/vm_extern.h> 54#ifdef KTRACE 55#include <sys/uio.h> 56#include <sys/ktrace.h> 57#endif 58 59#include <machine/cpu.h> 60#ifdef SMP 61#include <machine/smp.h> 62#endif 63#include <machine/limits.h> /* for UCHAR_MAX = typeof(p_priority)_MAX */ 64 65static void rqinit __P((void *)); 66SYSINIT(runqueue, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, rqinit, NULL) 67 68u_char curpriority; /* usrpri of curproc */ 69int lbolt; /* once a second sleep address */ 70 71static void endtsleep __P((void *)); 72static void roundrobin __P((void *arg)); 73static void schedcpu __P((void *arg)); 74static void updatepri __P((struct proc *p)); 75 76#define MAXIMUM_SCHEDULE_QUANTUM (1000000) /* arbitrary limit */ 77#ifndef DEFAULT_SCHEDULE_QUANTUM 78#define DEFAULT_SCHEDULE_QUANTUM 10 79#endif 80static int quantum = DEFAULT_SCHEDULE_QUANTUM; /* default value */ 81 82static int 83sysctl_kern_quantum SYSCTL_HANDLER_ARGS 84{ 85 int error; 86 int new_val = quantum; 87 88 new_val = quantum; 89 error = sysctl_handle_int(oidp, &new_val, 0, req); 90 if (error == 0) { 91 if ((new_val > 0) && (new_val < MAXIMUM_SCHEDULE_QUANTUM)) { 92 quantum = new_val; 93 } else { 94 error = EINVAL; 95 } 96 } 97 return (error); 98} 99 100SYSCTL_PROC(_kern, OID_AUTO, quantum, CTLTYPE_INT|CTLFLAG_RW, 101 0, sizeof quantum, sysctl_kern_quantum, "I", ""); 102 103/* maybe_resched: Decide if you need to reschedule or not 104 * taking the priorities and schedulers into account. 105 */ 106static void maybe_resched(struct proc *chk) 107{ 108 struct proc *p = curproc; /* XXX */ 109 110 /* 111 * Compare priorities if the new process is on the same scheduler, 112 * otherwise the one on the more realtimeish scheduler wins. 113 * 114 * XXX idle scheduler still broken because proccess stays on idle 115 * scheduler during waits (such as when getting FS locks). If a 116 * standard process becomes runaway cpu-bound, the system can lockup 117 * due to idle-scheduler processes in wakeup never getting any cpu. 118 */ 119 if (p == 0 || 120 (chk->p_priority < curpriority && RTP_PRIO_BASE(p->p_rtprio.type) == RTP_PRIO_BASE(chk->p_rtprio.type)) || 121 RTP_PRIO_BASE(chk->p_rtprio.type) < RTP_PRIO_BASE(p->p_rtprio.type) 122 ) { 123 need_resched(); 124 } 125} 126 127#define ROUNDROBIN_INTERVAL (hz / quantum) 128int roundrobin_interval(void) 129{ 130 return ROUNDROBIN_INTERVAL; 131} 132 133/* 134 * Force switch among equal priority processes every 100ms. 135 */ 136/* ARGSUSED */ 137static void 138roundrobin(arg) 139 void *arg; 140{ 141#ifndef SMP 142 struct proc *p = curproc; /* XXX */ 143#endif 144 145#ifdef SMP 146 need_resched(); 147 forward_roundrobin(); 148#else 149 if (p == 0 || RTP_PRIO_NEED_RR(p->p_rtprio.type)) 150 need_resched(); 151#endif 152 153 timeout(roundrobin, NULL, ROUNDROBIN_INTERVAL); 154} 155 156/* 157 * Constants for digital decay and forget: 158 * 90% of (p_estcpu) usage in 5 * loadav time 159 * 95% of (p_pctcpu) usage in 60 seconds (load insensitive) 160 * Note that, as ps(1) mentions, this can let percentages 161 * total over 100% (I've seen 137.9% for 3 processes). 162 * 163 * Note that statclock() updates p_estcpu and p_cpticks asynchronously. 164 * 165 * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds. 166 * That is, the system wants to compute a value of decay such 167 * that the following for loop: 168 * for (i = 0; i < (5 * loadavg); i++) 169 * p_estcpu *= decay; 170 * will compute 171 * p_estcpu *= 0.1; 172 * for all values of loadavg: 173 * 174 * Mathematically this loop can be expressed by saying: 175 * decay ** (5 * loadavg) ~= .1 176 * 177 * The system computes decay as: 178 * decay = (2 * loadavg) / (2 * loadavg + 1) 179 * 180 * We wish to prove that the system's computation of decay 181 * will always fulfill the equation: 182 * decay ** (5 * loadavg) ~= .1 183 * 184 * If we compute b as: 185 * b = 2 * loadavg 186 * then 187 * decay = b / (b + 1) 188 * 189 * We now need to prove two things: 190 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) 191 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) 192 * 193 * Facts: 194 * For x close to zero, exp(x) =~ 1 + x, since 195 * exp(x) = 0! + x**1/1! + x**2/2! + ... . 196 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. 197 * For x close to zero, ln(1+x) =~ x, since 198 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 199 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). 200 * ln(.1) =~ -2.30 201 * 202 * Proof of (1): 203 * Solve (factor)**(power) =~ .1 given power (5*loadav): 204 * solving for factor, 205 * ln(factor) =~ (-2.30/5*loadav), or 206 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = 207 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED 208 * 209 * Proof of (2): 210 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): 211 * solving for power, 212 * power*ln(b/(b+1)) =~ -2.30, or 213 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED 214 * 215 * Actual power values for the implemented algorithm are as follows: 216 * loadav: 1 2 3 4 217 * power: 5.68 10.32 14.94 19.55 218 */ 219 220/* calculations for digital decay to forget 90% of usage in 5*loadav sec */ 221#define loadfactor(loadav) (2 * (loadav)) 222#define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE)) 223 224/* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ 225static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ 226SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, ""); 227 228/* kernel uses `FSCALE', userland (SHOULD) use kern.fscale */ 229static int fscale __unused = FSCALE; 230SYSCTL_INT(_kern, OID_AUTO, fscale, CTLFLAG_RD, 0, FSCALE, ""); 231 232/* 233 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the 234 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below 235 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT). 236 * 237 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used: 238 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits). 239 * 240 * If you don't want to bother with the faster/more-accurate formula, you 241 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate 242 * (more general) method of calculating the %age of CPU used by a process. 243 */ 244#define CCPU_SHIFT 11 245 246/* 247 * Recompute process priorities, every hz ticks. 248 */ 249/* ARGSUSED */ 250static void 251schedcpu(arg) 252 void *arg; 253{ 254 register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 255 register struct proc *p; 256 register int realstathz, s; 257 register unsigned int newcpu; 258 259 realstathz = stathz ? stathz : hz; 260 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) { 261 /* 262 * Increment time in/out of memory and sleep time 263 * (if sleeping). We ignore overflow; with 16-bit int's 264 * (remember them?) overflow takes 45 days. 265 */ 266 p->p_swtime++; 267 if (p->p_stat == SSLEEP || p->p_stat == SSTOP) 268 p->p_slptime++; 269 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT; 270 /* 271 * If the process has slept the entire second, 272 * stop recalculating its priority until it wakes up. 273 */ 274 if (p->p_slptime > 1) 275 continue; 276 s = splhigh(); /* prevent state changes and protect run queue */ 277 /* 278 * p_pctcpu is only for ps. 279 */ 280#if (FSHIFT >= CCPU_SHIFT) 281 p->p_pctcpu += (realstathz == 100)? 282 ((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT): 283 100 * (((fixpt_t) p->p_cpticks) 284 << (FSHIFT - CCPU_SHIFT)) / realstathz; 285#else 286 p->p_pctcpu += ((FSCALE - ccpu) * 287 (p->p_cpticks * FSCALE / realstathz)) >> FSHIFT; 288#endif 289 p->p_cpticks = 0; 290 newcpu = (u_int) decay_cpu(loadfac, p->p_estcpu) + p->p_nice; 291 p->p_estcpu = min(newcpu, UCHAR_MAX); 292 resetpriority(p); 293 if (p->p_priority >= PUSER) { 294#define PPQ (128 / NQS) /* priorities per queue */ 295 if ((p != curproc) && 296#ifdef SMP 297 (u_char)p->p_oncpu == 0xff && /* idle */ 298#endif 299 p->p_stat == SRUN && 300 (p->p_flag & P_INMEM) && 301 (p->p_priority / PPQ) != (p->p_usrpri / PPQ)) { 302 remrq(p); 303 p->p_priority = p->p_usrpri; 304 setrunqueue(p); 305 } else 306 p->p_priority = p->p_usrpri; 307 } 308 splx(s); 309 } 310 vmmeter(); 311 wakeup((caddr_t)&lbolt); 312 timeout(schedcpu, (void *)0, hz); 313} 314 315/* 316 * Recalculate the priority of a process after it has slept for a while. 317 * For all load averages >= 1 and max p_estcpu of 255, sleeping for at 318 * least six times the loadfactor will decay p_estcpu to zero. 319 */ 320static void 321updatepri(p) 322 register struct proc *p; 323{ 324 register unsigned int newcpu = p->p_estcpu; 325 register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 326 327 if (p->p_slptime > 5 * loadfac) 328 p->p_estcpu = 0; 329 else { 330 p->p_slptime--; /* the first time was done in schedcpu */ 331 while (newcpu && --p->p_slptime) 332 newcpu = (int) decay_cpu(loadfac, newcpu); 333 p->p_estcpu = min(newcpu, UCHAR_MAX); 334 } 335 resetpriority(p); 336} 337 338/* 339 * We're only looking at 7 bits of the address; everything is 340 * aligned to 4, lots of things are aligned to greater powers 341 * of 2. Shift right by 8, i.e. drop the bottom 256 worth. 342 */ 343#define TABLESIZE 128 344static TAILQ_HEAD(slpquehead, proc) slpque[TABLESIZE]; 345#define LOOKUP(x) (((intptr_t)(x) >> 8) & (TABLESIZE - 1)) 346 347/* 348 * During autoconfiguration or after a panic, a sleep will simply 349 * lower the priority briefly to allow interrupts, then return. 350 * The priority to be used (safepri) is machine-dependent, thus this 351 * value is initialized and maintained in the machine-dependent layers. 352 * This priority will typically be 0, or the lowest priority 353 * that is safe for use on the interrupt stack; it can be made 354 * higher to block network software interrupts after panics. 355 */ 356int safepri; 357 358void 359sleepinit() 360{ 361 int i; 362 363 for (i = 0; i < TABLESIZE; i++) 364 TAILQ_INIT(&slpque[i]); 365} 366 367/* 368 * General sleep call. Suspends the current process until a wakeup is 369 * performed on the specified identifier. The process will then be made 370 * runnable with the specified priority. Sleeps at most timo/hz seconds 371 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 372 * before and after sleeping, else signals are not checked. Returns 0 if 373 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 374 * signal needs to be delivered, ERESTART is returned if the current system 375 * call should be restarted if possible, and EINTR is returned if the system 376 * call should be interrupted by the signal (return EINTR). 377 */ 378int 379tsleep(ident, priority, wmesg, timo) 380 void *ident; 381 int priority, timo; 382 const char *wmesg; 383{ 384 struct proc *p = curproc; 385 int s, sig, catch = priority & PCATCH; 386 struct callout_handle thandle; 387 388#ifdef KTRACE 389 if (KTRPOINT(p, KTR_CSW)) 390 ktrcsw(p->p_tracep, 1, 0); 391#endif 392 s = splhigh(); 393 if (cold || panicstr) { 394 /* 395 * After a panic, or during autoconfiguration, 396 * just give interrupts a chance, then just return; 397 * don't run any other procs or panic below, 398 * in case this is the idle process and already asleep. 399 */ 400 splx(safepri); 401 splx(s); 402 return (0); 403 } 404#ifdef DIAGNOSTIC 405 if(p == NULL) 406 panic("tsleep1"); 407 if (ident == NULL || p->p_stat != SRUN) 408 panic("tsleep"); 409 /* XXX This is not exhaustive, just the most common case */ 410#ifdef NOTDEF 411 /* 412 * This can happen legitimately now with asleep()/await() 413 */ 414 if ((p->p_procq.tqe_prev != NULL) && (*p->p_procq.tqe_prev == p)) 415 panic("sleeping process already on another queue"); 416#endif 417#endif 418 /* 419 * Process may be sitting on a slpque if asleep() was called, remove 420 * it before re-adding. 421 */ 422 if (p->p_wchan != NULL) 423 unsleep(p); 424 425 p->p_wchan = ident; 426 p->p_wmesg = wmesg; 427 p->p_slptime = 0; 428 p->p_priority = priority & PRIMASK; 429 TAILQ_INSERT_TAIL(&slpque[LOOKUP(ident)], p, p_procq); 430 if (timo) 431 thandle = timeout(endtsleep, (void *)p, timo); 432 /* 433 * We put ourselves on the sleep queue and start our timeout 434 * before calling CURSIG, as we could stop there, and a wakeup 435 * or a SIGCONT (or both) could occur while we were stopped. 436 * A SIGCONT would cause us to be marked as SSLEEP 437 * without resuming us, thus we must be ready for sleep 438 * when CURSIG is called. If the wakeup happens while we're 439 * stopped, p->p_wchan will be 0 upon return from CURSIG. 440 */ 441 if (catch) { 442 p->p_flag |= P_SINTR; 443 if ((sig = CURSIG(p))) { 444 if (p->p_wchan) 445 unsleep(p); 446 p->p_stat = SRUN; 447 goto resume; 448 } 449 if (p->p_wchan == 0) { 450 catch = 0; 451 goto resume; 452 } 453 } else 454 sig = 0; 455 p->p_stat = SSLEEP; 456 p->p_stats->p_ru.ru_nvcsw++; 457 mi_switch(); 458resume: 459 curpriority = p->p_usrpri; 460 splx(s); 461 p->p_flag &= ~P_SINTR; 462 if (p->p_flag & P_TIMEOUT) { 463 p->p_flag &= ~P_TIMEOUT; 464 if (sig == 0) { 465#ifdef KTRACE 466 if (KTRPOINT(p, KTR_CSW)) 467 ktrcsw(p->p_tracep, 0, 0); 468#endif 469 return (EWOULDBLOCK); 470 } 471 } else if (timo) 472 untimeout(endtsleep, (void *)p, thandle); 473 if (catch && (sig != 0 || (sig = CURSIG(p)))) { 474#ifdef KTRACE 475 if (KTRPOINT(p, KTR_CSW)) 476 ktrcsw(p->p_tracep, 0, 0); 477#endif 478 if (p->p_sigacts->ps_sigintr & sigmask(sig)) 479 return (EINTR); 480 return (ERESTART); 481 } 482#ifdef KTRACE 483 if (KTRPOINT(p, KTR_CSW)) 484 ktrcsw(p->p_tracep, 0, 0); 485#endif 486 return (0); 487} 488 489/* 490 * asleep() - async sleep call. Place process on wait queue and return 491 * immediately without blocking. The process stays runnable until await() 492 * is called. If ident is NULL, remove process from wait queue if it is still 493 * on one. 494 * 495 * Only the most recent sleep condition is effective when making successive 496 * calls to asleep() or when calling tsleep(). 497 * 498 * The timeout, if any, is not initiated until await() is called. The sleep 499 * priority, signal, and timeout is specified in the asleep() call but may be 500 * overriden in the await() call. 501 * 502 * <<<<<<<< EXPERIMENTAL, UNTESTED >>>>>>>>>> 503 */ 504 505int 506asleep(void *ident, int priority, const char *wmesg, int timo) 507{ 508 struct proc *p = curproc; 509 int s; 510 511 /* 512 * splhigh() while manipulating sleep structures and slpque. 513 * 514 * Remove preexisting wait condition (if any) and place process 515 * on appropriate slpque, but do not put process to sleep. 516 */ 517 518 s = splhigh(); 519 520 if (p->p_wchan != NULL) 521 unsleep(p); 522 523 if (ident) { 524 p->p_wchan = ident; 525 p->p_wmesg = wmesg; 526 p->p_slptime = 0; 527 p->p_asleep.as_priority = priority; 528 p->p_asleep.as_timo = timo; 529 TAILQ_INSERT_TAIL(&slpque[LOOKUP(ident)], p, p_procq); 530 } 531 532 splx(s); 533 534 return(0); 535} 536 537/* 538 * await() - wait for async condition to occur. The process blocks until 539 * wakeup() is called on the most recent asleep() address. If wakeup is called 540 * priority to await(), await() winds up being a NOP. 541 * 542 * If await() is called more then once (without an intervening asleep() call), 543 * await() is still effectively a NOP but it calls mi_switch() to give other 544 * processes some cpu before returning. The process is left runnable. 545 * 546 * <<<<<<<< EXPERIMENTAL, UNTESTED >>>>>>>>>> 547 */ 548 549int 550await(int priority, int timo) 551{ 552 struct proc *p = curproc; 553 int s; 554 555 s = splhigh(); 556 557 if (p->p_wchan != NULL) { 558 struct callout_handle thandle; 559 int sig; 560 int catch; 561 562 /* 563 * The call to await() can override defaults specified in 564 * the original asleep(). 565 */ 566 if (priority < 0) 567 priority = p->p_asleep.as_priority; 568 if (timo < 0) 569 timo = p->p_asleep.as_timo; 570 571 /* 572 * Install timeout 573 */ 574 575 if (timo) 576 thandle = timeout(endtsleep, (void *)p, timo); 577 578 sig = 0; 579 catch = priority & PCATCH; 580 581 if (catch) { 582 p->p_flag |= P_SINTR; 583 if ((sig = CURSIG(p))) { 584 if (p->p_wchan) 585 unsleep(p); 586 p->p_stat = SRUN; 587 goto resume; 588 } 589 if (p->p_wchan == NULL) { 590 catch = 0; 591 goto resume; 592 } 593 } 594 p->p_stat = SSLEEP; 595 p->p_stats->p_ru.ru_nvcsw++; 596 mi_switch(); 597resume: 598 curpriority = p->p_usrpri; 599 600 splx(s); 601 p->p_flag &= ~P_SINTR; 602 if (p->p_flag & P_TIMEOUT) { 603 p->p_flag &= ~P_TIMEOUT; 604 if (sig == 0) { 605#ifdef KTRACE 606 if (KTRPOINT(p, KTR_CSW)) 607 ktrcsw(p->p_tracep, 0, 0); 608#endif 609 return (EWOULDBLOCK); 610 } 611 } else if (timo) 612 untimeout(endtsleep, (void *)p, thandle); 613 if (catch && (sig != 0 || (sig = CURSIG(p)))) { 614#ifdef KTRACE 615 if (KTRPOINT(p, KTR_CSW)) 616 ktrcsw(p->p_tracep, 0, 0); 617#endif 618 if (p->p_sigacts->ps_sigintr & sigmask(sig)) 619 return (EINTR); 620 return (ERESTART); 621 } 622#ifdef KTRACE 623 if (KTRPOINT(p, KTR_CSW)) 624 ktrcsw(p->p_tracep, 0, 0); 625#endif 626 } else { 627 /* 628 * If as_priority is 0, await() has been called without an 629 * intervening asleep(). We are still effectively a NOP, 630 * but we call mi_switch() for safety. 631 */ 632 633 if (p->p_asleep.as_priority == 0) { 634 p->p_stats->p_ru.ru_nvcsw++; 635 mi_switch(); 636 } 637 splx(s); 638 } 639 640 /* 641 * clear p_asleep.as_priority as an indication that await() has been 642 * called. If await() is called again without an intervening asleep(), 643 * await() is still effectively a NOP but the above mi_switch() code 644 * is triggered as a safety. 645 */ 646 p->p_asleep.as_priority = 0; 647 648 return (0); 649} 650 651/* 652 * Implement timeout for tsleep or asleep()/await() 653 * 654 * If process hasn't been awakened (wchan non-zero), 655 * set timeout flag and undo the sleep. If proc 656 * is stopped, just unsleep so it will remain stopped. 657 */ 658static void 659endtsleep(arg) 660 void *arg; 661{ 662 register struct proc *p; 663 int s; 664 665 p = (struct proc *)arg; 666 s = splhigh(); 667 if (p->p_wchan) { 668 if (p->p_stat == SSLEEP) 669 setrunnable(p); 670 else 671 unsleep(p); 672 p->p_flag |= P_TIMEOUT; 673 } 674 splx(s); 675} 676 677/* 678 * Remove a process from its wait queue 679 */ 680void 681unsleep(p) 682 register struct proc *p; 683{ 684 int s; 685 686 s = splhigh(); 687 if (p->p_wchan) { 688 TAILQ_REMOVE(&slpque[LOOKUP(p->p_wchan)], p, p_procq); 689 p->p_wchan = 0; 690 } 691 splx(s); 692} 693 694/* 695 * Make all processes sleeping on the specified identifier runnable. 696 */ 697void 698wakeup(ident) 699 register void *ident; 700{ 701 register struct slpquehead *qp; 702 register struct proc *p; 703 int s; 704 705 s = splhigh(); 706 qp = &slpque[LOOKUP(ident)]; 707restart: 708 for (p = qp->tqh_first; p != NULL; p = p->p_procq.tqe_next) { 709#ifdef DIAGNOSTIC 710#ifdef NOTDEF 711 /* 712 * The process can legitimately be running now with 713 * asleep()/await(). 714 */ 715 if (p->p_stat != SSLEEP && p->p_stat != SSTOP) 716 panic("wakeup"); 717#endif 718#endif 719 if (p->p_wchan == ident) { 720 TAILQ_REMOVE(qp, p, p_procq); 721 p->p_wchan = 0; 722 if (p->p_stat == SSLEEP) { 723 /* OPTIMIZED EXPANSION OF setrunnable(p); */ 724 if (p->p_slptime > 1) 725 updatepri(p); 726 p->p_slptime = 0; 727 p->p_stat = SRUN; 728 if (p->p_flag & P_INMEM) { 729 setrunqueue(p); 730 maybe_resched(p); 731 } else { 732 p->p_flag |= P_SWAPINREQ; 733 wakeup((caddr_t)&proc0); 734 } 735 /* END INLINE EXPANSION */ 736 goto restart; 737 } 738 } 739 } 740 splx(s); 741} 742 743/* 744 * Make a process sleeping on the specified identifier runnable. 745 * May wake more than one process if a target prcoess is currently 746 * swapped out. 747 */ 748void 749wakeup_one(ident) 750 register void *ident; 751{ 752 register struct slpquehead *qp; 753 register struct proc *p; 754 int s; 755 756 s = splhigh(); 757 qp = &slpque[LOOKUP(ident)]; 758 759 for (p = qp->tqh_first; p != NULL; p = p->p_procq.tqe_next) { 760#ifdef DIAGNOSTIC 761#ifdef NOTDEF 762 /* 763 * The process can legitimately be running now with 764 * asleep()/await(). 765 */ 766 if (p->p_stat != SSLEEP && p->p_stat != SSTOP) 767 panic("wakeup_one"); 768#endif 769#endif 770 if (p->p_wchan == ident) { 771 TAILQ_REMOVE(qp, p, p_procq); 772 p->p_wchan = 0; 773 if (p->p_stat == SSLEEP) { 774 /* OPTIMIZED EXPANSION OF setrunnable(p); */ 775 if (p->p_slptime > 1) 776 updatepri(p); 777 p->p_slptime = 0; 778 p->p_stat = SRUN; 779 if (p->p_flag & P_INMEM) { 780 setrunqueue(p); 781 maybe_resched(p); 782 break; 783 } else { 784 p->p_flag |= P_SWAPINREQ; 785 wakeup((caddr_t)&proc0); 786 } 787 /* END INLINE EXPANSION */ 788 } 789 } 790 } 791 splx(s); 792} 793 794/* 795 * The machine independent parts of mi_switch(). 796 * Must be called at splstatclock() or higher. 797 */ 798void 799mi_switch() 800{ 801 register struct proc *p = curproc; /* XXX */ 802 register struct rlimit *rlim; 803 int x; 804 805 /* 806 * XXX this spl is almost unnecessary. It is partly to allow for 807 * sloppy callers that don't do it (issignal() via CURSIG() is the 808 * main offender). It is partly to work around a bug in the i386 809 * cpu_switch() (the ipl is not preserved). We ran for years 810 * without it. I think there was only a interrupt latency problem. 811 * The main caller, tsleep(), does an splx() a couple of instructions 812 * after calling here. The buggy caller, issignal(), usually calls 813 * here at spl0() and sometimes returns at splhigh(). The process 814 * then runs for a little too long at splhigh(). The ipl gets fixed 815 * when the process returns to user mode (or earlier). 816 * 817 * It would probably be better to always call here at spl0(). Callers 818 * are prepared to give up control to another process, so they must 819 * be prepared to be interrupted. The clock stuff here may not 820 * actually need splstatclock(). 821 */ 822 x = splstatclock(); 823 824#ifdef SIMPLELOCK_DEBUG 825 if (p->p_simple_locks) 826 printf("sleep: holding simple lock\n"); 827#endif 828 /* 829 * Compute the amount of time during which the current 830 * process was running, and add that to its total so far. 831 */ 832 microuptime(&switchtime); 833 p->p_runtime += (switchtime.tv_usec - p->p_switchtime.tv_usec) + 834 (switchtime.tv_sec - p->p_switchtime.tv_sec) * (int64_t)1000000; 835 836 /* 837 * Check if the process exceeds its cpu resource allocation. 838 * If over max, kill it. 839 */ 840 if (p->p_stat != SZOMB && p->p_limit->p_cpulimit != RLIM_INFINITY && 841 p->p_runtime > p->p_limit->p_cpulimit) { 842 rlim = &p->p_rlimit[RLIMIT_CPU]; 843 if (p->p_runtime / (rlim_t)1000000 >= rlim->rlim_max) { 844 killproc(p, "exceeded maximum CPU limit"); 845 } else { 846 psignal(p, SIGXCPU); 847 if (rlim->rlim_cur < rlim->rlim_max) { 848 /* XXX: we should make a private copy */ 849 rlim->rlim_cur += 5; 850 } 851 } 852 } 853 854 /* 855 * Pick a new current process and record its start time. 856 */ 857 cnt.v_swtch++; 858 cpu_switch(p); 859 if (switchtime.tv_sec) 860 p->p_switchtime = switchtime; 861 else 862 microuptime(&p->p_switchtime); 863 splx(x); 864} 865 866/* 867 * Initialize the (doubly-linked) run queues 868 * to be empty. 869 */ 870/* ARGSUSED*/ 871static void 872rqinit(dummy) 873 void *dummy; 874{ 875 register int i; 876 877 for (i = 0; i < NQS; i++) { 878 qs[i].ph_link = qs[i].ph_rlink = (struct proc *)&qs[i]; 879 rtqs[i].ph_link = rtqs[i].ph_rlink = (struct proc *)&rtqs[i]; 880 idqs[i].ph_link = idqs[i].ph_rlink = (struct proc *)&idqs[i]; 881 } 882} 883 884/* 885 * Change process state to be runnable, 886 * placing it on the run queue if it is in memory, 887 * and awakening the swapper if it isn't in memory. 888 */ 889void 890setrunnable(p) 891 register struct proc *p; 892{ 893 register int s; 894 895 s = splhigh(); 896 switch (p->p_stat) { 897 case 0: 898 case SRUN: 899 case SZOMB: 900 default: 901 panic("setrunnable"); 902 case SSTOP: 903 case SSLEEP: 904 unsleep(p); /* e.g. when sending signals */ 905 break; 906 907 case SIDL: 908 break; 909 } 910 p->p_stat = SRUN; 911 if (p->p_flag & P_INMEM) 912 setrunqueue(p); 913 splx(s); 914 if (p->p_slptime > 1) 915 updatepri(p); 916 p->p_slptime = 0; 917 if ((p->p_flag & P_INMEM) == 0) { 918 p->p_flag |= P_SWAPINREQ; 919 wakeup((caddr_t)&proc0); 920 } 921 else 922 maybe_resched(p); 923} 924 925/* 926 * Compute the priority of a process when running in user mode. 927 * Arrange to reschedule if the resulting priority is better 928 * than that of the current process. 929 */ 930void 931resetpriority(p) 932 register struct proc *p; 933{ 934 register unsigned int newpriority; 935 936 if (p->p_rtprio.type == RTP_PRIO_NORMAL) { 937 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice; 938 newpriority = min(newpriority, MAXPRI); 939 p->p_usrpri = newpriority; 940 } 941 maybe_resched(p); 942} 943 944/* ARGSUSED */ 945static void sched_setup __P((void *dummy)); 946static void 947sched_setup(dummy) 948 void *dummy; 949{ 950 /* Kick off timeout driven events by calling first time. */ 951 roundrobin(NULL); 952 schedcpu(NULL); 953} 954SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL) 955 956