kern_synch.c revision 2441
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 8.6 (Berkeley) 1/21/94 39 * $Id: kern_synch.c,v 1.3 1994/08/02 07:42:17 davidg Exp $ 40 */ 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/proc.h> 45#include <sys/kernel.h> 46#include <sys/buf.h> 47#include <sys/signalvar.h> 48#include <sys/resourcevar.h> 49#include <sys/vmmeter.h> 50#ifdef KTRACE 51#include <sys/ktrace.h> 52#endif 53 54#include <machine/cpu.h> 55 56u_char curpriority; /* usrpri of curproc */ 57int lbolt; /* once a second sleep address */ 58 59/* 60 * Force switch among equal priority processes every 100ms. 61 */ 62/* ARGSUSED */ 63void 64roundrobin(arg) 65 void *arg; 66{ 67 68 need_resched(); 69 timeout(roundrobin, NULL, hz / 10); 70} 71 72/* 73 * Constants for digital decay and forget: 74 * 90% of (p_estcpu) usage in 5 * loadav time 75 * 95% of (p_pctcpu) usage in 60 seconds (load insensitive) 76 * Note that, as ps(1) mentions, this can let percentages 77 * total over 100% (I've seen 137.9% for 3 processes). 78 * 79 * Note that hardclock updates p_estcpu and p_cpticks independently. 80 * 81 * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds. 82 * That is, the system wants to compute a value of decay such 83 * that the following for loop: 84 * for (i = 0; i < (5 * loadavg); i++) 85 * p_estcpu *= decay; 86 * will compute 87 * p_estcpu *= 0.1; 88 * for all values of loadavg: 89 * 90 * Mathematically this loop can be expressed by saying: 91 * decay ** (5 * loadavg) ~= .1 92 * 93 * The system computes decay as: 94 * decay = (2 * loadavg) / (2 * loadavg + 1) 95 * 96 * We wish to prove that the system's computation of decay 97 * will always fulfill the equation: 98 * decay ** (5 * loadavg) ~= .1 99 * 100 * If we compute b as: 101 * b = 2 * loadavg 102 * then 103 * decay = b / (b + 1) 104 * 105 * We now need to prove two things: 106 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) 107 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) 108 * 109 * Facts: 110 * For x close to zero, exp(x) =~ 1 + x, since 111 * exp(x) = 0! + x**1/1! + x**2/2! + ... . 112 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. 113 * For x close to zero, ln(1+x) =~ x, since 114 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 115 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). 116 * ln(.1) =~ -2.30 117 * 118 * Proof of (1): 119 * Solve (factor)**(power) =~ .1 given power (5*loadav): 120 * solving for factor, 121 * ln(factor) =~ (-2.30/5*loadav), or 122 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = 123 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED 124 * 125 * Proof of (2): 126 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): 127 * solving for power, 128 * power*ln(b/(b+1)) =~ -2.30, or 129 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED 130 * 131 * Actual power values for the implemented algorithm are as follows: 132 * loadav: 1 2 3 4 133 * power: 5.68 10.32 14.94 19.55 134 */ 135 136/* calculations for digital decay to forget 90% of usage in 5*loadav sec */ 137#define loadfactor(loadav) (2 * (loadav)) 138#define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE)) 139 140/* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ 141fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ 142 143/* 144 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the 145 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below 146 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT). 147 * 148 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used: 149 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits). 150 * 151 * If you dont want to bother with the faster/more-accurate formula, you 152 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate 153 * (more general) method of calculating the %age of CPU used by a process. 154 */ 155#define CCPU_SHIFT 11 156 157/* 158 * Recompute process priorities, every hz ticks. 159 */ 160/* ARGSUSED */ 161void 162schedcpu(arg) 163 void *arg; 164{ 165 register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 166 register struct proc *p; 167 register int s; 168 register unsigned int newcpu; 169 170 wakeup((caddr_t)&lbolt); 171 for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { 172 /* 173 * Increment time in/out of memory and sleep time 174 * (if sleeping). We ignore overflow; with 16-bit int's 175 * (remember them?) overflow takes 45 days. 176 */ 177 p->p_swtime++; 178 if (p->p_stat == SSLEEP || p->p_stat == SSTOP) 179 p->p_slptime++; 180 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT; 181 /* 182 * If the process has slept the entire second, 183 * stop recalculating its priority until it wakes up. 184 */ 185 if (p->p_slptime > 1) 186 continue; 187 s = splstatclock(); /* prevent state changes */ 188 /* 189 * p_pctcpu is only for ps. 190 */ 191#if (FSHIFT >= CCPU_SHIFT) 192 p->p_pctcpu += (hz == 100)? 193 ((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT): 194 100 * (((fixpt_t) p->p_cpticks) 195 << (FSHIFT - CCPU_SHIFT)) / hz; 196#else 197 p->p_pctcpu += ((FSCALE - ccpu) * 198 (p->p_cpticks * FSCALE / hz)) >> FSHIFT; 199#endif 200 p->p_cpticks = 0; 201 newcpu = (u_int) decay_cpu(loadfac, p->p_estcpu) + p->p_nice; 202 p->p_estcpu = min(newcpu, UCHAR_MAX); 203 resetpriority(p); 204 if (p->p_priority >= PUSER) { 205#define PPQ (128 / NQS) /* priorities per queue */ 206 if ((p != curproc) && 207 p->p_stat == SRUN && 208 (p->p_flag & P_INMEM) && 209 (p->p_priority / PPQ) != (p->p_usrpri / PPQ)) { 210 remrq(p); 211 p->p_priority = p->p_usrpri; 212 setrunqueue(p); 213 } else 214 p->p_priority = p->p_usrpri; 215 } 216 splx(s); 217 } 218 vmmeter(); 219 if (bclnlist != NULL) 220 wakeup((caddr_t)pageproc); 221 timeout(schedcpu, (void *)0, hz); 222} 223 224/* 225 * Recalculate the priority of a process after it has slept for a while. 226 * For all load averages >= 1 and max p_estcpu of 255, sleeping for at 227 * least six times the loadfactor will decay p_estcpu to zero. 228 */ 229void 230updatepri(p) 231 register struct proc *p; 232{ 233 register unsigned int newcpu = p->p_estcpu; 234 register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 235 236 if (p->p_slptime > 5 * loadfac) 237 p->p_estcpu = 0; 238 else { 239 p->p_slptime--; /* the first time was done in schedcpu */ 240 while (newcpu && --p->p_slptime) 241 newcpu = (int) decay_cpu(loadfac, newcpu); 242 p->p_estcpu = min(newcpu, UCHAR_MAX); 243 } 244 resetpriority(p); 245} 246 247/* 248 * We're only looking at 7 bits of the address; everything is 249 * aligned to 4, lots of things are aligned to greater powers 250 * of 2. Shift right by 8, i.e. drop the bottom 256 worth. 251 */ 252#define TABLESIZE 128 253#define LOOKUP(x) (((int)(x) >> 8) & (TABLESIZE - 1)) 254struct slpque { 255 struct proc *sq_head; 256 struct proc **sq_tailp; 257} slpque[TABLESIZE]; 258 259/* 260 * During autoconfiguration or after a panic, a sleep will simply 261 * lower the priority briefly to allow interrupts, then return. 262 * The priority to be used (safepri) is machine-dependent, thus this 263 * value is initialized and maintained in the machine-dependent layers. 264 * This priority will typically be 0, or the lowest priority 265 * that is safe for use on the interrupt stack; it can be made 266 * higher to block network software interrupts after panics. 267 */ 268int safepri; 269 270/* 271 * General sleep call. Suspends the current process until a wakeup is 272 * performed on the specified identifier. The process will then be made 273 * runnable with the specified priority. Sleeps at most timo/hz seconds 274 * (0 means no timeout). If pri includes PCATCH flag, signals are checked 275 * before and after sleeping, else signals are not checked. Returns 0 if 276 * awakened, EWOULDBLOCK if the timeout expires. If PCATCH is set and a 277 * signal needs to be delivered, ERESTART is returned if the current system 278 * call should be restarted if possible, and EINTR is returned if the system 279 * call should be interrupted by the signal (return EINTR). 280 */ 281int 282tsleep(ident, priority, wmesg, timo) 283 void *ident; 284 int priority, timo; 285 char *wmesg; 286{ 287 register struct proc *p = curproc; 288 register struct slpque *qp; 289 register s; 290 int sig, catch = priority & PCATCH; 291 extern int cold; 292 void endtsleep __P((void *)); 293 294#ifdef KTRACE 295 if (KTRPOINT(p, KTR_CSW)) 296 ktrcsw(p->p_tracep, 1, 0); 297#endif 298 s = splhigh(); 299 if (cold || panicstr) { 300 /* 301 * After a panic, or during autoconfiguration, 302 * just give interrupts a chance, then just return; 303 * don't run any other procs or panic below, 304 * in case this is the idle process and already asleep. 305 */ 306 splx(safepri); 307 splx(s); 308 return (0); 309 } 310#ifdef DIAGNOSTIC 311 if (ident == NULL || p->p_stat != SRUN || p->p_back) 312 panic("tsleep"); 313#endif 314 p->p_wchan = ident; 315 p->p_wmesg = wmesg; 316 p->p_slptime = 0; 317 p->p_priority = priority & PRIMASK; 318 qp = &slpque[LOOKUP(ident)]; 319 if (qp->sq_head == 0) 320 qp->sq_head = p; 321 else 322 *qp->sq_tailp = p; 323 *(qp->sq_tailp = &p->p_forw) = 0; 324 if (timo) 325 timeout(endtsleep, (void *)p, timo); 326 /* 327 * We put ourselves on the sleep queue and start our timeout 328 * before calling CURSIG, as we could stop there, and a wakeup 329 * or a SIGCONT (or both) could occur while we were stopped. 330 * A SIGCONT would cause us to be marked as SSLEEP 331 * without resuming us, thus we must be ready for sleep 332 * when CURSIG is called. If the wakeup happens while we're 333 * stopped, p->p_wchan will be 0 upon return from CURSIG. 334 */ 335 if (catch) { 336 p->p_flag |= P_SINTR; 337 if (sig = CURSIG(p)) { 338 if (p->p_wchan) 339 unsleep(p); 340 p->p_stat = SRUN; 341 goto resume; 342 } 343 if (p->p_wchan == 0) { 344 catch = 0; 345 goto resume; 346 } 347 } else 348 sig = 0; 349 p->p_stat = SSLEEP; 350 p->p_stats->p_ru.ru_nvcsw++; 351 mi_switch(); 352resume: 353 curpriority = p->p_usrpri; 354 splx(s); 355 p->p_flag &= ~P_SINTR; 356 if (p->p_flag & P_TIMEOUT) { 357 p->p_flag &= ~P_TIMEOUT; 358 if (sig == 0) { 359#ifdef KTRACE 360 if (KTRPOINT(p, KTR_CSW)) 361 ktrcsw(p->p_tracep, 0, 0); 362#endif 363 return (EWOULDBLOCK); 364 } 365 } else if (timo) 366 untimeout(endtsleep, (void *)p); 367 if (catch && (sig != 0 || (sig = CURSIG(p)))) { 368#ifdef KTRACE 369 if (KTRPOINT(p, KTR_CSW)) 370 ktrcsw(p->p_tracep, 0, 0); 371#endif 372 if (p->p_sigacts->ps_sigintr & sigmask(sig)) 373 return (EINTR); 374 return (ERESTART); 375 } 376#ifdef KTRACE 377 if (KTRPOINT(p, KTR_CSW)) 378 ktrcsw(p->p_tracep, 0, 0); 379#endif 380 return (0); 381} 382 383/* 384 * Implement timeout for tsleep. 385 * If process hasn't been awakened (wchan non-zero), 386 * set timeout flag and undo the sleep. If proc 387 * is stopped, just unsleep so it will remain stopped. 388 */ 389void 390endtsleep(arg) 391 void *arg; 392{ 393 register struct proc *p; 394 int s; 395 396 p = (struct proc *)arg; 397 s = splhigh(); 398 if (p->p_wchan) { 399 if (p->p_stat == SSLEEP) 400 setrunnable(p); 401 else 402 unsleep(p); 403 p->p_flag |= P_TIMEOUT; 404 } 405 splx(s); 406} 407 408/* 409 * Short-term, non-interruptable sleep. 410 */ 411void 412sleep(ident, priority) 413 void *ident; 414 int priority; 415{ 416 register struct proc *p = curproc; 417 register struct slpque *qp; 418 register s; 419 extern int cold; 420 421#ifdef DIAGNOSTIC 422 if (priority > PZERO) { 423 printf("sleep called with priority %d > PZERO, wchan: %x\n", 424 priority, ident); 425 panic("old sleep"); 426 } 427#endif 428 s = splhigh(); 429 if (cold || panicstr) { 430 /* 431 * After a panic, or during autoconfiguration, 432 * just give interrupts a chance, then just return; 433 * don't run any other procs or panic below, 434 * in case this is the idle process and already asleep. 435 */ 436 splx(safepri); 437 splx(s); 438 return; 439 } 440#ifdef DIAGNOSTIC 441 if (ident == NULL || p->p_stat != SRUN || p->p_back) 442 panic("sleep"); 443#endif 444 p->p_wchan = ident; 445 p->p_wmesg = NULL; 446 p->p_slptime = 0; 447 p->p_priority = priority; 448 qp = &slpque[LOOKUP(ident)]; 449 if (qp->sq_head == 0) 450 qp->sq_head = p; 451 else 452 *qp->sq_tailp = p; 453 *(qp->sq_tailp = &p->p_forw) = 0; 454 p->p_stat = SSLEEP; 455 p->p_stats->p_ru.ru_nvcsw++; 456#ifdef KTRACE 457 if (KTRPOINT(p, KTR_CSW)) 458 ktrcsw(p->p_tracep, 1, 0); 459#endif 460 mi_switch(); 461#ifdef KTRACE 462 if (KTRPOINT(p, KTR_CSW)) 463 ktrcsw(p->p_tracep, 0, 0); 464#endif 465 curpriority = p->p_usrpri; 466 splx(s); 467} 468 469/* 470 * Remove a process from its wait queue 471 */ 472void 473unsleep(p) 474 register struct proc *p; 475{ 476 register struct slpque *qp; 477 register struct proc **hp; 478 int s; 479 480 s = splhigh(); 481 if (p->p_wchan) { 482 hp = &(qp = &slpque[LOOKUP(p->p_wchan)])->sq_head; 483 while (*hp != p) 484 hp = &(*hp)->p_forw; 485 *hp = p->p_forw; 486 if (qp->sq_tailp == &p->p_forw) 487 qp->sq_tailp = hp; 488 p->p_wchan = 0; 489 } 490 splx(s); 491} 492 493/* 494 * Make all processes sleeping on the specified identifier runnable. 495 */ 496void 497wakeup(ident) 498 register void *ident; 499{ 500 register struct slpque *qp; 501 register struct proc *p, **q; 502 int s; 503 504 s = splhigh(); 505 qp = &slpque[LOOKUP(ident)]; 506restart: 507 for (q = &qp->sq_head; p = *q; ) { 508#ifdef DIAGNOSTIC 509 if (p->p_back || p->p_stat != SSLEEP && p->p_stat != SSTOP) 510 panic("wakeup"); 511#endif 512 if (p->p_wchan == ident) { 513 p->p_wchan = 0; 514 *q = p->p_forw; 515 if (qp->sq_tailp == &p->p_forw) 516 qp->sq_tailp = q; 517 if (p->p_stat == SSLEEP) { 518 /* OPTIMIZED EXPANSION OF setrunnable(p); */ 519 if (p->p_slptime > 1) 520 updatepri(p); 521 p->p_slptime = 0; 522 p->p_stat = SRUN; 523 if (p->p_flag & P_INMEM) 524 setrunqueue(p); 525 /* 526 * Since curpriority is a user priority, 527 * p->p_priority is always better than 528 * curpriority. 529 */ 530 if ((p->p_flag & P_INMEM) == 0) 531 wakeup((caddr_t)&proc0); 532 else 533 need_resched(); 534 /* END INLINE EXPANSION */ 535 goto restart; 536 } 537 } else 538 q = &p->p_forw; 539 } 540 splx(s); 541} 542 543/* 544 * The machine independent parts of mi_switch(). 545 * Must be called at splstatclock() or higher. 546 */ 547void 548mi_switch() 549{ 550 register struct proc *p = curproc; /* XXX */ 551 register struct rlimit *rlim; 552 register long s, u; 553 struct timeval tv; 554 555 /* 556 * Compute the amount of time during which the current 557 * process was running, and add that to its total so far. 558 */ 559 microtime(&tv); 560 u = p->p_rtime.tv_usec + (tv.tv_usec - runtime.tv_usec); 561 s = p->p_rtime.tv_sec + (tv.tv_sec - runtime.tv_sec); 562 if (u < 0) { 563 u += 1000000; 564 s--; 565 } else if (u >= 1000000) { 566 u -= 1000000; 567 s++; 568 } 569 p->p_rtime.tv_usec = u; 570 p->p_rtime.tv_sec = s; 571 572 /* 573 * Check if the process exceeds its cpu resource allocation. 574 * If over max, kill it. In any case, if it has run for more 575 * than 10 minutes, reduce priority to give others a chance. 576 */ 577 rlim = &p->p_rlimit[RLIMIT_CPU]; 578 if (s >= rlim->rlim_cur) { 579 if (s >= rlim->rlim_max) 580 psignal(p, SIGKILL); 581 else { 582 psignal(p, SIGXCPU); 583 if (rlim->rlim_cur < rlim->rlim_max) 584 rlim->rlim_cur += 5; 585 } 586 } 587 if (s > 10 * 60 && p->p_ucred->cr_uid && p->p_nice == NZERO) { 588 p->p_nice = NZERO + 4; 589 resetpriority(p); 590 } 591 592 /* 593 * Pick a new current process and record its start time. 594 */ 595 cnt.v_swtch++; 596 cpu_switch(p); 597 microtime(&runtime); 598} 599 600/* 601 * Initialize the (doubly-linked) run queues 602 * to be empty. 603 */ 604void 605rqinit() 606{ 607 register int i; 608 609 for (i = 0; i < NQS; i++) { 610 qs[i].ph_link = qs[i].ph_rlink = (struct proc *)&qs[i]; 611 rtqs[i].ph_link = rtqs[i].ph_rlink = (struct proc *)&rtqs[i]; 612 } 613} 614 615/* 616 * Change process state to be runnable, 617 * placing it on the run queue if it is in memory, 618 * and awakening the swapper if it isn't in memory. 619 */ 620void 621setrunnable(p) 622 register struct proc *p; 623{ 624 register int s; 625 626 s = splhigh(); 627 switch (p->p_stat) { 628 case 0: 629 case SRUN: 630 case SZOMB: 631 default: 632 panic("setrunnable"); 633 case SSTOP: 634 case SSLEEP: 635 unsleep(p); /* e.g. when sending signals */ 636 break; 637 638 case SIDL: 639 break; 640 } 641 p->p_stat = SRUN; 642 if (p->p_flag & P_INMEM) 643 setrunqueue(p); 644 splx(s); 645 if (p->p_slptime > 1) 646 updatepri(p); 647 p->p_slptime = 0; 648 if ((p->p_flag & P_INMEM) == 0) 649 wakeup((caddr_t)&proc0); 650 else if (p->p_priority < curpriority) 651 need_resched(); 652} 653 654/* 655 * Compute the priority of a process when running in user mode. 656 * Arrange to reschedule if the resulting priority is better 657 * than that of the current process. 658 */ 659void 660resetpriority(p) 661 register struct proc *p; 662{ 663 register unsigned int newpriority; 664 665 if (p->p_rtprio == RTPRIO_RTOFF) { 666 newpriority = PUSER + p->p_estcpu / 4 + 2 * p->p_nice; 667 newpriority = min(newpriority, MAXPRI); 668 p->p_usrpri = newpriority; 669 if (newpriority < curpriority) 670 need_resched(); 671 } else { 672 need_resched(); 673 } 674} 675