1 /* 2 * Copyright (c) 1982, 1986, 1990 Regents of the University of California. 3 * All rights reserved. The Berkeley software License Agreement 4 * specifies the terms and conditions for redistribution. 5 * 6 * @(#)kern_synch.c 7.14 (Berkeley) 03/17/91 7 */ 8 9 #include "param.h" 10 #include "systm.h" 11 #include "user.h" 12 #include "proc.h" 13 #include "kernel.h" 14 #include "buf.h" 15 16 #include "machine/cpu.h" 17 18 /* 19 * Force switch among equal priority processes every 100ms. 20 */ 21 roundrobin() 22 { 23 24 need_resched(); 25 timeout(roundrobin, (caddr_t)0, hz / 10); 26 } 27 28 /* 29 * constants for digital decay and forget 30 * 90% of (p_cpu) usage in 5*loadav time 31 * 95% of (p_pctcpu) usage in 60 seconds (load insensitive) 32 * Note that, as ps(1) mentions, this can let percentages 33 * total over 100% (I've seen 137.9% for 3 processes). 34 * 35 * Note that hardclock updates p_cpu and p_cpticks independently. 36 * 37 * We wish to decay away 90% of p_cpu in (5 * loadavg) seconds. 38 * That is, the system wants to compute a value of decay such 39 * that the following for loop: 40 * for (i = 0; i < (5 * loadavg); i++) 41 * p_cpu *= decay; 42 * will compute 43 * p_cpu *= 0.1; 44 * for all values of loadavg: 45 * 46 * Mathematically this loop can be expressed by saying: 47 * decay ** (5 * loadavg) ~= .1 48 * 49 * The system computes decay as: 50 * decay = (2 * loadavg) / (2 * loadavg + 1) 51 * 52 * We wish to prove that the system's computation of decay 53 * will always fulfill the equation: 54 * decay ** (5 * loadavg) ~= .1 55 * 56 * If we compute b as: 57 * b = 2 * loadavg 58 * then 59 * decay = b / (b + 1) 60 * 61 * We now need to prove two things: 62 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) 63 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) 64 * 65 * Facts: 66 * For x close to zero, exp(x) =~ 1 + x, since 67 * exp(x) = 0! + x**1/1! + x**2/2! + ... . 68 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. 69 * For x close to zero, ln(1+x) =~ x, since 70 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 71 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). 72 * ln(.1) =~ -2.30 73 * 74 * Proof of (1): 75 * Solve (factor)**(power) =~ .1 given power (5*loadav): 76 * solving for factor, 77 * ln(factor) =~ (-2.30/5*loadav), or 78 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = 79 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED 80 * 81 * Proof of (2): 82 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): 83 * solving for power, 84 * power*ln(b/(b+1)) =~ -2.30, or 85 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED 86 * 87 * Actual power values for the implemented algorithm are as follows: 88 * loadav: 1 2 3 4 89 * power: 5.68 10.32 14.94 19.55 90 */ 91 92 /* calculations for digital decay to forget 90% of usage in 5*loadav sec */ 93 #define loadfactor(loadav) (2 * (loadav)) 94 #define decay_cpu(loadfac, cpu) (((loadfac) * (cpu)) / ((loadfac) + FSCALE)) 95 96 /* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ 97 fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ 98 99 /* 100 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the 101 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below 102 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT). 103 * 104 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used: 105 * 1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits). 106 * 107 * If you dont want to bother with the faster/more-accurate formula, you 108 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate 109 * (more general) method of calculating the %age of CPU used by a process. 110 */ 111 #define CCPU_SHIFT 11 112 113 /* 114 * Recompute process priorities, once a second 115 */ 116 schedcpu() 117 { 118 register fixpt_t loadfac = loadfactor(averunnable[0]); 119 register struct proc *p; 120 register int s; 121 register unsigned int newcpu; 122 123 wakeup((caddr_t)&lbolt); 124 for (p = allproc; p != NULL; p = p->p_nxt) { 125 /* 126 * Increment time in/out of memory and sleep time 127 * (if sleeping). We ignore overflow; with 16-bit int's 128 * (remember them?) overflow takes 45 days. 129 */ 130 p->p_time++; 131 if (p->p_stat == SSLEEP || p->p_stat == SSTOP) 132 p->p_slptime++; 133 p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT; 134 /* 135 * If the process has slept the entire second, 136 * stop recalculating its priority until it wakes up. 137 */ 138 if (p->p_slptime > 1) 139 continue; 140 /* 141 * p_pctcpu is only for ps. 142 */ 143 #if (FSHIFT >= CCPU_SHIFT) 144 p->p_pctcpu += (hz == 100)? 145 ((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT): 146 100 * (((fixpt_t) p->p_cpticks) 147 << (FSHIFT - CCPU_SHIFT)) / hz; 148 #else 149 p->p_pctcpu += ((FSCALE - ccpu) * 150 (p->p_cpticks * FSCALE / hz)) >> FSHIFT; 151 #endif 152 p->p_cpticks = 0; 153 newcpu = (u_int) decay_cpu(loadfac, p->p_cpu) + p->p_nice; 154 p->p_cpu = min(newcpu, UCHAR_MAX); 155 setpri(p); 156 s = splhigh(); /* prevent state changes */ 157 if (p->p_pri >= PUSER) { 158 #define PPQ (128 / NQS) /* priorities per queue */ 159 if ((p != curproc || noproc) && 160 p->p_stat == SRUN && 161 (p->p_flag & SLOAD) && 162 (p->p_pri / PPQ) != (p->p_usrpri / PPQ)) { 163 remrq(p); 164 p->p_pri = p->p_usrpri; 165 setrq(p); 166 } else 167 p->p_pri = p->p_usrpri; 168 } 169 splx(s); 170 } 171 vmmeter(); 172 if (bclnlist != NULL) 173 wakeup((caddr_t)pageproc); 174 timeout(schedcpu, (caddr_t)0, hz); 175 } 176 177 /* 178 * Recalculate the priority of a process after it has slept for a while. 179 * For all load averages >= 1 and max p_cpu of 255, sleeping for at least 180 * six times the loadfactor will decay p_cpu to zero. 181 */ 182 updatepri(p) 183 register struct proc *p; 184 { 185 register unsigned int newcpu = p->p_cpu; 186 register fixpt_t loadfac = loadfactor(averunnable[0]); 187 188 if (p->p_slptime > 5 * loadfac) 189 p->p_cpu = 0; 190 else { 191 p->p_slptime--; /* the first time was done in schedcpu */ 192 while (newcpu && --p->p_slptime) 193 newcpu = (int) decay_cpu(loadfac, newcpu); 194 p->p_cpu = min(newcpu, UCHAR_MAX); 195 } 196 setpri(p); 197 } 198 199 #define SQSIZE 0100 /* Must be power of 2 */ 200 #define HASH(x) (( (int) x >> 5) & (SQSIZE-1)) 201 struct slpque { 202 struct proc *sq_head; 203 struct proc **sq_tailp; 204 } slpque[SQSIZE]; 205 206 /* 207 * During autoconfiguration or after a panic, a sleep will simply 208 * lower the priority briefly to allow interrupts, then return. 209 * The priority to be used (safepri) is machine-dependent, thus this 210 * value is initialized and maintained in the machine-dependent layers. 211 * This priority will typically be 0, or the lowest priority 212 * that is safe for use on the interrupt stack; it can be made 213 * higher to block network software interrupts after panics. 214 */ 215 int safepri; 216 217 /* 218 * General sleep call. 219 * Suspends current process until a wakeup is made on chan. 220 * The process will then be made runnable with priority pri. 221 * Sleeps at most timo/hz seconds (0 means no timeout). 222 * If pri includes PCATCH flag, signals are checked 223 * before and after sleeping, else signals are not checked. 224 * Returns 0 if awakened, EWOULDBLOCK if the timeout expires. 225 * If PCATCH is set and a signal needs to be delivered, 226 * ERESTART is returned if the current system call should be restarted 227 * if possible, and EINTR is returned if the system call should 228 * be interrupted by the signal (return EINTR). 229 */ 230 tsleep(chan, pri, wmesg, timo) 231 caddr_t chan; 232 int pri; 233 char *wmesg; 234 int timo; 235 { 236 register struct proc *p = curproc; /* XXX */ 237 register struct slpque *qp; 238 register s; 239 int sig, catch = pri & PCATCH; 240 extern int cold; 241 int endtsleep(); 242 243 s = splhigh(); 244 if (cold || panicstr) { 245 /* 246 * After a panic, or during autoconfiguration, 247 * just give interrupts a chance, then just return; 248 * don't run any other procs or panic below, 249 * in case this is the idle process and already asleep. 250 */ 251 splx(safepri); 252 splx(s); 253 return (0); 254 } 255 #ifdef DIAGNOSTIC 256 if (chan == 0 || p->p_stat != SRUN || p->p_rlink) 257 panic("tsleep"); 258 #endif 259 p->p_wchan = chan; 260 p->p_wmesg = wmesg; 261 p->p_slptime = 0; 262 p->p_pri = pri & PRIMASK; 263 qp = &slpque[HASH(chan)]; 264 if (qp->sq_head == 0) 265 qp->sq_head = p; 266 else 267 *qp->sq_tailp = p; 268 *(qp->sq_tailp = &p->p_link) = 0; 269 if (timo) 270 timeout(endtsleep, (caddr_t)p, timo); 271 /* 272 * We put ourselves on the sleep queue and start our timeout 273 * before calling CURSIG, as we could stop there, and a wakeup 274 * or a SIGCONT (or both) could occur while we were stopped. 275 * A SIGCONT would cause us to be marked as SSLEEP 276 * without resuming us, thus we must be ready for sleep 277 * when CURSIG is called. If the wakeup happens while we're 278 * stopped, p->p_wchan will be 0 upon return from CURSIG. 279 */ 280 if (catch) { 281 p->p_flag |= SSINTR; 282 if (sig = CURSIG(p)) { 283 if (p->p_wchan) 284 unsleep(p); 285 p->p_stat = SRUN; 286 goto resume; 287 } 288 if (p->p_wchan == 0) { 289 catch = 0; 290 goto resume; 291 } 292 } 293 p->p_stat = SSLEEP; 294 (void) spl0(); 295 p->p_stats->p_ru.ru_nvcsw++; 296 swtch(); 297 resume: 298 curpri = p->p_usrpri; 299 splx(s); 300 p->p_flag &= ~SSINTR; 301 if (p->p_flag & STIMO) { 302 p->p_flag &= ~STIMO; 303 if (catch == 0 || sig == 0) 304 return (EWOULDBLOCK); 305 } else if (timo) 306 untimeout(endtsleep, (caddr_t)p); 307 if (catch && (sig != 0 || (sig = CURSIG(p)))) { 308 if (p->p_sigacts->ps_sigintr & sigmask(sig)) 309 return (EINTR); 310 return (ERESTART); 311 } 312 return (0); 313 } 314 315 /* 316 * Implement timeout for tsleep. 317 * If process hasn't been awakened (wchan non-zero), 318 * set timeout flag and undo the sleep. If proc 319 * is stopped, just unsleep so it will remain stopped. 320 */ 321 endtsleep(p) 322 register struct proc *p; 323 { 324 int s = splhigh(); 325 326 if (p->p_wchan) { 327 if (p->p_stat == SSLEEP) 328 setrun(p); 329 else 330 unsleep(p); 331 p->p_flag |= STIMO; 332 } 333 splx(s); 334 } 335 336 /* 337 * Short-term, non-interruptable sleep. 338 */ 339 sleep(chan, pri) 340 caddr_t chan; 341 int pri; 342 { 343 register struct proc *p = curproc; /* XXX */ 344 register struct slpque *qp; 345 register s; 346 extern int cold; 347 348 #ifdef DIAGNOSTIC 349 if (pri > PZERO) { 350 printf("sleep called with pri %d > PZERO, wchan: %x\n", 351 pri, chan); 352 panic("old sleep"); 353 } 354 #endif 355 s = splhigh(); 356 if (cold || panicstr) { 357 /* 358 * After a panic, or during autoconfiguration, 359 * just give interrupts a chance, then just return; 360 * don't run any other procs or panic below, 361 * in case this is the idle process and already asleep. 362 */ 363 splx(safepri); 364 splx(s); 365 return; 366 } 367 #ifdef DIAGNOSTIC 368 if (chan==0 || p->p_stat != SRUN || p->p_rlink) 369 panic("sleep"); 370 #endif 371 p->p_wchan = chan; 372 p->p_wmesg = NULL; 373 p->p_slptime = 0; 374 p->p_pri = pri; 375 qp = &slpque[HASH(chan)]; 376 if (qp->sq_head == 0) 377 qp->sq_head = p; 378 else 379 *qp->sq_tailp = p; 380 *(qp->sq_tailp = &p->p_link) = 0; 381 p->p_stat = SSLEEP; 382 (void) spl0(); 383 p->p_stats->p_ru.ru_nvcsw++; 384 swtch(); 385 curpri = p->p_usrpri; 386 splx(s); 387 } 388 389 /* 390 * Remove a process from its wait queue 391 */ 392 unsleep(p) 393 register struct proc *p; 394 { 395 register struct slpque *qp; 396 register struct proc **hp; 397 int s; 398 399 s = splhigh(); 400 if (p->p_wchan) { 401 hp = &(qp = &slpque[HASH(p->p_wchan)])->sq_head; 402 while (*hp != p) 403 hp = &(*hp)->p_link; 404 *hp = p->p_link; 405 if (qp->sq_tailp == &p->p_link) 406 qp->sq_tailp = hp; 407 p->p_wchan = 0; 408 } 409 splx(s); 410 } 411 412 /* 413 * Wakeup on "chan"; set all processes 414 * sleeping on chan to run state. 415 */ 416 wakeup(chan) 417 register caddr_t chan; 418 { 419 register struct slpque *qp; 420 register struct proc *p, **q; 421 int s; 422 423 s = splhigh(); 424 qp = &slpque[HASH(chan)]; 425 restart: 426 for (q = &qp->sq_head; p = *q; ) { 427 #ifdef DIAGNOSTIC 428 if (p->p_rlink || p->p_stat != SSLEEP && p->p_stat != SSTOP) 429 panic("wakeup"); 430 #endif 431 if (p->p_wchan == chan) { 432 p->p_wchan = 0; 433 *q = p->p_link; 434 if (qp->sq_tailp == &p->p_link) 435 qp->sq_tailp = q; 436 if (p->p_stat == SSLEEP) { 437 /* OPTIMIZED INLINE EXPANSION OF setrun(p) */ 438 if (p->p_slptime > 1) 439 updatepri(p); 440 p->p_slptime = 0; 441 p->p_stat = SRUN; 442 if (p->p_flag & SLOAD) 443 setrq(p); 444 /* 445 * Since curpri is a usrpri, 446 * p->p_pri is always better than curpri. 447 */ 448 if ((p->p_flag&SLOAD) == 0) 449 wakeup((caddr_t)&proc0); 450 else 451 need_resched(); 452 /* END INLINE EXPANSION */ 453 goto restart; 454 } 455 } else 456 q = &p->p_link; 457 } 458 splx(s); 459 } 460 461 /* 462 * Initialize the (doubly-linked) run queues 463 * to be empty. 464 */ 465 rqinit() 466 { 467 register int i; 468 469 for (i = 0; i < NQS; i++) 470 qs[i].ph_link = qs[i].ph_rlink = (struct proc *)&qs[i]; 471 } 472 473 /* 474 * Change process state to be runnable, 475 * placing it on the run queue if it is in memory, 476 * and awakening the swapper if it isn't in memory. 477 */ 478 setrun(p) 479 register struct proc *p; 480 { 481 register int s; 482 483 s = splhigh(); 484 switch (p->p_stat) { 485 486 case 0: 487 case SWAIT: 488 case SRUN: 489 case SZOMB: 490 default: 491 panic("setrun"); 492 493 case SSTOP: 494 case SSLEEP: 495 unsleep(p); /* e.g. when sending signals */ 496 break; 497 498 case SIDL: 499 break; 500 } 501 p->p_stat = SRUN; 502 if (p->p_flag & SLOAD) 503 setrq(p); 504 splx(s); 505 if (p->p_slptime > 1) 506 updatepri(p); 507 p->p_slptime = 0; 508 if ((p->p_flag&SLOAD) == 0) 509 wakeup((caddr_t)&proc0); 510 else if (p->p_pri < curpri) 511 need_resched(); 512 } 513 514 /* 515 * Compute priority of process when running in user mode. 516 * Arrange to reschedule if the resulting priority 517 * is better than that of the current process. 518 */ 519 setpri(p) 520 register struct proc *p; 521 { 522 register unsigned int newpri; 523 524 newpri = PUSER + p->p_cpu / 4 + 2 * p->p_nice; 525 newpri = min(newpri, MAXPRI); 526 p->p_usrpri = newpri; 527 if (newpri < curpri) 528 need_resched(); 529 } 530