1 /* $NetBSD: sched_4bsd.c,v 1.28 2011/12/02 12:29:35 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2004, 2006, 2007, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, by Charles M. Hannum, Andrew Doran, and 10 * Daniel Sieger. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 23 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 /*- 35 * Copyright (c) 1982, 1986, 1990, 1991, 1993 36 * The Regents of the University of California. All rights reserved. 37 * (c) UNIX System Laboratories, Inc. 38 * All or some portions of this file are derived from material licensed 39 * to the University of California by American Telephone and Telegraph 40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 41 * the permission of UNIX System Laboratories, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)kern_synch.c 8.9 (Berkeley) 5/19/95 68 */ 69 70 #include <sys/cdefs.h> 71 __KERNEL_RCSID(0, "$NetBSD: sched_4bsd.c,v 1.28 2011/12/02 12:29:35 yamt Exp $"); 72 73 #include "opt_ddb.h" 74 #include "opt_lockdebug.h" 75 #include "opt_perfctrs.h" 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/callout.h> 80 #include <sys/cpu.h> 81 #include <sys/proc.h> 82 #include <sys/kernel.h> 83 #include <sys/signalvar.h> 84 #include <sys/resourcevar.h> 85 #include <sys/sched.h> 86 #include <sys/sysctl.h> 87 #include <sys/kauth.h> 88 #include <sys/lockdebug.h> 89 #include <sys/kmem.h> 90 #include <sys/intr.h> 91 92 static void updatepri(struct lwp *); 93 static void resetpriority(struct lwp *); 94 95 extern unsigned int sched_pstats_ticks; /* defined in kern_synch.c */ 96 97 /* Number of hardclock ticks per sched_tick() */ 98 static int rrticks; 99 100 /* 101 * Force switch among equal priority processes every 100ms. 102 * Called from hardclock every hz/10 == rrticks hardclock ticks. 103 * 104 * There's no need to lock anywhere in this routine, as it's 105 * CPU-local and runs at IPL_SCHED (called from clock interrupt). 106 */ 107 /* ARGSUSED */ 108 void 109 sched_tick(struct cpu_info *ci) 110 { 111 struct schedstate_percpu *spc = &ci->ci_schedstate; 112 lwp_t *l; 113 114 spc->spc_ticks = rrticks; 115 116 if (CURCPU_IDLE_P()) { 117 cpu_need_resched(ci, 0); 118 return; 119 } 120 l = ci->ci_data.cpu_onproc; 121 if (l == NULL) { 122 return; 123 } 124 switch (l->l_class) { 125 case SCHED_FIFO: 126 /* No timeslicing for FIFO jobs. */ 127 break; 128 case SCHED_RR: 129 /* Force it into mi_switch() to look for other jobs to run. */ 130 cpu_need_resched(ci, RESCHED_KPREEMPT); 131 break; 132 default: 133 if (spc->spc_flags & SPCF_SHOULDYIELD) { 134 /* 135 * Process is stuck in kernel somewhere, probably 136 * due to buggy or inefficient code. Force a 137 * kernel preemption. 138 */ 139 cpu_need_resched(ci, RESCHED_KPREEMPT); 140 } else if (spc->spc_flags & SPCF_SEENRR) { 141 /* 142 * The process has already been through a roundrobin 143 * without switching and may be hogging the CPU. 144 * Indicate that the process should yield. 145 */ 146 spc->spc_flags |= SPCF_SHOULDYIELD; 147 cpu_need_resched(ci, 0); 148 } else { 149 spc->spc_flags |= SPCF_SEENRR; 150 } 151 break; 152 } 153 } 154 155 /* 156 * Why PRIO_MAX - 2? From setpriority(2): 157 * 158 * prio is a value in the range -20 to 20. The default priority is 159 * 0; lower priorities cause more favorable scheduling. A value of 160 * 19 or 20 will schedule a process only when nothing at priority <= 161 * 0 is runnable. 162 * 163 * This gives estcpu influence over 18 priority levels, and leaves nice 164 * with 40 levels. One way to think about it is that nice has 20 levels 165 * either side of estcpu's 18. 166 */ 167 #define ESTCPU_SHIFT 11 168 #define ESTCPU_MAX ((PRIO_MAX - 2) << ESTCPU_SHIFT) 169 #define ESTCPU_ACCUM (1 << (ESTCPU_SHIFT - 1)) 170 #define ESTCPULIM(e) min((e), ESTCPU_MAX) 171 172 /* 173 * Constants for digital decay and forget: 174 * 90% of (l_estcpu) usage in 5 * loadav time 175 * 95% of (l_pctcpu) usage in 60 seconds (load insensitive) 176 * Note that, as ps(1) mentions, this can let percentages 177 * total over 100% (I've seen 137.9% for 3 processes). 178 * 179 * Note that hardclock updates l_estcpu and l_cpticks independently. 180 * 181 * We wish to decay away 90% of l_estcpu in (5 * loadavg) seconds. 182 * That is, the system wants to compute a value of decay such 183 * that the following for loop: 184 * for (i = 0; i < (5 * loadavg); i++) 185 * l_estcpu *= decay; 186 * will compute 187 * l_estcpu *= 0.1; 188 * for all values of loadavg: 189 * 190 * Mathematically this loop can be expressed by saying: 191 * decay ** (5 * loadavg) ~= .1 192 * 193 * The system computes decay as: 194 * decay = (2 * loadavg) / (2 * loadavg + 1) 195 * 196 * We wish to prove that the system's computation of decay 197 * will always fulfill the equation: 198 * decay ** (5 * loadavg) ~= .1 199 * 200 * If we compute b as: 201 * b = 2 * loadavg 202 * then 203 * decay = b / (b + 1) 204 * 205 * We now need to prove two things: 206 * 1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1) 207 * 2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg) 208 * 209 * Facts: 210 * For x close to zero, exp(x) =~ 1 + x, since 211 * exp(x) = 0! + x**1/1! + x**2/2! + ... . 212 * therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b. 213 * For x close to zero, ln(1+x) =~ x, since 214 * ln(1+x) = x - x**2/2 + x**3/3 - ... -1 < x < 1 215 * therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1). 216 * ln(.1) =~ -2.30 217 * 218 * Proof of (1): 219 * Solve (factor)**(power) =~ .1 given power (5*loadav): 220 * solving for factor, 221 * ln(factor) =~ (-2.30/5*loadav), or 222 * factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) = 223 * exp(-1/b) =~ (b-1)/b =~ b/(b+1). QED 224 * 225 * Proof of (2): 226 * Solve (factor)**(power) =~ .1 given factor == (b/(b+1)): 227 * solving for power, 228 * power*ln(b/(b+1)) =~ -2.30, or 229 * power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav. QED 230 * 231 * Actual power values for the implemented algorithm are as follows: 232 * loadav: 1 2 3 4 233 * power: 5.68 10.32 14.94 19.55 234 */ 235 236 /* calculations for digital decay to forget 90% of usage in 5*loadav sec */ 237 #define loadfactor(loadav) (2 * (loadav) / ncpu) 238 239 static fixpt_t 240 decay_cpu(fixpt_t loadfac, fixpt_t estcpu) 241 { 242 243 if (estcpu == 0) { 244 return 0; 245 } 246 247 #if !defined(_LP64) 248 /* avoid 64bit arithmetics. */ 249 #define FIXPT_MAX ((fixpt_t)((UINTMAX_C(1) << sizeof(fixpt_t) * CHAR_BIT) - 1)) 250 if (__predict_true(loadfac <= FIXPT_MAX / ESTCPU_MAX)) { 251 return estcpu * loadfac / (loadfac + FSCALE); 252 } 253 #endif /* !defined(_LP64) */ 254 255 return (uint64_t)estcpu * loadfac / (loadfac + FSCALE); 256 } 257 258 /* 259 * For all load averages >= 1 and max l_estcpu of (255 << ESTCPU_SHIFT), 260 * sleeping for at least seven times the loadfactor will decay l_estcpu to 261 * less than (1 << ESTCPU_SHIFT). 262 * 263 * note that our ESTCPU_MAX is actually much smaller than (255 << ESTCPU_SHIFT). 264 */ 265 static fixpt_t 266 decay_cpu_batch(fixpt_t loadfac, fixpt_t estcpu, unsigned int n) 267 { 268 269 if ((n << FSHIFT) >= 7 * loadfac) { 270 return 0; 271 } 272 273 while (estcpu != 0 && n > 1) { 274 estcpu = decay_cpu(loadfac, estcpu); 275 n--; 276 } 277 278 return estcpu; 279 } 280 281 /* 282 * sched_pstats_hook: 283 * 284 * Periodically called from sched_pstats(); used to recalculate priorities. 285 */ 286 void 287 sched_pstats_hook(struct lwp *l, int batch) 288 { 289 fixpt_t loadfac; 290 291 /* 292 * If the LWP has slept an entire second, stop recalculating 293 * its priority until it wakes up. 294 */ 295 KASSERT(lwp_locked(l, NULL)); 296 if (l->l_stat == LSSLEEP || l->l_stat == LSSTOP || 297 l->l_stat == LSSUSPENDED) { 298 if (l->l_slptime > 1) { 299 return; 300 } 301 } 302 loadfac = 2 * (averunnable.ldavg[0]); 303 l->l_estcpu = decay_cpu(loadfac, l->l_estcpu); 304 resetpriority(l); 305 } 306 307 /* 308 * Recalculate the priority of a process after it has slept for a while. 309 */ 310 static void 311 updatepri(struct lwp *l) 312 { 313 fixpt_t loadfac; 314 315 KASSERT(lwp_locked(l, NULL)); 316 KASSERT(l->l_slptime > 1); 317 318 loadfac = loadfactor(averunnable.ldavg[0]); 319 320 l->l_slptime--; /* the first time was done in sched_pstats */ 321 l->l_estcpu = decay_cpu_batch(loadfac, l->l_estcpu, l->l_slptime); 322 resetpriority(l); 323 } 324 325 void 326 sched_rqinit(void) 327 { 328 329 } 330 331 void 332 sched_setrunnable(struct lwp *l) 333 { 334 335 if (l->l_slptime > 1) 336 updatepri(l); 337 } 338 339 void 340 sched_nice(struct proc *p, int n) 341 { 342 struct lwp *l; 343 344 KASSERT(mutex_owned(p->p_lock)); 345 346 p->p_nice = n; 347 LIST_FOREACH(l, &p->p_lwps, l_sibling) { 348 lwp_lock(l); 349 resetpriority(l); 350 lwp_unlock(l); 351 } 352 } 353 354 /* 355 * Recompute the priority of an LWP. Arrange to reschedule if 356 * the resulting priority is better than that of the current LWP. 357 */ 358 static void 359 resetpriority(struct lwp *l) 360 { 361 pri_t pri; 362 struct proc *p = l->l_proc; 363 364 KASSERT(lwp_locked(l, NULL)); 365 366 if (l->l_class != SCHED_OTHER) 367 return; 368 369 /* See comments above ESTCPU_SHIFT definition. */ 370 pri = (PRI_KERNEL - 1) - (l->l_estcpu >> ESTCPU_SHIFT) - p->p_nice; 371 pri = imax(pri, 0); 372 if (pri != l->l_priority) 373 lwp_changepri(l, pri); 374 } 375 376 /* 377 * We adjust the priority of the current LWP. The priority of a LWP 378 * gets worse as it accumulates CPU time. The CPU usage estimator (l_estcpu) 379 * is increased here. The formula for computing priorities will compute a 380 * different value each time l_estcpu increases. This can cause a switch, 381 * but unless the priority crosses a PPQ boundary the actual queue will not 382 * change. The CPU usage estimator ramps up quite quickly when the process 383 * is running (linearly), and decays away exponentially, at a rate which is 384 * proportionally slower when the system is busy. The basic principle is 385 * that the system will 90% forget that the process used a lot of CPU time 386 * in 5 * loadav seconds. This causes the system to favor processes which 387 * haven't run much recently, and to round-robin among other processes. 388 */ 389 390 void 391 sched_schedclock(struct lwp *l) 392 { 393 394 if (l->l_class != SCHED_OTHER) 395 return; 396 397 KASSERT(!CURCPU_IDLE_P()); 398 l->l_estcpu = ESTCPULIM(l->l_estcpu + ESTCPU_ACCUM); 399 lwp_lock(l); 400 resetpriority(l); 401 lwp_unlock(l); 402 } 403 404 /* 405 * sched_proc_fork: 406 * 407 * Inherit the parent's scheduler history. 408 */ 409 void 410 sched_proc_fork(struct proc *parent, struct proc *child) 411 { 412 lwp_t *pl; 413 414 KASSERT(mutex_owned(parent->p_lock)); 415 416 pl = LIST_FIRST(&parent->p_lwps); 417 child->p_estcpu_inherited = pl->l_estcpu; 418 child->p_forktime = sched_pstats_ticks; 419 } 420 421 /* 422 * sched_proc_exit: 423 * 424 * Chargeback parents for the sins of their children. 425 */ 426 void 427 sched_proc_exit(struct proc *parent, struct proc *child) 428 { 429 fixpt_t loadfac = loadfactor(averunnable.ldavg[0]); 430 fixpt_t estcpu; 431 lwp_t *pl, *cl; 432 433 /* XXX Only if parent != init?? */ 434 435 mutex_enter(parent->p_lock); 436 pl = LIST_FIRST(&parent->p_lwps); 437 cl = LIST_FIRST(&child->p_lwps); 438 estcpu = decay_cpu_batch(loadfac, child->p_estcpu_inherited, 439 sched_pstats_ticks - child->p_forktime); 440 if (cl->l_estcpu > estcpu) { 441 lwp_lock(pl); 442 pl->l_estcpu = ESTCPULIM(pl->l_estcpu + cl->l_estcpu - estcpu); 443 lwp_unlock(pl); 444 } 445 mutex_exit(parent->p_lock); 446 } 447 448 void 449 sched_wakeup(struct lwp *l) 450 { 451 452 } 453 454 void 455 sched_slept(struct lwp *l) 456 { 457 458 } 459 460 void 461 sched_lwp_fork(struct lwp *l1, struct lwp *l2) 462 { 463 464 l2->l_estcpu = l1->l_estcpu; 465 } 466 467 void 468 sched_lwp_collect(struct lwp *t) 469 { 470 lwp_t *l; 471 472 /* Absorb estcpu value of collected LWP. */ 473 l = curlwp; 474 lwp_lock(l); 475 l->l_estcpu += t->l_estcpu; 476 lwp_unlock(l); 477 } 478 479 void 480 sched_oncpu(lwp_t *l) 481 { 482 483 } 484 485 void 486 sched_newts(lwp_t *l) 487 { 488 489 } 490 491 /* 492 * Sysctl nodes and initialization. 493 */ 494 495 static int 496 sysctl_sched_rtts(SYSCTLFN_ARGS) 497 { 498 struct sysctlnode node; 499 int rttsms = hztoms(rrticks); 500 501 node = *rnode; 502 node.sysctl_data = &rttsms; 503 return sysctl_lookup(SYSCTLFN_CALL(&node)); 504 } 505 506 SYSCTL_SETUP(sysctl_sched_4bsd_setup, "sysctl sched setup") 507 { 508 const struct sysctlnode *node = NULL; 509 510 sysctl_createv(clog, 0, NULL, NULL, 511 CTLFLAG_PERMANENT, 512 CTLTYPE_NODE, "kern", NULL, 513 NULL, 0, NULL, 0, 514 CTL_KERN, CTL_EOL); 515 sysctl_createv(clog, 0, NULL, &node, 516 CTLFLAG_PERMANENT, 517 CTLTYPE_NODE, "sched", 518 SYSCTL_DESCR("Scheduler options"), 519 NULL, 0, NULL, 0, 520 CTL_KERN, CTL_CREATE, CTL_EOL); 521 522 if (node == NULL) 523 return; 524 525 rrticks = hz / 10; 526 527 sysctl_createv(NULL, 0, &node, NULL, 528 CTLFLAG_PERMANENT, 529 CTLTYPE_STRING, "name", NULL, 530 NULL, 0, __UNCONST("4.4BSD"), 0, 531 CTL_CREATE, CTL_EOL); 532 sysctl_createv(NULL, 0, &node, NULL, 533 CTLFLAG_PERMANENT, 534 CTLTYPE_INT, "rtts", 535 SYSCTL_DESCR("Round-robin time quantum (in miliseconds)"), 536 sysctl_sched_rtts, 0, NULL, 0, 537 CTL_CREATE, CTL_EOL); 538 } 539