1 /*- 2 * Copyright (c) 2002-2003, Jeffrey Roberson <jeff@freebsd.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/ktr.h> 34 #include <sys/lock.h> 35 #include <sys/mutex.h> 36 #include <sys/proc.h> 37 #include <sys/resource.h> 38 #include <sys/sched.h> 39 #include <sys/smp.h> 40 #include <sys/sx.h> 41 #include <sys/sysctl.h> 42 #include <sys/sysproto.h> 43 #include <sys/vmmeter.h> 44 #ifdef DDB 45 #include <ddb/ddb.h> 46 #endif 47 #ifdef KTRACE 48 #include <sys/uio.h> 49 #include <sys/ktrace.h> 50 #endif 51 52 #include <machine/cpu.h> 53 54 #define KTR_ULE KTR_NFS 55 56 /* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */ 57 /* XXX This is bogus compatability crap for ps */ 58 static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */ 59 SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, ""); 60 61 static void sched_setup(void *dummy); 62 SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL) 63 64 static SYSCTL_NODE(_kern, OID_AUTO, sched, CTLFLAG_RW, 0, "SCHED"); 65 66 static int sched_strict; 67 SYSCTL_INT(_kern_sched, OID_AUTO, strict, CTLFLAG_RD, &sched_strict, 0, ""); 68 69 static int slice_min = 1; 70 SYSCTL_INT(_kern_sched, OID_AUTO, slice_min, CTLFLAG_RW, &slice_min, 0, ""); 71 72 static int slice_max = 10; 73 SYSCTL_INT(_kern_sched, OID_AUTO, slice_max, CTLFLAG_RW, &slice_max, 0, ""); 74 75 int realstathz; 76 int tickincr = 1; 77 78 #ifdef SMP 79 /* Callout to handle load balancing SMP systems. */ 80 static struct callout kseq_lb_callout; 81 #endif 82 83 /* 84 * These datastructures are allocated within their parent datastructure but 85 * are scheduler specific. 86 */ 87 88 struct ke_sched { 89 int ske_slice; 90 struct runq *ske_runq; 91 /* The following variables are only used for pctcpu calculation */ 92 int ske_ltick; /* Last tick that we were running on */ 93 int ske_ftick; /* First tick that we were running on */ 94 int ske_ticks; /* Tick count */ 95 /* CPU that we have affinity for. */ 96 u_char ske_cpu; 97 }; 98 #define ke_slice ke_sched->ske_slice 99 #define ke_runq ke_sched->ske_runq 100 #define ke_ltick ke_sched->ske_ltick 101 #define ke_ftick ke_sched->ske_ftick 102 #define ke_ticks ke_sched->ske_ticks 103 #define ke_cpu ke_sched->ske_cpu 104 105 struct kg_sched { 106 int skg_slptime; /* Number of ticks we vol. slept */ 107 int skg_runtime; /* Number of ticks we were running */ 108 }; 109 #define kg_slptime kg_sched->skg_slptime 110 #define kg_runtime kg_sched->skg_runtime 111 112 struct td_sched { 113 int std_slptime; 114 }; 115 #define td_slptime td_sched->std_slptime 116 117 struct td_sched td_sched; 118 struct ke_sched ke_sched; 119 struct kg_sched kg_sched; 120 121 struct ke_sched *kse0_sched = &ke_sched; 122 struct kg_sched *ksegrp0_sched = &kg_sched; 123 struct p_sched *proc0_sched = NULL; 124 struct td_sched *thread0_sched = &td_sched; 125 126 /* 127 * This priority range has 20 priorities on either end that are reachable 128 * only through nice values. 129 * 130 * PRI_RANGE: Total priority range for timeshare threads. 131 * PRI_NRESV: Reserved priorities for nice. 132 * PRI_BASE: The start of the dynamic range. 133 * DYN_RANGE: Number of priorities that are available int the dynamic 134 * priority range. 135 */ 136 #define SCHED_PRI_RANGE (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE + 1) 137 #define SCHED_PRI_NRESV PRIO_TOTAL 138 #define SCHED_PRI_NHALF (PRIO_TOTAL / 2) 139 #define SCHED_PRI_NTHRESH (SCHED_PRI_NHALF - 1) 140 #define SCHED_PRI_BASE ((SCHED_PRI_NRESV / 2) + PRI_MIN_TIMESHARE) 141 #define SCHED_DYN_RANGE (SCHED_PRI_RANGE - SCHED_PRI_NRESV) 142 #define SCHED_PRI_INTERACT(score) \ 143 ((score) * SCHED_DYN_RANGE / SCHED_INTERACT_MAX) 144 145 /* 146 * These determine the interactivity of a process. 147 * 148 * SLP_RUN_MAX: Maximum amount of sleep time + run time we'll accumulate 149 * before throttling back. 150 * SLP_RUN_THROTTLE: Divisor for reducing slp/run time. 151 * INTERACT_MAX: Maximum interactivity value. Smaller is better. 152 * INTERACT_THRESH: Threshhold for placement on the current runq. 153 */ 154 #define SCHED_SLP_RUN_MAX ((hz * 2) << 10) 155 #define SCHED_SLP_RUN_THROTTLE (2) 156 #define SCHED_INTERACT_MAX (100) 157 #define SCHED_INTERACT_HALF (SCHED_INTERACT_MAX / 2) 158 #define SCHED_INTERACT_THRESH (20) 159 160 /* 161 * These parameters and macros determine the size of the time slice that is 162 * granted to each thread. 163 * 164 * SLICE_MIN: Minimum time slice granted, in units of ticks. 165 * SLICE_MAX: Maximum time slice granted. 166 * SLICE_RANGE: Range of available time slices scaled by hz. 167 * SLICE_SCALE: The number slices granted per val in the range of [0, max]. 168 * SLICE_NICE: Determine the amount of slice granted to a scaled nice. 169 */ 170 #define SCHED_SLICE_MIN (slice_min) 171 #define SCHED_SLICE_MAX (slice_max) 172 #define SCHED_SLICE_RANGE (SCHED_SLICE_MAX - SCHED_SLICE_MIN + 1) 173 #define SCHED_SLICE_SCALE(val, max) (((val) * SCHED_SLICE_RANGE) / (max)) 174 #define SCHED_SLICE_NICE(nice) \ 175 (SCHED_SLICE_MAX - SCHED_SLICE_SCALE((nice), SCHED_PRI_NTHRESH)) 176 177 /* 178 * This macro determines whether or not the kse belongs on the current or 179 * next run queue. 180 * 181 * XXX nice value should effect how interactive a kg is. 182 */ 183 #define SCHED_INTERACTIVE(kg) \ 184 (sched_interact_score(kg) < SCHED_INTERACT_THRESH) 185 #define SCHED_CURR(kg, ke) \ 186 (ke->ke_thread->td_priority < PRI_MIN_TIMESHARE || SCHED_INTERACTIVE(kg)) 187 188 /* 189 * Cpu percentage computation macros and defines. 190 * 191 * SCHED_CPU_TIME: Number of seconds to average the cpu usage across. 192 * SCHED_CPU_TICKS: Number of hz ticks to average the cpu usage across. 193 */ 194 195 #define SCHED_CPU_TIME 10 196 #define SCHED_CPU_TICKS (hz * SCHED_CPU_TIME) 197 198 /* 199 * kseq - per processor runqs and statistics. 200 */ 201 202 #define KSEQ_NCLASS (PRI_IDLE + 1) /* Number of run classes. */ 203 204 struct kseq { 205 struct runq ksq_idle; /* Queue of IDLE threads. */ 206 struct runq ksq_timeshare[2]; /* Run queues for !IDLE. */ 207 struct runq *ksq_next; /* Next timeshare queue. */ 208 struct runq *ksq_curr; /* Current queue. */ 209 int ksq_loads[KSEQ_NCLASS]; /* Load for each class */ 210 int ksq_load; /* Aggregate load. */ 211 short ksq_nice[PRIO_TOTAL + 1]; /* KSEs in each nice bin. */ 212 short ksq_nicemin; /* Least nice. */ 213 #ifdef SMP 214 unsigned int ksq_rslices; /* Slices on run queue */ 215 #endif 216 }; 217 218 /* 219 * One kse queue per processor. 220 */ 221 #ifdef SMP 222 struct kseq kseq_cpu[MAXCPU]; 223 #define KSEQ_SELF() (&kseq_cpu[PCPU_GET(cpuid)]) 224 #define KSEQ_CPU(x) (&kseq_cpu[(x)]) 225 #else 226 struct kseq kseq_cpu; 227 #define KSEQ_SELF() (&kseq_cpu) 228 #define KSEQ_CPU(x) (&kseq_cpu) 229 #endif 230 231 static void sched_slice(struct kse *ke); 232 static void sched_priority(struct ksegrp *kg); 233 static int sched_interact_score(struct ksegrp *kg); 234 static void sched_interact_update(struct ksegrp *kg); 235 void sched_pctcpu_update(struct kse *ke); 236 int sched_pickcpu(void); 237 238 /* Operations on per processor queues */ 239 static struct kse * kseq_choose(struct kseq *kseq); 240 static void kseq_setup(struct kseq *kseq); 241 static void kseq_add(struct kseq *kseq, struct kse *ke); 242 static void kseq_rem(struct kseq *kseq, struct kse *ke); 243 static void kseq_nice_add(struct kseq *kseq, int nice); 244 static void kseq_nice_rem(struct kseq *kseq, int nice); 245 void kseq_print(int cpu); 246 #ifdef SMP 247 struct kseq * kseq_load_highest(void); 248 void kseq_balance(void *arg); 249 void kseq_move(struct kseq *from, int cpu); 250 #endif 251 252 void 253 kseq_print(int cpu) 254 { 255 struct kseq *kseq; 256 int i; 257 258 kseq = KSEQ_CPU(cpu); 259 260 printf("kseq:\n"); 261 printf("\tload: %d\n", kseq->ksq_load); 262 printf("\tload ITHD: %d\n", kseq->ksq_loads[PRI_ITHD]); 263 printf("\tload REALTIME: %d\n", kseq->ksq_loads[PRI_REALTIME]); 264 printf("\tload TIMESHARE: %d\n", kseq->ksq_loads[PRI_TIMESHARE]); 265 printf("\tload IDLE: %d\n", kseq->ksq_loads[PRI_IDLE]); 266 printf("\tnicemin:\t%d\n", kseq->ksq_nicemin); 267 printf("\tnice counts:\n"); 268 for (i = 0; i < PRIO_TOTAL + 1; i++) 269 if (kseq->ksq_nice[i]) 270 printf("\t\t%d = %d\n", 271 i - SCHED_PRI_NHALF, kseq->ksq_nice[i]); 272 } 273 274 static void 275 kseq_add(struct kseq *kseq, struct kse *ke) 276 { 277 mtx_assert(&sched_lock, MA_OWNED); 278 kseq->ksq_loads[PRI_BASE(ke->ke_ksegrp->kg_pri_class)]++; 279 kseq->ksq_load++; 280 if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) 281 CTR6(KTR_ULE, "Add kse %p to %p (slice: %d, pri: %d, nice: %d(%d))", 282 ke, ke->ke_runq, ke->ke_slice, ke->ke_thread->td_priority, 283 ke->ke_ksegrp->kg_nice, kseq->ksq_nicemin); 284 if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) 285 kseq_nice_add(kseq, ke->ke_ksegrp->kg_nice); 286 #ifdef SMP 287 kseq->ksq_rslices += ke->ke_slice; 288 #endif 289 } 290 291 static void 292 kseq_rem(struct kseq *kseq, struct kse *ke) 293 { 294 mtx_assert(&sched_lock, MA_OWNED); 295 kseq->ksq_loads[PRI_BASE(ke->ke_ksegrp->kg_pri_class)]--; 296 kseq->ksq_load--; 297 ke->ke_runq = NULL; 298 if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) 299 kseq_nice_rem(kseq, ke->ke_ksegrp->kg_nice); 300 #ifdef SMP 301 kseq->ksq_rslices -= ke->ke_slice; 302 #endif 303 } 304 305 static void 306 kseq_nice_add(struct kseq *kseq, int nice) 307 { 308 mtx_assert(&sched_lock, MA_OWNED); 309 /* Normalize to zero. */ 310 kseq->ksq_nice[nice + SCHED_PRI_NHALF]++; 311 if (nice < kseq->ksq_nicemin || kseq->ksq_loads[PRI_TIMESHARE] == 1) 312 kseq->ksq_nicemin = nice; 313 } 314 315 static void 316 kseq_nice_rem(struct kseq *kseq, int nice) 317 { 318 int n; 319 320 mtx_assert(&sched_lock, MA_OWNED); 321 /* Normalize to zero. */ 322 n = nice + SCHED_PRI_NHALF; 323 kseq->ksq_nice[n]--; 324 KASSERT(kseq->ksq_nice[n] >= 0, ("Negative nice count.")); 325 326 /* 327 * If this wasn't the smallest nice value or there are more in 328 * this bucket we can just return. Otherwise we have to recalculate 329 * the smallest nice. 330 */ 331 if (nice != kseq->ksq_nicemin || 332 kseq->ksq_nice[n] != 0 || 333 kseq->ksq_loads[PRI_TIMESHARE] == 0) 334 return; 335 336 for (; n < SCHED_PRI_NRESV + 1; n++) 337 if (kseq->ksq_nice[n]) { 338 kseq->ksq_nicemin = n - SCHED_PRI_NHALF; 339 return; 340 } 341 } 342 343 #ifdef SMP 344 /* 345 * kseq_balance is a simple CPU load balancing algorithm. It operates by 346 * finding the least loaded and most loaded cpu and equalizing their load 347 * by migrating some processes. 348 * 349 * Dealing only with two CPUs at a time has two advantages. Firstly, most 350 * installations will only have 2 cpus. Secondly, load balancing too much at 351 * once can have an unpleasant effect on the system. The scheduler rarely has 352 * enough information to make perfect decisions. So this algorithm chooses 353 * algorithm simplicity and more gradual effects on load in larger systems. 354 * 355 * It could be improved by considering the priorities and slices assigned to 356 * each task prior to balancing them. There are many pathological cases with 357 * any approach and so the semi random algorithm below may work as well as any. 358 * 359 */ 360 void 361 kseq_balance(void *arg) 362 { 363 struct kseq *kseq; 364 int high_load; 365 int low_load; 366 int high_cpu; 367 int low_cpu; 368 int move; 369 int diff; 370 int i; 371 372 high_cpu = 0; 373 low_cpu = 0; 374 high_load = 0; 375 low_load = -1; 376 377 mtx_lock_spin(&sched_lock); 378 for (i = 0; i < mp_maxid; i++) { 379 if (CPU_ABSENT(i)) 380 continue; 381 kseq = KSEQ_CPU(i); 382 if (kseq->ksq_load > high_load) { 383 high_load = kseq->ksq_load; 384 high_cpu = i; 385 } 386 if (low_load == -1 || kseq->ksq_load < low_load) { 387 low_load = kseq->ksq_load; 388 low_cpu = i; 389 } 390 } 391 392 /* 393 * Nothing to do. 394 */ 395 if (high_load < 2 || low_load == high_load) 396 goto out; 397 398 diff = high_load - low_load; 399 move = diff / 2; 400 if (diff & 0x1) 401 move++; 402 403 for (i = 0; i < move; i++) 404 kseq_move(KSEQ_CPU(high_cpu), low_cpu); 405 406 out: 407 mtx_unlock_spin(&sched_lock); 408 callout_reset(&kseq_lb_callout, hz, kseq_balance, NULL); 409 410 return; 411 } 412 413 struct kseq * 414 kseq_load_highest(void) 415 { 416 struct kseq *kseq; 417 int load; 418 int cpu; 419 int i; 420 421 mtx_assert(&sched_lock, MA_OWNED); 422 cpu = 0; 423 load = 0; 424 425 for (i = 0; i < mp_maxid; i++) { 426 if (CPU_ABSENT(i)) 427 continue; 428 kseq = KSEQ_CPU(i); 429 if (kseq->ksq_load > load) { 430 load = kseq->ksq_load; 431 cpu = i; 432 } 433 } 434 if (load > 1) 435 return (KSEQ_CPU(cpu)); 436 437 return (NULL); 438 } 439 440 void 441 kseq_move(struct kseq *from, int cpu) 442 { 443 struct kse *ke; 444 445 ke = kseq_choose(from); 446 runq_remove(ke->ke_runq, ke); 447 ke->ke_state = KES_THREAD; 448 kseq_rem(from, ke); 449 450 ke->ke_cpu = cpu; 451 sched_add(ke); 452 } 453 #endif 454 455 struct kse * 456 kseq_choose(struct kseq *kseq) 457 { 458 struct kse *ke; 459 struct runq *swap; 460 461 mtx_assert(&sched_lock, MA_OWNED); 462 swap = NULL; 463 464 for (;;) { 465 ke = runq_choose(kseq->ksq_curr); 466 if (ke == NULL) { 467 /* 468 * We already swaped once and didn't get anywhere. 469 */ 470 if (swap) 471 break; 472 swap = kseq->ksq_curr; 473 kseq->ksq_curr = kseq->ksq_next; 474 kseq->ksq_next = swap; 475 continue; 476 } 477 /* 478 * If we encounter a slice of 0 the kse is in a 479 * TIMESHARE kse group and its nice was too far out 480 * of the range that receives slices. 481 */ 482 if (ke->ke_slice == 0) { 483 runq_remove(ke->ke_runq, ke); 484 sched_slice(ke); 485 ke->ke_runq = kseq->ksq_next; 486 runq_add(ke->ke_runq, ke); 487 continue; 488 } 489 return (ke); 490 } 491 492 return (runq_choose(&kseq->ksq_idle)); 493 } 494 495 static void 496 kseq_setup(struct kseq *kseq) 497 { 498 runq_init(&kseq->ksq_timeshare[0]); 499 runq_init(&kseq->ksq_timeshare[1]); 500 runq_init(&kseq->ksq_idle); 501 502 kseq->ksq_curr = &kseq->ksq_timeshare[0]; 503 kseq->ksq_next = &kseq->ksq_timeshare[1]; 504 505 kseq->ksq_loads[PRI_ITHD] = 0; 506 kseq->ksq_loads[PRI_REALTIME] = 0; 507 kseq->ksq_loads[PRI_TIMESHARE] = 0; 508 kseq->ksq_loads[PRI_IDLE] = 0; 509 kseq->ksq_load = 0; 510 #ifdef SMP 511 kseq->ksq_rslices = 0; 512 #endif 513 } 514 515 static void 516 sched_setup(void *dummy) 517 { 518 int i; 519 520 slice_min = (hz/100); 521 slice_max = (hz/10); 522 523 mtx_lock_spin(&sched_lock); 524 /* init kseqs */ 525 for (i = 0; i < MAXCPU; i++) 526 kseq_setup(KSEQ_CPU(i)); 527 528 kseq_add(KSEQ_SELF(), &kse0); 529 mtx_unlock_spin(&sched_lock); 530 #ifdef SMP 531 callout_init(&kseq_lb_callout, 1); 532 kseq_balance(NULL); 533 #endif 534 } 535 536 /* 537 * Scale the scheduling priority according to the "interactivity" of this 538 * process. 539 */ 540 static void 541 sched_priority(struct ksegrp *kg) 542 { 543 int pri; 544 545 if (kg->kg_pri_class != PRI_TIMESHARE) 546 return; 547 548 pri = SCHED_PRI_INTERACT(sched_interact_score(kg)); 549 pri += SCHED_PRI_BASE; 550 pri += kg->kg_nice; 551 552 if (pri > PRI_MAX_TIMESHARE) 553 pri = PRI_MAX_TIMESHARE; 554 else if (pri < PRI_MIN_TIMESHARE) 555 pri = PRI_MIN_TIMESHARE; 556 557 kg->kg_user_pri = pri; 558 559 return; 560 } 561 562 /* 563 * Calculate a time slice based on the properties of the kseg and the runq 564 * that we're on. This is only for PRI_TIMESHARE ksegrps. 565 */ 566 static void 567 sched_slice(struct kse *ke) 568 { 569 struct kseq *kseq; 570 struct ksegrp *kg; 571 572 kg = ke->ke_ksegrp; 573 kseq = KSEQ_CPU(ke->ke_cpu); 574 575 /* 576 * Rationale: 577 * KSEs in interactive ksegs get the minimum slice so that we 578 * quickly notice if it abuses its advantage. 579 * 580 * KSEs in non-interactive ksegs are assigned a slice that is 581 * based on the ksegs nice value relative to the least nice kseg 582 * on the run queue for this cpu. 583 * 584 * If the KSE is less nice than all others it gets the maximum 585 * slice and other KSEs will adjust their slice relative to 586 * this when they first expire. 587 * 588 * There is 20 point window that starts relative to the least 589 * nice kse on the run queue. Slice size is determined by 590 * the kse distance from the last nice ksegrp. 591 * 592 * If you are outside of the window you will get no slice and 593 * you will be reevaluated each time you are selected on the 594 * run queue. 595 * 596 */ 597 598 if (!SCHED_INTERACTIVE(kg)) { 599 int nice; 600 601 nice = kg->kg_nice + (0 - kseq->ksq_nicemin); 602 if (kseq->ksq_loads[PRI_TIMESHARE] == 0 || 603 kg->kg_nice < kseq->ksq_nicemin) 604 ke->ke_slice = SCHED_SLICE_MAX; 605 else if (nice <= SCHED_PRI_NTHRESH) 606 ke->ke_slice = SCHED_SLICE_NICE(nice); 607 else 608 ke->ke_slice = 0; 609 } else 610 ke->ke_slice = SCHED_SLICE_MIN; 611 612 CTR6(KTR_ULE, 613 "Sliced %p(%d) (nice: %d, nicemin: %d, load: %d, interactive: %d)", 614 ke, ke->ke_slice, kg->kg_nice, kseq->ksq_nicemin, 615 kseq->ksq_loads[PRI_TIMESHARE], SCHED_INTERACTIVE(kg)); 616 617 /* 618 * Check to see if we need to scale back the slp and run time 619 * in the kg. This will cause us to forget old interactivity 620 * while maintaining the current ratio. 621 */ 622 sched_interact_update(kg); 623 624 return; 625 } 626 627 static void 628 sched_interact_update(struct ksegrp *kg) 629 { 630 if ((kg->kg_runtime + kg->kg_slptime) > SCHED_SLP_RUN_MAX) { 631 kg->kg_runtime = (kg->kg_runtime / 5) * 4; 632 kg->kg_slptime = (kg->kg_slptime / 5) * 4; 633 } 634 } 635 636 static int 637 sched_interact_score(struct ksegrp *kg) 638 { 639 int div; 640 641 if (kg->kg_runtime > kg->kg_slptime) { 642 div = max(1, kg->kg_runtime / SCHED_INTERACT_HALF); 643 return (SCHED_INTERACT_HALF + 644 (SCHED_INTERACT_HALF - (kg->kg_slptime / div))); 645 } if (kg->kg_slptime > kg->kg_runtime) { 646 div = max(1, kg->kg_slptime / SCHED_INTERACT_HALF); 647 return (kg->kg_runtime / div); 648 } 649 650 /* 651 * This can happen if slptime and runtime are 0. 652 */ 653 return (0); 654 655 } 656 657 /* 658 * This is only somewhat accurate since given many processes of the same 659 * priority they will switch when their slices run out, which will be 660 * at most SCHED_SLICE_MAX. 661 */ 662 int 663 sched_rr_interval(void) 664 { 665 return (SCHED_SLICE_MAX); 666 } 667 668 void 669 sched_pctcpu_update(struct kse *ke) 670 { 671 /* 672 * Adjust counters and watermark for pctcpu calc. 673 */ 674 675 /* 676 * Shift the tick count out so that the divide doesn't round away 677 * our results. 678 */ 679 ke->ke_ticks <<= 10; 680 ke->ke_ticks = (ke->ke_ticks / (ke->ke_ltick - ke->ke_ftick)) * 681 SCHED_CPU_TICKS; 682 ke->ke_ticks >>= 10; 683 ke->ke_ltick = ticks; 684 ke->ke_ftick = ke->ke_ltick - SCHED_CPU_TICKS; 685 } 686 687 #ifdef SMP 688 /* XXX Should be changed to kseq_load_lowest() */ 689 int 690 sched_pickcpu(void) 691 { 692 struct kseq *kseq; 693 int load; 694 int cpu; 695 int i; 696 697 mtx_assert(&sched_lock, MA_OWNED); 698 if (!smp_started) 699 return (0); 700 701 load = 0; 702 cpu = 0; 703 704 for (i = 0; i < mp_maxid; i++) { 705 if (CPU_ABSENT(i)) 706 continue; 707 kseq = KSEQ_CPU(i); 708 if (kseq->ksq_load < load) { 709 cpu = i; 710 load = kseq->ksq_load; 711 } 712 } 713 714 CTR1(KTR_RUNQ, "sched_pickcpu: %d", cpu); 715 return (cpu); 716 } 717 #else 718 int 719 sched_pickcpu(void) 720 { 721 return (0); 722 } 723 #endif 724 725 void 726 sched_prio(struct thread *td, u_char prio) 727 { 728 struct kse *ke; 729 struct runq *rq; 730 731 mtx_assert(&sched_lock, MA_OWNED); 732 ke = td->td_kse; 733 td->td_priority = prio; 734 735 if (TD_ON_RUNQ(td)) { 736 rq = ke->ke_runq; 737 738 runq_remove(rq, ke); 739 runq_add(rq, ke); 740 } 741 } 742 743 void 744 sched_switchout(struct thread *td) 745 { 746 struct kse *ke; 747 748 mtx_assert(&sched_lock, MA_OWNED); 749 750 ke = td->td_kse; 751 752 td->td_last_kse = ke; 753 td->td_lastcpu = td->td_oncpu; 754 td->td_oncpu = NOCPU; 755 td->td_flags &= ~TDF_NEEDRESCHED; 756 757 if (TD_IS_RUNNING(td)) { 758 /* 759 * This queue is always correct except for idle threads which 760 * have a higher priority due to priority propagation. 761 */ 762 if (ke->ke_ksegrp->kg_pri_class == PRI_IDLE && 763 ke->ke_thread->td_priority > PRI_MIN_IDLE) 764 ke->ke_runq = KSEQ_SELF()->ksq_curr; 765 runq_add(ke->ke_runq, ke); 766 /* setrunqueue(td); */ 767 return; 768 } 769 if (ke->ke_runq) 770 kseq_rem(KSEQ_CPU(ke->ke_cpu), ke); 771 /* 772 * We will not be on the run queue. So we must be 773 * sleeping or similar. 774 */ 775 if (td->td_proc->p_flag & P_SA) 776 kse_reassign(ke); 777 } 778 779 void 780 sched_switchin(struct thread *td) 781 { 782 /* struct kse *ke = td->td_kse; */ 783 mtx_assert(&sched_lock, MA_OWNED); 784 785 td->td_oncpu = PCPU_GET(cpuid); 786 } 787 788 void 789 sched_nice(struct ksegrp *kg, int nice) 790 { 791 struct kse *ke; 792 struct thread *td; 793 struct kseq *kseq; 794 795 PROC_LOCK_ASSERT(kg->kg_proc, MA_OWNED); 796 mtx_assert(&sched_lock, MA_OWNED); 797 /* 798 * We need to adjust the nice counts for running KSEs. 799 */ 800 if (kg->kg_pri_class == PRI_TIMESHARE) 801 FOREACH_KSE_IN_GROUP(kg, ke) { 802 if (ke->ke_state != KES_ONRUNQ && 803 ke->ke_state != KES_THREAD) 804 continue; 805 kseq = KSEQ_CPU(ke->ke_cpu); 806 kseq_nice_rem(kseq, kg->kg_nice); 807 kseq_nice_add(kseq, nice); 808 } 809 kg->kg_nice = nice; 810 sched_priority(kg); 811 FOREACH_THREAD_IN_GROUP(kg, td) 812 td->td_flags |= TDF_NEEDRESCHED; 813 } 814 815 void 816 sched_sleep(struct thread *td, u_char prio) 817 { 818 mtx_assert(&sched_lock, MA_OWNED); 819 820 td->td_slptime = ticks; 821 td->td_priority = prio; 822 823 CTR2(KTR_ULE, "sleep kse %p (tick: %d)", 824 td->td_kse, td->td_slptime); 825 } 826 827 void 828 sched_wakeup(struct thread *td) 829 { 830 mtx_assert(&sched_lock, MA_OWNED); 831 832 /* 833 * Let the kseg know how long we slept for. This is because process 834 * interactivity behavior is modeled in the kseg. 835 */ 836 if (td->td_slptime) { 837 struct ksegrp *kg; 838 int hzticks; 839 840 kg = td->td_ksegrp; 841 hzticks = ticks - td->td_slptime; 842 kg->kg_slptime += hzticks << 10; 843 sched_interact_update(kg); 844 sched_priority(kg); 845 if (td->td_kse) 846 sched_slice(td->td_kse); 847 CTR2(KTR_ULE, "wakeup kse %p (%d ticks)", 848 td->td_kse, hzticks); 849 td->td_slptime = 0; 850 } 851 setrunqueue(td); 852 if (td->td_priority < curthread->td_priority) 853 curthread->td_flags |= TDF_NEEDRESCHED; 854 } 855 856 /* 857 * Penalize the parent for creating a new child and initialize the child's 858 * priority. 859 */ 860 void 861 sched_fork(struct proc *p, struct proc *p1) 862 { 863 864 mtx_assert(&sched_lock, MA_OWNED); 865 866 sched_fork_ksegrp(FIRST_KSEGRP_IN_PROC(p), FIRST_KSEGRP_IN_PROC(p1)); 867 sched_fork_kse(FIRST_KSE_IN_PROC(p), FIRST_KSE_IN_PROC(p1)); 868 sched_fork_thread(FIRST_THREAD_IN_PROC(p), FIRST_THREAD_IN_PROC(p1)); 869 } 870 871 void 872 sched_fork_kse(struct kse *ke, struct kse *child) 873 { 874 875 child->ke_slice = 1; /* Attempt to quickly learn interactivity. */ 876 child->ke_cpu = ke->ke_cpu; /* sched_pickcpu(); */ 877 child->ke_runq = NULL; 878 879 /* 880 * Claim that we've been running for one second for statistical 881 * purposes. 882 */ 883 child->ke_ticks = 0; 884 child->ke_ltick = ticks; 885 child->ke_ftick = ticks - hz; 886 } 887 888 void 889 sched_fork_ksegrp(struct ksegrp *kg, struct ksegrp *child) 890 { 891 892 PROC_LOCK_ASSERT(child->kg_proc, MA_OWNED); 893 /* XXX Need something better here */ 894 895 child->kg_slptime = kg->kg_slptime; 896 child->kg_runtime = kg->kg_runtime; 897 kg->kg_runtime += tickincr << 10; 898 sched_interact_update(kg); 899 900 child->kg_user_pri = kg->kg_user_pri; 901 child->kg_nice = kg->kg_nice; 902 } 903 904 void 905 sched_fork_thread(struct thread *td, struct thread *child) 906 { 907 } 908 909 void 910 sched_class(struct ksegrp *kg, int class) 911 { 912 struct kseq *kseq; 913 struct kse *ke; 914 915 mtx_assert(&sched_lock, MA_OWNED); 916 if (kg->kg_pri_class == class) 917 return; 918 919 FOREACH_KSE_IN_GROUP(kg, ke) { 920 if (ke->ke_state != KES_ONRUNQ && 921 ke->ke_state != KES_THREAD) 922 continue; 923 kseq = KSEQ_CPU(ke->ke_cpu); 924 925 kseq->ksq_loads[PRI_BASE(kg->kg_pri_class)]--; 926 kseq->ksq_loads[PRI_BASE(class)]++; 927 928 if (kg->kg_pri_class == PRI_TIMESHARE) 929 kseq_nice_rem(kseq, kg->kg_nice); 930 else if (class == PRI_TIMESHARE) 931 kseq_nice_add(kseq, kg->kg_nice); 932 } 933 934 kg->kg_pri_class = class; 935 } 936 937 /* 938 * Return some of the child's priority and interactivity to the parent. 939 */ 940 void 941 sched_exit(struct proc *p, struct proc *child) 942 { 943 /* XXX Need something better here */ 944 mtx_assert(&sched_lock, MA_OWNED); 945 sched_exit_kse(FIRST_KSE_IN_PROC(p), FIRST_KSE_IN_PROC(child)); 946 sched_exit_ksegrp(FIRST_KSEGRP_IN_PROC(p), FIRST_KSEGRP_IN_PROC(child)); 947 } 948 949 void 950 sched_exit_kse(struct kse *ke, struct kse *child) 951 { 952 kseq_rem(KSEQ_CPU(child->ke_cpu), child); 953 } 954 955 void 956 sched_exit_ksegrp(struct ksegrp *kg, struct ksegrp *child) 957 { 958 /* kg->kg_slptime += child->kg_slptime; */ 959 kg->kg_runtime += child->kg_runtime; 960 sched_interact_update(kg); 961 } 962 963 void 964 sched_exit_thread(struct thread *td, struct thread *child) 965 { 966 } 967 968 void 969 sched_clock(struct kse *ke) 970 { 971 struct kseq *kseq; 972 struct ksegrp *kg; 973 struct thread *td; 974 #if 0 975 struct kse *nke; 976 #endif 977 978 /* 979 * sched_setup() apparently happens prior to stathz being set. We 980 * need to resolve the timers earlier in the boot so we can avoid 981 * calculating this here. 982 */ 983 if (realstathz == 0) { 984 realstathz = stathz ? stathz : hz; 985 tickincr = hz / realstathz; 986 /* 987 * XXX This does not work for values of stathz that are much 988 * larger than hz. 989 */ 990 if (tickincr == 0) 991 tickincr = 1; 992 } 993 994 td = ke->ke_thread; 995 kg = ke->ke_ksegrp; 996 997 mtx_assert(&sched_lock, MA_OWNED); 998 KASSERT((td != NULL), ("schedclock: null thread pointer")); 999 1000 /* Adjust ticks for pctcpu */ 1001 ke->ke_ticks++; 1002 ke->ke_ltick = ticks; 1003 1004 /* Go up to one second beyond our max and then trim back down */ 1005 if (ke->ke_ftick + SCHED_CPU_TICKS + hz < ke->ke_ltick) 1006 sched_pctcpu_update(ke); 1007 1008 if (td->td_flags & TDF_IDLETD) 1009 return; 1010 1011 CTR4(KTR_ULE, "Tick kse %p (slice: %d, slptime: %d, runtime: %d)", 1012 ke, ke->ke_slice, kg->kg_slptime >> 10, kg->kg_runtime >> 10); 1013 1014 /* 1015 * We only do slicing code for TIMESHARE ksegrps. 1016 */ 1017 if (kg->kg_pri_class != PRI_TIMESHARE) 1018 return; 1019 /* 1020 * Check for a higher priority task on the run queue. This can happen 1021 * on SMP if another processor woke up a process on our runq. 1022 */ 1023 kseq = KSEQ_SELF(); 1024 #if 0 1025 if (kseq->ksq_load > 1 && (nke = kseq_choose(kseq)) != NULL) { 1026 if (sched_strict && 1027 nke->ke_thread->td_priority < td->td_priority) 1028 td->td_flags |= TDF_NEEDRESCHED; 1029 else if (nke->ke_thread->td_priority < 1030 td->td_priority SCHED_PRIO_SLOP) 1031 1032 if (nke->ke_thread->td_priority < td->td_priority) 1033 td->td_flags |= TDF_NEEDRESCHED; 1034 } 1035 #endif 1036 /* 1037 * We used a tick charge it to the ksegrp so that we can compute our 1038 * interactivity. 1039 */ 1040 kg->kg_runtime += tickincr << 10; 1041 sched_interact_update(kg); 1042 1043 /* 1044 * We used up one time slice. 1045 */ 1046 ke->ke_slice--; 1047 #ifdef SMP 1048 kseq->ksq_rslices--; 1049 #endif 1050 1051 if (ke->ke_slice > 0) 1052 return; 1053 /* 1054 * We're out of time, recompute priorities and requeue. 1055 */ 1056 kseq_rem(kseq, ke); 1057 sched_priority(kg); 1058 sched_slice(ke); 1059 if (SCHED_CURR(kg, ke)) 1060 ke->ke_runq = kseq->ksq_curr; 1061 else 1062 ke->ke_runq = kseq->ksq_next; 1063 kseq_add(kseq, ke); 1064 td->td_flags |= TDF_NEEDRESCHED; 1065 } 1066 1067 int 1068 sched_runnable(void) 1069 { 1070 struct kseq *kseq; 1071 int load; 1072 1073 load = 1; 1074 1075 mtx_lock_spin(&sched_lock); 1076 kseq = KSEQ_SELF(); 1077 1078 if (kseq->ksq_load) 1079 goto out; 1080 #ifdef SMP 1081 /* 1082 * For SMP we may steal other processor's KSEs. Just search until we 1083 * verify that at least on other cpu has a runnable task. 1084 */ 1085 if (smp_started) { 1086 int i; 1087 1088 for (i = 0; i < mp_maxid; i++) { 1089 if (CPU_ABSENT(i)) 1090 continue; 1091 kseq = KSEQ_CPU(i); 1092 if (kseq->ksq_load > 1) 1093 goto out; 1094 } 1095 } 1096 #endif 1097 load = 0; 1098 out: 1099 mtx_unlock_spin(&sched_lock); 1100 return (load); 1101 } 1102 1103 void 1104 sched_userret(struct thread *td) 1105 { 1106 struct ksegrp *kg; 1107 struct kseq *kseq; 1108 struct kse *ke; 1109 1110 kg = td->td_ksegrp; 1111 1112 if (td->td_priority != kg->kg_user_pri) { 1113 mtx_lock_spin(&sched_lock); 1114 td->td_priority = kg->kg_user_pri; 1115 kseq = KSEQ_SELF(); 1116 if (td->td_ksegrp->kg_pri_class == PRI_TIMESHARE && 1117 kseq->ksq_load > 1 && 1118 (ke = kseq_choose(kseq)) != NULL && 1119 ke->ke_thread->td_priority < td->td_priority) 1120 curthread->td_flags |= TDF_NEEDRESCHED; 1121 mtx_unlock_spin(&sched_lock); 1122 } 1123 } 1124 1125 struct kse * 1126 sched_choose(void) 1127 { 1128 struct kseq *kseq; 1129 struct kse *ke; 1130 1131 mtx_assert(&sched_lock, MA_OWNED); 1132 #ifdef SMP 1133 retry: 1134 #endif 1135 kseq = KSEQ_SELF(); 1136 ke = kseq_choose(kseq); 1137 if (ke) { 1138 runq_remove(ke->ke_runq, ke); 1139 ke->ke_state = KES_THREAD; 1140 1141 if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) { 1142 CTR4(KTR_ULE, "Run kse %p from %p (slice: %d, pri: %d)", 1143 ke, ke->ke_runq, ke->ke_slice, 1144 ke->ke_thread->td_priority); 1145 } 1146 return (ke); 1147 } 1148 1149 #ifdef SMP 1150 if (smp_started) { 1151 /* 1152 * Find the cpu with the highest load and steal one proc. 1153 */ 1154 if ((kseq = kseq_load_highest()) == NULL) 1155 return (NULL); 1156 1157 /* 1158 * Remove this kse from this kseq and runq and then requeue 1159 * on the current processor. Then we will dequeue it 1160 * normally above. 1161 */ 1162 kseq_move(kseq, PCPU_GET(cpuid)); 1163 goto retry; 1164 } 1165 #endif 1166 1167 return (NULL); 1168 } 1169 1170 void 1171 sched_add(struct kse *ke) 1172 { 1173 struct kseq *kseq; 1174 struct ksegrp *kg; 1175 1176 mtx_assert(&sched_lock, MA_OWNED); 1177 KASSERT((ke->ke_thread != NULL), ("sched_add: No thread on KSE")); 1178 KASSERT((ke->ke_thread->td_kse != NULL), 1179 ("sched_add: No KSE on thread")); 1180 KASSERT(ke->ke_state != KES_ONRUNQ, 1181 ("sched_add: kse %p (%s) already in run queue", ke, 1182 ke->ke_proc->p_comm)); 1183 KASSERT(ke->ke_proc->p_sflag & PS_INMEM, 1184 ("sched_add: process swapped out")); 1185 KASSERT(ke->ke_runq == NULL, 1186 ("sched_add: KSE %p is still assigned to a run queue", ke)); 1187 1188 kg = ke->ke_ksegrp; 1189 1190 switch (PRI_BASE(kg->kg_pri_class)) { 1191 case PRI_ITHD: 1192 case PRI_REALTIME: 1193 kseq = KSEQ_SELF(); 1194 ke->ke_runq = kseq->ksq_curr; 1195 ke->ke_slice = SCHED_SLICE_MAX; 1196 ke->ke_cpu = PCPU_GET(cpuid); 1197 break; 1198 case PRI_TIMESHARE: 1199 kseq = KSEQ_CPU(ke->ke_cpu); 1200 if (SCHED_CURR(kg, ke)) 1201 ke->ke_runq = kseq->ksq_curr; 1202 else 1203 ke->ke_runq = kseq->ksq_next; 1204 break; 1205 case PRI_IDLE: 1206 kseq = KSEQ_CPU(ke->ke_cpu); 1207 /* 1208 * This is for priority prop. 1209 */ 1210 if (ke->ke_thread->td_priority > PRI_MIN_IDLE) 1211 ke->ke_runq = kseq->ksq_curr; 1212 else 1213 ke->ke_runq = &kseq->ksq_idle; 1214 ke->ke_slice = SCHED_SLICE_MIN; 1215 break; 1216 default: 1217 panic("Unknown pri class.\n"); 1218 break; 1219 } 1220 1221 ke->ke_ksegrp->kg_runq_kses++; 1222 ke->ke_state = KES_ONRUNQ; 1223 1224 runq_add(ke->ke_runq, ke); 1225 kseq_add(kseq, ke); 1226 } 1227 1228 void 1229 sched_rem(struct kse *ke) 1230 { 1231 struct kseq *kseq; 1232 1233 mtx_assert(&sched_lock, MA_OWNED); 1234 KASSERT((ke->ke_state == KES_ONRUNQ), ("KSE not on run queue")); 1235 1236 ke->ke_state = KES_THREAD; 1237 ke->ke_ksegrp->kg_runq_kses--; 1238 kseq = KSEQ_CPU(ke->ke_cpu); 1239 runq_remove(ke->ke_runq, ke); 1240 kseq_rem(kseq, ke); 1241 } 1242 1243 fixpt_t 1244 sched_pctcpu(struct kse *ke) 1245 { 1246 fixpt_t pctcpu; 1247 1248 pctcpu = 0; 1249 1250 mtx_lock_spin(&sched_lock); 1251 if (ke->ke_ticks) { 1252 int rtick; 1253 1254 /* Update to account for time potentially spent sleeping */ 1255 ke->ke_ltick = ticks; 1256 /* 1257 * Don't update more frequently than twice a second. Allowing 1258 * this causes the cpu usage to decay away too quickly due to 1259 * rounding errors. 1260 */ 1261 if (ke->ke_ltick < (ticks - (hz / 2))) 1262 sched_pctcpu_update(ke); 1263 1264 /* How many rtick per second ? */ 1265 rtick = min(ke->ke_ticks / SCHED_CPU_TIME, SCHED_CPU_TICKS); 1266 pctcpu = (FSCALE * ((FSCALE * rtick)/realstathz)) >> FSHIFT; 1267 } 1268 1269 ke->ke_proc->p_swtime = ke->ke_ltick - ke->ke_ftick; 1270 mtx_unlock_spin(&sched_lock); 1271 1272 return (pctcpu); 1273 } 1274 1275 int 1276 sched_sizeof_kse(void) 1277 { 1278 return (sizeof(struct kse) + sizeof(struct ke_sched)); 1279 } 1280 1281 int 1282 sched_sizeof_ksegrp(void) 1283 { 1284 return (sizeof(struct ksegrp) + sizeof(struct kg_sched)); 1285 } 1286 1287 int 1288 sched_sizeof_proc(void) 1289 { 1290 return (sizeof(struct proc)); 1291 } 1292 1293 int 1294 sched_sizeof_thread(void) 1295 { 1296 return (sizeof(struct thread) + sizeof(struct td_sched)); 1297 } 1298