1 /*- 2 * Copyright (c) 2001 Jake Burkholder <jake@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, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include "opt_sched.h" 32 33 #ifndef KERN_SWITCH_INCLUDE 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/kdb.h> 37 #include <sys/kernel.h> 38 #include <sys/ktr.h> 39 #include <sys/lock.h> 40 #include <sys/mutex.h> 41 #include <sys/proc.h> 42 #include <sys/queue.h> 43 #include <sys/sched.h> 44 #else /* KERN_SWITCH_INCLUDE */ 45 #if defined(SMP) && (defined(__i386__) || defined(__amd64__)) 46 #include <sys/smp.h> 47 #endif 48 #if defined(SMP) && defined(SCHED_4BSD) 49 #include <sys/sysctl.h> 50 #endif 51 52 /* Uncomment this to enable logging of critical_enter/exit. */ 53 #if 0 54 #define KTR_CRITICAL KTR_SCHED 55 #else 56 #define KTR_CRITICAL 0 57 #endif 58 59 #ifdef FULL_PREEMPTION 60 #ifndef PREEMPTION 61 #error "The FULL_PREEMPTION option requires the PREEMPTION option" 62 #endif 63 #endif 64 65 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS); 66 67 /* 68 * kern.sched.preemption allows user space to determine if preemption support 69 * is compiled in or not. It is not currently a boot or runtime flag that 70 * can be changed. 71 */ 72 #ifdef PREEMPTION 73 static int kern_sched_preemption = 1; 74 #else 75 static int kern_sched_preemption = 0; 76 #endif 77 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD, 78 &kern_sched_preemption, 0, "Kernel preemption enabled"); 79 80 /************************************************************************ 81 * Functions that manipulate runnability from a thread perspective. * 82 ************************************************************************/ 83 /* 84 * Select the thread that will be run next. 85 */ 86 struct thread * 87 choosethread(void) 88 { 89 struct td_sched *ts; 90 struct thread *td; 91 92 #if defined(SMP) && (defined(__i386__) || defined(__amd64__)) 93 if (smp_active == 0 && PCPU_GET(cpuid) != 0) { 94 /* Shutting down, run idlethread on AP's */ 95 td = PCPU_GET(idlethread); 96 ts = td->td_sched; 97 CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td); 98 ts->ts_flags |= TSF_DIDRUN; 99 TD_SET_RUNNING(td); 100 return (td); 101 } 102 #endif 103 104 retry: 105 ts = sched_choose(); 106 if (ts) { 107 td = ts->ts_thread; 108 CTR2(KTR_RUNQ, "choosethread: td=%p pri=%d", 109 td, td->td_priority); 110 } else { 111 /* Simulate runq_choose() having returned the idle thread */ 112 td = PCPU_GET(idlethread); 113 ts = td->td_sched; 114 CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td); 115 } 116 ts->ts_flags |= TSF_DIDRUN; 117 118 /* 119 * If we are in panic, only allow system threads, 120 * plus the one we are running in, to be run. 121 */ 122 if (panicstr && ((td->td_proc->p_flag & P_SYSTEM) == 0 && 123 (td->td_flags & TDF_INPANIC) == 0)) { 124 /* note that it is no longer on the run queue */ 125 TD_SET_CAN_RUN(td); 126 goto retry; 127 } 128 129 TD_SET_RUNNING(td); 130 return (td); 131 } 132 133 134 #if 0 135 /* 136 * currently not used.. threads remove themselves from the 137 * run queue by running. 138 */ 139 static void 140 remrunqueue(struct thread *td) 141 { 142 mtx_assert(&sched_lock, MA_OWNED); 143 KASSERT((TD_ON_RUNQ(td)), ("remrunqueue: Bad state on run queue")); 144 CTR1(KTR_RUNQ, "remrunqueue: td%p", td); 145 TD_SET_CAN_RUN(td); 146 /* remove from sys run queue */ 147 sched_rem(td); 148 return; 149 } 150 #endif 151 152 /* 153 * Change the priority of a thread that is on the run queue. 154 */ 155 void 156 adjustrunqueue( struct thread *td, int newpri) 157 { 158 struct td_sched *ts; 159 160 mtx_assert(&sched_lock, MA_OWNED); 161 KASSERT((TD_ON_RUNQ(td)), ("adjustrunqueue: Bad state on run queue")); 162 163 ts = td->td_sched; 164 CTR1(KTR_RUNQ, "adjustrunqueue: td%p", td); 165 /* We only care about the td_sched in the run queue. */ 166 td->td_priority = newpri; 167 #ifndef SCHED_CORE 168 if (ts->ts_rqindex != (newpri / RQ_PPQ)) 169 #else 170 if (ts->ts_rqindex != newpri) 171 #endif 172 { 173 sched_rem(td); 174 sched_add(td, SRQ_BORING); 175 } 176 } 177 178 void 179 setrunqueue(struct thread *td, int flags) 180 { 181 182 CTR2(KTR_RUNQ, "setrunqueue: td:%p pid:%d", 183 td, td->td_proc->p_pid); 184 CTR5(KTR_SCHED, "setrunqueue: %p(%s) prio %d by %p(%s)", 185 td, td->td_proc->p_comm, td->td_priority, curthread, 186 curthread->td_proc->p_comm); 187 mtx_assert(&sched_lock, MA_OWNED); 188 KASSERT((td->td_inhibitors == 0), 189 ("setrunqueue: trying to run inhibitted thread")); 190 KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)), 191 ("setrunqueue: bad thread state")); 192 TD_SET_RUNQ(td); 193 sched_add(td, flags); 194 } 195 196 /* 197 * Kernel thread preemption implementation. Critical sections mark 198 * regions of code in which preemptions are not allowed. 199 */ 200 void 201 critical_enter(void) 202 { 203 struct thread *td; 204 205 td = curthread; 206 td->td_critnest++; 207 CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td, 208 (long)td->td_proc->p_pid, td->td_proc->p_comm, td->td_critnest); 209 } 210 211 void 212 critical_exit(void) 213 { 214 struct thread *td; 215 216 td = curthread; 217 KASSERT(td->td_critnest != 0, 218 ("critical_exit: td_critnest == 0")); 219 #ifdef PREEMPTION 220 if (td->td_critnest == 1) { 221 td->td_critnest = 0; 222 mtx_assert(&sched_lock, MA_NOTOWNED); 223 if (td->td_owepreempt) { 224 td->td_critnest = 1; 225 mtx_lock_spin(&sched_lock); 226 td->td_critnest--; 227 mi_switch(SW_INVOL, NULL); 228 mtx_unlock_spin(&sched_lock); 229 } 230 } else 231 #endif 232 td->td_critnest--; 233 234 CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td, 235 (long)td->td_proc->p_pid, td->td_proc->p_comm, td->td_critnest); 236 } 237 238 /* 239 * This function is called when a thread is about to be put on run queue 240 * because it has been made runnable or its priority has been adjusted. It 241 * determines if the new thread should be immediately preempted to. If so, 242 * it switches to it and eventually returns true. If not, it returns false 243 * so that the caller may place the thread on an appropriate run queue. 244 */ 245 int 246 maybe_preempt(struct thread *td) 247 { 248 #ifdef PREEMPTION 249 struct thread *ctd; 250 int cpri, pri; 251 #endif 252 253 mtx_assert(&sched_lock, MA_OWNED); 254 #ifdef PREEMPTION 255 /* 256 * The new thread should not preempt the current thread if any of the 257 * following conditions are true: 258 * 259 * - The kernel is in the throes of crashing (panicstr). 260 * - The current thread has a higher (numerically lower) or 261 * equivalent priority. Note that this prevents curthread from 262 * trying to preempt to itself. 263 * - It is too early in the boot for context switches (cold is set). 264 * - The current thread has an inhibitor set or is in the process of 265 * exiting. In this case, the current thread is about to switch 266 * out anyways, so there's no point in preempting. If we did, 267 * the current thread would not be properly resumed as well, so 268 * just avoid that whole landmine. 269 * - If the new thread's priority is not a realtime priority and 270 * the current thread's priority is not an idle priority and 271 * FULL_PREEMPTION is disabled. 272 * 273 * If all of these conditions are false, but the current thread is in 274 * a nested critical section, then we have to defer the preemption 275 * until we exit the critical section. Otherwise, switch immediately 276 * to the new thread. 277 */ 278 ctd = curthread; 279 KASSERT ((ctd->td_sched != NULL && ctd->td_sched->ts_thread == ctd), 280 ("thread has no (or wrong) sched-private part.")); 281 KASSERT((td->td_inhibitors == 0), 282 ("maybe_preempt: trying to run inhibitted thread")); 283 pri = td->td_priority; 284 cpri = ctd->td_priority; 285 if (panicstr != NULL || pri >= cpri || cold /* || dumping */ || 286 TD_IS_INHIBITED(ctd) || td->td_sched->ts_state != TSS_THREAD) 287 return (0); 288 #ifndef FULL_PREEMPTION 289 if (pri > PRI_MAX_ITHD && cpri < PRI_MIN_IDLE) 290 return (0); 291 #endif 292 293 if (ctd->td_critnest > 1) { 294 CTR1(KTR_PROC, "maybe_preempt: in critical section %d", 295 ctd->td_critnest); 296 ctd->td_owepreempt = 1; 297 return (0); 298 } 299 300 /* 301 * Thread is runnable but not yet put on system run queue. 302 */ 303 MPASS(TD_ON_RUNQ(td)); 304 MPASS(td->td_sched->ts_state != TSS_ONRUNQ); 305 TD_SET_RUNNING(td); 306 CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td, 307 td->td_proc->p_pid, td->td_proc->p_comm); 308 mi_switch(SW_INVOL|SW_PREEMPT, td); 309 return (1); 310 #else 311 return (0); 312 #endif 313 } 314 315 #if 0 316 #ifndef PREEMPTION 317 /* XXX: There should be a non-static version of this. */ 318 static void 319 printf_caddr_t(void *data) 320 { 321 printf("%s", (char *)data); 322 } 323 static char preempt_warning[] = 324 "WARNING: Kernel preemption is disabled, expect reduced performance.\n"; 325 SYSINIT(preempt_warning, SI_SUB_COPYRIGHT, SI_ORDER_ANY, printf_caddr_t, 326 preempt_warning) 327 #endif 328 #endif 329 330 /************************************************************************ 331 * SYSTEM RUN QUEUE manipulations and tests * 332 ************************************************************************/ 333 /* 334 * Initialize a run structure. 335 */ 336 void 337 runq_init(struct runq *rq) 338 { 339 int i; 340 341 bzero(rq, sizeof *rq); 342 for (i = 0; i < RQ_NQS; i++) 343 TAILQ_INIT(&rq->rq_queues[i]); 344 } 345 346 /* 347 * Clear the status bit of the queue corresponding to priority level pri, 348 * indicating that it is empty. 349 */ 350 static __inline void 351 runq_clrbit(struct runq *rq, int pri) 352 { 353 struct rqbits *rqb; 354 355 rqb = &rq->rq_status; 356 CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d", 357 rqb->rqb_bits[RQB_WORD(pri)], 358 rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri), 359 RQB_BIT(pri), RQB_WORD(pri)); 360 rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri); 361 } 362 363 /* 364 * Find the index of the first non-empty run queue. This is done by 365 * scanning the status bits, a set bit indicates a non-empty queue. 366 */ 367 static __inline int 368 runq_findbit(struct runq *rq) 369 { 370 struct rqbits *rqb; 371 int pri; 372 int i; 373 374 rqb = &rq->rq_status; 375 for (i = 0; i < RQB_LEN; i++) 376 if (rqb->rqb_bits[i]) { 377 pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW); 378 CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d", 379 rqb->rqb_bits[i], i, pri); 380 return (pri); 381 } 382 383 return (-1); 384 } 385 386 /* 387 * Set the status bit of the queue corresponding to priority level pri, 388 * indicating that it is non-empty. 389 */ 390 static __inline void 391 runq_setbit(struct runq *rq, int pri) 392 { 393 struct rqbits *rqb; 394 395 rqb = &rq->rq_status; 396 CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d", 397 rqb->rqb_bits[RQB_WORD(pri)], 398 rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri), 399 RQB_BIT(pri), RQB_WORD(pri)); 400 rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri); 401 } 402 403 /* 404 * Add the thread to the queue specified by its priority, and set the 405 * corresponding status bit. 406 */ 407 void 408 runq_add(struct runq *rq, struct td_sched *ts, int flags) 409 { 410 struct rqhead *rqh; 411 int pri; 412 413 pri = ts->ts_thread->td_priority / RQ_PPQ; 414 ts->ts_rqindex = pri; 415 runq_setbit(rq, pri); 416 rqh = &rq->rq_queues[pri]; 417 CTR5(KTR_RUNQ, "runq_add: td=%p ts=%p pri=%d %d rqh=%p", 418 ts->ts_thread, ts, ts->ts_thread->td_priority, pri, rqh); 419 if (flags & SRQ_PREEMPTED) { 420 TAILQ_INSERT_HEAD(rqh, ts, ts_procq); 421 } else { 422 TAILQ_INSERT_TAIL(rqh, ts, ts_procq); 423 } 424 } 425 426 /* 427 * Return true if there are runnable processes of any priority on the run 428 * queue, false otherwise. Has no side effects, does not modify the run 429 * queue structure. 430 */ 431 int 432 runq_check(struct runq *rq) 433 { 434 struct rqbits *rqb; 435 int i; 436 437 rqb = &rq->rq_status; 438 for (i = 0; i < RQB_LEN; i++) 439 if (rqb->rqb_bits[i]) { 440 CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d", 441 rqb->rqb_bits[i], i); 442 return (1); 443 } 444 CTR0(KTR_RUNQ, "runq_check: empty"); 445 446 return (0); 447 } 448 449 #if defined(SMP) && defined(SCHED_4BSD) 450 int runq_fuzz = 1; 451 SYSCTL_INT(_kern_sched, OID_AUTO, runq_fuzz, CTLFLAG_RW, &runq_fuzz, 0, ""); 452 #endif 453 454 /* 455 * Find the highest priority process on the run queue. 456 */ 457 struct td_sched * 458 runq_choose(struct runq *rq) 459 { 460 struct rqhead *rqh; 461 struct td_sched *ts; 462 int pri; 463 464 mtx_assert(&sched_lock, MA_OWNED); 465 while ((pri = runq_findbit(rq)) != -1) { 466 rqh = &rq->rq_queues[pri]; 467 #if defined(SMP) && defined(SCHED_4BSD) 468 /* fuzz == 1 is normal.. 0 or less are ignored */ 469 if (runq_fuzz > 1) { 470 /* 471 * In the first couple of entries, check if 472 * there is one for our CPU as a preference. 473 */ 474 int count = runq_fuzz; 475 int cpu = PCPU_GET(cpuid); 476 struct td_sched *ts2; 477 ts2 = ts = TAILQ_FIRST(rqh); 478 479 while (count-- && ts2) { 480 if (ts->ts_thread->td_lastcpu == cpu) { 481 ts = ts2; 482 break; 483 } 484 ts2 = TAILQ_NEXT(ts2, ts_procq); 485 } 486 } else 487 #endif 488 ts = TAILQ_FIRST(rqh); 489 KASSERT(ts != NULL, ("runq_choose: no proc on busy queue")); 490 CTR3(KTR_RUNQ, 491 "runq_choose: pri=%d td_sched=%p rqh=%p", pri, ts, rqh); 492 return (ts); 493 } 494 CTR1(KTR_RUNQ, "runq_choose: idleproc pri=%d", pri); 495 496 return (NULL); 497 } 498 499 /* 500 * Remove the thread from the queue specified by its priority, and clear the 501 * corresponding status bit if the queue becomes empty. 502 * Caller must set ts->ts_state afterwards. 503 */ 504 void 505 runq_remove(struct runq *rq, struct td_sched *ts) 506 { 507 struct rqhead *rqh; 508 int pri; 509 510 KASSERT(ts->ts_thread->td_proc->p_sflag & PS_INMEM, 511 ("runq_remove: process swapped out")); 512 pri = ts->ts_rqindex; 513 rqh = &rq->rq_queues[pri]; 514 CTR5(KTR_RUNQ, "runq_remove: td=%p, ts=%p pri=%d %d rqh=%p", 515 ts->ts_thread, ts, ts->ts_thread->td_priority, pri, rqh); 516 KASSERT(ts != NULL, ("runq_remove: no proc on busy queue")); 517 TAILQ_REMOVE(rqh, ts, ts_procq); 518 if (TAILQ_EMPTY(rqh)) { 519 CTR0(KTR_RUNQ, "runq_remove: empty"); 520 runq_clrbit(rq, pri); 521 } 522 } 523 524 /****** functions that are temporarily here ***********/ 525 #include <vm/uma.h> 526 extern struct mtx kse_zombie_lock; 527 528 /* 529 * Allocate scheduler specific per-process resources. 530 * The thread and proc have already been linked in. 531 * 532 * Called from: 533 * proc_init() (UMA init method) 534 */ 535 void 536 sched_newproc(struct proc *p, struct thread *td) 537 { 538 } 539 540 /* 541 * thread is being either created or recycled. 542 * Fix up the per-scheduler resources associated with it. 543 * Called from: 544 * sched_fork_thread() 545 * thread_dtor() (*may go away) 546 * thread_init() (*may go away) 547 */ 548 void 549 sched_newthread(struct thread *td) 550 { 551 struct td_sched *ts; 552 553 ts = (struct td_sched *) (td + 1); 554 bzero(ts, sizeof(*ts)); 555 td->td_sched = ts; 556 ts->ts_thread = td; 557 ts->ts_state = TSS_THREAD; 558 } 559 560 /* 561 * Called from: 562 * thr_create() 563 * proc_init() (UMA) via sched_newproc() 564 */ 565 void 566 sched_init_concurrency(struct proc *p) 567 { 568 } 569 570 /* 571 * Change the concurrency of an existing proc to N 572 * Called from: 573 * kse_create() 574 * kse_exit() 575 * thread_exit() 576 * thread_single() 577 */ 578 void 579 sched_set_concurrency(struct proc *p, int concurrency) 580 { 581 } 582 583 /* 584 * Called from thread_exit() for all exiting thread 585 * 586 * Not to be confused with sched_exit_thread() 587 * that is only called from thread_exit() for threads exiting 588 * without the rest of the process exiting because it is also called from 589 * sched_exit() and we wouldn't want to call it twice. 590 * XXX This can probably be fixed. 591 */ 592 void 593 sched_thread_exit(struct thread *td) 594 { 595 } 596 597 #endif /* KERN_SWITCH_INCLUDE */ 598