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 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kdb.h> 36 #include <sys/kernel.h> 37 #include <sys/ktr.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/proc.h> 41 #include <sys/queue.h> 42 #include <sys/sched.h> 43 #include <sys/smp.h> 44 #include <sys/sysctl.h> 45 46 #include <machine/cpu.h> 47 48 /* Uncomment this to enable logging of critical_enter/exit. */ 49 #if 0 50 #define KTR_CRITICAL KTR_SCHED 51 #else 52 #define KTR_CRITICAL 0 53 #endif 54 55 #ifdef FULL_PREEMPTION 56 #ifndef PREEMPTION 57 #error "The FULL_PREEMPTION option requires the PREEMPTION option" 58 #endif 59 #endif 60 61 CTASSERT((RQB_BPW * RQB_LEN) == RQ_NQS); 62 63 /* 64 * kern.sched.preemption allows user space to determine if preemption support 65 * is compiled in or not. It is not currently a boot or runtime flag that 66 * can be changed. 67 */ 68 #ifdef PREEMPTION 69 static int kern_sched_preemption = 1; 70 #else 71 static int kern_sched_preemption = 0; 72 #endif 73 SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD, 74 &kern_sched_preemption, 0, "Kernel preemption enabled"); 75 76 /* 77 * Support for scheduler stats exported via kern.sched.stats. All stats may 78 * be reset with kern.sched.stats.reset = 1. Stats may be defined elsewhere 79 * with SCHED_STAT_DEFINE(). 80 */ 81 #ifdef SCHED_STATS 82 SYSCTL_NODE(_kern_sched, OID_AUTO, stats, CTLFLAG_RW, 0, "switch stats"); 83 84 /* Switch reasons from mi_switch(). */ 85 DPCPU_DEFINE(long, sched_switch_stats[SWT_COUNT]); 86 SCHED_STAT_DEFINE_VAR(uncategorized, 87 &DPCPU_NAME(sched_switch_stats[SWT_NONE]), ""); 88 SCHED_STAT_DEFINE_VAR(preempt, 89 &DPCPU_NAME(sched_switch_stats[SWT_PREEMPT]), ""); 90 SCHED_STAT_DEFINE_VAR(owepreempt, 91 &DPCPU_NAME(sched_switch_stats[SWT_OWEPREEMPT]), ""); 92 SCHED_STAT_DEFINE_VAR(turnstile, 93 &DPCPU_NAME(sched_switch_stats[SWT_TURNSTILE]), ""); 94 SCHED_STAT_DEFINE_VAR(sleepq, 95 &DPCPU_NAME(sched_switch_stats[SWT_SLEEPQ]), ""); 96 SCHED_STAT_DEFINE_VAR(sleepqtimo, 97 &DPCPU_NAME(sched_switch_stats[SWT_SLEEPQTIMO]), ""); 98 SCHED_STAT_DEFINE_VAR(relinquish, 99 &DPCPU_NAME(sched_switch_stats[SWT_RELINQUISH]), ""); 100 SCHED_STAT_DEFINE_VAR(needresched, 101 &DPCPU_NAME(sched_switch_stats[SWT_NEEDRESCHED]), ""); 102 SCHED_STAT_DEFINE_VAR(idle, 103 &DPCPU_NAME(sched_switch_stats[SWT_IDLE]), ""); 104 SCHED_STAT_DEFINE_VAR(iwait, 105 &DPCPU_NAME(sched_switch_stats[SWT_IWAIT]), ""); 106 SCHED_STAT_DEFINE_VAR(suspend, 107 &DPCPU_NAME(sched_switch_stats[SWT_SUSPEND]), ""); 108 SCHED_STAT_DEFINE_VAR(remotepreempt, 109 &DPCPU_NAME(sched_switch_stats[SWT_REMOTEPREEMPT]), ""); 110 SCHED_STAT_DEFINE_VAR(remotewakeidle, 111 &DPCPU_NAME(sched_switch_stats[SWT_REMOTEWAKEIDLE]), ""); 112 113 static int 114 sysctl_stats_reset(SYSCTL_HANDLER_ARGS) 115 { 116 struct sysctl_oid *p; 117 uintptr_t counter; 118 int error; 119 int val; 120 int i; 121 122 val = 0; 123 error = sysctl_handle_int(oidp, &val, 0, req); 124 if (error != 0 || req->newptr == NULL) 125 return (error); 126 if (val == 0) 127 return (0); 128 /* 129 * Traverse the list of children of _kern_sched_stats and reset each 130 * to 0. Skip the reset entry. 131 */ 132 SLIST_FOREACH(p, oidp->oid_parent, oid_link) { 133 if (p == oidp || p->oid_arg1 == NULL) 134 continue; 135 counter = (uintptr_t)p->oid_arg1; 136 CPU_FOREACH(i) { 137 *(long *)(dpcpu_off[i] + counter) = 0; 138 } 139 } 140 return (0); 141 } 142 143 SYSCTL_PROC(_kern_sched_stats, OID_AUTO, reset, CTLTYPE_INT | CTLFLAG_WR, NULL, 144 0, sysctl_stats_reset, "I", "Reset scheduler statistics"); 145 #endif 146 147 /************************************************************************ 148 * Functions that manipulate runnability from a thread perspective. * 149 ************************************************************************/ 150 /* 151 * Select the thread that will be run next. 152 */ 153 struct thread * 154 choosethread(void) 155 { 156 struct thread *td; 157 158 retry: 159 td = sched_choose(); 160 161 /* 162 * If we are in panic, only allow system threads, 163 * plus the one we are running in, to be run. 164 */ 165 if (panicstr && ((td->td_proc->p_flag & P_SYSTEM) == 0 && 166 (td->td_flags & TDF_INPANIC) == 0)) { 167 /* note that it is no longer on the run queue */ 168 TD_SET_CAN_RUN(td); 169 goto retry; 170 } 171 172 TD_SET_RUNNING(td); 173 return (td); 174 } 175 176 /* 177 * Kernel thread preemption implementation. Critical sections mark 178 * regions of code in which preemptions are not allowed. 179 * 180 * It might seem a good idea to inline critical_enter() but, in order 181 * to prevent instructions reordering by the compiler, a __compiler_membar() 182 * would have to be used here (the same as sched_pin()). The performance 183 * penalty imposed by the membar could, then, produce slower code than 184 * the function call itself, for most cases. 185 */ 186 void 187 critical_enter(void) 188 { 189 struct thread *td; 190 191 td = curthread; 192 td->td_critnest++; 193 CTR4(KTR_CRITICAL, "critical_enter by thread %p (%ld, %s) to %d", td, 194 (long)td->td_proc->p_pid, td->td_name, td->td_critnest); 195 } 196 197 void 198 critical_exit(void) 199 { 200 struct thread *td; 201 int flags; 202 203 td = curthread; 204 KASSERT(td->td_critnest != 0, 205 ("critical_exit: td_critnest == 0")); 206 207 if (td->td_critnest == 1) { 208 td->td_critnest = 0; 209 210 /* 211 * Interrupt handlers execute critical_exit() on 212 * leave, and td_owepreempt may be left set by an 213 * interrupt handler only when td_critnest > 0. If we 214 * are decrementing td_critnest from 1 to 0, read 215 * td_owepreempt after decrementing, to not miss the 216 * preempt. Disallow compiler to reorder operations. 217 */ 218 __compiler_membar(); 219 if (td->td_owepreempt && !kdb_active) { 220 /* 221 * Microoptimization: we committed to switch, 222 * disable preemption in interrupt handlers 223 * while spinning for the thread lock. 224 */ 225 td->td_critnest = 1; 226 thread_lock(td); 227 td->td_critnest--; 228 flags = SW_INVOL | SW_PREEMPT; 229 if (TD_IS_IDLETHREAD(td)) 230 flags |= SWT_IDLE; 231 else 232 flags |= SWT_OWEPREEMPT; 233 mi_switch(flags, NULL); 234 thread_unlock(td); 235 } 236 } else 237 td->td_critnest--; 238 239 CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td, 240 (long)td->td_proc->p_pid, td->td_name, td->td_critnest); 241 } 242 243 /************************************************************************ 244 * SYSTEM RUN QUEUE manipulations and tests * 245 ************************************************************************/ 246 /* 247 * Initialize a run structure. 248 */ 249 void 250 runq_init(struct runq *rq) 251 { 252 int i; 253 254 bzero(rq, sizeof *rq); 255 for (i = 0; i < RQ_NQS; i++) 256 TAILQ_INIT(&rq->rq_queues[i]); 257 } 258 259 /* 260 * Clear the status bit of the queue corresponding to priority level pri, 261 * indicating that it is empty. 262 */ 263 static __inline void 264 runq_clrbit(struct runq *rq, int pri) 265 { 266 struct rqbits *rqb; 267 268 rqb = &rq->rq_status; 269 CTR4(KTR_RUNQ, "runq_clrbit: bits=%#x %#x bit=%#x word=%d", 270 rqb->rqb_bits[RQB_WORD(pri)], 271 rqb->rqb_bits[RQB_WORD(pri)] & ~RQB_BIT(pri), 272 RQB_BIT(pri), RQB_WORD(pri)); 273 rqb->rqb_bits[RQB_WORD(pri)] &= ~RQB_BIT(pri); 274 } 275 276 /* 277 * Find the index of the first non-empty run queue. This is done by 278 * scanning the status bits, a set bit indicates a non-empty queue. 279 */ 280 static __inline int 281 runq_findbit(struct runq *rq) 282 { 283 struct rqbits *rqb; 284 int pri; 285 int i; 286 287 rqb = &rq->rq_status; 288 for (i = 0; i < RQB_LEN; i++) 289 if (rqb->rqb_bits[i]) { 290 pri = RQB_FFS(rqb->rqb_bits[i]) + (i << RQB_L2BPW); 291 CTR3(KTR_RUNQ, "runq_findbit: bits=%#x i=%d pri=%d", 292 rqb->rqb_bits[i], i, pri); 293 return (pri); 294 } 295 296 return (-1); 297 } 298 299 static __inline int 300 runq_findbit_from(struct runq *rq, u_char pri) 301 { 302 struct rqbits *rqb; 303 rqb_word_t mask; 304 int i; 305 306 /* 307 * Set the mask for the first word so we ignore priorities before 'pri'. 308 */ 309 mask = (rqb_word_t)-1 << (pri & (RQB_BPW - 1)); 310 rqb = &rq->rq_status; 311 again: 312 for (i = RQB_WORD(pri); i < RQB_LEN; mask = -1, i++) { 313 mask = rqb->rqb_bits[i] & mask; 314 if (mask == 0) 315 continue; 316 pri = RQB_FFS(mask) + (i << RQB_L2BPW); 317 CTR3(KTR_RUNQ, "runq_findbit_from: bits=%#x i=%d pri=%d", 318 mask, i, pri); 319 return (pri); 320 } 321 if (pri == 0) 322 return (-1); 323 /* 324 * Wrap back around to the beginning of the list just once so we 325 * scan the whole thing. 326 */ 327 pri = 0; 328 goto again; 329 } 330 331 /* 332 * Set the status bit of the queue corresponding to priority level pri, 333 * indicating that it is non-empty. 334 */ 335 static __inline void 336 runq_setbit(struct runq *rq, int pri) 337 { 338 struct rqbits *rqb; 339 340 rqb = &rq->rq_status; 341 CTR4(KTR_RUNQ, "runq_setbit: bits=%#x %#x bit=%#x word=%d", 342 rqb->rqb_bits[RQB_WORD(pri)], 343 rqb->rqb_bits[RQB_WORD(pri)] | RQB_BIT(pri), 344 RQB_BIT(pri), RQB_WORD(pri)); 345 rqb->rqb_bits[RQB_WORD(pri)] |= RQB_BIT(pri); 346 } 347 348 /* 349 * Add the thread to the queue specified by its priority, and set the 350 * corresponding status bit. 351 */ 352 void 353 runq_add(struct runq *rq, struct thread *td, int flags) 354 { 355 struct rqhead *rqh; 356 int pri; 357 358 pri = td->td_priority / RQ_PPQ; 359 td->td_rqindex = pri; 360 runq_setbit(rq, pri); 361 rqh = &rq->rq_queues[pri]; 362 CTR4(KTR_RUNQ, "runq_add: td=%p pri=%d %d rqh=%p", 363 td, td->td_priority, pri, rqh); 364 if (flags & SRQ_PREEMPTED) { 365 TAILQ_INSERT_HEAD(rqh, td, td_runq); 366 } else { 367 TAILQ_INSERT_TAIL(rqh, td, td_runq); 368 } 369 } 370 371 void 372 runq_add_pri(struct runq *rq, struct thread *td, u_char pri, int flags) 373 { 374 struct rqhead *rqh; 375 376 KASSERT(pri < RQ_NQS, ("runq_add_pri: %d out of range", pri)); 377 td->td_rqindex = pri; 378 runq_setbit(rq, pri); 379 rqh = &rq->rq_queues[pri]; 380 CTR4(KTR_RUNQ, "runq_add_pri: td=%p pri=%d idx=%d rqh=%p", 381 td, td->td_priority, pri, rqh); 382 if (flags & SRQ_PREEMPTED) { 383 TAILQ_INSERT_HEAD(rqh, td, td_runq); 384 } else { 385 TAILQ_INSERT_TAIL(rqh, td, td_runq); 386 } 387 } 388 /* 389 * Return true if there are runnable processes of any priority on the run 390 * queue, false otherwise. Has no side effects, does not modify the run 391 * queue structure. 392 */ 393 int 394 runq_check(struct runq *rq) 395 { 396 struct rqbits *rqb; 397 int i; 398 399 rqb = &rq->rq_status; 400 for (i = 0; i < RQB_LEN; i++) 401 if (rqb->rqb_bits[i]) { 402 CTR2(KTR_RUNQ, "runq_check: bits=%#x i=%d", 403 rqb->rqb_bits[i], i); 404 return (1); 405 } 406 CTR0(KTR_RUNQ, "runq_check: empty"); 407 408 return (0); 409 } 410 411 /* 412 * Find the highest priority process on the run queue. 413 */ 414 struct thread * 415 runq_choose_fuzz(struct runq *rq, int fuzz) 416 { 417 struct rqhead *rqh; 418 struct thread *td; 419 int pri; 420 421 while ((pri = runq_findbit(rq)) != -1) { 422 rqh = &rq->rq_queues[pri]; 423 /* fuzz == 1 is normal.. 0 or less are ignored */ 424 if (fuzz > 1) { 425 /* 426 * In the first couple of entries, check if 427 * there is one for our CPU as a preference. 428 */ 429 int count = fuzz; 430 int cpu = PCPU_GET(cpuid); 431 struct thread *td2; 432 td2 = td = TAILQ_FIRST(rqh); 433 434 while (count-- && td2) { 435 if (td2->td_lastcpu == cpu) { 436 td = td2; 437 break; 438 } 439 td2 = TAILQ_NEXT(td2, td_runq); 440 } 441 } else 442 td = TAILQ_FIRST(rqh); 443 KASSERT(td != NULL, ("runq_choose_fuzz: no proc on busy queue")); 444 CTR3(KTR_RUNQ, 445 "runq_choose_fuzz: pri=%d thread=%p rqh=%p", pri, td, rqh); 446 return (td); 447 } 448 CTR1(KTR_RUNQ, "runq_choose_fuzz: idleproc pri=%d", pri); 449 450 return (NULL); 451 } 452 453 /* 454 * Find the highest priority process on the run queue. 455 */ 456 struct thread * 457 runq_choose(struct runq *rq) 458 { 459 struct rqhead *rqh; 460 struct thread *td; 461 int pri; 462 463 while ((pri = runq_findbit(rq)) != -1) { 464 rqh = &rq->rq_queues[pri]; 465 td = TAILQ_FIRST(rqh); 466 KASSERT(td != NULL, ("runq_choose: no thread on busy queue")); 467 CTR3(KTR_RUNQ, 468 "runq_choose: pri=%d thread=%p rqh=%p", pri, td, rqh); 469 return (td); 470 } 471 CTR1(KTR_RUNQ, "runq_choose: idlethread pri=%d", pri); 472 473 return (NULL); 474 } 475 476 struct thread * 477 runq_choose_from(struct runq *rq, u_char idx) 478 { 479 struct rqhead *rqh; 480 struct thread *td; 481 int pri; 482 483 if ((pri = runq_findbit_from(rq, idx)) != -1) { 484 rqh = &rq->rq_queues[pri]; 485 td = TAILQ_FIRST(rqh); 486 KASSERT(td != NULL, ("runq_choose: no thread on busy queue")); 487 CTR4(KTR_RUNQ, 488 "runq_choose_from: pri=%d thread=%p idx=%d rqh=%p", 489 pri, td, td->td_rqindex, rqh); 490 return (td); 491 } 492 CTR1(KTR_RUNQ, "runq_choose_from: idlethread pri=%d", pri); 493 494 return (NULL); 495 } 496 /* 497 * Remove the thread from the queue specified by its priority, and clear the 498 * corresponding status bit if the queue becomes empty. 499 * Caller must set state afterwards. 500 */ 501 void 502 runq_remove(struct runq *rq, struct thread *td) 503 { 504 505 runq_remove_idx(rq, td, NULL); 506 } 507 508 void 509 runq_remove_idx(struct runq *rq, struct thread *td, u_char *idx) 510 { 511 struct rqhead *rqh; 512 u_char pri; 513 514 KASSERT(td->td_flags & TDF_INMEM, 515 ("runq_remove_idx: thread swapped out")); 516 pri = td->td_rqindex; 517 KASSERT(pri < RQ_NQS, ("runq_remove_idx: Invalid index %d\n", pri)); 518 rqh = &rq->rq_queues[pri]; 519 CTR4(KTR_RUNQ, "runq_remove_idx: td=%p, pri=%d %d rqh=%p", 520 td, td->td_priority, pri, rqh); 521 TAILQ_REMOVE(rqh, td, td_runq); 522 if (TAILQ_EMPTY(rqh)) { 523 CTR0(KTR_RUNQ, "runq_remove_idx: empty"); 524 runq_clrbit(rq, pri); 525 if (idx != NULL && *idx == pri) 526 *idx = (pri + 1) % RQ_NQS; 527 } 528 } 529