1 /* $OpenBSD: kern_timeout.c,v 1.101 2025/01/13 03:21:10 mvs Exp $ */ 2 /* 3 * Copyright (c) 2001 Thomas Nordin <nordin@openbsd.org> 4 * Copyright (c) 2000-2001 Artur Grabowski <art@openbsd.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. The name of the author may not be used to endorse or promote products 14 * derived from this software without specific prior written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, 17 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY 18 * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL 19 * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 20 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 22 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 23 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 24 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 25 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #include <sys/param.h> 29 #include <sys/systm.h> 30 #include <sys/kthread.h> 31 #include <sys/proc.h> 32 #include <sys/timeout.h> 33 #include <sys/mutex.h> 34 #include <sys/kernel.h> 35 #include <sys/queue.h> /* _Q_INVALIDATE */ 36 #include <sys/sysctl.h> 37 #include <sys/witness.h> 38 39 #ifdef DDB 40 #include <machine/db_machdep.h> 41 #include <ddb/db_interface.h> 42 #include <ddb/db_sym.h> 43 #include <ddb/db_output.h> 44 #endif 45 46 #include "kcov.h" 47 #if NKCOV > 0 48 #include <sys/kcov.h> 49 #endif 50 51 /* 52 * Locks used to protect global variables in this file: 53 * 54 * I immutable after initialization 55 * T timeout_mutex 56 */ 57 struct mutex timeout_mutex = MUTEX_INITIALIZER(IPL_HIGH); 58 59 void *softclock_si; /* [I] softclock() interrupt handle */ 60 struct timeoutstat tostat; /* [T] statistics and totals */ 61 62 /* 63 * Timeouts are kept in a hierarchical timing wheel. The to_time is the value 64 * of the global variable "ticks" when the timeout should be called. There are 65 * four levels with 256 buckets each. 66 */ 67 #define WHEELCOUNT 4 68 #define WHEELSIZE 256 69 #define WHEELMASK 255 70 #define WHEELBITS 8 71 #define BUCKETS (WHEELCOUNT * WHEELSIZE) 72 73 struct circq timeout_wheel[BUCKETS]; /* [T] Tick-based timeouts */ 74 struct circq timeout_wheel_kc[BUCKETS]; /* [T] Clock-based timeouts */ 75 struct circq timeout_new; /* [T] New, unscheduled timeouts */ 76 struct circq timeout_todo; /* [T] Due or needs rescheduling */ 77 struct circq timeout_proc; /* [T] Due + needs process context */ 78 #ifdef MULTIPROCESSOR 79 struct circq timeout_proc_mp; /* [T] Process ctx + no kernel lock */ 80 #endif 81 82 time_t timeout_level_width[WHEELCOUNT]; /* [I] Wheel level width (seconds) */ 83 struct timespec tick_ts; /* [I] Length of a tick (1/hz secs) */ 84 85 struct kclock { 86 struct timespec kc_lastscan; /* [T] Clock time at last wheel scan */ 87 struct timespec kc_late; /* [T] Late if due prior */ 88 struct timespec kc_offset; /* [T] Offset from primary kclock */ 89 } timeout_kclock[KCLOCK_MAX]; 90 91 #define MASKWHEEL(wheel, time) (((time) >> ((wheel)*WHEELBITS)) & WHEELMASK) 92 93 #define BUCKET(rel, abs) \ 94 (timeout_wheel[ \ 95 ((rel) <= (1 << (2*WHEELBITS))) \ 96 ? ((rel) <= (1 << WHEELBITS)) \ 97 ? MASKWHEEL(0, (abs)) \ 98 : MASKWHEEL(1, (abs)) + WHEELSIZE \ 99 : ((rel) <= (1 << (3*WHEELBITS))) \ 100 ? MASKWHEEL(2, (abs)) + 2*WHEELSIZE \ 101 : MASKWHEEL(3, (abs)) + 3*WHEELSIZE]) 102 103 #define MOVEBUCKET(wheel, time) \ 104 CIRCQ_CONCAT(&timeout_todo, \ 105 &timeout_wheel[MASKWHEEL((wheel), (time)) + (wheel)*WHEELSIZE]) 106 107 /* 108 * Circular queue definitions. 109 */ 110 111 #define CIRCQ_INIT(elem) do { \ 112 (elem)->next = (elem); \ 113 (elem)->prev = (elem); \ 114 } while (0) 115 116 #define CIRCQ_INSERT_TAIL(list, elem) do { \ 117 (elem)->prev = (list)->prev; \ 118 (elem)->next = (list); \ 119 (list)->prev->next = (elem); \ 120 (list)->prev = (elem); \ 121 tostat.tos_pending++; \ 122 } while (0) 123 124 #define CIRCQ_CONCAT(fst, snd) do { \ 125 if (!CIRCQ_EMPTY(snd)) { \ 126 (fst)->prev->next = (snd)->next;\ 127 (snd)->next->prev = (fst)->prev;\ 128 (snd)->prev->next = (fst); \ 129 (fst)->prev = (snd)->prev; \ 130 CIRCQ_INIT(snd); \ 131 } \ 132 } while (0) 133 134 #define CIRCQ_REMOVE(elem) do { \ 135 (elem)->next->prev = (elem)->prev; \ 136 (elem)->prev->next = (elem)->next; \ 137 _Q_INVALIDATE((elem)->prev); \ 138 _Q_INVALIDATE((elem)->next); \ 139 tostat.tos_pending--; \ 140 } while (0) 141 142 #define CIRCQ_FIRST(elem) ((elem)->next) 143 144 #define CIRCQ_EMPTY(elem) (CIRCQ_FIRST(elem) == (elem)) 145 146 #define CIRCQ_FOREACH(elem, list) \ 147 for ((elem) = CIRCQ_FIRST(list); \ 148 (elem) != (list); \ 149 (elem) = CIRCQ_FIRST(elem)) 150 151 #ifdef WITNESS 152 struct lock_object timeout_sleeplock_obj = { 153 .lo_name = "timeout", 154 .lo_flags = LO_WITNESS | LO_INITIALIZED | LO_SLEEPABLE | 155 (LO_CLASS_RWLOCK << LO_CLASSSHIFT) 156 }; 157 struct lock_object timeout_spinlock_obj = { 158 .lo_name = "timeout", 159 .lo_flags = LO_WITNESS | LO_INITIALIZED | 160 (LO_CLASS_MUTEX << LO_CLASSSHIFT) 161 }; 162 struct lock_type timeout_sleeplock_type = { 163 .lt_name = "timeout" 164 }; 165 struct lock_type timeout_spinlock_type = { 166 .lt_name = "timeout" 167 }; 168 #define TIMEOUT_LOCK_OBJ(needsproc) \ 169 ((needsproc) ? &timeout_sleeplock_obj : &timeout_spinlock_obj) 170 #endif 171 172 void softclock(void *); 173 void softclock_create_thread(void *); 174 void softclock_process_kclock_timeout(struct timeout *, int); 175 void softclock_process_tick_timeout(struct timeout *, int); 176 void softclock_thread(void *); 177 #ifdef MULTIPROCESSOR 178 void softclock_thread_mp(void *); 179 #endif 180 void timeout_barrier_timeout(void *); 181 uint32_t timeout_bucket(const struct timeout *); 182 uint32_t timeout_maskwheel(uint32_t, const struct timespec *); 183 void timeout_run(struct timeout *); 184 185 /* 186 * The first thing in a struct timeout is its struct circq, so we 187 * can get back from a pointer to the latter to a pointer to the 188 * whole timeout with just a cast. 189 */ 190 static inline struct timeout * 191 timeout_from_circq(struct circq *p) 192 { 193 return ((struct timeout *)(p)); 194 } 195 196 static inline void 197 timeout_sync_order(int needsproc) 198 { 199 WITNESS_CHECKORDER(TIMEOUT_LOCK_OBJ(needsproc), LOP_NEWORDER, NULL); 200 } 201 202 static inline void 203 timeout_sync_enter(int needsproc) 204 { 205 timeout_sync_order(needsproc); 206 WITNESS_LOCK(TIMEOUT_LOCK_OBJ(needsproc), 0); 207 } 208 209 static inline void 210 timeout_sync_leave(int needsproc) 211 { 212 WITNESS_UNLOCK(TIMEOUT_LOCK_OBJ(needsproc), 0); 213 } 214 215 /* 216 * Some of the "math" in here is a bit tricky. 217 * 218 * We have to beware of wrapping ints. 219 * We use the fact that any element added to the queue must be added with a 220 * positive time. That means that any element `to' on the queue cannot be 221 * scheduled to timeout further in time than INT_MAX, but to->to_time can 222 * be positive or negative so comparing it with anything is dangerous. 223 * The only way we can use the to->to_time value in any predictable way 224 * is when we calculate how far in the future `to' will timeout - 225 * "to->to_time - ticks". The result will always be positive for future 226 * timeouts and 0 or negative for due timeouts. 227 */ 228 229 void 230 timeout_startup(void) 231 { 232 int b, level; 233 234 CIRCQ_INIT(&timeout_new); 235 CIRCQ_INIT(&timeout_todo); 236 CIRCQ_INIT(&timeout_proc); 237 #ifdef MULTIPROCESSOR 238 CIRCQ_INIT(&timeout_proc_mp); 239 #endif 240 for (b = 0; b < nitems(timeout_wheel); b++) 241 CIRCQ_INIT(&timeout_wheel[b]); 242 for (b = 0; b < nitems(timeout_wheel_kc); b++) 243 CIRCQ_INIT(&timeout_wheel_kc[b]); 244 245 for (level = 0; level < nitems(timeout_level_width); level++) 246 timeout_level_width[level] = 2 << (level * WHEELBITS); 247 NSEC_TO_TIMESPEC(tick_nsec, &tick_ts); 248 } 249 250 void 251 timeout_proc_init(void) 252 { 253 softclock_si = softintr_establish(IPL_SOFTCLOCK, softclock, NULL); 254 if (softclock_si == NULL) 255 panic("%s: unable to register softclock interrupt", __func__); 256 257 WITNESS_INIT(&timeout_sleeplock_obj, &timeout_sleeplock_type); 258 WITNESS_INIT(&timeout_spinlock_obj, &timeout_spinlock_type); 259 260 kthread_create_deferred(softclock_create_thread, NULL); 261 } 262 263 void 264 timeout_set(struct timeout *new, void (*fn)(void *), void *arg) 265 { 266 timeout_set_flags(new, fn, arg, KCLOCK_NONE, 0); 267 } 268 269 void 270 timeout_set_flags(struct timeout *to, void (*fn)(void *), void *arg, int kclock, 271 int flags) 272 { 273 KASSERT(!ISSET(flags, ~(TIMEOUT_PROC | TIMEOUT_MPSAFE))); 274 KASSERT(kclock >= KCLOCK_NONE && kclock < KCLOCK_MAX); 275 276 to->to_func = fn; 277 to->to_arg = arg; 278 to->to_kclock = kclock; 279 to->to_flags = flags | TIMEOUT_INITIALIZED; 280 281 /* For now, only process context timeouts may be marked MP-safe. */ 282 if (ISSET(to->to_flags, TIMEOUT_MPSAFE)) 283 KASSERT(ISSET(to->to_flags, TIMEOUT_PROC)); 284 } 285 286 void 287 timeout_set_proc(struct timeout *new, void (*fn)(void *), void *arg) 288 { 289 timeout_set_flags(new, fn, arg, KCLOCK_NONE, TIMEOUT_PROC); 290 } 291 292 int 293 timeout_add(struct timeout *new, int to_ticks) 294 { 295 int old_time; 296 int ret = 1; 297 298 KASSERT(ISSET(new->to_flags, TIMEOUT_INITIALIZED)); 299 KASSERT(new->to_kclock == KCLOCK_NONE); 300 KASSERT(to_ticks >= 0); 301 302 mtx_enter(&timeout_mutex); 303 304 /* Initialize the time here, it won't change. */ 305 old_time = new->to_time; 306 new->to_time = to_ticks + ticks; 307 CLR(new->to_flags, TIMEOUT_TRIGGERED); 308 309 /* 310 * If this timeout already is scheduled and now is moved 311 * earlier, reschedule it now. Otherwise leave it in place 312 * and let it be rescheduled later. 313 */ 314 if (ISSET(new->to_flags, TIMEOUT_ONQUEUE)) { 315 if (new->to_time - ticks < old_time - ticks) { 316 CIRCQ_REMOVE(&new->to_list); 317 CIRCQ_INSERT_TAIL(&timeout_new, &new->to_list); 318 } 319 tostat.tos_readded++; 320 ret = 0; 321 } else { 322 SET(new->to_flags, TIMEOUT_ONQUEUE); 323 CIRCQ_INSERT_TAIL(&timeout_new, &new->to_list); 324 } 325 #if NKCOV > 0 326 if (!kcov_cold) 327 new->to_process = curproc->p_p; 328 #endif 329 tostat.tos_added++; 330 mtx_leave(&timeout_mutex); 331 332 return ret; 333 } 334 335 static inline int 336 timeout_add_ticks(struct timeout *to, uint64_t to_ticks, int notzero) 337 { 338 if (to_ticks > INT_MAX) 339 to_ticks = INT_MAX; 340 else if (to_ticks == 0 && notzero) 341 to_ticks = 1; 342 343 return timeout_add(to, (int)to_ticks); 344 } 345 346 int 347 timeout_add_tv(struct timeout *to, const struct timeval *tv) 348 { 349 uint64_t to_ticks; 350 351 to_ticks = (uint64_t)hz * tv->tv_sec + tv->tv_usec / tick; 352 353 return timeout_add_ticks(to, to_ticks, tv->tv_usec > 0); 354 } 355 356 int 357 timeout_add_sec(struct timeout *to, int secs) 358 { 359 uint64_t to_ticks; 360 361 to_ticks = (uint64_t)hz * secs; 362 363 return timeout_add_ticks(to, to_ticks, 1); 364 } 365 366 int 367 timeout_add_msec(struct timeout *to, uint64_t msecs) 368 { 369 uint64_t to_ticks; 370 371 to_ticks = msecs * 1000 / tick; 372 373 return timeout_add_ticks(to, to_ticks, msecs > 0); 374 } 375 376 int 377 timeout_add_usec(struct timeout *to, uint64_t usecs) 378 { 379 uint64_t to_ticks; 380 381 to_ticks = usecs / tick; 382 383 return timeout_add_ticks(to, to_ticks, usecs > 0); 384 } 385 386 int 387 timeout_add_nsec(struct timeout *to, uint64_t nsecs) 388 { 389 uint64_t to_ticks; 390 391 to_ticks = nsecs / (tick * 1000); 392 393 return timeout_add_ticks(to, to_ticks, nsecs > 0); 394 } 395 396 int 397 timeout_abs_ts(struct timeout *to, const struct timespec *abstime) 398 { 399 struct timespec old_abstime; 400 int ret = 1; 401 402 mtx_enter(&timeout_mutex); 403 404 KASSERT(ISSET(to->to_flags, TIMEOUT_INITIALIZED)); 405 KASSERT(to->to_kclock == KCLOCK_UPTIME); 406 407 old_abstime = to->to_abstime; 408 to->to_abstime = *abstime; 409 CLR(to->to_flags, TIMEOUT_TRIGGERED); 410 411 if (ISSET(to->to_flags, TIMEOUT_ONQUEUE)) { 412 if (timespeccmp(abstime, &old_abstime, <)) { 413 CIRCQ_REMOVE(&to->to_list); 414 CIRCQ_INSERT_TAIL(&timeout_new, &to->to_list); 415 } 416 tostat.tos_readded++; 417 ret = 0; 418 } else { 419 SET(to->to_flags, TIMEOUT_ONQUEUE); 420 CIRCQ_INSERT_TAIL(&timeout_new, &to->to_list); 421 } 422 #if NKCOV > 0 423 if (!kcov_cold) 424 to->to_process = curproc->p_p; 425 #endif 426 tostat.tos_added++; 427 428 mtx_leave(&timeout_mutex); 429 430 return ret; 431 } 432 433 int 434 timeout_del(struct timeout *to) 435 { 436 int ret = 0; 437 438 mtx_enter(&timeout_mutex); 439 if (ISSET(to->to_flags, TIMEOUT_ONQUEUE)) { 440 CIRCQ_REMOVE(&to->to_list); 441 CLR(to->to_flags, TIMEOUT_ONQUEUE); 442 tostat.tos_cancelled++; 443 ret = 1; 444 } 445 CLR(to->to_flags, TIMEOUT_TRIGGERED); 446 tostat.tos_deleted++; 447 mtx_leave(&timeout_mutex); 448 449 return ret; 450 } 451 452 int 453 timeout_del_barrier(struct timeout *to) 454 { 455 int removed; 456 457 timeout_sync_order(ISSET(to->to_flags, TIMEOUT_PROC)); 458 459 removed = timeout_del(to); 460 timeout_barrier(to); 461 462 return removed; 463 } 464 465 void 466 timeout_barrier(struct timeout *to) 467 { 468 struct timeout barrier; 469 struct cond c; 470 int flags; 471 472 flags = to->to_flags & (TIMEOUT_PROC | TIMEOUT_MPSAFE); 473 timeout_sync_order(ISSET(flags, TIMEOUT_PROC)); 474 475 timeout_set_flags(&barrier, timeout_barrier_timeout, &c, KCLOCK_NONE, 476 flags); 477 barrier.to_process = curproc->p_p; 478 cond_init(&c); 479 480 mtx_enter(&timeout_mutex); 481 482 barrier.to_time = ticks; 483 SET(barrier.to_flags, TIMEOUT_ONQUEUE); 484 if (ISSET(flags, TIMEOUT_PROC)) { 485 #ifdef MULTIPROCESSOR 486 if (ISSET(flags, TIMEOUT_MPSAFE)) 487 CIRCQ_INSERT_TAIL(&timeout_proc_mp, &barrier.to_list); 488 else 489 #endif 490 CIRCQ_INSERT_TAIL(&timeout_proc, &barrier.to_list); 491 } else 492 CIRCQ_INSERT_TAIL(&timeout_todo, &barrier.to_list); 493 494 mtx_leave(&timeout_mutex); 495 496 if (ISSET(flags, TIMEOUT_PROC)) { 497 #ifdef MULTIPROCESSOR 498 if (ISSET(flags, TIMEOUT_MPSAFE)) 499 wakeup_one(&timeout_proc_mp); 500 else 501 #endif 502 wakeup_one(&timeout_proc); 503 } else 504 softintr_schedule(softclock_si); 505 506 cond_wait(&c, "tmobar"); 507 } 508 509 void 510 timeout_barrier_timeout(void *arg) 511 { 512 struct cond *c = arg; 513 514 cond_signal(c); 515 } 516 517 uint32_t 518 timeout_bucket(const struct timeout *to) 519 { 520 struct timespec diff, shifted_abstime; 521 struct kclock *kc; 522 uint32_t level; 523 524 KASSERT(to->to_kclock == KCLOCK_UPTIME); 525 kc = &timeout_kclock[to->to_kclock]; 526 527 KASSERT(timespeccmp(&kc->kc_lastscan, &to->to_abstime, <)); 528 timespecsub(&to->to_abstime, &kc->kc_lastscan, &diff); 529 for (level = 0; level < nitems(timeout_level_width) - 1; level++) { 530 if (diff.tv_sec < timeout_level_width[level]) 531 break; 532 } 533 timespecadd(&to->to_abstime, &kc->kc_offset, &shifted_abstime); 534 return level * WHEELSIZE + timeout_maskwheel(level, &shifted_abstime); 535 } 536 537 /* 538 * Hash the absolute time into a bucket on a given level of the wheel. 539 * 540 * The complete hash is 32 bits. The upper 25 bits are seconds, the 541 * lower 7 bits are nanoseconds. tv_nsec is a positive value less 542 * than one billion so we need to divide it to isolate the desired 543 * bits. We can't just shift it. 544 * 545 * The level is used to isolate an 8-bit portion of the hash. The 546 * resulting number indicates which bucket the absolute time belongs 547 * in on the given level of the wheel. 548 */ 549 uint32_t 550 timeout_maskwheel(uint32_t level, const struct timespec *abstime) 551 { 552 uint32_t hi, lo; 553 554 hi = abstime->tv_sec << 7; 555 lo = abstime->tv_nsec / 7812500; 556 557 return ((hi | lo) >> (level * WHEELBITS)) & WHEELMASK; 558 } 559 560 /* 561 * This is called from hardclock() on the primary CPU at the start of 562 * every tick. 563 */ 564 void 565 timeout_hardclock_update(void) 566 { 567 struct timespec elapsed, now; 568 struct kclock *kc; 569 struct timespec *lastscan = &timeout_kclock[KCLOCK_UPTIME].kc_lastscan; 570 int b, done, first, i, last, level, need_softclock = 1, off; 571 572 mtx_enter(&timeout_mutex); 573 574 MOVEBUCKET(0, ticks); 575 if (MASKWHEEL(0, ticks) == 0) { 576 MOVEBUCKET(1, ticks); 577 if (MASKWHEEL(1, ticks) == 0) { 578 MOVEBUCKET(2, ticks); 579 if (MASKWHEEL(2, ticks) == 0) 580 MOVEBUCKET(3, ticks); 581 } 582 } 583 584 /* 585 * Dump the buckets that expired while we were away. 586 * 587 * If the elapsed time has exceeded a level's limit then we need 588 * to dump every bucket in the level. We have necessarily completed 589 * a lap of that level, too, so we need to process buckets in the 590 * next level. 591 * 592 * Otherwise we need to compare indices: if the index of the first 593 * expired bucket is greater than that of the last then we have 594 * completed a lap of the level and need to process buckets in the 595 * next level. 596 */ 597 nanouptime(&now); 598 timespecsub(&now, lastscan, &elapsed); 599 for (level = 0; level < nitems(timeout_level_width); level++) { 600 first = timeout_maskwheel(level, lastscan); 601 if (elapsed.tv_sec >= timeout_level_width[level]) { 602 last = (first == 0) ? WHEELSIZE - 1 : first - 1; 603 done = 0; 604 } else { 605 last = timeout_maskwheel(level, &now); 606 done = first <= last; 607 } 608 off = level * WHEELSIZE; 609 for (b = first;; b = (b + 1) % WHEELSIZE) { 610 CIRCQ_CONCAT(&timeout_todo, &timeout_wheel_kc[off + b]); 611 if (b == last) 612 break; 613 } 614 if (done) 615 break; 616 } 617 618 /* 619 * Update the cached state for each kclock. 620 */ 621 for (i = 0; i < nitems(timeout_kclock); i++) { 622 kc = &timeout_kclock[i]; 623 timespecadd(&now, &kc->kc_offset, &kc->kc_lastscan); 624 timespecsub(&kc->kc_lastscan, &tick_ts, &kc->kc_late); 625 } 626 627 if (CIRCQ_EMPTY(&timeout_new) && CIRCQ_EMPTY(&timeout_todo)) 628 need_softclock = 0; 629 630 mtx_leave(&timeout_mutex); 631 632 if (need_softclock) 633 softintr_schedule(softclock_si); 634 } 635 636 void 637 timeout_run(struct timeout *to) 638 { 639 void (*fn)(void *); 640 void *arg; 641 int needsproc; 642 643 MUTEX_ASSERT_LOCKED(&timeout_mutex); 644 645 CLR(to->to_flags, TIMEOUT_ONQUEUE); 646 SET(to->to_flags, TIMEOUT_TRIGGERED); 647 648 fn = to->to_func; 649 arg = to->to_arg; 650 needsproc = ISSET(to->to_flags, TIMEOUT_PROC); 651 #if NKCOV > 0 652 struct process *kcov_process = to->to_process; 653 #endif 654 655 mtx_leave(&timeout_mutex); 656 timeout_sync_enter(needsproc); 657 #if NKCOV > 0 658 kcov_remote_enter(KCOV_REMOTE_COMMON, kcov_process); 659 #endif 660 fn(arg); 661 #if NKCOV > 0 662 kcov_remote_leave(KCOV_REMOTE_COMMON, kcov_process); 663 #endif 664 timeout_sync_leave(needsproc); 665 mtx_enter(&timeout_mutex); 666 } 667 668 void 669 softclock_process_kclock_timeout(struct timeout *to, int new) 670 { 671 struct kclock *kc = &timeout_kclock[to->to_kclock]; 672 673 if (timespeccmp(&to->to_abstime, &kc->kc_lastscan, >)) { 674 tostat.tos_scheduled++; 675 if (!new) 676 tostat.tos_rescheduled++; 677 CIRCQ_INSERT_TAIL(&timeout_wheel_kc[timeout_bucket(to)], 678 &to->to_list); 679 return; 680 } 681 if (!new && timespeccmp(&to->to_abstime, &kc->kc_late, <=)) 682 tostat.tos_late++; 683 if (ISSET(to->to_flags, TIMEOUT_PROC)) { 684 #ifdef MULTIPROCESSOR 685 if (ISSET(to->to_flags, TIMEOUT_MPSAFE)) 686 CIRCQ_INSERT_TAIL(&timeout_proc_mp, &to->to_list); 687 else 688 #endif 689 CIRCQ_INSERT_TAIL(&timeout_proc, &to->to_list); 690 return; 691 } 692 timeout_run(to); 693 tostat.tos_run_softclock++; 694 } 695 696 void 697 softclock_process_tick_timeout(struct timeout *to, int new) 698 { 699 int delta = to->to_time - ticks; 700 701 if (delta > 0) { 702 tostat.tos_scheduled++; 703 if (!new) 704 tostat.tos_rescheduled++; 705 CIRCQ_INSERT_TAIL(&BUCKET(delta, to->to_time), &to->to_list); 706 return; 707 } 708 if (!new && delta < 0) 709 tostat.tos_late++; 710 if (ISSET(to->to_flags, TIMEOUT_PROC)) { 711 #ifdef MULTIPROCESSOR 712 if (ISSET(to->to_flags, TIMEOUT_MPSAFE)) 713 CIRCQ_INSERT_TAIL(&timeout_proc_mp, &to->to_list); 714 else 715 #endif 716 CIRCQ_INSERT_TAIL(&timeout_proc, &to->to_list); 717 return; 718 } 719 timeout_run(to); 720 tostat.tos_run_softclock++; 721 } 722 723 /* 724 * Timeouts are processed here instead of timeout_hardclock_update() 725 * to avoid doing any more work at IPL_CLOCK than absolutely necessary. 726 * Down here at IPL_SOFTCLOCK other interrupts can be serviced promptly 727 * so the system remains responsive even if there is a surge of timeouts. 728 */ 729 void 730 softclock(void *arg) 731 { 732 struct timeout *first_new, *to; 733 int needsproc, new; 734 #ifdef MULTIPROCESSOR 735 int need_proc_mp; 736 #endif 737 738 first_new = NULL; 739 new = 0; 740 741 mtx_enter(&timeout_mutex); 742 if (!CIRCQ_EMPTY(&timeout_new)) 743 first_new = timeout_from_circq(CIRCQ_FIRST(&timeout_new)); 744 CIRCQ_CONCAT(&timeout_todo, &timeout_new); 745 while (!CIRCQ_EMPTY(&timeout_todo)) { 746 to = timeout_from_circq(CIRCQ_FIRST(&timeout_todo)); 747 CIRCQ_REMOVE(&to->to_list); 748 if (to == first_new) 749 new = 1; 750 if (to->to_kclock == KCLOCK_NONE) 751 softclock_process_tick_timeout(to, new); 752 else if (to->to_kclock == KCLOCK_UPTIME) 753 softclock_process_kclock_timeout(to, new); 754 else { 755 panic("%s: invalid to_clock: %d", 756 __func__, to->to_kclock); 757 } 758 } 759 tostat.tos_softclocks++; 760 needsproc = !CIRCQ_EMPTY(&timeout_proc); 761 #ifdef MULTIPROCESSOR 762 need_proc_mp = !CIRCQ_EMPTY(&timeout_proc_mp); 763 #endif 764 mtx_leave(&timeout_mutex); 765 766 if (needsproc) 767 wakeup(&timeout_proc); 768 #ifdef MULTIPROCESSOR 769 if (need_proc_mp) 770 wakeup(&timeout_proc_mp); 771 #endif 772 } 773 774 void 775 softclock_create_thread(void *arg) 776 { 777 if (kthread_create(softclock_thread, NULL, NULL, "softclock")) 778 panic("fork softclock"); 779 #ifdef MULTIPROCESSOR 780 if (kthread_create(softclock_thread_mp, NULL, NULL, "softclockmp")) 781 panic("kthread_create softclock_thread_mp"); 782 #endif 783 } 784 785 void 786 softclock_thread(void *arg) 787 { 788 CPU_INFO_ITERATOR cii; 789 struct cpu_info *ci; 790 struct timeout *to; 791 int s; 792 793 KERNEL_ASSERT_LOCKED(); 794 795 /* Be conservative for the moment */ 796 CPU_INFO_FOREACH(cii, ci) { 797 if (CPU_IS_PRIMARY(ci)) 798 break; 799 } 800 KASSERT(ci != NULL); 801 sched_peg_curproc(ci); 802 803 s = splsoftclock(); 804 mtx_enter(&timeout_mutex); 805 for (;;) { 806 while (!CIRCQ_EMPTY(&timeout_proc)) { 807 to = timeout_from_circq(CIRCQ_FIRST(&timeout_proc)); 808 CIRCQ_REMOVE(&to->to_list); 809 timeout_run(to); 810 tostat.tos_run_thread++; 811 } 812 tostat.tos_thread_wakeups++; 813 msleep_nsec(&timeout_proc, &timeout_mutex, PSWP, "tmoslp", 814 INFSLP); 815 } 816 splx(s); 817 } 818 819 #ifdef MULTIPROCESSOR 820 void 821 softclock_thread_mp(void *arg) 822 { 823 struct timeout *to; 824 825 KERNEL_ASSERT_LOCKED(); 826 KERNEL_UNLOCK(); 827 828 mtx_enter(&timeout_mutex); 829 for (;;) { 830 while (!CIRCQ_EMPTY(&timeout_proc_mp)) { 831 to = timeout_from_circq(CIRCQ_FIRST(&timeout_proc_mp)); 832 CIRCQ_REMOVE(&to->to_list); 833 timeout_run(to); 834 tostat.tos_run_thread++; 835 } 836 tostat.tos_thread_wakeups++; 837 msleep_nsec(&timeout_proc_mp, &timeout_mutex, PSWP, "tmoslp", 838 INFSLP); 839 } 840 } 841 #endif /* MULTIPROCESSOR */ 842 843 #ifndef SMALL_KERNEL 844 void 845 timeout_adjust_ticks(int adj) 846 { 847 struct timeout *to; 848 struct circq *p; 849 int new_ticks, b; 850 851 /* adjusting the monotonic clock backwards would be a Bad Thing */ 852 if (adj <= 0) 853 return; 854 855 mtx_enter(&timeout_mutex); 856 new_ticks = ticks + adj; 857 for (b = 0; b < nitems(timeout_wheel); b++) { 858 p = CIRCQ_FIRST(&timeout_wheel[b]); 859 while (p != &timeout_wheel[b]) { 860 to = timeout_from_circq(p); 861 p = CIRCQ_FIRST(p); 862 863 /* when moving a timeout forward need to reinsert it */ 864 if (to->to_time - ticks < adj) 865 to->to_time = new_ticks; 866 CIRCQ_REMOVE(&to->to_list); 867 CIRCQ_INSERT_TAIL(&timeout_todo, &to->to_list); 868 } 869 } 870 ticks = new_ticks; 871 mtx_leave(&timeout_mutex); 872 } 873 #endif 874 875 int 876 timeout_sysctl(void *oldp, size_t *oldlenp, void *newp, size_t newlen) 877 { 878 struct timeoutstat status; 879 880 mtx_enter(&timeout_mutex); 881 memcpy(&status, &tostat, sizeof(status)); 882 mtx_leave(&timeout_mutex); 883 884 return sysctl_rdstruct(oldp, oldlenp, newp, &status, sizeof(status)); 885 } 886 887 #ifdef DDB 888 const char *db_kclock(int); 889 void db_show_callout_bucket(struct circq *); 890 void db_show_timeout(struct timeout *, struct circq *); 891 const char *db_timespec(const struct timespec *); 892 893 const char * 894 db_kclock(int kclock) 895 { 896 switch (kclock) { 897 case KCLOCK_UPTIME: 898 return "uptime"; 899 default: 900 return "invalid"; 901 } 902 } 903 904 const char * 905 db_timespec(const struct timespec *ts) 906 { 907 static char buf[32]; 908 struct timespec tmp, zero; 909 910 if (ts->tv_sec >= 0) { 911 snprintf(buf, sizeof(buf), "%lld.%09ld", 912 ts->tv_sec, ts->tv_nsec); 913 return buf; 914 } 915 916 timespecclear(&zero); 917 timespecsub(&zero, ts, &tmp); 918 snprintf(buf, sizeof(buf), "-%lld.%09ld", tmp.tv_sec, tmp.tv_nsec); 919 return buf; 920 } 921 922 void 923 db_show_callout_bucket(struct circq *bucket) 924 { 925 struct circq *p; 926 927 CIRCQ_FOREACH(p, bucket) 928 db_show_timeout(timeout_from_circq(p), bucket); 929 } 930 931 void 932 db_show_timeout(struct timeout *to, struct circq *bucket) 933 { 934 struct timespec remaining; 935 struct kclock *kc; 936 char buf[8]; 937 db_expr_t offset; 938 struct circq *wheel; 939 const char *name, *where; 940 int width = sizeof(long) * 2; 941 942 db_find_sym_and_offset((vaddr_t)to->to_func, &name, &offset); 943 name = name ? name : "?"; 944 if (bucket == &timeout_new) 945 where = "new"; 946 else if (bucket == &timeout_todo) 947 where = "softint"; 948 else if (bucket == &timeout_proc) 949 where = "thread"; 950 #ifdef MULTIPROCESSOR 951 else if (bucket == &timeout_proc_mp) 952 where = "thread-mp"; 953 #endif 954 else { 955 if (to->to_kclock == KCLOCK_UPTIME) 956 wheel = timeout_wheel_kc; 957 else if (to->to_kclock == KCLOCK_NONE) 958 wheel = timeout_wheel; 959 else 960 goto invalid; 961 snprintf(buf, sizeof(buf), "%3ld/%1ld", 962 (bucket - wheel) % WHEELSIZE, 963 (bucket - wheel) / WHEELSIZE); 964 where = buf; 965 } 966 if (to->to_kclock == KCLOCK_UPTIME) { 967 kc = &timeout_kclock[to->to_kclock]; 968 timespecsub(&to->to_abstime, &kc->kc_lastscan, &remaining); 969 db_printf("%20s %8s %9s 0x%0*lx %s\n", 970 db_timespec(&remaining), db_kclock(to->to_kclock), where, 971 width, (ulong)to->to_arg, name); 972 } else if (to->to_kclock == KCLOCK_NONE) { 973 db_printf("%20d %8s %9s 0x%0*lx %s\n", 974 to->to_time - ticks, "ticks", where, 975 width, (ulong)to->to_arg, name); 976 } else 977 goto invalid; 978 return; 979 980 invalid: 981 db_printf("%s: timeout 0x%p: invalid to_kclock: %d", 982 __func__, to, to->to_kclock); 983 } 984 985 void 986 db_show_callout(db_expr_t addr, int haddr, db_expr_t count, char *modif) 987 { 988 struct kclock *kc; 989 int width = sizeof(long) * 2 + 2; 990 int b, i; 991 992 db_printf("%20s %8s\n", "lastscan", "clock"); 993 db_printf("%20d %8s\n", ticks, "ticks"); 994 for (i = 0; i < nitems(timeout_kclock); i++) { 995 kc = &timeout_kclock[i]; 996 db_printf("%20s %8s\n", 997 db_timespec(&kc->kc_lastscan), db_kclock(i)); 998 } 999 db_printf("\n"); 1000 db_printf("%20s %8s %9s %*s %s\n", 1001 "remaining", "clock", "wheel", width, "arg", "func"); 1002 db_show_callout_bucket(&timeout_new); 1003 db_show_callout_bucket(&timeout_todo); 1004 db_show_callout_bucket(&timeout_proc); 1005 #ifdef MULTIPROCESSOR 1006 db_show_callout_bucket(&timeout_proc_mp); 1007 #endif 1008 for (b = 0; b < nitems(timeout_wheel); b++) 1009 db_show_callout_bucket(&timeout_wheel[b]); 1010 for (b = 0; b < nitems(timeout_wheel_kc); b++) 1011 db_show_callout_bucket(&timeout_wheel_kc[b]); 1012 } 1013 #endif 1014