1 /* $NetBSD: kern_timeout.c,v 1.58 2020/01/23 20:44:15 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2003, 2006, 2007, 2008, 2009, 2019 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe, and by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Copyright (c) 2001 Thomas Nordin <nordin@openbsd.org> 34 * Copyright (c) 2000-2001 Artur Grabowski <art@openbsd.org> 35 * All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. The name of the author may not be used to endorse or promote products 47 * derived from this software without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, 50 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY 51 * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL 52 * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 53 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 54 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 55 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 56 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 57 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 58 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 59 */ 60 61 #include <sys/cdefs.h> 62 __KERNEL_RCSID(0, "$NetBSD: kern_timeout.c,v 1.58 2020/01/23 20:44:15 ad Exp $"); 63 64 /* 65 * Timeouts are kept in a hierarchical timing wheel. The c_time is the 66 * value of c_cpu->cc_ticks when the timeout should be called. There are 67 * four levels with 256 buckets each. See 'Scheme 7' in "Hashed and 68 * Hierarchical Timing Wheels: Efficient Data Structures for Implementing 69 * a Timer Facility" by George Varghese and Tony Lauck. 70 * 71 * Some of the "math" in here is a bit tricky. We have to beware of 72 * wrapping ints. 73 * 74 * We use the fact that any element added to the queue must be added with 75 * a positive time. That means that any element `to' on the queue cannot 76 * be scheduled to timeout further in time than INT_MAX, but c->c_time can 77 * be positive or negative so comparing it with anything is dangerous. 78 * The only way we can use the c->c_time value in any predictable way is 79 * when we calculate how far in the future `to' will timeout - "c->c_time 80 * - c->c_cpu->cc_ticks". The result will always be positive for future 81 * timeouts and 0 or negative for due timeouts. 82 */ 83 84 #define _CALLOUT_PRIVATE 85 86 #include <sys/param.h> 87 #include <sys/systm.h> 88 #include <sys/kernel.h> 89 #include <sys/callout.h> 90 #include <sys/lwp.h> 91 #include <sys/mutex.h> 92 #include <sys/proc.h> 93 #include <sys/sleepq.h> 94 #include <sys/syncobj.h> 95 #include <sys/evcnt.h> 96 #include <sys/intr.h> 97 #include <sys/cpu.h> 98 #include <sys/kmem.h> 99 100 #ifdef DDB 101 #include <machine/db_machdep.h> 102 #include <ddb/db_interface.h> 103 #include <ddb/db_access.h> 104 #include <ddb/db_cpu.h> 105 #include <ddb/db_sym.h> 106 #include <ddb/db_output.h> 107 #endif 108 109 #define BUCKETS 1024 110 #define WHEELSIZE 256 111 #define WHEELMASK 255 112 #define WHEELBITS 8 113 114 #define MASKWHEEL(wheel, time) (((time) >> ((wheel)*WHEELBITS)) & WHEELMASK) 115 116 #define BUCKET(cc, rel, abs) \ 117 (((rel) <= (1 << (2*WHEELBITS))) \ 118 ? ((rel) <= (1 << WHEELBITS)) \ 119 ? &(cc)->cc_wheel[MASKWHEEL(0, (abs))] \ 120 : &(cc)->cc_wheel[MASKWHEEL(1, (abs)) + WHEELSIZE] \ 121 : ((rel) <= (1 << (3*WHEELBITS))) \ 122 ? &(cc)->cc_wheel[MASKWHEEL(2, (abs)) + 2*WHEELSIZE] \ 123 : &(cc)->cc_wheel[MASKWHEEL(3, (abs)) + 3*WHEELSIZE]) 124 125 #define MOVEBUCKET(cc, wheel, time) \ 126 CIRCQ_APPEND(&(cc)->cc_todo, \ 127 &(cc)->cc_wheel[MASKWHEEL((wheel), (time)) + (wheel)*WHEELSIZE]) 128 129 /* 130 * Circular queue definitions. 131 */ 132 133 #define CIRCQ_INIT(list) \ 134 do { \ 135 (list)->cq_next_l = (list); \ 136 (list)->cq_prev_l = (list); \ 137 } while (/*CONSTCOND*/0) 138 139 #define CIRCQ_INSERT(elem, list) \ 140 do { \ 141 (elem)->cq_prev_e = (list)->cq_prev_e; \ 142 (elem)->cq_next_l = (list); \ 143 (list)->cq_prev_l->cq_next_l = (elem); \ 144 (list)->cq_prev_l = (elem); \ 145 } while (/*CONSTCOND*/0) 146 147 #define CIRCQ_APPEND(fst, snd) \ 148 do { \ 149 if (!CIRCQ_EMPTY(snd)) { \ 150 (fst)->cq_prev_l->cq_next_l = (snd)->cq_next_l; \ 151 (snd)->cq_next_l->cq_prev_l = (fst)->cq_prev_l; \ 152 (snd)->cq_prev_l->cq_next_l = (fst); \ 153 (fst)->cq_prev_l = (snd)->cq_prev_l; \ 154 CIRCQ_INIT(snd); \ 155 } \ 156 } while (/*CONSTCOND*/0) 157 158 #define CIRCQ_REMOVE(elem) \ 159 do { \ 160 (elem)->cq_next_l->cq_prev_e = (elem)->cq_prev_e; \ 161 (elem)->cq_prev_l->cq_next_e = (elem)->cq_next_e; \ 162 } while (/*CONSTCOND*/0) 163 164 #define CIRCQ_FIRST(list) ((list)->cq_next_e) 165 #define CIRCQ_NEXT(elem) ((elem)->cq_next_e) 166 #define CIRCQ_LAST(elem,list) ((elem)->cq_next_l == (list)) 167 #define CIRCQ_EMPTY(list) ((list)->cq_next_l == (list)) 168 169 struct callout_cpu { 170 kmutex_t *cc_lock; 171 sleepq_t cc_sleepq; 172 u_int cc_nwait; 173 u_int cc_ticks; 174 lwp_t *cc_lwp; 175 callout_impl_t *cc_active; 176 callout_impl_t *cc_cancel; 177 struct evcnt cc_ev_late; 178 struct evcnt cc_ev_block; 179 struct callout_circq cc_todo; /* Worklist */ 180 struct callout_circq cc_wheel[BUCKETS]; /* Queues of timeouts */ 181 char cc_name1[12]; 182 char cc_name2[12]; 183 }; 184 185 #ifndef CRASH 186 187 static void callout_softclock(void *); 188 static void callout_wait(callout_impl_t *, void *, kmutex_t *); 189 190 static struct callout_cpu callout_cpu0 __cacheline_aligned; 191 static void *callout_sih __read_mostly; 192 193 static inline kmutex_t * 194 callout_lock(callout_impl_t *c) 195 { 196 struct callout_cpu *cc; 197 kmutex_t *lock; 198 199 for (;;) { 200 cc = c->c_cpu; 201 lock = cc->cc_lock; 202 mutex_spin_enter(lock); 203 if (__predict_true(cc == c->c_cpu)) 204 return lock; 205 mutex_spin_exit(lock); 206 } 207 } 208 209 /* 210 * callout_startup: 211 * 212 * Initialize the callout facility, called at system startup time. 213 * Do just enough to allow callouts to be safely registered. 214 */ 215 void 216 callout_startup(void) 217 { 218 struct callout_cpu *cc; 219 int b; 220 221 KASSERT(curcpu()->ci_data.cpu_callout == NULL); 222 223 cc = &callout_cpu0; 224 cc->cc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED); 225 CIRCQ_INIT(&cc->cc_todo); 226 for (b = 0; b < BUCKETS; b++) 227 CIRCQ_INIT(&cc->cc_wheel[b]); 228 curcpu()->ci_data.cpu_callout = cc; 229 } 230 231 /* 232 * callout_init_cpu: 233 * 234 * Per-CPU initialization. 235 */ 236 CTASSERT(sizeof(callout_impl_t) <= sizeof(callout_t)); 237 238 void 239 callout_init_cpu(struct cpu_info *ci) 240 { 241 struct callout_cpu *cc; 242 int b; 243 244 if ((cc = ci->ci_data.cpu_callout) == NULL) { 245 cc = kmem_zalloc(sizeof(*cc), KM_SLEEP); 246 cc->cc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED); 247 CIRCQ_INIT(&cc->cc_todo); 248 for (b = 0; b < BUCKETS; b++) 249 CIRCQ_INIT(&cc->cc_wheel[b]); 250 } else { 251 /* Boot CPU, one time only. */ 252 callout_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE, 253 callout_softclock, NULL); 254 if (callout_sih == NULL) 255 panic("callout_init_cpu (2)"); 256 } 257 258 sleepq_init(&cc->cc_sleepq); 259 260 snprintf(cc->cc_name1, sizeof(cc->cc_name1), "late/%u", 261 cpu_index(ci)); 262 evcnt_attach_dynamic(&cc->cc_ev_late, EVCNT_TYPE_MISC, 263 NULL, "callout", cc->cc_name1); 264 265 snprintf(cc->cc_name2, sizeof(cc->cc_name2), "wait/%u", 266 cpu_index(ci)); 267 evcnt_attach_dynamic(&cc->cc_ev_block, EVCNT_TYPE_MISC, 268 NULL, "callout", cc->cc_name2); 269 270 ci->ci_data.cpu_callout = cc; 271 } 272 273 /* 274 * callout_init: 275 * 276 * Initialize a callout structure. This must be quick, so we fill 277 * only the minimum number of fields. 278 */ 279 void 280 callout_init(callout_t *cs, u_int flags) 281 { 282 callout_impl_t *c = (callout_impl_t *)cs; 283 struct callout_cpu *cc; 284 285 KASSERT((flags & ~CALLOUT_FLAGMASK) == 0); 286 287 cc = curcpu()->ci_data.cpu_callout; 288 c->c_func = NULL; 289 c->c_magic = CALLOUT_MAGIC; 290 if (__predict_true((flags & CALLOUT_MPSAFE) != 0 && cc != NULL)) { 291 c->c_flags = flags; 292 c->c_cpu = cc; 293 return; 294 } 295 c->c_flags = flags | CALLOUT_BOUND; 296 c->c_cpu = &callout_cpu0; 297 } 298 299 /* 300 * callout_destroy: 301 * 302 * Destroy a callout structure. The callout must be stopped. 303 */ 304 void 305 callout_destroy(callout_t *cs) 306 { 307 callout_impl_t *c = (callout_impl_t *)cs; 308 309 KASSERTMSG(c->c_magic == CALLOUT_MAGIC, 310 "callout %p: c_magic (%#x) != CALLOUT_MAGIC (%#x)", 311 c, c->c_magic, CALLOUT_MAGIC); 312 /* 313 * It's not necessary to lock in order to see the correct value 314 * of c->c_flags. If the callout could potentially have been 315 * running, the current thread should have stopped it. 316 */ 317 KASSERTMSG((c->c_flags & CALLOUT_PENDING) == 0, 318 "callout %p: c_func (%p) c_flags (%#x) destroyed from %p", 319 c, c->c_func, c->c_flags, __builtin_return_address(0)); 320 KASSERT(c->c_cpu->cc_lwp == curlwp || c->c_cpu->cc_active != c); 321 c->c_magic = 0; 322 } 323 324 /* 325 * callout_schedule_locked: 326 * 327 * Schedule a callout to run. The function and argument must 328 * already be set in the callout structure. Must be called with 329 * callout_lock. 330 */ 331 static void 332 callout_schedule_locked(callout_impl_t *c, kmutex_t *lock, int to_ticks) 333 { 334 struct callout_cpu *cc, *occ; 335 int old_time; 336 337 KASSERT(to_ticks >= 0); 338 KASSERT(c->c_func != NULL); 339 340 /* Initialize the time here, it won't change. */ 341 occ = c->c_cpu; 342 c->c_flags &= ~(CALLOUT_FIRED | CALLOUT_INVOKING); 343 344 /* 345 * If this timeout is already scheduled and now is moved 346 * earlier, reschedule it now. Otherwise leave it in place 347 * and let it be rescheduled later. 348 */ 349 if ((c->c_flags & CALLOUT_PENDING) != 0) { 350 /* Leave on existing CPU. */ 351 old_time = c->c_time; 352 c->c_time = to_ticks + occ->cc_ticks; 353 if (c->c_time - old_time < 0) { 354 CIRCQ_REMOVE(&c->c_list); 355 CIRCQ_INSERT(&c->c_list, &occ->cc_todo); 356 } 357 mutex_spin_exit(lock); 358 return; 359 } 360 361 cc = curcpu()->ci_data.cpu_callout; 362 if ((c->c_flags & CALLOUT_BOUND) != 0 || cc == occ || 363 !mutex_tryenter(cc->cc_lock)) { 364 /* Leave on existing CPU. */ 365 c->c_time = to_ticks + occ->cc_ticks; 366 c->c_flags |= CALLOUT_PENDING; 367 CIRCQ_INSERT(&c->c_list, &occ->cc_todo); 368 } else { 369 /* Move to this CPU. */ 370 c->c_cpu = cc; 371 c->c_time = to_ticks + cc->cc_ticks; 372 c->c_flags |= CALLOUT_PENDING; 373 CIRCQ_INSERT(&c->c_list, &cc->cc_todo); 374 mutex_spin_exit(cc->cc_lock); 375 } 376 mutex_spin_exit(lock); 377 } 378 379 /* 380 * callout_reset: 381 * 382 * Reset a callout structure with a new function and argument, and 383 * schedule it to run. 384 */ 385 void 386 callout_reset(callout_t *cs, int to_ticks, void (*func)(void *), void *arg) 387 { 388 callout_impl_t *c = (callout_impl_t *)cs; 389 kmutex_t *lock; 390 391 KASSERT(c->c_magic == CALLOUT_MAGIC); 392 KASSERT(func != NULL); 393 394 lock = callout_lock(c); 395 c->c_func = func; 396 c->c_arg = arg; 397 callout_schedule_locked(c, lock, to_ticks); 398 } 399 400 /* 401 * callout_schedule: 402 * 403 * Schedule a callout to run. The function and argument must 404 * already be set in the callout structure. 405 */ 406 void 407 callout_schedule(callout_t *cs, int to_ticks) 408 { 409 callout_impl_t *c = (callout_impl_t *)cs; 410 kmutex_t *lock; 411 412 KASSERT(c->c_magic == CALLOUT_MAGIC); 413 414 lock = callout_lock(c); 415 callout_schedule_locked(c, lock, to_ticks); 416 } 417 418 /* 419 * callout_stop: 420 * 421 * Try to cancel a pending callout. It may be too late: the callout 422 * could be running on another CPU. If called from interrupt context, 423 * the callout could already be in progress at a lower priority. 424 */ 425 bool 426 callout_stop(callout_t *cs) 427 { 428 callout_impl_t *c = (callout_impl_t *)cs; 429 struct callout_cpu *cc; 430 kmutex_t *lock; 431 bool expired; 432 433 KASSERT(c->c_magic == CALLOUT_MAGIC); 434 435 lock = callout_lock(c); 436 437 if ((c->c_flags & CALLOUT_PENDING) != 0) 438 CIRCQ_REMOVE(&c->c_list); 439 expired = ((c->c_flags & CALLOUT_FIRED) != 0); 440 c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED); 441 442 cc = c->c_cpu; 443 if (cc->cc_active == c) { 444 /* 445 * This is for non-MPSAFE callouts only. To synchronize 446 * effectively we must be called with kernel_lock held. 447 * It's also taken in callout_softclock. 448 */ 449 cc->cc_cancel = c; 450 } 451 452 mutex_spin_exit(lock); 453 454 return expired; 455 } 456 457 /* 458 * callout_halt: 459 * 460 * Cancel a pending callout. If in-flight, block until it completes. 461 * May not be called from a hard interrupt handler. If the callout 462 * can take locks, the caller of callout_halt() must not hold any of 463 * those locks, otherwise the two could deadlock. If 'interlock' is 464 * non-NULL and we must wait for the callout to complete, it will be 465 * released and re-acquired before returning. 466 */ 467 bool 468 callout_halt(callout_t *cs, void *interlock) 469 { 470 callout_impl_t *c = (callout_impl_t *)cs; 471 kmutex_t *lock; 472 int flags; 473 474 KASSERT(c->c_magic == CALLOUT_MAGIC); 475 KASSERT(!cpu_intr_p()); 476 KASSERT(interlock == NULL || mutex_owned(interlock)); 477 478 /* Fast path. */ 479 lock = callout_lock(c); 480 flags = c->c_flags; 481 if ((flags & CALLOUT_PENDING) != 0) 482 CIRCQ_REMOVE(&c->c_list); 483 c->c_flags = flags & ~(CALLOUT_PENDING|CALLOUT_FIRED); 484 if (__predict_false(flags & CALLOUT_FIRED)) { 485 callout_wait(c, interlock, lock); 486 return true; 487 } 488 mutex_spin_exit(lock); 489 return false; 490 } 491 492 /* 493 * callout_wait: 494 * 495 * Slow path for callout_halt(). Deliberately marked __noinline to 496 * prevent unneeded overhead in the caller. 497 */ 498 static void __noinline 499 callout_wait(callout_impl_t *c, void *interlock, kmutex_t *lock) 500 { 501 struct callout_cpu *cc; 502 struct lwp *l; 503 kmutex_t *relock; 504 505 l = curlwp; 506 relock = NULL; 507 for (;;) { 508 /* 509 * At this point we know the callout is not pending, but it 510 * could be running on a CPU somewhere. That can be curcpu 511 * in a few cases: 512 * 513 * - curlwp is a higher priority soft interrupt 514 * - the callout blocked on a lock and is currently asleep 515 * - the callout itself has called callout_halt() (nice!) 516 */ 517 cc = c->c_cpu; 518 if (__predict_true(cc->cc_active != c || cc->cc_lwp == l)) 519 break; 520 521 /* It's running - need to wait for it to complete. */ 522 if (interlock != NULL) { 523 /* 524 * Avoid potential scheduler lock order problems by 525 * dropping the interlock without the callout lock 526 * held; then retry. 527 */ 528 mutex_spin_exit(lock); 529 mutex_exit(interlock); 530 relock = interlock; 531 interlock = NULL; 532 } else { 533 /* XXX Better to do priority inheritance. */ 534 KASSERT(l->l_wchan == NULL); 535 cc->cc_nwait++; 536 cc->cc_ev_block.ev_count++; 537 l->l_kpriority = true; 538 sleepq_enter(&cc->cc_sleepq, l, cc->cc_lock); 539 sleepq_enqueue(&cc->cc_sleepq, cc, "callout", 540 &sleep_syncobj); 541 sleepq_block(0, false); 542 } 543 544 /* 545 * Re-lock the callout and check the state of play again. 546 * It's a common design pattern for callouts to re-schedule 547 * themselves so put a stop to it again if needed. 548 */ 549 lock = callout_lock(c); 550 if ((c->c_flags & CALLOUT_PENDING) != 0) 551 CIRCQ_REMOVE(&c->c_list); 552 c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED); 553 } 554 555 mutex_spin_exit(lock); 556 if (__predict_false(relock != NULL)) 557 mutex_enter(relock); 558 } 559 560 #ifdef notyet 561 /* 562 * callout_bind: 563 * 564 * Bind a callout so that it will only execute on one CPU. 565 * The callout must be stopped, and must be MPSAFE. 566 * 567 * XXX Disabled for now until it is decided how to handle 568 * offlined CPUs. We may want weak+strong binding. 569 */ 570 void 571 callout_bind(callout_t *cs, struct cpu_info *ci) 572 { 573 callout_impl_t *c = (callout_impl_t *)cs; 574 struct callout_cpu *cc; 575 kmutex_t *lock; 576 577 KASSERT((c->c_flags & CALLOUT_PENDING) == 0); 578 KASSERT(c->c_cpu->cc_active != c); 579 KASSERT(c->c_magic == CALLOUT_MAGIC); 580 KASSERT((c->c_flags & CALLOUT_MPSAFE) != 0); 581 582 lock = callout_lock(c); 583 cc = ci->ci_data.cpu_callout; 584 c->c_flags |= CALLOUT_BOUND; 585 if (c->c_cpu != cc) { 586 /* 587 * Assigning c_cpu effectively unlocks the callout 588 * structure, as we don't hold the new CPU's lock. 589 * Issue memory barrier to prevent accesses being 590 * reordered. 591 */ 592 membar_exit(); 593 c->c_cpu = cc; 594 } 595 mutex_spin_exit(lock); 596 } 597 #endif 598 599 void 600 callout_setfunc(callout_t *cs, void (*func)(void *), void *arg) 601 { 602 callout_impl_t *c = (callout_impl_t *)cs; 603 kmutex_t *lock; 604 605 KASSERT(c->c_magic == CALLOUT_MAGIC); 606 KASSERT(func != NULL); 607 608 lock = callout_lock(c); 609 c->c_func = func; 610 c->c_arg = arg; 611 mutex_spin_exit(lock); 612 } 613 614 bool 615 callout_expired(callout_t *cs) 616 { 617 callout_impl_t *c = (callout_impl_t *)cs; 618 kmutex_t *lock; 619 bool rv; 620 621 KASSERT(c->c_magic == CALLOUT_MAGIC); 622 623 lock = callout_lock(c); 624 rv = ((c->c_flags & CALLOUT_FIRED) != 0); 625 mutex_spin_exit(lock); 626 627 return rv; 628 } 629 630 bool 631 callout_active(callout_t *cs) 632 { 633 callout_impl_t *c = (callout_impl_t *)cs; 634 kmutex_t *lock; 635 bool rv; 636 637 KASSERT(c->c_magic == CALLOUT_MAGIC); 638 639 lock = callout_lock(c); 640 rv = ((c->c_flags & (CALLOUT_PENDING|CALLOUT_FIRED)) != 0); 641 mutex_spin_exit(lock); 642 643 return rv; 644 } 645 646 bool 647 callout_pending(callout_t *cs) 648 { 649 callout_impl_t *c = (callout_impl_t *)cs; 650 kmutex_t *lock; 651 bool rv; 652 653 KASSERT(c->c_magic == CALLOUT_MAGIC); 654 655 lock = callout_lock(c); 656 rv = ((c->c_flags & CALLOUT_PENDING) != 0); 657 mutex_spin_exit(lock); 658 659 return rv; 660 } 661 662 bool 663 callout_invoking(callout_t *cs) 664 { 665 callout_impl_t *c = (callout_impl_t *)cs; 666 kmutex_t *lock; 667 bool rv; 668 669 KASSERT(c->c_magic == CALLOUT_MAGIC); 670 671 lock = callout_lock(c); 672 rv = ((c->c_flags & CALLOUT_INVOKING) != 0); 673 mutex_spin_exit(lock); 674 675 return rv; 676 } 677 678 void 679 callout_ack(callout_t *cs) 680 { 681 callout_impl_t *c = (callout_impl_t *)cs; 682 kmutex_t *lock; 683 684 KASSERT(c->c_magic == CALLOUT_MAGIC); 685 686 lock = callout_lock(c); 687 c->c_flags &= ~CALLOUT_INVOKING; 688 mutex_spin_exit(lock); 689 } 690 691 /* 692 * callout_hardclock: 693 * 694 * Called from hardclock() once every tick. We schedule a soft 695 * interrupt if there is work to be done. 696 */ 697 void 698 callout_hardclock(void) 699 { 700 struct callout_cpu *cc; 701 int needsoftclock, ticks; 702 703 cc = curcpu()->ci_data.cpu_callout; 704 mutex_spin_enter(cc->cc_lock); 705 706 ticks = ++cc->cc_ticks; 707 708 MOVEBUCKET(cc, 0, ticks); 709 if (MASKWHEEL(0, ticks) == 0) { 710 MOVEBUCKET(cc, 1, ticks); 711 if (MASKWHEEL(1, ticks) == 0) { 712 MOVEBUCKET(cc, 2, ticks); 713 if (MASKWHEEL(2, ticks) == 0) 714 MOVEBUCKET(cc, 3, ticks); 715 } 716 } 717 718 needsoftclock = !CIRCQ_EMPTY(&cc->cc_todo); 719 mutex_spin_exit(cc->cc_lock); 720 721 if (needsoftclock) 722 softint_schedule(callout_sih); 723 } 724 725 /* 726 * callout_softclock: 727 * 728 * Soft interrupt handler, scheduled above if there is work to 729 * be done. Callouts are made in soft interrupt context. 730 */ 731 static void 732 callout_softclock(void *v) 733 { 734 callout_impl_t *c; 735 struct callout_cpu *cc; 736 void (*func)(void *); 737 void *arg; 738 int mpsafe, count, ticks, delta; 739 lwp_t *l; 740 741 l = curlwp; 742 KASSERT(l->l_cpu == curcpu()); 743 cc = l->l_cpu->ci_data.cpu_callout; 744 745 mutex_spin_enter(cc->cc_lock); 746 cc->cc_lwp = l; 747 while (!CIRCQ_EMPTY(&cc->cc_todo)) { 748 c = CIRCQ_FIRST(&cc->cc_todo); 749 KASSERT(c->c_magic == CALLOUT_MAGIC); 750 KASSERT(c->c_func != NULL); 751 KASSERT(c->c_cpu == cc); 752 KASSERT((c->c_flags & CALLOUT_PENDING) != 0); 753 KASSERT((c->c_flags & CALLOUT_FIRED) == 0); 754 CIRCQ_REMOVE(&c->c_list); 755 756 /* If due run it, otherwise insert it into the right bucket. */ 757 ticks = cc->cc_ticks; 758 delta = (int)((unsigned)c->c_time - (unsigned)ticks); 759 if (delta > 0) { 760 CIRCQ_INSERT(&c->c_list, BUCKET(cc, delta, c->c_time)); 761 continue; 762 } 763 if (delta < 0) 764 cc->cc_ev_late.ev_count++; 765 766 c->c_flags = (c->c_flags & ~CALLOUT_PENDING) | 767 (CALLOUT_FIRED | CALLOUT_INVOKING); 768 mpsafe = (c->c_flags & CALLOUT_MPSAFE); 769 func = c->c_func; 770 arg = c->c_arg; 771 cc->cc_active = c; 772 773 mutex_spin_exit(cc->cc_lock); 774 KASSERT(func != NULL); 775 if (__predict_false(!mpsafe)) { 776 KERNEL_LOCK(1, NULL); 777 (*func)(arg); 778 KERNEL_UNLOCK_ONE(NULL); 779 } else 780 (*func)(arg); 781 mutex_spin_enter(cc->cc_lock); 782 783 /* 784 * We can't touch 'c' here because it might be 785 * freed already. If LWPs waiting for callout 786 * to complete, awaken them. 787 */ 788 cc->cc_active = NULL; 789 if ((count = cc->cc_nwait) != 0) { 790 cc->cc_nwait = 0; 791 /* sleepq_wake() drops the lock. */ 792 sleepq_wake(&cc->cc_sleepq, cc, count, cc->cc_lock); 793 mutex_spin_enter(cc->cc_lock); 794 } 795 } 796 cc->cc_lwp = NULL; 797 mutex_spin_exit(cc->cc_lock); 798 } 799 #endif 800 801 #ifdef DDB 802 static void 803 db_show_callout_bucket(struct callout_cpu *cc, struct callout_circq *kbucket, 804 struct callout_circq *bucket) 805 { 806 callout_impl_t *c, ci; 807 db_expr_t offset; 808 const char *name; 809 static char question[] = "?"; 810 int b; 811 812 if (CIRCQ_LAST(bucket, kbucket)) 813 return; 814 815 for (c = CIRCQ_FIRST(bucket); /*nothing*/; c = CIRCQ_NEXT(&c->c_list)) { 816 db_read_bytes((db_addr_t)c, sizeof(ci), (char *)&ci); 817 c = &ci; 818 db_find_sym_and_offset((db_addr_t)(intptr_t)c->c_func, &name, 819 &offset); 820 name = name ? name : question; 821 b = (bucket - cc->cc_wheel); 822 if (b < 0) 823 b = -WHEELSIZE; 824 db_printf("%9d %2d/%-4d %16lx %s\n", 825 c->c_time - cc->cc_ticks, b / WHEELSIZE, b, 826 (u_long)c->c_arg, name); 827 if (CIRCQ_LAST(&c->c_list, kbucket)) 828 break; 829 } 830 } 831 832 void 833 db_show_callout(db_expr_t addr, bool haddr, db_expr_t count, const char *modif) 834 { 835 struct callout_cpu *cc, ccb; 836 struct cpu_info *ci, cib; 837 int b; 838 839 #ifndef CRASH 840 db_printf("hardclock_ticks now: %d\n", hardclock_ticks); 841 #endif 842 db_printf(" ticks wheel arg func\n"); 843 844 /* 845 * Don't lock the callwheel; all the other CPUs are paused 846 * anyhow, and we might be called in a circumstance where 847 * some other CPU was paused while holding the lock. 848 */ 849 for (ci = db_cpu_first(); ci != NULL; ci = db_cpu_next(ci)) { 850 db_read_bytes((db_addr_t)ci, sizeof(cib), (char *)&cib); 851 cc = cib.ci_data.cpu_callout; 852 db_read_bytes((db_addr_t)cc, sizeof(ccb), (char *)&ccb); 853 db_show_callout_bucket(&ccb, &cc->cc_todo, &ccb.cc_todo); 854 } 855 for (b = 0; b < BUCKETS; b++) { 856 for (ci = db_cpu_first(); ci != NULL; ci = db_cpu_next(ci)) { 857 db_read_bytes((db_addr_t)ci, sizeof(cib), (char *)&cib); 858 cc = cib.ci_data.cpu_callout; 859 db_read_bytes((db_addr_t)cc, sizeof(ccb), (char *)&ccb); 860 db_show_callout_bucket(&ccb, &cc->cc_wheel[b], 861 &ccb.cc_wheel[b]); 862 } 863 } 864 } 865 #endif /* DDB */ 866