kern_timeout.c revision 1.17.20.4
1 /* $NetBSD: kern_timeout.c,v 1.17.20.4 2007/01/19 00:38:00 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 2003, 2006 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. 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 /* 40 * Copyright (c) 2001 Thomas Nordin <nordin (at) openbsd.org> 41 * Copyright (c) 2000-2001 Artur Grabowski <art (at) openbsd.org> 42 * All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 48 * 1. Redistributions of source code must retain the above copyright 49 * notice, this list of conditions and the following disclaimer. 50 * 2. Redistributions in binary form must reproduce the above copyright 51 * notice, this list of conditions and the following disclaimer in the 52 * documentation and/or other materials provided with the distribution. 53 * 3. The name of the author may not be used to endorse or promote products 54 * derived from this software without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, 57 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY 58 * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL 59 * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 60 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 61 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 62 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 63 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 64 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 65 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 66 */ 67 68 #include <sys/cdefs.h> 69 __KERNEL_RCSID(0, "$NetBSD: kern_timeout.c,v 1.17.20.4 2007/01/19 00:38:00 yamt Exp $"); 70 71 /* 72 * Adapted from OpenBSD: kern_timeout.c,v 1.15 2002/12/08 04:21:07 art Exp, 73 * modified to match NetBSD's pre-existing callout API. 74 */ 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/kernel.h> 79 #include <sys/lock.h> 80 #include <sys/callout.h> 81 #include <sys/mutex.h> 82 83 #ifdef DDB 84 #include <machine/db_machdep.h> 85 #include <ddb/db_interface.h> 86 #include <ddb/db_access.h> 87 #include <ddb/db_sym.h> 88 #include <ddb/db_output.h> 89 #endif 90 91 /* 92 * Timeouts are kept in a hierarchical timing wheel. The c_time is the value 93 * of the global variable "hardclock_ticks" when the timeout should be called. 94 * There are four levels with 256 buckets each. See 'Scheme 7' in 95 * "Hashed and Hierarchical Timing Wheels: Efficient Data Structures for 96 * Implementing a Timer Facility" by George Varghese and Tony Lauck. 97 */ 98 #define BUCKETS 1024 99 #define WHEELSIZE 256 100 #define WHEELMASK 255 101 #define WHEELBITS 8 102 103 static struct callout_circq timeout_wheel[BUCKETS]; /* Queues of timeouts */ 104 static struct callout_circq timeout_todo; /* Worklist */ 105 106 #define MASKWHEEL(wheel, time) (((time) >> ((wheel)*WHEELBITS)) & WHEELMASK) 107 108 #define BUCKET(rel, abs) \ 109 (((rel) <= (1 << (2*WHEELBITS))) \ 110 ? ((rel) <= (1 << WHEELBITS)) \ 111 ? &timeout_wheel[MASKWHEEL(0, (abs))] \ 112 : &timeout_wheel[MASKWHEEL(1, (abs)) + WHEELSIZE] \ 113 : ((rel) <= (1 << (3*WHEELBITS))) \ 114 ? &timeout_wheel[MASKWHEEL(2, (abs)) + 2*WHEELSIZE] \ 115 : &timeout_wheel[MASKWHEEL(3, (abs)) + 3*WHEELSIZE]) 116 117 #define MOVEBUCKET(wheel, time) \ 118 CIRCQ_APPEND(&timeout_todo, \ 119 &timeout_wheel[MASKWHEEL((wheel), (time)) + (wheel)*WHEELSIZE]) 120 121 /* 122 * All wheels are locked with the same lock (which must also block out all 123 * interrupts). 124 */ 125 kmutex_t callout_mutex; 126 127 /* 128 * Circular queue definitions. 129 */ 130 131 #define CIRCQ_INIT(list) \ 132 do { \ 133 (list)->cq_next_l = (list); \ 134 (list)->cq_prev_l = (list); \ 135 } while (/*CONSTCOND*/0) 136 137 #define CIRCQ_INSERT(elem, list) \ 138 do { \ 139 (elem)->cq_prev_e = (list)->cq_prev_e; \ 140 (elem)->cq_next_l = (list); \ 141 (list)->cq_prev_l->cq_next_l = (elem); \ 142 (list)->cq_prev_l = (elem); \ 143 } while (/*CONSTCOND*/0) 144 145 #define CIRCQ_APPEND(fst, snd) \ 146 do { \ 147 if (!CIRCQ_EMPTY(snd)) { \ 148 (fst)->cq_prev_l->cq_next_l = (snd)->cq_next_l; \ 149 (snd)->cq_next_l->cq_prev_l = (fst)->cq_prev_l; \ 150 (snd)->cq_prev_l->cq_next_l = (fst); \ 151 (fst)->cq_prev_l = (snd)->cq_prev_l; \ 152 CIRCQ_INIT(snd); \ 153 } \ 154 } while (/*CONSTCOND*/0) 155 156 #define CIRCQ_REMOVE(elem) \ 157 do { \ 158 (elem)->cq_next_l->cq_prev_e = (elem)->cq_prev_e; \ 159 (elem)->cq_prev_l->cq_next_e = (elem)->cq_next_e; \ 160 } while (/*CONSTCOND*/0) 161 162 #define CIRCQ_FIRST(list) ((list)->cq_next_e) 163 #define CIRCQ_NEXT(elem) ((elem)->cq_next_e) 164 #define CIRCQ_LAST(elem,list) ((elem)->cq_next_l == (list)) 165 #define CIRCQ_EMPTY(list) ((list)->cq_next_l == (list)) 166 167 /* 168 * Some of the "math" in here is a bit tricky. 169 * 170 * We have to beware of wrapping ints. 171 * We use the fact that any element added to the queue must be added with a 172 * positive time. That means that any element `to' on the queue cannot be 173 * scheduled to timeout further in time than INT_MAX, but c->c_time can 174 * be positive or negative so comparing it with anything is dangerous. 175 * The only way we can use the c->c_time value in any predictable way 176 * is when we calculate how far in the future `to' will timeout - 177 * "c->c_time - hardclock_ticks". The result will always be positive for 178 * future timeouts and 0 or negative for due timeouts. 179 */ 180 181 #ifdef CALLOUT_EVENT_COUNTERS 182 static struct evcnt callout_ev_late; 183 #endif 184 185 /* 186 * callout_barrier: 187 * 188 * If the callout is running on another CPU, busy wait until it 189 * completes. 190 */ 191 static inline void 192 callout_barrier(struct callout *c) 193 { 194 #ifdef MULTIPROCESSOR 195 struct cpu_info *ci; 196 197 LOCK_ASSERT(mutex_owned(&callout_mutex)); 198 199 /* 200 * The callout may have already been dispatched to run on the 201 * current CPU. It's possible for us to arrive here before it 202 * actually runs because the SPL is dropped from IPL_SCHED in 203 * softclock(), and IPL_SOFTCLOCK is low priority. We can't deal 204 * with that race easily, so for now the caller must deal with 205 * it. 206 */ 207 while ((ci = c->c_oncpu) != NULL && ci != curcpu() && 208 ci->ci_data.cpu_callout == c) { 209 smutex_exit(&callout_mutex); 210 while (ci->ci_data.cpu_callout == c) 211 ; 212 smutex_enter(&callout_mutex); 213 } 214 c->c_oncpu = NULL; 215 #endif 216 } 217 218 /* 219 * callout_startup: 220 * 221 * Initialize the callout facility, called at system startup time. 222 */ 223 void 224 callout_startup(void) 225 { 226 int b; 227 228 CIRCQ_INIT(&timeout_todo); 229 for (b = 0; b < BUCKETS; b++) 230 CIRCQ_INIT(&timeout_wheel[b]); 231 mutex_init(&callout_mutex, MUTEX_SPIN, IPL_SCHED); 232 233 #ifdef CALLOUT_EVENT_COUNTERS 234 evcnt_attach_dynamic(&callout_ev_late, EVCNT_TYPE_MISC, 235 NULL, "callout", "late"); 236 #endif 237 } 238 239 /* 240 * callout_init: 241 * 242 * Initialize a callout structure. 243 */ 244 void 245 callout_init(struct callout *c) 246 { 247 248 memset(c, 0, sizeof(*c)); 249 } 250 251 /* 252 * callout_reset: 253 * 254 * Reset a callout structure with a new function and argument, and 255 * schedule it to run. 256 */ 257 void 258 callout_reset(struct callout *c, int to_ticks, void (*func)(void *), void *arg) 259 { 260 int old_time; 261 262 KASSERT(to_ticks >= 0); 263 264 smutex_enter(&callout_mutex); 265 266 callout_barrier(c); 267 268 /* Initialize the time here, it won't change. */ 269 old_time = c->c_time; 270 c->c_time = to_ticks + hardclock_ticks; 271 c->c_flags &= ~(CALLOUT_FIRED|CALLOUT_INVOKING); 272 273 c->c_func = func; 274 c->c_arg = arg; 275 276 /* 277 * If this timeout is already scheduled and now is moved 278 * earlier, reschedule it now. Otherwise leave it in place 279 * and let it be rescheduled later. 280 */ 281 if (callout_pending(c)) { 282 if (c->c_time - old_time < 0) { 283 CIRCQ_REMOVE(&c->c_list); 284 CIRCQ_INSERT(&c->c_list, &timeout_todo); 285 } 286 } else { 287 c->c_flags |= CALLOUT_PENDING; 288 CIRCQ_INSERT(&c->c_list, &timeout_todo); 289 } 290 291 smutex_exit(&callout_mutex); 292 } 293 294 /* 295 * callout_schedule: 296 * 297 * Schedule a callout to run. The function and argument must 298 * already be set in the callout structure. 299 */ 300 void 301 callout_schedule(struct callout *c, int to_ticks) 302 { 303 int old_time; 304 305 KASSERT(to_ticks >= 0); 306 307 smutex_enter(&callout_mutex); 308 309 callout_barrier(c); 310 311 /* Initialize the time here, it won't change. */ 312 old_time = c->c_time; 313 c->c_time = to_ticks + hardclock_ticks; 314 c->c_flags &= ~(CALLOUT_FIRED|CALLOUT_INVOKING); 315 316 /* 317 * If this timeout is already scheduled and now is moved 318 * earlier, reschedule it now. Otherwise leave it in place 319 * and let it be rescheduled later. 320 */ 321 if (callout_pending(c)) { 322 if (c->c_time - old_time < 0) { 323 CIRCQ_REMOVE(&c->c_list); 324 CIRCQ_INSERT(&c->c_list, &timeout_todo); 325 } 326 } else { 327 c->c_flags |= CALLOUT_PENDING; 328 CIRCQ_INSERT(&c->c_list, &timeout_todo); 329 } 330 331 smutex_exit(&callout_mutex); 332 } 333 334 /* 335 * callout_stop: 336 * 337 * Cancel a pending callout. 338 */ 339 void 340 callout_stop(struct callout *c) 341 { 342 343 smutex_enter(&callout_mutex); 344 345 callout_barrier(c); 346 347 if (callout_pending(c)) 348 CIRCQ_REMOVE(&c->c_list); 349 350 c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED); 351 352 smutex_exit(&callout_mutex); 353 } 354 355 /* 356 * This is called from hardclock() once every tick. 357 * We return !0 if we need to schedule a softclock. 358 */ 359 int 360 callout_hardclock(void) 361 { 362 int needsoftclock; 363 364 smutex_enter(&callout_mutex); 365 366 MOVEBUCKET(0, hardclock_ticks); 367 if (MASKWHEEL(0, hardclock_ticks) == 0) { 368 MOVEBUCKET(1, hardclock_ticks); 369 if (MASKWHEEL(1, hardclock_ticks) == 0) { 370 MOVEBUCKET(2, hardclock_ticks); 371 if (MASKWHEEL(2, hardclock_ticks) == 0) 372 MOVEBUCKET(3, hardclock_ticks); 373 } 374 } 375 376 needsoftclock = !CIRCQ_EMPTY(&timeout_todo); 377 smutex_exit(&callout_mutex); 378 379 return needsoftclock; 380 } 381 382 /* ARGSUSED */ 383 void 384 softclock(void *v) 385 { 386 struct cpu_info *ci; 387 struct callout *c; 388 void (*func)(void *); 389 void *arg; 390 391 ci = curcpu(); 392 393 smutex_enter(&callout_mutex); 394 395 while (!CIRCQ_EMPTY(&timeout_todo)) { 396 c = CIRCQ_FIRST(&timeout_todo); 397 CIRCQ_REMOVE(&c->c_list); 398 399 /* If due run it, otherwise insert it into the right bucket. */ 400 if (c->c_time - hardclock_ticks > 0) { 401 CIRCQ_INSERT(&c->c_list, 402 BUCKET((c->c_time - hardclock_ticks), c->c_time)); 403 } else { 404 #ifdef CALLOUT_EVENT_COUNTERS 405 if (c->c_time - hardclock_ticks < 0) 406 callout_ev_late.ev_count++; 407 #endif 408 c->c_flags = (c->c_flags & ~CALLOUT_PENDING) | 409 (CALLOUT_FIRED|CALLOUT_INVOKING); 410 411 func = c->c_func; 412 arg = c->c_arg; 413 414 #ifdef MULTIPROCESSOR 415 c->c_oncpu = ci; 416 ci->ci_data.cpu_callout = c; 417 #endif 418 smutex_exit(&callout_mutex); 419 (*func)(arg); 420 smutex_enter(&callout_mutex); 421 #ifdef MULTIPROCESSOR 422 ci->ci_data.cpu_callout = NULL; 423 /* 424 * we can't touch 'c' here because it might be 425 * freed already. 426 */ 427 #endif 428 } 429 } 430 431 smutex_exit(&callout_mutex); 432 } 433 434 #ifdef DDB 435 static void 436 db_show_callout_bucket(struct callout_circq *bucket) 437 { 438 struct callout *c; 439 db_expr_t offset; 440 const char *name; 441 static char question[] = "?"; 442 443 if (CIRCQ_EMPTY(bucket)) 444 return; 445 446 for (c = CIRCQ_FIRST(bucket); /*nothing*/; c = CIRCQ_NEXT(&c->c_list)) { 447 db_find_sym_and_offset((db_addr_t)(intptr_t)c->c_func, &name, 448 &offset); 449 name = name ? name : question; 450 #ifdef _LP64 451 #define POINTER_WIDTH "%16lx" 452 #else 453 #define POINTER_WIDTH "%8lx" 454 #endif 455 db_printf("%9d %2d/%-4d " POINTER_WIDTH " %s\n", 456 c->c_time - hardclock_ticks, 457 (int)((bucket - timeout_wheel) / WHEELSIZE), 458 (int)(bucket - timeout_wheel), (u_long) c->c_arg, name); 459 460 if (CIRCQ_LAST(&c->c_list, bucket)) 461 break; 462 } 463 } 464 465 void 466 db_show_callout(db_expr_t addr, int haddr, db_expr_t count, const char *modif) 467 { 468 int b; 469 470 db_printf("hardclock_ticks now: %d\n", hardclock_ticks); 471 #ifdef _LP64 472 db_printf(" ticks wheel arg func\n"); 473 #else 474 db_printf(" ticks wheel arg func\n"); 475 #endif 476 477 /* 478 * Don't lock the callwheel; all the other CPUs are paused 479 * anyhow, and we might be called in a circumstance where 480 * some other CPU was paused while holding the lock. 481 */ 482 483 db_show_callout_bucket(&timeout_todo); 484 for (b = 0; b < BUCKETS; b++) 485 db_show_callout_bucket(&timeout_wheel[b]); 486 } 487 #endif /* DDB */ 488
Indexes created Sat Sep 20 22:09:52 GMT 2025