kern_timeout.c revision 116602
1/*- 2 * Copyright (c) 1982, 1986, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 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 University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * From: @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 39 */ 40 41#include <sys/cdefs.h> 42__FBSDID("$FreeBSD: head/sys/kern/kern_timeout.c 116602 2003-06-20 08:07:15Z phk $"); 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/callout.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/mutex.h> 50#include <sys/sysctl.h> 51 52static int avg_depth; 53SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0, 54 "Average number of items examined per softclock call. Units = 1/1000"); 55static int avg_gcalls; 56SYSCTL_INT(_debug, OID_AUTO, to_avg_gcalls, CTLFLAG_RD, &avg_gcalls, 0, 57 "Average number of Giant callouts made per softclock call. Units = 1/1000"); 58static int avg_mpcalls; 59SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0, 60 "Average number of MP callouts made per softclock call. Units = 1/1000"); 61/* 62 * TODO: 63 * allocate more timeout table slots when table overflows. 64 */ 65 66/* Exported to machdep.c and/or kern_clock.c. */ 67struct callout *callout; 68struct callout_list callfree; 69int callwheelsize, callwheelbits, callwheelmask; 70struct callout_tailq *callwheel; 71int softticks; /* Like ticks, but for softclock(). */ 72#ifdef DIAGNOSTIC 73struct mtx callout_lock, callout_dont_sleep; 74#endif 75 76static struct callout *nextsoftcheck; /* Next callout to be checked. */ 77 78/* 79 * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization 80 * 81 * This code is called very early in the kernel initialization sequence, 82 * and may be called more then once. 83 */ 84caddr_t 85kern_timeout_callwheel_alloc(caddr_t v) 86{ 87 /* 88 * Calculate callout wheel size 89 */ 90 for (callwheelsize = 1, callwheelbits = 0; 91 callwheelsize < ncallout; 92 callwheelsize <<= 1, ++callwheelbits) 93 ; 94 callwheelmask = callwheelsize - 1; 95 96 callout = (struct callout *)v; 97 v = (caddr_t)(callout + ncallout); 98 callwheel = (struct callout_tailq *)v; 99 v = (caddr_t)(callwheel + callwheelsize); 100 return(v); 101} 102 103/* 104 * kern_timeout_callwheel_init() - initialize previously reserved callwheel 105 * space. 106 * 107 * This code is called just once, after the space reserved for the 108 * callout wheel has been finalized. 109 */ 110void 111kern_timeout_callwheel_init(void) 112{ 113 int i; 114 115 SLIST_INIT(&callfree); 116 for (i = 0; i < ncallout; i++) { 117 callout_init(&callout[i], 0); 118 callout[i].c_flags = CALLOUT_LOCAL_ALLOC; 119 SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle); 120 } 121 for (i = 0; i < callwheelsize; i++) { 122 TAILQ_INIT(&callwheel[i]); 123 } 124 mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); 125#ifdef DIAGNOSTIC 126 mtx_init(&callout_dont_sleep, "callout_dont_sleep", NULL, MTX_DEF); 127#endif 128} 129 130/* 131 * The callout mechanism is based on the work of Adam M. Costello and 132 * George Varghese, published in a technical report entitled "Redesigning 133 * the BSD Callout and Timer Facilities" and modified slightly for inclusion 134 * in FreeBSD by Justin T. Gibbs. The original work on the data structures 135 * used in this implementation was published by G.Varghese and A. Lauck in 136 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for 137 * the Efficient Implementation of a Timer Facility" in the Proceedings of 138 * the 11th ACM Annual Symposium on Operating Systems Principles, 139 * Austin, Texas Nov 1987. 140 */ 141 142/* 143 * Software (low priority) clock interrupt. 144 * Run periodic events from timeout queue. 145 */ 146void 147softclock(void *dummy) 148{ 149 struct callout *c; 150 struct callout_tailq *bucket; 151 int curticks; 152 int steps; /* #steps since we last allowed interrupts */ 153 int depth; 154 int mpcalls; 155 int gcalls; 156#ifdef DIAGNOSTIC 157 struct bintime bt1, bt2; 158 struct timespec ts2; 159 static uint64_t maxdt = 18446744073709551LL; /* 1 msec */ 160#endif 161 162#ifndef MAX_SOFTCLOCK_STEPS 163#define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */ 164#endif /* MAX_SOFTCLOCK_STEPS */ 165 166 mpcalls = 0; 167 gcalls = 0; 168 depth = 0; 169 steps = 0; 170 mtx_lock_spin(&callout_lock); 171 while (softticks != ticks) { 172 softticks++; 173 /* 174 * softticks may be modified by hard clock, so cache 175 * it while we work on a given bucket. 176 */ 177 curticks = softticks; 178 bucket = &callwheel[curticks & callwheelmask]; 179 c = TAILQ_FIRST(bucket); 180 while (c) { 181 depth++; 182 if (c->c_time != curticks) { 183 c = TAILQ_NEXT(c, c_links.tqe); 184 ++steps; 185 if (steps >= MAX_SOFTCLOCK_STEPS) { 186 nextsoftcheck = c; 187 /* Give interrupts a chance. */ 188 mtx_unlock_spin(&callout_lock); 189 ; /* nothing */ 190 mtx_lock_spin(&callout_lock); 191 c = nextsoftcheck; 192 steps = 0; 193 } 194 } else { 195 void (*c_func)(void *); 196 void *c_arg; 197 int c_flags; 198 199 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 200 TAILQ_REMOVE(bucket, c, c_links.tqe); 201 c_func = c->c_func; 202 c_arg = c->c_arg; 203 c_flags = c->c_flags; 204 c->c_func = NULL; 205 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 206 c->c_flags = CALLOUT_LOCAL_ALLOC; 207 SLIST_INSERT_HEAD(&callfree, c, 208 c_links.sle); 209 } else { 210 c->c_flags = 211 (c->c_flags & ~CALLOUT_PENDING); 212 } 213 mtx_unlock_spin(&callout_lock); 214 if (!(c_flags & CALLOUT_MPSAFE)) { 215 mtx_lock(&Giant); 216 gcalls++; 217 } else { 218 mpcalls++; 219 } 220#ifdef DIAGNOSTIC 221 binuptime(&bt1); 222 mtx_lock(&callout_dont_sleep); 223#endif 224 c_func(c_arg); 225#ifdef DIAGNOSTIC 226 mtx_unlock(&callout_dont_sleep); 227 binuptime(&bt2); 228 bintime_sub(&bt2, &bt1); 229 if (bt2.frac > maxdt) { 230 maxdt = bt2.frac; 231 bintime2timespec(&bt2, &ts2); 232 printf( 233 "Expensive timeout(9) function: %p(%p) %d.%09ld s\n", 234 c_func, c_arg, 235 ts2.tv_sec, ts2.tv_nsec); 236 } 237#endif 238 if (!(c_flags & CALLOUT_MPSAFE)) 239 mtx_unlock(&Giant); 240 mtx_lock_spin(&callout_lock); 241 steps = 0; 242 c = nextsoftcheck; 243 } 244 } 245 } 246 avg_depth += (depth * 1000 - avg_depth) >> 8; 247 avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8; 248 avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8; 249 nextsoftcheck = NULL; 250 mtx_unlock_spin(&callout_lock); 251} 252 253/* 254 * timeout -- 255 * Execute a function after a specified length of time. 256 * 257 * untimeout -- 258 * Cancel previous timeout function call. 259 * 260 * callout_handle_init -- 261 * Initialize a handle so that using it with untimeout is benign. 262 * 263 * See AT&T BCI Driver Reference Manual for specification. This 264 * implementation differs from that one in that although an 265 * identification value is returned from timeout, the original 266 * arguments to timeout as well as the identifier are used to 267 * identify entries for untimeout. 268 */ 269struct callout_handle 270timeout(ftn, arg, to_ticks) 271 timeout_t *ftn; 272 void *arg; 273 int to_ticks; 274{ 275 struct callout *new; 276 struct callout_handle handle; 277 278 mtx_lock_spin(&callout_lock); 279 280 /* Fill in the next free callout structure. */ 281 new = SLIST_FIRST(&callfree); 282 if (new == NULL) 283 /* XXX Attempt to malloc first */ 284 panic("timeout table full"); 285 SLIST_REMOVE_HEAD(&callfree, c_links.sle); 286 287 callout_reset(new, to_ticks, ftn, arg); 288 289 handle.callout = new; 290 mtx_unlock_spin(&callout_lock); 291 return (handle); 292} 293 294void 295untimeout(ftn, arg, handle) 296 timeout_t *ftn; 297 void *arg; 298 struct callout_handle handle; 299{ 300 301 /* 302 * Check for a handle that was initialized 303 * by callout_handle_init, but never used 304 * for a real timeout. 305 */ 306 if (handle.callout == NULL) 307 return; 308 309 mtx_lock_spin(&callout_lock); 310 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) 311 callout_stop(handle.callout); 312 mtx_unlock_spin(&callout_lock); 313} 314 315void 316callout_handle_init(struct callout_handle *handle) 317{ 318 handle->callout = NULL; 319} 320 321/* 322 * New interface; clients allocate their own callout structures. 323 * 324 * callout_reset() - establish or change a timeout 325 * callout_stop() - disestablish a timeout 326 * callout_init() - initialize a callout structure so that it can 327 * safely be passed to callout_reset() and callout_stop() 328 * 329 * <sys/callout.h> defines three convenience macros: 330 * 331 * callout_active() - returns truth if callout has not been serviced 332 * callout_pending() - returns truth if callout is still waiting for timeout 333 * callout_deactivate() - marks the callout as having been serviced 334 */ 335void 336callout_reset(c, to_ticks, ftn, arg) 337 struct callout *c; 338 int to_ticks; 339 void (*ftn)(void *); 340 void *arg; 341{ 342 343 mtx_lock_spin(&callout_lock); 344 if (c->c_flags & CALLOUT_PENDING) 345 callout_stop(c); 346 347 /* 348 * We could unlock callout_lock here and lock it again before the 349 * TAILQ_INSERT_TAIL, but there's no point since doing this setup 350 * doesn't take much time. 351 */ 352 if (to_ticks <= 0) 353 to_ticks = 1; 354 355 c->c_arg = arg; 356 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 357 c->c_func = ftn; 358 c->c_time = ticks + to_ticks; 359 TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask], 360 c, c_links.tqe); 361 mtx_unlock_spin(&callout_lock); 362} 363 364int 365callout_stop(c) 366 struct callout *c; 367{ 368 369 mtx_lock_spin(&callout_lock); 370 /* 371 * Don't attempt to delete a callout that's not on the queue. 372 */ 373 if (!(c->c_flags & CALLOUT_PENDING)) { 374 c->c_flags &= ~CALLOUT_ACTIVE; 375 mtx_unlock_spin(&callout_lock); 376 return (0); 377 } 378 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 379 380 if (nextsoftcheck == c) { 381 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 382 } 383 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe); 384 c->c_func = NULL; 385 386 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 387 SLIST_INSERT_HEAD(&callfree, c, c_links.sle); 388 } 389 mtx_unlock_spin(&callout_lock); 390 return (1); 391} 392 393void 394callout_init(c, mpsafe) 395 struct callout *c; 396 int mpsafe; 397{ 398 bzero(c, sizeof *c); 399 if (mpsafe) 400 c->c_flags |= CALLOUT_MPSAFE; 401} 402 403#ifdef APM_FIXUP_CALLTODO 404/* 405 * Adjust the kernel calltodo timeout list. This routine is used after 406 * an APM resume to recalculate the calltodo timer list values with the 407 * number of hz's we have been sleeping. The next hardclock() will detect 408 * that there are fired timers and run softclock() to execute them. 409 * 410 * Please note, I have not done an exhaustive analysis of what code this 411 * might break. I am motivated to have my select()'s and alarm()'s that 412 * have expired during suspend firing upon resume so that the applications 413 * which set the timer can do the maintanence the timer was for as close 414 * as possible to the originally intended time. Testing this code for a 415 * week showed that resuming from a suspend resulted in 22 to 25 timers 416 * firing, which seemed independant on whether the suspend was 2 hours or 417 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 418 */ 419void 420adjust_timeout_calltodo(time_change) 421 struct timeval *time_change; 422{ 423 register struct callout *p; 424 unsigned long delta_ticks; 425 426 /* 427 * How many ticks were we asleep? 428 * (stolen from tvtohz()). 429 */ 430 431 /* Don't do anything */ 432 if (time_change->tv_sec < 0) 433 return; 434 else if (time_change->tv_sec <= LONG_MAX / 1000000) 435 delta_ticks = (time_change->tv_sec * 1000000 + 436 time_change->tv_usec + (tick - 1)) / tick + 1; 437 else if (time_change->tv_sec <= LONG_MAX / hz) 438 delta_ticks = time_change->tv_sec * hz + 439 (time_change->tv_usec + (tick - 1)) / tick + 1; 440 else 441 delta_ticks = LONG_MAX; 442 443 if (delta_ticks > INT_MAX) 444 delta_ticks = INT_MAX; 445 446 /* 447 * Now rip through the timer calltodo list looking for timers 448 * to expire. 449 */ 450 451 /* don't collide with softclock() */ 452 mtx_lock_spin(&callout_lock); 453 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 454 p->c_time -= delta_ticks; 455 456 /* Break if the timer had more time on it than delta_ticks */ 457 if (p->c_time > 0) 458 break; 459 460 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 461 delta_ticks = -p->c_time; 462 } 463 mtx_unlock_spin(&callout_lock); 464 465 return; 466} 467#endif /* APM_FIXUP_CALLTODO */ 468