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("$FreeBSD$"); 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 52 static int avg_depth; 53 SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0, 54 "Average number of items examined per softclock call. Units = 1/1000"); 55 static int avg_gcalls; 56 SYSCTL_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"); 58 static int avg_mpcalls; 59 SYSCTL_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. */ 67 struct callout *callout; 68 struct callout_list callfree; 69 int callwheelsize, callwheelbits, callwheelmask; 70 struct callout_tailq *callwheel; 71 int softticks; /* Like ticks, but for softclock(). */ 72 struct mtx callout_lock; 73 #ifdef DIAGNOSTIC 74 struct mtx callout_dont_sleep; 75 #endif 76 77 static struct callout *nextsoftcheck; /* Next callout to be checked. */ 78 79 /* 80 * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization 81 * 82 * This code is called very early in the kernel initialization sequence, 83 * and may be called more then once. 84 */ 85 caddr_t 86 kern_timeout_callwheel_alloc(caddr_t v) 87 { 88 /* 89 * Calculate callout wheel size 90 */ 91 for (callwheelsize = 1, callwheelbits = 0; 92 callwheelsize < ncallout; 93 callwheelsize <<= 1, ++callwheelbits) 94 ; 95 callwheelmask = callwheelsize - 1; 96 97 callout = (struct callout *)v; 98 v = (caddr_t)(callout + ncallout); 99 callwheel = (struct callout_tailq *)v; 100 v = (caddr_t)(callwheel + callwheelsize); 101 return(v); 102 } 103 104 /* 105 * kern_timeout_callwheel_init() - initialize previously reserved callwheel 106 * space. 107 * 108 * This code is called just once, after the space reserved for the 109 * callout wheel has been finalized. 110 */ 111 void 112 kern_timeout_callwheel_init(void) 113 { 114 int i; 115 116 SLIST_INIT(&callfree); 117 for (i = 0; i < ncallout; i++) { 118 callout_init(&callout[i], 0); 119 callout[i].c_flags = CALLOUT_LOCAL_ALLOC; 120 SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle); 121 } 122 for (i = 0; i < callwheelsize; i++) { 123 TAILQ_INIT(&callwheel[i]); 124 } 125 mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); 126 #ifdef DIAGNOSTIC 127 mtx_init(&callout_dont_sleep, "callout_dont_sleep", NULL, MTX_DEF); 128 #endif 129 } 130 131 /* 132 * The callout mechanism is based on the work of Adam M. Costello and 133 * George Varghese, published in a technical report entitled "Redesigning 134 * the BSD Callout and Timer Facilities" and modified slightly for inclusion 135 * in FreeBSD by Justin T. Gibbs. The original work on the data structures 136 * used in this implementation was published by G.Varghese and A. Lauck in 137 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for 138 * the Efficient Implementation of a Timer Facility" in the Proceedings of 139 * the 11th ACM Annual Symposium on Operating Systems Principles, 140 * Austin, Texas Nov 1987. 141 */ 142 143 /* 144 * Software (low priority) clock interrupt. 145 * Run periodic events from timeout queue. 146 */ 147 void 148 softclock(void *dummy) 149 { 150 struct callout *c; 151 struct callout_tailq *bucket; 152 int curticks; 153 int steps; /* #steps since we last allowed interrupts */ 154 int depth; 155 int mpcalls; 156 int gcalls; 157 #ifdef DIAGNOSTIC 158 struct bintime bt1, bt2; 159 struct timespec ts2; 160 static uint64_t maxdt = 18446744073709551LL; /* 1 msec */ 161 #endif 162 163 #ifndef MAX_SOFTCLOCK_STEPS 164 #define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */ 165 #endif /* MAX_SOFTCLOCK_STEPS */ 166 167 mpcalls = 0; 168 gcalls = 0; 169 depth = 0; 170 steps = 0; 171 mtx_lock_spin(&callout_lock); 172 while (softticks != ticks) { 173 softticks++; 174 /* 175 * softticks may be modified by hard clock, so cache 176 * it while we work on a given bucket. 177 */ 178 curticks = softticks; 179 bucket = &callwheel[curticks & callwheelmask]; 180 c = TAILQ_FIRST(bucket); 181 while (c) { 182 depth++; 183 if (c->c_time != curticks) { 184 c = TAILQ_NEXT(c, c_links.tqe); 185 ++steps; 186 if (steps >= MAX_SOFTCLOCK_STEPS) { 187 nextsoftcheck = c; 188 /* Give interrupts a chance. */ 189 mtx_unlock_spin(&callout_lock); 190 ; /* nothing */ 191 mtx_lock_spin(&callout_lock); 192 c = nextsoftcheck; 193 steps = 0; 194 } 195 } else { 196 void (*c_func)(void *); 197 void *c_arg; 198 int c_flags; 199 200 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 201 TAILQ_REMOVE(bucket, c, c_links.tqe); 202 c_func = c->c_func; 203 c_arg = c->c_arg; 204 c_flags = c->c_flags; 205 c->c_func = NULL; 206 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 207 c->c_flags = CALLOUT_LOCAL_ALLOC; 208 SLIST_INSERT_HEAD(&callfree, c, 209 c_links.sle); 210 } else { 211 c->c_flags = 212 (c->c_flags & ~CALLOUT_PENDING); 213 } 214 mtx_unlock_spin(&callout_lock); 215 if (!(c_flags & CALLOUT_MPSAFE)) { 216 mtx_lock(&Giant); 217 gcalls++; 218 } else { 219 mpcalls++; 220 } 221 #ifdef DIAGNOSTIC 222 binuptime(&bt1); 223 mtx_lock(&callout_dont_sleep); 224 #endif 225 c_func(c_arg); 226 #ifdef DIAGNOSTIC 227 mtx_unlock(&callout_dont_sleep); 228 binuptime(&bt2); 229 bintime_sub(&bt2, &bt1); 230 if (bt2.frac > maxdt) { 231 maxdt = bt2.frac; 232 bintime2timespec(&bt2, &ts2); 233 printf( 234 "Expensive timeout(9) function: %p(%p) %ld.%09ld s\n", 235 c_func, c_arg, 236 (long)ts2.tv_sec, ts2.tv_nsec); 237 } 238 #endif 239 if (!(c_flags & CALLOUT_MPSAFE)) 240 mtx_unlock(&Giant); 241 mtx_lock_spin(&callout_lock); 242 steps = 0; 243 c = nextsoftcheck; 244 } 245 } 246 } 247 avg_depth += (depth * 1000 - avg_depth) >> 8; 248 avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8; 249 avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8; 250 nextsoftcheck = NULL; 251 mtx_unlock_spin(&callout_lock); 252 } 253 254 /* 255 * timeout -- 256 * Execute a function after a specified length of time. 257 * 258 * untimeout -- 259 * Cancel previous timeout function call. 260 * 261 * callout_handle_init -- 262 * Initialize a handle so that using it with untimeout is benign. 263 * 264 * See AT&T BCI Driver Reference Manual for specification. This 265 * implementation differs from that one in that although an 266 * identification value is returned from timeout, the original 267 * arguments to timeout as well as the identifier are used to 268 * identify entries for untimeout. 269 */ 270 struct callout_handle 271 timeout(ftn, arg, to_ticks) 272 timeout_t *ftn; 273 void *arg; 274 int to_ticks; 275 { 276 struct callout *new; 277 struct callout_handle handle; 278 279 mtx_lock_spin(&callout_lock); 280 281 /* Fill in the next free callout structure. */ 282 new = SLIST_FIRST(&callfree); 283 if (new == NULL) 284 /* XXX Attempt to malloc first */ 285 panic("timeout table full"); 286 SLIST_REMOVE_HEAD(&callfree, c_links.sle); 287 288 callout_reset(new, to_ticks, ftn, arg); 289 290 handle.callout = new; 291 mtx_unlock_spin(&callout_lock); 292 return (handle); 293 } 294 295 void 296 untimeout(ftn, arg, handle) 297 timeout_t *ftn; 298 void *arg; 299 struct callout_handle handle; 300 { 301 302 /* 303 * Check for a handle that was initialized 304 * by callout_handle_init, but never used 305 * for a real timeout. 306 */ 307 if (handle.callout == NULL) 308 return; 309 310 mtx_lock_spin(&callout_lock); 311 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) 312 callout_stop(handle.callout); 313 mtx_unlock_spin(&callout_lock); 314 } 315 316 void 317 callout_handle_init(struct callout_handle *handle) 318 { 319 handle->callout = NULL; 320 } 321 322 /* 323 * New interface; clients allocate their own callout structures. 324 * 325 * callout_reset() - establish or change a timeout 326 * callout_stop() - disestablish a timeout 327 * callout_init() - initialize a callout structure so that it can 328 * safely be passed to callout_reset() and callout_stop() 329 * 330 * <sys/callout.h> defines three convenience macros: 331 * 332 * callout_active() - returns truth if callout has not been serviced 333 * callout_pending() - returns truth if callout is still waiting for timeout 334 * callout_deactivate() - marks the callout as having been serviced 335 */ 336 void 337 callout_reset(c, to_ticks, ftn, arg) 338 struct callout *c; 339 int to_ticks; 340 void (*ftn)(void *); 341 void *arg; 342 { 343 344 mtx_lock_spin(&callout_lock); 345 if (c->c_flags & CALLOUT_PENDING) 346 callout_stop(c); 347 348 /* 349 * We could unlock callout_lock here and lock it again before the 350 * TAILQ_INSERT_TAIL, but there's no point since doing this setup 351 * doesn't take much time. 352 */ 353 if (to_ticks <= 0) 354 to_ticks = 1; 355 356 c->c_arg = arg; 357 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 358 c->c_func = ftn; 359 c->c_time = ticks + to_ticks; 360 TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask], 361 c, c_links.tqe); 362 mtx_unlock_spin(&callout_lock); 363 } 364 365 int 366 callout_stop(c) 367 struct callout *c; 368 { 369 370 mtx_lock_spin(&callout_lock); 371 /* 372 * Don't attempt to delete a callout that's not on the queue. 373 */ 374 if (!(c->c_flags & CALLOUT_PENDING)) { 375 c->c_flags &= ~CALLOUT_ACTIVE; 376 mtx_unlock_spin(&callout_lock); 377 return (0); 378 } 379 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 380 381 if (nextsoftcheck == c) { 382 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 383 } 384 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe); 385 c->c_func = NULL; 386 387 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 388 SLIST_INSERT_HEAD(&callfree, c, c_links.sle); 389 } 390 mtx_unlock_spin(&callout_lock); 391 return (1); 392 } 393 394 void 395 callout_init(c, mpsafe) 396 struct callout *c; 397 int mpsafe; 398 { 399 bzero(c, sizeof *c); 400 if (mpsafe) 401 c->c_flags |= CALLOUT_MPSAFE; 402 } 403 404 #ifdef APM_FIXUP_CALLTODO 405 /* 406 * Adjust the kernel calltodo timeout list. This routine is used after 407 * an APM resume to recalculate the calltodo timer list values with the 408 * number of hz's we have been sleeping. The next hardclock() will detect 409 * that there are fired timers and run softclock() to execute them. 410 * 411 * Please note, I have not done an exhaustive analysis of what code this 412 * might break. I am motivated to have my select()'s and alarm()'s that 413 * have expired during suspend firing upon resume so that the applications 414 * which set the timer can do the maintanence the timer was for as close 415 * as possible to the originally intended time. Testing this code for a 416 * week showed that resuming from a suspend resulted in 22 to 25 timers 417 * firing, which seemed independant on whether the suspend was 2 hours or 418 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 419 */ 420 void 421 adjust_timeout_calltodo(time_change) 422 struct timeval *time_change; 423 { 424 register struct callout *p; 425 unsigned long delta_ticks; 426 427 /* 428 * How many ticks were we asleep? 429 * (stolen from tvtohz()). 430 */ 431 432 /* Don't do anything */ 433 if (time_change->tv_sec < 0) 434 return; 435 else if (time_change->tv_sec <= LONG_MAX / 1000000) 436 delta_ticks = (time_change->tv_sec * 1000000 + 437 time_change->tv_usec + (tick - 1)) / tick + 1; 438 else if (time_change->tv_sec <= LONG_MAX / hz) 439 delta_ticks = time_change->tv_sec * hz + 440 (time_change->tv_usec + (tick - 1)) / tick + 1; 441 else 442 delta_ticks = LONG_MAX; 443 444 if (delta_ticks > INT_MAX) 445 delta_ticks = INT_MAX; 446 447 /* 448 * Now rip through the timer calltodo list looking for timers 449 * to expire. 450 */ 451 452 /* don't collide with softclock() */ 453 mtx_lock_spin(&callout_lock); 454 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 455 p->c_time -= delta_ticks; 456 457 /* Break if the timer had more time on it than delta_ticks */ 458 if (p->c_time > 0) 459 break; 460 461 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 462 delta_ticks = -p->c_time; 463 } 464 mtx_unlock_spin(&callout_lock); 465 466 return; 467 } 468 #endif /* APM_FIXUP_CALLTODO */ 469