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 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * From: @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/callout.h> 43 #include <sys/condvar.h> 44 #include <sys/kernel.h> 45 #include <sys/ktr.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/sysctl.h> 49 50 static int avg_depth; 51 SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0, 52 "Average number of items examined per softclock call. Units = 1/1000"); 53 static int avg_gcalls; 54 SYSCTL_INT(_debug, OID_AUTO, to_avg_gcalls, CTLFLAG_RD, &avg_gcalls, 0, 55 "Average number of Giant callouts made per softclock call. Units = 1/1000"); 56 static int avg_mpcalls; 57 SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0, 58 "Average number of MP callouts made per softclock call. Units = 1/1000"); 59 /* 60 * TODO: 61 * allocate more timeout table slots when table overflows. 62 */ 63 64 /* Exported to machdep.c and/or kern_clock.c. */ 65 struct callout *callout; 66 struct callout_list callfree; 67 int callwheelsize, callwheelbits, callwheelmask; 68 struct callout_tailq *callwheel; 69 int softticks; /* Like ticks, but for softclock(). */ 70 struct mtx callout_lock; 71 #ifdef DIAGNOSTIC 72 struct mtx dont_sleep_in_callout; 73 #endif 74 75 static struct callout *nextsoftcheck; /* Next callout to be checked. */ 76 77 /*- 78 * Locked by callout_lock: 79 * curr_callout - If a callout is in progress, it is curr_callout. 80 * If curr_callout is non-NULL, threads waiting on 81 * callout_wait will be woken up as soon as the 82 * relevant callout completes. 83 * wakeup_ctr - Incremented every time a thread wants to wait 84 * for a callout to complete. Modified only when 85 * curr_callout is non-NULL. 86 * wakeup_needed - If a thread is waiting on callout_wait, then 87 * wakeup_needed is nonzero. Increased only when 88 * cutt_callout is non-NULL. 89 */ 90 static struct callout *curr_callout; 91 static int wakeup_ctr; 92 static int wakeup_needed; 93 94 /*- 95 * Locked by callout_wait_lock: 96 * callout_wait - If wakeup_needed is set, callout_wait will be 97 * triggered after the current callout finishes. 98 * wakeup_done_ctr - Set to the current value of wakeup_ctr after 99 * callout_wait is triggered. 100 */ 101 static struct mtx callout_wait_lock; 102 static struct cv callout_wait; 103 static int wakeup_done_ctr; 104 105 /* 106 * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization 107 * 108 * This code is called very early in the kernel initialization sequence, 109 * and may be called more then once. 110 */ 111 caddr_t 112 kern_timeout_callwheel_alloc(caddr_t v) 113 { 114 /* 115 * Calculate callout wheel size 116 */ 117 for (callwheelsize = 1, callwheelbits = 0; 118 callwheelsize < ncallout; 119 callwheelsize <<= 1, ++callwheelbits) 120 ; 121 callwheelmask = callwheelsize - 1; 122 123 callout = (struct callout *)v; 124 v = (caddr_t)(callout + ncallout); 125 callwheel = (struct callout_tailq *)v; 126 v = (caddr_t)(callwheel + callwheelsize); 127 return(v); 128 } 129 130 /* 131 * kern_timeout_callwheel_init() - initialize previously reserved callwheel 132 * space. 133 * 134 * This code is called just once, after the space reserved for the 135 * callout wheel has been finalized. 136 */ 137 void 138 kern_timeout_callwheel_init(void) 139 { 140 int i; 141 142 SLIST_INIT(&callfree); 143 for (i = 0; i < ncallout; i++) { 144 callout_init(&callout[i], 0); 145 callout[i].c_flags = CALLOUT_LOCAL_ALLOC; 146 SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle); 147 } 148 for (i = 0; i < callwheelsize; i++) { 149 TAILQ_INIT(&callwheel[i]); 150 } 151 mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE); 152 #ifdef DIAGNOSTIC 153 mtx_init(&dont_sleep_in_callout, "dont_sleep_in_callout", NULL, MTX_DEF); 154 #endif 155 mtx_init(&callout_wait_lock, "callout_wait_lock", NULL, MTX_DEF); 156 cv_init(&callout_wait, "callout_wait"); 157 } 158 159 /* 160 * The callout mechanism is based on the work of Adam M. Costello and 161 * George Varghese, published in a technical report entitled "Redesigning 162 * the BSD Callout and Timer Facilities" and modified slightly for inclusion 163 * in FreeBSD by Justin T. Gibbs. The original work on the data structures 164 * used in this implementation was published by G. Varghese and T. Lauck in 165 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for 166 * the Efficient Implementation of a Timer Facility" in the Proceedings of 167 * the 11th ACM Annual Symposium on Operating Systems Principles, 168 * Austin, Texas Nov 1987. 169 */ 170 171 /* 172 * Software (low priority) clock interrupt. 173 * Run periodic events from timeout queue. 174 */ 175 void 176 softclock(void *dummy) 177 { 178 struct callout *c; 179 struct callout_tailq *bucket; 180 int curticks; 181 int steps; /* #steps since we last allowed interrupts */ 182 int depth; 183 int mpcalls; 184 int gcalls; 185 int wakeup_cookie; 186 #ifdef DIAGNOSTIC 187 struct bintime bt1, bt2; 188 struct timespec ts2; 189 static uint64_t maxdt = 36893488147419102LL; /* 2 msec */ 190 static timeout_t *lastfunc; 191 #endif 192 193 #ifndef MAX_SOFTCLOCK_STEPS 194 #define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */ 195 #endif /* MAX_SOFTCLOCK_STEPS */ 196 197 mpcalls = 0; 198 gcalls = 0; 199 depth = 0; 200 steps = 0; 201 mtx_lock_spin(&callout_lock); 202 while (softticks != ticks) { 203 softticks++; 204 /* 205 * softticks may be modified by hard clock, so cache 206 * it while we work on a given bucket. 207 */ 208 curticks = softticks; 209 bucket = &callwheel[curticks & callwheelmask]; 210 c = TAILQ_FIRST(bucket); 211 while (c) { 212 depth++; 213 if (c->c_time != curticks) { 214 c = TAILQ_NEXT(c, c_links.tqe); 215 ++steps; 216 if (steps >= MAX_SOFTCLOCK_STEPS) { 217 nextsoftcheck = c; 218 /* Give interrupts a chance. */ 219 mtx_unlock_spin(&callout_lock); 220 ; /* nothing */ 221 mtx_lock_spin(&callout_lock); 222 c = nextsoftcheck; 223 steps = 0; 224 } 225 } else { 226 void (*c_func)(void *); 227 void *c_arg; 228 int c_flags; 229 230 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 231 TAILQ_REMOVE(bucket, c, c_links.tqe); 232 c_func = c->c_func; 233 c_arg = c->c_arg; 234 c_flags = c->c_flags; 235 c->c_func = NULL; 236 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 237 c->c_flags = CALLOUT_LOCAL_ALLOC; 238 SLIST_INSERT_HEAD(&callfree, c, 239 c_links.sle); 240 } else { 241 c->c_flags = 242 (c->c_flags & ~CALLOUT_PENDING); 243 } 244 curr_callout = c; 245 mtx_unlock_spin(&callout_lock); 246 if (!(c_flags & CALLOUT_MPSAFE)) { 247 mtx_lock(&Giant); 248 gcalls++; 249 CTR1(KTR_CALLOUT, "callout %p", c_func); 250 } else { 251 mpcalls++; 252 CTR1(KTR_CALLOUT, "callout mpsafe %p", 253 c_func); 254 } 255 #ifdef DIAGNOSTIC 256 binuptime(&bt1); 257 mtx_lock(&dont_sleep_in_callout); 258 #endif 259 c_func(c_arg); 260 #ifdef DIAGNOSTIC 261 mtx_unlock(&dont_sleep_in_callout); 262 binuptime(&bt2); 263 bintime_sub(&bt2, &bt1); 264 if (bt2.frac > maxdt) { 265 if (lastfunc != c_func || 266 bt2.frac > maxdt * 2) { 267 bintime2timespec(&bt2, &ts2); 268 printf( 269 "Expensive timeout(9) function: %p(%p) %jd.%09ld s\n", 270 c_func, c_arg, 271 (intmax_t)ts2.tv_sec, 272 ts2.tv_nsec); 273 } 274 maxdt = bt2.frac; 275 lastfunc = c_func; 276 } 277 #endif 278 if (!(c_flags & CALLOUT_MPSAFE)) 279 mtx_unlock(&Giant); 280 mtx_lock_spin(&callout_lock); 281 curr_callout = NULL; 282 if (wakeup_needed) { 283 /* 284 * There might be someone waiting 285 * for the callout to complete. 286 */ 287 wakeup_cookie = wakeup_ctr; 288 mtx_unlock_spin(&callout_lock); 289 mtx_lock(&callout_wait_lock); 290 cv_broadcast(&callout_wait); 291 wakeup_done_ctr = wakeup_cookie; 292 mtx_unlock(&callout_wait_lock); 293 mtx_lock_spin(&callout_lock); 294 wakeup_needed = 0; 295 } 296 steps = 0; 297 c = nextsoftcheck; 298 } 299 } 300 } 301 avg_depth += (depth * 1000 - avg_depth) >> 8; 302 avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8; 303 avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8; 304 nextsoftcheck = NULL; 305 mtx_unlock_spin(&callout_lock); 306 } 307 308 /* 309 * timeout -- 310 * Execute a function after a specified length of time. 311 * 312 * untimeout -- 313 * Cancel previous timeout function call. 314 * 315 * callout_handle_init -- 316 * Initialize a handle so that using it with untimeout is benign. 317 * 318 * See AT&T BCI Driver Reference Manual for specification. This 319 * implementation differs from that one in that although an 320 * identification value is returned from timeout, the original 321 * arguments to timeout as well as the identifier are used to 322 * identify entries for untimeout. 323 */ 324 struct callout_handle 325 timeout(ftn, arg, to_ticks) 326 timeout_t *ftn; 327 void *arg; 328 int to_ticks; 329 { 330 struct callout *new; 331 struct callout_handle handle; 332 333 mtx_lock_spin(&callout_lock); 334 335 /* Fill in the next free callout structure. */ 336 new = SLIST_FIRST(&callfree); 337 if (new == NULL) 338 /* XXX Attempt to malloc first */ 339 panic("timeout table full"); 340 SLIST_REMOVE_HEAD(&callfree, c_links.sle); 341 342 callout_reset(new, to_ticks, ftn, arg); 343 344 handle.callout = new; 345 mtx_unlock_spin(&callout_lock); 346 return (handle); 347 } 348 349 void 350 untimeout(ftn, arg, handle) 351 timeout_t *ftn; 352 void *arg; 353 struct callout_handle handle; 354 { 355 356 /* 357 * Check for a handle that was initialized 358 * by callout_handle_init, but never used 359 * for a real timeout. 360 */ 361 if (handle.callout == NULL) 362 return; 363 364 mtx_lock_spin(&callout_lock); 365 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) 366 callout_stop(handle.callout); 367 mtx_unlock_spin(&callout_lock); 368 } 369 370 void 371 callout_handle_init(struct callout_handle *handle) 372 { 373 handle->callout = NULL; 374 } 375 376 /* 377 * New interface; clients allocate their own callout structures. 378 * 379 * callout_reset() - establish or change a timeout 380 * callout_stop() - disestablish a timeout 381 * callout_init() - initialize a callout structure so that it can 382 * safely be passed to callout_reset() and callout_stop() 383 * 384 * <sys/callout.h> defines three convenience macros: 385 * 386 * callout_active() - returns truth if callout has not been serviced 387 * callout_pending() - returns truth if callout is still waiting for timeout 388 * callout_deactivate() - marks the callout as having been serviced 389 */ 390 void 391 callout_reset(c, to_ticks, ftn, arg) 392 struct callout *c; 393 int to_ticks; 394 void (*ftn)(void *); 395 void *arg; 396 { 397 398 mtx_lock_spin(&callout_lock); 399 if (c == curr_callout && wakeup_needed) { 400 /* 401 * We're being asked to reschedule a callout which is 402 * currently in progress, and someone has called 403 * callout_drain to kill that callout. Don't reschedule. 404 */ 405 mtx_unlock_spin(&callout_lock); 406 return; 407 } 408 if (c->c_flags & CALLOUT_PENDING) { 409 if (nextsoftcheck == c) { 410 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 411 } 412 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, 413 c_links.tqe); 414 415 /* 416 * Part of the normal "stop a pending callout" process 417 * is to clear the CALLOUT_ACTIVE and CALLOUT_PENDING 418 * flags. We're not going to bother doing that here, 419 * because we're going to be setting those flags ten lines 420 * after this point, and we're holding callout_lock 421 * between now and then. 422 */ 423 } 424 425 /* 426 * We could unlock callout_lock here and lock it again before the 427 * TAILQ_INSERT_TAIL, but there's no point since doing this setup 428 * doesn't take much time. 429 */ 430 if (to_ticks <= 0) 431 to_ticks = 1; 432 433 c->c_arg = arg; 434 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 435 c->c_func = ftn; 436 c->c_time = ticks + to_ticks; 437 TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask], 438 c, c_links.tqe); 439 mtx_unlock_spin(&callout_lock); 440 } 441 442 int 443 _callout_stop_safe(c, safe) 444 struct callout *c; 445 int safe; 446 { 447 int wakeup_cookie; 448 449 mtx_lock_spin(&callout_lock); 450 /* 451 * Don't attempt to delete a callout that's not on the queue. 452 */ 453 if (!(c->c_flags & CALLOUT_PENDING)) { 454 c->c_flags &= ~CALLOUT_ACTIVE; 455 if (c == curr_callout && safe) { 456 /* We need to wait until the callout is finished. */ 457 wakeup_needed = 1; 458 wakeup_cookie = wakeup_ctr++; 459 mtx_unlock_spin(&callout_lock); 460 mtx_lock(&callout_wait_lock); 461 462 /* 463 * Check to make sure that softclock() didn't 464 * do the wakeup in between our dropping 465 * callout_lock and picking up callout_wait_lock 466 */ 467 if (wakeup_cookie - wakeup_done_ctr > 0) 468 cv_wait(&callout_wait, &callout_wait_lock); 469 470 mtx_unlock(&callout_wait_lock); 471 } else 472 mtx_unlock_spin(&callout_lock); 473 return (0); 474 } 475 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 476 477 if (nextsoftcheck == c) { 478 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe); 479 } 480 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe); 481 c->c_func = NULL; 482 483 if (c->c_flags & CALLOUT_LOCAL_ALLOC) { 484 SLIST_INSERT_HEAD(&callfree, c, c_links.sle); 485 } 486 mtx_unlock_spin(&callout_lock); 487 return (1); 488 } 489 490 void 491 callout_init(c, mpsafe) 492 struct callout *c; 493 int mpsafe; 494 { 495 bzero(c, sizeof *c); 496 if (mpsafe) 497 c->c_flags |= CALLOUT_MPSAFE; 498 } 499 500 #ifdef APM_FIXUP_CALLTODO 501 /* 502 * Adjust the kernel calltodo timeout list. This routine is used after 503 * an APM resume to recalculate the calltodo timer list values with the 504 * number of hz's we have been sleeping. The next hardclock() will detect 505 * that there are fired timers and run softclock() to execute them. 506 * 507 * Please note, I have not done an exhaustive analysis of what code this 508 * might break. I am motivated to have my select()'s and alarm()'s that 509 * have expired during suspend firing upon resume so that the applications 510 * which set the timer can do the maintanence the timer was for as close 511 * as possible to the originally intended time. Testing this code for a 512 * week showed that resuming from a suspend resulted in 22 to 25 timers 513 * firing, which seemed independant on whether the suspend was 2 hours or 514 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 515 */ 516 void 517 adjust_timeout_calltodo(time_change) 518 struct timeval *time_change; 519 { 520 register struct callout *p; 521 unsigned long delta_ticks; 522 523 /* 524 * How many ticks were we asleep? 525 * (stolen from tvtohz()). 526 */ 527 528 /* Don't do anything */ 529 if (time_change->tv_sec < 0) 530 return; 531 else if (time_change->tv_sec <= LONG_MAX / 1000000) 532 delta_ticks = (time_change->tv_sec * 1000000 + 533 time_change->tv_usec + (tick - 1)) / tick + 1; 534 else if (time_change->tv_sec <= LONG_MAX / hz) 535 delta_ticks = time_change->tv_sec * hz + 536 (time_change->tv_usec + (tick - 1)) / tick + 1; 537 else 538 delta_ticks = LONG_MAX; 539 540 if (delta_ticks > INT_MAX) 541 delta_ticks = INT_MAX; 542 543 /* 544 * Now rip through the timer calltodo list looking for timers 545 * to expire. 546 */ 547 548 /* don't collide with softclock() */ 549 mtx_lock_spin(&callout_lock); 550 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 551 p->c_time -= delta_ticks; 552 553 /* Break if the timer had more time on it than delta_ticks */ 554 if (p->c_time > 0) 555 break; 556 557 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 558 delta_ticks = -p->c_time; 559 } 560 mtx_unlock_spin(&callout_lock); 561 562 return; 563 } 564 #endif /* APM_FIXUP_CALLTODO */ 565