1 /* 2 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 /* 35 * Copyright (c) 1982, 1986, 1991, 1993 36 * The Regents of the University of California. All rights reserved. 37 * (c) UNIX System Laboratories, Inc. 38 * All or some portions of this file are derived from material licensed 39 * to the University of California by American Telephone and Telegraph 40 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 41 * the permission of UNIX System Laboratories, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by the University of 54 * California, Berkeley and its contributors. 55 * 4. Neither the name of the University nor the names of its contributors 56 * may be used to endorse or promote products derived from this software 57 * without specific prior written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 * 71 * From: @(#)kern_clock.c 8.5 (Berkeley) 1/21/94 72 * $FreeBSD: src/sys/kern/kern_timeout.c,v 1.59.2.1 2001/11/13 18:24:52 archie Exp $ 73 * $DragonFly: src/sys/kern/kern_timeout.c,v 1.27 2007/11/14 18:27:52 swildner Exp $ 74 */ 75 /* 76 * DRAGONFLY BGL STATUS 77 * 78 * All the API functions should be MP safe. 79 * 80 * The callback functions will be flagged as being MP safe if the 81 * timeout structure is initialized with callout_init_mp() instead of 82 * callout_init(). 83 * 84 * The helper threads cannot be made preempt-capable until after we 85 * clean up all the uses of splsoftclock() and related interlocks (which 86 * require the related functions to be MP safe as well). 87 */ 88 /* 89 * The callout mechanism is based on the work of Adam M. Costello and 90 * George Varghese, published in a technical report entitled "Redesigning 91 * the BSD Callout and Timer Facilities" and modified slightly for inclusion 92 * in FreeBSD by Justin T. Gibbs. The original work on the data structures 93 * used in this implementation was published by G. Varghese and T. Lauck in 94 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for 95 * the Efficient Implementation of a Timer Facility" in the Proceedings of 96 * the 11th ACM Annual Symposium on Operating Systems Principles, 97 * Austin, Texas Nov 1987. 98 * 99 * The per-cpu augmentation was done by Matthew Dillon. 100 */ 101 102 #include "opt_ddb.h" 103 104 #include <sys/param.h> 105 #include <sys/systm.h> 106 #include <sys/callout.h> 107 #include <sys/kernel.h> 108 #include <sys/interrupt.h> 109 #include <sys/thread.h> 110 #include <sys/thread2.h> 111 #include <ddb/ddb.h> 112 113 #ifndef MAX_SOFTCLOCK_STEPS 114 #define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */ 115 #endif 116 117 118 struct softclock_pcpu { 119 struct callout_tailq *callwheel; 120 struct callout * volatile next; 121 int softticks; /* softticks index */ 122 int curticks; /* per-cpu ticks counter */ 123 int isrunning; 124 struct thread thread; 125 126 }; 127 128 typedef struct softclock_pcpu *softclock_pcpu_t; 129 130 /* 131 * TODO: 132 * allocate more timeout table slots when table overflows. 133 */ 134 static MALLOC_DEFINE(M_CALLOUT, "callout", "callout structures"); 135 static int callwheelsize; 136 static int callwheelbits; 137 static int callwheelmask; 138 static struct softclock_pcpu softclock_pcpu_ary[MAXCPU]; 139 140 static void softclock_handler(void *arg); 141 142 static void 143 swi_softclock_setup(void *arg) 144 { 145 int cpu; 146 int i; 147 148 /* 149 * Figure out how large a callwheel we need. It must be a power of 2. 150 */ 151 callwheelsize = 1; 152 callwheelbits = 0; 153 while (callwheelsize < ncallout) { 154 callwheelsize <<= 1; 155 ++callwheelbits; 156 } 157 callwheelmask = callwheelsize - 1; 158 159 /* 160 * Initialize per-cpu data structures. 161 */ 162 for (cpu = 0; cpu < ncpus; ++cpu) { 163 softclock_pcpu_t sc; 164 165 sc = &softclock_pcpu_ary[cpu]; 166 167 sc->callwheel = kmalloc(sizeof(*sc->callwheel) * callwheelsize, 168 M_CALLOUT, M_WAITOK|M_ZERO); 169 for (i = 0; i < callwheelsize; ++i) 170 TAILQ_INIT(&sc->callwheel[i]); 171 172 /* 173 * Create a preemption-capable thread for each cpu to handle 174 * softclock timeouts on that cpu. The preemption can only 175 * be blocked by a critical section. The thread can itself 176 * be preempted by normal interrupts. 177 */ 178 lwkt_create(softclock_handler, sc, NULL, 179 &sc->thread, TDF_STOPREQ|TDF_INTTHREAD, cpu, 180 "softclock %d", cpu); 181 #if 0 182 /* 183 * Do not make the thread preemptable until we clean up all 184 * the splsoftclock() calls in the system. Since the threads 185 * are no longer operated as a software interrupt, the 186 * splsoftclock() calls will not have any effect on them. 187 */ 188 sc->thread.td_preemptable = lwkt_preempt; 189 #endif 190 } 191 } 192 193 /* 194 * Must occur after ncpus has been initialized. 195 */ 196 SYSINIT(softclock_setup, SI_BOOT2_SOFTCLOCK, SI_ORDER_SECOND, 197 swi_softclock_setup, NULL); 198 199 /* 200 * This routine is called from the hardclock() (basically a FASTint/IPI) on 201 * each cpu in the system. sc->curticks is this cpu's notion of the timebase. 202 * It IS NOT NECESSARILY SYNCHRONIZED WITH 'ticks'! sc->softticks is where 203 * the callwheel is currently indexed. 204 * 205 * WARNING! The MP lock is not necessarily held on call, nor can it be 206 * safely obtained. 207 * 208 * sc->softticks is adjusted by either this routine or our helper thread 209 * depending on whether the helper thread is running or not. 210 */ 211 void 212 hardclock_softtick(globaldata_t gd) 213 { 214 softclock_pcpu_t sc; 215 216 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 217 ++sc->curticks; 218 if (sc->isrunning) 219 return; 220 if (sc->softticks == sc->curticks) { 221 /* 222 * in sync, only wakeup the thread if there is something to 223 * do. 224 */ 225 if (TAILQ_FIRST(&sc->callwheel[sc->softticks & callwheelmask])) 226 { 227 sc->isrunning = 1; 228 lwkt_schedule(&sc->thread); 229 } else { 230 ++sc->softticks; 231 } 232 } else { 233 /* 234 * out of sync, wakeup the thread unconditionally so it can 235 * catch up. 236 */ 237 sc->isrunning = 1; 238 lwkt_schedule(&sc->thread); 239 } 240 } 241 242 /* 243 * This procedure is the main loop of our per-cpu helper thread. The 244 * sc->isrunning flag prevents us from racing hardclock_softtick() and 245 * a critical section is sufficient to interlock sc->curticks and protect 246 * us from remote IPI's / list removal. 247 * 248 * The thread starts with the MP lock held and not in a critical section. 249 * The loop itself is MP safe while individual callbacks may or may not 250 * be, so we obtain or release the MP lock as appropriate. 251 */ 252 static void 253 softclock_handler(void *arg) 254 { 255 softclock_pcpu_t sc; 256 struct callout *c; 257 struct callout_tailq *bucket; 258 void (*c_func)(void *); 259 void *c_arg; 260 #ifdef SMP 261 int mpsafe = 0; 262 #endif 263 264 lwkt_setpri_self(TDPRI_SOFT_NORM); 265 266 sc = arg; 267 crit_enter(); 268 loop: 269 while (sc->softticks != (int)(sc->curticks + 1)) { 270 bucket = &sc->callwheel[sc->softticks & callwheelmask]; 271 272 for (c = TAILQ_FIRST(bucket); c; c = sc->next) { 273 if (c->c_time != sc->softticks) { 274 sc->next = TAILQ_NEXT(c, c_links.tqe); 275 continue; 276 } 277 #ifdef SMP 278 if (c->c_flags & CALLOUT_MPSAFE) { 279 if (mpsafe == 0) { 280 mpsafe = 1; 281 rel_mplock(); 282 } 283 } else { 284 /* 285 * The request might be removed while we 286 * are waiting to get the MP lock. If it 287 * was removed sc->next will point to the 288 * next valid request or NULL, loop up. 289 */ 290 if (mpsafe) { 291 mpsafe = 0; 292 sc->next = c; 293 get_mplock(); 294 if (c != sc->next) 295 continue; 296 } 297 } 298 #endif 299 sc->next = TAILQ_NEXT(c, c_links.tqe); 300 TAILQ_REMOVE(bucket, c, c_links.tqe); 301 302 c_func = c->c_func; 303 c_arg = c->c_arg; 304 c->c_func = NULL; 305 KKASSERT(c->c_flags & CALLOUT_DID_INIT); 306 c->c_flags &= ~CALLOUT_PENDING; 307 crit_exit(); 308 c_func(c_arg); 309 crit_enter(); 310 /* NOTE: list may have changed */ 311 } 312 ++sc->softticks; 313 } 314 sc->isrunning = 0; 315 lwkt_deschedule_self(&sc->thread); /* == curthread */ 316 lwkt_switch(); 317 goto loop; 318 /* NOT REACHED */ 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 * callout_init_mp() - same but any installed functions must be MP safe. 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 */ 336 337 /* 338 * Start or restart a timeout. Install the callout structure in the 339 * callwheel. Callers may legally pass any value, even if 0 or negative, 340 * but since the sc->curticks index may have already been processed a 341 * minimum timeout of 1 tick will be enforced. 342 * 343 * The callout is installed on and will be processed on the current cpu's 344 * callout wheel. 345 * 346 * WARNING! This function may be called from any cpu but the caller must 347 * serialize callout_stop() and callout_reset() calls on the passed 348 * structure regardless of cpu. 349 */ 350 void 351 callout_reset(struct callout *c, int to_ticks, void (*ftn)(void *), 352 void *arg) 353 { 354 softclock_pcpu_t sc; 355 globaldata_t gd; 356 357 #ifdef INVARIANTS 358 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 359 callout_init(c); 360 kprintf( 361 "callout_reset(%p) from %p: callout was not initialized\n", 362 c, ((int **)&c)[-1]); 363 #ifdef DDB 364 db_print_backtrace(); 365 #endif 366 } 367 #endif 368 gd = mycpu; 369 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 370 crit_enter_gd(gd); 371 372 if (c->c_flags & CALLOUT_PENDING) 373 callout_stop(c); 374 375 if (to_ticks <= 0) 376 to_ticks = 1; 377 378 c->c_arg = arg; 379 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 380 c->c_func = ftn; 381 c->c_time = sc->curticks + to_ticks; 382 #ifdef SMP 383 c->c_gd = gd; 384 #endif 385 386 TAILQ_INSERT_TAIL(&sc->callwheel[c->c_time & callwheelmask], 387 c, c_links.tqe); 388 crit_exit_gd(gd); 389 } 390 391 /* 392 * Stop a running timer. WARNING! If called on a cpu other then the one 393 * the callout was started on this function will liveloop on its IPI to 394 * the target cpu to process the request. It is possible for the callout 395 * to execute in that case. 396 * 397 * WARNING! This function may be called from any cpu but the caller must 398 * serialize callout_stop() and callout_reset() calls on the passed 399 * structure regardless of cpu. 400 * 401 * WARNING! This routine may be called from an IPI 402 */ 403 int 404 callout_stop(struct callout *c) 405 { 406 globaldata_t gd = mycpu; 407 #ifdef SMP 408 globaldata_t tgd; 409 #endif 410 softclock_pcpu_t sc; 411 412 #ifdef INVARIANTS 413 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 414 callout_init(c); 415 kprintf( 416 "callout_stop(%p) from %p: callout was not initialized\n", 417 c, ((int **)&c)[-1]); 418 #ifdef DDB 419 db_print_backtrace(); 420 #endif 421 } 422 #endif 423 crit_enter_gd(gd); 424 425 /* 426 * Don't attempt to delete a callout that's not on the queue. The 427 * callout may not have a cpu assigned to it. Callers do not have 428 * to be on the issuing cpu but must still serialize access to the 429 * callout structure. 430 * 431 * We are not cpu-localized here and cannot safely modify the 432 * flags field in the callout structure. Note that most of the 433 * time CALLOUT_ACTIVE will be 0 if CALLOUT_PENDING is also 0. 434 * 435 * If we race another cpu's dispatch of this callout it is possible 436 * for CALLOUT_ACTIVE to be set with CALLOUT_PENDING unset. This 437 * will cause us to fall through and synchronize with the other 438 * cpu. 439 */ 440 if ((c->c_flags & CALLOUT_PENDING) == 0) { 441 #ifdef SMP 442 if ((c->c_flags & CALLOUT_ACTIVE) == 0) { 443 crit_exit_gd(gd); 444 return (0); 445 } 446 if (c->c_gd == NULL || c->c_gd == gd) { 447 c->c_flags &= ~CALLOUT_ACTIVE; 448 crit_exit_gd(gd); 449 return (0); 450 } 451 /* fall-through to the cpu-localization code. */ 452 #else 453 c->c_flags &= ~CALLOUT_ACTIVE; 454 crit_exit_gd(gd); 455 return (0); 456 #endif 457 } 458 #ifdef SMP 459 if ((tgd = c->c_gd) != gd) { 460 /* 461 * If the callout is owned by a different CPU we have to 462 * execute the function synchronously on the target cpu. 463 */ 464 int seq; 465 466 cpu_ccfence(); /* don't let tgd alias c_gd */ 467 seq = lwkt_send_ipiq(tgd, (void *)callout_stop, c); 468 lwkt_wait_ipiq(tgd, seq); 469 } else 470 #endif 471 { 472 /* 473 * If the callout is owned by the same CPU we can 474 * process it directly, but if we are racing our helper 475 * thread (sc->next), we have to adjust sc->next. The 476 * race is interlocked by a critical section. 477 */ 478 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 479 480 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 481 if (sc->next == c) 482 sc->next = TAILQ_NEXT(c, c_links.tqe); 483 484 TAILQ_REMOVE(&sc->callwheel[c->c_time & callwheelmask], 485 c, c_links.tqe); 486 c->c_func = NULL; 487 } 488 crit_exit_gd(gd); 489 return (1); 490 } 491 492 /* 493 * Prepare a callout structure for use by callout_reset() and/or 494 * callout_stop(). The MP version of this routine requires that the callback 495 * function installed by callout_reset() be MP safe. 496 */ 497 void 498 callout_init(struct callout *c) 499 { 500 bzero(c, sizeof *c); 501 c->c_flags = CALLOUT_DID_INIT; 502 } 503 504 void 505 callout_init_mp(struct callout *c) 506 { 507 callout_init(c); 508 c->c_flags |= CALLOUT_MPSAFE; 509 } 510 511 /* What, are you joking? This is nuts! -Matt */ 512 #if 0 513 #ifdef APM_FIXUP_CALLTODO 514 /* 515 * Adjust the kernel calltodo timeout list. This routine is used after 516 * an APM resume to recalculate the calltodo timer list values with the 517 * number of hz's we have been sleeping. The next hardclock() will detect 518 * that there are fired timers and run softclock() to execute them. 519 * 520 * Please note, I have not done an exhaustive analysis of what code this 521 * might break. I am motivated to have my select()'s and alarm()'s that 522 * have expired during suspend firing upon resume so that the applications 523 * which set the timer can do the maintanence the timer was for as close 524 * as possible to the originally intended time. Testing this code for a 525 * week showed that resuming from a suspend resulted in 22 to 25 timers 526 * firing, which seemed independant on whether the suspend was 2 hours or 527 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 528 */ 529 void 530 adjust_timeout_calltodo(struct timeval *time_change) 531 { 532 struct callout *p; 533 unsigned long delta_ticks; 534 535 /* 536 * How many ticks were we asleep? 537 * (stolen from tvtohz()). 538 */ 539 540 /* Don't do anything */ 541 if (time_change->tv_sec < 0) 542 return; 543 else if (time_change->tv_sec <= LONG_MAX / 1000000) 544 delta_ticks = (time_change->tv_sec * 1000000 + 545 time_change->tv_usec + (tick - 1)) / tick + 1; 546 else if (time_change->tv_sec <= LONG_MAX / hz) 547 delta_ticks = time_change->tv_sec * hz + 548 (time_change->tv_usec + (tick - 1)) / tick + 1; 549 else 550 delta_ticks = LONG_MAX; 551 552 if (delta_ticks > INT_MAX) 553 delta_ticks = INT_MAX; 554 555 /* 556 * Now rip through the timer calltodo list looking for timers 557 * to expire. 558 */ 559 560 /* don't collide with softclock() */ 561 crit_enter(); 562 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 563 p->c_time -= delta_ticks; 564 565 /* Break if the timer had more time on it than delta_ticks */ 566 if (p->c_time > 0) 567 break; 568 569 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 570 delta_ticks = -p->c_time; 571 } 572 crit_exit(); 573 574 return; 575 } 576 #endif /* APM_FIXUP_CALLTODO */ 577 #endif 578 579