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 1.14 2004/09/19 02:52:26 dillon 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 <machine/ipl.h> 112 #include <ddb/ddb.h> 113 114 #ifndef MAX_SOFTCLOCK_STEPS 115 #define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */ 116 #endif 117 118 119 struct softclock_pcpu { 120 struct callout_tailq *callwheel; 121 struct callout * volatile next; 122 int softticks; /* softticks index */ 123 int curticks; /* per-cpu ticks counter */ 124 int isrunning; 125 struct thread thread; 126 127 }; 128 129 typedef struct softclock_pcpu *softclock_pcpu_t; 130 131 /* 132 * TODO: 133 * allocate more timeout table slots when table overflows. 134 */ 135 static MALLOC_DEFINE(M_CALLOUT, "callout", "callout structures"); 136 static int callwheelsize; 137 static int callwheelbits; 138 static int callwheelmask; 139 static struct softclock_pcpu softclock_pcpu_ary[MAXCPU]; 140 141 static void softclock_handler(void *arg); 142 143 static void 144 swi_softclock_setup(void *arg) 145 { 146 int cpu; 147 int i; 148 149 /* 150 * Figure out how large a callwheel we need. It must be a power of 2. 151 */ 152 callwheelsize = 1; 153 callwheelbits = 0; 154 while (callwheelsize < ncallout) { 155 callwheelsize <<= 1; 156 ++callwheelbits; 157 } 158 callwheelmask = callwheelsize - 1; 159 160 /* 161 * Initialize per-cpu data structures. 162 */ 163 for (cpu = 0; cpu < ncpus; ++cpu) { 164 softclock_pcpu_t sc; 165 166 sc = &softclock_pcpu_ary[cpu]; 167 168 sc->callwheel = malloc(sizeof(*sc->callwheel) * callwheelsize, 169 M_CALLOUT, M_WAITOK|M_ZERO); 170 for (i = 0; i < callwheelsize; ++i) 171 TAILQ_INIT(&sc->callwheel[i]); 172 173 /* 174 * Create a preemption-capable thread for each cpu to handle 175 * softclock timeouts on that cpu. The preemption can only 176 * be blocked by a critical section. The thread can itself 177 * be preempted by normal interrupts. 178 */ 179 lwkt_create(softclock_handler, sc, NULL, 180 &sc->thread, TDF_STOPREQ|TDF_INTTHREAD, -1, 181 "softclock %d", cpu); 182 lwkt_setpri(&sc->thread, TDPRI_SOFT_NORM); 183 #if 0 184 /* 185 * Do not make the thread preemptable until we clean up all 186 * the splsoftclock() calls in the system. Since the threads 187 * are no longer operated as a software interrupt, the 188 * splsoftclock() calls will not have any effect on them. 189 */ 190 sc->thread.td_preemptable = lwkt_preempt; 191 #endif 192 } 193 } 194 195 SYSINIT(softclock_setup, SI_SUB_CPU, SI_ORDER_ANY, swi_softclock_setup, NULL); 196 197 /* 198 * This routine is called from the hardclock() (basically a FASTint/IPI) on 199 * each cpu in the system. sc->curticks is this cpu's notion of the timebase. 200 * It IS NOT NECESSARILY SYNCHRONIZED WITH 'ticks'! sc->softticks is where 201 * the callwheel is currently indexed. 202 * 203 * WARNING! The MP lock is not necessarily held on call, nor can it be 204 * safely obtained. 205 * 206 * sc->softticks is adjusted by either this routine or our helper thread 207 * depending on whether the helper thread is running or not. 208 */ 209 void 210 hardclock_softtick(globaldata_t gd) 211 { 212 softclock_pcpu_t sc; 213 214 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 215 ++sc->curticks; 216 if (sc->isrunning) 217 return; 218 if (sc->softticks == sc->curticks) { 219 /* 220 * in sync, only wakeup the thread if there is something to 221 * do. 222 */ 223 if (TAILQ_FIRST(&sc->callwheel[sc->softticks & callwheelmask])) 224 { 225 sc->isrunning = 1; 226 lwkt_schedule(&sc->thread); 227 } else { 228 ++sc->softticks; 229 } 230 } else { 231 /* 232 * out of sync, wakeup the thread unconditionally so it can 233 * catch up. 234 */ 235 sc->isrunning = 1; 236 lwkt_schedule(&sc->thread); 237 } 238 } 239 240 /* 241 * This procedure is the main loop of our per-cpu helper thread. The 242 * sc->isrunning flag prevents us from racing hardclock_softtick() and 243 * a critical section is sufficient to interlock sc->curticks and protect 244 * us from remote IPI's / list removal. 245 * 246 * The thread starts with the MP lock held and not in a critical section. 247 * The loop itself is MP safe while individual callbacks may or may not 248 * be, so we obtain or release the MP lock as appropriate. 249 */ 250 static void 251 softclock_handler(void *arg) 252 { 253 softclock_pcpu_t sc; 254 struct callout *c; 255 struct callout_tailq *bucket; 256 void (*c_func)(void *); 257 void *c_arg; 258 #ifdef SMP 259 int mpsafe = 0; 260 #endif 261 262 sc = arg; 263 crit_enter(); 264 loop: 265 while (sc->softticks != (int)(sc->curticks + 1)) { 266 bucket = &sc->callwheel[sc->softticks & callwheelmask]; 267 268 for (c = TAILQ_FIRST(bucket); c; c = sc->next) { 269 if (c->c_time != sc->softticks) { 270 sc->next = TAILQ_NEXT(c, c_links.tqe); 271 continue; 272 } 273 #ifdef SMP 274 if (c->c_flags & CALLOUT_MPSAFE) { 275 if (mpsafe == 0) { 276 mpsafe = 1; 277 rel_mplock(); 278 } 279 } else { 280 /* 281 * The request might be removed while we 282 * are waiting to get the MP lock. If it 283 * was removed sc->next will point to the 284 * next valid request or NULL, loop up. 285 */ 286 if (mpsafe) { 287 mpsafe = 0; 288 sc->next = c; 289 get_mplock(); 290 if (c != sc->next) 291 continue; 292 } 293 } 294 #endif 295 sc->next = TAILQ_NEXT(c, c_links.tqe); 296 TAILQ_REMOVE(bucket, c, c_links.tqe); 297 298 c_func = c->c_func; 299 c_arg = c->c_arg; 300 c->c_func = NULL; 301 KKASSERT(c->c_flags & CALLOUT_DID_INIT); 302 c->c_flags &= ~CALLOUT_PENDING; 303 crit_exit(); 304 c_func(c_arg); 305 crit_enter(); 306 /* NOTE: list may have changed */ 307 } 308 ++sc->softticks; 309 } 310 sc->isrunning = 0; 311 lwkt_deschedule_self(&sc->thread); /* == curthread */ 312 lwkt_switch(); 313 goto loop; 314 /* NOT REACHED */ 315 } 316 317 #if 0 318 319 /* 320 * timeout -- 321 * Execute a function after a specified length of time. 322 * 323 * untimeout -- 324 * Cancel previous timeout function call. 325 * 326 * callout_handle_init -- 327 * Initialize a handle so that using it with untimeout is benign. 328 * 329 * See AT&T BCI Driver Reference Manual for specification. This 330 * implementation differs from that one in that although an 331 * identification value is returned from timeout, the original 332 * arguments to timeout as well as the identifier are used to 333 * identify entries for untimeout. 334 */ 335 struct callout_handle 336 timeout(timeout_t *ftn, void *arg, int to_ticks) 337 { 338 softclock_pcpu_t sc; 339 struct callout *new; 340 struct callout_handle handle; 341 342 sc = &softclock_pcpu_ary[mycpu->gd_cpuid]; 343 crit_enter(); 344 345 /* Fill in the next free callout structure. */ 346 new = SLIST_FIRST(&sc->callfree); 347 if (new == NULL) { 348 /* XXX Attempt to malloc first */ 349 panic("timeout table full"); 350 } 351 SLIST_REMOVE_HEAD(&sc->callfree, c_links.sle); 352 353 callout_reset(new, to_ticks, ftn, arg); 354 355 handle.callout = new; 356 crit_exit(); 357 return (handle); 358 } 359 360 void 361 untimeout(timeout_t *ftn, void *arg, struct callout_handle handle) 362 { 363 /* 364 * Check for a handle that was initialized 365 * by callout_handle_init, but never used 366 * for a real timeout. 367 */ 368 if (handle.callout == NULL) 369 return; 370 371 crit_enter(); 372 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) 373 callout_stop(handle.callout); 374 crit_exit(); 375 } 376 377 void 378 callout_handle_init(struct callout_handle *handle) 379 { 380 handle->callout = NULL; 381 } 382 383 #endif 384 385 /* 386 * New interface; clients allocate their own callout structures. 387 * 388 * callout_reset() - establish or change a timeout 389 * callout_stop() - disestablish a timeout 390 * callout_init() - initialize a callout structure so that it can 391 * safely be passed to callout_reset() and callout_stop() 392 * callout_init_mp() - same but any installed functions must be MP safe. 393 * 394 * <sys/callout.h> defines three convenience macros: 395 * 396 * callout_active() - returns truth if callout has not been serviced 397 * callout_pending() - returns truth if callout is still waiting for timeout 398 * callout_deactivate() - marks the callout as having been serviced 399 */ 400 401 /* 402 * Start or restart a timeout. Install the callout structure in the 403 * callwheel. Callers may legally pass any value, even if 0 or negative, 404 * but since the sc->curticks index may have already been processed a 405 * minimum timeout of 1 tick will be enforced. 406 * 407 * The callout is installed on and will be processed on the current cpu's 408 * callout wheel. 409 */ 410 void 411 callout_reset(struct callout *c, int to_ticks, void (*ftn)(void *), 412 void *arg) 413 { 414 softclock_pcpu_t sc; 415 globaldata_t gd; 416 417 #ifdef INVARIANTS 418 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 419 callout_init(c); 420 printf( 421 "callout_reset(%p) from %p: callout was not initialized\n", 422 c, ((int **)&c)[-1]); 423 #ifdef DDB 424 db_print_backtrace(); 425 #endif 426 } 427 #endif 428 gd = mycpu; 429 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 430 crit_enter_gd(gd); 431 432 if (c->c_flags & CALLOUT_PENDING) 433 callout_stop(c); 434 435 if (to_ticks <= 0) 436 to_ticks = 1; 437 438 c->c_arg = arg; 439 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 440 c->c_func = ftn; 441 c->c_time = sc->curticks + to_ticks; 442 #ifdef SMP 443 c->c_gd = gd; 444 #endif 445 446 TAILQ_INSERT_TAIL(&sc->callwheel[c->c_time & callwheelmask], 447 c, c_links.tqe); 448 crit_exit_gd(gd); 449 } 450 451 /* 452 * Stop a running timer. WARNING! If called on a cpu other then the one 453 * the callout was started on this function will liveloop on its IPI to 454 * the target cpu to process the request. It is possible for the callout 455 * to execute in that case. 456 * 457 * WARNING! This routine may be called from an IPI 458 */ 459 int 460 callout_stop(struct callout *c) 461 { 462 globaldata_t gd = mycpu; 463 #ifdef SMP 464 globaldata_t tgd; 465 #endif 466 softclock_pcpu_t sc; 467 468 #ifdef INVARIANTS 469 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 470 callout_init(c); 471 printf( 472 "callout_reset(%p) from %p: callout was not initialized\n", 473 c, ((int **)&c)[-1]); 474 #ifdef DDB 475 db_print_backtrace(); 476 #endif 477 } 478 #endif 479 crit_enter_gd(gd); 480 481 /* 482 * Don't attempt to delete a callout that's not on the queue. 483 */ 484 if ((c->c_flags & CALLOUT_PENDING) == 0) { 485 c->c_flags &= ~CALLOUT_ACTIVE; 486 crit_exit_gd(gd); 487 return (0); 488 } 489 #ifdef SMP 490 if ((tgd = c->c_gd) != gd) { 491 /* 492 * If the callout is owned by a different CPU we have to 493 * execute the function synchronously on the target cpu. 494 */ 495 int seq; 496 497 cpu_mb1(); /* don't let tgd alias c_gd */ 498 seq = lwkt_send_ipiq(tgd, (void *)callout_stop, c); 499 lwkt_wait_ipiq(tgd, seq); 500 } else 501 #endif 502 { 503 /* 504 * If the callout is owned by the same CPU we can 505 * process it directly, but if we are racing our helper 506 * thread (sc->next), we have to adjust sc->next. The 507 * race is interlocked by a critical section. 508 */ 509 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 510 511 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 512 if (sc->next == c) 513 sc->next = TAILQ_NEXT(c, c_links.tqe); 514 515 TAILQ_REMOVE(&sc->callwheel[c->c_time & callwheelmask], 516 c, c_links.tqe); 517 c->c_func = NULL; 518 } 519 crit_exit_gd(gd); 520 return (1); 521 } 522 523 /* 524 * Prepare a callout structure for use by callout_reset() and/or 525 * callout_stop(). The MP version of this routine requires that the callback 526 * function installed by callout_reset() by MP safe. 527 */ 528 void 529 callout_init(struct callout *c) 530 { 531 bzero(c, sizeof *c); 532 c->c_flags = CALLOUT_DID_INIT; 533 } 534 535 void 536 callout_init_mp(struct callout *c) 537 { 538 callout_init(c); 539 c->c_flags |= CALLOUT_MPSAFE; 540 } 541 542 /* What, are you joking? This is nuts! -Matt */ 543 #if 0 544 #ifdef APM_FIXUP_CALLTODO 545 /* 546 * Adjust the kernel calltodo timeout list. This routine is used after 547 * an APM resume to recalculate the calltodo timer list values with the 548 * number of hz's we have been sleeping. The next hardclock() will detect 549 * that there are fired timers and run softclock() to execute them. 550 * 551 * Please note, I have not done an exhaustive analysis of what code this 552 * might break. I am motivated to have my select()'s and alarm()'s that 553 * have expired during suspend firing upon resume so that the applications 554 * which set the timer can do the maintanence the timer was for as close 555 * as possible to the originally intended time. Testing this code for a 556 * week showed that resuming from a suspend resulted in 22 to 25 timers 557 * firing, which seemed independant on whether the suspend was 2 hours or 558 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 559 */ 560 void 561 adjust_timeout_calltodo(struct timeval *time_change) 562 { 563 struct callout *p; 564 unsigned long delta_ticks; 565 int s; 566 567 /* 568 * How many ticks were we asleep? 569 * (stolen from tvtohz()). 570 */ 571 572 /* Don't do anything */ 573 if (time_change->tv_sec < 0) 574 return; 575 else if (time_change->tv_sec <= LONG_MAX / 1000000) 576 delta_ticks = (time_change->tv_sec * 1000000 + 577 time_change->tv_usec + (tick - 1)) / tick + 1; 578 else if (time_change->tv_sec <= LONG_MAX / hz) 579 delta_ticks = time_change->tv_sec * hz + 580 (time_change->tv_usec + (tick - 1)) / tick + 1; 581 else 582 delta_ticks = LONG_MAX; 583 584 if (delta_ticks > INT_MAX) 585 delta_ticks = INT_MAX; 586 587 /* 588 * Now rip through the timer calltodo list looking for timers 589 * to expire. 590 */ 591 592 /* don't collide with softclock() */ 593 s = splhigh(); 594 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 595 p->c_time -= delta_ticks; 596 597 /* Break if the timer had more time on it than delta_ticks */ 598 if (p->c_time > 0) 599 break; 600 601 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 602 delta_ticks = -p->c_time; 603 } 604 splx(s); 605 606 return; 607 } 608 #endif /* APM_FIXUP_CALLTODO */ 609 #endif 610 611