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 1.19 2005/06/18 12:56:41 eirikn 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 /* 196 * Must occur after ncpus has been initialized. 197 */ 198 SYSINIT(softclock_setup, SI_SUB_CPU, SI_ORDER_SECOND, swi_softclock_setup, NULL); 199 200 /* 201 * This routine is called from the hardclock() (basically a FASTint/IPI) on 202 * each cpu in the system. sc->curticks is this cpu's notion of the timebase. 203 * It IS NOT NECESSARILY SYNCHRONIZED WITH 'ticks'! sc->softticks is where 204 * the callwheel is currently indexed. 205 * 206 * WARNING! The MP lock is not necessarily held on call, nor can it be 207 * safely obtained. 208 * 209 * sc->softticks is adjusted by either this routine or our helper thread 210 * depending on whether the helper thread is running or not. 211 */ 212 void 213 hardclock_softtick(globaldata_t gd) 214 { 215 softclock_pcpu_t sc; 216 217 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 218 ++sc->curticks; 219 if (sc->isrunning) 220 return; 221 if (sc->softticks == sc->curticks) { 222 /* 223 * in sync, only wakeup the thread if there is something to 224 * do. 225 */ 226 if (TAILQ_FIRST(&sc->callwheel[sc->softticks & callwheelmask])) 227 { 228 sc->isrunning = 1; 229 lwkt_schedule(&sc->thread); 230 } else { 231 ++sc->softticks; 232 } 233 } else { 234 /* 235 * out of sync, wakeup the thread unconditionally so it can 236 * catch up. 237 */ 238 sc->isrunning = 1; 239 lwkt_schedule(&sc->thread); 240 } 241 } 242 243 /* 244 * This procedure is the main loop of our per-cpu helper thread. The 245 * sc->isrunning flag prevents us from racing hardclock_softtick() and 246 * a critical section is sufficient to interlock sc->curticks and protect 247 * us from remote IPI's / list removal. 248 * 249 * The thread starts with the MP lock held and not in a critical section. 250 * The loop itself is MP safe while individual callbacks may or may not 251 * be, so we obtain or release the MP lock as appropriate. 252 */ 253 static void 254 softclock_handler(void *arg) 255 { 256 softclock_pcpu_t sc; 257 struct callout *c; 258 struct callout_tailq *bucket; 259 void (*c_func)(void *); 260 void *c_arg; 261 #ifdef SMP 262 int mpsafe = 0; 263 #endif 264 265 sc = arg; 266 crit_enter(); 267 loop: 268 while (sc->softticks != (int)(sc->curticks + 1)) { 269 bucket = &sc->callwheel[sc->softticks & callwheelmask]; 270 271 for (c = TAILQ_FIRST(bucket); c; c = sc->next) { 272 if (c->c_time != sc->softticks) { 273 sc->next = TAILQ_NEXT(c, c_links.tqe); 274 continue; 275 } 276 #ifdef SMP 277 if (c->c_flags & CALLOUT_MPSAFE) { 278 if (mpsafe == 0) { 279 mpsafe = 1; 280 rel_mplock(); 281 } 282 } else { 283 /* 284 * The request might be removed while we 285 * are waiting to get the MP lock. If it 286 * was removed sc->next will point to the 287 * next valid request or NULL, loop up. 288 */ 289 if (mpsafe) { 290 mpsafe = 0; 291 sc->next = c; 292 get_mplock(); 293 if (c != sc->next) 294 continue; 295 } 296 } 297 #endif 298 sc->next = TAILQ_NEXT(c, c_links.tqe); 299 TAILQ_REMOVE(bucket, c, c_links.tqe); 300 301 c_func = c->c_func; 302 c_arg = c->c_arg; 303 c->c_func = NULL; 304 KKASSERT(c->c_flags & CALLOUT_DID_INIT); 305 c->c_flags &= ~CALLOUT_PENDING; 306 crit_exit(); 307 c_func(c_arg); 308 crit_enter(); 309 /* NOTE: list may have changed */ 310 } 311 ++sc->softticks; 312 } 313 sc->isrunning = 0; 314 lwkt_deschedule_self(&sc->thread); /* == curthread */ 315 lwkt_switch(); 316 goto loop; 317 /* NOT REACHED */ 318 } 319 320 #if 0 321 322 /* 323 * timeout -- 324 * Execute a function after a specified length of time. 325 * 326 * untimeout -- 327 * Cancel previous timeout function call. 328 * 329 * callout_handle_init -- 330 * Initialize a handle so that using it with untimeout is benign. 331 * 332 * See AT&T BCI Driver Reference Manual for specification. This 333 * implementation differs from that one in that although an 334 * identification value is returned from timeout, the original 335 * arguments to timeout as well as the identifier are used to 336 * identify entries for untimeout. 337 */ 338 struct callout_handle 339 timeout(timeout_t *ftn, void *arg, int to_ticks) 340 { 341 softclock_pcpu_t sc; 342 struct callout *new; 343 struct callout_handle handle; 344 345 sc = &softclock_pcpu_ary[mycpu->gd_cpuid]; 346 crit_enter(); 347 348 /* Fill in the next free callout structure. */ 349 new = SLIST_FIRST(&sc->callfree); 350 if (new == NULL) { 351 /* XXX Attempt to malloc first */ 352 panic("timeout table full"); 353 } 354 SLIST_REMOVE_HEAD(&sc->callfree, c_links.sle); 355 356 callout_reset(new, to_ticks, ftn, arg); 357 358 handle.callout = new; 359 crit_exit(); 360 return (handle); 361 } 362 363 void 364 untimeout(timeout_t *ftn, void *arg, struct callout_handle handle) 365 { 366 /* 367 * Check for a handle that was initialized 368 * by callout_handle_init, but never used 369 * for a real timeout. 370 */ 371 if (handle.callout == NULL) 372 return; 373 374 crit_enter(); 375 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg) 376 callout_stop(handle.callout); 377 crit_exit(); 378 } 379 380 void 381 callout_handle_init(struct callout_handle *handle) 382 { 383 handle->callout = NULL; 384 } 385 386 #endif 387 388 /* 389 * New interface; clients allocate their own callout structures. 390 * 391 * callout_reset() - establish or change a timeout 392 * callout_stop() - disestablish a timeout 393 * callout_init() - initialize a callout structure so that it can 394 * safely be passed to callout_reset() and callout_stop() 395 * callout_init_mp() - same but any installed functions must be MP safe. 396 * 397 * <sys/callout.h> defines three convenience macros: 398 * 399 * callout_active() - returns truth if callout has not been serviced 400 * callout_pending() - returns truth if callout is still waiting for timeout 401 * callout_deactivate() - marks the callout as having been serviced 402 */ 403 404 /* 405 * Start or restart a timeout. Install the callout structure in the 406 * callwheel. Callers may legally pass any value, even if 0 or negative, 407 * but since the sc->curticks index may have already been processed a 408 * minimum timeout of 1 tick will be enforced. 409 * 410 * The callout is installed on and will be processed on the current cpu's 411 * callout wheel. 412 */ 413 void 414 callout_reset(struct callout *c, int to_ticks, void (*ftn)(void *), 415 void *arg) 416 { 417 softclock_pcpu_t sc; 418 globaldata_t gd; 419 420 #ifdef INVARIANTS 421 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 422 callout_init(c); 423 printf( 424 "callout_reset(%p) from %p: callout was not initialized\n", 425 c, ((int **)&c)[-1]); 426 #ifdef DDB 427 db_print_backtrace(); 428 #endif 429 } 430 #endif 431 gd = mycpu; 432 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 433 crit_enter_gd(gd); 434 435 if (c->c_flags & CALLOUT_PENDING) 436 callout_stop(c); 437 438 if (to_ticks <= 0) 439 to_ticks = 1; 440 441 c->c_arg = arg; 442 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING); 443 c->c_func = ftn; 444 c->c_time = sc->curticks + to_ticks; 445 #ifdef SMP 446 c->c_gd = gd; 447 #endif 448 449 TAILQ_INSERT_TAIL(&sc->callwheel[c->c_time & callwheelmask], 450 c, c_links.tqe); 451 crit_exit_gd(gd); 452 } 453 454 /* 455 * Stop a running timer. WARNING! If called on a cpu other then the one 456 * the callout was started on this function will liveloop on its IPI to 457 * the target cpu to process the request. It is possible for the callout 458 * to execute in that case. 459 * 460 * WARNING! This routine may be called from an IPI 461 */ 462 int 463 callout_stop(struct callout *c) 464 { 465 globaldata_t gd = mycpu; 466 #ifdef SMP 467 globaldata_t tgd; 468 #endif 469 softclock_pcpu_t sc; 470 471 #ifdef INVARIANTS 472 if ((c->c_flags & CALLOUT_DID_INIT) == 0) { 473 callout_init(c); 474 printf( 475 "callout_stop(%p) from %p: callout was not initialized\n", 476 c, ((int **)&c)[-1]); 477 #ifdef DDB 478 db_print_backtrace(); 479 #endif 480 } 481 #endif 482 crit_enter_gd(gd); 483 484 /* 485 * Don't attempt to delete a callout that's not on the queue. 486 */ 487 if ((c->c_flags & CALLOUT_PENDING) == 0) { 488 c->c_flags &= ~CALLOUT_ACTIVE; 489 crit_exit_gd(gd); 490 return (0); 491 } 492 #ifdef SMP 493 if ((tgd = c->c_gd) != gd) { 494 /* 495 * If the callout is owned by a different CPU we have to 496 * execute the function synchronously on the target cpu. 497 */ 498 int seq; 499 500 cpu_ccfence(); /* don't let tgd alias c_gd */ 501 seq = lwkt_send_ipiq(tgd, (void *)callout_stop, c); 502 lwkt_wait_ipiq(tgd, seq); 503 } else 504 #endif 505 { 506 /* 507 * If the callout is owned by the same CPU we can 508 * process it directly, but if we are racing our helper 509 * thread (sc->next), we have to adjust sc->next. The 510 * race is interlocked by a critical section. 511 */ 512 sc = &softclock_pcpu_ary[gd->gd_cpuid]; 513 514 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING); 515 if (sc->next == c) 516 sc->next = TAILQ_NEXT(c, c_links.tqe); 517 518 TAILQ_REMOVE(&sc->callwheel[c->c_time & callwheelmask], 519 c, c_links.tqe); 520 c->c_func = NULL; 521 } 522 crit_exit_gd(gd); 523 return (1); 524 } 525 526 /* 527 * Prepare a callout structure for use by callout_reset() and/or 528 * callout_stop(). The MP version of this routine requires that the callback 529 * function installed by callout_reset() by MP safe. 530 */ 531 void 532 callout_init(struct callout *c) 533 { 534 bzero(c, sizeof *c); 535 c->c_flags = CALLOUT_DID_INIT; 536 } 537 538 void 539 callout_init_mp(struct callout *c) 540 { 541 callout_init(c); 542 c->c_flags |= CALLOUT_MPSAFE; 543 } 544 545 /* What, are you joking? This is nuts! -Matt */ 546 #if 0 547 #ifdef APM_FIXUP_CALLTODO 548 /* 549 * Adjust the kernel calltodo timeout list. This routine is used after 550 * an APM resume to recalculate the calltodo timer list values with the 551 * number of hz's we have been sleeping. The next hardclock() will detect 552 * that there are fired timers and run softclock() to execute them. 553 * 554 * Please note, I have not done an exhaustive analysis of what code this 555 * might break. I am motivated to have my select()'s and alarm()'s that 556 * have expired during suspend firing upon resume so that the applications 557 * which set the timer can do the maintanence the timer was for as close 558 * as possible to the originally intended time. Testing this code for a 559 * week showed that resuming from a suspend resulted in 22 to 25 timers 560 * firing, which seemed independant on whether the suspend was 2 hours or 561 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu> 562 */ 563 void 564 adjust_timeout_calltodo(struct timeval *time_change) 565 { 566 struct callout *p; 567 unsigned long delta_ticks; 568 569 /* 570 * How many ticks were we asleep? 571 * (stolen from tvtohz()). 572 */ 573 574 /* Don't do anything */ 575 if (time_change->tv_sec < 0) 576 return; 577 else if (time_change->tv_sec <= LONG_MAX / 1000000) 578 delta_ticks = (time_change->tv_sec * 1000000 + 579 time_change->tv_usec + (tick - 1)) / tick + 1; 580 else if (time_change->tv_sec <= LONG_MAX / hz) 581 delta_ticks = time_change->tv_sec * hz + 582 (time_change->tv_usec + (tick - 1)) / tick + 1; 583 else 584 delta_ticks = LONG_MAX; 585 586 if (delta_ticks > INT_MAX) 587 delta_ticks = INT_MAX; 588 589 /* 590 * Now rip through the timer calltodo list looking for timers 591 * to expire. 592 */ 593 594 /* don't collide with softclock() */ 595 crit_enter(); 596 for (p = calltodo.c_next; p != NULL; p = p->c_next) { 597 p->c_time -= delta_ticks; 598 599 /* Break if the timer had more time on it than delta_ticks */ 600 if (p->c_time > 0) 601 break; 602 603 /* take back the ticks the timer didn't use (p->c_time <= 0) */ 604 delta_ticks = -p->c_time; 605 } 606 crit_exit(); 607 608 return; 609 } 610 #endif /* APM_FIXUP_CALLTODO */ 611 #endif 612 613