1 /* 2 * linux/kernel/timer.c 3 * 4 * Kernel internal timers 5 * 6 * Copyright (C) 1991, 1992 Linus Torvalds 7 * 8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better. 9 * 10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 11 * "A Kernel Model for Precision Timekeeping" by Dave Mills 12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to 13 * serialize accesses to xtime/lost_ticks). 14 * Copyright (C) 1998 Andrea Arcangeli 15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl 16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love 17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling. 18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar 19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar 20 */ 21 22 #include <linux/kernel_stat.h> 23 #include <linux/export.h> 24 #include <linux/interrupt.h> 25 #include <linux/percpu.h> 26 #include <linux/init.h> 27 #include <linux/mm.h> 28 #include <linux/swap.h> 29 #include <linux/pid_namespace.h> 30 #include <linux/notifier.h> 31 #include <linux/thread_info.h> 32 #include <linux/time.h> 33 #include <linux/jiffies.h> 34 #include <linux/posix-timers.h> 35 #include <linux/cpu.h> 36 #include <linux/syscalls.h> 37 #include <linux/delay.h> 38 #include <linux/tick.h> 39 #include <linux/kallsyms.h> 40 #include <linux/irq_work.h> 41 #include <linux/sched/signal.h> 42 #include <linux/sched/sysctl.h> 43 #include <linux/sched/nohz.h> 44 #include <linux/sched/debug.h> 45 #include <linux/slab.h> 46 #include <linux/compat.h> 47 48 #include <linux/uaccess.h> 49 #include <asm/unistd.h> 50 #include <asm/div64.h> 51 #include <asm/timex.h> 52 #include <asm/io.h> 53 54 #include "tick-internal.h" 55 56 #define CREATE_TRACE_POINTS 57 #include <trace/events/timer.h> 58 59 __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; 60 61 EXPORT_SYMBOL(jiffies_64); 62 63 /* 64 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of 65 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each 66 * level has a different granularity. 67 * 68 * The level granularity is: LVL_CLK_DIV ^ lvl 69 * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level) 70 * 71 * The array level of a newly armed timer depends on the relative expiry 72 * time. The farther the expiry time is away the higher the array level and 73 * therefor the granularity becomes. 74 * 75 * Contrary to the original timer wheel implementation, which aims for 'exact' 76 * expiry of the timers, this implementation removes the need for recascading 77 * the timers into the lower array levels. The previous 'classic' timer wheel 78 * implementation of the kernel already violated the 'exact' expiry by adding 79 * slack to the expiry time to provide batched expiration. The granularity 80 * levels provide implicit batching. 81 * 82 * This is an optimization of the original timer wheel implementation for the 83 * majority of the timer wheel use cases: timeouts. The vast majority of 84 * timeout timers (networking, disk I/O ...) are canceled before expiry. If 85 * the timeout expires it indicates that normal operation is disturbed, so it 86 * does not matter much whether the timeout comes with a slight delay. 87 * 88 * The only exception to this are networking timers with a small expiry 89 * time. They rely on the granularity. Those fit into the first wheel level, 90 * which has HZ granularity. 91 * 92 * We don't have cascading anymore. timers with a expiry time above the 93 * capacity of the last wheel level are force expired at the maximum timeout 94 * value of the last wheel level. From data sampling we know that the maximum 95 * value observed is 5 days (network connection tracking), so this should not 96 * be an issue. 97 * 98 * The currently chosen array constants values are a good compromise between 99 * array size and granularity. 100 * 101 * This results in the following granularity and range levels: 102 * 103 * HZ 1000 steps 104 * Level Offset Granularity Range 105 * 0 0 1 ms 0 ms - 63 ms 106 * 1 64 8 ms 64 ms - 511 ms 107 * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s) 108 * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s) 109 * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m) 110 * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m) 111 * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h) 112 * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d) 113 * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d) 114 * 115 * HZ 300 116 * Level Offset Granularity Range 117 * 0 0 3 ms 0 ms - 210 ms 118 * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s) 119 * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s) 120 * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m) 121 * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m) 122 * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h) 123 * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h) 124 * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d) 125 * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d) 126 * 127 * HZ 250 128 * Level Offset Granularity Range 129 * 0 0 4 ms 0 ms - 255 ms 130 * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s) 131 * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s) 132 * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m) 133 * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m) 134 * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h) 135 * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h) 136 * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d) 137 * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d) 138 * 139 * HZ 100 140 * Level Offset Granularity Range 141 * 0 0 10 ms 0 ms - 630 ms 142 * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s) 143 * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s) 144 * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m) 145 * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m) 146 * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h) 147 * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d) 148 * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d) 149 */ 150 151 /* Clock divisor for the next level */ 152 #define LVL_CLK_SHIFT 3 153 #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT) 154 #define LVL_CLK_MASK (LVL_CLK_DIV - 1) 155 #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT) 156 #define LVL_GRAN(n) (1UL << LVL_SHIFT(n)) 157 158 /* 159 * The time start value for each level to select the bucket at enqueue 160 * time. 161 */ 162 #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT)) 163 164 /* Size of each clock level */ 165 #define LVL_BITS 6 166 #define LVL_SIZE (1UL << LVL_BITS) 167 #define LVL_MASK (LVL_SIZE - 1) 168 #define LVL_OFFS(n) ((n) * LVL_SIZE) 169 170 /* Level depth */ 171 #if HZ > 100 172 # define LVL_DEPTH 9 173 # else 174 # define LVL_DEPTH 8 175 #endif 176 177 /* The cutoff (max. capacity of the wheel) */ 178 #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH)) 179 #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1)) 180 181 /* 182 * The resulting wheel size. If NOHZ is configured we allocate two 183 * wheels so we have a separate storage for the deferrable timers. 184 */ 185 #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH) 186 187 #ifdef CONFIG_NO_HZ_COMMON 188 # define NR_BASES 2 189 # define BASE_STD 0 190 # define BASE_DEF 1 191 #else 192 # define NR_BASES 1 193 # define BASE_STD 0 194 # define BASE_DEF 0 195 #endif 196 197 struct timer_base { 198 raw_spinlock_t lock; 199 struct timer_list *running_timer; 200 unsigned long clk; 201 unsigned long next_expiry; 202 unsigned int cpu; 203 bool migration_enabled; 204 bool nohz_active; 205 bool is_idle; 206 DECLARE_BITMAP(pending_map, WHEEL_SIZE); 207 struct hlist_head vectors[WHEEL_SIZE]; 208 } ____cacheline_aligned; 209 210 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]); 211 212 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) 213 unsigned int sysctl_timer_migration = 1; 214 215 void timers_update_migration(bool update_nohz) 216 { 217 bool on = sysctl_timer_migration && tick_nohz_active; 218 unsigned int cpu; 219 220 /* Avoid the loop, if nothing to update */ 221 if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on) 222 return; 223 224 for_each_possible_cpu(cpu) { 225 per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on; 226 per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on; 227 per_cpu(hrtimer_bases.migration_enabled, cpu) = on; 228 if (!update_nohz) 229 continue; 230 per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true; 231 per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true; 232 per_cpu(hrtimer_bases.nohz_active, cpu) = true; 233 } 234 } 235 236 int timer_migration_handler(struct ctl_table *table, int write, 237 void __user *buffer, size_t *lenp, 238 loff_t *ppos) 239 { 240 static DEFINE_MUTEX(mutex); 241 int ret; 242 243 mutex_lock(&mutex); 244 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 245 if (!ret && write) 246 timers_update_migration(false); 247 mutex_unlock(&mutex); 248 return ret; 249 } 250 #endif 251 252 static unsigned long round_jiffies_common(unsigned long j, int cpu, 253 bool force_up) 254 { 255 int rem; 256 unsigned long original = j; 257 258 /* 259 * We don't want all cpus firing their timers at once hitting the 260 * same lock or cachelines, so we skew each extra cpu with an extra 261 * 3 jiffies. This 3 jiffies came originally from the mm/ code which 262 * already did this. 263 * The skew is done by adding 3*cpunr, then round, then subtract this 264 * extra offset again. 265 */ 266 j += cpu * 3; 267 268 rem = j % HZ; 269 270 /* 271 * If the target jiffie is just after a whole second (which can happen 272 * due to delays of the timer irq, long irq off times etc etc) then 273 * we should round down to the whole second, not up. Use 1/4th second 274 * as cutoff for this rounding as an extreme upper bound for this. 275 * But never round down if @force_up is set. 276 */ 277 if (rem < HZ/4 && !force_up) /* round down */ 278 j = j - rem; 279 else /* round up */ 280 j = j - rem + HZ; 281 282 /* now that we have rounded, subtract the extra skew again */ 283 j -= cpu * 3; 284 285 /* 286 * Make sure j is still in the future. Otherwise return the 287 * unmodified value. 288 */ 289 return time_is_after_jiffies(j) ? j : original; 290 } 291 292 /** 293 * __round_jiffies - function to round jiffies to a full second 294 * @j: the time in (absolute) jiffies that should be rounded 295 * @cpu: the processor number on which the timeout will happen 296 * 297 * __round_jiffies() rounds an absolute time in the future (in jiffies) 298 * up or down to (approximately) full seconds. This is useful for timers 299 * for which the exact time they fire does not matter too much, as long as 300 * they fire approximately every X seconds. 301 * 302 * By rounding these timers to whole seconds, all such timers will fire 303 * at the same time, rather than at various times spread out. The goal 304 * of this is to have the CPU wake up less, which saves power. 305 * 306 * The exact rounding is skewed for each processor to avoid all 307 * processors firing at the exact same time, which could lead 308 * to lock contention or spurious cache line bouncing. 309 * 310 * The return value is the rounded version of the @j parameter. 311 */ 312 unsigned long __round_jiffies(unsigned long j, int cpu) 313 { 314 return round_jiffies_common(j, cpu, false); 315 } 316 EXPORT_SYMBOL_GPL(__round_jiffies); 317 318 /** 319 * __round_jiffies_relative - function to round jiffies to a full second 320 * @j: the time in (relative) jiffies that should be rounded 321 * @cpu: the processor number on which the timeout will happen 322 * 323 * __round_jiffies_relative() rounds a time delta in the future (in jiffies) 324 * up or down to (approximately) full seconds. This is useful for timers 325 * for which the exact time they fire does not matter too much, as long as 326 * they fire approximately every X seconds. 327 * 328 * By rounding these timers to whole seconds, all such timers will fire 329 * at the same time, rather than at various times spread out. The goal 330 * of this is to have the CPU wake up less, which saves power. 331 * 332 * The exact rounding is skewed for each processor to avoid all 333 * processors firing at the exact same time, which could lead 334 * to lock contention or spurious cache line bouncing. 335 * 336 * The return value is the rounded version of the @j parameter. 337 */ 338 unsigned long __round_jiffies_relative(unsigned long j, int cpu) 339 { 340 unsigned long j0 = jiffies; 341 342 /* Use j0 because jiffies might change while we run */ 343 return round_jiffies_common(j + j0, cpu, false) - j0; 344 } 345 EXPORT_SYMBOL_GPL(__round_jiffies_relative); 346 347 /** 348 * round_jiffies - function to round jiffies to a full second 349 * @j: the time in (absolute) jiffies that should be rounded 350 * 351 * round_jiffies() rounds an absolute time in the future (in jiffies) 352 * up or down to (approximately) full seconds. This is useful for timers 353 * for which the exact time they fire does not matter too much, as long as 354 * they fire approximately every X seconds. 355 * 356 * By rounding these timers to whole seconds, all such timers will fire 357 * at the same time, rather than at various times spread out. The goal 358 * of this is to have the CPU wake up less, which saves power. 359 * 360 * The return value is the rounded version of the @j parameter. 361 */ 362 unsigned long round_jiffies(unsigned long j) 363 { 364 return round_jiffies_common(j, raw_smp_processor_id(), false); 365 } 366 EXPORT_SYMBOL_GPL(round_jiffies); 367 368 /** 369 * round_jiffies_relative - function to round jiffies to a full second 370 * @j: the time in (relative) jiffies that should be rounded 371 * 372 * round_jiffies_relative() rounds a time delta in the future (in jiffies) 373 * up or down to (approximately) full seconds. This is useful for timers 374 * for which the exact time they fire does not matter too much, as long as 375 * they fire approximately every X seconds. 376 * 377 * By rounding these timers to whole seconds, all such timers will fire 378 * at the same time, rather than at various times spread out. The goal 379 * of this is to have the CPU wake up less, which saves power. 380 * 381 * The return value is the rounded version of the @j parameter. 382 */ 383 unsigned long round_jiffies_relative(unsigned long j) 384 { 385 return __round_jiffies_relative(j, raw_smp_processor_id()); 386 } 387 EXPORT_SYMBOL_GPL(round_jiffies_relative); 388 389 /** 390 * __round_jiffies_up - function to round jiffies up to a full second 391 * @j: the time in (absolute) jiffies that should be rounded 392 * @cpu: the processor number on which the timeout will happen 393 * 394 * This is the same as __round_jiffies() except that it will never 395 * round down. This is useful for timeouts for which the exact time 396 * of firing does not matter too much, as long as they don't fire too 397 * early. 398 */ 399 unsigned long __round_jiffies_up(unsigned long j, int cpu) 400 { 401 return round_jiffies_common(j, cpu, true); 402 } 403 EXPORT_SYMBOL_GPL(__round_jiffies_up); 404 405 /** 406 * __round_jiffies_up_relative - function to round jiffies up to a full second 407 * @j: the time in (relative) jiffies that should be rounded 408 * @cpu: the processor number on which the timeout will happen 409 * 410 * This is the same as __round_jiffies_relative() except that it will never 411 * round down. This is useful for timeouts for which the exact time 412 * of firing does not matter too much, as long as they don't fire too 413 * early. 414 */ 415 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) 416 { 417 unsigned long j0 = jiffies; 418 419 /* Use j0 because jiffies might change while we run */ 420 return round_jiffies_common(j + j0, cpu, true) - j0; 421 } 422 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); 423 424 /** 425 * round_jiffies_up - function to round jiffies up to a full second 426 * @j: the time in (absolute) jiffies that should be rounded 427 * 428 * This is the same as round_jiffies() except that it will never 429 * round down. This is useful for timeouts for which the exact time 430 * of firing does not matter too much, as long as they don't fire too 431 * early. 432 */ 433 unsigned long round_jiffies_up(unsigned long j) 434 { 435 return round_jiffies_common(j, raw_smp_processor_id(), true); 436 } 437 EXPORT_SYMBOL_GPL(round_jiffies_up); 438 439 /** 440 * round_jiffies_up_relative - function to round jiffies up to a full second 441 * @j: the time in (relative) jiffies that should be rounded 442 * 443 * This is the same as round_jiffies_relative() except that it will never 444 * round down. This is useful for timeouts for which the exact time 445 * of firing does not matter too much, as long as they don't fire too 446 * early. 447 */ 448 unsigned long round_jiffies_up_relative(unsigned long j) 449 { 450 return __round_jiffies_up_relative(j, raw_smp_processor_id()); 451 } 452 EXPORT_SYMBOL_GPL(round_jiffies_up_relative); 453 454 455 static inline unsigned int timer_get_idx(struct timer_list *timer) 456 { 457 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT; 458 } 459 460 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx) 461 { 462 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) | 463 idx << TIMER_ARRAYSHIFT; 464 } 465 466 /* 467 * Helper function to calculate the array index for a given expiry 468 * time. 469 */ 470 static inline unsigned calc_index(unsigned expires, unsigned lvl) 471 { 472 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl); 473 return LVL_OFFS(lvl) + (expires & LVL_MASK); 474 } 475 476 static int calc_wheel_index(unsigned long expires, unsigned long clk) 477 { 478 unsigned long delta = expires - clk; 479 unsigned int idx; 480 481 if (delta < LVL_START(1)) { 482 idx = calc_index(expires, 0); 483 } else if (delta < LVL_START(2)) { 484 idx = calc_index(expires, 1); 485 } else if (delta < LVL_START(3)) { 486 idx = calc_index(expires, 2); 487 } else if (delta < LVL_START(4)) { 488 idx = calc_index(expires, 3); 489 } else if (delta < LVL_START(5)) { 490 idx = calc_index(expires, 4); 491 } else if (delta < LVL_START(6)) { 492 idx = calc_index(expires, 5); 493 } else if (delta < LVL_START(7)) { 494 idx = calc_index(expires, 6); 495 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { 496 idx = calc_index(expires, 7); 497 } else if ((long) delta < 0) { 498 idx = clk & LVL_MASK; 499 } else { 500 /* 501 * Force expire obscene large timeouts to expire at the 502 * capacity limit of the wheel. 503 */ 504 if (expires >= WHEEL_TIMEOUT_CUTOFF) 505 expires = WHEEL_TIMEOUT_MAX; 506 507 idx = calc_index(expires, LVL_DEPTH - 1); 508 } 509 return idx; 510 } 511 512 /* 513 * Enqueue the timer into the hash bucket, mark it pending in 514 * the bitmap and store the index in the timer flags. 515 */ 516 static void enqueue_timer(struct timer_base *base, struct timer_list *timer, 517 unsigned int idx) 518 { 519 hlist_add_head(&timer->entry, base->vectors + idx); 520 __set_bit(idx, base->pending_map); 521 timer_set_idx(timer, idx); 522 } 523 524 static void 525 __internal_add_timer(struct timer_base *base, struct timer_list *timer) 526 { 527 unsigned int idx; 528 529 idx = calc_wheel_index(timer->expires, base->clk); 530 enqueue_timer(base, timer, idx); 531 } 532 533 static void 534 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer) 535 { 536 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active) 537 return; 538 539 /* 540 * TODO: This wants some optimizing similar to the code below, but we 541 * will do that when we switch from push to pull for deferrable timers. 542 */ 543 if (timer->flags & TIMER_DEFERRABLE) { 544 if (tick_nohz_full_cpu(base->cpu)) 545 wake_up_nohz_cpu(base->cpu); 546 return; 547 } 548 549 /* 550 * We might have to IPI the remote CPU if the base is idle and the 551 * timer is not deferrable. If the other CPU is on the way to idle 552 * then it can't set base->is_idle as we hold the base lock: 553 */ 554 if (!base->is_idle) 555 return; 556 557 /* Check whether this is the new first expiring timer: */ 558 if (time_after_eq(timer->expires, base->next_expiry)) 559 return; 560 561 /* 562 * Set the next expiry time and kick the CPU so it can reevaluate the 563 * wheel: 564 */ 565 base->next_expiry = timer->expires; 566 wake_up_nohz_cpu(base->cpu); 567 } 568 569 static void 570 internal_add_timer(struct timer_base *base, struct timer_list *timer) 571 { 572 __internal_add_timer(base, timer); 573 trigger_dyntick_cpu(base, timer); 574 } 575 576 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS 577 578 static struct debug_obj_descr timer_debug_descr; 579 580 static void *timer_debug_hint(void *addr) 581 { 582 return ((struct timer_list *) addr)->function; 583 } 584 585 static bool timer_is_static_object(void *addr) 586 { 587 struct timer_list *timer = addr; 588 589 return (timer->entry.pprev == NULL && 590 timer->entry.next == TIMER_ENTRY_STATIC); 591 } 592 593 /* 594 * fixup_init is called when: 595 * - an active object is initialized 596 */ 597 static bool timer_fixup_init(void *addr, enum debug_obj_state state) 598 { 599 struct timer_list *timer = addr; 600 601 switch (state) { 602 case ODEBUG_STATE_ACTIVE: 603 del_timer_sync(timer); 604 debug_object_init(timer, &timer_debug_descr); 605 return true; 606 default: 607 return false; 608 } 609 } 610 611 /* Stub timer callback for improperly used timers. */ 612 static void stub_timer(unsigned long data) 613 { 614 WARN_ON(1); 615 } 616 617 /* 618 * fixup_activate is called when: 619 * - an active object is activated 620 * - an unknown non-static object is activated 621 */ 622 static bool timer_fixup_activate(void *addr, enum debug_obj_state state) 623 { 624 struct timer_list *timer = addr; 625 626 switch (state) { 627 case ODEBUG_STATE_NOTAVAILABLE: 628 setup_timer(timer, stub_timer, 0); 629 return true; 630 631 case ODEBUG_STATE_ACTIVE: 632 WARN_ON(1); 633 634 default: 635 return false; 636 } 637 } 638 639 /* 640 * fixup_free is called when: 641 * - an active object is freed 642 */ 643 static bool timer_fixup_free(void *addr, enum debug_obj_state state) 644 { 645 struct timer_list *timer = addr; 646 647 switch (state) { 648 case ODEBUG_STATE_ACTIVE: 649 del_timer_sync(timer); 650 debug_object_free(timer, &timer_debug_descr); 651 return true; 652 default: 653 return false; 654 } 655 } 656 657 /* 658 * fixup_assert_init is called when: 659 * - an untracked/uninit-ed object is found 660 */ 661 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) 662 { 663 struct timer_list *timer = addr; 664 665 switch (state) { 666 case ODEBUG_STATE_NOTAVAILABLE: 667 setup_timer(timer, stub_timer, 0); 668 return true; 669 default: 670 return false; 671 } 672 } 673 674 static struct debug_obj_descr timer_debug_descr = { 675 .name = "timer_list", 676 .debug_hint = timer_debug_hint, 677 .is_static_object = timer_is_static_object, 678 .fixup_init = timer_fixup_init, 679 .fixup_activate = timer_fixup_activate, 680 .fixup_free = timer_fixup_free, 681 .fixup_assert_init = timer_fixup_assert_init, 682 }; 683 684 static inline void debug_timer_init(struct timer_list *timer) 685 { 686 debug_object_init(timer, &timer_debug_descr); 687 } 688 689 static inline void debug_timer_activate(struct timer_list *timer) 690 { 691 debug_object_activate(timer, &timer_debug_descr); 692 } 693 694 static inline void debug_timer_deactivate(struct timer_list *timer) 695 { 696 debug_object_deactivate(timer, &timer_debug_descr); 697 } 698 699 static inline void debug_timer_free(struct timer_list *timer) 700 { 701 debug_object_free(timer, &timer_debug_descr); 702 } 703 704 static inline void debug_timer_assert_init(struct timer_list *timer) 705 { 706 debug_object_assert_init(timer, &timer_debug_descr); 707 } 708 709 static void do_init_timer(struct timer_list *timer, unsigned int flags, 710 const char *name, struct lock_class_key *key); 711 712 void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags, 713 const char *name, struct lock_class_key *key) 714 { 715 debug_object_init_on_stack(timer, &timer_debug_descr); 716 do_init_timer(timer, flags, name, key); 717 } 718 EXPORT_SYMBOL_GPL(init_timer_on_stack_key); 719 720 void destroy_timer_on_stack(struct timer_list *timer) 721 { 722 debug_object_free(timer, &timer_debug_descr); 723 } 724 EXPORT_SYMBOL_GPL(destroy_timer_on_stack); 725 726 #else 727 static inline void debug_timer_init(struct timer_list *timer) { } 728 static inline void debug_timer_activate(struct timer_list *timer) { } 729 static inline void debug_timer_deactivate(struct timer_list *timer) { } 730 static inline void debug_timer_assert_init(struct timer_list *timer) { } 731 #endif 732 733 static inline void debug_init(struct timer_list *timer) 734 { 735 debug_timer_init(timer); 736 trace_timer_init(timer); 737 } 738 739 static inline void 740 debug_activate(struct timer_list *timer, unsigned long expires) 741 { 742 debug_timer_activate(timer); 743 trace_timer_start(timer, expires, timer->flags); 744 } 745 746 static inline void debug_deactivate(struct timer_list *timer) 747 { 748 debug_timer_deactivate(timer); 749 trace_timer_cancel(timer); 750 } 751 752 static inline void debug_assert_init(struct timer_list *timer) 753 { 754 debug_timer_assert_init(timer); 755 } 756 757 static void do_init_timer(struct timer_list *timer, unsigned int flags, 758 const char *name, struct lock_class_key *key) 759 { 760 timer->entry.pprev = NULL; 761 timer->flags = flags | raw_smp_processor_id(); 762 lockdep_init_map(&timer->lockdep_map, name, key, 0); 763 } 764 765 /** 766 * init_timer_key - initialize a timer 767 * @timer: the timer to be initialized 768 * @flags: timer flags 769 * @name: name of the timer 770 * @key: lockdep class key of the fake lock used for tracking timer 771 * sync lock dependencies 772 * 773 * init_timer_key() must be done to a timer prior calling *any* of the 774 * other timer functions. 775 */ 776 void init_timer_key(struct timer_list *timer, unsigned int flags, 777 const char *name, struct lock_class_key *key) 778 { 779 debug_init(timer); 780 do_init_timer(timer, flags, name, key); 781 } 782 EXPORT_SYMBOL(init_timer_key); 783 784 static inline void detach_timer(struct timer_list *timer, bool clear_pending) 785 { 786 struct hlist_node *entry = &timer->entry; 787 788 debug_deactivate(timer); 789 790 __hlist_del(entry); 791 if (clear_pending) 792 entry->pprev = NULL; 793 entry->next = LIST_POISON2; 794 } 795 796 static int detach_if_pending(struct timer_list *timer, struct timer_base *base, 797 bool clear_pending) 798 { 799 unsigned idx = timer_get_idx(timer); 800 801 if (!timer_pending(timer)) 802 return 0; 803 804 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) 805 __clear_bit(idx, base->pending_map); 806 807 detach_timer(timer, clear_pending); 808 return 1; 809 } 810 811 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu) 812 { 813 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu); 814 815 /* 816 * If the timer is deferrable and nohz is active then we need to use 817 * the deferrable base. 818 */ 819 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active && 820 (tflags & TIMER_DEFERRABLE)) 821 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu); 822 return base; 823 } 824 825 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags) 826 { 827 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 828 829 /* 830 * If the timer is deferrable and nohz is active then we need to use 831 * the deferrable base. 832 */ 833 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active && 834 (tflags & TIMER_DEFERRABLE)) 835 base = this_cpu_ptr(&timer_bases[BASE_DEF]); 836 return base; 837 } 838 839 static inline struct timer_base *get_timer_base(u32 tflags) 840 { 841 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK); 842 } 843 844 #ifdef CONFIG_NO_HZ_COMMON 845 static inline struct timer_base * 846 get_target_base(struct timer_base *base, unsigned tflags) 847 { 848 #ifdef CONFIG_SMP 849 if ((tflags & TIMER_PINNED) || !base->migration_enabled) 850 return get_timer_this_cpu_base(tflags); 851 return get_timer_cpu_base(tflags, get_nohz_timer_target()); 852 #else 853 return get_timer_this_cpu_base(tflags); 854 #endif 855 } 856 857 static inline void forward_timer_base(struct timer_base *base) 858 { 859 unsigned long jnow = READ_ONCE(jiffies); 860 861 /* 862 * We only forward the base when it's idle and we have a delta between 863 * base clock and jiffies. 864 */ 865 if (!base->is_idle || (long) (jnow - base->clk) < 2) 866 return; 867 868 /* 869 * If the next expiry value is > jiffies, then we fast forward to 870 * jiffies otherwise we forward to the next expiry value. 871 */ 872 if (time_after(base->next_expiry, jnow)) 873 base->clk = jnow; 874 else 875 base->clk = base->next_expiry; 876 } 877 #else 878 static inline struct timer_base * 879 get_target_base(struct timer_base *base, unsigned tflags) 880 { 881 return get_timer_this_cpu_base(tflags); 882 } 883 884 static inline void forward_timer_base(struct timer_base *base) { } 885 #endif 886 887 888 /* 889 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means 890 * that all timers which are tied to this base are locked, and the base itself 891 * is locked too. 892 * 893 * So __run_timers/migrate_timers can safely modify all timers which could 894 * be found in the base->vectors array. 895 * 896 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need 897 * to wait until the migration is done. 898 */ 899 static struct timer_base *lock_timer_base(struct timer_list *timer, 900 unsigned long *flags) 901 __acquires(timer->base->lock) 902 { 903 for (;;) { 904 struct timer_base *base; 905 u32 tf; 906 907 /* 908 * We need to use READ_ONCE() here, otherwise the compiler 909 * might re-read @tf between the check for TIMER_MIGRATING 910 * and spin_lock(). 911 */ 912 tf = READ_ONCE(timer->flags); 913 914 if (!(tf & TIMER_MIGRATING)) { 915 base = get_timer_base(tf); 916 raw_spin_lock_irqsave(&base->lock, *flags); 917 if (timer->flags == tf) 918 return base; 919 raw_spin_unlock_irqrestore(&base->lock, *flags); 920 } 921 cpu_relax(); 922 } 923 } 924 925 static inline int 926 __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only) 927 { 928 struct timer_base *base, *new_base; 929 unsigned int idx = UINT_MAX; 930 unsigned long clk = 0, flags; 931 int ret = 0; 932 933 BUG_ON(!timer->function); 934 935 /* 936 * This is a common optimization triggered by the networking code - if 937 * the timer is re-modified to have the same timeout or ends up in the 938 * same array bucket then just return: 939 */ 940 if (timer_pending(timer)) { 941 if (timer->expires == expires) 942 return 1; 943 944 /* 945 * We lock timer base and calculate the bucket index right 946 * here. If the timer ends up in the same bucket, then we 947 * just update the expiry time and avoid the whole 948 * dequeue/enqueue dance. 949 */ 950 base = lock_timer_base(timer, &flags); 951 952 clk = base->clk; 953 idx = calc_wheel_index(expires, clk); 954 955 /* 956 * Retrieve and compare the array index of the pending 957 * timer. If it matches set the expiry to the new value so a 958 * subsequent call will exit in the expires check above. 959 */ 960 if (idx == timer_get_idx(timer)) { 961 timer->expires = expires; 962 ret = 1; 963 goto out_unlock; 964 } 965 } else { 966 base = lock_timer_base(timer, &flags); 967 } 968 969 ret = detach_if_pending(timer, base, false); 970 if (!ret && pending_only) 971 goto out_unlock; 972 973 debug_activate(timer, expires); 974 975 new_base = get_target_base(base, timer->flags); 976 977 if (base != new_base) { 978 /* 979 * We are trying to schedule the timer on the new base. 980 * However we can't change timer's base while it is running, 981 * otherwise del_timer_sync() can't detect that the timer's 982 * handler yet has not finished. This also guarantees that the 983 * timer is serialized wrt itself. 984 */ 985 if (likely(base->running_timer != timer)) { 986 /* See the comment in lock_timer_base() */ 987 timer->flags |= TIMER_MIGRATING; 988 989 raw_spin_unlock(&base->lock); 990 base = new_base; 991 raw_spin_lock(&base->lock); 992 WRITE_ONCE(timer->flags, 993 (timer->flags & ~TIMER_BASEMASK) | base->cpu); 994 } 995 } 996 997 /* Try to forward a stale timer base clock */ 998 forward_timer_base(base); 999 1000 timer->expires = expires; 1001 /* 1002 * If 'idx' was calculated above and the base time did not advance 1003 * between calculating 'idx' and possibly switching the base, only 1004 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise 1005 * we need to (re)calculate the wheel index via 1006 * internal_add_timer(). 1007 */ 1008 if (idx != UINT_MAX && clk == base->clk) { 1009 enqueue_timer(base, timer, idx); 1010 trigger_dyntick_cpu(base, timer); 1011 } else { 1012 internal_add_timer(base, timer); 1013 } 1014 1015 out_unlock: 1016 raw_spin_unlock_irqrestore(&base->lock, flags); 1017 1018 return ret; 1019 } 1020 1021 /** 1022 * mod_timer_pending - modify a pending timer's timeout 1023 * @timer: the pending timer to be modified 1024 * @expires: new timeout in jiffies 1025 * 1026 * mod_timer_pending() is the same for pending timers as mod_timer(), 1027 * but will not re-activate and modify already deleted timers. 1028 * 1029 * It is useful for unserialized use of timers. 1030 */ 1031 int mod_timer_pending(struct timer_list *timer, unsigned long expires) 1032 { 1033 return __mod_timer(timer, expires, true); 1034 } 1035 EXPORT_SYMBOL(mod_timer_pending); 1036 1037 /** 1038 * mod_timer - modify a timer's timeout 1039 * @timer: the timer to be modified 1040 * @expires: new timeout in jiffies 1041 * 1042 * mod_timer() is a more efficient way to update the expire field of an 1043 * active timer (if the timer is inactive it will be activated) 1044 * 1045 * mod_timer(timer, expires) is equivalent to: 1046 * 1047 * del_timer(timer); timer->expires = expires; add_timer(timer); 1048 * 1049 * Note that if there are multiple unserialized concurrent users of the 1050 * same timer, then mod_timer() is the only safe way to modify the timeout, 1051 * since add_timer() cannot modify an already running timer. 1052 * 1053 * The function returns whether it has modified a pending timer or not. 1054 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an 1055 * active timer returns 1.) 1056 */ 1057 int mod_timer(struct timer_list *timer, unsigned long expires) 1058 { 1059 return __mod_timer(timer, expires, false); 1060 } 1061 EXPORT_SYMBOL(mod_timer); 1062 1063 /** 1064 * add_timer - start a timer 1065 * @timer: the timer to be added 1066 * 1067 * The kernel will do a ->function(->data) callback from the 1068 * timer interrupt at the ->expires point in the future. The 1069 * current time is 'jiffies'. 1070 * 1071 * The timer's ->expires, ->function (and if the handler uses it, ->data) 1072 * fields must be set prior calling this function. 1073 * 1074 * Timers with an ->expires field in the past will be executed in the next 1075 * timer tick. 1076 */ 1077 void add_timer(struct timer_list *timer) 1078 { 1079 BUG_ON(timer_pending(timer)); 1080 mod_timer(timer, timer->expires); 1081 } 1082 EXPORT_SYMBOL(add_timer); 1083 1084 /** 1085 * add_timer_on - start a timer on a particular CPU 1086 * @timer: the timer to be added 1087 * @cpu: the CPU to start it on 1088 * 1089 * This is not very scalable on SMP. Double adds are not possible. 1090 */ 1091 void add_timer_on(struct timer_list *timer, int cpu) 1092 { 1093 struct timer_base *new_base, *base; 1094 unsigned long flags; 1095 1096 BUG_ON(timer_pending(timer) || !timer->function); 1097 1098 new_base = get_timer_cpu_base(timer->flags, cpu); 1099 1100 /* 1101 * If @timer was on a different CPU, it should be migrated with the 1102 * old base locked to prevent other operations proceeding with the 1103 * wrong base locked. See lock_timer_base(). 1104 */ 1105 base = lock_timer_base(timer, &flags); 1106 if (base != new_base) { 1107 timer->flags |= TIMER_MIGRATING; 1108 1109 raw_spin_unlock(&base->lock); 1110 base = new_base; 1111 raw_spin_lock(&base->lock); 1112 WRITE_ONCE(timer->flags, 1113 (timer->flags & ~TIMER_BASEMASK) | cpu); 1114 } 1115 1116 debug_activate(timer, timer->expires); 1117 internal_add_timer(base, timer); 1118 raw_spin_unlock_irqrestore(&base->lock, flags); 1119 } 1120 EXPORT_SYMBOL_GPL(add_timer_on); 1121 1122 /** 1123 * del_timer - deactivate a timer. 1124 * @timer: the timer to be deactivated 1125 * 1126 * del_timer() deactivates a timer - this works on both active and inactive 1127 * timers. 1128 * 1129 * The function returns whether it has deactivated a pending timer or not. 1130 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an 1131 * active timer returns 1.) 1132 */ 1133 int del_timer(struct timer_list *timer) 1134 { 1135 struct timer_base *base; 1136 unsigned long flags; 1137 int ret = 0; 1138 1139 debug_assert_init(timer); 1140 1141 if (timer_pending(timer)) { 1142 base = lock_timer_base(timer, &flags); 1143 ret = detach_if_pending(timer, base, true); 1144 raw_spin_unlock_irqrestore(&base->lock, flags); 1145 } 1146 1147 return ret; 1148 } 1149 EXPORT_SYMBOL(del_timer); 1150 1151 /** 1152 * try_to_del_timer_sync - Try to deactivate a timer 1153 * @timer: timer to delete 1154 * 1155 * This function tries to deactivate a timer. Upon successful (ret >= 0) 1156 * exit the timer is not queued and the handler is not running on any CPU. 1157 */ 1158 int try_to_del_timer_sync(struct timer_list *timer) 1159 { 1160 struct timer_base *base; 1161 unsigned long flags; 1162 int ret = -1; 1163 1164 debug_assert_init(timer); 1165 1166 base = lock_timer_base(timer, &flags); 1167 1168 if (base->running_timer != timer) 1169 ret = detach_if_pending(timer, base, true); 1170 1171 raw_spin_unlock_irqrestore(&base->lock, flags); 1172 1173 return ret; 1174 } 1175 EXPORT_SYMBOL(try_to_del_timer_sync); 1176 1177 #ifdef CONFIG_SMP 1178 /** 1179 * del_timer_sync - deactivate a timer and wait for the handler to finish. 1180 * @timer: the timer to be deactivated 1181 * 1182 * This function only differs from del_timer() on SMP: besides deactivating 1183 * the timer it also makes sure the handler has finished executing on other 1184 * CPUs. 1185 * 1186 * Synchronization rules: Callers must prevent restarting of the timer, 1187 * otherwise this function is meaningless. It must not be called from 1188 * interrupt contexts unless the timer is an irqsafe one. The caller must 1189 * not hold locks which would prevent completion of the timer's 1190 * handler. The timer's handler must not call add_timer_on(). Upon exit the 1191 * timer is not queued and the handler is not running on any CPU. 1192 * 1193 * Note: For !irqsafe timers, you must not hold locks that are held in 1194 * interrupt context while calling this function. Even if the lock has 1195 * nothing to do with the timer in question. Here's why: 1196 * 1197 * CPU0 CPU1 1198 * ---- ---- 1199 * <SOFTIRQ> 1200 * call_timer_fn(); 1201 * base->running_timer = mytimer; 1202 * spin_lock_irq(somelock); 1203 * <IRQ> 1204 * spin_lock(somelock); 1205 * del_timer_sync(mytimer); 1206 * while (base->running_timer == mytimer); 1207 * 1208 * Now del_timer_sync() will never return and never release somelock. 1209 * The interrupt on the other CPU is waiting to grab somelock but 1210 * it has interrupted the softirq that CPU0 is waiting to finish. 1211 * 1212 * The function returns whether it has deactivated a pending timer or not. 1213 */ 1214 int del_timer_sync(struct timer_list *timer) 1215 { 1216 #ifdef CONFIG_LOCKDEP 1217 unsigned long flags; 1218 1219 /* 1220 * If lockdep gives a backtrace here, please reference 1221 * the synchronization rules above. 1222 */ 1223 local_irq_save(flags); 1224 lock_map_acquire(&timer->lockdep_map); 1225 lock_map_release(&timer->lockdep_map); 1226 local_irq_restore(flags); 1227 #endif 1228 /* 1229 * don't use it in hardirq context, because it 1230 * could lead to deadlock. 1231 */ 1232 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); 1233 for (;;) { 1234 int ret = try_to_del_timer_sync(timer); 1235 if (ret >= 0) 1236 return ret; 1237 cpu_relax(); 1238 } 1239 } 1240 EXPORT_SYMBOL(del_timer_sync); 1241 #endif 1242 1243 static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long), 1244 unsigned long data) 1245 { 1246 int count = preempt_count(); 1247 1248 #ifdef CONFIG_LOCKDEP 1249 /* 1250 * It is permissible to free the timer from inside the 1251 * function that is called from it, this we need to take into 1252 * account for lockdep too. To avoid bogus "held lock freed" 1253 * warnings as well as problems when looking into 1254 * timer->lockdep_map, make a copy and use that here. 1255 */ 1256 struct lockdep_map lockdep_map; 1257 1258 lockdep_copy_map(&lockdep_map, &timer->lockdep_map); 1259 #endif 1260 /* 1261 * Couple the lock chain with the lock chain at 1262 * del_timer_sync() by acquiring the lock_map around the fn() 1263 * call here and in del_timer_sync(). 1264 */ 1265 lock_map_acquire(&lockdep_map); 1266 1267 trace_timer_expire_entry(timer); 1268 fn(data); 1269 trace_timer_expire_exit(timer); 1270 1271 lock_map_release(&lockdep_map); 1272 1273 if (count != preempt_count()) { 1274 WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", 1275 fn, count, preempt_count()); 1276 /* 1277 * Restore the preempt count. That gives us a decent 1278 * chance to survive and extract information. If the 1279 * callback kept a lock held, bad luck, but not worse 1280 * than the BUG() we had. 1281 */ 1282 preempt_count_set(count); 1283 } 1284 } 1285 1286 static void expire_timers(struct timer_base *base, struct hlist_head *head) 1287 { 1288 while (!hlist_empty(head)) { 1289 struct timer_list *timer; 1290 void (*fn)(unsigned long); 1291 unsigned long data; 1292 1293 timer = hlist_entry(head->first, struct timer_list, entry); 1294 1295 base->running_timer = timer; 1296 detach_timer(timer, true); 1297 1298 fn = timer->function; 1299 data = timer->data; 1300 1301 if (timer->flags & TIMER_IRQSAFE) { 1302 raw_spin_unlock(&base->lock); 1303 call_timer_fn(timer, fn, data); 1304 raw_spin_lock(&base->lock); 1305 } else { 1306 raw_spin_unlock_irq(&base->lock); 1307 call_timer_fn(timer, fn, data); 1308 raw_spin_lock_irq(&base->lock); 1309 } 1310 } 1311 } 1312 1313 static int __collect_expired_timers(struct timer_base *base, 1314 struct hlist_head *heads) 1315 { 1316 unsigned long clk = base->clk; 1317 struct hlist_head *vec; 1318 int i, levels = 0; 1319 unsigned int idx; 1320 1321 for (i = 0; i < LVL_DEPTH; i++) { 1322 idx = (clk & LVL_MASK) + i * LVL_SIZE; 1323 1324 if (__test_and_clear_bit(idx, base->pending_map)) { 1325 vec = base->vectors + idx; 1326 hlist_move_list(vec, heads++); 1327 levels++; 1328 } 1329 /* Is it time to look at the next level? */ 1330 if (clk & LVL_CLK_MASK) 1331 break; 1332 /* Shift clock for the next level granularity */ 1333 clk >>= LVL_CLK_SHIFT; 1334 } 1335 return levels; 1336 } 1337 1338 #ifdef CONFIG_NO_HZ_COMMON 1339 /* 1340 * Find the next pending bucket of a level. Search from level start (@offset) 1341 * + @clk upwards and if nothing there, search from start of the level 1342 * (@offset) up to @offset + clk. 1343 */ 1344 static int next_pending_bucket(struct timer_base *base, unsigned offset, 1345 unsigned clk) 1346 { 1347 unsigned pos, start = offset + clk; 1348 unsigned end = offset + LVL_SIZE; 1349 1350 pos = find_next_bit(base->pending_map, end, start); 1351 if (pos < end) 1352 return pos - start; 1353 1354 pos = find_next_bit(base->pending_map, start, offset); 1355 return pos < start ? pos + LVL_SIZE - start : -1; 1356 } 1357 1358 /* 1359 * Search the first expiring timer in the various clock levels. Caller must 1360 * hold base->lock. 1361 */ 1362 static unsigned long __next_timer_interrupt(struct timer_base *base) 1363 { 1364 unsigned long clk, next, adj; 1365 unsigned lvl, offset = 0; 1366 1367 next = base->clk + NEXT_TIMER_MAX_DELTA; 1368 clk = base->clk; 1369 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) { 1370 int pos = next_pending_bucket(base, offset, clk & LVL_MASK); 1371 1372 if (pos >= 0) { 1373 unsigned long tmp = clk + (unsigned long) pos; 1374 1375 tmp <<= LVL_SHIFT(lvl); 1376 if (time_before(tmp, next)) 1377 next = tmp; 1378 } 1379 /* 1380 * Clock for the next level. If the current level clock lower 1381 * bits are zero, we look at the next level as is. If not we 1382 * need to advance it by one because that's going to be the 1383 * next expiring bucket in that level. base->clk is the next 1384 * expiring jiffie. So in case of: 1385 * 1386 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1387 * 0 0 0 0 0 0 1388 * 1389 * we have to look at all levels @index 0. With 1390 * 1391 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1392 * 0 0 0 0 0 2 1393 * 1394 * LVL0 has the next expiring bucket @index 2. The upper 1395 * levels have the next expiring bucket @index 1. 1396 * 1397 * In case that the propagation wraps the next level the same 1398 * rules apply: 1399 * 1400 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1401 * 0 0 0 0 F 2 1402 * 1403 * So after looking at LVL0 we get: 1404 * 1405 * LVL5 LVL4 LVL3 LVL2 LVL1 1406 * 0 0 0 1 0 1407 * 1408 * So no propagation from LVL1 to LVL2 because that happened 1409 * with the add already, but then we need to propagate further 1410 * from LVL2 to LVL3. 1411 * 1412 * So the simple check whether the lower bits of the current 1413 * level are 0 or not is sufficient for all cases. 1414 */ 1415 adj = clk & LVL_CLK_MASK ? 1 : 0; 1416 clk >>= LVL_CLK_SHIFT; 1417 clk += adj; 1418 } 1419 return next; 1420 } 1421 1422 /* 1423 * Check, if the next hrtimer event is before the next timer wheel 1424 * event: 1425 */ 1426 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) 1427 { 1428 u64 nextevt = hrtimer_get_next_event(); 1429 1430 /* 1431 * If high resolution timers are enabled 1432 * hrtimer_get_next_event() returns KTIME_MAX. 1433 */ 1434 if (expires <= nextevt) 1435 return expires; 1436 1437 /* 1438 * If the next timer is already expired, return the tick base 1439 * time so the tick is fired immediately. 1440 */ 1441 if (nextevt <= basem) 1442 return basem; 1443 1444 /* 1445 * Round up to the next jiffie. High resolution timers are 1446 * off, so the hrtimers are expired in the tick and we need to 1447 * make sure that this tick really expires the timer to avoid 1448 * a ping pong of the nohz stop code. 1449 * 1450 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 1451 */ 1452 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; 1453 } 1454 1455 /** 1456 * get_next_timer_interrupt - return the time (clock mono) of the next timer 1457 * @basej: base time jiffies 1458 * @basem: base time clock monotonic 1459 * 1460 * Returns the tick aligned clock monotonic time of the next pending 1461 * timer or KTIME_MAX if no timer is pending. 1462 */ 1463 u64 get_next_timer_interrupt(unsigned long basej, u64 basem) 1464 { 1465 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1466 u64 expires = KTIME_MAX; 1467 unsigned long nextevt; 1468 bool is_max_delta; 1469 1470 /* 1471 * Pretend that there is no timer pending if the cpu is offline. 1472 * Possible pending timers will be migrated later to an active cpu. 1473 */ 1474 if (cpu_is_offline(smp_processor_id())) 1475 return expires; 1476 1477 raw_spin_lock(&base->lock); 1478 nextevt = __next_timer_interrupt(base); 1479 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA); 1480 base->next_expiry = nextevt; 1481 /* 1482 * We have a fresh next event. Check whether we can forward the 1483 * base. We can only do that when @basej is past base->clk 1484 * otherwise we might rewind base->clk. 1485 */ 1486 if (time_after(basej, base->clk)) { 1487 if (time_after(nextevt, basej)) 1488 base->clk = basej; 1489 else if (time_after(nextevt, base->clk)) 1490 base->clk = nextevt; 1491 } 1492 1493 if (time_before_eq(nextevt, basej)) { 1494 expires = basem; 1495 base->is_idle = false; 1496 } else { 1497 if (!is_max_delta) 1498 expires = basem + (u64)(nextevt - basej) * TICK_NSEC; 1499 /* 1500 * If we expect to sleep more than a tick, mark the base idle: 1501 */ 1502 if ((expires - basem) > TICK_NSEC) 1503 base->is_idle = true; 1504 } 1505 raw_spin_unlock(&base->lock); 1506 1507 return cmp_next_hrtimer_event(basem, expires); 1508 } 1509 1510 /** 1511 * timer_clear_idle - Clear the idle state of the timer base 1512 * 1513 * Called with interrupts disabled 1514 */ 1515 void timer_clear_idle(void) 1516 { 1517 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1518 1519 /* 1520 * We do this unlocked. The worst outcome is a remote enqueue sending 1521 * a pointless IPI, but taking the lock would just make the window for 1522 * sending the IPI a few instructions smaller for the cost of taking 1523 * the lock in the exit from idle path. 1524 */ 1525 base->is_idle = false; 1526 } 1527 1528 static int collect_expired_timers(struct timer_base *base, 1529 struct hlist_head *heads) 1530 { 1531 /* 1532 * NOHZ optimization. After a long idle sleep we need to forward the 1533 * base to current jiffies. Avoid a loop by searching the bitfield for 1534 * the next expiring timer. 1535 */ 1536 if ((long)(jiffies - base->clk) > 2) { 1537 unsigned long next = __next_timer_interrupt(base); 1538 1539 /* 1540 * If the next timer is ahead of time forward to current 1541 * jiffies, otherwise forward to the next expiry time: 1542 */ 1543 if (time_after(next, jiffies)) { 1544 /* The call site will increment clock! */ 1545 base->clk = jiffies - 1; 1546 return 0; 1547 } 1548 base->clk = next; 1549 } 1550 return __collect_expired_timers(base, heads); 1551 } 1552 #else 1553 static inline int collect_expired_timers(struct timer_base *base, 1554 struct hlist_head *heads) 1555 { 1556 return __collect_expired_timers(base, heads); 1557 } 1558 #endif 1559 1560 /* 1561 * Called from the timer interrupt handler to charge one tick to the current 1562 * process. user_tick is 1 if the tick is user time, 0 for system. 1563 */ 1564 void update_process_times(int user_tick) 1565 { 1566 struct task_struct *p = current; 1567 1568 /* Note: this timer irq context must be accounted for as well. */ 1569 account_process_tick(p, user_tick); 1570 run_local_timers(); 1571 rcu_check_callbacks(user_tick); 1572 #ifdef CONFIG_IRQ_WORK 1573 if (in_irq()) 1574 irq_work_tick(); 1575 #endif 1576 scheduler_tick(); 1577 if (IS_ENABLED(CONFIG_POSIX_TIMERS)) 1578 run_posix_cpu_timers(p); 1579 } 1580 1581 /** 1582 * __run_timers - run all expired timers (if any) on this CPU. 1583 * @base: the timer vector to be processed. 1584 */ 1585 static inline void __run_timers(struct timer_base *base) 1586 { 1587 struct hlist_head heads[LVL_DEPTH]; 1588 int levels; 1589 1590 if (!time_after_eq(jiffies, base->clk)) 1591 return; 1592 1593 raw_spin_lock_irq(&base->lock); 1594 1595 while (time_after_eq(jiffies, base->clk)) { 1596 1597 levels = collect_expired_timers(base, heads); 1598 base->clk++; 1599 1600 while (levels--) 1601 expire_timers(base, heads + levels); 1602 } 1603 base->running_timer = NULL; 1604 raw_spin_unlock_irq(&base->lock); 1605 } 1606 1607 /* 1608 * This function runs timers and the timer-tq in bottom half context. 1609 */ 1610 static __latent_entropy void run_timer_softirq(struct softirq_action *h) 1611 { 1612 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1613 1614 __run_timers(base); 1615 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && base->nohz_active) 1616 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); 1617 } 1618 1619 /* 1620 * Called by the local, per-CPU timer interrupt on SMP. 1621 */ 1622 void run_local_timers(void) 1623 { 1624 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1625 1626 hrtimer_run_queues(); 1627 /* Raise the softirq only if required. */ 1628 if (time_before(jiffies, base->clk)) { 1629 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active) 1630 return; 1631 /* CPU is awake, so check the deferrable base. */ 1632 base++; 1633 if (time_before(jiffies, base->clk)) 1634 return; 1635 } 1636 raise_softirq(TIMER_SOFTIRQ); 1637 } 1638 1639 static void process_timeout(unsigned long __data) 1640 { 1641 wake_up_process((struct task_struct *)__data); 1642 } 1643 1644 /** 1645 * schedule_timeout - sleep until timeout 1646 * @timeout: timeout value in jiffies 1647 * 1648 * Make the current task sleep until @timeout jiffies have 1649 * elapsed. The routine will return immediately unless 1650 * the current task state has been set (see set_current_state()). 1651 * 1652 * You can set the task state as follows - 1653 * 1654 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to 1655 * pass before the routine returns unless the current task is explicitly 1656 * woken up, (e.g. by wake_up_process())". 1657 * 1658 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is 1659 * delivered to the current task or the current task is explicitly woken 1660 * up. 1661 * 1662 * The current task state is guaranteed to be TASK_RUNNING when this 1663 * routine returns. 1664 * 1665 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule 1666 * the CPU away without a bound on the timeout. In this case the return 1667 * value will be %MAX_SCHEDULE_TIMEOUT. 1668 * 1669 * Returns 0 when the timer has expired otherwise the remaining time in 1670 * jiffies will be returned. In all cases the return value is guaranteed 1671 * to be non-negative. 1672 */ 1673 signed long __sched schedule_timeout(signed long timeout) 1674 { 1675 struct timer_list timer; 1676 unsigned long expire; 1677 1678 switch (timeout) 1679 { 1680 case MAX_SCHEDULE_TIMEOUT: 1681 /* 1682 * These two special cases are useful to be comfortable 1683 * in the caller. Nothing more. We could take 1684 * MAX_SCHEDULE_TIMEOUT from one of the negative value 1685 * but I' d like to return a valid offset (>=0) to allow 1686 * the caller to do everything it want with the retval. 1687 */ 1688 schedule(); 1689 goto out; 1690 default: 1691 /* 1692 * Another bit of PARANOID. Note that the retval will be 1693 * 0 since no piece of kernel is supposed to do a check 1694 * for a negative retval of schedule_timeout() (since it 1695 * should never happens anyway). You just have the printk() 1696 * that will tell you if something is gone wrong and where. 1697 */ 1698 if (timeout < 0) { 1699 printk(KERN_ERR "schedule_timeout: wrong timeout " 1700 "value %lx\n", timeout); 1701 dump_stack(); 1702 current->state = TASK_RUNNING; 1703 goto out; 1704 } 1705 } 1706 1707 expire = timeout + jiffies; 1708 1709 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); 1710 __mod_timer(&timer, expire, false); 1711 schedule(); 1712 del_singleshot_timer_sync(&timer); 1713 1714 /* Remove the timer from the object tracker */ 1715 destroy_timer_on_stack(&timer); 1716 1717 timeout = expire - jiffies; 1718 1719 out: 1720 return timeout < 0 ? 0 : timeout; 1721 } 1722 EXPORT_SYMBOL(schedule_timeout); 1723 1724 /* 1725 * We can use __set_current_state() here because schedule_timeout() calls 1726 * schedule() unconditionally. 1727 */ 1728 signed long __sched schedule_timeout_interruptible(signed long timeout) 1729 { 1730 __set_current_state(TASK_INTERRUPTIBLE); 1731 return schedule_timeout(timeout); 1732 } 1733 EXPORT_SYMBOL(schedule_timeout_interruptible); 1734 1735 signed long __sched schedule_timeout_killable(signed long timeout) 1736 { 1737 __set_current_state(TASK_KILLABLE); 1738 return schedule_timeout(timeout); 1739 } 1740 EXPORT_SYMBOL(schedule_timeout_killable); 1741 1742 signed long __sched schedule_timeout_uninterruptible(signed long timeout) 1743 { 1744 __set_current_state(TASK_UNINTERRUPTIBLE); 1745 return schedule_timeout(timeout); 1746 } 1747 EXPORT_SYMBOL(schedule_timeout_uninterruptible); 1748 1749 /* 1750 * Like schedule_timeout_uninterruptible(), except this task will not contribute 1751 * to load average. 1752 */ 1753 signed long __sched schedule_timeout_idle(signed long timeout) 1754 { 1755 __set_current_state(TASK_IDLE); 1756 return schedule_timeout(timeout); 1757 } 1758 EXPORT_SYMBOL(schedule_timeout_idle); 1759 1760 #ifdef CONFIG_HOTPLUG_CPU 1761 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head) 1762 { 1763 struct timer_list *timer; 1764 int cpu = new_base->cpu; 1765 1766 while (!hlist_empty(head)) { 1767 timer = hlist_entry(head->first, struct timer_list, entry); 1768 detach_timer(timer, false); 1769 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; 1770 internal_add_timer(new_base, timer); 1771 } 1772 } 1773 1774 int timers_dead_cpu(unsigned int cpu) 1775 { 1776 struct timer_base *old_base; 1777 struct timer_base *new_base; 1778 int b, i; 1779 1780 BUG_ON(cpu_online(cpu)); 1781 1782 for (b = 0; b < NR_BASES; b++) { 1783 old_base = per_cpu_ptr(&timer_bases[b], cpu); 1784 new_base = get_cpu_ptr(&timer_bases[b]); 1785 /* 1786 * The caller is globally serialized and nobody else 1787 * takes two locks at once, deadlock is not possible. 1788 */ 1789 raw_spin_lock_irq(&new_base->lock); 1790 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); 1791 1792 BUG_ON(old_base->running_timer); 1793 1794 for (i = 0; i < WHEEL_SIZE; i++) 1795 migrate_timer_list(new_base, old_base->vectors + i); 1796 1797 raw_spin_unlock(&old_base->lock); 1798 raw_spin_unlock_irq(&new_base->lock); 1799 put_cpu_ptr(&timer_bases); 1800 } 1801 return 0; 1802 } 1803 1804 #endif /* CONFIG_HOTPLUG_CPU */ 1805 1806 static void __init init_timer_cpu(int cpu) 1807 { 1808 struct timer_base *base; 1809 int i; 1810 1811 for (i = 0; i < NR_BASES; i++) { 1812 base = per_cpu_ptr(&timer_bases[i], cpu); 1813 base->cpu = cpu; 1814 raw_spin_lock_init(&base->lock); 1815 base->clk = jiffies; 1816 } 1817 } 1818 1819 static void __init init_timer_cpus(void) 1820 { 1821 int cpu; 1822 1823 for_each_possible_cpu(cpu) 1824 init_timer_cpu(cpu); 1825 } 1826 1827 void __init init_timers(void) 1828 { 1829 init_timer_cpus(); 1830 open_softirq(TIMER_SOFTIRQ, run_timer_softirq); 1831 } 1832 1833 /** 1834 * msleep - sleep safely even with waitqueue interruptions 1835 * @msecs: Time in milliseconds to sleep for 1836 */ 1837 void msleep(unsigned int msecs) 1838 { 1839 unsigned long timeout = msecs_to_jiffies(msecs) + 1; 1840 1841 while (timeout) 1842 timeout = schedule_timeout_uninterruptible(timeout); 1843 } 1844 1845 EXPORT_SYMBOL(msleep); 1846 1847 /** 1848 * msleep_interruptible - sleep waiting for signals 1849 * @msecs: Time in milliseconds to sleep for 1850 */ 1851 unsigned long msleep_interruptible(unsigned int msecs) 1852 { 1853 unsigned long timeout = msecs_to_jiffies(msecs) + 1; 1854 1855 while (timeout && !signal_pending(current)) 1856 timeout = schedule_timeout_interruptible(timeout); 1857 return jiffies_to_msecs(timeout); 1858 } 1859 1860 EXPORT_SYMBOL(msleep_interruptible); 1861 1862 /** 1863 * usleep_range - Sleep for an approximate time 1864 * @min: Minimum time in usecs to sleep 1865 * @max: Maximum time in usecs to sleep 1866 * 1867 * In non-atomic context where the exact wakeup time is flexible, use 1868 * usleep_range() instead of udelay(). The sleep improves responsiveness 1869 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces 1870 * power usage by allowing hrtimers to take advantage of an already- 1871 * scheduled interrupt instead of scheduling a new one just for this sleep. 1872 */ 1873 void __sched usleep_range(unsigned long min, unsigned long max) 1874 { 1875 ktime_t exp = ktime_add_us(ktime_get(), min); 1876 u64 delta = (u64)(max - min) * NSEC_PER_USEC; 1877 1878 for (;;) { 1879 __set_current_state(TASK_UNINTERRUPTIBLE); 1880 /* Do not return before the requested sleep time has elapsed */ 1881 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS)) 1882 break; 1883 } 1884 } 1885 EXPORT_SYMBOL(usleep_range); 1886