1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Kernel internal timers 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better. 8 * 9 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 10 * "A Kernel Model for Precision Timekeeping" by Dave Mills 11 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to 12 * serialize accesses to xtime/lost_ticks). 13 * Copyright (C) 1998 Andrea Arcangeli 14 * 1999-03-10 Improved NTP compatibility by Ulrich Windl 15 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love 16 * 2000-10-05 Implemented scalable SMP per-CPU timer handling. 17 * Copyright (C) 2000, 2001, 2002 Ingo Molnar 18 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar 19 */ 20 21 #include <linux/kernel_stat.h> 22 #include <linux/export.h> 23 #include <linux/interrupt.h> 24 #include <linux/percpu.h> 25 #include <linux/init.h> 26 #include <linux/mm.h> 27 #include <linux/swap.h> 28 #include <linux/pid_namespace.h> 29 #include <linux/notifier.h> 30 #include <linux/thread_info.h> 31 #include <linux/time.h> 32 #include <linux/jiffies.h> 33 #include <linux/posix-timers.h> 34 #include <linux/cpu.h> 35 #include <linux/syscalls.h> 36 #include <linux/delay.h> 37 #include <linux/tick.h> 38 #include <linux/kallsyms.h> 39 #include <linux/irq_work.h> 40 #include <linux/sched/signal.h> 41 #include <linux/sched/sysctl.h> 42 #include <linux/sched/nohz.h> 43 #include <linux/sched/debug.h> 44 #include <linux/slab.h> 45 #include <linux/compat.h> 46 #include <linux/random.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 #ifdef CONFIG_PREEMPT_RT 201 spinlock_t expiry_lock; 202 atomic_t timer_waiters; 203 #endif 204 unsigned long clk; 205 unsigned long next_expiry; 206 unsigned int cpu; 207 bool is_idle; 208 bool must_forward_clk; 209 DECLARE_BITMAP(pending_map, WHEEL_SIZE); 210 struct hlist_head vectors[WHEEL_SIZE]; 211 } ____cacheline_aligned; 212 213 static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]); 214 215 #ifdef CONFIG_NO_HZ_COMMON 216 217 static DEFINE_STATIC_KEY_FALSE(timers_nohz_active); 218 static DEFINE_MUTEX(timer_keys_mutex); 219 220 static void timer_update_keys(struct work_struct *work); 221 static DECLARE_WORK(timer_update_work, timer_update_keys); 222 223 #ifdef CONFIG_SMP 224 unsigned int sysctl_timer_migration = 1; 225 226 DEFINE_STATIC_KEY_FALSE(timers_migration_enabled); 227 228 static void timers_update_migration(void) 229 { 230 if (sysctl_timer_migration && tick_nohz_active) 231 static_branch_enable(&timers_migration_enabled); 232 else 233 static_branch_disable(&timers_migration_enabled); 234 } 235 #else 236 static inline void timers_update_migration(void) { } 237 #endif /* !CONFIG_SMP */ 238 239 static void timer_update_keys(struct work_struct *work) 240 { 241 mutex_lock(&timer_keys_mutex); 242 timers_update_migration(); 243 static_branch_enable(&timers_nohz_active); 244 mutex_unlock(&timer_keys_mutex); 245 } 246 247 void timers_update_nohz(void) 248 { 249 schedule_work(&timer_update_work); 250 } 251 252 int timer_migration_handler(struct ctl_table *table, int write, 253 void *buffer, size_t *lenp, loff_t *ppos) 254 { 255 int ret; 256 257 mutex_lock(&timer_keys_mutex); 258 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 259 if (!ret && write) 260 timers_update_migration(); 261 mutex_unlock(&timer_keys_mutex); 262 return ret; 263 } 264 265 static inline bool is_timers_nohz_active(void) 266 { 267 return static_branch_unlikely(&timers_nohz_active); 268 } 269 #else 270 static inline bool is_timers_nohz_active(void) { return false; } 271 #endif /* NO_HZ_COMMON */ 272 273 static unsigned long round_jiffies_common(unsigned long j, int cpu, 274 bool force_up) 275 { 276 int rem; 277 unsigned long original = j; 278 279 /* 280 * We don't want all cpus firing their timers at once hitting the 281 * same lock or cachelines, so we skew each extra cpu with an extra 282 * 3 jiffies. This 3 jiffies came originally from the mm/ code which 283 * already did this. 284 * The skew is done by adding 3*cpunr, then round, then subtract this 285 * extra offset again. 286 */ 287 j += cpu * 3; 288 289 rem = j % HZ; 290 291 /* 292 * If the target jiffie is just after a whole second (which can happen 293 * due to delays of the timer irq, long irq off times etc etc) then 294 * we should round down to the whole second, not up. Use 1/4th second 295 * as cutoff for this rounding as an extreme upper bound for this. 296 * But never round down if @force_up is set. 297 */ 298 if (rem < HZ/4 && !force_up) /* round down */ 299 j = j - rem; 300 else /* round up */ 301 j = j - rem + HZ; 302 303 /* now that we have rounded, subtract the extra skew again */ 304 j -= cpu * 3; 305 306 /* 307 * Make sure j is still in the future. Otherwise return the 308 * unmodified value. 309 */ 310 return time_is_after_jiffies(j) ? j : original; 311 } 312 313 /** 314 * __round_jiffies - function to round jiffies to a full second 315 * @j: the time in (absolute) jiffies that should be rounded 316 * @cpu: the processor number on which the timeout will happen 317 * 318 * __round_jiffies() rounds an absolute time in the future (in jiffies) 319 * up or down to (approximately) full seconds. This is useful for timers 320 * for which the exact time they fire does not matter too much, as long as 321 * they fire approximately every X seconds. 322 * 323 * By rounding these timers to whole seconds, all such timers will fire 324 * at the same time, rather than at various times spread out. The goal 325 * of this is to have the CPU wake up less, which saves power. 326 * 327 * The exact rounding is skewed for each processor to avoid all 328 * processors firing at the exact same time, which could lead 329 * to lock contention or spurious cache line bouncing. 330 * 331 * The return value is the rounded version of the @j parameter. 332 */ 333 unsigned long __round_jiffies(unsigned long j, int cpu) 334 { 335 return round_jiffies_common(j, cpu, false); 336 } 337 EXPORT_SYMBOL_GPL(__round_jiffies); 338 339 /** 340 * __round_jiffies_relative - function to round jiffies to a full second 341 * @j: the time in (relative) jiffies that should be rounded 342 * @cpu: the processor number on which the timeout will happen 343 * 344 * __round_jiffies_relative() rounds a time delta in the future (in jiffies) 345 * up or down to (approximately) full seconds. This is useful for timers 346 * for which the exact time they fire does not matter too much, as long as 347 * they fire approximately every X seconds. 348 * 349 * By rounding these timers to whole seconds, all such timers will fire 350 * at the same time, rather than at various times spread out. The goal 351 * of this is to have the CPU wake up less, which saves power. 352 * 353 * The exact rounding is skewed for each processor to avoid all 354 * processors firing at the exact same time, which could lead 355 * to lock contention or spurious cache line bouncing. 356 * 357 * The return value is the rounded version of the @j parameter. 358 */ 359 unsigned long __round_jiffies_relative(unsigned long j, int cpu) 360 { 361 unsigned long j0 = jiffies; 362 363 /* Use j0 because jiffies might change while we run */ 364 return round_jiffies_common(j + j0, cpu, false) - j0; 365 } 366 EXPORT_SYMBOL_GPL(__round_jiffies_relative); 367 368 /** 369 * round_jiffies - function to round jiffies to a full second 370 * @j: the time in (absolute) jiffies that should be rounded 371 * 372 * round_jiffies() rounds an absolute time 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(unsigned long j) 384 { 385 return round_jiffies_common(j, raw_smp_processor_id(), false); 386 } 387 EXPORT_SYMBOL_GPL(round_jiffies); 388 389 /** 390 * round_jiffies_relative - function to round jiffies to a full second 391 * @j: the time in (relative) jiffies that should be rounded 392 * 393 * round_jiffies_relative() rounds a time delta in the future (in jiffies) 394 * up or down to (approximately) full seconds. This is useful for timers 395 * for which the exact time they fire does not matter too much, as long as 396 * they fire approximately every X seconds. 397 * 398 * By rounding these timers to whole seconds, all such timers will fire 399 * at the same time, rather than at various times spread out. The goal 400 * of this is to have the CPU wake up less, which saves power. 401 * 402 * The return value is the rounded version of the @j parameter. 403 */ 404 unsigned long round_jiffies_relative(unsigned long j) 405 { 406 return __round_jiffies_relative(j, raw_smp_processor_id()); 407 } 408 EXPORT_SYMBOL_GPL(round_jiffies_relative); 409 410 /** 411 * __round_jiffies_up - function to round jiffies up to a full second 412 * @j: the time in (absolute) jiffies that should be rounded 413 * @cpu: the processor number on which the timeout will happen 414 * 415 * This is the same as __round_jiffies() except that it will never 416 * round down. This is useful for timeouts for which the exact time 417 * of firing does not matter too much, as long as they don't fire too 418 * early. 419 */ 420 unsigned long __round_jiffies_up(unsigned long j, int cpu) 421 { 422 return round_jiffies_common(j, cpu, true); 423 } 424 EXPORT_SYMBOL_GPL(__round_jiffies_up); 425 426 /** 427 * __round_jiffies_up_relative - function to round jiffies up to a full second 428 * @j: the time in (relative) jiffies that should be rounded 429 * @cpu: the processor number on which the timeout will happen 430 * 431 * This is the same as __round_jiffies_relative() except that it will never 432 * round down. This is useful for timeouts for which the exact time 433 * of firing does not matter too much, as long as they don't fire too 434 * early. 435 */ 436 unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) 437 { 438 unsigned long j0 = jiffies; 439 440 /* Use j0 because jiffies might change while we run */ 441 return round_jiffies_common(j + j0, cpu, true) - j0; 442 } 443 EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); 444 445 /** 446 * round_jiffies_up - function to round jiffies up to a full second 447 * @j: the time in (absolute) jiffies that should be rounded 448 * 449 * This is the same as round_jiffies() except that it will never 450 * round down. This is useful for timeouts for which the exact time 451 * of firing does not matter too much, as long as they don't fire too 452 * early. 453 */ 454 unsigned long round_jiffies_up(unsigned long j) 455 { 456 return round_jiffies_common(j, raw_smp_processor_id(), true); 457 } 458 EXPORT_SYMBOL_GPL(round_jiffies_up); 459 460 /** 461 * round_jiffies_up_relative - function to round jiffies up to a full second 462 * @j: the time in (relative) jiffies that should be rounded 463 * 464 * This is the same as round_jiffies_relative() except that it will never 465 * round down. This is useful for timeouts for which the exact time 466 * of firing does not matter too much, as long as they don't fire too 467 * early. 468 */ 469 unsigned long round_jiffies_up_relative(unsigned long j) 470 { 471 return __round_jiffies_up_relative(j, raw_smp_processor_id()); 472 } 473 EXPORT_SYMBOL_GPL(round_jiffies_up_relative); 474 475 476 static inline unsigned int timer_get_idx(struct timer_list *timer) 477 { 478 return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT; 479 } 480 481 static inline void timer_set_idx(struct timer_list *timer, unsigned int idx) 482 { 483 timer->flags = (timer->flags & ~TIMER_ARRAYMASK) | 484 idx << TIMER_ARRAYSHIFT; 485 } 486 487 /* 488 * Helper function to calculate the array index for a given expiry 489 * time. 490 */ 491 static inline unsigned calc_index(unsigned expires, unsigned lvl) 492 { 493 expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl); 494 return LVL_OFFS(lvl) + (expires & LVL_MASK); 495 } 496 497 static int calc_wheel_index(unsigned long expires, unsigned long clk) 498 { 499 unsigned long delta = expires - clk; 500 unsigned int idx; 501 502 if (delta < LVL_START(1)) { 503 idx = calc_index(expires, 0); 504 } else if (delta < LVL_START(2)) { 505 idx = calc_index(expires, 1); 506 } else if (delta < LVL_START(3)) { 507 idx = calc_index(expires, 2); 508 } else if (delta < LVL_START(4)) { 509 idx = calc_index(expires, 3); 510 } else if (delta < LVL_START(5)) { 511 idx = calc_index(expires, 4); 512 } else if (delta < LVL_START(6)) { 513 idx = calc_index(expires, 5); 514 } else if (delta < LVL_START(7)) { 515 idx = calc_index(expires, 6); 516 } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { 517 idx = calc_index(expires, 7); 518 } else if ((long) delta < 0) { 519 idx = clk & LVL_MASK; 520 } else { 521 /* 522 * Force expire obscene large timeouts to expire at the 523 * capacity limit of the wheel. 524 */ 525 if (delta >= WHEEL_TIMEOUT_CUTOFF) 526 expires = clk + WHEEL_TIMEOUT_MAX; 527 528 idx = calc_index(expires, LVL_DEPTH - 1); 529 } 530 return idx; 531 } 532 533 /* 534 * Enqueue the timer into the hash bucket, mark it pending in 535 * the bitmap and store the index in the timer flags. 536 */ 537 static void enqueue_timer(struct timer_base *base, struct timer_list *timer, 538 unsigned int idx) 539 { 540 hlist_add_head(&timer->entry, base->vectors + idx); 541 __set_bit(idx, base->pending_map); 542 timer_set_idx(timer, idx); 543 544 trace_timer_start(timer, timer->expires, timer->flags); 545 } 546 547 static void 548 __internal_add_timer(struct timer_base *base, struct timer_list *timer) 549 { 550 unsigned int idx; 551 552 idx = calc_wheel_index(timer->expires, base->clk); 553 enqueue_timer(base, timer, idx); 554 } 555 556 static void 557 trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer) 558 { 559 if (!is_timers_nohz_active()) 560 return; 561 562 /* 563 * TODO: This wants some optimizing similar to the code below, but we 564 * will do that when we switch from push to pull for deferrable timers. 565 */ 566 if (timer->flags & TIMER_DEFERRABLE) { 567 if (tick_nohz_full_cpu(base->cpu)) 568 wake_up_nohz_cpu(base->cpu); 569 return; 570 } 571 572 /* 573 * We might have to IPI the remote CPU if the base is idle and the 574 * timer is not deferrable. If the other CPU is on the way to idle 575 * then it can't set base->is_idle as we hold the base lock: 576 */ 577 if (!base->is_idle) 578 return; 579 580 /* Check whether this is the new first expiring timer: */ 581 if (time_after_eq(timer->expires, base->next_expiry)) 582 return; 583 584 /* 585 * Set the next expiry time and kick the CPU so it can reevaluate the 586 * wheel: 587 */ 588 if (time_before(timer->expires, base->clk)) { 589 /* 590 * Prevent from forward_timer_base() moving the base->clk 591 * backward 592 */ 593 base->next_expiry = base->clk; 594 } else { 595 base->next_expiry = timer->expires; 596 } 597 wake_up_nohz_cpu(base->cpu); 598 } 599 600 static void 601 internal_add_timer(struct timer_base *base, struct timer_list *timer) 602 { 603 __internal_add_timer(base, timer); 604 trigger_dyntick_cpu(base, timer); 605 } 606 607 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS 608 609 static struct debug_obj_descr timer_debug_descr; 610 611 static void *timer_debug_hint(void *addr) 612 { 613 return ((struct timer_list *) addr)->function; 614 } 615 616 static bool timer_is_static_object(void *addr) 617 { 618 struct timer_list *timer = addr; 619 620 return (timer->entry.pprev == NULL && 621 timer->entry.next == TIMER_ENTRY_STATIC); 622 } 623 624 /* 625 * fixup_init is called when: 626 * - an active object is initialized 627 */ 628 static bool timer_fixup_init(void *addr, enum debug_obj_state state) 629 { 630 struct timer_list *timer = addr; 631 632 switch (state) { 633 case ODEBUG_STATE_ACTIVE: 634 del_timer_sync(timer); 635 debug_object_init(timer, &timer_debug_descr); 636 return true; 637 default: 638 return false; 639 } 640 } 641 642 /* Stub timer callback for improperly used timers. */ 643 static void stub_timer(struct timer_list *unused) 644 { 645 WARN_ON(1); 646 } 647 648 /* 649 * fixup_activate is called when: 650 * - an active object is activated 651 * - an unknown non-static object is activated 652 */ 653 static bool timer_fixup_activate(void *addr, enum debug_obj_state state) 654 { 655 struct timer_list *timer = addr; 656 657 switch (state) { 658 case ODEBUG_STATE_NOTAVAILABLE: 659 timer_setup(timer, stub_timer, 0); 660 return true; 661 662 case ODEBUG_STATE_ACTIVE: 663 WARN_ON(1); 664 /* fall through */ 665 default: 666 return false; 667 } 668 } 669 670 /* 671 * fixup_free is called when: 672 * - an active object is freed 673 */ 674 static bool timer_fixup_free(void *addr, enum debug_obj_state state) 675 { 676 struct timer_list *timer = addr; 677 678 switch (state) { 679 case ODEBUG_STATE_ACTIVE: 680 del_timer_sync(timer); 681 debug_object_free(timer, &timer_debug_descr); 682 return true; 683 default: 684 return false; 685 } 686 } 687 688 /* 689 * fixup_assert_init is called when: 690 * - an untracked/uninit-ed object is found 691 */ 692 static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) 693 { 694 struct timer_list *timer = addr; 695 696 switch (state) { 697 case ODEBUG_STATE_NOTAVAILABLE: 698 timer_setup(timer, stub_timer, 0); 699 return true; 700 default: 701 return false; 702 } 703 } 704 705 static struct debug_obj_descr timer_debug_descr = { 706 .name = "timer_list", 707 .debug_hint = timer_debug_hint, 708 .is_static_object = timer_is_static_object, 709 .fixup_init = timer_fixup_init, 710 .fixup_activate = timer_fixup_activate, 711 .fixup_free = timer_fixup_free, 712 .fixup_assert_init = timer_fixup_assert_init, 713 }; 714 715 static inline void debug_timer_init(struct timer_list *timer) 716 { 717 debug_object_init(timer, &timer_debug_descr); 718 } 719 720 static inline void debug_timer_activate(struct timer_list *timer) 721 { 722 debug_object_activate(timer, &timer_debug_descr); 723 } 724 725 static inline void debug_timer_deactivate(struct timer_list *timer) 726 { 727 debug_object_deactivate(timer, &timer_debug_descr); 728 } 729 730 static inline void debug_timer_free(struct timer_list *timer) 731 { 732 debug_object_free(timer, &timer_debug_descr); 733 } 734 735 static inline void debug_timer_assert_init(struct timer_list *timer) 736 { 737 debug_object_assert_init(timer, &timer_debug_descr); 738 } 739 740 static void do_init_timer(struct timer_list *timer, 741 void (*func)(struct timer_list *), 742 unsigned int flags, 743 const char *name, struct lock_class_key *key); 744 745 void init_timer_on_stack_key(struct timer_list *timer, 746 void (*func)(struct timer_list *), 747 unsigned int flags, 748 const char *name, struct lock_class_key *key) 749 { 750 debug_object_init_on_stack(timer, &timer_debug_descr); 751 do_init_timer(timer, func, flags, name, key); 752 } 753 EXPORT_SYMBOL_GPL(init_timer_on_stack_key); 754 755 void destroy_timer_on_stack(struct timer_list *timer) 756 { 757 debug_object_free(timer, &timer_debug_descr); 758 } 759 EXPORT_SYMBOL_GPL(destroy_timer_on_stack); 760 761 #else 762 static inline void debug_timer_init(struct timer_list *timer) { } 763 static inline void debug_timer_activate(struct timer_list *timer) { } 764 static inline void debug_timer_deactivate(struct timer_list *timer) { } 765 static inline void debug_timer_assert_init(struct timer_list *timer) { } 766 #endif 767 768 static inline void debug_init(struct timer_list *timer) 769 { 770 debug_timer_init(timer); 771 trace_timer_init(timer); 772 } 773 774 static inline void debug_deactivate(struct timer_list *timer) 775 { 776 debug_timer_deactivate(timer); 777 trace_timer_cancel(timer); 778 } 779 780 static inline void debug_assert_init(struct timer_list *timer) 781 { 782 debug_timer_assert_init(timer); 783 } 784 785 static void do_init_timer(struct timer_list *timer, 786 void (*func)(struct timer_list *), 787 unsigned int flags, 788 const char *name, struct lock_class_key *key) 789 { 790 timer->entry.pprev = NULL; 791 timer->function = func; 792 timer->flags = flags | raw_smp_processor_id(); 793 lockdep_init_map(&timer->lockdep_map, name, key, 0); 794 } 795 796 /** 797 * init_timer_key - initialize a timer 798 * @timer: the timer to be initialized 799 * @func: timer callback function 800 * @flags: timer flags 801 * @name: name of the timer 802 * @key: lockdep class key of the fake lock used for tracking timer 803 * sync lock dependencies 804 * 805 * init_timer_key() must be done to a timer prior calling *any* of the 806 * other timer functions. 807 */ 808 void init_timer_key(struct timer_list *timer, 809 void (*func)(struct timer_list *), unsigned int flags, 810 const char *name, struct lock_class_key *key) 811 { 812 debug_init(timer); 813 do_init_timer(timer, func, flags, name, key); 814 } 815 EXPORT_SYMBOL(init_timer_key); 816 817 static inline void detach_timer(struct timer_list *timer, bool clear_pending) 818 { 819 struct hlist_node *entry = &timer->entry; 820 821 debug_deactivate(timer); 822 823 __hlist_del(entry); 824 if (clear_pending) 825 entry->pprev = NULL; 826 entry->next = LIST_POISON2; 827 } 828 829 static int detach_if_pending(struct timer_list *timer, struct timer_base *base, 830 bool clear_pending) 831 { 832 unsigned idx = timer_get_idx(timer); 833 834 if (!timer_pending(timer)) 835 return 0; 836 837 if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) 838 __clear_bit(idx, base->pending_map); 839 840 detach_timer(timer, clear_pending); 841 return 1; 842 } 843 844 static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu) 845 { 846 struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu); 847 848 /* 849 * If the timer is deferrable and NO_HZ_COMMON is set then we need 850 * to use the deferrable base. 851 */ 852 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) 853 base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu); 854 return base; 855 } 856 857 static inline struct timer_base *get_timer_this_cpu_base(u32 tflags) 858 { 859 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 860 861 /* 862 * If the timer is deferrable and NO_HZ_COMMON is set then we need 863 * to use the deferrable base. 864 */ 865 if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) 866 base = this_cpu_ptr(&timer_bases[BASE_DEF]); 867 return base; 868 } 869 870 static inline struct timer_base *get_timer_base(u32 tflags) 871 { 872 return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK); 873 } 874 875 static inline struct timer_base * 876 get_target_base(struct timer_base *base, unsigned tflags) 877 { 878 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) 879 if (static_branch_likely(&timers_migration_enabled) && 880 !(tflags & TIMER_PINNED)) 881 return get_timer_cpu_base(tflags, get_nohz_timer_target()); 882 #endif 883 return get_timer_this_cpu_base(tflags); 884 } 885 886 static inline void forward_timer_base(struct timer_base *base) 887 { 888 #ifdef CONFIG_NO_HZ_COMMON 889 unsigned long jnow; 890 891 /* 892 * We only forward the base when we are idle or have just come out of 893 * idle (must_forward_clk logic), and have a delta between base clock 894 * and jiffies. In the common case, run_timers will take care of it. 895 */ 896 if (likely(!base->must_forward_clk)) 897 return; 898 899 jnow = READ_ONCE(jiffies); 900 base->must_forward_clk = base->is_idle; 901 if ((long)(jnow - base->clk) < 2) 902 return; 903 904 /* 905 * If the next expiry value is > jiffies, then we fast forward to 906 * jiffies otherwise we forward to the next expiry value. 907 */ 908 if (time_after(base->next_expiry, jnow)) { 909 base->clk = jnow; 910 } else { 911 if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk))) 912 return; 913 base->clk = base->next_expiry; 914 } 915 #endif 916 } 917 918 919 /* 920 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means 921 * that all timers which are tied to this base are locked, and the base itself 922 * is locked too. 923 * 924 * So __run_timers/migrate_timers can safely modify all timers which could 925 * be found in the base->vectors array. 926 * 927 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need 928 * to wait until the migration is done. 929 */ 930 static struct timer_base *lock_timer_base(struct timer_list *timer, 931 unsigned long *flags) 932 __acquires(timer->base->lock) 933 { 934 for (;;) { 935 struct timer_base *base; 936 u32 tf; 937 938 /* 939 * We need to use READ_ONCE() here, otherwise the compiler 940 * might re-read @tf between the check for TIMER_MIGRATING 941 * and spin_lock(). 942 */ 943 tf = READ_ONCE(timer->flags); 944 945 if (!(tf & TIMER_MIGRATING)) { 946 base = get_timer_base(tf); 947 raw_spin_lock_irqsave(&base->lock, *flags); 948 if (timer->flags == tf) 949 return base; 950 raw_spin_unlock_irqrestore(&base->lock, *flags); 951 } 952 cpu_relax(); 953 } 954 } 955 956 #define MOD_TIMER_PENDING_ONLY 0x01 957 #define MOD_TIMER_REDUCE 0x02 958 #define MOD_TIMER_NOTPENDING 0x04 959 960 static inline int 961 __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options) 962 { 963 struct timer_base *base, *new_base; 964 unsigned int idx = UINT_MAX; 965 unsigned long clk = 0, flags; 966 int ret = 0; 967 968 BUG_ON(!timer->function); 969 970 /* 971 * This is a common optimization triggered by the networking code - if 972 * the timer is re-modified to have the same timeout or ends up in the 973 * same array bucket then just return: 974 */ 975 if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) { 976 /* 977 * The downside of this optimization is that it can result in 978 * larger granularity than you would get from adding a new 979 * timer with this expiry. 980 */ 981 long diff = timer->expires - expires; 982 983 if (!diff) 984 return 1; 985 if (options & MOD_TIMER_REDUCE && diff <= 0) 986 return 1; 987 988 /* 989 * We lock timer base and calculate the bucket index right 990 * here. If the timer ends up in the same bucket, then we 991 * just update the expiry time and avoid the whole 992 * dequeue/enqueue dance. 993 */ 994 base = lock_timer_base(timer, &flags); 995 forward_timer_base(base); 996 997 if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) && 998 time_before_eq(timer->expires, expires)) { 999 ret = 1; 1000 goto out_unlock; 1001 } 1002 1003 clk = base->clk; 1004 idx = calc_wheel_index(expires, clk); 1005 1006 /* 1007 * Retrieve and compare the array index of the pending 1008 * timer. If it matches set the expiry to the new value so a 1009 * subsequent call will exit in the expires check above. 1010 */ 1011 if (idx == timer_get_idx(timer)) { 1012 if (!(options & MOD_TIMER_REDUCE)) 1013 timer->expires = expires; 1014 else if (time_after(timer->expires, expires)) 1015 timer->expires = expires; 1016 ret = 1; 1017 goto out_unlock; 1018 } 1019 } else { 1020 base = lock_timer_base(timer, &flags); 1021 forward_timer_base(base); 1022 } 1023 1024 ret = detach_if_pending(timer, base, false); 1025 if (!ret && (options & MOD_TIMER_PENDING_ONLY)) 1026 goto out_unlock; 1027 1028 new_base = get_target_base(base, timer->flags); 1029 1030 if (base != new_base) { 1031 /* 1032 * We are trying to schedule the timer on the new base. 1033 * However we can't change timer's base while it is running, 1034 * otherwise del_timer_sync() can't detect that the timer's 1035 * handler yet has not finished. This also guarantees that the 1036 * timer is serialized wrt itself. 1037 */ 1038 if (likely(base->running_timer != timer)) { 1039 /* See the comment in lock_timer_base() */ 1040 timer->flags |= TIMER_MIGRATING; 1041 1042 raw_spin_unlock(&base->lock); 1043 base = new_base; 1044 raw_spin_lock(&base->lock); 1045 WRITE_ONCE(timer->flags, 1046 (timer->flags & ~TIMER_BASEMASK) | base->cpu); 1047 forward_timer_base(base); 1048 } 1049 } 1050 1051 debug_timer_activate(timer); 1052 1053 timer->expires = expires; 1054 /* 1055 * If 'idx' was calculated above and the base time did not advance 1056 * between calculating 'idx' and possibly switching the base, only 1057 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise 1058 * we need to (re)calculate the wheel index via 1059 * internal_add_timer(). 1060 */ 1061 if (idx != UINT_MAX && clk == base->clk) { 1062 enqueue_timer(base, timer, idx); 1063 trigger_dyntick_cpu(base, timer); 1064 } else { 1065 internal_add_timer(base, timer); 1066 } 1067 1068 out_unlock: 1069 raw_spin_unlock_irqrestore(&base->lock, flags); 1070 1071 return ret; 1072 } 1073 1074 /** 1075 * mod_timer_pending - modify a pending timer's timeout 1076 * @timer: the pending timer to be modified 1077 * @expires: new timeout in jiffies 1078 * 1079 * mod_timer_pending() is the same for pending timers as mod_timer(), 1080 * but will not re-activate and modify already deleted timers. 1081 * 1082 * It is useful for unserialized use of timers. 1083 */ 1084 int mod_timer_pending(struct timer_list *timer, unsigned long expires) 1085 { 1086 return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY); 1087 } 1088 EXPORT_SYMBOL(mod_timer_pending); 1089 1090 /** 1091 * mod_timer - modify a timer's timeout 1092 * @timer: the timer to be modified 1093 * @expires: new timeout in jiffies 1094 * 1095 * mod_timer() is a more efficient way to update the expire field of an 1096 * active timer (if the timer is inactive it will be activated) 1097 * 1098 * mod_timer(timer, expires) is equivalent to: 1099 * 1100 * del_timer(timer); timer->expires = expires; add_timer(timer); 1101 * 1102 * Note that if there are multiple unserialized concurrent users of the 1103 * same timer, then mod_timer() is the only safe way to modify the timeout, 1104 * since add_timer() cannot modify an already running timer. 1105 * 1106 * The function returns whether it has modified a pending timer or not. 1107 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an 1108 * active timer returns 1.) 1109 */ 1110 int mod_timer(struct timer_list *timer, unsigned long expires) 1111 { 1112 return __mod_timer(timer, expires, 0); 1113 } 1114 EXPORT_SYMBOL(mod_timer); 1115 1116 /** 1117 * timer_reduce - Modify a timer's timeout if it would reduce the timeout 1118 * @timer: The timer to be modified 1119 * @expires: New timeout in jiffies 1120 * 1121 * timer_reduce() is very similar to mod_timer(), except that it will only 1122 * modify a running timer if that would reduce the expiration time (it will 1123 * start a timer that isn't running). 1124 */ 1125 int timer_reduce(struct timer_list *timer, unsigned long expires) 1126 { 1127 return __mod_timer(timer, expires, MOD_TIMER_REDUCE); 1128 } 1129 EXPORT_SYMBOL(timer_reduce); 1130 1131 /** 1132 * add_timer - start a timer 1133 * @timer: the timer to be added 1134 * 1135 * The kernel will do a ->function(@timer) callback from the 1136 * timer interrupt at the ->expires point in the future. The 1137 * current time is 'jiffies'. 1138 * 1139 * The timer's ->expires, ->function fields must be set prior calling this 1140 * function. 1141 * 1142 * Timers with an ->expires field in the past will be executed in the next 1143 * timer tick. 1144 */ 1145 void add_timer(struct timer_list *timer) 1146 { 1147 BUG_ON(timer_pending(timer)); 1148 __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING); 1149 } 1150 EXPORT_SYMBOL(add_timer); 1151 1152 /** 1153 * add_timer_on - start a timer on a particular CPU 1154 * @timer: the timer to be added 1155 * @cpu: the CPU to start it on 1156 * 1157 * This is not very scalable on SMP. Double adds are not possible. 1158 */ 1159 void add_timer_on(struct timer_list *timer, int cpu) 1160 { 1161 struct timer_base *new_base, *base; 1162 unsigned long flags; 1163 1164 BUG_ON(timer_pending(timer) || !timer->function); 1165 1166 new_base = get_timer_cpu_base(timer->flags, cpu); 1167 1168 /* 1169 * If @timer was on a different CPU, it should be migrated with the 1170 * old base locked to prevent other operations proceeding with the 1171 * wrong base locked. See lock_timer_base(). 1172 */ 1173 base = lock_timer_base(timer, &flags); 1174 if (base != new_base) { 1175 timer->flags |= TIMER_MIGRATING; 1176 1177 raw_spin_unlock(&base->lock); 1178 base = new_base; 1179 raw_spin_lock(&base->lock); 1180 WRITE_ONCE(timer->flags, 1181 (timer->flags & ~TIMER_BASEMASK) | cpu); 1182 } 1183 forward_timer_base(base); 1184 1185 debug_timer_activate(timer); 1186 internal_add_timer(base, timer); 1187 raw_spin_unlock_irqrestore(&base->lock, flags); 1188 } 1189 EXPORT_SYMBOL_GPL(add_timer_on); 1190 1191 /** 1192 * del_timer - deactivate a timer. 1193 * @timer: the timer to be deactivated 1194 * 1195 * del_timer() deactivates a timer - this works on both active and inactive 1196 * timers. 1197 * 1198 * The function returns whether it has deactivated a pending timer or not. 1199 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an 1200 * active timer returns 1.) 1201 */ 1202 int del_timer(struct timer_list *timer) 1203 { 1204 struct timer_base *base; 1205 unsigned long flags; 1206 int ret = 0; 1207 1208 debug_assert_init(timer); 1209 1210 if (timer_pending(timer)) { 1211 base = lock_timer_base(timer, &flags); 1212 ret = detach_if_pending(timer, base, true); 1213 raw_spin_unlock_irqrestore(&base->lock, flags); 1214 } 1215 1216 return ret; 1217 } 1218 EXPORT_SYMBOL(del_timer); 1219 1220 /** 1221 * try_to_del_timer_sync - Try to deactivate a timer 1222 * @timer: timer to delete 1223 * 1224 * This function tries to deactivate a timer. Upon successful (ret >= 0) 1225 * exit the timer is not queued and the handler is not running on any CPU. 1226 */ 1227 int try_to_del_timer_sync(struct timer_list *timer) 1228 { 1229 struct timer_base *base; 1230 unsigned long flags; 1231 int ret = -1; 1232 1233 debug_assert_init(timer); 1234 1235 base = lock_timer_base(timer, &flags); 1236 1237 if (base->running_timer != timer) 1238 ret = detach_if_pending(timer, base, true); 1239 1240 raw_spin_unlock_irqrestore(&base->lock, flags); 1241 1242 return ret; 1243 } 1244 EXPORT_SYMBOL(try_to_del_timer_sync); 1245 1246 #ifdef CONFIG_PREEMPT_RT 1247 static __init void timer_base_init_expiry_lock(struct timer_base *base) 1248 { 1249 spin_lock_init(&base->expiry_lock); 1250 } 1251 1252 static inline void timer_base_lock_expiry(struct timer_base *base) 1253 { 1254 spin_lock(&base->expiry_lock); 1255 } 1256 1257 static inline void timer_base_unlock_expiry(struct timer_base *base) 1258 { 1259 spin_unlock(&base->expiry_lock); 1260 } 1261 1262 /* 1263 * The counterpart to del_timer_wait_running(). 1264 * 1265 * If there is a waiter for base->expiry_lock, then it was waiting for the 1266 * timer callback to finish. Drop expiry_lock and reaquire it. That allows 1267 * the waiter to acquire the lock and make progress. 1268 */ 1269 static void timer_sync_wait_running(struct timer_base *base) 1270 { 1271 if (atomic_read(&base->timer_waiters)) { 1272 spin_unlock(&base->expiry_lock); 1273 spin_lock(&base->expiry_lock); 1274 } 1275 } 1276 1277 /* 1278 * This function is called on PREEMPT_RT kernels when the fast path 1279 * deletion of a timer failed because the timer callback function was 1280 * running. 1281 * 1282 * This prevents priority inversion, if the softirq thread on a remote CPU 1283 * got preempted, and it prevents a life lock when the task which tries to 1284 * delete a timer preempted the softirq thread running the timer callback 1285 * function. 1286 */ 1287 static void del_timer_wait_running(struct timer_list *timer) 1288 { 1289 u32 tf; 1290 1291 tf = READ_ONCE(timer->flags); 1292 if (!(tf & TIMER_MIGRATING)) { 1293 struct timer_base *base = get_timer_base(tf); 1294 1295 /* 1296 * Mark the base as contended and grab the expiry lock, 1297 * which is held by the softirq across the timer 1298 * callback. Drop the lock immediately so the softirq can 1299 * expire the next timer. In theory the timer could already 1300 * be running again, but that's more than unlikely and just 1301 * causes another wait loop. 1302 */ 1303 atomic_inc(&base->timer_waiters); 1304 spin_lock_bh(&base->expiry_lock); 1305 atomic_dec(&base->timer_waiters); 1306 spin_unlock_bh(&base->expiry_lock); 1307 } 1308 } 1309 #else 1310 static inline void timer_base_init_expiry_lock(struct timer_base *base) { } 1311 static inline void timer_base_lock_expiry(struct timer_base *base) { } 1312 static inline void timer_base_unlock_expiry(struct timer_base *base) { } 1313 static inline void timer_sync_wait_running(struct timer_base *base) { } 1314 static inline void del_timer_wait_running(struct timer_list *timer) { } 1315 #endif 1316 1317 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT) 1318 /** 1319 * del_timer_sync - deactivate a timer and wait for the handler to finish. 1320 * @timer: the timer to be deactivated 1321 * 1322 * This function only differs from del_timer() on SMP: besides deactivating 1323 * the timer it also makes sure the handler has finished executing on other 1324 * CPUs. 1325 * 1326 * Synchronization rules: Callers must prevent restarting of the timer, 1327 * otherwise this function is meaningless. It must not be called from 1328 * interrupt contexts unless the timer is an irqsafe one. The caller must 1329 * not hold locks which would prevent completion of the timer's 1330 * handler. The timer's handler must not call add_timer_on(). Upon exit the 1331 * timer is not queued and the handler is not running on any CPU. 1332 * 1333 * Note: For !irqsafe timers, you must not hold locks that are held in 1334 * interrupt context while calling this function. Even if the lock has 1335 * nothing to do with the timer in question. Here's why:: 1336 * 1337 * CPU0 CPU1 1338 * ---- ---- 1339 * <SOFTIRQ> 1340 * call_timer_fn(); 1341 * base->running_timer = mytimer; 1342 * spin_lock_irq(somelock); 1343 * <IRQ> 1344 * spin_lock(somelock); 1345 * del_timer_sync(mytimer); 1346 * while (base->running_timer == mytimer); 1347 * 1348 * Now del_timer_sync() will never return and never release somelock. 1349 * The interrupt on the other CPU is waiting to grab somelock but 1350 * it has interrupted the softirq that CPU0 is waiting to finish. 1351 * 1352 * The function returns whether it has deactivated a pending timer or not. 1353 */ 1354 int del_timer_sync(struct timer_list *timer) 1355 { 1356 int ret; 1357 1358 #ifdef CONFIG_LOCKDEP 1359 unsigned long flags; 1360 1361 /* 1362 * If lockdep gives a backtrace here, please reference 1363 * the synchronization rules above. 1364 */ 1365 local_irq_save(flags); 1366 lock_map_acquire(&timer->lockdep_map); 1367 lock_map_release(&timer->lockdep_map); 1368 local_irq_restore(flags); 1369 #endif 1370 /* 1371 * don't use it in hardirq context, because it 1372 * could lead to deadlock. 1373 */ 1374 WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); 1375 1376 do { 1377 ret = try_to_del_timer_sync(timer); 1378 1379 if (unlikely(ret < 0)) { 1380 del_timer_wait_running(timer); 1381 cpu_relax(); 1382 } 1383 } while (ret < 0); 1384 1385 return ret; 1386 } 1387 EXPORT_SYMBOL(del_timer_sync); 1388 #endif 1389 1390 static void call_timer_fn(struct timer_list *timer, 1391 void (*fn)(struct timer_list *), 1392 unsigned long baseclk) 1393 { 1394 int count = preempt_count(); 1395 1396 #ifdef CONFIG_LOCKDEP 1397 /* 1398 * It is permissible to free the timer from inside the 1399 * function that is called from it, this we need to take into 1400 * account for lockdep too. To avoid bogus "held lock freed" 1401 * warnings as well as problems when looking into 1402 * timer->lockdep_map, make a copy and use that here. 1403 */ 1404 struct lockdep_map lockdep_map; 1405 1406 lockdep_copy_map(&lockdep_map, &timer->lockdep_map); 1407 #endif 1408 /* 1409 * Couple the lock chain with the lock chain at 1410 * del_timer_sync() by acquiring the lock_map around the fn() 1411 * call here and in del_timer_sync(). 1412 */ 1413 lock_map_acquire(&lockdep_map); 1414 1415 trace_timer_expire_entry(timer, baseclk); 1416 fn(timer); 1417 trace_timer_expire_exit(timer); 1418 1419 lock_map_release(&lockdep_map); 1420 1421 if (count != preempt_count()) { 1422 WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n", 1423 fn, count, preempt_count()); 1424 /* 1425 * Restore the preempt count. That gives us a decent 1426 * chance to survive and extract information. If the 1427 * callback kept a lock held, bad luck, but not worse 1428 * than the BUG() we had. 1429 */ 1430 preempt_count_set(count); 1431 } 1432 } 1433 1434 static void expire_timers(struct timer_base *base, struct hlist_head *head) 1435 { 1436 /* 1437 * This value is required only for tracing. base->clk was 1438 * incremented directly before expire_timers was called. But expiry 1439 * is related to the old base->clk value. 1440 */ 1441 unsigned long baseclk = base->clk - 1; 1442 1443 while (!hlist_empty(head)) { 1444 struct timer_list *timer; 1445 void (*fn)(struct timer_list *); 1446 1447 timer = hlist_entry(head->first, struct timer_list, entry); 1448 1449 base->running_timer = timer; 1450 detach_timer(timer, true); 1451 1452 fn = timer->function; 1453 1454 if (timer->flags & TIMER_IRQSAFE) { 1455 raw_spin_unlock(&base->lock); 1456 call_timer_fn(timer, fn, baseclk); 1457 base->running_timer = NULL; 1458 raw_spin_lock(&base->lock); 1459 } else { 1460 raw_spin_unlock_irq(&base->lock); 1461 call_timer_fn(timer, fn, baseclk); 1462 base->running_timer = NULL; 1463 timer_sync_wait_running(base); 1464 raw_spin_lock_irq(&base->lock); 1465 } 1466 } 1467 } 1468 1469 static int __collect_expired_timers(struct timer_base *base, 1470 struct hlist_head *heads) 1471 { 1472 unsigned long clk = base->clk; 1473 struct hlist_head *vec; 1474 int i, levels = 0; 1475 unsigned int idx; 1476 1477 for (i = 0; i < LVL_DEPTH; i++) { 1478 idx = (clk & LVL_MASK) + i * LVL_SIZE; 1479 1480 if (__test_and_clear_bit(idx, base->pending_map)) { 1481 vec = base->vectors + idx; 1482 hlist_move_list(vec, heads++); 1483 levels++; 1484 } 1485 /* Is it time to look at the next level? */ 1486 if (clk & LVL_CLK_MASK) 1487 break; 1488 /* Shift clock for the next level granularity */ 1489 clk >>= LVL_CLK_SHIFT; 1490 } 1491 return levels; 1492 } 1493 1494 #ifdef CONFIG_NO_HZ_COMMON 1495 /* 1496 * Find the next pending bucket of a level. Search from level start (@offset) 1497 * + @clk upwards and if nothing there, search from start of the level 1498 * (@offset) up to @offset + clk. 1499 */ 1500 static int next_pending_bucket(struct timer_base *base, unsigned offset, 1501 unsigned clk) 1502 { 1503 unsigned pos, start = offset + clk; 1504 unsigned end = offset + LVL_SIZE; 1505 1506 pos = find_next_bit(base->pending_map, end, start); 1507 if (pos < end) 1508 return pos - start; 1509 1510 pos = find_next_bit(base->pending_map, start, offset); 1511 return pos < start ? pos + LVL_SIZE - start : -1; 1512 } 1513 1514 /* 1515 * Search the first expiring timer in the various clock levels. Caller must 1516 * hold base->lock. 1517 */ 1518 static unsigned long __next_timer_interrupt(struct timer_base *base) 1519 { 1520 unsigned long clk, next, adj; 1521 unsigned lvl, offset = 0; 1522 1523 next = base->clk + NEXT_TIMER_MAX_DELTA; 1524 clk = base->clk; 1525 for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) { 1526 int pos = next_pending_bucket(base, offset, clk & LVL_MASK); 1527 1528 if (pos >= 0) { 1529 unsigned long tmp = clk + (unsigned long) pos; 1530 1531 tmp <<= LVL_SHIFT(lvl); 1532 if (time_before(tmp, next)) 1533 next = tmp; 1534 } 1535 /* 1536 * Clock for the next level. If the current level clock lower 1537 * bits are zero, we look at the next level as is. If not we 1538 * need to advance it by one because that's going to be the 1539 * next expiring bucket in that level. base->clk is the next 1540 * expiring jiffie. So in case of: 1541 * 1542 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1543 * 0 0 0 0 0 0 1544 * 1545 * we have to look at all levels @index 0. With 1546 * 1547 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1548 * 0 0 0 0 0 2 1549 * 1550 * LVL0 has the next expiring bucket @index 2. The upper 1551 * levels have the next expiring bucket @index 1. 1552 * 1553 * In case that the propagation wraps the next level the same 1554 * rules apply: 1555 * 1556 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 1557 * 0 0 0 0 F 2 1558 * 1559 * So after looking at LVL0 we get: 1560 * 1561 * LVL5 LVL4 LVL3 LVL2 LVL1 1562 * 0 0 0 1 0 1563 * 1564 * So no propagation from LVL1 to LVL2 because that happened 1565 * with the add already, but then we need to propagate further 1566 * from LVL2 to LVL3. 1567 * 1568 * So the simple check whether the lower bits of the current 1569 * level are 0 or not is sufficient for all cases. 1570 */ 1571 adj = clk & LVL_CLK_MASK ? 1 : 0; 1572 clk >>= LVL_CLK_SHIFT; 1573 clk += adj; 1574 } 1575 return next; 1576 } 1577 1578 /* 1579 * Check, if the next hrtimer event is before the next timer wheel 1580 * event: 1581 */ 1582 static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) 1583 { 1584 u64 nextevt = hrtimer_get_next_event(); 1585 1586 /* 1587 * If high resolution timers are enabled 1588 * hrtimer_get_next_event() returns KTIME_MAX. 1589 */ 1590 if (expires <= nextevt) 1591 return expires; 1592 1593 /* 1594 * If the next timer is already expired, return the tick base 1595 * time so the tick is fired immediately. 1596 */ 1597 if (nextevt <= basem) 1598 return basem; 1599 1600 /* 1601 * Round up to the next jiffie. High resolution timers are 1602 * off, so the hrtimers are expired in the tick and we need to 1603 * make sure that this tick really expires the timer to avoid 1604 * a ping pong of the nohz stop code. 1605 * 1606 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 1607 */ 1608 return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; 1609 } 1610 1611 /** 1612 * get_next_timer_interrupt - return the time (clock mono) of the next timer 1613 * @basej: base time jiffies 1614 * @basem: base time clock monotonic 1615 * 1616 * Returns the tick aligned clock monotonic time of the next pending 1617 * timer or KTIME_MAX if no timer is pending. 1618 */ 1619 u64 get_next_timer_interrupt(unsigned long basej, u64 basem) 1620 { 1621 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1622 u64 expires = KTIME_MAX; 1623 unsigned long nextevt; 1624 bool is_max_delta; 1625 1626 /* 1627 * Pretend that there is no timer pending if the cpu is offline. 1628 * Possible pending timers will be migrated later to an active cpu. 1629 */ 1630 if (cpu_is_offline(smp_processor_id())) 1631 return expires; 1632 1633 raw_spin_lock(&base->lock); 1634 nextevt = __next_timer_interrupt(base); 1635 is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA); 1636 base->next_expiry = nextevt; 1637 /* 1638 * We have a fresh next event. Check whether we can forward the 1639 * base. We can only do that when @basej is past base->clk 1640 * otherwise we might rewind base->clk. 1641 */ 1642 if (time_after(basej, base->clk)) { 1643 if (time_after(nextevt, basej)) 1644 base->clk = basej; 1645 else if (time_after(nextevt, base->clk)) 1646 base->clk = nextevt; 1647 } 1648 1649 if (time_before_eq(nextevt, basej)) { 1650 expires = basem; 1651 base->is_idle = false; 1652 } else { 1653 if (!is_max_delta) 1654 expires = basem + (u64)(nextevt - basej) * TICK_NSEC; 1655 /* 1656 * If we expect to sleep more than a tick, mark the base idle. 1657 * Also the tick is stopped so any added timer must forward 1658 * the base clk itself to keep granularity small. This idle 1659 * logic is only maintained for the BASE_STD base, deferrable 1660 * timers may still see large granularity skew (by design). 1661 */ 1662 if ((expires - basem) > TICK_NSEC) { 1663 base->must_forward_clk = true; 1664 base->is_idle = true; 1665 } 1666 } 1667 raw_spin_unlock(&base->lock); 1668 1669 return cmp_next_hrtimer_event(basem, expires); 1670 } 1671 1672 /** 1673 * timer_clear_idle - Clear the idle state of the timer base 1674 * 1675 * Called with interrupts disabled 1676 */ 1677 void timer_clear_idle(void) 1678 { 1679 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1680 1681 /* 1682 * We do this unlocked. The worst outcome is a remote enqueue sending 1683 * a pointless IPI, but taking the lock would just make the window for 1684 * sending the IPI a few instructions smaller for the cost of taking 1685 * the lock in the exit from idle path. 1686 */ 1687 base->is_idle = false; 1688 } 1689 1690 static int collect_expired_timers(struct timer_base *base, 1691 struct hlist_head *heads) 1692 { 1693 unsigned long now = READ_ONCE(jiffies); 1694 1695 /* 1696 * NOHZ optimization. After a long idle sleep we need to forward the 1697 * base to current jiffies. Avoid a loop by searching the bitfield for 1698 * the next expiring timer. 1699 */ 1700 if ((long)(now - base->clk) > 2) { 1701 unsigned long next = __next_timer_interrupt(base); 1702 1703 /* 1704 * If the next timer is ahead of time forward to current 1705 * jiffies, otherwise forward to the next expiry time: 1706 */ 1707 if (time_after(next, now)) { 1708 /* 1709 * The call site will increment base->clk and then 1710 * terminate the expiry loop immediately. 1711 */ 1712 base->clk = now; 1713 return 0; 1714 } 1715 base->clk = next; 1716 } 1717 return __collect_expired_timers(base, heads); 1718 } 1719 #else 1720 static inline int collect_expired_timers(struct timer_base *base, 1721 struct hlist_head *heads) 1722 { 1723 return __collect_expired_timers(base, heads); 1724 } 1725 #endif 1726 1727 /* 1728 * Called from the timer interrupt handler to charge one tick to the current 1729 * process. user_tick is 1 if the tick is user time, 0 for system. 1730 */ 1731 void update_process_times(int user_tick) 1732 { 1733 struct task_struct *p = current; 1734 1735 /* Note: this timer irq context must be accounted for as well. */ 1736 account_process_tick(p, user_tick); 1737 run_local_timers(); 1738 rcu_sched_clock_irq(user_tick); 1739 #ifdef CONFIG_IRQ_WORK 1740 if (in_irq()) 1741 irq_work_tick(); 1742 #endif 1743 scheduler_tick(); 1744 if (IS_ENABLED(CONFIG_POSIX_TIMERS)) 1745 run_posix_cpu_timers(); 1746 1747 /* The current CPU might make use of net randoms without receiving IRQs 1748 * to renew them often enough. Let's update the net_rand_state from a 1749 * non-constant value that's not affine to the number of calls to make 1750 * sure it's updated when there's some activity (we don't care in idle). 1751 */ 1752 this_cpu_add(net_rand_state.s1, rol32(jiffies, 24) + user_tick); 1753 } 1754 1755 /** 1756 * __run_timers - run all expired timers (if any) on this CPU. 1757 * @base: the timer vector to be processed. 1758 */ 1759 static inline void __run_timers(struct timer_base *base) 1760 { 1761 struct hlist_head heads[LVL_DEPTH]; 1762 int levels; 1763 1764 if (!time_after_eq(jiffies, base->clk)) 1765 return; 1766 1767 timer_base_lock_expiry(base); 1768 raw_spin_lock_irq(&base->lock); 1769 1770 /* 1771 * timer_base::must_forward_clk must be cleared before running 1772 * timers so that any timer functions that call mod_timer() will 1773 * not try to forward the base. Idle tracking / clock forwarding 1774 * logic is only used with BASE_STD timers. 1775 * 1776 * The must_forward_clk flag is cleared unconditionally also for 1777 * the deferrable base. The deferrable base is not affected by idle 1778 * tracking and never forwarded, so clearing the flag is a NOOP. 1779 * 1780 * The fact that the deferrable base is never forwarded can cause 1781 * large variations in granularity for deferrable timers, but they 1782 * can be deferred for long periods due to idle anyway. 1783 */ 1784 base->must_forward_clk = false; 1785 1786 while (time_after_eq(jiffies, base->clk)) { 1787 1788 levels = collect_expired_timers(base, heads); 1789 base->clk++; 1790 1791 while (levels--) 1792 expire_timers(base, heads + levels); 1793 } 1794 raw_spin_unlock_irq(&base->lock); 1795 timer_base_unlock_expiry(base); 1796 } 1797 1798 /* 1799 * This function runs timers and the timer-tq in bottom half context. 1800 */ 1801 static __latent_entropy void run_timer_softirq(struct softirq_action *h) 1802 { 1803 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1804 1805 __run_timers(base); 1806 if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) 1807 __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); 1808 } 1809 1810 /* 1811 * Called by the local, per-CPU timer interrupt on SMP. 1812 */ 1813 void run_local_timers(void) 1814 { 1815 struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); 1816 1817 hrtimer_run_queues(); 1818 /* Raise the softirq only if required. */ 1819 if (time_before(jiffies, base->clk)) { 1820 if (!IS_ENABLED(CONFIG_NO_HZ_COMMON)) 1821 return; 1822 /* CPU is awake, so check the deferrable base. */ 1823 base++; 1824 if (time_before(jiffies, base->clk)) 1825 return; 1826 } 1827 raise_softirq(TIMER_SOFTIRQ); 1828 } 1829 1830 /* 1831 * Since schedule_timeout()'s timer is defined on the stack, it must store 1832 * the target task on the stack as well. 1833 */ 1834 struct process_timer { 1835 struct timer_list timer; 1836 struct task_struct *task; 1837 }; 1838 1839 static void process_timeout(struct timer_list *t) 1840 { 1841 struct process_timer *timeout = from_timer(timeout, t, timer); 1842 1843 wake_up_process(timeout->task); 1844 } 1845 1846 /** 1847 * schedule_timeout - sleep until timeout 1848 * @timeout: timeout value in jiffies 1849 * 1850 * Make the current task sleep until @timeout jiffies have elapsed. 1851 * The function behavior depends on the current task state 1852 * (see also set_current_state() description): 1853 * 1854 * %TASK_RUNNING - the scheduler is called, but the task does not sleep 1855 * at all. That happens because sched_submit_work() does nothing for 1856 * tasks in %TASK_RUNNING state. 1857 * 1858 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to 1859 * pass before the routine returns unless the current task is explicitly 1860 * woken up, (e.g. by wake_up_process()). 1861 * 1862 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is 1863 * delivered to the current task or the current task is explicitly woken 1864 * up. 1865 * 1866 * The current task state is guaranteed to be %TASK_RUNNING when this 1867 * routine returns. 1868 * 1869 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule 1870 * the CPU away without a bound on the timeout. In this case the return 1871 * value will be %MAX_SCHEDULE_TIMEOUT. 1872 * 1873 * Returns 0 when the timer has expired otherwise the remaining time in 1874 * jiffies will be returned. In all cases the return value is guaranteed 1875 * to be non-negative. 1876 */ 1877 signed long __sched schedule_timeout(signed long timeout) 1878 { 1879 struct process_timer timer; 1880 unsigned long expire; 1881 1882 switch (timeout) 1883 { 1884 case MAX_SCHEDULE_TIMEOUT: 1885 /* 1886 * These two special cases are useful to be comfortable 1887 * in the caller. Nothing more. We could take 1888 * MAX_SCHEDULE_TIMEOUT from one of the negative value 1889 * but I' d like to return a valid offset (>=0) to allow 1890 * the caller to do everything it want with the retval. 1891 */ 1892 schedule(); 1893 goto out; 1894 default: 1895 /* 1896 * Another bit of PARANOID. Note that the retval will be 1897 * 0 since no piece of kernel is supposed to do a check 1898 * for a negative retval of schedule_timeout() (since it 1899 * should never happens anyway). You just have the printk() 1900 * that will tell you if something is gone wrong and where. 1901 */ 1902 if (timeout < 0) { 1903 printk(KERN_ERR "schedule_timeout: wrong timeout " 1904 "value %lx\n", timeout); 1905 dump_stack(); 1906 current->state = TASK_RUNNING; 1907 goto out; 1908 } 1909 } 1910 1911 expire = timeout + jiffies; 1912 1913 timer.task = current; 1914 timer_setup_on_stack(&timer.timer, process_timeout, 0); 1915 __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING); 1916 schedule(); 1917 del_singleshot_timer_sync(&timer.timer); 1918 1919 /* Remove the timer from the object tracker */ 1920 destroy_timer_on_stack(&timer.timer); 1921 1922 timeout = expire - jiffies; 1923 1924 out: 1925 return timeout < 0 ? 0 : timeout; 1926 } 1927 EXPORT_SYMBOL(schedule_timeout); 1928 1929 /* 1930 * We can use __set_current_state() here because schedule_timeout() calls 1931 * schedule() unconditionally. 1932 */ 1933 signed long __sched schedule_timeout_interruptible(signed long timeout) 1934 { 1935 __set_current_state(TASK_INTERRUPTIBLE); 1936 return schedule_timeout(timeout); 1937 } 1938 EXPORT_SYMBOL(schedule_timeout_interruptible); 1939 1940 signed long __sched schedule_timeout_killable(signed long timeout) 1941 { 1942 __set_current_state(TASK_KILLABLE); 1943 return schedule_timeout(timeout); 1944 } 1945 EXPORT_SYMBOL(schedule_timeout_killable); 1946 1947 signed long __sched schedule_timeout_uninterruptible(signed long timeout) 1948 { 1949 __set_current_state(TASK_UNINTERRUPTIBLE); 1950 return schedule_timeout(timeout); 1951 } 1952 EXPORT_SYMBOL(schedule_timeout_uninterruptible); 1953 1954 /* 1955 * Like schedule_timeout_uninterruptible(), except this task will not contribute 1956 * to load average. 1957 */ 1958 signed long __sched schedule_timeout_idle(signed long timeout) 1959 { 1960 __set_current_state(TASK_IDLE); 1961 return schedule_timeout(timeout); 1962 } 1963 EXPORT_SYMBOL(schedule_timeout_idle); 1964 1965 #ifdef CONFIG_HOTPLUG_CPU 1966 static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head) 1967 { 1968 struct timer_list *timer; 1969 int cpu = new_base->cpu; 1970 1971 while (!hlist_empty(head)) { 1972 timer = hlist_entry(head->first, struct timer_list, entry); 1973 detach_timer(timer, false); 1974 timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; 1975 internal_add_timer(new_base, timer); 1976 } 1977 } 1978 1979 int timers_prepare_cpu(unsigned int cpu) 1980 { 1981 struct timer_base *base; 1982 int b; 1983 1984 for (b = 0; b < NR_BASES; b++) { 1985 base = per_cpu_ptr(&timer_bases[b], cpu); 1986 base->clk = jiffies; 1987 base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA; 1988 base->is_idle = false; 1989 base->must_forward_clk = true; 1990 } 1991 return 0; 1992 } 1993 1994 int timers_dead_cpu(unsigned int cpu) 1995 { 1996 struct timer_base *old_base; 1997 struct timer_base *new_base; 1998 int b, i; 1999 2000 BUG_ON(cpu_online(cpu)); 2001 2002 for (b = 0; b < NR_BASES; b++) { 2003 old_base = per_cpu_ptr(&timer_bases[b], cpu); 2004 new_base = get_cpu_ptr(&timer_bases[b]); 2005 /* 2006 * The caller is globally serialized and nobody else 2007 * takes two locks at once, deadlock is not possible. 2008 */ 2009 raw_spin_lock_irq(&new_base->lock); 2010 raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); 2011 2012 /* 2013 * The current CPUs base clock might be stale. Update it 2014 * before moving the timers over. 2015 */ 2016 forward_timer_base(new_base); 2017 2018 BUG_ON(old_base->running_timer); 2019 2020 for (i = 0; i < WHEEL_SIZE; i++) 2021 migrate_timer_list(new_base, old_base->vectors + i); 2022 2023 raw_spin_unlock(&old_base->lock); 2024 raw_spin_unlock_irq(&new_base->lock); 2025 put_cpu_ptr(&timer_bases); 2026 } 2027 return 0; 2028 } 2029 2030 #endif /* CONFIG_HOTPLUG_CPU */ 2031 2032 static void __init init_timer_cpu(int cpu) 2033 { 2034 struct timer_base *base; 2035 int i; 2036 2037 for (i = 0; i < NR_BASES; i++) { 2038 base = per_cpu_ptr(&timer_bases[i], cpu); 2039 base->cpu = cpu; 2040 raw_spin_lock_init(&base->lock); 2041 base->clk = jiffies; 2042 timer_base_init_expiry_lock(base); 2043 } 2044 } 2045 2046 static void __init init_timer_cpus(void) 2047 { 2048 int cpu; 2049 2050 for_each_possible_cpu(cpu) 2051 init_timer_cpu(cpu); 2052 } 2053 2054 void __init init_timers(void) 2055 { 2056 init_timer_cpus(); 2057 open_softirq(TIMER_SOFTIRQ, run_timer_softirq); 2058 } 2059 2060 /** 2061 * msleep - sleep safely even with waitqueue interruptions 2062 * @msecs: Time in milliseconds to sleep for 2063 */ 2064 void msleep(unsigned int msecs) 2065 { 2066 unsigned long timeout = msecs_to_jiffies(msecs) + 1; 2067 2068 while (timeout) 2069 timeout = schedule_timeout_uninterruptible(timeout); 2070 } 2071 2072 EXPORT_SYMBOL(msleep); 2073 2074 /** 2075 * msleep_interruptible - sleep waiting for signals 2076 * @msecs: Time in milliseconds to sleep for 2077 */ 2078 unsigned long msleep_interruptible(unsigned int msecs) 2079 { 2080 unsigned long timeout = msecs_to_jiffies(msecs) + 1; 2081 2082 while (timeout && !signal_pending(current)) 2083 timeout = schedule_timeout_interruptible(timeout); 2084 return jiffies_to_msecs(timeout); 2085 } 2086 2087 EXPORT_SYMBOL(msleep_interruptible); 2088 2089 /** 2090 * usleep_range - Sleep for an approximate time 2091 * @min: Minimum time in usecs to sleep 2092 * @max: Maximum time in usecs to sleep 2093 * 2094 * In non-atomic context where the exact wakeup time is flexible, use 2095 * usleep_range() instead of udelay(). The sleep improves responsiveness 2096 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces 2097 * power usage by allowing hrtimers to take advantage of an already- 2098 * scheduled interrupt instead of scheduling a new one just for this sleep. 2099 */ 2100 void __sched usleep_range(unsigned long min, unsigned long max) 2101 { 2102 ktime_t exp = ktime_add_us(ktime_get(), min); 2103 u64 delta = (u64)(max - min) * NSEC_PER_USEC; 2104 2105 for (;;) { 2106 __set_current_state(TASK_UNINTERRUPTIBLE); 2107 /* Do not return before the requested sleep time has elapsed */ 2108 if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS)) 2109 break; 2110 } 2111 } 2112 EXPORT_SYMBOL(usleep_range); 2113