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
  47#include <linux/uaccess.h>
  48#include <asm/unistd.h>
  49#include <asm/div64.h>
  50#include <asm/timex.h>
  51#include <asm/io.h>
  52
  53#include "tick-internal.h"
  54
  55#define CREATE_TRACE_POINTS
  56#include <trace/events/timer.h>
  57
  58__visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
  59
  60EXPORT_SYMBOL(jiffies_64);
  61
  62/*
  63 * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
  64 * LVL_SIZE buckets. Each level is driven by its own clock and therefor each
  65 * level has a different granularity.
  66 *
  67 * The level granularity is:		LVL_CLK_DIV ^ lvl
  68 * The level clock frequency is:	HZ / (LVL_CLK_DIV ^ level)
  69 *
  70 * The array level of a newly armed timer depends on the relative expiry
  71 * time. The farther the expiry time is away the higher the array level and
  72 * therefor the granularity becomes.
  73 *
  74 * Contrary to the original timer wheel implementation, which aims for 'exact'
  75 * expiry of the timers, this implementation removes the need for recascading
  76 * the timers into the lower array levels. The previous 'classic' timer wheel
  77 * implementation of the kernel already violated the 'exact' expiry by adding
  78 * slack to the expiry time to provide batched expiration. The granularity
  79 * levels provide implicit batching.
  80 *
  81 * This is an optimization of the original timer wheel implementation for the
  82 * majority of the timer wheel use cases: timeouts. The vast majority of
  83 * timeout timers (networking, disk I/O ...) are canceled before expiry. If
  84 * the timeout expires it indicates that normal operation is disturbed, so it
  85 * does not matter much whether the timeout comes with a slight delay.
  86 *
  87 * The only exception to this are networking timers with a small expiry
  88 * time. They rely on the granularity. Those fit into the first wheel level,
  89 * which has HZ granularity.
  90 *
  91 * We don't have cascading anymore. timers with a expiry time above the
  92 * capacity of the last wheel level are force expired at the maximum timeout
  93 * value of the last wheel level. From data sampling we know that the maximum
  94 * value observed is 5 days (network connection tracking), so this should not
  95 * be an issue.
  96 *
  97 * The currently chosen array constants values are a good compromise between
  98 * array size and granularity.
  99 *
 100 * This results in the following granularity and range levels:
 101 *
 102 * HZ 1000 steps
 103 * Level Offset  Granularity            Range
 104 *  0      0         1 ms                0 ms -         63 ms
 105 *  1     64         8 ms               64 ms -        511 ms
 106 *  2    128        64 ms              512 ms -       4095 ms (512ms - ~4s)
 107 *  3    192       512 ms             4096 ms -      32767 ms (~4s - ~32s)
 108 *  4    256      4096 ms (~4s)      32768 ms -     262143 ms (~32s - ~4m)
 109 *  5    320     32768 ms (~32s)    262144 ms -    2097151 ms (~4m - ~34m)
 110 *  6    384    262144 ms (~4m)    2097152 ms -   16777215 ms (~34m - ~4h)
 111 *  7    448   2097152 ms (~34m)  16777216 ms -  134217727 ms (~4h - ~1d)
 112 *  8    512  16777216 ms (~4h)  134217728 ms - 1073741822 ms (~1d - ~12d)
 113 *
 114 * HZ  300
 115 * Level Offset  Granularity            Range
 116 *  0	   0         3 ms                0 ms -        210 ms
 117 *  1	  64        26 ms              213 ms -       1703 ms (213ms - ~1s)
 118 *  2	 128       213 ms             1706 ms -      13650 ms (~1s - ~13s)
 119 *  3	 192      1706 ms (~1s)      13653 ms -     109223 ms (~13s - ~1m)
 120 *  4	 256     13653 ms (~13s)    109226 ms -     873810 ms (~1m - ~14m)
 121 *  5	 320    109226 ms (~1m)     873813 ms -    6990503 ms (~14m - ~1h)
 122 *  6	 384    873813 ms (~14m)   6990506 ms -   55924050 ms (~1h - ~15h)
 123 *  7	 448   6990506 ms (~1h)   55924053 ms -  447392423 ms (~15h - ~5d)
 124 *  8    512  55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
 125 *
 126 * HZ  250
 127 * Level Offset  Granularity            Range
 128 *  0	   0         4 ms                0 ms -        255 ms
 129 *  1	  64        32 ms              256 ms -       2047 ms (256ms - ~2s)
 130 *  2	 128       256 ms             2048 ms -      16383 ms (~2s - ~16s)
 131 *  3	 192      2048 ms (~2s)      16384 ms -     131071 ms (~16s - ~2m)
 132 *  4	 256     16384 ms (~16s)    131072 ms -    1048575 ms (~2m - ~17m)
 133 *  5	 320    131072 ms (~2m)    1048576 ms -    8388607 ms (~17m - ~2h)
 134 *  6	 384   1048576 ms (~17m)   8388608 ms -   67108863 ms (~2h - ~18h)
 135 *  7	 448   8388608 ms (~2h)   67108864 ms -  536870911 ms (~18h - ~6d)
 136 *  8    512  67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
 137 *
 138 * HZ  100
 139 * Level Offset  Granularity            Range
 140 *  0	   0         10 ms               0 ms -        630 ms
 141 *  1	  64         80 ms             640 ms -       5110 ms (640ms - ~5s)
 142 *  2	 128        640 ms            5120 ms -      40950 ms (~5s - ~40s)
 143 *  3	 192       5120 ms (~5s)     40960 ms -     327670 ms (~40s - ~5m)
 144 *  4	 256      40960 ms (~40s)   327680 ms -    2621430 ms (~5m - ~43m)
 145 *  5	 320     327680 ms (~5m)   2621440 ms -   20971510 ms (~43m - ~5h)
 146 *  6	 384    2621440 ms (~43m) 20971520 ms -  167772150 ms (~5h - ~1d)
 147 *  7	 448   20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
 148 */
 149
 150/* Clock divisor for the next level */
 151#define LVL_CLK_SHIFT	3
 152#define LVL_CLK_DIV	(1UL << LVL_CLK_SHIFT)
 153#define LVL_CLK_MASK	(LVL_CLK_DIV - 1)
 154#define LVL_SHIFT(n)	((n) * LVL_CLK_SHIFT)
 155#define LVL_GRAN(n)	(1UL << LVL_SHIFT(n))
 156
 157/*
 158 * The time start value for each level to select the bucket at enqueue
 159 * time.
 
 160 */
 161#define LVL_START(n)	((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
 162
 163/* Size of each clock level */
 164#define LVL_BITS	6
 165#define LVL_SIZE	(1UL << LVL_BITS)
 166#define LVL_MASK	(LVL_SIZE - 1)
 167#define LVL_OFFS(n)	((n) * LVL_SIZE)
 168
 169/* Level depth */
 170#if HZ > 100
 171# define LVL_DEPTH	9
 172# else
 173# define LVL_DEPTH	8
 174#endif
 175
 176/* The cutoff (max. capacity of the wheel) */
 177#define WHEEL_TIMEOUT_CUTOFF	(LVL_START(LVL_DEPTH))
 178#define WHEEL_TIMEOUT_MAX	(WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
 179
 180/*
 181 * The resulting wheel size. If NOHZ is configured we allocate two
 182 * wheels so we have a separate storage for the deferrable timers.
 183 */
 184#define WHEEL_SIZE	(LVL_SIZE * LVL_DEPTH)
 185
 186#ifdef CONFIG_NO_HZ_COMMON
 187# define NR_BASES	2
 188# define BASE_STD	0
 189# define BASE_DEF	1
 190#else
 191# define NR_BASES	1
 192# define BASE_STD	0
 193# define BASE_DEF	0
 194#endif
 195
 196struct timer_base {
 197	raw_spinlock_t		lock;
 198	struct timer_list	*running_timer;
 199#ifdef CONFIG_PREEMPT_RT
 200	spinlock_t		expiry_lock;
 201	atomic_t		timer_waiters;
 202#endif
 203	unsigned long		clk;
 204	unsigned long		next_expiry;
 205	unsigned int		cpu;
 
 206	bool			is_idle;
 207	bool			must_forward_clk;
 208	DECLARE_BITMAP(pending_map, WHEEL_SIZE);
 209	struct hlist_head	vectors[WHEEL_SIZE];
 210} ____cacheline_aligned;
 211
 212static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
 213
 214#ifdef CONFIG_NO_HZ_COMMON
 215
 216static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
 217static DEFINE_MUTEX(timer_keys_mutex);
 218
 219static void timer_update_keys(struct work_struct *work);
 220static DECLARE_WORK(timer_update_work, timer_update_keys);
 221
 222#ifdef CONFIG_SMP
 223unsigned int sysctl_timer_migration = 1;
 224
 225DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
 226
 227static void timers_update_migration(void)
 228{
 229	if (sysctl_timer_migration && tick_nohz_active)
 230		static_branch_enable(&timers_migration_enabled);
 231	else
 232		static_branch_disable(&timers_migration_enabled);
 233}
 234#else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 235static inline void timers_update_migration(void) { }
 236#endif /* !CONFIG_SMP */
 237
 238static void timer_update_keys(struct work_struct *work)
 239{
 240	mutex_lock(&timer_keys_mutex);
 241	timers_update_migration();
 242	static_branch_enable(&timers_nohz_active);
 243	mutex_unlock(&timer_keys_mutex);
 244}
 245
 246void timers_update_nohz(void)
 247{
 248	schedule_work(&timer_update_work);
 249}
 250
 251int timer_migration_handler(struct ctl_table *table, int write,
 252			    void __user *buffer, size_t *lenp,
 253			    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
 265static inline bool is_timers_nohz_active(void)
 266{
 267	return static_branch_unlikely(&timers_nohz_active);
 268}
 269#else
 270static inline bool is_timers_nohz_active(void) { return false; }
 271#endif /* NO_HZ_COMMON */
 272
 273static 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 */
 333unsigned long __round_jiffies(unsigned long j, int cpu)
 334{
 335	return round_jiffies_common(j, cpu, false);
 336}
 337EXPORT_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 */
 359unsigned 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}
 366EXPORT_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 */
 383unsigned long round_jiffies(unsigned long j)
 384{
 385	return round_jiffies_common(j, raw_smp_processor_id(), false);
 386}
 387EXPORT_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 */
 404unsigned long round_jiffies_relative(unsigned long j)
 405{
 406	return __round_jiffies_relative(j, raw_smp_processor_id());
 407}
 408EXPORT_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 */
 420unsigned long __round_jiffies_up(unsigned long j, int cpu)
 421{
 422	return round_jiffies_common(j, cpu, true);
 423}
 424EXPORT_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 */
 436unsigned 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}
 443EXPORT_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 */
 454unsigned long round_jiffies_up(unsigned long j)
 455{
 456	return round_jiffies_common(j, raw_smp_processor_id(), true);
 457}
 458EXPORT_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 */
 469unsigned long round_jiffies_up_relative(unsigned long j)
 470{
 471	return __round_jiffies_up_relative(j, raw_smp_processor_id());
 472}
 473EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
 474
 475
 476static inline unsigned int timer_get_idx(struct timer_list *timer)
 477{
 478	return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
 479}
 480
 481static 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 */
 491static 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
 497static 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 (expires >= WHEEL_TIMEOUT_CUTOFF)
 526			expires = 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 */
 537static 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
 547static 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
 556static void
 557trigger_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	base->next_expiry = timer->expires;
 589	wake_up_nohz_cpu(base->cpu);
 
 
 
 
 
 
 
 
 590}
 591
 592static void
 593internal_add_timer(struct timer_base *base, struct timer_list *timer)
 594{
 595	__internal_add_timer(base, timer);
 596	trigger_dyntick_cpu(base, timer);
 
 
 
 597}
 598
 599#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 600
 601static struct debug_obj_descr timer_debug_descr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 602
 603static void *timer_debug_hint(void *addr)
 604{
 605	return ((struct timer_list *) addr)->function;
 
 
 
 
 
 
 
 
 
 
 
 606}
 607
 608static bool timer_is_static_object(void *addr)
 609{
 610	struct timer_list *timer = addr;
 611
 612	return (timer->entry.pprev == NULL &&
 613		timer->entry.next == TIMER_ENTRY_STATIC);
 614}
 615
 616/*
 617 * fixup_init is called when:
 618 * - an active object is initialized
 619 */
 620static bool timer_fixup_init(void *addr, enum debug_obj_state state)
 621{
 622	struct timer_list *timer = addr;
 623
 624	switch (state) {
 625	case ODEBUG_STATE_ACTIVE:
 626		del_timer_sync(timer);
 627		debug_object_init(timer, &timer_debug_descr);
 628		return true;
 629	default:
 630		return false;
 631	}
 632}
 633
 634/* Stub timer callback for improperly used timers. */
 635static void stub_timer(struct timer_list *unused)
 636{
 637	WARN_ON(1);
 638}
 639
 640/*
 641 * fixup_activate is called when:
 642 * - an active object is activated
 643 * - an unknown non-static object is activated
 644 */
 645static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
 646{
 647	struct timer_list *timer = addr;
 648
 649	switch (state) {
 650	case ODEBUG_STATE_NOTAVAILABLE:
 651		timer_setup(timer, stub_timer, 0);
 652		return true;
 653
 654	case ODEBUG_STATE_ACTIVE:
 655		WARN_ON(1);
 656		/* fall through */
 657	default:
 658		return false;
 659	}
 660}
 661
 662/*
 663 * fixup_free is called when:
 664 * - an active object is freed
 665 */
 666static bool timer_fixup_free(void *addr, enum debug_obj_state state)
 667{
 668	struct timer_list *timer = addr;
 669
 670	switch (state) {
 671	case ODEBUG_STATE_ACTIVE:
 672		del_timer_sync(timer);
 673		debug_object_free(timer, &timer_debug_descr);
 674		return true;
 675	default:
 676		return false;
 677	}
 678}
 679
 680/*
 681 * fixup_assert_init is called when:
 682 * - an untracked/uninit-ed object is found
 683 */
 684static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
 685{
 686	struct timer_list *timer = addr;
 687
 688	switch (state) {
 689	case ODEBUG_STATE_NOTAVAILABLE:
 690		timer_setup(timer, stub_timer, 0);
 691		return true;
 692	default:
 693		return false;
 694	}
 695}
 696
 697static struct debug_obj_descr timer_debug_descr = {
 698	.name			= "timer_list",
 699	.debug_hint		= timer_debug_hint,
 700	.is_static_object	= timer_is_static_object,
 701	.fixup_init		= timer_fixup_init,
 702	.fixup_activate		= timer_fixup_activate,
 703	.fixup_free		= timer_fixup_free,
 704	.fixup_assert_init	= timer_fixup_assert_init,
 705};
 706
 707static inline void debug_timer_init(struct timer_list *timer)
 708{
 709	debug_object_init(timer, &timer_debug_descr);
 710}
 711
 712static inline void debug_timer_activate(struct timer_list *timer)
 713{
 714	debug_object_activate(timer, &timer_debug_descr);
 715}
 716
 717static inline void debug_timer_deactivate(struct timer_list *timer)
 718{
 719	debug_object_deactivate(timer, &timer_debug_descr);
 720}
 721
 722static inline void debug_timer_free(struct timer_list *timer)
 723{
 724	debug_object_free(timer, &timer_debug_descr);
 725}
 726
 727static inline void debug_timer_assert_init(struct timer_list *timer)
 728{
 729	debug_object_assert_init(timer, &timer_debug_descr);
 730}
 731
 732static void do_init_timer(struct timer_list *timer,
 733			  void (*func)(struct timer_list *),
 734			  unsigned int flags,
 735			  const char *name, struct lock_class_key *key);
 736
 737void init_timer_on_stack_key(struct timer_list *timer,
 738			     void (*func)(struct timer_list *),
 739			     unsigned int flags,
 740			     const char *name, struct lock_class_key *key)
 741{
 742	debug_object_init_on_stack(timer, &timer_debug_descr);
 743	do_init_timer(timer, func, flags, name, key);
 744}
 745EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
 746
 747void destroy_timer_on_stack(struct timer_list *timer)
 748{
 749	debug_object_free(timer, &timer_debug_descr);
 750}
 751EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
 752
 753#else
 754static inline void debug_timer_init(struct timer_list *timer) { }
 755static inline void debug_timer_activate(struct timer_list *timer) { }
 756static inline void debug_timer_deactivate(struct timer_list *timer) { }
 757static inline void debug_timer_assert_init(struct timer_list *timer) { }
 758#endif
 759
 760static inline void debug_init(struct timer_list *timer)
 761{
 762	debug_timer_init(timer);
 763	trace_timer_init(timer);
 764}
 765
 766static inline void debug_deactivate(struct timer_list *timer)
 767{
 768	debug_timer_deactivate(timer);
 769	trace_timer_cancel(timer);
 770}
 771
 772static inline void debug_assert_init(struct timer_list *timer)
 773{
 774	debug_timer_assert_init(timer);
 775}
 776
 777static void do_init_timer(struct timer_list *timer,
 778			  void (*func)(struct timer_list *),
 779			  unsigned int flags,
 780			  const char *name, struct lock_class_key *key)
 781{
 782	timer->entry.pprev = NULL;
 783	timer->function = func;
 
 
 784	timer->flags = flags | raw_smp_processor_id();
 785	lockdep_init_map(&timer->lockdep_map, name, key, 0);
 786}
 787
 788/**
 789 * init_timer_key - initialize a timer
 790 * @timer: the timer to be initialized
 791 * @func: timer callback function
 792 * @flags: timer flags
 793 * @name: name of the timer
 794 * @key: lockdep class key of the fake lock used for tracking timer
 795 *       sync lock dependencies
 796 *
 797 * init_timer_key() must be done to a timer prior calling *any* of the
 798 * other timer functions.
 799 */
 800void init_timer_key(struct timer_list *timer,
 801		    void (*func)(struct timer_list *), unsigned int flags,
 802		    const char *name, struct lock_class_key *key)
 803{
 804	debug_init(timer);
 805	do_init_timer(timer, func, flags, name, key);
 806}
 807EXPORT_SYMBOL(init_timer_key);
 808
 809static inline void detach_timer(struct timer_list *timer, bool clear_pending)
 810{
 811	struct hlist_node *entry = &timer->entry;
 812
 813	debug_deactivate(timer);
 814
 815	__hlist_del(entry);
 816	if (clear_pending)
 817		entry->pprev = NULL;
 818	entry->next = LIST_POISON2;
 819}
 820
 821static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
 822			     bool clear_pending)
 823{
 824	unsigned idx = timer_get_idx(timer);
 825
 826	if (!timer_pending(timer))
 827		return 0;
 828
 829	if (hlist_is_singular_node(&timer->entry, base->vectors + idx))
 830		__clear_bit(idx, base->pending_map);
 
 
 831
 832	detach_timer(timer, clear_pending);
 833	return 1;
 834}
 835
 836static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
 837{
 838	struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu);
 839
 840	/*
 841	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
 842	 * to use the deferrable base.
 843	 */
 844	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
 845		base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
 846	return base;
 847}
 848
 849static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
 850{
 851	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
 852
 853	/*
 854	 * If the timer is deferrable and NO_HZ_COMMON is set then we need
 855	 * to use the deferrable base.
 856	 */
 857	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
 858		base = this_cpu_ptr(&timer_bases[BASE_DEF]);
 859	return base;
 860}
 861
 862static inline struct timer_base *get_timer_base(u32 tflags)
 863{
 864	return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
 865}
 866
 867static inline struct timer_base *
 868get_target_base(struct timer_base *base, unsigned tflags)
 869{
 870#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 871	if (static_branch_likely(&timers_migration_enabled) &&
 872	    !(tflags & TIMER_PINNED))
 873		return get_timer_cpu_base(tflags, get_nohz_timer_target());
 874#endif
 875	return get_timer_this_cpu_base(tflags);
 876}
 877
 878static inline void forward_timer_base(struct timer_base *base)
 879{
 880#ifdef CONFIG_NO_HZ_COMMON
 881	unsigned long jnow;
 882
 883	/*
 884	 * We only forward the base when we are idle or have just come out of
 885	 * idle (must_forward_clk logic), and have a delta between base clock
 886	 * and jiffies. In the common case, run_timers will take care of it.
 887	 */
 888	if (likely(!base->must_forward_clk))
 889		return;
 890
 891	jnow = READ_ONCE(jiffies);
 892	base->must_forward_clk = base->is_idle;
 893	if ((long)(jnow - base->clk) < 2)
 894		return;
 895
 896	/*
 897	 * If the next expiry value is > jiffies, then we fast forward to
 898	 * jiffies otherwise we forward to the next expiry value.
 899	 */
 900	if (time_after(base->next_expiry, jnow))
 901		base->clk = jnow;
 902	else
 
 
 903		base->clk = base->next_expiry;
 904#endif
 905}
 906
 907
 908/*
 909 * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
 910 * that all timers which are tied to this base are locked, and the base itself
 911 * is locked too.
 912 *
 913 * So __run_timers/migrate_timers can safely modify all timers which could
 914 * be found in the base->vectors array.
 915 *
 916 * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
 917 * to wait until the migration is done.
 918 */
 919static struct timer_base *lock_timer_base(struct timer_list *timer,
 920					  unsigned long *flags)
 921	__acquires(timer->base->lock)
 922{
 923	for (;;) {
 924		struct timer_base *base;
 925		u32 tf;
 926
 927		/*
 928		 * We need to use READ_ONCE() here, otherwise the compiler
 929		 * might re-read @tf between the check for TIMER_MIGRATING
 930		 * and spin_lock().
 931		 */
 932		tf = READ_ONCE(timer->flags);
 933
 934		if (!(tf & TIMER_MIGRATING)) {
 935			base = get_timer_base(tf);
 936			raw_spin_lock_irqsave(&base->lock, *flags);
 937			if (timer->flags == tf)
 938				return base;
 939			raw_spin_unlock_irqrestore(&base->lock, *flags);
 940		}
 941		cpu_relax();
 942	}
 943}
 944
 945#define MOD_TIMER_PENDING_ONLY		0x01
 946#define MOD_TIMER_REDUCE		0x02
 
 947
 948static inline int
 949__mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
 950{
 
 951	struct timer_base *base, *new_base;
 952	unsigned int idx = UINT_MAX;
 953	unsigned long clk = 0, flags;
 954	int ret = 0;
 955
 956	BUG_ON(!timer->function);
 957
 958	/*
 959	 * This is a common optimization triggered by the networking code - if
 960	 * the timer is re-modified to have the same timeout or ends up in the
 961	 * same array bucket then just return:
 962	 */
 963	if (timer_pending(timer)) {
 964		/*
 965		 * The downside of this optimization is that it can result in
 966		 * larger granularity than you would get from adding a new
 967		 * timer with this expiry.
 968		 */
 969		long diff = timer->expires - expires;
 970
 971		if (!diff)
 972			return 1;
 973		if (options & MOD_TIMER_REDUCE && diff <= 0)
 974			return 1;
 975
 976		/*
 977		 * We lock timer base and calculate the bucket index right
 978		 * here. If the timer ends up in the same bucket, then we
 979		 * just update the expiry time and avoid the whole
 980		 * dequeue/enqueue dance.
 981		 */
 982		base = lock_timer_base(timer, &flags);
 
 
 
 
 
 
 
 
 983		forward_timer_base(base);
 984
 985		if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
 986		    time_before_eq(timer->expires, expires)) {
 987			ret = 1;
 988			goto out_unlock;
 989		}
 990
 991		clk = base->clk;
 992		idx = calc_wheel_index(expires, clk);
 993
 994		/*
 995		 * Retrieve and compare the array index of the pending
 996		 * timer. If it matches set the expiry to the new value so a
 997		 * subsequent call will exit in the expires check above.
 998		 */
 999		if (idx == timer_get_idx(timer)) {
1000			if (!(options & MOD_TIMER_REDUCE))
1001				timer->expires = expires;
1002			else if (time_after(timer->expires, expires))
1003				timer->expires = expires;
1004			ret = 1;
1005			goto out_unlock;
1006		}
1007	} else {
1008		base = lock_timer_base(timer, &flags);
 
 
 
 
 
 
 
 
1009		forward_timer_base(base);
1010	}
1011
1012	ret = detach_if_pending(timer, base, false);
1013	if (!ret && (options & MOD_TIMER_PENDING_ONLY))
1014		goto out_unlock;
1015
1016	new_base = get_target_base(base, timer->flags);
1017
1018	if (base != new_base) {
1019		/*
1020		 * We are trying to schedule the timer on the new base.
1021		 * However we can't change timer's base while it is running,
1022		 * otherwise del_timer_sync() can't detect that the timer's
1023		 * handler yet has not finished. This also guarantees that the
1024		 * timer is serialized wrt itself.
1025		 */
1026		if (likely(base->running_timer != timer)) {
1027			/* See the comment in lock_timer_base() */
1028			timer->flags |= TIMER_MIGRATING;
1029
1030			raw_spin_unlock(&base->lock);
1031			base = new_base;
1032			raw_spin_lock(&base->lock);
1033			WRITE_ONCE(timer->flags,
1034				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
1035			forward_timer_base(base);
1036		}
1037	}
1038
1039	debug_timer_activate(timer);
1040
1041	timer->expires = expires;
1042	/*
1043	 * If 'idx' was calculated above and the base time did not advance
1044	 * between calculating 'idx' and possibly switching the base, only
1045	 * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise
1046	 * we need to (re)calculate the wheel index via
1047	 * internal_add_timer().
1048	 */
1049	if (idx != UINT_MAX && clk == base->clk) {
1050		enqueue_timer(base, timer, idx);
1051		trigger_dyntick_cpu(base, timer);
1052	} else {
1053		internal_add_timer(base, timer);
1054	}
1055
1056out_unlock:
1057	raw_spin_unlock_irqrestore(&base->lock, flags);
1058
1059	return ret;
1060}
1061
1062/**
1063 * mod_timer_pending - modify a pending timer's timeout
1064 * @timer: the pending timer to be modified
1065 * @expires: new timeout in jiffies
1066 *
1067 * mod_timer_pending() is the same for pending timers as mod_timer(),
1068 * but will not re-activate and modify already deleted timers.
1069 *
1070 * It is useful for unserialized use of timers.
 
 
 
 
 
 
1071 */
1072int mod_timer_pending(struct timer_list *timer, unsigned long expires)
1073{
1074	return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
1075}
1076EXPORT_SYMBOL(mod_timer_pending);
1077
1078/**
1079 * mod_timer - modify a timer's timeout
1080 * @timer: the timer to be modified
1081 * @expires: new timeout in jiffies
1082 *
1083 * mod_timer() is a more efficient way to update the expire field of an
1084 * active timer (if the timer is inactive it will be activated)
1085 *
1086 * mod_timer(timer, expires) is equivalent to:
1087 *
1088 *     del_timer(timer); timer->expires = expires; add_timer(timer);
1089 *
 
 
 
 
1090 * Note that if there are multiple unserialized concurrent users of the
1091 * same timer, then mod_timer() is the only safe way to modify the timeout,
1092 * since add_timer() cannot modify an already running timer.
1093 *
1094 * The function returns whether it has modified a pending timer or not.
1095 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
1096 * active timer returns 1.)
 
 
 
 
 
 
1097 */
1098int mod_timer(struct timer_list *timer, unsigned long expires)
1099{
1100	return __mod_timer(timer, expires, 0);
1101}
1102EXPORT_SYMBOL(mod_timer);
1103
1104/**
1105 * timer_reduce - Modify a timer's timeout if it would reduce the timeout
1106 * @timer:	The timer to be modified
1107 * @expires:	New timeout in jiffies
1108 *
1109 * timer_reduce() is very similar to mod_timer(), except that it will only
1110 * modify a running timer if that would reduce the expiration time (it will
1111 * start a timer that isn't running).
 
 
 
 
 
 
 
 
 
 
 
1112 */
1113int timer_reduce(struct timer_list *timer, unsigned long expires)
1114{
1115	return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
1116}
1117EXPORT_SYMBOL(timer_reduce);
1118
1119/**
1120 * add_timer - start a timer
1121 * @timer: the timer to be added
1122 *
1123 * The kernel will do a ->function(@timer) callback from the
1124 * timer interrupt at the ->expires point in the future. The
1125 * current time is 'jiffies'.
1126 *
1127 * The timer's ->expires, ->function fields must be set prior calling this
1128 * function.
 
 
 
1129 *
1130 * Timers with an ->expires field in the past will be executed in the next
1131 * timer tick.
 
 
 
1132 */
1133void add_timer(struct timer_list *timer)
1134{
1135	BUG_ON(timer_pending(timer));
1136	mod_timer(timer, timer->expires);
 
1137}
1138EXPORT_SYMBOL(add_timer);
1139
1140/**
1141 * add_timer_on - start a timer on a particular CPU
1142 * @timer: the timer to be added
1143 * @cpu: the CPU to start it on
 
 
1144 *
1145 * This is not very scalable on SMP. Double adds are not possible.
1146 */
1147void add_timer_on(struct timer_list *timer, int cpu)
1148{
1149	struct timer_base *new_base, *base;
1150	unsigned long flags;
1151
1152	BUG_ON(timer_pending(timer) || !timer->function);
 
 
 
1153
1154	new_base = get_timer_cpu_base(timer->flags, cpu);
1155
1156	/*
1157	 * If @timer was on a different CPU, it should be migrated with the
1158	 * old base locked to prevent other operations proceeding with the
1159	 * wrong base locked.  See lock_timer_base().
1160	 */
1161	base = lock_timer_base(timer, &flags);
 
 
 
 
 
 
 
1162	if (base != new_base) {
1163		timer->flags |= TIMER_MIGRATING;
1164
1165		raw_spin_unlock(&base->lock);
1166		base = new_base;
1167		raw_spin_lock(&base->lock);
1168		WRITE_ONCE(timer->flags,
1169			   (timer->flags & ~TIMER_BASEMASK) | cpu);
1170	}
1171	forward_timer_base(base);
1172
1173	debug_timer_activate(timer);
1174	internal_add_timer(base, timer);
 
1175	raw_spin_unlock_irqrestore(&base->lock, flags);
1176}
1177EXPORT_SYMBOL_GPL(add_timer_on);
1178
1179/**
1180 * del_timer - deactivate a timer.
1181 * @timer: the timer to be deactivated
1182 *
1183 * del_timer() deactivates a timer - this works on both active and inactive
1184 * timers.
1185 *
1186 * The function returns whether it has deactivated a pending timer or not.
1187 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
1188 * active timer returns 1.)
 
 
 
 
1189 */
1190int del_timer(struct timer_list *timer)
1191{
1192	struct timer_base *base;
1193	unsigned long flags;
1194	int ret = 0;
1195
1196	debug_assert_init(timer);
1197
1198	if (timer_pending(timer)) {
 
 
 
 
 
 
 
 
 
 
 
1199		base = lock_timer_base(timer, &flags);
1200		ret = detach_if_pending(timer, base, true);
 
 
1201		raw_spin_unlock_irqrestore(&base->lock, flags);
1202	}
1203
1204	return ret;
1205}
1206EXPORT_SYMBOL(del_timer);
1207
1208/**
1209 * try_to_del_timer_sync - Try to deactivate a timer
1210 * @timer: timer to delete
 
 
 
 
 
 
1211 *
1212 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1213 * exit the timer is not queued and the handler is not running on any CPU.
 
1214 */
1215int try_to_del_timer_sync(struct timer_list *timer)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1216{
1217	struct timer_base *base;
1218	unsigned long flags;
1219	int ret = -1;
1220
1221	debug_assert_init(timer);
1222
1223	base = lock_timer_base(timer, &flags);
1224
1225	if (base->running_timer != timer)
1226		ret = detach_if_pending(timer, base, true);
 
 
1227
1228	raw_spin_unlock_irqrestore(&base->lock, flags);
1229
1230	return ret;
1231}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1232EXPORT_SYMBOL(try_to_del_timer_sync);
1233
1234#ifdef CONFIG_PREEMPT_RT
1235static __init void timer_base_init_expiry_lock(struct timer_base *base)
1236{
1237	spin_lock_init(&base->expiry_lock);
1238}
1239
1240static inline void timer_base_lock_expiry(struct timer_base *base)
1241{
1242	spin_lock(&base->expiry_lock);
1243}
1244
1245static inline void timer_base_unlock_expiry(struct timer_base *base)
1246{
1247	spin_unlock(&base->expiry_lock);
1248}
1249
1250/*
1251 * The counterpart to del_timer_wait_running().
1252 *
1253 * If there is a waiter for base->expiry_lock, then it was waiting for the
1254 * timer callback to finish. Drop expiry_lock and reaquire it. That allows
1255 * the waiter to acquire the lock and make progress.
1256 */
1257static void timer_sync_wait_running(struct timer_base *base)
1258{
1259	if (atomic_read(&base->timer_waiters)) {
 
1260		spin_unlock(&base->expiry_lock);
1261		spin_lock(&base->expiry_lock);
 
1262	}
1263}
1264
1265/*
1266 * This function is called on PREEMPT_RT kernels when the fast path
1267 * deletion of a timer failed because the timer callback function was
1268 * running.
1269 *
1270 * This prevents priority inversion, if the softirq thread on a remote CPU
1271 * got preempted, and it prevents a life lock when the task which tries to
1272 * delete a timer preempted the softirq thread running the timer callback
1273 * function.
1274 */
1275static void del_timer_wait_running(struct timer_list *timer)
1276{
1277	u32 tf;
1278
1279	tf = READ_ONCE(timer->flags);
1280	if (!(tf & TIMER_MIGRATING)) {
1281		struct timer_base *base = get_timer_base(tf);
1282
1283		/*
1284		 * Mark the base as contended and grab the expiry lock,
1285		 * which is held by the softirq across the timer
1286		 * callback. Drop the lock immediately so the softirq can
1287		 * expire the next timer. In theory the timer could already
1288		 * be running again, but that's more than unlikely and just
1289		 * causes another wait loop.
1290		 */
1291		atomic_inc(&base->timer_waiters);
1292		spin_lock_bh(&base->expiry_lock);
1293		atomic_dec(&base->timer_waiters);
1294		spin_unlock_bh(&base->expiry_lock);
1295	}
1296}
1297#else
1298static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
1299static inline void timer_base_lock_expiry(struct timer_base *base) { }
1300static inline void timer_base_unlock_expiry(struct timer_base *base) { }
1301static inline void timer_sync_wait_running(struct timer_base *base) { }
1302static inline void del_timer_wait_running(struct timer_list *timer) { }
1303#endif
1304
1305#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
1306/**
1307 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1308 * @timer: the timer to be deactivated
1309 *
1310 * This function only differs from del_timer() on SMP: besides deactivating
1311 * the timer it also makes sure the handler has finished executing on other
1312 * CPUs.
1313 *
1314 * Synchronization rules: Callers must prevent restarting of the timer,
1315 * otherwise this function is meaningless. It must not be called from
1316 * interrupt contexts unless the timer is an irqsafe one. The caller must
1317 * not hold locks which would prevent completion of the timer's
1318 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1319 * timer is not queued and the handler is not running on any CPU.
1320 *
1321 * Note: For !irqsafe timers, you must not hold locks that are held in
1322 *   interrupt context while calling this function. Even if the lock has
1323 *   nothing to do with the timer in question.  Here's why::
1324 *
1325 *    CPU0                             CPU1
1326 *    ----                             ----
1327 *                                     <SOFTIRQ>
1328 *                                       call_timer_fn();
1329 *                                       base->running_timer = mytimer;
1330 *    spin_lock_irq(somelock);
1331 *                                     <IRQ>
1332 *                                        spin_lock(somelock);
1333 *    del_timer_sync(mytimer);
1334 *    while (base->running_timer == mytimer);
1335 *
1336 * Now del_timer_sync() will never return and never release somelock.
1337 * The interrupt on the other CPU is waiting to grab somelock but
1338 * it has interrupted the softirq that CPU0 is waiting to finish.
1339 *
1340 * The function returns whether it has deactivated a pending timer or not.
1341 */
1342int del_timer_sync(struct timer_list *timer)
1343{
1344	int ret;
1345
1346#ifdef CONFIG_LOCKDEP
1347	unsigned long flags;
1348
1349	/*
1350	 * If lockdep gives a backtrace here, please reference
1351	 * the synchronization rules above.
1352	 */
1353	local_irq_save(flags);
1354	lock_map_acquire(&timer->lockdep_map);
1355	lock_map_release(&timer->lockdep_map);
1356	local_irq_restore(flags);
1357#endif
1358	/*
1359	 * don't use it in hardirq context, because it
1360	 * could lead to deadlock.
1361	 */
1362	WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE));
 
 
 
 
 
 
 
1363
1364	do {
1365		ret = try_to_del_timer_sync(timer);
1366
1367		if (unlikely(ret < 0)) {
1368			del_timer_wait_running(timer);
1369			cpu_relax();
1370		}
1371	} while (ret < 0);
1372
1373	return ret;
1374}
1375EXPORT_SYMBOL(del_timer_sync);
1376#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1377
1378static void call_timer_fn(struct timer_list *timer,
1379			  void (*fn)(struct timer_list *),
1380			  unsigned long baseclk)
1381{
1382	int count = preempt_count();
1383
1384#ifdef CONFIG_LOCKDEP
1385	/*
1386	 * It is permissible to free the timer from inside the
1387	 * function that is called from it, this we need to take into
1388	 * account for lockdep too. To avoid bogus "held lock freed"
1389	 * warnings as well as problems when looking into
1390	 * timer->lockdep_map, make a copy and use that here.
1391	 */
1392	struct lockdep_map lockdep_map;
1393
1394	lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1395#endif
1396	/*
1397	 * Couple the lock chain with the lock chain at
1398	 * del_timer_sync() by acquiring the lock_map around the fn()
1399	 * call here and in del_timer_sync().
1400	 */
1401	lock_map_acquire(&lockdep_map);
1402
1403	trace_timer_expire_entry(timer, baseclk);
1404	fn(timer);
1405	trace_timer_expire_exit(timer);
1406
1407	lock_map_release(&lockdep_map);
1408
1409	if (count != preempt_count()) {
1410		WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
1411			  fn, count, preempt_count());
1412		/*
1413		 * Restore the preempt count. That gives us a decent
1414		 * chance to survive and extract information. If the
1415		 * callback kept a lock held, bad luck, but not worse
1416		 * than the BUG() we had.
1417		 */
1418		preempt_count_set(count);
1419	}
1420}
1421
1422static void expire_timers(struct timer_base *base, struct hlist_head *head)
1423{
1424	/*
1425	 * This value is required only for tracing. base->clk was
1426	 * incremented directly before expire_timers was called. But expiry
1427	 * is related to the old base->clk value.
1428	 */
1429	unsigned long baseclk = base->clk - 1;
1430
1431	while (!hlist_empty(head)) {
1432		struct timer_list *timer;
1433		void (*fn)(struct timer_list *);
1434
1435		timer = hlist_entry(head->first, struct timer_list, entry);
1436
1437		base->running_timer = timer;
1438		detach_timer(timer, true);
1439
1440		fn = timer->function;
1441
 
 
 
 
 
 
1442		if (timer->flags & TIMER_IRQSAFE) {
1443			raw_spin_unlock(&base->lock);
1444			call_timer_fn(timer, fn, baseclk);
1445			base->running_timer = NULL;
1446			raw_spin_lock(&base->lock);
 
1447		} else {
1448			raw_spin_unlock_irq(&base->lock);
1449			call_timer_fn(timer, fn, baseclk);
 
1450			base->running_timer = NULL;
1451			timer_sync_wait_running(base);
1452			raw_spin_lock_irq(&base->lock);
1453		}
1454	}
1455}
1456
1457static int __collect_expired_timers(struct timer_base *base,
1458				    struct hlist_head *heads)
1459{
1460	unsigned long clk = base->clk;
1461	struct hlist_head *vec;
1462	int i, levels = 0;
1463	unsigned int idx;
1464
1465	for (i = 0; i < LVL_DEPTH; i++) {
1466		idx = (clk & LVL_MASK) + i * LVL_SIZE;
1467
1468		if (__test_and_clear_bit(idx, base->pending_map)) {
1469			vec = base->vectors + idx;
1470			hlist_move_list(vec, heads++);
1471			levels++;
1472		}
1473		/* Is it time to look at the next level? */
1474		if (clk & LVL_CLK_MASK)
1475			break;
1476		/* Shift clock for the next level granularity */
1477		clk >>= LVL_CLK_SHIFT;
1478	}
1479	return levels;
1480}
1481
1482#ifdef CONFIG_NO_HZ_COMMON
1483/*
1484 * Find the next pending bucket of a level. Search from level start (@offset)
1485 * + @clk upwards and if nothing there, search from start of the level
1486 * (@offset) up to @offset + clk.
1487 */
1488static int next_pending_bucket(struct timer_base *base, unsigned offset,
1489			       unsigned clk)
1490{
1491	unsigned pos, start = offset + clk;
1492	unsigned end = offset + LVL_SIZE;
1493
1494	pos = find_next_bit(base->pending_map, end, start);
1495	if (pos < end)
1496		return pos - start;
1497
1498	pos = find_next_bit(base->pending_map, start, offset);
1499	return pos < start ? pos + LVL_SIZE - start : -1;
1500}
1501
1502/*
1503 * Search the first expiring timer in the various clock levels. Caller must
1504 * hold base->lock.
1505 */
1506static unsigned long __next_timer_interrupt(struct timer_base *base)
1507{
1508	unsigned long clk, next, adj;
1509	unsigned lvl, offset = 0;
1510
1511	next = base->clk + NEXT_TIMER_MAX_DELTA;
1512	clk = base->clk;
1513	for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
1514		int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
 
1515
1516		if (pos >= 0) {
1517			unsigned long tmp = clk + (unsigned long) pos;
1518
1519			tmp <<= LVL_SHIFT(lvl);
1520			if (time_before(tmp, next))
1521				next = tmp;
 
 
 
 
 
 
 
1522		}
1523		/*
1524		 * Clock for the next level. If the current level clock lower
1525		 * bits are zero, we look at the next level as is. If not we
1526		 * need to advance it by one because that's going to be the
1527		 * next expiring bucket in that level. base->clk is the next
1528		 * expiring jiffie. So in case of:
1529		 *
1530		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1531		 *  0    0    0    0    0    0
1532		 *
1533		 * we have to look at all levels @index 0. With
1534		 *
1535		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1536		 *  0    0    0    0    0    2
1537		 *
1538		 * LVL0 has the next expiring bucket @index 2. The upper
1539		 * levels have the next expiring bucket @index 1.
1540		 *
1541		 * In case that the propagation wraps the next level the same
1542		 * rules apply:
1543		 *
1544		 * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
1545		 *  0    0    0    0    F    2
1546		 *
1547		 * So after looking at LVL0 we get:
1548		 *
1549		 * LVL5 LVL4 LVL3 LVL2 LVL1
1550		 *  0    0    0    1    0
1551		 *
1552		 * So no propagation from LVL1 to LVL2 because that happened
1553		 * with the add already, but then we need to propagate further
1554		 * from LVL2 to LVL3.
1555		 *
1556		 * So the simple check whether the lower bits of the current
1557		 * level are 0 or not is sufficient for all cases.
1558		 */
1559		adj = clk & LVL_CLK_MASK ? 1 : 0;
1560		clk >>= LVL_CLK_SHIFT;
1561		clk += adj;
1562	}
 
 
 
 
1563	return next;
1564}
1565
 
1566/*
1567 * Check, if the next hrtimer event is before the next timer wheel
1568 * event:
1569 */
1570static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
1571{
1572	u64 nextevt = hrtimer_get_next_event();
1573
1574	/*
1575	 * If high resolution timers are enabled
1576	 * hrtimer_get_next_event() returns KTIME_MAX.
1577	 */
1578	if (expires <= nextevt)
1579		return expires;
1580
1581	/*
1582	 * If the next timer is already expired, return the tick base
1583	 * time so the tick is fired immediately.
1584	 */
1585	if (nextevt <= basem)
1586		return basem;
1587
1588	/*
1589	 * Round up to the next jiffie. High resolution timers are
1590	 * off, so the hrtimers are expired in the tick and we need to
1591	 * make sure that this tick really expires the timer to avoid
1592	 * a ping pong of the nohz stop code.
1593	 *
1594	 * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
1595	 */
1596	return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
1597}
1598
1599/**
1600 * get_next_timer_interrupt - return the time (clock mono) of the next timer
1601 * @basej:	base time jiffies
1602 * @basem:	base time clock monotonic
1603 *
1604 * Returns the tick aligned clock monotonic time of the next pending
1605 * timer or KTIME_MAX if no timer is pending.
1606 */
1607u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
1608{
1609	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1610	u64 expires = KTIME_MAX;
1611	unsigned long nextevt;
1612	bool is_max_delta;
1613
1614	/*
1615	 * Pretend that there is no timer pending if the cpu is offline.
1616	 * Possible pending timers will be migrated later to an active cpu.
1617	 */
1618	if (cpu_is_offline(smp_processor_id()))
1619		return expires;
1620
1621	raw_spin_lock(&base->lock);
1622	nextevt = __next_timer_interrupt(base);
1623	is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA);
1624	base->next_expiry = nextevt;
 
1625	/*
1626	 * We have a fresh next event. Check whether we can forward the
1627	 * base. We can only do that when @basej is past base->clk
1628	 * otherwise we might rewind base->clk.
1629	 */
1630	if (time_after(basej, base->clk)) {
1631		if (time_after(nextevt, basej))
1632			base->clk = basej;
1633		else if (time_after(nextevt, base->clk))
1634			base->clk = nextevt;
1635	}
1636
1637	if (time_before_eq(nextevt, basej)) {
1638		expires = basem;
1639		base->is_idle = false;
1640	} else {
1641		if (!is_max_delta)
1642			expires = basem + (u64)(nextevt - basej) * TICK_NSEC;
1643		/*
1644		 * If we expect to sleep more than a tick, mark the base idle.
1645		 * Also the tick is stopped so any added timer must forward
1646		 * the base clk itself to keep granularity small. This idle
1647		 * logic is only maintained for the BASE_STD base, deferrable
1648		 * timers may still see large granularity skew (by design).
1649		 */
1650		if ((expires - basem) > TICK_NSEC) {
1651			base->must_forward_clk = true;
1652			base->is_idle = true;
1653		}
1654	}
1655	raw_spin_unlock(&base->lock);
1656
1657	return cmp_next_hrtimer_event(basem, expires);
1658}
1659
1660/**
1661 * timer_clear_idle - Clear the idle state of the timer base
1662 *
1663 * Called with interrupts disabled
1664 */
1665void timer_clear_idle(void)
1666{
1667	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1668
1669	/*
1670	 * We do this unlocked. The worst outcome is a remote enqueue sending
1671	 * a pointless IPI, but taking the lock would just make the window for
1672	 * sending the IPI a few instructions smaller for the cost of taking
1673	 * the lock in the exit from idle path.
1674	 */
1675	base->is_idle = false;
1676}
1677
1678static int collect_expired_timers(struct timer_base *base,
1679				  struct hlist_head *heads)
1680{
1681	unsigned long now = READ_ONCE(jiffies);
1682
1683	/*
1684	 * NOHZ optimization. After a long idle sleep we need to forward the
1685	 * base to current jiffies. Avoid a loop by searching the bitfield for
1686	 * the next expiring timer.
1687	 */
1688	if ((long)(now - base->clk) > 2) {
1689		unsigned long next = __next_timer_interrupt(base);
1690
1691		/*
1692		 * If the next timer is ahead of time forward to current
1693		 * jiffies, otherwise forward to the next expiry time:
1694		 */
1695		if (time_after(next, now)) {
1696			/*
1697			 * The call site will increment base->clk and then
1698			 * terminate the expiry loop immediately.
1699			 */
1700			base->clk = now;
1701			return 0;
1702		}
1703		base->clk = next;
1704	}
1705	return __collect_expired_timers(base, heads);
1706}
1707#else
1708static inline int collect_expired_timers(struct timer_base *base,
1709					 struct hlist_head *heads)
1710{
1711	return __collect_expired_timers(base, heads);
1712}
1713#endif
1714
1715/*
1716 * Called from the timer interrupt handler to charge one tick to the current
1717 * process.  user_tick is 1 if the tick is user time, 0 for system.
1718 */
1719void update_process_times(int user_tick)
1720{
1721	struct task_struct *p = current;
1722
1723	/* Note: this timer irq context must be accounted for as well. */
1724	account_process_tick(p, user_tick);
1725	run_local_timers();
1726	rcu_sched_clock_irq(user_tick);
1727#ifdef CONFIG_IRQ_WORK
1728	if (in_irq())
1729		irq_work_tick();
1730#endif
1731	scheduler_tick();
1732	if (IS_ENABLED(CONFIG_POSIX_TIMERS))
1733		run_posix_cpu_timers();
1734}
1735
1736/**
1737 * __run_timers - run all expired timers (if any) on this CPU.
1738 * @base: the timer vector to be processed.
1739 */
1740static inline void __run_timers(struct timer_base *base)
1741{
1742	struct hlist_head heads[LVL_DEPTH];
1743	int levels;
1744
1745	if (!time_after_eq(jiffies, base->clk))
1746		return;
1747
1748	timer_base_lock_expiry(base);
1749	raw_spin_lock_irq(&base->lock);
1750
1751	/*
1752	 * timer_base::must_forward_clk must be cleared before running
1753	 * timers so that any timer functions that call mod_timer() will
1754	 * not try to forward the base. Idle tracking / clock forwarding
1755	 * logic is only used with BASE_STD timers.
1756	 *
1757	 * The must_forward_clk flag is cleared unconditionally also for
1758	 * the deferrable base. The deferrable base is not affected by idle
1759	 * tracking and never forwarded, so clearing the flag is a NOOP.
1760	 *
1761	 * The fact that the deferrable base is never forwarded can cause
1762	 * large variations in granularity for deferrable timers, but they
1763	 * can be deferred for long periods due to idle anyway.
1764	 */
1765	base->must_forward_clk = false;
1766
1767	while (time_after_eq(jiffies, base->clk)) {
1768
1769		levels = collect_expired_timers(base, heads);
 
 
 
 
 
 
 
 
 
1770		base->clk++;
 
1771
1772		while (levels--)
1773			expire_timers(base, heads + levels);
1774	}
1775	raw_spin_unlock_irq(&base->lock);
1776	timer_base_unlock_expiry(base);
1777}
1778
1779/*
1780 * This function runs timers and the timer-tq in bottom half context.
1781 */
1782static __latent_entropy void run_timer_softirq(struct softirq_action *h)
1783{
1784	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1785
1786	__run_timers(base);
1787	if (IS_ENABLED(CONFIG_NO_HZ_COMMON))
1788		__run_timers(this_cpu_ptr(&timer_bases[BASE_DEF]));
1789}
1790
1791/*
1792 * Called by the local, per-CPU timer interrupt on SMP.
1793 */
1794void run_local_timers(void)
1795{
1796	struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]);
1797
1798	hrtimer_run_queues();
1799	/* Raise the softirq only if required. */
1800	if (time_before(jiffies, base->clk)) {
1801		if (!IS_ENABLED(CONFIG_NO_HZ_COMMON))
1802			return;
1803		/* CPU is awake, so check the deferrable base. */
1804		base++;
1805		if (time_before(jiffies, base->clk))
1806			return;
1807	}
1808	raise_softirq(TIMER_SOFTIRQ);
1809}
1810
1811/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1812 * Since schedule_timeout()'s timer is defined on the stack, it must store
1813 * the target task on the stack as well.
1814 */
1815struct process_timer {
1816	struct timer_list timer;
1817	struct task_struct *task;
1818};
1819
1820static void process_timeout(struct timer_list *t)
1821{
1822	struct process_timer *timeout = from_timer(timeout, t, timer);
1823
1824	wake_up_process(timeout->task);
1825}
1826
1827/**
1828 * schedule_timeout - sleep until timeout
1829 * @timeout: timeout value in jiffies
1830 *
1831 * Make the current task sleep until @timeout jiffies have
1832 * elapsed. The routine will return immediately unless
1833 * the current task state has been set (see set_current_state()).
1834 *
1835 * You can set the task state as follows -
 
 
1836 *
1837 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1838 * pass before the routine returns unless the current task is explicitly
1839 * woken up, (e.g. by wake_up_process())".
1840 *
1841 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1842 * delivered to the current task or the current task is explicitly woken
1843 * up.
1844 *
1845 * The current task state is guaranteed to be TASK_RUNNING when this
1846 * routine returns.
1847 *
1848 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1849 * the CPU away without a bound on the timeout. In this case the return
1850 * value will be %MAX_SCHEDULE_TIMEOUT.
1851 *
1852 * Returns 0 when the timer has expired otherwise the remaining time in
1853 * jiffies will be returned.  In all cases the return value is guaranteed
1854 * to be non-negative.
1855 */
1856signed long __sched schedule_timeout(signed long timeout)
1857{
1858	struct process_timer timer;
1859	unsigned long expire;
1860
1861	switch (timeout)
1862	{
1863	case MAX_SCHEDULE_TIMEOUT:
1864		/*
1865		 * These two special cases are useful to be comfortable
1866		 * in the caller. Nothing more. We could take
1867		 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1868		 * but I' d like to return a valid offset (>=0) to allow
1869		 * the caller to do everything it want with the retval.
1870		 */
1871		schedule();
1872		goto out;
1873	default:
1874		/*
1875		 * Another bit of PARANOID. Note that the retval will be
1876		 * 0 since no piece of kernel is supposed to do a check
1877		 * for a negative retval of schedule_timeout() (since it
1878		 * should never happens anyway). You just have the printk()
1879		 * that will tell you if something is gone wrong and where.
1880		 */
1881		if (timeout < 0) {
1882			printk(KERN_ERR "schedule_timeout: wrong timeout "
1883				"value %lx\n", timeout);
1884			dump_stack();
1885			current->state = TASK_RUNNING;
1886			goto out;
1887		}
1888	}
1889
1890	expire = timeout + jiffies;
1891
1892	timer.task = current;
1893	timer_setup_on_stack(&timer.timer, process_timeout, 0);
1894	__mod_timer(&timer.timer, expire, 0);
1895	schedule();
1896	del_singleshot_timer_sync(&timer.timer);
1897
1898	/* Remove the timer from the object tracker */
1899	destroy_timer_on_stack(&timer.timer);
1900
1901	timeout = expire - jiffies;
1902
1903 out:
1904	return timeout < 0 ? 0 : timeout;
1905}
1906EXPORT_SYMBOL(schedule_timeout);
1907
1908/*
1909 * We can use __set_current_state() here because schedule_timeout() calls
1910 * schedule() unconditionally.
1911 */
1912signed long __sched schedule_timeout_interruptible(signed long timeout)
1913{
1914	__set_current_state(TASK_INTERRUPTIBLE);
1915	return schedule_timeout(timeout);
1916}
1917EXPORT_SYMBOL(schedule_timeout_interruptible);
1918
1919signed long __sched schedule_timeout_killable(signed long timeout)
1920{
1921	__set_current_state(TASK_KILLABLE);
1922	return schedule_timeout(timeout);
1923}
1924EXPORT_SYMBOL(schedule_timeout_killable);
1925
1926signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1927{
1928	__set_current_state(TASK_UNINTERRUPTIBLE);
1929	return schedule_timeout(timeout);
1930}
1931EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1932
1933/*
1934 * Like schedule_timeout_uninterruptible(), except this task will not contribute
1935 * to load average.
1936 */
1937signed long __sched schedule_timeout_idle(signed long timeout)
1938{
1939	__set_current_state(TASK_IDLE);
1940	return schedule_timeout(timeout);
1941}
1942EXPORT_SYMBOL(schedule_timeout_idle);
1943
1944#ifdef CONFIG_HOTPLUG_CPU
1945static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
1946{
1947	struct timer_list *timer;
1948	int cpu = new_base->cpu;
1949
1950	while (!hlist_empty(head)) {
1951		timer = hlist_entry(head->first, struct timer_list, entry);
1952		detach_timer(timer, false);
1953		timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
1954		internal_add_timer(new_base, timer);
1955	}
1956}
1957
1958int timers_prepare_cpu(unsigned int cpu)
1959{
1960	struct timer_base *base;
1961	int b;
1962
1963	for (b = 0; b < NR_BASES; b++) {
1964		base = per_cpu_ptr(&timer_bases[b], cpu);
1965		base->clk = jiffies;
1966		base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
 
 
1967		base->is_idle = false;
1968		base->must_forward_clk = true;
1969	}
1970	return 0;
1971}
1972
1973int timers_dead_cpu(unsigned int cpu)
1974{
1975	struct timer_base *old_base;
1976	struct timer_base *new_base;
1977	int b, i;
1978
1979	BUG_ON(cpu_online(cpu));
1980
1981	for (b = 0; b < NR_BASES; b++) {
1982		old_base = per_cpu_ptr(&timer_bases[b], cpu);
1983		new_base = get_cpu_ptr(&timer_bases[b]);
1984		/*
1985		 * The caller is globally serialized and nobody else
1986		 * takes two locks at once, deadlock is not possible.
1987		 */
1988		raw_spin_lock_irq(&new_base->lock);
1989		raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1990
1991		/*
1992		 * The current CPUs base clock might be stale. Update it
1993		 * before moving the timers over.
1994		 */
1995		forward_timer_base(new_base);
1996
1997		BUG_ON(old_base->running_timer);
 
1998
1999		for (i = 0; i < WHEEL_SIZE; i++)
2000			migrate_timer_list(new_base, old_base->vectors + i);
2001
2002		raw_spin_unlock(&old_base->lock);
2003		raw_spin_unlock_irq(&new_base->lock);
2004		put_cpu_ptr(&timer_bases);
2005	}
2006	return 0;
2007}
2008
2009#endif /* CONFIG_HOTPLUG_CPU */
2010
2011static void __init init_timer_cpu(int cpu)
2012{
2013	struct timer_base *base;
2014	int i;
2015
2016	for (i = 0; i < NR_BASES; i++) {
2017		base = per_cpu_ptr(&timer_bases[i], cpu);
2018		base->cpu = cpu;
2019		raw_spin_lock_init(&base->lock);
2020		base->clk = jiffies;
 
2021		timer_base_init_expiry_lock(base);
2022	}
2023}
2024
2025static void __init init_timer_cpus(void)
2026{
2027	int cpu;
2028
2029	for_each_possible_cpu(cpu)
2030		init_timer_cpu(cpu);
2031}
2032
2033void __init init_timers(void)
2034{
2035	init_timer_cpus();
 
2036	open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
2037}
2038
2039/**
2040 * msleep - sleep safely even with waitqueue interruptions
2041 * @msecs: Time in milliseconds to sleep for
2042 */
2043void msleep(unsigned int msecs)
2044{
2045	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2046
2047	while (timeout)
2048		timeout = schedule_timeout_uninterruptible(timeout);
2049}
2050
2051EXPORT_SYMBOL(msleep);
2052
2053/**
2054 * msleep_interruptible - sleep waiting for signals
2055 * @msecs: Time in milliseconds to sleep for
2056 */
2057unsigned long msleep_interruptible(unsigned int msecs)
2058{
2059	unsigned long timeout = msecs_to_jiffies(msecs) + 1;
2060
2061	while (timeout && !signal_pending(current))
2062		timeout = schedule_timeout_interruptible(timeout);
2063	return jiffies_to_msecs(timeout);
2064}
2065
2066EXPORT_SYMBOL(msleep_interruptible);
2067
2068/**
2069 * usleep_range - Sleep for an approximate time
2070 * @min: Minimum time in usecs to sleep
2071 * @max: Maximum time in usecs to sleep
 
2072 *
2073 * In non-atomic context where the exact wakeup time is flexible, use
2074 * usleep_range() instead of udelay().  The sleep improves responsiveness
2075 * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
2076 * power usage by allowing hrtimers to take advantage of an already-
2077 * scheduled interrupt instead of scheduling a new one just for this sleep.
2078 */
2079void __sched usleep_range(unsigned long min, unsigned long max)
 
2080{
2081	ktime_t exp = ktime_add_us(ktime_get(), min);
2082	u64 delta = (u64)(max - min) * NSEC_PER_USEC;
2083
2084	for (;;) {
2085		__set_current_state(TASK_UNINTERRUPTIBLE);
2086		/* Do not return before the requested sleep time has elapsed */
2087		if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
2088			break;
2089	}
2090}
2091EXPORT_SYMBOL(usleep_range);