| 1 | |
| 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 | #include <linux/sysctl.h>
|
| 48 |
|
| 49 | #include <linux/uaccess.h>
|
| 50 | #include <asm/unistd.h>
|
| 51 | #include <asm/div64.h>
|
| 52 | #include <asm/timex.h>
|
| 53 | #include <asm/io.h>
|
| 54 |
|
| 55 | #include "tick-internal.h"
|
| 56 | #include "timer_migration.h"
|
| 57 |
|
| 58 | #define CREATE_TRACE_POINTS
|
| 59 | #include <trace/events/timer.h>
|
| 60 |
|
| 61 | __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
|
| 62 |
|
| 63 | EXPORT_SYMBOL(jiffies_64);
|
| 64 |
|
| 65 | |
| 66 | * The timer wheel has LVL_DEPTH array levels. Each level provides an array of
|
| 67 | * LVL_SIZE buckets. Each level is driven by its own clock and therefore each
|
| 68 | * level has a different granularity.
|
| 69 | *
|
| 70 | * The level granularity is: LVL_CLK_DIV ^ level
|
| 71 | * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level)
|
| 72 | *
|
| 73 | * The array level of a newly armed timer depends on the relative expiry
|
| 74 | * time. The farther the expiry time is away the higher the array level and
|
| 75 | * therefore the granularity becomes.
|
| 76 | *
|
| 77 | * Contrary to the original timer wheel implementation, which aims for 'exact'
|
| 78 | * expiry of the timers, this implementation removes the need for recascading
|
| 79 | * the timers into the lower array levels. The previous 'classic' timer wheel
|
| 80 | * implementation of the kernel already violated the 'exact' expiry by adding
|
| 81 | * slack to the expiry time to provide batched expiration. The granularity
|
| 82 | * levels provide implicit batching.
|
| 83 | *
|
| 84 | * This is an optimization of the original timer wheel implementation for the
|
| 85 | * majority of the timer wheel use cases: timeouts. The vast majority of
|
| 86 | * timeout timers (networking, disk I/O ...) are canceled before expiry. If
|
| 87 | * the timeout expires it indicates that normal operation is disturbed, so it
|
| 88 | * does not matter much whether the timeout comes with a slight delay.
|
| 89 | *
|
| 90 | * The only exception to this are networking timers with a small expiry
|
| 91 | * time. They rely on the granularity. Those fit into the first wheel level,
|
| 92 | * which has HZ granularity.
|
| 93 | *
|
| 94 | * We don't have cascading anymore. timers with a expiry time above the
|
| 95 | * capacity of the last wheel level are force expired at the maximum timeout
|
| 96 | * value of the last wheel level. From data sampling we know that the maximum
|
| 97 | * value observed is 5 days (network connection tracking), so this should not
|
| 98 | * be an issue.
|
| 99 | *
|
| 100 | * The currently chosen array constants values are a good compromise between
|
| 101 | * array size and granularity.
|
| 102 | *
|
| 103 | * This results in the following granularity and range levels:
|
| 104 | *
|
| 105 | * HZ 1000 steps
|
| 106 | * Level Offset Granularity Range
|
| 107 | * 0 0 1 ms 0 ms - 63 ms
|
| 108 | * 1 64 8 ms 64 ms - 511 ms
|
| 109 | * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s)
|
| 110 | * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s)
|
| 111 | * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m)
|
| 112 | * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m)
|
| 113 | * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h)
|
| 114 | * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d)
|
| 115 | * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d)
|
| 116 | *
|
| 117 | * HZ 300
|
| 118 | * Level Offset Granularity Range
|
| 119 | * 0 0 3 ms 0 ms - 210 ms
|
| 120 | * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s)
|
| 121 | * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s)
|
| 122 | * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m)
|
| 123 | * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m)
|
| 124 | * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h)
|
| 125 | * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h)
|
| 126 | * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d)
|
| 127 | * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d)
|
| 128 | *
|
| 129 | * HZ 250
|
| 130 | * Level Offset Granularity Range
|
| 131 | * 0 0 4 ms 0 ms - 255 ms
|
| 132 | * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s)
|
| 133 | * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s)
|
| 134 | * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m)
|
| 135 | * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m)
|
| 136 | * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h)
|
| 137 | * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h)
|
| 138 | * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d)
|
| 139 | * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d)
|
| 140 | *
|
| 141 | * HZ 100
|
| 142 | * Level Offset Granularity Range
|
| 143 | * 0 0 10 ms 0 ms - 630 ms
|
| 144 | * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s)
|
| 145 | * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s)
|
| 146 | * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m)
|
| 147 | * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m)
|
| 148 | * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h)
|
| 149 | * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d)
|
| 150 | * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d)
|
| 151 |
|
| 152 |
|
| 153 |
|
| 154 | #define LVL_CLK_SHIFT 3
|
| 155 | #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT)
|
| 156 | #define LVL_CLK_MASK (LVL_CLK_DIV - 1)
|
| 157 | #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT)
|
| 158 | #define LVL_GRAN(n) (1UL << LVL_SHIFT(n))
|
| 159 |
|
| 160 | |
| 161 | * The time start value for each level to select the bucket at enqueue
|
| 162 | * time. We start from the last possible delta of the previous level
|
| 163 | * so that we can later add an extra LVL_GRAN(n) to n (see calc_index()).
|
| 164 |
|
| 165 | #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT))
|
| 166 |
|
| 167 |
|
| 168 | #define LVL_BITS 6
|
| 169 | #define LVL_SIZE (1UL << LVL_BITS)
|
| 170 | #define LVL_MASK (LVL_SIZE - 1)
|
| 171 | #define LVL_OFFS(n) ((n) * LVL_SIZE)
|
| 172 |
|
| 173 |
|
| 174 | #if HZ > 100
|
| 175 | # define LVL_DEPTH 9
|
| 176 | # else
|
| 177 | # define LVL_DEPTH 8
|
| 178 | #endif
|
| 179 |
|
| 180 |
|
| 181 | #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH))
|
| 182 | #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1))
|
| 183 |
|
| 184 | |
| 185 | * The resulting wheel size. If NOHZ is configured we allocate two
|
| 186 | * wheels so we have a separate storage for the deferrable timers.
|
| 187 |
|
| 188 | #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH)
|
| 189 |
|
| 190 | #ifdef CONFIG_NO_HZ_COMMON
|
| 191 | |
| 192 | * If multiple bases need to be locked, use the base ordering for lock
|
| 193 | * nesting, i.e. lowest number first.
|
| 194 |
|
| 195 | # define NR_BASES 3
|
| 196 | # define BASE_LOCAL 0
|
| 197 | # define BASE_GLOBAL 1
|
| 198 | # define BASE_DEF 2
|
| 199 | #else
|
| 200 | # define NR_BASES 1
|
| 201 | # define BASE_LOCAL 0
|
| 202 | # define BASE_GLOBAL 0
|
| 203 | # define BASE_DEF 0
|
| 204 | #endif
|
| 205 |
|
| 206 | |
| 207 | * struct timer_base - Per CPU timer base (number of base depends on config)
|
| 208 | * @lock: Lock protecting the timer_base
|
| 209 | * @running_timer: When expiring timers, the lock is dropped. To make
|
| 210 | * sure not to race against deleting/modifying a
|
| 211 | * currently running timer, the pointer is set to the
|
| 212 | * timer, which expires at the moment. If no timer is
|
| 213 | * running, the pointer is NULL.
|
| 214 | * @expiry_lock: PREEMPT_RT only: Lock is taken in softirq around
|
| 215 | * timer expiry callback execution and when trying to
|
| 216 | * delete a running timer and it wasn't successful in
|
| 217 | * the first glance. It prevents priority inversion
|
| 218 | * when callback was preempted on a remote CPU and a
|
| 219 | * caller tries to delete the running timer. It also
|
| 220 | * prevents a life lock, when the task which tries to
|
| 221 | * delete a timer preempted the softirq thread which
|
| 222 | * is running the timer callback function.
|
| 223 | * @timer_waiters: PREEMPT_RT only: Tells, if there is a waiter
|
| 224 | * waiting for the end of the timer callback function
|
| 225 | * execution.
|
| 226 | * @clk: clock of the timer base; is updated before enqueue
|
| 227 | * of a timer; during expiry, it is 1 offset ahead of
|
| 228 | * jiffies to avoid endless requeuing to current
|
| 229 | * jiffies
|
| 230 | * @next_expiry: expiry value of the first timer; it is updated when
|
| 231 | * finding the next timer and during enqueue; the
|
| 232 | * value is not valid, when next_expiry_recalc is set
|
| 233 | * @cpu: Number of CPU the timer base belongs to
|
| 234 | * @next_expiry_recalc: States, whether a recalculation of next_expiry is
|
| 235 | * required. Value is set true, when a timer was
|
| 236 | * deleted.
|
| 237 | * @is_idle: Is set, when timer_base is idle. It is triggered by NOHZ
|
| 238 | * code. This state is only used in standard
|
| 239 | * base. Deferrable timers, which are enqueued remotely
|
| 240 | * never wake up an idle CPU. So no matter of supporting it
|
| 241 | * for this base.
|
| 242 | * @timers_pending: Is set, when a timer is pending in the base. It is only
|
| 243 | * reliable when next_expiry_recalc is not set.
|
| 244 | * @pending_map: bitmap of the timer wheel; each bit reflects a
|
| 245 | * bucket of the wheel. When a bit is set, at least a
|
| 246 | * single timer is enqueued in the related bucket.
|
| 247 | * @vectors: Array of lists; Each array member reflects a bucket
|
| 248 | * of the timer wheel. The list contains all timers
|
| 249 | * which are enqueued into a specific bucket.
|
| 250 |
|
| 251 | struct timer_base {
|
| 252 | raw_spinlock_t lock;
|
| 253 | struct timer_list *running_timer;
|
| 254 | #ifdef CONFIG_PREEMPT_RT
|
| 255 | spinlock_t expiry_lock;
|
| 256 | atomic_t timer_waiters;
|
| 257 | #endif
|
| 258 | unsigned long clk;
|
| 259 | unsigned long next_expiry;
|
| 260 | unsigned int cpu;
|
| 261 | bool next_expiry_recalc;
|
| 262 | bool is_idle;
|
| 263 | bool timers_pending;
|
| 264 | DECLARE_BITMAP(pending_map, WHEEL_SIZE);
|
| 265 | struct hlist_head vectors[WHEEL_SIZE];
|
| 266 | } ____cacheline_aligned;
|
| 267 |
|
| 268 | static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]);
|
| 269 |
|
| 270 | #ifdef CONFIG_NO_HZ_COMMON
|
| 271 |
|
| 272 | static DEFINE_STATIC_KEY_FALSE(timers_nohz_active);
|
| 273 | static DEFINE_MUTEX(timer_keys_mutex);
|
| 274 |
|
| 275 | static void timer_update_keys(struct work_struct *work);
|
| 276 | static DECLARE_WORK(timer_update_work, timer_update_keys);
|
| 277 |
|
| 278 | #ifdef CONFIG_SMP
|
| 279 | static unsigned int sysctl_timer_migration = 1;
|
| 280 |
|
| 281 | DEFINE_STATIC_KEY_FALSE(timers_migration_enabled);
|
| 282 |
|
| 283 | static void timers_update_migration(void)
|
| 284 | {
|
| 285 | if (sysctl_timer_migration && tick_nohz_active)
|
| 286 | static_branch_enable(&timers_migration_enabled);
|
| 287 | else
|
| 288 | static_branch_disable(&timers_migration_enabled);
|
| 289 | }
|
| 290 |
|
| 291 | #ifdef CONFIG_SYSCTL
|
| 292 | static int timer_migration_handler(const struct ctl_table *table, int write,
|
| 293 | void *buffer, size_t *lenp, loff_t *ppos)
|
| 294 | {
|
| 295 | int ret;
|
| 296 |
|
| 297 | mutex_lock(&timer_keys_mutex);
|
| 298 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
|
| 299 | if (!ret && write)
|
| 300 | timers_update_migration();
|
| 301 | mutex_unlock(&timer_keys_mutex);
|
| 302 | return ret;
|
| 303 | }
|
| 304 |
|
| 305 | static struct ctl_table timer_sysctl[] = {
|
| 306 | {
|
| 307 | .procname = "timer_migration",
|
| 308 | .data = &sysctl_timer_migration,
|
| 309 | .maxlen = sizeof(unsigned int),
|
| 310 | .mode = 0644,
|
| 311 | .proc_handler = timer_migration_handler,
|
| 312 | .extra1 = SYSCTL_ZERO,
|
| 313 | .extra2 = SYSCTL_ONE,
|
| 314 | },
|
| 315 | };
|
| 316 |
|
| 317 | static int __init timer_sysctl_init(void)
|
| 318 | {
|
| 319 | register_sysctl("kernel", timer_sysctl);
|
| 320 | return 0;
|
| 321 | }
|
| 322 | device_initcall(timer_sysctl_init);
|
| 323 | #endif
|
| 324 | #else
|
| 325 | static inline void timers_update_migration(void) { }
|
| 326 | #endif
|
| 327 |
|
| 328 | static void timer_update_keys(struct work_struct *work)
|
| 329 | {
|
| 330 | mutex_lock(&timer_keys_mutex);
|
| 331 | timers_update_migration();
|
| 332 | static_branch_enable(&timers_nohz_active);
|
| 333 | mutex_unlock(&timer_keys_mutex);
|
| 334 | }
|
| 335 |
|
| 336 | void timers_update_nohz(void)
|
| 337 | {
|
| 338 | schedule_work(&timer_update_work);
|
| 339 | }
|
| 340 |
|
| 341 | static inline bool is_timers_nohz_active(void)
|
| 342 | {
|
| 343 | return static_branch_unlikely(&timers_nohz_active);
|
| 344 | }
|
| 345 | #else
|
| 346 | static inline bool is_timers_nohz_active(void) { return false; }
|
| 347 | #endif
|
| 348 |
|
| 349 | static unsigned long round_jiffies_common(unsigned long j, int cpu,
|
| 350 | bool force_up)
|
| 351 | {
|
| 352 | int rem;
|
| 353 | unsigned long original = j;
|
| 354 |
|
| 355 | |
| 356 | * We don't want all cpus firing their timers at once hitting the
|
| 357 | * same lock or cachelines, so we skew each extra cpu with an extra
|
| 358 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which
|
| 359 | * already did this.
|
| 360 | * The skew is done by adding 3*cpunr, then round, then subtract this
|
| 361 | * extra offset again.
|
| 362 |
|
| 363 | j += cpu * 3;
|
| 364 |
|
| 365 | rem = j % HZ;
|
| 366 |
|
| 367 | |
| 368 | * If the target jiffy is just after a whole second (which can happen
|
| 369 | * due to delays of the timer irq, long irq off times etc etc) then
|
| 370 | * we should round down to the whole second, not up. Use 1/4th second
|
| 371 | * as cutoff for this rounding as an extreme upper bound for this.
|
| 372 | * But never round down if @force_up is set.
|
| 373 |
|
| 374 | if (rem < HZ/4 && !force_up)
|
| 375 | j = j - rem;
|
| 376 | else
|
| 377 | j = j - rem + HZ;
|
| 378 |
|
| 379 |
|
| 380 | j -= cpu * 3;
|
| 381 |
|
| 382 | |
| 383 | * Make sure j is still in the future. Otherwise return the
|
| 384 | * unmodified value.
|
| 385 |
|
| 386 | return time_is_after_jiffies(j) ? j : original;
|
| 387 | }
|
| 388 |
|
| 389 | |
| 390 | * __round_jiffies - function to round jiffies to a full second
|
| 391 | * @j: the time in (absolute) jiffies that should be rounded
|
| 392 | * @cpu: the processor number on which the timeout will happen
|
| 393 | *
|
| 394 | * __round_jiffies() rounds an absolute time in the future (in jiffies)
|
| 395 | * up or down to (approximately) full seconds. This is useful for timers
|
| 396 | * for which the exact time they fire does not matter too much, as long as
|
| 397 | * they fire approximately every X seconds.
|
| 398 | *
|
| 399 | * By rounding these timers to whole seconds, all such timers will fire
|
| 400 | * at the same time, rather than at various times spread out. The goal
|
| 401 | * of this is to have the CPU wake up less, which saves power.
|
| 402 | *
|
| 403 | * The exact rounding is skewed for each processor to avoid all
|
| 404 | * processors firing at the exact same time, which could lead
|
| 405 | * to lock contention or spurious cache line bouncing.
|
| 406 | *
|
| 407 | * The return value is the rounded version of the @j parameter.
|
| 408 |
|
| 409 | unsigned long __round_jiffies(unsigned long j, int cpu)
|
| 410 | {
|
| 411 | return round_jiffies_common(j, cpu, false);
|
| 412 | }
|
| 413 | EXPORT_SYMBOL_GPL(__round_jiffies);
|
| 414 |
|
| 415 | |
| 416 | * __round_jiffies_relative - function to round jiffies to a full second
|
| 417 | * @j: the time in (relative) jiffies that should be rounded
|
| 418 | * @cpu: the processor number on which the timeout will happen
|
| 419 | *
|
| 420 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
|
| 421 | * up or down to (approximately) full seconds. This is useful for timers
|
| 422 | * for which the exact time they fire does not matter too much, as long as
|
| 423 | * they fire approximately every X seconds.
|
| 424 | *
|
| 425 | * By rounding these timers to whole seconds, all such timers will fire
|
| 426 | * at the same time, rather than at various times spread out. The goal
|
| 427 | * of this is to have the CPU wake up less, which saves power.
|
| 428 | *
|
| 429 | * The exact rounding is skewed for each processor to avoid all
|
| 430 | * processors firing at the exact same time, which could lead
|
| 431 | * to lock contention or spurious cache line bouncing.
|
| 432 | *
|
| 433 | * The return value is the rounded version of the @j parameter.
|
| 434 |
|
| 435 | unsigned long __round_jiffies_relative(unsigned long j, int cpu)
|
| 436 | {
|
| 437 | unsigned long j0 = jiffies;
|
| 438 |
|
| 439 |
|
| 440 | return round_jiffies_common(j + j0, cpu, false) - j0;
|
| 441 | }
|
| 442 | EXPORT_SYMBOL_GPL(__round_jiffies_relative);
|
| 443 |
|
| 444 | |
| 445 | * round_jiffies - function to round jiffies to a full second
|
| 446 | * @j: the time in (absolute) jiffies that should be rounded
|
| 447 | *
|
| 448 | * round_jiffies() rounds an absolute time in the future (in jiffies)
|
| 449 | * up or down to (approximately) full seconds. This is useful for timers
|
| 450 | * for which the exact time they fire does not matter too much, as long as
|
| 451 | * they fire approximately every X seconds.
|
| 452 | *
|
| 453 | * By rounding these timers to whole seconds, all such timers will fire
|
| 454 | * at the same time, rather than at various times spread out. The goal
|
| 455 | * of this is to have the CPU wake up less, which saves power.
|
| 456 | *
|
| 457 | * The return value is the rounded version of the @j parameter.
|
| 458 |
|
| 459 | unsigned long round_jiffies(unsigned long j)
|
| 460 | {
|
| 461 | return round_jiffies_common(j, raw_smp_processor_id(), false);
|
| 462 | }
|
| 463 | EXPORT_SYMBOL_GPL(round_jiffies);
|
| 464 |
|
| 465 | |
| 466 | * round_jiffies_relative - function to round jiffies to a full second
|
| 467 | * @j: the time in (relative) jiffies that should be rounded
|
| 468 | *
|
| 469 | * round_jiffies_relative() rounds a time delta in the future (in jiffies)
|
| 470 | * up or down to (approximately) full seconds. This is useful for timers
|
| 471 | * for which the exact time they fire does not matter too much, as long as
|
| 472 | * they fire approximately every X seconds.
|
| 473 | *
|
| 474 | * By rounding these timers to whole seconds, all such timers will fire
|
| 475 | * at the same time, rather than at various times spread out. The goal
|
| 476 | * of this is to have the CPU wake up less, which saves power.
|
| 477 | *
|
| 478 | * The return value is the rounded version of the @j parameter.
|
| 479 |
|
| 480 | unsigned long round_jiffies_relative(unsigned long j)
|
| 481 | {
|
| 482 | return __round_jiffies_relative(j, raw_smp_processor_id());
|
| 483 | }
|
| 484 | EXPORT_SYMBOL_GPL(round_jiffies_relative);
|
| 485 |
|
| 486 | |
| 487 | * __round_jiffies_up - function to round jiffies up to a full second
|
| 488 | * @j: the time in (absolute) jiffies that should be rounded
|
| 489 | * @cpu: the processor number on which the timeout will happen
|
| 490 | *
|
| 491 | * This is the same as __round_jiffies() except that it will never
|
| 492 | * round down. This is useful for timeouts for which the exact time
|
| 493 | * of firing does not matter too much, as long as they don't fire too
|
| 494 | * early.
|
| 495 |
|
| 496 | unsigned long __round_jiffies_up(unsigned long j, int cpu)
|
| 497 | {
|
| 498 | return round_jiffies_common(j, cpu, true);
|
| 499 | }
|
| 500 | EXPORT_SYMBOL_GPL(__round_jiffies_up);
|
| 501 |
|
| 502 | |
| 503 | * __round_jiffies_up_relative - function to round jiffies up to a full second
|
| 504 | * @j: the time in (relative) jiffies that should be rounded
|
| 505 | * @cpu: the processor number on which the timeout will happen
|
| 506 | *
|
| 507 | * This is the same as __round_jiffies_relative() except that it will never
|
| 508 | * round down. This is useful for timeouts for which the exact time
|
| 509 | * of firing does not matter too much, as long as they don't fire too
|
| 510 | * early.
|
| 511 |
|
| 512 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
|
| 513 | {
|
| 514 | unsigned long j0 = jiffies;
|
| 515 |
|
| 516 |
|
| 517 | return round_jiffies_common(j + j0, cpu, true) - j0;
|
| 518 | }
|
| 519 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
|
| 520 |
|
| 521 | |
| 522 | * round_jiffies_up - function to round jiffies up to a full second
|
| 523 | * @j: the time in (absolute) jiffies that should be rounded
|
| 524 | *
|
| 525 | * This is the same as round_jiffies() except that it will never
|
| 526 | * round down. This is useful for timeouts for which the exact time
|
| 527 | * of firing does not matter too much, as long as they don't fire too
|
| 528 | * early.
|
| 529 |
|
| 530 | unsigned long round_jiffies_up(unsigned long j)
|
| 531 | {
|
| 532 | return round_jiffies_common(j, raw_smp_processor_id(), true);
|
| 533 | }
|
| 534 | EXPORT_SYMBOL_GPL(round_jiffies_up);
|
| 535 |
|
| 536 | |
| 537 | * round_jiffies_up_relative - function to round jiffies up to a full second
|
| 538 | * @j: the time in (relative) jiffies that should be rounded
|
| 539 | *
|
| 540 | * This is the same as round_jiffies_relative() except that it will never
|
| 541 | * round down. This is useful for timeouts for which the exact time
|
| 542 | * of firing does not matter too much, as long as they don't fire too
|
| 543 | * early.
|
| 544 |
|
| 545 | unsigned long round_jiffies_up_relative(unsigned long j)
|
| 546 | {
|
| 547 | return __round_jiffies_up_relative(j, raw_smp_processor_id());
|
| 548 | }
|
| 549 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
|
| 550 |
|
| 551 |
|
| 552 | static inline unsigned int timer_get_idx(struct timer_list *timer)
|
| 553 | {
|
| 554 | return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT;
|
| 555 | }
|
| 556 |
|
| 557 | static inline void timer_set_idx(struct timer_list *timer, unsigned int idx)
|
| 558 | {
|
| 559 | timer->flags = (timer->flags & ~TIMER_ARRAYMASK) |
|
| 560 | idx << TIMER_ARRAYSHIFT;
|
| 561 | }
|
| 562 |
|
| 563 | |
| 564 | * Helper function to calculate the array index for a given expiry
|
| 565 | * time.
|
| 566 |
|
| 567 | static inline unsigned calc_index(unsigned long expires, unsigned lvl,
|
| 568 | unsigned long *bucket_expiry)
|
| 569 | {
|
| 570 |
|
| 571 | |
| 572 | * The timer wheel has to guarantee that a timer does not fire
|
| 573 | * early. Early expiry can happen due to:
|
| 574 | * - Timer is armed at the edge of a tick
|
| 575 | * - Truncation of the expiry time in the outer wheel levels
|
| 576 | *
|
| 577 | * Round up with level granularity to prevent this.
|
| 578 |
|
| 579 | expires = (expires >> LVL_SHIFT(lvl)) + 1;
|
| 580 | *bucket_expiry = expires << LVL_SHIFT(lvl);
|
| 581 | return LVL_OFFS(lvl) + (expires & LVL_MASK);
|
| 582 | }
|
| 583 |
|
| 584 | static int calc_wheel_index(unsigned long expires, unsigned long clk,
|
| 585 | unsigned long *bucket_expiry)
|
| 586 | {
|
| 587 | unsigned long delta = expires - clk;
|
| 588 | unsigned int idx;
|
| 589 |
|
| 590 | if (delta < LVL_START(1)) {
|
| 591 | idx = calc_index(expires, 0, bucket_expiry);
|
| 592 | } else if (delta < LVL_START(2)) {
|
| 593 | idx = calc_index(expires, 1, bucket_expiry);
|
| 594 | } else if (delta < LVL_START(3)) {
|
| 595 | idx = calc_index(expires, 2, bucket_expiry);
|
| 596 | } else if (delta < LVL_START(4)) {
|
| 597 | idx = calc_index(expires, 3, bucket_expiry);
|
| 598 | } else if (delta < LVL_START(5)) {
|
| 599 | idx = calc_index(expires, 4, bucket_expiry);
|
| 600 | } else if (delta < LVL_START(6)) {
|
| 601 | idx = calc_index(expires, 5, bucket_expiry);
|
| 602 | } else if (delta < LVL_START(7)) {
|
| 603 | idx = calc_index(expires, 6, bucket_expiry);
|
| 604 | } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) {
|
| 605 | idx = calc_index(expires, 7, bucket_expiry);
|
| 606 | } else if ((long) delta < 0) {
|
| 607 | idx = clk & LVL_MASK;
|
| 608 | *bucket_expiry = clk;
|
| 609 | } else {
|
| 610 | |
| 611 | * Force expire obscene large timeouts to expire at the
|
| 612 | * capacity limit of the wheel.
|
| 613 |
|
| 614 | if (delta >= WHEEL_TIMEOUT_CUTOFF)
|
| 615 | expires = clk + WHEEL_TIMEOUT_MAX;
|
| 616 |
|
| 617 | idx = calc_index(expires, LVL_DEPTH - 1, bucket_expiry);
|
| 618 | }
|
| 619 | return idx;
|
| 620 | }
|
| 621 |
|
| 622 | static void
|
| 623 | trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer)
|
| 624 | {
|
| 625 | |
| 626 | * Deferrable timers do not prevent the CPU from entering dynticks and
|
| 627 | * are not taken into account on the idle/nohz_full path. An IPI when a
|
| 628 | * new deferrable timer is enqueued will wake up the remote CPU but
|
| 629 | * nothing will be done with the deferrable timer base. Therefore skip
|
| 630 | * the remote IPI for deferrable timers completely.
|
| 631 |
|
| 632 | if (!is_timers_nohz_active() || timer->flags & TIMER_DEFERRABLE)
|
| 633 | return;
|
| 634 |
|
| 635 | |
| 636 | * We might have to IPI the remote CPU if the base is idle and the
|
| 637 | * timer is pinned. If it is a non pinned timer, it is only queued
|
| 638 | * on the remote CPU, when timer was running during queueing. Then
|
| 639 | * everything is handled by remote CPU anyway. If the other CPU is
|
| 640 | * on the way to idle then it can't set base->is_idle as we hold
|
| 641 | * the base lock:
|
| 642 |
|
| 643 | if (base->is_idle) {
|
| 644 | WARN_ON_ONCE(!(timer->flags & TIMER_PINNED ||
|
| 645 | tick_nohz_full_cpu(base->cpu)));
|
| 646 | wake_up_nohz_cpu(base->cpu);
|
| 647 | }
|
| 648 | }
|
| 649 |
|
| 650 | |
| 651 | * Enqueue the timer into the hash bucket, mark it pending in
|
| 652 | * the bitmap, store the index in the timer flags then wake up
|
| 653 | * the target CPU if needed.
|
| 654 |
|
| 655 | static void enqueue_timer(struct timer_base *base, struct timer_list *timer,
|
| 656 | unsigned int idx, unsigned long bucket_expiry)
|
| 657 | {
|
| 658 |
|
| 659 | hlist_add_head(&timer->entry, base->vectors + idx);
|
| 660 | __set_bit(idx, base->pending_map);
|
| 661 | timer_set_idx(timer, idx);
|
| 662 |
|
| 663 | trace_timer_start(timer, bucket_expiry);
|
| 664 |
|
| 665 | |
| 666 | * Check whether this is the new first expiring timer. The
|
| 667 | * effective expiry time of the timer is required here
|
| 668 | * (bucket_expiry) instead of timer->expires.
|
| 669 |
|
| 670 | if (time_before(bucket_expiry, base->next_expiry)) {
|
| 671 | |
| 672 | * Set the next expiry time and kick the CPU so it
|
| 673 | * can reevaluate the wheel:
|
| 674 |
|
| 675 | WRITE_ONCE(base->next_expiry, bucket_expiry);
|
| 676 | base->timers_pending = true;
|
| 677 | base->next_expiry_recalc = false;
|
| 678 | trigger_dyntick_cpu(base, timer);
|
| 679 | }
|
| 680 | }
|
| 681 |
|
| 682 | static void internal_add_timer(struct timer_base *base, struct timer_list *timer)
|
| 683 | {
|
| 684 | unsigned long bucket_expiry;
|
| 685 | unsigned int idx;
|
| 686 |
|
| 687 | idx = calc_wheel_index(timer->expires, base->clk, &bucket_expiry);
|
| 688 | enqueue_timer(base, timer, idx, bucket_expiry);
|
| 689 | }
|
| 690 |
|
| 691 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
|
| 692 |
|
| 693 | static const struct debug_obj_descr timer_debug_descr;
|
| 694 |
|
| 695 | struct timer_hint {
|
| 696 | void (*function)(struct timer_list *t);
|
| 697 | long offset;
|
| 698 | };
|
| 699 |
|
| 700 | #define TIMER_HINT(fn, container, timr, hintfn) \
|
| 701 | { \
|
| 702 | .function = fn, \
|
| 703 | .offset = offsetof(container, hintfn) - \
|
| 704 | offsetof(container, timr) \
|
| 705 | }
|
| 706 |
|
| 707 | static const struct timer_hint timer_hints[] = {
|
| 708 | TIMER_HINT(delayed_work_timer_fn,
|
| 709 | struct delayed_work, timer, work.func),
|
| 710 | TIMER_HINT(kthread_delayed_work_timer_fn,
|
| 711 | struct kthread_delayed_work, timer, work.func),
|
| 712 | };
|
| 713 |
|
| 714 | static void *timer_debug_hint(void *addr)
|
| 715 | {
|
| 716 | struct timer_list *timer = addr;
|
| 717 | int i;
|
| 718 |
|
| 719 | for (i = 0; i < ARRAY_SIZE(timer_hints); i++) {
|
| 720 | if (timer_hints[i].function == timer->function) {
|
| 721 | void (**fn)(void) = addr + timer_hints[i].offset;
|
| 722 |
|
| 723 | return *fn;
|
| 724 | }
|
| 725 | }
|
| 726 |
|
| 727 | return timer->function;
|
| 728 | }
|
| 729 |
|
| 730 | static bool timer_is_static_object(void *addr)
|
| 731 | {
|
| 732 | struct timer_list *timer = addr;
|
| 733 |
|
| 734 | return (timer->entry.pprev == NULL &&
|
| 735 | timer->entry.next == TIMER_ENTRY_STATIC);
|
| 736 | }
|
| 737 |
|
| 738 | |
| 739 | * timer_fixup_init is called when:
|
| 740 | * - an active object is initialized
|
| 741 |
|
| 742 | static bool timer_fixup_init(void *addr, enum debug_obj_state state)
|
| 743 | {
|
| 744 | struct timer_list *timer = addr;
|
| 745 |
|
| 746 | switch (state) {
|
| 747 | case ODEBUG_STATE_ACTIVE:
|
| 748 | del_timer_sync(timer);
|
| 749 | debug_object_init(timer, &timer_debug_descr);
|
| 750 | return true;
|
| 751 | default:
|
| 752 | return false;
|
| 753 | }
|
| 754 | }
|
| 755 |
|
| 756 |
|
| 757 | static void stub_timer(struct timer_list *unused)
|
| 758 | {
|
| 759 | WARN_ON(1);
|
| 760 | }
|
| 761 |
|
| 762 | |
| 763 | * timer_fixup_activate is called when:
|
| 764 | * - an active object is activated
|
| 765 | * - an unknown non-static object is activated
|
| 766 |
|
| 767 | static bool timer_fixup_activate(void *addr, enum debug_obj_state state)
|
| 768 | {
|
| 769 | struct timer_list *timer = addr;
|
| 770 |
|
| 771 | switch (state) {
|
| 772 | case ODEBUG_STATE_NOTAVAILABLE:
|
| 773 | timer_setup(timer, stub_timer, 0);
|
| 774 | return true;
|
| 775 |
|
| 776 | case ODEBUG_STATE_ACTIVE:
|
| 777 | WARN_ON(1);
|
| 778 | fallthrough;
|
| 779 | default:
|
| 780 | return false;
|
| 781 | }
|
| 782 | }
|
| 783 |
|
| 784 | |
| 785 | * timer_fixup_free is called when:
|
| 786 | * - an active object is freed
|
| 787 |
|
| 788 | static bool timer_fixup_free(void *addr, enum debug_obj_state state)
|
| 789 | {
|
| 790 | struct timer_list *timer = addr;
|
| 791 |
|
| 792 | switch (state) {
|
| 793 | case ODEBUG_STATE_ACTIVE:
|
| 794 | del_timer_sync(timer);
|
| 795 | debug_object_free(timer, &timer_debug_descr);
|
| 796 | return true;
|
| 797 | default:
|
| 798 | return false;
|
| 799 | }
|
| 800 | }
|
| 801 |
|
| 802 | |
| 803 | * timer_fixup_assert_init is called when:
|
| 804 | * - an untracked/uninit-ed object is found
|
| 805 |
|
| 806 | static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state)
|
| 807 | {
|
| 808 | struct timer_list *timer = addr;
|
| 809 |
|
| 810 | switch (state) {
|
| 811 | case ODEBUG_STATE_NOTAVAILABLE:
|
| 812 | timer_setup(timer, stub_timer, 0);
|
| 813 | return true;
|
| 814 | default:
|
| 815 | return false;
|
| 816 | }
|
| 817 | }
|
| 818 |
|
| 819 | static const struct debug_obj_descr timer_debug_descr = {
|
| 820 | .name = "timer_list",
|
| 821 | .debug_hint = timer_debug_hint,
|
| 822 | .is_static_object = timer_is_static_object,
|
| 823 | .fixup_init = timer_fixup_init,
|
| 824 | .fixup_activate = timer_fixup_activate,
|
| 825 | .fixup_free = timer_fixup_free,
|
| 826 | .fixup_assert_init = timer_fixup_assert_init,
|
| 827 | };
|
| 828 |
|
| 829 | static inline void debug_timer_init(struct timer_list *timer)
|
| 830 | {
|
| 831 | debug_object_init(timer, &timer_debug_descr);
|
| 832 | }
|
| 833 |
|
| 834 | static inline void debug_timer_activate(struct timer_list *timer)
|
| 835 | {
|
| 836 | debug_object_activate(timer, &timer_debug_descr);
|
| 837 | }
|
| 838 |
|
| 839 | static inline void debug_timer_deactivate(struct timer_list *timer)
|
| 840 | {
|
| 841 | debug_object_deactivate(timer, &timer_debug_descr);
|
| 842 | }
|
| 843 |
|
| 844 | static inline void debug_timer_assert_init(struct timer_list *timer)
|
| 845 | {
|
| 846 | debug_object_assert_init(timer, &timer_debug_descr);
|
| 847 | }
|
| 848 |
|
| 849 | static void do_init_timer(struct timer_list *timer,
|
| 850 | void (*func)(struct timer_list *),
|
| 851 | unsigned int flags,
|
| 852 | const char *name, struct lock_class_key *key);
|
| 853 |
|
| 854 | void init_timer_on_stack_key(struct timer_list *timer,
|
| 855 | void (*func)(struct timer_list *),
|
| 856 | unsigned int flags,
|
| 857 | const char *name, struct lock_class_key *key)
|
| 858 | {
|
| 859 | debug_object_init_on_stack(timer, &timer_debug_descr);
|
| 860 | do_init_timer(timer, func, flags, name, key);
|
| 861 | }
|
| 862 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
|
| 863 |
|
| 864 | void destroy_timer_on_stack(struct timer_list *timer)
|
| 865 | {
|
| 866 | debug_object_free(timer, &timer_debug_descr);
|
| 867 | }
|
| 868 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
|
| 869 |
|
| 870 | #else
|
| 871 | static inline void debug_timer_init(struct timer_list *timer) { }
|
| 872 | static inline void debug_timer_activate(struct timer_list *timer) { }
|
| 873 | static inline void debug_timer_deactivate(struct timer_list *timer) { }
|
| 874 | static inline void debug_timer_assert_init(struct timer_list *timer) { }
|
| 875 | #endif
|
| 876 |
|
| 877 | static inline void debug_init(struct timer_list *timer)
|
| 878 | {
|
| 879 | debug_timer_init(timer);
|
| 880 | trace_timer_init(timer);
|
| 881 | }
|
| 882 |
|
| 883 | static inline void debug_deactivate(struct timer_list *timer)
|
| 884 | {
|
| 885 | debug_timer_deactivate(timer);
|
| 886 | trace_timer_cancel(timer);
|
| 887 | }
|
| 888 |
|
| 889 | static inline void debug_assert_init(struct timer_list *timer)
|
| 890 | {
|
| 891 | debug_timer_assert_init(timer);
|
| 892 | }
|
| 893 |
|
| 894 | static void do_init_timer(struct timer_list *timer,
|
| 895 | void (*func)(struct timer_list *),
|
| 896 | unsigned int flags,
|
| 897 | const char *name, struct lock_class_key *key)
|
| 898 | {
|
| 899 | timer->entry.pprev = NULL;
|
| 900 | timer->function = func;
|
| 901 | if (WARN_ON_ONCE(flags & ~TIMER_INIT_FLAGS))
|
| 902 | flags &= TIMER_INIT_FLAGS;
|
| 903 | timer->flags = flags | raw_smp_processor_id();
|
| 904 | lockdep_init_map(&timer->lockdep_map, name, key, 0);
|
| 905 | }
|
| 906 |
|
| 907 | |
| 908 | * init_timer_key - initialize a timer
|
| 909 | * @timer: the timer to be initialized
|
| 910 | * @func: timer callback function
|
| 911 | * @flags: timer flags
|
| 912 | * @name: name of the timer
|
| 913 | * @key: lockdep class key of the fake lock used for tracking timer
|
| 914 | * sync lock dependencies
|
| 915 | *
|
| 916 | * init_timer_key() must be done to a timer prior to calling *any* of the
|
| 917 | * other timer functions.
|
| 918 |
|
| 919 | void init_timer_key(struct timer_list *timer,
|
| 920 | void (*func)(struct timer_list *), unsigned int flags,
|
| 921 | const char *name, struct lock_class_key *key)
|
| 922 | {
|
| 923 | debug_init(timer);
|
| 924 | do_init_timer(timer, func, flags, name, key);
|
| 925 | }
|
| 926 | EXPORT_SYMBOL(init_timer_key);
|
| 927 |
|
| 928 | static inline void detach_timer(struct timer_list *timer, bool clear_pending)
|
| 929 | {
|
| 930 | struct hlist_node *entry = &timer->entry;
|
| 931 |
|
| 932 | debug_deactivate(timer);
|
| 933 |
|
| 934 | __hlist_del(entry);
|
| 935 | if (clear_pending)
|
| 936 | entry->pprev = NULL;
|
| 937 | entry->next = LIST_POISON2;
|
| 938 | }
|
| 939 |
|
| 940 | static int detach_if_pending(struct timer_list *timer, struct timer_base *base,
|
| 941 | bool clear_pending)
|
| 942 | {
|
| 943 | unsigned idx = timer_get_idx(timer);
|
| 944 |
|
| 945 | if (!timer_pending(timer))
|
| 946 | return 0;
|
| 947 |
|
| 948 | if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) {
|
| 949 | __clear_bit(idx, base->pending_map);
|
| 950 | base->next_expiry_recalc = true;
|
| 951 | }
|
| 952 |
|
| 953 | detach_timer(timer, clear_pending);
|
| 954 | return 1;
|
| 955 | }
|
| 956 |
|
| 957 | static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu)
|
| 958 | {
|
| 959 | int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
|
| 960 | struct timer_base *base;
|
| 961 |
|
| 962 | base = per_cpu_ptr(&timer_bases[index], cpu);
|
| 963 |
|
| 964 | |
| 965 | * If the timer is deferrable and NO_HZ_COMMON is set then we need
|
| 966 | * to use the deferrable base.
|
| 967 |
|
| 968 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
|
| 969 | base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu);
|
| 970 | return base;
|
| 971 | }
|
| 972 |
|
| 973 | static inline struct timer_base *get_timer_this_cpu_base(u32 tflags)
|
| 974 | {
|
| 975 | int index = tflags & TIMER_PINNED ? BASE_LOCAL : BASE_GLOBAL;
|
| 976 | struct timer_base *base;
|
| 977 |
|
| 978 | base = this_cpu_ptr(&timer_bases[index]);
|
| 979 |
|
| 980 | |
| 981 | * If the timer is deferrable and NO_HZ_COMMON is set then we need
|
| 982 | * to use the deferrable base.
|
| 983 |
|
| 984 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE))
|
| 985 | base = this_cpu_ptr(&timer_bases[BASE_DEF]);
|
| 986 | return base;
|
| 987 | }
|
| 988 |
|
| 989 | static inline struct timer_base *get_timer_base(u32 tflags)
|
| 990 | {
|
| 991 | return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK);
|
| 992 | }
|
| 993 |
|
| 994 | static inline void __forward_timer_base(struct timer_base *base,
|
| 995 | unsigned long basej)
|
| 996 | {
|
| 997 | |
| 998 | * Check whether we can forward the base. We can only do that when
|
| 999 | * @basej is past base->clk otherwise we might rewind base->clk.
|
| 1000 |
|
| 1001 | if (time_before_eq(basej, base->clk))
|
| 1002 | return;
|
| 1003 |
|
| 1004 | |
| 1005 | * If the next expiry value is > jiffies, then we fast forward to
|
| 1006 | * jiffies otherwise we forward to the next expiry value.
|
| 1007 |
|
| 1008 | if (time_after(base->next_expiry, basej)) {
|
| 1009 | base->clk = basej;
|
| 1010 | } else {
|
| 1011 | if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
|
| 1012 | return;
|
| 1013 | base->clk = base->next_expiry;
|
| 1014 | }
|
| 1015 |
|
| 1016 | }
|
| 1017 |
|
| 1018 | static inline void forward_timer_base(struct timer_base *base)
|
| 1019 | {
|
| 1020 | __forward_timer_base(base, READ_ONCE(jiffies));
|
| 1021 | }
|
| 1022 |
|
| 1023 | |
| 1024 | * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means
|
| 1025 | * that all timers which are tied to this base are locked, and the base itself
|
| 1026 | * is locked too.
|
| 1027 | *
|
| 1028 | * So __run_timers/migrate_timers can safely modify all timers which could
|
| 1029 | * be found in the base->vectors array.
|
| 1030 | *
|
| 1031 | * When a timer is migrating then the TIMER_MIGRATING flag is set and we need
|
| 1032 | * to wait until the migration is done.
|
| 1033 |
|
| 1034 | static struct timer_base *lock_timer_base(struct timer_list *timer,
|
| 1035 | unsigned long *flags)
|
| 1036 | __acquires(timer->base->lock)
|
| 1037 | {
|
| 1038 | for (;;) {
|
| 1039 | struct timer_base *base;
|
| 1040 | u32 tf;
|
| 1041 |
|
| 1042 | |
| 1043 | * We need to use READ_ONCE() here, otherwise the compiler
|
| 1044 | * might re-read @tf between the check for TIMER_MIGRATING
|
| 1045 | * and spin_lock().
|
| 1046 |
|
| 1047 | tf = READ_ONCE(timer->flags);
|
| 1048 |
|
| 1049 | if (!(tf & TIMER_MIGRATING)) {
|
| 1050 | base = get_timer_base(tf);
|
| 1051 | raw_spin_lock_irqsave(&base->lock, *flags);
|
| 1052 | if (timer->flags == tf)
|
| 1053 | return base;
|
| 1054 | raw_spin_unlock_irqrestore(&base->lock, *flags);
|
| 1055 | }
|
| 1056 | cpu_relax();
|
| 1057 | }
|
| 1058 | }
|
| 1059 |
|
| 1060 | #define MOD_TIMER_PENDING_ONLY 0x01
|
| 1061 | #define MOD_TIMER_REDUCE 0x02
|
| 1062 | #define MOD_TIMER_NOTPENDING 0x04
|
| 1063 |
|
| 1064 | static inline int
|
| 1065 | __mod_timer(struct timer_list *timer, unsigned long expires, unsigned int options)
|
| 1066 | {
|
| 1067 | unsigned long clk = 0, flags, bucket_expiry;
|
| 1068 | struct timer_base *base, *new_base;
|
| 1069 | unsigned int idx = UINT_MAX;
|
| 1070 | int ret = 0;
|
| 1071 |
|
| 1072 | debug_assert_init(timer);
|
| 1073 |
|
| 1074 | |
| 1075 | * This is a common optimization triggered by the networking code - if
|
| 1076 | * the timer is re-modified to have the same timeout or ends up in the
|
| 1077 | * same array bucket then just return:
|
| 1078 |
|
| 1079 | if (!(options & MOD_TIMER_NOTPENDING) && timer_pending(timer)) {
|
| 1080 | |
| 1081 | * The downside of this optimization is that it can result in
|
| 1082 | * larger granularity than you would get from adding a new
|
| 1083 | * timer with this expiry.
|
| 1084 |
|
| 1085 | long diff = timer->expires - expires;
|
| 1086 |
|
| 1087 | if (!diff)
|
| 1088 | return 1;
|
| 1089 | if (options & MOD_TIMER_REDUCE && diff <= 0)
|
| 1090 | return 1;
|
| 1091 |
|
| 1092 | |
| 1093 | * We lock timer base and calculate the bucket index right
|
| 1094 | * here. If the timer ends up in the same bucket, then we
|
| 1095 | * just update the expiry time and avoid the whole
|
| 1096 | * dequeue/enqueue dance.
|
| 1097 |
|
| 1098 | base = lock_timer_base(timer, &flags);
|
| 1099 | |
| 1100 | * Has @timer been shutdown? This needs to be evaluated
|
| 1101 | * while holding base lock to prevent a race against the
|
| 1102 | * shutdown code.
|
| 1103 |
|
| 1104 | if (!timer->function)
|
| 1105 | goto out_unlock;
|
| 1106 |
|
| 1107 | forward_timer_base(base);
|
| 1108 |
|
| 1109 | if (timer_pending(timer) && (options & MOD_TIMER_REDUCE) &&
|
| 1110 | time_before_eq(timer->expires, expires)) {
|
| 1111 | ret = 1;
|
| 1112 | goto out_unlock;
|
| 1113 | }
|
| 1114 |
|
| 1115 | clk = base->clk;
|
| 1116 | idx = calc_wheel_index(expires, clk, &bucket_expiry);
|
| 1117 |
|
| 1118 | |
| 1119 | * Retrieve and compare the array index of the pending
|
| 1120 | * timer. If it matches set the expiry to the new value so a
|
| 1121 | * subsequent call will exit in the expires check above.
|
| 1122 |
|
| 1123 | if (idx == timer_get_idx(timer)) {
|
| 1124 | if (!(options & MOD_TIMER_REDUCE))
|
| 1125 | timer->expires = expires;
|
| 1126 | else if (time_after(timer->expires, expires))
|
| 1127 | timer->expires = expires;
|
| 1128 | ret = 1;
|
| 1129 | goto out_unlock;
|
| 1130 | }
|
| 1131 | } else {
|
| 1132 | base = lock_timer_base(timer, &flags);
|
| 1133 | |
| 1134 | * Has @timer been shutdown? This needs to be evaluated
|
| 1135 | * while holding base lock to prevent a race against the
|
| 1136 | * shutdown code.
|
| 1137 |
|
| 1138 | if (!timer->function)
|
| 1139 | goto out_unlock;
|
| 1140 |
|
| 1141 | forward_timer_base(base);
|
| 1142 | }
|
| 1143 |
|
| 1144 | ret = detach_if_pending(timer, base, false);
|
| 1145 | if (!ret && (options & MOD_TIMER_PENDING_ONLY))
|
| 1146 | goto out_unlock;
|
| 1147 |
|
| 1148 | new_base = get_timer_this_cpu_base(timer->flags);
|
| 1149 |
|
| 1150 | if (base != new_base) {
|
| 1151 | |
| 1152 | * We are trying to schedule the timer on the new base.
|
| 1153 | * However we can't change timer's base while it is running,
|
| 1154 | * otherwise timer_delete_sync() can't detect that the timer's
|
| 1155 | * handler yet has not finished. This also guarantees that the
|
| 1156 | * timer is serialized wrt itself.
|
| 1157 |
|
| 1158 | if (likely(base->running_timer != timer)) {
|
| 1159 |
|
| 1160 | timer->flags |= TIMER_MIGRATING;
|
| 1161 |
|
| 1162 | raw_spin_unlock(&base->lock);
|
| 1163 | base = new_base;
|
| 1164 | raw_spin_lock(&base->lock);
|
| 1165 | WRITE_ONCE(timer->flags,
|
| 1166 | (timer->flags & ~TIMER_BASEMASK) | base->cpu);
|
| 1167 | forward_timer_base(base);
|
| 1168 | }
|
| 1169 | }
|
| 1170 |
|
| 1171 | debug_timer_activate(timer);
|
| 1172 |
|
| 1173 | timer->expires = expires;
|
| 1174 | |
| 1175 | * If 'idx' was calculated above and the base time did not advance
|
| 1176 | * between calculating 'idx' and possibly switching the base, only
|
| 1177 | * enqueue_timer() is required. Otherwise we need to (re)calculate
|
| 1178 | * the wheel index via internal_add_timer().
|
| 1179 |
|
| 1180 | if (idx != UINT_MAX && clk == base->clk)
|
| 1181 | enqueue_timer(base, timer, idx, bucket_expiry);
|
| 1182 | else
|
| 1183 | internal_add_timer(base, timer);
|
| 1184 |
|
| 1185 | out_unlock:
|
| 1186 | raw_spin_unlock_irqrestore(&base->lock, flags);
|
| 1187 |
|
| 1188 | return ret;
|
| 1189 | }
|
| 1190 |
|
| 1191 | |
| 1192 | * mod_timer_pending - Modify a pending timer's timeout
|
| 1193 | * @timer: The pending timer to be modified
|
| 1194 | * @expires: New absolute timeout in jiffies
|
| 1195 | *
|
| 1196 | * mod_timer_pending() is the same for pending timers as mod_timer(), but
|
| 1197 | * will not activate inactive timers.
|
| 1198 | *
|
| 1199 | * If @timer->function == NULL then the start operation is silently
|
| 1200 | * discarded.
|
| 1201 | *
|
| 1202 | * Return:
|
| 1203 | * * %0 - The timer was inactive and not modified or was in
|
| 1204 | * shutdown state and the operation was discarded
|
| 1205 | * * %1 - The timer was active and requeued to expire at @expires
|
| 1206 |
|
| 1207 | int mod_timer_pending(struct timer_list *timer, unsigned long expires)
|
| 1208 | {
|
| 1209 | return __mod_timer(timer, expires, MOD_TIMER_PENDING_ONLY);
|
| 1210 | }
|
| 1211 | EXPORT_SYMBOL(mod_timer_pending);
|
| 1212 |
|
| 1213 | |
| 1214 | * mod_timer - Modify a timer's timeout
|
| 1215 | * @timer: The timer to be modified
|
| 1216 | * @expires: New absolute timeout in jiffies
|
| 1217 | *
|
| 1218 | * mod_timer(timer, expires) is equivalent to:
|
| 1219 | *
|
| 1220 | * del_timer(timer); timer->expires = expires; add_timer(timer);
|
| 1221 | *
|
| 1222 | * mod_timer() is more efficient than the above open coded sequence. In
|
| 1223 | * case that the timer is inactive, the del_timer() part is a NOP. The
|
| 1224 | * timer is in any case activated with the new expiry time @expires.
|
| 1225 | *
|
| 1226 | * Note that if there are multiple unserialized concurrent users of the
|
| 1227 | * same timer, then mod_timer() is the only safe way to modify the timeout,
|
| 1228 | * since add_timer() cannot modify an already running timer.
|
| 1229 | *
|
| 1230 | * If @timer->function == NULL then the start operation is silently
|
| 1231 | * discarded. In this case the return value is 0 and meaningless.
|
| 1232 | *
|
| 1233 | * Return:
|
| 1234 | * * %0 - The timer was inactive and started or was in shutdown
|
| 1235 | * state and the operation was discarded
|
| 1236 | * * %1 - The timer was active and requeued to expire at @expires or
|
| 1237 | * the timer was active and not modified because @expires did
|
| 1238 | * not change the effective expiry time
|
| 1239 |
|
| 1240 | int mod_timer(struct timer_list *timer, unsigned long expires)
|
| 1241 | {
|
| 1242 | return __mod_timer(timer, expires, 0);
|
| 1243 | }
|
| 1244 | EXPORT_SYMBOL(mod_timer);
|
| 1245 |
|
| 1246 | |
| 1247 | * timer_reduce - Modify a timer's timeout if it would reduce the timeout
|
| 1248 | * @timer: The timer to be modified
|
| 1249 | * @expires: New absolute timeout in jiffies
|
| 1250 | *
|
| 1251 | * timer_reduce() is very similar to mod_timer(), except that it will only
|
| 1252 | * modify an enqueued timer if that would reduce the expiration time. If
|
| 1253 | * @timer is not enqueued it starts the timer.
|
| 1254 | *
|
| 1255 | * If @timer->function == NULL then the start operation is silently
|
| 1256 | * discarded.
|
| 1257 | *
|
| 1258 | * Return:
|
| 1259 | * * %0 - The timer was inactive and started or was in shutdown
|
| 1260 | * state and the operation was discarded
|
| 1261 | * * %1 - The timer was active and requeued to expire at @expires or
|
| 1262 | * the timer was active and not modified because @expires
|
| 1263 | * did not change the effective expiry time such that the
|
| 1264 | * timer would expire earlier than already scheduled
|
| 1265 |
|
| 1266 | int timer_reduce(struct timer_list *timer, unsigned long expires)
|
| 1267 | {
|
| 1268 | return __mod_timer(timer, expires, MOD_TIMER_REDUCE);
|
| 1269 | }
|
| 1270 | EXPORT_SYMBOL(timer_reduce);
|
| 1271 |
|
| 1272 | |
| 1273 | * add_timer - Start a timer
|
| 1274 | * @timer: The timer to be started
|
| 1275 | *
|
| 1276 | * Start @timer to expire at @timer->expires in the future. @timer->expires
|
| 1277 | * is the absolute expiry time measured in 'jiffies'. When the timer expires
|
| 1278 | * timer->function(timer) will be invoked from soft interrupt context.
|
| 1279 | *
|
| 1280 | * The @timer->expires and @timer->function fields must be set prior
|
| 1281 | * to calling this function.
|
| 1282 | *
|
| 1283 | * If @timer->function == NULL then the start operation is silently
|
| 1284 | * discarded.
|
| 1285 | *
|
| 1286 | * If @timer->expires is already in the past @timer will be queued to
|
| 1287 | * expire at the next timer tick.
|
| 1288 | *
|
| 1289 | * This can only operate on an inactive timer. Attempts to invoke this on
|
| 1290 | * an active timer are rejected with a warning.
|
| 1291 |
|
| 1292 | void add_timer(struct timer_list *timer)
|
| 1293 | {
|
| 1294 | if (WARN_ON_ONCE(timer_pending(timer)))
|
| 1295 | return;
|
| 1296 | __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
|
| 1297 | }
|
| 1298 | EXPORT_SYMBOL(add_timer);
|
| 1299 |
|
| 1300 | |
| 1301 | * add_timer_local() - Start a timer on the local CPU
|
| 1302 | * @timer: The timer to be started
|
| 1303 | *
|
| 1304 | * Same as add_timer() except that the timer flag TIMER_PINNED is set.
|
| 1305 | *
|
| 1306 | * See add_timer() for further details.
|
| 1307 |
|
| 1308 | void add_timer_local(struct timer_list *timer)
|
| 1309 | {
|
| 1310 | if (WARN_ON_ONCE(timer_pending(timer)))
|
| 1311 | return;
|
| 1312 | timer->flags |= TIMER_PINNED;
|
| 1313 | __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
|
| 1314 | }
|
| 1315 | EXPORT_SYMBOL(add_timer_local);
|
| 1316 |
|
| 1317 | |
| 1318 | * add_timer_global() - Start a timer without TIMER_PINNED flag set
|
| 1319 | * @timer: The timer to be started
|
| 1320 | *
|
| 1321 | * Same as add_timer() except that the timer flag TIMER_PINNED is unset.
|
| 1322 | *
|
| 1323 | * See add_timer() for further details.
|
| 1324 |
|
| 1325 | void add_timer_global(struct timer_list *timer)
|
| 1326 | {
|
| 1327 | if (WARN_ON_ONCE(timer_pending(timer)))
|
| 1328 | return;
|
| 1329 | timer->flags &= ~TIMER_PINNED;
|
| 1330 | __mod_timer(timer, timer->expires, MOD_TIMER_NOTPENDING);
|
| 1331 | }
|
| 1332 | EXPORT_SYMBOL(add_timer_global);
|
| 1333 |
|
| 1334 | |
| 1335 | * add_timer_on - Start a timer on a particular CPU
|
| 1336 | * @timer: The timer to be started
|
| 1337 | * @cpu: The CPU to start it on
|
| 1338 | *
|
| 1339 | * Same as add_timer() except that it starts the timer on the given CPU and
|
| 1340 | * the TIMER_PINNED flag is set. When timer shouldn't be a pinned timer in
|
| 1341 | * the next round, add_timer_global() should be used instead as it unsets
|
| 1342 | * the TIMER_PINNED flag.
|
| 1343 | *
|
| 1344 | * See add_timer() for further details.
|
| 1345 |
|
| 1346 | void add_timer_on(struct timer_list *timer, int cpu)
|
| 1347 | {
|
| 1348 | struct timer_base *new_base, *base;
|
| 1349 | unsigned long flags;
|
| 1350 |
|
| 1351 | debug_assert_init(timer);
|
| 1352 |
|
| 1353 | if (WARN_ON_ONCE(timer_pending(timer)))
|
| 1354 | return;
|
| 1355 |
|
| 1356 |
|
| 1357 | timer->flags |= TIMER_PINNED;
|
| 1358 |
|
| 1359 | new_base = get_timer_cpu_base(timer->flags, cpu);
|
| 1360 |
|
| 1361 | |
| 1362 | * If @timer was on a different CPU, it should be migrated with the
|
| 1363 | * old base locked to prevent other operations proceeding with the
|
| 1364 | * wrong base locked. See lock_timer_base().
|
| 1365 |
|
| 1366 | base = lock_timer_base(timer, &flags);
|
| 1367 | |
| 1368 | * Has @timer been shutdown? This needs to be evaluated while
|
| 1369 | * holding base lock to prevent a race against the shutdown code.
|
| 1370 |
|
| 1371 | if (!timer->function)
|
| 1372 | goto out_unlock;
|
| 1373 |
|
| 1374 | if (base != new_base) {
|
| 1375 | timer->flags |= TIMER_MIGRATING;
|
| 1376 |
|
| 1377 | raw_spin_unlock(&base->lock);
|
| 1378 | base = new_base;
|
| 1379 | raw_spin_lock(&base->lock);
|
| 1380 | WRITE_ONCE(timer->flags,
|
| 1381 | (timer->flags & ~TIMER_BASEMASK) | cpu);
|
| 1382 | }
|
| 1383 | forward_timer_base(base);
|
| 1384 |
|
| 1385 | debug_timer_activate(timer);
|
| 1386 | internal_add_timer(base, timer);
|
| 1387 | out_unlock:
|
| 1388 | raw_spin_unlock_irqrestore(&base->lock, flags);
|
| 1389 | }
|
| 1390 | EXPORT_SYMBOL_GPL(add_timer_on);
|
| 1391 |
|
| 1392 | |
| 1393 | * __timer_delete - Internal function: Deactivate a timer
|
| 1394 | * @timer: The timer to be deactivated
|
| 1395 | * @shutdown: If true, this indicates that the timer is about to be
|
| 1396 | * shutdown permanently.
|
| 1397 | *
|
| 1398 | * If @shutdown is true then @timer->function is set to NULL under the
|
| 1399 | * timer base lock which prevents further rearming of the time. In that
|
| 1400 | * case any attempt to rearm @timer after this function returns will be
|
| 1401 | * silently ignored.
|
| 1402 | *
|
| 1403 | * Return:
|
| 1404 | * * %0 - The timer was not pending
|
| 1405 | * * %1 - The timer was pending and deactivated
|
| 1406 |
|
| 1407 | static int __timer_delete(struct timer_list *timer, bool shutdown)
|
| 1408 | {
|
| 1409 | struct timer_base *base;
|
| 1410 | unsigned long flags;
|
| 1411 | int ret = 0;
|
| 1412 |
|
| 1413 | debug_assert_init(timer);
|
| 1414 |
|
| 1415 | |
| 1416 | * If @shutdown is set then the lock has to be taken whether the
|
| 1417 | * timer is pending or not to protect against a concurrent rearm
|
| 1418 | * which might hit between the lockless pending check and the lock
|
| 1419 | * acquisition. By taking the lock it is ensured that such a newly
|
| 1420 | * enqueued timer is dequeued and cannot end up with
|
| 1421 | * timer->function == NULL in the expiry code.
|
| 1422 | *
|
| 1423 | * If timer->function is currently executed, then this makes sure
|
| 1424 | * that the callback cannot requeue the timer.
|
| 1425 |
|
| 1426 | if (timer_pending(timer) || shutdown) {
|
| 1427 | base = lock_timer_base(timer, &flags);
|
| 1428 | ret = detach_if_pending(timer, base, true);
|
| 1429 | if (shutdown)
|
| 1430 | timer->function = NULL;
|
| 1431 | raw_spin_unlock_irqrestore(&base->lock, flags);
|
| 1432 | }
|
| 1433 |
|
| 1434 | return ret;
|
| 1435 | }
|
| 1436 |
|
| 1437 | |
| 1438 | * timer_delete - Deactivate a timer
|
| 1439 | * @timer: The timer to be deactivated
|
| 1440 | *
|
| 1441 | * The function only deactivates a pending timer, but contrary to
|
| 1442 | * timer_delete_sync() it does not take into account whether the timer's
|
| 1443 | * callback function is concurrently executed on a different CPU or not.
|
| 1444 | * It neither prevents rearming of the timer. If @timer can be rearmed
|
| 1445 | * concurrently then the return value of this function is meaningless.
|
| 1446 | *
|
| 1447 | * Return:
|
| 1448 | * * %0 - The timer was not pending
|
| 1449 | * * %1 - The timer was pending and deactivated
|
| 1450 |
|
| 1451 | int timer_delete(struct timer_list *timer)
|
| 1452 | {
|
| 1453 | return __timer_delete(timer, false);
|
| 1454 | }
|
| 1455 | EXPORT_SYMBOL(timer_delete);
|
| 1456 |
|
| 1457 | |
| 1458 | * timer_shutdown - Deactivate a timer and prevent rearming
|
| 1459 | * @timer: The timer to be deactivated
|
| 1460 | *
|
| 1461 | * The function does not wait for an eventually running timer callback on a
|
| 1462 | * different CPU but it prevents rearming of the timer. Any attempt to arm
|
| 1463 | * @timer after this function returns will be silently ignored.
|
| 1464 | *
|
| 1465 | * This function is useful for teardown code and should only be used when
|
| 1466 | * timer_shutdown_sync() cannot be invoked due to locking or context constraints.
|
| 1467 | *
|
| 1468 | * Return:
|
| 1469 | * * %0 - The timer was not pending
|
| 1470 | * * %1 - The timer was pending
|
| 1471 |
|
| 1472 | int timer_shutdown(struct timer_list *timer)
|
| 1473 | {
|
| 1474 | return __timer_delete(timer, true);
|
| 1475 | }
|
| 1476 | EXPORT_SYMBOL_GPL(timer_shutdown);
|
| 1477 |
|
| 1478 | |
| 1479 | * __try_to_del_timer_sync - Internal function: Try to deactivate a timer
|
| 1480 | * @timer: Timer to deactivate
|
| 1481 | * @shutdown: If true, this indicates that the timer is about to be
|
| 1482 | * shutdown permanently.
|
| 1483 | *
|
| 1484 | * If @shutdown is true then @timer->function is set to NULL under the
|
| 1485 | * timer base lock which prevents further rearming of the timer. Any
|
| 1486 | * attempt to rearm @timer after this function returns will be silently
|
| 1487 | * ignored.
|
| 1488 | *
|
| 1489 | * This function cannot guarantee that the timer cannot be rearmed
|
| 1490 | * right after dropping the base lock if @shutdown is false. That
|
| 1491 | * needs to be prevented by the calling code if necessary.
|
| 1492 | *
|
| 1493 | * Return:
|
| 1494 | * * %0 - The timer was not pending
|
| 1495 | * * %1 - The timer was pending and deactivated
|
| 1496 | * * %-1 - The timer callback function is running on a different CPU
|
| 1497 |
|
| 1498 | static int __try_to_del_timer_sync(struct timer_list *timer, bool shutdown)
|
| 1499 | {
|
| 1500 | struct timer_base *base;
|
| 1501 | unsigned long flags;
|
| 1502 | int ret = -1;
|
| 1503 |
|
| 1504 | debug_assert_init(timer);
|
| 1505 |
|
| 1506 | base = lock_timer_base(timer, &flags);
|
| 1507 |
|
| 1508 | if (base->running_timer != timer)
|
| 1509 | ret = detach_if_pending(timer, base, true);
|
| 1510 | if (shutdown)
|
| 1511 | timer->function = NULL;
|
| 1512 |
|
| 1513 | raw_spin_unlock_irqrestore(&base->lock, flags);
|
| 1514 |
|
| 1515 | return ret;
|
| 1516 | }
|
| 1517 |
|
| 1518 | |
| 1519 | * try_to_del_timer_sync - Try to deactivate a timer
|
| 1520 | * @timer: Timer to deactivate
|
| 1521 | *
|
| 1522 | * This function tries to deactivate a timer. On success the timer is not
|
| 1523 | * queued and the timer callback function is not running on any CPU.
|
| 1524 | *
|
| 1525 | * This function does not guarantee that the timer cannot be rearmed right
|
| 1526 | * after dropping the base lock. That needs to be prevented by the calling
|
| 1527 | * code if necessary.
|
| 1528 | *
|
| 1529 | * Return:
|
| 1530 | * * %0 - The timer was not pending
|
| 1531 | * * %1 - The timer was pending and deactivated
|
| 1532 | * * %-1 - The timer callback function is running on a different CPU
|
| 1533 |
|
| 1534 | int try_to_del_timer_sync(struct timer_list *timer)
|
| 1535 | {
|
| 1536 | return __try_to_del_timer_sync(timer, false);
|
| 1537 | }
|
| 1538 | EXPORT_SYMBOL(try_to_del_timer_sync);
|
| 1539 |
|
| 1540 | #ifdef CONFIG_PREEMPT_RT
|
| 1541 | static __init void timer_base_init_expiry_lock(struct timer_base *base)
|
| 1542 | {
|
| 1543 | spin_lock_init(&base->expiry_lock);
|
| 1544 | }
|
| 1545 |
|
| 1546 | static inline void timer_base_lock_expiry(struct timer_base *base)
|
| 1547 | {
|
| 1548 | spin_lock(&base->expiry_lock);
|
| 1549 | }
|
| 1550 |
|
| 1551 | static inline void timer_base_unlock_expiry(struct timer_base *base)
|
| 1552 | {
|
| 1553 | spin_unlock(&base->expiry_lock);
|
| 1554 | }
|
| 1555 |
|
| 1556 | |
| 1557 | * The counterpart to del_timer_wait_running().
|
| 1558 | *
|
| 1559 | * If there is a waiter for base->expiry_lock, then it was waiting for the
|
| 1560 | * timer callback to finish. Drop expiry_lock and reacquire it. That allows
|
| 1561 | * the waiter to acquire the lock and make progress.
|
| 1562 |
|
| 1563 | static void timer_sync_wait_running(struct timer_base *base)
|
| 1564 | __releases(&base->lock) __releases(&base->expiry_lock)
|
| 1565 | __acquires(&base->expiry_lock) __acquires(&base->lock)
|
| 1566 | {
|
| 1567 | if (atomic_read(&base->timer_waiters)) {
|
| 1568 | raw_spin_unlock_irq(&base->lock);
|
| 1569 | spin_unlock(&base->expiry_lock);
|
| 1570 | spin_lock(&base->expiry_lock);
|
| 1571 | raw_spin_lock_irq(&base->lock);
|
| 1572 | }
|
| 1573 | }
|
| 1574 |
|
| 1575 | |
| 1576 | * This function is called on PREEMPT_RT kernels when the fast path
|
| 1577 | * deletion of a timer failed because the timer callback function was
|
| 1578 | * running.
|
| 1579 | *
|
| 1580 | * This prevents priority inversion, if the softirq thread on a remote CPU
|
| 1581 | * got preempted, and it prevents a life lock when the task which tries to
|
| 1582 | * delete a timer preempted the softirq thread running the timer callback
|
| 1583 | * function.
|
| 1584 |
|
| 1585 | static void del_timer_wait_running(struct timer_list *timer)
|
| 1586 | {
|
| 1587 | u32 tf;
|
| 1588 |
|
| 1589 | tf = READ_ONCE(timer->flags);
|
| 1590 | if (!(tf & (TIMER_MIGRATING | TIMER_IRQSAFE))) {
|
| 1591 | struct timer_base *base = get_timer_base(tf);
|
| 1592 |
|
| 1593 | |
| 1594 | * Mark the base as contended and grab the expiry lock,
|
| 1595 | * which is held by the softirq across the timer
|
| 1596 | * callback. Drop the lock immediately so the softirq can
|
| 1597 | * expire the next timer. In theory the timer could already
|
| 1598 | * be running again, but that's more than unlikely and just
|
| 1599 | * causes another wait loop.
|
| 1600 |
|
| 1601 | atomic_inc(&base->timer_waiters);
|
| 1602 | spin_lock_bh(&base->expiry_lock);
|
| 1603 | atomic_dec(&base->timer_waiters);
|
| 1604 | spin_unlock_bh(&base->expiry_lock);
|
| 1605 | }
|
| 1606 | }
|
| 1607 | #else
|
| 1608 | static inline void timer_base_init_expiry_lock(struct timer_base *base) { }
|
| 1609 | static inline void timer_base_lock_expiry(struct timer_base *base) { }
|
| 1610 | static inline void timer_base_unlock_expiry(struct timer_base *base) { }
|
| 1611 | static inline void timer_sync_wait_running(struct timer_base *base) { }
|
| 1612 | static inline void del_timer_wait_running(struct timer_list *timer) { }
|
| 1613 | #endif
|
| 1614 |
|
| 1615 | |
| 1616 | * __timer_delete_sync - Internal function: Deactivate a timer and wait
|
| 1617 | * for the handler to finish.
|
| 1618 | * @timer: The timer to be deactivated
|
| 1619 | * @shutdown: If true, @timer->function will be set to NULL under the
|
| 1620 | * timer base lock which prevents rearming of @timer
|
| 1621 | *
|
| 1622 | * If @shutdown is not set the timer can be rearmed later. If the timer can
|
| 1623 | * be rearmed concurrently, i.e. after dropping the base lock then the
|
| 1624 | * return value is meaningless.
|
| 1625 | *
|
| 1626 | * If @shutdown is set then @timer->function is set to NULL under timer
|
| 1627 | * base lock which prevents rearming of the timer. Any attempt to rearm
|
| 1628 | * a shutdown timer is silently ignored.
|
| 1629 | *
|
| 1630 | * If the timer should be reused after shutdown it has to be initialized
|
| 1631 | * again.
|
| 1632 | *
|
| 1633 | * Return:
|
| 1634 | * * %0 - The timer was not pending
|
| 1635 | * * %1 - The timer was pending and deactivated
|
| 1636 |
|
| 1637 | static int __timer_delete_sync(struct timer_list *timer, bool shutdown)
|
| 1638 | {
|
| 1639 | int ret;
|
| 1640 |
|
| 1641 | #ifdef CONFIG_LOCKDEP
|
| 1642 | unsigned long flags;
|
| 1643 |
|
| 1644 | |
| 1645 | * If lockdep gives a backtrace here, please reference
|
| 1646 | * the synchronization rules above.
|
| 1647 |
|
| 1648 | local_irq_save(flags);
|
| 1649 | lock_map_acquire(&timer->lockdep_map);
|
| 1650 | lock_map_release(&timer->lockdep_map);
|
| 1651 | local_irq_restore(flags);
|
| 1652 | #endif
|
| 1653 | |
| 1654 | * don't use it in hardirq context, because it
|
| 1655 | * could lead to deadlock.
|
| 1656 |
|
| 1657 | WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));
|
| 1658 |
|
| 1659 | |
| 1660 | * Must be able to sleep on PREEMPT_RT because of the slowpath in
|
| 1661 | * del_timer_wait_running().
|
| 1662 |
|
| 1663 | if (IS_ENABLED(CONFIG_PREEMPT_RT) && !(timer->flags & TIMER_IRQSAFE))
|
| 1664 | lockdep_assert_preemption_enabled();
|
| 1665 |
|
| 1666 | do {
|
| 1667 | ret = __try_to_del_timer_sync(timer, shutdown);
|
| 1668 |
|
| 1669 | if (unlikely(ret < 0)) {
|
| 1670 | del_timer_wait_running(timer);
|
| 1671 | cpu_relax();
|
| 1672 | }
|
| 1673 | } while (ret < 0);
|
| 1674 |
|
| 1675 | return ret;
|
| 1676 | }
|
| 1677 |
|
| 1678 | |
| 1679 | * timer_delete_sync - Deactivate a timer and wait for the handler to finish.
|
| 1680 | * @timer: The timer to be deactivated
|
| 1681 | *
|
| 1682 | * Synchronization rules: Callers must prevent restarting of the timer,
|
| 1683 | * otherwise this function is meaningless. It must not be called from
|
| 1684 | * interrupt contexts unless the timer is an irqsafe one. The caller must
|
| 1685 | * not hold locks which would prevent completion of the timer's callback
|
| 1686 | * function. The timer's handler must not call add_timer_on(). Upon exit
|
| 1687 | * the timer is not queued and the handler is not running on any CPU.
|
| 1688 | *
|
| 1689 | * For !irqsafe timers, the caller must not hold locks that are held in
|
| 1690 | * interrupt context. Even if the lock has nothing to do with the timer in
|
| 1691 | * question. Here's why::
|
| 1692 | *
|
| 1693 | * CPU0 CPU1
|
| 1694 | * ---- ----
|
| 1695 | * <SOFTIRQ>
|
| 1696 | * call_timer_fn();
|
| 1697 | * base->running_timer = mytimer;
|
| 1698 | * spin_lock_irq(somelock);
|
| 1699 | * <IRQ>
|
| 1700 | * spin_lock(somelock);
|
| 1701 | * timer_delete_sync(mytimer);
|
| 1702 | * while (base->running_timer == mytimer);
|
| 1703 | *
|
| 1704 | * Now timer_delete_sync() will never return and never release somelock.
|
| 1705 | * The interrupt on the other CPU is waiting to grab somelock but it has
|
| 1706 | * interrupted the softirq that CPU0 is waiting to finish.
|
| 1707 | *
|
| 1708 | * This function cannot guarantee that the timer is not rearmed again by
|
| 1709 | * some concurrent or preempting code, right after it dropped the base
|
| 1710 | * lock. If there is the possibility of a concurrent rearm then the return
|
| 1711 | * value of the function is meaningless.
|
| 1712 | *
|
| 1713 | * If such a guarantee is needed, e.g. for teardown situations then use
|
| 1714 | * timer_shutdown_sync() instead.
|
| 1715 | *
|
| 1716 | * Return:
|
| 1717 | * * %0 - The timer was not pending
|
| 1718 | * * %1 - The timer was pending and deactivated
|
| 1719 |
|
| 1720 | int timer_delete_sync(struct timer_list *timer)
|
| 1721 | {
|
| 1722 | return __timer_delete_sync(timer, false);
|
| 1723 | }
|
| 1724 | EXPORT_SYMBOL(timer_delete_sync);
|
| 1725 |
|
| 1726 | |
| 1727 | * timer_shutdown_sync - Shutdown a timer and prevent rearming
|
| 1728 | * @timer: The timer to be shutdown
|
| 1729 | *
|
| 1730 | * When the function returns it is guaranteed that:
|
| 1731 | * - @timer is not queued
|
| 1732 | * - The callback function of @timer is not running
|
| 1733 | * - @timer cannot be enqueued again. Any attempt to rearm
|
| 1734 | * @timer is silently ignored.
|
| 1735 | *
|
| 1736 | * See timer_delete_sync() for synchronization rules.
|
| 1737 | *
|
| 1738 | * This function is useful for final teardown of an infrastructure where
|
| 1739 | * the timer is subject to a circular dependency problem.
|
| 1740 | *
|
| 1741 | * A common pattern for this is a timer and a workqueue where the timer can
|
| 1742 | * schedule work and work can arm the timer. On shutdown the workqueue must
|
| 1743 | * be destroyed and the timer must be prevented from rearming. Unless the
|
| 1744 | * code has conditionals like 'if (mything->in_shutdown)' to prevent that
|
| 1745 | * there is no way to get this correct with timer_delete_sync().
|
| 1746 | *
|
| 1747 | * timer_shutdown_sync() is solving the problem. The correct ordering of
|
| 1748 | * calls in this case is:
|
| 1749 | *
|
| 1750 | * timer_shutdown_sync(&mything->timer);
|
| 1751 | * workqueue_destroy(&mything->workqueue);
|
| 1752 | *
|
| 1753 | * After this 'mything' can be safely freed.
|
| 1754 | *
|
| 1755 | * This obviously implies that the timer is not required to be functional
|
| 1756 | * for the rest of the shutdown operation.
|
| 1757 | *
|
| 1758 | * Return:
|
| 1759 | * * %0 - The timer was not pending
|
| 1760 | * * %1 - The timer was pending
|
| 1761 |
|
| 1762 | int timer_shutdown_sync(struct timer_list *timer)
|
| 1763 | {
|
| 1764 | return __timer_delete_sync(timer, true);
|
| 1765 | }
|
| 1766 | EXPORT_SYMBOL_GPL(timer_shutdown_sync);
|
| 1767 |
|
| 1768 | static void call_timer_fn(struct timer_list *timer,
|
| 1769 | void (*fn)(struct timer_list *),
|
| 1770 | unsigned long baseclk)
|
| 1771 | {
|
| 1772 | int count = preempt_count();
|
| 1773 |
|
| 1774 | #ifdef CONFIG_LOCKDEP
|
| 1775 | |
| 1776 | * It is permissible to free the timer from inside the
|
| 1777 | * function that is called from it, this we need to take into
|
| 1778 | * account for lockdep too. To avoid bogus "held lock freed"
|
| 1779 | * warnings as well as problems when looking into
|
| 1780 | * timer->lockdep_map, make a copy and use that here.
|
| 1781 |
|
| 1782 | struct lockdep_map lockdep_map;
|
| 1783 |
|
| 1784 | lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
|
| 1785 | #endif
|
| 1786 | |
| 1787 | * Couple the lock chain with the lock chain at
|
| 1788 | * timer_delete_sync() by acquiring the lock_map around the fn()
|
| 1789 | * call here and in timer_delete_sync().
|
| 1790 |
|
| 1791 | lock_map_acquire(&lockdep_map);
|
| 1792 |
|
| 1793 | trace_timer_expire_entry(timer, baseclk);
|
| 1794 | fn(timer);
|
| 1795 | trace_timer_expire_exit(timer);
|
| 1796 |
|
| 1797 | lock_map_release(&lockdep_map);
|
| 1798 |
|
| 1799 | if (count != preempt_count()) {
|
| 1800 | WARN_ONCE(1, "timer: %pS preempt leak: %08x -> %08x\n",
|
| 1801 | fn, count, preempt_count());
|
| 1802 | |
| 1803 | * Restore the preempt count. That gives us a decent
|
| 1804 | * chance to survive and extract information. If the
|
| 1805 | * callback kept a lock held, bad luck, but not worse
|
| 1806 | * than the BUG() we had.
|
| 1807 |
|
| 1808 | preempt_count_set(count);
|
| 1809 | }
|
| 1810 | }
|
| 1811 |
|
| 1812 | static void expire_timers(struct timer_base *base, struct hlist_head *head)
|
| 1813 | {
|
| 1814 | |
| 1815 | * This value is required only for tracing. base->clk was
|
| 1816 | * incremented directly before expire_timers was called. But expiry
|
| 1817 | * is related to the old base->clk value.
|
| 1818 |
|
| 1819 | unsigned long baseclk = base->clk - 1;
|
| 1820 |
|
| 1821 | while (!hlist_empty(head)) {
|
| 1822 | struct timer_list *timer;
|
| 1823 | void (*fn)(struct timer_list *);
|
| 1824 |
|
| 1825 | timer = hlist_entry(head->first, struct timer_list, entry);
|
| 1826 |
|
| 1827 | base->running_timer = timer;
|
| 1828 | detach_timer(timer, true);
|
| 1829 |
|
| 1830 | fn = timer->function;
|
| 1831 |
|
| 1832 | if (WARN_ON_ONCE(!fn)) {
|
| 1833 |
|
| 1834 | base->running_timer = NULL;
|
| 1835 | continue;
|
| 1836 | }
|
| 1837 |
|
| 1838 | if (timer->flags & TIMER_IRQSAFE) {
|
| 1839 | raw_spin_unlock(&base->lock);
|
| 1840 | call_timer_fn(timer, fn, baseclk);
|
| 1841 | raw_spin_lock(&base->lock);
|
| 1842 | base->running_timer = NULL;
|
| 1843 | } else {
|
| 1844 | raw_spin_unlock_irq(&base->lock);
|
| 1845 | call_timer_fn(timer, fn, baseclk);
|
| 1846 | raw_spin_lock_irq(&base->lock);
|
| 1847 | base->running_timer = NULL;
|
| 1848 | timer_sync_wait_running(base);
|
| 1849 | }
|
| 1850 | }
|
| 1851 | }
|
| 1852 |
|
| 1853 | static int collect_expired_timers(struct timer_base *base,
|
| 1854 | struct hlist_head *heads)
|
| 1855 | {
|
| 1856 | unsigned long clk = base->clk = base->next_expiry;
|
| 1857 | struct hlist_head *vec;
|
| 1858 | int i, levels = 0;
|
| 1859 | unsigned int idx;
|
| 1860 |
|
| 1861 | for (i = 0; i < LVL_DEPTH; i++) {
|
| 1862 | idx = (clk & LVL_MASK) + i * LVL_SIZE;
|
| 1863 |
|
| 1864 | if (__test_and_clear_bit(idx, base->pending_map)) {
|
| 1865 | vec = base->vectors + idx;
|
| 1866 | hlist_move_list(vec, heads++);
|
| 1867 | levels++;
|
| 1868 | }
|
| 1869 |
|
| 1870 | if (clk & LVL_CLK_MASK)
|
| 1871 | break;
|
| 1872 |
|
| 1873 | clk >>= LVL_CLK_SHIFT;
|
| 1874 | }
|
| 1875 | return levels;
|
| 1876 | }
|
| 1877 |
|
| 1878 | |
| 1879 | * Find the next pending bucket of a level. Search from level start (@offset)
|
| 1880 | * + @clk upwards and if nothing there, search from start of the level
|
| 1881 | * (@offset) up to @offset + clk.
|
| 1882 |
|
| 1883 | static int next_pending_bucket(struct timer_base *base, unsigned offset,
|
| 1884 | unsigned clk)
|
| 1885 | {
|
| 1886 | unsigned pos, start = offset + clk;
|
| 1887 | unsigned end = offset + LVL_SIZE;
|
| 1888 |
|
| 1889 | pos = find_next_bit(base->pending_map, end, start);
|
| 1890 | if (pos < end)
|
| 1891 | return pos - start;
|
| 1892 |
|
| 1893 | pos = find_next_bit(base->pending_map, start, offset);
|
| 1894 | return pos < start ? pos + LVL_SIZE - start : -1;
|
| 1895 | }
|
| 1896 |
|
| 1897 | |
| 1898 | * Search the first expiring timer in the various clock levels. Caller must
|
| 1899 | * hold base->lock.
|
| 1900 | *
|
| 1901 | * Store next expiry time in base->next_expiry.
|
| 1902 |
|
| 1903 | static void timer_recalc_next_expiry(struct timer_base *base)
|
| 1904 | {
|
| 1905 | unsigned long clk, next, adj;
|
| 1906 | unsigned lvl, offset = 0;
|
| 1907 |
|
| 1908 | next = base->clk + NEXT_TIMER_MAX_DELTA;
|
| 1909 | clk = base->clk;
|
| 1910 | for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) {
|
| 1911 | int pos = next_pending_bucket(base, offset, clk & LVL_MASK);
|
| 1912 | unsigned long lvl_clk = clk & LVL_CLK_MASK;
|
| 1913 |
|
| 1914 | if (pos >= 0) {
|
| 1915 | unsigned long tmp = clk + (unsigned long) pos;
|
| 1916 |
|
| 1917 | tmp <<= LVL_SHIFT(lvl);
|
| 1918 | if (time_before(tmp, next))
|
| 1919 | next = tmp;
|
| 1920 |
|
| 1921 | |
| 1922 | * If the next expiration happens before we reach
|
| 1923 | * the next level, no need to check further.
|
| 1924 |
|
| 1925 | if (pos <= ((LVL_CLK_DIV - lvl_clk) & LVL_CLK_MASK))
|
| 1926 | break;
|
| 1927 | }
|
| 1928 | |
| 1929 | * Clock for the next level. If the current level clock lower
|
| 1930 | * bits are zero, we look at the next level as is. If not we
|
| 1931 | * need to advance it by one because that's going to be the
|
| 1932 | * next expiring bucket in that level. base->clk is the next
|
| 1933 | * expiring jiffy. So in case of:
|
| 1934 | *
|
| 1935 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
|
| 1936 | * 0 0 0 0 0 0
|
| 1937 | *
|
| 1938 | * we have to look at all levels @index 0. With
|
| 1939 | *
|
| 1940 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
|
| 1941 | * 0 0 0 0 0 2
|
| 1942 | *
|
| 1943 | * LVL0 has the next expiring bucket @index 2. The upper
|
| 1944 | * levels have the next expiring bucket @index 1.
|
| 1945 | *
|
| 1946 | * In case that the propagation wraps the next level the same
|
| 1947 | * rules apply:
|
| 1948 | *
|
| 1949 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0
|
| 1950 | * 0 0 0 0 F 2
|
| 1951 | *
|
| 1952 | * So after looking at LVL0 we get:
|
| 1953 | *
|
| 1954 | * LVL5 LVL4 LVL3 LVL2 LVL1
|
| 1955 | * 0 0 0 1 0
|
| 1956 | *
|
| 1957 | * So no propagation from LVL1 to LVL2 because that happened
|
| 1958 | * with the add already, but then we need to propagate further
|
| 1959 | * from LVL2 to LVL3.
|
| 1960 | *
|
| 1961 | * So the simple check whether the lower bits of the current
|
| 1962 | * level are 0 or not is sufficient for all cases.
|
| 1963 |
|
| 1964 | adj = lvl_clk ? 1 : 0;
|
| 1965 | clk >>= LVL_CLK_SHIFT;
|
| 1966 | clk += adj;
|
| 1967 | }
|
| 1968 |
|
| 1969 | WRITE_ONCE(base->next_expiry, next);
|
| 1970 | base->next_expiry_recalc = false;
|
| 1971 | base->timers_pending = !(next == base->clk + NEXT_TIMER_MAX_DELTA);
|
| 1972 | }
|
| 1973 |
|
| 1974 | #ifdef CONFIG_NO_HZ_COMMON
|
| 1975 | |
| 1976 | * Check, if the next hrtimer event is before the next timer wheel
|
| 1977 | * event:
|
| 1978 |
|
| 1979 | static u64 cmp_next_hrtimer_event(u64 basem, u64 expires)
|
| 1980 | {
|
| 1981 | u64 nextevt = hrtimer_get_next_event();
|
| 1982 |
|
| 1983 | |
| 1984 | * If high resolution timers are enabled
|
| 1985 | * hrtimer_get_next_event() returns KTIME_MAX.
|
| 1986 |
|
| 1987 | if (expires <= nextevt)
|
| 1988 | return expires;
|
| 1989 |
|
| 1990 | |
| 1991 | * If the next timer is already expired, return the tick base
|
| 1992 | * time so the tick is fired immediately.
|
| 1993 |
|
| 1994 | if (nextevt <= basem)
|
| 1995 | return basem;
|
| 1996 |
|
| 1997 | |
| 1998 | * Round up to the next jiffy. High resolution timers are
|
| 1999 | * off, so the hrtimers are expired in the tick and we need to
|
| 2000 | * make sure that this tick really expires the timer to avoid
|
| 2001 | * a ping pong of the nohz stop code.
|
| 2002 | *
|
| 2003 | * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3
|
| 2004 |
|
| 2005 | return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC;
|
| 2006 | }
|
| 2007 |
|
| 2008 | static unsigned long next_timer_interrupt(struct timer_base *base,
|
| 2009 | unsigned long basej)
|
| 2010 | {
|
| 2011 | if (base->next_expiry_recalc)
|
| 2012 | timer_recalc_next_expiry(base);
|
| 2013 |
|
| 2014 | |
| 2015 | * Move next_expiry for the empty base into the future to prevent an
|
| 2016 | * unnecessary raise of the timer softirq when the next_expiry value
|
| 2017 | * will be reached even if there is no timer pending.
|
| 2018 | *
|
| 2019 | * This update is also required to make timer_base::next_expiry values
|
| 2020 | * easy comparable to find out which base holds the first pending timer.
|
| 2021 |
|
| 2022 | if (!base->timers_pending)
|
| 2023 | WRITE_ONCE(base->next_expiry, basej + NEXT_TIMER_MAX_DELTA);
|
| 2024 |
|
| 2025 | return base->next_expiry;
|
| 2026 | }
|
| 2027 |
|
| 2028 | static unsigned long fetch_next_timer_interrupt(unsigned long basej, u64 basem,
|
| 2029 | struct timer_base *base_local,
|
| 2030 | struct timer_base *base_global,
|
| 2031 | struct timer_events *tevt)
|
| 2032 | {
|
| 2033 | unsigned long nextevt, nextevt_local, nextevt_global;
|
| 2034 | bool local_first;
|
| 2035 |
|
| 2036 | nextevt_local = next_timer_interrupt(base_local, basej);
|
| 2037 | nextevt_global = next_timer_interrupt(base_global, basej);
|
| 2038 |
|
| 2039 | local_first = time_before_eq(nextevt_local, nextevt_global);
|
| 2040 |
|
| 2041 | nextevt = local_first ? nextevt_local : nextevt_global;
|
| 2042 |
|
| 2043 | |
| 2044 | * If the @nextevt is at max. one tick away, use @nextevt and store
|
| 2045 | * it in the local expiry value. The next global event is irrelevant in
|
| 2046 | * this case and can be left as KTIME_MAX.
|
| 2047 |
|
| 2048 | if (time_before_eq(nextevt, basej + 1)) {
|
| 2049 |
|
| 2050 | if (time_before(nextevt, basej))
|
| 2051 | nextevt = basej;
|
| 2052 | tevt->local = basem + (u64)(nextevt - basej) * TICK_NSEC;
|
| 2053 |
|
| 2054 | |
| 2055 | * This is required for the remote check only but it doesn't
|
| 2056 | * hurt, when it is done for both call sites:
|
| 2057 | *
|
| 2058 | * * The remote callers will only take care of the global timers
|
| 2059 | * as local timers will be handled by CPU itself. When not
|
| 2060 | * updating tevt->global with the already missed first global
|
| 2061 | * timer, it is possible that it will be missed completely.
|
| 2062 | *
|
| 2063 | * * The local callers will ignore the tevt->global anyway, when
|
| 2064 | * nextevt is max. one tick away.
|
| 2065 |
|
| 2066 | if (!local_first)
|
| 2067 | tevt->global = tevt->local;
|
| 2068 | return nextevt;
|
| 2069 | }
|
| 2070 |
|
| 2071 | |
| 2072 | * Update tevt.* values:
|
| 2073 | *
|
| 2074 | * If the local queue expires first, then the global event can be
|
| 2075 | * ignored. If the global queue is empty, nothing to do either.
|
| 2076 |
|
| 2077 | if (!local_first && base_global->timers_pending)
|
| 2078 | tevt->global = basem + (u64)(nextevt_global - basej) * TICK_NSEC;
|
| 2079 |
|
| 2080 | if (base_local->timers_pending)
|
| 2081 | tevt->local = basem + (u64)(nextevt_local - basej) * TICK_NSEC;
|
| 2082 |
|
| 2083 | return nextevt;
|
| 2084 | }
|
| 2085 |
|
| 2086 | # ifdef CONFIG_SMP
|
| 2087 | |
| 2088 | * fetch_next_timer_interrupt_remote() - Store next timers into @tevt
|
| 2089 | * @basej: base time jiffies
|
| 2090 | * @basem: base time clock monotonic
|
| 2091 | * @tevt: Pointer to the storage for the expiry values
|
| 2092 | * @cpu: Remote CPU
|
| 2093 | *
|
| 2094 | * Stores the next pending local and global timer expiry values in the
|
| 2095 | * struct pointed to by @tevt. If a queue is empty the corresponding
|
| 2096 | * field is set to KTIME_MAX. If local event expires before global
|
| 2097 | * event, global event is set to KTIME_MAX as well.
|
| 2098 | *
|
| 2099 | * Caller needs to make sure timer base locks are held (use
|
| 2100 | * timer_lock_remote_bases() for this purpose).
|
| 2101 |
|
| 2102 | void fetch_next_timer_interrupt_remote(unsigned long basej, u64 basem,
|
| 2103 | struct timer_events *tevt,
|
| 2104 | unsigned int cpu)
|
| 2105 | {
|
| 2106 | struct timer_base *base_local, *base_global;
|
| 2107 |
|
| 2108 |
|
| 2109 | tevt->local = tevt->global = KTIME_MAX;
|
| 2110 |
|
| 2111 | base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
|
| 2112 | base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
|
| 2113 |
|
| 2114 | lockdep_assert_held(&base_local->lock);
|
| 2115 | lockdep_assert_held(&base_global->lock);
|
| 2116 |
|
| 2117 | fetch_next_timer_interrupt(basej, basem, base_local, base_global, tevt);
|
| 2118 | }
|
| 2119 |
|
| 2120 | |
| 2121 | * timer_unlock_remote_bases - unlock timer bases of cpu
|
| 2122 | * @cpu: Remote CPU
|
| 2123 | *
|
| 2124 | * Unlocks the remote timer bases.
|
| 2125 |
|
| 2126 | void timer_unlock_remote_bases(unsigned int cpu)
|
| 2127 | __releases(timer_bases[BASE_LOCAL]->lock)
|
| 2128 | __releases(timer_bases[BASE_GLOBAL]->lock)
|
| 2129 | {
|
| 2130 | struct timer_base *base_local, *base_global;
|
| 2131 |
|
| 2132 | base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
|
| 2133 | base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
|
| 2134 |
|
| 2135 | raw_spin_unlock(&base_global->lock);
|
| 2136 | raw_spin_unlock(&base_local->lock);
|
| 2137 | }
|
| 2138 |
|
| 2139 | |
| 2140 | * timer_lock_remote_bases - lock timer bases of cpu
|
| 2141 | * @cpu: Remote CPU
|
| 2142 | *
|
| 2143 | * Locks the remote timer bases.
|
| 2144 |
|
| 2145 | void timer_lock_remote_bases(unsigned int cpu)
|
| 2146 | __acquires(timer_bases[BASE_LOCAL]->lock)
|
| 2147 | __acquires(timer_bases[BASE_GLOBAL]->lock)
|
| 2148 | {
|
| 2149 | struct timer_base *base_local, *base_global;
|
| 2150 |
|
| 2151 | base_local = per_cpu_ptr(&timer_bases[BASE_LOCAL], cpu);
|
| 2152 | base_global = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
|
| 2153 |
|
| 2154 | lockdep_assert_irqs_disabled();
|
| 2155 |
|
| 2156 | raw_spin_lock(&base_local->lock);
|
| 2157 | raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
|
| 2158 | }
|
| 2159 |
|
| 2160 | |
| 2161 | * timer_base_is_idle() - Return whether timer base is set idle
|
| 2162 | *
|
| 2163 | * Returns value of local timer base is_idle value.
|
| 2164 |
|
| 2165 | bool timer_base_is_idle(void)
|
| 2166 | {
|
| 2167 | return __this_cpu_read(timer_bases[BASE_LOCAL].is_idle);
|
| 2168 | }
|
| 2169 |
|
| 2170 | static void __run_timer_base(struct timer_base *base);
|
| 2171 |
|
| 2172 | |
| 2173 | * timer_expire_remote() - expire global timers of cpu
|
| 2174 | * @cpu: Remote CPU
|
| 2175 | *
|
| 2176 | * Expire timers of global base of remote CPU.
|
| 2177 |
|
| 2178 | void timer_expire_remote(unsigned int cpu)
|
| 2179 | {
|
| 2180 | struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_GLOBAL], cpu);
|
| 2181 |
|
| 2182 | __run_timer_base(base);
|
| 2183 | }
|
| 2184 |
|
| 2185 | static void timer_use_tmigr(unsigned long basej, u64 basem,
|
| 2186 | unsigned long *nextevt, bool *tick_stop_path,
|
| 2187 | bool timer_base_idle, struct timer_events *tevt)
|
| 2188 | {
|
| 2189 | u64 next_tmigr;
|
| 2190 |
|
| 2191 | if (timer_base_idle)
|
| 2192 | next_tmigr = tmigr_cpu_new_timer(tevt->global);
|
| 2193 | else if (tick_stop_path)
|
| 2194 | next_tmigr = tmigr_cpu_deactivate(tevt->global);
|
| 2195 | else
|
| 2196 | next_tmigr = tmigr_quick_check(tevt->global);
|
| 2197 |
|
| 2198 | |
| 2199 | * If the CPU is the last going idle in timer migration hierarchy, make
|
| 2200 | * sure the CPU will wake up in time to handle remote timers.
|
| 2201 | * next_tmigr == KTIME_MAX if other CPUs are still active.
|
| 2202 |
|
| 2203 | if (next_tmigr < tevt->local) {
|
| 2204 | u64 tmp;
|
| 2205 |
|
| 2206 |
|
| 2207 | if (next_tmigr < basem)
|
| 2208 | next_tmigr = basem;
|
| 2209 |
|
| 2210 | tmp = div_u64(next_tmigr - basem, TICK_NSEC);
|
| 2211 |
|
| 2212 | *nextevt = basej + (unsigned long)tmp;
|
| 2213 | tevt->local = next_tmigr;
|
| 2214 | }
|
| 2215 | }
|
| 2216 | # else
|
| 2217 | static void timer_use_tmigr(unsigned long basej, u64 basem,
|
| 2218 | unsigned long *nextevt, bool *tick_stop_path,
|
| 2219 | bool timer_base_idle, struct timer_events *tevt)
|
| 2220 | {
|
| 2221 | |
| 2222 | * Make sure first event is written into tevt->local to not miss a
|
| 2223 | * timer on !SMP systems.
|
| 2224 |
|
| 2225 | tevt->local = min_t(u64, tevt->local, tevt->global);
|
| 2226 | }
|
| 2227 | # endif
|
| 2228 |
|
| 2229 | static inline u64 __get_next_timer_interrupt(unsigned long basej, u64 basem,
|
| 2230 | bool *idle)
|
| 2231 | {
|
| 2232 | struct timer_events tevt = { .local = KTIME_MAX, .global = KTIME_MAX };
|
| 2233 | struct timer_base *base_local, *base_global;
|
| 2234 | unsigned long nextevt;
|
| 2235 | bool idle_is_possible;
|
| 2236 |
|
| 2237 | |
| 2238 | * When the CPU is offline, the tick is cancelled and nothing is supposed
|
| 2239 | * to try to stop it.
|
| 2240 |
|
| 2241 | if (WARN_ON_ONCE(cpu_is_offline(smp_processor_id()))) {
|
| 2242 | if (idle)
|
| 2243 | *idle = true;
|
| 2244 | return tevt.local;
|
| 2245 | }
|
| 2246 |
|
| 2247 | base_local = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
|
| 2248 | base_global = this_cpu_ptr(&timer_bases[BASE_GLOBAL]);
|
| 2249 |
|
| 2250 | raw_spin_lock(&base_local->lock);
|
| 2251 | raw_spin_lock_nested(&base_global->lock, SINGLE_DEPTH_NESTING);
|
| 2252 |
|
| 2253 | nextevt = fetch_next_timer_interrupt(basej, basem, base_local,
|
| 2254 | base_global, &tevt);
|
| 2255 |
|
| 2256 | |
| 2257 | * If the next event is only one jiffy ahead there is no need to call
|
| 2258 | * timer migration hierarchy related functions. The value for the next
|
| 2259 | * global timer in @tevt struct equals then KTIME_MAX. This is also
|
| 2260 | * true, when the timer base is idle.
|
| 2261 | *
|
| 2262 | * The proper timer migration hierarchy function depends on the callsite
|
| 2263 | * and whether timer base is idle or not. @nextevt will be updated when
|
| 2264 | * this CPU needs to handle the first timer migration hierarchy
|
| 2265 | * event. See timer_use_tmigr() for detailed information.
|
| 2266 |
|
| 2267 | idle_is_possible = time_after(nextevt, basej + 1);
|
| 2268 | if (idle_is_possible)
|
| 2269 | timer_use_tmigr(basej, basem, &nextevt, idle,
|
| 2270 | base_local->is_idle, &tevt);
|
| 2271 |
|
| 2272 | |
| 2273 | * We have a fresh next event. Check whether we can forward the
|
| 2274 | * base.
|
| 2275 |
|
| 2276 | __forward_timer_base(base_local, basej);
|
| 2277 | __forward_timer_base(base_global, basej);
|
| 2278 |
|
| 2279 | |
| 2280 | * Set base->is_idle only when caller is timer_base_try_to_set_idle()
|
| 2281 |
|
| 2282 | if (idle) {
|
| 2283 | |
| 2284 | * Bases are idle if the next event is more than a tick
|
| 2285 | * away. Caution: @nextevt could have changed by enqueueing a
|
| 2286 | * global timer into timer migration hierarchy. Therefore a new
|
| 2287 | * check is required here.
|
| 2288 | *
|
| 2289 | * If the base is marked idle then any timer add operation must
|
| 2290 | * forward the base clk itself to keep granularity small. This
|
| 2291 | * idle logic is only maintained for the BASE_LOCAL and
|
| 2292 | * BASE_GLOBAL base, deferrable timers may still see large
|
| 2293 | * granularity skew (by design).
|
| 2294 |
|
| 2295 | if (!base_local->is_idle && time_after(nextevt, basej + 1)) {
|
| 2296 | base_local->is_idle = true;
|
| 2297 | |
| 2298 | * Global timers queued locally while running in a task
|
| 2299 | * in nohz_full mode need a self-IPI to kick reprogramming
|
| 2300 | * in IRQ tail.
|
| 2301 |
|
| 2302 | if (tick_nohz_full_cpu(base_local->cpu))
|
| 2303 | base_global->is_idle = true;
|
| 2304 | trace_timer_base_idle(true, base_local->cpu);
|
| 2305 | }
|
| 2306 | *idle = base_local->is_idle;
|
| 2307 |
|
| 2308 | |
| 2309 | * When timer base is not set idle, undo the effect of
|
| 2310 | * tmigr_cpu_deactivate() to prevent inconsistent states - active
|
| 2311 | * timer base but inactive timer migration hierarchy.
|
| 2312 | *
|
| 2313 | * When timer base was already marked idle, nothing will be
|
| 2314 | * changed here.
|
| 2315 |
|
| 2316 | if (!base_local->is_idle && idle_is_possible)
|
| 2317 | tmigr_cpu_activate();
|
| 2318 | }
|
| 2319 |
|
| 2320 | raw_spin_unlock(&base_global->lock);
|
| 2321 | raw_spin_unlock(&base_local->lock);
|
| 2322 |
|
| 2323 | return cmp_next_hrtimer_event(basem, tevt.local);
|
| 2324 | }
|
| 2325 |
|
| 2326 | |
| 2327 | * get_next_timer_interrupt() - return the time (clock mono) of the next timer
|
| 2328 | * @basej: base time jiffies
|
| 2329 | * @basem: base time clock monotonic
|
| 2330 | *
|
| 2331 | * Returns the tick aligned clock monotonic time of the next pending timer or
|
| 2332 | * KTIME_MAX if no timer is pending. If timer of global base was queued into
|
| 2333 | * timer migration hierarchy, first global timer is not taken into account. If
|
| 2334 | * it was the last CPU of timer migration hierarchy going idle, first global
|
| 2335 | * event is taken into account.
|
| 2336 |
|
| 2337 | u64 get_next_timer_interrupt(unsigned long basej, u64 basem)
|
| 2338 | {
|
| 2339 | return __get_next_timer_interrupt(basej, basem, NULL);
|
| 2340 | }
|
| 2341 |
|
| 2342 | |
| 2343 | * timer_base_try_to_set_idle() - Try to set the idle state of the timer bases
|
| 2344 | * @basej: base time jiffies
|
| 2345 | * @basem: base time clock monotonic
|
| 2346 | * @idle: pointer to store the value of timer_base->is_idle on return;
|
| 2347 | * *idle contains the information whether tick was already stopped
|
| 2348 | *
|
| 2349 | * Returns the tick aligned clock monotonic time of the next pending timer or
|
| 2350 | * KTIME_MAX if no timer is pending. When tick was already stopped KTIME_MAX is
|
| 2351 | * returned as well.
|
| 2352 |
|
| 2353 | u64 timer_base_try_to_set_idle(unsigned long basej, u64 basem, bool *idle)
|
| 2354 | {
|
| 2355 | if (*idle)
|
| 2356 | return KTIME_MAX;
|
| 2357 |
|
| 2358 | return __get_next_timer_interrupt(basej, basem, idle);
|
| 2359 | }
|
| 2360 |
|
| 2361 | |
| 2362 | * timer_clear_idle - Clear the idle state of the timer base
|
| 2363 | *
|
| 2364 | * Called with interrupts disabled
|
| 2365 |
|
| 2366 | void timer_clear_idle(void)
|
| 2367 | {
|
| 2368 | |
| 2369 | * We do this unlocked. The worst outcome is a remote pinned timer
|
| 2370 | * enqueue sending a pointless IPI, but taking the lock would just
|
| 2371 | * make the window for sending the IPI a few instructions smaller
|
| 2372 | * for the cost of taking the lock in the exit from idle
|
| 2373 | * path. Required for BASE_LOCAL only.
|
| 2374 |
|
| 2375 | __this_cpu_write(timer_bases[BASE_LOCAL].is_idle, false);
|
| 2376 | if (tick_nohz_full_cpu(smp_processor_id()))
|
| 2377 | __this_cpu_write(timer_bases[BASE_GLOBAL].is_idle, false);
|
| 2378 | trace_timer_base_idle(false, smp_processor_id());
|
| 2379 |
|
| 2380 |
|
| 2381 | tmigr_cpu_activate();
|
| 2382 | }
|
| 2383 | #endif
|
| 2384 |
|
| 2385 | |
| 2386 | * __run_timers - run all expired timers (if any) on this CPU.
|
| 2387 | * @base: the timer vector to be processed.
|
| 2388 |
|
| 2389 | static inline void __run_timers(struct timer_base *base)
|
| 2390 | {
|
| 2391 | struct hlist_head heads[LVL_DEPTH];
|
| 2392 | int levels;
|
| 2393 |
|
| 2394 | lockdep_assert_held(&base->lock);
|
| 2395 |
|
| 2396 | if (base->running_timer)
|
| 2397 | return;
|
| 2398 |
|
| 2399 | while (time_after_eq(jiffies, base->clk) &&
|
| 2400 | time_after_eq(jiffies, base->next_expiry)) {
|
| 2401 | levels = collect_expired_timers(base, heads);
|
| 2402 | |
| 2403 | * The two possible reasons for not finding any expired
|
| 2404 | * timer at this clk are that all matching timers have been
|
| 2405 | * dequeued or no timer has been queued since
|
| 2406 | * base::next_expiry was set to base::clk +
|
| 2407 | * NEXT_TIMER_MAX_DELTA.
|
| 2408 |
|
| 2409 | WARN_ON_ONCE(!levels && !base->next_expiry_recalc
|
| 2410 | && base->timers_pending);
|
| 2411 | |
| 2412 | * While executing timers, base->clk is set 1 offset ahead of
|
| 2413 | * jiffies to avoid endless requeuing to current jiffies.
|
| 2414 |
|
| 2415 | base->clk++;
|
| 2416 | timer_recalc_next_expiry(base);
|
| 2417 |
|
| 2418 | while (levels--)
|
| 2419 | expire_timers(base, heads + levels);
|
| 2420 | }
|
| 2421 | }
|
| 2422 |
|
| 2423 | static void __run_timer_base(struct timer_base *base)
|
| 2424 | {
|
| 2425 | if (time_before(jiffies, base->next_expiry))
|
| 2426 | return;
|
| 2427 |
|
| 2428 | timer_base_lock_expiry(base);
|
| 2429 | raw_spin_lock_irq(&base->lock);
|
| 2430 | __run_timers(base);
|
| 2431 | raw_spin_unlock_irq(&base->lock);
|
| 2432 | timer_base_unlock_expiry(base);
|
| 2433 | }
|
| 2434 |
|
| 2435 | static void run_timer_base(int index)
|
| 2436 | {
|
| 2437 | struct timer_base *base = this_cpu_ptr(&timer_bases[index]);
|
| 2438 |
|
| 2439 | __run_timer_base(base);
|
| 2440 | }
|
| 2441 |
|
| 2442 | |
| 2443 | * This function runs timers and the timer-tq in bottom half context.
|
| 2444 |
|
| 2445 | static __latent_entropy void run_timer_softirq(void)
|
| 2446 | {
|
| 2447 | run_timer_base(BASE_LOCAL);
|
| 2448 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) {
|
| 2449 | run_timer_base(BASE_GLOBAL);
|
| 2450 | run_timer_base(BASE_DEF);
|
| 2451 |
|
| 2452 | if (is_timers_nohz_active())
|
| 2453 | tmigr_handle_remote();
|
| 2454 | }
|
| 2455 | }
|
| 2456 |
|
| 2457 | |
| 2458 | * Called by the local, per-CPU timer interrupt on SMP.
|
| 2459 |
|
| 2460 | static void run_local_timers(void)
|
| 2461 | {
|
| 2462 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_LOCAL]);
|
| 2463 |
|
| 2464 | hrtimer_run_queues();
|
| 2465 |
|
| 2466 | for (int i = 0; i < NR_BASES; i++, base++) {
|
| 2467 | |
| 2468 | * Raise the softirq only if required.
|
| 2469 | *
|
| 2470 | * timer_base::next_expiry can be written by a remote CPU while
|
| 2471 | * holding the lock. If this write happens at the same time than
|
| 2472 | * the lockless local read, sanity checker could complain about
|
| 2473 | * data corruption.
|
| 2474 | *
|
| 2475 | * There are two possible situations where
|
| 2476 | * timer_base::next_expiry is written by a remote CPU:
|
| 2477 | *
|
| 2478 | * 1. Remote CPU expires global timers of this CPU and updates
|
| 2479 | * timer_base::next_expiry of BASE_GLOBAL afterwards in
|
| 2480 | * next_timer_interrupt() or timer_recalc_next_expiry(). The
|
| 2481 | * worst outcome is a superfluous raise of the timer softirq
|
| 2482 | * when the not yet updated value is read.
|
| 2483 | *
|
| 2484 | * 2. A new first pinned timer is enqueued by a remote CPU
|
| 2485 | * and therefore timer_base::next_expiry of BASE_LOCAL is
|
| 2486 | * updated. When this update is missed, this isn't a
|
| 2487 | * problem, as an IPI is executed nevertheless when the CPU
|
| 2488 | * was idle before. When the CPU wasn't idle but the update
|
| 2489 | * is missed, then the timer would expire one jiffy late -
|
| 2490 | * bad luck.
|
| 2491 | *
|
| 2492 | * Those unlikely corner cases where the worst outcome is only a
|
| 2493 | * one jiffy delay or a superfluous raise of the softirq are
|
| 2494 | * not that expensive as doing the check always while holding
|
| 2495 | * the lock.
|
| 2496 | *
|
| 2497 | * Possible remote writers are using WRITE_ONCE(). Local reader
|
| 2498 | * uses therefore READ_ONCE().
|
| 2499 |
|
| 2500 | if (time_after_eq(jiffies, READ_ONCE(base->next_expiry)) ||
|
| 2501 | (i == BASE_DEF && tmigr_requires_handle_remote())) {
|
| 2502 | raise_softirq(TIMER_SOFTIRQ);
|
| 2503 | return;
|
| 2504 | }
|
| 2505 | }
|
| 2506 | }
|
| 2507 |
|
| 2508 | |
| 2509 | * Called from the timer interrupt handler to charge one tick to the current
|
| 2510 | * process. user_tick is 1 if the tick is user time, 0 for system.
|
| 2511 |
|
| 2512 | void update_process_times(int user_tick)
|
| 2513 | {
|
| 2514 | struct task_struct *p = current;
|
| 2515 |
|
| 2516 |
|
| 2517 | account_process_tick(p, user_tick);
|
| 2518 | run_local_timers();
|
| 2519 | rcu_sched_clock_irq(user_tick);
|
| 2520 | #ifdef CONFIG_IRQ_WORK
|
| 2521 | if (in_irq())
|
| 2522 | irq_work_tick();
|
| 2523 | #endif
|
| 2524 | sched_tick();
|
| 2525 | if (IS_ENABLED(CONFIG_POSIX_TIMERS))
|
| 2526 | run_posix_cpu_timers();
|
| 2527 | }
|
| 2528 |
|
| 2529 | |
| 2530 | * Since schedule_timeout()'s timer is defined on the stack, it must store
|
| 2531 | * the target task on the stack as well.
|
| 2532 |
|
| 2533 | struct process_timer {
|
| 2534 | struct timer_list timer;
|
| 2535 | struct task_struct *task;
|
| 2536 | };
|
| 2537 |
|
| 2538 | static void process_timeout(struct timer_list *t)
|
| 2539 | {
|
| 2540 | struct process_timer *timeout = from_timer(timeout, t, timer);
|
| 2541 |
|
| 2542 | wake_up_process(timeout->task);
|
| 2543 | }
|
| 2544 |
|
| 2545 | |
| 2546 | * schedule_timeout - sleep until timeout
|
| 2547 | * @timeout: timeout value in jiffies
|
| 2548 | *
|
| 2549 | * Make the current task sleep until @timeout jiffies have elapsed.
|
| 2550 | * The function behavior depends on the current task state
|
| 2551 | * (see also set_current_state() description):
|
| 2552 | *
|
| 2553 | * %TASK_RUNNING - the scheduler is called, but the task does not sleep
|
| 2554 | * at all. That happens because sched_submit_work() does nothing for
|
| 2555 | * tasks in %TASK_RUNNING state.
|
| 2556 | *
|
| 2557 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
|
| 2558 | * pass before the routine returns unless the current task is explicitly
|
| 2559 | * woken up, (e.g. by wake_up_process()).
|
| 2560 | *
|
| 2561 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
|
| 2562 | * delivered to the current task or the current task is explicitly woken
|
| 2563 | * up.
|
| 2564 | *
|
| 2565 | * The current task state is guaranteed to be %TASK_RUNNING when this
|
| 2566 | * routine returns.
|
| 2567 | *
|
| 2568 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
|
| 2569 | * the CPU away without a bound on the timeout. In this case the return
|
| 2570 | * value will be %MAX_SCHEDULE_TIMEOUT.
|
| 2571 | *
|
| 2572 | * Returns 0 when the timer has expired otherwise the remaining time in
|
| 2573 | * jiffies will be returned. In all cases the return value is guaranteed
|
| 2574 | * to be non-negative.
|
| 2575 |
|
| 2576 | signed long __sched schedule_timeout(signed long timeout)
|
| 2577 | {
|
| 2578 | struct process_timer timer;
|
| 2579 | unsigned long expire;
|
| 2580 |
|
| 2581 | switch (timeout)
|
| 2582 | {
|
| 2583 | case MAX_SCHEDULE_TIMEOUT:
|
| 2584 | |
| 2585 | * These two special cases are useful to be comfortable
|
| 2586 | * in the caller. Nothing more. We could take
|
| 2587 | * MAX_SCHEDULE_TIMEOUT from one of the negative value
|
| 2588 | * but I' d like to return a valid offset (>=0) to allow
|
| 2589 | * the caller to do everything it want with the retval.
|
| 2590 |
|
| 2591 | schedule();
|
| 2592 | goto out;
|
| 2593 | default:
|
| 2594 | |
| 2595 | * Another bit of PARANOID. Note that the retval will be
|
| 2596 | * 0 since no piece of kernel is supposed to do a check
|
| 2597 | * for a negative retval of schedule_timeout() (since it
|
| 2598 | * should never happens anyway). You just have the printk()
|
| 2599 | * that will tell you if something is gone wrong and where.
|
| 2600 |
|
| 2601 | if (timeout < 0) {
|
| 2602 | printk(KERN_ERR "schedule_timeout: wrong timeout "
|
| 2603 | "value %lx\n", timeout);
|
| 2604 | dump_stack();
|
| 2605 | __set_current_state(TASK_RUNNING);
|
| 2606 | goto out;
|
| 2607 | }
|
| 2608 | }
|
| 2609 |
|
| 2610 | expire = timeout + jiffies;
|
| 2611 |
|
| 2612 | timer.task = current;
|
| 2613 | timer_setup_on_stack(&timer.timer, process_timeout, 0);
|
| 2614 | __mod_timer(&timer.timer, expire, MOD_TIMER_NOTPENDING);
|
| 2615 | schedule();
|
| 2616 | del_timer_sync(&timer.timer);
|
| 2617 |
|
| 2618 |
|
| 2619 | destroy_timer_on_stack(&timer.timer);
|
| 2620 |
|
| 2621 | timeout = expire - jiffies;
|
| 2622 |
|
| 2623 | out:
|
| 2624 | return timeout < 0 ? 0 : timeout;
|
| 2625 | }
|
| 2626 | EXPORT_SYMBOL(schedule_timeout);
|
| 2627 |
|
| 2628 | |
| 2629 | * We can use __set_current_state() here because schedule_timeout() calls
|
| 2630 | * schedule() unconditionally.
|
| 2631 |
|
| 2632 | signed long __sched schedule_timeout_interruptible(signed long timeout)
|
| 2633 | {
|
| 2634 | __set_current_state(TASK_INTERRUPTIBLE);
|
| 2635 | return schedule_timeout(timeout);
|
| 2636 | }
|
| 2637 | EXPORT_SYMBOL(schedule_timeout_interruptible);
|
| 2638 |
|
| 2639 | signed long __sched schedule_timeout_killable(signed long timeout)
|
| 2640 | {
|
| 2641 | __set_current_state(TASK_KILLABLE);
|
| 2642 | return schedule_timeout(timeout);
|
| 2643 | }
|
| 2644 | EXPORT_SYMBOL(schedule_timeout_killable);
|
| 2645 |
|
| 2646 | signed long __sched schedule_timeout_uninterruptible(signed long timeout)
|
| 2647 | {
|
| 2648 | __set_current_state(TASK_UNINTERRUPTIBLE);
|
| 2649 | return schedule_timeout(timeout);
|
| 2650 | }
|
| 2651 | EXPORT_SYMBOL(schedule_timeout_uninterruptible);
|
| 2652 |
|
| 2653 | |
| 2654 | * Like schedule_timeout_uninterruptible(), except this task will not contribute
|
| 2655 | * to load average.
|
| 2656 |
|
| 2657 | signed long __sched schedule_timeout_idle(signed long timeout)
|
| 2658 | {
|
| 2659 | __set_current_state(TASK_IDLE);
|
| 2660 | return schedule_timeout(timeout);
|
| 2661 | }
|
| 2662 | EXPORT_SYMBOL(schedule_timeout_idle);
|
| 2663 |
|
| 2664 | #ifdef CONFIG_HOTPLUG_CPU
|
| 2665 | static void migrate_timer_list(struct timer_base *new_base, struct hlist_head *head)
|
| 2666 | {
|
| 2667 | struct timer_list *timer;
|
| 2668 | int cpu = new_base->cpu;
|
| 2669 |
|
| 2670 | while (!hlist_empty(head)) {
|
| 2671 | timer = hlist_entry(head->first, struct timer_list, entry);
|
| 2672 | detach_timer(timer, false);
|
| 2673 | timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
|
| 2674 | internal_add_timer(new_base, timer);
|
| 2675 | }
|
| 2676 | }
|
| 2677 |
|
| 2678 | int timers_prepare_cpu(unsigned int cpu)
|
| 2679 | {
|
| 2680 | struct timer_base *base;
|
| 2681 | int b;
|
| 2682 |
|
| 2683 | for (b = 0; b < NR_BASES; b++) {
|
| 2684 | base = per_cpu_ptr(&timer_bases[b], cpu);
|
| 2685 | base->clk = jiffies;
|
| 2686 | base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
|
| 2687 | base->next_expiry_recalc = false;
|
| 2688 | base->timers_pending = false;
|
| 2689 | base->is_idle = false;
|
| 2690 | }
|
| 2691 | return 0;
|
| 2692 | }
|
| 2693 |
|
| 2694 | int timers_dead_cpu(unsigned int cpu)
|
| 2695 | {
|
| 2696 | struct timer_base *old_base;
|
| 2697 | struct timer_base *new_base;
|
| 2698 | int b, i;
|
| 2699 |
|
| 2700 | for (b = 0; b < NR_BASES; b++) {
|
| 2701 | old_base = per_cpu_ptr(&timer_bases[b], cpu);
|
| 2702 | new_base = get_cpu_ptr(&timer_bases[b]);
|
| 2703 | |
| 2704 | * The caller is globally serialized and nobody else
|
| 2705 | * takes two locks at once, deadlock is not possible.
|
| 2706 |
|
| 2707 | raw_spin_lock_irq(&new_base->lock);
|
| 2708 | raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
|
| 2709 |
|
| 2710 | |
| 2711 | * The current CPUs base clock might be stale. Update it
|
| 2712 | * before moving the timers over.
|
| 2713 |
|
| 2714 | forward_timer_base(new_base);
|
| 2715 |
|
| 2716 | WARN_ON_ONCE(old_base->running_timer);
|
| 2717 | old_base->running_timer = NULL;
|
| 2718 |
|
| 2719 | for (i = 0; i < WHEEL_SIZE; i++)
|
| 2720 | migrate_timer_list(new_base, old_base->vectors + i);
|
| 2721 |
|
| 2722 | raw_spin_unlock(&old_base->lock);
|
| 2723 | raw_spin_unlock_irq(&new_base->lock);
|
| 2724 | put_cpu_ptr(&timer_bases);
|
| 2725 | }
|
| 2726 | return 0;
|
| 2727 | }
|
| 2728 |
|
| 2729 | #endif
|
| 2730 |
|
| 2731 | static void __init init_timer_cpu(int cpu)
|
| 2732 | {
|
| 2733 | struct timer_base *base;
|
| 2734 | int i;
|
| 2735 |
|
| 2736 | for (i = 0; i < NR_BASES; i++) {
|
| 2737 | base = per_cpu_ptr(&timer_bases[i], cpu);
|
| 2738 | base->cpu = cpu;
|
| 2739 | raw_spin_lock_init(&base->lock);
|
| 2740 | base->clk = jiffies;
|
| 2741 | base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA;
|
| 2742 | timer_base_init_expiry_lock(base);
|
| 2743 | }
|
| 2744 | }
|
| 2745 |
|
| 2746 | static void __init init_timer_cpus(void)
|
| 2747 | {
|
| 2748 | int cpu;
|
| 2749 |
|
| 2750 | for_each_possible_cpu(cpu)
|
| 2751 | init_timer_cpu(cpu);
|
| 2752 | }
|
| 2753 |
|
| 2754 | void __init init_timers(void)
|
| 2755 | {
|
| 2756 | init_timer_cpus();
|
| 2757 | posix_cputimers_init_work();
|
| 2758 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
|
| 2759 | }
|
| 2760 |
|
| 2761 | |
| 2762 | * msleep - sleep safely even with waitqueue interruptions
|
| 2763 | * @msecs: Time in milliseconds to sleep for
|
| 2764 |
|
| 2765 | void msleep(unsigned int msecs)
|
| 2766 | {
|
| 2767 | unsigned long timeout = msecs_to_jiffies(msecs);
|
| 2768 |
|
| 2769 | while (timeout)
|
| 2770 | timeout = schedule_timeout_uninterruptible(timeout);
|
| 2771 | }
|
| 2772 |
|
| 2773 | EXPORT_SYMBOL(msleep);
|
| 2774 |
|
| 2775 | |
| 2776 | * msleep_interruptible - sleep waiting for signals
|
| 2777 | * @msecs: Time in milliseconds to sleep for
|
| 2778 |
|
| 2779 | unsigned long msleep_interruptible(unsigned int msecs)
|
| 2780 | {
|
| 2781 | unsigned long timeout = msecs_to_jiffies(msecs);
|
| 2782 |
|
| 2783 | while (timeout && !signal_pending(current))
|
| 2784 | timeout = schedule_timeout_interruptible(timeout);
|
| 2785 | return jiffies_to_msecs(timeout);
|
| 2786 | }
|
| 2787 |
|
| 2788 | EXPORT_SYMBOL(msleep_interruptible);
|
| 2789 |
|
| 2790 | |
| 2791 | * usleep_range_state - Sleep for an approximate time in a given state
|
| 2792 | * @min: Minimum time in usecs to sleep
|
| 2793 | * @max: Maximum time in usecs to sleep
|
| 2794 | * @state: State of the current task that will be while sleeping
|
| 2795 | *
|
| 2796 | * In non-atomic context where the exact wakeup time is flexible, use
|
| 2797 | * usleep_range_state() instead of udelay(). The sleep improves responsiveness
|
| 2798 | * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces
|
| 2799 | * power usage by allowing hrtimers to take advantage of an already-
|
| 2800 | * scheduled interrupt instead of scheduling a new one just for this sleep.
|
| 2801 |
|
| 2802 | void __sched usleep_range_state(unsigned long min, unsigned long max,
|
| 2803 | unsigned int state)
|
| 2804 | {
|
| 2805 | ktime_t exp = ktime_add_us(ktime_get(), min);
|
| 2806 | u64 delta = (u64)(max - min) * NSEC_PER_USEC;
|
| 2807 |
|
| 2808 | for (;;) {
|
| 2809 | __set_current_state(state);
|
| 2810 |
|
| 2811 | if (!schedule_hrtimeout_range(&exp, delta, HRTIMER_MODE_ABS))
|
| 2812 | break;
|
| 2813 | }
|
| 2814 | }
|
| 2815 | EXPORT_SYMBOL(usleep_range_state);
|