Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/kernel/timer.c |
| 3 | * |
Stephen Rothwell | 4a22f16 | 2013-04-30 15:27:37 -0700 | [diff] [blame] | 4 | * Kernel internal timers |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * |
| 6 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 7 | * |
| 8 | * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better. |
| 9 | * |
| 10 | * 1997-09-10 Updated NTP code according to technical memorandum Jan '96 |
| 11 | * "A Kernel Model for Precision Timekeeping" by Dave Mills |
| 12 | * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to |
| 13 | * serialize accesses to xtime/lost_ticks). |
| 14 | * Copyright (C) 1998 Andrea Arcangeli |
| 15 | * 1999-03-10 Improved NTP compatibility by Ulrich Windl |
| 16 | * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love |
| 17 | * 2000-10-05 Implemented scalable SMP per-CPU timer handling. |
| 18 | * Copyright (C) 2000, 2001, 2002 Ingo Molnar |
| 19 | * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar |
| 20 | */ |
| 21 | |
| 22 | #include <linux/kernel_stat.h> |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 23 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/interrupt.h> |
| 25 | #include <linux/percpu.h> |
| 26 | #include <linux/init.h> |
| 27 | #include <linux/mm.h> |
| 28 | #include <linux/swap.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 29 | #include <linux/pid_namespace.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/notifier.h> |
| 31 | #include <linux/thread_info.h> |
| 32 | #include <linux/time.h> |
| 33 | #include <linux/jiffies.h> |
| 34 | #include <linux/posix-timers.h> |
| 35 | #include <linux/cpu.h> |
| 36 | #include <linux/syscalls.h> |
Adrian Bunk | 97a41e2 | 2006-01-08 01:02:17 -0800 | [diff] [blame] | 37 | #include <linux/delay.h> |
Thomas Gleixner | 79bf2bb | 2007-02-16 01:28:03 -0800 | [diff] [blame] | 38 | #include <linux/tick.h> |
Ingo Molnar | 82f67cd | 2007-02-16 01:28:13 -0800 | [diff] [blame] | 39 | #include <linux/kallsyms.h> |
Peter Zijlstra | e360adb | 2010-10-14 14:01:34 +0800 | [diff] [blame] | 40 | #include <linux/irq_work.h> |
Arun R Bharadwaj | eea08f3 | 2009-04-16 12:16:41 +0530 | [diff] [blame] | 41 | #include <linux/sched.h> |
Clark Williams | cf4aebc2 | 2013-02-07 09:46:59 -0600 | [diff] [blame] | 42 | #include <linux/sched/sysctl.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 43 | #include <linux/slab.h> |
Stephen Rothwell | 1a0df59 | 2013-04-30 15:27:34 -0700 | [diff] [blame] | 44 | #include <linux/compat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | |
| 46 | #include <asm/uaccess.h> |
| 47 | #include <asm/unistd.h> |
| 48 | #include <asm/div64.h> |
| 49 | #include <asm/timex.h> |
| 50 | #include <asm/io.h> |
| 51 | |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 52 | #include "tick-internal.h" |
| 53 | |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 54 | #define CREATE_TRACE_POINTS |
| 55 | #include <trace/events/timer.h> |
| 56 | |
Andi Kleen | 40747ff | 2014-02-08 08:51:59 +0100 | [diff] [blame] | 57 | __visible u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES; |
Thomas Gleixner | ecea8d1 | 2005-10-30 15:03:00 -0800 | [diff] [blame] | 58 | |
| 59 | EXPORT_SYMBOL(jiffies_64); |
| 60 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | /* |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 62 | * The timer wheel has LVL_DEPTH array levels. Each level provides an array of |
| 63 | * LVL_SIZE buckets. Each level is driven by its own clock and therefor each |
| 64 | * level has a different granularity. |
| 65 | * |
| 66 | * The level granularity is: LVL_CLK_DIV ^ lvl |
| 67 | * The level clock frequency is: HZ / (LVL_CLK_DIV ^ level) |
| 68 | * |
| 69 | * The array level of a newly armed timer depends on the relative expiry |
| 70 | * time. The farther the expiry time is away the higher the array level and |
| 71 | * therefor the granularity becomes. |
| 72 | * |
| 73 | * Contrary to the original timer wheel implementation, which aims for 'exact' |
| 74 | * expiry of the timers, this implementation removes the need for recascading |
| 75 | * the timers into the lower array levels. The previous 'classic' timer wheel |
| 76 | * implementation of the kernel already violated the 'exact' expiry by adding |
| 77 | * slack to the expiry time to provide batched expiration. The granularity |
| 78 | * levels provide implicit batching. |
| 79 | * |
| 80 | * This is an optimization of the original timer wheel implementation for the |
| 81 | * majority of the timer wheel use cases: timeouts. The vast majority of |
| 82 | * timeout timers (networking, disk I/O ...) are canceled before expiry. If |
| 83 | * the timeout expires it indicates that normal operation is disturbed, so it |
| 84 | * does not matter much whether the timeout comes with a slight delay. |
| 85 | * |
| 86 | * The only exception to this are networking timers with a small expiry |
| 87 | * time. They rely on the granularity. Those fit into the first wheel level, |
| 88 | * which has HZ granularity. |
| 89 | * |
| 90 | * We don't have cascading anymore. timers with a expiry time above the |
| 91 | * capacity of the last wheel level are force expired at the maximum timeout |
| 92 | * value of the last wheel level. From data sampling we know that the maximum |
| 93 | * value observed is 5 days (network connection tracking), so this should not |
| 94 | * be an issue. |
| 95 | * |
| 96 | * The currently chosen array constants values are a good compromise between |
| 97 | * array size and granularity. |
| 98 | * |
| 99 | * This results in the following granularity and range levels: |
| 100 | * |
| 101 | * HZ 1000 steps |
| 102 | * Level Offset Granularity Range |
| 103 | * 0 0 1 ms 0 ms - 63 ms |
| 104 | * 1 64 8 ms 64 ms - 511 ms |
| 105 | * 2 128 64 ms 512 ms - 4095 ms (512ms - ~4s) |
| 106 | * 3 192 512 ms 4096 ms - 32767 ms (~4s - ~32s) |
| 107 | * 4 256 4096 ms (~4s) 32768 ms - 262143 ms (~32s - ~4m) |
| 108 | * 5 320 32768 ms (~32s) 262144 ms - 2097151 ms (~4m - ~34m) |
| 109 | * 6 384 262144 ms (~4m) 2097152 ms - 16777215 ms (~34m - ~4h) |
| 110 | * 7 448 2097152 ms (~34m) 16777216 ms - 134217727 ms (~4h - ~1d) |
| 111 | * 8 512 16777216 ms (~4h) 134217728 ms - 1073741822 ms (~1d - ~12d) |
| 112 | * |
| 113 | * HZ 300 |
| 114 | * Level Offset Granularity Range |
| 115 | * 0 0 3 ms 0 ms - 210 ms |
| 116 | * 1 64 26 ms 213 ms - 1703 ms (213ms - ~1s) |
| 117 | * 2 128 213 ms 1706 ms - 13650 ms (~1s - ~13s) |
| 118 | * 3 192 1706 ms (~1s) 13653 ms - 109223 ms (~13s - ~1m) |
| 119 | * 4 256 13653 ms (~13s) 109226 ms - 873810 ms (~1m - ~14m) |
| 120 | * 5 320 109226 ms (~1m) 873813 ms - 6990503 ms (~14m - ~1h) |
| 121 | * 6 384 873813 ms (~14m) 6990506 ms - 55924050 ms (~1h - ~15h) |
| 122 | * 7 448 6990506 ms (~1h) 55924053 ms - 447392423 ms (~15h - ~5d) |
| 123 | * 8 512 55924053 ms (~15h) 447392426 ms - 3579139406 ms (~5d - ~41d) |
| 124 | * |
| 125 | * HZ 250 |
| 126 | * Level Offset Granularity Range |
| 127 | * 0 0 4 ms 0 ms - 255 ms |
| 128 | * 1 64 32 ms 256 ms - 2047 ms (256ms - ~2s) |
| 129 | * 2 128 256 ms 2048 ms - 16383 ms (~2s - ~16s) |
| 130 | * 3 192 2048 ms (~2s) 16384 ms - 131071 ms (~16s - ~2m) |
| 131 | * 4 256 16384 ms (~16s) 131072 ms - 1048575 ms (~2m - ~17m) |
| 132 | * 5 320 131072 ms (~2m) 1048576 ms - 8388607 ms (~17m - ~2h) |
| 133 | * 6 384 1048576 ms (~17m) 8388608 ms - 67108863 ms (~2h - ~18h) |
| 134 | * 7 448 8388608 ms (~2h) 67108864 ms - 536870911 ms (~18h - ~6d) |
| 135 | * 8 512 67108864 ms (~18h) 536870912 ms - 4294967288 ms (~6d - ~49d) |
| 136 | * |
| 137 | * HZ 100 |
| 138 | * Level Offset Granularity Range |
| 139 | * 0 0 10 ms 0 ms - 630 ms |
| 140 | * 1 64 80 ms 640 ms - 5110 ms (640ms - ~5s) |
| 141 | * 2 128 640 ms 5120 ms - 40950 ms (~5s - ~40s) |
| 142 | * 3 192 5120 ms (~5s) 40960 ms - 327670 ms (~40s - ~5m) |
| 143 | * 4 256 40960 ms (~40s) 327680 ms - 2621430 ms (~5m - ~43m) |
| 144 | * 5 320 327680 ms (~5m) 2621440 ms - 20971510 ms (~43m - ~5h) |
| 145 | * 6 384 2621440 ms (~43m) 20971520 ms - 167772150 ms (~5h - ~1d) |
| 146 | * 7 448 20971520 ms (~5h) 167772160 ms - 1342177270 ms (~1d - ~15d) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 149 | /* Clock divisor for the next level */ |
| 150 | #define LVL_CLK_SHIFT 3 |
| 151 | #define LVL_CLK_DIV (1UL << LVL_CLK_SHIFT) |
| 152 | #define LVL_CLK_MASK (LVL_CLK_DIV - 1) |
| 153 | #define LVL_SHIFT(n) ((n) * LVL_CLK_SHIFT) |
| 154 | #define LVL_GRAN(n) (1UL << LVL_SHIFT(n)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 156 | /* |
| 157 | * The time start value for each level to select the bucket at enqueue |
| 158 | * time. |
| 159 | */ |
| 160 | #define LVL_START(n) ((LVL_SIZE - 1) << (((n) - 1) * LVL_CLK_SHIFT)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 162 | /* Size of each clock level */ |
| 163 | #define LVL_BITS 6 |
| 164 | #define LVL_SIZE (1UL << LVL_BITS) |
| 165 | #define LVL_MASK (LVL_SIZE - 1) |
| 166 | #define LVL_OFFS(n) ((n) * LVL_SIZE) |
| 167 | |
| 168 | /* Level depth */ |
| 169 | #if HZ > 100 |
| 170 | # define LVL_DEPTH 9 |
| 171 | # else |
| 172 | # define LVL_DEPTH 8 |
| 173 | #endif |
| 174 | |
| 175 | /* The cutoff (max. capacity of the wheel) */ |
| 176 | #define WHEEL_TIMEOUT_CUTOFF (LVL_START(LVL_DEPTH)) |
| 177 | #define WHEEL_TIMEOUT_MAX (WHEEL_TIMEOUT_CUTOFF - LVL_GRAN(LVL_DEPTH - 1)) |
| 178 | |
| 179 | /* |
| 180 | * The resulting wheel size. If NOHZ is configured we allocate two |
| 181 | * wheels so we have a separate storage for the deferrable timers. |
| 182 | */ |
| 183 | #define WHEEL_SIZE (LVL_SIZE * LVL_DEPTH) |
| 184 | |
| 185 | #ifdef CONFIG_NO_HZ_COMMON |
| 186 | # define NR_BASES 2 |
| 187 | # define BASE_STD 0 |
| 188 | # define BASE_DEF 1 |
| 189 | #else |
| 190 | # define NR_BASES 1 |
| 191 | # define BASE_STD 0 |
| 192 | # define BASE_DEF 0 |
| 193 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 195 | struct timer_base { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 196 | spinlock_t lock; |
| 197 | struct timer_list *running_timer; |
| 198 | unsigned long clk; |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 199 | unsigned long next_expiry; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 200 | unsigned int cpu; |
| 201 | bool migration_enabled; |
| 202 | bool nohz_active; |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 203 | bool is_idle; |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 204 | bool must_forward_clk; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 205 | DECLARE_BITMAP(pending_map, WHEEL_SIZE); |
| 206 | struct hlist_head vectors[WHEEL_SIZE]; |
Venki Pallipadi | 6e453a6 | 2007-05-08 00:27:44 -0700 | [diff] [blame] | 207 | } ____cacheline_aligned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 209 | static DEFINE_PER_CPU(struct timer_base, timer_bases[NR_BASES]); |
Kyle Yan | c1f109c | 2017-02-24 16:49:27 -0800 | [diff] [blame] | 210 | struct timer_base timer_base_deferrable; |
Lingutla Chandrasekhar | ea6a324 | 2018-03-01 20:41:57 +0530 | [diff] [blame] | 211 | static atomic_t deferrable_pending; |
Venki Pallipadi | 6e453a6 | 2007-05-08 00:27:44 -0700 | [diff] [blame] | 212 | |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 213 | #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON) |
| 214 | unsigned int sysctl_timer_migration = 1; |
| 215 | |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 216 | void timers_update_migration(bool update_nohz) |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 217 | { |
| 218 | bool on = sysctl_timer_migration && tick_nohz_active; |
| 219 | unsigned int cpu; |
| 220 | |
| 221 | /* Avoid the loop, if nothing to update */ |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 222 | if (this_cpu_read(timer_bases[BASE_STD].migration_enabled) == on) |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 223 | return; |
| 224 | |
| 225 | for_each_possible_cpu(cpu) { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 226 | per_cpu(timer_bases[BASE_STD].migration_enabled, cpu) = on; |
| 227 | per_cpu(timer_bases[BASE_DEF].migration_enabled, cpu) = on; |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 228 | per_cpu(hrtimer_bases.migration_enabled, cpu) = on; |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 229 | if (!update_nohz) |
| 230 | continue; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 231 | per_cpu(timer_bases[BASE_STD].nohz_active, cpu) = true; |
| 232 | per_cpu(timer_bases[BASE_DEF].nohz_active, cpu) = true; |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 233 | per_cpu(hrtimer_bases.nohz_active, cpu) = true; |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 234 | } |
Kyle Yan | c1f109c | 2017-02-24 16:49:27 -0800 | [diff] [blame] | 235 | |
| 236 | timer_base_deferrable.migration_enabled = on; |
| 237 | timer_base_deferrable.nohz_active = true; |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 238 | } |
| 239 | |
| 240 | int timer_migration_handler(struct ctl_table *table, int write, |
| 241 | void __user *buffer, size_t *lenp, |
| 242 | loff_t *ppos) |
| 243 | { |
| 244 | static DEFINE_MUTEX(mutex); |
| 245 | int ret; |
| 246 | |
| 247 | mutex_lock(&mutex); |
Myungho Jung | 4c00015 | 2017-04-19 15:24:50 -0700 | [diff] [blame] | 248 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 249 | if (!ret && write) |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 250 | timers_update_migration(false); |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 251 | mutex_unlock(&mutex); |
| 252 | return ret; |
| 253 | } |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 254 | #endif |
| 255 | |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 256 | static unsigned long round_jiffies_common(unsigned long j, int cpu, |
| 257 | bool force_up) |
| 258 | { |
| 259 | int rem; |
| 260 | unsigned long original = j; |
| 261 | |
| 262 | /* |
| 263 | * We don't want all cpus firing their timers at once hitting the |
| 264 | * same lock or cachelines, so we skew each extra cpu with an extra |
| 265 | * 3 jiffies. This 3 jiffies came originally from the mm/ code which |
| 266 | * already did this. |
| 267 | * The skew is done by adding 3*cpunr, then round, then subtract this |
| 268 | * extra offset again. |
| 269 | */ |
| 270 | j += cpu * 3; |
| 271 | |
| 272 | rem = j % HZ; |
| 273 | |
| 274 | /* |
| 275 | * If the target jiffie is just after a whole second (which can happen |
| 276 | * due to delays of the timer irq, long irq off times etc etc) then |
| 277 | * we should round down to the whole second, not up. Use 1/4th second |
| 278 | * as cutoff for this rounding as an extreme upper bound for this. |
| 279 | * But never round down if @force_up is set. |
| 280 | */ |
| 281 | if (rem < HZ/4 && !force_up) /* round down */ |
| 282 | j = j - rem; |
| 283 | else /* round up */ |
| 284 | j = j - rem + HZ; |
| 285 | |
| 286 | /* now that we have rounded, subtract the extra skew again */ |
| 287 | j -= cpu * 3; |
| 288 | |
Bart Van Assche | 9e04d38 | 2013-05-21 20:43:50 +0200 | [diff] [blame] | 289 | /* |
| 290 | * Make sure j is still in the future. Otherwise return the |
| 291 | * unmodified value. |
| 292 | */ |
| 293 | return time_is_after_jiffies(j) ? j : original; |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 294 | } |
| 295 | |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 296 | /** |
| 297 | * __round_jiffies - function to round jiffies to a full second |
| 298 | * @j: the time in (absolute) jiffies that should be rounded |
| 299 | * @cpu: the processor number on which the timeout will happen |
| 300 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 301 | * __round_jiffies() rounds an absolute time in the future (in jiffies) |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 302 | * up or down to (approximately) full seconds. This is useful for timers |
| 303 | * for which the exact time they fire does not matter too much, as long as |
| 304 | * they fire approximately every X seconds. |
| 305 | * |
| 306 | * By rounding these timers to whole seconds, all such timers will fire |
| 307 | * at the same time, rather than at various times spread out. The goal |
| 308 | * of this is to have the CPU wake up less, which saves power. |
| 309 | * |
| 310 | * The exact rounding is skewed for each processor to avoid all |
| 311 | * processors firing at the exact same time, which could lead |
| 312 | * to lock contention or spurious cache line bouncing. |
| 313 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 314 | * The return value is the rounded version of the @j parameter. |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 315 | */ |
| 316 | unsigned long __round_jiffies(unsigned long j, int cpu) |
| 317 | { |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 318 | return round_jiffies_common(j, cpu, false); |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 319 | } |
| 320 | EXPORT_SYMBOL_GPL(__round_jiffies); |
| 321 | |
| 322 | /** |
| 323 | * __round_jiffies_relative - function to round jiffies to a full second |
| 324 | * @j: the time in (relative) jiffies that should be rounded |
| 325 | * @cpu: the processor number on which the timeout will happen |
| 326 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 327 | * __round_jiffies_relative() rounds a time delta in the future (in jiffies) |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 328 | * up or down to (approximately) full seconds. This is useful for timers |
| 329 | * for which the exact time they fire does not matter too much, as long as |
| 330 | * they fire approximately every X seconds. |
| 331 | * |
| 332 | * By rounding these timers to whole seconds, all such timers will fire |
| 333 | * at the same time, rather than at various times spread out. The goal |
| 334 | * of this is to have the CPU wake up less, which saves power. |
| 335 | * |
| 336 | * The exact rounding is skewed for each processor to avoid all |
| 337 | * processors firing at the exact same time, which could lead |
| 338 | * to lock contention or spurious cache line bouncing. |
| 339 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 340 | * The return value is the rounded version of the @j parameter. |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 341 | */ |
| 342 | unsigned long __round_jiffies_relative(unsigned long j, int cpu) |
| 343 | { |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 344 | unsigned long j0 = jiffies; |
| 345 | |
| 346 | /* Use j0 because jiffies might change while we run */ |
| 347 | return round_jiffies_common(j + j0, cpu, false) - j0; |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 348 | } |
| 349 | EXPORT_SYMBOL_GPL(__round_jiffies_relative); |
| 350 | |
| 351 | /** |
| 352 | * round_jiffies - function to round jiffies to a full second |
| 353 | * @j: the time in (absolute) jiffies that should be rounded |
| 354 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 355 | * round_jiffies() rounds an absolute time in the future (in jiffies) |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 356 | * up or down to (approximately) full seconds. This is useful for timers |
| 357 | * for which the exact time they fire does not matter too much, as long as |
| 358 | * they fire approximately every X seconds. |
| 359 | * |
| 360 | * By rounding these timers to whole seconds, all such timers will fire |
| 361 | * at the same time, rather than at various times spread out. The goal |
| 362 | * of this is to have the CPU wake up less, which saves power. |
| 363 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 364 | * The return value is the rounded version of the @j parameter. |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 365 | */ |
| 366 | unsigned long round_jiffies(unsigned long j) |
| 367 | { |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 368 | return round_jiffies_common(j, raw_smp_processor_id(), false); |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 369 | } |
| 370 | EXPORT_SYMBOL_GPL(round_jiffies); |
| 371 | |
| 372 | /** |
| 373 | * round_jiffies_relative - function to round jiffies to a full second |
| 374 | * @j: the time in (relative) jiffies that should be rounded |
| 375 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 376 | * round_jiffies_relative() rounds a time delta in the future (in jiffies) |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 377 | * up or down to (approximately) full seconds. This is useful for timers |
| 378 | * for which the exact time they fire does not matter too much, as long as |
| 379 | * they fire approximately every X seconds. |
| 380 | * |
| 381 | * By rounding these timers to whole seconds, all such timers will fire |
| 382 | * at the same time, rather than at various times spread out. The goal |
| 383 | * of this is to have the CPU wake up less, which saves power. |
| 384 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 385 | * The return value is the rounded version of the @j parameter. |
Arjan van de Ven | 4c36a5d | 2006-12-10 02:21:24 -0800 | [diff] [blame] | 386 | */ |
| 387 | unsigned long round_jiffies_relative(unsigned long j) |
| 388 | { |
| 389 | return __round_jiffies_relative(j, raw_smp_processor_id()); |
| 390 | } |
| 391 | EXPORT_SYMBOL_GPL(round_jiffies_relative); |
| 392 | |
Alan Stern | 9c133c4 | 2008-11-06 08:42:48 +0100 | [diff] [blame] | 393 | /** |
| 394 | * __round_jiffies_up - function to round jiffies up to a full second |
| 395 | * @j: the time in (absolute) jiffies that should be rounded |
| 396 | * @cpu: the processor number on which the timeout will happen |
| 397 | * |
| 398 | * This is the same as __round_jiffies() except that it will never |
| 399 | * round down. This is useful for timeouts for which the exact time |
| 400 | * of firing does not matter too much, as long as they don't fire too |
| 401 | * early. |
| 402 | */ |
| 403 | unsigned long __round_jiffies_up(unsigned long j, int cpu) |
| 404 | { |
| 405 | return round_jiffies_common(j, cpu, true); |
| 406 | } |
| 407 | EXPORT_SYMBOL_GPL(__round_jiffies_up); |
| 408 | |
| 409 | /** |
| 410 | * __round_jiffies_up_relative - function to round jiffies up to a full second |
| 411 | * @j: the time in (relative) jiffies that should be rounded |
| 412 | * @cpu: the processor number on which the timeout will happen |
| 413 | * |
| 414 | * This is the same as __round_jiffies_relative() except that it will never |
| 415 | * round down. This is useful for timeouts for which the exact time |
| 416 | * of firing does not matter too much, as long as they don't fire too |
| 417 | * early. |
| 418 | */ |
| 419 | unsigned long __round_jiffies_up_relative(unsigned long j, int cpu) |
| 420 | { |
| 421 | unsigned long j0 = jiffies; |
| 422 | |
| 423 | /* Use j0 because jiffies might change while we run */ |
| 424 | return round_jiffies_common(j + j0, cpu, true) - j0; |
| 425 | } |
| 426 | EXPORT_SYMBOL_GPL(__round_jiffies_up_relative); |
| 427 | |
| 428 | /** |
| 429 | * round_jiffies_up - function to round jiffies up to a full second |
| 430 | * @j: the time in (absolute) jiffies that should be rounded |
| 431 | * |
| 432 | * This is the same as round_jiffies() except that it will never |
| 433 | * round down. This is useful for timeouts for which the exact time |
| 434 | * of firing does not matter too much, as long as they don't fire too |
| 435 | * early. |
| 436 | */ |
| 437 | unsigned long round_jiffies_up(unsigned long j) |
| 438 | { |
| 439 | return round_jiffies_common(j, raw_smp_processor_id(), true); |
| 440 | } |
| 441 | EXPORT_SYMBOL_GPL(round_jiffies_up); |
| 442 | |
| 443 | /** |
| 444 | * round_jiffies_up_relative - function to round jiffies up to a full second |
| 445 | * @j: the time in (relative) jiffies that should be rounded |
| 446 | * |
| 447 | * This is the same as round_jiffies_relative() except that it will never |
| 448 | * round down. This is useful for timeouts for which the exact time |
| 449 | * of firing does not matter too much, as long as they don't fire too |
| 450 | * early. |
| 451 | */ |
| 452 | unsigned long round_jiffies_up_relative(unsigned long j) |
| 453 | { |
| 454 | return __round_jiffies_up_relative(j, raw_smp_processor_id()); |
| 455 | } |
| 456 | EXPORT_SYMBOL_GPL(round_jiffies_up_relative); |
| 457 | |
Arjan van de Ven | 3bbb9ec | 2010-03-11 14:04:36 -0800 | [diff] [blame] | 458 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 459 | static inline unsigned int timer_get_idx(struct timer_list *timer) |
Venki Pallipadi | c5c061b8 | 2007-07-15 23:40:30 -0700 | [diff] [blame] | 460 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 461 | return (timer->flags & TIMER_ARRAYMASK) >> TIMER_ARRAYSHIFT; |
Venki Pallipadi | c5c061b8 | 2007-07-15 23:40:30 -0700 | [diff] [blame] | 462 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 463 | |
| 464 | static inline void timer_set_idx(struct timer_list *timer, unsigned int idx) |
| 465 | { |
| 466 | timer->flags = (timer->flags & ~TIMER_ARRAYMASK) | |
| 467 | idx << TIMER_ARRAYSHIFT; |
| 468 | } |
| 469 | |
| 470 | /* |
| 471 | * Helper function to calculate the array index for a given expiry |
| 472 | * time. |
| 473 | */ |
| 474 | static inline unsigned calc_index(unsigned expires, unsigned lvl) |
| 475 | { |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 476 | if (expires & ~(UINT_MAX << LVL_SHIFT(lvl))) |
| 477 | expires = (expires + LVL_GRAN(lvl)) >> LVL_SHIFT(lvl); |
| 478 | else |
| 479 | expires = expires >> LVL_SHIFT(lvl); |
| 480 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 481 | return LVL_OFFS(lvl) + (expires & LVL_MASK); |
| 482 | } |
| 483 | |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 484 | static inline unsigned int calc_index_min_granularity(unsigned int expires) |
| 485 | { |
| 486 | return LVL_OFFS(0) + ((expires >> LVL_SHIFT(0)) & LVL_MASK); |
| 487 | } |
| 488 | |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 489 | static int calc_wheel_index(unsigned long expires, unsigned long clk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | { |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 491 | unsigned long delta = expires - clk; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 492 | unsigned int idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 494 | if (delta < LVL_START(1)) { |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 495 | idx = calc_index_min_granularity(expires); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 496 | } else if (delta < LVL_START(2)) { |
| 497 | idx = calc_index(expires, 1); |
| 498 | } else if (delta < LVL_START(3)) { |
| 499 | idx = calc_index(expires, 2); |
| 500 | } else if (delta < LVL_START(4)) { |
| 501 | idx = calc_index(expires, 3); |
| 502 | } else if (delta < LVL_START(5)) { |
| 503 | idx = calc_index(expires, 4); |
| 504 | } else if (delta < LVL_START(6)) { |
| 505 | idx = calc_index(expires, 5); |
| 506 | } else if (delta < LVL_START(7)) { |
| 507 | idx = calc_index(expires, 6); |
| 508 | } else if (LVL_DEPTH > 8 && delta < LVL_START(8)) { |
| 509 | idx = calc_index(expires, 7); |
| 510 | } else if ((long) delta < 0) { |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 511 | idx = clk & LVL_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | } else { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 513 | /* |
| 514 | * Force expire obscene large timeouts to expire at the |
| 515 | * capacity limit of the wheel. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | */ |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 517 | if (expires >= WHEEL_TIMEOUT_CUTOFF) |
| 518 | expires = WHEEL_TIMEOUT_MAX; |
Thomas Gleixner | 1bd04bf | 2015-05-26 22:50:26 +0000 | [diff] [blame] | 519 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 520 | idx = calc_index(expires, LVL_DEPTH - 1); |
| 521 | } |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 522 | return idx; |
| 523 | } |
| 524 | |
| 525 | /* |
| 526 | * Enqueue the timer into the hash bucket, mark it pending in |
| 527 | * the bitmap and store the index in the timer flags. |
| 528 | */ |
| 529 | static void enqueue_timer(struct timer_base *base, struct timer_list *timer, |
| 530 | unsigned int idx) |
| 531 | { |
| 532 | hlist_add_head(&timer->entry, base->vectors + idx); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 533 | __set_bit(idx, base->pending_map); |
| 534 | timer_set_idx(timer, idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 535 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | |
| 537 | static void |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 538 | __internal_add_timer(struct timer_base *base, struct timer_list *timer) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | { |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 540 | unsigned int idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 542 | idx = calc_wheel_index(timer->expires, base->clk); |
| 543 | enqueue_timer(base, timer, idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | } |
| 545 | |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 546 | static void |
| 547 | trigger_dyntick_cpu(struct timer_base *base, struct timer_list *timer) |
Thomas Gleixner | facbb4a | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 548 | { |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 549 | if (!IS_ENABLED(CONFIG_NO_HZ_COMMON) || !base->nohz_active) |
| 550 | return; |
Viresh Kumar | 9f6d9ba | 2014-06-22 01:29:14 +0200 | [diff] [blame] | 551 | |
| 552 | /* |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 553 | * TODO: This wants some optimizing similar to the code below, but we |
| 554 | * will do that when we switch from push to pull for deferrable timers. |
Viresh Kumar | 9f6d9ba | 2014-06-22 01:29:14 +0200 | [diff] [blame] | 555 | */ |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 556 | if (timer->flags & TIMER_DEFERRABLE) { |
| 557 | if (tick_nohz_full_cpu(base->cpu)) |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 558 | wake_up_nohz_cpu(base->cpu); |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 559 | return; |
Thomas Gleixner | 683be13 | 2015-05-26 22:50:35 +0000 | [diff] [blame] | 560 | } |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 561 | |
| 562 | /* |
| 563 | * We might have to IPI the remote CPU if the base is idle and the |
| 564 | * timer is not deferrable. If the other CPU is on the way to idle |
| 565 | * then it can't set base->is_idle as we hold the base lock: |
| 566 | */ |
| 567 | if (!base->is_idle) |
| 568 | return; |
| 569 | |
| 570 | /* Check whether this is the new first expiring timer: */ |
| 571 | if (time_after_eq(timer->expires, base->next_expiry)) |
| 572 | return; |
| 573 | |
| 574 | /* |
| 575 | * Set the next expiry time and kick the CPU so it can reevaluate the |
| 576 | * wheel: |
| 577 | */ |
| 578 | base->next_expiry = timer->expires; |
Anna-Maria Gleixner | ffdf047 | 2016-07-04 09:50:39 +0000 | [diff] [blame] | 579 | wake_up_nohz_cpu(base->cpu); |
| 580 | } |
| 581 | |
| 582 | static void |
| 583 | internal_add_timer(struct timer_base *base, struct timer_list *timer) |
| 584 | { |
| 585 | __internal_add_timer(base, timer); |
| 586 | trigger_dyntick_cpu(base, timer); |
Thomas Gleixner | facbb4a | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 587 | } |
| 588 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 589 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
| 590 | |
| 591 | static struct debug_obj_descr timer_debug_descr; |
| 592 | |
Stanislaw Gruszka | 9977728 | 2011-03-07 09:58:33 +0100 | [diff] [blame] | 593 | static void *timer_debug_hint(void *addr) |
| 594 | { |
| 595 | return ((struct timer_list *) addr)->function; |
| 596 | } |
| 597 | |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 598 | static bool timer_is_static_object(void *addr) |
| 599 | { |
| 600 | struct timer_list *timer = addr; |
| 601 | |
| 602 | return (timer->entry.pprev == NULL && |
| 603 | timer->entry.next == TIMER_ENTRY_STATIC); |
| 604 | } |
| 605 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 606 | /* |
| 607 | * fixup_init is called when: |
| 608 | * - an active object is initialized |
| 609 | */ |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 610 | static bool timer_fixup_init(void *addr, enum debug_obj_state state) |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 611 | { |
| 612 | struct timer_list *timer = addr; |
| 613 | |
| 614 | switch (state) { |
| 615 | case ODEBUG_STATE_ACTIVE: |
| 616 | del_timer_sync(timer); |
| 617 | debug_object_init(timer, &timer_debug_descr); |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 618 | return true; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 619 | default: |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 620 | return false; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 621 | } |
| 622 | } |
| 623 | |
Stephen Boyd | fb16b8c | 2011-11-07 19:48:26 -0800 | [diff] [blame] | 624 | /* Stub timer callback for improperly used timers. */ |
| 625 | static void stub_timer(unsigned long data) |
| 626 | { |
| 627 | WARN_ON(1); |
| 628 | } |
| 629 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 630 | /* |
| 631 | * fixup_activate is called when: |
| 632 | * - an active object is activated |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 633 | * - an unknown non-static object is activated |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 634 | */ |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 635 | static bool timer_fixup_activate(void *addr, enum debug_obj_state state) |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 636 | { |
| 637 | struct timer_list *timer = addr; |
| 638 | |
| 639 | switch (state) { |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 640 | case ODEBUG_STATE_NOTAVAILABLE: |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 641 | setup_timer(timer, stub_timer, 0); |
| 642 | return true; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 643 | |
| 644 | case ODEBUG_STATE_ACTIVE: |
| 645 | WARN_ON(1); |
| 646 | |
| 647 | default: |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 648 | return false; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 649 | } |
| 650 | } |
| 651 | |
| 652 | /* |
| 653 | * fixup_free is called when: |
| 654 | * - an active object is freed |
| 655 | */ |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 656 | static bool timer_fixup_free(void *addr, enum debug_obj_state state) |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 657 | { |
| 658 | struct timer_list *timer = addr; |
| 659 | |
| 660 | switch (state) { |
| 661 | case ODEBUG_STATE_ACTIVE: |
| 662 | del_timer_sync(timer); |
| 663 | debug_object_free(timer, &timer_debug_descr); |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 664 | return true; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 665 | default: |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 666 | return false; |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 667 | } |
| 668 | } |
| 669 | |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 670 | /* |
| 671 | * fixup_assert_init is called when: |
| 672 | * - an untracked/uninit-ed object is found |
| 673 | */ |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 674 | static bool timer_fixup_assert_init(void *addr, enum debug_obj_state state) |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 675 | { |
| 676 | struct timer_list *timer = addr; |
| 677 | |
| 678 | switch (state) { |
| 679 | case ODEBUG_STATE_NOTAVAILABLE: |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 680 | setup_timer(timer, stub_timer, 0); |
| 681 | return true; |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 682 | default: |
Du, Changbin | e325246 | 2016-05-19 17:09:29 -0700 | [diff] [blame] | 683 | return false; |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 684 | } |
| 685 | } |
| 686 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 687 | static struct debug_obj_descr timer_debug_descr = { |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 688 | .name = "timer_list", |
| 689 | .debug_hint = timer_debug_hint, |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 690 | .is_static_object = timer_is_static_object, |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 691 | .fixup_init = timer_fixup_init, |
| 692 | .fixup_activate = timer_fixup_activate, |
| 693 | .fixup_free = timer_fixup_free, |
| 694 | .fixup_assert_init = timer_fixup_assert_init, |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 695 | }; |
| 696 | |
| 697 | static inline void debug_timer_init(struct timer_list *timer) |
| 698 | { |
| 699 | debug_object_init(timer, &timer_debug_descr); |
| 700 | } |
| 701 | |
| 702 | static inline void debug_timer_activate(struct timer_list *timer) |
| 703 | { |
| 704 | debug_object_activate(timer, &timer_debug_descr); |
| 705 | } |
| 706 | |
| 707 | static inline void debug_timer_deactivate(struct timer_list *timer) |
| 708 | { |
| 709 | debug_object_deactivate(timer, &timer_debug_descr); |
| 710 | } |
| 711 | |
| 712 | static inline void debug_timer_free(struct timer_list *timer) |
| 713 | { |
| 714 | debug_object_free(timer, &timer_debug_descr); |
| 715 | } |
| 716 | |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 717 | static inline void debug_timer_assert_init(struct timer_list *timer) |
| 718 | { |
| 719 | debug_object_assert_init(timer, &timer_debug_descr); |
| 720 | } |
| 721 | |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 722 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
| 723 | const char *name, struct lock_class_key *key); |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 724 | |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 725 | void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags, |
| 726 | const char *name, struct lock_class_key *key) |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 727 | { |
| 728 | debug_object_init_on_stack(timer, &timer_debug_descr); |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 729 | do_init_timer(timer, flags, name, key); |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 730 | } |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 731 | EXPORT_SYMBOL_GPL(init_timer_on_stack_key); |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 732 | |
| 733 | void destroy_timer_on_stack(struct timer_list *timer) |
| 734 | { |
| 735 | debug_object_free(timer, &timer_debug_descr); |
| 736 | } |
| 737 | EXPORT_SYMBOL_GPL(destroy_timer_on_stack); |
| 738 | |
| 739 | #else |
| 740 | static inline void debug_timer_init(struct timer_list *timer) { } |
| 741 | static inline void debug_timer_activate(struct timer_list *timer) { } |
| 742 | static inline void debug_timer_deactivate(struct timer_list *timer) { } |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 743 | static inline void debug_timer_assert_init(struct timer_list *timer) { } |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 744 | #endif |
| 745 | |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 746 | static inline void debug_init(struct timer_list *timer) |
| 747 | { |
| 748 | debug_timer_init(timer); |
| 749 | trace_timer_init(timer); |
| 750 | } |
| 751 | |
| 752 | static inline void |
| 753 | debug_activate(struct timer_list *timer, unsigned long expires) |
| 754 | { |
| 755 | debug_timer_activate(timer); |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 756 | trace_timer_start(timer, expires, timer->flags); |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 757 | } |
| 758 | |
| 759 | static inline void debug_deactivate(struct timer_list *timer) |
| 760 | { |
| 761 | debug_timer_deactivate(timer); |
| 762 | trace_timer_cancel(timer); |
| 763 | } |
| 764 | |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 765 | static inline void debug_assert_init(struct timer_list *timer) |
| 766 | { |
| 767 | debug_timer_assert_init(timer); |
| 768 | } |
| 769 | |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 770 | static void do_init_timer(struct timer_list *timer, unsigned int flags, |
| 771 | const char *name, struct lock_class_key *key) |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 772 | { |
Thomas Gleixner | 1dabbce | 2015-05-26 22:50:28 +0000 | [diff] [blame] | 773 | timer->entry.pprev = NULL; |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 774 | timer->flags = flags | raw_smp_processor_id(); |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 775 | lockdep_init_map(&timer->lockdep_map, name, key, 0); |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 776 | } |
| 777 | |
Rolf Eike Beer | 2aae4a1 | 2006-09-29 01:59:46 -0700 | [diff] [blame] | 778 | /** |
Randy Dunlap | 633fe79 | 2009-04-01 17:47:23 -0700 | [diff] [blame] | 779 | * init_timer_key - initialize a timer |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 780 | * @timer: the timer to be initialized |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 781 | * @flags: timer flags |
Randy Dunlap | 633fe79 | 2009-04-01 17:47:23 -0700 | [diff] [blame] | 782 | * @name: name of the timer |
| 783 | * @key: lockdep class key of the fake lock used for tracking timer |
| 784 | * sync lock dependencies |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 785 | * |
Randy Dunlap | 633fe79 | 2009-04-01 17:47:23 -0700 | [diff] [blame] | 786 | * init_timer_key() must be done to a timer prior calling *any* of the |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 787 | * other timer functions. |
| 788 | */ |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 789 | void init_timer_key(struct timer_list *timer, unsigned int flags, |
| 790 | const char *name, struct lock_class_key *key) |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 791 | { |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 792 | debug_init(timer); |
Tejun Heo | fc68399 | 2012-08-08 11:10:27 -0700 | [diff] [blame] | 793 | do_init_timer(timer, flags, name, key); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 794 | } |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 795 | EXPORT_SYMBOL(init_timer_key); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 796 | |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 797 | static inline void detach_timer(struct timer_list *timer, bool clear_pending) |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 798 | { |
Thomas Gleixner | 1dabbce | 2015-05-26 22:50:28 +0000 | [diff] [blame] | 799 | struct hlist_node *entry = &timer->entry; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 800 | |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 801 | debug_deactivate(timer); |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 802 | |
Thomas Gleixner | 1dabbce | 2015-05-26 22:50:28 +0000 | [diff] [blame] | 803 | __hlist_del(entry); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 804 | if (clear_pending) |
Thomas Gleixner | 1dabbce | 2015-05-26 22:50:28 +0000 | [diff] [blame] | 805 | entry->pprev = NULL; |
| 806 | entry->next = LIST_POISON2; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 807 | } |
| 808 | |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 809 | static int detach_if_pending(struct timer_list *timer, struct timer_base *base, |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 810 | bool clear_pending) |
| 811 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 812 | unsigned idx = timer_get_idx(timer); |
| 813 | |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 814 | if (!timer_pending(timer)) |
| 815 | return 0; |
| 816 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 817 | if (hlist_is_singular_node(&timer->entry, base->vectors + idx)) |
| 818 | __clear_bit(idx, base->pending_map); |
| 819 | |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 820 | detach_timer(timer, clear_pending); |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 821 | return 1; |
| 822 | } |
| 823 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 824 | static inline struct timer_base *get_timer_cpu_base(u32 tflags, u32 cpu) |
| 825 | { |
| 826 | struct timer_base *base = per_cpu_ptr(&timer_bases[BASE_STD], cpu); |
| 827 | |
| 828 | /* |
Anna-Maria Gleixner | d840687 | 2017-12-22 15:51:12 +0100 | [diff] [blame] | 829 | * If the timer is deferrable and NO_HZ_COMMON is set then we need |
| 830 | * to use the deferrable base. |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 831 | */ |
Blagovest Kolenichev | 42d4259 | 2018-01-18 06:58:43 -0800 | [diff] [blame] | 832 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) { |
Kyle Yan | c1f109c | 2017-02-24 16:49:27 -0800 | [diff] [blame] | 833 | base = &timer_base_deferrable; |
| 834 | if (tflags & TIMER_PINNED) |
| 835 | base = per_cpu_ptr(&timer_bases[BASE_DEF], cpu); |
| 836 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 837 | return base; |
| 838 | } |
| 839 | |
| 840 | static inline struct timer_base *get_timer_this_cpu_base(u32 tflags) |
| 841 | { |
| 842 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
| 843 | |
| 844 | /* |
Anna-Maria Gleixner | d840687 | 2017-12-22 15:51:12 +0100 | [diff] [blame] | 845 | * If the timer is deferrable and NO_HZ_COMMON is set then we need |
| 846 | * to use the deferrable base. |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 847 | */ |
Blagovest Kolenichev | 42d4259 | 2018-01-18 06:58:43 -0800 | [diff] [blame] | 848 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON) && (tflags & TIMER_DEFERRABLE)) { |
Kyle Yan | c1f109c | 2017-02-24 16:49:27 -0800 | [diff] [blame] | 849 | base = &timer_base_deferrable; |
| 850 | if (tflags & TIMER_PINNED) |
| 851 | base = this_cpu_ptr(&timer_bases[BASE_DEF]); |
Kyle Yan | df06519 | 2017-03-13 17:09:41 -0700 | [diff] [blame] | 852 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 853 | return base; |
| 854 | } |
| 855 | |
| 856 | static inline struct timer_base *get_timer_base(u32 tflags) |
| 857 | { |
| 858 | return get_timer_cpu_base(tflags, tflags & TIMER_CPUMASK); |
| 859 | } |
| 860 | |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 861 | #ifdef CONFIG_NO_HZ_COMMON |
| 862 | static inline struct timer_base * |
Thomas Gleixner | 6bad6bc | 2016-10-22 11:07:37 +0000 | [diff] [blame] | 863 | get_target_base(struct timer_base *base, unsigned tflags) |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 864 | { |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 865 | #ifdef CONFIG_SMP |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 866 | if ((tflags & TIMER_PINNED) || !base->migration_enabled) |
| 867 | return get_timer_this_cpu_base(tflags); |
| 868 | return get_timer_cpu_base(tflags, get_nohz_timer_target()); |
| 869 | #else |
| 870 | return get_timer_this_cpu_base(tflags); |
| 871 | #endif |
| 872 | } |
| 873 | |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 874 | static inline void forward_timer_base(struct timer_base *base) |
| 875 | { |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 876 | unsigned long jnow; |
Thomas Gleixner | 6bad6bc | 2016-10-22 11:07:37 +0000 | [diff] [blame] | 877 | |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 878 | /* |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 879 | * We only forward the base when we are idle or have just come out of |
| 880 | * idle (must_forward_clk logic), and have a delta between base clock |
| 881 | * and jiffies. In the common case, run_timers will take care of it. |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 882 | */ |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 883 | if (likely(!base->must_forward_clk)) |
| 884 | return; |
| 885 | |
| 886 | jnow = READ_ONCE(jiffies); |
| 887 | base->must_forward_clk = base->is_idle; |
| 888 | if ((long)(jnow - base->clk) < 2) |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 889 | return; |
| 890 | |
| 891 | /* |
| 892 | * If the next expiry value is > jiffies, then we fast forward to |
| 893 | * jiffies otherwise we forward to the next expiry value. |
| 894 | */ |
Thomas Gleixner | 6bad6bc | 2016-10-22 11:07:37 +0000 | [diff] [blame] | 895 | if (time_after(base->next_expiry, jnow)) |
| 896 | base->clk = jnow; |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 897 | else |
| 898 | base->clk = base->next_expiry; |
| 899 | } |
| 900 | #else |
| 901 | static inline struct timer_base * |
Thomas Gleixner | 6bad6bc | 2016-10-22 11:07:37 +0000 | [diff] [blame] | 902 | get_target_base(struct timer_base *base, unsigned tflags) |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 903 | { |
| 904 | return get_timer_this_cpu_base(tflags); |
| 905 | } |
| 906 | |
| 907 | static inline void forward_timer_base(struct timer_base *base) { } |
| 908 | #endif |
| 909 | |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 910 | |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 911 | /* |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 912 | * We are using hashed locking: Holding per_cpu(timer_bases[x]).lock means |
| 913 | * that all timers which are tied to this base are locked, and the base itself |
| 914 | * is locked too. |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 915 | * |
| 916 | * So __run_timers/migrate_timers can safely modify all timers which could |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 917 | * be found in the base->vectors array. |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 918 | * |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 919 | * When a timer is migrating then the TIMER_MIGRATING flag is set and we need |
| 920 | * to wait until the migration is done. |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 921 | */ |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 922 | static struct timer_base *lock_timer_base(struct timer_list *timer, |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 923 | unsigned long *flags) |
Josh Triplett | 89e7e374 | 2006-09-29 01:59:36 -0700 | [diff] [blame] | 924 | __acquires(timer->base->lock) |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 925 | { |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 926 | for (;;) { |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 927 | struct timer_base *base; |
Thomas Gleixner | b831275 | 2016-10-24 11:41:56 +0200 | [diff] [blame] | 928 | u32 tf; |
| 929 | |
| 930 | /* |
| 931 | * We need to use READ_ONCE() here, otherwise the compiler |
| 932 | * might re-read @tf between the check for TIMER_MIGRATING |
| 933 | * and spin_lock(). |
| 934 | */ |
| 935 | tf = READ_ONCE(timer->flags); |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 936 | |
| 937 | if (!(tf & TIMER_MIGRATING)) { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 938 | base = get_timer_base(tf); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 939 | spin_lock_irqsave(&base->lock, *flags); |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 940 | if (timer->flags == tf) |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 941 | return base; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 942 | spin_unlock_irqrestore(&base->lock, *flags); |
| 943 | } |
| 944 | cpu_relax(); |
| 945 | } |
| 946 | } |
| 947 | |
Ingo Molnar | 7401922 | 2009-02-18 12:23:29 +0100 | [diff] [blame] | 948 | static inline int |
Thomas Gleixner | 177ec0a | 2016-07-04 09:50:24 +0000 | [diff] [blame] | 949 | __mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | { |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 951 | struct timer_base *base, *new_base; |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 952 | unsigned int idx = UINT_MAX; |
| 953 | unsigned long clk = 0, flags; |
Thomas Gleixner | bc7a34b | 2015-05-26 22:50:33 +0000 | [diff] [blame] | 954 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 956 | BUG_ON(!timer->function); |
| 957 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 958 | /* |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 959 | * This is a common optimization triggered by the networking code - if |
| 960 | * the timer is re-modified to have the same timeout or ends up in the |
| 961 | * same array bucket then just return: |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 962 | */ |
| 963 | if (timer_pending(timer)) { |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 964 | /* |
| 965 | * The downside of this optimization is that it can result in |
| 966 | * larger granularity than you would get from adding a new |
| 967 | * timer with this expiry. |
| 968 | */ |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 969 | if (timer->expires == expires) |
| 970 | return 1; |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 971 | |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 972 | /* |
| 973 | * We lock timer base and calculate the bucket index right |
| 974 | * here. If the timer ends up in the same bucket, then we |
| 975 | * just update the expiry time and avoid the whole |
| 976 | * dequeue/enqueue dance. |
| 977 | */ |
| 978 | base = lock_timer_base(timer, &flags); |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 979 | forward_timer_base(base); |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 980 | |
| 981 | clk = base->clk; |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 982 | idx = calc_wheel_index(expires, clk); |
| 983 | |
| 984 | /* |
| 985 | * Retrieve and compare the array index of the pending |
| 986 | * timer. If it matches set the expiry to the new value so a |
| 987 | * subsequent call will exit in the expires check above. |
| 988 | */ |
| 989 | if (idx == timer_get_idx(timer)) { |
| 990 | timer->expires = expires; |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 991 | ret = 1; |
| 992 | goto out_unlock; |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 993 | } |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 994 | } else { |
| 995 | base = lock_timer_base(timer, &flags); |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 996 | forward_timer_base(base); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 997 | } |
| 998 | |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 999 | ret = detach_if_pending(timer, base, false); |
| 1000 | if (!ret && pending_only) |
| 1001 | goto out_unlock; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1002 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1003 | new_base = get_target_base(base, timer->flags); |
Arun R Bharadwaj | eea08f3 | 2009-04-16 12:16:41 +0530 | [diff] [blame] | 1004 | |
Oleg Nesterov | 3691c51 | 2006-03-31 02:30:30 -0800 | [diff] [blame] | 1005 | if (base != new_base) { |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1006 | /* |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1007 | * We are trying to schedule the timer on the new base. |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1008 | * However we can't change timer's base while it is running, |
| 1009 | * otherwise del_timer_sync() can't detect that the timer's |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1010 | * handler yet has not finished. This also guarantees that the |
| 1011 | * timer is serialized wrt itself. |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1012 | */ |
Oleg Nesterov | a2c348f | 2006-03-31 02:30:31 -0800 | [diff] [blame] | 1013 | if (likely(base->running_timer != timer)) { |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1014 | /* See the comment in lock_timer_base() */ |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1015 | timer->flags |= TIMER_MIGRATING; |
| 1016 | |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1017 | spin_unlock(&base->lock); |
Oleg Nesterov | a2c348f | 2006-03-31 02:30:31 -0800 | [diff] [blame] | 1018 | base = new_base; |
| 1019 | spin_lock(&base->lock); |
Eric Dumazet | d0023a1 | 2015-08-17 10:18:48 -0700 | [diff] [blame] | 1020 | WRITE_ONCE(timer->flags, |
| 1021 | (timer->flags & ~TIMER_BASEMASK) | base->cpu); |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 1022 | forward_timer_base(base); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1023 | } |
| 1024 | } |
| 1025 | |
Thomas Gleixner | 574e543 | 2017-12-22 15:51:14 +0100 | [diff] [blame] | 1026 | debug_activate(timer, expires); |
| 1027 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1028 | timer->expires = expires; |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 1029 | /* |
| 1030 | * If 'idx' was calculated above and the base time did not advance |
Thomas Gleixner | 4da9152 | 2016-10-24 11:55:10 +0200 | [diff] [blame] | 1031 | * between calculating 'idx' and possibly switching the base, only |
| 1032 | * enqueue_timer() and trigger_dyntick_cpu() is required. Otherwise |
| 1033 | * we need to (re)calculate the wheel index via |
| 1034 | * internal_add_timer(). |
Anna-Maria Gleixner | f00c0af | 2016-07-04 09:50:40 +0000 | [diff] [blame] | 1035 | */ |
| 1036 | if (idx != UINT_MAX && clk == base->clk) { |
| 1037 | enqueue_timer(base, timer, idx); |
| 1038 | trigger_dyntick_cpu(base, timer); |
| 1039 | } else { |
| 1040 | internal_add_timer(base, timer); |
| 1041 | } |
Ingo Molnar | 7401922 | 2009-02-18 12:23:29 +0100 | [diff] [blame] | 1042 | |
| 1043 | out_unlock: |
Oleg Nesterov | a2c348f | 2006-03-31 02:30:31 -0800 | [diff] [blame] | 1044 | spin_unlock_irqrestore(&base->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | |
| 1046 | return ret; |
| 1047 | } |
| 1048 | |
Ingo Molnar | 7401922 | 2009-02-18 12:23:29 +0100 | [diff] [blame] | 1049 | /** |
| 1050 | * mod_timer_pending - modify a pending timer's timeout |
| 1051 | * @timer: the pending timer to be modified |
| 1052 | * @expires: new timeout in jiffies |
| 1053 | * |
| 1054 | * mod_timer_pending() is the same for pending timers as mod_timer(), |
| 1055 | * but will not re-activate and modify already deleted timers. |
| 1056 | * |
| 1057 | * It is useful for unserialized use of timers. |
| 1058 | */ |
| 1059 | int mod_timer_pending(struct timer_list *timer, unsigned long expires) |
| 1060 | { |
Thomas Gleixner | 177ec0a | 2016-07-04 09:50:24 +0000 | [diff] [blame] | 1061 | return __mod_timer(timer, expires, true); |
Ingo Molnar | 7401922 | 2009-02-18 12:23:29 +0100 | [diff] [blame] | 1062 | } |
| 1063 | EXPORT_SYMBOL(mod_timer_pending); |
| 1064 | |
| 1065 | /** |
| 1066 | * mod_timer - modify a timer's timeout |
| 1067 | * @timer: the timer to be modified |
| 1068 | * @expires: new timeout in jiffies |
| 1069 | * |
| 1070 | * mod_timer() is a more efficient way to update the expire field of an |
| 1071 | * active timer (if the timer is inactive it will be activated) |
| 1072 | * |
| 1073 | * mod_timer(timer, expires) is equivalent to: |
| 1074 | * |
| 1075 | * del_timer(timer); timer->expires = expires; add_timer(timer); |
| 1076 | * |
| 1077 | * Note that if there are multiple unserialized concurrent users of the |
| 1078 | * same timer, then mod_timer() is the only safe way to modify the timeout, |
| 1079 | * since add_timer() cannot modify an already running timer. |
| 1080 | * |
| 1081 | * The function returns whether it has modified a pending timer or not. |
| 1082 | * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an |
| 1083 | * active timer returns 1.) |
| 1084 | */ |
| 1085 | int mod_timer(struct timer_list *timer, unsigned long expires) |
| 1086 | { |
Thomas Gleixner | 177ec0a | 2016-07-04 09:50:24 +0000 | [diff] [blame] | 1087 | return __mod_timer(timer, expires, false); |
Ingo Molnar | 7401922 | 2009-02-18 12:23:29 +0100 | [diff] [blame] | 1088 | } |
| 1089 | EXPORT_SYMBOL(mod_timer); |
| 1090 | |
| 1091 | /** |
| 1092 | * add_timer - start a timer |
| 1093 | * @timer: the timer to be added |
| 1094 | * |
| 1095 | * The kernel will do a ->function(->data) callback from the |
| 1096 | * timer interrupt at the ->expires point in the future. The |
| 1097 | * current time is 'jiffies'. |
| 1098 | * |
| 1099 | * The timer's ->expires, ->function (and if the handler uses it, ->data) |
| 1100 | * fields must be set prior calling this function. |
| 1101 | * |
| 1102 | * Timers with an ->expires field in the past will be executed in the next |
| 1103 | * timer tick. |
| 1104 | */ |
| 1105 | void add_timer(struct timer_list *timer) |
| 1106 | { |
| 1107 | BUG_ON(timer_pending(timer)); |
| 1108 | mod_timer(timer, timer->expires); |
| 1109 | } |
| 1110 | EXPORT_SYMBOL(add_timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | |
Rolf Eike Beer | 2aae4a1 | 2006-09-29 01:59:46 -0700 | [diff] [blame] | 1112 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | * add_timer_on - start a timer on a particular CPU |
| 1114 | * @timer: the timer to be added |
| 1115 | * @cpu: the CPU to start it on |
| 1116 | * |
| 1117 | * This is not very scalable on SMP. Double adds are not possible. |
| 1118 | */ |
| 1119 | void add_timer_on(struct timer_list *timer, int cpu) |
| 1120 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1121 | struct timer_base *new_base, *base; |
Thomas Gleixner | 6819457 | 2007-07-19 01:49:16 -0700 | [diff] [blame] | 1122 | unsigned long flags; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1123 | |
Thomas Gleixner | 6819457 | 2007-07-19 01:49:16 -0700 | [diff] [blame] | 1124 | BUG_ON(timer_pending(timer) || !timer->function); |
Tejun Heo | 22b886d | 2015-11-04 12:15:33 -0500 | [diff] [blame] | 1125 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1126 | new_base = get_timer_cpu_base(timer->flags, cpu); |
| 1127 | |
Tejun Heo | 22b886d | 2015-11-04 12:15:33 -0500 | [diff] [blame] | 1128 | /* |
| 1129 | * If @timer was on a different CPU, it should be migrated with the |
| 1130 | * old base locked to prevent other operations proceeding with the |
| 1131 | * wrong base locked. See lock_timer_base(). |
| 1132 | */ |
| 1133 | base = lock_timer_base(timer, &flags); |
| 1134 | if (base != new_base) { |
| 1135 | timer->flags |= TIMER_MIGRATING; |
| 1136 | |
| 1137 | spin_unlock(&base->lock); |
| 1138 | base = new_base; |
| 1139 | spin_lock(&base->lock); |
| 1140 | WRITE_ONCE(timer->flags, |
| 1141 | (timer->flags & ~TIMER_BASEMASK) | cpu); |
| 1142 | } |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 1143 | forward_timer_base(base); |
Tejun Heo | 22b886d | 2015-11-04 12:15:33 -0500 | [diff] [blame] | 1144 | |
Xiao Guangrong | 2b022e3 | 2009-08-10 10:48:59 +0800 | [diff] [blame] | 1145 | debug_activate(timer, timer->expires); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | internal_add_timer(base, timer); |
Oleg Nesterov | 3691c51 | 2006-03-31 02:30:30 -0800 | [diff] [blame] | 1147 | spin_unlock_irqrestore(&base->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | } |
Andi Kleen | a9862e0 | 2009-05-19 22:49:07 +0200 | [diff] [blame] | 1149 | EXPORT_SYMBOL_GPL(add_timer_on); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | |
Rolf Eike Beer | 2aae4a1 | 2006-09-29 01:59:46 -0700 | [diff] [blame] | 1151 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | * del_timer - deactive a timer. |
| 1153 | * @timer: the timer to be deactivated |
| 1154 | * |
| 1155 | * del_timer() deactivates a timer - this works on both active and inactive |
| 1156 | * timers. |
| 1157 | * |
| 1158 | * The function returns whether it has deactivated a pending timer or not. |
| 1159 | * (ie. del_timer() of an inactive timer returns 0, del_timer() of an |
| 1160 | * active timer returns 1.) |
| 1161 | */ |
| 1162 | int del_timer(struct timer_list *timer) |
| 1163 | { |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 1164 | struct timer_base *base; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | unsigned long flags; |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1166 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1167 | |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 1168 | debug_assert_init(timer); |
| 1169 | |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1170 | if (timer_pending(timer)) { |
| 1171 | base = lock_timer_base(timer, &flags); |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 1172 | ret = detach_if_pending(timer, base, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1173 | spin_unlock_irqrestore(&base->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1174 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1175 | |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1176 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | EXPORT_SYMBOL(del_timer); |
| 1179 | |
Rolf Eike Beer | 2aae4a1 | 2006-09-29 01:59:46 -0700 | [diff] [blame] | 1180 | /** |
| 1181 | * try_to_del_timer_sync - Try to deactivate a timer |
| 1182 | * @timer: timer do del |
| 1183 | * |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1184 | * This function tries to deactivate a timer. Upon successful (ret >= 0) |
| 1185 | * exit the timer is not queued and the handler is not running on any CPU. |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1186 | */ |
| 1187 | int try_to_del_timer_sync(struct timer_list *timer) |
| 1188 | { |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 1189 | struct timer_base *base; |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1190 | unsigned long flags; |
| 1191 | int ret = -1; |
| 1192 | |
Christine Chan | dc4218b | 2011-11-07 19:48:28 -0800 | [diff] [blame] | 1193 | debug_assert_init(timer); |
| 1194 | |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1195 | base = lock_timer_base(timer, &flags); |
| 1196 | |
Kees Cook | a967be8 | 2017-02-08 11:26:59 -0800 | [diff] [blame] | 1197 | if (base->running_timer != timer) |
Thomas Gleixner | ec44bc7 | 2012-05-25 22:08:57 +0000 | [diff] [blame] | 1198 | ret = detach_if_pending(timer, base, true); |
Kees Cook | a967be8 | 2017-02-08 11:26:59 -0800 | [diff] [blame] | 1199 | |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1200 | spin_unlock_irqrestore(&base->lock, flags); |
| 1201 | |
| 1202 | return ret; |
| 1203 | } |
David Howells | e19dff1 | 2007-04-26 15:46:56 -0700 | [diff] [blame] | 1204 | EXPORT_SYMBOL(try_to_del_timer_sync); |
| 1205 | |
Yong Zhang | 6f1bc45 | 2010-10-20 15:57:31 -0700 | [diff] [blame] | 1206 | #ifdef CONFIG_SMP |
Rolf Eike Beer | 2aae4a1 | 2006-09-29 01:59:46 -0700 | [diff] [blame] | 1207 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | * del_timer_sync - deactivate a timer and wait for the handler to finish. |
| 1209 | * @timer: the timer to be deactivated |
| 1210 | * |
| 1211 | * This function only differs from del_timer() on SMP: besides deactivating |
| 1212 | * the timer it also makes sure the handler has finished executing on other |
| 1213 | * CPUs. |
| 1214 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 1215 | * Synchronization rules: Callers must prevent restarting of the timer, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | * otherwise this function is meaningless. It must not be called from |
Tejun Heo | c5f66e9 | 2012-08-08 11:10:28 -0700 | [diff] [blame] | 1217 | * interrupt contexts unless the timer is an irqsafe one. The caller must |
| 1218 | * not hold locks which would prevent completion of the timer's |
| 1219 | * handler. The timer's handler must not call add_timer_on(). Upon exit the |
| 1220 | * timer is not queued and the handler is not running on any CPU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1221 | * |
Tejun Heo | c5f66e9 | 2012-08-08 11:10:28 -0700 | [diff] [blame] | 1222 | * Note: For !irqsafe timers, you must not hold locks that are held in |
| 1223 | * interrupt context while calling this function. Even if the lock has |
| 1224 | * nothing to do with the timer in question. Here's why: |
Steven Rostedt | 48228f7 | 2011-02-08 12:39:54 -0500 | [diff] [blame] | 1225 | * |
| 1226 | * CPU0 CPU1 |
| 1227 | * ---- ---- |
| 1228 | * <SOFTIRQ> |
| 1229 | * call_timer_fn(); |
| 1230 | * base->running_timer = mytimer; |
| 1231 | * spin_lock_irq(somelock); |
| 1232 | * <IRQ> |
| 1233 | * spin_lock(somelock); |
| 1234 | * del_timer_sync(mytimer); |
| 1235 | * while (base->running_timer == mytimer); |
| 1236 | * |
| 1237 | * Now del_timer_sync() will never return and never release somelock. |
| 1238 | * The interrupt on the other CPU is waiting to grab somelock but |
| 1239 | * it has interrupted the softirq that CPU0 is waiting to finish. |
| 1240 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1241 | * The function returns whether it has deactivated a pending timer or not. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | */ |
| 1243 | int del_timer_sync(struct timer_list *timer) |
| 1244 | { |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 1245 | #ifdef CONFIG_LOCKDEP |
Peter Zijlstra | f266a51 | 2011-02-03 15:09:41 +0100 | [diff] [blame] | 1246 | unsigned long flags; |
| 1247 | |
Steven Rostedt | 48228f7 | 2011-02-08 12:39:54 -0500 | [diff] [blame] | 1248 | /* |
| 1249 | * If lockdep gives a backtrace here, please reference |
| 1250 | * the synchronization rules above. |
| 1251 | */ |
Peter Zijlstra | 7ff2079 | 2011-02-08 15:18:00 +0100 | [diff] [blame] | 1252 | local_irq_save(flags); |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 1253 | lock_map_acquire(&timer->lockdep_map); |
| 1254 | lock_map_release(&timer->lockdep_map); |
Peter Zijlstra | 7ff2079 | 2011-02-08 15:18:00 +0100 | [diff] [blame] | 1255 | local_irq_restore(flags); |
Johannes Berg | 6f2b9b9 | 2009-01-29 16:03:20 +0100 | [diff] [blame] | 1256 | #endif |
Yong Zhang | 466bd30 | 2010-10-20 15:57:33 -0700 | [diff] [blame] | 1257 | /* |
| 1258 | * don't use it in hardirq context, because it |
| 1259 | * could lead to deadlock. |
| 1260 | */ |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1261 | WARN_ON(in_irq() && !(timer->flags & TIMER_IRQSAFE)); |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1262 | for (;;) { |
| 1263 | int ret = try_to_del_timer_sync(timer); |
| 1264 | if (ret >= 0) |
| 1265 | return ret; |
Andrew Morton | a000965 | 2006-07-14 00:24:06 -0700 | [diff] [blame] | 1266 | cpu_relax(); |
Oleg Nesterov | fd450b7 | 2005-06-23 00:08:59 -0700 | [diff] [blame] | 1267 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | } |
| 1269 | EXPORT_SYMBOL(del_timer_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1270 | #endif |
| 1271 | |
Thomas Gleixner | 576da12 | 2010-03-12 21:10:29 +0100 | [diff] [blame] | 1272 | static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long), |
| 1273 | unsigned long data) |
| 1274 | { |
Peter Zijlstra | 4a2b4b2 | 2013-08-14 14:55:24 +0200 | [diff] [blame] | 1275 | int count = preempt_count(); |
Thomas Gleixner | 576da12 | 2010-03-12 21:10:29 +0100 | [diff] [blame] | 1276 | |
| 1277 | #ifdef CONFIG_LOCKDEP |
| 1278 | /* |
| 1279 | * It is permissible to free the timer from inside the |
| 1280 | * function that is called from it, this we need to take into |
| 1281 | * account for lockdep too. To avoid bogus "held lock freed" |
| 1282 | * warnings as well as problems when looking into |
| 1283 | * timer->lockdep_map, make a copy and use that here. |
| 1284 | */ |
Peter Zijlstra | 4d82a1d | 2012-05-15 08:06:19 -0700 | [diff] [blame] | 1285 | struct lockdep_map lockdep_map; |
| 1286 | |
| 1287 | lockdep_copy_map(&lockdep_map, &timer->lockdep_map); |
Thomas Gleixner | 576da12 | 2010-03-12 21:10:29 +0100 | [diff] [blame] | 1288 | #endif |
| 1289 | /* |
| 1290 | * Couple the lock chain with the lock chain at |
| 1291 | * del_timer_sync() by acquiring the lock_map around the fn() |
| 1292 | * call here and in del_timer_sync(). |
| 1293 | */ |
| 1294 | lock_map_acquire(&lockdep_map); |
| 1295 | |
| 1296 | trace_timer_expire_entry(timer); |
| 1297 | fn(data); |
| 1298 | trace_timer_expire_exit(timer); |
| 1299 | |
| 1300 | lock_map_release(&lockdep_map); |
| 1301 | |
Peter Zijlstra | 4a2b4b2 | 2013-08-14 14:55:24 +0200 | [diff] [blame] | 1302 | if (count != preempt_count()) { |
Thomas Gleixner | 802702e | 2010-03-12 20:13:23 +0100 | [diff] [blame] | 1303 | WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n", |
Peter Zijlstra | 4a2b4b2 | 2013-08-14 14:55:24 +0200 | [diff] [blame] | 1304 | fn, count, preempt_count()); |
Thomas Gleixner | 802702e | 2010-03-12 20:13:23 +0100 | [diff] [blame] | 1305 | /* |
| 1306 | * Restore the preempt count. That gives us a decent |
| 1307 | * chance to survive and extract information. If the |
| 1308 | * callback kept a lock held, bad luck, but not worse |
| 1309 | * than the BUG() we had. |
| 1310 | */ |
Peter Zijlstra | 4a2b4b2 | 2013-08-14 14:55:24 +0200 | [diff] [blame] | 1311 | preempt_count_set(count); |
Thomas Gleixner | 576da12 | 2010-03-12 21:10:29 +0100 | [diff] [blame] | 1312 | } |
| 1313 | } |
| 1314 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1315 | static void expire_timers(struct timer_base *base, struct hlist_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1317 | while (!hlist_empty(head)) { |
| 1318 | struct timer_list *timer; |
| 1319 | void (*fn)(unsigned long); |
| 1320 | unsigned long data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1322 | timer = hlist_entry(head->first, struct timer_list, entry); |
Thomas Gleixner | 3bb475a | 2015-05-26 22:50:24 +0000 | [diff] [blame] | 1323 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1324 | base->running_timer = timer; |
| 1325 | detach_timer(timer, true); |
Thomas Gleixner | 3bb475a | 2015-05-26 22:50:24 +0000 | [diff] [blame] | 1326 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1327 | fn = timer->function; |
| 1328 | data = timer->data; |
Thomas Gleixner | 3bb475a | 2015-05-26 22:50:24 +0000 | [diff] [blame] | 1329 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1330 | if (timer->flags & TIMER_IRQSAFE) { |
| 1331 | spin_unlock(&base->lock); |
| 1332 | call_timer_fn(timer, fn, data); |
| 1333 | spin_lock(&base->lock); |
| 1334 | } else { |
| 1335 | spin_unlock_irq(&base->lock); |
| 1336 | call_timer_fn(timer, fn, data); |
| 1337 | spin_lock_irq(&base->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | } |
| 1339 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1340 | } |
| 1341 | |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1342 | static int __collect_expired_timers(struct timer_base *base, |
| 1343 | struct hlist_head *heads) |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1344 | { |
| 1345 | unsigned long clk = base->clk; |
| 1346 | struct hlist_head *vec; |
| 1347 | int i, levels = 0; |
| 1348 | unsigned int idx; |
| 1349 | |
| 1350 | for (i = 0; i < LVL_DEPTH; i++) { |
| 1351 | idx = (clk & LVL_MASK) + i * LVL_SIZE; |
| 1352 | |
| 1353 | if (__test_and_clear_bit(idx, base->pending_map)) { |
| 1354 | vec = base->vectors + idx; |
| 1355 | hlist_move_list(vec, heads++); |
| 1356 | levels++; |
| 1357 | } |
| 1358 | /* Is it time to look at the next level? */ |
| 1359 | if (clk & LVL_CLK_MASK) |
| 1360 | break; |
| 1361 | /* Shift clock for the next level granularity */ |
| 1362 | clk >>= LVL_CLK_SHIFT; |
| 1363 | } |
| 1364 | return levels; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | } |
| 1366 | |
Frederic Weisbecker | 3451d02 | 2011-08-10 23:21:01 +0200 | [diff] [blame] | 1367 | #ifdef CONFIG_NO_HZ_COMMON |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | /* |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1369 | * Find the next pending bucket of a level. Search from level start (@offset) |
| 1370 | * + @clk upwards and if nothing there, search from start of the level |
| 1371 | * (@offset) up to @offset + clk. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | */ |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1373 | static int next_pending_bucket(struct timer_base *base, unsigned offset, |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1374 | unsigned int clk, int lvl) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | { |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1376 | unsigned int pos_up = -1, pos_down, start = offset + clk; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1377 | unsigned end = offset + LVL_SIZE; |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1378 | unsigned int pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1379 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1380 | pos = find_next_bit(base->pending_map, end, start); |
| 1381 | if (pos < end) |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1382 | pos_up = pos - start; |
Venki Pallipadi | 6e453a6 | 2007-05-08 00:27:44 -0700 | [diff] [blame] | 1383 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1384 | pos = find_next_bit(base->pending_map, start, offset); |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1385 | pos_down = pos < start ? pos + LVL_SIZE - start : -1; |
Maria Yu | 219fe50 | 2017-10-25 14:26:07 +0800 | [diff] [blame] | 1386 | if (((pos_up + (u64)base->clk) << LVL_SHIFT(lvl)) > |
| 1387 | ((pos_down + (u64)base->clk) << LVL_SHIFT(lvl))) |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1388 | return pos_down; |
| 1389 | return pos_up; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1390 | } |
| 1391 | |
| 1392 | /* |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1393 | * Search the first expiring timer in the various clock levels. Caller must |
| 1394 | * hold base->lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | */ |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 1396 | static unsigned long __next_timer_interrupt(struct timer_base *base) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1398 | unsigned long clk, next, adj; |
| 1399 | unsigned lvl, offset = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1400 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1401 | next = base->clk + NEXT_TIMER_MAX_DELTA; |
| 1402 | clk = base->clk; |
| 1403 | for (lvl = 0; lvl < LVL_DEPTH; lvl++, offset += LVL_SIZE) { |
Channagoud Kadabi | ce49c27 | 2017-08-18 13:22:34 -0700 | [diff] [blame] | 1404 | int pos = next_pending_bucket(base, offset, clk & LVL_MASK, |
| 1405 | lvl); |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1406 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1407 | if (pos >= 0) { |
| 1408 | unsigned long tmp = clk + (unsigned long) pos; |
| 1409 | |
| 1410 | tmp <<= LVL_SHIFT(lvl); |
| 1411 | if (time_before(tmp, next)) |
| 1412 | next = tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1414 | /* |
| 1415 | * Clock for the next level. If the current level clock lower |
| 1416 | * bits are zero, we look at the next level as is. If not we |
| 1417 | * need to advance it by one because that's going to be the |
| 1418 | * next expiring bucket in that level. base->clk is the next |
| 1419 | * expiring jiffie. So in case of: |
| 1420 | * |
| 1421 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 |
| 1422 | * 0 0 0 0 0 0 |
| 1423 | * |
| 1424 | * we have to look at all levels @index 0. With |
| 1425 | * |
| 1426 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 |
| 1427 | * 0 0 0 0 0 2 |
| 1428 | * |
| 1429 | * LVL0 has the next expiring bucket @index 2. The upper |
| 1430 | * levels have the next expiring bucket @index 1. |
| 1431 | * |
| 1432 | * In case that the propagation wraps the next level the same |
| 1433 | * rules apply: |
| 1434 | * |
| 1435 | * LVL5 LVL4 LVL3 LVL2 LVL1 LVL0 |
| 1436 | * 0 0 0 0 F 2 |
| 1437 | * |
| 1438 | * So after looking at LVL0 we get: |
| 1439 | * |
| 1440 | * LVL5 LVL4 LVL3 LVL2 LVL1 |
| 1441 | * 0 0 0 1 0 |
| 1442 | * |
| 1443 | * So no propagation from LVL1 to LVL2 because that happened |
| 1444 | * with the add already, but then we need to propagate further |
| 1445 | * from LVL2 to LVL3. |
| 1446 | * |
| 1447 | * So the simple check whether the lower bits of the current |
| 1448 | * level are 0 or not is sufficient for all cases. |
| 1449 | */ |
| 1450 | adj = clk & LVL_CLK_MASK ? 1 : 0; |
| 1451 | clk >>= LVL_CLK_SHIFT; |
| 1452 | clk += adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | } |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1454 | return next; |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1455 | } |
| 1456 | |
| 1457 | /* |
| 1458 | * Check, if the next hrtimer event is before the next timer wheel |
| 1459 | * event: |
| 1460 | */ |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1461 | static u64 cmp_next_hrtimer_event(u64 basem, u64 expires) |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1462 | { |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1463 | u64 nextevt = hrtimer_get_next_event(); |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1464 | |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1465 | /* |
| 1466 | * If high resolution timers are enabled |
| 1467 | * hrtimer_get_next_event() returns KTIME_MAX. |
| 1468 | */ |
| 1469 | if (expires <= nextevt) |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1470 | return expires; |
| 1471 | |
Thomas Gleixner | 9501b6c | 2007-03-25 14:31:17 +0200 | [diff] [blame] | 1472 | /* |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1473 | * If the next timer is already expired, return the tick base |
| 1474 | * time so the tick is fired immediately. |
Thomas Gleixner | 9501b6c | 2007-03-25 14:31:17 +0200 | [diff] [blame] | 1475 | */ |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1476 | if (nextevt <= basem) |
| 1477 | return basem; |
Thomas Gleixner | eaad084 | 2007-05-29 23:47:39 +0200 | [diff] [blame] | 1478 | |
| 1479 | /* |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1480 | * Round up to the next jiffie. High resolution timers are |
| 1481 | * off, so the hrtimers are expired in the tick and we need to |
| 1482 | * make sure that this tick really expires the timer to avoid |
| 1483 | * a ping pong of the nohz stop code. |
| 1484 | * |
| 1485 | * Use DIV_ROUND_UP_ULL to prevent gcc calling __divdi3 |
Thomas Gleixner | eaad084 | 2007-05-29 23:47:39 +0200 | [diff] [blame] | 1486 | */ |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1487 | return DIV_ROUND_UP_ULL(nextevt, TICK_NSEC) * TICK_NSEC; |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1488 | } |
| 1489 | |
Prasad Sodagudi | 602c4e2 | 2017-05-17 23:26:09 -0700 | [diff] [blame] | 1490 | |
| 1491 | #ifdef CONFIG_SMP |
| 1492 | /* |
| 1493 | * check_pending_deferrable_timers - Check for unbound deferrable timer expiry |
| 1494 | * @cpu - Current CPU |
| 1495 | * |
| 1496 | * The function checks whether any global deferrable pending timers |
| 1497 | * are exipired or not. This function does not check cpu bounded |
| 1498 | * diferrable pending timers expiry. |
| 1499 | * |
| 1500 | * The function returns true when a cpu unbounded deferrable timer is expired. |
| 1501 | */ |
| 1502 | bool check_pending_deferrable_timers(int cpu) |
| 1503 | { |
| 1504 | if (cpu == tick_do_timer_cpu || |
| 1505 | tick_do_timer_cpu == TICK_DO_TIMER_NONE) { |
| 1506 | if (time_after_eq(jiffies, timer_base_deferrable.clk) |
| 1507 | && !atomic_cmpxchg(&deferrable_pending, 0, 1)) { |
| 1508 | return true; |
| 1509 | } |
| 1510 | } |
| 1511 | return false; |
| 1512 | } |
| 1513 | #endif |
| 1514 | |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1515 | /** |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1516 | * get_next_timer_interrupt - return the time (clock mono) of the next timer |
| 1517 | * @basej: base time jiffies |
| 1518 | * @basem: base time clock monotonic |
| 1519 | * |
| 1520 | * Returns the tick aligned clock monotonic time of the next pending |
| 1521 | * timer or KTIME_MAX if no timer is pending. |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1522 | */ |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1523 | u64 get_next_timer_interrupt(unsigned long basej, u64 basem) |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1524 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1525 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1526 | u64 expires = KTIME_MAX; |
| 1527 | unsigned long nextevt; |
Chris Metcalf | 46c8f0b | 2016-08-08 16:29:07 -0400 | [diff] [blame] | 1528 | bool is_max_delta; |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1529 | |
Heiko Carstens | dbd87b5 | 2010-12-01 10:11:09 +0100 | [diff] [blame] | 1530 | /* |
| 1531 | * Pretend that there is no timer pending if the cpu is offline. |
| 1532 | * Possible pending timers will be migrated later to an active cpu. |
| 1533 | */ |
| 1534 | if (cpu_is_offline(smp_processor_id())) |
Thomas Gleixner | e40468a | 2012-05-25 22:08:59 +0000 | [diff] [blame] | 1535 | return expires; |
| 1536 | |
Thomas Gleixner | 1cfd684 | 2007-02-16 01:27:46 -0800 | [diff] [blame] | 1537 | spin_lock(&base->lock); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1538 | nextevt = __next_timer_interrupt(base); |
Chris Metcalf | 46c8f0b | 2016-08-08 16:29:07 -0400 | [diff] [blame] | 1539 | is_max_delta = (nextevt == base->clk + NEXT_TIMER_MAX_DELTA); |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1540 | base->next_expiry = nextevt; |
| 1541 | /* |
Thomas Gleixner | 041ad7b | 2016-10-22 11:07:35 +0000 | [diff] [blame] | 1542 | * We have a fresh next event. Check whether we can forward the |
| 1543 | * base. We can only do that when @basej is past base->clk |
| 1544 | * otherwise we might rewind base->clk. |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1545 | */ |
Thomas Gleixner | 041ad7b | 2016-10-22 11:07:35 +0000 | [diff] [blame] | 1546 | if (time_after(basej, base->clk)) { |
| 1547 | if (time_after(nextevt, basej)) |
| 1548 | base->clk = basej; |
| 1549 | else if (time_after(nextevt, base->clk)) |
| 1550 | base->clk = nextevt; |
| 1551 | } |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1552 | |
| 1553 | if (time_before_eq(nextevt, basej)) { |
| 1554 | expires = basem; |
| 1555 | base->is_idle = false; |
| 1556 | } else { |
Chris Metcalf | 46c8f0b | 2016-08-08 16:29:07 -0400 | [diff] [blame] | 1557 | if (!is_max_delta) |
Matija Glavinic Pecotic | 9ef8b23 | 2017-08-01 09:11:52 +0200 | [diff] [blame] | 1558 | expires = basem + (u64)(nextevt - basej) * TICK_NSEC; |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1559 | /* |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 1560 | * If we expect to sleep more than a tick, mark the base idle. |
| 1561 | * Also the tick is stopped so any added timer must forward |
| 1562 | * the base clk itself to keep granularity small. This idle |
| 1563 | * logic is only maintained for the BASE_STD base, deferrable |
| 1564 | * timers may still see large granularity skew (by design). |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1565 | */ |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 1566 | if ((expires - basem) > TICK_NSEC) { |
| 1567 | base->must_forward_clk = true; |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1568 | base->is_idle = true; |
Nicholas Piggin | 70b3fd5 | 2017-08-22 18:43:48 +1000 | [diff] [blame] | 1569 | } |
Thomas Gleixner | e40468a | 2012-05-25 22:08:59 +0000 | [diff] [blame] | 1570 | } |
Oleg Nesterov | 3691c51 | 2006-03-31 02:30:30 -0800 | [diff] [blame] | 1571 | spin_unlock(&base->lock); |
Tony Lindgren | 6923974 | 2006-03-06 15:42:45 -0800 | [diff] [blame] | 1572 | |
Thomas Gleixner | c1ad348 | 2015-04-14 21:08:58 +0000 | [diff] [blame] | 1573 | return cmp_next_hrtimer_event(basem, expires); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | } |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1575 | |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1576 | /** |
| 1577 | * timer_clear_idle - Clear the idle state of the timer base |
| 1578 | * |
| 1579 | * Called with interrupts disabled |
| 1580 | */ |
| 1581 | void timer_clear_idle(void) |
| 1582 | { |
| 1583 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
| 1584 | |
| 1585 | /* |
| 1586 | * We do this unlocked. The worst outcome is a remote enqueue sending |
| 1587 | * a pointless IPI, but taking the lock would just make the window for |
| 1588 | * sending the IPI a few instructions smaller for the cost of taking |
| 1589 | * the lock in the exit from idle path. |
| 1590 | */ |
| 1591 | base->is_idle = false; |
| 1592 | } |
| 1593 | |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1594 | static int collect_expired_timers(struct timer_base *base, |
| 1595 | struct hlist_head *heads) |
| 1596 | { |
| 1597 | /* |
| 1598 | * NOHZ optimization. After a long idle sleep we need to forward the |
| 1599 | * base to current jiffies. Avoid a loop by searching the bitfield for |
| 1600 | * the next expiring timer. |
| 1601 | */ |
| 1602 | if ((long)(jiffies - base->clk) > 2) { |
| 1603 | unsigned long next = __next_timer_interrupt(base); |
| 1604 | |
| 1605 | /* |
| 1606 | * If the next timer is ahead of time forward to current |
Thomas Gleixner | a683f39 | 2016-07-04 09:50:36 +0000 | [diff] [blame] | 1607 | * jiffies, otherwise forward to the next expiry time: |
Anna-Maria Gleixner | 2369683 | 2016-07-04 09:50:34 +0000 | [diff] [blame] | 1608 | */ |
| 1609 | if (time_after(next, jiffies)) { |
| 1610 | /* The call site will increment clock! */ |
| 1611 | base->clk = jiffies - 1; |
| 1612 | return 0; |
| 1613 | } |
| 1614 | base->clk = next; |
| 1615 | } |
| 1616 | return __collect_expired_timers(base, heads); |
| 1617 | } |
| 1618 | #else |
| 1619 | static inline int collect_expired_timers(struct timer_base *base, |
| 1620 | struct hlist_head *heads) |
| 1621 | { |
| 1622 | return __collect_expired_timers(base, heads); |
| 1623 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | #endif |
| 1625 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | /* |
Daniel Walker | 5b4db0c | 2007-10-18 03:06:11 -0700 | [diff] [blame] | 1627 | * Called from the timer interrupt handler to charge one tick to the current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | * process. user_tick is 1 if the tick is user time, 0 for system. |
| 1629 | */ |
| 1630 | void update_process_times(int user_tick) |
| 1631 | { |
| 1632 | struct task_struct *p = current; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | |
| 1634 | /* Note: this timer irq context must be accounted for as well. */ |
Paul Mackerras | fa13a5a | 2007-11-09 22:39:38 +0100 | [diff] [blame] | 1635 | account_process_tick(p, user_tick); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | run_local_timers(); |
Paul E. McKenney | c3377c2d | 2014-10-21 07:53:02 -0700 | [diff] [blame] | 1637 | rcu_check_callbacks(user_tick); |
Peter Zijlstra | e360adb | 2010-10-14 14:01:34 +0800 | [diff] [blame] | 1638 | #ifdef CONFIG_IRQ_WORK |
| 1639 | if (in_irq()) |
Frederic Weisbecker | 76a3306 | 2014-08-16 18:37:19 +0200 | [diff] [blame] | 1640 | irq_work_tick(); |
Peter Zijlstra | e360adb | 2010-10-14 14:01:34 +0800 | [diff] [blame] | 1641 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | scheduler_tick(); |
Thomas Gleixner | 6819457 | 2007-07-19 01:49:16 -0700 | [diff] [blame] | 1643 | run_posix_cpu_timers(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1644 | } |
| 1645 | |
Anna-Maria Gleixner | 73420fe | 2016-07-04 09:50:33 +0000 | [diff] [blame] | 1646 | /** |
| 1647 | * __run_timers - run all expired timers (if any) on this CPU. |
| 1648 | * @base: the timer vector to be processed. |
| 1649 | */ |
| 1650 | static inline void __run_timers(struct timer_base *base) |
| 1651 | { |
| 1652 | struct hlist_head heads[LVL_DEPTH]; |
| 1653 | int levels; |
| 1654 | |
| 1655 | if (!time_after_eq(jiffies, base->clk)) |
| 1656 | return; |
| 1657 | |
| 1658 | spin_lock_irq(&base->lock); |
| 1659 | |
Gaurav Kohli | cd3ede4 | 2018-08-02 14:21:03 +0530 | [diff] [blame^] | 1660 | /* |
| 1661 | * timer_base::must_forward_clk must be cleared before running |
| 1662 | * timers so that any timer functions that call mod_timer() will |
| 1663 | * not try to forward the base. Idle tracking / clock forwarding |
| 1664 | * logic is only used with BASE_STD timers. |
| 1665 | * |
| 1666 | * The must_forward_clk flag is cleared unconditionally also for |
| 1667 | * the deferrable base. The deferrable base is not affected by idle |
| 1668 | * tracking and never forwarded, so clearing the flag is a NOOP. |
| 1669 | * |
| 1670 | * The fact that the deferrable base is never forwarded can cause |
| 1671 | * large variations in granularity for deferrable timers, but they |
| 1672 | * can be deferred for long periods due to idle anyway. |
| 1673 | */ |
| 1674 | base->must_forward_clk = false; |
| 1675 | |
Anna-Maria Gleixner | 73420fe | 2016-07-04 09:50:33 +0000 | [diff] [blame] | 1676 | while (time_after_eq(jiffies, base->clk)) { |
| 1677 | |
| 1678 | levels = collect_expired_timers(base, heads); |
| 1679 | base->clk++; |
| 1680 | |
| 1681 | while (levels--) |
| 1682 | expire_timers(base, heads + levels); |
| 1683 | } |
| 1684 | base->running_timer = NULL; |
| 1685 | spin_unlock_irq(&base->lock); |
| 1686 | } |
| 1687 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1688 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1689 | * This function runs timers and the timer-tq in bottom half context. |
| 1690 | */ |
Emese Revfy | 0766f78 | 2016-06-20 20:42:34 +0200 | [diff] [blame] | 1691 | static __latent_entropy void run_timer_softirq(struct softirq_action *h) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1693 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1695 | __run_timers(base); |
Anna-Maria Gleixner | d840687 | 2017-12-22 15:51:12 +0100 | [diff] [blame] | 1696 | if (IS_ENABLED(CONFIG_NO_HZ_COMMON)) |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1697 | __run_timers(this_cpu_ptr(&timer_bases[BASE_DEF])); |
Prasad Sodagudi | 602c4e2 | 2017-05-17 23:26:09 -0700 | [diff] [blame] | 1698 | |
Lingutla Chandrasekhar | ea6a324 | 2018-03-01 20:41:57 +0530 | [diff] [blame] | 1699 | if ((atomic_cmpxchg(&deferrable_pending, 1, 0) && |
| 1700 | tick_do_timer_cpu == TICK_DO_TIMER_NONE) || |
| 1701 | tick_do_timer_cpu == smp_processor_id()) |
Prasad Sodagudi | 602c4e2 | 2017-05-17 23:26:09 -0700 | [diff] [blame] | 1702 | __run_timers(&timer_base_deferrable); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1703 | } |
| 1704 | |
| 1705 | /* |
| 1706 | * Called by the local, per-CPU timer interrupt on SMP. |
| 1707 | */ |
| 1708 | void run_local_timers(void) |
| 1709 | { |
Thomas Gleixner | 4e85876 | 2016-07-04 09:50:37 +0000 | [diff] [blame] | 1710 | struct timer_base *base = this_cpu_ptr(&timer_bases[BASE_STD]); |
| 1711 | |
Peter Zijlstra | d3d7445 | 2008-01-25 21:08:31 +0100 | [diff] [blame] | 1712 | hrtimer_run_queues(); |
Thomas Gleixner | 4e85876 | 2016-07-04 09:50:37 +0000 | [diff] [blame] | 1713 | /* Raise the softirq only if required. */ |
| 1714 | if (time_before(jiffies, base->clk)) { |
Thomas Gleixner | 676109b | 2018-01-14 23:19:49 +0100 | [diff] [blame] | 1715 | if (!IS_ENABLED(CONFIG_NO_HZ_COMMON)) |
Thomas Gleixner | 4e85876 | 2016-07-04 09:50:37 +0000 | [diff] [blame] | 1716 | return; |
| 1717 | /* CPU is awake, so check the deferrable base. */ |
| 1718 | base++; |
| 1719 | if (time_before(jiffies, base->clk)) |
| 1720 | return; |
| 1721 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | raise_softirq(TIMER_SOFTIRQ); |
| 1723 | } |
| 1724 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | #ifdef __ARCH_WANT_SYS_ALARM |
| 1726 | |
| 1727 | /* |
| 1728 | * For backwards compatibility? This can be done in libc so Alpha |
| 1729 | * and all newer ports shouldn't need it. |
| 1730 | */ |
Heiko Carstens | 58fd3aa | 2009-01-14 14:14:03 +0100 | [diff] [blame] | 1731 | SYSCALL_DEFINE1(alarm, unsigned int, seconds) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1732 | { |
Thomas Gleixner | c08b8a4 | 2006-03-25 03:06:33 -0800 | [diff] [blame] | 1733 | return alarm_setitimer(seconds); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1734 | } |
| 1735 | |
| 1736 | #endif |
| 1737 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | static void process_timeout(unsigned long __data) |
| 1739 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1740 | wake_up_process((struct task_struct *)__data); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1741 | } |
| 1742 | |
| 1743 | /** |
| 1744 | * schedule_timeout - sleep until timeout |
| 1745 | * @timeout: timeout value in jiffies |
| 1746 | * |
| 1747 | * Make the current task sleep until @timeout jiffies have |
| 1748 | * elapsed. The routine will return immediately unless |
| 1749 | * the current task state has been set (see set_current_state()). |
| 1750 | * |
| 1751 | * You can set the task state as follows - |
| 1752 | * |
| 1753 | * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to |
| 1754 | * pass before the routine returns. The routine will return 0 |
| 1755 | * |
| 1756 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is |
| 1757 | * delivered to the current task. In this case the remaining time |
| 1758 | * in jiffies will be returned, or 0 if the timer expired in time |
| 1759 | * |
| 1760 | * The current task state is guaranteed to be TASK_RUNNING when this |
| 1761 | * routine returns. |
| 1762 | * |
| 1763 | * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule |
| 1764 | * the CPU away without a bound on the timeout. In this case the return |
| 1765 | * value will be %MAX_SCHEDULE_TIMEOUT. |
| 1766 | * |
| 1767 | * In all cases the return value is guaranteed to be non-negative. |
| 1768 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 1769 | signed long __sched schedule_timeout(signed long timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1770 | { |
| 1771 | struct timer_list timer; |
| 1772 | unsigned long expire; |
| 1773 | |
| 1774 | switch (timeout) |
| 1775 | { |
| 1776 | case MAX_SCHEDULE_TIMEOUT: |
| 1777 | /* |
| 1778 | * These two special cases are useful to be comfortable |
| 1779 | * in the caller. Nothing more. We could take |
| 1780 | * MAX_SCHEDULE_TIMEOUT from one of the negative value |
| 1781 | * but I' d like to return a valid offset (>=0) to allow |
| 1782 | * the caller to do everything it want with the retval. |
| 1783 | */ |
| 1784 | schedule(); |
| 1785 | goto out; |
| 1786 | default: |
| 1787 | /* |
| 1788 | * Another bit of PARANOID. Note that the retval will be |
| 1789 | * 0 since no piece of kernel is supposed to do a check |
| 1790 | * for a negative retval of schedule_timeout() (since it |
| 1791 | * should never happens anyway). You just have the printk() |
| 1792 | * that will tell you if something is gone wrong and where. |
| 1793 | */ |
Andrew Morton | 5b149bc | 2006-12-22 01:10:14 -0800 | [diff] [blame] | 1794 | if (timeout < 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1795 | printk(KERN_ERR "schedule_timeout: wrong timeout " |
Andrew Morton | 5b149bc | 2006-12-22 01:10:14 -0800 | [diff] [blame] | 1796 | "value %lx\n", timeout); |
| 1797 | dump_stack(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1798 | current->state = TASK_RUNNING; |
| 1799 | goto out; |
| 1800 | } |
| 1801 | } |
| 1802 | |
| 1803 | expire = timeout + jiffies; |
| 1804 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 1805 | setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); |
Thomas Gleixner | 177ec0a | 2016-07-04 09:50:24 +0000 | [diff] [blame] | 1806 | __mod_timer(&timer, expire, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1807 | schedule(); |
| 1808 | del_singleshot_timer_sync(&timer); |
| 1809 | |
Thomas Gleixner | c6f3a97 | 2008-04-30 00:55:03 -0700 | [diff] [blame] | 1810 | /* Remove the timer from the object tracker */ |
| 1811 | destroy_timer_on_stack(&timer); |
| 1812 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | timeout = expire - jiffies; |
| 1814 | |
| 1815 | out: |
| 1816 | return timeout < 0 ? 0 : timeout; |
| 1817 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1818 | EXPORT_SYMBOL(schedule_timeout); |
| 1819 | |
Andrew Morton | 8a1c175 | 2005-09-13 01:25:15 -0700 | [diff] [blame] | 1820 | /* |
| 1821 | * We can use __set_current_state() here because schedule_timeout() calls |
| 1822 | * schedule() unconditionally. |
| 1823 | */ |
Nishanth Aravamudan | 64ed93a | 2005-09-10 00:27:21 -0700 | [diff] [blame] | 1824 | signed long __sched schedule_timeout_interruptible(signed long timeout) |
| 1825 | { |
Andrew Morton | a5a0d52 | 2005-10-30 15:01:42 -0800 | [diff] [blame] | 1826 | __set_current_state(TASK_INTERRUPTIBLE); |
| 1827 | return schedule_timeout(timeout); |
Nishanth Aravamudan | 64ed93a | 2005-09-10 00:27:21 -0700 | [diff] [blame] | 1828 | } |
| 1829 | EXPORT_SYMBOL(schedule_timeout_interruptible); |
| 1830 | |
Matthew Wilcox | 294d5cc | 2007-12-06 11:59:46 -0500 | [diff] [blame] | 1831 | signed long __sched schedule_timeout_killable(signed long timeout) |
| 1832 | { |
| 1833 | __set_current_state(TASK_KILLABLE); |
| 1834 | return schedule_timeout(timeout); |
| 1835 | } |
| 1836 | EXPORT_SYMBOL(schedule_timeout_killable); |
| 1837 | |
Nishanth Aravamudan | 64ed93a | 2005-09-10 00:27:21 -0700 | [diff] [blame] | 1838 | signed long __sched schedule_timeout_uninterruptible(signed long timeout) |
| 1839 | { |
Andrew Morton | a5a0d52 | 2005-10-30 15:01:42 -0800 | [diff] [blame] | 1840 | __set_current_state(TASK_UNINTERRUPTIBLE); |
| 1841 | return schedule_timeout(timeout); |
Nishanth Aravamudan | 64ed93a | 2005-09-10 00:27:21 -0700 | [diff] [blame] | 1842 | } |
| 1843 | EXPORT_SYMBOL(schedule_timeout_uninterruptible); |
| 1844 | |
Andrew Morton | 69b27ba | 2016-03-25 14:20:21 -0700 | [diff] [blame] | 1845 | /* |
| 1846 | * Like schedule_timeout_uninterruptible(), except this task will not contribute |
| 1847 | * to load average. |
| 1848 | */ |
| 1849 | signed long __sched schedule_timeout_idle(signed long timeout) |
| 1850 | { |
| 1851 | __set_current_state(TASK_IDLE); |
| 1852 | return schedule_timeout(timeout); |
| 1853 | } |
| 1854 | EXPORT_SYMBOL(schedule_timeout_idle); |
| 1855 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | #ifdef CONFIG_HOTPLUG_CPU |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1857 | static void migrate_timer_list(struct timer_base *new_base, |
| 1858 | struct hlist_head *head, bool remove_pinned) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1859 | { |
| 1860 | struct timer_list *timer; |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1861 | int cpu = new_base->cpu; |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1862 | struct hlist_node *n; |
| 1863 | int is_pinned; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1864 | |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1865 | hlist_for_each_entry_safe(timer, n, head, entry) { |
| 1866 | is_pinned = timer->flags & TIMER_PINNED; |
| 1867 | if (!remove_pinned && is_pinned) |
| 1868 | continue; |
| 1869 | |
| 1870 | detach_if_pending(timer, get_timer_base(timer->flags), false); |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1871 | timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1872 | internal_add_timer(new_base, timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1873 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1874 | } |
| 1875 | |
Thomas Gleixner | 249d4a9 | 2017-12-27 21:37:25 +0100 | [diff] [blame] | 1876 | int timers_prepare_cpu(unsigned int cpu) |
| 1877 | { |
| 1878 | struct timer_base *base; |
| 1879 | int b; |
| 1880 | |
| 1881 | for (b = 0; b < NR_BASES; b++) { |
| 1882 | base = per_cpu_ptr(&timer_bases[b], cpu); |
| 1883 | base->clk = jiffies; |
| 1884 | base->next_expiry = base->clk + NEXT_TIMER_MAX_DELTA; |
| 1885 | base->is_idle = false; |
| 1886 | base->must_forward_clk = true; |
| 1887 | } |
| 1888 | return 0; |
| 1889 | } |
| 1890 | |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1891 | static void __migrate_timers(unsigned int cpu, bool remove_pinned) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | { |
Thomas Gleixner | 494af3e | 2016-07-04 09:50:28 +0000 | [diff] [blame] | 1893 | struct timer_base *old_base; |
| 1894 | struct timer_base *new_base; |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1895 | unsigned long flags; |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1896 | int b, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1897 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1898 | for (b = 0; b < NR_BASES; b++) { |
| 1899 | old_base = per_cpu_ptr(&timer_bases[b], cpu); |
| 1900 | new_base = get_cpu_ptr(&timer_bases[b]); |
| 1901 | /* |
| 1902 | * The caller is globally serialized and nobody else |
| 1903 | * takes two locks at once, deadlock is not possible. |
| 1904 | */ |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1905 | spin_lock_irqsave(&new_base->lock, flags); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1906 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
Oleg Nesterov | 3691c51 | 2006-03-31 02:30:30 -0800 | [diff] [blame] | 1907 | |
Lingutla Chandrasekhar | 13e75c7 | 2018-01-18 17:20:22 +0530 | [diff] [blame] | 1908 | /* |
| 1909 | * The current CPUs base clock might be stale. Update it |
| 1910 | * before moving the timers over. |
| 1911 | */ |
| 1912 | forward_timer_base(new_base); |
| 1913 | |
Vikram Mulukutla | 78a643e | 2017-04-20 17:12:58 -0700 | [diff] [blame] | 1914 | if (!cpu_online(cpu)) |
| 1915 | BUG_ON(old_base->running_timer); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1916 | |
| 1917 | for (i = 0; i < WHEEL_SIZE; i++) |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1918 | migrate_timer_list(new_base, old_base->vectors + i, |
| 1919 | remove_pinned); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1920 | |
| 1921 | spin_unlock(&old_base->lock); |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1922 | spin_unlock_irqrestore(&new_base->lock, flags); |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1923 | put_cpu_ptr(&timer_bases); |
Oleg Nesterov | 55c888d | 2005-06-23 00:08:56 -0700 | [diff] [blame] | 1924 | } |
Viresh Kumar | 9536efe | 2015-03-25 11:47:53 +0530 | [diff] [blame] | 1925 | } |
| 1926 | |
| 1927 | int timers_dead_cpu(unsigned int cpu) |
| 1928 | { |
| 1929 | BUG_ON(cpu_online(cpu)); |
| 1930 | __migrate_timers(cpu, true); |
Richard Cochran | 24f73b9 | 2016-07-13 17:16:59 +0000 | [diff] [blame] | 1931 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1932 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1933 | |
Santosh Shukla | e92935e2 | 2015-03-25 16:09:32 +0530 | [diff] [blame] | 1934 | void timer_quiesce_cpu(void *cpup) |
| 1935 | { |
| 1936 | __migrate_timers(*(unsigned int *)cpup, false); |
| 1937 | } |
Santosh Shukla | e92935e2 | 2015-03-25 16:09:32 +0530 | [diff] [blame] | 1938 | |
Peter Zijlstra | 3650b57 | 2015-03-31 20:49:02 +0530 | [diff] [blame] | 1939 | #endif /* CONFIG_HOTPLUG_CPU */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1940 | |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1941 | static void __init init_timer_cpu(int cpu) |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1942 | { |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1943 | struct timer_base *base; |
| 1944 | int i; |
Peter Zijlstra | 3650b57 | 2015-03-31 20:49:02 +0530 | [diff] [blame] | 1945 | |
Thomas Gleixner | 500462a | 2016-07-04 09:50:30 +0000 | [diff] [blame] | 1946 | for (i = 0; i < NR_BASES; i++) { |
| 1947 | base = per_cpu_ptr(&timer_bases[i], cpu); |
| 1948 | base->cpu = cpu; |
| 1949 | spin_lock_init(&base->lock); |
| 1950 | base->clk = jiffies; |
| 1951 | } |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1952 | } |
| 1953 | |
Kyle Yan | e980f1e | 2017-03-07 11:51:38 -0800 | [diff] [blame] | 1954 | static inline void init_timer_deferrable_global(void) |
| 1955 | { |
| 1956 | timer_base_deferrable.cpu = nr_cpu_ids; |
| 1957 | spin_lock_init(&timer_base_deferrable.lock); |
| 1958 | timer_base_deferrable.clk = jiffies; |
| 1959 | } |
| 1960 | |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1961 | static void __init init_timer_cpus(void) |
| 1962 | { |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1963 | int cpu; |
| 1964 | |
Kyle Yan | e980f1e | 2017-03-07 11:51:38 -0800 | [diff] [blame] | 1965 | init_timer_deferrable_global(); |
| 1966 | |
Thomas Gleixner | 0eeda71 | 2015-05-26 22:50:29 +0000 | [diff] [blame] | 1967 | for_each_possible_cpu(cpu) |
| 1968 | init_timer_cpu(cpu); |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1969 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1970 | |
| 1971 | void __init init_timers(void) |
| 1972 | { |
Viresh Kumar | 8def906 | 2015-03-31 20:49:01 +0530 | [diff] [blame] | 1973 | init_timer_cpus(); |
Carlos R. Mafra | 962cf36 | 2008-05-15 11:15:37 -0300 | [diff] [blame] | 1974 | open_softirq(TIMER_SOFTIRQ, run_timer_softirq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1975 | } |
| 1976 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1977 | /** |
| 1978 | * msleep - sleep safely even with waitqueue interruptions |
| 1979 | * @msecs: Time in milliseconds to sleep for |
| 1980 | */ |
| 1981 | void msleep(unsigned int msecs) |
| 1982 | { |
| 1983 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; |
| 1984 | |
Nishanth Aravamudan | 75bcc8c | 2005-09-10 00:27:24 -0700 | [diff] [blame] | 1985 | while (timeout) |
| 1986 | timeout = schedule_timeout_uninterruptible(timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | } |
| 1988 | |
| 1989 | EXPORT_SYMBOL(msleep); |
| 1990 | |
| 1991 | /** |
Domen Puncer | 96ec3ef | 2005-06-25 14:58:43 -0700 | [diff] [blame] | 1992 | * msleep_interruptible - sleep waiting for signals |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | * @msecs: Time in milliseconds to sleep for |
| 1994 | */ |
| 1995 | unsigned long msleep_interruptible(unsigned int msecs) |
| 1996 | { |
| 1997 | unsigned long timeout = msecs_to_jiffies(msecs) + 1; |
| 1998 | |
Nishanth Aravamudan | 75bcc8c | 2005-09-10 00:27:24 -0700 | [diff] [blame] | 1999 | while (timeout && !signal_pending(current)) |
| 2000 | timeout = schedule_timeout_interruptible(timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | return jiffies_to_msecs(timeout); |
| 2002 | } |
| 2003 | |
| 2004 | EXPORT_SYMBOL(msleep_interruptible); |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2005 | |
Thomas Gleixner | 6deba08 | 2015-04-14 21:09:28 +0000 | [diff] [blame] | 2006 | static void __sched do_usleep_range(unsigned long min, unsigned long max) |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2007 | { |
| 2008 | ktime_t kmin; |
John Stultz | da8b44d | 2016-03-17 14:20:51 -0700 | [diff] [blame] | 2009 | u64 delta; |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2010 | |
| 2011 | kmin = ktime_set(0, min * NSEC_PER_USEC); |
John Stultz | da8b44d | 2016-03-17 14:20:51 -0700 | [diff] [blame] | 2012 | delta = (u64)(max - min) * NSEC_PER_USEC; |
Thomas Gleixner | 6deba08 | 2015-04-14 21:09:28 +0000 | [diff] [blame] | 2013 | schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL); |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2014 | } |
| 2015 | |
| 2016 | /** |
Bjorn Helgaas | b5227d0 | 2016-05-31 16:23:02 -0500 | [diff] [blame] | 2017 | * usleep_range - Sleep for an approximate time |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2018 | * @min: Minimum time in usecs to sleep |
| 2019 | * @max: Maximum time in usecs to sleep |
Bjorn Helgaas | b5227d0 | 2016-05-31 16:23:02 -0500 | [diff] [blame] | 2020 | * |
| 2021 | * In non-atomic context where the exact wakeup time is flexible, use |
| 2022 | * usleep_range() instead of udelay(). The sleep improves responsiveness |
| 2023 | * by avoiding the CPU-hogging busy-wait of udelay(), and the range reduces |
| 2024 | * power usage by allowing hrtimers to take advantage of an already- |
| 2025 | * scheduled interrupt instead of scheduling a new one just for this sleep. |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2026 | */ |
Thomas Gleixner | 2ad5d32 | 2015-04-14 21:09:30 +0000 | [diff] [blame] | 2027 | void __sched usleep_range(unsigned long min, unsigned long max) |
Patrick Pannuto | 5e7f5a1 | 2010-08-02 15:01:04 -0700 | [diff] [blame] | 2028 | { |
| 2029 | __set_current_state(TASK_UNINTERRUPTIBLE); |
| 2030 | do_usleep_range(min, max); |
| 2031 | } |
| 2032 | EXPORT_SYMBOL(usleep_range); |