linux/kernel/timer.c
<<
>>
Prefs
   1/*
   2 *  linux/kernel/timer.c
   3 *
   4 *  Kernel internal timers
   5 *
   6 *  Copyright (C) 1991, 1992  Linus Torvalds
   7 *
   8 *  1997-01-28  Modified by Finn Arne Gangstad to make timers scale better.
   9 *
  10 *  1997-09-10  Updated NTP code according to technical memorandum Jan '96
  11 *              "A Kernel Model for Precision Timekeeping" by Dave Mills
  12 *  1998-12-24  Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
  13 *              serialize accesses to xtime/lost_ticks).
  14 *                              Copyright (C) 1998  Andrea Arcangeli
  15 *  1999-03-10  Improved NTP compatibility by Ulrich Windl
  16 *  2002-05-31  Move sys_sysinfo here and make its locking sane, Robert Love
  17 *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
  18 *                              Copyright (C) 2000, 2001, 2002  Ingo Molnar
  19 *              Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
  20 */
  21
  22#include <linux/kernel_stat.h>
  23#include <linux/export.h>
  24#include <linux/interrupt.h>
  25#include <linux/percpu.h>
  26#include <linux/init.h>
  27#include <linux/mm.h>
  28#include <linux/swap.h>
  29#include <linux/pid_namespace.h>
  30#include <linux/notifier.h>
  31#include <linux/thread_info.h>
  32#include <linux/time.h>
  33#include <linux/jiffies.h>
  34#include <linux/posix-timers.h>
  35#include <linux/cpu.h>
  36#include <linux/syscalls.h>
  37#include <linux/delay.h>
  38#include <linux/tick.h>
  39#include <linux/kallsyms.h>
  40#include <linux/irq_work.h>
  41#include <linux/sched.h>
  42#include <linux/sched/sysctl.h>
  43#include <linux/slab.h>
  44#include <linux/compat.h>
  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
  52#define CREATE_TRACE_POINTS
  53#include <trace/events/timer.h>
  54
  55u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
  56
  57EXPORT_SYMBOL(jiffies_64);
  58
  59/*
  60 * per-CPU timer vector definitions:
  61 */
  62#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
  63#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
  64#define TVN_SIZE (1 << TVN_BITS)
  65#define TVR_SIZE (1 << TVR_BITS)
  66#define TVN_MASK (TVN_SIZE - 1)
  67#define TVR_MASK (TVR_SIZE - 1)
  68#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
  69
  70struct tvec {
  71        struct list_head vec[TVN_SIZE];
  72};
  73
  74struct tvec_root {
  75        struct list_head vec[TVR_SIZE];
  76};
  77
  78struct tvec_base {
  79        spinlock_t lock;
  80        struct timer_list *running_timer;
  81        unsigned long timer_jiffies;
  82        unsigned long next_timer;
  83        unsigned long active_timers;
  84        struct tvec_root tv1;
  85        struct tvec tv2;
  86        struct tvec tv3;
  87        struct tvec tv4;
  88        struct tvec tv5;
  89} ____cacheline_aligned;
  90
  91struct tvec_base boot_tvec_bases;
  92EXPORT_SYMBOL(boot_tvec_bases);
  93static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
  94
  95/* Functions below help us manage 'deferrable' flag */
  96static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
  97{
  98        return ((unsigned int)(unsigned long)base & TIMER_DEFERRABLE);
  99}
 100
 101static inline unsigned int tbase_get_irqsafe(struct tvec_base *base)
 102{
 103        return ((unsigned int)(unsigned long)base & TIMER_IRQSAFE);
 104}
 105
 106static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
 107{
 108        return ((struct tvec_base *)((unsigned long)base & ~TIMER_FLAG_MASK));
 109}
 110
 111static inline void
 112timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
 113{
 114        unsigned long flags = (unsigned long)timer->base & TIMER_FLAG_MASK;
 115
 116        timer->base = (struct tvec_base *)((unsigned long)(new_base) | flags);
 117}
 118
 119static unsigned long round_jiffies_common(unsigned long j, int cpu,
 120                bool force_up)
 121{
 122        int rem;
 123        unsigned long original = j;
 124
 125        /*
 126         * We don't want all cpus firing their timers at once hitting the
 127         * same lock or cachelines, so we skew each extra cpu with an extra
 128         * 3 jiffies. This 3 jiffies came originally from the mm/ code which
 129         * already did this.
 130         * The skew is done by adding 3*cpunr, then round, then subtract this
 131         * extra offset again.
 132         */
 133        j += cpu * 3;
 134
 135        rem = j % HZ;
 136
 137        /*
 138         * If the target jiffie is just after a whole second (which can happen
 139         * due to delays of the timer irq, long irq off times etc etc) then
 140         * we should round down to the whole second, not up. Use 1/4th second
 141         * as cutoff for this rounding as an extreme upper bound for this.
 142         * But never round down if @force_up is set.
 143         */
 144        if (rem < HZ/4 && !force_up) /* round down */
 145                j = j - rem;
 146        else /* round up */
 147                j = j - rem + HZ;
 148
 149        /* now that we have rounded, subtract the extra skew again */
 150        j -= cpu * 3;
 151
 152        /*
 153         * Make sure j is still in the future. Otherwise return the
 154         * unmodified value.
 155         */
 156        return time_is_after_jiffies(j) ? j : original;
 157}
 158
 159/**
 160 * __round_jiffies - function to round jiffies to a full second
 161 * @j: the time in (absolute) jiffies that should be rounded
 162 * @cpu: the processor number on which the timeout will happen
 163 *
 164 * __round_jiffies() rounds an absolute time in the future (in jiffies)
 165 * up or down to (approximately) full seconds. This is useful for timers
 166 * for which the exact time they fire does not matter too much, as long as
 167 * they fire approximately every X seconds.
 168 *
 169 * By rounding these timers to whole seconds, all such timers will fire
 170 * at the same time, rather than at various times spread out. The goal
 171 * of this is to have the CPU wake up less, which saves power.
 172 *
 173 * The exact rounding is skewed for each processor to avoid all
 174 * processors firing at the exact same time, which could lead
 175 * to lock contention or spurious cache line bouncing.
 176 *
 177 * The return value is the rounded version of the @j parameter.
 178 */
 179unsigned long __round_jiffies(unsigned long j, int cpu)
 180{
 181        return round_jiffies_common(j, cpu, false);
 182}
 183EXPORT_SYMBOL_GPL(__round_jiffies);
 184
 185/**
 186 * __round_jiffies_relative - function to round jiffies to a full second
 187 * @j: the time in (relative) jiffies that should be rounded
 188 * @cpu: the processor number on which the timeout will happen
 189 *
 190 * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 191 * up or down to (approximately) full seconds. This is useful for timers
 192 * for which the exact time they fire does not matter too much, as long as
 193 * they fire approximately every X seconds.
 194 *
 195 * By rounding these timers to whole seconds, all such timers will fire
 196 * at the same time, rather than at various times spread out. The goal
 197 * of this is to have the CPU wake up less, which saves power.
 198 *
 199 * The exact rounding is skewed for each processor to avoid all
 200 * processors firing at the exact same time, which could lead
 201 * to lock contention or spurious cache line bouncing.
 202 *
 203 * The return value is the rounded version of the @j parameter.
 204 */
 205unsigned long __round_jiffies_relative(unsigned long j, int cpu)
 206{
 207        unsigned long j0 = jiffies;
 208
 209        /* Use j0 because jiffies might change while we run */
 210        return round_jiffies_common(j + j0, cpu, false) - j0;
 211}
 212EXPORT_SYMBOL_GPL(__round_jiffies_relative);
 213
 214/**
 215 * round_jiffies - function to round jiffies to a full second
 216 * @j: the time in (absolute) jiffies that should be rounded
 217 *
 218 * round_jiffies() rounds an absolute time in the future (in jiffies)
 219 * up or down to (approximately) full seconds. This is useful for timers
 220 * for which the exact time they fire does not matter too much, as long as
 221 * they fire approximately every X seconds.
 222 *
 223 * By rounding these timers to whole seconds, all such timers will fire
 224 * at the same time, rather than at various times spread out. The goal
 225 * of this is to have the CPU wake up less, which saves power.
 226 *
 227 * The return value is the rounded version of the @j parameter.
 228 */
 229unsigned long round_jiffies(unsigned long j)
 230{
 231        return round_jiffies_common(j, raw_smp_processor_id(), false);
 232}
 233EXPORT_SYMBOL_GPL(round_jiffies);
 234
 235/**
 236 * round_jiffies_relative - function to round jiffies to a full second
 237 * @j: the time in (relative) jiffies that should be rounded
 238 *
 239 * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 240 * up or down to (approximately) full seconds. This is useful for timers
 241 * for which the exact time they fire does not matter too much, as long as
 242 * they fire approximately every X seconds.
 243 *
 244 * By rounding these timers to whole seconds, all such timers will fire
 245 * at the same time, rather than at various times spread out. The goal
 246 * of this is to have the CPU wake up less, which saves power.
 247 *
 248 * The return value is the rounded version of the @j parameter.
 249 */
 250unsigned long round_jiffies_relative(unsigned long j)
 251{
 252        return __round_jiffies_relative(j, raw_smp_processor_id());
 253}
 254EXPORT_SYMBOL_GPL(round_jiffies_relative);
 255
 256/**
 257 * __round_jiffies_up - function to round jiffies up to a full second
 258 * @j: the time in (absolute) jiffies that should be rounded
 259 * @cpu: the processor number on which the timeout will happen
 260 *
 261 * This is the same as __round_jiffies() except that it will never
 262 * round down.  This is useful for timeouts for which the exact time
 263 * of firing does not matter too much, as long as they don't fire too
 264 * early.
 265 */
 266unsigned long __round_jiffies_up(unsigned long j, int cpu)
 267{
 268        return round_jiffies_common(j, cpu, true);
 269}
 270EXPORT_SYMBOL_GPL(__round_jiffies_up);
 271
 272/**
 273 * __round_jiffies_up_relative - function to round jiffies up to a full second
 274 * @j: the time in (relative) jiffies that should be rounded
 275 * @cpu: the processor number on which the timeout will happen
 276 *
 277 * This is the same as __round_jiffies_relative() except that it will never
 278 * round down.  This is useful for timeouts for which the exact time
 279 * of firing does not matter too much, as long as they don't fire too
 280 * early.
 281 */
 282unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
 283{
 284        unsigned long j0 = jiffies;
 285
 286        /* Use j0 because jiffies might change while we run */
 287        return round_jiffies_common(j + j0, cpu, true) - j0;
 288}
 289EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
 290
 291/**
 292 * round_jiffies_up - function to round jiffies up to a full second
 293 * @j: the time in (absolute) jiffies that should be rounded
 294 *
 295 * This is the same as round_jiffies() except that it will never
 296 * round down.  This is useful for timeouts for which the exact time
 297 * of firing does not matter too much, as long as they don't fire too
 298 * early.
 299 */
 300unsigned long round_jiffies_up(unsigned long j)
 301{
 302        return round_jiffies_common(j, raw_smp_processor_id(), true);
 303}
 304EXPORT_SYMBOL_GPL(round_jiffies_up);
 305
 306/**
 307 * round_jiffies_up_relative - function to round jiffies up to a full second
 308 * @j: the time in (relative) jiffies that should be rounded
 309 *
 310 * This is the same as round_jiffies_relative() except that it will never
 311 * round down.  This is useful for timeouts for which the exact time
 312 * of firing does not matter too much, as long as they don't fire too
 313 * early.
 314 */
 315unsigned long round_jiffies_up_relative(unsigned long j)
 316{
 317        return __round_jiffies_up_relative(j, raw_smp_processor_id());
 318}
 319EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
 320
 321/**
 322 * set_timer_slack - set the allowed slack for a timer
 323 * @timer: the timer to be modified
 324 * @slack_hz: the amount of time (in jiffies) allowed for rounding
 325 *
 326 * Set the amount of time, in jiffies, that a certain timer has
 327 * in terms of slack. By setting this value, the timer subsystem
 328 * will schedule the actual timer somewhere between
 329 * the time mod_timer() asks for, and that time plus the slack.
 330 *
 331 * By setting the slack to -1, a percentage of the delay is used
 332 * instead.
 333 */
 334void set_timer_slack(struct timer_list *timer, int slack_hz)
 335{
 336        timer->slack = slack_hz;
 337}
 338EXPORT_SYMBOL_GPL(set_timer_slack);
 339
 340static void
 341__internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 342{
 343        unsigned long expires = timer->expires;
 344        unsigned long idx = expires - base->timer_jiffies;
 345        struct list_head *vec;
 346
 347        if (idx < TVR_SIZE) {
 348                int i = expires & TVR_MASK;
 349                vec = base->tv1.vec + i;
 350        } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
 351                int i = (expires >> TVR_BITS) & TVN_MASK;
 352                vec = base->tv2.vec + i;
 353        } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
 354                int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
 355                vec = base->tv3.vec + i;
 356        } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
 357                int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
 358                vec = base->tv4.vec + i;
 359        } else if ((signed long) idx < 0) {
 360                /*
 361                 * Can happen if you add a timer with expires == jiffies,
 362                 * or you set a timer to go off in the past
 363                 */
 364                vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
 365        } else {
 366                int i;
 367                /* If the timeout is larger than MAX_TVAL (on 64-bit
 368                 * architectures or with CONFIG_BASE_SMALL=1) then we
 369                 * use the maximum timeout.
 370                 */
 371                if (idx > MAX_TVAL) {
 372                        idx = MAX_TVAL;
 373                        expires = idx + base->timer_jiffies;
 374                }
 375                i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
 376                vec = base->tv5.vec + i;
 377        }
 378        /*
 379         * Timers are FIFO:
 380         */
 381        list_add_tail(&timer->entry, vec);
 382}
 383
 384static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 385{
 386        __internal_add_timer(base, timer);
 387        /*
 388         * Update base->active_timers and base->next_timer
 389         */
 390        if (!tbase_get_deferrable(timer->base)) {
 391                if (time_before(timer->expires, base->next_timer))
 392                        base->next_timer = timer->expires;
 393                base->active_timers++;
 394        }
 395}
 396
 397#ifdef CONFIG_TIMER_STATS
 398void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
 399{
 400        if (timer->start_site)
 401                return;
 402
 403        timer->start_site = addr;
 404        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 405        timer->start_pid = current->pid;
 406}
 407
 408static void timer_stats_account_timer(struct timer_list *timer)
 409{
 410        unsigned int flag = 0;
 411
 412        if (likely(!timer->start_site))
 413                return;
 414        if (unlikely(tbase_get_deferrable(timer->base)))
 415                flag |= TIMER_STATS_FLAG_DEFERRABLE;
 416
 417        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 418                                 timer->function, timer->start_comm, flag);
 419}
 420
 421#else
 422static void timer_stats_account_timer(struct timer_list *timer) {}
 423#endif
 424
 425#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 426
 427static struct debug_obj_descr timer_debug_descr;
 428
 429static void *timer_debug_hint(void *addr)
 430{
 431        return ((struct timer_list *) addr)->function;
 432}
 433
 434/*
 435 * fixup_init is called when:
 436 * - an active object is initialized
 437 */
 438static int timer_fixup_init(void *addr, enum debug_obj_state state)
 439{
 440        struct timer_list *timer = addr;
 441
 442        switch (state) {
 443        case ODEBUG_STATE_ACTIVE:
 444                del_timer_sync(timer);
 445                debug_object_init(timer, &timer_debug_descr);
 446                return 1;
 447        default:
 448                return 0;
 449        }
 450}
 451
 452/* Stub timer callback for improperly used timers. */
 453static void stub_timer(unsigned long data)
 454{
 455        WARN_ON(1);
 456}
 457
 458/*
 459 * fixup_activate is called when:
 460 * - an active object is activated
 461 * - an unknown object is activated (might be a statically initialized object)
 462 */
 463static int timer_fixup_activate(void *addr, enum debug_obj_state state)
 464{
 465        struct timer_list *timer = addr;
 466
 467        switch (state) {
 468
 469        case ODEBUG_STATE_NOTAVAILABLE:
 470                /*
 471                 * This is not really a fixup. The timer was
 472                 * statically initialized. We just make sure that it
 473                 * is tracked in the object tracker.
 474                 */
 475                if (timer->entry.next == NULL &&
 476                    timer->entry.prev == TIMER_ENTRY_STATIC) {
 477                        debug_object_init(timer, &timer_debug_descr);
 478                        debug_object_activate(timer, &timer_debug_descr);
 479                        return 0;
 480                } else {
 481                        setup_timer(timer, stub_timer, 0);
 482                        return 1;
 483                }
 484                return 0;
 485
 486        case ODEBUG_STATE_ACTIVE:
 487                WARN_ON(1);
 488
 489        default:
 490                return 0;
 491        }
 492}
 493
 494/*
 495 * fixup_free is called when:
 496 * - an active object is freed
 497 */
 498static int timer_fixup_free(void *addr, enum debug_obj_state state)
 499{
 500        struct timer_list *timer = addr;
 501
 502        switch (state) {
 503        case ODEBUG_STATE_ACTIVE:
 504                del_timer_sync(timer);
 505                debug_object_free(timer, &timer_debug_descr);
 506                return 1;
 507        default:
 508                return 0;
 509        }
 510}
 511
 512/*
 513 * fixup_assert_init is called when:
 514 * - an untracked/uninit-ed object is found
 515 */
 516static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
 517{
 518        struct timer_list *timer = addr;
 519
 520        switch (state) {
 521        case ODEBUG_STATE_NOTAVAILABLE:
 522                if (timer->entry.prev == TIMER_ENTRY_STATIC) {
 523                        /*
 524                         * This is not really a fixup. The timer was
 525                         * statically initialized. We just make sure that it
 526                         * is tracked in the object tracker.
 527                         */
 528                        debug_object_init(timer, &timer_debug_descr);
 529                        return 0;
 530                } else {
 531                        setup_timer(timer, stub_timer, 0);
 532                        return 1;
 533                }
 534        default:
 535                return 0;
 536        }
 537}
 538
 539static struct debug_obj_descr timer_debug_descr = {
 540        .name                   = "timer_list",
 541        .debug_hint             = timer_debug_hint,
 542        .fixup_init             = timer_fixup_init,
 543        .fixup_activate         = timer_fixup_activate,
 544        .fixup_free             = timer_fixup_free,
 545        .fixup_assert_init      = timer_fixup_assert_init,
 546};
 547
 548static inline void debug_timer_init(struct timer_list *timer)
 549{
 550        debug_object_init(timer, &timer_debug_descr);
 551}
 552
 553static inline void debug_timer_activate(struct timer_list *timer)
 554{
 555        debug_object_activate(timer, &timer_debug_descr);
 556}
 557
 558static inline void debug_timer_deactivate(struct timer_list *timer)
 559{
 560        debug_object_deactivate(timer, &timer_debug_descr);
 561}
 562
 563static inline void debug_timer_free(struct timer_list *timer)
 564{
 565        debug_object_free(timer, &timer_debug_descr);
 566}
 567
 568static inline void debug_timer_assert_init(struct timer_list *timer)
 569{
 570        debug_object_assert_init(timer, &timer_debug_descr);
 571}
 572
 573static void do_init_timer(struct timer_list *timer, unsigned int flags,
 574                          const char *name, struct lock_class_key *key);
 575
 576void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
 577                             const char *name, struct lock_class_key *key)
 578{
 579        debug_object_init_on_stack(timer, &timer_debug_descr);
 580        do_init_timer(timer, flags, name, key);
 581}
 582EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
 583
 584void destroy_timer_on_stack(struct timer_list *timer)
 585{
 586        debug_object_free(timer, &timer_debug_descr);
 587}
 588EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
 589
 590#else
 591static inline void debug_timer_init(struct timer_list *timer) { }
 592static inline void debug_timer_activate(struct timer_list *timer) { }
 593static inline void debug_timer_deactivate(struct timer_list *timer) { }
 594static inline void debug_timer_assert_init(struct timer_list *timer) { }
 595#endif
 596
 597static inline void debug_init(struct timer_list *timer)
 598{
 599        debug_timer_init(timer);
 600        trace_timer_init(timer);
 601}
 602
 603static inline void
 604debug_activate(struct timer_list *timer, unsigned long expires)
 605{
 606        debug_timer_activate(timer);
 607        trace_timer_start(timer, expires);
 608}
 609
 610static inline void debug_deactivate(struct timer_list *timer)
 611{
 612        debug_timer_deactivate(timer);
 613        trace_timer_cancel(timer);
 614}
 615
 616static inline void debug_assert_init(struct timer_list *timer)
 617{
 618        debug_timer_assert_init(timer);
 619}
 620
 621static void do_init_timer(struct timer_list *timer, unsigned int flags,
 622                          const char *name, struct lock_class_key *key)
 623{
 624        struct tvec_base *base = __raw_get_cpu_var(tvec_bases);
 625
 626        timer->entry.next = NULL;
 627        timer->base = (void *)((unsigned long)base | flags);
 628        timer->slack = -1;
 629#ifdef CONFIG_TIMER_STATS
 630        timer->start_site = NULL;
 631        timer->start_pid = -1;
 632        memset(timer->start_comm, 0, TASK_COMM_LEN);
 633#endif
 634        lockdep_init_map(&timer->lockdep_map, name, key, 0);
 635}
 636
 637/**
 638 * init_timer_key - initialize a timer
 639 * @timer: the timer to be initialized
 640 * @flags: timer flags
 641 * @name: name of the timer
 642 * @key: lockdep class key of the fake lock used for tracking timer
 643 *       sync lock dependencies
 644 *
 645 * init_timer_key() must be done to a timer prior calling *any* of the
 646 * other timer functions.
 647 */
 648void init_timer_key(struct timer_list *timer, unsigned int flags,
 649                    const char *name, struct lock_class_key *key)
 650{
 651        debug_init(timer);
 652        do_init_timer(timer, flags, name, key);
 653}
 654EXPORT_SYMBOL(init_timer_key);
 655
 656static inline void detach_timer(struct timer_list *timer, bool clear_pending)
 657{
 658        struct list_head *entry = &timer->entry;
 659
 660        debug_deactivate(timer);
 661
 662        __list_del(entry->prev, entry->next);
 663        if (clear_pending)
 664                entry->next = NULL;
 665        entry->prev = LIST_POISON2;
 666}
 667
 668static inline void
 669detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
 670{
 671        detach_timer(timer, true);
 672        if (!tbase_get_deferrable(timer->base))
 673                base->active_timers--;
 674}
 675
 676static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
 677                             bool clear_pending)
 678{
 679        if (!timer_pending(timer))
 680                return 0;
 681
 682        detach_timer(timer, clear_pending);
 683        if (!tbase_get_deferrable(timer->base)) {
 684                base->active_timers--;
 685                if (timer->expires == base->next_timer)
 686                        base->next_timer = base->timer_jiffies;
 687        }
 688        return 1;
 689}
 690
 691/*
 692 * We are using hashed locking: holding per_cpu(tvec_bases).lock
 693 * means that all timers which are tied to this base via timer->base are
 694 * locked, and the base itself is locked too.
 695 *
 696 * So __run_timers/migrate_timers can safely modify all timers which could
 697 * be found on ->tvX lists.
 698 *
 699 * When the timer's base is locked, and the timer removed from list, it is
 700 * possible to set timer->base = NULL and drop the lock: the timer remains
 701 * locked.
 702 */
 703static struct tvec_base *lock_timer_base(struct timer_list *timer,
 704                                        unsigned long *flags)
 705        __acquires(timer->base->lock)
 706{
 707        struct tvec_base *base;
 708
 709        for (;;) {
 710                struct tvec_base *prelock_base = timer->base;
 711                base = tbase_get_base(prelock_base);
 712                if (likely(base != NULL)) {
 713                        spin_lock_irqsave(&base->lock, *flags);
 714                        if (likely(prelock_base == timer->base))
 715                                return base;
 716                        /* The timer has migrated to another CPU */
 717                        spin_unlock_irqrestore(&base->lock, *flags);
 718                }
 719                cpu_relax();
 720        }
 721}
 722
 723static inline int
 724__mod_timer(struct timer_list *timer, unsigned long expires,
 725                                                bool pending_only, int pinned)
 726{
 727        struct tvec_base *base, *new_base;
 728        unsigned long flags;
 729        int ret = 0 , cpu;
 730
 731        timer_stats_timer_set_start_info(timer);
 732        BUG_ON(!timer->function);
 733
 734        base = lock_timer_base(timer, &flags);
 735
 736        ret = detach_if_pending(timer, base, false);
 737        if (!ret && pending_only)
 738                goto out_unlock;
 739
 740        debug_activate(timer, expires);
 741
 742        cpu = smp_processor_id();
 743
 744#if defined(CONFIG_NO_HZ_COMMON) && defined(CONFIG_SMP)
 745        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
 746                cpu = get_nohz_timer_target();
 747#endif
 748        new_base = per_cpu(tvec_bases, cpu);
 749
 750        if (base != new_base) {
 751                /*
 752                 * We are trying to schedule the timer on the local CPU.
 753                 * However we can't change timer's base while it is running,
 754                 * otherwise del_timer_sync() can't detect that the timer's
 755                 * handler yet has not finished. This also guarantees that
 756                 * the timer is serialized wrt itself.
 757                 */
 758                if (likely(base->running_timer != timer)) {
 759                        /* See the comment in lock_timer_base() */
 760                        timer_set_base(timer, NULL);
 761                        spin_unlock(&base->lock);
 762                        base = new_base;
 763                        spin_lock(&base->lock);
 764                        timer_set_base(timer, base);
 765                }
 766        }
 767
 768        timer->expires = expires;
 769        internal_add_timer(base, timer);
 770
 771out_unlock:
 772        spin_unlock_irqrestore(&base->lock, flags);
 773
 774        return ret;
 775}
 776
 777/**
 778 * mod_timer_pending - modify a pending timer's timeout
 779 * @timer: the pending timer to be modified
 780 * @expires: new timeout in jiffies
 781 *
 782 * mod_timer_pending() is the same for pending timers as mod_timer(),
 783 * but will not re-activate and modify already deleted timers.
 784 *
 785 * It is useful for unserialized use of timers.
 786 */
 787int mod_timer_pending(struct timer_list *timer, unsigned long expires)
 788{
 789        return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
 790}
 791EXPORT_SYMBOL(mod_timer_pending);
 792
 793/*
 794 * Decide where to put the timer while taking the slack into account
 795 *
 796 * Algorithm:
 797 *   1) calculate the maximum (absolute) time
 798 *   2) calculate the highest bit where the expires and new max are different
 799 *   3) use this bit to make a mask
 800 *   4) use the bitmask to round down the maximum time, so that all last
 801 *      bits are zeros
 802 */
 803static inline
 804unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
 805{
 806        unsigned long expires_limit, mask;
 807        int bit;
 808
 809        if (timer->slack >= 0) {
 810                expires_limit = expires + timer->slack;
 811        } else {
 812                long delta = expires - jiffies;
 813
 814                if (delta < 256)
 815                        return expires;
 816
 817                expires_limit = expires + delta / 256;
 818        }
 819        mask = expires ^ expires_limit;
 820        if (mask == 0)
 821                return expires;
 822
 823        bit = find_last_bit(&mask, BITS_PER_LONG);
 824
 825        mask = (1 << bit) - 1;
 826
 827        expires_limit = expires_limit & ~(mask);
 828
 829        return expires_limit;
 830}
 831
 832/**
 833 * mod_timer - modify a timer's timeout
 834 * @timer: the timer to be modified
 835 * @expires: new timeout in jiffies
 836 *
 837 * mod_timer() is a more efficient way to update the expire field of an
 838 * active timer (if the timer is inactive it will be activated)
 839 *
 840 * mod_timer(timer, expires) is equivalent to:
 841 *
 842 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 843 *
 844 * Note that if there are multiple unserialized concurrent users of the
 845 * same timer, then mod_timer() is the only safe way to modify the timeout,
 846 * since add_timer() cannot modify an already running timer.
 847 *
 848 * The function returns whether it has modified a pending timer or not.
 849 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
 850 * active timer returns 1.)
 851 */
 852int mod_timer(struct timer_list *timer, unsigned long expires)
 853{
 854        expires = apply_slack(timer, expires);
 855
 856        /*
 857         * This is a common optimization triggered by the
 858         * networking code - if the timer is re-modified
 859         * to be the same thing then just return:
 860         */
 861        if (timer_pending(timer) && timer->expires == expires)
 862                return 1;
 863
 864        return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
 865}
 866EXPORT_SYMBOL(mod_timer);
 867
 868/**
 869 * mod_timer_pinned - modify a timer's timeout
 870 * @timer: the timer to be modified
 871 * @expires: new timeout in jiffies
 872 *
 873 * mod_timer_pinned() is a way to update the expire field of an
 874 * active timer (if the timer is inactive it will be activated)
 875 * and to ensure that the timer is scheduled on the current CPU.
 876 *
 877 * Note that this does not prevent the timer from being migrated
 878 * when the current CPU goes offline.  If this is a problem for
 879 * you, use CPU-hotplug notifiers to handle it correctly, for
 880 * example, cancelling the timer when the corresponding CPU goes
 881 * offline.
 882 *
 883 * mod_timer_pinned(timer, expires) is equivalent to:
 884 *
 885 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 886 */
 887int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
 888{
 889        if (timer->expires == expires && timer_pending(timer))
 890                return 1;
 891
 892        return __mod_timer(timer, expires, false, TIMER_PINNED);
 893}
 894EXPORT_SYMBOL(mod_timer_pinned);
 895
 896/**
 897 * add_timer - start a timer
 898 * @timer: the timer to be added
 899 *
 900 * The kernel will do a ->function(->data) callback from the
 901 * timer interrupt at the ->expires point in the future. The
 902 * current time is 'jiffies'.
 903 *
 904 * The timer's ->expires, ->function (and if the handler uses it, ->data)
 905 * fields must be set prior calling this function.
 906 *
 907 * Timers with an ->expires field in the past will be executed in the next
 908 * timer tick.
 909 */
 910void add_timer(struct timer_list *timer)
 911{
 912        BUG_ON(timer_pending(timer));
 913        mod_timer(timer, timer->expires);
 914}
 915EXPORT_SYMBOL(add_timer);
 916
 917/**
 918 * add_timer_on - start a timer on a particular CPU
 919 * @timer: the timer to be added
 920 * @cpu: the CPU to start it on
 921 *
 922 * This is not very scalable on SMP. Double adds are not possible.
 923 */
 924void add_timer_on(struct timer_list *timer, int cpu)
 925{
 926        struct tvec_base *base = per_cpu(tvec_bases, cpu);
 927        unsigned long flags;
 928
 929        timer_stats_timer_set_start_info(timer);
 930        BUG_ON(timer_pending(timer) || !timer->function);
 931        spin_lock_irqsave(&base->lock, flags);
 932        timer_set_base(timer, base);
 933        debug_activate(timer, timer->expires);
 934        internal_add_timer(base, timer);
 935        /*
 936         * Check whether the other CPU is in dynticks mode and needs
 937         * to be triggered to reevaluate the timer wheel.
 938         * We are protected against the other CPU fiddling
 939         * with the timer by holding the timer base lock. This also
 940         * makes sure that a CPU on the way to stop its tick can not
 941         * evaluate the timer wheel.
 942         */
 943        wake_up_nohz_cpu(cpu);
 944        spin_unlock_irqrestore(&base->lock, flags);
 945}
 946EXPORT_SYMBOL_GPL(add_timer_on);
 947
 948/**
 949 * del_timer - deactive a timer.
 950 * @timer: the timer to be deactivated
 951 *
 952 * del_timer() deactivates a timer - this works on both active and inactive
 953 * timers.
 954 *
 955 * The function returns whether it has deactivated a pending timer or not.
 956 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
 957 * active timer returns 1.)
 958 */
 959int del_timer(struct timer_list *timer)
 960{
 961        struct tvec_base *base;
 962        unsigned long flags;
 963        int ret = 0;
 964
 965        debug_assert_init(timer);
 966
 967        timer_stats_timer_clear_start_info(timer);
 968        if (timer_pending(timer)) {
 969                base = lock_timer_base(timer, &flags);
 970                ret = detach_if_pending(timer, base, true);
 971                spin_unlock_irqrestore(&base->lock, flags);
 972        }
 973
 974        return ret;
 975}
 976EXPORT_SYMBOL(del_timer);
 977
 978/**
 979 * try_to_del_timer_sync - Try to deactivate a timer
 980 * @timer: timer do del
 981 *
 982 * This function tries to deactivate a timer. Upon successful (ret >= 0)
 983 * exit the timer is not queued and the handler is not running on any CPU.
 984 */
 985int try_to_del_timer_sync(struct timer_list *timer)
 986{
 987        struct tvec_base *base;
 988        unsigned long flags;
 989        int ret = -1;
 990
 991        debug_assert_init(timer);
 992
 993        base = lock_timer_base(timer, &flags);
 994
 995        if (base->running_timer != timer) {
 996                timer_stats_timer_clear_start_info(timer);
 997                ret = detach_if_pending(timer, base, true);
 998        }
 999        spin_unlock_irqrestore(&base->lock, flags);
1000
1001        return ret;
1002}
1003EXPORT_SYMBOL(try_to_del_timer_sync);
1004
1005#ifdef CONFIG_SMP
1006/**
1007 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1008 * @timer: the timer to be deactivated
1009 *
1010 * This function only differs from del_timer() on SMP: besides deactivating
1011 * the timer it also makes sure the handler has finished executing on other
1012 * CPUs.
1013 *
1014 * Synchronization rules: Callers must prevent restarting of the timer,
1015 * otherwise this function is meaningless. It must not be called from
1016 * interrupt contexts unless the timer is an irqsafe one. The caller must
1017 * not hold locks which would prevent completion of the timer's
1018 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1019 * timer is not queued and the handler is not running on any CPU.
1020 *
1021 * Note: For !irqsafe timers, you must not hold locks that are held in
1022 *   interrupt context while calling this function. Even if the lock has
1023 *   nothing to do with the timer in question.  Here's why:
1024 *
1025 *    CPU0                             CPU1
1026 *    ----                             ----
1027 *                                   <SOFTIRQ>
1028 *                                   call_timer_fn();
1029 *                                     base->running_timer = mytimer;
1030 *  spin_lock_irq(somelock);
1031 *                                     <IRQ>
1032 *                                        spin_lock(somelock);
1033 *  del_timer_sync(mytimer);
1034 *   while (base->running_timer == mytimer);
1035 *
1036 * Now del_timer_sync() will never return and never release somelock.
1037 * The interrupt on the other CPU is waiting to grab somelock but
1038 * it has interrupted the softirq that CPU0 is waiting to finish.
1039 *
1040 * The function returns whether it has deactivated a pending timer or not.
1041 */
1042int del_timer_sync(struct timer_list *timer)
1043{
1044#ifdef CONFIG_LOCKDEP
1045        unsigned long flags;
1046
1047        /*
1048         * If lockdep gives a backtrace here, please reference
1049         * the synchronization rules above.
1050         */
1051        local_irq_save(flags);
1052        lock_map_acquire(&timer->lockdep_map);
1053        lock_map_release(&timer->lockdep_map);
1054        local_irq_restore(flags);
1055#endif
1056        /*
1057         * don't use it in hardirq context, because it
1058         * could lead to deadlock.
1059         */
1060        WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
1061        for (;;) {
1062                int ret = try_to_del_timer_sync(timer);
1063                if (ret >= 0)
1064                        return ret;
1065                cpu_relax();
1066        }
1067}
1068EXPORT_SYMBOL(del_timer_sync);
1069#endif
1070
1071static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1072{
1073        /* cascade all the timers from tv up one level */
1074        struct timer_list *timer, *tmp;
1075        struct list_head tv_list;
1076
1077        list_replace_init(tv->vec + index, &tv_list);
1078
1079        /*
1080         * We are removing _all_ timers from the list, so we
1081         * don't have to detach them individually.
1082         */
1083        list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1084                BUG_ON(tbase_get_base(timer->base) != base);
1085                /* No accounting, while moving them */
1086                __internal_add_timer(base, timer);
1087        }
1088
1089        return index;
1090}
1091
1092static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1093                          unsigned long data)
1094{
1095        int count = preempt_count();
1096
1097#ifdef CONFIG_LOCKDEP
1098        /*
1099         * It is permissible to free the timer from inside the
1100         * function that is called from it, this we need to take into
1101         * account for lockdep too. To avoid bogus "held lock freed"
1102         * warnings as well as problems when looking into
1103         * timer->lockdep_map, make a copy and use that here.
1104         */
1105        struct lockdep_map lockdep_map;
1106
1107        lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1108#endif
1109        /*
1110         * Couple the lock chain with the lock chain at
1111         * del_timer_sync() by acquiring the lock_map around the fn()
1112         * call here and in del_timer_sync().
1113         */
1114        lock_map_acquire(&lockdep_map);
1115
1116        trace_timer_expire_entry(timer);
1117        fn(data);
1118        trace_timer_expire_exit(timer);
1119
1120        lock_map_release(&lockdep_map);
1121
1122        if (count != preempt_count()) {
1123                WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1124                          fn, count, preempt_count());
1125                /*
1126                 * Restore the preempt count. That gives us a decent
1127                 * chance to survive and extract information. If the
1128                 * callback kept a lock held, bad luck, but not worse
1129                 * than the BUG() we had.
1130                 */
1131                preempt_count_set(count);
1132        }
1133}
1134
1135#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1136
1137/**
1138 * __run_timers - run all expired timers (if any) on this CPU.
1139 * @base: the timer vector to be processed.
1140 *
1141 * This function cascades all vectors and executes all expired timer
1142 * vectors.
1143 */
1144static inline void __run_timers(struct tvec_base *base)
1145{
1146        struct timer_list *timer;
1147
1148        spin_lock_irq(&base->lock);
1149        while (time_after_eq(jiffies, base->timer_jiffies)) {
1150                struct list_head work_list;
1151                struct list_head *head = &work_list;
1152                int index = base->timer_jiffies & TVR_MASK;
1153
1154                /*
1155                 * Cascade timers:
1156                 */
1157                if (!index &&
1158                        (!cascade(base, &base->tv2, INDEX(0))) &&
1159                                (!cascade(base, &base->tv3, INDEX(1))) &&
1160                                        !cascade(base, &base->tv4, INDEX(2)))
1161                        cascade(base, &base->tv5, INDEX(3));
1162                ++base->timer_jiffies;
1163                list_replace_init(base->tv1.vec + index, &work_list);
1164                while (!list_empty(head)) {
1165                        void (*fn)(unsigned long);
1166                        unsigned long data;
1167                        bool irqsafe;
1168
1169                        timer = list_first_entry(head, struct timer_list,entry);
1170                        fn = timer->function;
1171                        data = timer->data;
1172                        irqsafe = tbase_get_irqsafe(timer->base);
1173
1174                        timer_stats_account_timer(timer);
1175
1176                        base->running_timer = timer;
1177                        detach_expired_timer(timer, base);
1178
1179                        if (irqsafe) {
1180                                spin_unlock(&base->lock);
1181                                call_timer_fn(timer, fn, data);
1182                                spin_lock(&base->lock);
1183                        } else {
1184                                spin_unlock_irq(&base->lock);
1185                                call_timer_fn(timer, fn, data);
1186                                spin_lock_irq(&base->lock);
1187                        }
1188                }
1189        }
1190        base->running_timer = NULL;
1191        spin_unlock_irq(&base->lock);
1192}
1193
1194#ifdef CONFIG_NO_HZ_COMMON
1195/*
1196 * Find out when the next timer event is due to happen. This
1197 * is used on S/390 to stop all activity when a CPU is idle.
1198 * This function needs to be called with interrupts disabled.
1199 */
1200static unsigned long __next_timer_interrupt(struct tvec_base *base)
1201{
1202        unsigned long timer_jiffies = base->timer_jiffies;
1203        unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1204        int index, slot, array, found = 0;
1205        struct timer_list *nte;
1206        struct tvec *varray[4];
1207
1208        /* Look for timer events in tv1. */
1209        index = slot = timer_jiffies & TVR_MASK;
1210        do {
1211                list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1212                        if (tbase_get_deferrable(nte->base))
1213                                continue;
1214
1215                        found = 1;
1216                        expires = nte->expires;
1217                        /* Look at the cascade bucket(s)? */
1218                        if (!index || slot < index)
1219                                goto cascade;
1220                        return expires;
1221                }
1222                slot = (slot + 1) & TVR_MASK;
1223        } while (slot != index);
1224
1225cascade:
1226        /* Calculate the next cascade event */
1227        if (index)
1228                timer_jiffies += TVR_SIZE - index;
1229        timer_jiffies >>= TVR_BITS;
1230
1231        /* Check tv2-tv5. */
1232        varray[0] = &base->tv2;
1233        varray[1] = &base->tv3;
1234        varray[2] = &base->tv4;
1235        varray[3] = &base->tv5;
1236
1237        for (array = 0; array < 4; array++) {
1238                struct tvec *varp = varray[array];
1239
1240                index = slot = timer_jiffies & TVN_MASK;
1241                do {
1242                        list_for_each_entry(nte, varp->vec + slot, entry) {
1243                                if (tbase_get_deferrable(nte->base))
1244                                        continue;
1245
1246                                found = 1;
1247                                if (time_before(nte->expires, expires))
1248                                        expires = nte->expires;
1249                        }
1250                        /*
1251                         * Do we still search for the first timer or are
1252                         * we looking up the cascade buckets ?
1253                         */
1254                        if (found) {
1255                                /* Look at the cascade bucket(s)? */
1256                                if (!index || slot < index)
1257                                        break;
1258                                return expires;
1259                        }
1260                        slot = (slot + 1) & TVN_MASK;
1261                } while (slot != index);
1262
1263                if (index)
1264                        timer_jiffies += TVN_SIZE - index;
1265                timer_jiffies >>= TVN_BITS;
1266        }
1267        return expires;
1268}
1269
1270/*
1271 * Check, if the next hrtimer event is before the next timer wheel
1272 * event:
1273 */
1274static unsigned long cmp_next_hrtimer_event(unsigned long now,
1275                                            unsigned long expires)
1276{
1277        ktime_t hr_delta = hrtimer_get_next_event();
1278        struct timespec tsdelta;
1279        unsigned long delta;
1280
1281        if (hr_delta.tv64 == KTIME_MAX)
1282                return expires;
1283
1284        /*
1285         * Expired timer available, let it expire in the next tick
1286         */
1287        if (hr_delta.tv64 <= 0)
1288                return now + 1;
1289
1290        tsdelta = ktime_to_timespec(hr_delta);
1291        delta = timespec_to_jiffies(&tsdelta);
1292
1293        /*
1294         * Limit the delta to the max value, which is checked in
1295         * tick_nohz_stop_sched_tick():
1296         */
1297        if (delta > NEXT_TIMER_MAX_DELTA)
1298                delta = NEXT_TIMER_MAX_DELTA;
1299
1300        /*
1301         * Take rounding errors in to account and make sure, that it
1302         * expires in the next tick. Otherwise we go into an endless
1303         * ping pong due to tick_nohz_stop_sched_tick() retriggering
1304         * the timer softirq
1305         */
1306        if (delta < 1)
1307                delta = 1;
1308        now += delta;
1309        if (time_before(now, expires))
1310                return now;
1311        return expires;
1312}
1313
1314/**
1315 * get_next_timer_interrupt - return the jiffy of the next pending timer
1316 * @now: current time (in jiffies)
1317 */
1318unsigned long get_next_timer_interrupt(unsigned long now)
1319{
1320        struct tvec_base *base = __this_cpu_read(tvec_bases);
1321        unsigned long expires = now + NEXT_TIMER_MAX_DELTA;
1322
1323        /*
1324         * Pretend that there is no timer pending if the cpu is offline.
1325         * Possible pending timers will be migrated later to an active cpu.
1326         */
1327        if (cpu_is_offline(smp_processor_id()))
1328                return expires;
1329
1330        spin_lock(&base->lock);
1331        if (base->active_timers) {
1332                if (time_before_eq(base->next_timer, base->timer_jiffies))
1333                        base->next_timer = __next_timer_interrupt(base);
1334                expires = base->next_timer;
1335        }
1336        spin_unlock(&base->lock);
1337
1338        if (time_before_eq(expires, now))
1339                return now;
1340
1341        return cmp_next_hrtimer_event(now, expires);
1342}
1343#endif
1344
1345/*
1346 * Called from the timer interrupt handler to charge one tick to the current
1347 * process.  user_tick is 1 if the tick is user time, 0 for system.
1348 */
1349void update_process_times(int user_tick)
1350{
1351        struct task_struct *p = current;
1352        int cpu = smp_processor_id();
1353
1354        /* Note: this timer irq context must be accounted for as well. */
1355        account_process_tick(p, user_tick);
1356        run_local_timers();
1357        rcu_check_callbacks(cpu, user_tick);
1358#ifdef CONFIG_IRQ_WORK
1359        if (in_irq())
1360                irq_work_run();
1361#endif
1362        scheduler_tick();
1363        run_posix_cpu_timers(p);
1364}
1365
1366/*
1367 * This function runs timers and the timer-tq in bottom half context.
1368 */
1369static void run_timer_softirq(struct softirq_action *h)
1370{
1371        struct tvec_base *base = __this_cpu_read(tvec_bases);
1372
1373        hrtimer_run_pending();
1374
1375        if (time_after_eq(jiffies, base->timer_jiffies))
1376                __run_timers(base);
1377}
1378
1379/*
1380 * Called by the local, per-CPU timer interrupt on SMP.
1381 */
1382void run_local_timers(void)
1383{
1384        hrtimer_run_queues();
1385        raise_softirq(TIMER_SOFTIRQ);
1386}
1387
1388#ifdef __ARCH_WANT_SYS_ALARM
1389
1390/*
1391 * For backwards compatibility?  This can be done in libc so Alpha
1392 * and all newer ports shouldn't need it.
1393 */
1394SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1395{
1396        return alarm_setitimer(seconds);
1397}
1398
1399#endif
1400
1401static void process_timeout(unsigned long __data)
1402{
1403        wake_up_process((struct task_struct *)__data);
1404}
1405
1406/**
1407 * schedule_timeout - sleep until timeout
1408 * @timeout: timeout value in jiffies
1409 *
1410 * Make the current task sleep until @timeout jiffies have
1411 * elapsed. The routine will return immediately unless
1412 * the current task state has been set (see set_current_state()).
1413 *
1414 * You can set the task state as follows -
1415 *
1416 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1417 * pass before the routine returns. The routine will return 0
1418 *
1419 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1420 * delivered to the current task. In this case the remaining time
1421 * in jiffies will be returned, or 0 if the timer expired in time
1422 *
1423 * The current task state is guaranteed to be TASK_RUNNING when this
1424 * routine returns.
1425 *
1426 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1427 * the CPU away without a bound on the timeout. In this case the return
1428 * value will be %MAX_SCHEDULE_TIMEOUT.
1429 *
1430 * In all cases the return value is guaranteed to be non-negative.
1431 */
1432signed long __sched schedule_timeout(signed long timeout)
1433{
1434        struct timer_list timer;
1435        unsigned long expire;
1436
1437        switch (timeout)
1438        {
1439        case MAX_SCHEDULE_TIMEOUT:
1440                /*
1441                 * These two special cases are useful to be comfortable
1442                 * in the caller. Nothing more. We could take
1443                 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1444                 * but I' d like to return a valid offset (>=0) to allow
1445                 * the caller to do everything it want with the retval.
1446                 */
1447                schedule();
1448                goto out;
1449        default:
1450                /*
1451                 * Another bit of PARANOID. Note that the retval will be
1452                 * 0 since no piece of kernel is supposed to do a check
1453                 * for a negative retval of schedule_timeout() (since it
1454                 * should never happens anyway). You just have the printk()
1455                 * that will tell you if something is gone wrong and where.
1456                 */
1457                if (timeout < 0) {
1458                        printk(KERN_ERR "schedule_timeout: wrong timeout "
1459                                "value %lx\n", timeout);
1460                        dump_stack();
1461                        current->state = TASK_RUNNING;
1462                        goto out;
1463                }
1464        }
1465
1466        expire = timeout + jiffies;
1467
1468        setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1469        __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1470        schedule();
1471        del_singleshot_timer_sync(&timer);
1472
1473        /* Remove the timer from the object tracker */
1474        destroy_timer_on_stack(&timer);
1475
1476        timeout = expire - jiffies;
1477
1478 out:
1479        return timeout < 0 ? 0 : timeout;
1480}
1481EXPORT_SYMBOL(schedule_timeout);
1482
1483/*
1484 * We can use __set_current_state() here because schedule_timeout() calls
1485 * schedule() unconditionally.
1486 */
1487signed long __sched schedule_timeout_interruptible(signed long timeout)
1488{
1489        __set_current_state(TASK_INTERRUPTIBLE);
1490        return schedule_timeout(timeout);
1491}
1492EXPORT_SYMBOL(schedule_timeout_interruptible);
1493
1494signed long __sched schedule_timeout_killable(signed long timeout)
1495{
1496        __set_current_state(TASK_KILLABLE);
1497        return schedule_timeout(timeout);
1498}
1499EXPORT_SYMBOL(schedule_timeout_killable);
1500
1501signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1502{
1503        __set_current_state(TASK_UNINTERRUPTIBLE);
1504        return schedule_timeout(timeout);
1505}
1506EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1507
1508static int init_timers_cpu(int cpu)
1509{
1510        int j;
1511        struct tvec_base *base;
1512        static char tvec_base_done[NR_CPUS];
1513
1514        if (!tvec_base_done[cpu]) {
1515                static char boot_done;
1516
1517                if (boot_done) {
1518                        /*
1519                         * The APs use this path later in boot
1520                         */
1521                        base = kzalloc_node(sizeof(*base), GFP_KERNEL,
1522                                            cpu_to_node(cpu));
1523                        if (!base)
1524                                return -ENOMEM;
1525
1526                        /* Make sure that tvec_base is 2 byte aligned */
1527                        if (tbase_get_deferrable(base)) {
1528                                WARN_ON(1);
1529                                kfree(base);
1530                                return -ENOMEM;
1531                        }
1532                        per_cpu(tvec_bases, cpu) = base;
1533                } else {
1534                        /*
1535                         * This is for the boot CPU - we use compile-time
1536                         * static initialisation because per-cpu memory isn't
1537                         * ready yet and because the memory allocators are not
1538                         * initialised either.
1539                         */
1540                        boot_done = 1;
1541                        base = &boot_tvec_bases;
1542                }
1543                spin_lock_init(&base->lock);
1544                tvec_base_done[cpu] = 1;
1545        } else {
1546                base = per_cpu(tvec_bases, cpu);
1547        }
1548
1549
1550        for (j = 0; j < TVN_SIZE; j++) {
1551                INIT_LIST_HEAD(base->tv5.vec + j);
1552                INIT_LIST_HEAD(base->tv4.vec + j);
1553                INIT_LIST_HEAD(base->tv3.vec + j);
1554                INIT_LIST_HEAD(base->tv2.vec + j);
1555        }
1556        for (j = 0; j < TVR_SIZE; j++)
1557                INIT_LIST_HEAD(base->tv1.vec + j);
1558
1559        base->timer_jiffies = jiffies;
1560        base->next_timer = base->timer_jiffies;
1561        base->active_timers = 0;
1562        return 0;
1563}
1564
1565#ifdef CONFIG_HOTPLUG_CPU
1566static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1567{
1568        struct timer_list *timer;
1569
1570        while (!list_empty(head)) {
1571                timer = list_first_entry(head, struct timer_list, entry);
1572                /* We ignore the accounting on the dying cpu */
1573                detach_timer(timer, false);
1574                timer_set_base(timer, new_base);
1575                internal_add_timer(new_base, timer);
1576        }
1577}
1578
1579static void migrate_timers(int cpu)
1580{
1581        struct tvec_base *old_base;
1582        struct tvec_base *new_base;
1583        int i;
1584
1585        BUG_ON(cpu_online(cpu));
1586        old_base = per_cpu(tvec_bases, cpu);
1587        new_base = get_cpu_var(tvec_bases);
1588        /*
1589         * The caller is globally serialized and nobody else
1590         * takes two locks at once, deadlock is not possible.
1591         */
1592        spin_lock_irq(&new_base->lock);
1593        spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1594
1595        BUG_ON(old_base->running_timer);
1596
1597        for (i = 0; i < TVR_SIZE; i++)
1598                migrate_timer_list(new_base, old_base->tv1.vec + i);
1599        for (i = 0; i < TVN_SIZE; i++) {
1600                migrate_timer_list(new_base, old_base->tv2.vec + i);
1601                migrate_timer_list(new_base, old_base->tv3.vec + i);
1602                migrate_timer_list(new_base, old_base->tv4.vec + i);
1603                migrate_timer_list(new_base, old_base->tv5.vec + i);
1604        }
1605
1606        spin_unlock(&old_base->lock);
1607        spin_unlock_irq(&new_base->lock);
1608        put_cpu_var(tvec_bases);
1609}
1610#endif /* CONFIG_HOTPLUG_CPU */
1611
1612static int timer_cpu_notify(struct notifier_block *self,
1613                                unsigned long action, void *hcpu)
1614{
1615        long cpu = (long)hcpu;
1616        int err;
1617
1618        switch(action) {
1619        case CPU_UP_PREPARE:
1620        case CPU_UP_PREPARE_FROZEN:
1621                err = init_timers_cpu(cpu);
1622                if (err < 0)
1623                        return notifier_from_errno(err);
1624                break;
1625#ifdef CONFIG_HOTPLUG_CPU
1626        case CPU_DEAD:
1627        case CPU_DEAD_FROZEN:
1628                migrate_timers(cpu);
1629                break;
1630#endif
1631        default:
1632                break;
1633        }
1634        return NOTIFY_OK;
1635}
1636
1637static struct notifier_block timers_nb = {
1638        .notifier_call  = timer_cpu_notify,
1639};
1640
1641
1642void __init init_timers(void)
1643{
1644        int err;
1645
1646        /* ensure there are enough low bits for flags in timer->base pointer */
1647        BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);
1648
1649        err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1650                               (void *)(long)smp_processor_id());
1651        init_timer_stats();
1652
1653        BUG_ON(err != NOTIFY_OK);
1654        register_cpu_notifier(&timers_nb);
1655        open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1656}
1657
1658/**
1659 * msleep - sleep safely even with waitqueue interruptions
1660 * @msecs: Time in milliseconds to sleep for
1661 */
1662void msleep(unsigned int msecs)
1663{
1664        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1665
1666        while (timeout)
1667                timeout = schedule_timeout_uninterruptible(timeout);
1668}
1669
1670EXPORT_SYMBOL(msleep);
1671
1672/**
1673 * msleep_interruptible - sleep waiting for signals
1674 * @msecs: Time in milliseconds to sleep for
1675 */
1676unsigned long msleep_interruptible(unsigned int msecs)
1677{
1678        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1679
1680        while (timeout && !signal_pending(current))
1681                timeout = schedule_timeout_interruptible(timeout);
1682        return jiffies_to_msecs(timeout);
1683}
1684
1685EXPORT_SYMBOL(msleep_interruptible);
1686
1687static int __sched do_usleep_range(unsigned long min, unsigned long max)
1688{
1689        ktime_t kmin;
1690        unsigned long delta;
1691
1692        kmin = ktime_set(0, min * NSEC_PER_USEC);
1693        delta = (max - min) * NSEC_PER_USEC;
1694        return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1695}
1696
1697/**
1698 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1699 * @min: Minimum time in usecs to sleep
1700 * @max: Maximum time in usecs to sleep
1701 */
1702void usleep_range(unsigned long min, unsigned long max)
1703{
1704        __set_current_state(TASK_UNINTERRUPTIBLE);
1705        do_usleep_range(min, max);
1706}
1707EXPORT_SYMBOL(usleep_range);
1708