linux/kernel/hrtimer.c
<<
>>
Prefs
   1/*
   2 *  linux/kernel/hrtimer.c
   3 *
   4 *  Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
   5 *  Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
   6 *  Copyright(C) 2006-2007  Timesys Corp., Thomas Gleixner
   7 *
   8 *  High-resolution kernel timers
   9 *
  10 *  In contrast to the low-resolution timeout API implemented in
  11 *  kernel/timer.c, hrtimers provide finer resolution and accuracy
  12 *  depending on system configuration and capabilities.
  13 *
  14 *  These timers are currently used for:
  15 *   - itimers
  16 *   - POSIX timers
  17 *   - nanosleep
  18 *   - precise in-kernel timing
  19 *
  20 *  Started by: Thomas Gleixner and Ingo Molnar
  21 *
  22 *  Credits:
  23 *      based on kernel/timer.c
  24 *
  25 *      Help, testing, suggestions, bugfixes, improvements were
  26 *      provided by:
  27 *
  28 *      George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel
  29 *      et. al.
  30 *
  31 *  For licencing details see kernel-base/COPYING
  32 */
  33
  34#include <linux/cpu.h>
  35#include <linux/module.h>
  36#include <linux/percpu.h>
  37#include <linux/hrtimer.h>
  38#include <linux/notifier.h>
  39#include <linux/syscalls.h>
  40#include <linux/kallsyms.h>
  41#include <linux/interrupt.h>
  42#include <linux/tick.h>
  43#include <linux/seq_file.h>
  44#include <linux/err.h>
  45#include <linux/debugobjects.h>
  46#include <linux/sched.h>
  47#include <linux/timer.h>
  48
  49#include <asm/uaccess.h>
  50
  51#include <trace/events/timer.h>
  52
  53/*
  54 * The timer bases:
  55 *
  56 * There are more clockids then hrtimer bases. Thus, we index
  57 * into the timer bases by the hrtimer_base_type enum. When trying
  58 * to reach a base using a clockid, hrtimer_clockid_to_base()
  59 * is used to convert from clockid to the proper hrtimer_base_type.
  60 */
  61DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
  62{
  63
  64        .clock_base =
  65        {
  66                {
  67                        .index = HRTIMER_BASE_MONOTONIC,
  68                        .clockid = CLOCK_MONOTONIC,
  69                        .get_time = &ktime_get,
  70                        .resolution = KTIME_LOW_RES,
  71                },
  72                {
  73                        .index = HRTIMER_BASE_REALTIME,
  74                        .clockid = CLOCK_REALTIME,
  75                        .get_time = &ktime_get_real,
  76                        .resolution = KTIME_LOW_RES,
  77                },
  78                {
  79                        .index = HRTIMER_BASE_BOOTTIME,
  80                        .clockid = CLOCK_BOOTTIME,
  81                        .get_time = &ktime_get_boottime,
  82                        .resolution = KTIME_LOW_RES,
  83                },
  84        }
  85};
  86
  87static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
  88        [CLOCK_REALTIME]        = HRTIMER_BASE_REALTIME,
  89        [CLOCK_MONOTONIC]       = HRTIMER_BASE_MONOTONIC,
  90        [CLOCK_BOOTTIME]        = HRTIMER_BASE_BOOTTIME,
  91};
  92
  93static inline int hrtimer_clockid_to_base(clockid_t clock_id)
  94{
  95        return hrtimer_clock_to_base_table[clock_id];
  96}
  97
  98
  99/*
 100 * Get the coarse grained time at the softirq based on xtime and
 101 * wall_to_monotonic.
 102 */
 103static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 104{
 105        ktime_t xtim, mono, boot;
 106        struct timespec xts, tom, slp;
 107
 108        get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
 109
 110        xtim = timespec_to_ktime(xts);
 111        mono = ktime_add(xtim, timespec_to_ktime(tom));
 112        boot = ktime_add(mono, timespec_to_ktime(slp));
 113        base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
 114        base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
 115        base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
 116}
 117
 118/*
 119 * Functions and macros which are different for UP/SMP systems are kept in a
 120 * single place
 121 */
 122#ifdef CONFIG_SMP
 123
 124/*
 125 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
 126 * means that all timers which are tied to this base via timer->base are
 127 * locked, and the base itself is locked too.
 128 *
 129 * So __run_timers/migrate_timers can safely modify all timers which could
 130 * be found on the lists/queues.
 131 *
 132 * When the timer's base is locked, and the timer removed from list, it is
 133 * possible to set timer->base = NULL and drop the lock: the timer remains
 134 * locked.
 135 */
 136static
 137struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 138                                             unsigned long *flags)
 139{
 140        struct hrtimer_clock_base *base;
 141
 142        for (;;) {
 143                base = timer->base;
 144                if (likely(base != NULL)) {
 145                        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 146                        if (likely(base == timer->base))
 147                                return base;
 148                        /* The timer has migrated to another CPU: */
 149                        raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
 150                }
 151                cpu_relax();
 152        }
 153}
 154
 155
 156/*
 157 * Get the preferred target CPU for NOHZ
 158 */
 159static int hrtimer_get_target(int this_cpu, int pinned)
 160{
 161#ifdef CONFIG_NO_HZ
 162        if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
 163                return get_nohz_timer_target();
 164#endif
 165        return this_cpu;
 166}
 167
 168/*
 169 * With HIGHRES=y we do not migrate the timer when it is expiring
 170 * before the next event on the target cpu because we cannot reprogram
 171 * the target cpu hardware and we would cause it to fire late.
 172 *
 173 * Called with cpu_base->lock of target cpu held.
 174 */
 175static int
 176hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 177{
 178#ifdef CONFIG_HIGH_RES_TIMERS
 179        ktime_t expires;
 180
 181        if (!new_base->cpu_base->hres_active)
 182                return 0;
 183
 184        expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
 185        return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
 186#else
 187        return 0;
 188#endif
 189}
 190
 191/*
 192 * Switch the timer base to the current CPU when possible.
 193 */
 194static inline struct hrtimer_clock_base *
 195switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 196                    int pinned)
 197{
 198        struct hrtimer_clock_base *new_base;
 199        struct hrtimer_cpu_base *new_cpu_base;
 200        int this_cpu = smp_processor_id();
 201        int cpu = hrtimer_get_target(this_cpu, pinned);
 202        int basenum = base->index;
 203
 204again:
 205        new_cpu_base = &per_cpu(hrtimer_bases, cpu);
 206        new_base = &new_cpu_base->clock_base[basenum];
 207
 208        if (base != new_base) {
 209                /*
 210                 * We are trying to move timer to new_base.
 211                 * However we can't change timer's base while it is running,
 212                 * so we keep it on the same CPU. No hassle vs. reprogramming
 213                 * the event source in the high resolution case. The softirq
 214                 * code will take care of this when the timer function has
 215                 * completed. There is no conflict as we hold the lock until
 216                 * the timer is enqueued.
 217                 */
 218                if (unlikely(hrtimer_callback_running(timer)))
 219                        return base;
 220
 221                /* See the comment in lock_timer_base() */
 222                timer->base = NULL;
 223                raw_spin_unlock(&base->cpu_base->lock);
 224                raw_spin_lock(&new_base->cpu_base->lock);
 225
 226                if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
 227                        cpu = this_cpu;
 228                        raw_spin_unlock(&new_base->cpu_base->lock);
 229                        raw_spin_lock(&base->cpu_base->lock);
 230                        timer->base = base;
 231                        goto again;
 232                }
 233                timer->base = new_base;
 234        }
 235        return new_base;
 236}
 237
 238#else /* CONFIG_SMP */
 239
 240static inline struct hrtimer_clock_base *
 241lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 242{
 243        struct hrtimer_clock_base *base = timer->base;
 244
 245        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 246
 247        return base;
 248}
 249
 250# define switch_hrtimer_base(t, b, p)   (b)
 251
 252#endif  /* !CONFIG_SMP */
 253
 254/*
 255 * Functions for the union type storage format of ktime_t which are
 256 * too large for inlining:
 257 */
 258#if BITS_PER_LONG < 64
 259# ifndef CONFIG_KTIME_SCALAR
 260/**
 261 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
 262 * @kt:         addend
 263 * @nsec:       the scalar nsec value to add
 264 *
 265 * Returns the sum of kt and nsec in ktime_t format
 266 */
 267ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
 268{
 269        ktime_t tmp;
 270
 271        if (likely(nsec < NSEC_PER_SEC)) {
 272                tmp.tv64 = nsec;
 273        } else {
 274                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 275
 276                tmp = ktime_set((long)nsec, rem);
 277        }
 278
 279        return ktime_add(kt, tmp);
 280}
 281
 282EXPORT_SYMBOL_GPL(ktime_add_ns);
 283
 284/**
 285 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
 286 * @kt:         minuend
 287 * @nsec:       the scalar nsec value to subtract
 288 *
 289 * Returns the subtraction of @nsec from @kt in ktime_t format
 290 */
 291ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
 292{
 293        ktime_t tmp;
 294
 295        if (likely(nsec < NSEC_PER_SEC)) {
 296                tmp.tv64 = nsec;
 297        } else {
 298                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 299
 300                tmp = ktime_set((long)nsec, rem);
 301        }
 302
 303        return ktime_sub(kt, tmp);
 304}
 305
 306EXPORT_SYMBOL_GPL(ktime_sub_ns);
 307# endif /* !CONFIG_KTIME_SCALAR */
 308
 309/*
 310 * Divide a ktime value by a nanosecond value
 311 */
 312u64 ktime_divns(const ktime_t kt, s64 div)
 313{
 314        u64 dclc;
 315        int sft = 0;
 316
 317        dclc = ktime_to_ns(kt);
 318        /* Make sure the divisor is less than 2^32: */
 319        while (div >> 32) {
 320                sft++;
 321                div >>= 1;
 322        }
 323        dclc >>= sft;
 324        do_div(dclc, (unsigned long) div);
 325
 326        return dclc;
 327}
 328#endif /* BITS_PER_LONG >= 64 */
 329
 330/*
 331 * Add two ktime values and do a safety check for overflow:
 332 */
 333ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
 334{
 335        ktime_t res = ktime_add(lhs, rhs);
 336
 337        /*
 338         * We use KTIME_SEC_MAX here, the maximum timeout which we can
 339         * return to user space in a timespec:
 340         */
 341        if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
 342                res = ktime_set(KTIME_SEC_MAX, 0);
 343
 344        return res;
 345}
 346
 347EXPORT_SYMBOL_GPL(ktime_add_safe);
 348
 349#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 350
 351static struct debug_obj_descr hrtimer_debug_descr;
 352
 353static void *hrtimer_debug_hint(void *addr)
 354{
 355        return ((struct hrtimer *) addr)->function;
 356}
 357
 358/*
 359 * fixup_init is called when:
 360 * - an active object is initialized
 361 */
 362static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
 363{
 364        struct hrtimer *timer = addr;
 365
 366        switch (state) {
 367        case ODEBUG_STATE_ACTIVE:
 368                hrtimer_cancel(timer);
 369                debug_object_init(timer, &hrtimer_debug_descr);
 370                return 1;
 371        default:
 372                return 0;
 373        }
 374}
 375
 376/*
 377 * fixup_activate is called when:
 378 * - an active object is activated
 379 * - an unknown object is activated (might be a statically initialized object)
 380 */
 381static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
 382{
 383        switch (state) {
 384
 385        case ODEBUG_STATE_NOTAVAILABLE:
 386                WARN_ON_ONCE(1);
 387                return 0;
 388
 389        case ODEBUG_STATE_ACTIVE:
 390                WARN_ON(1);
 391
 392        default:
 393                return 0;
 394        }
 395}
 396
 397/*
 398 * fixup_free is called when:
 399 * - an active object is freed
 400 */
 401static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
 402{
 403        struct hrtimer *timer = addr;
 404
 405        switch (state) {
 406        case ODEBUG_STATE_ACTIVE:
 407                hrtimer_cancel(timer);
 408                debug_object_free(timer, &hrtimer_debug_descr);
 409                return 1;
 410        default:
 411                return 0;
 412        }
 413}
 414
 415static struct debug_obj_descr hrtimer_debug_descr = {
 416        .name           = "hrtimer",
 417        .debug_hint     = hrtimer_debug_hint,
 418        .fixup_init     = hrtimer_fixup_init,
 419        .fixup_activate = hrtimer_fixup_activate,
 420        .fixup_free     = hrtimer_fixup_free,
 421};
 422
 423static inline void debug_hrtimer_init(struct hrtimer *timer)
 424{
 425        debug_object_init(timer, &hrtimer_debug_descr);
 426}
 427
 428static inline void debug_hrtimer_activate(struct hrtimer *timer)
 429{
 430        debug_object_activate(timer, &hrtimer_debug_descr);
 431}
 432
 433static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
 434{
 435        debug_object_deactivate(timer, &hrtimer_debug_descr);
 436}
 437
 438static inline void debug_hrtimer_free(struct hrtimer *timer)
 439{
 440        debug_object_free(timer, &hrtimer_debug_descr);
 441}
 442
 443static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 444                           enum hrtimer_mode mode);
 445
 446void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
 447                           enum hrtimer_mode mode)
 448{
 449        debug_object_init_on_stack(timer, &hrtimer_debug_descr);
 450        __hrtimer_init(timer, clock_id, mode);
 451}
 452EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 453
 454void destroy_hrtimer_on_stack(struct hrtimer *timer)
 455{
 456        debug_object_free(timer, &hrtimer_debug_descr);
 457}
 458
 459#else
 460static inline void debug_hrtimer_init(struct hrtimer *timer) { }
 461static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
 462static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 463#endif
 464
 465static inline void
 466debug_init(struct hrtimer *timer, clockid_t clockid,
 467           enum hrtimer_mode mode)
 468{
 469        debug_hrtimer_init(timer);
 470        trace_hrtimer_init(timer, clockid, mode);
 471}
 472
 473static inline void debug_activate(struct hrtimer *timer)
 474{
 475        debug_hrtimer_activate(timer);
 476        trace_hrtimer_start(timer);
 477}
 478
 479static inline void debug_deactivate(struct hrtimer *timer)
 480{
 481        debug_hrtimer_deactivate(timer);
 482        trace_hrtimer_cancel(timer);
 483}
 484
 485/* High resolution timer related functions */
 486#ifdef CONFIG_HIGH_RES_TIMERS
 487
 488/*
 489 * High resolution timer enabled ?
 490 */
 491static int hrtimer_hres_enabled __read_mostly  = 1;
 492
 493/*
 494 * Enable / Disable high resolution mode
 495 */
 496static int __init setup_hrtimer_hres(char *str)
 497{
 498        if (!strcmp(str, "off"))
 499                hrtimer_hres_enabled = 0;
 500        else if (!strcmp(str, "on"))
 501                hrtimer_hres_enabled = 1;
 502        else
 503                return 0;
 504        return 1;
 505}
 506
 507__setup("highres=", setup_hrtimer_hres);
 508
 509/*
 510 * hrtimer_high_res_enabled - query, if the highres mode is enabled
 511 */
 512static inline int hrtimer_is_hres_enabled(void)
 513{
 514        return hrtimer_hres_enabled;
 515}
 516
 517/*
 518 * Is the high resolution mode active ?
 519 */
 520static inline int hrtimer_hres_active(void)
 521{
 522        return __this_cpu_read(hrtimer_bases.hres_active);
 523}
 524
 525/*
 526 * Reprogram the event source with checking both queues for the
 527 * next event
 528 * Called with interrupts disabled and base->lock held
 529 */
 530static void
 531hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 532{
 533        int i;
 534        struct hrtimer_clock_base *base = cpu_base->clock_base;
 535        ktime_t expires, expires_next;
 536
 537        expires_next.tv64 = KTIME_MAX;
 538
 539        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
 540                struct hrtimer *timer;
 541                struct timerqueue_node *next;
 542
 543                next = timerqueue_getnext(&base->active);
 544                if (!next)
 545                        continue;
 546                timer = container_of(next, struct hrtimer, node);
 547
 548                expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 549                /*
 550                 * clock_was_set() has changed base->offset so the
 551                 * result might be negative. Fix it up to prevent a
 552                 * false positive in clockevents_program_event()
 553                 */
 554                if (expires.tv64 < 0)
 555                        expires.tv64 = 0;
 556                if (expires.tv64 < expires_next.tv64)
 557                        expires_next = expires;
 558        }
 559
 560        if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
 561                return;
 562
 563        cpu_base->expires_next.tv64 = expires_next.tv64;
 564
 565        if (cpu_base->expires_next.tv64 != KTIME_MAX)
 566                tick_program_event(cpu_base->expires_next, 1);
 567}
 568
 569/*
 570 * Shared reprogramming for clock_realtime and clock_monotonic
 571 *
 572 * When a timer is enqueued and expires earlier than the already enqueued
 573 * timers, we have to check, whether it expires earlier than the timer for
 574 * which the clock event device was armed.
 575 *
 576 * Called with interrupts disabled and base->cpu_base.lock held
 577 */
 578static int hrtimer_reprogram(struct hrtimer *timer,
 579                             struct hrtimer_clock_base *base)
 580{
 581        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
 582        ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 583        int res;
 584
 585        WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
 586
 587        /*
 588         * When the callback is running, we do not reprogram the clock event
 589         * device. The timer callback is either running on a different CPU or
 590         * the callback is executed in the hrtimer_interrupt context. The
 591         * reprogramming is handled either by the softirq, which called the
 592         * callback or at the end of the hrtimer_interrupt.
 593         */
 594        if (hrtimer_callback_running(timer))
 595                return 0;
 596
 597        /*
 598         * CLOCK_REALTIME timer might be requested with an absolute
 599         * expiry time which is less than base->offset. Nothing wrong
 600         * about that, just avoid to call into the tick code, which
 601         * has now objections against negative expiry values.
 602         */
 603        if (expires.tv64 < 0)
 604                return -ETIME;
 605
 606        if (expires.tv64 >= cpu_base->expires_next.tv64)
 607                return 0;
 608
 609        /*
 610         * If a hang was detected in the last timer interrupt then we
 611         * do not schedule a timer which is earlier than the expiry
 612         * which we enforced in the hang detection. We want the system
 613         * to make progress.
 614         */
 615        if (cpu_base->hang_detected)
 616                return 0;
 617
 618        /*
 619         * Clockevents returns -ETIME, when the event was in the past.
 620         */
 621        res = tick_program_event(expires, 0);
 622        if (!IS_ERR_VALUE(res))
 623                cpu_base->expires_next = expires;
 624        return res;
 625}
 626
 627/*
 628 * Initialize the high resolution related parts of cpu_base
 629 */
 630static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
 631{
 632        base->expires_next.tv64 = KTIME_MAX;
 633        base->hres_active = 0;
 634}
 635
 636/*
 637 * When High resolution timers are active, try to reprogram. Note, that in case
 638 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
 639 * check happens. The timer gets enqueued into the rbtree. The reprogramming
 640 * and expiry check is done in the hrtimer_interrupt or in the softirq.
 641 */
 642static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 643                                            struct hrtimer_clock_base *base,
 644                                            int wakeup)
 645{
 646        if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
 647                if (wakeup) {
 648                        raw_spin_unlock(&base->cpu_base->lock);
 649                        raise_softirq_irqoff(HRTIMER_SOFTIRQ);
 650                        raw_spin_lock(&base->cpu_base->lock);
 651                } else
 652                        __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
 653
 654                return 1;
 655        }
 656
 657        return 0;
 658}
 659
 660/*
 661 * Retrigger next event is called after clock was set
 662 *
 663 * Called with interrupts disabled via on_each_cpu()
 664 */
 665static void retrigger_next_event(void *arg)
 666{
 667        struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
 668        struct timespec realtime_offset, xtim, wtm, sleep;
 669
 670        if (!hrtimer_hres_active())
 671                return;
 672
 673        /* Optimized out for !HIGH_RES */
 674        get_xtime_and_monotonic_and_sleep_offset(&xtim, &wtm, &sleep);
 675        set_normalized_timespec(&realtime_offset, -wtm.tv_sec, -wtm.tv_nsec);
 676
 677        /* Adjust CLOCK_REALTIME offset */
 678        raw_spin_lock(&base->lock);
 679        base->clock_base[HRTIMER_BASE_REALTIME].offset =
 680                timespec_to_ktime(realtime_offset);
 681        base->clock_base[HRTIMER_BASE_BOOTTIME].offset =
 682                timespec_to_ktime(sleep);
 683
 684        hrtimer_force_reprogram(base, 0);
 685        raw_spin_unlock(&base->lock);
 686}
 687
 688/*
 689 * Switch to high resolution mode
 690 */
 691static int hrtimer_switch_to_hres(void)
 692{
 693        int i, cpu = smp_processor_id();
 694        struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
 695        unsigned long flags;
 696
 697        if (base->hres_active)
 698                return 1;
 699
 700        local_irq_save(flags);
 701
 702        if (tick_init_highres()) {
 703                local_irq_restore(flags);
 704                printk(KERN_WARNING "Could not switch to high resolution "
 705                                    "mode on CPU %d\n", cpu);
 706                return 0;
 707        }
 708        base->hres_active = 1;
 709        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
 710                base->clock_base[i].resolution = KTIME_HIGH_RES;
 711
 712        tick_setup_sched_timer();
 713
 714        /* "Retrigger" the interrupt to get things going */
 715        retrigger_next_event(NULL);
 716        local_irq_restore(flags);
 717        return 1;
 718}
 719
 720#else
 721
 722static inline int hrtimer_hres_active(void) { return 0; }
 723static inline int hrtimer_is_hres_enabled(void) { return 0; }
 724static inline int hrtimer_switch_to_hres(void) { return 0; }
 725static inline void
 726hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
 727static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 728                                            struct hrtimer_clock_base *base,
 729                                            int wakeup)
 730{
 731        return 0;
 732}
 733static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
 734static inline void retrigger_next_event(void *arg) { }
 735
 736#endif /* CONFIG_HIGH_RES_TIMERS */
 737
 738/*
 739 * Clock realtime was set
 740 *
 741 * Change the offset of the realtime clock vs. the monotonic
 742 * clock.
 743 *
 744 * We might have to reprogram the high resolution timer interrupt. On
 745 * SMP we call the architecture specific code to retrigger _all_ high
 746 * resolution timer interrupts. On UP we just disable interrupts and
 747 * call the high resolution interrupt code.
 748 */
 749void clock_was_set(void)
 750{
 751#ifdef CONFIG_HIGH_RES_TIMERS
 752        /* Retrigger the CPU local events everywhere */
 753        on_each_cpu(retrigger_next_event, NULL, 1);
 754#endif
 755        timerfd_clock_was_set();
 756}
 757
 758/*
 759 * During resume we might have to reprogram the high resolution timer
 760 * interrupt (on the local CPU):
 761 */
 762void hrtimers_resume(void)
 763{
 764        WARN_ONCE(!irqs_disabled(),
 765                  KERN_INFO "hrtimers_resume() called with IRQs enabled!");
 766
 767        retrigger_next_event(NULL);
 768        timerfd_clock_was_set();
 769}
 770
 771static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
 772{
 773#ifdef CONFIG_TIMER_STATS
 774        if (timer->start_site)
 775                return;
 776        timer->start_site = __builtin_return_address(0);
 777        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 778        timer->start_pid = current->pid;
 779#endif
 780}
 781
 782static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
 783{
 784#ifdef CONFIG_TIMER_STATS
 785        timer->start_site = NULL;
 786#endif
 787}
 788
 789static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
 790{
 791#ifdef CONFIG_TIMER_STATS
 792        if (likely(!timer_stats_active))
 793                return;
 794        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 795                                 timer->function, timer->start_comm, 0);
 796#endif
 797}
 798
 799/*
 800 * Counterpart to lock_hrtimer_base above:
 801 */
 802static inline
 803void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 804{
 805        raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
 806}
 807
 808/**
 809 * hrtimer_forward - forward the timer expiry
 810 * @timer:      hrtimer to forward
 811 * @now:        forward past this time
 812 * @interval:   the interval to forward
 813 *
 814 * Forward the timer expiry so it will expire in the future.
 815 * Returns the number of overruns.
 816 */
 817u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
 818{
 819        u64 orun = 1;
 820        ktime_t delta;
 821
 822        delta = ktime_sub(now, hrtimer_get_expires(timer));
 823
 824        if (delta.tv64 < 0)
 825                return 0;
 826
 827        if (interval.tv64 < timer->base->resolution.tv64)
 828                interval.tv64 = timer->base->resolution.tv64;
 829
 830        if (unlikely(delta.tv64 >= interval.tv64)) {
 831                s64 incr = ktime_to_ns(interval);
 832
 833                orun = ktime_divns(delta, incr);
 834                hrtimer_add_expires_ns(timer, incr * orun);
 835                if (hrtimer_get_expires_tv64(timer) > now.tv64)
 836                        return orun;
 837                /*
 838                 * This (and the ktime_add() below) is the
 839                 * correction for exact:
 840                 */
 841                orun++;
 842        }
 843        hrtimer_add_expires(timer, interval);
 844
 845        return orun;
 846}
 847EXPORT_SYMBOL_GPL(hrtimer_forward);
 848
 849/*
 850 * enqueue_hrtimer - internal function to (re)start a timer
 851 *
 852 * The timer is inserted in expiry order. Insertion into the
 853 * red black tree is O(log(n)). Must hold the base lock.
 854 *
 855 * Returns 1 when the new timer is the leftmost timer in the tree.
 856 */
 857static int enqueue_hrtimer(struct hrtimer *timer,
 858                           struct hrtimer_clock_base *base)
 859{
 860        debug_activate(timer);
 861
 862        timerqueue_add(&base->active, &timer->node);
 863        base->cpu_base->active_bases |= 1 << base->index;
 864
 865        /*
 866         * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
 867         * state of a possibly running callback.
 868         */
 869        timer->state |= HRTIMER_STATE_ENQUEUED;
 870
 871        return (&timer->node == base->active.next);
 872}
 873
 874/*
 875 * __remove_hrtimer - internal function to remove a timer
 876 *
 877 * Caller must hold the base lock.
 878 *
 879 * High resolution timer mode reprograms the clock event device when the
 880 * timer is the one which expires next. The caller can disable this by setting
 881 * reprogram to zero. This is useful, when the context does a reprogramming
 882 * anyway (e.g. timer interrupt)
 883 */
 884static void __remove_hrtimer(struct hrtimer *timer,
 885                             struct hrtimer_clock_base *base,
 886                             unsigned long newstate, int reprogram)
 887{
 888        struct timerqueue_node *next_timer;
 889        if (!(timer->state & HRTIMER_STATE_ENQUEUED))
 890                goto out;
 891
 892        next_timer = timerqueue_getnext(&base->active);
 893        timerqueue_del(&base->active, &timer->node);
 894        if (&timer->node == next_timer) {
 895#ifdef CONFIG_HIGH_RES_TIMERS
 896                /* Reprogram the clock event device. if enabled */
 897                if (reprogram && hrtimer_hres_active()) {
 898                        ktime_t expires;
 899
 900                        expires = ktime_sub(hrtimer_get_expires(timer),
 901                                            base->offset);
 902                        if (base->cpu_base->expires_next.tv64 == expires.tv64)
 903                                hrtimer_force_reprogram(base->cpu_base, 1);
 904                }
 905#endif
 906        }
 907        if (!timerqueue_getnext(&base->active))
 908                base->cpu_base->active_bases &= ~(1 << base->index);
 909out:
 910        timer->state = newstate;
 911}
 912
 913/*
 914 * remove hrtimer, called with base lock held
 915 */
 916static inline int
 917remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
 918{
 919        if (hrtimer_is_queued(timer)) {
 920                unsigned long state;
 921                int reprogram;
 922
 923                /*
 924                 * Remove the timer and force reprogramming when high
 925                 * resolution mode is active and the timer is on the current
 926                 * CPU. If we remove a timer on another CPU, reprogramming is
 927                 * skipped. The interrupt event on this CPU is fired and
 928                 * reprogramming happens in the interrupt handler. This is a
 929                 * rare case and less expensive than a smp call.
 930                 */
 931                debug_deactivate(timer);
 932                timer_stats_hrtimer_clear_start_info(timer);
 933                reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
 934                /*
 935                 * We must preserve the CALLBACK state flag here,
 936                 * otherwise we could move the timer base in
 937                 * switch_hrtimer_base.
 938                 */
 939                state = timer->state & HRTIMER_STATE_CALLBACK;
 940                __remove_hrtimer(timer, base, state, reprogram);
 941                return 1;
 942        }
 943        return 0;
 944}
 945
 946int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 947                unsigned long delta_ns, const enum hrtimer_mode mode,
 948                int wakeup)
 949{
 950        struct hrtimer_clock_base *base, *new_base;
 951        unsigned long flags;
 952        int ret, leftmost;
 953
 954        base = lock_hrtimer_base(timer, &flags);
 955
 956        /* Remove an active timer from the queue: */
 957        ret = remove_hrtimer(timer, base);
 958
 959        /* Switch the timer base, if necessary: */
 960        new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
 961
 962        if (mode & HRTIMER_MODE_REL) {
 963                tim = ktime_add_safe(tim, new_base->get_time());
 964                /*
 965                 * CONFIG_TIME_LOW_RES is a temporary way for architectures
 966                 * to signal that they simply return xtime in
 967                 * do_gettimeoffset(). In this case we want to round up by
 968                 * resolution when starting a relative timer, to avoid short
 969                 * timeouts. This will go away with the GTOD framework.
 970                 */
 971#ifdef CONFIG_TIME_LOW_RES
 972                tim = ktime_add_safe(tim, base->resolution);
 973#endif
 974        }
 975
 976        hrtimer_set_expires_range_ns(timer, tim, delta_ns);
 977
 978        timer_stats_hrtimer_set_start_info(timer);
 979
 980        leftmost = enqueue_hrtimer(timer, new_base);
 981
 982        /*
 983         * Only allow reprogramming if the new base is on this CPU.
 984         * (it might still be on another CPU if the timer was pending)
 985         *
 986         * XXX send_remote_softirq() ?
 987         */
 988        if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
 989                hrtimer_enqueue_reprogram(timer, new_base, wakeup);
 990
 991        unlock_hrtimer_base(timer, &flags);
 992
 993        return ret;
 994}
 995
 996/**
 997 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
 998 * @timer:      the timer to be added
 999 * @tim:        expiry time
1000 * @delta_ns:   "slack" range for the timer
1001 * @mode:       expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1002 *
1003 * Returns:
1004 *  0 on success
1005 *  1 when the timer was active
1006 */
1007int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1008                unsigned long delta_ns, const enum hrtimer_mode mode)
1009{
1010        return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1011}
1012EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1013
1014/**
1015 * hrtimer_start - (re)start an hrtimer on the current CPU
1016 * @timer:      the timer to be added
1017 * @tim:        expiry time
1018 * @mode:       expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1019 *
1020 * Returns:
1021 *  0 on success
1022 *  1 when the timer was active
1023 */
1024int
1025hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1026{
1027        return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1028}
1029EXPORT_SYMBOL_GPL(hrtimer_start);
1030
1031
1032/**
1033 * hrtimer_try_to_cancel - try to deactivate a timer
1034 * @timer:      hrtimer to stop
1035 *
1036 * Returns:
1037 *  0 when the timer was not active
1038 *  1 when the timer was active
1039 * -1 when the timer is currently excuting the callback function and
1040 *    cannot be stopped
1041 */
1042int hrtimer_try_to_cancel(struct hrtimer *timer)
1043{
1044        struct hrtimer_clock_base *base;
1045        unsigned long flags;
1046        int ret = -1;
1047
1048        base = lock_hrtimer_base(timer, &flags);
1049
1050        if (!hrtimer_callback_running(timer))
1051                ret = remove_hrtimer(timer, base);
1052
1053        unlock_hrtimer_base(timer, &flags);
1054
1055        return ret;
1056
1057}
1058EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1059
1060/**
1061 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1062 * @timer:      the timer to be cancelled
1063 *
1064 * Returns:
1065 *  0 when the timer was not active
1066 *  1 when the timer was active
1067 */
1068int hrtimer_cancel(struct hrtimer *timer)
1069{
1070        for (;;) {
1071                int ret = hrtimer_try_to_cancel(timer);
1072
1073                if (ret >= 0)
1074                        return ret;
1075                cpu_relax();
1076        }
1077}
1078EXPORT_SYMBOL_GPL(hrtimer_cancel);
1079
1080/**
1081 * hrtimer_get_remaining - get remaining time for the timer
1082 * @timer:      the timer to read
1083 */
1084ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1085{
1086        unsigned long flags;
1087        ktime_t rem;
1088
1089        lock_hrtimer_base(timer, &flags);
1090        rem = hrtimer_expires_remaining(timer);
1091        unlock_hrtimer_base(timer, &flags);
1092
1093        return rem;
1094}
1095EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1096
1097#ifdef CONFIG_NO_HZ
1098/**
1099 * hrtimer_get_next_event - get the time until next expiry event
1100 *
1101 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1102 * is pending.
1103 */
1104ktime_t hrtimer_get_next_event(void)
1105{
1106        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1107        struct hrtimer_clock_base *base = cpu_base->clock_base;
1108        ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1109        unsigned long flags;
1110        int i;
1111
1112        raw_spin_lock_irqsave(&cpu_base->lock, flags);
1113
1114        if (!hrtimer_hres_active()) {
1115                for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1116                        struct hrtimer *timer;
1117                        struct timerqueue_node *next;
1118
1119                        next = timerqueue_getnext(&base->active);
1120                        if (!next)
1121                                continue;
1122
1123                        timer = container_of(next, struct hrtimer, node);
1124                        delta.tv64 = hrtimer_get_expires_tv64(timer);
1125                        delta = ktime_sub(delta, base->get_time());
1126                        if (delta.tv64 < mindelta.tv64)
1127                                mindelta.tv64 = delta.tv64;
1128                }
1129        }
1130
1131        raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1132
1133        if (mindelta.tv64 < 0)
1134                mindelta.tv64 = 0;
1135        return mindelta;
1136}
1137#endif
1138
1139static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1140                           enum hrtimer_mode mode)
1141{
1142        struct hrtimer_cpu_base *cpu_base;
1143        int base;
1144
1145        memset(timer, 0, sizeof(struct hrtimer));
1146
1147        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1148
1149        if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1150                clock_id = CLOCK_MONOTONIC;
1151
1152        base = hrtimer_clockid_to_base(clock_id);
1153        timer->base = &cpu_base->clock_base[base];
1154        timerqueue_init(&timer->node);
1155
1156#ifdef CONFIG_TIMER_STATS
1157        timer->start_site = NULL;
1158        timer->start_pid = -1;
1159        memset(timer->start_comm, 0, TASK_COMM_LEN);
1160#endif
1161}
1162
1163/**
1164 * hrtimer_init - initialize a timer to the given clock
1165 * @timer:      the timer to be initialized
1166 * @clock_id:   the clock to be used
1167 * @mode:       timer mode abs/rel
1168 */
1169void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1170                  enum hrtimer_mode mode)
1171{
1172        debug_init(timer, clock_id, mode);
1173        __hrtimer_init(timer, clock_id, mode);
1174}
1175EXPORT_SYMBOL_GPL(hrtimer_init);
1176
1177/**
1178 * hrtimer_get_res - get the timer resolution for a clock
1179 * @which_clock: which clock to query
1180 * @tp:          pointer to timespec variable to store the resolution
1181 *
1182 * Store the resolution of the clock selected by @which_clock in the
1183 * variable pointed to by @tp.
1184 */
1185int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1186{
1187        struct hrtimer_cpu_base *cpu_base;
1188        int base = hrtimer_clockid_to_base(which_clock);
1189
1190        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1191        *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1192
1193        return 0;
1194}
1195EXPORT_SYMBOL_GPL(hrtimer_get_res);
1196
1197static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1198{
1199        struct hrtimer_clock_base *base = timer->base;
1200        struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1201        enum hrtimer_restart (*fn)(struct hrtimer *);
1202        int restart;
1203
1204        WARN_ON(!irqs_disabled());
1205
1206        debug_deactivate(timer);
1207        __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1208        timer_stats_account_hrtimer(timer);
1209        fn = timer->function;
1210
1211        /*
1212         * Because we run timers from hardirq context, there is no chance
1213         * they get migrated to another cpu, therefore its safe to unlock
1214         * the timer base.
1215         */
1216        raw_spin_unlock(&cpu_base->lock);
1217        trace_hrtimer_expire_entry(timer, now);
1218        restart = fn(timer);
1219        trace_hrtimer_expire_exit(timer);
1220        raw_spin_lock(&cpu_base->lock);
1221
1222        /*
1223         * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1224         * we do not reprogramm the event hardware. Happens either in
1225         * hrtimer_start_range_ns() or in hrtimer_interrupt()
1226         */
1227        if (restart != HRTIMER_NORESTART) {
1228                BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1229                enqueue_hrtimer(timer, base);
1230        }
1231
1232        WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1233
1234        timer->state &= ~HRTIMER_STATE_CALLBACK;
1235}
1236
1237#ifdef CONFIG_HIGH_RES_TIMERS
1238
1239/*
1240 * High resolution timer interrupt
1241 * Called with interrupts disabled
1242 */
1243void hrtimer_interrupt(struct clock_event_device *dev)
1244{
1245        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1246        ktime_t expires_next, now, entry_time, delta;
1247        int i, retries = 0;
1248
1249        BUG_ON(!cpu_base->hres_active);
1250        cpu_base->nr_events++;
1251        dev->next_event.tv64 = KTIME_MAX;
1252
1253        entry_time = now = ktime_get();
1254retry:
1255        expires_next.tv64 = KTIME_MAX;
1256
1257        raw_spin_lock(&cpu_base->lock);
1258        /*
1259         * We set expires_next to KTIME_MAX here with cpu_base->lock
1260         * held to prevent that a timer is enqueued in our queue via
1261         * the migration code. This does not affect enqueueing of
1262         * timers which run their callback and need to be requeued on
1263         * this CPU.
1264         */
1265        cpu_base->expires_next.tv64 = KTIME_MAX;
1266
1267        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1268                struct hrtimer_clock_base *base;
1269                struct timerqueue_node *node;
1270                ktime_t basenow;
1271
1272                if (!(cpu_base->active_bases & (1 << i)))
1273                        continue;
1274
1275                base = cpu_base->clock_base + i;
1276                basenow = ktime_add(now, base->offset);
1277
1278                while ((node = timerqueue_getnext(&base->active))) {
1279                        struct hrtimer *timer;
1280
1281                        timer = container_of(node, struct hrtimer, node);
1282
1283                        /*
1284                         * The immediate goal for using the softexpires is
1285                         * minimizing wakeups, not running timers at the
1286                         * earliest interrupt after their soft expiration.
1287                         * This allows us to avoid using a Priority Search
1288                         * Tree, which can answer a stabbing querry for
1289                         * overlapping intervals and instead use the simple
1290                         * BST we already have.
1291                         * We don't add extra wakeups by delaying timers that
1292                         * are right-of a not yet expired timer, because that
1293                         * timer will have to trigger a wakeup anyway.
1294                         */
1295
1296                        if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1297                                ktime_t expires;
1298
1299                                expires = ktime_sub(hrtimer_get_expires(timer),
1300                                                    base->offset);
1301                                if (expires.tv64 < expires_next.tv64)
1302                                        expires_next = expires;
1303                                break;
1304                        }
1305
1306                        __run_hrtimer(timer, &basenow);
1307                }
1308        }
1309
1310        /*
1311         * Store the new expiry value so the migration code can verify
1312         * against it.
1313         */
1314        cpu_base->expires_next = expires_next;
1315        raw_spin_unlock(&cpu_base->lock);
1316
1317        /* Reprogramming necessary ? */
1318        if (expires_next.tv64 == KTIME_MAX ||
1319            !tick_program_event(expires_next, 0)) {
1320                cpu_base->hang_detected = 0;
1321                return;
1322        }
1323
1324        /*
1325         * The next timer was already expired due to:
1326         * - tracing
1327         * - long lasting callbacks
1328         * - being scheduled away when running in a VM
1329         *
1330         * We need to prevent that we loop forever in the hrtimer
1331         * interrupt routine. We give it 3 attempts to avoid
1332         * overreacting on some spurious event.
1333         */
1334        now = ktime_get();
1335        cpu_base->nr_retries++;
1336        if (++retries < 3)
1337                goto retry;
1338        /*
1339         * Give the system a chance to do something else than looping
1340         * here. We stored the entry time, so we know exactly how long
1341         * we spent here. We schedule the next event this amount of
1342         * time away.
1343         */
1344        cpu_base->nr_hangs++;
1345        cpu_base->hang_detected = 1;
1346        delta = ktime_sub(now, entry_time);
1347        if (delta.tv64 > cpu_base->max_hang_time.tv64)
1348                cpu_base->max_hang_time = delta;
1349        /*
1350         * Limit it to a sensible value as we enforce a longer
1351         * delay. Give the CPU at least 100ms to catch up.
1352         */
1353        if (delta.tv64 > 100 * NSEC_PER_MSEC)
1354                expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1355        else
1356                expires_next = ktime_add(now, delta);
1357        tick_program_event(expires_next, 1);
1358        printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1359                    ktime_to_ns(delta));
1360}
1361
1362/*
1363 * local version of hrtimer_peek_ahead_timers() called with interrupts
1364 * disabled.
1365 */
1366static void __hrtimer_peek_ahead_timers(void)
1367{
1368        struct tick_device *td;
1369
1370        if (!hrtimer_hres_active())
1371                return;
1372
1373        td = &__get_cpu_var(tick_cpu_device);
1374        if (td && td->evtdev)
1375                hrtimer_interrupt(td->evtdev);
1376}
1377
1378/**
1379 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1380 *
1381 * hrtimer_peek_ahead_timers will peek at the timer queue of
1382 * the current cpu and check if there are any timers for which
1383 * the soft expires time has passed. If any such timers exist,
1384 * they are run immediately and then removed from the timer queue.
1385 *
1386 */
1387void hrtimer_peek_ahead_timers(void)
1388{
1389        unsigned long flags;
1390
1391        local_irq_save(flags);
1392        __hrtimer_peek_ahead_timers();
1393        local_irq_restore(flags);
1394}
1395
1396static void run_hrtimer_softirq(struct softirq_action *h)
1397{
1398        hrtimer_peek_ahead_timers();
1399}
1400
1401#else /* CONFIG_HIGH_RES_TIMERS */
1402
1403static inline void __hrtimer_peek_ahead_timers(void) { }
1404
1405#endif  /* !CONFIG_HIGH_RES_TIMERS */
1406
1407/*
1408 * Called from timer softirq every jiffy, expire hrtimers:
1409 *
1410 * For HRT its the fall back code to run the softirq in the timer
1411 * softirq context in case the hrtimer initialization failed or has
1412 * not been done yet.
1413 */
1414void hrtimer_run_pending(void)
1415{
1416        if (hrtimer_hres_active())
1417                return;
1418
1419        /*
1420         * This _is_ ugly: We have to check in the softirq context,
1421         * whether we can switch to highres and / or nohz mode. The
1422         * clocksource switch happens in the timer interrupt with
1423         * xtime_lock held. Notification from there only sets the
1424         * check bit in the tick_oneshot code, otherwise we might
1425         * deadlock vs. xtime_lock.
1426         */
1427        if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1428                hrtimer_switch_to_hres();
1429}
1430
1431/*
1432 * Called from hardirq context every jiffy
1433 */
1434void hrtimer_run_queues(void)
1435{
1436        struct timerqueue_node *node;
1437        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1438        struct hrtimer_clock_base *base;
1439        int index, gettime = 1;
1440
1441        if (hrtimer_hres_active())
1442                return;
1443
1444        for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1445                base = &cpu_base->clock_base[index];
1446                if (!timerqueue_getnext(&base->active))
1447                        continue;
1448
1449                if (gettime) {
1450                        hrtimer_get_softirq_time(cpu_base);
1451                        gettime = 0;
1452                }
1453
1454                raw_spin_lock(&cpu_base->lock);
1455
1456                while ((node = timerqueue_getnext(&base->active))) {
1457                        struct hrtimer *timer;
1458
1459                        timer = container_of(node, struct hrtimer, node);
1460                        if (base->softirq_time.tv64 <=
1461                                        hrtimer_get_expires_tv64(timer))
1462                                break;
1463
1464                        __run_hrtimer(timer, &base->softirq_time);
1465                }
1466                raw_spin_unlock(&cpu_base->lock);
1467        }
1468}
1469
1470/*
1471 * Sleep related functions:
1472 */
1473static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1474{
1475        struct hrtimer_sleeper *t =
1476                container_of(timer, struct hrtimer_sleeper, timer);
1477        struct task_struct *task = t->task;
1478
1479        t->task = NULL;
1480        if (task)
1481                wake_up_process(task);
1482
1483        return HRTIMER_NORESTART;
1484}
1485
1486void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1487{
1488        sl->timer.function = hrtimer_wakeup;
1489        sl->task = task;
1490}
1491EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1492
1493static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1494{
1495        hrtimer_init_sleeper(t, current);
1496
1497        do {
1498                set_current_state(TASK_INTERRUPTIBLE);
1499                hrtimer_start_expires(&t->timer, mode);
1500                if (!hrtimer_active(&t->timer))
1501                        t->task = NULL;
1502
1503                if (likely(t->task))
1504                        schedule();
1505
1506                hrtimer_cancel(&t->timer);
1507                mode = HRTIMER_MODE_ABS;
1508
1509        } while (t->task && !signal_pending(current));
1510
1511        __set_current_state(TASK_RUNNING);
1512
1513        return t->task == NULL;
1514}
1515
1516static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1517{
1518        struct timespec rmt;
1519        ktime_t rem;
1520
1521        rem = hrtimer_expires_remaining(timer);
1522        if (rem.tv64 <= 0)
1523                return 0;
1524        rmt = ktime_to_timespec(rem);
1525
1526        if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1527                return -EFAULT;
1528
1529        return 1;
1530}
1531
1532long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1533{
1534        struct hrtimer_sleeper t;
1535        struct timespec __user  *rmtp;
1536        int ret = 0;
1537
1538        hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1539                                HRTIMER_MODE_ABS);
1540        hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1541
1542        if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1543                goto out;
1544
1545        rmtp = restart->nanosleep.rmtp;
1546        if (rmtp) {
1547                ret = update_rmtp(&t.timer, rmtp);
1548                if (ret <= 0)
1549                        goto out;
1550        }
1551
1552        /* The other values in restart are already filled in */
1553        ret = -ERESTART_RESTARTBLOCK;
1554out:
1555        destroy_hrtimer_on_stack(&t.timer);
1556        return ret;
1557}
1558
1559long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1560                       const enum hrtimer_mode mode, const clockid_t clockid)
1561{
1562        struct restart_block *restart;
1563        struct hrtimer_sleeper t;
1564        int ret = 0;
1565        unsigned long slack;
1566
1567        slack = current->timer_slack_ns;
1568        if (rt_task(current))
1569                slack = 0;
1570
1571        hrtimer_init_on_stack(&t.timer, clockid, mode);
1572        hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1573        if (do_nanosleep(&t, mode))
1574                goto out;
1575
1576        /* Absolute timers do not update the rmtp value and restart: */
1577        if (mode == HRTIMER_MODE_ABS) {
1578                ret = -ERESTARTNOHAND;
1579                goto out;
1580        }
1581
1582        if (rmtp) {
1583                ret = update_rmtp(&t.timer, rmtp);
1584                if (ret <= 0)
1585                        goto out;
1586        }
1587
1588        restart = &current_thread_info()->restart_block;
1589        restart->fn = hrtimer_nanosleep_restart;
1590        restart->nanosleep.clockid = t.timer.base->clockid;
1591        restart->nanosleep.rmtp = rmtp;
1592        restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1593
1594        ret = -ERESTART_RESTARTBLOCK;
1595out:
1596        destroy_hrtimer_on_stack(&t.timer);
1597        return ret;
1598}
1599
1600SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1601                struct timespec __user *, rmtp)
1602{
1603        struct timespec tu;
1604
1605        if (copy_from_user(&tu, rqtp, sizeof(tu)))
1606                return -EFAULT;
1607
1608        if (!timespec_valid(&tu))
1609                return -EINVAL;
1610
1611        return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1612}
1613
1614/*
1615 * Functions related to boot-time initialization:
1616 */
1617static void __cpuinit init_hrtimers_cpu(int cpu)
1618{
1619        struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1620        int i;
1621
1622        raw_spin_lock_init(&cpu_base->lock);
1623
1624        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1625                cpu_base->clock_base[i].cpu_base = cpu_base;
1626                timerqueue_init_head(&cpu_base->clock_base[i].active);
1627        }
1628
1629        hrtimer_init_hres(cpu_base);
1630}
1631
1632#ifdef CONFIG_HOTPLUG_CPU
1633
1634static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1635                                struct hrtimer_clock_base *new_base)
1636{
1637        struct hrtimer *timer;
1638        struct timerqueue_node *node;
1639
1640        while ((node = timerqueue_getnext(&old_base->active))) {
1641                timer = container_of(node, struct hrtimer, node);
1642                BUG_ON(hrtimer_callback_running(timer));
1643                debug_deactivate(timer);
1644
1645                /*
1646                 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1647                 * timer could be seen as !active and just vanish away
1648                 * under us on another CPU
1649                 */
1650                __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1651                timer->base = new_base;
1652                /*
1653                 * Enqueue the timers on the new cpu. This does not
1654                 * reprogram the event device in case the timer
1655                 * expires before the earliest on this CPU, but we run
1656                 * hrtimer_interrupt after we migrated everything to
1657                 * sort out already expired timers and reprogram the
1658                 * event device.
1659                 */
1660                enqueue_hrtimer(timer, new_base);
1661
1662                /* Clear the migration state bit */
1663                timer->state &= ~HRTIMER_STATE_MIGRATE;
1664        }
1665}
1666
1667static void migrate_hrtimers(int scpu)
1668{
1669        struct hrtimer_cpu_base *old_base, *new_base;
1670        int i;
1671
1672        BUG_ON(cpu_online(scpu));
1673        tick_cancel_sched_timer(scpu);
1674
1675        local_irq_disable();
1676        old_base = &per_cpu(hrtimer_bases, scpu);
1677        new_base = &__get_cpu_var(hrtimer_bases);
1678        /*
1679         * The caller is globally serialized and nobody else
1680         * takes two locks at once, deadlock is not possible.
1681         */
1682        raw_spin_lock(&new_base->lock);
1683        raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1684
1685        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1686                migrate_hrtimer_list(&old_base->clock_base[i],
1687                                     &new_base->clock_base[i]);
1688        }
1689
1690        raw_spin_unlock(&old_base->lock);
1691        raw_spin_unlock(&new_base->lock);
1692
1693        /* Check, if we got expired work to do */
1694        __hrtimer_peek_ahead_timers();
1695        local_irq_enable();
1696}
1697
1698#endif /* CONFIG_HOTPLUG_CPU */
1699
1700static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1701                                        unsigned long action, void *hcpu)
1702{
1703        int scpu = (long)hcpu;
1704
1705        switch (action) {
1706
1707        case CPU_UP_PREPARE:
1708        case CPU_UP_PREPARE_FROZEN:
1709                init_hrtimers_cpu(scpu);
1710                break;
1711
1712#ifdef CONFIG_HOTPLUG_CPU
1713        case CPU_DYING:
1714        case CPU_DYING_FROZEN:
1715                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1716                break;
1717        case CPU_DEAD:
1718        case CPU_DEAD_FROZEN:
1719        {
1720                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1721                migrate_hrtimers(scpu);
1722                break;
1723        }
1724#endif
1725
1726        default:
1727                break;
1728        }
1729
1730        return NOTIFY_OK;
1731}
1732
1733static struct notifier_block __cpuinitdata hrtimers_nb = {
1734        .notifier_call = hrtimer_cpu_notify,
1735};
1736
1737void __init hrtimers_init(void)
1738{
1739        hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1740                          (void *)(long)smp_processor_id());
1741        register_cpu_notifier(&hrtimers_nb);
1742#ifdef CONFIG_HIGH_RES_TIMERS
1743        open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1744#endif
1745}
1746
1747/**
1748 * schedule_hrtimeout_range_clock - sleep until timeout
1749 * @expires:    timeout value (ktime_t)
1750 * @delta:      slack in expires timeout (ktime_t)
1751 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1752 * @clock:      timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1753 */
1754int __sched
1755schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1756                               const enum hrtimer_mode mode, int clock)
1757{
1758        struct hrtimer_sleeper t;
1759
1760        /*
1761         * Optimize when a zero timeout value is given. It does not
1762         * matter whether this is an absolute or a relative time.
1763         */
1764        if (expires && !expires->tv64) {
1765                __set_current_state(TASK_RUNNING);
1766                return 0;
1767        }
1768
1769        /*
1770         * A NULL parameter means "infinite"
1771         */
1772        if (!expires) {
1773                schedule();
1774                __set_current_state(TASK_RUNNING);
1775                return -EINTR;
1776        }
1777
1778        hrtimer_init_on_stack(&t.timer, clock, mode);
1779        hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1780
1781        hrtimer_init_sleeper(&t, current);
1782
1783        hrtimer_start_expires(&t.timer, mode);
1784        if (!hrtimer_active(&t.timer))
1785                t.task = NULL;
1786
1787        if (likely(t.task))
1788                schedule();
1789
1790        hrtimer_cancel(&t.timer);
1791        destroy_hrtimer_on_stack(&t.timer);
1792
1793        __set_current_state(TASK_RUNNING);
1794
1795        return !t.task ? 0 : -EINTR;
1796}
1797
1798/**
1799 * schedule_hrtimeout_range - sleep until timeout
1800 * @expires:    timeout value (ktime_t)
1801 * @delta:      slack in expires timeout (ktime_t)
1802 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1803 *
1804 * Make the current task sleep until the given expiry time has
1805 * elapsed. The routine will return immediately unless
1806 * the current task state has been set (see set_current_state()).
1807 *
1808 * The @delta argument gives the kernel the freedom to schedule the
1809 * actual wakeup to a time that is both power and performance friendly.
1810 * The kernel give the normal best effort behavior for "@expires+@delta",
1811 * but may decide to fire the timer earlier, but no earlier than @expires.
1812 *
1813 * You can set the task state as follows -
1814 *
1815 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1816 * pass before the routine returns.
1817 *
1818 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1819 * delivered to the current task.
1820 *
1821 * The current task state is guaranteed to be TASK_RUNNING when this
1822 * routine returns.
1823 *
1824 * Returns 0 when the timer has expired otherwise -EINTR
1825 */
1826int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1827                                     const enum hrtimer_mode mode)
1828{
1829        return schedule_hrtimeout_range_clock(expires, delta, mode,
1830                                              CLOCK_MONOTONIC);
1831}
1832EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1833
1834/**
1835 * schedule_hrtimeout - sleep until timeout
1836 * @expires:    timeout value (ktime_t)
1837 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1838 *
1839 * Make the current task sleep until the given expiry time has
1840 * elapsed. The routine will return immediately unless
1841 * the current task state has been set (see set_current_state()).
1842 *
1843 * You can set the task state as follows -
1844 *
1845 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1846 * pass before the routine returns.
1847 *
1848 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1849 * delivered to the current task.
1850 *
1851 * The current task state is guaranteed to be TASK_RUNNING when this
1852 * routine returns.
1853 *
1854 * Returns 0 when the timer has expired otherwise -EINTR
1855 */
1856int __sched schedule_hrtimeout(ktime_t *expires,
1857                               const enum hrtimer_mode mode)
1858{
1859        return schedule_hrtimeout_range(expires, 0, mode);
1860}
1861EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1862
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.