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/export.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/sched/sysctl.h>
  48#include <linux/sched/rt.h>
  49#include <linux/timer.h>
  50
  51#include <asm/uaccess.h>
  52
  53#include <trace/events/timer.h>
  54
  55/*
  56 * The timer bases:
  57 *
  58 * There are more clockids then hrtimer bases. Thus, we index
  59 * into the timer bases by the hrtimer_base_type enum. When trying
  60 * to reach a base using a clockid, hrtimer_clockid_to_base()
  61 * is used to convert from clockid to the proper hrtimer_base_type.
  62 */
  63DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
  64{
  65
  66        .lock = __RAW_SPIN_LOCK_UNLOCKED(hrtimer_bases.lock),
  67        .clock_base =
  68        {
  69                {
  70                        .index = HRTIMER_BASE_MONOTONIC,
  71                        .clockid = CLOCK_MONOTONIC,
  72                        .get_time = &ktime_get,
  73                        .resolution = KTIME_LOW_RES,
  74                },
  75                {
  76                        .index = HRTIMER_BASE_REALTIME,
  77                        .clockid = CLOCK_REALTIME,
  78                        .get_time = &ktime_get_real,
  79                        .resolution = KTIME_LOW_RES,
  80                },
  81                {
  82                        .index = HRTIMER_BASE_BOOTTIME,
  83                        .clockid = CLOCK_BOOTTIME,
  84                        .get_time = &ktime_get_boottime,
  85                        .resolution = KTIME_LOW_RES,
  86                },
  87        }
  88};
  89
  90static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
  91        [CLOCK_REALTIME]        = HRTIMER_BASE_REALTIME,
  92        [CLOCK_MONOTONIC]       = HRTIMER_BASE_MONOTONIC,
  93        [CLOCK_BOOTTIME]        = HRTIMER_BASE_BOOTTIME,
  94};
  95
  96static inline int hrtimer_clockid_to_base(clockid_t clock_id)
  97{
  98        return hrtimer_clock_to_base_table[clock_id];
  99}
 100
 101
 102/*
 103 * Get the coarse grained time at the softirq based on xtime and
 104 * wall_to_monotonic.
 105 */
 106static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 107{
 108        ktime_t xtim, mono, boot;
 109        struct timespec xts, tom, slp;
 110
 111        get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
 112
 113        xtim = timespec_to_ktime(xts);
 114        mono = ktime_add(xtim, timespec_to_ktime(tom));
 115        boot = ktime_add(mono, timespec_to_ktime(slp));
 116        base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
 117        base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
 118        base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
 119}
 120
 121/*
 122 * Functions and macros which are different for UP/SMP systems are kept in a
 123 * single place
 124 */
 125#ifdef CONFIG_SMP
 126
 127/*
 128 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
 129 * means that all timers which are tied to this base via timer->base are
 130 * locked, and the base itself is locked too.
 131 *
 132 * So __run_timers/migrate_timers can safely modify all timers which could
 133 * be found on the lists/queues.
 134 *
 135 * When the timer's base is locked, and the timer removed from list, it is
 136 * possible to set timer->base = NULL and drop the lock: the timer remains
 137 * locked.
 138 */
 139static
 140struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 141                                             unsigned long *flags)
 142{
 143        struct hrtimer_clock_base *base;
 144
 145        for (;;) {
 146                base = timer->base;
 147                if (likely(base != NULL)) {
 148                        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 149                        if (likely(base == timer->base))
 150                                return base;
 151                        /* The timer has migrated to another CPU: */
 152                        raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
 153                }
 154                cpu_relax();
 155        }
 156}
 157
 158
 159/*
 160 * Get the preferred target CPU for NOHZ
 161 */
 162static int hrtimer_get_target(int this_cpu, int pinned)
 163{
 164#ifdef CONFIG_NO_HZ
 165        if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
 166                return get_nohz_timer_target();
 167#endif
 168        return this_cpu;
 169}
 170
 171/*
 172 * With HIGHRES=y we do not migrate the timer when it is expiring
 173 * before the next event on the target cpu because we cannot reprogram
 174 * the target cpu hardware and we would cause it to fire late.
 175 *
 176 * Called with cpu_base->lock of target cpu held.
 177 */
 178static int
 179hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 180{
 181#ifdef CONFIG_HIGH_RES_TIMERS
 182        ktime_t expires;
 183
 184        if (!new_base->cpu_base->hres_active)
 185                return 0;
 186
 187        expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
 188        return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
 189#else
 190        return 0;
 191#endif
 192}
 193
 194/*
 195 * Switch the timer base to the current CPU when possible.
 196 */
 197static inline struct hrtimer_clock_base *
 198switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 199                    int pinned)
 200{
 201        struct hrtimer_clock_base *new_base;
 202        struct hrtimer_cpu_base *new_cpu_base;
 203        int this_cpu = smp_processor_id();
 204        int cpu = hrtimer_get_target(this_cpu, pinned);
 205        int basenum = base->index;
 206
 207again:
 208        new_cpu_base = &per_cpu(hrtimer_bases, cpu);
 209        new_base = &new_cpu_base->clock_base[basenum];
 210
 211        if (base != new_base) {
 212                /*
 213                 * We are trying to move timer to new_base.
 214                 * However we can't change timer's base while it is running,
 215                 * so we keep it on the same CPU. No hassle vs. reprogramming
 216                 * the event source in the high resolution case. The softirq
 217                 * code will take care of this when the timer function has
 218                 * completed. There is no conflict as we hold the lock until
 219                 * the timer is enqueued.
 220                 */
 221                if (unlikely(hrtimer_callback_running(timer)))
 222                        return base;
 223
 224                /* See the comment in lock_timer_base() */
 225                timer->base = NULL;
 226                raw_spin_unlock(&base->cpu_base->lock);
 227                raw_spin_lock(&new_base->cpu_base->lock);
 228
 229                if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
 230                        cpu = this_cpu;
 231                        raw_spin_unlock(&new_base->cpu_base->lock);
 232                        raw_spin_lock(&base->cpu_base->lock);
 233                        timer->base = base;
 234                        goto again;
 235                }
 236                timer->base = new_base;
 237        }
 238        return new_base;
 239}
 240
 241#else /* CONFIG_SMP */
 242
 243static inline struct hrtimer_clock_base *
 244lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 245{
 246        struct hrtimer_clock_base *base = timer->base;
 247
 248        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 249
 250        return base;
 251}
 252
 253# define switch_hrtimer_base(t, b, p)   (b)
 254
 255#endif  /* !CONFIG_SMP */
 256
 257/*
 258 * Functions for the union type storage format of ktime_t which are
 259 * too large for inlining:
 260 */
 261#if BITS_PER_LONG < 64
 262# ifndef CONFIG_KTIME_SCALAR
 263/**
 264 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
 265 * @kt:         addend
 266 * @nsec:       the scalar nsec value to add
 267 *
 268 * Returns the sum of kt and nsec in ktime_t format
 269 */
 270ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
 271{
 272        ktime_t tmp;
 273
 274        if (likely(nsec < NSEC_PER_SEC)) {
 275                tmp.tv64 = nsec;
 276        } else {
 277                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 278
 279                tmp = ktime_set((long)nsec, rem);
 280        }
 281
 282        return ktime_add(kt, tmp);
 283}
 284
 285EXPORT_SYMBOL_GPL(ktime_add_ns);
 286
 287/**
 288 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
 289 * @kt:         minuend
 290 * @nsec:       the scalar nsec value to subtract
 291 *
 292 * Returns the subtraction of @nsec from @kt in ktime_t format
 293 */
 294ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
 295{
 296        ktime_t tmp;
 297
 298        if (likely(nsec < NSEC_PER_SEC)) {
 299                tmp.tv64 = nsec;
 300        } else {
 301                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 302
 303                tmp = ktime_set((long)nsec, rem);
 304        }
 305
 306        return ktime_sub(kt, tmp);
 307}
 308
 309EXPORT_SYMBOL_GPL(ktime_sub_ns);
 310# endif /* !CONFIG_KTIME_SCALAR */
 311
 312/*
 313 * Divide a ktime value by a nanosecond value
 314 */
 315u64 ktime_divns(const ktime_t kt, s64 div)
 316{
 317        u64 dclc;
 318        int sft = 0;
 319
 320        dclc = ktime_to_ns(kt);
 321        /* Make sure the divisor is less than 2^32: */
 322        while (div >> 32) {
 323                sft++;
 324                div >>= 1;
 325        }
 326        dclc >>= sft;
 327        do_div(dclc, (unsigned long) div);
 328
 329        return dclc;
 330}
 331#endif /* BITS_PER_LONG >= 64 */
 332
 333/*
 334 * Add two ktime values and do a safety check for overflow:
 335 */
 336ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
 337{
 338        ktime_t res = ktime_add(lhs, rhs);
 339
 340        /*
 341         * We use KTIME_SEC_MAX here, the maximum timeout which we can
 342         * return to user space in a timespec:
 343         */
 344        if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
 345                res = ktime_set(KTIME_SEC_MAX, 0);
 346
 347        return res;
 348}
 349
 350EXPORT_SYMBOL_GPL(ktime_add_safe);
 351
 352#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 353
 354static struct debug_obj_descr hrtimer_debug_descr;
 355
 356static void *hrtimer_debug_hint(void *addr)
 357{
 358        return ((struct hrtimer *) addr)->function;
 359}
 360
 361/*
 362 * fixup_init is called when:
 363 * - an active object is initialized
 364 */
 365static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
 366{
 367        struct hrtimer *timer = addr;
 368
 369        switch (state) {
 370        case ODEBUG_STATE_ACTIVE:
 371                hrtimer_cancel(timer);
 372                debug_object_init(timer, &hrtimer_debug_descr);
 373                return 1;
 374        default:
 375                return 0;
 376        }
 377}
 378
 379/*
 380 * fixup_activate is called when:
 381 * - an active object is activated
 382 * - an unknown object is activated (might be a statically initialized object)
 383 */
 384static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
 385{
 386        switch (state) {
 387
 388        case ODEBUG_STATE_NOTAVAILABLE:
 389                WARN_ON_ONCE(1);
 390                return 0;
 391
 392        case ODEBUG_STATE_ACTIVE:
 393                WARN_ON(1);
 394
 395        default:
 396                return 0;
 397        }
 398}
 399
 400/*
 401 * fixup_free is called when:
 402 * - an active object is freed
 403 */
 404static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
 405{
 406        struct hrtimer *timer = addr;
 407
 408        switch (state) {
 409        case ODEBUG_STATE_ACTIVE:
 410                hrtimer_cancel(timer);
 411                debug_object_free(timer, &hrtimer_debug_descr);
 412                return 1;
 413        default:
 414                return 0;
 415        }
 416}
 417
 418static struct debug_obj_descr hrtimer_debug_descr = {
 419        .name           = "hrtimer",
 420        .debug_hint     = hrtimer_debug_hint,
 421        .fixup_init     = hrtimer_fixup_init,
 422        .fixup_activate = hrtimer_fixup_activate,
 423        .fixup_free     = hrtimer_fixup_free,
 424};
 425
 426static inline void debug_hrtimer_init(struct hrtimer *timer)
 427{
 428        debug_object_init(timer, &hrtimer_debug_descr);
 429}
 430
 431static inline void debug_hrtimer_activate(struct hrtimer *timer)
 432{
 433        debug_object_activate(timer, &hrtimer_debug_descr);
 434}
 435
 436static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
 437{
 438        debug_object_deactivate(timer, &hrtimer_debug_descr);
 439}
 440
 441static inline void debug_hrtimer_free(struct hrtimer *timer)
 442{
 443        debug_object_free(timer, &hrtimer_debug_descr);
 444}
 445
 446static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 447                           enum hrtimer_mode mode);
 448
 449void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
 450                           enum hrtimer_mode mode)
 451{
 452        debug_object_init_on_stack(timer, &hrtimer_debug_descr);
 453        __hrtimer_init(timer, clock_id, mode);
 454}
 455EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 456
 457void destroy_hrtimer_on_stack(struct hrtimer *timer)
 458{
 459        debug_object_free(timer, &hrtimer_debug_descr);
 460}
 461
 462#else
 463static inline void debug_hrtimer_init(struct hrtimer *timer) { }
 464static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
 465static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 466#endif
 467
 468static inline void
 469debug_init(struct hrtimer *timer, clockid_t clockid,
 470           enum hrtimer_mode mode)
 471{
 472        debug_hrtimer_init(timer);
 473        trace_hrtimer_init(timer, clockid, mode);
 474}
 475
 476static inline void debug_activate(struct hrtimer *timer)
 477{
 478        debug_hrtimer_activate(timer);
 479        trace_hrtimer_start(timer);
 480}
 481
 482static inline void debug_deactivate(struct hrtimer *timer)
 483{
 484        debug_hrtimer_deactivate(timer);
 485        trace_hrtimer_cancel(timer);
 486}
 487
 488/* High resolution timer related functions */
 489#ifdef CONFIG_HIGH_RES_TIMERS
 490
 491/*
 492 * High resolution timer enabled ?
 493 */
 494static int hrtimer_hres_enabled __read_mostly  = 1;
 495
 496/*
 497 * Enable / Disable high resolution mode
 498 */
 499static int __init setup_hrtimer_hres(char *str)
 500{
 501        if (!strcmp(str, "off"))
 502                hrtimer_hres_enabled = 0;
 503        else if (!strcmp(str, "on"))
 504                hrtimer_hres_enabled = 1;
 505        else
 506                return 0;
 507        return 1;
 508}
 509
 510__setup("highres=", setup_hrtimer_hres);
 511
 512/*
 513 * hrtimer_high_res_enabled - query, if the highres mode is enabled
 514 */
 515static inline int hrtimer_is_hres_enabled(void)
 516{
 517        return hrtimer_hres_enabled;
 518}
 519
 520/*
 521 * Is the high resolution mode active ?
 522 */
 523static inline int hrtimer_hres_active(void)
 524{
 525        return __this_cpu_read(hrtimer_bases.hres_active);
 526}
 527
 528/*
 529 * Reprogram the event source with checking both queues for the
 530 * next event
 531 * Called with interrupts disabled and base->lock held
 532 */
 533static void
 534hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 535{
 536        int i;
 537        struct hrtimer_clock_base *base = cpu_base->clock_base;
 538        ktime_t expires, expires_next;
 539
 540        expires_next.tv64 = KTIME_MAX;
 541
 542        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
 543                struct hrtimer *timer;
 544                struct timerqueue_node *next;
 545
 546                next = timerqueue_getnext(&base->active);
 547                if (!next)
 548                        continue;
 549                timer = container_of(next, struct hrtimer, node);
 550
 551                expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 552                /*
 553                 * clock_was_set() has changed base->offset so the
 554                 * result might be negative. Fix it up to prevent a
 555                 * false positive in clockevents_program_event()
 556                 */
 557                if (expires.tv64 < 0)
 558                        expires.tv64 = 0;
 559                if (expires.tv64 < expires_next.tv64)
 560                        expires_next = expires;
 561        }
 562
 563        if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
 564                return;
 565
 566        cpu_base->expires_next.tv64 = expires_next.tv64;
 567
 568        if (cpu_base->expires_next.tv64 != KTIME_MAX)
 569                tick_program_event(cpu_base->expires_next, 1);
 570}
 571
 572/*
 573 * Shared reprogramming for clock_realtime and clock_monotonic
 574 *
 575 * When a timer is enqueued and expires earlier than the already enqueued
 576 * timers, we have to check, whether it expires earlier than the timer for
 577 * which the clock event device was armed.
 578 *
 579 * Called with interrupts disabled and base->cpu_base.lock held
 580 */
 581static int hrtimer_reprogram(struct hrtimer *timer,
 582                             struct hrtimer_clock_base *base)
 583{
 584        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
 585        ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 586        int res;
 587
 588        WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
 589
 590        /*
 591         * When the callback is running, we do not reprogram the clock event
 592         * device. The timer callback is either running on a different CPU or
 593         * the callback is executed in the hrtimer_interrupt context. The
 594         * reprogramming is handled either by the softirq, which called the
 595         * callback or at the end of the hrtimer_interrupt.
 596         */
 597        if (hrtimer_callback_running(timer))
 598                return 0;
 599
 600        /*
 601         * CLOCK_REALTIME timer might be requested with an absolute
 602         * expiry time which is less than base->offset. Nothing wrong
 603         * about that, just avoid to call into the tick code, which
 604         * has now objections against negative expiry values.
 605         */
 606        if (expires.tv64 < 0)
 607                return -ETIME;
 608
 609        if (expires.tv64 >= cpu_base->expires_next.tv64)
 610                return 0;
 611
 612        /*
 613         * If a hang was detected in the last timer interrupt then we
 614         * do not schedule a timer which is earlier than the expiry
 615         * which we enforced in the hang detection. We want the system
 616         * to make progress.
 617         */
 618        if (cpu_base->hang_detected)
 619                return 0;
 620
 621        /*
 622         * Clockevents returns -ETIME, when the event was in the past.
 623         */
 624        res = tick_program_event(expires, 0);
 625        if (!IS_ERR_VALUE(res))
 626                cpu_base->expires_next = expires;
 627        return res;
 628}
 629
 630/*
 631 * Initialize the high resolution related parts of cpu_base
 632 */
 633static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
 634{
 635        base->expires_next.tv64 = KTIME_MAX;
 636        base->hres_active = 0;
 637}
 638
 639/*
 640 * When High resolution timers are active, try to reprogram. Note, that in case
 641 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
 642 * check happens. The timer gets enqueued into the rbtree. The reprogramming
 643 * and expiry check is done in the hrtimer_interrupt or in the softirq.
 644 */
 645static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 646                                            struct hrtimer_clock_base *base)
 647{
 648        return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
 649}
 650
 651static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
 652{
 653        ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
 654        ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
 655
 656        return ktime_get_update_offsets(offs_real, offs_boot);
 657}
 658
 659/*
 660 * Retrigger next event is called after clock was set
 661 *
 662 * Called with interrupts disabled via on_each_cpu()
 663 */
 664static void retrigger_next_event(void *arg)
 665{
 666        struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
 667
 668        if (!hrtimer_hres_active())
 669                return;
 670
 671        raw_spin_lock(&base->lock);
 672        hrtimer_update_base(base);
 673        hrtimer_force_reprogram(base, 0);
 674        raw_spin_unlock(&base->lock);
 675}
 676
 677/*
 678 * Switch to high resolution mode
 679 */
 680static int hrtimer_switch_to_hres(void)
 681{
 682        int i, cpu = smp_processor_id();
 683        struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
 684        unsigned long flags;
 685
 686        if (base->hres_active)
 687                return 1;
 688
 689        local_irq_save(flags);
 690
 691        if (tick_init_highres()) {
 692                local_irq_restore(flags);
 693                printk(KERN_WARNING "Could not switch to high resolution "
 694                                    "mode on CPU %d\n", cpu);
 695                return 0;
 696        }
 697        base->hres_active = 1;
 698        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
 699                base->clock_base[i].resolution = KTIME_HIGH_RES;
 700
 701        tick_setup_sched_timer();
 702        /* "Retrigger" the interrupt to get things going */
 703        retrigger_next_event(NULL);
 704        local_irq_restore(flags);
 705        return 1;
 706}
 707
 708/*
 709 * Called from timekeeping code to reprogramm the hrtimer interrupt
 710 * device. If called from the timer interrupt context we defer it to
 711 * softirq context.
 712 */
 713void clock_was_set_delayed(void)
 714{
 715        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
 716
 717        cpu_base->clock_was_set = 1;
 718        __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
 719}
 720
 721#else
 722
 723static inline int hrtimer_hres_active(void) { return 0; }
 724static inline int hrtimer_is_hres_enabled(void) { return 0; }
 725static inline int hrtimer_switch_to_hres(void) { return 0; }
 726static inline void
 727hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
 728static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 729                                            struct hrtimer_clock_base *base)
 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)) {
 990                if (wakeup) {
 991                        /*
 992                         * We need to drop cpu_base->lock to avoid a
 993                         * lock ordering issue vs. rq->lock.
 994                         */
 995                        raw_spin_unlock(&new_base->cpu_base->lock);
 996                        raise_softirq_irqoff(HRTIMER_SOFTIRQ);
 997                        local_irq_restore(flags);
 998                        return ret;
 999                } else {
1000                        __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1001                }
1002        }
1003
1004        unlock_hrtimer_base(timer, &flags);
1005
1006        return ret;
1007}
1008
1009/**
1010 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1011 * @timer:      the timer to be added
1012 * @tim:        expiry time
1013 * @delta_ns:   "slack" range for the timer
1014 * @mode:       expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1015 *
1016 * Returns:
1017 *  0 on success
1018 *  1 when the timer was active
1019 */
1020int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1021                unsigned long delta_ns, const enum hrtimer_mode mode)
1022{
1023        return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1024}
1025EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1026
1027/**
1028 * hrtimer_start - (re)start an hrtimer on the current CPU
1029 * @timer:      the timer to be added
1030 * @tim:        expiry time
1031 * @mode:       expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
1032 *
1033 * Returns:
1034 *  0 on success
1035 *  1 when the timer was active
1036 */
1037int
1038hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1039{
1040        return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1041}
1042EXPORT_SYMBOL_GPL(hrtimer_start);
1043
1044
1045/**
1046 * hrtimer_try_to_cancel - try to deactivate a timer
1047 * @timer:      hrtimer to stop
1048 *
1049 * Returns:
1050 *  0 when the timer was not active
1051 *  1 when the timer was active
1052 * -1 when the timer is currently excuting the callback function and
1053 *    cannot be stopped
1054 */
1055int hrtimer_try_to_cancel(struct hrtimer *timer)
1056{
1057        struct hrtimer_clock_base *base;
1058        unsigned long flags;
1059        int ret = -1;
1060
1061        base = lock_hrtimer_base(timer, &flags);
1062
1063        if (!hrtimer_callback_running(timer))
1064                ret = remove_hrtimer(timer, base);
1065
1066        unlock_hrtimer_base(timer, &flags);
1067
1068        return ret;
1069
1070}
1071EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1072
1073/**
1074 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1075 * @timer:      the timer to be cancelled
1076 *
1077 * Returns:
1078 *  0 when the timer was not active
1079 *  1 when the timer was active
1080 */
1081int hrtimer_cancel(struct hrtimer *timer)
1082{
1083        for (;;) {
1084                int ret = hrtimer_try_to_cancel(timer);
1085
1086                if (ret >= 0)
1087                        return ret;
1088                cpu_relax();
1089        }
1090}
1091EXPORT_SYMBOL_GPL(hrtimer_cancel);
1092
1093/**
1094 * hrtimer_get_remaining - get remaining time for the timer
1095 * @timer:      the timer to read
1096 */
1097ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1098{
1099        unsigned long flags;
1100        ktime_t rem;
1101
1102        lock_hrtimer_base(timer, &flags);
1103        rem = hrtimer_expires_remaining(timer);
1104        unlock_hrtimer_base(timer, &flags);
1105
1106        return rem;
1107}
1108EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1109
1110#ifdef CONFIG_NO_HZ
1111/**
1112 * hrtimer_get_next_event - get the time until next expiry event
1113 *
1114 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1115 * is pending.
1116 */
1117ktime_t hrtimer_get_next_event(void)
1118{
1119        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1120        struct hrtimer_clock_base *base = cpu_base->clock_base;
1121        ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1122        unsigned long flags;
1123        int i;
1124
1125        raw_spin_lock_irqsave(&cpu_base->lock, flags);
1126
1127        if (!hrtimer_hres_active()) {
1128                for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1129                        struct hrtimer *timer;
1130                        struct timerqueue_node *next;
1131
1132                        next = timerqueue_getnext(&base->active);
1133                        if (!next)
1134                                continue;
1135
1136                        timer = container_of(next, struct hrtimer, node);
1137                        delta.tv64 = hrtimer_get_expires_tv64(timer);
1138                        delta = ktime_sub(delta, base->get_time());
1139                        if (delta.tv64 < mindelta.tv64)
1140                                mindelta.tv64 = delta.tv64;
1141                }
1142        }
1143
1144        raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1145
1146        if (mindelta.tv64 < 0)
1147                mindelta.tv64 = 0;
1148        return mindelta;
1149}
1150#endif
1151
1152static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1153                           enum hrtimer_mode mode)
1154{
1155        struct hrtimer_cpu_base *cpu_base;
1156        int base;
1157
1158        memset(timer, 0, sizeof(struct hrtimer));
1159
1160        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1161
1162        if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1163                clock_id = CLOCK_MONOTONIC;
1164
1165        base = hrtimer_clockid_to_base(clock_id);
1166        timer->base = &cpu_base->clock_base[base];
1167        timerqueue_init(&timer->node);
1168
1169#ifdef CONFIG_TIMER_STATS
1170        timer->start_site = NULL;
1171        timer->start_pid = -1;
1172        memset(timer->start_comm, 0, TASK_COMM_LEN);
1173#endif
1174}
1175
1176/**
1177 * hrtimer_init - initialize a timer to the given clock
1178 * @timer:      the timer to be initialized
1179 * @clock_id:   the clock to be used
1180 * @mode:       timer mode abs/rel
1181 */
1182void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1183                  enum hrtimer_mode mode)
1184{
1185        debug_init(timer, clock_id, mode);
1186        __hrtimer_init(timer, clock_id, mode);
1187}
1188EXPORT_SYMBOL_GPL(hrtimer_init);
1189
1190/**
1191 * hrtimer_get_res - get the timer resolution for a clock
1192 * @which_clock: which clock to query
1193 * @tp:          pointer to timespec variable to store the resolution
1194 *
1195 * Store the resolution of the clock selected by @which_clock in the
1196 * variable pointed to by @tp.
1197 */
1198int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1199{
1200        struct hrtimer_cpu_base *cpu_base;
1201        int base = hrtimer_clockid_to_base(which_clock);
1202
1203        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1204        *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1205
1206        return 0;
1207}
1208EXPORT_SYMBOL_GPL(hrtimer_get_res);
1209
1210static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1211{
1212        struct hrtimer_clock_base *base = timer->base;
1213        struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1214        enum hrtimer_restart (*fn)(struct hrtimer *);
1215        int restart;
1216
1217        WARN_ON(!irqs_disabled());
1218
1219        debug_deactivate(timer);
1220        __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1221        timer_stats_account_hrtimer(timer);
1222        fn = timer->function;
1223
1224        /*
1225         * Because we run timers from hardirq context, there is no chance
1226         * they get migrated to another cpu, therefore its safe to unlock
1227         * the timer base.
1228         */
1229        raw_spin_unlock(&cpu_base->lock);
1230        trace_hrtimer_expire_entry(timer, now);
1231        restart = fn(timer);
1232        trace_hrtimer_expire_exit(timer);
1233        raw_spin_lock(&cpu_base->lock);
1234
1235        /*
1236         * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1237         * we do not reprogramm the event hardware. Happens either in
1238         * hrtimer_start_range_ns() or in hrtimer_interrupt()
1239         */
1240        if (restart != HRTIMER_NORESTART) {
1241                BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1242                enqueue_hrtimer(timer, base);
1243        }
1244
1245        WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1246
1247        timer->state &= ~HRTIMER_STATE_CALLBACK;
1248}
1249
1250#ifdef CONFIG_HIGH_RES_TIMERS
1251
1252/*
1253 * High resolution timer interrupt
1254 * Called with interrupts disabled
1255 */
1256void hrtimer_interrupt(struct clock_event_device *dev)
1257{
1258        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1259        ktime_t expires_next, now, entry_time, delta;
1260        int i, retries = 0;
1261
1262        BUG_ON(!cpu_base->hres_active);
1263        cpu_base->nr_events++;
1264        dev->next_event.tv64 = KTIME_MAX;
1265
1266        raw_spin_lock(&cpu_base->lock);
1267        entry_time = now = hrtimer_update_base(cpu_base);
1268retry:
1269        expires_next.tv64 = KTIME_MAX;
1270        /*
1271         * We set expires_next to KTIME_MAX here with cpu_base->lock
1272         * held to prevent that a timer is enqueued in our queue via
1273         * the migration code. This does not affect enqueueing of
1274         * timers which run their callback and need to be requeued on
1275         * this CPU.
1276         */
1277        cpu_base->expires_next.tv64 = KTIME_MAX;
1278
1279        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1280                struct hrtimer_clock_base *base;
1281                struct timerqueue_node *node;
1282                ktime_t basenow;
1283
1284                if (!(cpu_base->active_bases & (1 << i)))
1285                        continue;
1286
1287                base = cpu_base->clock_base + i;
1288                basenow = ktime_add(now, base->offset);
1289
1290                while ((node = timerqueue_getnext(&base->active))) {
1291                        struct hrtimer *timer;
1292
1293                        timer = container_of(node, struct hrtimer, node);
1294
1295                        /*
1296                         * The immediate goal for using the softexpires is
1297                         * minimizing wakeups, not running timers at the
1298                         * earliest interrupt after their soft expiration.
1299                         * This allows us to avoid using a Priority Search
1300                         * Tree, which can answer a stabbing querry for
1301                         * overlapping intervals and instead use the simple
1302                         * BST we already have.
1303                         * We don't add extra wakeups by delaying timers that
1304                         * are right-of a not yet expired timer, because that
1305                         * timer will have to trigger a wakeup anyway.
1306                         */
1307
1308                        if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1309                                ktime_t expires;
1310
1311                                expires = ktime_sub(hrtimer_get_expires(timer),
1312                                                    base->offset);
1313                                if (expires.tv64 < expires_next.tv64)
1314                                        expires_next = expires;
1315                                break;
1316                        }
1317
1318                        __run_hrtimer(timer, &basenow);
1319                }
1320        }
1321
1322        /*
1323         * Store the new expiry value so the migration code can verify
1324         * against it.
1325         */
1326        cpu_base->expires_next = expires_next;
1327        raw_spin_unlock(&cpu_base->lock);
1328
1329        /* Reprogramming necessary ? */
1330        if (expires_next.tv64 == KTIME_MAX ||
1331            !tick_program_event(expires_next, 0)) {
1332                cpu_base->hang_detected = 0;
1333                return;
1334        }
1335
1336        /*
1337         * The next timer was already expired due to:
1338         * - tracing
1339         * - long lasting callbacks
1340         * - being scheduled away when running in a VM
1341         *
1342         * We need to prevent that we loop forever in the hrtimer
1343         * interrupt routine. We give it 3 attempts to avoid
1344         * overreacting on some spurious event.
1345         *
1346         * Acquire base lock for updating the offsets and retrieving
1347         * the current time.
1348         */
1349        raw_spin_lock(&cpu_base->lock);
1350        now = hrtimer_update_base(cpu_base);
1351        cpu_base->nr_retries++;
1352        if (++retries < 3)
1353                goto retry;
1354        /*
1355         * Give the system a chance to do something else than looping
1356         * here. We stored the entry time, so we know exactly how long
1357         * we spent here. We schedule the next event this amount of
1358         * time away.
1359         */
1360        cpu_base->nr_hangs++;
1361        cpu_base->hang_detected = 1;
1362        raw_spin_unlock(&cpu_base->lock);
1363        delta = ktime_sub(now, entry_time);
1364        if (delta.tv64 > cpu_base->max_hang_time.tv64)
1365                cpu_base->max_hang_time = delta;
1366        /*
1367         * Limit it to a sensible value as we enforce a longer
1368         * delay. Give the CPU at least 100ms to catch up.
1369         */
1370        if (delta.tv64 > 100 * NSEC_PER_MSEC)
1371                expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1372        else
1373                expires_next = ktime_add(now, delta);
1374        tick_program_event(expires_next, 1);
1375        printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1376                    ktime_to_ns(delta));
1377}
1378
1379/*
1380 * local version of hrtimer_peek_ahead_timers() called with interrupts
1381 * disabled.
1382 */
1383static void __hrtimer_peek_ahead_timers(void)
1384{
1385        struct tick_device *td;
1386
1387        if (!hrtimer_hres_active())
1388                return;
1389
1390        td = &__get_cpu_var(tick_cpu_device);
1391        if (td && td->evtdev)
1392                hrtimer_interrupt(td->evtdev);
1393}
1394
1395/**
1396 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1397 *
1398 * hrtimer_peek_ahead_timers will peek at the timer queue of
1399 * the current cpu and check if there are any timers for which
1400 * the soft expires time has passed. If any such timers exist,
1401 * they are run immediately and then removed from the timer queue.
1402 *
1403 */
1404void hrtimer_peek_ahead_timers(void)
1405{
1406        unsigned long flags;
1407
1408        local_irq_save(flags);
1409        __hrtimer_peek_ahead_timers();
1410        local_irq_restore(flags);
1411}
1412
1413static void run_hrtimer_softirq(struct softirq_action *h)
1414{
1415        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1416
1417        if (cpu_base->clock_was_set) {
1418                cpu_base->clock_was_set = 0;
1419                clock_was_set();
1420        }
1421
1422        hrtimer_peek_ahead_timers();
1423}
1424
1425#else /* CONFIG_HIGH_RES_TIMERS */
1426
1427static inline void __hrtimer_peek_ahead_timers(void) { }
1428
1429#endif  /* !CONFIG_HIGH_RES_TIMERS */
1430
1431/*
1432 * Called from timer softirq every jiffy, expire hrtimers:
1433 *
1434 * For HRT its the fall back code to run the softirq in the timer
1435 * softirq context in case the hrtimer initialization failed or has
1436 * not been done yet.
1437 */
1438void hrtimer_run_pending(void)
1439{
1440        if (hrtimer_hres_active())
1441                return;
1442
1443        /*
1444         * This _is_ ugly: We have to check in the softirq context,
1445         * whether we can switch to highres and / or nohz mode. The
1446         * clocksource switch happens in the timer interrupt with
1447         * xtime_lock held. Notification from there only sets the
1448         * check bit in the tick_oneshot code, otherwise we might
1449         * deadlock vs. xtime_lock.
1450         */
1451        if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1452                hrtimer_switch_to_hres();
1453}
1454
1455/*
1456 * Called from hardirq context every jiffy
1457 */
1458void hrtimer_run_queues(void)
1459{
1460        struct timerqueue_node *node;
1461        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1462        struct hrtimer_clock_base *base;
1463        int index, gettime = 1;
1464
1465        if (hrtimer_hres_active())
1466                return;
1467
1468        for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1469                base = &cpu_base->clock_base[index];
1470                if (!timerqueue_getnext(&base->active))
1471                        continue;
1472
1473                if (gettime) {
1474                        hrtimer_get_softirq_time(cpu_base);
1475                        gettime = 0;
1476                }
1477
1478                raw_spin_lock(&cpu_base->lock);
1479
1480                while ((node = timerqueue_getnext(&base->active))) {
1481                        struct hrtimer *timer;
1482
1483                        timer = container_of(node, struct hrtimer, node);
1484                        if (base->softirq_time.tv64 <=
1485                                        hrtimer_get_expires_tv64(timer))
1486                                break;
1487
1488                        __run_hrtimer(timer, &base->softirq_time);
1489                }
1490                raw_spin_unlock(&cpu_base->lock);
1491        }
1492}
1493
1494/*
1495 * Sleep related functions:
1496 */
1497static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1498{
1499        struct hrtimer_sleeper *t =
1500                container_of(timer, struct hrtimer_sleeper, timer);
1501        struct task_struct *task = t->task;
1502
1503        t->task = NULL;
1504        if (task)
1505                wake_up_process(task);
1506
1507        return HRTIMER_NORESTART;
1508}
1509
1510void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1511{
1512        sl->timer.function = hrtimer_wakeup;
1513        sl->task = task;
1514}
1515EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1516
1517static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1518{
1519        hrtimer_init_sleeper(t, current);
1520
1521        do {
1522                set_current_state(TASK_INTERRUPTIBLE);
1523                hrtimer_start_expires(&t->timer, mode);
1524                if (!hrtimer_active(&t->timer))
1525                        t->task = NULL;
1526
1527                if (likely(t->task))
1528                        schedule();
1529
1530                hrtimer_cancel(&t->timer);
1531                mode = HRTIMER_MODE_ABS;
1532
1533        } while (t->task && !signal_pending(current));
1534
1535        __set_current_state(TASK_RUNNING);
1536
1537        return t->task == NULL;
1538}
1539
1540static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1541{
1542        struct timespec rmt;
1543        ktime_t rem;
1544
1545        rem = hrtimer_expires_remaining(timer);
1546        if (rem.tv64 <= 0)
1547                return 0;
1548        rmt = ktime_to_timespec(rem);
1549
1550        if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1551                return -EFAULT;
1552
1553        return 1;
1554}
1555
1556long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1557{
1558        struct hrtimer_sleeper t;
1559        struct timespec __user  *rmtp;
1560        int ret = 0;
1561
1562        hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1563                                HRTIMER_MODE_ABS);
1564        hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1565
1566        if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1567                goto out;
1568
1569        rmtp = restart->nanosleep.rmtp;
1570        if (rmtp) {
1571                ret = update_rmtp(&t.timer, rmtp);
1572                if (ret <= 0)
1573                        goto out;
1574        }
1575
1576        /* The other values in restart are already filled in */
1577        ret = -ERESTART_RESTARTBLOCK;
1578out:
1579        destroy_hrtimer_on_stack(&t.timer);
1580        return ret;
1581}
1582
1583long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1584                       const enum hrtimer_mode mode, const clockid_t clockid)
1585{
1586        struct restart_block *restart;
1587        struct hrtimer_sleeper t;
1588        int ret = 0;
1589        unsigned long slack;
1590
1591        slack = current->timer_slack_ns;
1592        if (rt_task(current))
1593                slack = 0;
1594
1595        hrtimer_init_on_stack(&t.timer, clockid, mode);
1596        hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1597        if (do_nanosleep(&t, mode))
1598                goto out;
1599
1600        /* Absolute timers do not update the rmtp value and restart: */
1601        if (mode == HRTIMER_MODE_ABS) {
1602                ret = -ERESTARTNOHAND;
1603                goto out;
1604        }
1605
1606        if (rmtp) {
1607                ret = update_rmtp(&t.timer, rmtp);
1608                if (ret <= 0)
1609                        goto out;
1610        }
1611
1612        restart = &current_thread_info()->restart_block;
1613        restart->fn = hrtimer_nanosleep_restart;
1614        restart->nanosleep.clockid = t.timer.base->clockid;
1615        restart->nanosleep.rmtp = rmtp;
1616        restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1617
1618        ret = -ERESTART_RESTARTBLOCK;
1619out:
1620        destroy_hrtimer_on_stack(&t.timer);
1621        return ret;
1622}
1623
1624SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1625                struct timespec __user *, rmtp)
1626{
1627        struct timespec tu;
1628
1629        if (copy_from_user(&tu, rqtp, sizeof(tu)))
1630                return -EFAULT;
1631
1632        if (!timespec_valid(&tu))
1633                return -EINVAL;
1634
1635        return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1636}
1637
1638/*
1639 * Functions related to boot-time initialization:
1640 */
1641static void __cpuinit init_hrtimers_cpu(int cpu)
1642{
1643        struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1644        int i;
1645
1646        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1647                cpu_base->clock_base[i].cpu_base = cpu_base;
1648                timerqueue_init_head(&cpu_base->clock_base[i].active);
1649        }
1650
1651        hrtimer_init_hres(cpu_base);
1652}
1653
1654#ifdef CONFIG_HOTPLUG_CPU
1655
1656static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1657                                struct hrtimer_clock_base *new_base)
1658{
1659        struct hrtimer *timer;
1660        struct timerqueue_node *node;
1661
1662        while ((node = timerqueue_getnext(&old_base->active))) {
1663                timer = container_of(node, struct hrtimer, node);
1664                BUG_ON(hrtimer_callback_running(timer));
1665                debug_deactivate(timer);
1666
1667                /*
1668                 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1669                 * timer could be seen as !active and just vanish away
1670                 * under us on another CPU
1671                 */
1672                __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1673                timer->base = new_base;
1674                /*
1675                 * Enqueue the timers on the new cpu. This does not
1676                 * reprogram the event device in case the timer
1677                 * expires before the earliest on this CPU, but we run
1678                 * hrtimer_interrupt after we migrated everything to
1679                 * sort out already expired timers and reprogram the
1680                 * event device.
1681                 */
1682                enqueue_hrtimer(timer, new_base);
1683
1684                /* Clear the migration state bit */
1685                timer->state &= ~HRTIMER_STATE_MIGRATE;
1686        }
1687}
1688
1689static void migrate_hrtimers(int scpu)
1690{
1691        struct hrtimer_cpu_base *old_base, *new_base;
1692        int i;
1693
1694        BUG_ON(cpu_online(scpu));
1695        tick_cancel_sched_timer(scpu);
1696
1697        local_irq_disable();
1698        old_base = &per_cpu(hrtimer_bases, scpu);
1699        new_base = &__get_cpu_var(hrtimer_bases);
1700        /*
1701         * The caller is globally serialized and nobody else
1702         * takes two locks at once, deadlock is not possible.
1703         */
1704        raw_spin_lock(&new_base->lock);
1705        raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1706
1707        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1708                migrate_hrtimer_list(&old_base->clock_base[i],
1709                                     &new_base->clock_base[i]);
1710        }
1711
1712        raw_spin_unlock(&old_base->lock);
1713        raw_spin_unlock(&new_base->lock);
1714
1715        /* Check, if we got expired work to do */
1716        __hrtimer_peek_ahead_timers();
1717        local_irq_enable();
1718}
1719
1720#endif /* CONFIG_HOTPLUG_CPU */
1721
1722static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1723                                        unsigned long action, void *hcpu)
1724{
1725        int scpu = (long)hcpu;
1726
1727        switch (action) {
1728
1729        case CPU_UP_PREPARE:
1730        case CPU_UP_PREPARE_FROZEN:
1731                init_hrtimers_cpu(scpu);
1732                break;
1733
1734#ifdef CONFIG_HOTPLUG_CPU
1735        case CPU_DYING:
1736        case CPU_DYING_FROZEN:
1737                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1738                break;
1739        case CPU_DEAD:
1740        case CPU_DEAD_FROZEN:
1741        {
1742                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1743                migrate_hrtimers(scpu);
1744                break;
1745        }
1746#endif
1747
1748        default:
1749                break;
1750        }
1751
1752        return NOTIFY_OK;
1753}
1754
1755static struct notifier_block __cpuinitdata hrtimers_nb = {
1756        .notifier_call = hrtimer_cpu_notify,
1757};
1758
1759void __init hrtimers_init(void)
1760{
1761        hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1762                          (void *)(long)smp_processor_id());
1763        register_cpu_notifier(&hrtimers_nb);
1764#ifdef CONFIG_HIGH_RES_TIMERS
1765        open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1766#endif
1767}
1768
1769/**
1770 * schedule_hrtimeout_range_clock - sleep until timeout
1771 * @expires:    timeout value (ktime_t)
1772 * @delta:      slack in expires timeout (ktime_t)
1773 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1774 * @clock:      timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1775 */
1776int __sched
1777schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1778                               const enum hrtimer_mode mode, int clock)
1779{
1780        struct hrtimer_sleeper t;
1781
1782        /*
1783         * Optimize when a zero timeout value is given. It does not
1784         * matter whether this is an absolute or a relative time.
1785         */
1786        if (expires && !expires->tv64) {
1787                __set_current_state(TASK_RUNNING);
1788                return 0;
1789        }
1790
1791        /*
1792         * A NULL parameter means "infinite"
1793         */
1794        if (!expires) {
1795                schedule();
1796                __set_current_state(TASK_RUNNING);
1797                return -EINTR;
1798        }
1799
1800        hrtimer_init_on_stack(&t.timer, clock, mode);
1801        hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1802
1803        hrtimer_init_sleeper(&t, current);
1804
1805        hrtimer_start_expires(&t.timer, mode);
1806        if (!hrtimer_active(&t.timer))
1807                t.task = NULL;
1808
1809        if (likely(t.task))
1810                schedule();
1811
1812        hrtimer_cancel(&t.timer);
1813        destroy_hrtimer_on_stack(&t.timer);
1814
1815        __set_current_state(TASK_RUNNING);
1816
1817        return !t.task ? 0 : -EINTR;
1818}
1819
1820/**
1821 * schedule_hrtimeout_range - sleep until timeout
1822 * @expires:    timeout value (ktime_t)
1823 * @delta:      slack in expires timeout (ktime_t)
1824 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1825 *
1826 * Make the current task sleep until the given expiry time has
1827 * elapsed. The routine will return immediately unless
1828 * the current task state has been set (see set_current_state()).
1829 *
1830 * The @delta argument gives the kernel the freedom to schedule the
1831 * actual wakeup to a time that is both power and performance friendly.
1832 * The kernel give the normal best effort behavior for "@expires+@delta",
1833 * but may decide to fire the timer earlier, but no earlier than @expires.
1834 *
1835 * You can set the task state as follows -
1836 *
1837 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1838 * pass before the routine returns.
1839 *
1840 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1841 * delivered to the current task.
1842 *
1843 * The current task state is guaranteed to be TASK_RUNNING when this
1844 * routine returns.
1845 *
1846 * Returns 0 when the timer has expired otherwise -EINTR
1847 */
1848int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1849                                     const enum hrtimer_mode mode)
1850{
1851        return schedule_hrtimeout_range_clock(expires, delta, mode,
1852                                              CLOCK_MONOTONIC);
1853}
1854EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1855
1856/**
1857 * schedule_hrtimeout - sleep until timeout
1858 * @expires:    timeout value (ktime_t)
1859 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1860 *
1861 * Make the current task sleep until the given expiry time has
1862 * elapsed. The routine will return immediately unless
1863 * the current task state has been set (see set_current_state()).
1864 *
1865 * You can set the task state as follows -
1866 *
1867 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1868 * pass before the routine returns.
1869 *
1870 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1871 * delivered to the current task.
1872 *
1873 * The current task state is guaranteed to be TASK_RUNNING when this
1874 * routine returns.
1875 *
1876 * Returns 0 when the timer has expired otherwise -EINTR
1877 */
1878int __sched schedule_hrtimeout(ktime_t *expires,
1879                               const enum hrtimer_mode mode)
1880{
1881        return schedule_hrtimeout_range(expires, 0, mode);
1882}
1883EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1884
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.