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                        .index = HRTIMER_BASE_TAI,
  89                        .clockid = CLOCK_TAI,
  90                        .get_time = &ktime_get_clocktai,
  91                        .resolution = KTIME_LOW_RES,
  92                },
  93        }
  94};
  95
  96static const int hrtimer_clock_to_base_table[MAX_CLOCKS] = {
  97        [CLOCK_REALTIME]        = HRTIMER_BASE_REALTIME,
  98        [CLOCK_MONOTONIC]       = HRTIMER_BASE_MONOTONIC,
  99        [CLOCK_BOOTTIME]        = HRTIMER_BASE_BOOTTIME,
 100        [CLOCK_TAI]             = HRTIMER_BASE_TAI,
 101};
 102
 103static inline int hrtimer_clockid_to_base(clockid_t clock_id)
 104{
 105        return hrtimer_clock_to_base_table[clock_id];
 106}
 107
 108
 109/*
 110 * Get the coarse grained time at the softirq based on xtime and
 111 * wall_to_monotonic.
 112 */
 113static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 114{
 115        ktime_t xtim, mono, boot;
 116        struct timespec xts, tom, slp;
 117        s32 tai_offset;
 118
 119        get_xtime_and_monotonic_and_sleep_offset(&xts, &tom, &slp);
 120        tai_offset = timekeeping_get_tai_offset();
 121
 122        xtim = timespec_to_ktime(xts);
 123        mono = ktime_add(xtim, timespec_to_ktime(tom));
 124        boot = ktime_add(mono, timespec_to_ktime(slp));
 125        base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim;
 126        base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono;
 127        base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot;
 128        base->clock_base[HRTIMER_BASE_TAI].softirq_time =
 129                                ktime_add(xtim, ktime_set(tai_offset, 0));
 130}
 131
 132/*
 133 * Functions and macros which are different for UP/SMP systems are kept in a
 134 * single place
 135 */
 136#ifdef CONFIG_SMP
 137
 138/*
 139 * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock
 140 * means that all timers which are tied to this base via timer->base are
 141 * locked, and the base itself is locked too.
 142 *
 143 * So __run_timers/migrate_timers can safely modify all timers which could
 144 * be found on the lists/queues.
 145 *
 146 * When the timer's base is locked, and the timer removed from list, it is
 147 * possible to set timer->base = NULL and drop the lock: the timer remains
 148 * locked.
 149 */
 150static
 151struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
 152                                             unsigned long *flags)
 153{
 154        struct hrtimer_clock_base *base;
 155
 156        for (;;) {
 157                base = timer->base;
 158                if (likely(base != NULL)) {
 159                        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 160                        if (likely(base == timer->base))
 161                                return base;
 162                        /* The timer has migrated to another CPU: */
 163                        raw_spin_unlock_irqrestore(&base->cpu_base->lock, *flags);
 164                }
 165                cpu_relax();
 166        }
 167}
 168
 169
 170/*
 171 * Get the preferred target CPU for NOHZ
 172 */
 173static int hrtimer_get_target(int this_cpu, int pinned)
 174{
 175#ifdef CONFIG_NO_HZ_COMMON
 176        if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu))
 177                return get_nohz_timer_target();
 178#endif
 179        return this_cpu;
 180}
 181
 182/*
 183 * With HIGHRES=y we do not migrate the timer when it is expiring
 184 * before the next event on the target cpu because we cannot reprogram
 185 * the target cpu hardware and we would cause it to fire late.
 186 *
 187 * Called with cpu_base->lock of target cpu held.
 188 */
 189static int
 190hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 191{
 192#ifdef CONFIG_HIGH_RES_TIMERS
 193        ktime_t expires;
 194
 195        if (!new_base->cpu_base->hres_active)
 196                return 0;
 197
 198        expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
 199        return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
 200#else
 201        return 0;
 202#endif
 203}
 204
 205/*
 206 * Switch the timer base to the current CPU when possible.
 207 */
 208static inline struct hrtimer_clock_base *
 209switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 210                    int pinned)
 211{
 212        struct hrtimer_clock_base *new_base;
 213        struct hrtimer_cpu_base *new_cpu_base;
 214        int this_cpu = smp_processor_id();
 215        int cpu = hrtimer_get_target(this_cpu, pinned);
 216        int basenum = base->index;
 217
 218again:
 219        new_cpu_base = &per_cpu(hrtimer_bases, cpu);
 220        new_base = &new_cpu_base->clock_base[basenum];
 221
 222        if (base != new_base) {
 223                /*
 224                 * We are trying to move timer to new_base.
 225                 * However we can't change timer's base while it is running,
 226                 * so we keep it on the same CPU. No hassle vs. reprogramming
 227                 * the event source in the high resolution case. The softirq
 228                 * code will take care of this when the timer function has
 229                 * completed. There is no conflict as we hold the lock until
 230                 * the timer is enqueued.
 231                 */
 232                if (unlikely(hrtimer_callback_running(timer)))
 233                        return base;
 234
 235                /* See the comment in lock_timer_base() */
 236                timer->base = NULL;
 237                raw_spin_unlock(&base->cpu_base->lock);
 238                raw_spin_lock(&new_base->cpu_base->lock);
 239
 240                if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
 241                        cpu = this_cpu;
 242                        raw_spin_unlock(&new_base->cpu_base->lock);
 243                        raw_spin_lock(&base->cpu_base->lock);
 244                        timer->base = base;
 245                        goto again;
 246                }
 247                timer->base = new_base;
 248        }
 249        return new_base;
 250}
 251
 252#else /* CONFIG_SMP */
 253
 254static inline struct hrtimer_clock_base *
 255lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 256{
 257        struct hrtimer_clock_base *base = timer->base;
 258
 259        raw_spin_lock_irqsave(&base->cpu_base->lock, *flags);
 260
 261        return base;
 262}
 263
 264# define switch_hrtimer_base(t, b, p)   (b)
 265
 266#endif  /* !CONFIG_SMP */
 267
 268/*
 269 * Functions for the union type storage format of ktime_t which are
 270 * too large for inlining:
 271 */
 272#if BITS_PER_LONG < 64
 273# ifndef CONFIG_KTIME_SCALAR
 274/**
 275 * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable
 276 * @kt:         addend
 277 * @nsec:       the scalar nsec value to add
 278 *
 279 * Returns the sum of kt and nsec in ktime_t format
 280 */
 281ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
 282{
 283        ktime_t tmp;
 284
 285        if (likely(nsec < NSEC_PER_SEC)) {
 286                tmp.tv64 = nsec;
 287        } else {
 288                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 289
 290                /* Make sure nsec fits into long */
 291                if (unlikely(nsec > KTIME_SEC_MAX))
 292                        return (ktime_t){ .tv64 = KTIME_MAX };
 293
 294                tmp = ktime_set((long)nsec, rem);
 295        }
 296
 297        return ktime_add(kt, tmp);
 298}
 299
 300EXPORT_SYMBOL_GPL(ktime_add_ns);
 301
 302/**
 303 * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
 304 * @kt:         minuend
 305 * @nsec:       the scalar nsec value to subtract
 306 *
 307 * Returns the subtraction of @nsec from @kt in ktime_t format
 308 */
 309ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
 310{
 311        ktime_t tmp;
 312
 313        if (likely(nsec < NSEC_PER_SEC)) {
 314                tmp.tv64 = nsec;
 315        } else {
 316                unsigned long rem = do_div(nsec, NSEC_PER_SEC);
 317
 318                tmp = ktime_set((long)nsec, rem);
 319        }
 320
 321        return ktime_sub(kt, tmp);
 322}
 323
 324EXPORT_SYMBOL_GPL(ktime_sub_ns);
 325# endif /* !CONFIG_KTIME_SCALAR */
 326
 327/*
 328 * Divide a ktime value by a nanosecond value
 329 */
 330u64 ktime_divns(const ktime_t kt, s64 div)
 331{
 332        u64 dclc;
 333        int sft = 0;
 334
 335        dclc = ktime_to_ns(kt);
 336        /* Make sure the divisor is less than 2^32: */
 337        while (div >> 32) {
 338                sft++;
 339                div >>= 1;
 340        }
 341        dclc >>= sft;
 342        do_div(dclc, (unsigned long) div);
 343
 344        return dclc;
 345}
 346#endif /* BITS_PER_LONG >= 64 */
 347
 348/*
 349 * Add two ktime values and do a safety check for overflow:
 350 */
 351ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
 352{
 353        ktime_t res = ktime_add(lhs, rhs);
 354
 355        /*
 356         * We use KTIME_SEC_MAX here, the maximum timeout which we can
 357         * return to user space in a timespec:
 358         */
 359        if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
 360                res = ktime_set(KTIME_SEC_MAX, 0);
 361
 362        return res;
 363}
 364
 365EXPORT_SYMBOL_GPL(ktime_add_safe);
 366
 367#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 368
 369static struct debug_obj_descr hrtimer_debug_descr;
 370
 371static void *hrtimer_debug_hint(void *addr)
 372{
 373        return ((struct hrtimer *) addr)->function;
 374}
 375
 376/*
 377 * fixup_init is called when:
 378 * - an active object is initialized
 379 */
 380static int hrtimer_fixup_init(void *addr, enum debug_obj_state state)
 381{
 382        struct hrtimer *timer = addr;
 383
 384        switch (state) {
 385        case ODEBUG_STATE_ACTIVE:
 386                hrtimer_cancel(timer);
 387                debug_object_init(timer, &hrtimer_debug_descr);
 388                return 1;
 389        default:
 390                return 0;
 391        }
 392}
 393
 394/*
 395 * fixup_activate is called when:
 396 * - an active object is activated
 397 * - an unknown object is activated (might be a statically initialized object)
 398 */
 399static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state)
 400{
 401        switch (state) {
 402
 403        case ODEBUG_STATE_NOTAVAILABLE:
 404                WARN_ON_ONCE(1);
 405                return 0;
 406
 407        case ODEBUG_STATE_ACTIVE:
 408                WARN_ON(1);
 409
 410        default:
 411                return 0;
 412        }
 413}
 414
 415/*
 416 * fixup_free is called when:
 417 * - an active object is freed
 418 */
 419static int hrtimer_fixup_free(void *addr, enum debug_obj_state state)
 420{
 421        struct hrtimer *timer = addr;
 422
 423        switch (state) {
 424        case ODEBUG_STATE_ACTIVE:
 425                hrtimer_cancel(timer);
 426                debug_object_free(timer, &hrtimer_debug_descr);
 427                return 1;
 428        default:
 429                return 0;
 430        }
 431}
 432
 433static struct debug_obj_descr hrtimer_debug_descr = {
 434        .name           = "hrtimer",
 435        .debug_hint     = hrtimer_debug_hint,
 436        .fixup_init     = hrtimer_fixup_init,
 437        .fixup_activate = hrtimer_fixup_activate,
 438        .fixup_free     = hrtimer_fixup_free,
 439};
 440
 441static inline void debug_hrtimer_init(struct hrtimer *timer)
 442{
 443        debug_object_init(timer, &hrtimer_debug_descr);
 444}
 445
 446static inline void debug_hrtimer_activate(struct hrtimer *timer)
 447{
 448        debug_object_activate(timer, &hrtimer_debug_descr);
 449}
 450
 451static inline void debug_hrtimer_deactivate(struct hrtimer *timer)
 452{
 453        debug_object_deactivate(timer, &hrtimer_debug_descr);
 454}
 455
 456static inline void debug_hrtimer_free(struct hrtimer *timer)
 457{
 458        debug_object_free(timer, &hrtimer_debug_descr);
 459}
 460
 461static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 462                           enum hrtimer_mode mode);
 463
 464void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
 465                           enum hrtimer_mode mode)
 466{
 467        debug_object_init_on_stack(timer, &hrtimer_debug_descr);
 468        __hrtimer_init(timer, clock_id, mode);
 469}
 470EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 471
 472void destroy_hrtimer_on_stack(struct hrtimer *timer)
 473{
 474        debug_object_free(timer, &hrtimer_debug_descr);
 475}
 476
 477#else
 478static inline void debug_hrtimer_init(struct hrtimer *timer) { }
 479static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
 480static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 481#endif
 482
 483static inline void
 484debug_init(struct hrtimer *timer, clockid_t clockid,
 485           enum hrtimer_mode mode)
 486{
 487        debug_hrtimer_init(timer);
 488        trace_hrtimer_init(timer, clockid, mode);
 489}
 490
 491static inline void debug_activate(struct hrtimer *timer)
 492{
 493        debug_hrtimer_activate(timer);
 494        trace_hrtimer_start(timer);
 495}
 496
 497static inline void debug_deactivate(struct hrtimer *timer)
 498{
 499        debug_hrtimer_deactivate(timer);
 500        trace_hrtimer_cancel(timer);
 501}
 502
 503/* High resolution timer related functions */
 504#ifdef CONFIG_HIGH_RES_TIMERS
 505
 506/*
 507 * High resolution timer enabled ?
 508 */
 509static int hrtimer_hres_enabled __read_mostly  = 1;
 510
 511/*
 512 * Enable / Disable high resolution mode
 513 */
 514static int __init setup_hrtimer_hres(char *str)
 515{
 516        if (!strcmp(str, "off"))
 517                hrtimer_hres_enabled = 0;
 518        else if (!strcmp(str, "on"))
 519                hrtimer_hres_enabled = 1;
 520        else
 521                return 0;
 522        return 1;
 523}
 524
 525__setup("highres=", setup_hrtimer_hres);
 526
 527/*
 528 * hrtimer_high_res_enabled - query, if the highres mode is enabled
 529 */
 530static inline int hrtimer_is_hres_enabled(void)
 531{
 532        return hrtimer_hres_enabled;
 533}
 534
 535/*
 536 * Is the high resolution mode active ?
 537 */
 538static inline int hrtimer_hres_active(void)
 539{
 540        return __this_cpu_read(hrtimer_bases.hres_active);
 541}
 542
 543/*
 544 * Reprogram the event source with checking both queues for the
 545 * next event
 546 * Called with interrupts disabled and base->lock held
 547 */
 548static void
 549hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 550{
 551        int i;
 552        struct hrtimer_clock_base *base = cpu_base->clock_base;
 553        ktime_t expires, expires_next;
 554
 555        expires_next.tv64 = KTIME_MAX;
 556
 557        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
 558                struct hrtimer *timer;
 559                struct timerqueue_node *next;
 560
 561                next = timerqueue_getnext(&base->active);
 562                if (!next)
 563                        continue;
 564                timer = container_of(next, struct hrtimer, node);
 565
 566                expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 567                /*
 568                 * clock_was_set() has changed base->offset so the
 569                 * result might be negative. Fix it up to prevent a
 570                 * false positive in clockevents_program_event()
 571                 */
 572                if (expires.tv64 < 0)
 573                        expires.tv64 = 0;
 574                if (expires.tv64 < expires_next.tv64)
 575                        expires_next = expires;
 576        }
 577
 578        if (skip_equal && expires_next.tv64 == cpu_base->expires_next.tv64)
 579                return;
 580
 581        cpu_base->expires_next.tv64 = expires_next.tv64;
 582
 583        if (cpu_base->expires_next.tv64 != KTIME_MAX)
 584                tick_program_event(cpu_base->expires_next, 1);
 585}
 586
 587/*
 588 * Shared reprogramming for clock_realtime and clock_monotonic
 589 *
 590 * When a timer is enqueued and expires earlier than the already enqueued
 591 * timers, we have to check, whether it expires earlier than the timer for
 592 * which the clock event device was armed.
 593 *
 594 * Called with interrupts disabled and base->cpu_base.lock held
 595 */
 596static int hrtimer_reprogram(struct hrtimer *timer,
 597                             struct hrtimer_clock_base *base)
 598{
 599        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
 600        ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 601        int res;
 602
 603        WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);
 604
 605        /*
 606         * When the callback is running, we do not reprogram the clock event
 607         * device. The timer callback is either running on a different CPU or
 608         * the callback is executed in the hrtimer_interrupt context. The
 609         * reprogramming is handled either by the softirq, which called the
 610         * callback or at the end of the hrtimer_interrupt.
 611         */
 612        if (hrtimer_callback_running(timer))
 613                return 0;
 614
 615        /*
 616         * CLOCK_REALTIME timer might be requested with an absolute
 617         * expiry time which is less than base->offset. Nothing wrong
 618         * about that, just avoid to call into the tick code, which
 619         * has now objections against negative expiry values.
 620         */
 621        if (expires.tv64 < 0)
 622                return -ETIME;
 623
 624        if (expires.tv64 >= cpu_base->expires_next.tv64)
 625                return 0;
 626
 627        /*
 628         * If a hang was detected in the last timer interrupt then we
 629         * do not schedule a timer which is earlier than the expiry
 630         * which we enforced in the hang detection. We want the system
 631         * to make progress.
 632         */
 633        if (cpu_base->hang_detected)
 634                return 0;
 635
 636        /*
 637         * Clockevents returns -ETIME, when the event was in the past.
 638         */
 639        res = tick_program_event(expires, 0);
 640        if (!IS_ERR_VALUE(res))
 641                cpu_base->expires_next = expires;
 642        return res;
 643}
 644
 645/*
 646 * Initialize the high resolution related parts of cpu_base
 647 */
 648static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
 649{
 650        base->expires_next.tv64 = KTIME_MAX;
 651        base->hres_active = 0;
 652}
 653
 654/*
 655 * When High resolution timers are active, try to reprogram. Note, that in case
 656 * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry
 657 * check happens. The timer gets enqueued into the rbtree. The reprogramming
 658 * and expiry check is done in the hrtimer_interrupt or in the softirq.
 659 */
 660static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 661                                            struct hrtimer_clock_base *base)
 662{
 663        return base->cpu_base->hres_active && hrtimer_reprogram(timer, base);
 664}
 665
 666static inline ktime_t hrtimer_update_base(struct hrtimer_cpu_base *base)
 667{
 668        ktime_t *offs_real = &base->clock_base[HRTIMER_BASE_REALTIME].offset;
 669        ktime_t *offs_boot = &base->clock_base[HRTIMER_BASE_BOOTTIME].offset;
 670        ktime_t *offs_tai = &base->clock_base[HRTIMER_BASE_TAI].offset;
 671
 672        return ktime_get_update_offsets(offs_real, offs_boot, offs_tai);
 673}
 674
 675/*
 676 * Retrigger next event is called after clock was set
 677 *
 678 * Called with interrupts disabled via on_each_cpu()
 679 */
 680static void retrigger_next_event(void *arg)
 681{
 682        struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
 683
 684        if (!hrtimer_hres_active())
 685                return;
 686
 687        raw_spin_lock(&base->lock);
 688        hrtimer_update_base(base);
 689        hrtimer_force_reprogram(base, 0);
 690        raw_spin_unlock(&base->lock);
 691}
 692
 693/*
 694 * Switch to high resolution mode
 695 */
 696static int hrtimer_switch_to_hres(void)
 697{
 698        int i, cpu = smp_processor_id();
 699        struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
 700        unsigned long flags;
 701
 702        if (base->hres_active)
 703                return 1;
 704
 705        local_irq_save(flags);
 706
 707        if (tick_init_highres()) {
 708                local_irq_restore(flags);
 709                printk(KERN_WARNING "Could not switch to high resolution "
 710                                    "mode on CPU %d\n", cpu);
 711                return 0;
 712        }
 713        base->hres_active = 1;
 714        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
 715                base->clock_base[i].resolution = KTIME_HIGH_RES;
 716
 717        tick_setup_sched_timer();
 718        /* "Retrigger" the interrupt to get things going */
 719        retrigger_next_event(NULL);
 720        local_irq_restore(flags);
 721        return 1;
 722}
 723
 724/*
 725 * Called from timekeeping code to reprogramm the hrtimer interrupt
 726 * device. If called from the timer interrupt context we defer it to
 727 * softirq context.
 728 */
 729void clock_was_set_delayed(void)
 730{
 731        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
 732
 733        cpu_base->clock_was_set = 1;
 734        __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
 735}
 736
 737#else
 738
 739static inline int hrtimer_hres_active(void) { return 0; }
 740static inline int hrtimer_is_hres_enabled(void) { return 0; }
 741static inline int hrtimer_switch_to_hres(void) { return 0; }
 742static inline void
 743hrtimer_force_reprogram(struct hrtimer_cpu_base *base, int skip_equal) { }
 744static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 745                                            struct hrtimer_clock_base *base)
 746{
 747        return 0;
 748}
 749static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
 750static inline void retrigger_next_event(void *arg) { }
 751
 752#endif /* CONFIG_HIGH_RES_TIMERS */
 753
 754/*
 755 * Clock realtime was set
 756 *
 757 * Change the offset of the realtime clock vs. the monotonic
 758 * clock.
 759 *
 760 * We might have to reprogram the high resolution timer interrupt. On
 761 * SMP we call the architecture specific code to retrigger _all_ high
 762 * resolution timer interrupts. On UP we just disable interrupts and
 763 * call the high resolution interrupt code.
 764 */
 765void clock_was_set(void)
 766{
 767#ifdef CONFIG_HIGH_RES_TIMERS
 768        /* Retrigger the CPU local events everywhere */
 769        on_each_cpu(retrigger_next_event, NULL, 1);
 770#endif
 771        timerfd_clock_was_set();
 772}
 773
 774/*
 775 * During resume we might have to reprogram the high resolution timer
 776 * interrupt (on the local CPU):
 777 */
 778void hrtimers_resume(void)
 779{
 780        WARN_ONCE(!irqs_disabled(),
 781                  KERN_INFO "hrtimers_resume() called with IRQs enabled!");
 782
 783        retrigger_next_event(NULL);
 784        timerfd_clock_was_set();
 785}
 786
 787static inline void timer_stats_hrtimer_set_start_info(struct hrtimer *timer)
 788{
 789#ifdef CONFIG_TIMER_STATS
 790        if (timer->start_site)
 791                return;
 792        timer->start_site = __builtin_return_address(0);
 793        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 794        timer->start_pid = current->pid;
 795#endif
 796}
 797
 798static inline void timer_stats_hrtimer_clear_start_info(struct hrtimer *timer)
 799{
 800#ifdef CONFIG_TIMER_STATS
 801        timer->start_site = NULL;
 802#endif
 803}
 804
 805static inline void timer_stats_account_hrtimer(struct hrtimer *timer)
 806{
 807#ifdef CONFIG_TIMER_STATS
 808        if (likely(!timer_stats_active))
 809                return;
 810        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 811                                 timer->function, timer->start_comm, 0);
 812#endif
 813}
 814
 815/*
 816 * Counterpart to lock_hrtimer_base above:
 817 */
 818static inline
 819void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
 820{
 821        raw_spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags);
 822}
 823
 824/**
 825 * hrtimer_forward - forward the timer expiry
 826 * @timer:      hrtimer to forward
 827 * @now:        forward past this time
 828 * @interval:   the interval to forward
 829 *
 830 * Forward the timer expiry so it will expire in the future.
 831 * Returns the number of overruns.
 832 */
 833u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
 834{
 835        u64 orun = 1;
 836        ktime_t delta;
 837
 838        delta = ktime_sub(now, hrtimer_get_expires(timer));
 839
 840        if (delta.tv64 < 0)
 841                return 0;
 842
 843        if (interval.tv64 < timer->base->resolution.tv64)
 844                interval.tv64 = timer->base->resolution.tv64;
 845
 846        if (unlikely(delta.tv64 >= interval.tv64)) {
 847                s64 incr = ktime_to_ns(interval);
 848
 849                orun = ktime_divns(delta, incr);
 850                hrtimer_add_expires_ns(timer, incr * orun);
 851                if (hrtimer_get_expires_tv64(timer) > now.tv64)
 852                        return orun;
 853                /*
 854                 * This (and the ktime_add() below) is the
 855                 * correction for exact:
 856                 */
 857                orun++;
 858        }
 859        hrtimer_add_expires(timer, interval);
 860
 861        return orun;
 862}
 863EXPORT_SYMBOL_GPL(hrtimer_forward);
 864
 865/*
 866 * enqueue_hrtimer - internal function to (re)start a timer
 867 *
 868 * The timer is inserted in expiry order. Insertion into the
 869 * red black tree is O(log(n)). Must hold the base lock.
 870 *
 871 * Returns 1 when the new timer is the leftmost timer in the tree.
 872 */
 873static int enqueue_hrtimer(struct hrtimer *timer,
 874                           struct hrtimer_clock_base *base)
 875{
 876        debug_activate(timer);
 877
 878        timerqueue_add(&base->active, &timer->node);
 879        base->cpu_base->active_bases |= 1 << base->index;
 880
 881        /*
 882         * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the
 883         * state of a possibly running callback.
 884         */
 885        timer->state |= HRTIMER_STATE_ENQUEUED;
 886
 887        return (&timer->node == base->active.next);
 888}
 889
 890/*
 891 * __remove_hrtimer - internal function to remove a timer
 892 *
 893 * Caller must hold the base lock.
 894 *
 895 * High resolution timer mode reprograms the clock event device when the
 896 * timer is the one which expires next. The caller can disable this by setting
 897 * reprogram to zero. This is useful, when the context does a reprogramming
 898 * anyway (e.g. timer interrupt)
 899 */
 900static void __remove_hrtimer(struct hrtimer *timer,
 901                             struct hrtimer_clock_base *base,
 902                             unsigned long newstate, int reprogram)
 903{
 904        struct timerqueue_node *next_timer;
 905        if (!(timer->state & HRTIMER_STATE_ENQUEUED))
 906                goto out;
 907
 908        next_timer = timerqueue_getnext(&base->active);
 909        timerqueue_del(&base->active, &timer->node);
 910        if (&timer->node == next_timer) {
 911#ifdef CONFIG_HIGH_RES_TIMERS
 912                /* Reprogram the clock event device. if enabled */
 913                if (reprogram && hrtimer_hres_active()) {
 914                        ktime_t expires;
 915
 916                        expires = ktime_sub(hrtimer_get_expires(timer),
 917                                            base->offset);
 918                        if (base->cpu_base->expires_next.tv64 == expires.tv64)
 919                                hrtimer_force_reprogram(base->cpu_base, 1);
 920                }
 921#endif
 922        }
 923        if (!timerqueue_getnext(&base->active))
 924                base->cpu_base->active_bases &= ~(1 << base->index);
 925out:
 926        timer->state = newstate;
 927}
 928
 929/*
 930 * remove hrtimer, called with base lock held
 931 */
 932static inline int
 933remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
 934{
 935        if (hrtimer_is_queued(timer)) {
 936                unsigned long state;
 937                int reprogram;
 938
 939                /*
 940                 * Remove the timer and force reprogramming when high
 941                 * resolution mode is active and the timer is on the current
 942                 * CPU. If we remove a timer on another CPU, reprogramming is
 943                 * skipped. The interrupt event on this CPU is fired and
 944                 * reprogramming happens in the interrupt handler. This is a
 945                 * rare case and less expensive than a smp call.
 946                 */
 947                debug_deactivate(timer);
 948                timer_stats_hrtimer_clear_start_info(timer);
 949                reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
 950                /*
 951                 * We must preserve the CALLBACK state flag here,
 952                 * otherwise we could move the timer base in
 953                 * switch_hrtimer_base.
 954                 */
 955                state = timer->state & HRTIMER_STATE_CALLBACK;
 956                __remove_hrtimer(timer, base, state, reprogram);
 957                return 1;
 958        }
 959        return 0;
 960}
 961
 962int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 963                unsigned long delta_ns, const enum hrtimer_mode mode,
 964                int wakeup)
 965{
 966        struct hrtimer_clock_base *base, *new_base;
 967        unsigned long flags;
 968        int ret, leftmost;
 969
 970        base = lock_hrtimer_base(timer, &flags);
 971
 972        /* Remove an active timer from the queue: */
 973        ret = remove_hrtimer(timer, base);
 974
 975        /* Switch the timer base, if necessary: */
 976        new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
 977
 978        if (mode & HRTIMER_MODE_REL) {
 979                tim = ktime_add_safe(tim, new_base->get_time());
 980                /*
 981                 * CONFIG_TIME_LOW_RES is a temporary way for architectures
 982                 * to signal that they simply return xtime in
 983                 * do_gettimeoffset(). In this case we want to round up by
 984                 * resolution when starting a relative timer, to avoid short
 985                 * timeouts. This will go away with the GTOD framework.
 986                 */
 987#ifdef CONFIG_TIME_LOW_RES
 988                tim = ktime_add_safe(tim, base->resolution);
 989#endif
 990        }
 991
 992        hrtimer_set_expires_range_ns(timer, tim, delta_ns);
 993
 994        timer_stats_hrtimer_set_start_info(timer);
 995
 996        leftmost = enqueue_hrtimer(timer, new_base);
 997
 998        /*
 999         * Only allow reprogramming if the new base is on this CPU.
1000         * (it might still be on another CPU if the timer was pending)
1001         *
1002         * XXX send_remote_softirq() ?
1003         */
1004        if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
1005                && hrtimer_enqueue_reprogram(timer, new_base)) {
1006                if (wakeup) {
1007                        /*
1008                         * We need to drop cpu_base->lock to avoid a
1009                         * lock ordering issue vs. rq->lock.
1010                         */
1011                        raw_spin_unlock(&new_base->cpu_base->lock);
1012                        raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1013                        local_irq_restore(flags);
1014                        return ret;
1015                } else {
1016                        __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
1017                }
1018        }
1019
1020        unlock_hrtimer_base(timer, &flags);
1021
1022        return ret;
1023}
1024
1025/**
1026 * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
1027 * @timer:      the timer to be added
1028 * @tim:        expiry time
1029 * @delta_ns:   "slack" range for the timer
1030 * @mode:       expiry mode: absolute (HRTIMER_MODE_ABS) or
1031 *              relative (HRTIMER_MODE_REL)
1032 *
1033 * Returns:
1034 *  0 on success
1035 *  1 when the timer was active
1036 */
1037int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
1038                unsigned long delta_ns, const enum hrtimer_mode mode)
1039{
1040        return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
1041}
1042EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
1043
1044/**
1045 * hrtimer_start - (re)start an hrtimer on the current CPU
1046 * @timer:      the timer to be added
1047 * @tim:        expiry time
1048 * @mode:       expiry mode: absolute (HRTIMER_MODE_ABS) or
1049 *              relative (HRTIMER_MODE_REL)
1050 *
1051 * Returns:
1052 *  0 on success
1053 *  1 when the timer was active
1054 */
1055int
1056hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
1057{
1058        return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
1059}
1060EXPORT_SYMBOL_GPL(hrtimer_start);
1061
1062
1063/**
1064 * hrtimer_try_to_cancel - try to deactivate a timer
1065 * @timer:      hrtimer to stop
1066 *
1067 * Returns:
1068 *  0 when the timer was not active
1069 *  1 when the timer was active
1070 * -1 when the timer is currently excuting the callback function and
1071 *    cannot be stopped
1072 */
1073int hrtimer_try_to_cancel(struct hrtimer *timer)
1074{
1075        struct hrtimer_clock_base *base;
1076        unsigned long flags;
1077        int ret = -1;
1078
1079        base = lock_hrtimer_base(timer, &flags);
1080
1081        if (!hrtimer_callback_running(timer))
1082                ret = remove_hrtimer(timer, base);
1083
1084        unlock_hrtimer_base(timer, &flags);
1085
1086        return ret;
1087
1088}
1089EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel);
1090
1091/**
1092 * hrtimer_cancel - cancel a timer and wait for the handler to finish.
1093 * @timer:      the timer to be cancelled
1094 *
1095 * Returns:
1096 *  0 when the timer was not active
1097 *  1 when the timer was active
1098 */
1099int hrtimer_cancel(struct hrtimer *timer)
1100{
1101        for (;;) {
1102                int ret = hrtimer_try_to_cancel(timer);
1103
1104                if (ret >= 0)
1105                        return ret;
1106                cpu_relax();
1107        }
1108}
1109EXPORT_SYMBOL_GPL(hrtimer_cancel);
1110
1111/**
1112 * hrtimer_get_remaining - get remaining time for the timer
1113 * @timer:      the timer to read
1114 */
1115ktime_t hrtimer_get_remaining(const struct hrtimer *timer)
1116{
1117        unsigned long flags;
1118        ktime_t rem;
1119
1120        lock_hrtimer_base(timer, &flags);
1121        rem = hrtimer_expires_remaining(timer);
1122        unlock_hrtimer_base(timer, &flags);
1123
1124        return rem;
1125}
1126EXPORT_SYMBOL_GPL(hrtimer_get_remaining);
1127
1128#ifdef CONFIG_NO_HZ_COMMON
1129/**
1130 * hrtimer_get_next_event - get the time until next expiry event
1131 *
1132 * Returns the delta to the next expiry event or KTIME_MAX if no timer
1133 * is pending.
1134 */
1135ktime_t hrtimer_get_next_event(void)
1136{
1137        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1138        struct hrtimer_clock_base *base = cpu_base->clock_base;
1139        ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
1140        unsigned long flags;
1141        int i;
1142
1143        raw_spin_lock_irqsave(&cpu_base->lock, flags);
1144
1145        if (!hrtimer_hres_active()) {
1146                for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) {
1147                        struct hrtimer *timer;
1148                        struct timerqueue_node *next;
1149
1150                        next = timerqueue_getnext(&base->active);
1151                        if (!next)
1152                                continue;
1153
1154                        timer = container_of(next, struct hrtimer, node);
1155                        delta.tv64 = hrtimer_get_expires_tv64(timer);
1156                        delta = ktime_sub(delta, base->get_time());
1157                        if (delta.tv64 < mindelta.tv64)
1158                                mindelta.tv64 = delta.tv64;
1159                }
1160        }
1161
1162        raw_spin_unlock_irqrestore(&cpu_base->lock, flags);
1163
1164        if (mindelta.tv64 < 0)
1165                mindelta.tv64 = 0;
1166        return mindelta;
1167}
1168#endif
1169
1170static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1171                           enum hrtimer_mode mode)
1172{
1173        struct hrtimer_cpu_base *cpu_base;
1174        int base;
1175
1176        memset(timer, 0, sizeof(struct hrtimer));
1177
1178        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1179
1180        if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
1181                clock_id = CLOCK_MONOTONIC;
1182
1183        base = hrtimer_clockid_to_base(clock_id);
1184        timer->base = &cpu_base->clock_base[base];
1185        timerqueue_init(&timer->node);
1186
1187#ifdef CONFIG_TIMER_STATS
1188        timer->start_site = NULL;
1189        timer->start_pid = -1;
1190        memset(timer->start_comm, 0, TASK_COMM_LEN);
1191#endif
1192}
1193
1194/**
1195 * hrtimer_init - initialize a timer to the given clock
1196 * @timer:      the timer to be initialized
1197 * @clock_id:   the clock to be used
1198 * @mode:       timer mode abs/rel
1199 */
1200void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
1201                  enum hrtimer_mode mode)
1202{
1203        debug_init(timer, clock_id, mode);
1204        __hrtimer_init(timer, clock_id, mode);
1205}
1206EXPORT_SYMBOL_GPL(hrtimer_init);
1207
1208/**
1209 * hrtimer_get_res - get the timer resolution for a clock
1210 * @which_clock: which clock to query
1211 * @tp:          pointer to timespec variable to store the resolution
1212 *
1213 * Store the resolution of the clock selected by @which_clock in the
1214 * variable pointed to by @tp.
1215 */
1216int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
1217{
1218        struct hrtimer_cpu_base *cpu_base;
1219        int base = hrtimer_clockid_to_base(which_clock);
1220
1221        cpu_base = &__raw_get_cpu_var(hrtimer_bases);
1222        *tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
1223
1224        return 0;
1225}
1226EXPORT_SYMBOL_GPL(hrtimer_get_res);
1227
1228static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1229{
1230        struct hrtimer_clock_base *base = timer->base;
1231        struct hrtimer_cpu_base *cpu_base = base->cpu_base;
1232        enum hrtimer_restart (*fn)(struct hrtimer *);
1233        int restart;
1234
1235        WARN_ON(!irqs_disabled());
1236
1237        debug_deactivate(timer);
1238        __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
1239        timer_stats_account_hrtimer(timer);
1240        fn = timer->function;
1241
1242        /*
1243         * Because we run timers from hardirq context, there is no chance
1244         * they get migrated to another cpu, therefore its safe to unlock
1245         * the timer base.
1246         */
1247        raw_spin_unlock(&cpu_base->lock);
1248        trace_hrtimer_expire_entry(timer, now);
1249        restart = fn(timer);
1250        trace_hrtimer_expire_exit(timer);
1251        raw_spin_lock(&cpu_base->lock);
1252
1253        /*
1254         * Note: We clear the CALLBACK bit after enqueue_hrtimer and
1255         * we do not reprogramm the event hardware. Happens either in
1256         * hrtimer_start_range_ns() or in hrtimer_interrupt()
1257         */
1258        if (restart != HRTIMER_NORESTART) {
1259                BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1260                enqueue_hrtimer(timer, base);
1261        }
1262
1263        WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1264
1265        timer->state &= ~HRTIMER_STATE_CALLBACK;
1266}
1267
1268#ifdef CONFIG_HIGH_RES_TIMERS
1269
1270/*
1271 * High resolution timer interrupt
1272 * Called with interrupts disabled
1273 */
1274void hrtimer_interrupt(struct clock_event_device *dev)
1275{
1276        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1277        ktime_t expires_next, now, entry_time, delta;
1278        int i, retries = 0;
1279
1280        BUG_ON(!cpu_base->hres_active);
1281        cpu_base->nr_events++;
1282        dev->next_event.tv64 = KTIME_MAX;
1283
1284        raw_spin_lock(&cpu_base->lock);
1285        entry_time = now = hrtimer_update_base(cpu_base);
1286retry:
1287        expires_next.tv64 = KTIME_MAX;
1288        /*
1289         * We set expires_next to KTIME_MAX here with cpu_base->lock
1290         * held to prevent that a timer is enqueued in our queue via
1291         * the migration code. This does not affect enqueueing of
1292         * timers which run their callback and need to be requeued on
1293         * this CPU.
1294         */
1295        cpu_base->expires_next.tv64 = KTIME_MAX;
1296
1297        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1298                struct hrtimer_clock_base *base;
1299                struct timerqueue_node *node;
1300                ktime_t basenow;
1301
1302                if (!(cpu_base->active_bases & (1 << i)))
1303                        continue;
1304
1305                base = cpu_base->clock_base + i;
1306                basenow = ktime_add(now, base->offset);
1307
1308                while ((node = timerqueue_getnext(&base->active))) {
1309                        struct hrtimer *timer;
1310
1311                        timer = container_of(node, struct hrtimer, node);
1312
1313                        /*
1314                         * The immediate goal for using the softexpires is
1315                         * minimizing wakeups, not running timers at the
1316                         * earliest interrupt after their soft expiration.
1317                         * This allows us to avoid using a Priority Search
1318                         * Tree, which can answer a stabbing querry for
1319                         * overlapping intervals and instead use the simple
1320                         * BST we already have.
1321                         * We don't add extra wakeups by delaying timers that
1322                         * are right-of a not yet expired timer, because that
1323                         * timer will have to trigger a wakeup anyway.
1324                         */
1325
1326                        if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) {
1327                                ktime_t expires;
1328
1329                                expires = ktime_sub(hrtimer_get_expires(timer),
1330                                                    base->offset);
1331                                if (expires.tv64 < 0)
1332                                        expires.tv64 = KTIME_MAX;
1333                                if (expires.tv64 < expires_next.tv64)
1334                                        expires_next = expires;
1335                                break;
1336                        }
1337
1338                        __run_hrtimer(timer, &basenow);
1339                }
1340        }
1341
1342        /*
1343         * Store the new expiry value so the migration code can verify
1344         * against it.
1345         */
1346        cpu_base->expires_next = expires_next;
1347        raw_spin_unlock(&cpu_base->lock);
1348
1349        /* Reprogramming necessary ? */
1350        if (expires_next.tv64 == KTIME_MAX ||
1351            !tick_program_event(expires_next, 0)) {
1352                cpu_base->hang_detected = 0;
1353                return;
1354        }
1355
1356        /*
1357         * The next timer was already expired due to:
1358         * - tracing
1359         * - long lasting callbacks
1360         * - being scheduled away when running in a VM
1361         *
1362         * We need to prevent that we loop forever in the hrtimer
1363         * interrupt routine. We give it 3 attempts to avoid
1364         * overreacting on some spurious event.
1365         *
1366         * Acquire base lock for updating the offsets and retrieving
1367         * the current time.
1368         */
1369        raw_spin_lock(&cpu_base->lock);
1370        now = hrtimer_update_base(cpu_base);
1371        cpu_base->nr_retries++;
1372        if (++retries < 3)
1373                goto retry;
1374        /*
1375         * Give the system a chance to do something else than looping
1376         * here. We stored the entry time, so we know exactly how long
1377         * we spent here. We schedule the next event this amount of
1378         * time away.
1379         */
1380        cpu_base->nr_hangs++;
1381        cpu_base->hang_detected = 1;
1382        raw_spin_unlock(&cpu_base->lock);
1383        delta = ktime_sub(now, entry_time);
1384        if (delta.tv64 > cpu_base->max_hang_time.tv64)
1385                cpu_base->max_hang_time = delta;
1386        /*
1387         * Limit it to a sensible value as we enforce a longer
1388         * delay. Give the CPU at least 100ms to catch up.
1389         */
1390        if (delta.tv64 > 100 * NSEC_PER_MSEC)
1391                expires_next = ktime_add_ns(now, 100 * NSEC_PER_MSEC);
1392        else
1393                expires_next = ktime_add(now, delta);
1394        tick_program_event(expires_next, 1);
1395        printk_once(KERN_WARNING "hrtimer: interrupt took %llu ns\n",
1396                    ktime_to_ns(delta));
1397}
1398
1399/*
1400 * local version of hrtimer_peek_ahead_timers() called with interrupts
1401 * disabled.
1402 */
1403static void __hrtimer_peek_ahead_timers(void)
1404{
1405        struct tick_device *td;
1406
1407        if (!hrtimer_hres_active())
1408                return;
1409
1410        td = &__get_cpu_var(tick_cpu_device);
1411        if (td && td->evtdev)
1412                hrtimer_interrupt(td->evtdev);
1413}
1414
1415/**
1416 * hrtimer_peek_ahead_timers -- run soft-expired timers now
1417 *
1418 * hrtimer_peek_ahead_timers will peek at the timer queue of
1419 * the current cpu and check if there are any timers for which
1420 * the soft expires time has passed. If any such timers exist,
1421 * they are run immediately and then removed from the timer queue.
1422 *
1423 */
1424void hrtimer_peek_ahead_timers(void)
1425{
1426        unsigned long flags;
1427
1428        local_irq_save(flags);
1429        __hrtimer_peek_ahead_timers();
1430        local_irq_restore(flags);
1431}
1432
1433static void run_hrtimer_softirq(struct softirq_action *h)
1434{
1435        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1436
1437        if (cpu_base->clock_was_set) {
1438                cpu_base->clock_was_set = 0;
1439                clock_was_set();
1440        }
1441
1442        hrtimer_peek_ahead_timers();
1443}
1444
1445#else /* CONFIG_HIGH_RES_TIMERS */
1446
1447static inline void __hrtimer_peek_ahead_timers(void) { }
1448
1449#endif  /* !CONFIG_HIGH_RES_TIMERS */
1450
1451/*
1452 * Called from timer softirq every jiffy, expire hrtimers:
1453 *
1454 * For HRT its the fall back code to run the softirq in the timer
1455 * softirq context in case the hrtimer initialization failed or has
1456 * not been done yet.
1457 */
1458void hrtimer_run_pending(void)
1459{
1460        if (hrtimer_hres_active())
1461                return;
1462
1463        /*
1464         * This _is_ ugly: We have to check in the softirq context,
1465         * whether we can switch to highres and / or nohz mode. The
1466         * clocksource switch happens in the timer interrupt with
1467         * xtime_lock held. Notification from there only sets the
1468         * check bit in the tick_oneshot code, otherwise we might
1469         * deadlock vs. xtime_lock.
1470         */
1471        if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
1472                hrtimer_switch_to_hres();
1473}
1474
1475/*
1476 * Called from hardirq context every jiffy
1477 */
1478void hrtimer_run_queues(void)
1479{
1480        struct timerqueue_node *node;
1481        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
1482        struct hrtimer_clock_base *base;
1483        int index, gettime = 1;
1484
1485        if (hrtimer_hres_active())
1486                return;
1487
1488        for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
1489                base = &cpu_base->clock_base[index];
1490                if (!timerqueue_getnext(&base->active))
1491                        continue;
1492
1493                if (gettime) {
1494                        hrtimer_get_softirq_time(cpu_base);
1495                        gettime = 0;
1496                }
1497
1498                raw_spin_lock(&cpu_base->lock);
1499
1500                while ((node = timerqueue_getnext(&base->active))) {
1501                        struct hrtimer *timer;
1502
1503                        timer = container_of(node, struct hrtimer, node);
1504                        if (base->softirq_time.tv64 <=
1505                                        hrtimer_get_expires_tv64(timer))
1506                                break;
1507
1508                        __run_hrtimer(timer, &base->softirq_time);
1509                }
1510                raw_spin_unlock(&cpu_base->lock);
1511        }
1512}
1513
1514/*
1515 * Sleep related functions:
1516 */
1517static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer)
1518{
1519        struct hrtimer_sleeper *t =
1520                container_of(timer, struct hrtimer_sleeper, timer);
1521        struct task_struct *task = t->task;
1522
1523        t->task = NULL;
1524        if (task)
1525                wake_up_process(task);
1526
1527        return HRTIMER_NORESTART;
1528}
1529
1530void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
1531{
1532        sl->timer.function = hrtimer_wakeup;
1533        sl->task = task;
1534}
1535EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
1536
1537static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
1538{
1539        hrtimer_init_sleeper(t, current);
1540
1541        do {
1542                set_current_state(TASK_INTERRUPTIBLE);
1543                hrtimer_start_expires(&t->timer, mode);
1544                if (!hrtimer_active(&t->timer))
1545                        t->task = NULL;
1546
1547                if (likely(t->task))
1548                        schedule();
1549
1550                hrtimer_cancel(&t->timer);
1551                mode = HRTIMER_MODE_ABS;
1552
1553        } while (t->task && !signal_pending(current));
1554
1555        __set_current_state(TASK_RUNNING);
1556
1557        return t->task == NULL;
1558}
1559
1560static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
1561{
1562        struct timespec rmt;
1563        ktime_t rem;
1564
1565        rem = hrtimer_expires_remaining(timer);
1566        if (rem.tv64 <= 0)
1567                return 0;
1568        rmt = ktime_to_timespec(rem);
1569
1570        if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
1571                return -EFAULT;
1572
1573        return 1;
1574}
1575
1576long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
1577{
1578        struct hrtimer_sleeper t;
1579        struct timespec __user  *rmtp;
1580        int ret = 0;
1581
1582        hrtimer_init_on_stack(&t.timer, restart->nanosleep.clockid,
1583                                HRTIMER_MODE_ABS);
1584        hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
1585
1586        if (do_nanosleep(&t, HRTIMER_MODE_ABS))
1587                goto out;
1588
1589        rmtp = restart->nanosleep.rmtp;
1590        if (rmtp) {
1591                ret = update_rmtp(&t.timer, rmtp);
1592                if (ret <= 0)
1593                        goto out;
1594        }
1595
1596        /* The other values in restart are already filled in */
1597        ret = -ERESTART_RESTARTBLOCK;
1598out:
1599        destroy_hrtimer_on_stack(&t.timer);
1600        return ret;
1601}
1602
1603long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
1604                       const enum hrtimer_mode mode, const clockid_t clockid)
1605{
1606        struct restart_block *restart;
1607        struct hrtimer_sleeper t;
1608        int ret = 0;
1609        unsigned long slack;
1610
1611        slack = current->timer_slack_ns;
1612        if (rt_task(current))
1613                slack = 0;
1614
1615        hrtimer_init_on_stack(&t.timer, clockid, mode);
1616        hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
1617        if (do_nanosleep(&t, mode))
1618                goto out;
1619
1620        /* Absolute timers do not update the rmtp value and restart: */
1621        if (mode == HRTIMER_MODE_ABS) {
1622                ret = -ERESTARTNOHAND;
1623                goto out;
1624        }
1625
1626        if (rmtp) {
1627                ret = update_rmtp(&t.timer, rmtp);
1628                if (ret <= 0)
1629                        goto out;
1630        }
1631
1632        restart = &current_thread_info()->restart_block;
1633        restart->fn = hrtimer_nanosleep_restart;
1634        restart->nanosleep.clockid = t.timer.base->clockid;
1635        restart->nanosleep.rmtp = rmtp;
1636        restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer);
1637
1638        ret = -ERESTART_RESTARTBLOCK;
1639out:
1640        destroy_hrtimer_on_stack(&t.timer);
1641        return ret;
1642}
1643
1644SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp,
1645                struct timespec __user *, rmtp)
1646{
1647        struct timespec tu;
1648
1649        if (copy_from_user(&tu, rqtp, sizeof(tu)))
1650                return -EFAULT;
1651
1652        if (!timespec_valid(&tu))
1653                return -EINVAL;
1654
1655        return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC);
1656}
1657
1658/*
1659 * Functions related to boot-time initialization:
1660 */
1661static void __cpuinit init_hrtimers_cpu(int cpu)
1662{
1663        struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
1664        int i;
1665
1666        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1667                cpu_base->clock_base[i].cpu_base = cpu_base;
1668                timerqueue_init_head(&cpu_base->clock_base[i].active);
1669        }
1670
1671        hrtimer_init_hres(cpu_base);
1672}
1673
1674#ifdef CONFIG_HOTPLUG_CPU
1675
1676static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
1677                                struct hrtimer_clock_base *new_base)
1678{
1679        struct hrtimer *timer;
1680        struct timerqueue_node *node;
1681
1682        while ((node = timerqueue_getnext(&old_base->active))) {
1683                timer = container_of(node, struct hrtimer, node);
1684                BUG_ON(hrtimer_callback_running(timer));
1685                debug_deactivate(timer);
1686
1687                /*
1688                 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
1689                 * timer could be seen as !active and just vanish away
1690                 * under us on another CPU
1691                 */
1692                __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0);
1693                timer->base = new_base;
1694                /*
1695                 * Enqueue the timers on the new cpu. This does not
1696                 * reprogram the event device in case the timer
1697                 * expires before the earliest on this CPU, but we run
1698                 * hrtimer_interrupt after we migrated everything to
1699                 * sort out already expired timers and reprogram the
1700                 * event device.
1701                 */
1702                enqueue_hrtimer(timer, new_base);
1703
1704                /* Clear the migration state bit */
1705                timer->state &= ~HRTIMER_STATE_MIGRATE;
1706        }
1707}
1708
1709static void migrate_hrtimers(int scpu)
1710{
1711        struct hrtimer_cpu_base *old_base, *new_base;
1712        int i;
1713
1714        BUG_ON(cpu_online(scpu));
1715        tick_cancel_sched_timer(scpu);
1716
1717        local_irq_disable();
1718        old_base = &per_cpu(hrtimer_bases, scpu);
1719        new_base = &__get_cpu_var(hrtimer_bases);
1720        /*
1721         * The caller is globally serialized and nobody else
1722         * takes two locks at once, deadlock is not possible.
1723         */
1724        raw_spin_lock(&new_base->lock);
1725        raw_spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1726
1727        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
1728                migrate_hrtimer_list(&old_base->clock_base[i],
1729                                     &new_base->clock_base[i]);
1730        }
1731
1732        raw_spin_unlock(&old_base->lock);
1733        raw_spin_unlock(&new_base->lock);
1734
1735        /* Check, if we got expired work to do */
1736        __hrtimer_peek_ahead_timers();
1737        local_irq_enable();
1738}
1739
1740#endif /* CONFIG_HOTPLUG_CPU */
1741
1742static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
1743                                        unsigned long action, void *hcpu)
1744{
1745        int scpu = (long)hcpu;
1746
1747        switch (action) {
1748
1749        case CPU_UP_PREPARE:
1750        case CPU_UP_PREPARE_FROZEN:
1751                init_hrtimers_cpu(scpu);
1752                break;
1753
1754#ifdef CONFIG_HOTPLUG_CPU
1755        case CPU_DYING:
1756        case CPU_DYING_FROZEN:
1757                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
1758                break;
1759        case CPU_DEAD:
1760        case CPU_DEAD_FROZEN:
1761        {
1762                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu);
1763                migrate_hrtimers(scpu);
1764                break;
1765        }
1766#endif
1767
1768        default:
1769                break;
1770        }
1771
1772        return NOTIFY_OK;
1773}
1774
1775static struct notifier_block __cpuinitdata hrtimers_nb = {
1776        .notifier_call = hrtimer_cpu_notify,
1777};
1778
1779void __init hrtimers_init(void)
1780{
1781        hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE,
1782                          (void *)(long)smp_processor_id());
1783        register_cpu_notifier(&hrtimers_nb);
1784#ifdef CONFIG_HIGH_RES_TIMERS
1785        open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq);
1786#endif
1787}
1788
1789/**
1790 * schedule_hrtimeout_range_clock - sleep until timeout
1791 * @expires:    timeout value (ktime_t)
1792 * @delta:      slack in expires timeout (ktime_t)
1793 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1794 * @clock:      timer clock, CLOCK_MONOTONIC or CLOCK_REALTIME
1795 */
1796int __sched
1797schedule_hrtimeout_range_clock(ktime_t *expires, unsigned long delta,
1798                               const enum hrtimer_mode mode, int clock)
1799{
1800        struct hrtimer_sleeper t;
1801
1802        /*
1803         * Optimize when a zero timeout value is given. It does not
1804         * matter whether this is an absolute or a relative time.
1805         */
1806        if (expires && !expires->tv64) {
1807                __set_current_state(TASK_RUNNING);
1808                return 0;
1809        }
1810
1811        /*
1812         * A NULL parameter means "infinite"
1813         */
1814        if (!expires) {
1815                schedule();
1816                __set_current_state(TASK_RUNNING);
1817                return -EINTR;
1818        }
1819
1820        hrtimer_init_on_stack(&t.timer, clock, mode);
1821        hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
1822
1823        hrtimer_init_sleeper(&t, current);
1824
1825        hrtimer_start_expires(&t.timer, mode);
1826        if (!hrtimer_active(&t.timer))
1827                t.task = NULL;
1828
1829        if (likely(t.task))
1830                schedule();
1831
1832        hrtimer_cancel(&t.timer);
1833        destroy_hrtimer_on_stack(&t.timer);
1834
1835        __set_current_state(TASK_RUNNING);
1836
1837        return !t.task ? 0 : -EINTR;
1838}
1839
1840/**
1841 * schedule_hrtimeout_range - sleep until timeout
1842 * @expires:    timeout value (ktime_t)
1843 * @delta:      slack in expires timeout (ktime_t)
1844 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1845 *
1846 * Make the current task sleep until the given expiry time has
1847 * elapsed. The routine will return immediately unless
1848 * the current task state has been set (see set_current_state()).
1849 *
1850 * The @delta argument gives the kernel the freedom to schedule the
1851 * actual wakeup to a time that is both power and performance friendly.
1852 * The kernel give the normal best effort behavior for "@expires+@delta",
1853 * but may decide to fire the timer earlier, but no earlier than @expires.
1854 *
1855 * You can set the task state as follows -
1856 *
1857 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1858 * pass before the routine returns.
1859 *
1860 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1861 * delivered to the current task.
1862 *
1863 * The current task state is guaranteed to be TASK_RUNNING when this
1864 * routine returns.
1865 *
1866 * Returns 0 when the timer has expired otherwise -EINTR
1867 */
1868int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta,
1869                                     const enum hrtimer_mode mode)
1870{
1871        return schedule_hrtimeout_range_clock(expires, delta, mode,
1872                                              CLOCK_MONOTONIC);
1873}
1874EXPORT_SYMBOL_GPL(schedule_hrtimeout_range);
1875
1876/**
1877 * schedule_hrtimeout - sleep until timeout
1878 * @expires:    timeout value (ktime_t)
1879 * @mode:       timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL
1880 *
1881 * Make the current task sleep until the given expiry time has
1882 * elapsed. The routine will return immediately unless
1883 * the current task state has been set (see set_current_state()).
1884 *
1885 * You can set the task state as follows -
1886 *
1887 * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to
1888 * pass before the routine returns.
1889 *
1890 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1891 * delivered to the current task.
1892 *
1893 * The current task state is guaranteed to be TASK_RUNNING when this
1894 * routine returns.
1895 *
1896 * Returns 0 when the timer has expired otherwise -EINTR
1897 */
1898int __sched schedule_hrtimeout(ktime_t *expires,
1899                               const enum hrtimer_mode mode)
1900{
1901        return schedule_hrtimeout_range(expires, 0, mode);
1902}
1903EXPORT_SYMBOL_GPL(schedule_hrtimeout);
1904
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.