linux/kernel/timer.c History
<<
>>
Prefs
   1/*
   2 *  linux/kernel/timer.c
   3 *
   4 *  Kernel internal timers, basic process system calls
   5 *
   6 *  Copyright (C) 1991, 1992  Linus Torvalds
   7 *
   8 *  1997-01-28  Modified by Finn Arne Gangstad to make timers scale better.
   9 *
  10 *  1997-09-10  Updated NTP code according to technical memorandum Jan '96
  11 *              "A Kernel Model for Precision Timekeeping" by Dave Mills
  12 *  1998-12-24  Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
  13 *              serialize accesses to xtime/lost_ticks).
  14 *                              Copyright (C) 1998  Andrea Arcangeli
  15 *  1999-03-10  Improved NTP compatibility by Ulrich Windl
  16 *  2002-05-31  Move sys_sysinfo here and make its locking sane, Robert Love
  17 *  2000-10-05  Implemented scalable SMP per-CPU timer handling.
  18 *                              Copyright (C) 2000, 2001, 2002  Ingo Molnar
  19 *              Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
  20 */
  21
  22#include <linux/kernel_stat.h>
  23#include <linux/module.h>
  24#include <linux/interrupt.h>
  25#include <linux/percpu.h>
  26#include <linux/init.h>
  27#include <linux/mm.h>
  28#include <linux/swap.h>
  29#include <linux/pid_namespace.h>
  30#include <linux/notifier.h>
  31#include <linux/thread_info.h>
  32#include <linux/time.h>
  33#include <linux/jiffies.h>
  34#include <linux/posix-timers.h>
  35#include <linux/cpu.h>
  36#include <linux/syscalls.h>
  37#include <linux/delay.h>
  38#include <linux/tick.h>
  39#include <linux/kallsyms.h>
  40#include <linux/perf_event.h>
  41#include <linux/sched.h>
  42#include <linux/slab.h>
  43
  44#include <asm/uaccess.h>
  45#include <asm/unistd.h>
  46#include <asm/div64.h>
  47#include <asm/timex.h>
  48#include <asm/io.h>
  49
  50#define CREATE_TRACE_POINTS
  51#include <trace/events/timer.h>
  52
  53u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
  54
  55EXPORT_SYMBOL(jiffies_64);
  56
  57/*
  58 * per-CPU timer vector definitions:
  59 */
  60#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
  61#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
  62#define TVN_SIZE (1 << TVN_BITS)
  63#define TVR_SIZE (1 << TVR_BITS)
  64#define TVN_MASK (TVN_SIZE - 1)
  65#define TVR_MASK (TVR_SIZE - 1)
  66
  67struct tvec {
  68        struct list_head vec[TVN_SIZE];
  69};
  70
  71struct tvec_root {
  72        struct list_head vec[TVR_SIZE];
  73};
  74
  75struct tvec_base {
  76        spinlock_t lock;
  77        struct timer_list *running_timer;
  78        unsigned long timer_jiffies;
  79        unsigned long next_timer;
  80        struct tvec_root tv1;
  81        struct tvec tv2;
  82        struct tvec tv3;
  83        struct tvec tv4;
  84        struct tvec tv5;
  85} ____cacheline_aligned;
  86
  87struct tvec_base boot_tvec_bases;
  88EXPORT_SYMBOL(boot_tvec_bases);
  89static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
  90
  91/*
  92 * Note that all tvec_bases are 2 byte aligned and lower bit of
  93 * base in timer_list is guaranteed to be zero. Use the LSB for
  94 * the new flag to indicate whether the timer is deferrable
  95 */
  96#define TBASE_DEFERRABLE_FLAG           (0x1)
  97
  98/* Functions below help us manage 'deferrable' flag */
  99static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
 100{
 101        return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
 102}
 103
 104static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
 105{
 106        return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
 107}
 108
 109static inline void timer_set_deferrable(struct timer_list *timer)
 110{
 111        timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
 112                                       TBASE_DEFERRABLE_FLAG));
 113}
 114
 115static inline void
 116timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
 117{
 118        timer->base = (struct tvec_base *)((unsigned long)(new_base) |
 119                                      tbase_get_deferrable(timer->base));
 120}
 121
 122static unsigned long round_jiffies_common(unsigned long j, int cpu,
 123                bool force_up)
 124{
 125        int rem;
 126        unsigned long original = j;
 127
 128        /*
 129         * We don't want all cpus firing their timers at once hitting the
 130         * same lock or cachelines, so we skew each extra cpu with an extra
 131         * 3 jiffies. This 3 jiffies came originally from the mm/ code which
 132         * already did this.
 133         * The skew is done by adding 3*cpunr, then round, then subtract this
 134         * extra offset again.
 135         */
 136        j += cpu * 3;
 137
 138        rem = j % HZ;
 139
 140        /*
 141         * If the target jiffie is just after a whole second (which can happen
 142         * due to delays of the timer irq, long irq off times etc etc) then
 143         * we should round down to the whole second, not up. Use 1/4th second
 144         * as cutoff for this rounding as an extreme upper bound for this.
 145         * But never round down if @force_up is set.
 146         */
 147        if (rem < HZ/4 && !force_up) /* round down */
 148                j = j - rem;
 149        else /* round up */
 150                j = j - rem + HZ;
 151
 152        /* now that we have rounded, subtract the extra skew again */
 153        j -= cpu * 3;
 154
 155        if (j <= jiffies) /* rounding ate our timeout entirely; */
 156                return original;
 157        return j;
 158}
 159
 160/**
 161 * __round_jiffies - function to round jiffies to a full second
 162 * @j: the time in (absolute) jiffies that should be rounded
 163 * @cpu: the processor number on which the timeout will happen
 164 *
 165 * __round_jiffies() rounds an absolute time in the future (in jiffies)
 166 * up or down to (approximately) full seconds. This is useful for timers
 167 * for which the exact time they fire does not matter too much, as long as
 168 * they fire approximately every X seconds.
 169 *
 170 * By rounding these timers to whole seconds, all such timers will fire
 171 * at the same time, rather than at various times spread out. The goal
 172 * of this is to have the CPU wake up less, which saves power.
 173 *
 174 * The exact rounding is skewed for each processor to avoid all
 175 * processors firing at the exact same time, which could lead
 176 * to lock contention or spurious cache line bouncing.
 177 *
 178 * The return value is the rounded version of the @j parameter.
 179 */
 180unsigned long __round_jiffies(unsigned long j, int cpu)
 181{
 182        return round_jiffies_common(j, cpu, false);
 183}
 184EXPORT_SYMBOL_GPL(__round_jiffies);
 185
 186/**
 187 * __round_jiffies_relative - function to round jiffies to a full second
 188 * @j: the time in (relative) jiffies that should be rounded
 189 * @cpu: the processor number on which the timeout will happen
 190 *
 191 * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 192 * up or down to (approximately) full seconds. This is useful for timers
 193 * for which the exact time they fire does not matter too much, as long as
 194 * they fire approximately every X seconds.
 195 *
 196 * By rounding these timers to whole seconds, all such timers will fire
 197 * at the same time, rather than at various times spread out. The goal
 198 * of this is to have the CPU wake up less, which saves power.
 199 *
 200 * The exact rounding is skewed for each processor to avoid all
 201 * processors firing at the exact same time, which could lead
 202 * to lock contention or spurious cache line bouncing.
 203 *
 204 * The return value is the rounded version of the @j parameter.
 205 */
 206unsigned long __round_jiffies_relative(unsigned long j, int cpu)
 207{
 208        unsigned long j0 = jiffies;
 209
 210        /* Use j0 because jiffies might change while we run */
 211        return round_jiffies_common(j + j0, cpu, false) - j0;
 212}
 213EXPORT_SYMBOL_GPL(__round_jiffies_relative);
 214
 215/**
 216 * round_jiffies - function to round jiffies to a full second
 217 * @j: the time in (absolute) jiffies that should be rounded
 218 *
 219 * round_jiffies() rounds an absolute time in the future (in jiffies)
 220 * up or down to (approximately) full seconds. This is useful for timers
 221 * for which the exact time they fire does not matter too much, as long as
 222 * they fire approximately every X seconds.
 223 *
 224 * By rounding these timers to whole seconds, all such timers will fire
 225 * at the same time, rather than at various times spread out. The goal
 226 * of this is to have the CPU wake up less, which saves power.
 227 *
 228 * The return value is the rounded version of the @j parameter.
 229 */
 230unsigned long round_jiffies(unsigned long j)
 231{
 232        return round_jiffies_common(j, raw_smp_processor_id(), false);
 233}
 234EXPORT_SYMBOL_GPL(round_jiffies);
 235
 236/**
 237 * round_jiffies_relative - function to round jiffies to a full second
 238 * @j: the time in (relative) jiffies that should be rounded
 239 *
 240 * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 241 * up or down to (approximately) full seconds. This is useful for timers
 242 * for which the exact time they fire does not matter too much, as long as
 243 * they fire approximately every X seconds.
 244 *
 245 * By rounding these timers to whole seconds, all such timers will fire
 246 * at the same time, rather than at various times spread out. The goal
 247 * of this is to have the CPU wake up less, which saves power.
 248 *
 249 * The return value is the rounded version of the @j parameter.
 250 */
 251unsigned long round_jiffies_relative(unsigned long j)
 252{
 253        return __round_jiffies_relative(j, raw_smp_processor_id());
 254}
 255EXPORT_SYMBOL_GPL(round_jiffies_relative);
 256
 257/**
 258 * __round_jiffies_up - function to round jiffies up to a full second
 259 * @j: the time in (absolute) jiffies that should be rounded
 260 * @cpu: the processor number on which the timeout will happen
 261 *
 262 * This is the same as __round_jiffies() except that it will never
 263 * round down.  This is useful for timeouts for which the exact time
 264 * of firing does not matter too much, as long as they don't fire too
 265 * early.
 266 */
 267unsigned long __round_jiffies_up(unsigned long j, int cpu)
 268{
 269        return round_jiffies_common(j, cpu, true);
 270}
 271EXPORT_SYMBOL_GPL(__round_jiffies_up);
 272
 273/**
 274 * __round_jiffies_up_relative - function to round jiffies up to a full second
 275 * @j: the time in (relative) jiffies that should be rounded
 276 * @cpu: the processor number on which the timeout will happen
 277 *
 278 * This is the same as __round_jiffies_relative() except that it will never
 279 * round down.  This is useful for timeouts for which the exact time
 280 * of firing does not matter too much, as long as they don't fire too
 281 * early.
 282 */
 283unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
 284{
 285        unsigned long j0 = jiffies;
 286
 287        /* Use j0 because jiffies might change while we run */
 288        return round_jiffies_common(j + j0, cpu, true) - j0;
 289}
 290EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
 291
 292/**
 293 * round_jiffies_up - function to round jiffies up to a full second
 294 * @j: the time in (absolute) jiffies that should be rounded
 295 *
 296 * This is the same as round_jiffies() except that it will never
 297 * round down.  This is useful for timeouts for which the exact time
 298 * of firing does not matter too much, as long as they don't fire too
 299 * early.
 300 */
 301unsigned long round_jiffies_up(unsigned long j)
 302{
 303        return round_jiffies_common(j, raw_smp_processor_id(), true);
 304}
 305EXPORT_SYMBOL_GPL(round_jiffies_up);
 306
 307/**
 308 * round_jiffies_up_relative - function to round jiffies up to a full second
 309 * @j: the time in (relative) jiffies that should be rounded
 310 *
 311 * This is the same as round_jiffies_relative() except that it will never
 312 * round down.  This is useful for timeouts for which the exact time
 313 * of firing does not matter too much, as long as they don't fire too
 314 * early.
 315 */
 316unsigned long round_jiffies_up_relative(unsigned long j)
 317{
 318        return __round_jiffies_up_relative(j, raw_smp_processor_id());
 319}
 320EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
 321
 322
 323static inline void set_running_timer(struct tvec_base *base,
 324                                        struct timer_list *timer)
 325{
 326#ifdef CONFIG_SMP
 327        base->running_timer = timer;
 328#endif
 329}
 330
 331static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 332{
 333        unsigned long expires = timer->expires;
 334        unsigned long idx = expires - base->timer_jiffies;
 335        struct list_head *vec;
 336
 337        if (idx < TVR_SIZE) {
 338                int i = expires & TVR_MASK;
 339                vec = base->tv1.vec + i;
 340        } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
 341                int i = (expires >> TVR_BITS) & TVN_MASK;
 342                vec = base->tv2.vec + i;
 343        } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
 344                int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
 345                vec = base->tv3.vec + i;
 346        } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
 347                int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
 348                vec = base->tv4.vec + i;
 349        } else if ((signed long) idx < 0) {
 350                /*
 351                 * Can happen if you add a timer with expires == jiffies,
 352                 * or you set a timer to go off in the past
 353                 */
 354                vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
 355        } else {
 356                int i;
 357                /* If the timeout is larger than 0xffffffff on 64-bit
 358                 * architectures then we use the maximum timeout:
 359                 */
 360                if (idx > 0xffffffffUL) {
 361                        idx = 0xffffffffUL;
 362                        expires = idx + base->timer_jiffies;
 363                }
 364                i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
 365                vec = base->tv5.vec + i;
 366        }
 367        /*
 368         * Timers are FIFO:
 369         */
 370        list_add_tail(&timer->entry, vec);
 371}
 372
 373#ifdef CONFIG_TIMER_STATS
 374void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
 375{
 376        if (timer->start_site)
 377                return;
 378
 379        timer->start_site = addr;
 380        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 381        timer->start_pid = current->pid;
 382}
 383
 384static void timer_stats_account_timer(struct timer_list *timer)
 385{
 386        unsigned int flag = 0;
 387
 388        if (likely(!timer->start_site))
 389                return;
 390        if (unlikely(tbase_get_deferrable(timer->base)))
 391                flag |= TIMER_STATS_FLAG_DEFERRABLE;
 392
 393        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 394                                 timer->function, timer->start_comm, flag);
 395}
 396
 397#else
 398static void timer_stats_account_timer(struct timer_list *timer) {}
 399#endif
 400
 401#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 402
 403static struct debug_obj_descr timer_debug_descr;
 404
 405/*
 406 * fixup_init is called when:
 407 * - an active object is initialized
 408 */
 409static int timer_fixup_init(void *addr, enum debug_obj_state state)
 410{
 411        struct timer_list *timer = addr;
 412
 413        switch (state) {
 414        case ODEBUG_STATE_ACTIVE:
 415                del_timer_sync(timer);
 416                debug_object_init(timer, &timer_debug_descr);
 417                return 1;
 418        default:
 419                return 0;
 420        }
 421}
 422
 423/*
 424 * fixup_activate is called when:
 425 * - an active object is activated
 426 * - an unknown object is activated (might be a statically initialized object)
 427 */
 428static int timer_fixup_activate(void *addr, enum debug_obj_state state)
 429{
 430        struct timer_list *timer = addr;
 431
 432        switch (state) {
 433
 434        case ODEBUG_STATE_NOTAVAILABLE:
 435                /*
 436                 * This is not really a fixup. The timer was
 437                 * statically initialized. We just make sure that it
 438                 * is tracked in the object tracker.
 439                 */
 440                if (timer->entry.next == NULL &&
 441                    timer->entry.prev == TIMER_ENTRY_STATIC) {
 442                        debug_object_init(timer, &timer_debug_descr);
 443                        debug_object_activate(timer, &timer_debug_descr);
 444                        return 0;
 445                } else {
 446                        WARN_ON_ONCE(1);
 447                }
 448                return 0;
 449
 450        case ODEBUG_STATE_ACTIVE:
 451                WARN_ON(1);
 452
 453        default:
 454                return 0;
 455        }
 456}
 457
 458/*
 459 * fixup_free is called when:
 460 * - an active object is freed
 461 */
 462static int timer_fixup_free(void *addr, enum debug_obj_state state)
 463{
 464        struct timer_list *timer = addr;
 465
 466        switch (state) {
 467        case ODEBUG_STATE_ACTIVE:
 468                del_timer_sync(timer);
 469                debug_object_free(timer, &timer_debug_descr);
 470                return 1;
 471        default:
 472                return 0;
 473        }
 474}
 475
 476static struct debug_obj_descr timer_debug_descr = {
 477        .name           = "timer_list",
 478        .fixup_init     = timer_fixup_init,
 479        .fixup_activate = timer_fixup_activate,
 480        .fixup_free     = timer_fixup_free,
 481};
 482
 483static inline void debug_timer_init(struct timer_list *timer)
 484{
 485        debug_object_init(timer, &timer_debug_descr);
 486}
 487
 488static inline void debug_timer_activate(struct timer_list *timer)
 489{
 490        debug_object_activate(timer, &timer_debug_descr);
 491}
 492
 493static inline void debug_timer_deactivate(struct timer_list *timer)
 494{
 495        debug_object_deactivate(timer, &timer_debug_descr);
 496}
 497
 498static inline void debug_timer_free(struct timer_list *timer)
 499{
 500        debug_object_free(timer, &timer_debug_descr);
 501}
 502
 503static void __init_timer(struct timer_list *timer,
 504                         const char *name,
 505                         struct lock_class_key *key);
 506
 507void init_timer_on_stack_key(struct timer_list *timer,
 508                             const char *name,
 509                             struct lock_class_key *key)
 510{
 511        debug_object_init_on_stack(timer, &timer_debug_descr);
 512        __init_timer(timer, name, key);
 513}
 514EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
 515
 516void destroy_timer_on_stack(struct timer_list *timer)
 517{
 518        debug_object_free(timer, &timer_debug_descr);
 519}
 520EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
 521
 522#else
 523static inline void debug_timer_init(struct timer_list *timer) { }
 524static inline void debug_timer_activate(struct timer_list *timer) { }
 525static inline void debug_timer_deactivate(struct timer_list *timer) { }
 526#endif
 527
 528static inline void debug_init(struct timer_list *timer)
 529{
 530        debug_timer_init(timer);
 531        trace_timer_init(timer);
 532}
 533
 534static inline void
 535debug_activate(struct timer_list *timer, unsigned long expires)
 536{
 537        debug_timer_activate(timer);
 538        trace_timer_start(timer, expires);
 539}
 540
 541static inline void debug_deactivate(struct timer_list *timer)
 542{
 543        debug_timer_deactivate(timer);
 544        trace_timer_cancel(timer);
 545}
 546
 547static void __init_timer(struct timer_list *timer,
 548                         const char *name,
 549                         struct lock_class_key *key)
 550{
 551        timer->entry.next = NULL;
 552        timer->base = __raw_get_cpu_var(tvec_bases);
 553#ifdef CONFIG_TIMER_STATS
 554        timer->start_site = NULL;
 555        timer->start_pid = -1;
 556        memset(timer->start_comm, 0, TASK_COMM_LEN);
 557#endif
 558        lockdep_init_map(&timer->lockdep_map, name, key, 0);
 559}
 560
 561/**
 562 * init_timer_key - initialize a timer
 563 * @timer: the timer to be initialized
 564 * @name: name of the timer
 565 * @key: lockdep class key of the fake lock used for tracking timer
 566 *       sync lock dependencies
 567 *
 568 * init_timer_key() must be done to a timer prior calling *any* of the
 569 * other timer functions.
 570 */
 571void init_timer_key(struct timer_list *timer,
 572                    const char *name,
 573                    struct lock_class_key *key)
 574{
 575        debug_init(timer);
 576        __init_timer(timer, name, key);
 577}
 578EXPORT_SYMBOL(init_timer_key);
 579
 580void init_timer_deferrable_key(struct timer_list *timer,
 581                               const char *name,
 582                               struct lock_class_key *key)
 583{
 584        init_timer_key(timer, name, key);
 585        timer_set_deferrable(timer);
 586}
 587EXPORT_SYMBOL(init_timer_deferrable_key);
 588
 589static inline void detach_timer(struct timer_list *timer,
 590                                int clear_pending)
 591{
 592        struct list_head *entry = &timer->entry;
 593
 594        debug_deactivate(timer);
 595
 596        __list_del(entry->prev, entry->next);
 597        if (clear_pending)
 598                entry->next = NULL;
 599        entry->prev = LIST_POISON2;
 600}
 601
 602/*
 603 * We are using hashed locking: holding per_cpu(tvec_bases).lock
 604 * means that all timers which are tied to this base via timer->base are
 605 * locked, and the base itself is locked too.
 606 *
 607 * So __run_timers/migrate_timers can safely modify all timers which could
 608 * be found on ->tvX lists.
 609 *
 610 * When the timer's base is locked, and the timer removed from list, it is
 611 * possible to set timer->base = NULL and drop the lock: the timer remains
 612 * locked.
 613 */
 614static struct tvec_base *lock_timer_base(struct timer_list *timer,
 615                                        unsigned long *flags)
 616        __acquires(timer->base->lock)
 617{
 618        struct tvec_base *base;
 619
 620        for (;;) {
 621                struct tvec_base *prelock_base = timer->base;
 622                base = tbase_get_base(prelock_base);
 623                if (likely(base != NULL)) {
 624                        spin_lock_irqsave(&base->lock, *flags);
 625                        if (likely(prelock_base == timer->base))
 626                                return base;
 627                        /* The timer has migrated to another CPU */
 628                        spin_unlock_irqrestore(&base->lock, *flags);
 629                }
 630                cpu_relax();
 631        }
 632}
 633
 634static inline int
 635__mod_timer(struct timer_list *timer, unsigned long expires,
 636                                                bool pending_only, int pinned)
 637{
 638        struct tvec_base *base, *new_base;
 639        unsigned long flags;
 640        int ret = 0 , cpu;
 641
 642        timer_stats_timer_set_start_info(timer);
 643        BUG_ON(!timer->function);
 644
 645        base = lock_timer_base(timer, &flags);
 646
 647        if (timer_pending(timer)) {
 648                detach_timer(timer, 0);
 649                if (timer->expires == base->next_timer &&
 650                    !tbase_get_deferrable(timer->base))
 651                        base->next_timer = base->timer_jiffies;
 652                ret = 1;
 653        } else {
 654                if (pending_only)
 655                        goto out_unlock;
 656        }
 657
 658        debug_activate(timer, expires);
 659
 660        cpu = smp_processor_id();
 661
 662#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
 663        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
 664                int preferred_cpu = get_nohz_load_balancer();
 665
 666                if (preferred_cpu >= 0)
 667                        cpu = preferred_cpu;
 668        }
 669#endif
 670        new_base = per_cpu(tvec_bases, cpu);
 671
 672        if (base != new_base) {
 673                /*
 674                 * We are trying to schedule the timer on the local CPU.
 675                 * However we can't change timer's base while it is running,
 676                 * otherwise del_timer_sync() can't detect that the timer's
 677                 * handler yet has not finished. This also guarantees that
 678                 * the timer is serialized wrt itself.
 679                 */
 680                if (likely(base->running_timer != timer)) {
 681                        /* See the comment in lock_timer_base() */
 682                        timer_set_base(timer, NULL);
 683                        spin_unlock(&base->lock);
 684                        base = new_base;
 685                        spin_lock(&base->lock);
 686                        timer_set_base(timer, base);
 687                }
 688        }
 689
 690        timer->expires = expires;
 691        if (time_before(timer->expires, base->next_timer) &&
 692            !tbase_get_deferrable(timer->base))
 693                base->next_timer = timer->expires;
 694        internal_add_timer(base, timer);
 695
 696out_unlock:
 697        spin_unlock_irqrestore(&base->lock, flags);
 698
 699        return ret;
 700}
 701
 702/**
 703 * mod_timer_pending - modify a pending timer's timeout
 704 * @timer: the pending timer to be modified
 705 * @expires: new timeout in jiffies
 706 *
 707 * mod_timer_pending() is the same for pending timers as mod_timer(),
 708 * but will not re-activate and modify already deleted timers.
 709 *
 710 * It is useful for unserialized use of timers.
 711 */
 712int mod_timer_pending(struct timer_list *timer, unsigned long expires)
 713{
 714        return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
 715}
 716EXPORT_SYMBOL(mod_timer_pending);
 717
 718/**
 719 * mod_timer - modify a timer's timeout
 720 * @timer: the timer to be modified
 721 * @expires: new timeout in jiffies
 722 *
 723 * mod_timer() is a more efficient way to update the expire field of an
 724 * active timer (if the timer is inactive it will be activated)
 725 *
 726 * mod_timer(timer, expires) is equivalent to:
 727 *
 728 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 729 *
 730 * Note that if there are multiple unserialized concurrent users of the
 731 * same timer, then mod_timer() is the only safe way to modify the timeout,
 732 * since add_timer() cannot modify an already running timer.
 733 *
 734 * The function returns whether it has modified a pending timer or not.
 735 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
 736 * active timer returns 1.)
 737 */
 738int mod_timer(struct timer_list *timer, unsigned long expires)
 739{
 740        /*
 741         * This is a common optimization triggered by the
 742         * networking code - if the timer is re-modified
 743         * to be the same thing then just return:
 744         */
 745        if (timer_pending(timer) && timer->expires == expires)
 746                return 1;
 747
 748        return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
 749}
 750EXPORT_SYMBOL(mod_timer);
 751
 752/**
 753 * mod_timer_pinned - modify a timer's timeout
 754 * @timer: the timer to be modified
 755 * @expires: new timeout in jiffies
 756 *
 757 * mod_timer_pinned() is a way to update the expire field of an
 758 * active timer (if the timer is inactive it will be activated)
 759 * and not allow the timer to be migrated to a different CPU.
 760 *
 761 * mod_timer_pinned(timer, expires) is equivalent to:
 762 *
 763 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 764 */
 765int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
 766{
 767        if (timer->expires == expires && timer_pending(timer))
 768                return 1;
 769
 770        return __mod_timer(timer, expires, false, TIMER_PINNED);
 771}
 772EXPORT_SYMBOL(mod_timer_pinned);
 773
 774/**
 775 * add_timer - start a timer
 776 * @timer: the timer to be added
 777 *
 778 * The kernel will do a ->function(->data) callback from the
 779 * timer interrupt at the ->expires point in the future. The
 780 * current time is 'jiffies'.
 781 *
 782 * The timer's ->expires, ->function (and if the handler uses it, ->data)
 783 * fields must be set prior calling this function.
 784 *
 785 * Timers with an ->expires field in the past will be executed in the next
 786 * timer tick.
 787 */
 788void add_timer(struct timer_list *timer)
 789{
 790        BUG_ON(timer_pending(timer));
 791        mod_timer(timer, timer->expires);
 792}
 793EXPORT_SYMBOL(add_timer);
 794
 795/**
 796 * add_timer_on - start a timer on a particular CPU
 797 * @timer: the timer to be added
 798 * @cpu: the CPU to start it on
 799 *
 800 * This is not very scalable on SMP. Double adds are not possible.
 801 */
 802void add_timer_on(struct timer_list *timer, int cpu)
 803{
 804        struct tvec_base *base = per_cpu(tvec_bases, cpu);
 805        unsigned long flags;
 806
 807        timer_stats_timer_set_start_info(timer);
 808        BUG_ON(timer_pending(timer) || !timer->function);
 809        spin_lock_irqsave(&base->lock, flags);
 810        timer_set_base(timer, base);
 811        debug_activate(timer, timer->expires);
 812        if (time_before(timer->expires, base->next_timer) &&
 813            !tbase_get_deferrable(timer->base))
 814                base->next_timer = timer->expires;
 815        internal_add_timer(base, timer);
 816        /*
 817         * Check whether the other CPU is idle and needs to be
 818         * triggered to reevaluate the timer wheel when nohz is
 819         * active. We are protected against the other CPU fiddling
 820         * with the timer by holding the timer base lock. This also
 821         * makes sure that a CPU on the way to idle can not evaluate
 822         * the timer wheel.
 823         */
 824        wake_up_idle_cpu(cpu);
 825        spin_unlock_irqrestore(&base->lock, flags);
 826}
 827EXPORT_SYMBOL_GPL(add_timer_on);
 828
 829/**
 830 * del_timer - deactive a timer.
 831 * @timer: the timer to be deactivated
 832 *
 833 * del_timer() deactivates a timer - this works on both active and inactive
 834 * timers.
 835 *
 836 * The function returns whether it has deactivated a pending timer or not.
 837 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
 838 * active timer returns 1.)
 839 */
 840int del_timer(struct timer_list *timer)
 841{
 842        struct tvec_base *base;
 843        unsigned long flags;
 844        int ret = 0;
 845
 846        timer_stats_timer_clear_start_info(timer);
 847        if (timer_pending(timer)) {
 848                base = lock_timer_base(timer, &flags);
 849                if (timer_pending(timer)) {
 850                        detach_timer(timer, 1);
 851                        if (timer->expires == base->next_timer &&
 852                            !tbase_get_deferrable(timer->base))
 853                                base->next_timer = base->timer_jiffies;
 854                        ret = 1;
 855                }
 856                spin_unlock_irqrestore(&base->lock, flags);
 857        }
 858
 859        return ret;
 860}
 861EXPORT_SYMBOL(del_timer);
 862
 863#ifdef CONFIG_SMP
 864/**
 865 * try_to_del_timer_sync - Try to deactivate a timer
 866 * @timer: timer do del
 867 *
 868 * This function tries to deactivate a timer. Upon successful (ret >= 0)
 869 * exit the timer is not queued and the handler is not running on any CPU.
 870 *
 871 * It must not be called from interrupt contexts.
 872 */
 873int try_to_del_timer_sync(struct timer_list *timer)
 874{
 875        struct tvec_base *base;
 876        unsigned long flags;
 877        int ret = -1;
 878
 879        base = lock_timer_base(timer, &flags);
 880
 881        if (base->running_timer == timer)
 882                goto out;
 883
 884        timer_stats_timer_clear_start_info(timer);
 885        ret = 0;
 886        if (timer_pending(timer)) {
 887                detach_timer(timer, 1);
 888                if (timer->expires == base->next_timer &&
 889                    !tbase_get_deferrable(timer->base))
 890                        base->next_timer = base->timer_jiffies;
 891                ret = 1;
 892        }
 893out:
 894        spin_unlock_irqrestore(&base->lock, flags);
 895
 896        return ret;
 897}
 898EXPORT_SYMBOL(try_to_del_timer_sync);
 899
 900/**
 901 * del_timer_sync - deactivate a timer and wait for the handler to finish.
 902 * @timer: the timer to be deactivated
 903 *
 904 * This function only differs from del_timer() on SMP: besides deactivating
 905 * the timer it also makes sure the handler has finished executing on other
 906 * CPUs.
 907 *
 908 * Synchronization rules: Callers must prevent restarting of the timer,
 909 * otherwise this function is meaningless. It must not be called from
 910 * interrupt contexts. The caller must not hold locks which would prevent
 911 * completion of the timer's handler. The timer's handler must not call
 912 * add_timer_on(). Upon exit the timer is not queued and the handler is
 913 * not running on any CPU.
 914 *
 915 * The function returns whether it has deactivated a pending timer or not.
 916 */
 917int del_timer_sync(struct timer_list *timer)
 918{
 919#ifdef CONFIG_LOCKDEP
 920        unsigned long flags;
 921
 922        local_irq_save(flags);
 923        lock_map_acquire(&timer->lockdep_map);
 924        lock_map_release(&timer->lockdep_map);
 925        local_irq_restore(flags);
 926#endif
 927
 928        for (;;) {
 929                int ret = try_to_del_timer_sync(timer);
 930                if (ret >= 0)
 931                        return ret;
 932                cpu_relax();
 933        }
 934}
 935EXPORT_SYMBOL(del_timer_sync);
 936#endif
 937
 938static int cascade(struct tvec_base *base, struct tvec *tv, int index)
 939{
 940        /* cascade all the timers from tv up one level */
 941        struct timer_list *timer, *tmp;
 942        struct list_head tv_list;
 943
 944        list_replace_init(tv->vec + index, &tv_list);
 945
 946        /*
 947         * We are removing _all_ timers from the list, so we
 948         * don't have to detach them individually.
 949         */
 950        list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
 951                BUG_ON(tbase_get_base(timer->base) != base);
 952                internal_add_timer(base, timer);
 953        }
 954
 955        return index;
 956}
 957
 958#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
 959
 960/**
 961 * __run_timers - run all expired timers (if any) on this CPU.
 962 * @base: the timer vector to be processed.
 963 *
 964 * This function cascades all vectors and executes all expired timer
 965 * vectors.
 966 */
 967static inline void __run_timers(struct tvec_base *base)
 968{
 969        struct timer_list *timer;
 970
 971        spin_lock_irq(&base->lock);
 972        while (time_after_eq(jiffies, base->timer_jiffies)) {
 973                struct list_head work_list;
 974                struct list_head *head = &work_list;
 975                int index = base->timer_jiffies & TVR_MASK;
 976
 977                /*
 978                 * Cascade timers:
 979                 */
 980                if (!index &&
 981                        (!cascade(base, &base->tv2, INDEX(0))) &&
 982                                (!cascade(base, &base->tv3, INDEX(1))) &&
 983                                        !cascade(base, &base->tv4, INDEX(2)))
 984                        cascade(base, &base->tv5, INDEX(3));
 985                ++base->timer_jiffies;
 986                list_replace_init(base->tv1.vec + index, &work_list);
 987                while (!list_empty(head)) {
 988                        void (*fn)(unsigned long);
 989                        unsigned long data;
 990
 991                        timer = list_first_entry(head, struct timer_list,entry);
 992                        fn = timer->function;
 993                        data = timer->data;
 994
 995                        timer_stats_account_timer(timer);
 996
 997                        set_running_timer(base, timer);
 998                        detach_timer(timer, 1);
 999
1000                        spin_unlock_irq(&base->lock);
1001                        {
1002                                int preempt_count = preempt_count();
1003
1004#ifdef CONFIG_LOCKDEP
1005                                /*
1006                                 * It is permissible to free the timer from
1007                                 * inside the function that is called from
1008                                 * it, this we need to take into account for
1009                                 * lockdep too. To avoid bogus "held lock
1010                                 * freed" warnings as well as problems when
1011                                 * looking into timer->lockdep_map, make a
1012                                 * copy and use that here.
1013                                 */
1014                                struct lockdep_map lockdep_map =
1015                                        timer->lockdep_map;
1016#endif
1017                                /*
1018                                 * Couple the lock chain with the lock chain at
1019                                 * del_timer_sync() by acquiring the lock_map
1020                                 * around the fn() call here and in
1021                                 * del_timer_sync().
1022                                 */
1023                                lock_map_acquire(&lockdep_map);
1024
1025                                trace_timer_expire_entry(timer);
1026                                fn(data);
1027                                trace_timer_expire_exit(timer);
1028
1029                                lock_map_release(&lockdep_map);
1030
1031                                if (preempt_count != preempt_count()) {
1032                                        printk(KERN_ERR "huh, entered %p "
1033                                               "with preempt_count %08x, exited"
1034                                               " with %08x?\n",
1035                                               fn, preempt_count,
1036                                               preempt_count());
1037                                        BUG();
1038                                }
1039                        }
1040                        spin_lock_irq(&base->lock);
1041                }
1042        }
1043        set_running_timer(base, NULL);
1044        spin_unlock_irq(&base->lock);
1045}
1046
1047#ifdef CONFIG_NO_HZ
1048/*
1049 * Find out when the next timer event is due to happen. This
1050 * is used on S/390 to stop all activity when a CPU is idle.
1051 * This function needs to be called with interrupts disabled.
1052 */
1053static unsigned long __next_timer_interrupt(struct tvec_base *base)
1054{
1055        unsigned long timer_jiffies = base->timer_jiffies;
1056        unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1057        int index, slot, array, found = 0;
1058        struct timer_list *nte;
1059        struct tvec *varray[4];
1060
1061        /* Look for timer events in tv1. */
1062        index = slot = timer_jiffies & TVR_MASK;
1063        do {
1064                list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1065                        if (tbase_get_deferrable(nte->base))
1066                                continue;
1067
1068                        found = 1;
1069                        expires = nte->expires;
1070                        /* Look at the cascade bucket(s)? */
1071                        if (!index || slot < index)
1072                                goto cascade;
1073                        return expires;
1074                }
1075                slot = (slot + 1) & TVR_MASK;
1076        } while (slot != index);
1077
1078cascade:
1079        /* Calculate the next cascade event */
1080        if (index)
1081                timer_jiffies += TVR_SIZE - index;
1082        timer_jiffies >>= TVR_BITS;
1083
1084        /* Check tv2-tv5. */
1085        varray[0] = &base->tv2;
1086        varray[1] = &base->tv3;
1087        varray[2] = &base->tv4;
1088        varray[3] = &base->tv5;
1089
1090        for (array = 0; array < 4; array++) {
1091                struct tvec *varp = varray[array];
1092
1093                index = slot = timer_jiffies & TVN_MASK;
1094                do {
1095                        list_for_each_entry(nte, varp->vec + slot, entry) {
1096                                if (tbase_get_deferrable(nte->base))
1097                                        continue;
1098
1099                                found = 1;
1100                                if (time_before(nte->expires, expires))
1101                                        expires = nte->expires;
1102                        }
1103                        /*
1104                         * Do we still search for the first timer or are
1105                         * we looking up the cascade buckets ?
1106                         */
1107                        if (found) {
1108                                /* Look at the cascade bucket(s)? */
1109                                if (!index || slot < index)
1110                                        break;
1111                                return expires;
1112                        }
1113                        slot = (slot + 1) & TVN_MASK;
1114                } while (slot != index);
1115
1116                if (index)
1117                        timer_jiffies += TVN_SIZE - index;
1118                timer_jiffies >>= TVN_BITS;
1119        }
1120        return expires;
1121}
1122
1123/*
1124 * Check, if the next hrtimer event is before the next timer wheel
1125 * event:
1126 */
1127static unsigned long cmp_next_hrtimer_event(unsigned long now,
1128                                            unsigned long expires)
1129{
1130        ktime_t hr_delta = hrtimer_get_next_event();
1131        struct timespec tsdelta;
1132        unsigned long delta;
1133
1134        if (hr_delta.tv64 == KTIME_MAX)
1135                return expires;
1136
1137        /*
1138         * Expired timer available, let it expire in the next tick
1139         */
1140        if (hr_delta.tv64 <= 0)
1141                return now + 1;
1142
1143        tsdelta = ktime_to_timespec(hr_delta);
1144        delta = timespec_to_jiffies(&tsdelta);
1145
1146        /*
1147         * Limit the delta to the max value, which is checked in
1148         * tick_nohz_stop_sched_tick():
1149         */
1150        if (delta > NEXT_TIMER_MAX_DELTA)
1151                delta = NEXT_TIMER_MAX_DELTA;
1152
1153        /*
1154         * Take rounding errors in to account and make sure, that it
1155         * expires in the next tick. Otherwise we go into an endless
1156         * ping pong due to tick_nohz_stop_sched_tick() retriggering
1157         * the timer softirq
1158         */
1159        if (delta < 1)
1160                delta = 1;
1161        now += delta;
1162        if (time_before(now, expires))
1163                return now;
1164        return expires;
1165}
1166
1167/**
1168 * get_next_timer_interrupt - return the jiffy of the next pending timer
1169 * @now: current time (in jiffies)
1170 */
1171unsigned long get_next_timer_interrupt(unsigned long now)
1172{
1173        struct tvec_base *base = __get_cpu_var(tvec_bases);
1174        unsigned long expires;
1175
1176        spin_lock(&base->lock);
1177        if (time_before_eq(base->next_timer, base->timer_jiffies))
1178                base->next_timer = __next_timer_interrupt(base);
1179        expires = base->next_timer;
1180        spin_unlock(&base->lock);
1181
1182        if (time_before_eq(expires, now))
1183                return now;
1184
1185        return cmp_next_hrtimer_event(now, expires);
1186}
1187#endif
1188
1189/*
1190 * Called from the timer interrupt handler to charge one tick to the current
1191 * process.  user_tick is 1 if the tick is user time, 0 for system.
1192 */
1193void update_process_times(int user_tick)
1194{
1195        struct task_struct *p = current;
1196        int cpu = smp_processor_id();
1197
1198        /* Note: this timer irq context must be accounted for as well. */
1199        account_process_tick(p, user_tick);
1200        run_local_timers();
1201        rcu_check_callbacks(cpu, user_tick);
1202        printk_tick();
1203        perf_event_do_pending();
1204        scheduler_tick();
1205        run_posix_cpu_timers(p);
1206}
1207
1208/*
1209 * This function runs timers and the timer-tq in bottom half context.
1210 */
1211static void run_timer_softirq(struct softirq_action *h)
1212{
1213        struct tvec_base *base = __get_cpu_var(tvec_bases);
1214
1215        hrtimer_run_pending();
1216
1217        if (time_after_eq(jiffies, base->timer_jiffies))
1218                __run_timers(base);
1219}
1220
1221/*
1222 * Called by the local, per-CPU timer interrupt on SMP.
1223 */
1224void run_local_timers(void)
1225{
1226        hrtimer_run_queues();
1227        raise_softirq(TIMER_SOFTIRQ);
1228        softlockup_tick();
1229}
1230
1231/*
1232 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1233 * without sampling the sequence number in xtime_lock.
1234 * jiffies is defined in the linker script...
1235 */
1236
1237void do_timer(unsigned long ticks)
1238{
1239        jiffies_64 += ticks;
1240        update_wall_time();
1241        calc_global_load();
1242}
1243
1244#ifdef __ARCH_WANT_SYS_ALARM
1245
1246/*
1247 * For backwards compatibility?  This can be done in libc so Alpha
1248 * and all newer ports shouldn't need it.
1249 */
1250SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1251{
1252        return alarm_setitimer(seconds);
1253}
1254
1255#endif
1256
1257#ifndef __alpha__
1258
1259/*
1260 * The Alpha uses getxpid, getxuid, and getxgid instead.  Maybe this
1261 * should be moved into arch/i386 instead?
1262 */
1263
1264/**
1265 * sys_getpid - return the thread group id of the current process
1266 *
1267 * Note, despite the name, this returns the tgid not the pid.  The tgid and
1268 * the pid are identical unless CLONE_THREAD was specified on clone() in
1269 * which case the tgid is the same in all threads of the same group.
1270 *
1271 * This is SMP safe as current->tgid does not change.
1272 */
1273SYSCALL_DEFINE0(getpid)
1274{
1275        return task_tgid_vnr(current);
1276}
1277
1278/*
1279 * Accessing ->real_parent is not SMP-safe, it could
1280 * change from under us. However, we can use a stale
1281 * value of ->real_parent under rcu_read_lock(), see
1282 * release_task()->call_rcu(delayed_put_task_struct).
1283 */
1284SYSCALL_DEFINE0(getppid)
1285{
1286        int pid;
1287
1288        rcu_read_lock();
1289        pid = task_tgid_vnr(current->real_parent);
1290        rcu_read_unlock();
1291
1292        return pid;
1293}
1294
1295SYSCALL_DEFINE0(getuid)
1296{
1297        /* Only we change this so SMP safe */
1298        return current_uid();
1299}
1300
1301SYSCALL_DEFINE0(geteuid)
1302{
1303        /* Only we change this so SMP safe */
1304        return current_euid();
1305}
1306
1307SYSCALL_DEFINE0(getgid)
1308{
1309        /* Only we change this so SMP safe */
1310        return current_gid();
1311}
1312
1313SYSCALL_DEFINE0(getegid)
1314{
1315        /* Only we change this so SMP safe */
1316        return  current_egid();
1317}
1318
1319#endif
1320
1321static void process_timeout(unsigned long __data)
1322{
1323        wake_up_process((struct task_struct *)__data);
1324}
1325
1326/**
1327 * schedule_timeout - sleep until timeout
1328 * @timeout: timeout value in jiffies
1329 *
1330 * Make the current task sleep until @timeout jiffies have
1331 * elapsed. The routine will return immediately unless
1332 * the current task state has been set (see set_current_state()).
1333 *
1334 * You can set the task state as follows -
1335 *
1336 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1337 * pass before the routine returns. The routine will return 0
1338 *
1339 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1340 * delivered to the current task. In this case the remaining time
1341 * in jiffies will be returned, or 0 if the timer expired in time
1342 *
1343 * The current task state is guaranteed to be TASK_RUNNING when this
1344 * routine returns.
1345 *
1346 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1347 * the CPU away without a bound on the timeout. In this case the return
1348 * value will be %MAX_SCHEDULE_TIMEOUT.
1349 *
1350 * In all cases the return value is guaranteed to be non-negative.
1351 */
1352signed long __sched schedule_timeout(signed long timeout)
1353{
1354        struct timer_list timer;
1355        unsigned long expire;
1356
1357        switch (timeout)
1358        {
1359        case MAX_SCHEDULE_TIMEOUT:
1360                /*
1361                 * These two special cases are useful to be comfortable
1362                 * in the caller. Nothing more. We could take
1363                 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1364                 * but I' d like to return a valid offset (>=0) to allow
1365                 * the caller to do everything it want with the retval.
1366                 */
1367                schedule();
1368                goto out;
1369        default:
1370                /*
1371                 * Another bit of PARANOID. Note that the retval will be
1372                 * 0 since no piece of kernel is supposed to do a check
1373                 * for a negative retval of schedule_timeout() (since it
1374                 * should never happens anyway). You just have the printk()
1375                 * that will tell you if something is gone wrong and where.
1376                 */
1377                if (timeout < 0) {
1378                        printk(KERN_ERR "schedule_timeout: wrong timeout "
1379                                "value %lx\n", timeout);
1380                        dump_stack();
1381                        current->state = TASK_RUNNING;
1382                        goto out;
1383                }
1384        }
1385
1386        expire = timeout + jiffies;
1387
1388        setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1389        __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1390        schedule();
1391        del_singleshot_timer_sync(&timer);
1392
1393        /* Remove the timer from the object tracker */
1394        destroy_timer_on_stack(&timer);
1395
1396        timeout = expire - jiffies;
1397
1398 out:
1399        return timeout < 0 ? 0 : timeout;
1400}
1401EXPORT_SYMBOL(schedule_timeout);
1402
1403/*
1404 * We can use __set_current_state() here because schedule_timeout() calls
1405 * schedule() unconditionally.
1406 */
1407signed long __sched schedule_timeout_interruptible(signed long timeout)
1408{
1409        __set_current_state(TASK_INTERRUPTIBLE);
1410        return schedule_timeout(timeout);
1411}
1412EXPORT_SYMBOL(schedule_timeout_interruptible);
1413
1414signed long __sched schedule_timeout_killable(signed long timeout)
1415{
1416        __set_current_state(TASK_KILLABLE);
1417        return schedule_timeout(timeout);
1418}
1419EXPORT_SYMBOL(schedule_timeout_killable);
1420
1421signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1422{
1423        __set_current_state(TASK_UNINTERRUPTIBLE);
1424        return schedule_timeout(timeout);
1425}
1426EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1427
1428/* Thread ID - the internal kernel "pid" */
1429SYSCALL_DEFINE0(gettid)
1430{
1431        return task_pid_vnr(current);
1432}
1433
1434/**
1435 * do_sysinfo - fill in sysinfo struct
1436 * @info: pointer to buffer to fill
1437 */
1438int do_sysinfo(struct sysinfo *info)
1439{
1440        unsigned long mem_total, sav_total;
1441        unsigned int mem_unit, bitcount;
1442        struct timespec tp;
1443
1444        memset(info, 0, sizeof(struct sysinfo));
1445
1446        ktime_get_ts(&tp);
1447        monotonic_to_bootbased(&tp);
1448        info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1449
1450        get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1451
1452        info->procs = nr_threads;
1453
1454        si_meminfo(info);
1455        si_swapinfo(info);
1456
1457        /*
1458         * If the sum of all the available memory (i.e. ram + swap)
1459         * is less than can be stored in a 32 bit unsigned long then
1460         * we can be binary compatible with 2.2.x kernels.  If not,
1461         * well, in that case 2.2.x was broken anyways...
1462         *
1463         *  -Erik Andersen <andersee@debian.org>
1464         */
1465
1466        mem_total = info->totalram + info->totalswap;
1467        if (mem_total < info->totalram || mem_total < info->totalswap)
1468                goto out;
1469        bitcount = 0;
1470        mem_unit = info->mem_unit;
1471        while (mem_unit > 1) {
1472                bitcount++;
1473                mem_unit >>= 1;
1474                sav_total = mem_total;
1475                mem_total <<= 1;
1476                if (mem_total < sav_total)
1477                        goto out;
1478        }
1479
1480        /*
1481         * If mem_total did not overflow, multiply all memory values by
1482         * info->mem_unit and set it to 1.  This leaves things compatible
1483         * with 2.2.x, and also retains compatibility with earlier 2.4.x
1484         * kernels...
1485         */
1486
1487        info->mem_unit = 1;
1488        info->totalram <<= bitcount;
1489        info->freeram <<= bitcount;
1490        info->sharedram <<= bitcount;
1491        info->bufferram <<= bitcount;
1492        info->totalswap <<= bitcount;
1493        info->freeswap <<= bitcount;
1494        info->totalhigh <<= bitcount;
1495        info->freehigh <<= bitcount;
1496
1497out:
1498        return 0;
1499}
1500
1501SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1502{
1503        struct sysinfo val;
1504
1505        do_sysinfo(&val);
1506
1507        if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1508                return -EFAULT;
1509
1510        return 0;
1511}
1512
1513static int __cpuinit init_timers_cpu(int cpu)
1514{
1515        int j;
1516        struct tvec_base *base;
1517        static char __cpuinitdata tvec_base_done[NR_CPUS];
1518
1519        if (!tvec_base_done[cpu]) {
1520                static char boot_done;
1521
1522                if (boot_done) {
1523                        /*
1524                         * The APs use this path later in boot
1525                         */
1526                        base = kmalloc_node(sizeof(*base),
1527                                                GFP_KERNEL | __GFP_ZERO,
1528                                                cpu_to_node(cpu));
1529                        if (!base)
1530                                return -ENOMEM;
1531
1532                        /* Make sure that tvec_base is 2 byte aligned */
1533                        if (tbase_get_deferrable(base)) {
1534                                WARN_ON(1);
1535                                kfree(base);
1536                                return -ENOMEM;
1537                        }
1538                        per_cpu(tvec_bases, cpu) = base;
1539                } else {
1540                        /*
1541                         * This is for the boot CPU - we use compile-time
1542                         * static initialisation because per-cpu memory isn't
1543                         * ready yet and because the memory allocators are not
1544                         * initialised either.
1545                         */
1546                        boot_done = 1;
1547                        base = &boot_tvec_bases;
1548                }
1549                tvec_base_done[cpu] = 1;
1550        } else {
1551                base = per_cpu(tvec_bases, cpu);
1552        }
1553
1554        spin_lock_init(&base->lock);
1555
1556        for (j = 0; j < TVN_SIZE; j++) {
1557                INIT_LIST_HEAD(base->tv5.vec + j);
1558                INIT_LIST_HEAD(base->tv4.vec + j);
1559                INIT_LIST_HEAD(base->tv3.vec + j);
1560                INIT_LIST_HEAD(base->tv2.vec + j);
1561        }
1562        for (j = 0; j < TVR_SIZE; j++)
1563                INIT_LIST_HEAD(base->tv1.vec + j);
1564
1565        base->timer_jiffies = jiffies;
1566        base->next_timer = base->timer_jiffies;
1567        return 0;
1568}
1569
1570#ifdef CONFIG_HOTPLUG_CPU
1571static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1572{
1573        struct timer_list *timer;
1574
1575        while (!list_empty(head)) {
1576                timer = list_first_entry(head, struct timer_list, entry);
1577                detach_timer(timer, 0);
1578                timer_set_base(timer, new_base);
1579                if (time_before(timer->expires, new_base->next_timer) &&
1580                    !tbase_get_deferrable(timer->base))
1581                        new_base->next_timer = timer->expires;
1582                internal_add_timer(new_base, timer);
1583        }
1584}
1585
1586static void __cpuinit migrate_timers(int cpu)
1587{
1588        struct tvec_base *old_base;
1589        struct tvec_base *new_base;
1590        int i;
1591
1592        BUG_ON(cpu_online(cpu));
1593        old_base = per_cpu(tvec_bases, cpu);
1594        new_base = get_cpu_var(tvec_bases);
1595        /*
1596         * The caller is globally serialized and nobody else
1597         * takes two locks at once, deadlock is not possible.
1598         */
1599        spin_lock_irq(&new_base->lock);
1600        spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1601
1602        BUG_ON(old_base->running_timer);
1603
1604        for (i = 0; i < TVR_SIZE; i++)
1605                migrate_timer_list(new_base, old_base->tv1.vec + i);
1606        for (i = 0; i < TVN_SIZE; i++) {
1607                migrate_timer_list(new_base, old_base->tv2.vec + i);
1608                migrate_timer_list(new_base, old_base->tv3.vec + i);
1609                migrate_timer_list(new_base, old_base->tv4.vec + i);
1610                migrate_timer_list(new_base, old_base->tv5.vec + i);
1611        }
1612
1613        spin_unlock(&old_base->lock);
1614        spin_unlock_irq(&new_base->lock);
1615        put_cpu_var(tvec_bases);
1616}
1617#endif /* CONFIG_HOTPLUG_CPU */
1618
1619static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1620                                unsigned long action, void *hcpu)
1621{
1622        long cpu = (long)hcpu;
1623        switch(action) {
1624        case CPU_UP_PREPARE:
1625        case CPU_UP_PREPARE_FROZEN:
1626                if (init_timers_cpu(cpu) < 0)
1627                        return NOTIFY_BAD;
1628                break;
1629#ifdef CONFIG_HOTPLUG_CPU
1630        case CPU_DEAD:
1631        case CPU_DEAD_FROZEN:
1632                migrate_timers(cpu);
1633                break;
1634#endif
1635        default:
1636                break;
1637        }
1638        return NOTIFY_OK;
1639}
1640
1641static struct notifier_block __cpuinitdata timers_nb = {
1642        .notifier_call  = timer_cpu_notify,
1643};
1644
1645
1646void __init init_timers(void)
1647{
1648        int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1649                                (void *)(long)smp_processor_id());
1650
1651        init_timer_stats();
1652
1653        BUG_ON(err == NOTIFY_BAD);
1654        register_cpu_notifier(&timers_nb);
1655        open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1656}
1657
1658/**
1659 * msleep - sleep safely even with waitqueue interruptions
1660 * @msecs: Time in milliseconds to sleep for
1661 */
1662void msleep(unsigned int msecs)
1663{
1664        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1665
1666        while (timeout)
1667                timeout = schedule_timeout_uninterruptible(timeout);
1668}
1669
1670EXPORT_SYMBOL(msleep);
1671
1672/**
1673 * msleep_interruptible - sleep waiting for signals
1674 * @msecs: Time in milliseconds to sleep for
1675 */
1676unsigned long msleep_interruptible(unsigned int msecs)
1677{
1678        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1679
1680        while (timeout && !signal_pending(current))
1681                timeout = schedule_timeout_interruptible(timeout);
1682        return jiffies_to_msecs(timeout);
1683}
1684
1685EXPORT_SYMBOL(msleep_interruptible);
1686
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.