linux/kernel/timer.c
<<
>>
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
  43#include <asm/uaccess.h>
  44#include <asm/unistd.h>
  45#include <asm/div64.h>
  46#include <asm/timex.h>
  47#include <asm/io.h>
  48
  49#define CREATE_TRACE_POINTS
  50#include <trace/events/timer.h>
  51
  52u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
  53
  54EXPORT_SYMBOL(jiffies_64);
  55
  56/*
  57 * per-CPU timer vector definitions:
  58 */
  59#define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
  60#define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
  61#define TVN_SIZE (1 << TVN_BITS)
  62#define TVR_SIZE (1 << TVR_BITS)
  63#define TVN_MASK (TVN_SIZE - 1)
  64#define TVR_MASK (TVR_SIZE - 1)
  65
  66struct tvec {
  67        struct list_head vec[TVN_SIZE];
  68};
  69
  70struct tvec_root {
  71        struct list_head vec[TVR_SIZE];
  72};
  73
  74struct tvec_base {
  75        spinlock_t lock;
  76        struct timer_list *running_timer;
  77        unsigned long timer_jiffies;
  78        unsigned long next_timer;
  79        struct tvec_root tv1;
  80        struct tvec tv2;
  81        struct tvec tv3;
  82        struct tvec tv4;
  83        struct tvec tv5;
  84} ____cacheline_aligned;
  85
  86struct tvec_base boot_tvec_bases;
  87EXPORT_SYMBOL(boot_tvec_bases);
  88static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
  89
  90/*
  91 * Note that all tvec_bases are 2 byte aligned and lower bit of
  92 * base in timer_list is guaranteed to be zero. Use the LSB for
  93 * the new flag to indicate whether the timer is deferrable
  94 */
  95#define TBASE_DEFERRABLE_FLAG           (0x1)
  96
  97/* Functions below help us manage 'deferrable' flag */
  98static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
  99{
 100        return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
 101}
 102
 103static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
 104{
 105        return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
 106}
 107
 108static inline void timer_set_deferrable(struct timer_list *timer)
 109{
 110        timer->base = ((struct tvec_base *)((unsigned long)(timer->base) |
 111                                       TBASE_DEFERRABLE_FLAG));
 112}
 113
 114static inline void
 115timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
 116{
 117        timer->base = (struct tvec_base *)((unsigned long)(new_base) |
 118                                      tbase_get_deferrable(timer->base));
 119}
 120
 121static unsigned long round_jiffies_common(unsigned long j, int cpu,
 122                bool force_up)
 123{
 124        int rem;
 125        unsigned long original = j;
 126
 127        /*
 128         * We don't want all cpus firing their timers at once hitting the
 129         * same lock or cachelines, so we skew each extra cpu with an extra
 130         * 3 jiffies. This 3 jiffies came originally from the mm/ code which
 131         * already did this.
 132         * The skew is done by adding 3*cpunr, then round, then subtract this
 133         * extra offset again.
 134         */
 135        j += cpu * 3;
 136
 137        rem = j % HZ;
 138
 139        /*
 140         * If the target jiffie is just after a whole second (which can happen
 141         * due to delays of the timer irq, long irq off times etc etc) then
 142         * we should round down to the whole second, not up. Use 1/4th second
 143         * as cutoff for this rounding as an extreme upper bound for this.
 144         * But never round down if @force_up is set.
 145         */
 146        if (rem < HZ/4 && !force_up) /* round down */
 147                j = j - rem;
 148        else /* round up */
 149                j = j - rem + HZ;
 150
 151        /* now that we have rounded, subtract the extra skew again */
 152        j -= cpu * 3;
 153
 154        if (j <= jiffies) /* rounding ate our timeout entirely; */
 155                return original;
 156        return j;
 157}
 158
 159/**
 160 * __round_jiffies - function to round jiffies to a full second
 161 * @j: the time in (absolute) jiffies that should be rounded
 162 * @cpu: the processor number on which the timeout will happen
 163 *
 164 * __round_jiffies() rounds an absolute time in the future (in jiffies)
 165 * up or down to (approximately) full seconds. This is useful for timers
 166 * for which the exact time they fire does not matter too much, as long as
 167 * they fire approximately every X seconds.
 168 *
 169 * By rounding these timers to whole seconds, all such timers will fire
 170 * at the same time, rather than at various times spread out. The goal
 171 * of this is to have the CPU wake up less, which saves power.
 172 *
 173 * The exact rounding is skewed for each processor to avoid all
 174 * processors firing at the exact same time, which could lead
 175 * to lock contention or spurious cache line bouncing.
 176 *
 177 * The return value is the rounded version of the @j parameter.
 178 */
 179unsigned long __round_jiffies(unsigned long j, int cpu)
 180{
 181        return round_jiffies_common(j, cpu, false);
 182}
 183EXPORT_SYMBOL_GPL(__round_jiffies);
 184
 185/**
 186 * __round_jiffies_relative - function to round jiffies to a full second
 187 * @j: the time in (relative) jiffies that should be rounded
 188 * @cpu: the processor number on which the timeout will happen
 189 *
 190 * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 191 * up or down to (approximately) full seconds. This is useful for timers
 192 * for which the exact time they fire does not matter too much, as long as
 193 * they fire approximately every X seconds.
 194 *
 195 * By rounding these timers to whole seconds, all such timers will fire
 196 * at the same time, rather than at various times spread out. The goal
 197 * of this is to have the CPU wake up less, which saves power.
 198 *
 199 * The exact rounding is skewed for each processor to avoid all
 200 * processors firing at the exact same time, which could lead
 201 * to lock contention or spurious cache line bouncing.
 202 *
 203 * The return value is the rounded version of the @j parameter.
 204 */
 205unsigned long __round_jiffies_relative(unsigned long j, int cpu)
 206{
 207        unsigned long j0 = jiffies;
 208
 209        /* Use j0 because jiffies might change while we run */
 210        return round_jiffies_common(j + j0, cpu, false) - j0;
 211}
 212EXPORT_SYMBOL_GPL(__round_jiffies_relative);
 213
 214/**
 215 * round_jiffies - function to round jiffies to a full second
 216 * @j: the time in (absolute) jiffies that should be rounded
 217 *
 218 * round_jiffies() rounds an absolute time in the future (in jiffies)
 219 * up or down to (approximately) full seconds. This is useful for timers
 220 * for which the exact time they fire does not matter too much, as long as
 221 * they fire approximately every X seconds.
 222 *
 223 * By rounding these timers to whole seconds, all such timers will fire
 224 * at the same time, rather than at various times spread out. The goal
 225 * of this is to have the CPU wake up less, which saves power.
 226 *
 227 * The return value is the rounded version of the @j parameter.
 228 */
 229unsigned long round_jiffies(unsigned long j)
 230{
 231        return round_jiffies_common(j, raw_smp_processor_id(), false);
 232}
 233EXPORT_SYMBOL_GPL(round_jiffies);
 234
 235/**
 236 * round_jiffies_relative - function to round jiffies to a full second
 237 * @j: the time in (relative) jiffies that should be rounded
 238 *
 239 * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 240 * up or down to (approximately) full seconds. This is useful for timers
 241 * for which the exact time they fire does not matter too much, as long as
 242 * they fire approximately every X seconds.
 243 *
 244 * By rounding these timers to whole seconds, all such timers will fire
 245 * at the same time, rather than at various times spread out. The goal
 246 * of this is to have the CPU wake up less, which saves power.
 247 *
 248 * The return value is the rounded version of the @j parameter.
 249 */
 250unsigned long round_jiffies_relative(unsigned long j)
 251{
 252        return __round_jiffies_relative(j, raw_smp_processor_id());
 253}
 254EXPORT_SYMBOL_GPL(round_jiffies_relative);
 255
 256/**
 257 * __round_jiffies_up - function to round jiffies up to a full second
 258 * @j: the time in (absolute) jiffies that should be rounded
 259 * @cpu: the processor number on which the timeout will happen
 260 *
 261 * This is the same as __round_jiffies() except that it will never
 262 * round down.  This is useful for timeouts for which the exact time
 263 * of firing does not matter too much, as long as they don't fire too
 264 * early.
 265 */
 266unsigned long __round_jiffies_up(unsigned long j, int cpu)
 267{
 268        return round_jiffies_common(j, cpu, true);
 269}
 270EXPORT_SYMBOL_GPL(__round_jiffies_up);
 271
 272/**
 273 * __round_jiffies_up_relative - function to round jiffies up to a full second
 274 * @j: the time in (relative) jiffies that should be rounded
 275 * @cpu: the processor number on which the timeout will happen
 276 *
 277 * This is the same as __round_jiffies_relative() except that it will never
 278 * round down.  This is useful for timeouts for which the exact time
 279 * of firing does not matter too much, as long as they don't fire too
 280 * early.
 281 */
 282unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
 283{
 284        unsigned long j0 = jiffies;
 285
 286        /* Use j0 because jiffies might change while we run */
 287        return round_jiffies_common(j + j0, cpu, true) - j0;
 288}
 289EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
 290
 291/**
 292 * round_jiffies_up - function to round jiffies up to a full second
 293 * @j: the time in (absolute) jiffies that should be rounded
 294 *
 295 * This is the same as round_jiffies() except that it will never
 296 * round down.  This is useful for timeouts for which the exact time
 297 * of firing does not matter too much, as long as they don't fire too
 298 * early.
 299 */
 300unsigned long round_jiffies_up(unsigned long j)
 301{
 302        return round_jiffies_common(j, raw_smp_processor_id(), true);
 303}
 304EXPORT_SYMBOL_GPL(round_jiffies_up);
 305
 306/**
 307 * round_jiffies_up_relative - function to round jiffies up to a full second
 308 * @j: the time in (relative) jiffies that should be rounded
 309 *
 310 * This is the same as round_jiffies_relative() except that it will never
 311 * round down.  This is useful for timeouts for which the exact time
 312 * of firing does not matter too much, as long as they don't fire too
 313 * early.
 314 */
 315unsigned long round_jiffies_up_relative(unsigned long j)
 316{
 317        return __round_jiffies_up_relative(j, raw_smp_processor_id());
 318}
 319EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
 320
 321
 322static inline void set_running_timer(struct tvec_base *base,
 323                                        struct timer_list *timer)
 324{
 325#ifdef CONFIG_SMP
 326        base->running_timer = timer;
 327#endif
 328}
 329
 330static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 331{
 332        unsigned long expires = timer->expires;
 333        unsigned long idx = expires - base->timer_jiffies;
 334        struct list_head *vec;
 335
 336        if (idx < TVR_SIZE) {
 337                int i = expires & TVR_MASK;
 338                vec = base->tv1.vec + i;
 339        } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
 340                int i = (expires >> TVR_BITS) & TVN_MASK;
 341                vec = base->tv2.vec + i;
 342        } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
 343                int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
 344                vec = base->tv3.vec + i;
 345        } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
 346                int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
 347                vec = base->tv4.vec + i;
 348        } else if ((signed long) idx < 0) {
 349                /*
 350                 * Can happen if you add a timer with expires == jiffies,
 351                 * or you set a timer to go off in the past
 352                 */
 353                vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
 354        } else {
 355                int i;
 356                /* If the timeout is larger than 0xffffffff on 64-bit
 357                 * architectures then we use the maximum timeout:
 358                 */
 359                if (idx > 0xffffffffUL) {
 360                        idx = 0xffffffffUL;
 361                        expires = idx + base->timer_jiffies;
 362                }
 363                i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
 364                vec = base->tv5.vec + i;
 365        }
 366        /*
 367         * Timers are FIFO:
 368         */
 369        list_add_tail(&timer->entry, vec);
 370}
 371
 372#ifdef CONFIG_TIMER_STATS
 373void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
 374{
 375        if (timer->start_site)
 376                return;
 377
 378        timer->start_site = addr;
 379        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 380        timer->start_pid = current->pid;
 381}
 382
 383static void timer_stats_account_timer(struct timer_list *timer)
 384{
 385        unsigned int flag = 0;
 386
 387        if (likely(!timer->start_site))
 388                return;
 389        if (unlikely(tbase_get_deferrable(timer->base)))
 390                flag |= TIMER_STATS_FLAG_DEFERRABLE;
 391
 392        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 393                                 timer->function, timer->start_comm, flag);
 394}
 395
 396#else
 397static void timer_stats_account_timer(struct timer_list *timer) {}
 398#endif
 399
 400#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 401
 402static struct debug_obj_descr timer_debug_descr;
 403
 404/*
 405 * fixup_init is called when:
 406 * - an active object is initialized
 407 */
 408static int timer_fixup_init(void *addr, enum debug_obj_state state)
 409{
 410        struct timer_list *timer = addr;
 411
 412        switch (state) {
 413        case ODEBUG_STATE_ACTIVE:
 414                del_timer_sync(timer);
 415                debug_object_init(timer, &timer_debug_descr);
 416                return 1;
 417        default:
 418                return 0;
 419        }
 420}
 421
 422/*
 423 * fixup_activate is called when:
 424 * - an active object is activated
 425 * - an unknown object is activated (might be a statically initialized object)
 426 */
 427static int timer_fixup_activate(void *addr, enum debug_obj_state state)
 428{
 429        struct timer_list *timer = addr;
 430
 431        switch (state) {
 432
 433        case ODEBUG_STATE_NOTAVAILABLE:
 434                /*
 435                 * This is not really a fixup. The timer was
 436                 * statically initialized. We just make sure that it
 437                 * is tracked in the object tracker.
 438                 */
 439                if (timer->entry.next == NULL &&
 440                    timer->entry.prev == TIMER_ENTRY_STATIC) {
 441                        debug_object_init(timer, &timer_debug_descr);
 442                        debug_object_activate(timer, &timer_debug_descr);
 443                        return 0;
 444                } else {
 445                        WARN_ON_ONCE(1);
 446                }
 447                return 0;
 448
 449        case ODEBUG_STATE_ACTIVE:
 450                WARN_ON(1);
 451
 452        default:
 453                return 0;
 454        }
 455}
 456
 457/*
 458 * fixup_free is called when:
 459 * - an active object is freed
 460 */
 461static int timer_fixup_free(void *addr, enum debug_obj_state state)
 462{
 463        struct timer_list *timer = addr;
 464
 465        switch (state) {
 466        case ODEBUG_STATE_ACTIVE:
 467                del_timer_sync(timer);
 468                debug_object_free(timer, &timer_debug_descr);
 469                return 1;
 470        default:
 471                return 0;
 472        }
 473}
 474
 475static struct debug_obj_descr timer_debug_descr = {
 476        .name           = "timer_list",
 477        .fixup_init     = timer_fixup_init,
 478        .fixup_activate = timer_fixup_activate,
 479        .fixup_free     = timer_fixup_free,
 480};
 481
 482static inline void debug_timer_init(struct timer_list *timer)
 483{
 484        debug_object_init(timer, &timer_debug_descr);
 485}
 486
 487static inline void debug_timer_activate(struct timer_list *timer)
 488{
 489        debug_object_activate(timer, &timer_debug_descr);
 490}
 491
 492static inline void debug_timer_deactivate(struct timer_list *timer)
 493{
 494        debug_object_deactivate(timer, &timer_debug_descr);
 495}
 496
 497static inline void debug_timer_free(struct timer_list *timer)
 498{
 499        debug_object_free(timer, &timer_debug_descr);
 500}
 501
 502static void __init_timer(struct timer_list *timer,
 503                         const char *name,
 504                         struct lock_class_key *key);
 505
 506void init_timer_on_stack_key(struct timer_list *timer,
 507                             const char *name,
 508                             struct lock_class_key *key)
 509{
 510        debug_object_init_on_stack(timer, &timer_debug_descr);
 511        __init_timer(timer, name, key);
 512}
 513EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
 514
 515void destroy_timer_on_stack(struct timer_list *timer)
 516{
 517        debug_object_free(timer, &timer_debug_descr);
 518}
 519EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
 520
 521#else
 522static inline void debug_timer_init(struct timer_list *timer) { }
 523static inline void debug_timer_activate(struct timer_list *timer) { }
 524static inline void debug_timer_deactivate(struct timer_list *timer) { }
 525#endif
 526
 527static inline void debug_init(struct timer_list *timer)
 528{
 529        debug_timer_init(timer);
 530        trace_timer_init(timer);
 531}
 532
 533static inline void
 534debug_activate(struct timer_list *timer, unsigned long expires)
 535{
 536        debug_timer_activate(timer);
 537        trace_timer_start(timer, expires);
 538}
 539
 540static inline void debug_deactivate(struct timer_list *timer)
 541{
 542        debug_timer_deactivate(timer);
 543        trace_timer_cancel(timer);
 544}
 545
 546static void __init_timer(struct timer_list *timer,
 547                         const char *name,
 548                         struct lock_class_key *key)
 549{
 550        timer->entry.next = NULL;
 551        timer->base = __raw_get_cpu_var(tvec_bases);
 552#ifdef CONFIG_TIMER_STATS
 553        timer->start_site = NULL;
 554        timer->start_pid = -1;
 555        memset(timer->start_comm, 0, TASK_COMM_LEN);
 556#endif
 557        lockdep_init_map(&timer->lockdep_map, name, key, 0);
 558}
 559
 560/**
 561 * init_timer_key - initialize a timer
 562 * @timer: the timer to be initialized
 563 * @name: name of the timer
 564 * @key: lockdep class key of the fake lock used for tracking timer
 565 *       sync lock dependencies
 566 *
 567 * init_timer_key() must be done to a timer prior calling *any* of the
 568 * other timer functions.
 569 */
 570void init_timer_key(struct timer_list *timer,
 571                    const char *name,
 572                    struct lock_class_key *key)
 573{
 574        debug_init(timer);
 575        __init_timer(timer, name, key);
 576}
 577EXPORT_SYMBOL(init_timer_key);
 578
 579void init_timer_deferrable_key(struct timer_list *timer,
 580                               const char *name,
 581                               struct lock_class_key *key)
 582{
 583        init_timer_key(timer, name, key);
 584        timer_set_deferrable(timer);
 585}
 586EXPORT_SYMBOL(init_timer_deferrable_key);
 587
 588static inline void detach_timer(struct timer_list *timer,
 589                                int clear_pending)
 590{
 591        struct list_head *entry = &timer->entry;
 592
 593        debug_deactivate(timer);
 594
 595        __list_del(entry->prev, entry->next);
 596        if (clear_pending)
 597                entry->next = NULL;
 598        entry->prev = LIST_POISON2;
 599}
 600
 601/*
 602 * We are using hashed locking: holding per_cpu(tvec_bases).lock
 603 * means that all timers which are tied to this base via timer->base are
 604 * locked, and the base itself is locked too.
 605 *
 606 * So __run_timers/migrate_timers can safely modify all timers which could
 607 * be found on ->tvX lists.
 608 *
 609 * When the timer's base is locked, and the timer removed from list, it is
 610 * possible to set timer->base = NULL and drop the lock: the timer remains
 611 * locked.
 612 */
 613static struct tvec_base *lock_timer_base(struct timer_list *timer,
 614                                        unsigned long *flags)
 615        __acquires(timer->base->lock)
 616{
 617        struct tvec_base *base;
 618
 619        for (;;) {
 620                struct tvec_base *prelock_base = timer->base;
 621                base = tbase_get_base(prelock_base);
 622                if (likely(base != NULL)) {
 623                        spin_lock_irqsave(&base->lock, *flags);
 624                        if (likely(prelock_base == timer->base))
 625                                return base;
 626                        /* The timer has migrated to another CPU */
 627                        spin_unlock_irqrestore(&base->lock, *flags);
 628                }
 629                cpu_relax();
 630        }
 631}
 632
 633static inline int
 634__mod_timer(struct timer_list *timer, unsigned long expires,
 635                                                bool pending_only, int pinned)
 636{
 637        struct tvec_base *base, *new_base;
 638        unsigned long flags;
 639        int ret = 0 , cpu;
 640
 641        timer_stats_timer_set_start_info(timer);
 642        BUG_ON(!timer->function);
 643
 644        base = lock_timer_base(timer, &flags);
 645
 646        if (timer_pending(timer)) {
 647                detach_timer(timer, 0);
 648                if (timer->expires == base->next_timer &&
 649                    !tbase_get_deferrable(timer->base))
 650                        base->next_timer = base->timer_jiffies;
 651                ret = 1;
 652        } else {
 653                if (pending_only)
 654                        goto out_unlock;
 655        }
 656
 657        debug_activate(timer, expires);
 658
 659        new_base = __get_cpu_var(tvec_bases);
 660
 661        cpu = smp_processor_id();
 662
 663#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
 664        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
 665                int preferred_cpu = get_nohz_load_balancer();
 666
 667                if (preferred_cpu >= 0)
 668                        cpu = preferred_cpu;
 669        }
 670#endif
 671        new_base = per_cpu(tvec_bases, cpu);
 672
 673        if (base != new_base) {
 674                /*
 675                 * We are trying to schedule the timer on the local CPU.
 676                 * However we can't change timer's base while it is running,
 677                 * otherwise del_timer_sync() can't detect that the timer's
 678                 * handler yet has not finished. This also guarantees that
 679                 * the timer is serialized wrt itself.
 680                 */
 681                if (likely(base->running_timer != timer)) {
 682                        /* See the comment in lock_timer_base() */
 683                        timer_set_base(timer, NULL);
 684                        spin_unlock(&base->lock);
 685                        base = new_base;
 686                        spin_lock(&base->lock);
 687                        timer_set_base(timer, base);
 688                }
 689        }
 690
 691        timer->expires = expires;
 692        if (time_before(timer->expires, base->next_timer) &&
 693            !tbase_get_deferrable(timer->base))
 694                base->next_timer = timer->expires;
 695        internal_add_timer(base, timer);
 696
 697out_unlock:
 698        spin_unlock_irqrestore(&base->lock, flags);
 699
 700        return ret;
 701}
 702
 703/**
 704 * mod_timer_pending - modify a pending timer's timeout
 705 * @timer: the pending timer to be modified
 706 * @expires: new timeout in jiffies
 707 *
 708 * mod_timer_pending() is the same for pending timers as mod_timer(),
 709 * but will not re-activate and modify already deleted timers.
 710 *
 711 * It is useful for unserialized use of timers.
 712 */
 713int mod_timer_pending(struct timer_list *timer, unsigned long expires)
 714{
 715        return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
 716}
 717EXPORT_SYMBOL(mod_timer_pending);
 718
 719/**
 720 * mod_timer - modify a timer's timeout
 721 * @timer: the timer to be modified
 722 * @expires: new timeout in jiffies
 723 *
 724 * mod_timer() is a more efficient way to update the expire field of an
 725 * active timer (if the timer is inactive it will be activated)
 726 *
 727 * mod_timer(timer, expires) is equivalent to:
 728 *
 729 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 730 *
 731 * Note that if there are multiple unserialized concurrent users of the
 732 * same timer, then mod_timer() is the only safe way to modify the timeout,
 733 * since add_timer() cannot modify an already running timer.
 734 *
 735 * The function returns whether it has modified a pending timer or not.
 736 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
 737 * active timer returns 1.)
 738 */
 739int mod_timer(struct timer_list *timer, unsigned long expires)
 740{
 741        /*
 742         * This is a common optimization triggered by the
 743         * networking code - if the timer is re-modified
 744         * to be the same thing then just return:
 745         */
 746        if (timer_pending(timer) && timer->expires == expires)
 747                return 1;
 748
 749        return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
 750}
 751EXPORT_SYMBOL(mod_timer);
 752
 753/**
 754 * mod_timer_pinned - modify a timer's timeout
 755 * @timer: the timer to be modified
 756 * @expires: new timeout in jiffies
 757 *
 758 * mod_timer_pinned() is a way to update the expire field of an
 759 * active timer (if the timer is inactive it will be activated)
 760 * and not allow the timer to be migrated to a different CPU.
 761 *
 762 * mod_timer_pinned(timer, expires) is equivalent to:
 763 *
 764 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 765 */
 766int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
 767{
 768        if (timer->expires == expires && timer_pending(timer))
 769                return 1;
 770
 771        return __mod_timer(timer, expires, false, TIMER_PINNED);
 772}
 773EXPORT_SYMBOL(mod_timer_pinned);
 774
 775/**
 776 * add_timer - start a timer
 777 * @timer: the timer to be added
 778 *
 779 * The kernel will do a ->function(->data) callback from the
 780 * timer interrupt at the ->expires point in the future. The
 781 * current time is 'jiffies'.
 782 *
 783 * The timer's ->expires, ->function (and if the handler uses it, ->data)
 784 * fields must be set prior calling this function.
 785 *
 786 * Timers with an ->expires field in the past will be executed in the next
 787 * timer tick.
 788 */
 789void add_timer(struct timer_list *timer)
 790{
 791        BUG_ON(timer_pending(timer));
 792        mod_timer(timer, timer->expires);
 793}
 794EXPORT_SYMBOL(add_timer);
 795
 796/**
 797 * add_timer_on - start a timer on a particular CPU
 798 * @timer: the timer to be added
 799 * @cpu: the CPU to start it on
 800 *
 801 * This is not very scalable on SMP. Double adds are not possible.
 802 */
 803void add_timer_on(struct timer_list *timer, int cpu)
 804{
 805        struct tvec_base *base = per_cpu(tvec_bases, cpu);
 806        unsigned long flags;
 807
 808        timer_stats_timer_set_start_info(timer);
 809        BUG_ON(timer_pending(timer) || !timer->function);
 810        spin_lock_irqsave(&base->lock, flags);
 811        timer_set_base(timer, base);
 812        debug_activate(timer, timer->expires);
 813        if (time_before(timer->expires, base->next_timer) &&
 814            !tbase_get_deferrable(timer->base))
 815                base->next_timer = timer->expires;
 816        internal_add_timer(base, timer);
 817        /*
 818         * Check whether the other CPU is idle and needs to be
 819         * triggered to reevaluate the timer wheel when nohz is
 820         * active. We are protected against the other CPU fiddling
 821         * with the timer by holding the timer base lock. This also
 822         * makes sure that a CPU on the way to idle can not evaluate
 823         * the timer wheel.
 824         */
 825        wake_up_idle_cpu(cpu);
 826        spin_unlock_irqrestore(&base->lock, flags);
 827}
 828EXPORT_SYMBOL_GPL(add_timer_on);
 829
 830/**
 831 * del_timer - deactive a timer.
 832 * @timer: the timer to be deactivated
 833 *
 834 * del_timer() deactivates a timer - this works on both active and inactive
 835 * timers.
 836 *
 837 * The function returns whether it has deactivated a pending timer or not.
 838 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
 839 * active timer returns 1.)
 840 */
 841int del_timer(struct timer_list *timer)
 842{
 843        struct tvec_base *base;
 844        unsigned long flags;
 845        int ret = 0;
 846
 847        timer_stats_timer_clear_start_info(timer);
 848        if (timer_pending(timer)) {
 849                base = lock_timer_base(timer, &flags);
 850                if (timer_pending(timer)) {
 851                        detach_timer(timer, 1);
 852                        if (timer->expires == base->next_timer &&
 853                            !tbase_get_deferrable(timer->base))
 854                                base->next_timer = base->timer_jiffies;
 855                        ret = 1;
 856                }
 857                spin_unlock_irqrestore(&base->lock, flags);
 858        }
 859
 860        return ret;
 861}
 862EXPORT_SYMBOL(del_timer);
 863
 864#ifdef CONFIG_SMP
 865/**
 866 * try_to_del_timer_sync - Try to deactivate a timer
 867 * @timer: timer do del
 868 *
 869 * This function tries to deactivate a timer. Upon successful (ret >= 0)
 870 * exit the timer is not queued and the handler is not running on any CPU.
 871 *
 872 * It must not be called from interrupt contexts.
 873 */
 874int try_to_del_timer_sync(struct timer_list *timer)
 875{
 876        struct tvec_base *base;
 877        unsigned long flags;
 878        int ret = -1;
 879
 880        base = lock_timer_base(timer, &flags);
 881
 882        if (base->running_timer == timer)
 883                goto out;
 884
 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        scheduler_tick();
1204        run_posix_cpu_timers(p);
1205}
1206
1207/*
1208 * This function runs timers and the timer-tq in bottom half context.
1209 */
1210static void run_timer_softirq(struct softirq_action *h)
1211{
1212        struct tvec_base *base = __get_cpu_var(tvec_bases);
1213
1214        perf_event_do_pending();
1215
1216        hrtimer_run_pending();
1217
1218        if (time_after_eq(jiffies, base->timer_jiffies))
1219                __run_timers(base);
1220}
1221
1222/*
1223 * Called by the local, per-CPU timer interrupt on SMP.
1224 */
1225void run_local_timers(void)
1226{
1227        hrtimer_run_queues();
1228        raise_softirq(TIMER_SOFTIRQ);
1229        softlockup_tick();
1230}
1231
1232/*
1233 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1234 * without sampling the sequence number in xtime_lock.
1235 * jiffies is defined in the linker script...
1236 */
1237
1238void do_timer(unsigned long ticks)
1239{
1240        jiffies_64 += ticks;
1241        update_wall_time();
1242        calc_global_load();
1243}
1244
1245#ifdef __ARCH_WANT_SYS_ALARM
1246
1247/*
1248 * For backwards compatibility?  This can be done in libc so Alpha
1249 * and all newer ports shouldn't need it.
1250 */
1251SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1252{
1253        return alarm_setitimer(seconds);
1254}
1255
1256#endif
1257
1258#ifndef __alpha__
1259
1260/*
1261 * The Alpha uses getxpid, getxuid, and getxgid instead.  Maybe this
1262 * should be moved into arch/i386 instead?
1263 */
1264
1265/**
1266 * sys_getpid - return the thread group id of the current process
1267 *
1268 * Note, despite the name, this returns the tgid not the pid.  The tgid and
1269 * the pid are identical unless CLONE_THREAD was specified on clone() in
1270 * which case the tgid is the same in all threads of the same group.
1271 *
1272 * This is SMP safe as current->tgid does not change.
1273 */
1274SYSCALL_DEFINE0(getpid)
1275{
1276        return task_tgid_vnr(current);
1277}
1278
1279/*
1280 * Accessing ->real_parent is not SMP-safe, it could
1281 * change from under us. However, we can use a stale
1282 * value of ->real_parent under rcu_read_lock(), see
1283 * release_task()->call_rcu(delayed_put_task_struct).
1284 */
1285SYSCALL_DEFINE0(getppid)
1286{
1287        int pid;
1288
1289        rcu_read_lock();
1290        pid = task_tgid_vnr(current->real_parent);
1291        rcu_read_unlock();
1292
1293        return pid;
1294}
1295
1296SYSCALL_DEFINE0(getuid)
1297{
1298        /* Only we change this so SMP safe */
1299        return current_uid();
1300}
1301
1302SYSCALL_DEFINE0(geteuid)
1303{
1304        /* Only we change this so SMP safe */
1305        return current_euid();
1306}
1307
1308SYSCALL_DEFINE0(getgid)
1309{
1310        /* Only we change this so SMP safe */
1311        return current_gid();
1312}
1313
1314SYSCALL_DEFINE0(getegid)
1315{
1316        /* Only we change this so SMP safe */
1317        return  current_egid();
1318}
1319
1320#endif
1321
1322static void process_timeout(unsigned long __data)
1323{
1324        wake_up_process((struct task_struct *)__data);
1325}
1326
1327/**
1328 * schedule_timeout - sleep until timeout
1329 * @timeout: timeout value in jiffies
1330 *
1331 * Make the current task sleep until @timeout jiffies have
1332 * elapsed. The routine will return immediately unless
1333 * the current task state has been set (see set_current_state()).
1334 *
1335 * You can set the task state as follows -
1336 *
1337 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1338 * pass before the routine returns. The routine will return 0
1339 *
1340 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1341 * delivered to the current task. In this case the remaining time
1342 * in jiffies will be returned, or 0 if the timer expired in time
1343 *
1344 * The current task state is guaranteed to be TASK_RUNNING when this
1345 * routine returns.
1346 *
1347 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1348 * the CPU away without a bound on the timeout. In this case the return
1349 * value will be %MAX_SCHEDULE_TIMEOUT.
1350 *
1351 * In all cases the return value is guaranteed to be non-negative.
1352 */
1353signed long __sched schedule_timeout(signed long timeout)
1354{
1355        struct timer_list timer;
1356        unsigned long expire;
1357
1358        switch (timeout)
1359        {
1360        case MAX_SCHEDULE_TIMEOUT:
1361                /*
1362                 * These two special cases are useful to be comfortable
1363                 * in the caller. Nothing more. We could take
1364                 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1365                 * but I' d like to return a valid offset (>=0) to allow
1366                 * the caller to do everything it want with the retval.
1367                 */
1368                schedule();
1369                goto out;
1370        default:
1371                /*
1372                 * Another bit of PARANOID. Note that the retval will be
1373                 * 0 since no piece of kernel is supposed to do a check
1374                 * for a negative retval of schedule_timeout() (since it
1375                 * should never happens anyway). You just have the printk()
1376                 * that will tell you if something is gone wrong and where.
1377                 */
1378                if (timeout < 0) {
1379                        printk(KERN_ERR "schedule_timeout: wrong timeout "
1380                                "value %lx\n", timeout);
1381                        dump_stack();
1382                        current->state = TASK_RUNNING;
1383                        goto out;
1384                }
1385        }
1386
1387        expire = timeout + jiffies;
1388
1389        setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1390        __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1391        schedule();
1392        del_singleshot_timer_sync(&timer);
1393
1394        /* Remove the timer from the object tracker */
1395        destroy_timer_on_stack(&timer);
1396
1397        timeout = expire - jiffies;
1398
1399 out:
1400        return timeout < 0 ? 0 : timeout;
1401}
1402EXPORT_SYMBOL(schedule_timeout);
1403
1404/*
1405 * We can use __set_current_state() here because schedule_timeout() calls
1406 * schedule() unconditionally.
1407 */
1408signed long __sched schedule_timeout_interruptible(signed long timeout)
1409{
1410        __set_current_state(TASK_INTERRUPTIBLE);
1411        return schedule_timeout(timeout);
1412}
1413EXPORT_SYMBOL(schedule_timeout_interruptible);
1414
1415signed long __sched schedule_timeout_killable(signed long timeout)
1416{
1417        __set_current_state(TASK_KILLABLE);
1418        return schedule_timeout(timeout);
1419}
1420EXPORT_SYMBOL(schedule_timeout_killable);
1421
1422signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1423{
1424        __set_current_state(TASK_UNINTERRUPTIBLE);
1425        return schedule_timeout(timeout);
1426}
1427EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1428
1429/* Thread ID - the internal kernel "pid" */
1430SYSCALL_DEFINE0(gettid)
1431{
1432        return task_pid_vnr(current);
1433}
1434
1435/**
1436 * do_sysinfo - fill in sysinfo struct
1437 * @info: pointer to buffer to fill
1438 */
1439int do_sysinfo(struct sysinfo *info)
1440{
1441        unsigned long mem_total, sav_total;
1442        unsigned int mem_unit, bitcount;
1443        struct timespec tp;
1444
1445        memset(info, 0, sizeof(struct sysinfo));
1446
1447        ktime_get_ts(&tp);
1448        monotonic_to_bootbased(&tp);
1449        info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1450
1451        get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1452
1453        info->procs = nr_threads;
1454
1455        si_meminfo(info);
1456        si_swapinfo(info);
1457
1458        /*
1459         * If the sum of all the available memory (i.e. ram + swap)
1460         * is less than can be stored in a 32 bit unsigned long then
1461         * we can be binary compatible with 2.2.x kernels.  If not,
1462         * well, in that case 2.2.x was broken anyways...
1463         *
1464         *  -Erik Andersen <andersee@debian.org>
1465         */
1466
1467        mem_total = info->totalram + info->totalswap;
1468        if (mem_total < info->totalram || mem_total < info->totalswap)
1469                goto out;
1470        bitcount = 0;
1471        mem_unit = info->mem_unit;
1472        while (mem_unit > 1) {
1473                bitcount++;
1474                mem_unit >>= 1;
1475                sav_total = mem_total;
1476                mem_total <<= 1;
1477                if (mem_total < sav_total)
1478                        goto out;
1479        }
1480
1481        /*
1482         * If mem_total did not overflow, multiply all memory values by
1483         * info->mem_unit and set it to 1.  This leaves things compatible
1484         * with 2.2.x, and also retains compatibility with earlier 2.4.x
1485         * kernels...
1486         */
1487
1488        info->mem_unit = 1;
1489        info->totalram <<= bitcount;
1490        info->freeram <<= bitcount;
1491        info->sharedram <<= bitcount;
1492        info->bufferram <<= bitcount;
1493        info->totalswap <<= bitcount;
1494        info->freeswap <<= bitcount;
1495        info->totalhigh <<= bitcount;
1496        info->freehigh <<= bitcount;
1497
1498out:
1499        return 0;
1500}
1501
1502SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1503{
1504        struct sysinfo val;
1505
1506        do_sysinfo(&val);
1507
1508        if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1509                return -EFAULT;
1510
1511        return 0;
1512}
1513
1514static int __cpuinit init_timers_cpu(int cpu)
1515{
1516        int j;
1517        struct tvec_base *base;
1518        static char __cpuinitdata tvec_base_done[NR_CPUS];
1519
1520        if (!tvec_base_done[cpu]) {
1521                static char boot_done;
1522
1523                if (boot_done) {
1524                        /*
1525                         * The APs use this path later in boot
1526                         */
1527                        base = kmalloc_node(sizeof(*base),
1528                                                GFP_KERNEL | __GFP_ZERO,
1529                                                cpu_to_node(cpu));
1530                        if (!base)
1531                                return -ENOMEM;
1532
1533                        /* Make sure that tvec_base is 2 byte aligned */
1534                        if (tbase_get_deferrable(base)) {
1535                                WARN_ON(1);
1536                                kfree(base);
1537                                return -ENOMEM;
1538                        }
1539                        per_cpu(tvec_bases, cpu) = base;
1540                } else {
1541                        /*
1542                         * This is for the boot CPU - we use compile-time
1543                         * static initialisation because per-cpu memory isn't
1544                         * ready yet and because the memory allocators are not
1545                         * initialised either.
1546                         */
1547                        boot_done = 1;
1548                        base = &boot_tvec_bases;
1549                }
1550                tvec_base_done[cpu] = 1;
1551        } else {
1552                base = per_cpu(tvec_bases, cpu);
1553        }
1554
1555        spin_lock_init(&base->lock);
1556
1557        for (j = 0; j < TVN_SIZE; j++) {
1558                INIT_LIST_HEAD(base->tv5.vec + j);
1559                INIT_LIST_HEAD(base->tv4.vec + j);
1560                INIT_LIST_HEAD(base->tv3.vec + j);
1561                INIT_LIST_HEAD(base->tv2.vec + j);
1562        }
1563        for (j = 0; j < TVR_SIZE; j++)
1564                INIT_LIST_HEAD(base->tv1.vec + j);
1565
1566        base->timer_jiffies = jiffies;
1567        base->next_timer = base->timer_jiffies;
1568        return 0;
1569}
1570
1571#ifdef CONFIG_HOTPLUG_CPU
1572static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1573{
1574        struct timer_list *timer;
1575
1576        while (!list_empty(head)) {
1577                timer = list_first_entry(head, struct timer_list, entry);
1578                detach_timer(timer, 0);
1579                timer_set_base(timer, new_base);
1580                if (time_before(timer->expires, new_base->next_timer) &&
1581                    !tbase_get_deferrable(timer->base))
1582                        new_base->next_timer = timer->expires;
1583                internal_add_timer(new_base, timer);
1584        }
1585}
1586
1587static void __cpuinit migrate_timers(int cpu)
1588{
1589        struct tvec_base *old_base;
1590        struct tvec_base *new_base;
1591        int i;
1592
1593        BUG_ON(cpu_online(cpu));
1594        old_base = per_cpu(tvec_bases, cpu);
1595        new_base = get_cpu_var(tvec_bases);
1596        /*
1597         * The caller is globally serialized and nobody else
1598         * takes two locks at once, deadlock is not possible.
1599         */
1600        spin_lock_irq(&new_base->lock);
1601        spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1602
1603        BUG_ON(old_base->running_timer);
1604
1605        for (i = 0; i < TVR_SIZE; i++)
1606                migrate_timer_list(new_base, old_base->tv1.vec + i);
1607        for (i = 0; i < TVN_SIZE; i++) {
1608                migrate_timer_list(new_base, old_base->tv2.vec + i);
1609                migrate_timer_list(new_base, old_base->tv3.vec + i);
1610                migrate_timer_list(new_base, old_base->tv4.vec + i);
1611                migrate_timer_list(new_base, old_base->tv5.vec + i);
1612        }
1613
1614        spin_unlock(&old_base->lock);
1615        spin_unlock_irq(&new_base->lock);
1616        put_cpu_var(tvec_bases);
1617}
1618#endif /* CONFIG_HOTPLUG_CPU */
1619
1620static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1621                                unsigned long action, void *hcpu)
1622{
1623        long cpu = (long)hcpu;
1624        switch(action) {
1625        case CPU_UP_PREPARE:
1626        case CPU_UP_PREPARE_FROZEN:
1627                if (init_timers_cpu(cpu) < 0)
1628                        return NOTIFY_BAD;
1629                break;
1630#ifdef CONFIG_HOTPLUG_CPU
1631        case CPU_DEAD:
1632        case CPU_DEAD_FROZEN:
1633                migrate_timers(cpu);
1634                break;
1635#endif
1636        default:
1637                break;
1638        }
1639        return NOTIFY_OK;
1640}
1641
1642static struct notifier_block __cpuinitdata timers_nb = {
1643        .notifier_call  = timer_cpu_notify,
1644};
1645
1646
1647void __init init_timers(void)
1648{
1649        int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1650                                (void *)(long)smp_processor_id());
1651
1652        init_timer_stats();
1653
1654        BUG_ON(err == NOTIFY_BAD);
1655        register_cpu_notifier(&timers_nb);
1656        open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1657}
1658
1659/**
1660 * msleep - sleep safely even with waitqueue interruptions
1661 * @msecs: Time in milliseconds to sleep for
1662 */
1663void msleep(unsigned int msecs)
1664{
1665        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1666
1667        while (timeout)
1668                timeout = schedule_timeout_uninterruptible(timeout);
1669}
1670
1671EXPORT_SYMBOL(msleep);
1672
1673/**
1674 * msleep_interruptible - sleep waiting for signals
1675 * @msecs: Time in milliseconds to sleep for
1676 */
1677unsigned long msleep_interruptible(unsigned int msecs)
1678{
1679        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1680
1681        while (timeout && !signal_pending(current))
1682                timeout = schedule_timeout_interruptible(timeout);
1683        return jiffies_to_msecs(timeout);
1684}
1685
1686EXPORT_SYMBOL(msleep_interruptible);
1687