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/export.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/irq_work.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#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
  67
  68struct tvec {
  69        struct list_head vec[TVN_SIZE];
  70};
  71
  72struct tvec_root {
  73        struct list_head vec[TVR_SIZE];
  74};
  75
  76struct tvec_base {
  77        spinlock_t lock;
  78        struct timer_list *running_timer;
  79        unsigned long timer_jiffies;
  80        unsigned long next_timer;
  81        unsigned long active_timers;
  82        struct tvec_root tv1;
  83        struct tvec tv2;
  84        struct tvec tv3;
  85        struct tvec tv4;
  86        struct tvec tv5;
  87} ____cacheline_aligned;
  88
  89struct tvec_base boot_tvec_bases;
  90EXPORT_SYMBOL(boot_tvec_bases);
  91static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
  92
  93/* Functions below help us manage 'deferrable' flag */
  94static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
  95{
  96        return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG);
  97}
  98
  99static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
 100{
 101        return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG));
 102}
 103
 104static inline void timer_set_deferrable(struct timer_list *timer)
 105{
 106        timer->base = TBASE_MAKE_DEFERRED(timer->base);
 107}
 108
 109static inline void
 110timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
 111{
 112        timer->base = (struct tvec_base *)((unsigned long)(new_base) |
 113                                      tbase_get_deferrable(timer->base));
 114}
 115
 116static unsigned long round_jiffies_common(unsigned long j, int cpu,
 117                bool force_up)
 118{
 119        int rem;
 120        unsigned long original = j;
 121
 122        /*
 123         * We don't want all cpus firing their timers at once hitting the
 124         * same lock or cachelines, so we skew each extra cpu with an extra
 125         * 3 jiffies. This 3 jiffies came originally from the mm/ code which
 126         * already did this.
 127         * The skew is done by adding 3*cpunr, then round, then subtract this
 128         * extra offset again.
 129         */
 130        j += cpu * 3;
 131
 132        rem = j % HZ;
 133
 134        /*
 135         * If the target jiffie is just after a whole second (which can happen
 136         * due to delays of the timer irq, long irq off times etc etc) then
 137         * we should round down to the whole second, not up. Use 1/4th second
 138         * as cutoff for this rounding as an extreme upper bound for this.
 139         * But never round down if @force_up is set.
 140         */
 141        if (rem < HZ/4 && !force_up) /* round down */
 142                j = j - rem;
 143        else /* round up */
 144                j = j - rem + HZ;
 145
 146        /* now that we have rounded, subtract the extra skew again */
 147        j -= cpu * 3;
 148
 149        if (j <= jiffies) /* rounding ate our timeout entirely; */
 150                return original;
 151        return j;
 152}
 153
 154/**
 155 * __round_jiffies - function to round jiffies to a full second
 156 * @j: the time in (absolute) jiffies that should be rounded
 157 * @cpu: the processor number on which the timeout will happen
 158 *
 159 * __round_jiffies() rounds an absolute time in the future (in jiffies)
 160 * up or down to (approximately) full seconds. This is useful for timers
 161 * for which the exact time they fire does not matter too much, as long as
 162 * they fire approximately every X seconds.
 163 *
 164 * By rounding these timers to whole seconds, all such timers will fire
 165 * at the same time, rather than at various times spread out. The goal
 166 * of this is to have the CPU wake up less, which saves power.
 167 *
 168 * The exact rounding is skewed for each processor to avoid all
 169 * processors firing at the exact same time, which could lead
 170 * to lock contention or spurious cache line bouncing.
 171 *
 172 * The return value is the rounded version of the @j parameter.
 173 */
 174unsigned long __round_jiffies(unsigned long j, int cpu)
 175{
 176        return round_jiffies_common(j, cpu, false);
 177}
 178EXPORT_SYMBOL_GPL(__round_jiffies);
 179
 180/**
 181 * __round_jiffies_relative - function to round jiffies to a full second
 182 * @j: the time in (relative) jiffies that should be rounded
 183 * @cpu: the processor number on which the timeout will happen
 184 *
 185 * __round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 186 * up or down to (approximately) full seconds. This is useful for timers
 187 * for which the exact time they fire does not matter too much, as long as
 188 * they fire approximately every X seconds.
 189 *
 190 * By rounding these timers to whole seconds, all such timers will fire
 191 * at the same time, rather than at various times spread out. The goal
 192 * of this is to have the CPU wake up less, which saves power.
 193 *
 194 * The exact rounding is skewed for each processor to avoid all
 195 * processors firing at the exact same time, which could lead
 196 * to lock contention or spurious cache line bouncing.
 197 *
 198 * The return value is the rounded version of the @j parameter.
 199 */
 200unsigned long __round_jiffies_relative(unsigned long j, int cpu)
 201{
 202        unsigned long j0 = jiffies;
 203
 204        /* Use j0 because jiffies might change while we run */
 205        return round_jiffies_common(j + j0, cpu, false) - j0;
 206}
 207EXPORT_SYMBOL_GPL(__round_jiffies_relative);
 208
 209/**
 210 * round_jiffies - function to round jiffies to a full second
 211 * @j: the time in (absolute) jiffies that should be rounded
 212 *
 213 * round_jiffies() rounds an absolute time in the future (in jiffies)
 214 * up or down to (approximately) full seconds. This is useful for timers
 215 * for which the exact time they fire does not matter too much, as long as
 216 * they fire approximately every X seconds.
 217 *
 218 * By rounding these timers to whole seconds, all such timers will fire
 219 * at the same time, rather than at various times spread out. The goal
 220 * of this is to have the CPU wake up less, which saves power.
 221 *
 222 * The return value is the rounded version of the @j parameter.
 223 */
 224unsigned long round_jiffies(unsigned long j)
 225{
 226        return round_jiffies_common(j, raw_smp_processor_id(), false);
 227}
 228EXPORT_SYMBOL_GPL(round_jiffies);
 229
 230/**
 231 * round_jiffies_relative - function to round jiffies to a full second
 232 * @j: the time in (relative) jiffies that should be rounded
 233 *
 234 * round_jiffies_relative() rounds a time delta  in the future (in jiffies)
 235 * up or down to (approximately) full seconds. This is useful for timers
 236 * for which the exact time they fire does not matter too much, as long as
 237 * they fire approximately every X seconds.
 238 *
 239 * By rounding these timers to whole seconds, all such timers will fire
 240 * at the same time, rather than at various times spread out. The goal
 241 * of this is to have the CPU wake up less, which saves power.
 242 *
 243 * The return value is the rounded version of the @j parameter.
 244 */
 245unsigned long round_jiffies_relative(unsigned long j)
 246{
 247        return __round_jiffies_relative(j, raw_smp_processor_id());
 248}
 249EXPORT_SYMBOL_GPL(round_jiffies_relative);
 250
 251/**
 252 * __round_jiffies_up - function to round jiffies up to a full second
 253 * @j: the time in (absolute) jiffies that should be rounded
 254 * @cpu: the processor number on which the timeout will happen
 255 *
 256 * This is the same as __round_jiffies() except that it will never
 257 * round down.  This is useful for timeouts for which the exact time
 258 * of firing does not matter too much, as long as they don't fire too
 259 * early.
 260 */
 261unsigned long __round_jiffies_up(unsigned long j, int cpu)
 262{
 263        return round_jiffies_common(j, cpu, true);
 264}
 265EXPORT_SYMBOL_GPL(__round_jiffies_up);
 266
 267/**
 268 * __round_jiffies_up_relative - function to round jiffies up to a full second
 269 * @j: the time in (relative) jiffies that should be rounded
 270 * @cpu: the processor number on which the timeout will happen
 271 *
 272 * This is the same as __round_jiffies_relative() except that it will never
 273 * round down.  This is useful for timeouts for which the exact time
 274 * of firing does not matter too much, as long as they don't fire too
 275 * early.
 276 */
 277unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
 278{
 279        unsigned long j0 = jiffies;
 280
 281        /* Use j0 because jiffies might change while we run */
 282        return round_jiffies_common(j + j0, cpu, true) - j0;
 283}
 284EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
 285
 286/**
 287 * round_jiffies_up - function to round jiffies up to a full second
 288 * @j: the time in (absolute) jiffies that should be rounded
 289 *
 290 * This is the same as round_jiffies() except that it will never
 291 * round down.  This is useful for timeouts for which the exact time
 292 * of firing does not matter too much, as long as they don't fire too
 293 * early.
 294 */
 295unsigned long round_jiffies_up(unsigned long j)
 296{
 297        return round_jiffies_common(j, raw_smp_processor_id(), true);
 298}
 299EXPORT_SYMBOL_GPL(round_jiffies_up);
 300
 301/**
 302 * round_jiffies_up_relative - function to round jiffies up to a full second
 303 * @j: the time in (relative) jiffies that should be rounded
 304 *
 305 * This is the same as round_jiffies_relative() except that it will never
 306 * round down.  This is useful for timeouts for which the exact time
 307 * of firing does not matter too much, as long as they don't fire too
 308 * early.
 309 */
 310unsigned long round_jiffies_up_relative(unsigned long j)
 311{
 312        return __round_jiffies_up_relative(j, raw_smp_processor_id());
 313}
 314EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
 315
 316/**
 317 * set_timer_slack - set the allowed slack for a timer
 318 * @timer: the timer to be modified
 319 * @slack_hz: the amount of time (in jiffies) allowed for rounding
 320 *
 321 * Set the amount of time, in jiffies, that a certain timer has
 322 * in terms of slack. By setting this value, the timer subsystem
 323 * will schedule the actual timer somewhere between
 324 * the time mod_timer() asks for, and that time plus the slack.
 325 *
 326 * By setting the slack to -1, a percentage of the delay is used
 327 * instead.
 328 */
 329void set_timer_slack(struct timer_list *timer, int slack_hz)
 330{
 331        timer->slack = slack_hz;
 332}
 333EXPORT_SYMBOL_GPL(set_timer_slack);
 334
 335static void
 336__internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 337{
 338        unsigned long expires = timer->expires;
 339        unsigned long idx = expires - base->timer_jiffies;
 340        struct list_head *vec;
 341
 342        if (idx < TVR_SIZE) {
 343                int i = expires & TVR_MASK;
 344                vec = base->tv1.vec + i;
 345        } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
 346                int i = (expires >> TVR_BITS) & TVN_MASK;
 347                vec = base->tv2.vec + i;
 348        } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
 349                int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
 350                vec = base->tv3.vec + i;
 351        } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
 352                int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
 353                vec = base->tv4.vec + i;
 354        } else if ((signed long) idx < 0) {
 355                /*
 356                 * Can happen if you add a timer with expires == jiffies,
 357                 * or you set a timer to go off in the past
 358                 */
 359                vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
 360        } else {
 361                int i;
 362                /* If the timeout is larger than MAX_TVAL (on 64-bit
 363                 * architectures or with CONFIG_BASE_SMALL=1) then we
 364                 * use the maximum timeout.
 365                 */
 366                if (idx > MAX_TVAL) {
 367                        idx = MAX_TVAL;
 368                        expires = idx + base->timer_jiffies;
 369                }
 370                i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
 371                vec = base->tv5.vec + i;
 372        }
 373        /*
 374         * Timers are FIFO:
 375         */
 376        list_add_tail(&timer->entry, vec);
 377}
 378
 379static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 380{
 381        __internal_add_timer(base, timer);
 382        /*
 383         * Update base->active_timers and base->next_timer
 384         */
 385        if (!tbase_get_deferrable(timer->base)) {
 386                if (time_before(timer->expires, base->next_timer))
 387                        base->next_timer = timer->expires;
 388                base->active_timers++;
 389        }
 390}
 391
 392#ifdef CONFIG_TIMER_STATS
 393void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
 394{
 395        if (timer->start_site)
 396                return;
 397
 398        timer->start_site = addr;
 399        memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
 400        timer->start_pid = current->pid;
 401}
 402
 403static void timer_stats_account_timer(struct timer_list *timer)
 404{
 405        unsigned int flag = 0;
 406
 407        if (likely(!timer->start_site))
 408                return;
 409        if (unlikely(tbase_get_deferrable(timer->base)))
 410                flag |= TIMER_STATS_FLAG_DEFERRABLE;
 411
 412        timer_stats_update_stats(timer, timer->start_pid, timer->start_site,
 413                                 timer->function, timer->start_comm, flag);
 414}
 415
 416#else
 417static void timer_stats_account_timer(struct timer_list *timer) {}
 418#endif
 419
 420#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 421
 422static struct debug_obj_descr timer_debug_descr;
 423
 424static void *timer_debug_hint(void *addr)
 425{
 426        return ((struct timer_list *) addr)->function;
 427}
 428
 429/*
 430 * fixup_init is called when:
 431 * - an active object is initialized
 432 */
 433static int timer_fixup_init(void *addr, enum debug_obj_state state)
 434{
 435        struct timer_list *timer = addr;
 436
 437        switch (state) {
 438        case ODEBUG_STATE_ACTIVE:
 439                del_timer_sync(timer);
 440                debug_object_init(timer, &timer_debug_descr);
 441                return 1;
 442        default:
 443                return 0;
 444        }
 445}
 446
 447/* Stub timer callback for improperly used timers. */
 448static void stub_timer(unsigned long data)
 449{
 450        WARN_ON(1);
 451}
 452
 453/*
 454 * fixup_activate is called when:
 455 * - an active object is activated
 456 * - an unknown object is activated (might be a statically initialized object)
 457 */
 458static int timer_fixup_activate(void *addr, enum debug_obj_state state)
 459{
 460        struct timer_list *timer = addr;
 461
 462        switch (state) {
 463
 464        case ODEBUG_STATE_NOTAVAILABLE:
 465                /*
 466                 * This is not really a fixup. The timer was
 467                 * statically initialized. We just make sure that it
 468                 * is tracked in the object tracker.
 469                 */
 470                if (timer->entry.next == NULL &&
 471                    timer->entry.prev == TIMER_ENTRY_STATIC) {
 472                        debug_object_init(timer, &timer_debug_descr);
 473                        debug_object_activate(timer, &timer_debug_descr);
 474                        return 0;
 475                } else {
 476                        setup_timer(timer, stub_timer, 0);
 477                        return 1;
 478                }
 479                return 0;
 480
 481        case ODEBUG_STATE_ACTIVE:
 482                WARN_ON(1);
 483
 484        default:
 485                return 0;
 486        }
 487}
 488
 489/*
 490 * fixup_free is called when:
 491 * - an active object is freed
 492 */
 493static int timer_fixup_free(void *addr, enum debug_obj_state state)
 494{
 495        struct timer_list *timer = addr;
 496
 497        switch (state) {
 498        case ODEBUG_STATE_ACTIVE:
 499                del_timer_sync(timer);
 500                debug_object_free(timer, &timer_debug_descr);
 501                return 1;
 502        default:
 503                return 0;
 504        }
 505}
 506
 507/*
 508 * fixup_assert_init is called when:
 509 * - an untracked/uninit-ed object is found
 510 */
 511static int timer_fixup_assert_init(void *addr, enum debug_obj_state state)
 512{
 513        struct timer_list *timer = addr;
 514
 515        switch (state) {
 516        case ODEBUG_STATE_NOTAVAILABLE:
 517                if (timer->entry.prev == TIMER_ENTRY_STATIC) {
 518                        /*
 519                         * This is not really a fixup. The timer was
 520                         * statically initialized. We just make sure that it
 521                         * is tracked in the object tracker.
 522                         */
 523                        debug_object_init(timer, &timer_debug_descr);
 524                        return 0;
 525                } else {
 526                        setup_timer(timer, stub_timer, 0);
 527                        return 1;
 528                }
 529        default:
 530                return 0;
 531        }
 532}
 533
 534static struct debug_obj_descr timer_debug_descr = {
 535        .name                   = "timer_list",
 536        .debug_hint             = timer_debug_hint,
 537        .fixup_init             = timer_fixup_init,
 538        .fixup_activate         = timer_fixup_activate,
 539        .fixup_free             = timer_fixup_free,
 540        .fixup_assert_init      = timer_fixup_assert_init,
 541};
 542
 543static inline void debug_timer_init(struct timer_list *timer)
 544{
 545        debug_object_init(timer, &timer_debug_descr);
 546}
 547
 548static inline void debug_timer_activate(struct timer_list *timer)
 549{
 550        debug_object_activate(timer, &timer_debug_descr);
 551}
 552
 553static inline void debug_timer_deactivate(struct timer_list *timer)
 554{
 555        debug_object_deactivate(timer, &timer_debug_descr);
 556}
 557
 558static inline void debug_timer_free(struct timer_list *timer)
 559{
 560        debug_object_free(timer, &timer_debug_descr);
 561}
 562
 563static inline void debug_timer_assert_init(struct timer_list *timer)
 564{
 565        debug_object_assert_init(timer, &timer_debug_descr);
 566}
 567
 568static void __init_timer(struct timer_list *timer,
 569                         const char *name,
 570                         struct lock_class_key *key);
 571
 572void init_timer_on_stack_key(struct timer_list *timer,
 573                             const char *name,
 574                             struct lock_class_key *key)
 575{
 576        debug_object_init_on_stack(timer, &timer_debug_descr);
 577        __init_timer(timer, name, key);
 578}
 579EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
 580
 581void destroy_timer_on_stack(struct timer_list *timer)
 582{
 583        debug_object_free(timer, &timer_debug_descr);
 584}
 585EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
 586
 587#else
 588static inline void debug_timer_init(struct timer_list *timer) { }
 589static inline void debug_timer_activate(struct timer_list *timer) { }
 590static inline void debug_timer_deactivate(struct timer_list *timer) { }
 591static inline void debug_timer_assert_init(struct timer_list *timer) { }
 592#endif
 593
 594static inline void debug_init(struct timer_list *timer)
 595{
 596        debug_timer_init(timer);
 597        trace_timer_init(timer);
 598}
 599
 600static inline void
 601debug_activate(struct timer_list *timer, unsigned long expires)
 602{
 603        debug_timer_activate(timer);
 604        trace_timer_start(timer, expires);
 605}
 606
 607static inline void debug_deactivate(struct timer_list *timer)
 608{
 609        debug_timer_deactivate(timer);
 610        trace_timer_cancel(timer);
 611}
 612
 613static inline void debug_assert_init(struct timer_list *timer)
 614{
 615        debug_timer_assert_init(timer);
 616}
 617
 618static void __init_timer(struct timer_list *timer,
 619                         const char *name,
 620                         struct lock_class_key *key)
 621{
 622        timer->entry.next = NULL;
 623        timer->base = __raw_get_cpu_var(tvec_bases);
 624        timer->slack = -1;
 625#ifdef CONFIG_TIMER_STATS
 626        timer->start_site = NULL;
 627        timer->start_pid = -1;
 628        memset(timer->start_comm, 0, TASK_COMM_LEN);
 629#endif
 630        lockdep_init_map(&timer->lockdep_map, name, key, 0);
 631}
 632
 633void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
 634                                         const char *name,
 635                                         struct lock_class_key *key,
 636                                         void (*function)(unsigned long),
 637                                         unsigned long data)
 638{
 639        timer->function = function;
 640        timer->data = data;
 641        init_timer_on_stack_key(timer, name, key);
 642        timer_set_deferrable(timer);
 643}
 644EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
 645
 646/**
 647 * init_timer_key - initialize a timer
 648 * @timer: the timer to be initialized
 649 * @name: name of the timer
 650 * @key: lockdep class key of the fake lock used for tracking timer
 651 *       sync lock dependencies
 652 *
 653 * init_timer_key() must be done to a timer prior calling *any* of the
 654 * other timer functions.
 655 */
 656void init_timer_key(struct timer_list *timer,
 657                    const char *name,
 658                    struct lock_class_key *key)
 659{
 660        debug_init(timer);
 661        __init_timer(timer, name, key);
 662}
 663EXPORT_SYMBOL(init_timer_key);
 664
 665void init_timer_deferrable_key(struct timer_list *timer,
 666                               const char *name,
 667                               struct lock_class_key *key)
 668{
 669        init_timer_key(timer, name, key);
 670        timer_set_deferrable(timer);
 671}
 672EXPORT_SYMBOL(init_timer_deferrable_key);
 673
 674static inline void detach_timer(struct timer_list *timer, bool clear_pending)
 675{
 676        struct list_head *entry = &timer->entry;
 677
 678        debug_deactivate(timer);
 679
 680        __list_del(entry->prev, entry->next);
 681        if (clear_pending)
 682                entry->next = NULL;
 683        entry->prev = LIST_POISON2;
 684}
 685
 686static inline void
 687detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
 688{
 689        detach_timer(timer, true);
 690        if (!tbase_get_deferrable(timer->base))
 691                timer->base->active_timers--;
 692}
 693
 694static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
 695                             bool clear_pending)
 696{
 697        if (!timer_pending(timer))
 698                return 0;
 699
 700        detach_timer(timer, clear_pending);
 701        if (!tbase_get_deferrable(timer->base)) {
 702                timer->base->active_timers--;
 703                if (timer->expires == base->next_timer)
 704                        base->next_timer = base->timer_jiffies;
 705        }
 706        return 1;
 707}
 708
 709/*
 710 * We are using hashed locking: holding per_cpu(tvec_bases).lock
 711 * means that all timers which are tied to this base via timer->base are
 712 * locked, and the base itself is locked too.
 713 *
 714 * So __run_timers/migrate_timers can safely modify all timers which could
 715 * be found on ->tvX lists.
 716 *
 717 * When the timer's base is locked, and the timer removed from list, it is
 718 * possible to set timer->base = NULL and drop the lock: the timer remains
 719 * locked.
 720 */
 721static struct tvec_base *lock_timer_base(struct timer_list *timer,
 722                                        unsigned long *flags)
 723        __acquires(timer->base->lock)
 724{
 725        struct tvec_base *base;
 726
 727        for (;;) {
 728                struct tvec_base *prelock_base = timer->base;
 729                base = tbase_get_base(prelock_base);
 730                if (likely(base != NULL)) {
 731                        spin_lock_irqsave(&base->lock, *flags);
 732                        if (likely(prelock_base == timer->base))
 733                                return base;
 734                        /* The timer has migrated to another CPU */
 735                        spin_unlock_irqrestore(&base->lock, *flags);
 736                }
 737                cpu_relax();
 738        }
 739}
 740
 741static inline int
 742__mod_timer(struct timer_list *timer, unsigned long expires,
 743                                                bool pending_only, int pinned)
 744{
 745        struct tvec_base *base, *new_base;
 746        unsigned long flags;
 747        int ret = 0 , cpu;
 748
 749        timer_stats_timer_set_start_info(timer);
 750        BUG_ON(!timer->function);
 751
 752        base = lock_timer_base(timer, &flags);
 753
 754        ret = detach_if_pending(timer, base, false);
 755        if (!ret && pending_only)
 756                goto out_unlock;
 757
 758        debug_activate(timer, expires);
 759
 760        cpu = smp_processor_id();
 761
 762#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
 763        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu))
 764                cpu = get_nohz_timer_target();
 765#endif
 766        new_base = per_cpu(tvec_bases, cpu);
 767
 768        if (base != new_base) {
 769                /*
 770                 * We are trying to schedule the timer on the local CPU.
 771                 * However we can't change timer's base while it is running,
 772                 * otherwise del_timer_sync() can't detect that the timer's
 773                 * handler yet has not finished. This also guarantees that
 774                 * the timer is serialized wrt itself.
 775                 */
 776                if (likely(base->running_timer != timer)) {
 777                        /* See the comment in lock_timer_base() */
 778                        timer_set_base(timer, NULL);
 779                        spin_unlock(&base->lock);
 780                        base = new_base;
 781                        spin_lock(&base->lock);
 782                        timer_set_base(timer, base);
 783                }
 784        }
 785
 786        timer->expires = expires;
 787        internal_add_timer(base, timer);
 788
 789out_unlock:
 790        spin_unlock_irqrestore(&base->lock, flags);
 791
 792        return ret;
 793}
 794
 795/**
 796 * mod_timer_pending - modify a pending timer's timeout
 797 * @timer: the pending timer to be modified
 798 * @expires: new timeout in jiffies
 799 *
 800 * mod_timer_pending() is the same for pending timers as mod_timer(),
 801 * but will not re-activate and modify already deleted timers.
 802 *
 803 * It is useful for unserialized use of timers.
 804 */
 805int mod_timer_pending(struct timer_list *timer, unsigned long expires)
 806{
 807        return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
 808}
 809EXPORT_SYMBOL(mod_timer_pending);
 810
 811/*
 812 * Decide where to put the timer while taking the slack into account
 813 *
 814 * Algorithm:
 815 *   1) calculate the maximum (absolute) time
 816 *   2) calculate the highest bit where the expires and new max are different
 817 *   3) use this bit to make a mask
 818 *   4) use the bitmask to round down the maximum time, so that all last
 819 *      bits are zeros
 820 */
 821static inline
 822unsigned long apply_slack(struct timer_list *timer, unsigned long expires)
 823{
 824        unsigned long expires_limit, mask;
 825        int bit;
 826
 827        if (timer->slack >= 0) {
 828                expires_limit = expires + timer->slack;
 829        } else {
 830                long delta = expires - jiffies;
 831
 832                if (delta < 256)
 833                        return expires;
 834
 835                expires_limit = expires + delta / 256;
 836        }
 837        mask = expires ^ expires_limit;
 838        if (mask == 0)
 839                return expires;
 840
 841        bit = find_last_bit(&mask, BITS_PER_LONG);
 842
 843        mask = (1 << bit) - 1;
 844
 845        expires_limit = expires_limit & ~(mask);
 846
 847        return expires_limit;
 848}
 849
 850/**
 851 * mod_timer - modify a timer's timeout
 852 * @timer: the timer to be modified
 853 * @expires: new timeout in jiffies
 854 *
 855 * mod_timer() is a more efficient way to update the expire field of an
 856 * active timer (if the timer is inactive it will be activated)
 857 *
 858 * mod_timer(timer, expires) is equivalent to:
 859 *
 860 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 861 *
 862 * Note that if there are multiple unserialized concurrent users of the
 863 * same timer, then mod_timer() is the only safe way to modify the timeout,
 864 * since add_timer() cannot modify an already running timer.
 865 *
 866 * The function returns whether it has modified a pending timer or not.
 867 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
 868 * active timer returns 1.)
 869 */
 870int mod_timer(struct timer_list *timer, unsigned long expires)
 871{
 872        expires = apply_slack(timer, expires);
 873
 874        /*
 875         * This is a common optimization triggered by the
 876         * networking code - if the timer is re-modified
 877         * to be the same thing then just return:
 878         */
 879        if (timer_pending(timer) && timer->expires == expires)
 880                return 1;
 881
 882        return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
 883}
 884EXPORT_SYMBOL(mod_timer);
 885
 886/**
 887 * mod_timer_pinned - modify a timer's timeout
 888 * @timer: the timer to be modified
 889 * @expires: new timeout in jiffies
 890 *
 891 * mod_timer_pinned() is a way to update the expire field of an
 892 * active timer (if the timer is inactive it will be activated)
 893 * and to ensure that the timer is scheduled on the current CPU.
 894 *
 895 * Note that this does not prevent the timer from being migrated
 896 * when the current CPU goes offline.  If this is a problem for
 897 * you, use CPU-hotplug notifiers to handle it correctly, for
 898 * example, cancelling the timer when the corresponding CPU goes
 899 * offline.
 900 *
 901 * mod_timer_pinned(timer, expires) is equivalent to:
 902 *
 903 *     del_timer(timer); timer->expires = expires; add_timer(timer);
 904 */
 905int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
 906{
 907        if (timer->expires == expires && timer_pending(timer))
 908                return 1;
 909
 910        return __mod_timer(timer, expires, false, TIMER_PINNED);
 911}
 912EXPORT_SYMBOL(mod_timer_pinned);
 913
 914/**
 915 * add_timer - start a timer
 916 * @timer: the timer to be added
 917 *
 918 * The kernel will do a ->function(->data) callback from the
 919 * timer interrupt at the ->expires point in the future. The
 920 * current time is 'jiffies'.
 921 *
 922 * The timer's ->expires, ->function (and if the handler uses it, ->data)
 923 * fields must be set prior calling this function.
 924 *
 925 * Timers with an ->expires field in the past will be executed in the next
 926 * timer tick.
 927 */
 928void add_timer(struct timer_list *timer)
 929{
 930        BUG_ON(timer_pending(timer));
 931        mod_timer(timer, timer->expires);
 932}
 933EXPORT_SYMBOL(add_timer);
 934
 935/**
 936 * add_timer_on - start a timer on a particular CPU
 937 * @timer: the timer to be added
 938 * @cpu: the CPU to start it on
 939 *
 940 * This is not very scalable on SMP. Double adds are not possible.
 941 */
 942void add_timer_on(struct timer_list *timer, int cpu)
 943{
 944        struct tvec_base *base = per_cpu(tvec_bases, cpu);
 945        unsigned long flags;
 946
 947        timer_stats_timer_set_start_info(timer);
 948        BUG_ON(timer_pending(timer) || !timer->function);
 949        spin_lock_irqsave(&base->lock, flags);
 950        timer_set_base(timer, base);
 951        debug_activate(timer, timer->expires);
 952        internal_add_timer(base, timer);
 953        /*
 954         * Check whether the other CPU is idle and needs to be
 955         * triggered to reevaluate the timer wheel when nohz is
 956         * active. We are protected against the other CPU fiddling
 957         * with the timer by holding the timer base lock. This also
 958         * makes sure that a CPU on the way to idle can not evaluate
 959         * the timer wheel.
 960         */
 961        wake_up_idle_cpu(cpu);
 962        spin_unlock_irqrestore(&base->lock, flags);
 963}
 964EXPORT_SYMBOL_GPL(add_timer_on);
 965
 966/**
 967 * del_timer - deactive a timer.
 968 * @timer: the timer to be deactivated
 969 *
 970 * del_timer() deactivates a timer - this works on both active and inactive
 971 * timers.
 972 *
 973 * The function returns whether it has deactivated a pending timer or not.
 974 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
 975 * active timer returns 1.)
 976 */
 977int del_timer(struct timer_list *timer)
 978{
 979        struct tvec_base *base;
 980        unsigned long flags;
 981        int ret = 0;
 982
 983        debug_assert_init(timer);
 984
 985        timer_stats_timer_clear_start_info(timer);
 986        if (timer_pending(timer)) {
 987                base = lock_timer_base(timer, &flags);
 988                ret = detach_if_pending(timer, base, true);
 989                spin_unlock_irqrestore(&base->lock, flags);
 990        }
 991
 992        return ret;
 993}
 994EXPORT_SYMBOL(del_timer);
 995
 996/**
 997 * try_to_del_timer_sync - Try to deactivate a timer
 998 * @timer: timer do del
 999 *
1000 * This function tries to deactivate a timer. Upon successful (ret >= 0)
1001 * exit the timer is not queued and the handler is not running on any CPU.
1002 */
1003int try_to_del_timer_sync(struct timer_list *timer)
1004{
1005        struct tvec_base *base;
1006        unsigned long flags;
1007        int ret = -1;
1008
1009        debug_assert_init(timer);
1010
1011        base = lock_timer_base(timer, &flags);
1012
1013        if (base->running_timer != timer) {
1014                timer_stats_timer_clear_start_info(timer);
1015                ret = detach_if_pending(timer, base, true);
1016        }
1017        spin_unlock_irqrestore(&base->lock, flags);
1018
1019        return ret;
1020}
1021EXPORT_SYMBOL(try_to_del_timer_sync);
1022
1023#ifdef CONFIG_SMP
1024/**
1025 * del_timer_sync - deactivate a timer and wait for the handler to finish.
1026 * @timer: the timer to be deactivated
1027 *
1028 * This function only differs from del_timer() on SMP: besides deactivating
1029 * the timer it also makes sure the handler has finished executing on other
1030 * CPUs.
1031 *
1032 * Synchronization rules: Callers must prevent restarting of the timer,
1033 * otherwise this function is meaningless. It must not be called from
1034 * interrupt contexts. The caller must not hold locks which would prevent
1035 * completion of the timer's handler. The timer's handler must not call
1036 * add_timer_on(). Upon exit the timer is not queued and the handler is
1037 * not running on any CPU.
1038 *
1039 * Note: You must not hold locks that are held in interrupt context
1040 *   while calling this function. Even if the lock has nothing to do
1041 *   with the timer in question.  Here's why:
1042 *
1043 *    CPU0                             CPU1
1044 *    ----                             ----
1045 *                                   <SOFTIRQ>
1046 *                                   call_timer_fn();
1047 *                                     base->running_timer = mytimer;
1048 *  spin_lock_irq(somelock);
1049 *                                     <IRQ>
1050 *                                        spin_lock(somelock);
1051 *  del_timer_sync(mytimer);
1052 *   while (base->running_timer == mytimer);
1053 *
1054 * Now del_timer_sync() will never return and never release somelock.
1055 * The interrupt on the other CPU is waiting to grab somelock but
1056 * it has interrupted the softirq that CPU0 is waiting to finish.
1057 *
1058 * The function returns whether it has deactivated a pending timer or not.
1059 */
1060int del_timer_sync(struct timer_list *timer)
1061{
1062#ifdef CONFIG_LOCKDEP
1063        unsigned long flags;
1064
1065        /*
1066         * If lockdep gives a backtrace here, please reference
1067         * the synchronization rules above.
1068         */
1069        local_irq_save(flags);
1070        lock_map_acquire(&timer->lockdep_map);
1071        lock_map_release(&timer->lockdep_map);
1072        local_irq_restore(flags);
1073#endif
1074        /*
1075         * don't use it in hardirq context, because it
1076         * could lead to deadlock.
1077         */
1078        WARN_ON(in_irq());
1079        for (;;) {
1080                int ret = try_to_del_timer_sync(timer);
1081                if (ret >= 0)
1082                        return ret;
1083                cpu_relax();
1084        }
1085}
1086EXPORT_SYMBOL(del_timer_sync);
1087#endif
1088
1089static int cascade(struct tvec_base *base, struct tvec *tv, int index)
1090{
1091        /* cascade all the timers from tv up one level */
1092        struct timer_list *timer, *tmp;
1093        struct list_head tv_list;
1094
1095        list_replace_init(tv->vec + index, &tv_list);
1096
1097        /*
1098         * We are removing _all_ timers from the list, so we
1099         * don't have to detach them individually.
1100         */
1101        list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
1102                BUG_ON(tbase_get_base(timer->base) != base);
1103                /* No accounting, while moving them */
1104                __internal_add_timer(base, timer);
1105        }
1106
1107        return index;
1108}
1109
1110static void call_timer_fn(struct timer_list *timer, void (*fn)(unsigned long),
1111                          unsigned long data)
1112{
1113        int preempt_count = preempt_count();
1114
1115#ifdef CONFIG_LOCKDEP
1116        /*
1117         * It is permissible to free the timer from inside the
1118         * function that is called from it, this we need to take into
1119         * account for lockdep too. To avoid bogus "held lock freed"
1120         * warnings as well as problems when looking into
1121         * timer->lockdep_map, make a copy and use that here.
1122         */
1123        struct lockdep_map lockdep_map;
1124
1125        lockdep_copy_map(&lockdep_map, &timer->lockdep_map);
1126#endif
1127        /*
1128         * Couple the lock chain with the lock chain at
1129         * del_timer_sync() by acquiring the lock_map around the fn()
1130         * call here and in del_timer_sync().
1131         */
1132        lock_map_acquire(&lockdep_map);
1133
1134        trace_timer_expire_entry(timer);
1135        fn(data);
1136        trace_timer_expire_exit(timer);
1137
1138        lock_map_release(&lockdep_map);
1139
1140        if (preempt_count != preempt_count()) {
1141                WARN_ONCE(1, "timer: %pF preempt leak: %08x -> %08x\n",
1142                          fn, preempt_count, preempt_count());
1143                /*
1144                 * Restore the preempt count. That gives us a decent
1145                 * chance to survive and extract information. If the
1146                 * callback kept a lock held, bad luck, but not worse
1147                 * than the BUG() we had.
1148                 */
1149                preempt_count() = preempt_count;
1150        }
1151}
1152
1153#define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
1154
1155/**
1156 * __run_timers - run all expired timers (if any) on this CPU.
1157 * @base: the timer vector to be processed.
1158 *
1159 * This function cascades all vectors and executes all expired timer
1160 * vectors.
1161 */
1162static inline void __run_timers(struct tvec_base *base)
1163{
1164        struct timer_list *timer;
1165
1166        spin_lock_irq(&base->lock);
1167        while (time_after_eq(jiffies, base->timer_jiffies)) {
1168                struct list_head work_list;
1169                struct list_head *head = &work_list;
1170                int index = base->timer_jiffies & TVR_MASK;
1171
1172                /*
1173                 * Cascade timers:
1174                 */
1175                if (!index &&
1176                        (!cascade(base, &base->tv2, INDEX(0))) &&
1177                                (!cascade(base, &base->tv3, INDEX(1))) &&
1178                                        !cascade(base, &base->tv4, INDEX(2)))
1179                        cascade(base, &base->tv5, INDEX(3));
1180                ++base->timer_jiffies;
1181                list_replace_init(base->tv1.vec + index, &work_list);
1182                while (!list_empty(head)) {
1183                        void (*fn)(unsigned long);
1184                        unsigned long data;
1185
1186                        timer = list_first_entry(head, struct timer_list,entry);
1187                        fn = timer->function;
1188                        data = timer->data;
1189
1190                        timer_stats_account_timer(timer);
1191
1192                        base->running_timer = timer;
1193                        detach_expired_timer(timer, base);
1194
1195                        spin_unlock_irq(&base->lock);
1196                        call_timer_fn(timer, fn, data);
1197                        spin_lock_irq(&base->lock);
1198                }
1199        }
1200        base->running_timer = NULL;
1201        spin_unlock_irq(&base->lock);
1202}
1203
1204#ifdef CONFIG_NO_HZ
1205/*
1206 * Find out when the next timer event is due to happen. This
1207 * is used on S/390 to stop all activity when a CPU is idle.
1208 * This function needs to be called with interrupts disabled.
1209 */
1210static unsigned long __next_timer_interrupt(struct tvec_base *base)
1211{
1212        unsigned long timer_jiffies = base->timer_jiffies;
1213        unsigned long expires = timer_jiffies + NEXT_TIMER_MAX_DELTA;
1214        int index, slot, array, found = 0;
1215        struct timer_list *nte;
1216        struct tvec *varray[4];
1217
1218        /* Look for timer events in tv1. */
1219        index = slot = timer_jiffies & TVR_MASK;
1220        do {
1221                list_for_each_entry(nte, base->tv1.vec + slot, entry) {
1222                        if (tbase_get_deferrable(nte->base))
1223                                continue;
1224
1225                        found = 1;
1226                        expires = nte->expires;
1227                        /* Look at the cascade bucket(s)? */
1228                        if (!index || slot < index)
1229                                goto cascade;
1230                        return expires;
1231                }
1232                slot = (slot + 1) & TVR_MASK;
1233        } while (slot != index);
1234
1235cascade:
1236        /* Calculate the next cascade event */
1237        if (index)
1238                timer_jiffies += TVR_SIZE - index;
1239        timer_jiffies >>= TVR_BITS;
1240
1241        /* Check tv2-tv5. */
1242        varray[0] = &base->tv2;
1243        varray[1] = &base->tv3;
1244        varray[2] = &base->tv4;
1245        varray[3] = &base->tv5;
1246
1247        for (array = 0; array < 4; array++) {
1248                struct tvec *varp = varray[array];
1249
1250                index = slot = timer_jiffies & TVN_MASK;
1251                do {
1252                        list_for_each_entry(nte, varp->vec + slot, entry) {
1253                                if (tbase_get_deferrable(nte->base))
1254                                        continue;
1255
1256                                found = 1;
1257                                if (time_before(nte->expires, expires))
1258                                        expires = nte->expires;
1259                        }
1260                        /*
1261                         * Do we still search for the first timer or are
1262                         * we looking up the cascade buckets ?
1263                         */
1264                        if (found) {
1265                                /* Look at the cascade bucket(s)? */
1266                                if (!index || slot < index)
1267                                        break;
1268                                return expires;
1269                        }
1270                        slot = (slot + 1) & TVN_MASK;
1271                } while (slot != index);
1272
1273                if (index)
1274                        timer_jiffies += TVN_SIZE - index;
1275                timer_jiffies >>= TVN_BITS;
1276        }
1277        return expires;
1278}
1279
1280/*
1281 * Check, if the next hrtimer event is before the next timer wheel
1282 * event:
1283 */
1284static unsigned long cmp_next_hrtimer_event(unsigned long now,
1285                                            unsigned long expires)
1286{
1287        ktime_t hr_delta = hrtimer_get_next_event();
1288        struct timespec tsdelta;
1289        unsigned long delta;
1290
1291        if (hr_delta.tv64 == KTIME_MAX)
1292                return expires;
1293
1294        /*
1295         * Expired timer available, let it expire in the next tick
1296         */
1297        if (hr_delta.tv64 <= 0)
1298                return now + 1;
1299
1300        tsdelta = ktime_to_timespec(hr_delta);
1301        delta = timespec_to_jiffies(&tsdelta);
1302
1303        /*
1304         * Limit the delta to the max value, which is checked in
1305         * tick_nohz_stop_sched_tick():
1306         */
1307        if (delta > NEXT_TIMER_MAX_DELTA)
1308                delta = NEXT_TIMER_MAX_DELTA;
1309
1310        /*
1311         * Take rounding errors in to account and make sure, that it
1312         * expires in the next tick. Otherwise we go into an endless
1313         * ping pong due to tick_nohz_stop_sched_tick() retriggering
1314         * the timer softirq
1315         */
1316        if (delta < 1)
1317                delta = 1;
1318        now += delta;
1319        if (time_before(now, expires))
1320                return now;
1321        return expires;
1322}
1323
1324/**
1325 * get_next_timer_interrupt - return the jiffy of the next pending timer
1326 * @now: current time (in jiffies)
1327 */
1328unsigned long get_next_timer_interrupt(unsigned long now)
1329{
1330        struct tvec_base *base = __this_cpu_read(tvec_bases);
1331        unsigned long expires = now + NEXT_TIMER_MAX_DELTA;
1332
1333        /*
1334         * Pretend that there is no timer pending if the cpu is offline.
1335         * Possible pending timers will be migrated later to an active cpu.
1336         */
1337        if (cpu_is_offline(smp_processor_id()))
1338                return expires;
1339
1340        spin_lock(&base->lock);
1341        if (base->active_timers) {
1342                if (time_before_eq(base->next_timer, base->timer_jiffies))
1343                        base->next_timer = __next_timer_interrupt(base);
1344                expires = base->next_timer;
1345        }
1346        spin_unlock(&base->lock);
1347
1348        if (time_before_eq(expires, now))
1349                return now;
1350
1351        return cmp_next_hrtimer_event(now, expires);
1352}
1353#endif
1354
1355/*
1356 * Called from the timer interrupt handler to charge one tick to the current
1357 * process.  user_tick is 1 if the tick is user time, 0 for system.
1358 */
1359void update_process_times(int user_tick)
1360{
1361        struct task_struct *p = current;
1362        int cpu = smp_processor_id();
1363
1364        /* Note: this timer irq context must be accounted for as well. */
1365        account_process_tick(p, user_tick);
1366        run_local_timers();
1367        rcu_check_callbacks(cpu, user_tick);
1368        printk_tick();
1369#ifdef CONFIG_IRQ_WORK
1370        if (in_irq())
1371                irq_work_run();
1372#endif
1373        scheduler_tick();
1374        run_posix_cpu_timers(p);
1375}
1376
1377/*
1378 * This function runs timers and the timer-tq in bottom half context.
1379 */
1380static void run_timer_softirq(struct softirq_action *h)
1381{
1382        struct tvec_base *base = __this_cpu_read(tvec_bases);
1383
1384        hrtimer_run_pending();
1385
1386        if (time_after_eq(jiffies, base->timer_jiffies))
1387                __run_timers(base);
1388}
1389
1390/*
1391 * Called by the local, per-CPU timer interrupt on SMP.
1392 */
1393void run_local_timers(void)
1394{
1395        hrtimer_run_queues();
1396        raise_softirq(TIMER_SOFTIRQ);
1397}
1398
1399#ifdef __ARCH_WANT_SYS_ALARM
1400
1401/*
1402 * For backwards compatibility?  This can be done in libc so Alpha
1403 * and all newer ports shouldn't need it.
1404 */
1405SYSCALL_DEFINE1(alarm, unsigned int, seconds)
1406{
1407        return alarm_setitimer(seconds);
1408}
1409
1410#endif
1411
1412/**
1413 * sys_getpid - return the thread group id of the current process
1414 *
1415 * Note, despite the name, this returns the tgid not the pid.  The tgid and
1416 * the pid are identical unless CLONE_THREAD was specified on clone() in
1417 * which case the tgid is the same in all threads of the same group.
1418 *
1419 * This is SMP safe as current->tgid does not change.
1420 */
1421SYSCALL_DEFINE0(getpid)
1422{
1423        return task_tgid_vnr(current);
1424}
1425
1426/*
1427 * Accessing ->real_parent is not SMP-safe, it could
1428 * change from under us. However, we can use a stale
1429 * value of ->real_parent under rcu_read_lock(), see
1430 * release_task()->call_rcu(delayed_put_task_struct).
1431 */
1432SYSCALL_DEFINE0(getppid)
1433{
1434        int pid;
1435
1436        rcu_read_lock();
1437        pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1438        rcu_read_unlock();
1439
1440        return pid;
1441}
1442
1443SYSCALL_DEFINE0(getuid)
1444{
1445        /* Only we change this so SMP safe */
1446        return from_kuid_munged(current_user_ns(), current_uid());
1447}
1448
1449SYSCALL_DEFINE0(geteuid)
1450{
1451        /* Only we change this so SMP safe */
1452        return from_kuid_munged(current_user_ns(), current_euid());
1453}
1454
1455SYSCALL_DEFINE0(getgid)
1456{
1457        /* Only we change this so SMP safe */
1458        return from_kgid_munged(current_user_ns(), current_gid());
1459}
1460
1461SYSCALL_DEFINE0(getegid)
1462{
1463        /* Only we change this so SMP safe */
1464        return from_kgid_munged(current_user_ns(), current_egid());
1465}
1466
1467static void process_timeout(unsigned long __data)
1468{
1469        wake_up_process((struct task_struct *)__data);
1470}
1471
1472/**
1473 * schedule_timeout - sleep until timeout
1474 * @timeout: timeout value in jiffies
1475 *
1476 * Make the current task sleep until @timeout jiffies have
1477 * elapsed. The routine will return immediately unless
1478 * the current task state has been set (see set_current_state()).
1479 *
1480 * You can set the task state as follows -
1481 *
1482 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1483 * pass before the routine returns. The routine will return 0
1484 *
1485 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1486 * delivered to the current task. In this case the remaining time
1487 * in jiffies will be returned, or 0 if the timer expired in time
1488 *
1489 * The current task state is guaranteed to be TASK_RUNNING when this
1490 * routine returns.
1491 *
1492 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1493 * the CPU away without a bound on the timeout. In this case the return
1494 * value will be %MAX_SCHEDULE_TIMEOUT.
1495 *
1496 * In all cases the return value is guaranteed to be non-negative.
1497 */
1498signed long __sched schedule_timeout(signed long timeout)
1499{
1500        struct timer_list timer;
1501        unsigned long expire;
1502
1503        switch (timeout)
1504        {
1505        case MAX_SCHEDULE_TIMEOUT:
1506                /*
1507                 * These two special cases are useful to be comfortable
1508                 * in the caller. Nothing more. We could take
1509                 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1510                 * but I' d like to return a valid offset (>=0) to allow
1511                 * the caller to do everything it want with the retval.
1512                 */
1513                schedule();
1514                goto out;
1515        default:
1516                /*
1517                 * Another bit of PARANOID. Note that the retval will be
1518                 * 0 since no piece of kernel is supposed to do a check
1519                 * for a negative retval of schedule_timeout() (since it
1520                 * should never happens anyway). You just have the printk()
1521                 * that will tell you if something is gone wrong and where.
1522                 */
1523                if (timeout < 0) {
1524                        printk(KERN_ERR "schedule_timeout: wrong timeout "
1525                                "value %lx\n", timeout);
1526                        dump_stack();
1527                        current->state = TASK_RUNNING;
1528                        goto out;
1529                }
1530        }
1531
1532        expire = timeout + jiffies;
1533
1534        setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1535        __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
1536        schedule();
1537        del_singleshot_timer_sync(&timer);
1538
1539        /* Remove the timer from the object tracker */
1540        destroy_timer_on_stack(&timer);
1541
1542        timeout = expire - jiffies;
1543
1544 out:
1545        return timeout < 0 ? 0 : timeout;
1546}
1547EXPORT_SYMBOL(schedule_timeout);
1548
1549/*
1550 * We can use __set_current_state() here because schedule_timeout() calls
1551 * schedule() unconditionally.
1552 */
1553signed long __sched schedule_timeout_interruptible(signed long timeout)
1554{
1555        __set_current_state(TASK_INTERRUPTIBLE);
1556        return schedule_timeout(timeout);
1557}
1558EXPORT_SYMBOL(schedule_timeout_interruptible);
1559
1560signed long __sched schedule_timeout_killable(signed long timeout)
1561{
1562        __set_current_state(TASK_KILLABLE);
1563        return schedule_timeout(timeout);
1564}
1565EXPORT_SYMBOL(schedule_timeout_killable);
1566
1567signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1568{
1569        __set_current_state(TASK_UNINTERRUPTIBLE);
1570        return schedule_timeout(timeout);
1571}
1572EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1573
1574/* Thread ID - the internal kernel "pid" */
1575SYSCALL_DEFINE0(gettid)
1576{
1577        return task_pid_vnr(current);
1578}
1579
1580/**
1581 * do_sysinfo - fill in sysinfo struct
1582 * @info: pointer to buffer to fill
1583 */
1584int do_sysinfo(struct sysinfo *info)
1585{
1586        unsigned long mem_total, sav_total;
1587        unsigned int mem_unit, bitcount;
1588        struct timespec tp;
1589
1590        memset(info, 0, sizeof(struct sysinfo));
1591
1592        ktime_get_ts(&tp);
1593        monotonic_to_bootbased(&tp);
1594        info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1595
1596        get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
1597
1598        info->procs = nr_threads;
1599
1600        si_meminfo(info);
1601        si_swapinfo(info);
1602
1603        /*
1604         * If the sum of all the available memory (i.e. ram + swap)
1605         * is less than can be stored in a 32 bit unsigned long then
1606         * we can be binary compatible with 2.2.x kernels.  If not,
1607         * well, in that case 2.2.x was broken anyways...
1608         *
1609         *  -Erik Andersen <andersee@debian.org>
1610         */
1611
1612        mem_total = info->totalram + info->totalswap;
1613        if (mem_total < info->totalram || mem_total < info->totalswap)
1614                goto out;
1615        bitcount = 0;
1616        mem_unit = info->mem_unit;
1617        while (mem_unit > 1) {
1618                bitcount++;
1619                mem_unit >>= 1;
1620                sav_total = mem_total;
1621                mem_total <<= 1;
1622                if (mem_total < sav_total)
1623                        goto out;
1624        }
1625
1626        /*
1627         * If mem_total did not overflow, multiply all memory values by
1628         * info->mem_unit and set it to 1.  This leaves things compatible
1629         * with 2.2.x, and also retains compatibility with earlier 2.4.x
1630         * kernels...
1631         */
1632
1633        info->mem_unit = 1;
1634        info->totalram <<= bitcount;
1635        info->freeram <<= bitcount;
1636        info->sharedram <<= bitcount;
1637        info->bufferram <<= bitcount;
1638        info->totalswap <<= bitcount;
1639        info->freeswap <<= bitcount;
1640        info->totalhigh <<= bitcount;
1641        info->freehigh <<= bitcount;
1642
1643out:
1644        return 0;
1645}
1646
1647SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
1648{
1649        struct sysinfo val;
1650
1651        do_sysinfo(&val);
1652
1653        if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1654                return -EFAULT;
1655
1656        return 0;
1657}
1658
1659static int __cpuinit init_timers_cpu(int cpu)
1660{
1661        int j;
1662        struct tvec_base *base;
1663        static char __cpuinitdata tvec_base_done[NR_CPUS];
1664
1665        if (!tvec_base_done[cpu]) {
1666                static char boot_done;
1667
1668                if (boot_done) {
1669                        /*
1670                         * The APs use this path later in boot
1671                         */
1672                        base = kmalloc_node(sizeof(*base),
1673                                                GFP_KERNEL | __GFP_ZERO,
1674                                                cpu_to_node(cpu));
1675                        if (!base)
1676                                return -ENOMEM;
1677
1678                        /* Make sure that tvec_base is 2 byte aligned */
1679                        if (tbase_get_deferrable(base)) {
1680                                WARN_ON(1);
1681                                kfree(base);
1682                                return -ENOMEM;
1683                        }
1684                        per_cpu(tvec_bases, cpu) = base;
1685                } else {
1686                        /*
1687                         * This is for the boot CPU - we use compile-time
1688                         * static initialisation because per-cpu memory isn't
1689                         * ready yet and because the memory allocators are not
1690                         * initialised either.
1691                         */
1692                        boot_done = 1;
1693                        base = &boot_tvec_bases;
1694                }
1695                tvec_base_done[cpu] = 1;
1696        } else {
1697                base = per_cpu(tvec_bases, cpu);
1698        }
1699
1700        spin_lock_init(&base->lock);
1701
1702        for (j = 0; j < TVN_SIZE; j++) {
1703                INIT_LIST_HEAD(base->tv5.vec + j);
1704                INIT_LIST_HEAD(base->tv4.vec + j);
1705                INIT_LIST_HEAD(base->tv3.vec + j);
1706                INIT_LIST_HEAD(base->tv2.vec + j);
1707        }
1708        for (j = 0; j < TVR_SIZE; j++)
1709                INIT_LIST_HEAD(base->tv1.vec + j);
1710
1711        base->timer_jiffies = jiffies;
1712        base->next_timer = base->timer_jiffies;
1713        base->active_timers = 0;
1714        return 0;
1715}
1716
1717#ifdef CONFIG_HOTPLUG_CPU
1718static void migrate_timer_list(struct tvec_base *new_base, struct list_head *head)
1719{
1720        struct timer_list *timer;
1721
1722        while (!list_empty(head)) {
1723                timer = list_first_entry(head, struct timer_list, entry);
1724                /* We ignore the accounting on the dying cpu */
1725                detach_timer(timer, false);
1726                timer_set_base(timer, new_base);
1727                internal_add_timer(new_base, timer);
1728        }
1729}
1730
1731static void __cpuinit migrate_timers(int cpu)
1732{
1733        struct tvec_base *old_base;
1734        struct tvec_base *new_base;
1735        int i;
1736
1737        BUG_ON(cpu_online(cpu));
1738        old_base = per_cpu(tvec_bases, cpu);
1739        new_base = get_cpu_var(tvec_bases);
1740        /*
1741         * The caller is globally serialized and nobody else
1742         * takes two locks at once, deadlock is not possible.
1743         */
1744        spin_lock_irq(&new_base->lock);
1745        spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
1746
1747        BUG_ON(old_base->running_timer);
1748
1749        for (i = 0; i < TVR_SIZE; i++)
1750                migrate_timer_list(new_base, old_base->tv1.vec + i);
1751        for (i = 0; i < TVN_SIZE; i++) {
1752                migrate_timer_list(new_base, old_base->tv2.vec + i);
1753                migrate_timer_list(new_base, old_base->tv3.vec + i);
1754                migrate_timer_list(new_base, old_base->tv4.vec + i);
1755                migrate_timer_list(new_base, old_base->tv5.vec + i);
1756        }
1757
1758        spin_unlock(&old_base->lock);
1759        spin_unlock_irq(&new_base->lock);
1760        put_cpu_var(tvec_bases);
1761}
1762#endif /* CONFIG_HOTPLUG_CPU */
1763
1764static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1765                                unsigned long action, void *hcpu)
1766{
1767        long cpu = (long)hcpu;
1768        int err;
1769
1770        switch(action) {
1771        case CPU_UP_PREPARE:
1772        case CPU_UP_PREPARE_FROZEN:
1773                err = init_timers_cpu(cpu);
1774                if (err < 0)
1775                        return notifier_from_errno(err);
1776                break;
1777#ifdef CONFIG_HOTPLUG_CPU
1778        case CPU_DEAD:
1779        case CPU_DEAD_FROZEN:
1780                migrate_timers(cpu);
1781                break;
1782#endif
1783        default:
1784                break;
1785        }
1786        return NOTIFY_OK;
1787}
1788
1789static struct notifier_block __cpuinitdata timers_nb = {
1790        .notifier_call  = timer_cpu_notify,
1791};
1792
1793
1794void __init init_timers(void)
1795{
1796        int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1797                                (void *)(long)smp_processor_id());
1798
1799        init_timer_stats();
1800
1801        BUG_ON(err != NOTIFY_OK);
1802        register_cpu_notifier(&timers_nb);
1803        open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
1804}
1805
1806/**
1807 * msleep - sleep safely even with waitqueue interruptions
1808 * @msecs: Time in milliseconds to sleep for
1809 */
1810void msleep(unsigned int msecs)
1811{
1812        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1813
1814        while (timeout)
1815                timeout = schedule_timeout_uninterruptible(timeout);
1816}
1817
1818EXPORT_SYMBOL(msleep);
1819
1820/**
1821 * msleep_interruptible - sleep waiting for signals
1822 * @msecs: Time in milliseconds to sleep for
1823 */
1824unsigned long msleep_interruptible(unsigned int msecs)
1825{
1826        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1827
1828        while (timeout && !signal_pending(current))
1829                timeout = schedule_timeout_interruptible(timeout);
1830        return jiffies_to_msecs(timeout);
1831}
1832
1833EXPORT_SYMBOL(msleep_interruptible);
1834
1835static int __sched do_usleep_range(unsigned long min, unsigned long max)
1836{
1837        ktime_t kmin;
1838        unsigned long delta;
1839
1840        kmin = ktime_set(0, min * NSEC_PER_USEC);
1841        delta = (max - min) * NSEC_PER_USEC;
1842        return schedule_hrtimeout_range(&kmin, delta, HRTIMER_MODE_REL);
1843}
1844
1845/**
1846 * usleep_range - Drop in replacement for udelay where wakeup is flexible
1847 * @min: Minimum time in usecs to sleep
1848 * @max: Maximum time in usecs to sleep
1849 */
1850void usleep_range(unsigned long min, unsigned long max)
1851{
1852        __set_current_state(TASK_UNINTERRUPTIBLE);
1853        do_usleep_range(min, max);
1854}
1855EXPORT_SYMBOL(usleep_range);
1856
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.