kernel/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/posix-timers.h>
   7 #include <linux/errno.h>
   8 #include <linux/math64.h>
   9 #include <asm/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11 #include <trace/events/timer.h>
  12 #include <linux/random.h>
  13 #include <linux/tick.h>
  14 #include <linux/workqueue.h>
  15
  16 /*
  17  * Called after updating RLIMIT_CPU to run cpu timer and update
  18  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  19  * siglock protection since other code may update expiration cache as
  20  * well.
  21  */
  22 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  23 {
  24         cputime_t cputime = secs_to_cputime(rlim_new);
  25
  26         spin_lock_irq(&task->sighand->siglock);
  27         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  28         spin_unlock_irq(&task->sighand->siglock);
  29 }
  30
  31 static int check_clock(const clockid_t which_clock)
  32 {
  33         int error = 0;
  34         struct task_struct *p;
  35         const pid_t pid = CPUCLOCK_PID(which_clock);
  36
  37         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  38                 return -EINVAL;
  39
  40         if (pid == 0)
  41                 return 0;
  42
  43         rcu_read_lock();
  44         p = find_task_by_vpid(pid);
  45         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  46                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  47                 error = -EINVAL;
  48         }
  49         rcu_read_unlock();
  50
  51         return error;
  52 }
  53
  54 static inline unsigned long long
  55 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  56 {
  57         unsigned long long ret;
  58
  59         ret = 0;                /* high half always zero when .cpu used */
  60         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  61                 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  62         } else {
  63                 ret = cputime_to_expires(timespec_to_cputime(tp));
  64         }
  65         return ret;
  66 }
  67
  68 static void sample_to_timespec(const clockid_t which_clock,
  69                                unsigned long long expires,
  70                                struct timespec *tp)
  71 {
  72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  73                 *tp = ns_to_timespec(expires);
  74         else
  75                 cputime_to_timespec((__force cputime_t)expires, tp);
  76 }
  77
  78 /*
  79  * Update expiry time from increment, and increase overrun count,
  80  * given the current clock sample.
  81  */
  82 static void bump_cpu_timer(struct k_itimer *timer,
  83                            unsigned long long now)
  84 {
  85         int i;
  86         unsigned long long delta, incr;
  87
  88         if (timer->it.cpu.incr == 0)
  89                 return;
  90
  91         if (now < timer->it.cpu.expires)
  92                 return;
  93
  94         incr = timer->it.cpu.incr;
  95         delta = now + incr - timer->it.cpu.expires;
  96
  97         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  98         for (i = 0; incr < delta - incr; i++)
  99                 incr = incr << 1;
 100
 101         for (; i >= 0; incr >>= 1, i--) {
 102                 if (delta < incr)
 103                         continue;
 104
 105                 timer->it.cpu.expires += incr;
 106                 timer->it_overrun += 1 << i;
 107                 delta -= incr;
 108         }
 109 }
 110
 111 /**
 112  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 113  *
 114  * @cputime:    The struct to compare.
 115  *
 116  * Checks @cputime to see if all fields are zero.  Returns true if all fields
 117  * are zero, false if any field is nonzero.
 118  */
 119 static inline int task_cputime_zero(const struct task_cputime *cputime)
 120 {
 121         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 static inline unsigned long long prof_ticks(struct task_struct *p)
 127 {
 128         cputime_t utime, stime;
 129
 130         task_cputime(p, &utime, &stime);
 131
 132         return cputime_to_expires(utime + stime);
 133 }
 134 static inline unsigned long long virt_ticks(struct task_struct *p)
 135 {
 136         cputime_t utime;
 137
 138         task_cputime(p, &utime, NULL);
 139
 140         return cputime_to_expires(utime);
 141 }
 142
 143 static int
 144 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 145 {
 146         int error = check_clock(which_clock);
 147         if (!error) {
 148                 tp->tv_sec = 0;
 149                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 150                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 151                         /*
 152                          * If sched_clock is using a cycle counter, we
 153                          * don't have any idea of its true resolution
 154                          * exported, but it is much more than 1s/HZ.
 155                          */
 156                         tp->tv_nsec = 1;
 157                 }
 158         }
 159         return error;
 160 }
 161
 162 static int
 163 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 164 {
 165         /*
 166          * You can never reset a CPU clock, but we check for other errors
 167          * in the call before failing with EPERM.
 168          */
 169         int error = check_clock(which_clock);
 170         if (error == 0) {
 171                 error = -EPERM;
 172         }
 173         return error;
 174 }
 175
 176
 177 /*
 178  * Sample a per-thread clock for the given task.
 179  */
 180 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 181                             unsigned long long *sample)
 182 {
 183         switch (CPUCLOCK_WHICH(which_clock)) {
 184         default:
 185                 return -EINVAL;
 186         case CPUCLOCK_PROF:
 187                 *sample = prof_ticks(p);
 188                 break;
 189         case CPUCLOCK_VIRT:
 190                 *sample = virt_ticks(p);
 191                 break;
 192         case CPUCLOCK_SCHED:
 193                 *sample = task_sched_runtime(p);
 194                 break;
 195         }
 196         return 0;
 197 }
 198
 199 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 200 {
 201         if (b->utime > a->utime)
 202                 a->utime = b->utime;
 203
 204         if (b->stime > a->stime)
 205                 a->stime = b->stime;
 206
 207         if (b->sum_exec_runtime > a->sum_exec_runtime)
 208                 a->sum_exec_runtime = b->sum_exec_runtime;
 209 }
 210
 211 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 212 {
 213         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 214         struct task_cputime sum;
 215         unsigned long flags;
 216
 217         if (!cputimer->running) {
 218                 /*
 219                  * The POSIX timer interface allows for absolute time expiry
 220                  * values through the TIMER_ABSTIME flag, therefore we have
 221                  * to synchronize the timer to the clock every time we start
 222                  * it.
 223                  */
 224                 thread_group_cputime(tsk, &sum);
 225                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 226                 cputimer->running = 1;
 227                 update_gt_cputime(&cputimer->cputime, &sum);
 228         } else
 229                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 230         *times = cputimer->cputime;
 231         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 232 }
 233
 234 /*
 235  * Sample a process (thread group) clock for the given group_leader task.
 236  * Must be called with task sighand lock held for safe while_each_thread()
 237  * traversal.
 238  */
 239 static int cpu_clock_sample_group(const clockid_t which_clock,
 240                                   struct task_struct *p,
 241                                   unsigned long long *sample)
 242 {
 243         struct task_cputime cputime;
 244
 245         switch (CPUCLOCK_WHICH(which_clock)) {
 246         default:
 247                 return -EINVAL;
 248         case CPUCLOCK_PROF:
 249                 thread_group_cputime(p, &cputime);
 250                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 251                 break;
 252         case CPUCLOCK_VIRT:
 253                 thread_group_cputime(p, &cputime);
 254                 *sample = cputime_to_expires(cputime.utime);
 255                 break;
 256         case CPUCLOCK_SCHED:
 257                 thread_group_cputime(p, &cputime);
 258                 *sample = cputime.sum_exec_runtime;
 259                 break;
 260         }
 261         return 0;
 262 }
 263
 264 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 265                                     const clockid_t which_clock,
 266                                     struct timespec *tp)
 267 {
 268         int err = -EINVAL;
 269         unsigned long long rtn;
 270
 271         if (CPUCLOCK_PERTHREAD(which_clock)) {
 272                 if (same_thread_group(tsk, current))
 273                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 274         } else {
 275                 unsigned long flags;
 276                 struct sighand_struct *sighand;
 277
 278                 /*
 279                  * while_each_thread() is not yet entirely RCU safe,
 280                  * keep locking the group while sampling process
 281                  * clock for now.
 282                  */
 283                 sighand = lock_task_sighand(tsk, &flags);
 284                 if (!sighand)
 285                         return err;
 286
 287                 if (tsk == current || thread_group_leader(tsk))
 288                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 289
 290                 unlock_task_sighand(tsk, &flags);
 291         }
 292
 293         if (!err)
 294                 sample_to_timespec(which_clock, rtn, tp);
 295
 296         return err;
 297 }
 298
 299
 300 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 301 {
 302         const pid_t pid = CPUCLOCK_PID(which_clock);
 303         int err = -EINVAL;
 304
 305         if (pid == 0) {
 306                 /*
 307                  * Special case constant value for our own clocks.
 308                  * We don't have to do any lookup to find ourselves.
 309                  */
 310                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 311         } else {
 312                 /*
 313                  * Find the given PID, and validate that the caller
 314                  * should be able to see it.
 315                  */
 316                 struct task_struct *p;
 317                 rcu_read_lock();
 318                 p = find_task_by_vpid(pid);
 319                 if (p)
 320                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 321                 rcu_read_unlock();
 322         }
 323
 324         return err;
 325 }
 326
 327
 328 /*
 329  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 330  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 331  * new timer already all-zeros initialized.
 332  */
 333 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 334 {
 335         int ret = 0;
 336         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 337         struct task_struct *p;
 338
 339         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 340                 return -EINVAL;
 341
 342         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 343
 344         rcu_read_lock();
 345         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 346                 if (pid == 0) {
 347                         p = current;
 348                 } else {
 349                         p = find_task_by_vpid(pid);
 350                         if (p && !same_thread_group(p, current))
 351                                 p = NULL;
 352                 }
 353         } else {
 354                 if (pid == 0) {
 355                         p = current->group_leader;
 356                 } else {
 357                         p = find_task_by_vpid(pid);
 358                         if (p && !has_group_leader_pid(p))
 359                                 p = NULL;
 360                 }
 361         }
 362         new_timer->it.cpu.task = p;
 363         if (p) {
 364                 get_task_struct(p);
 365         } else {
 366                 ret = -EINVAL;
 367         }
 368         rcu_read_unlock();
 369
 370         return ret;
 371 }
 372
 373 /*
 374  * Clean up a CPU-clock timer that is about to be destroyed.
 375  * This is called from timer deletion with the timer already locked.
 376  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 377  * and try again.  (This happens when the timer is in the middle of firing.)
 378  */
 379 static int posix_cpu_timer_del(struct k_itimer *timer)
 380 {
 381         int ret = 0;
 382         unsigned long flags;
 383         struct sighand_struct *sighand;
 384         struct task_struct *p = timer->it.cpu.task;
 385
 386         WARN_ON_ONCE(p == NULL);
 387
 388         /*
 389          * Protect against sighand release/switch in exit/exec and process/
 390          * thread timer list entry concurrent read/writes.
 391          */
 392         sighand = lock_task_sighand(p, &flags);
 393         if (unlikely(sighand == NULL)) {
 394                 /*
 395                  * We raced with the reaping of the task.
 396                  * The deletion should have cleared us off the list.
 397                  */
 398                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 399         } else {
 400                 if (timer->it.cpu.firing)
 401                         ret = TIMER_RETRY;
 402                 else
 403                         list_del(&timer->it.cpu.entry);
 404
 405                 unlock_task_sighand(p, &flags);
 406         }
 407
 408         if (!ret)
 409                 put_task_struct(p);
 410
 411         return ret;
 412 }
 413
 414 static void cleanup_timers_list(struct list_head *head)
 415 {
 416         struct cpu_timer_list *timer, *next;
 417
 418         list_for_each_entry_safe(timer, next, head, entry)
 419                 list_del_init(&timer->entry);
 420 }
 421
 422 /*
 423  * Clean out CPU timers still ticking when a thread exited.  The task
 424  * pointer is cleared, and the expiry time is replaced with the residual
 425  * time for later timer_gettime calls to return.
 426  * This must be called with the siglock held.
 427  */
 428 static void cleanup_timers(struct list_head *head)
 429 {
 430         cleanup_timers_list(head);
 431         cleanup_timers_list(++head);
 432         cleanup_timers_list(++head);
 433 }
 434
 435 /*
 436  * These are both called with the siglock held, when the current thread
 437  * is being reaped.  When the final (leader) thread in the group is reaped,
 438  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 439  */
 440 void posix_cpu_timers_exit(struct task_struct *tsk)
 441 {
 442         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 443                                                 sizeof(unsigned long long));
 444         cleanup_timers(tsk->cpu_timers);
 445
 446 }
 447 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 448 {
 449         cleanup_timers(tsk->signal->cpu_timers);
 450 }
 451
 452 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 453 {
 454         return expires == 0 || expires > new_exp;
 455 }
 456
 457 /*
 458  * Insert the timer on the appropriate list before any timers that
 459  * expire later.  This must be called with the sighand lock held.
 460  */
 461 static void arm_timer(struct k_itimer *timer)
 462 {
 463         struct task_struct *p = timer->it.cpu.task;
 464         struct list_head *head, *listpos;
 465         struct task_cputime *cputime_expires;
 466         struct cpu_timer_list *const nt = &timer->it.cpu;
 467         struct cpu_timer_list *next;
 468
 469         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 470                 head = p->cpu_timers;
 471                 cputime_expires = &p->cputime_expires;
 472         } else {
 473                 head = p->signal->cpu_timers;
 474                 cputime_expires = &p->signal->cputime_expires;
 475         }
 476         head += CPUCLOCK_WHICH(timer->it_clock);
 477
 478         listpos = head;
 479         list_for_each_entry(next, head, entry) {
 480                 if (nt->expires < next->expires)
 481                         break;
 482                 listpos = &next->entry;
 483         }
 484         list_add(&nt->entry, listpos);
 485
 486         if (listpos == head) {
 487                 unsigned long long exp = nt->expires;
 488
 489                 /*
 490                  * We are the new earliest-expiring POSIX 1.b timer, hence
 491                  * need to update expiration cache. Take into account that
 492                  * for process timers we share expiration cache with itimers
 493                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 494                  */
 495
 496                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 497                 case CPUCLOCK_PROF:
 498                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 499                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 500                         break;
 501                 case CPUCLOCK_VIRT:
 502                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 503                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 504                         break;
 505                 case CPUCLOCK_SCHED:
 506                         if (cputime_expires->sched_exp == 0 ||
 507                             cputime_expires->sched_exp > exp)
 508                                 cputime_expires->sched_exp = exp;
 509                         break;
 510                 }
 511         }
 512 }
 513
 514 /*
 515  * The timer is locked, fire it and arrange for its reload.
 516  */
 517 static void cpu_timer_fire(struct k_itimer *timer)
 518 {
 519         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 520                 /*
 521                  * User don't want any signal.
 522                  */
 523                 timer->it.cpu.expires = 0;
 524         } else if (unlikely(timer->sigq == NULL)) {
 525                 /*
 526                  * This a special case for clock_nanosleep,
 527                  * not a normal timer from sys_timer_create.
 528                  */
 529                 wake_up_process(timer->it_process);
 530                 timer->it.cpu.expires = 0;
 531         } else if (timer->it.cpu.incr == 0) {
 532                 /*
 533                  * One-shot timer.  Clear it as soon as it's fired.
 534                  */
 535                 posix_timer_event(timer, 0);
 536                 timer->it.cpu.expires = 0;
 537         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 538                 /*
 539                  * The signal did not get queued because the signal
 540                  * was ignored, so we won't get any callback to
 541                  * reload the timer.  But we need to keep it
 542                  * ticking in case the signal is deliverable next time.
 543                  */
 544                 posix_cpu_timer_schedule(timer);
 545         }
 546 }
 547
 548 /*
 549  * Sample a process (thread group) timer for the given group_leader task.
 550  * Must be called with task sighand lock held for safe while_each_thread()
 551  * traversal.
 552  */
 553 static int cpu_timer_sample_group(const clockid_t which_clock,
 554                                   struct task_struct *p,
 555                                   unsigned long long *sample)
 556 {
 557         struct task_cputime cputime;
 558
 559         thread_group_cputimer(p, &cputime);
 560         switch (CPUCLOCK_WHICH(which_clock)) {
 561         default:
 562                 return -EINVAL;
 563         case CPUCLOCK_PROF:
 564                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 565                 break;
 566         case CPUCLOCK_VIRT:
 567                 *sample = cputime_to_expires(cputime.utime);
 568                 break;
 569         case CPUCLOCK_SCHED:
 570                 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
 571                 break;
 572         }
 573         return 0;
 574 }
 575
 576 #ifdef CONFIG_NO_HZ_FULL
 577 static void nohz_kick_work_fn(struct work_struct *work)
 578 {
 579         tick_nohz_full_kick_all();
 580 }
 581
 582 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
 583
 584 /*
 585  * We need the IPIs to be sent from sane process context.
 586  * The posix cpu timers are always set with irqs disabled.
 587  */
 588 static void posix_cpu_timer_kick_nohz(void)
 589 {
 590         if (context_tracking_is_enabled())
 591                 schedule_work(&nohz_kick_work);
 592 }
 593
 594 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
 595 {
 596         if (!task_cputime_zero(&tsk->cputime_expires))
 597                 return false;
 598
 599         if (tsk->signal->cputimer.running)
 600                 return false;
 601
 602         return true;
 603 }
 604 #else
 605 static inline void posix_cpu_timer_kick_nohz(void) { }
 606 #endif
 607
 608 /*
 609  * Guts of sys_timer_settime for CPU timers.
 610  * This is called with the timer locked and interrupts disabled.
 611  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 612  * and try again.  (This happens when the timer is in the middle of firing.)
 613  */
 614 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 615                                struct itimerspec *new, struct itimerspec *old)
 616 {
 617         unsigned long flags;
 618         struct sighand_struct *sighand;
 619         struct task_struct *p = timer->it.cpu.task;
 620         unsigned long long old_expires, new_expires, old_incr, val;
 621         int ret;
 622
 623         WARN_ON_ONCE(p == NULL);
 624
 625         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 626
 627         /*
 628          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 629          * and p->signal->cpu_timers read/write in arm_timer()
 630          */
 631         sighand = lock_task_sighand(p, &flags);
 632         /*
 633          * If p has just been reaped, we can no
 634          * longer get any information about it at all.
 635          */
 636         if (unlikely(sighand == NULL)) {
 637                 return -ESRCH;
 638         }
 639
 640         /*
 641          * Disarm any old timer after extracting its expiry time.
 642          */
 643         WARN_ON_ONCE(!irqs_disabled());
 644
 645         ret = 0;
 646         old_incr = timer->it.cpu.incr;
 647         old_expires = timer->it.cpu.expires;
 648         if (unlikely(timer->it.cpu.firing)) {
 649                 timer->it.cpu.firing = -1;
 650                 ret = TIMER_RETRY;
 651         } else
 652                 list_del_init(&timer->it.cpu.entry);
 653
 654         /*
 655          * We need to sample the current value to convert the new
 656          * value from to relative and absolute, and to convert the
 657          * old value from absolute to relative.  To set a process
 658          * timer, we need a sample to balance the thread expiry
 659          * times (in arm_timer).  With an absolute time, we must
 660          * check if it's already passed.  In short, we need a sample.
 661          */
 662         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 663                 cpu_clock_sample(timer->it_clock, p, &val);
 664         } else {
 665                 cpu_timer_sample_group(timer->it_clock, p, &val);
 666         }
 667
 668         if (old) {
 669                 if (old_expires == 0) {
 670                         old->it_value.tv_sec = 0;
 671                         old->it_value.tv_nsec = 0;
 672                 } else {
 673                         /*
 674                          * Update the timer in case it has
 675                          * overrun already.  If it has,
 676                          * we'll report it as having overrun
 677                          * and with the next reloaded timer
 678                          * already ticking, though we are
 679                          * swallowing that pending
 680                          * notification here to install the
 681                          * new setting.
 682                          */
 683                         bump_cpu_timer(timer, val);
 684                         if (val < timer->it.cpu.expires) {
 685                                 old_expires = timer->it.cpu.expires - val;
 686                                 sample_to_timespec(timer->it_clock,
 687                                                    old_expires,
 688                                                    &old->it_value);
 689                         } else {
 690                                 old->it_value.tv_nsec = 1;
 691                                 old->it_value.tv_sec = 0;
 692                         }
 693                 }
 694         }
 695
 696         if (unlikely(ret)) {
 697                 /*
 698                  * We are colliding with the timer actually firing.
 699                  * Punt after filling in the timer's old value, and
 700                  * disable this firing since we are already reporting
 701                  * it as an overrun (thanks to bump_cpu_timer above).
 702                  */
 703                 unlock_task_sighand(p, &flags);
 704                 goto out;
 705         }
 706
 707         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 708                 new_expires += val;
 709         }
 710
 711         /*
 712          * Install the new expiry time (or zero).
 713          * For a timer with no notification action, we don't actually
 714          * arm the timer (we'll just fake it for timer_gettime).
 715          */
 716         timer->it.cpu.expires = new_expires;
 717         if (new_expires != 0 && val < new_expires) {
 718                 arm_timer(timer);
 719         }
 720
 721         unlock_task_sighand(p, &flags);
 722         /*
 723          * Install the new reload setting, and
 724          * set up the signal and overrun bookkeeping.
 725          */
 726         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 727                                                 &new->it_interval);
 728
 729         /*
 730          * This acts as a modification timestamp for the timer,
 731          * so any automatic reload attempt will punt on seeing
 732          * that we have reset the timer manually.
 733          */
 734         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 735                 ~REQUEUE_PENDING;
 736         timer->it_overrun_last = 0;
 737         timer->it_overrun = -1;
 738
 739         if (new_expires != 0 && !(val < new_expires)) {
 740                 /*
 741                  * The designated time already passed, so we notify
 742                  * immediately, even if the thread never runs to
 743                  * accumulate more time on this clock.
 744                  */
 745                 cpu_timer_fire(timer);
 746         }
 747
 748         ret = 0;
 749  out:
 750         if (old) {
 751                 sample_to_timespec(timer->it_clock,
 752                                    old_incr, &old->it_interval);
 753         }
 754         if (!ret)
 755                 posix_cpu_timer_kick_nohz();
 756         return ret;
 757 }
 758
 759 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 760 {
 761         unsigned long long now;
 762         struct task_struct *p = timer->it.cpu.task;
 763
 764         WARN_ON_ONCE(p == NULL);
 765
 766         /*
 767          * Easy part: convert the reload time.
 768          */
 769         sample_to_timespec(timer->it_clock,
 770                            timer->it.cpu.incr, &itp->it_interval);
 771
 772         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 773                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 774                 return;
 775         }
 776
 777         /*
 778          * Sample the clock to take the difference with the expiry time.
 779          */
 780         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 781                 cpu_clock_sample(timer->it_clock, p, &now);
 782         } else {
 783                 struct sighand_struct *sighand;
 784                 unsigned long flags;
 785
 786                 /*
 787                  * Protect against sighand release/switch in exit/exec and
 788                  * also make timer sampling safe if it ends up calling
 789                  * thread_group_cputime().
 790                  */
 791                 sighand = lock_task_sighand(p, &flags);
 792                 if (unlikely(sighand == NULL)) {
 793                         /*
 794                          * The process has been reaped.
 795                          * We can't even collect a sample any more.
 796                          * Call the timer disarmed, nothing else to do.
 797                          */
 798                         timer->it.cpu.expires = 0;
 799                         sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 800                                            &itp->it_value);
 801                 } else {
 802                         cpu_timer_sample_group(timer->it_clock, p, &now);
 803                         unlock_task_sighand(p, &flags);
 804                 }
 805         }
 806
 807         if (now < timer->it.cpu.expires) {
 808                 sample_to_timespec(timer->it_clock,
 809                                    timer->it.cpu.expires - now,
 810                                    &itp->it_value);
 811         } else {
 812                 /*
 813                  * The timer should have expired already, but the firing
 814                  * hasn't taken place yet.  Say it's just about to expire.
 815                  */
 816                 itp->it_value.tv_nsec = 1;
 817                 itp->it_value.tv_sec = 0;
 818         }
 819 }
 820
 821 static unsigned long long
 822 check_timers_list(struct list_head *timers,
 823                   struct list_head *firing,
 824                   unsigned long long curr)
 825 {
 826         int maxfire = 20;
 827
 828         while (!list_empty(timers)) {
 829                 struct cpu_timer_list *t;
 830
 831                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 832
 833                 if (!--maxfire || curr < t->expires)
 834                         return t->expires;
 835
 836                 t->firing = 1;
 837                 list_move_tail(&t->entry, firing);
 838         }
 839
 840         return 0;
 841 }
 842
 843 /*
 844  * Check for any per-thread CPU timers that have fired and move them off
 845  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 846  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 847  */
 848 static void check_thread_timers(struct task_struct *tsk,
 849                                 struct list_head *firing)
 850 {
 851         struct list_head *timers = tsk->cpu_timers;
 852         struct signal_struct *const sig = tsk->signal;
 853         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 854         unsigned long long expires;
 855         unsigned long soft;
 856
 857         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 858         tsk_expires->prof_exp = expires_to_cputime(expires);
 859
 860         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 861         tsk_expires->virt_exp = expires_to_cputime(expires);
 862
 863         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 864                                                    tsk->se.sum_exec_runtime);
 865
 866         /*
 867          * Check for the special case thread timers.
 868          */
 869         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 870         if (soft != RLIM_INFINITY) {
 871                 unsigned long hard =
 872                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 873
 874                 if (hard != RLIM_INFINITY &&
 875                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 876                         /*
 877                          * At the hard limit, we just die.
 878                          * No need to calculate anything else now.
 879                          */
 880                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 881                         return;
 882                 }
 883                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 884                         /*
 885                          * At the soft limit, send a SIGXCPU every second.
 886                          */
 887                         if (soft < hard) {
 888                                 soft += USEC_PER_SEC;
 889                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 890                         }
 891                         printk(KERN_INFO
 892                                 "RT Watchdog Timeout: %s[%d]\n",
 893                                 tsk->comm, task_pid_nr(tsk));
 894                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 895                 }
 896         }
 897 }
 898
 899 static void stop_process_timers(struct signal_struct *sig)
 900 {
 901         struct thread_group_cputimer *cputimer = &sig->cputimer;
 902         unsigned long flags;
 903
 904         raw_spin_lock_irqsave(&cputimer->lock, flags);
 905         cputimer->running = 0;
 906         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 907 }
 908
 909 static u32 onecputick;
 910
 911 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 912                              unsigned long long *expires,
 913                              unsigned long long cur_time, int signo)
 914 {
 915         if (!it->expires)
 916                 return;
 917
 918         if (cur_time >= it->expires) {
 919                 if (it->incr) {
 920                         it->expires += it->incr;
 921                         it->error += it->incr_error;
 922                         if (it->error >= onecputick) {
 923                                 it->expires -= cputime_one_jiffy;
 924                                 it->error -= onecputick;
 925                         }
 926                 } else {
 927                         it->expires = 0;
 928                 }
 929
 930                 trace_itimer_expire(signo == SIGPROF ?
 931                                     ITIMER_PROF : ITIMER_VIRTUAL,
 932                                     tsk->signal->leader_pid, cur_time);
 933                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 934         }
 935
 936         if (it->expires && (!*expires || it->expires < *expires)) {
 937                 *expires = it->expires;
 938         }
 939 }
 940
 941 /*
 942  * Check for any per-thread CPU timers that have fired and move them
 943  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 944  * have already been taken off.
 945  */
 946 static void check_process_timers(struct task_struct *tsk,
 947                                  struct list_head *firing)
 948 {
 949         struct signal_struct *const sig = tsk->signal;
 950         unsigned long long utime, ptime, virt_expires, prof_expires;
 951         unsigned long long sum_sched_runtime, sched_expires;
 952         struct list_head *timers = sig->cpu_timers;
 953         struct task_cputime cputime;
 954         unsigned long soft;
 955
 956         /*
 957          * Collect the current process totals.
 958          */
 959         thread_group_cputimer(tsk, &cputime);
 960         utime = cputime_to_expires(cputime.utime);
 961         ptime = utime + cputime_to_expires(cputime.stime);
 962         sum_sched_runtime = cputime.sum_exec_runtime;
 963
 964         prof_expires = check_timers_list(timers, firing, ptime);
 965         virt_expires = check_timers_list(++timers, firing, utime);
 966         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 967
 968         /*
 969          * Check for the special case process timers.
 970          */
 971         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 972                          SIGPROF);
 973         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 974                          SIGVTALRM);
 975         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 976         if (soft != RLIM_INFINITY) {
 977                 unsigned long psecs = cputime_to_secs(ptime);
 978                 unsigned long hard =
 979                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 980                 cputime_t x;
 981                 if (psecs >= hard) {
 982                         /*
 983                          * At the hard limit, we just die.
 984                          * No need to calculate anything else now.
 985                          */
 986                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 987                         return;
 988                 }
 989                 if (psecs >= soft) {
 990                         /*
 991                          * At the soft limit, send a SIGXCPU every second.
 992                          */
 993                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 994                         if (soft < hard) {
 995                                 soft++;
 996                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 997                         }
 998                 }
 999                 x = secs_to_cputime(soft);
1000                 if (!prof_expires || x < prof_expires) {
1001                         prof_expires = x;
1002                 }
1003         }
1004
1005         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1006         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1007         sig->cputime_expires.sched_exp = sched_expires;
1008         if (task_cputime_zero(&sig->cputime_expires))
1009                 stop_process_timers(sig);
1010 }
1011
1012 /*
1013  * This is called from the signal code (via do_schedule_next_timer)
1014  * when the last timer signal was delivered and we have to reload the timer.
1015  */
1016 void posix_cpu_timer_schedule(struct k_itimer *timer)
1017 {
1018         struct sighand_struct *sighand;
1019         unsigned long flags;
1020         struct task_struct *p = timer->it.cpu.task;
1021         unsigned long long now;
1022
1023         WARN_ON_ONCE(p == NULL);
1024
1025         /*
1026          * Fetch the current sample and update the timer's expiry time.
1027          */
1028         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1029                 cpu_clock_sample(timer->it_clock, p, &now);
1030                 bump_cpu_timer(timer, now);
1031                 if (unlikely(p->exit_state))
1032                         goto out;
1033
1034                 /* Protect timer list r/w in arm_timer() */
1035                 sighand = lock_task_sighand(p, &flags);
1036                 if (!sighand)
1037                         goto out;
1038         } else {
1039                 /*
1040                  * Protect arm_timer() and timer sampling in case of call to
1041                  * thread_group_cputime().
1042                  */
1043                 sighand = lock_task_sighand(p, &flags);
1044                 if (unlikely(sighand == NULL)) {
1045                         /*
1046                          * The process has been reaped.
1047                          * We can't even collect a sample any more.
1048                          */
1049                         timer->it.cpu.expires = 0;
1050                         goto out;
1051                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1052                         unlock_task_sighand(p, &flags);
1053                         /* Optimizations: if the process is dying, no need to rearm */
1054                         goto out;
1055                 }
1056                 cpu_timer_sample_group(timer->it_clock, p, &now);
1057                 bump_cpu_timer(timer, now);
1058                 /* Leave the sighand locked for the call below.  */
1059         }
1060
1061         /*
1062          * Now re-arm for the new expiry time.
1063          */
1064         WARN_ON_ONCE(!irqs_disabled());
1065         arm_timer(timer);
1066         unlock_task_sighand(p, &flags);
1067
1068         /* Kick full dynticks CPUs in case they need to tick on the new timer */
1069         posix_cpu_timer_kick_nohz();
1070 out:
1071         timer->it_overrun_last = timer->it_overrun;
1072         timer->it_overrun = -1;
1073         ++timer->it_requeue_pending;
1074 }
1075
1076 /**
1077  * task_cputime_expired - Compare two task_cputime entities.
1078  *
1079  * @sample:     The task_cputime structure to be checked for expiration.
1080  * @expires:    Expiration times, against which @sample will be checked.
1081  *
1082  * Checks @sample against @expires to see if any field of @sample has expired.
1083  * Returns true if any field of the former is greater than the corresponding
1084  * field of the latter if the latter field is set.  Otherwise returns false.
1085  */
1086 static inline int task_cputime_expired(const struct task_cputime *sample,
1087                                         const struct task_cputime *expires)
1088 {
1089         if (expires->utime && sample->utime >= expires->utime)
1090                 return 1;
1091         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1092                 return 1;
1093         if (expires->sum_exec_runtime != 0 &&
1094             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1095                 return 1;
1096         return 0;
1097 }
1098
1099 /**
1100  * fastpath_timer_check - POSIX CPU timers fast path.
1101  *
1102  * @tsk:        The task (thread) being checked.
1103  *
1104  * Check the task and thread group timers.  If both are zero (there are no
1105  * timers set) return false.  Otherwise snapshot the task and thread group
1106  * timers and compare them with the corresponding expiration times.  Return
1107  * true if a timer has expired, else return false.
1108  */
1109 static inline int fastpath_timer_check(struct task_struct *tsk)
1110 {
1111         struct signal_struct *sig;
1112         cputime_t utime, stime;
1113
1114         task_cputime(tsk, &utime, &stime);
1115
1116         if (!task_cputime_zero(&tsk->cputime_expires)) {
1117                 struct task_cputime task_sample = {
1118                         .utime = utime,
1119                         .stime = stime,
1120                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1121                 };
1122
1123                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1124                         return 1;
1125         }
1126
1127         sig = tsk->signal;
1128         if (sig->cputimer.running) {
1129                 struct task_cputime group_sample;
1130
1131                 raw_spin_lock(&sig->cputimer.lock);
1132                 group_sample = sig->cputimer.cputime;
1133                 raw_spin_unlock(&sig->cputimer.lock);
1134
1135                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1136                         return 1;
1137         }
1138
1139         return 0;
1140 }
1141
1142 /*
1143  * This is called from the timer interrupt handler.  The irq handler has
1144  * already updated our counts.  We need to check if any timers fire now.
1145  * Interrupts are disabled.
1146  */
1147 void run_posix_cpu_timers(struct task_struct *tsk)
1148 {
1149         LIST_HEAD(firing);
1150         struct k_itimer *timer, *next;
1151         unsigned long flags;
1152
1153         WARN_ON_ONCE(!irqs_disabled());
1154
1155         /*
1156          * The fast path checks that there are no expired thread or thread
1157          * group timers.  If that's so, just return.
1158          */
1159         if (!fastpath_timer_check(tsk))
1160                 return;
1161
1162         if (!lock_task_sighand(tsk, &flags))
1163                 return;
1164         /*
1165          * Here we take off tsk->signal->cpu_timers[N] and
1166          * tsk->cpu_timers[N] all the timers that are firing, and
1167          * put them on the firing list.
1168          */
1169         check_thread_timers(tsk, &firing);
1170         /*
1171          * If there are any active process wide timers (POSIX 1.b, itimers,
1172          * RLIMIT_CPU) cputimer must be running.
1173          */
1174         if (tsk->signal->cputimer.running)
1175                 check_process_timers(tsk, &firing);
1176
1177         /*
1178          * We must release these locks before taking any timer's lock.
1179          * There is a potential race with timer deletion here, as the
1180          * siglock now protects our private firing list.  We have set
1181          * the firing flag in each timer, so that a deletion attempt
1182          * that gets the timer lock before we do will give it up and
1183          * spin until we've taken care of that timer below.
1184          */
1185         unlock_task_sighand(tsk, &flags);
1186
1187         /*
1188          * Now that all the timers on our list have the firing flag,
1189          * no one will touch their list entries but us.  We'll take
1190          * each timer's lock before clearing its firing flag, so no
1191          * timer call will interfere.
1192          */
1193         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1194                 int cpu_firing;
1195
1196                 spin_lock(&timer->it_lock);
1197                 list_del_init(&timer->it.cpu.entry);
1198                 cpu_firing = timer->it.cpu.firing;
1199                 timer->it.cpu.firing = 0;
1200                 /*
1201                  * The firing flag is -1 if we collided with a reset
1202                  * of the timer, which already reported this
1203                  * almost-firing as an overrun.  So don't generate an event.
1204                  */
1205                 if (likely(cpu_firing >= 0))
1206                         cpu_timer_fire(timer);
1207                 spin_unlock(&timer->it_lock);
1208         }
1209 }
1210
1211 /*
1212  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1213  * The tsk->sighand->siglock must be held by the caller.
1214  */
1215 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1216                            cputime_t *newval, cputime_t *oldval)
1217 {
1218         unsigned long long now;
1219
1220         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1221         cpu_timer_sample_group(clock_idx, tsk, &now);
1222
1223         if (oldval) {
1224                 /*
1225                  * We are setting itimer. The *oldval is absolute and we update
1226                  * it to be relative, *newval argument is relative and we update
1227                  * it to be absolute.
1228                  */
1229                 if (*oldval) {
1230                         if (*oldval <= now) {
1231                                 /* Just about to fire. */
1232                                 *oldval = cputime_one_jiffy;
1233                         } else {
1234                                 *oldval -= now;
1235                         }
1236                 }
1237
1238                 if (!*newval)
1239                         goto out;
1240                 *newval += now;
1241         }
1242
1243         /*
1244          * Update expiration cache if we are the earliest timer, or eventually
1245          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1246          */
1247         switch (clock_idx) {
1248         case CPUCLOCK_PROF:
1249                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1250                         tsk->signal->cputime_expires.prof_exp = *newval;
1251                 break;
1252         case CPUCLOCK_VIRT:
1253                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1254                         tsk->signal->cputime_expires.virt_exp = *newval;
1255                 break;
1256         }
1257 out:
1258         posix_cpu_timer_kick_nohz();
1259 }
1260
1261 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1262                             struct timespec *rqtp, struct itimerspec *it)
1263 {
1264         struct k_itimer timer;
1265         int error;
1266
1267         /*
1268          * Set up a temporary timer and then wait for it to go off.
1269          */
1270         memset(&timer, 0, sizeof timer);
1271         spin_lock_init(&timer.it_lock);
1272         timer.it_clock = which_clock;
1273         timer.it_overrun = -1;
1274         error = posix_cpu_timer_create(&timer);
1275         timer.it_process = current;
1276         if (!error) {
1277                 static struct itimerspec zero_it;
1278
1279                 memset(it, 0, sizeof *it);
1280                 it->it_value = *rqtp;
1281
1282                 spin_lock_irq(&timer.it_lock);
1283                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1284                 if (error) {
1285                         spin_unlock_irq(&timer.it_lock);
1286                         return error;
1287                 }
1288
1289                 while (!signal_pending(current)) {
1290                         if (timer.it.cpu.expires == 0) {
1291                                 /*
1292                                  * Our timer fired and was reset, below
1293                                  * deletion can not fail.
1294                                  */
1295                                 posix_cpu_timer_del(&timer);
1296                                 spin_unlock_irq(&timer.it_lock);
1297                                 return 0;
1298                         }
1299
1300                         /*
1301                          * Block until cpu_timer_fire (or a signal) wakes us.
1302                          */
1303                         __set_current_state(TASK_INTERRUPTIBLE);
1304                         spin_unlock_irq(&timer.it_lock);
1305                         schedule();
1306                         spin_lock_irq(&timer.it_lock);
1307                 }
1308
1309                 /*
1310                  * We were interrupted by a signal.
1311                  */
1312                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1313                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1314                 if (!error) {
1315                         /*
1316                          * Timer is now unarmed, deletion can not fail.
1317                          */
1318                         posix_cpu_timer_del(&timer);
1319                 }
1320                 spin_unlock_irq(&timer.it_lock);
1321
1322                 while (error == TIMER_RETRY) {
1323                         /*
1324                          * We need to handle case when timer was or is in the
1325                          * middle of firing. In other cases we already freed
1326                          * resources.
1327                          */
1328                         spin_lock_irq(&timer.it_lock);
1329                         error = posix_cpu_timer_del(&timer);
1330                         spin_unlock_irq(&timer.it_lock);
1331                 }
1332
1333                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1334                         /*
1335                          * It actually did fire already.
1336                          */
1337                         return 0;
1338                 }
1339
1340                 error = -ERESTART_RESTARTBLOCK;
1341         }
1342
1343         return error;
1344 }
1345
1346 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1347
1348 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1349                             struct timespec *rqtp, struct timespec __user *rmtp)
1350 {
1351         struct restart_block *restart_block =
1352                 &current_thread_info()->restart_block;
1353         struct itimerspec it;
1354         int error;
1355
1356         /*
1357          * Diagnose required errors first.
1358          */
1359         if (CPUCLOCK_PERTHREAD(which_clock) &&
1360             (CPUCLOCK_PID(which_clock) == 0 ||
1361              CPUCLOCK_PID(which_clock) == current->pid))
1362                 return -EINVAL;
1363
1364         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1365
1366         if (error == -ERESTART_RESTARTBLOCK) {
1367
1368                 if (flags & TIMER_ABSTIME)
1369                         return -ERESTARTNOHAND;
1370                 /*
1371                  * Report back to the user the time still remaining.
1372                  */
1373                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1374                         return -EFAULT;
1375
1376                 restart_block->fn = posix_cpu_nsleep_restart;
1377                 restart_block->nanosleep.clockid = which_clock;
1378                 restart_block->nanosleep.rmtp = rmtp;
1379                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1380         }
1381         return error;
1382 }
1383
1384 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1385 {
1386         clockid_t which_clock = restart_block->nanosleep.clockid;
1387         struct timespec t;
1388         struct itimerspec it;
1389         int error;
1390
1391         t = ns_to_timespec(restart_block->nanosleep.expires);
1392
1393         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1394
1395         if (error == -ERESTART_RESTARTBLOCK) {
1396                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1397                 /*
1398                  * Report back to the user the time still remaining.
1399                  */
1400                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1401                         return -EFAULT;
1402
1403                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1404         }
1405         return error;
1406
1407 }
1408
1409 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1410 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1411
1412 static int process_cpu_clock_getres(const clockid_t which_clock,
1413                                     struct timespec *tp)
1414 {
1415         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1416 }
1417 static int process_cpu_clock_get(const clockid_t which_clock,
1418                                  struct timespec *tp)
1419 {
1420         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1421 }
1422 static int process_cpu_timer_create(struct k_itimer *timer)
1423 {
1424         timer->it_clock = PROCESS_CLOCK;
1425         return posix_cpu_timer_create(timer);
1426 }
1427 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1428                               struct timespec *rqtp,
1429                               struct timespec __user *rmtp)
1430 {
1431         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1432 }
1433 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1434 {
1435         return -EINVAL;
1436 }
1437 static int thread_cpu_clock_getres(const clockid_t which_clock,
1438                                    struct timespec *tp)
1439 {
1440         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1441 }
1442 static int thread_cpu_clock_get(const clockid_t which_clock,
1443                                 struct timespec *tp)
1444 {
1445         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1446 }
1447 static int thread_cpu_timer_create(struct k_itimer *timer)
1448 {
1449         timer->it_clock = THREAD_CLOCK;
1450         return posix_cpu_timer_create(timer);
1451 }
1452
1453 struct k_clock clock_posix_cpu = {
1454         .clock_getres   = posix_cpu_clock_getres,
1455         .clock_set      = posix_cpu_clock_set,
1456         .clock_get      = posix_cpu_clock_get,
1457         .timer_create   = posix_cpu_timer_create,
1458         .nsleep         = posix_cpu_nsleep,
1459         .nsleep_restart = posix_cpu_nsleep_restart,
1460         .timer_set      = posix_cpu_timer_set,
1461         .timer_del      = posix_cpu_timer_del,
1462         .timer_get      = posix_cpu_timer_get,
1463 };
1464
1465 static __init int init_posix_cpu_timers(void)
1466 {
1467         struct k_clock process = {
1468                 .clock_getres   = process_cpu_clock_getres,
1469                 .clock_get      = process_cpu_clock_get,
1470                 .timer_create   = process_cpu_timer_create,
1471                 .nsleep         = process_cpu_nsleep,
1472                 .nsleep_restart = process_cpu_nsleep_restart,
1473         };
1474         struct k_clock thread = {
1475                 .clock_getres   = thread_cpu_clock_getres,
1476                 .clock_get      = thread_cpu_clock_get,
1477                 .timer_create   = thread_cpu_timer_create,
1478         };
1479         struct timespec ts;
1480
1481         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1482         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1483
1484         cputime_to_timespec(cputime_one_jiffy, &ts);
1485         onecputick = ts.tv_nsec;
1486         WARN_ON(ts.tv_sec != 0);
1487
1488         return 0;
1489 }
1490 __initcall(init_posix_cpu_timers);