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