Pileus Git - ~andy/linux/blob - 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         struct task_struct *p = timer->it.cpu.task;
 381         int ret = 0;
 382
 383         WARN_ON_ONCE(p == NULL);
 384
 385         read_lock(&tasklist_lock);
 386         if (unlikely(p->sighand == NULL)) {
 387                 /*
 388                  * We raced with the reaping of the task.
 389                  * The deletion should have cleared us off the list.
 390                  */
 391                 BUG_ON(!list_empty(&timer->it.cpu.entry));
 392         } else {
 393                 spin_lock(&p->sighand->siglock);
 394                 if (timer->it.cpu.firing)
 395                         ret = TIMER_RETRY;
 396                 else
 397                         list_del(&timer->it.cpu.entry);
 398                 spin_unlock(&p->sighand->siglock);
 399         }
 400         read_unlock(&tasklist_lock);
 401
 402         if (!ret)
 403                 put_task_struct(p);
 404
 405         return ret;
 406 }
 407
 408 static void cleanup_timers_list(struct list_head *head)
 409 {
 410         struct cpu_timer_list *timer, *next;
 411
 412         list_for_each_entry_safe(timer, next, head, entry)
 413                 list_del_init(&timer->entry);
 414 }
 415
 416 /*
 417  * Clean out CPU timers still ticking when a thread exited.  The task
 418  * pointer is cleared, and the expiry time is replaced with the residual
 419  * time for later timer_gettime calls to return.
 420  * This must be called with the siglock held.
 421  */
 422 static void cleanup_timers(struct list_head *head)
 423 {
 424         cleanup_timers_list(head);
 425         cleanup_timers_list(++head);
 426         cleanup_timers_list(++head);
 427 }
 428
 429 /*
 430  * These are both called with the siglock held, when the current thread
 431  * is being reaped.  When the final (leader) thread in the group is reaped,
 432  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 433  */
 434 void posix_cpu_timers_exit(struct task_struct *tsk)
 435 {
 436         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 437                                                 sizeof(unsigned long long));
 438         cleanup_timers(tsk->cpu_timers);
 439
 440 }
 441 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 442 {
 443         cleanup_timers(tsk->signal->cpu_timers);
 444 }
 445
 446 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 447 {
 448         return expires == 0 || expires > new_exp;
 449 }
 450
 451 /*
 452  * Insert the timer on the appropriate list before any timers that
 453  * expire later.  This must be called with the tasklist_lock held
 454  * for reading, interrupts disabled and p->sighand->siglock taken.
 455  */
 456 static void arm_timer(struct k_itimer *timer)
 457 {
 458         struct task_struct *p = timer->it.cpu.task;
 459         struct list_head *head, *listpos;
 460         struct task_cputime *cputime_expires;
 461         struct cpu_timer_list *const nt = &timer->it.cpu;
 462         struct cpu_timer_list *next;
 463
 464         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 465                 head = p->cpu_timers;
 466                 cputime_expires = &p->cputime_expires;
 467         } else {
 468                 head = p->signal->cpu_timers;
 469                 cputime_expires = &p->signal->cputime_expires;
 470         }
 471         head += CPUCLOCK_WHICH(timer->it_clock);
 472
 473         listpos = head;
 474         list_for_each_entry(next, head, entry) {
 475                 if (nt->expires < next->expires)
 476                         break;
 477                 listpos = &next->entry;
 478         }
 479         list_add(&nt->entry, listpos);
 480
 481         if (listpos == head) {
 482                 unsigned long long exp = nt->expires;
 483
 484                 /*
 485                  * We are the new earliest-expiring POSIX 1.b timer, hence
 486                  * need to update expiration cache. Take into account that
 487                  * for process timers we share expiration cache with itimers
 488                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 489                  */
 490
 491                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 492                 case CPUCLOCK_PROF:
 493                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 494                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 495                         break;
 496                 case CPUCLOCK_VIRT:
 497                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 498                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 499                         break;
 500                 case CPUCLOCK_SCHED:
 501                         if (cputime_expires->sched_exp == 0 ||
 502                             cputime_expires->sched_exp > exp)
 503                                 cputime_expires->sched_exp = exp;
 504                         break;
 505                 }
 506         }
 507 }
 508
 509 /*
 510  * The timer is locked, fire it and arrange for its reload.
 511  */
 512 static void cpu_timer_fire(struct k_itimer *timer)
 513 {
 514         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 515                 /*
 516                  * User don't want any signal.
 517                  */
 518                 timer->it.cpu.expires = 0;
 519         } else if (unlikely(timer->sigq == NULL)) {
 520                 /*
 521                  * This a special case for clock_nanosleep,
 522                  * not a normal timer from sys_timer_create.
 523                  */
 524                 wake_up_process(timer->it_process);
 525                 timer->it.cpu.expires = 0;
 526         } else if (timer->it.cpu.incr == 0) {
 527                 /*
 528                  * One-shot timer.  Clear it as soon as it's fired.
 529                  */
 530                 posix_timer_event(timer, 0);
 531                 timer->it.cpu.expires = 0;
 532         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 533                 /*
 534                  * The signal did not get queued because the signal
 535                  * was ignored, so we won't get any callback to
 536                  * reload the timer.  But we need to keep it
 537                  * ticking in case the signal is deliverable next time.
 538                  */
 539                 posix_cpu_timer_schedule(timer);
 540         }
 541 }
 542
 543 /*
 544  * Sample a process (thread group) timer for the given group_leader task.
 545  * Must be called with tasklist_lock held for reading.
 546  */
 547 static int cpu_timer_sample_group(const clockid_t which_clock,
 548                                   struct task_struct *p,
 549                                   unsigned long long *sample)
 550 {
 551         struct task_cputime cputime;
 552
 553         thread_group_cputimer(p, &cputime);
 554         switch (CPUCLOCK_WHICH(which_clock)) {
 555         default:
 556                 return -EINVAL;
 557         case CPUCLOCK_PROF:
 558                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 559                 break;
 560         case CPUCLOCK_VIRT:
 561                 *sample = cputime_to_expires(cputime.utime);
 562                 break;
 563         case CPUCLOCK_SCHED:
 564                 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
 565                 break;
 566         }
 567         return 0;
 568 }
 569
 570 #ifdef CONFIG_NO_HZ_FULL
 571 static void nohz_kick_work_fn(struct work_struct *work)
 572 {
 573         tick_nohz_full_kick_all();
 574 }
 575
 576 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
 577
 578 /*
 579  * We need the IPIs to be sent from sane process context.
 580  * The posix cpu timers are always set with irqs disabled.
 581  */
 582 static void posix_cpu_timer_kick_nohz(void)
 583 {
 584         if (context_tracking_is_enabled())
 585                 schedule_work(&nohz_kick_work);
 586 }
 587
 588 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
 589 {
 590         if (!task_cputime_zero(&tsk->cputime_expires))
 591                 return false;
 592
 593         if (tsk->signal->cputimer.running)
 594                 return false;
 595
 596         return true;
 597 }
 598 #else
 599 static inline void posix_cpu_timer_kick_nohz(void) { }
 600 #endif
 601
 602 /*
 603  * Guts of sys_timer_settime for CPU timers.
 604  * This is called with the timer locked and interrupts disabled.
 605  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 606  * and try again.  (This happens when the timer is in the middle of firing.)
 607  */
 608 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 609                                struct itimerspec *new, struct itimerspec *old)
 610 {
 611         struct task_struct *p = timer->it.cpu.task;
 612         unsigned long long old_expires, new_expires, old_incr, val;
 613         int ret;
 614
 615         WARN_ON_ONCE(p == NULL);
 616
 617         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 618
 619         read_lock(&tasklist_lock);
 620         /*
 621          * We need the tasklist_lock to protect against reaping that
 622          * clears p->sighand.  If p has just been reaped, we can no
 623          * longer get any information about it at all.
 624          */
 625         if (unlikely(p->sighand == NULL)) {
 626                 read_unlock(&tasklist_lock);
 627                 return -ESRCH;
 628         }
 629
 630         /*
 631          * Disarm any old timer after extracting its expiry time.
 632          */
 633         BUG_ON(!irqs_disabled());
 634
 635         ret = 0;
 636         old_incr = timer->it.cpu.incr;
 637         spin_lock(&p->sighand->siglock);
 638         old_expires = timer->it.cpu.expires;
 639         if (unlikely(timer->it.cpu.firing)) {
 640                 timer->it.cpu.firing = -1;
 641                 ret = TIMER_RETRY;
 642         } else
 643                 list_del_init(&timer->it.cpu.entry);
 644
 645         /*
 646          * We need to sample the current value to convert the new
 647          * value from to relative and absolute, and to convert the
 648          * old value from absolute to relative.  To set a process
 649          * timer, we need a sample to balance the thread expiry
 650          * times (in arm_timer).  With an absolute time, we must
 651          * check if it's already passed.  In short, we need a sample.
 652          */
 653         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 654                 cpu_clock_sample(timer->it_clock, p, &val);
 655         } else {
 656                 cpu_timer_sample_group(timer->it_clock, p, &val);
 657         }
 658
 659         if (old) {
 660                 if (old_expires == 0) {
 661                         old->it_value.tv_sec = 0;
 662                         old->it_value.tv_nsec = 0;
 663                 } else {
 664                         /*
 665                          * Update the timer in case it has
 666                          * overrun already.  If it has,
 667                          * we'll report it as having overrun
 668                          * and with the next reloaded timer
 669                          * already ticking, though we are
 670                          * swallowing that pending
 671                          * notification here to install the
 672                          * new setting.
 673                          */
 674                         bump_cpu_timer(timer, val);
 675                         if (val < timer->it.cpu.expires) {
 676                                 old_expires = timer->it.cpu.expires - val;
 677                                 sample_to_timespec(timer->it_clock,
 678                                                    old_expires,
 679                                                    &old->it_value);
 680                         } else {
 681                                 old->it_value.tv_nsec = 1;
 682                                 old->it_value.tv_sec = 0;
 683                         }
 684                 }
 685         }
 686
 687         if (unlikely(ret)) {
 688                 /*
 689                  * We are colliding with the timer actually firing.
 690                  * Punt after filling in the timer's old value, and
 691                  * disable this firing since we are already reporting
 692                  * it as an overrun (thanks to bump_cpu_timer above).
 693                  */
 694                 spin_unlock(&p->sighand->siglock);
 695                 read_unlock(&tasklist_lock);
 696                 goto out;
 697         }
 698
 699         if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
 700                 new_expires += val;
 701         }
 702
 703         /*
 704          * Install the new expiry time (or zero).
 705          * For a timer with no notification action, we don't actually
 706          * arm the timer (we'll just fake it for timer_gettime).
 707          */
 708         timer->it.cpu.expires = new_expires;
 709         if (new_expires != 0 && val < new_expires) {
 710                 arm_timer(timer);
 711         }
 712
 713         spin_unlock(&p->sighand->siglock);
 714         read_unlock(&tasklist_lock);
 715
 716         /*
 717          * Install the new reload setting, and
 718          * set up the signal and overrun bookkeeping.
 719          */
 720         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 721                                                 &new->it_interval);
 722
 723         /*
 724          * This acts as a modification timestamp for the timer,
 725          * so any automatic reload attempt will punt on seeing
 726          * that we have reset the timer manually.
 727          */
 728         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 729                 ~REQUEUE_PENDING;
 730         timer->it_overrun_last = 0;
 731         timer->it_overrun = -1;
 732
 733         if (new_expires != 0 && !(val < new_expires)) {
 734                 /*
 735                  * The designated time already passed, so we notify
 736                  * immediately, even if the thread never runs to
 737                  * accumulate more time on this clock.
 738                  */
 739                 cpu_timer_fire(timer);
 740         }
 741
 742         ret = 0;
 743  out:
 744         if (old) {
 745                 sample_to_timespec(timer->it_clock,
 746                                    old_incr, &old->it_interval);
 747         }
 748         if (!ret)
 749                 posix_cpu_timer_kick_nohz();
 750         return ret;
 751 }
 752
 753 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 754 {
 755         unsigned long long now;
 756         struct task_struct *p = timer->it.cpu.task;
 757
 758         WARN_ON_ONCE(p == NULL);
 759
 760         /*
 761          * Easy part: convert the reload time.
 762          */
 763         sample_to_timespec(timer->it_clock,
 764                            timer->it.cpu.incr, &itp->it_interval);
 765
 766         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 767                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 768                 return;
 769         }
 770
 771         /*
 772          * Sample the clock to take the difference with the expiry time.
 773          */
 774         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 775                 cpu_clock_sample(timer->it_clock, p, &now);
 776         } else {
 777                 read_lock(&tasklist_lock);
 778                 if (unlikely(p->sighand == NULL)) {
 779                         /*
 780                          * The process has been reaped.
 781                          * We can't even collect a sample any more.
 782                          * Call the timer disarmed, nothing else to do.
 783                          */
 784                         timer->it.cpu.expires = 0;
 785                         sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 786                                            &itp->it_value);
 787                         read_unlock(&tasklist_lock);
 788                 } else {
 789                         cpu_timer_sample_group(timer->it_clock, p, &now);
 790                 }
 791                 read_unlock(&tasklist_lock);
 792         }
 793
 794         if (now < timer->it.cpu.expires) {
 795                 sample_to_timespec(timer->it_clock,
 796                                    timer->it.cpu.expires - now,
 797                                    &itp->it_value);
 798         } else {
 799                 /*
 800                  * The timer should have expired already, but the firing
 801                  * hasn't taken place yet.  Say it's just about to expire.
 802                  */
 803                 itp->it_value.tv_nsec = 1;
 804                 itp->it_value.tv_sec = 0;
 805         }
 806 }
 807
 808 static unsigned long long
 809 check_timers_list(struct list_head *timers,
 810                   struct list_head *firing,
 811                   unsigned long long curr)
 812 {
 813         int maxfire = 20;
 814
 815         while (!list_empty(timers)) {
 816                 struct cpu_timer_list *t;
 817
 818                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 819
 820                 if (!--maxfire || curr < t->expires)
 821                         return t->expires;
 822
 823                 t->firing = 1;
 824                 list_move_tail(&t->entry, firing);
 825         }
 826
 827         return 0;
 828 }
 829
 830 /*
 831  * Check for any per-thread CPU timers that have fired and move them off
 832  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 833  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 834  */
 835 static void check_thread_timers(struct task_struct *tsk,
 836                                 struct list_head *firing)
 837 {
 838         struct list_head *timers = tsk->cpu_timers;
 839         struct signal_struct *const sig = tsk->signal;
 840         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 841         unsigned long long expires;
 842         unsigned long soft;
 843
 844         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 845         tsk_expires->prof_exp = expires_to_cputime(expires);
 846
 847         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 848         tsk_expires->virt_exp = expires_to_cputime(expires);
 849
 850         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 851                                                    tsk->se.sum_exec_runtime);
 852
 853         /*
 854          * Check for the special case thread timers.
 855          */
 856         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 857         if (soft != RLIM_INFINITY) {
 858                 unsigned long hard =
 859                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 860
 861                 if (hard != RLIM_INFINITY &&
 862                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 863                         /*
 864                          * At the hard limit, we just die.
 865                          * No need to calculate anything else now.
 866                          */
 867                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 868                         return;
 869                 }
 870                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 871                         /*
 872                          * At the soft limit, send a SIGXCPU every second.
 873                          */
 874                         if (soft < hard) {
 875                                 soft += USEC_PER_SEC;
 876                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 877                         }
 878                         printk(KERN_INFO
 879                                 "RT Watchdog Timeout: %s[%d]\n",
 880                                 tsk->comm, task_pid_nr(tsk));
 881                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 882                 }
 883         }
 884 }
 885
 886 static void stop_process_timers(struct signal_struct *sig)
 887 {
 888         struct thread_group_cputimer *cputimer = &sig->cputimer;
 889         unsigned long flags;
 890
 891         raw_spin_lock_irqsave(&cputimer->lock, flags);
 892         cputimer->running = 0;
 893         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 894 }
 895
 896 static u32 onecputick;
 897
 898 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 899                              unsigned long long *expires,
 900                              unsigned long long cur_time, int signo)
 901 {
 902         if (!it->expires)
 903                 return;
 904
 905         if (cur_time >= it->expires) {
 906                 if (it->incr) {
 907                         it->expires += it->incr;
 908                         it->error += it->incr_error;
 909                         if (it->error >= onecputick) {
 910                                 it->expires -= cputime_one_jiffy;
 911                                 it->error -= onecputick;
 912                         }
 913                 } else {
 914                         it->expires = 0;
 915                 }
 916
 917                 trace_itimer_expire(signo == SIGPROF ?
 918                                     ITIMER_PROF : ITIMER_VIRTUAL,
 919                                     tsk->signal->leader_pid, cur_time);
 920                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 921         }
 922
 923         if (it->expires && (!*expires || it->expires < *expires)) {
 924                 *expires = it->expires;
 925         }
 926 }
 927
 928 /*
 929  * Check for any per-thread CPU timers that have fired and move them
 930  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 931  * have already been taken off.
 932  */
 933 static void check_process_timers(struct task_struct *tsk,
 934                                  struct list_head *firing)
 935 {
 936         struct signal_struct *const sig = tsk->signal;
 937         unsigned long long utime, ptime, virt_expires, prof_expires;
 938         unsigned long long sum_sched_runtime, sched_expires;
 939         struct list_head *timers = sig->cpu_timers;
 940         struct task_cputime cputime;
 941         unsigned long soft;
 942
 943         /*
 944          * Collect the current process totals.
 945          */
 946         thread_group_cputimer(tsk, &cputime);
 947         utime = cputime_to_expires(cputime.utime);
 948         ptime = utime + cputime_to_expires(cputime.stime);
 949         sum_sched_runtime = cputime.sum_exec_runtime;
 950
 951         prof_expires = check_timers_list(timers, firing, ptime);
 952         virt_expires = check_timers_list(++timers, firing, utime);
 953         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 954
 955         /*
 956          * Check for the special case process timers.
 957          */
 958         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 959                          SIGPROF);
 960         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 961                          SIGVTALRM);
 962         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 963         if (soft != RLIM_INFINITY) {
 964                 unsigned long psecs = cputime_to_secs(ptime);
 965                 unsigned long hard =
 966                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 967                 cputime_t x;
 968                 if (psecs >= hard) {
 969                         /*
 970                          * At the hard limit, we just die.
 971                          * No need to calculate anything else now.
 972                          */
 973                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 974                         return;
 975                 }
 976                 if (psecs >= soft) {
 977                         /*
 978                          * At the soft limit, send a SIGXCPU every second.
 979                          */
 980                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 981                         if (soft < hard) {
 982                                 soft++;
 983                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 984                         }
 985                 }
 986                 x = secs_to_cputime(soft);
 987                 if (!prof_expires || x < prof_expires) {
 988                         prof_expires = x;
 989                 }
 990         }
 991
 992         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
 993         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
 994         sig->cputime_expires.sched_exp = sched_expires;
 995         if (task_cputime_zero(&sig->cputime_expires))
 996                 stop_process_timers(sig);
 997 }
 998
 999 /*
1000  * This is called from the signal code (via do_schedule_next_timer)
1001  * when the last timer signal was delivered and we have to reload the timer.
1002  */
1003 void posix_cpu_timer_schedule(struct k_itimer *timer)
1004 {
1005         struct task_struct *p = timer->it.cpu.task;
1006         unsigned long long now;
1007
1008         WARN_ON_ONCE(p == NULL);
1009
1010         /*
1011          * Fetch the current sample and update the timer's expiry time.
1012          */
1013         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1014                 cpu_clock_sample(timer->it_clock, p, &now);
1015                 bump_cpu_timer(timer, now);
1016                 if (unlikely(p->exit_state))
1017                         goto out;
1018
1019                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1020                 spin_lock(&p->sighand->siglock);
1021         } else {
1022                 read_lock(&tasklist_lock);
1023                 if (unlikely(p->sighand == NULL)) {
1024                         /*
1025                          * The process has been reaped.
1026                          * We can't even collect a sample any more.
1027                          */
1028                         timer->it.cpu.expires = 0;
1029                         read_unlock(&tasklist_lock);
1030                         goto out;
1031                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1032                         read_unlock(&tasklist_lock);
1033                         /* Optimizations: if the process is dying, no need to rearm */
1034                         goto out;
1035                 }
1036                 spin_lock(&p->sighand->siglock);
1037                 cpu_timer_sample_group(timer->it_clock, p, &now);
1038                 bump_cpu_timer(timer, now);
1039                 /* Leave the tasklist_lock locked for the call below.  */
1040         }
1041
1042         /*
1043          * Now re-arm for the new expiry time.
1044          */
1045         BUG_ON(!irqs_disabled());
1046         arm_timer(timer);
1047         spin_unlock(&p->sighand->siglock);
1048         read_unlock(&tasklist_lock);
1049
1050         /* Kick full dynticks CPUs in case they need to tick on the new timer */
1051         posix_cpu_timer_kick_nohz();
1052
1053 out:
1054         timer->it_overrun_last = timer->it_overrun;
1055         timer->it_overrun = -1;
1056         ++timer->it_requeue_pending;
1057 }
1058
1059 /**
1060  * task_cputime_expired - Compare two task_cputime entities.
1061  *
1062  * @sample:     The task_cputime structure to be checked for expiration.
1063  * @expires:    Expiration times, against which @sample will be checked.
1064  *
1065  * Checks @sample against @expires to see if any field of @sample has expired.
1066  * Returns true if any field of the former is greater than the corresponding
1067  * field of the latter if the latter field is set.  Otherwise returns false.
1068  */
1069 static inline int task_cputime_expired(const struct task_cputime *sample,
1070                                         const struct task_cputime *expires)
1071 {
1072         if (expires->utime && sample->utime >= expires->utime)
1073                 return 1;
1074         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1075                 return 1;
1076         if (expires->sum_exec_runtime != 0 &&
1077             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1078                 return 1;
1079         return 0;
1080 }
1081
1082 /**
1083  * fastpath_timer_check - POSIX CPU timers fast path.
1084  *
1085  * @tsk:        The task (thread) being checked.
1086  *
1087  * Check the task and thread group timers.  If both are zero (there are no
1088  * timers set) return false.  Otherwise snapshot the task and thread group
1089  * timers and compare them with the corresponding expiration times.  Return
1090  * true if a timer has expired, else return false.
1091  */
1092 static inline int fastpath_timer_check(struct task_struct *tsk)
1093 {
1094         struct signal_struct *sig;
1095         cputime_t utime, stime;
1096
1097         task_cputime(tsk, &utime, &stime);
1098
1099         if (!task_cputime_zero(&tsk->cputime_expires)) {
1100                 struct task_cputime task_sample = {
1101                         .utime = utime,
1102                         .stime = stime,
1103                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1104                 };
1105
1106                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1107                         return 1;
1108         }
1109
1110         sig = tsk->signal;
1111         if (sig->cputimer.running) {
1112                 struct task_cputime group_sample;
1113
1114                 raw_spin_lock(&sig->cputimer.lock);
1115                 group_sample = sig->cputimer.cputime;
1116                 raw_spin_unlock(&sig->cputimer.lock);
1117
1118                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1119                         return 1;
1120         }
1121
1122         return 0;
1123 }
1124
1125 /*
1126  * This is called from the timer interrupt handler.  The irq handler has
1127  * already updated our counts.  We need to check if any timers fire now.
1128  * Interrupts are disabled.
1129  */
1130 void run_posix_cpu_timers(struct task_struct *tsk)
1131 {
1132         LIST_HEAD(firing);
1133         struct k_itimer *timer, *next;
1134         unsigned long flags;
1135
1136         BUG_ON(!irqs_disabled());
1137
1138         /*
1139          * The fast path checks that there are no expired thread or thread
1140          * group timers.  If that's so, just return.
1141          */
1142         if (!fastpath_timer_check(tsk))
1143                 return;
1144
1145         if (!lock_task_sighand(tsk, &flags))
1146                 return;
1147         /*
1148          * Here we take off tsk->signal->cpu_timers[N] and
1149          * tsk->cpu_timers[N] all the timers that are firing, and
1150          * put them on the firing list.
1151          */
1152         check_thread_timers(tsk, &firing);
1153         /*
1154          * If there are any active process wide timers (POSIX 1.b, itimers,
1155          * RLIMIT_CPU) cputimer must be running.
1156          */
1157         if (tsk->signal->cputimer.running)
1158                 check_process_timers(tsk, &firing);
1159
1160         /*
1161          * We must release these locks before taking any timer's lock.
1162          * There is a potential race with timer deletion here, as the
1163          * siglock now protects our private firing list.  We have set
1164          * the firing flag in each timer, so that a deletion attempt
1165          * that gets the timer lock before we do will give it up and
1166          * spin until we've taken care of that timer below.
1167          */
1168         unlock_task_sighand(tsk, &flags);
1169
1170         /*
1171          * Now that all the timers on our list have the firing flag,
1172          * no one will touch their list entries but us.  We'll take
1173          * each timer's lock before clearing its firing flag, so no
1174          * timer call will interfere.
1175          */
1176         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1177                 int cpu_firing;
1178
1179                 spin_lock(&timer->it_lock);
1180                 list_del_init(&timer->it.cpu.entry);
1181                 cpu_firing = timer->it.cpu.firing;
1182                 timer->it.cpu.firing = 0;
1183                 /*
1184                  * The firing flag is -1 if we collided with a reset
1185                  * of the timer, which already reported this
1186                  * almost-firing as an overrun.  So don't generate an event.
1187                  */
1188                 if (likely(cpu_firing >= 0))
1189                         cpu_timer_fire(timer);
1190                 spin_unlock(&timer->it_lock);
1191         }
1192 }
1193
1194 /*
1195  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1196  * The tsk->sighand->siglock must be held by the caller.
1197  */
1198 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1199                            cputime_t *newval, cputime_t *oldval)
1200 {
1201         unsigned long long now;
1202
1203         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1204         cpu_timer_sample_group(clock_idx, tsk, &now);
1205
1206         if (oldval) {
1207                 /*
1208                  * We are setting itimer. The *oldval is absolute and we update
1209                  * it to be relative, *newval argument is relative and we update
1210                  * it to be absolute.
1211                  */
1212                 if (*oldval) {
1213                         if (*oldval <= now) {
1214                                 /* Just about to fire. */
1215                                 *oldval = cputime_one_jiffy;
1216                         } else {
1217                                 *oldval -= now;
1218                         }
1219                 }
1220
1221                 if (!*newval)
1222                         goto out;
1223                 *newval += now;
1224         }
1225
1226         /*
1227          * Update expiration cache if we are the earliest timer, or eventually
1228          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1229          */
1230         switch (clock_idx) {
1231         case CPUCLOCK_PROF:
1232                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1233                         tsk->signal->cputime_expires.prof_exp = *newval;
1234                 break;
1235         case CPUCLOCK_VIRT:
1236                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1237                         tsk->signal->cputime_expires.virt_exp = *newval;
1238                 break;
1239         }
1240 out:
1241         posix_cpu_timer_kick_nohz();
1242 }
1243
1244 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1245                             struct timespec *rqtp, struct itimerspec *it)
1246 {
1247         struct k_itimer timer;
1248         int error;
1249
1250         /*
1251          * Set up a temporary timer and then wait for it to go off.
1252          */
1253         memset(&timer, 0, sizeof timer);
1254         spin_lock_init(&timer.it_lock);
1255         timer.it_clock = which_clock;
1256         timer.it_overrun = -1;
1257         error = posix_cpu_timer_create(&timer);
1258         timer.it_process = current;
1259         if (!error) {
1260                 static struct itimerspec zero_it;
1261
1262                 memset(it, 0, sizeof *it);
1263                 it->it_value = *rqtp;
1264
1265                 spin_lock_irq(&timer.it_lock);
1266                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1267                 if (error) {
1268                         spin_unlock_irq(&timer.it_lock);
1269                         return error;
1270                 }
1271
1272                 while (!signal_pending(current)) {
1273                         if (timer.it.cpu.expires == 0) {
1274                                 /*
1275                                  * Our timer fired and was reset, below
1276                                  * deletion can not fail.
1277                                  */
1278                                 posix_cpu_timer_del(&timer);
1279                                 spin_unlock_irq(&timer.it_lock);
1280                                 return 0;
1281                         }
1282
1283                         /*
1284                          * Block until cpu_timer_fire (or a signal) wakes us.
1285                          */
1286                         __set_current_state(TASK_INTERRUPTIBLE);
1287                         spin_unlock_irq(&timer.it_lock);
1288                         schedule();
1289                         spin_lock_irq(&timer.it_lock);
1290                 }
1291
1292                 /*
1293                  * We were interrupted by a signal.
1294                  */
1295                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1296                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1297                 if (!error) {
1298                         /*
1299                          * Timer is now unarmed, deletion can not fail.
1300                          */
1301                         posix_cpu_timer_del(&timer);
1302                 }
1303                 spin_unlock_irq(&timer.it_lock);
1304
1305                 while (error == TIMER_RETRY) {
1306                         /*
1307                          * We need to handle case when timer was or is in the
1308                          * middle of firing. In other cases we already freed
1309                          * resources.
1310                          */
1311                         spin_lock_irq(&timer.it_lock);
1312                         error = posix_cpu_timer_del(&timer);
1313                         spin_unlock_irq(&timer.it_lock);
1314                 }
1315
1316                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1317                         /*
1318                          * It actually did fire already.
1319                          */
1320                         return 0;
1321                 }
1322
1323                 error = -ERESTART_RESTARTBLOCK;
1324         }
1325
1326         return error;
1327 }
1328
1329 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1330
1331 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1332                             struct timespec *rqtp, struct timespec __user *rmtp)
1333 {
1334         struct restart_block *restart_block =
1335                 &current_thread_info()->restart_block;
1336         struct itimerspec it;
1337         int error;
1338
1339         /*
1340          * Diagnose required errors first.
1341          */
1342         if (CPUCLOCK_PERTHREAD(which_clock) &&
1343             (CPUCLOCK_PID(which_clock) == 0 ||
1344              CPUCLOCK_PID(which_clock) == current->pid))
1345                 return -EINVAL;
1346
1347         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1348
1349         if (error == -ERESTART_RESTARTBLOCK) {
1350
1351                 if (flags & TIMER_ABSTIME)
1352                         return -ERESTARTNOHAND;
1353                 /*
1354                  * Report back to the user the time still remaining.
1355                  */
1356                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1357                         return -EFAULT;
1358
1359                 restart_block->fn = posix_cpu_nsleep_restart;
1360                 restart_block->nanosleep.clockid = which_clock;
1361                 restart_block->nanosleep.rmtp = rmtp;
1362                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1363         }
1364         return error;
1365 }
1366
1367 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1368 {
1369         clockid_t which_clock = restart_block->nanosleep.clockid;
1370         struct timespec t;
1371         struct itimerspec it;
1372         int error;
1373
1374         t = ns_to_timespec(restart_block->nanosleep.expires);
1375
1376         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1377
1378         if (error == -ERESTART_RESTARTBLOCK) {
1379                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1380                 /*
1381                  * Report back to the user the time still remaining.
1382                  */
1383                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1384                         return -EFAULT;
1385
1386                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1387         }
1388         return error;
1389
1390 }
1391
1392 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1393 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1394
1395 static int process_cpu_clock_getres(const clockid_t which_clock,
1396                                     struct timespec *tp)
1397 {
1398         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1399 }
1400 static int process_cpu_clock_get(const clockid_t which_clock,
1401                                  struct timespec *tp)
1402 {
1403         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1404 }
1405 static int process_cpu_timer_create(struct k_itimer *timer)
1406 {
1407         timer->it_clock = PROCESS_CLOCK;
1408         return posix_cpu_timer_create(timer);
1409 }
1410 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1411                               struct timespec *rqtp,
1412                               struct timespec __user *rmtp)
1413 {
1414         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1415 }
1416 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1417 {
1418         return -EINVAL;
1419 }
1420 static int thread_cpu_clock_getres(const clockid_t which_clock,
1421                                    struct timespec *tp)
1422 {
1423         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1424 }
1425 static int thread_cpu_clock_get(const clockid_t which_clock,
1426                                 struct timespec *tp)
1427 {
1428         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1429 }
1430 static int thread_cpu_timer_create(struct k_itimer *timer)
1431 {
1432         timer->it_clock = THREAD_CLOCK;
1433         return posix_cpu_timer_create(timer);
1434 }
1435
1436 struct k_clock clock_posix_cpu = {
1437         .clock_getres   = posix_cpu_clock_getres,
1438         .clock_set      = posix_cpu_clock_set,
1439         .clock_get      = posix_cpu_clock_get,
1440         .timer_create   = posix_cpu_timer_create,
1441         .nsleep         = posix_cpu_nsleep,
1442         .nsleep_restart = posix_cpu_nsleep_restart,
1443         .timer_set      = posix_cpu_timer_set,
1444         .timer_del      = posix_cpu_timer_del,
1445         .timer_get      = posix_cpu_timer_get,
1446 };
1447
1448 static __init int init_posix_cpu_timers(void)
1449 {
1450         struct k_clock process = {
1451                 .clock_getres   = process_cpu_clock_getres,
1452                 .clock_get      = process_cpu_clock_get,
1453                 .timer_create   = process_cpu_timer_create,
1454                 .nsleep         = process_cpu_nsleep,
1455                 .nsleep_restart = process_cpu_nsleep_restart,
1456         };
1457         struct k_clock thread = {
1458                 .clock_getres   = thread_cpu_clock_getres,
1459                 .clock_get      = thread_cpu_clock_get,
1460                 .timer_create   = thread_cpu_timer_create,
1461         };
1462         struct timespec ts;
1463
1464         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1465         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1466
1467         cputime_to_timespec(cputime_one_jiffy, &ts);
1468         onecputick = ts.tv_nsec;
1469         WARN_ON(ts.tv_sec != 0);
1470
1471         return 0;
1472 }
1473 __initcall(init_posix_cpu_timers);