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