kernel/time/posix-cpu-timers.c

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