kernel/posix-cpu-timers.c

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