kernel/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/posix-timers.h>
   7 #include <linux/errno.h>
   8 #include <linux/math64.h>
   9 #include <asm/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11 #include <trace/events/timer.h>
  12 #include <linux/random.h>
  13 #include <linux/tick.h>
  14 #include <linux/workqueue.h>
  15
  16 /*
  17  * Called after updating RLIMIT_CPU to run cpu timer and update
  18  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  19  * siglock protection since other code may update expiration cache as
  20  * well.
  21  */
  22 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  23 {
  24         cputime_t cputime = secs_to_cputime(rlim_new);
  25
  26         spin_lock_irq(&task->sighand->siglock);
  27         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  28         spin_unlock_irq(&task->sighand->siglock);
  29 }
  30
  31 static int check_clock(const clockid_t which_clock)
  32 {
  33         int error = 0;
  34         struct task_struct *p;
  35         const pid_t pid = CPUCLOCK_PID(which_clock);
  36
  37         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  38                 return -EINVAL;
  39
  40         if (pid == 0)
  41                 return 0;
  42
  43         rcu_read_lock();
  44         p = find_task_by_vpid(pid);
  45         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  46                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  47                 error = -EINVAL;
  48         }
  49         rcu_read_unlock();
  50
  51         return error;
  52 }
  53
  54 static inline unsigned long long
  55 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  56 {
  57         unsigned long long ret;
  58
  59         ret = 0;                /* high half always zero when .cpu used */
  60         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  61                 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  62         } else {
  63                 ret = cputime_to_expires(timespec_to_cputime(tp));
  64         }
  65         return ret;
  66 }
  67
  68 static void sample_to_timespec(const clockid_t which_clock,
  69                                unsigned long long expires,
  70                                struct timespec *tp)
  71 {
  72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  73                 *tp = ns_to_timespec(expires);
  74         else
  75                 cputime_to_timespec((__force cputime_t)expires, tp);
  76 }
  77
  78 /*
  79  * Update expiry time from increment, and increase overrun count,
  80  * given the current clock sample.
  81  */
  82 static void bump_cpu_timer(struct k_itimer *timer,
  83                            unsigned long long now)
  84 {
  85         int i;
  86         unsigned long long delta, incr;
  87
  88         if (timer->it.cpu.incr == 0)
  89                 return;
  90
  91         if (now < timer->it.cpu.expires)
  92                 return;
  93
  94         incr = timer->it.cpu.incr;
  95         delta = now + incr - timer->it.cpu.expires;
  96
  97         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  98         for (i = 0; incr < delta - incr; i++)
  99                 incr = incr << 1;
 100
 101         for (; i >= 0; incr >>= 1, i--) {
 102                 if (delta < incr)
 103                         continue;
 104
 105                 timer->it.cpu.expires += incr;
 106                 timer->it_overrun += 1 << i;
 107                 delta -= incr;
 108         }
 109 }
 110
 111 /**
 112  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 113  *
 114  * @cputime:    The struct to compare.
 115  *
 116  * Checks @cputime to see if all fields are zero.  Returns true if all fields
 117  * are zero, false if any field is nonzero.
 118  */
 119 static inline int task_cputime_zero(const struct task_cputime *cputime)
 120 {
 121         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 static inline unsigned long long prof_ticks(struct task_struct *p)
 127 {
 128         cputime_t utime, stime;
 129
 130         task_cputime(p, &utime, &stime);
 131
 132         return cputime_to_expires(utime + stime);
 133 }
 134 static inline unsigned long long virt_ticks(struct task_struct *p)
 135 {
 136         cputime_t utime;
 137
 138         task_cputime(p, &utime, NULL);
 139
 140         return cputime_to_expires(utime);
 141 }
 142
 143 static int
 144 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 145 {
 146         int error = check_clock(which_clock);
 147         if (!error) {
 148                 tp->tv_sec = 0;
 149                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 150                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 151                         /*
 152                          * If sched_clock is using a cycle counter, we
 153                          * don't have any idea of its true resolution
 154                          * exported, but it is much more than 1s/HZ.
 155                          */
 156                         tp->tv_nsec = 1;
 157                 }
 158         }
 159         return error;
 160 }
 161
 162 static int
 163 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 164 {
 165         /*
 166          * You can never reset a CPU clock, but we check for other errors
 167          * in the call before failing with EPERM.
 168          */
 169         int error = check_clock(which_clock);
 170         if (error == 0) {
 171                 error = -EPERM;
 172         }
 173         return error;
 174 }
 175
 176
 177 /*
 178  * Sample a per-thread clock for the given task.
 179  */
 180 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 181                             unsigned long long *sample)
 182 {
 183         switch (CPUCLOCK_WHICH(which_clock)) {
 184         default:
 185                 return -EINVAL;
 186         case CPUCLOCK_PROF:
 187                 *sample = prof_ticks(p);
 188                 break;
 189         case CPUCLOCK_VIRT:
 190                 *sample = virt_ticks(p);
 191                 break;
 192         case CPUCLOCK_SCHED:
 193                 *sample = task_sched_runtime(p);
 194                 break;
 195         }
 196         return 0;
 197 }
 198
 199 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 200 {
 201         if (b->utime > a->utime)
 202                 a->utime = b->utime;
 203
 204         if (b->stime > a->stime)
 205                 a->stime = b->stime;
 206
 207         if (b->sum_exec_runtime > a->sum_exec_runtime)
 208                 a->sum_exec_runtime = b->sum_exec_runtime;
 209 }
 210
 211 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 212 {
 213         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 214         struct task_cputime sum;
 215         unsigned long flags;
 216
 217         if (!cputimer->running) {
 218                 /*
 219                  * The POSIX timer interface allows for absolute time expiry
 220                  * values through the TIMER_ABSTIME flag, therefore we have
 221                  * to synchronize the timer to the clock every time we start
 222                  * it.
 223                  */
 224                 thread_group_cputime(tsk, &sum);
 225                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 226                 cputimer->running = 1;
 227                 update_gt_cputime(&cputimer->cputime, &sum);
 228         } else
 229                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 230         *times = cputimer->cputime;
 231         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 232 }
 233
 234 /*
 235  * Sample a process (thread group) clock for the given group_leader task.
 236  * Must be called with tasklist_lock held for reading.
 237  */
 238 static int cpu_clock_sample_group(const clockid_t which_clock,
 239                                   struct task_struct *p,
 240                                   unsigned long long *sample)
 241 {
 242         struct task_cputime cputime;
 243
 244         switch (CPUCLOCK_WHICH(which_clock)) {
 245         default:
 246                 return -EINVAL;
 247         case CPUCLOCK_PROF:
 248                 thread_group_cputime(p, &cputime);
 249                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 250                 break;
 251         case CPUCLOCK_VIRT:
 252                 thread_group_cputime(p, &cputime);
 253                 *sample = cputime_to_expires(cputime.utime);
 254                 break;
 255         case CPUCLOCK_SCHED:
 256                 thread_group_cputime(p, &cputime);
 257                 *sample = cputime.sum_exec_runtime;
 258                 break;
 259         }
 260         return 0;
 261 }
 262
 263
 264 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 265 {
 266         const pid_t pid = CPUCLOCK_PID(which_clock);
 267         int error = -EINVAL;
 268         unsigned long long rtn;
 269
 270         if (pid == 0) {
 271                 /*
 272                  * Special case constant value for our own clocks.
 273                  * We don't have to do any lookup to find ourselves.
 274                  */
 275                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 276                         /*
 277                          * Sampling just ourselves we can do with no locking.
 278                          */
 279                         error = cpu_clock_sample(which_clock,
 280                                                  current, &rtn);
 281                 } else {
 282                         read_lock(&tasklist_lock);
 283                         error = cpu_clock_sample_group(which_clock,
 284                                                        current, &rtn);
 285                         read_unlock(&tasklist_lock);
 286                 }
 287         } else {
 288                 /*
 289                  * Find the given PID, and validate that the caller
 290                  * should be able to see it.
 291                  */
 292                 struct task_struct *p;
 293                 rcu_read_lock();
 294                 p = find_task_by_vpid(pid);
 295                 if (p) {
 296                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 297                                 if (same_thread_group(p, current)) {
 298                                         error = cpu_clock_sample(which_clock,
 299                                                                  p, &rtn);
 300                                 }
 301                         } else {
 302                                 read_lock(&tasklist_lock);
 303                                 if (thread_group_leader(p) && p->sighand) {
 304                                         error =
 305                                             cpu_clock_sample_group(which_clock,
 306                                                                    p, &rtn);
 307                                 }
 308                                 read_unlock(&tasklist_lock);
 309                         }
 310                 }
 311                 rcu_read_unlock();
 312         }
 313
 314         if (error)
 315                 return error;
 316         sample_to_timespec(which_clock, rtn, tp);
 317         return 0;
 318 }
 319
 320
 321 /*
 322  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 323  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 324  * new timer already all-zeros initialized.
 325  */
 326 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 327 {
 328         int ret = 0;
 329         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 330         struct task_struct *p;
 331
 332         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 333                 return -EINVAL;
 334
 335         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 336
 337         rcu_read_lock();
 338         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 339                 if (pid == 0) {
 340                         p = current;
 341                 } else {
 342                         p = find_task_by_vpid(pid);
 343                         if (p && !same_thread_group(p, current))
 344                                 p = NULL;
 345                 }
 346         } else {
 347                 if (pid == 0) {
 348                         p = current->group_leader;
 349                 } else {
 350                         p = find_task_by_vpid(pid);
 351                         if (p && !has_group_leader_pid(p))
 352                                 p = NULL;
 353                 }
 354         }
 355         new_timer->it.cpu.task = p;
 356         if (p) {
 357                 get_task_struct(p);
 358         } else {
 359                 ret = -EINVAL;
 360         }
 361         rcu_read_unlock();
 362
 363         return ret;
 364 }
 365
 366 /*
 367  * Clean up a CPU-clock timer that is about to be destroyed.
 368  * This is called from timer deletion with the timer already locked.
 369  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 370  * and try again.  (This happens when the timer is in the middle of firing.)
 371  */
 372 static int posix_cpu_timer_del(struct k_itimer *timer)
 373 {
 374         struct task_struct *p = timer->it.cpu.task;
 375         int ret = 0;
 376
 377         WARN_ON_ONCE(p == NULL);
 378
 379         read_lock(&tasklist_lock);
 380         if (unlikely(p->sighand == NULL)) {
 381                 /*
 382                  * We raced with the reaping of the task.
 383                  * The deletion should have cleared us off the list.
 384                  */
 385                 BUG_ON(!list_empty(&timer->it.cpu.entry));
 386         } else {
 387                 spin_lock(&p->sighand->siglock);
 388                 if (timer->it.cpu.firing)
 389                         ret = TIMER_RETRY;
 390                 else
 391                         list_del(&timer->it.cpu.entry);
 392                 spin_unlock(&p->sighand->siglock);
 393         }
 394         read_unlock(&tasklist_lock);
 395
 396         if (!ret)
 397                 put_task_struct(p);
 398
 399         return ret;
 400 }
 401
 402 static void cleanup_timers_list(struct list_head *head,
 403                                 unsigned long long curr)
 404 {
 405         struct cpu_timer_list *timer, *next;
 406
 407         list_for_each_entry_safe(timer, next, head, entry)
 408                 list_del_init(&timer->entry);
 409 }
 410
 411 /*
 412  * Clean out CPU timers still ticking when a thread exited.  The task
 413  * pointer is cleared, and the expiry time is replaced with the residual
 414  * time for later timer_gettime calls to return.
 415  * This must be called with the siglock held.
 416  */
 417 static void cleanup_timers(struct list_head *head,
 418                            cputime_t utime, cputime_t stime,
 419                            unsigned long long sum_exec_runtime)
 420 {
 421
 422         cputime_t ptime = utime + stime;
 423
 424         cleanup_timers_list(head, cputime_to_expires(ptime));
 425         cleanup_timers_list(++head, cputime_to_expires(utime));
 426         cleanup_timers_list(++head, sum_exec_runtime);
 427 }
 428
 429 /*
 430  * These are both called with the siglock held, when the current thread
 431  * is being reaped.  When the final (leader) thread in the group is reaped,
 432  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 433  */
 434 void posix_cpu_timers_exit(struct task_struct *tsk)
 435 {
 436         cputime_t utime, stime;
 437
 438         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 439                                                 sizeof(unsigned long long));
 440         task_cputime(tsk, &utime, &stime);
 441         cleanup_timers(tsk->cpu_timers,
 442                        utime, stime, tsk->se.sum_exec_runtime);
 443
 444 }
 445 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 446 {
 447         struct signal_struct *const sig = tsk->signal;
 448         cputime_t utime, stime;
 449
 450         task_cputime(tsk, &utime, &stime);
 451         cleanup_timers(tsk->signal->cpu_timers,
 452                        utime + sig->utime, stime + sig->stime,
 453                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 454 }
 455
 456 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 457 {
 458         return expires == 0 || expires > new_exp;
 459 }
 460
 461 /*
 462  * Insert the timer on the appropriate list before any timers that
 463  * expire later.  This must be called with the tasklist_lock held
 464  * for reading, interrupts disabled and p->sighand->siglock taken.
 465  */
 466 static void arm_timer(struct k_itimer *timer)
 467 {
 468         struct task_struct *p = timer->it.cpu.task;
 469         struct list_head *head, *listpos;
 470         struct task_cputime *cputime_expires;
 471         struct cpu_timer_list *const nt = &timer->it.cpu;
 472         struct cpu_timer_list *next;
 473
 474         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 475                 head = p->cpu_timers;
 476                 cputime_expires = &p->cputime_expires;
 477         } else {
 478                 head = p->signal->cpu_timers;
 479                 cputime_expires = &p->signal->cputime_expires;
 480         }
 481         head += CPUCLOCK_WHICH(timer->it_clock);
 482
 483         listpos = head;
 484         list_for_each_entry(next, head, entry) {
 485                 if (nt->expires < next->expires)
 486                         break;
 487                 listpos = &next->entry;
 488         }
 489         list_add(&nt->entry, listpos);
 490
 491         if (listpos == head) {
 492                 unsigned long long exp = nt->expires;
 493
 494                 /*
 495                  * We are the new earliest-expiring POSIX 1.b timer, hence
 496                  * need to update expiration cache. Take into account that
 497                  * for process timers we share expiration cache with itimers
 498                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 499                  */
 500
 501                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 502                 case CPUCLOCK_PROF:
 503                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 504                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 505                         break;
 506                 case CPUCLOCK_VIRT:
 507                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 508                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 509                         break;
 510                 case CPUCLOCK_SCHED:
 511                         if (cputime_expires->sched_exp == 0 ||
 512                             cputime_expires->sched_exp > exp)
 513                                 cputime_expires->sched_exp = exp;
 514                         break;
 515                 }
 516         }
 517 }
 518
 519 /*
 520  * The timer is locked, fire it and arrange for its reload.
 521  */
 522 static void cpu_timer_fire(struct k_itimer *timer)
 523 {
 524         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 525                 /*
 526                  * User don't want any signal.
 527                  */
 528                 timer->it.cpu.expires = 0;
 529         } else if (unlikely(timer->sigq == NULL)) {
 530                 /*
 531                  * This a special case for clock_nanosleep,
 532                  * not a normal timer from sys_timer_create.
 533                  */
 534                 wake_up_process(timer->it_process);
 535                 timer->it.cpu.expires = 0;
 536         } else if (timer->it.cpu.incr == 0) {
 537                 /*
 538                  * One-shot timer.  Clear it as soon as it's fired.
 539                  */
 540                 posix_timer_event(timer, 0);
 541                 timer->it.cpu.expires = 0;
 542         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 543                 /*
 544                  * The signal did not get queued because the signal
 545                  * was ignored, so we won't get any callback to
 546                  * reload the timer.  But we need to keep it
 547                  * ticking in case the signal is deliverable next time.
 548                  */
 549                 posix_cpu_timer_schedule(timer);
 550         }
 551 }
 552
 553 /*
 554  * Sample a process (thread group) timer for the given group_leader task.
 555  * Must be called with tasklist_lock held for reading.
 556  */
 557 static int cpu_timer_sample_group(const clockid_t which_clock,
 558                                   struct task_struct *p,
 559                                   unsigned long long *sample)
 560 {
 561         struct task_cputime cputime;
 562
 563         thread_group_cputimer(p, &cputime);
 564         switch (CPUCLOCK_WHICH(which_clock)) {
 565         default:
 566                 return -EINVAL;
 567         case CPUCLOCK_PROF:
 568                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 569                 break;
 570         case CPUCLOCK_VIRT:
 571                 *sample = cputime_to_expires(cputime.utime);
 572                 break;
 573         case CPUCLOCK_SCHED:
 574                 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
 575                 break;
 576         }
 577         return 0;
 578 }
 579
 580 #ifdef CONFIG_NO_HZ_FULL
 581 static void nohz_kick_work_fn(struct work_struct *work)
 582 {
 583         tick_nohz_full_kick_all();
 584 }
 585
 586 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
 587
 588 /*
 589  * We need the IPIs to be sent from sane process context.
 590  * The posix cpu timers are always set with irqs disabled.
 591  */
 592 static void posix_cpu_timer_kick_nohz(void)
 593 {
 594         if (context_tracking_is_enabled())
 595                 schedule_work(&nohz_kick_work);
 596 }
 597
 598 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
 599 {
 600         if (!task_cputime_zero(&tsk->cputime_expires))
 601                 return false;
 602
 603         if (tsk->signal->cputimer.running)
 604                 return false;
 605
 606         return true;
 607 }
 608 #else
 609 static inline void posix_cpu_timer_kick_nohz(void) { }
 610 #endif
 611
 612 /*
 613  * Guts of sys_timer_settime for CPU timers.
 614  * This is called with the timer locked and interrupts disabled.
 615  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 616  * and try again.  (This happens when the timer is in the middle of firing.)
 617  */
 618 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 619                                struct itimerspec *new, struct itimerspec *old)
 620 {
 621         struct task_struct *p = timer->it.cpu.task;
 622         unsigned long long old_expires, new_expires, old_incr, val;
 623         int ret;
 624
 625         WARN_ON_ONCE(p == NULL);
 626
 627         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 628
 629         read_lock(&tasklist_lock);
 630         /*
 631          * We need the tasklist_lock to protect against reaping that
 632          * clears p->sighand.  If p has just been reaped, we can no
 633          * longer get any information about it at all.
 634          */
 635         if (unlikely(p->sighand == NULL)) {
 636                 read_unlock(&tasklist_lock);
 637                 return -ESRCH;
 638         }
 639
 640         /*
 641          * Disarm any old timer after extracting its expiry time.
 642          */
 643         BUG_ON(!irqs_disabled());
 644
 645         ret = 0;
 646         old_incr = timer->it.cpu.incr;
 647         spin_lock(&p->sighand->siglock);
 648         old_expires = timer->it.cpu.expires;
 649         if (unlikely(timer->it.cpu.firing)) {
 650                 timer->it.cpu.firing = -1;
 651                 ret = TIMER_RETRY;
 652         } else
 653                 list_del_init(&timer->it.cpu.entry);
 654
 655         /*
 656          * We need to sample the current value to convert the new
 657          * value from to relative and absolute, and to convert the
 658          * old value from absolute to relative.  To set a process
 659          * timer, we need a sample to balance the thread expiry
 660          * times (in arm_timer).  With an absolute time, we must
 661          * check if it's already passed.  In short, we need a sample.
 662          */
 663         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 664                 cpu_clock_sample(timer->it_clock, p, &val);
 665         } else {
 666                 cpu_timer_sample_group(timer->it_clock, p, &val);
 667         }
 668
 669         if (old) {
 670                 if (old_expires == 0) {
 671                         old->it_value.tv_sec = 0;
 672                         old->it_value.tv_nsec = 0;
 673                 } else {
 674                         /*
 675                          * Update the timer in case it has
 676                          * overrun already.  If it has,
 677                          * we'll report it as having overrun
 678                          * and with the next reloaded timer
 679                          * already ticking, though we are
 680                          * swallowing that pending
 681                          * notification here to install the
 682                          * new setting.
 683                          */
 684                         bump_cpu_timer(timer, val);
 685                         if (val < timer->it.cpu.expires) {
 686                                 old_expires = timer->it.cpu.expires - val;
 687                                 sample_to_timespec(timer->it_clock,
 688                                                    old_expires,
 689                                                    &old->it_value);
 690                         } else {
 691                                 old->it_value.tv_nsec = 1;
 692                                 old->it_value.tv_sec = 0;
 693                         }
 694                 }
 695         }
 696
 697         if (unlikely(ret)) {
 698                 /*
 699                  * We are colliding with the timer actually firing.
 700                  * Punt after filling in the timer's old value, and
 701                  * disable this firing since we are already reporting
 702                  * it as an overrun (thanks to bump_cpu_timer above).
 703                  */
 704                 spin_unlock(&p->sighand->siglock);
 705                 read_unlock(&tasklist_lock);
 706                 goto out;
 707         }
 708
 709         if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
 710                 new_expires += val;
 711         }
 712
 713         /*
 714          * Install the new expiry time (or zero).
 715          * For a timer with no notification action, we don't actually
 716          * arm the timer (we'll just fake it for timer_gettime).
 717          */
 718         timer->it.cpu.expires = new_expires;
 719         if (new_expires != 0 && val < new_expires) {
 720                 arm_timer(timer);
 721         }
 722
 723         spin_unlock(&p->sighand->siglock);
 724         read_unlock(&tasklist_lock);
 725
 726         /*
 727          * Install the new reload setting, and
 728          * set up the signal and overrun bookkeeping.
 729          */
 730         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 731                                                 &new->it_interval);
 732
 733         /*
 734          * This acts as a modification timestamp for the timer,
 735          * so any automatic reload attempt will punt on seeing
 736          * that we have reset the timer manually.
 737          */
 738         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 739                 ~REQUEUE_PENDING;
 740         timer->it_overrun_last = 0;
 741         timer->it_overrun = -1;
 742
 743         if (new_expires != 0 && !(val < new_expires)) {
 744                 /*
 745                  * The designated time already passed, so we notify
 746                  * immediately, even if the thread never runs to
 747                  * accumulate more time on this clock.
 748                  */
 749                 cpu_timer_fire(timer);
 750         }
 751
 752         ret = 0;
 753  out:
 754         if (old) {
 755                 sample_to_timespec(timer->it_clock,
 756                                    old_incr, &old->it_interval);
 757         }
 758         if (!ret)
 759                 posix_cpu_timer_kick_nohz();
 760         return ret;
 761 }
 762
 763 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 764 {
 765         unsigned long long now;
 766         struct task_struct *p = timer->it.cpu.task;
 767
 768         WARN_ON_ONCE(p == NULL);
 769
 770         /*
 771          * Easy part: convert the reload time.
 772          */
 773         sample_to_timespec(timer->it_clock,
 774                            timer->it.cpu.incr, &itp->it_interval);
 775
 776         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 777                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 778                 return;
 779         }
 780
 781         /*
 782          * Sample the clock to take the difference with the expiry time.
 783          */
 784         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 785                 cpu_clock_sample(timer->it_clock, p, &now);
 786         } else {
 787                 read_lock(&tasklist_lock);
 788                 if (unlikely(p->sighand == NULL)) {
 789                         /*
 790                          * The process has been reaped.
 791                          * We can't even collect a sample any more.
 792                          * Call the timer disarmed, nothing else to do.
 793                          */
 794                         timer->it.cpu.expires = 0;
 795                         sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 796                                            &itp->it_value);
 797                         read_unlock(&tasklist_lock);
 798                 } else {
 799                         cpu_timer_sample_group(timer->it_clock, p, &now);
 800                 }
 801                 read_unlock(&tasklist_lock);
 802         }
 803
 804         if (now < timer->it.cpu.expires) {
 805                 sample_to_timespec(timer->it_clock,
 806                                    timer->it.cpu.expires - now,
 807                                    &itp->it_value);
 808         } else {
 809                 /*
 810                  * The timer should have expired already, but the firing
 811                  * hasn't taken place yet.  Say it's just about to expire.
 812                  */
 813                 itp->it_value.tv_nsec = 1;
 814                 itp->it_value.tv_sec = 0;
 815         }
 816 }
 817
 818 static unsigned long long
 819 check_timers_list(struct list_head *timers,
 820                   struct list_head *firing,
 821                   unsigned long long curr)
 822 {
 823         int maxfire = 20;
 824
 825         while (!list_empty(timers)) {
 826                 struct cpu_timer_list *t;
 827
 828                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 829
 830                 if (!--maxfire || curr < t->expires)
 831                         return t->expires;
 832
 833                 t->firing = 1;
 834                 list_move_tail(&t->entry, firing);
 835         }
 836
 837         return 0;
 838 }
 839
 840 /*
 841  * Check for any per-thread CPU timers that have fired and move them off
 842  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 843  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 844  */
 845 static void check_thread_timers(struct task_struct *tsk,
 846                                 struct list_head *firing)
 847 {
 848         struct list_head *timers = tsk->cpu_timers;
 849         struct signal_struct *const sig = tsk->signal;
 850         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 851         unsigned long long expires;
 852         unsigned long soft;
 853
 854         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 855         tsk_expires->prof_exp = expires_to_cputime(expires);
 856
 857         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 858         tsk_expires->virt_exp = expires_to_cputime(expires);
 859
 860         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 861                                                    tsk->se.sum_exec_runtime);
 862
 863         /*
 864          * Check for the special case thread timers.
 865          */
 866         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 867         if (soft != RLIM_INFINITY) {
 868                 unsigned long hard =
 869                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 870
 871                 if (hard != RLIM_INFINITY &&
 872                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 873                         /*
 874                          * At the hard limit, we just die.
 875                          * No need to calculate anything else now.
 876                          */
 877                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 878                         return;
 879                 }
 880                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 881                         /*
 882                          * At the soft limit, send a SIGXCPU every second.
 883                          */
 884                         if (soft < hard) {
 885                                 soft += USEC_PER_SEC;
 886                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 887                         }
 888                         printk(KERN_INFO
 889                                 "RT Watchdog Timeout: %s[%d]\n",
 890                                 tsk->comm, task_pid_nr(tsk));
 891                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 892                 }
 893         }
 894 }
 895
 896 static void stop_process_timers(struct signal_struct *sig)
 897 {
 898         struct thread_group_cputimer *cputimer = &sig->cputimer;
 899         unsigned long flags;
 900
 901         raw_spin_lock_irqsave(&cputimer->lock, flags);
 902         cputimer->running = 0;
 903         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 904 }
 905
 906 static u32 onecputick;
 907
 908 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 909                              unsigned long long *expires,
 910                              unsigned long long cur_time, int signo)
 911 {
 912         if (!it->expires)
 913                 return;
 914
 915         if (cur_time >= it->expires) {
 916                 if (it->incr) {
 917                         it->expires += it->incr;
 918                         it->error += it->incr_error;
 919                         if (it->error >= onecputick) {
 920                                 it->expires -= cputime_one_jiffy;
 921                                 it->error -= onecputick;
 922                         }
 923                 } else {
 924                         it->expires = 0;
 925                 }
 926
 927                 trace_itimer_expire(signo == SIGPROF ?
 928                                     ITIMER_PROF : ITIMER_VIRTUAL,
 929                                     tsk->signal->leader_pid, cur_time);
 930                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 931         }
 932
 933         if (it->expires && (!*expires || it->expires < *expires)) {
 934                 *expires = it->expires;
 935         }
 936 }
 937
 938 /*
 939  * Check for any per-thread CPU timers that have fired and move them
 940  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 941  * have already been taken off.
 942  */
 943 static void check_process_timers(struct task_struct *tsk,
 944                                  struct list_head *firing)
 945 {
 946         struct signal_struct *const sig = tsk->signal;
 947         unsigned long long utime, ptime, virt_expires, prof_expires;
 948         unsigned long long sum_sched_runtime, sched_expires;
 949         struct list_head *timers = sig->cpu_timers;
 950         struct task_cputime cputime;
 951         unsigned long soft;
 952
 953         /*
 954          * Collect the current process totals.
 955          */
 956         thread_group_cputimer(tsk, &cputime);
 957         utime = cputime_to_expires(cputime.utime);
 958         ptime = utime + cputime_to_expires(cputime.stime);
 959         sum_sched_runtime = cputime.sum_exec_runtime;
 960
 961         prof_expires = check_timers_list(timers, firing, ptime);
 962         virt_expires = check_timers_list(++timers, firing, utime);
 963         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 964
 965         /*
 966          * Check for the special case process timers.
 967          */
 968         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 969                          SIGPROF);
 970         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 971                          SIGVTALRM);
 972         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 973         if (soft != RLIM_INFINITY) {
 974                 unsigned long psecs = cputime_to_secs(ptime);
 975                 unsigned long hard =
 976                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 977                 cputime_t x;
 978                 if (psecs >= hard) {
 979                         /*
 980                          * At the hard limit, we just die.
 981                          * No need to calculate anything else now.
 982                          */
 983                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 984                         return;
 985                 }
 986                 if (psecs >= soft) {
 987                         /*
 988                          * At the soft limit, send a SIGXCPU every second.
 989                          */
 990                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 991                         if (soft < hard) {
 992                                 soft++;
 993                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 994                         }
 995                 }
 996                 x = secs_to_cputime(soft);
 997                 if (!prof_expires || x < prof_expires) {
 998                         prof_expires = x;
 999                 }
1000         }
1001
1002         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1003         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1004         sig->cputime_expires.sched_exp = sched_expires;
1005         if (task_cputime_zero(&sig->cputime_expires))
1006                 stop_process_timers(sig);
1007 }
1008
1009 /*
1010  * This is called from the signal code (via do_schedule_next_timer)
1011  * when the last timer signal was delivered and we have to reload the timer.
1012  */
1013 void posix_cpu_timer_schedule(struct k_itimer *timer)
1014 {
1015         struct task_struct *p = timer->it.cpu.task;
1016         unsigned long long now;
1017
1018         WARN_ON_ONCE(p == NULL);
1019
1020         /*
1021          * Fetch the current sample and update the timer's expiry time.
1022          */
1023         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1024                 cpu_clock_sample(timer->it_clock, p, &now);
1025                 bump_cpu_timer(timer, now);
1026                 if (unlikely(p->exit_state))
1027                         goto out;
1028
1029                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1030                 spin_lock(&p->sighand->siglock);
1031         } else {
1032                 read_lock(&tasklist_lock);
1033                 if (unlikely(p->sighand == NULL)) {
1034                         /*
1035                          * The process has been reaped.
1036                          * We can't even collect a sample any more.
1037                          */
1038                         timer->it.cpu.expires = 0;
1039                         read_unlock(&tasklist_lock);
1040                         goto out;
1041                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1042                         read_unlock(&tasklist_lock);
1043                         /* Optimizations: if the process is dying, no need to rearm */
1044                         goto out;
1045                 }
1046                 spin_lock(&p->sighand->siglock);
1047                 cpu_timer_sample_group(timer->it_clock, p, &now);
1048                 bump_cpu_timer(timer, now);
1049                 /* Leave the tasklist_lock locked for the call below.  */
1050         }
1051
1052         /*
1053          * Now re-arm for the new expiry time.
1054          */
1055         BUG_ON(!irqs_disabled());
1056         arm_timer(timer);
1057         spin_unlock(&p->sighand->siglock);
1058         read_unlock(&tasklist_lock);
1059
1060         /* Kick full dynticks CPUs in case they need to tick on the new timer */
1061         posix_cpu_timer_kick_nohz();
1062
1063 out:
1064         timer->it_overrun_last = timer->it_overrun;
1065         timer->it_overrun = -1;
1066         ++timer->it_requeue_pending;
1067 }
1068
1069 /**
1070  * task_cputime_expired - Compare two task_cputime entities.
1071  *
1072  * @sample:     The task_cputime structure to be checked for expiration.
1073  * @expires:    Expiration times, against which @sample will be checked.
1074  *
1075  * Checks @sample against @expires to see if any field of @sample has expired.
1076  * Returns true if any field of the former is greater than the corresponding
1077  * field of the latter if the latter field is set.  Otherwise returns false.
1078  */
1079 static inline int task_cputime_expired(const struct task_cputime *sample,
1080                                         const struct task_cputime *expires)
1081 {
1082         if (expires->utime && sample->utime >= expires->utime)
1083                 return 1;
1084         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1085                 return 1;
1086         if (expires->sum_exec_runtime != 0 &&
1087             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1088                 return 1;
1089         return 0;
1090 }
1091
1092 /**
1093  * fastpath_timer_check - POSIX CPU timers fast path.
1094  *
1095  * @tsk:        The task (thread) being checked.
1096  *
1097  * Check the task and thread group timers.  If both are zero (there are no
1098  * timers set) return false.  Otherwise snapshot the task and thread group
1099  * timers and compare them with the corresponding expiration times.  Return
1100  * true if a timer has expired, else return false.
1101  */
1102 static inline int fastpath_timer_check(struct task_struct *tsk)
1103 {
1104         struct signal_struct *sig;
1105         cputime_t utime, stime;
1106
1107         task_cputime(tsk, &utime, &stime);
1108
1109         if (!task_cputime_zero(&tsk->cputime_expires)) {
1110                 struct task_cputime task_sample = {
1111                         .utime = utime,
1112                         .stime = stime,
1113                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1114                 };
1115
1116                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1117                         return 1;
1118         }
1119
1120         sig = tsk->signal;
1121         if (sig->cputimer.running) {
1122                 struct task_cputime group_sample;
1123
1124                 raw_spin_lock(&sig->cputimer.lock);
1125                 group_sample = sig->cputimer.cputime;
1126                 raw_spin_unlock(&sig->cputimer.lock);
1127
1128                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1129                         return 1;
1130         }
1131
1132         return 0;
1133 }
1134
1135 /*
1136  * This is called from the timer interrupt handler.  The irq handler has
1137  * already updated our counts.  We need to check if any timers fire now.
1138  * Interrupts are disabled.
1139  */
1140 void run_posix_cpu_timers(struct task_struct *tsk)
1141 {
1142         LIST_HEAD(firing);
1143         struct k_itimer *timer, *next;
1144         unsigned long flags;
1145
1146         BUG_ON(!irqs_disabled());
1147
1148         /*
1149          * The fast path checks that there are no expired thread or thread
1150          * group timers.  If that's so, just return.
1151          */
1152         if (!fastpath_timer_check(tsk))
1153                 return;
1154
1155         if (!lock_task_sighand(tsk, &flags))
1156                 return;
1157         /*
1158          * Here we take off tsk->signal->cpu_timers[N] and
1159          * tsk->cpu_timers[N] all the timers that are firing, and
1160          * put them on the firing list.
1161          */
1162         check_thread_timers(tsk, &firing);
1163         /*
1164          * If there are any active process wide timers (POSIX 1.b, itimers,
1165          * RLIMIT_CPU) cputimer must be running.
1166          */
1167         if (tsk->signal->cputimer.running)
1168                 check_process_timers(tsk, &firing);
1169
1170         /*
1171          * We must release these locks before taking any timer's lock.
1172          * There is a potential race with timer deletion here, as the
1173          * siglock now protects our private firing list.  We have set
1174          * the firing flag in each timer, so that a deletion attempt
1175          * that gets the timer lock before we do will give it up and
1176          * spin until we've taken care of that timer below.
1177          */
1178         unlock_task_sighand(tsk, &flags);
1179
1180         /*
1181          * Now that all the timers on our list have the firing flag,
1182          * no one will touch their list entries but us.  We'll take
1183          * each timer's lock before clearing its firing flag, so no
1184          * timer call will interfere.
1185          */
1186         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1187                 int cpu_firing;
1188
1189                 spin_lock(&timer->it_lock);
1190                 list_del_init(&timer->it.cpu.entry);
1191                 cpu_firing = timer->it.cpu.firing;
1192                 timer->it.cpu.firing = 0;
1193                 /*
1194                  * The firing flag is -1 if we collided with a reset
1195                  * of the timer, which already reported this
1196                  * almost-firing as an overrun.  So don't generate an event.
1197                  */
1198                 if (likely(cpu_firing >= 0))
1199                         cpu_timer_fire(timer);
1200                 spin_unlock(&timer->it_lock);
1201         }
1202 }
1203
1204 /*
1205  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1206  * The tsk->sighand->siglock must be held by the caller.
1207  */
1208 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1209                            cputime_t *newval, cputime_t *oldval)
1210 {
1211         unsigned long long now;
1212
1213         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1214         cpu_timer_sample_group(clock_idx, tsk, &now);
1215
1216         if (oldval) {
1217                 /*
1218                  * We are setting itimer. The *oldval is absolute and we update
1219                  * it to be relative, *newval argument is relative and we update
1220                  * it to be absolute.
1221                  */
1222                 if (*oldval) {
1223                         if (*oldval <= now) {
1224                                 /* Just about to fire. */
1225                                 *oldval = cputime_one_jiffy;
1226                         } else {
1227                                 *oldval -= now;
1228                         }
1229                 }
1230
1231                 if (!*newval)
1232                         goto out;
1233                 *newval += now;
1234         }
1235
1236         /*
1237          * Update expiration cache if we are the earliest timer, or eventually
1238          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1239          */
1240         switch (clock_idx) {
1241         case CPUCLOCK_PROF:
1242                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1243                         tsk->signal->cputime_expires.prof_exp = *newval;
1244                 break;
1245         case CPUCLOCK_VIRT:
1246                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1247                         tsk->signal->cputime_expires.virt_exp = *newval;
1248                 break;
1249         }
1250 out:
1251         posix_cpu_timer_kick_nohz();
1252 }
1253
1254 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1255                             struct timespec *rqtp, struct itimerspec *it)
1256 {
1257         struct k_itimer timer;
1258         int error;
1259
1260         /*
1261          * Set up a temporary timer and then wait for it to go off.
1262          */
1263         memset(&timer, 0, sizeof timer);
1264         spin_lock_init(&timer.it_lock);
1265         timer.it_clock = which_clock;
1266         timer.it_overrun = -1;
1267         error = posix_cpu_timer_create(&timer);
1268         timer.it_process = current;
1269         if (!error) {
1270                 static struct itimerspec zero_it;
1271
1272                 memset(it, 0, sizeof *it);
1273                 it->it_value = *rqtp;
1274
1275                 spin_lock_irq(&timer.it_lock);
1276                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1277                 if (error) {
1278                         spin_unlock_irq(&timer.it_lock);
1279                         return error;
1280                 }
1281
1282                 while (!signal_pending(current)) {
1283                         if (timer.it.cpu.expires == 0) {
1284                                 /*
1285                                  * Our timer fired and was reset, below
1286                                  * deletion can not fail.
1287                                  */
1288                                 posix_cpu_timer_del(&timer);
1289                                 spin_unlock_irq(&timer.it_lock);
1290                                 return 0;
1291                         }
1292
1293                         /*
1294                          * Block until cpu_timer_fire (or a signal) wakes us.
1295                          */
1296                         __set_current_state(TASK_INTERRUPTIBLE);
1297                         spin_unlock_irq(&timer.it_lock);
1298                         schedule();
1299                         spin_lock_irq(&timer.it_lock);
1300                 }
1301
1302                 /*
1303                  * We were interrupted by a signal.
1304                  */
1305                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1306                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1307                 if (!error) {
1308                         /*
1309                          * Timer is now unarmed, deletion can not fail.
1310                          */
1311                         posix_cpu_timer_del(&timer);
1312                 }
1313                 spin_unlock_irq(&timer.it_lock);
1314
1315                 while (error == TIMER_RETRY) {
1316                         /*
1317                          * We need to handle case when timer was or is in the
1318                          * middle of firing. In other cases we already freed
1319                          * resources.
1320                          */
1321                         spin_lock_irq(&timer.it_lock);
1322                         error = posix_cpu_timer_del(&timer);
1323                         spin_unlock_irq(&timer.it_lock);
1324                 }
1325
1326                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1327                         /*
1328                          * It actually did fire already.
1329                          */
1330                         return 0;
1331                 }
1332
1333                 error = -ERESTART_RESTARTBLOCK;
1334         }
1335
1336         return error;
1337 }
1338
1339 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1340
1341 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1342                             struct timespec *rqtp, struct timespec __user *rmtp)
1343 {
1344         struct restart_block *restart_block =
1345                 &current_thread_info()->restart_block;
1346         struct itimerspec it;
1347         int error;
1348
1349         /*
1350          * Diagnose required errors first.
1351          */
1352         if (CPUCLOCK_PERTHREAD(which_clock) &&
1353             (CPUCLOCK_PID(which_clock) == 0 ||
1354              CPUCLOCK_PID(which_clock) == current->pid))
1355                 return -EINVAL;
1356
1357         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1358
1359         if (error == -ERESTART_RESTARTBLOCK) {
1360
1361                 if (flags & TIMER_ABSTIME)
1362                         return -ERESTARTNOHAND;
1363                 /*
1364                  * Report back to the user the time still remaining.
1365                  */
1366                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1367                         return -EFAULT;
1368
1369                 restart_block->fn = posix_cpu_nsleep_restart;
1370                 restart_block->nanosleep.clockid = which_clock;
1371                 restart_block->nanosleep.rmtp = rmtp;
1372                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1373         }
1374         return error;
1375 }
1376
1377 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1378 {
1379         clockid_t which_clock = restart_block->nanosleep.clockid;
1380         struct timespec t;
1381         struct itimerspec it;
1382         int error;
1383
1384         t = ns_to_timespec(restart_block->nanosleep.expires);
1385
1386         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1387
1388         if (error == -ERESTART_RESTARTBLOCK) {
1389                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1390                 /*
1391                  * Report back to the user the time still remaining.
1392                  */
1393                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1394                         return -EFAULT;
1395
1396                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1397         }
1398         return error;
1399
1400 }
1401
1402 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1403 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1404
1405 static int process_cpu_clock_getres(const clockid_t which_clock,
1406                                     struct timespec *tp)
1407 {
1408         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1409 }
1410 static int process_cpu_clock_get(const clockid_t which_clock,
1411                                  struct timespec *tp)
1412 {
1413         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1414 }
1415 static int process_cpu_timer_create(struct k_itimer *timer)
1416 {
1417         timer->it_clock = PROCESS_CLOCK;
1418         return posix_cpu_timer_create(timer);
1419 }
1420 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1421                               struct timespec *rqtp,
1422                               struct timespec __user *rmtp)
1423 {
1424         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1425 }
1426 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1427 {
1428         return -EINVAL;
1429 }
1430 static int thread_cpu_clock_getres(const clockid_t which_clock,
1431                                    struct timespec *tp)
1432 {
1433         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1434 }
1435 static int thread_cpu_clock_get(const clockid_t which_clock,
1436                                 struct timespec *tp)
1437 {
1438         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1439 }
1440 static int thread_cpu_timer_create(struct k_itimer *timer)
1441 {
1442         timer->it_clock = THREAD_CLOCK;
1443         return posix_cpu_timer_create(timer);
1444 }
1445
1446 struct k_clock clock_posix_cpu = {
1447         .clock_getres   = posix_cpu_clock_getres,
1448         .clock_set      = posix_cpu_clock_set,
1449         .clock_get      = posix_cpu_clock_get,
1450         .timer_create   = posix_cpu_timer_create,
1451         .nsleep         = posix_cpu_nsleep,
1452         .nsleep_restart = posix_cpu_nsleep_restart,
1453         .timer_set      = posix_cpu_timer_set,
1454         .timer_del      = posix_cpu_timer_del,
1455         .timer_get      = posix_cpu_timer_get,
1456 };
1457
1458 static __init int init_posix_cpu_timers(void)
1459 {
1460         struct k_clock process = {
1461                 .clock_getres   = process_cpu_clock_getres,
1462                 .clock_get      = process_cpu_clock_get,
1463                 .timer_create   = process_cpu_timer_create,
1464                 .nsleep         = process_cpu_nsleep,
1465                 .nsleep_restart = process_cpu_nsleep_restart,
1466         };
1467         struct k_clock thread = {
1468                 .clock_getres   = thread_cpu_clock_getres,
1469                 .clock_get      = thread_cpu_clock_get,
1470                 .timer_create   = thread_cpu_timer_create,
1471         };
1472         struct timespec ts;
1473
1474         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1475         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1476
1477         cputime_to_timespec(cputime_one_jiffy, &ts);
1478         onecputick = ts.tv_nsec;
1479         WARN_ON(ts.tv_sec != 0);
1480
1481         return 0;
1482 }
1483 __initcall(init_posix_cpu_timers);