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         if (likely(p != NULL)) {
 378                 read_lock(&tasklist_lock);
 379                 if (unlikely(p->sighand == NULL)) {
 380                         /*
 381                          * We raced with the reaping of the task.
 382                          * The deletion should have cleared us off the list.
 383                          */
 384                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 385                 } else {
 386                         spin_lock(&p->sighand->siglock);
 387                         if (timer->it.cpu.firing)
 388                                 ret = TIMER_RETRY;
 389                         else
 390                                 list_del(&timer->it.cpu.entry);
 391                         spin_unlock(&p->sighand->siglock);
 392                 }
 393                 read_unlock(&tasklist_lock);
 394
 395                 if (!ret)
 396                         put_task_struct(p);
 397         }
 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         if (unlikely(p == NULL)) {
 626                 /*
 627                  * Timer refers to a dead task's clock.
 628                  */
 629                 return -ESRCH;
 630         }
 631
 632         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 633
 634         read_lock(&tasklist_lock);
 635         /*
 636          * We need the tasklist_lock to protect against reaping that
 637          * clears p->sighand.  If p has just been reaped, we can no
 638          * longer get any information about it at all.
 639          */
 640         if (unlikely(p->sighand == NULL)) {
 641                 read_unlock(&tasklist_lock);
 642                 put_task_struct(p);
 643                 timer->it.cpu.task = NULL;
 644                 return -ESRCH;
 645         }
 646
 647         /*
 648          * Disarm any old timer after extracting its expiry time.
 649          */
 650         BUG_ON(!irqs_disabled());
 651
 652         ret = 0;
 653         old_incr = timer->it.cpu.incr;
 654         spin_lock(&p->sighand->siglock);
 655         old_expires = timer->it.cpu.expires;
 656         if (unlikely(timer->it.cpu.firing)) {
 657                 timer->it.cpu.firing = -1;
 658                 ret = TIMER_RETRY;
 659         } else
 660                 list_del_init(&timer->it.cpu.entry);
 661
 662         /*
 663          * We need to sample the current value to convert the new
 664          * value from to relative and absolute, and to convert the
 665          * old value from absolute to relative.  To set a process
 666          * timer, we need a sample to balance the thread expiry
 667          * times (in arm_timer).  With an absolute time, we must
 668          * check if it's already passed.  In short, we need a sample.
 669          */
 670         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 671                 cpu_clock_sample(timer->it_clock, p, &val);
 672         } else {
 673                 cpu_timer_sample_group(timer->it_clock, p, &val);
 674         }
 675
 676         if (old) {
 677                 if (old_expires == 0) {
 678                         old->it_value.tv_sec = 0;
 679                         old->it_value.tv_nsec = 0;
 680                 } else {
 681                         /*
 682                          * Update the timer in case it has
 683                          * overrun already.  If it has,
 684                          * we'll report it as having overrun
 685                          * and with the next reloaded timer
 686                          * already ticking, though we are
 687                          * swallowing that pending
 688                          * notification here to install the
 689                          * new setting.
 690                          */
 691                         bump_cpu_timer(timer, val);
 692                         if (val < timer->it.cpu.expires) {
 693                                 old_expires = timer->it.cpu.expires - val;
 694                                 sample_to_timespec(timer->it_clock,
 695                                                    old_expires,
 696                                                    &old->it_value);
 697                         } else {
 698                                 old->it_value.tv_nsec = 1;
 699                                 old->it_value.tv_sec = 0;
 700                         }
 701                 }
 702         }
 703
 704         if (unlikely(ret)) {
 705                 /*
 706                  * We are colliding with the timer actually firing.
 707                  * Punt after filling in the timer's old value, and
 708                  * disable this firing since we are already reporting
 709                  * it as an overrun (thanks to bump_cpu_timer above).
 710                  */
 711                 spin_unlock(&p->sighand->siglock);
 712                 read_unlock(&tasklist_lock);
 713                 goto out;
 714         }
 715
 716         if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
 717                 new_expires += val;
 718         }
 719
 720         /*
 721          * Install the new expiry time (or zero).
 722          * For a timer with no notification action, we don't actually
 723          * arm the timer (we'll just fake it for timer_gettime).
 724          */
 725         timer->it.cpu.expires = new_expires;
 726         if (new_expires != 0 && val < new_expires) {
 727                 arm_timer(timer);
 728         }
 729
 730         spin_unlock(&p->sighand->siglock);
 731         read_unlock(&tasklist_lock);
 732
 733         /*
 734          * Install the new reload setting, and
 735          * set up the signal and overrun bookkeeping.
 736          */
 737         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 738                                                 &new->it_interval);
 739
 740         /*
 741          * This acts as a modification timestamp for the timer,
 742          * so any automatic reload attempt will punt on seeing
 743          * that we have reset the timer manually.
 744          */
 745         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 746                 ~REQUEUE_PENDING;
 747         timer->it_overrun_last = 0;
 748         timer->it_overrun = -1;
 749
 750         if (new_expires != 0 && !(val < new_expires)) {
 751                 /*
 752                  * The designated time already passed, so we notify
 753                  * immediately, even if the thread never runs to
 754                  * accumulate more time on this clock.
 755                  */
 756                 cpu_timer_fire(timer);
 757         }
 758
 759         ret = 0;
 760  out:
 761         if (old) {
 762                 sample_to_timespec(timer->it_clock,
 763                                    old_incr, &old->it_interval);
 764         }
 765         if (!ret)
 766                 posix_cpu_timer_kick_nohz();
 767         return ret;
 768 }
 769
 770 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 771 {
 772         unsigned long long now;
 773         struct task_struct *p = timer->it.cpu.task;
 774
 775         /*
 776          * Easy part: convert the reload time.
 777          */
 778         sample_to_timespec(timer->it_clock,
 779                            timer->it.cpu.incr, &itp->it_interval);
 780
 781         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 782                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 783                 return;
 784         }
 785
 786         if (unlikely(p == NULL)) {
 787                 WARN_ON_ONCE(CPUCLOCK_PERTHREAD(timer->it_clock));
 788                 /*
 789                  * This task already died and the timer will never fire.
 790                  * In this case, expires is actually the dead value.
 791                  */
 792         dead:
 793                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 794                                    &itp->it_value);
 795                 return;
 796         }
 797
 798         /*
 799          * Sample the clock to take the difference with the expiry time.
 800          */
 801         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 802                 cpu_clock_sample(timer->it_clock, p, &now);
 803         } else {
 804                 read_lock(&tasklist_lock);
 805                 if (unlikely(p->sighand == NULL)) {
 806                         /*
 807                          * The process has been reaped.
 808                          * We can't even collect a sample any more.
 809                          * Call the timer disarmed, nothing else to do.
 810                          */
 811                         put_task_struct(p);
 812                         timer->it.cpu.task = NULL;
 813                         timer->it.cpu.expires = 0;
 814                         read_unlock(&tasklist_lock);
 815                         goto dead;
 816                 } else {
 817                         cpu_timer_sample_group(timer->it_clock, p, &now);
 818                 }
 819                 read_unlock(&tasklist_lock);
 820         }
 821
 822         if (now < timer->it.cpu.expires) {
 823                 sample_to_timespec(timer->it_clock,
 824                                    timer->it.cpu.expires - now,
 825                                    &itp->it_value);
 826         } else {
 827                 /*
 828                  * The timer should have expired already, but the firing
 829                  * hasn't taken place yet.  Say it's just about to expire.
 830                  */
 831                 itp->it_value.tv_nsec = 1;
 832                 itp->it_value.tv_sec = 0;
 833         }
 834 }
 835
 836 static unsigned long long
 837 check_timers_list(struct list_head *timers,
 838                   struct list_head *firing,
 839                   unsigned long long curr)
 840 {
 841         int maxfire = 20;
 842
 843         while (!list_empty(timers)) {
 844                 struct cpu_timer_list *t;
 845
 846                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 847
 848                 if (!--maxfire || curr < t->expires)
 849                         return t->expires;
 850
 851                 t->firing = 1;
 852                 list_move_tail(&t->entry, firing);
 853         }
 854
 855         return 0;
 856 }
 857
 858 /*
 859  * Check for any per-thread CPU timers that have fired and move them off
 860  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 861  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 862  */
 863 static void check_thread_timers(struct task_struct *tsk,
 864                                 struct list_head *firing)
 865 {
 866         struct list_head *timers = tsk->cpu_timers;
 867         struct signal_struct *const sig = tsk->signal;
 868         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 869         unsigned long long expires;
 870         unsigned long soft;
 871
 872         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 873         tsk_expires->prof_exp = expires_to_cputime(expires);
 874
 875         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 876         tsk_expires->virt_exp = expires_to_cputime(expires);
 877
 878         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 879                                                    tsk->se.sum_exec_runtime);
 880
 881         /*
 882          * Check for the special case thread timers.
 883          */
 884         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 885         if (soft != RLIM_INFINITY) {
 886                 unsigned long hard =
 887                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 888
 889                 if (hard != RLIM_INFINITY &&
 890                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 891                         /*
 892                          * At the hard limit, we just die.
 893                          * No need to calculate anything else now.
 894                          */
 895                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 896                         return;
 897                 }
 898                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 899                         /*
 900                          * At the soft limit, send a SIGXCPU every second.
 901                          */
 902                         if (soft < hard) {
 903                                 soft += USEC_PER_SEC;
 904                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 905                         }
 906                         printk(KERN_INFO
 907                                 "RT Watchdog Timeout: %s[%d]\n",
 908                                 tsk->comm, task_pid_nr(tsk));
 909                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 910                 }
 911         }
 912 }
 913
 914 static void stop_process_timers(struct signal_struct *sig)
 915 {
 916         struct thread_group_cputimer *cputimer = &sig->cputimer;
 917         unsigned long flags;
 918
 919         raw_spin_lock_irqsave(&cputimer->lock, flags);
 920         cputimer->running = 0;
 921         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 922 }
 923
 924 static u32 onecputick;
 925
 926 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 927                              unsigned long long *expires,
 928                              unsigned long long cur_time, int signo)
 929 {
 930         if (!it->expires)
 931                 return;
 932
 933         if (cur_time >= it->expires) {
 934                 if (it->incr) {
 935                         it->expires += it->incr;
 936                         it->error += it->incr_error;
 937                         if (it->error >= onecputick) {
 938                                 it->expires -= cputime_one_jiffy;
 939                                 it->error -= onecputick;
 940                         }
 941                 } else {
 942                         it->expires = 0;
 943                 }
 944
 945                 trace_itimer_expire(signo == SIGPROF ?
 946                                     ITIMER_PROF : ITIMER_VIRTUAL,
 947                                     tsk->signal->leader_pid, cur_time);
 948                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 949         }
 950
 951         if (it->expires && (!*expires || it->expires < *expires)) {
 952                 *expires = it->expires;
 953         }
 954 }
 955
 956 /*
 957  * Check for any per-thread CPU timers that have fired and move them
 958  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 959  * have already been taken off.
 960  */
 961 static void check_process_timers(struct task_struct *tsk,
 962                                  struct list_head *firing)
 963 {
 964         struct signal_struct *const sig = tsk->signal;
 965         unsigned long long utime, ptime, virt_expires, prof_expires;
 966         unsigned long long sum_sched_runtime, sched_expires;
 967         struct list_head *timers = sig->cpu_timers;
 968         struct task_cputime cputime;
 969         unsigned long soft;
 970
 971         /*
 972          * Collect the current process totals.
 973          */
 974         thread_group_cputimer(tsk, &cputime);
 975         utime = cputime_to_expires(cputime.utime);
 976         ptime = utime + cputime_to_expires(cputime.stime);
 977         sum_sched_runtime = cputime.sum_exec_runtime;
 978
 979         prof_expires = check_timers_list(timers, firing, ptime);
 980         virt_expires = check_timers_list(++timers, firing, utime);
 981         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 982
 983         /*
 984          * Check for the special case process timers.
 985          */
 986         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 987                          SIGPROF);
 988         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 989                          SIGVTALRM);
 990         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 991         if (soft != RLIM_INFINITY) {
 992                 unsigned long psecs = cputime_to_secs(ptime);
 993                 unsigned long hard =
 994                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 995                 cputime_t x;
 996                 if (psecs >= hard) {
 997                         /*
 998                          * At the hard limit, we just die.
 999                          * No need to calculate anything else now.
1000                          */
1001                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1002                         return;
1003                 }
1004                 if (psecs >= soft) {
1005                         /*
1006                          * At the soft limit, send a SIGXCPU every second.
1007                          */
1008                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1009                         if (soft < hard) {
1010                                 soft++;
1011                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1012                         }
1013                 }
1014                 x = secs_to_cputime(soft);
1015                 if (!prof_expires || x < prof_expires) {
1016                         prof_expires = x;
1017                 }
1018         }
1019
1020         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1021         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1022         sig->cputime_expires.sched_exp = sched_expires;
1023         if (task_cputime_zero(&sig->cputime_expires))
1024                 stop_process_timers(sig);
1025 }
1026
1027 /*
1028  * This is called from the signal code (via do_schedule_next_timer)
1029  * when the last timer signal was delivered and we have to reload the timer.
1030  */
1031 void posix_cpu_timer_schedule(struct k_itimer *timer)
1032 {
1033         struct task_struct *p = timer->it.cpu.task;
1034         unsigned long long now;
1035
1036         if (unlikely(p == NULL)) {
1037                 WARN_ON_ONCE(CPUCLOCK_PERTHREAD(timer->it_clock));
1038                 /*
1039                  * The task was cleaned up already, no future firings.
1040                  */
1041                 goto out;
1042         }
1043
1044         /*
1045          * Fetch the current sample and update the timer's expiry time.
1046          */
1047         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1048                 cpu_clock_sample(timer->it_clock, p, &now);
1049                 bump_cpu_timer(timer, now);
1050                 if (unlikely(p->exit_state))
1051                         goto out;
1052
1053                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1054                 spin_lock(&p->sighand->siglock);
1055         } else {
1056                 read_lock(&tasklist_lock);
1057                 if (unlikely(p->sighand == NULL)) {
1058                         /*
1059                          * The process has been reaped.
1060                          * We can't even collect a sample any more.
1061                          */
1062                         put_task_struct(p);
1063                         timer->it.cpu.task = p = NULL;
1064                         timer->it.cpu.expires = 0;
1065                         read_unlock(&tasklist_lock);
1066                         goto out;
1067                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1068                         read_unlock(&tasklist_lock);
1069                         /* Optimizations: if the process is dying, no need to rearm */
1070                         goto out;
1071                 }
1072                 spin_lock(&p->sighand->siglock);
1073                 cpu_timer_sample_group(timer->it_clock, p, &now);
1074                 bump_cpu_timer(timer, now);
1075                 /* Leave the tasklist_lock locked for the call below.  */
1076         }
1077
1078         /*
1079          * Now re-arm for the new expiry time.
1080          */
1081         BUG_ON(!irqs_disabled());
1082         arm_timer(timer);
1083         spin_unlock(&p->sighand->siglock);
1084         read_unlock(&tasklist_lock);
1085
1086         /* Kick full dynticks CPUs in case they need to tick on the new timer */
1087         posix_cpu_timer_kick_nohz();
1088
1089 out:
1090         timer->it_overrun_last = timer->it_overrun;
1091         timer->it_overrun = -1;
1092         ++timer->it_requeue_pending;
1093 }
1094
1095 /**
1096  * task_cputime_expired - Compare two task_cputime entities.
1097  *
1098  * @sample:     The task_cputime structure to be checked for expiration.
1099  * @expires:    Expiration times, against which @sample will be checked.
1100  *
1101  * Checks @sample against @expires to see if any field of @sample has expired.
1102  * Returns true if any field of the former is greater than the corresponding
1103  * field of the latter if the latter field is set.  Otherwise returns false.
1104  */
1105 static inline int task_cputime_expired(const struct task_cputime *sample,
1106                                         const struct task_cputime *expires)
1107 {
1108         if (expires->utime && sample->utime >= expires->utime)
1109                 return 1;
1110         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1111                 return 1;
1112         if (expires->sum_exec_runtime != 0 &&
1113             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1114                 return 1;
1115         return 0;
1116 }
1117
1118 /**
1119  * fastpath_timer_check - POSIX CPU timers fast path.
1120  *
1121  * @tsk:        The task (thread) being checked.
1122  *
1123  * Check the task and thread group timers.  If both are zero (there are no
1124  * timers set) return false.  Otherwise snapshot the task and thread group
1125  * timers and compare them with the corresponding expiration times.  Return
1126  * true if a timer has expired, else return false.
1127  */
1128 static inline int fastpath_timer_check(struct task_struct *tsk)
1129 {
1130         struct signal_struct *sig;
1131         cputime_t utime, stime;
1132
1133         task_cputime(tsk, &utime, &stime);
1134
1135         if (!task_cputime_zero(&tsk->cputime_expires)) {
1136                 struct task_cputime task_sample = {
1137                         .utime = utime,
1138                         .stime = stime,
1139                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1140                 };
1141
1142                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1143                         return 1;
1144         }
1145
1146         sig = tsk->signal;
1147         if (sig->cputimer.running) {
1148                 struct task_cputime group_sample;
1149
1150                 raw_spin_lock(&sig->cputimer.lock);
1151                 group_sample = sig->cputimer.cputime;
1152                 raw_spin_unlock(&sig->cputimer.lock);
1153
1154                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1155                         return 1;
1156         }
1157
1158         return 0;
1159 }
1160
1161 /*
1162  * This is called from the timer interrupt handler.  The irq handler has
1163  * already updated our counts.  We need to check if any timers fire now.
1164  * Interrupts are disabled.
1165  */
1166 void run_posix_cpu_timers(struct task_struct *tsk)
1167 {
1168         LIST_HEAD(firing);
1169         struct k_itimer *timer, *next;
1170         unsigned long flags;
1171
1172         BUG_ON(!irqs_disabled());
1173
1174         /*
1175          * The fast path checks that there are no expired thread or thread
1176          * group timers.  If that's so, just return.
1177          */
1178         if (!fastpath_timer_check(tsk))
1179                 return;
1180
1181         if (!lock_task_sighand(tsk, &flags))
1182                 return;
1183         /*
1184          * Here we take off tsk->signal->cpu_timers[N] and
1185          * tsk->cpu_timers[N] all the timers that are firing, and
1186          * put them on the firing list.
1187          */
1188         check_thread_timers(tsk, &firing);
1189         /*
1190          * If there are any active process wide timers (POSIX 1.b, itimers,
1191          * RLIMIT_CPU) cputimer must be running.
1192          */
1193         if (tsk->signal->cputimer.running)
1194                 check_process_timers(tsk, &firing);
1195
1196         /*
1197          * We must release these locks before taking any timer's lock.
1198          * There is a potential race with timer deletion here, as the
1199          * siglock now protects our private firing list.  We have set
1200          * the firing flag in each timer, so that a deletion attempt
1201          * that gets the timer lock before we do will give it up and
1202          * spin until we've taken care of that timer below.
1203          */
1204         unlock_task_sighand(tsk, &flags);
1205
1206         /*
1207          * Now that all the timers on our list have the firing flag,
1208          * no one will touch their list entries but us.  We'll take
1209          * each timer's lock before clearing its firing flag, so no
1210          * timer call will interfere.
1211          */
1212         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1213                 int cpu_firing;
1214
1215                 spin_lock(&timer->it_lock);
1216                 list_del_init(&timer->it.cpu.entry);
1217                 cpu_firing = timer->it.cpu.firing;
1218                 timer->it.cpu.firing = 0;
1219                 /*
1220                  * The firing flag is -1 if we collided with a reset
1221                  * of the timer, which already reported this
1222                  * almost-firing as an overrun.  So don't generate an event.
1223                  */
1224                 if (likely(cpu_firing >= 0))
1225                         cpu_timer_fire(timer);
1226                 spin_unlock(&timer->it_lock);
1227         }
1228 }
1229
1230 /*
1231  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1232  * The tsk->sighand->siglock must be held by the caller.
1233  */
1234 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1235                            cputime_t *newval, cputime_t *oldval)
1236 {
1237         unsigned long long now;
1238
1239         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1240         cpu_timer_sample_group(clock_idx, tsk, &now);
1241
1242         if (oldval) {
1243                 /*
1244                  * We are setting itimer. The *oldval is absolute and we update
1245                  * it to be relative, *newval argument is relative and we update
1246                  * it to be absolute.
1247                  */
1248                 if (*oldval) {
1249                         if (*oldval <= now) {
1250                                 /* Just about to fire. */
1251                                 *oldval = cputime_one_jiffy;
1252                         } else {
1253                                 *oldval -= now;
1254                         }
1255                 }
1256
1257                 if (!*newval)
1258                         goto out;
1259                 *newval += now;
1260         }
1261
1262         /*
1263          * Update expiration cache if we are the earliest timer, or eventually
1264          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1265          */
1266         switch (clock_idx) {
1267         case CPUCLOCK_PROF:
1268                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1269                         tsk->signal->cputime_expires.prof_exp = *newval;
1270                 break;
1271         case CPUCLOCK_VIRT:
1272                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1273                         tsk->signal->cputime_expires.virt_exp = *newval;
1274                 break;
1275         }
1276 out:
1277         posix_cpu_timer_kick_nohz();
1278 }
1279
1280 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1281                             struct timespec *rqtp, struct itimerspec *it)
1282 {
1283         struct k_itimer timer;
1284         int error;
1285
1286         /*
1287          * Set up a temporary timer and then wait for it to go off.
1288          */
1289         memset(&timer, 0, sizeof timer);
1290         spin_lock_init(&timer.it_lock);
1291         timer.it_clock = which_clock;
1292         timer.it_overrun = -1;
1293         error = posix_cpu_timer_create(&timer);
1294         timer.it_process = current;
1295         if (!error) {
1296                 static struct itimerspec zero_it;
1297
1298                 memset(it, 0, sizeof *it);
1299                 it->it_value = *rqtp;
1300
1301                 spin_lock_irq(&timer.it_lock);
1302                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1303                 if (error) {
1304                         spin_unlock_irq(&timer.it_lock);
1305                         return error;
1306                 }
1307
1308                 while (!signal_pending(current)) {
1309                         if (timer.it.cpu.expires == 0) {
1310                                 /*
1311                                  * Our timer fired and was reset, below
1312                                  * deletion can not fail.
1313                                  */
1314                                 posix_cpu_timer_del(&timer);
1315                                 spin_unlock_irq(&timer.it_lock);
1316                                 return 0;
1317                         }
1318
1319                         /*
1320                          * Block until cpu_timer_fire (or a signal) wakes us.
1321                          */
1322                         __set_current_state(TASK_INTERRUPTIBLE);
1323                         spin_unlock_irq(&timer.it_lock);
1324                         schedule();
1325                         spin_lock_irq(&timer.it_lock);
1326                 }
1327
1328                 /*
1329                  * We were interrupted by a signal.
1330                  */
1331                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1332                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1333                 if (!error) {
1334                         /*
1335                          * Timer is now unarmed, deletion can not fail.
1336                          */
1337                         posix_cpu_timer_del(&timer);
1338                 }
1339                 spin_unlock_irq(&timer.it_lock);
1340
1341                 while (error == TIMER_RETRY) {
1342                         /*
1343                          * We need to handle case when timer was or is in the
1344                          * middle of firing. In other cases we already freed
1345                          * resources.
1346                          */
1347                         spin_lock_irq(&timer.it_lock);
1348                         error = posix_cpu_timer_del(&timer);
1349                         spin_unlock_irq(&timer.it_lock);
1350                 }
1351
1352                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1353                         /*
1354                          * It actually did fire already.
1355                          */
1356                         return 0;
1357                 }
1358
1359                 error = -ERESTART_RESTARTBLOCK;
1360         }
1361
1362         return error;
1363 }
1364
1365 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1366
1367 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1368                             struct timespec *rqtp, struct timespec __user *rmtp)
1369 {
1370         struct restart_block *restart_block =
1371                 &current_thread_info()->restart_block;
1372         struct itimerspec it;
1373         int error;
1374
1375         /*
1376          * Diagnose required errors first.
1377          */
1378         if (CPUCLOCK_PERTHREAD(which_clock) &&
1379             (CPUCLOCK_PID(which_clock) == 0 ||
1380              CPUCLOCK_PID(which_clock) == current->pid))
1381                 return -EINVAL;
1382
1383         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1384
1385         if (error == -ERESTART_RESTARTBLOCK) {
1386
1387                 if (flags & TIMER_ABSTIME)
1388                         return -ERESTARTNOHAND;
1389                 /*
1390                  * Report back to the user the time still remaining.
1391                  */
1392                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1393                         return -EFAULT;
1394
1395                 restart_block->fn = posix_cpu_nsleep_restart;
1396                 restart_block->nanosleep.clockid = which_clock;
1397                 restart_block->nanosleep.rmtp = rmtp;
1398                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1399         }
1400         return error;
1401 }
1402
1403 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1404 {
1405         clockid_t which_clock = restart_block->nanosleep.clockid;
1406         struct timespec t;
1407         struct itimerspec it;
1408         int error;
1409
1410         t = ns_to_timespec(restart_block->nanosleep.expires);
1411
1412         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1413
1414         if (error == -ERESTART_RESTARTBLOCK) {
1415                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1416                 /*
1417                  * Report back to the user the time still remaining.
1418                  */
1419                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1420                         return -EFAULT;
1421
1422                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1423         }
1424         return error;
1425
1426 }
1427
1428 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1429 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1430
1431 static int process_cpu_clock_getres(const clockid_t which_clock,
1432                                     struct timespec *tp)
1433 {
1434         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1435 }
1436 static int process_cpu_clock_get(const clockid_t which_clock,
1437                                  struct timespec *tp)
1438 {
1439         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1440 }
1441 static int process_cpu_timer_create(struct k_itimer *timer)
1442 {
1443         timer->it_clock = PROCESS_CLOCK;
1444         return posix_cpu_timer_create(timer);
1445 }
1446 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1447                               struct timespec *rqtp,
1448                               struct timespec __user *rmtp)
1449 {
1450         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1451 }
1452 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1453 {
1454         return -EINVAL;
1455 }
1456 static int thread_cpu_clock_getres(const clockid_t which_clock,
1457                                    struct timespec *tp)
1458 {
1459         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1460 }
1461 static int thread_cpu_clock_get(const clockid_t which_clock,
1462                                 struct timespec *tp)
1463 {
1464         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1465 }
1466 static int thread_cpu_timer_create(struct k_itimer *timer)
1467 {
1468         timer->it_clock = THREAD_CLOCK;
1469         return posix_cpu_timer_create(timer);
1470 }
1471
1472 struct k_clock clock_posix_cpu = {
1473         .clock_getres   = posix_cpu_clock_getres,
1474         .clock_set      = posix_cpu_clock_set,
1475         .clock_get      = posix_cpu_clock_get,
1476         .timer_create   = posix_cpu_timer_create,
1477         .nsleep         = posix_cpu_nsleep,
1478         .nsleep_restart = posix_cpu_nsleep_restart,
1479         .timer_set      = posix_cpu_timer_set,
1480         .timer_del      = posix_cpu_timer_del,
1481         .timer_get      = posix_cpu_timer_get,
1482 };
1483
1484 static __init int init_posix_cpu_timers(void)
1485 {
1486         struct k_clock process = {
1487                 .clock_getres   = process_cpu_clock_getres,
1488                 .clock_get      = process_cpu_clock_get,
1489                 .timer_create   = process_cpu_timer_create,
1490                 .nsleep         = process_cpu_nsleep,
1491                 .nsleep_restart = process_cpu_nsleep_restart,
1492         };
1493         struct k_clock thread = {
1494                 .clock_getres   = thread_cpu_clock_getres,
1495                 .clock_get      = thread_cpu_clock_get,
1496                 .timer_create   = thread_cpu_timer_create,
1497         };
1498         struct timespec ts;
1499
1500         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1501         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1502
1503         cputime_to_timespec(cputime_one_jiffy, &ts);
1504         onecputick = ts.tv_nsec;
1505         WARN_ON(ts.tv_sec != 0);
1506
1507         return 0;
1508 }
1509 __initcall(init_posix_cpu_timers);