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                 if (timer->expires < curr) {
 410                         timer->expires = 0;
 411                 } else {
 412                         timer->expires -= curr;
 413                 }
 414         }
 415 }
 416
 417 /*
 418  * Clean out CPU timers still ticking when a thread exited.  The task
 419  * pointer is cleared, and the expiry time is replaced with the residual
 420  * time for later timer_gettime calls to return.
 421  * This must be called with the siglock held.
 422  */
 423 static void cleanup_timers(struct list_head *head,
 424                            cputime_t utime, cputime_t stime,
 425                            unsigned long long sum_exec_runtime)
 426 {
 427
 428         cputime_t ptime = utime + stime;
 429
 430         cleanup_timers_list(head, cputime_to_expires(ptime));
 431         cleanup_timers_list(++head, cputime_to_expires(utime));
 432         cleanup_timers_list(++head, sum_exec_runtime);
 433 }
 434
 435 /*
 436  * These are both called with the siglock held, when the current thread
 437  * is being reaped.  When the final (leader) thread in the group is reaped,
 438  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 439  */
 440 void posix_cpu_timers_exit(struct task_struct *tsk)
 441 {
 442         cputime_t utime, stime;
 443
 444         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 445                                                 sizeof(unsigned long long));
 446         task_cputime(tsk, &utime, &stime);
 447         cleanup_timers(tsk->cpu_timers,
 448                        utime, stime, tsk->se.sum_exec_runtime);
 449
 450 }
 451 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 452 {
 453         struct signal_struct *const sig = tsk->signal;
 454         cputime_t utime, stime;
 455
 456         task_cputime(tsk, &utime, &stime);
 457         cleanup_timers(tsk->signal->cpu_timers,
 458                        utime + sig->utime, stime + sig->stime,
 459                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 460 }
 461
 462 static void clear_dead_task(struct k_itimer *timer, unsigned long long now)
 463 {
 464         /*
 465          * That's all for this thread or process.
 466          * We leave our residual in expires to be reported.
 467          */
 468         put_task_struct(timer->it.cpu.task);
 469         timer->it.cpu.task = NULL;
 470         timer->it.cpu.expires -= now;
 471 }
 472
 473 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 474 {
 475         return expires == 0 || expires > new_exp;
 476 }
 477
 478 /*
 479  * Insert the timer on the appropriate list before any timers that
 480  * expire later.  This must be called with the tasklist_lock held
 481  * for reading, interrupts disabled and p->sighand->siglock taken.
 482  */
 483 static void arm_timer(struct k_itimer *timer)
 484 {
 485         struct task_struct *p = timer->it.cpu.task;
 486         struct list_head *head, *listpos;
 487         struct task_cputime *cputime_expires;
 488         struct cpu_timer_list *const nt = &timer->it.cpu;
 489         struct cpu_timer_list *next;
 490
 491         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 492                 head = p->cpu_timers;
 493                 cputime_expires = &p->cputime_expires;
 494         } else {
 495                 head = p->signal->cpu_timers;
 496                 cputime_expires = &p->signal->cputime_expires;
 497         }
 498         head += CPUCLOCK_WHICH(timer->it_clock);
 499
 500         listpos = head;
 501         list_for_each_entry(next, head, entry) {
 502                 if (nt->expires < next->expires)
 503                         break;
 504                 listpos = &next->entry;
 505         }
 506         list_add(&nt->entry, listpos);
 507
 508         if (listpos == head) {
 509                 unsigned long long exp = nt->expires;
 510
 511                 /*
 512                  * We are the new earliest-expiring POSIX 1.b timer, hence
 513                  * need to update expiration cache. Take into account that
 514                  * for process timers we share expiration cache with itimers
 515                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 516                  */
 517
 518                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 519                 case CPUCLOCK_PROF:
 520                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 521                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 522                         break;
 523                 case CPUCLOCK_VIRT:
 524                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 525                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 526                         break;
 527                 case CPUCLOCK_SCHED:
 528                         if (cputime_expires->sched_exp == 0 ||
 529                             cputime_expires->sched_exp > exp)
 530                                 cputime_expires->sched_exp = exp;
 531                         break;
 532                 }
 533         }
 534 }
 535
 536 /*
 537  * The timer is locked, fire it and arrange for its reload.
 538  */
 539 static void cpu_timer_fire(struct k_itimer *timer)
 540 {
 541         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 542                 /*
 543                  * User don't want any signal.
 544                  */
 545                 timer->it.cpu.expires = 0;
 546         } else if (unlikely(timer->sigq == NULL)) {
 547                 /*
 548                  * This a special case for clock_nanosleep,
 549                  * not a normal timer from sys_timer_create.
 550                  */
 551                 wake_up_process(timer->it_process);
 552                 timer->it.cpu.expires = 0;
 553         } else if (timer->it.cpu.incr == 0) {
 554                 /*
 555                  * One-shot timer.  Clear it as soon as it's fired.
 556                  */
 557                 posix_timer_event(timer, 0);
 558                 timer->it.cpu.expires = 0;
 559         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 560                 /*
 561                  * The signal did not get queued because the signal
 562                  * was ignored, so we won't get any callback to
 563                  * reload the timer.  But we need to keep it
 564                  * ticking in case the signal is deliverable next time.
 565                  */
 566                 posix_cpu_timer_schedule(timer);
 567         }
 568 }
 569
 570 /*
 571  * Sample a process (thread group) timer for the given group_leader task.
 572  * Must be called with tasklist_lock held for reading.
 573  */
 574 static int cpu_timer_sample_group(const clockid_t which_clock,
 575                                   struct task_struct *p,
 576                                   unsigned long long *sample)
 577 {
 578         struct task_cputime cputime;
 579
 580         thread_group_cputimer(p, &cputime);
 581         switch (CPUCLOCK_WHICH(which_clock)) {
 582         default:
 583                 return -EINVAL;
 584         case CPUCLOCK_PROF:
 585                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 586                 break;
 587         case CPUCLOCK_VIRT:
 588                 *sample = cputime_to_expires(cputime.utime);
 589                 break;
 590         case CPUCLOCK_SCHED:
 591                 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
 592                 break;
 593         }
 594         return 0;
 595 }
 596
 597 #ifdef CONFIG_NO_HZ_FULL
 598 static void nohz_kick_work_fn(struct work_struct *work)
 599 {
 600         tick_nohz_full_kick_all();
 601 }
 602
 603 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
 604
 605 /*
 606  * We need the IPIs to be sent from sane process context.
 607  * The posix cpu timers are always set with irqs disabled.
 608  */
 609 static void posix_cpu_timer_kick_nohz(void)
 610 {
 611         schedule_work(&nohz_kick_work);
 612 }
 613
 614 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
 615 {
 616         if (!task_cputime_zero(&tsk->cputime_expires))
 617                 return false;
 618
 619         if (tsk->signal->cputimer.running)
 620                 return false;
 621
 622         return true;
 623 }
 624 #else
 625 static inline void posix_cpu_timer_kick_nohz(void) { }
 626 #endif
 627
 628 /*
 629  * Guts of sys_timer_settime for CPU timers.
 630  * This is called with the timer locked and interrupts disabled.
 631  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 632  * and try again.  (This happens when the timer is in the middle of firing.)
 633  */
 634 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 635                                struct itimerspec *new, struct itimerspec *old)
 636 {
 637         struct task_struct *p = timer->it.cpu.task;
 638         unsigned long long old_expires, new_expires, old_incr, val;
 639         int ret;
 640
 641         if (unlikely(p == NULL)) {
 642                 /*
 643                  * Timer refers to a dead task's clock.
 644                  */
 645                 return -ESRCH;
 646         }
 647
 648         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 649
 650         read_lock(&tasklist_lock);
 651         /*
 652          * We need the tasklist_lock to protect against reaping that
 653          * clears p->sighand.  If p has just been reaped, we can no
 654          * longer get any information about it at all.
 655          */
 656         if (unlikely(p->sighand == NULL)) {
 657                 read_unlock(&tasklist_lock);
 658                 put_task_struct(p);
 659                 timer->it.cpu.task = NULL;
 660                 return -ESRCH;
 661         }
 662
 663         /*
 664          * Disarm any old timer after extracting its expiry time.
 665          */
 666         BUG_ON(!irqs_disabled());
 667
 668         ret = 0;
 669         old_incr = timer->it.cpu.incr;
 670         spin_lock(&p->sighand->siglock);
 671         old_expires = timer->it.cpu.expires;
 672         if (unlikely(timer->it.cpu.firing)) {
 673                 timer->it.cpu.firing = -1;
 674                 ret = TIMER_RETRY;
 675         } else
 676                 list_del_init(&timer->it.cpu.entry);
 677
 678         /*
 679          * We need to sample the current value to convert the new
 680          * value from to relative and absolute, and to convert the
 681          * old value from absolute to relative.  To set a process
 682          * timer, we need a sample to balance the thread expiry
 683          * times (in arm_timer).  With an absolute time, we must
 684          * check if it's already passed.  In short, we need a sample.
 685          */
 686         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 687                 cpu_clock_sample(timer->it_clock, p, &val);
 688         } else {
 689                 cpu_timer_sample_group(timer->it_clock, p, &val);
 690         }
 691
 692         if (old) {
 693                 if (old_expires == 0) {
 694                         old->it_value.tv_sec = 0;
 695                         old->it_value.tv_nsec = 0;
 696                 } else {
 697                         /*
 698                          * Update the timer in case it has
 699                          * overrun already.  If it has,
 700                          * we'll report it as having overrun
 701                          * and with the next reloaded timer
 702                          * already ticking, though we are
 703                          * swallowing that pending
 704                          * notification here to install the
 705                          * new setting.
 706                          */
 707                         bump_cpu_timer(timer, val);
 708                         if (val < timer->it.cpu.expires) {
 709                                 old_expires = timer->it.cpu.expires - val;
 710                                 sample_to_timespec(timer->it_clock,
 711                                                    old_expires,
 712                                                    &old->it_value);
 713                         } else {
 714                                 old->it_value.tv_nsec = 1;
 715                                 old->it_value.tv_sec = 0;
 716                         }
 717                 }
 718         }
 719
 720         if (unlikely(ret)) {
 721                 /*
 722                  * We are colliding with the timer actually firing.
 723                  * Punt after filling in the timer's old value, and
 724                  * disable this firing since we are already reporting
 725                  * it as an overrun (thanks to bump_cpu_timer above).
 726                  */
 727                 spin_unlock(&p->sighand->siglock);
 728                 read_unlock(&tasklist_lock);
 729                 goto out;
 730         }
 731
 732         if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
 733                 new_expires += val;
 734         }
 735
 736         /*
 737          * Install the new expiry time (or zero).
 738          * For a timer with no notification action, we don't actually
 739          * arm the timer (we'll just fake it for timer_gettime).
 740          */
 741         timer->it.cpu.expires = new_expires;
 742         if (new_expires != 0 && val < new_expires) {
 743                 arm_timer(timer);
 744         }
 745
 746         spin_unlock(&p->sighand->siglock);
 747         read_unlock(&tasklist_lock);
 748
 749         /*
 750          * Install the new reload setting, and
 751          * set up the signal and overrun bookkeeping.
 752          */
 753         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 754                                                 &new->it_interval);
 755
 756         /*
 757          * This acts as a modification timestamp for the timer,
 758          * so any automatic reload attempt will punt on seeing
 759          * that we have reset the timer manually.
 760          */
 761         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 762                 ~REQUEUE_PENDING;
 763         timer->it_overrun_last = 0;
 764         timer->it_overrun = -1;
 765
 766         if (new_expires != 0 && !(val < new_expires)) {
 767                 /*
 768                  * The designated time already passed, so we notify
 769                  * immediately, even if the thread never runs to
 770                  * accumulate more time on this clock.
 771                  */
 772                 cpu_timer_fire(timer);
 773         }
 774
 775         ret = 0;
 776  out:
 777         if (old) {
 778                 sample_to_timespec(timer->it_clock,
 779                                    old_incr, &old->it_interval);
 780         }
 781         if (!ret)
 782                 posix_cpu_timer_kick_nohz();
 783         return ret;
 784 }
 785
 786 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 787 {
 788         unsigned long long now;
 789         struct task_struct *p = timer->it.cpu.task;
 790         int clear_dead;
 791
 792         /*
 793          * Easy part: convert the reload time.
 794          */
 795         sample_to_timespec(timer->it_clock,
 796                            timer->it.cpu.incr, &itp->it_interval);
 797
 798         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 799                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 800                 return;
 801         }
 802
 803         if (unlikely(p == NULL)) {
 804                 /*
 805                  * This task already died and the timer will never fire.
 806                  * In this case, expires is actually the dead value.
 807                  */
 808         dead:
 809                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 810                                    &itp->it_value);
 811                 return;
 812         }
 813
 814         /*
 815          * Sample the clock to take the difference with the expiry time.
 816          */
 817         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 818                 cpu_clock_sample(timer->it_clock, p, &now);
 819                 clear_dead = p->exit_state;
 820         } else {
 821                 read_lock(&tasklist_lock);
 822                 if (unlikely(p->sighand == NULL)) {
 823                         /*
 824                          * The process has been reaped.
 825                          * We can't even collect a sample any more.
 826                          * Call the timer disarmed, nothing else to do.
 827                          */
 828                         put_task_struct(p);
 829                         timer->it.cpu.task = NULL;
 830                         timer->it.cpu.expires = 0;
 831                         read_unlock(&tasklist_lock);
 832                         goto dead;
 833                 } else {
 834                         cpu_timer_sample_group(timer->it_clock, p, &now);
 835                         clear_dead = (unlikely(p->exit_state) &&
 836                                       thread_group_empty(p));
 837                 }
 838                 read_unlock(&tasklist_lock);
 839         }
 840
 841         if (unlikely(clear_dead)) {
 842                 /*
 843                  * We've noticed that the thread is dead, but
 844                  * not yet reaped.  Take this opportunity to
 845                  * drop our task ref.
 846                  */
 847                 clear_dead_task(timer, now);
 848                 goto dead;
 849         }
 850
 851         if (now < timer->it.cpu.expires) {
 852                 sample_to_timespec(timer->it_clock,
 853                                    timer->it.cpu.expires - now,
 854                                    &itp->it_value);
 855         } else {
 856                 /*
 857                  * The timer should have expired already, but the firing
 858                  * hasn't taken place yet.  Say it's just about to expire.
 859                  */
 860                 itp->it_value.tv_nsec = 1;
 861                 itp->it_value.tv_sec = 0;
 862         }
 863 }
 864
 865 /*
 866  * Check for any per-thread CPU timers that have fired and move them off
 867  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 868  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 869  */
 870 static void check_thread_timers(struct task_struct *tsk,
 871                                 struct list_head *firing)
 872 {
 873         int maxfire;
 874         struct list_head *timers = tsk->cpu_timers;
 875         struct signal_struct *const sig = tsk->signal;
 876         unsigned long soft;
 877
 878         maxfire = 20;
 879         tsk->cputime_expires.prof_exp = 0;
 880         while (!list_empty(timers)) {
 881                 struct cpu_timer_list *t = list_first_entry(timers,
 882                                                       struct cpu_timer_list,
 883                                                       entry);
 884                 if (!--maxfire || prof_ticks(tsk) < t->expires) {
 885                         tsk->cputime_expires.prof_exp = expires_to_cputime(t->expires);
 886                         break;
 887                 }
 888                 t->firing = 1;
 889                 list_move_tail(&t->entry, firing);
 890         }
 891
 892         ++timers;
 893         maxfire = 20;
 894         tsk->cputime_expires.virt_exp = 0;
 895         while (!list_empty(timers)) {
 896                 struct cpu_timer_list *t = list_first_entry(timers,
 897                                                       struct cpu_timer_list,
 898                                                       entry);
 899                 if (!--maxfire || virt_ticks(tsk) < t->expires) {
 900                         tsk->cputime_expires.virt_exp = expires_to_cputime(t->expires);
 901                         break;
 902                 }
 903                 t->firing = 1;
 904                 list_move_tail(&t->entry, firing);
 905         }
 906
 907         ++timers;
 908         maxfire = 20;
 909         tsk->cputime_expires.sched_exp = 0;
 910         while (!list_empty(timers)) {
 911                 struct cpu_timer_list *t = list_first_entry(timers,
 912                                                       struct cpu_timer_list,
 913                                                       entry);
 914                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires) {
 915                         tsk->cputime_expires.sched_exp = t->expires;
 916                         break;
 917                 }
 918                 t->firing = 1;
 919                 list_move_tail(&t->entry, firing);
 920         }
 921
 922         /*
 923          * Check for the special case thread timers.
 924          */
 925         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 926         if (soft != RLIM_INFINITY) {
 927                 unsigned long hard =
 928                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 929
 930                 if (hard != RLIM_INFINITY &&
 931                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 932                         /*
 933                          * At the hard limit, we just die.
 934                          * No need to calculate anything else now.
 935                          */
 936                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 937                         return;
 938                 }
 939                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 940                         /*
 941                          * At the soft limit, send a SIGXCPU every second.
 942                          */
 943                         if (soft < hard) {
 944                                 soft += USEC_PER_SEC;
 945                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 946                         }
 947                         printk(KERN_INFO
 948                                 "RT Watchdog Timeout: %s[%d]\n",
 949                                 tsk->comm, task_pid_nr(tsk));
 950                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 951                 }
 952         }
 953 }
 954
 955 static void stop_process_timers(struct signal_struct *sig)
 956 {
 957         struct thread_group_cputimer *cputimer = &sig->cputimer;
 958         unsigned long flags;
 959
 960         raw_spin_lock_irqsave(&cputimer->lock, flags);
 961         cputimer->running = 0;
 962         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 963 }
 964
 965 static u32 onecputick;
 966
 967 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 968                              unsigned long long *expires,
 969                              unsigned long long cur_time, int signo)
 970 {
 971         if (!it->expires)
 972                 return;
 973
 974         if (cur_time >= it->expires) {
 975                 if (it->incr) {
 976                         it->expires += it->incr;
 977                         it->error += it->incr_error;
 978                         if (it->error >= onecputick) {
 979                                 it->expires -= cputime_one_jiffy;
 980                                 it->error -= onecputick;
 981                         }
 982                 } else {
 983                         it->expires = 0;
 984                 }
 985
 986                 trace_itimer_expire(signo == SIGPROF ?
 987                                     ITIMER_PROF : ITIMER_VIRTUAL,
 988                                     tsk->signal->leader_pid, cur_time);
 989                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 990         }
 991
 992         if (it->expires && (!*expires || it->expires < *expires)) {
 993                 *expires = it->expires;
 994         }
 995 }
 996
 997 /*
 998  * Check for any per-thread CPU timers that have fired and move them
 999  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1000  * have already been taken off.
1001  */
1002 static void check_process_timers(struct task_struct *tsk,
1003                                  struct list_head *firing)
1004 {
1005         int maxfire;
1006         struct signal_struct *const sig = tsk->signal;
1007         unsigned long long utime, ptime, virt_expires, prof_expires;
1008         unsigned long long sum_sched_runtime, sched_expires;
1009         struct list_head *timers = sig->cpu_timers;
1010         struct task_cputime cputime;
1011         unsigned long soft;
1012
1013         /*
1014          * Collect the current process totals.
1015          */
1016         thread_group_cputimer(tsk, &cputime);
1017         utime = cputime_to_expires(cputime.utime);
1018         ptime = utime + cputime_to_expires(cputime.stime);
1019         sum_sched_runtime = cputime.sum_exec_runtime;
1020         maxfire = 20;
1021         prof_expires = 0;
1022         while (!list_empty(timers)) {
1023                 struct cpu_timer_list *tl = list_first_entry(timers,
1024                                                       struct cpu_timer_list,
1025                                                       entry);
1026                 if (!--maxfire || ptime < tl->expires) {
1027                         prof_expires = tl->expires;
1028                         break;
1029                 }
1030                 tl->firing = 1;
1031                 list_move_tail(&tl->entry, firing);
1032         }
1033
1034         ++timers;
1035         maxfire = 20;
1036         virt_expires = 0;
1037         while (!list_empty(timers)) {
1038                 struct cpu_timer_list *tl = list_first_entry(timers,
1039                                                       struct cpu_timer_list,
1040                                                       entry);
1041                 if (!--maxfire || utime < tl->expires) {
1042                         virt_expires = tl->expires;
1043                         break;
1044                 }
1045                 tl->firing = 1;
1046                 list_move_tail(&tl->entry, firing);
1047         }
1048
1049         ++timers;
1050         maxfire = 20;
1051         sched_expires = 0;
1052         while (!list_empty(timers)) {
1053                 struct cpu_timer_list *tl = list_first_entry(timers,
1054                                                       struct cpu_timer_list,
1055                                                       entry);
1056                 if (!--maxfire || sum_sched_runtime < tl->expires) {
1057                         sched_expires = tl->expires;
1058                         break;
1059                 }
1060                 tl->firing = 1;
1061                 list_move_tail(&tl->entry, firing);
1062         }
1063
1064         /*
1065          * Check for the special case process timers.
1066          */
1067         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1068                          SIGPROF);
1069         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1070                          SIGVTALRM);
1071         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1072         if (soft != RLIM_INFINITY) {
1073                 unsigned long psecs = cputime_to_secs(ptime);
1074                 unsigned long hard =
1075                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1076                 cputime_t x;
1077                 if (psecs >= hard) {
1078                         /*
1079                          * At the hard limit, we just die.
1080                          * No need to calculate anything else now.
1081                          */
1082                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1083                         return;
1084                 }
1085                 if (psecs >= soft) {
1086                         /*
1087                          * At the soft limit, send a SIGXCPU every second.
1088                          */
1089                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1090                         if (soft < hard) {
1091                                 soft++;
1092                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1093                         }
1094                 }
1095                 x = secs_to_cputime(soft);
1096                 if (!prof_expires || x < prof_expires) {
1097                         prof_expires = x;
1098                 }
1099         }
1100
1101         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1102         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1103         sig->cputime_expires.sched_exp = sched_expires;
1104         if (task_cputime_zero(&sig->cputime_expires))
1105                 stop_process_timers(sig);
1106 }
1107
1108 /*
1109  * This is called from the signal code (via do_schedule_next_timer)
1110  * when the last timer signal was delivered and we have to reload the timer.
1111  */
1112 void posix_cpu_timer_schedule(struct k_itimer *timer)
1113 {
1114         struct task_struct *p = timer->it.cpu.task;
1115         unsigned long long now;
1116
1117         if (unlikely(p == NULL))
1118                 /*
1119                  * The task was cleaned up already, no future firings.
1120                  */
1121                 goto out;
1122
1123         /*
1124          * Fetch the current sample and update the timer's expiry time.
1125          */
1126         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1127                 cpu_clock_sample(timer->it_clock, p, &now);
1128                 bump_cpu_timer(timer, now);
1129                 if (unlikely(p->exit_state)) {
1130                         clear_dead_task(timer, now);
1131                         goto out;
1132                 }
1133                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1134                 spin_lock(&p->sighand->siglock);
1135         } else {
1136                 read_lock(&tasklist_lock);
1137                 if (unlikely(p->sighand == NULL)) {
1138                         /*
1139                          * The process has been reaped.
1140                          * We can't even collect a sample any more.
1141                          */
1142                         put_task_struct(p);
1143                         timer->it.cpu.task = p = NULL;
1144                         timer->it.cpu.expires = 0;
1145                         goto out_unlock;
1146                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1147                         /*
1148                          * We've noticed that the thread is dead, but
1149                          * not yet reaped.  Take this opportunity to
1150                          * drop our task ref.
1151                          */
1152                         clear_dead_task(timer, now);
1153                         goto out_unlock;
1154                 }
1155                 spin_lock(&p->sighand->siglock);
1156                 cpu_timer_sample_group(timer->it_clock, p, &now);
1157                 bump_cpu_timer(timer, now);
1158                 /* Leave the tasklist_lock locked for the call below.  */
1159         }
1160
1161         /*
1162          * Now re-arm for the new expiry time.
1163          */
1164         BUG_ON(!irqs_disabled());
1165         arm_timer(timer);
1166         spin_unlock(&p->sighand->siglock);
1167
1168 out_unlock:
1169         read_unlock(&tasklist_lock);
1170
1171 out:
1172         timer->it_overrun_last = timer->it_overrun;
1173         timer->it_overrun = -1;
1174         ++timer->it_requeue_pending;
1175 }
1176
1177 /**
1178  * task_cputime_expired - Compare two task_cputime entities.
1179  *
1180  * @sample:     The task_cputime structure to be checked for expiration.
1181  * @expires:    Expiration times, against which @sample will be checked.
1182  *
1183  * Checks @sample against @expires to see if any field of @sample has expired.
1184  * Returns true if any field of the former is greater than the corresponding
1185  * field of the latter if the latter field is set.  Otherwise returns false.
1186  */
1187 static inline int task_cputime_expired(const struct task_cputime *sample,
1188                                         const struct task_cputime *expires)
1189 {
1190         if (expires->utime && sample->utime >= expires->utime)
1191                 return 1;
1192         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1193                 return 1;
1194         if (expires->sum_exec_runtime != 0 &&
1195             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1196                 return 1;
1197         return 0;
1198 }
1199
1200 /**
1201  * fastpath_timer_check - POSIX CPU timers fast path.
1202  *
1203  * @tsk:        The task (thread) being checked.
1204  *
1205  * Check the task and thread group timers.  If both are zero (there are no
1206  * timers set) return false.  Otherwise snapshot the task and thread group
1207  * timers and compare them with the corresponding expiration times.  Return
1208  * true if a timer has expired, else return false.
1209  */
1210 static inline int fastpath_timer_check(struct task_struct *tsk)
1211 {
1212         struct signal_struct *sig;
1213         cputime_t utime, stime;
1214
1215         task_cputime(tsk, &utime, &stime);
1216
1217         if (!task_cputime_zero(&tsk->cputime_expires)) {
1218                 struct task_cputime task_sample = {
1219                         .utime = utime,
1220                         .stime = stime,
1221                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1222                 };
1223
1224                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1225                         return 1;
1226         }
1227
1228         sig = tsk->signal;
1229         if (sig->cputimer.running) {
1230                 struct task_cputime group_sample;
1231
1232                 raw_spin_lock(&sig->cputimer.lock);
1233                 group_sample = sig->cputimer.cputime;
1234                 raw_spin_unlock(&sig->cputimer.lock);
1235
1236                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1237                         return 1;
1238         }
1239
1240         return 0;
1241 }
1242
1243 /*
1244  * This is called from the timer interrupt handler.  The irq handler has
1245  * already updated our counts.  We need to check if any timers fire now.
1246  * Interrupts are disabled.
1247  */
1248 void run_posix_cpu_timers(struct task_struct *tsk)
1249 {
1250         LIST_HEAD(firing);
1251         struct k_itimer *timer, *next;
1252         unsigned long flags;
1253
1254         BUG_ON(!irqs_disabled());
1255
1256         /*
1257          * The fast path checks that there are no expired thread or thread
1258          * group timers.  If that's so, just return.
1259          */
1260         if (!fastpath_timer_check(tsk))
1261                 return;
1262
1263         if (!lock_task_sighand(tsk, &flags))
1264                 return;
1265         /*
1266          * Here we take off tsk->signal->cpu_timers[N] and
1267          * tsk->cpu_timers[N] all the timers that are firing, and
1268          * put them on the firing list.
1269          */
1270         check_thread_timers(tsk, &firing);
1271         /*
1272          * If there are any active process wide timers (POSIX 1.b, itimers,
1273          * RLIMIT_CPU) cputimer must be running.
1274          */
1275         if (tsk->signal->cputimer.running)
1276                 check_process_timers(tsk, &firing);
1277
1278         /*
1279          * We must release these locks before taking any timer's lock.
1280          * There is a potential race with timer deletion here, as the
1281          * siglock now protects our private firing list.  We have set
1282          * the firing flag in each timer, so that a deletion attempt
1283          * that gets the timer lock before we do will give it up and
1284          * spin until we've taken care of that timer below.
1285          */
1286         unlock_task_sighand(tsk, &flags);
1287
1288         /*
1289          * Now that all the timers on our list have the firing flag,
1290          * no one will touch their list entries but us.  We'll take
1291          * each timer's lock before clearing its firing flag, so no
1292          * timer call will interfere.
1293          */
1294         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1295                 int cpu_firing;
1296
1297                 spin_lock(&timer->it_lock);
1298                 list_del_init(&timer->it.cpu.entry);
1299                 cpu_firing = timer->it.cpu.firing;
1300                 timer->it.cpu.firing = 0;
1301                 /*
1302                  * The firing flag is -1 if we collided with a reset
1303                  * of the timer, which already reported this
1304                  * almost-firing as an overrun.  So don't generate an event.
1305                  */
1306                 if (likely(cpu_firing >= 0))
1307                         cpu_timer_fire(timer);
1308                 spin_unlock(&timer->it_lock);
1309         }
1310
1311         /*
1312          * In case some timers were rescheduled after the queue got emptied,
1313          * wake up full dynticks CPUs.
1314          */
1315         if (tsk->signal->cputimer.running)
1316                 posix_cpu_timer_kick_nohz();
1317 }
1318
1319 /*
1320  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1321  * The tsk->sighand->siglock must be held by the caller.
1322  */
1323 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1324                            cputime_t *newval, cputime_t *oldval)
1325 {
1326         unsigned long long now;
1327
1328         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1329         cpu_timer_sample_group(clock_idx, tsk, &now);
1330
1331         if (oldval) {
1332                 /*
1333                  * We are setting itimer. The *oldval is absolute and we update
1334                  * it to be relative, *newval argument is relative and we update
1335                  * it to be absolute.
1336                  */
1337                 if (*oldval) {
1338                         if (*oldval <= now) {
1339                                 /* Just about to fire. */
1340                                 *oldval = cputime_one_jiffy;
1341                         } else {
1342                                 *oldval -= now;
1343                         }
1344                 }
1345
1346                 if (!*newval)
1347                         goto out;
1348                 *newval += now;
1349         }
1350
1351         /*
1352          * Update expiration cache if we are the earliest timer, or eventually
1353          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1354          */
1355         switch (clock_idx) {
1356         case CPUCLOCK_PROF:
1357                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1358                         tsk->signal->cputime_expires.prof_exp = *newval;
1359                 break;
1360         case CPUCLOCK_VIRT:
1361                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1362                         tsk->signal->cputime_expires.virt_exp = *newval;
1363                 break;
1364         }
1365 out:
1366         posix_cpu_timer_kick_nohz();
1367 }
1368
1369 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1370                             struct timespec *rqtp, struct itimerspec *it)
1371 {
1372         struct k_itimer timer;
1373         int error;
1374
1375         /*
1376          * Set up a temporary timer and then wait for it to go off.
1377          */
1378         memset(&timer, 0, sizeof timer);
1379         spin_lock_init(&timer.it_lock);
1380         timer.it_clock = which_clock;
1381         timer.it_overrun = -1;
1382         error = posix_cpu_timer_create(&timer);
1383         timer.it_process = current;
1384         if (!error) {
1385                 static struct itimerspec zero_it;
1386
1387                 memset(it, 0, sizeof *it);
1388                 it->it_value = *rqtp;
1389
1390                 spin_lock_irq(&timer.it_lock);
1391                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1392                 if (error) {
1393                         spin_unlock_irq(&timer.it_lock);
1394                         return error;
1395                 }
1396
1397                 while (!signal_pending(current)) {
1398                         if (timer.it.cpu.expires == 0) {
1399                                 /*
1400                                  * Our timer fired and was reset, below
1401                                  * deletion can not fail.
1402                                  */
1403                                 posix_cpu_timer_del(&timer);
1404                                 spin_unlock_irq(&timer.it_lock);
1405                                 return 0;
1406                         }
1407
1408                         /*
1409                          * Block until cpu_timer_fire (or a signal) wakes us.
1410                          */
1411                         __set_current_state(TASK_INTERRUPTIBLE);
1412                         spin_unlock_irq(&timer.it_lock);
1413                         schedule();
1414                         spin_lock_irq(&timer.it_lock);
1415                 }
1416
1417                 /*
1418                  * We were interrupted by a signal.
1419                  */
1420                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1421                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1422                 if (!error) {
1423                         /*
1424                          * Timer is now unarmed, deletion can not fail.
1425                          */
1426                         posix_cpu_timer_del(&timer);
1427                 }
1428                 spin_unlock_irq(&timer.it_lock);
1429
1430                 while (error == TIMER_RETRY) {
1431                         /*
1432                          * We need to handle case when timer was or is in the
1433                          * middle of firing. In other cases we already freed
1434                          * resources.
1435                          */
1436                         spin_lock_irq(&timer.it_lock);
1437                         error = posix_cpu_timer_del(&timer);
1438                         spin_unlock_irq(&timer.it_lock);
1439                 }
1440
1441                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1442                         /*
1443                          * It actually did fire already.
1444                          */
1445                         return 0;
1446                 }
1447
1448                 error = -ERESTART_RESTARTBLOCK;
1449         }
1450
1451         return error;
1452 }
1453
1454 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1455
1456 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1457                             struct timespec *rqtp, struct timespec __user *rmtp)
1458 {
1459         struct restart_block *restart_block =
1460                 &current_thread_info()->restart_block;
1461         struct itimerspec it;
1462         int error;
1463
1464         /*
1465          * Diagnose required errors first.
1466          */
1467         if (CPUCLOCK_PERTHREAD(which_clock) &&
1468             (CPUCLOCK_PID(which_clock) == 0 ||
1469              CPUCLOCK_PID(which_clock) == current->pid))
1470                 return -EINVAL;
1471
1472         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1473
1474         if (error == -ERESTART_RESTARTBLOCK) {
1475
1476                 if (flags & TIMER_ABSTIME)
1477                         return -ERESTARTNOHAND;
1478                 /*
1479                  * Report back to the user the time still remaining.
1480                  */
1481                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1482                         return -EFAULT;
1483
1484                 restart_block->fn = posix_cpu_nsleep_restart;
1485                 restart_block->nanosleep.clockid = which_clock;
1486                 restart_block->nanosleep.rmtp = rmtp;
1487                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1488         }
1489         return error;
1490 }
1491
1492 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1493 {
1494         clockid_t which_clock = restart_block->nanosleep.clockid;
1495         struct timespec t;
1496         struct itimerspec it;
1497         int error;
1498
1499         t = ns_to_timespec(restart_block->nanosleep.expires);
1500
1501         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1502
1503         if (error == -ERESTART_RESTARTBLOCK) {
1504                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1505                 /*
1506                  * Report back to the user the time still remaining.
1507                  */
1508                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1509                         return -EFAULT;
1510
1511                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1512         }
1513         return error;
1514
1515 }
1516
1517 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1518 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1519
1520 static int process_cpu_clock_getres(const clockid_t which_clock,
1521                                     struct timespec *tp)
1522 {
1523         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1524 }
1525 static int process_cpu_clock_get(const clockid_t which_clock,
1526                                  struct timespec *tp)
1527 {
1528         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1529 }
1530 static int process_cpu_timer_create(struct k_itimer *timer)
1531 {
1532         timer->it_clock = PROCESS_CLOCK;
1533         return posix_cpu_timer_create(timer);
1534 }
1535 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1536                               struct timespec *rqtp,
1537                               struct timespec __user *rmtp)
1538 {
1539         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1540 }
1541 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1542 {
1543         return -EINVAL;
1544 }
1545 static int thread_cpu_clock_getres(const clockid_t which_clock,
1546                                    struct timespec *tp)
1547 {
1548         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1549 }
1550 static int thread_cpu_clock_get(const clockid_t which_clock,
1551                                 struct timespec *tp)
1552 {
1553         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1554 }
1555 static int thread_cpu_timer_create(struct k_itimer *timer)
1556 {
1557         timer->it_clock = THREAD_CLOCK;
1558         return posix_cpu_timer_create(timer);
1559 }
1560
1561 struct k_clock clock_posix_cpu = {
1562         .clock_getres   = posix_cpu_clock_getres,
1563         .clock_set      = posix_cpu_clock_set,
1564         .clock_get      = posix_cpu_clock_get,
1565         .timer_create   = posix_cpu_timer_create,
1566         .nsleep         = posix_cpu_nsleep,
1567         .nsleep_restart = posix_cpu_nsleep_restart,
1568         .timer_set      = posix_cpu_timer_set,
1569         .timer_del      = posix_cpu_timer_del,
1570         .timer_get      = posix_cpu_timer_get,
1571 };
1572
1573 static __init int init_posix_cpu_timers(void)
1574 {
1575         struct k_clock process = {
1576                 .clock_getres   = process_cpu_clock_getres,
1577                 .clock_get      = process_cpu_clock_get,
1578                 .timer_create   = process_cpu_timer_create,
1579                 .nsleep         = process_cpu_nsleep,
1580                 .nsleep_restart = process_cpu_nsleep_restart,
1581         };
1582         struct k_clock thread = {
1583                 .clock_getres   = thread_cpu_clock_getres,
1584                 .clock_get      = thread_cpu_clock_get,
1585                 .timer_create   = thread_cpu_timer_create,
1586         };
1587         struct timespec ts;
1588
1589         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1590         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1591
1592         cputime_to_timespec(cputime_one_jiffy, &ts);
1593         onecputick = ts.tv_nsec;
1594         WARN_ON(ts.tv_sec != 0);
1595
1596         return 0;
1597 }
1598 __initcall(init_posix_cpu_timers);