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