kernel/time/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 <linux/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11 #include <trace/events/timer.h>
  12 #include <linux/tick.h>
  13 #include <linux/workqueue.h>
  14
  15 /*
  16  * Called after updating RLIMIT_CPU to run cpu timer and update
  17  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  18  * siglock protection since other code may update expiration cache as
  19  * well.
  20  */
  21 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  22 {
  23         cputime_t cputime = secs_to_cputime(rlim_new);
  24
  25         spin_lock_irq(&task->sighand->siglock);
  26         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  27         spin_unlock_irq(&task->sighand->siglock);
  28 }
  29
  30 static int check_clock(const clockid_t which_clock)
  31 {
  32         int error = 0;
  33         struct task_struct *p;
  34         const pid_t pid = CPUCLOCK_PID(which_clock);
  35
  36         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  37                 return -EINVAL;
  38
  39         if (pid == 0)
  40                 return 0;
  41
  42         rcu_read_lock();
  43         p = find_task_by_vpid(pid);
  44         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  45                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  46                 error = -EINVAL;
  47         }
  48         rcu_read_unlock();
  49
  50         return error;
  51 }
  52
  53 /*
  54  * Update expiry time from increment, and increase overrun count,
  55  * given the current clock sample.
  56  */
  57 static void bump_cpu_timer(struct k_itimer *timer, u64 now)
  58 {
  59         int i;
  60         u64 delta, incr;
  61
  62         if (timer->it.cpu.incr == 0)
  63                 return;
  64
  65         if (now < timer->it.cpu.expires)
  66                 return;
  67
  68         incr = timer->it.cpu.incr;
  69         delta = now + incr - timer->it.cpu.expires;
  70
  71         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  72         for (i = 0; incr < delta - incr; i++)
  73                 incr = incr << 1;
  74
  75         for (; i >= 0; incr >>= 1, i--) {
  76                 if (delta < incr)
  77                         continue;
  78
  79                 timer->it.cpu.expires += incr;
  80                 timer->it_overrun += 1 << i;
  81                 delta -= incr;
  82         }
  83 }
  84
  85 /**
  86  * task_cputime_zero - Check a task_cputime struct for all zero fields.
  87  *
  88  * @cputime:    The struct to compare.
  89  *
  90  * Checks @cputime to see if all fields are zero.  Returns true if all fields
  91  * are zero, false if any field is nonzero.
  92  */
  93 static inline int task_cputime_zero(const struct task_cputime *cputime)
  94 {
  95         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
  96                 return 1;
  97         return 0;
  98 }
  99
 100 static inline u64 prof_ticks(struct task_struct *p)
 101 {
 102         u64 utime, stime;
 103
 104         task_cputime(p, &utime, &stime);
 105
 106         return utime + stime;
 107 }
 108 static inline u64 virt_ticks(struct task_struct *p)
 109 {
 110         u64 utime, stime;
 111
 112         task_cputime(p, &utime, &stime);
 113
 114         return utime;
 115 }
 116
 117 static int
 118 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 119 {
 120         int error = check_clock(which_clock);
 121         if (!error) {
 122                 tp->tv_sec = 0;
 123                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 124                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 125                         /*
 126                          * If sched_clock is using a cycle counter, we
 127                          * don't have any idea of its true resolution
 128                          * exported, but it is much more than 1s/HZ.
 129                          */
 130                         tp->tv_nsec = 1;
 131                 }
 132         }
 133         return error;
 134 }
 135
 136 static int
 137 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 138 {
 139         /*
 140          * You can never reset a CPU clock, but we check for other errors
 141          * in the call before failing with EPERM.
 142          */
 143         int error = check_clock(which_clock);
 144         if (error == 0) {
 145                 error = -EPERM;
 146         }
 147         return error;
 148 }
 149
 150
 151 /*
 152  * Sample a per-thread clock for the given task.
 153  */
 154 static int cpu_clock_sample(const clockid_t which_clock,
 155                             struct task_struct *p, u64 *sample)
 156 {
 157         switch (CPUCLOCK_WHICH(which_clock)) {
 158         default:
 159                 return -EINVAL;
 160         case CPUCLOCK_PROF:
 161                 *sample = prof_ticks(p);
 162                 break;
 163         case CPUCLOCK_VIRT:
 164                 *sample = virt_ticks(p);
 165                 break;
 166         case CPUCLOCK_SCHED:
 167                 *sample = task_sched_runtime(p);
 168                 break;
 169         }
 170         return 0;
 171 }
 172
 173 /*
 174  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 175  * to avoid race conditions with concurrent updates to cputime.
 176  */
 177 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 178 {
 179         u64 curr_cputime;
 180 retry:
 181         curr_cputime = atomic64_read(cputime);
 182         if (sum_cputime > curr_cputime) {
 183                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 184                         goto retry;
 185         }
 186 }
 187
 188 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
 189 {
 190         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 191         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 192         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 193 }
 194
 195 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 196 static inline void sample_cputime_atomic(struct task_cputime *times,
 197                                          struct task_cputime_atomic *atomic_times)
 198 {
 199         times->utime = atomic64_read(&atomic_times->utime);
 200         times->stime = atomic64_read(&atomic_times->stime);
 201         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 202 }
 203
 204 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 205 {
 206         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 207         struct task_cputime sum;
 208
 209         /* Check if cputimer isn't running. This is accessed without locking. */
 210         if (!READ_ONCE(cputimer->running)) {
 211                 /*
 212                  * The POSIX timer interface allows for absolute time expiry
 213                  * values through the TIMER_ABSTIME flag, therefore we have
 214                  * to synchronize the timer to the clock every time we start it.
 215                  */
 216                 thread_group_cputime(tsk, &sum);
 217                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 218
 219                 /*
 220                  * We're setting cputimer->running without a lock. Ensure
 221                  * this only gets written to in one operation. We set
 222                  * running after update_gt_cputime() as a small optimization,
 223                  * but barriers are not required because update_gt_cputime()
 224                  * can handle concurrent updates.
 225                  */
 226                 WRITE_ONCE(cputimer->running, true);
 227         }
 228         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 229 }
 230
 231 /*
 232  * Sample a process (thread group) clock for the given group_leader task.
 233  * Must be called with task sighand lock held for safe while_each_thread()
 234  * traversal.
 235  */
 236 static int cpu_clock_sample_group(const clockid_t which_clock,
 237                                   struct task_struct *p,
 238                                   u64 *sample)
 239 {
 240         struct task_cputime cputime;
 241
 242         switch (CPUCLOCK_WHICH(which_clock)) {
 243         default:
 244                 return -EINVAL;
 245         case CPUCLOCK_PROF:
 246                 thread_group_cputime(p, &cputime);
 247                 *sample = cputime.utime + cputime.stime;
 248                 break;
 249         case CPUCLOCK_VIRT:
 250                 thread_group_cputime(p, &cputime);
 251                 *sample = cputime.utime;
 252                 break;
 253         case CPUCLOCK_SCHED:
 254                 thread_group_cputime(p, &cputime);
 255                 *sample = cputime.sum_exec_runtime;
 256                 break;
 257         }
 258         return 0;
 259 }
 260
 261 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 262                                     const clockid_t which_clock,
 263                                     struct timespec *tp)
 264 {
 265         int err = -EINVAL;
 266         u64 rtn;
 267
 268         if (CPUCLOCK_PERTHREAD(which_clock)) {
 269                 if (same_thread_group(tsk, current))
 270                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 271         } else {
 272                 if (tsk == current || thread_group_leader(tsk))
 273                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 274         }
 275
 276         if (!err)
 277                 *tp = ns_to_timespec(rtn);
 278
 279         return err;
 280 }
 281
 282
 283 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 284 {
 285         const pid_t pid = CPUCLOCK_PID(which_clock);
 286         int err = -EINVAL;
 287
 288         if (pid == 0) {
 289                 /*
 290                  * Special case constant value for our own clocks.
 291                  * We don't have to do any lookup to find ourselves.
 292                  */
 293                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 294         } else {
 295                 /*
 296                  * Find the given PID, and validate that the caller
 297                  * should be able to see it.
 298                  */
 299                 struct task_struct *p;
 300                 rcu_read_lock();
 301                 p = find_task_by_vpid(pid);
 302                 if (p)
 303                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 304                 rcu_read_unlock();
 305         }
 306
 307         return err;
 308 }
 309
 310 /*
 311  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 312  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 313  * new timer already all-zeros initialized.
 314  */
 315 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 316 {
 317         int ret = 0;
 318         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 319         struct task_struct *p;
 320
 321         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 322                 return -EINVAL;
 323
 324         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 325
 326         rcu_read_lock();
 327         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 328                 if (pid == 0) {
 329                         p = current;
 330                 } else {
 331                         p = find_task_by_vpid(pid);
 332                         if (p && !same_thread_group(p, current))
 333                                 p = NULL;
 334                 }
 335         } else {
 336                 if (pid == 0) {
 337                         p = current->group_leader;
 338                 } else {
 339                         p = find_task_by_vpid(pid);
 340                         if (p && !has_group_leader_pid(p))
 341                                 p = NULL;
 342                 }
 343         }
 344         new_timer->it.cpu.task = p;
 345         if (p) {
 346                 get_task_struct(p);
 347         } else {
 348                 ret = -EINVAL;
 349         }
 350         rcu_read_unlock();
 351
 352         return ret;
 353 }
 354
 355 /*
 356  * Clean up a CPU-clock timer that is about to be destroyed.
 357  * This is called from timer deletion with the timer already locked.
 358  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 359  * and try again.  (This happens when the timer is in the middle of firing.)
 360  */
 361 static int posix_cpu_timer_del(struct k_itimer *timer)
 362 {
 363         int ret = 0;
 364         unsigned long flags;
 365         struct sighand_struct *sighand;
 366         struct task_struct *p = timer->it.cpu.task;
 367
 368         WARN_ON_ONCE(p == NULL);
 369
 370         /*
 371          * Protect against sighand release/switch in exit/exec and process/
 372          * thread timer list entry concurrent read/writes.
 373          */
 374         sighand = lock_task_sighand(p, &flags);
 375         if (unlikely(sighand == NULL)) {
 376                 /*
 377                  * We raced with the reaping of the task.
 378                  * The deletion should have cleared us off the list.
 379                  */
 380                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 381         } else {
 382                 if (timer->it.cpu.firing)
 383                         ret = TIMER_RETRY;
 384                 else
 385                         list_del(&timer->it.cpu.entry);
 386
 387                 unlock_task_sighand(p, &flags);
 388         }
 389
 390         if (!ret)
 391                 put_task_struct(p);
 392
 393         return ret;
 394 }
 395
 396 static void cleanup_timers_list(struct list_head *head)
 397 {
 398         struct cpu_timer_list *timer, *next;
 399
 400         list_for_each_entry_safe(timer, next, head, entry)
 401                 list_del_init(&timer->entry);
 402 }
 403
 404 /*
 405  * Clean out CPU timers still ticking when a thread exited.  The task
 406  * pointer is cleared, and the expiry time is replaced with the residual
 407  * time for later timer_gettime calls to return.
 408  * This must be called with the siglock held.
 409  */
 410 static void cleanup_timers(struct list_head *head)
 411 {
 412         cleanup_timers_list(head);
 413         cleanup_timers_list(++head);
 414         cleanup_timers_list(++head);
 415 }
 416
 417 /*
 418  * These are both called with the siglock held, when the current thread
 419  * is being reaped.  When the final (leader) thread in the group is reaped,
 420  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 421  */
 422 void posix_cpu_timers_exit(struct task_struct *tsk)
 423 {
 424         cleanup_timers(tsk->cpu_timers);
 425 }
 426 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 427 {
 428         cleanup_timers(tsk->signal->cpu_timers);
 429 }
 430
 431 static inline int expires_gt(u64 expires, u64 new_exp)
 432 {
 433         return expires == 0 || expires > new_exp;
 434 }
 435
 436 /*
 437  * Insert the timer on the appropriate list before any timers that
 438  * expire later.  This must be called with the sighand lock held.
 439  */
 440 static void arm_timer(struct k_itimer *timer)
 441 {
 442         struct task_struct *p = timer->it.cpu.task;
 443         struct list_head *head, *listpos;
 444         struct task_cputime *cputime_expires;
 445         struct cpu_timer_list *const nt = &timer->it.cpu;
 446         struct cpu_timer_list *next;
 447
 448         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 449                 head = p->cpu_timers;
 450                 cputime_expires = &p->cputime_expires;
 451         } else {
 452                 head = p->signal->cpu_timers;
 453                 cputime_expires = &p->signal->cputime_expires;
 454         }
 455         head += CPUCLOCK_WHICH(timer->it_clock);
 456
 457         listpos = head;
 458         list_for_each_entry(next, head, entry) {
 459                 if (nt->expires < next->expires)
 460                         break;
 461                 listpos = &next->entry;
 462         }
 463         list_add(&nt->entry, listpos);
 464
 465         if (listpos == head) {
 466                 u64 exp = nt->expires;
 467
 468                 /*
 469                  * We are the new earliest-expiring POSIX 1.b timer, hence
 470                  * need to update expiration cache. Take into account that
 471                  * for process timers we share expiration cache with itimers
 472                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 473                  */
 474
 475                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 476                 case CPUCLOCK_PROF:
 477                         if (expires_gt(cputime_expires->prof_exp, exp))
 478                                 cputime_expires->prof_exp = exp;
 479                         break;
 480                 case CPUCLOCK_VIRT:
 481                         if (expires_gt(cputime_expires->virt_exp, exp))
 482                                 cputime_expires->virt_exp = exp;
 483                         break;
 484                 case CPUCLOCK_SCHED:
 485                         if (expires_gt(cputime_expires->sched_exp, exp))
 486                                 cputime_expires->sched_exp = exp;
 487                         break;
 488                 }
 489                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 490                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 491                 else
 492                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 493         }
 494 }
 495
 496 /*
 497  * The timer is locked, fire it and arrange for its reload.
 498  */
 499 static void cpu_timer_fire(struct k_itimer *timer)
 500 {
 501         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 502                 /*
 503                  * User don't want any signal.
 504                  */
 505                 timer->it.cpu.expires = 0;
 506         } else if (unlikely(timer->sigq == NULL)) {
 507                 /*
 508                  * This a special case for clock_nanosleep,
 509                  * not a normal timer from sys_timer_create.
 510                  */
 511                 wake_up_process(timer->it_process);
 512                 timer->it.cpu.expires = 0;
 513         } else if (timer->it.cpu.incr == 0) {
 514                 /*
 515                  * One-shot timer.  Clear it as soon as it's fired.
 516                  */
 517                 posix_timer_event(timer, 0);
 518                 timer->it.cpu.expires = 0;
 519         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 520                 /*
 521                  * The signal did not get queued because the signal
 522                  * was ignored, so we won't get any callback to
 523                  * reload the timer.  But we need to keep it
 524                  * ticking in case the signal is deliverable next time.
 525                  */
 526                 posix_cpu_timer_schedule(timer);
 527         }
 528 }
 529
 530 /*
 531  * Sample a process (thread group) timer for the given group_leader task.
 532  * Must be called with task sighand lock held for safe while_each_thread()
 533  * traversal.
 534  */
 535 static int cpu_timer_sample_group(const clockid_t which_clock,
 536                                   struct task_struct *p, u64 *sample)
 537 {
 538         struct task_cputime cputime;
 539
 540         thread_group_cputimer(p, &cputime);
 541         switch (CPUCLOCK_WHICH(which_clock)) {
 542         default:
 543                 return -EINVAL;
 544         case CPUCLOCK_PROF:
 545                 *sample = cputime.utime + cputime.stime;
 546                 break;
 547         case CPUCLOCK_VIRT:
 548                 *sample = cputime.utime;
 549                 break;
 550         case CPUCLOCK_SCHED:
 551                 *sample = cputime.sum_exec_runtime;
 552                 break;
 553         }
 554         return 0;
 555 }
 556
 557 /*
 558  * Guts of sys_timer_settime for CPU timers.
 559  * This is called with the timer locked and interrupts disabled.
 560  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 561  * and try again.  (This happens when the timer is in the middle of firing.)
 562  */
 563 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 564                                struct itimerspec *new, struct itimerspec *old)
 565 {
 566         unsigned long flags;
 567         struct sighand_struct *sighand;
 568         struct task_struct *p = timer->it.cpu.task;
 569         u64 old_expires, new_expires, old_incr, val;
 570         int ret;
 571
 572         WARN_ON_ONCE(p == NULL);
 573
 574         new_expires = timespec_to_ns(&new->it_value);
 575
 576         /*
 577          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 578          * and p->signal->cpu_timers read/write in arm_timer()
 579          */
 580         sighand = lock_task_sighand(p, &flags);
 581         /*
 582          * If p has just been reaped, we can no
 583          * longer get any information about it at all.
 584          */
 585         if (unlikely(sighand == NULL)) {
 586                 return -ESRCH;
 587         }
 588
 589         /*
 590          * Disarm any old timer after extracting its expiry time.
 591          */
 592         WARN_ON_ONCE(!irqs_disabled());
 593
 594         ret = 0;
 595         old_incr = timer->it.cpu.incr;
 596         old_expires = timer->it.cpu.expires;
 597         if (unlikely(timer->it.cpu.firing)) {
 598                 timer->it.cpu.firing = -1;
 599                 ret = TIMER_RETRY;
 600         } else
 601                 list_del_init(&timer->it.cpu.entry);
 602
 603         /*
 604          * We need to sample the current value to convert the new
 605          * value from to relative and absolute, and to convert the
 606          * old value from absolute to relative.  To set a process
 607          * timer, we need a sample to balance the thread expiry
 608          * times (in arm_timer).  With an absolute time, we must
 609          * check if it's already passed.  In short, we need a sample.
 610          */
 611         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 612                 cpu_clock_sample(timer->it_clock, p, &val);
 613         } else {
 614                 cpu_timer_sample_group(timer->it_clock, p, &val);
 615         }
 616
 617         if (old) {
 618                 if (old_expires == 0) {
 619                         old->it_value.tv_sec = 0;
 620                         old->it_value.tv_nsec = 0;
 621                 } else {
 622                         /*
 623                          * Update the timer in case it has
 624                          * overrun already.  If it has,
 625                          * we'll report it as having overrun
 626                          * and with the next reloaded timer
 627                          * already ticking, though we are
 628                          * swallowing that pending
 629                          * notification here to install the
 630                          * new setting.
 631                          */
 632                         bump_cpu_timer(timer, val);
 633                         if (val < timer->it.cpu.expires) {
 634                                 old_expires = timer->it.cpu.expires - val;
 635                                 old->it_value = ns_to_timespec(old_expires);
 636                         } else {
 637                                 old->it_value.tv_nsec = 1;
 638                                 old->it_value.tv_sec = 0;
 639                         }
 640                 }
 641         }
 642
 643         if (unlikely(ret)) {
 644                 /*
 645                  * We are colliding with the timer actually firing.
 646                  * Punt after filling in the timer's old value, and
 647                  * disable this firing since we are already reporting
 648                  * it as an overrun (thanks to bump_cpu_timer above).
 649                  */
 650                 unlock_task_sighand(p, &flags);
 651                 goto out;
 652         }
 653
 654         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 655                 new_expires += val;
 656         }
 657
 658         /*
 659          * Install the new expiry time (or zero).
 660          * For a timer with no notification action, we don't actually
 661          * arm the timer (we'll just fake it for timer_gettime).
 662          */
 663         timer->it.cpu.expires = new_expires;
 664         if (new_expires != 0 && val < new_expires) {
 665                 arm_timer(timer);
 666         }
 667
 668         unlock_task_sighand(p, &flags);
 669         /*
 670          * Install the new reload setting, and
 671          * set up the signal and overrun bookkeeping.
 672          */
 673         timer->it.cpu.incr = timespec_to_ns(&new->it_interval);
 674
 675         /*
 676          * This acts as a modification timestamp for the timer,
 677          * so any automatic reload attempt will punt on seeing
 678          * that we have reset the timer manually.
 679          */
 680         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 681                 ~REQUEUE_PENDING;
 682         timer->it_overrun_last = 0;
 683         timer->it_overrun = -1;
 684
 685         if (new_expires != 0 && !(val < new_expires)) {
 686                 /*
 687                  * The designated time already passed, so we notify
 688                  * immediately, even if the thread never runs to
 689                  * accumulate more time on this clock.
 690                  */
 691                 cpu_timer_fire(timer);
 692         }
 693
 694         ret = 0;
 695  out:
 696         if (old)
 697                 old->it_interval = ns_to_timespec(old_incr);
 698
 699         return ret;
 700 }
 701
 702 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 703 {
 704         u64 now;
 705         struct task_struct *p = timer->it.cpu.task;
 706
 707         WARN_ON_ONCE(p == NULL);
 708
 709         /*
 710          * Easy part: convert the reload time.
 711          */
 712         itp->it_interval = ns_to_timespec(timer->it.cpu.incr);
 713
 714         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 715                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 716                 return;
 717         }
 718
 719         /*
 720          * Sample the clock to take the difference with the expiry time.
 721          */
 722         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 723                 cpu_clock_sample(timer->it_clock, p, &now);
 724         } else {
 725                 struct sighand_struct *sighand;
 726                 unsigned long flags;
 727
 728                 /*
 729                  * Protect against sighand release/switch in exit/exec and
 730                  * also make timer sampling safe if it ends up calling
 731                  * thread_group_cputime().
 732                  */
 733                 sighand = lock_task_sighand(p, &flags);
 734                 if (unlikely(sighand == NULL)) {
 735                         /*
 736                          * The process has been reaped.
 737                          * We can't even collect a sample any more.
 738                          * Call the timer disarmed, nothing else to do.
 739                          */
 740                         timer->it.cpu.expires = 0;
 741                         itp->it_value = ns_to_timespec(timer->it.cpu.expires);
 742                         return;
 743                 } else {
 744                         cpu_timer_sample_group(timer->it_clock, p, &now);
 745                         unlock_task_sighand(p, &flags);
 746                 }
 747         }
 748
 749         if (now < timer->it.cpu.expires) {
 750                 itp->it_value = ns_to_timespec(timer->it.cpu.expires - now);
 751         } else {
 752                 /*
 753                  * The timer should have expired already, but the firing
 754                  * hasn't taken place yet.  Say it's just about to expire.
 755                  */
 756                 itp->it_value.tv_nsec = 1;
 757                 itp->it_value.tv_sec = 0;
 758         }
 759 }
 760
 761 static unsigned long long
 762 check_timers_list(struct list_head *timers,
 763                   struct list_head *firing,
 764                   unsigned long long curr)
 765 {
 766         int maxfire = 20;
 767
 768         while (!list_empty(timers)) {
 769                 struct cpu_timer_list *t;
 770
 771                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 772
 773                 if (!--maxfire || curr < t->expires)
 774                         return t->expires;
 775
 776                 t->firing = 1;
 777                 list_move_tail(&t->entry, firing);
 778         }
 779
 780         return 0;
 781 }
 782
 783 /*
 784  * Check for any per-thread CPU timers that have fired and move them off
 785  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 786  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 787  */
 788 static void check_thread_timers(struct task_struct *tsk,
 789                                 struct list_head *firing)
 790 {
 791         struct list_head *timers = tsk->cpu_timers;
 792         struct signal_struct *const sig = tsk->signal;
 793         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 794         u64 expires;
 795         unsigned long soft;
 796
 797         /*
 798          * If cputime_expires is zero, then there are no active
 799          * per thread CPU timers.
 800          */
 801         if (task_cputime_zero(&tsk->cputime_expires))
 802                 return;
 803
 804         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 805         tsk_expires->prof_exp = expires;
 806
 807         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 808         tsk_expires->virt_exp = expires;
 809
 810         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 811                                                    tsk->se.sum_exec_runtime);
 812
 813         /*
 814          * Check for the special case thread timers.
 815          */
 816         soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 817         if (soft != RLIM_INFINITY) {
 818                 unsigned long hard =
 819                         READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 820
 821                 if (hard != RLIM_INFINITY &&
 822                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 823                         /*
 824                          * At the hard limit, we just die.
 825                          * No need to calculate anything else now.
 826                          */
 827                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 828                         return;
 829                 }
 830                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 831                         /*
 832                          * At the soft limit, send a SIGXCPU every second.
 833                          */
 834                         if (soft < hard) {
 835                                 soft += USEC_PER_SEC;
 836                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 837                         }
 838                         printk(KERN_INFO
 839                                 "RT Watchdog Timeout: %s[%d]\n",
 840                                 tsk->comm, task_pid_nr(tsk));
 841                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 842                 }
 843         }
 844         if (task_cputime_zero(tsk_expires))
 845                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 846 }
 847
 848 static inline void stop_process_timers(struct signal_struct *sig)
 849 {
 850         struct thread_group_cputimer *cputimer = &sig->cputimer;
 851
 852         /* Turn off cputimer->running. This is done without locking. */
 853         WRITE_ONCE(cputimer->running, false);
 854         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 855 }
 856
 857 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 858                              u64 *expires, u64 cur_time, int signo)
 859 {
 860         if (!it->expires)
 861                 return;
 862
 863         if (cur_time >= cputime_to_nsecs(it->expires)) {
 864                 if (it->incr) {
 865                         it->expires += it->incr;
 866                         it->error += it->incr_error;
 867                         if (it->error >= TICK_NSEC) {
 868                                 it->expires -= cputime_one_jiffy;
 869                                 it->error -= TICK_NSEC;
 870                         }
 871                 } else {
 872                         it->expires = 0;
 873                 }
 874
 875                 trace_itimer_expire(signo == SIGPROF ?
 876                                     ITIMER_PROF : ITIMER_VIRTUAL,
 877                                     tsk->signal->leader_pid, cur_time);
 878                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 879         }
 880
 881         if (it->expires && (!*expires || cputime_to_nsecs(it->expires) < *expires)) {
 882                 *expires = cputime_to_nsecs(it->expires);
 883         }
 884 }
 885
 886 /*
 887  * Check for any per-thread CPU timers that have fired and move them
 888  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 889  * have already been taken off.
 890  */
 891 static void check_process_timers(struct task_struct *tsk,
 892                                  struct list_head *firing)
 893 {
 894         struct signal_struct *const sig = tsk->signal;
 895         u64 utime, ptime, virt_expires, prof_expires;
 896         u64 sum_sched_runtime, sched_expires;
 897         struct list_head *timers = sig->cpu_timers;
 898         struct task_cputime cputime;
 899         unsigned long soft;
 900
 901         /*
 902          * If cputimer is not running, then there are no active
 903          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 904          */
 905         if (!READ_ONCE(tsk->signal->cputimer.running))
 906                 return;
 907
 908         /*
 909          * Signify that a thread is checking for process timers.
 910          * Write access to this field is protected by the sighand lock.
 911          */
 912         sig->cputimer.checking_timer = true;
 913
 914         /*
 915          * Collect the current process totals.
 916          */
 917         thread_group_cputimer(tsk, &cputime);
 918         utime = cputime.utime;
 919         ptime = utime + cputime.stime;
 920         sum_sched_runtime = cputime.sum_exec_runtime;
 921
 922         prof_expires = check_timers_list(timers, firing, ptime);
 923         virt_expires = check_timers_list(++timers, firing, utime);
 924         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 925
 926         /*
 927          * Check for the special case process timers.
 928          */
 929         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 930                          SIGPROF);
 931         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 932                          SIGVTALRM);
 933         soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 934         if (soft != RLIM_INFINITY) {
 935                 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
 936                 unsigned long hard =
 937                         READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 938                 u64 x;
 939                 if (psecs >= hard) {
 940                         /*
 941                          * At the hard limit, we just die.
 942                          * No need to calculate anything else now.
 943                          */
 944                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 945                         return;
 946                 }
 947                 if (psecs >= soft) {
 948                         /*
 949                          * At the soft limit, send a SIGXCPU every second.
 950                          */
 951                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 952                         if (soft < hard) {
 953                                 soft++;
 954                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 955                         }
 956                 }
 957                 x = soft * NSEC_PER_SEC;
 958                 if (!prof_expires || x < prof_expires)
 959                         prof_expires = x;
 960         }
 961
 962         sig->cputime_expires.prof_exp = prof_expires;
 963         sig->cputime_expires.virt_exp = virt_expires;
 964         sig->cputime_expires.sched_exp = sched_expires;
 965         if (task_cputime_zero(&sig->cputime_expires))
 966                 stop_process_timers(sig);
 967
 968         sig->cputimer.checking_timer = false;
 969 }
 970
 971 /*
 972  * This is called from the signal code (via do_schedule_next_timer)
 973  * when the last timer signal was delivered and we have to reload the timer.
 974  */
 975 void posix_cpu_timer_schedule(struct k_itimer *timer)
 976 {
 977         struct sighand_struct *sighand;
 978         unsigned long flags;
 979         struct task_struct *p = timer->it.cpu.task;
 980         u64 now;
 981
 982         WARN_ON_ONCE(p == NULL);
 983
 984         /*
 985          * Fetch the current sample and update the timer's expiry time.
 986          */
 987         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 988                 cpu_clock_sample(timer->it_clock, p, &now);
 989                 bump_cpu_timer(timer, now);
 990                 if (unlikely(p->exit_state))
 991                         goto out;
 992
 993                 /* Protect timer list r/w in arm_timer() */
 994                 sighand = lock_task_sighand(p, &flags);
 995                 if (!sighand)
 996                         goto out;
 997         } else {
 998                 /*
 999                  * Protect arm_timer() and timer sampling in case of call to
1000                  * thread_group_cputime().
1001                  */
1002                 sighand = lock_task_sighand(p, &flags);
1003                 if (unlikely(sighand == NULL)) {
1004                         /*
1005                          * The process has been reaped.
1006                          * We can't even collect a sample any more.
1007                          */
1008                         timer->it.cpu.expires = 0;
1009                         goto out;
1010                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1011                         unlock_task_sighand(p, &flags);
1012                         /* Optimizations: if the process is dying, no need to rearm */
1013                         goto out;
1014                 }
1015                 cpu_timer_sample_group(timer->it_clock, p, &now);
1016                 bump_cpu_timer(timer, now);
1017                 /* Leave the sighand locked for the call below.  */
1018         }
1019
1020         /*
1021          * Now re-arm for the new expiry time.
1022          */
1023         WARN_ON_ONCE(!irqs_disabled());
1024         arm_timer(timer);
1025         unlock_task_sighand(p, &flags);
1026
1027 out:
1028         timer->it_overrun_last = timer->it_overrun;
1029         timer->it_overrun = -1;
1030         ++timer->it_requeue_pending;
1031 }
1032
1033 /**
1034  * task_cputime_expired - Compare two task_cputime entities.
1035  *
1036  * @sample:     The task_cputime structure to be checked for expiration.
1037  * @expires:    Expiration times, against which @sample will be checked.
1038  *
1039  * Checks @sample against @expires to see if any field of @sample has expired.
1040  * Returns true if any field of the former is greater than the corresponding
1041  * field of the latter if the latter field is set.  Otherwise returns false.
1042  */
1043 static inline int task_cputime_expired(const struct task_cputime *sample,
1044                                         const struct task_cputime *expires)
1045 {
1046         if (expires->utime && sample->utime >= expires->utime)
1047                 return 1;
1048         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1049                 return 1;
1050         if (expires->sum_exec_runtime != 0 &&
1051             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1052                 return 1;
1053         return 0;
1054 }
1055
1056 /**
1057  * fastpath_timer_check - POSIX CPU timers fast path.
1058  *
1059  * @tsk:        The task (thread) being checked.
1060  *
1061  * Check the task and thread group timers.  If both are zero (there are no
1062  * timers set) return false.  Otherwise snapshot the task and thread group
1063  * timers and compare them with the corresponding expiration times.  Return
1064  * true if a timer has expired, else return false.
1065  */
1066 static inline int fastpath_timer_check(struct task_struct *tsk)
1067 {
1068         struct signal_struct *sig;
1069
1070         if (!task_cputime_zero(&tsk->cputime_expires)) {
1071                 struct task_cputime task_sample;
1072
1073                 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1074                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1075                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1076                         return 1;
1077         }
1078
1079         sig = tsk->signal;
1080         /*
1081          * Check if thread group timers expired when the cputimer is
1082          * running and no other thread in the group is already checking
1083          * for thread group cputimers. These fields are read without the
1084          * sighand lock. However, this is fine because this is meant to
1085          * be a fastpath heuristic to determine whether we should try to
1086          * acquire the sighand lock to check/handle timers.
1087          *
1088          * In the worst case scenario, if 'running' or 'checking_timer' gets
1089          * set but the current thread doesn't see the change yet, we'll wait
1090          * until the next thread in the group gets a scheduler interrupt to
1091          * handle the timer. This isn't an issue in practice because these
1092          * types of delays with signals actually getting sent are expected.
1093          */
1094         if (READ_ONCE(sig->cputimer.running) &&
1095             !READ_ONCE(sig->cputimer.checking_timer)) {
1096                 struct task_cputime group_sample;
1097
1098                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1099
1100                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1101                         return 1;
1102         }
1103
1104         return 0;
1105 }
1106
1107 /*
1108  * This is called from the timer interrupt handler.  The irq handler has
1109  * already updated our counts.  We need to check if any timers fire now.
1110  * Interrupts are disabled.
1111  */
1112 void run_posix_cpu_timers(struct task_struct *tsk)
1113 {
1114         LIST_HEAD(firing);
1115         struct k_itimer *timer, *next;
1116         unsigned long flags;
1117
1118         WARN_ON_ONCE(!irqs_disabled());
1119
1120         /*
1121          * The fast path checks that there are no expired thread or thread
1122          * group timers.  If that's so, just return.
1123          */
1124         if (!fastpath_timer_check(tsk))
1125                 return;
1126
1127         if (!lock_task_sighand(tsk, &flags))
1128                 return;
1129         /*
1130          * Here we take off tsk->signal->cpu_timers[N] and
1131          * tsk->cpu_timers[N] all the timers that are firing, and
1132          * put them on the firing list.
1133          */
1134         check_thread_timers(tsk, &firing);
1135
1136         check_process_timers(tsk, &firing);
1137
1138         /*
1139          * We must release these locks before taking any timer's lock.
1140          * There is a potential race with timer deletion here, as the
1141          * siglock now protects our private firing list.  We have set
1142          * the firing flag in each timer, so that a deletion attempt
1143          * that gets the timer lock before we do will give it up and
1144          * spin until we've taken care of that timer below.
1145          */
1146         unlock_task_sighand(tsk, &flags);
1147
1148         /*
1149          * Now that all the timers on our list have the firing flag,
1150          * no one will touch their list entries but us.  We'll take
1151          * each timer's lock before clearing its firing flag, so no
1152          * timer call will interfere.
1153          */
1154         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1155                 int cpu_firing;
1156
1157                 spin_lock(&timer->it_lock);
1158                 list_del_init(&timer->it.cpu.entry);
1159                 cpu_firing = timer->it.cpu.firing;
1160                 timer->it.cpu.firing = 0;
1161                 /*
1162                  * The firing flag is -1 if we collided with a reset
1163                  * of the timer, which already reported this
1164                  * almost-firing as an overrun.  So don't generate an event.
1165                  */
1166                 if (likely(cpu_firing >= 0))
1167                         cpu_timer_fire(timer);
1168                 spin_unlock(&timer->it_lock);
1169         }
1170 }
1171
1172 /*
1173  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1174  * The tsk->sighand->siglock must be held by the caller.
1175  */
1176 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1177                            cputime_t *newval, cputime_t *oldval)
1178 {
1179         u64 now, new;
1180
1181         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1182         cpu_timer_sample_group(clock_idx, tsk, &now);
1183
1184         if (oldval) {
1185                 /*
1186                  * We are setting itimer. The *oldval is absolute and we update
1187                  * it to be relative, *newval argument is relative and we update
1188                  * it to be absolute.
1189                  */
1190                 if (*oldval) {
1191                         if (cputime_to_nsecs(*oldval) <= now) {
1192                                 /* Just about to fire. */
1193                                 *oldval = cputime_one_jiffy;
1194                         } else {
1195                                 *oldval -= nsecs_to_cputime(now);
1196                         }
1197                 }
1198
1199                 if (!*newval)
1200                         return;
1201                 *newval += nsecs_to_cputime(now);
1202         }
1203
1204         new = cputime_to_nsecs(*newval);
1205
1206         /*
1207          * Update expiration cache if we are the earliest timer, or eventually
1208          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1209          */
1210         switch (clock_idx) {
1211         case CPUCLOCK_PROF:
1212                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, new))
1213                         tsk->signal->cputime_expires.prof_exp = new;
1214                 break;
1215         case CPUCLOCK_VIRT:
1216                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, new))
1217                         tsk->signal->cputime_expires.virt_exp = new;
1218                 break;
1219         }
1220
1221         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1222 }
1223
1224 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1225                             struct timespec *rqtp, struct itimerspec *it)
1226 {
1227         struct k_itimer timer;
1228         int error;
1229
1230         /*
1231          * Set up a temporary timer and then wait for it to go off.
1232          */
1233         memset(&timer, 0, sizeof timer);
1234         spin_lock_init(&timer.it_lock);
1235         timer.it_clock = which_clock;
1236         timer.it_overrun = -1;
1237         error = posix_cpu_timer_create(&timer);
1238         timer.it_process = current;
1239         if (!error) {
1240                 static struct itimerspec zero_it;
1241
1242                 memset(it, 0, sizeof *it);
1243                 it->it_value = *rqtp;
1244
1245                 spin_lock_irq(&timer.it_lock);
1246                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1247                 if (error) {
1248                         spin_unlock_irq(&timer.it_lock);
1249                         return error;
1250                 }
1251
1252                 while (!signal_pending(current)) {
1253                         if (timer.it.cpu.expires == 0) {
1254                                 /*
1255                                  * Our timer fired and was reset, below
1256                                  * deletion can not fail.
1257                                  */
1258                                 posix_cpu_timer_del(&timer);
1259                                 spin_unlock_irq(&timer.it_lock);
1260                                 return 0;
1261                         }
1262
1263                         /*
1264                          * Block until cpu_timer_fire (or a signal) wakes us.
1265                          */
1266                         __set_current_state(TASK_INTERRUPTIBLE);
1267                         spin_unlock_irq(&timer.it_lock);
1268                         schedule();
1269                         spin_lock_irq(&timer.it_lock);
1270                 }
1271
1272                 /*
1273                  * We were interrupted by a signal.
1274                  */
1275                 *rqtp = ns_to_timespec(timer.it.cpu.expires);
1276                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1277                 if (!error) {
1278                         /*
1279                          * Timer is now unarmed, deletion can not fail.
1280                          */
1281                         posix_cpu_timer_del(&timer);
1282                 }
1283                 spin_unlock_irq(&timer.it_lock);
1284
1285                 while (error == TIMER_RETRY) {
1286                         /*
1287                          * We need to handle case when timer was or is in the
1288                          * middle of firing. In other cases we already freed
1289                          * resources.
1290                          */
1291                         spin_lock_irq(&timer.it_lock);
1292                         error = posix_cpu_timer_del(&timer);
1293                         spin_unlock_irq(&timer.it_lock);
1294                 }
1295
1296                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1297                         /*
1298                          * It actually did fire already.
1299                          */
1300                         return 0;
1301                 }
1302
1303                 error = -ERESTART_RESTARTBLOCK;
1304         }
1305
1306         return error;
1307 }
1308
1309 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1310
1311 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1312                             struct timespec *rqtp, struct timespec __user *rmtp)
1313 {
1314         struct restart_block *restart_block = &current->restart_block;
1315         struct itimerspec it;
1316         int error;
1317
1318         /*
1319          * Diagnose required errors first.
1320          */
1321         if (CPUCLOCK_PERTHREAD(which_clock) &&
1322             (CPUCLOCK_PID(which_clock) == 0 ||
1323              CPUCLOCK_PID(which_clock) == current->pid))
1324                 return -EINVAL;
1325
1326         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1327
1328         if (error == -ERESTART_RESTARTBLOCK) {
1329
1330                 if (flags & TIMER_ABSTIME)
1331                         return -ERESTARTNOHAND;
1332                 /*
1333                  * Report back to the user the time still remaining.
1334                  */
1335                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1336                         return -EFAULT;
1337
1338                 restart_block->fn = posix_cpu_nsleep_restart;
1339                 restart_block->nanosleep.clockid = which_clock;
1340                 restart_block->nanosleep.rmtp = rmtp;
1341                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1342         }
1343         return error;
1344 }
1345
1346 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1347 {
1348         clockid_t which_clock = restart_block->nanosleep.clockid;
1349         struct timespec t;
1350         struct itimerspec it;
1351         int error;
1352
1353         t = ns_to_timespec(restart_block->nanosleep.expires);
1354
1355         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1356
1357         if (error == -ERESTART_RESTARTBLOCK) {
1358                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1359                 /*
1360                  * Report back to the user the time still remaining.
1361                  */
1362                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1363                         return -EFAULT;
1364
1365                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1366         }
1367         return error;
1368
1369 }
1370
1371 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1372 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1373
1374 static int process_cpu_clock_getres(const clockid_t which_clock,
1375                                     struct timespec *tp)
1376 {
1377         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1378 }
1379 static int process_cpu_clock_get(const clockid_t which_clock,
1380                                  struct timespec *tp)
1381 {
1382         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1383 }
1384 static int process_cpu_timer_create(struct k_itimer *timer)
1385 {
1386         timer->it_clock = PROCESS_CLOCK;
1387         return posix_cpu_timer_create(timer);
1388 }
1389 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1390                               struct timespec *rqtp,
1391                               struct timespec __user *rmtp)
1392 {
1393         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1394 }
1395 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1396 {
1397         return -EINVAL;
1398 }
1399 static int thread_cpu_clock_getres(const clockid_t which_clock,
1400                                    struct timespec *tp)
1401 {
1402         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1403 }
1404 static int thread_cpu_clock_get(const clockid_t which_clock,
1405                                 struct timespec *tp)
1406 {
1407         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1408 }
1409 static int thread_cpu_timer_create(struct k_itimer *timer)
1410 {
1411         timer->it_clock = THREAD_CLOCK;
1412         return posix_cpu_timer_create(timer);
1413 }
1414
1415 struct k_clock clock_posix_cpu = {
1416         .clock_getres   = posix_cpu_clock_getres,
1417         .clock_set      = posix_cpu_clock_set,
1418         .clock_get      = posix_cpu_clock_get,
1419         .timer_create   = posix_cpu_timer_create,
1420         .nsleep         = posix_cpu_nsleep,
1421         .nsleep_restart = posix_cpu_nsleep_restart,
1422         .timer_set      = posix_cpu_timer_set,
1423         .timer_del      = posix_cpu_timer_del,
1424         .timer_get      = posix_cpu_timer_get,
1425 };
1426
1427 static __init int init_posix_cpu_timers(void)
1428 {
1429         struct k_clock process = {
1430                 .clock_getres   = process_cpu_clock_getres,
1431                 .clock_get      = process_cpu_clock_get,
1432                 .timer_create   = process_cpu_timer_create,
1433                 .nsleep         = process_cpu_nsleep,
1434                 .nsleep_restart = process_cpu_nsleep_restart,
1435         };
1436         struct k_clock thread = {
1437                 .clock_getres   = thread_cpu_clock_getres,
1438                 .clock_get      = thread_cpu_clock_get,
1439                 .timer_create   = thread_cpu_timer_create,
1440         };
1441
1442         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1443         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1444
1445         return 0;
1446 }
1447 __initcall(init_posix_cpu_timers);