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 static inline unsigned long long
  54 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  55 {
  56         unsigned long long ret;
  57
  58         ret = 0;                /* high half always zero when .cpu used */
  59         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  60                 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  61         } else {
  62                 ret = cputime_to_expires(timespec_to_cputime(tp));
  63         }
  64         return ret;
  65 }
  66
  67 static void sample_to_timespec(const clockid_t which_clock,
  68                                unsigned long long expires,
  69                                struct timespec *tp)
  70 {
  71         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  72                 *tp = ns_to_timespec(expires);
  73         else
  74                 cputime_to_timespec((__force cputime_t)expires, tp);
  75 }
  76
  77 /*
  78  * Update expiry time from increment, and increase overrun count,
  79  * given the current clock sample.
  80  */
  81 static void bump_cpu_timer(struct k_itimer *timer,
  82                            unsigned long long now)
  83 {
  84         int i;
  85         unsigned long long delta, incr;
  86
  87         if (timer->it.cpu.incr == 0)
  88                 return;
  89
  90         if (now < timer->it.cpu.expires)
  91                 return;
  92
  93         incr = timer->it.cpu.incr;
  94         delta = now + incr - timer->it.cpu.expires;
  95
  96         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  97         for (i = 0; incr < delta - incr; i++)
  98                 incr = incr << 1;
  99
 100         for (; i >= 0; incr >>= 1, i--) {
 101                 if (delta < incr)
 102                         continue;
 103
 104                 timer->it.cpu.expires += incr;
 105                 timer->it_overrun += 1 << i;
 106                 delta -= incr;
 107         }
 108 }
 109
 110 /**
 111  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 112  *
 113  * @cputime:    The struct to compare.
 114  *
 115  * Checks @cputime to see if all fields are zero.  Returns true if all fields
 116  * are zero, false if any field is nonzero.
 117  */
 118 static inline int task_cputime_zero(const struct task_cputime_t *cputime)
 119 {
 120         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 121                 return 1;
 122         return 0;
 123 }
 124
 125 static inline unsigned long long prof_ticks(struct task_struct *p)
 126 {
 127         cputime_t utime, stime;
 128
 129         task_cputime_t(p, &utime, &stime);
 130
 131         return cputime_to_expires(utime + stime);
 132 }
 133 static inline unsigned long long virt_ticks(struct task_struct *p)
 134 {
 135         cputime_t utime, stime;
 136
 137         task_cputime_t(p, &utime, &stime);
 138
 139         return cputime_to_expires(utime);
 140 }
 141
 142 static int
 143 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 144 {
 145         int error = check_clock(which_clock);
 146         if (!error) {
 147                 tp->tv_sec = 0;
 148                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 149                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 150                         /*
 151                          * If sched_clock is using a cycle counter, we
 152                          * don't have any idea of its true resolution
 153                          * exported, but it is much more than 1s/HZ.
 154                          */
 155                         tp->tv_nsec = 1;
 156                 }
 157         }
 158         return error;
 159 }
 160
 161 static int
 162 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 163 {
 164         /*
 165          * You can never reset a CPU clock, but we check for other errors
 166          * in the call before failing with EPERM.
 167          */
 168         int error = check_clock(which_clock);
 169         if (error == 0) {
 170                 error = -EPERM;
 171         }
 172         return error;
 173 }
 174
 175
 176 /*
 177  * Sample a per-thread clock for the given task.
 178  */
 179 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 180                             unsigned long long *sample)
 181 {
 182         switch (CPUCLOCK_WHICH(which_clock)) {
 183         default:
 184                 return -EINVAL;
 185         case CPUCLOCK_PROF:
 186                 *sample = prof_ticks(p);
 187                 break;
 188         case CPUCLOCK_VIRT:
 189                 *sample = virt_ticks(p);
 190                 break;
 191         case CPUCLOCK_SCHED:
 192                 *sample = task_sched_runtime(p);
 193                 break;
 194         }
 195         return 0;
 196 }
 197
 198 /*
 199  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 200  * to avoid race conditions with concurrent updates to cputime.
 201  */
 202 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 203 {
 204         u64 curr_cputime;
 205 retry:
 206         curr_cputime = atomic64_read(cputime);
 207         if (sum_cputime > curr_cputime) {
 208                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 209                         goto retry;
 210         }
 211 }
 212
 213 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime_t *sum)
 214 {
 215         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 216         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 217         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 218 }
 219
 220 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 221 static inline void sample_cputime_atomic(struct task_cputime_t *times,
 222                                          struct task_cputime_atomic *atomic_times)
 223 {
 224         times->utime = atomic64_read(&atomic_times->utime);
 225         times->stime = atomic64_read(&atomic_times->stime);
 226         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 227 }
 228
 229 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime_t *times)
 230 {
 231         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 232         struct task_cputime_t sum;
 233
 234         /* Check if cputimer isn't running. This is accessed without locking. */
 235         if (!READ_ONCE(cputimer->running)) {
 236                 /*
 237                  * The POSIX timer interface allows for absolute time expiry
 238                  * values through the TIMER_ABSTIME flag, therefore we have
 239                  * to synchronize the timer to the clock every time we start it.
 240                  */
 241                 thread_group_cputime_t(tsk, &sum);
 242                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 243
 244                 /*
 245                  * We're setting cputimer->running without a lock. Ensure
 246                  * this only gets written to in one operation. We set
 247                  * running after update_gt_cputime() as a small optimization,
 248                  * but barriers are not required because update_gt_cputime()
 249                  * can handle concurrent updates.
 250                  */
 251                 WRITE_ONCE(cputimer->running, true);
 252         }
 253         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 254 }
 255
 256 /*
 257  * Sample a process (thread group) clock for the given group_leader task.
 258  * Must be called with task sighand lock held for safe while_each_thread()
 259  * traversal.
 260  */
 261 static int cpu_clock_sample_group(const clockid_t which_clock,
 262                                   struct task_struct *p,
 263                                   unsigned long long *sample)
 264 {
 265         struct task_cputime_t cputime;
 266
 267         switch (CPUCLOCK_WHICH(which_clock)) {
 268         default:
 269                 return -EINVAL;
 270         case CPUCLOCK_PROF:
 271                 thread_group_cputime_t(p, &cputime);
 272                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 273                 break;
 274         case CPUCLOCK_VIRT:
 275                 thread_group_cputime_t(p, &cputime);
 276                 *sample = cputime_to_expires(cputime.utime);
 277                 break;
 278         case CPUCLOCK_SCHED:
 279                 thread_group_cputime_t(p, &cputime);
 280                 *sample = cputime.sum_exec_runtime;
 281                 break;
 282         }
 283         return 0;
 284 }
 285
 286 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 287                                     const clockid_t which_clock,
 288                                     struct timespec *tp)
 289 {
 290         int err = -EINVAL;
 291         unsigned long long rtn;
 292
 293         if (CPUCLOCK_PERTHREAD(which_clock)) {
 294                 if (same_thread_group(tsk, current))
 295                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 296         } else {
 297                 if (tsk == current || thread_group_leader(tsk))
 298                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 299         }
 300
 301         if (!err)
 302                 sample_to_timespec(which_clock, rtn, tp);
 303
 304         return err;
 305 }
 306
 307
 308 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 309 {
 310         const pid_t pid = CPUCLOCK_PID(which_clock);
 311         int err = -EINVAL;
 312
 313         if (pid == 0) {
 314                 /*
 315                  * Special case constant value for our own clocks.
 316                  * We don't have to do any lookup to find ourselves.
 317                  */
 318                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 319         } else {
 320                 /*
 321                  * Find the given PID, and validate that the caller
 322                  * should be able to see it.
 323                  */
 324                 struct task_struct *p;
 325                 rcu_read_lock();
 326                 p = find_task_by_vpid(pid);
 327                 if (p)
 328                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 329                 rcu_read_unlock();
 330         }
 331
 332         return err;
 333 }
 334
 335 /*
 336  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 337  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 338  * new timer already all-zeros initialized.
 339  */
 340 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 341 {
 342         int ret = 0;
 343         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 344         struct task_struct *p;
 345
 346         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 347                 return -EINVAL;
 348
 349         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 350
 351         rcu_read_lock();
 352         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 353                 if (pid == 0) {
 354                         p = current;
 355                 } else {
 356                         p = find_task_by_vpid(pid);
 357                         if (p && !same_thread_group(p, current))
 358                                 p = NULL;
 359                 }
 360         } else {
 361                 if (pid == 0) {
 362                         p = current->group_leader;
 363                 } else {
 364                         p = find_task_by_vpid(pid);
 365                         if (p && !has_group_leader_pid(p))
 366                                 p = NULL;
 367                 }
 368         }
 369         new_timer->it.cpu.task = p;
 370         if (p) {
 371                 get_task_struct(p);
 372         } else {
 373                 ret = -EINVAL;
 374         }
 375         rcu_read_unlock();
 376
 377         return ret;
 378 }
 379
 380 /*
 381  * Clean up a CPU-clock timer that is about to be destroyed.
 382  * This is called from timer deletion with the timer already locked.
 383  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 384  * and try again.  (This happens when the timer is in the middle of firing.)
 385  */
 386 static int posix_cpu_timer_del(struct k_itimer *timer)
 387 {
 388         int ret = 0;
 389         unsigned long flags;
 390         struct sighand_struct *sighand;
 391         struct task_struct *p = timer->it.cpu.task;
 392
 393         WARN_ON_ONCE(p == NULL);
 394
 395         /*
 396          * Protect against sighand release/switch in exit/exec and process/
 397          * thread timer list entry concurrent read/writes.
 398          */
 399         sighand = lock_task_sighand(p, &flags);
 400         if (unlikely(sighand == NULL)) {
 401                 /*
 402                  * We raced with the reaping of the task.
 403                  * The deletion should have cleared us off the list.
 404                  */
 405                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 406         } else {
 407                 if (timer->it.cpu.firing)
 408                         ret = TIMER_RETRY;
 409                 else
 410                         list_del(&timer->it.cpu.entry);
 411
 412                 unlock_task_sighand(p, &flags);
 413         }
 414
 415         if (!ret)
 416                 put_task_struct(p);
 417
 418         return ret;
 419 }
 420
 421 static void cleanup_timers_list(struct list_head *head)
 422 {
 423         struct cpu_timer_list *timer, *next;
 424
 425         list_for_each_entry_safe(timer, next, head, entry)
 426                 list_del_init(&timer->entry);
 427 }
 428
 429 /*
 430  * Clean out CPU timers still ticking when a thread exited.  The task
 431  * pointer is cleared, and the expiry time is replaced with the residual
 432  * time for later timer_gettime calls to return.
 433  * This must be called with the siglock held.
 434  */
 435 static void cleanup_timers(struct list_head *head)
 436 {
 437         cleanup_timers_list(head);
 438         cleanup_timers_list(++head);
 439         cleanup_timers_list(++head);
 440 }
 441
 442 /*
 443  * These are both called with the siglock held, when the current thread
 444  * is being reaped.  When the final (leader) thread in the group is reaped,
 445  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 446  */
 447 void posix_cpu_timers_exit(struct task_struct *tsk)
 448 {
 449         cleanup_timers(tsk->cpu_timers);
 450 }
 451 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 452 {
 453         cleanup_timers(tsk->signal->cpu_timers);
 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 sighand lock held.
 464  */
 465 static void arm_timer(struct k_itimer *timer)
 466 {
 467         struct task_struct *p = timer->it.cpu.task;
 468         struct list_head *head, *listpos;
 469         struct task_cputime_t *cputime_expires;
 470         struct cpu_timer_list *const nt = &timer->it.cpu;
 471         struct cpu_timer_list *next;
 472
 473         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 474                 head = p->cpu_timers;
 475                 cputime_expires = &p->cputime_expires;
 476         } else {
 477                 head = p->signal->cpu_timers;
 478                 cputime_expires = &p->signal->cputime_expires;
 479         }
 480         head += CPUCLOCK_WHICH(timer->it_clock);
 481
 482         listpos = head;
 483         list_for_each_entry(next, head, entry) {
 484                 if (nt->expires < next->expires)
 485                         break;
 486                 listpos = &next->entry;
 487         }
 488         list_add(&nt->entry, listpos);
 489
 490         if (listpos == head) {
 491                 unsigned long long exp = nt->expires;
 492
 493                 /*
 494                  * We are the new earliest-expiring POSIX 1.b timer, hence
 495                  * need to update expiration cache. Take into account that
 496                  * for process timers we share expiration cache with itimers
 497                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 498                  */
 499
 500                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 501                 case CPUCLOCK_PROF:
 502                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 503                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 504                         break;
 505                 case CPUCLOCK_VIRT:
 506                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 507                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 508                         break;
 509                 case CPUCLOCK_SCHED:
 510                         if (cputime_expires->sched_exp == 0 ||
 511                             cputime_expires->sched_exp > exp)
 512                                 cputime_expires->sched_exp = exp;
 513                         break;
 514                 }
 515                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 516                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 517                 else
 518                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 519         }
 520 }
 521
 522 /*
 523  * The timer is locked, fire it and arrange for its reload.
 524  */
 525 static void cpu_timer_fire(struct k_itimer *timer)
 526 {
 527         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 528                 /*
 529                  * User don't want any signal.
 530                  */
 531                 timer->it.cpu.expires = 0;
 532         } else if (unlikely(timer->sigq == NULL)) {
 533                 /*
 534                  * This a special case for clock_nanosleep,
 535                  * not a normal timer from sys_timer_create.
 536                  */
 537                 wake_up_process(timer->it_process);
 538                 timer->it.cpu.expires = 0;
 539         } else if (timer->it.cpu.incr == 0) {
 540                 /*
 541                  * One-shot timer.  Clear it as soon as it's fired.
 542                  */
 543                 posix_timer_event(timer, 0);
 544                 timer->it.cpu.expires = 0;
 545         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 546                 /*
 547                  * The signal did not get queued because the signal
 548                  * was ignored, so we won't get any callback to
 549                  * reload the timer.  But we need to keep it
 550                  * ticking in case the signal is deliverable next time.
 551                  */
 552                 posix_cpu_timer_schedule(timer);
 553         }
 554 }
 555
 556 /*
 557  * Sample a process (thread group) timer for the given group_leader task.
 558  * Must be called with task sighand lock held for safe while_each_thread()
 559  * traversal.
 560  */
 561 static int cpu_timer_sample_group(const clockid_t which_clock,
 562                                   struct task_struct *p,
 563                                   unsigned long long *sample)
 564 {
 565         struct task_cputime_t cputime;
 566
 567         thread_group_cputimer(p, &cputime);
 568         switch (CPUCLOCK_WHICH(which_clock)) {
 569         default:
 570                 return -EINVAL;
 571         case CPUCLOCK_PROF:
 572                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 573                 break;
 574         case CPUCLOCK_VIRT:
 575                 *sample = cputime_to_expires(cputime.utime);
 576                 break;
 577         case CPUCLOCK_SCHED:
 578                 *sample = cputime.sum_exec_runtime;
 579                 break;
 580         }
 581         return 0;
 582 }
 583
 584 /*
 585  * Guts of sys_timer_settime for CPU timers.
 586  * This is called with the timer locked and interrupts disabled.
 587  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 588  * and try again.  (This happens when the timer is in the middle of firing.)
 589  */
 590 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 591                                struct itimerspec *new, struct itimerspec *old)
 592 {
 593         unsigned long flags;
 594         struct sighand_struct *sighand;
 595         struct task_struct *p = timer->it.cpu.task;
 596         unsigned long long old_expires, new_expires, old_incr, val;
 597         int ret;
 598
 599         WARN_ON_ONCE(p == NULL);
 600
 601         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 602
 603         /*
 604          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 605          * and p->signal->cpu_timers read/write in arm_timer()
 606          */
 607         sighand = lock_task_sighand(p, &flags);
 608         /*
 609          * If p has just been reaped, we can no
 610          * longer get any information about it at all.
 611          */
 612         if (unlikely(sighand == NULL)) {
 613                 return -ESRCH;
 614         }
 615
 616         /*
 617          * Disarm any old timer after extracting its expiry time.
 618          */
 619         WARN_ON_ONCE(!irqs_disabled());
 620
 621         ret = 0;
 622         old_incr = timer->it.cpu.incr;
 623         old_expires = timer->it.cpu.expires;
 624         if (unlikely(timer->it.cpu.firing)) {
 625                 timer->it.cpu.firing = -1;
 626                 ret = TIMER_RETRY;
 627         } else
 628                 list_del_init(&timer->it.cpu.entry);
 629
 630         /*
 631          * We need to sample the current value to convert the new
 632          * value from to relative and absolute, and to convert the
 633          * old value from absolute to relative.  To set a process
 634          * timer, we need a sample to balance the thread expiry
 635          * times (in arm_timer).  With an absolute time, we must
 636          * check if it's already passed.  In short, we need a sample.
 637          */
 638         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 639                 cpu_clock_sample(timer->it_clock, p, &val);
 640         } else {
 641                 cpu_timer_sample_group(timer->it_clock, p, &val);
 642         }
 643
 644         if (old) {
 645                 if (old_expires == 0) {
 646                         old->it_value.tv_sec = 0;
 647                         old->it_value.tv_nsec = 0;
 648                 } else {
 649                         /*
 650                          * Update the timer in case it has
 651                          * overrun already.  If it has,
 652                          * we'll report it as having overrun
 653                          * and with the next reloaded timer
 654                          * already ticking, though we are
 655                          * swallowing that pending
 656                          * notification here to install the
 657                          * new setting.
 658                          */
 659                         bump_cpu_timer(timer, val);
 660                         if (val < timer->it.cpu.expires) {
 661                                 old_expires = timer->it.cpu.expires - val;
 662                                 sample_to_timespec(timer->it_clock,
 663                                                    old_expires,
 664                                                    &old->it_value);
 665                         } else {
 666                                 old->it_value.tv_nsec = 1;
 667                                 old->it_value.tv_sec = 0;
 668                         }
 669                 }
 670         }
 671
 672         if (unlikely(ret)) {
 673                 /*
 674                  * We are colliding with the timer actually firing.
 675                  * Punt after filling in the timer's old value, and
 676                  * disable this firing since we are already reporting
 677                  * it as an overrun (thanks to bump_cpu_timer above).
 678                  */
 679                 unlock_task_sighand(p, &flags);
 680                 goto out;
 681         }
 682
 683         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 684                 new_expires += val;
 685         }
 686
 687         /*
 688          * Install the new expiry time (or zero).
 689          * For a timer with no notification action, we don't actually
 690          * arm the timer (we'll just fake it for timer_gettime).
 691          */
 692         timer->it.cpu.expires = new_expires;
 693         if (new_expires != 0 && val < new_expires) {
 694                 arm_timer(timer);
 695         }
 696
 697         unlock_task_sighand(p, &flags);
 698         /*
 699          * Install the new reload setting, and
 700          * set up the signal and overrun bookkeeping.
 701          */
 702         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 703                                                 &new->it_interval);
 704
 705         /*
 706          * This acts as a modification timestamp for the timer,
 707          * so any automatic reload attempt will punt on seeing
 708          * that we have reset the timer manually.
 709          */
 710         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 711                 ~REQUEUE_PENDING;
 712         timer->it_overrun_last = 0;
 713         timer->it_overrun = -1;
 714
 715         if (new_expires != 0 && !(val < new_expires)) {
 716                 /*
 717                  * The designated time already passed, so we notify
 718                  * immediately, even if the thread never runs to
 719                  * accumulate more time on this clock.
 720                  */
 721                 cpu_timer_fire(timer);
 722         }
 723
 724         ret = 0;
 725  out:
 726         if (old) {
 727                 sample_to_timespec(timer->it_clock,
 728                                    old_incr, &old->it_interval);
 729         }
 730
 731         return ret;
 732 }
 733
 734 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 735 {
 736         unsigned long long now;
 737         struct task_struct *p = timer->it.cpu.task;
 738
 739         WARN_ON_ONCE(p == NULL);
 740
 741         /*
 742          * Easy part: convert the reload time.
 743          */
 744         sample_to_timespec(timer->it_clock,
 745                            timer->it.cpu.incr, &itp->it_interval);
 746
 747         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 748                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 749                 return;
 750         }
 751
 752         /*
 753          * Sample the clock to take the difference with the expiry time.
 754          */
 755         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 756                 cpu_clock_sample(timer->it_clock, p, &now);
 757         } else {
 758                 struct sighand_struct *sighand;
 759                 unsigned long flags;
 760
 761                 /*
 762                  * Protect against sighand release/switch in exit/exec and
 763                  * also make timer sampling safe if it ends up calling
 764                  * thread_group_cputime_t().
 765                  */
 766                 sighand = lock_task_sighand(p, &flags);
 767                 if (unlikely(sighand == NULL)) {
 768                         /*
 769                          * The process has been reaped.
 770                          * We can't even collect a sample any more.
 771                          * Call the timer disarmed, nothing else to do.
 772                          */
 773                         timer->it.cpu.expires = 0;
 774                         sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 775                                            &itp->it_value);
 776                         return;
 777                 } else {
 778                         cpu_timer_sample_group(timer->it_clock, p, &now);
 779                         unlock_task_sighand(p, &flags);
 780                 }
 781         }
 782
 783         if (now < timer->it.cpu.expires) {
 784                 sample_to_timespec(timer->it_clock,
 785                                    timer->it.cpu.expires - now,
 786                                    &itp->it_value);
 787         } else {
 788                 /*
 789                  * The timer should have expired already, but the firing
 790                  * hasn't taken place yet.  Say it's just about to expire.
 791                  */
 792                 itp->it_value.tv_nsec = 1;
 793                 itp->it_value.tv_sec = 0;
 794         }
 795 }
 796
 797 static unsigned long long
 798 check_timers_list(struct list_head *timers,
 799                   struct list_head *firing,
 800                   unsigned long long curr)
 801 {
 802         int maxfire = 20;
 803
 804         while (!list_empty(timers)) {
 805                 struct cpu_timer_list *t;
 806
 807                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 808
 809                 if (!--maxfire || curr < t->expires)
 810                         return t->expires;
 811
 812                 t->firing = 1;
 813                 list_move_tail(&t->entry, firing);
 814         }
 815
 816         return 0;
 817 }
 818
 819 /*
 820  * Check for any per-thread CPU timers that have fired and move them off
 821  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 822  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 823  */
 824 static void check_thread_timers(struct task_struct *tsk,
 825                                 struct list_head *firing)
 826 {
 827         struct list_head *timers = tsk->cpu_timers;
 828         struct signal_struct *const sig = tsk->signal;
 829         struct task_cputime_t *tsk_expires = &tsk->cputime_expires;
 830         unsigned long long expires;
 831         unsigned long soft;
 832
 833         /*
 834          * If cputime_expires is zero, then there are no active
 835          * per thread CPU timers.
 836          */
 837         if (task_cputime_zero(&tsk->cputime_expires))
 838                 return;
 839
 840         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 841         tsk_expires->prof_exp = expires_to_cputime(expires);
 842
 843         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 844         tsk_expires->virt_exp = expires_to_cputime(expires);
 845
 846         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 847                                                    tsk->se.sum_exec_runtime);
 848
 849         /*
 850          * Check for the special case thread timers.
 851          */
 852         soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 853         if (soft != RLIM_INFINITY) {
 854                 unsigned long hard =
 855                         READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 856
 857                 if (hard != RLIM_INFINITY &&
 858                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 859                         /*
 860                          * At the hard limit, we just die.
 861                          * No need to calculate anything else now.
 862                          */
 863                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 864                         return;
 865                 }
 866                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 867                         /*
 868                          * At the soft limit, send a SIGXCPU every second.
 869                          */
 870                         if (soft < hard) {
 871                                 soft += USEC_PER_SEC;
 872                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 873                         }
 874                         printk(KERN_INFO
 875                                 "RT Watchdog Timeout: %s[%d]\n",
 876                                 tsk->comm, task_pid_nr(tsk));
 877                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 878                 }
 879         }
 880         if (task_cputime_zero(tsk_expires))
 881                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 882 }
 883
 884 static inline void stop_process_timers(struct signal_struct *sig)
 885 {
 886         struct thread_group_cputimer *cputimer = &sig->cputimer;
 887
 888         /* Turn off cputimer->running. This is done without locking. */
 889         WRITE_ONCE(cputimer->running, false);
 890         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 891 }
 892
 893 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 894                              unsigned long long *expires,
 895                              unsigned long long cur_time, int signo)
 896 {
 897         if (!it->expires)
 898                 return;
 899
 900         if (cur_time >= it->expires) {
 901                 if (it->incr) {
 902                         it->expires += it->incr;
 903                         it->error += it->incr_error;
 904                         if (it->error >= TICK_NSEC) {
 905                                 it->expires -= cputime_one_jiffy;
 906                                 it->error -= TICK_NSEC;
 907                         }
 908                 } else {
 909                         it->expires = 0;
 910                 }
 911
 912                 trace_itimer_expire(signo == SIGPROF ?
 913                                     ITIMER_PROF : ITIMER_VIRTUAL,
 914                                     tsk->signal->leader_pid, cur_time);
 915                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 916         }
 917
 918         if (it->expires && (!*expires || it->expires < *expires)) {
 919                 *expires = it->expires;
 920         }
 921 }
 922
 923 /*
 924  * Check for any per-thread CPU timers that have fired and move them
 925  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 926  * have already been taken off.
 927  */
 928 static void check_process_timers(struct task_struct *tsk,
 929                                  struct list_head *firing)
 930 {
 931         struct signal_struct *const sig = tsk->signal;
 932         unsigned long long utime, ptime, virt_expires, prof_expires;
 933         unsigned long long sum_sched_runtime, sched_expires;
 934         struct list_head *timers = sig->cpu_timers;
 935         struct task_cputime_t cputime;
 936         unsigned long soft;
 937
 938         /*
 939          * If cputimer is not running, then there are no active
 940          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 941          */
 942         if (!READ_ONCE(tsk->signal->cputimer.running))
 943                 return;
 944
 945         /*
 946          * Signify that a thread is checking for process timers.
 947          * Write access to this field is protected by the sighand lock.
 948          */
 949         sig->cputimer.checking_timer = true;
 950
 951         /*
 952          * Collect the current process totals.
 953          */
 954         thread_group_cputimer(tsk, &cputime);
 955         utime = cputime_to_expires(cputime.utime);
 956         ptime = utime + cputime_to_expires(cputime.stime);
 957         sum_sched_runtime = cputime.sum_exec_runtime;
 958
 959         prof_expires = check_timers_list(timers, firing, ptime);
 960         virt_expires = check_timers_list(++timers, firing, utime);
 961         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 962
 963         /*
 964          * Check for the special case process timers.
 965          */
 966         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 967                          SIGPROF);
 968         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 969                          SIGVTALRM);
 970         soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 971         if (soft != RLIM_INFINITY) {
 972                 unsigned long psecs = cputime_to_secs(ptime);
 973                 unsigned long hard =
 974                         READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 975                 cputime_t x;
 976                 if (psecs >= hard) {
 977                         /*
 978                          * At the hard limit, we just die.
 979                          * No need to calculate anything else now.
 980                          */
 981                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 982                         return;
 983                 }
 984                 if (psecs >= soft) {
 985                         /*
 986                          * At the soft limit, send a SIGXCPU every second.
 987                          */
 988                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 989                         if (soft < hard) {
 990                                 soft++;
 991                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 992                         }
 993                 }
 994                 x = secs_to_cputime(soft);
 995                 if (!prof_expires || x < prof_expires) {
 996                         prof_expires = x;
 997                 }
 998         }
 999
1000         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1001         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1002         sig->cputime_expires.sched_exp = sched_expires;
1003         if (task_cputime_zero(&sig->cputime_expires))
1004                 stop_process_timers(sig);
1005
1006         sig->cputimer.checking_timer = false;
1007 }
1008
1009 /*
1010  * This is called from the signal code (via do_schedule_next_timer)
1011  * when the last timer signal was delivered and we have to reload the timer.
1012  */
1013 void posix_cpu_timer_schedule(struct k_itimer *timer)
1014 {
1015         struct sighand_struct *sighand;
1016         unsigned long flags;
1017         struct task_struct *p = timer->it.cpu.task;
1018         unsigned long long now;
1019
1020         WARN_ON_ONCE(p == NULL);
1021
1022         /*
1023          * Fetch the current sample and update the timer's expiry time.
1024          */
1025         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1026                 cpu_clock_sample(timer->it_clock, p, &now);
1027                 bump_cpu_timer(timer, now);
1028                 if (unlikely(p->exit_state))
1029                         goto out;
1030
1031                 /* Protect timer list r/w in arm_timer() */
1032                 sighand = lock_task_sighand(p, &flags);
1033                 if (!sighand)
1034                         goto out;
1035         } else {
1036                 /*
1037                  * Protect arm_timer() and timer sampling in case of call to
1038                  * thread_group_cputime_t().
1039                  */
1040                 sighand = lock_task_sighand(p, &flags);
1041                 if (unlikely(sighand == NULL)) {
1042                         /*
1043                          * The process has been reaped.
1044                          * We can't even collect a sample any more.
1045                          */
1046                         timer->it.cpu.expires = 0;
1047                         goto out;
1048                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1049                         unlock_task_sighand(p, &flags);
1050                         /* Optimizations: if the process is dying, no need to rearm */
1051                         goto out;
1052                 }
1053                 cpu_timer_sample_group(timer->it_clock, p, &now);
1054                 bump_cpu_timer(timer, now);
1055                 /* Leave the sighand locked for the call below.  */
1056         }
1057
1058         /*
1059          * Now re-arm for the new expiry time.
1060          */
1061         WARN_ON_ONCE(!irqs_disabled());
1062         arm_timer(timer);
1063         unlock_task_sighand(p, &flags);
1064
1065 out:
1066         timer->it_overrun_last = timer->it_overrun;
1067         timer->it_overrun = -1;
1068         ++timer->it_requeue_pending;
1069 }
1070
1071 /**
1072  * task_cputime_expired - Compare two task_cputime entities.
1073  *
1074  * @sample:     The task_cputime structure to be checked for expiration.
1075  * @expires:    Expiration times, against which @sample will be checked.
1076  *
1077  * Checks @sample against @expires to see if any field of @sample has expired.
1078  * Returns true if any field of the former is greater than the corresponding
1079  * field of the latter if the latter field is set.  Otherwise returns false.
1080  */
1081 static inline int task_cputime_expired(const struct task_cputime_t *sample,
1082                                         const struct task_cputime_t *expires)
1083 {
1084         if (expires->utime && sample->utime >= expires->utime)
1085                 return 1;
1086         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1087                 return 1;
1088         if (expires->sum_exec_runtime != 0 &&
1089             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1090                 return 1;
1091         return 0;
1092 }
1093
1094 /**
1095  * fastpath_timer_check - POSIX CPU timers fast path.
1096  *
1097  * @tsk:        The task (thread) being checked.
1098  *
1099  * Check the task and thread group timers.  If both are zero (there are no
1100  * timers set) return false.  Otherwise snapshot the task and thread group
1101  * timers and compare them with the corresponding expiration times.  Return
1102  * true if a timer has expired, else return false.
1103  */
1104 static inline int fastpath_timer_check(struct task_struct *tsk)
1105 {
1106         struct signal_struct *sig;
1107
1108         if (!task_cputime_zero(&tsk->cputime_expires)) {
1109                 struct task_cputime_t task_sample;
1110
1111                 task_cputime_t(tsk, &task_sample.utime, &task_sample.stime);
1112                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1113                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1114                         return 1;
1115         }
1116
1117         sig = tsk->signal;
1118         /*
1119          * Check if thread group timers expired when the cputimer is
1120          * running and no other thread in the group is already checking
1121          * for thread group cputimers. These fields are read without the
1122          * sighand lock. However, this is fine because this is meant to
1123          * be a fastpath heuristic to determine whether we should try to
1124          * acquire the sighand lock to check/handle timers.
1125          *
1126          * In the worst case scenario, if 'running' or 'checking_timer' gets
1127          * set but the current thread doesn't see the change yet, we'll wait
1128          * until the next thread in the group gets a scheduler interrupt to
1129          * handle the timer. This isn't an issue in practice because these
1130          * types of delays with signals actually getting sent are expected.
1131          */
1132         if (READ_ONCE(sig->cputimer.running) &&
1133             !READ_ONCE(sig->cputimer.checking_timer)) {
1134                 struct task_cputime_t group_sample;
1135
1136                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1137
1138                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1139                         return 1;
1140         }
1141
1142         return 0;
1143 }
1144
1145 /*
1146  * This is called from the timer interrupt handler.  The irq handler has
1147  * already updated our counts.  We need to check if any timers fire now.
1148  * Interrupts are disabled.
1149  */
1150 void run_posix_cpu_timers(struct task_struct *tsk)
1151 {
1152         LIST_HEAD(firing);
1153         struct k_itimer *timer, *next;
1154         unsigned long flags;
1155
1156         WARN_ON_ONCE(!irqs_disabled());
1157
1158         /*
1159          * The fast path checks that there are no expired thread or thread
1160          * group timers.  If that's so, just return.
1161          */
1162         if (!fastpath_timer_check(tsk))
1163                 return;
1164
1165         if (!lock_task_sighand(tsk, &flags))
1166                 return;
1167         /*
1168          * Here we take off tsk->signal->cpu_timers[N] and
1169          * tsk->cpu_timers[N] all the timers that are firing, and
1170          * put them on the firing list.
1171          */
1172         check_thread_timers(tsk, &firing);
1173
1174         check_process_timers(tsk, &firing);
1175
1176         /*
1177          * We must release these locks before taking any timer's lock.
1178          * There is a potential race with timer deletion here, as the
1179          * siglock now protects our private firing list.  We have set
1180          * the firing flag in each timer, so that a deletion attempt
1181          * that gets the timer lock before we do will give it up and
1182          * spin until we've taken care of that timer below.
1183          */
1184         unlock_task_sighand(tsk, &flags);
1185
1186         /*
1187          * Now that all the timers on our list have the firing flag,
1188          * no one will touch their list entries but us.  We'll take
1189          * each timer's lock before clearing its firing flag, so no
1190          * timer call will interfere.
1191          */
1192         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1193                 int cpu_firing;
1194
1195                 spin_lock(&timer->it_lock);
1196                 list_del_init(&timer->it.cpu.entry);
1197                 cpu_firing = timer->it.cpu.firing;
1198                 timer->it.cpu.firing = 0;
1199                 /*
1200                  * The firing flag is -1 if we collided with a reset
1201                  * of the timer, which already reported this
1202                  * almost-firing as an overrun.  So don't generate an event.
1203                  */
1204                 if (likely(cpu_firing >= 0))
1205                         cpu_timer_fire(timer);
1206                 spin_unlock(&timer->it_lock);
1207         }
1208 }
1209
1210 /*
1211  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1212  * The tsk->sighand->siglock must be held by the caller.
1213  */
1214 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1215                            cputime_t *newval, cputime_t *oldval)
1216 {
1217         unsigned long long now;
1218
1219         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1220         cpu_timer_sample_group(clock_idx, tsk, &now);
1221
1222         if (oldval) {
1223                 /*
1224                  * We are setting itimer. The *oldval is absolute and we update
1225                  * it to be relative, *newval argument is relative and we update
1226                  * it to be absolute.
1227                  */
1228                 if (*oldval) {
1229                         if (*oldval <= now) {
1230                                 /* Just about to fire. */
1231                                 *oldval = cputime_one_jiffy;
1232                         } else {
1233                                 *oldval -= now;
1234                         }
1235                 }
1236
1237                 if (!*newval)
1238                         return;
1239                 *newval += now;
1240         }
1241
1242         /*
1243          * Update expiration cache if we are the earliest timer, or eventually
1244          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1245          */
1246         switch (clock_idx) {
1247         case CPUCLOCK_PROF:
1248                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1249                         tsk->signal->cputime_expires.prof_exp = *newval;
1250                 break;
1251         case CPUCLOCK_VIRT:
1252                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1253                         tsk->signal->cputime_expires.virt_exp = *newval;
1254                 break;
1255         }
1256
1257         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1258 }
1259
1260 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1261                             struct timespec *rqtp, struct itimerspec *it)
1262 {
1263         struct k_itimer timer;
1264         int error;
1265
1266         /*
1267          * Set up a temporary timer and then wait for it to go off.
1268          */
1269         memset(&timer, 0, sizeof timer);
1270         spin_lock_init(&timer.it_lock);
1271         timer.it_clock = which_clock;
1272         timer.it_overrun = -1;
1273         error = posix_cpu_timer_create(&timer);
1274         timer.it_process = current;
1275         if (!error) {
1276                 static struct itimerspec zero_it;
1277
1278                 memset(it, 0, sizeof *it);
1279                 it->it_value = *rqtp;
1280
1281                 spin_lock_irq(&timer.it_lock);
1282                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1283                 if (error) {
1284                         spin_unlock_irq(&timer.it_lock);
1285                         return error;
1286                 }
1287
1288                 while (!signal_pending(current)) {
1289                         if (timer.it.cpu.expires == 0) {
1290                                 /*
1291                                  * Our timer fired and was reset, below
1292                                  * deletion can not fail.
1293                                  */
1294                                 posix_cpu_timer_del(&timer);
1295                                 spin_unlock_irq(&timer.it_lock);
1296                                 return 0;
1297                         }
1298
1299                         /*
1300                          * Block until cpu_timer_fire (or a signal) wakes us.
1301                          */
1302                         __set_current_state(TASK_INTERRUPTIBLE);
1303                         spin_unlock_irq(&timer.it_lock);
1304                         schedule();
1305                         spin_lock_irq(&timer.it_lock);
1306                 }
1307
1308                 /*
1309                  * We were interrupted by a signal.
1310                  */
1311                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1312                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1313                 if (!error) {
1314                         /*
1315                          * Timer is now unarmed, deletion can not fail.
1316                          */
1317                         posix_cpu_timer_del(&timer);
1318                 }
1319                 spin_unlock_irq(&timer.it_lock);
1320
1321                 while (error == TIMER_RETRY) {
1322                         /*
1323                          * We need to handle case when timer was or is in the
1324                          * middle of firing. In other cases we already freed
1325                          * resources.
1326                          */
1327                         spin_lock_irq(&timer.it_lock);
1328                         error = posix_cpu_timer_del(&timer);
1329                         spin_unlock_irq(&timer.it_lock);
1330                 }
1331
1332                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1333                         /*
1334                          * It actually did fire already.
1335                          */
1336                         return 0;
1337                 }
1338
1339                 error = -ERESTART_RESTARTBLOCK;
1340         }
1341
1342         return error;
1343 }
1344
1345 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1346
1347 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1348                             struct timespec *rqtp, struct timespec __user *rmtp)
1349 {
1350         struct restart_block *restart_block = &current->restart_block;
1351         struct itimerspec it;
1352         int error;
1353
1354         /*
1355          * Diagnose required errors first.
1356          */
1357         if (CPUCLOCK_PERTHREAD(which_clock) &&
1358             (CPUCLOCK_PID(which_clock) == 0 ||
1359              CPUCLOCK_PID(which_clock) == current->pid))
1360                 return -EINVAL;
1361
1362         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1363
1364         if (error == -ERESTART_RESTARTBLOCK) {
1365
1366                 if (flags & TIMER_ABSTIME)
1367                         return -ERESTARTNOHAND;
1368                 /*
1369                  * Report back to the user the time still remaining.
1370                  */
1371                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1372                         return -EFAULT;
1373
1374                 restart_block->fn = posix_cpu_nsleep_restart;
1375                 restart_block->nanosleep.clockid = which_clock;
1376                 restart_block->nanosleep.rmtp = rmtp;
1377                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1378         }
1379         return error;
1380 }
1381
1382 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1383 {
1384         clockid_t which_clock = restart_block->nanosleep.clockid;
1385         struct timespec t;
1386         struct itimerspec it;
1387         int error;
1388
1389         t = ns_to_timespec(restart_block->nanosleep.expires);
1390
1391         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1392
1393         if (error == -ERESTART_RESTARTBLOCK) {
1394                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1395                 /*
1396                  * Report back to the user the time still remaining.
1397                  */
1398                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1399                         return -EFAULT;
1400
1401                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1402         }
1403         return error;
1404
1405 }
1406
1407 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1408 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1409
1410 static int process_cpu_clock_getres(const clockid_t which_clock,
1411                                     struct timespec *tp)
1412 {
1413         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1414 }
1415 static int process_cpu_clock_get(const clockid_t which_clock,
1416                                  struct timespec *tp)
1417 {
1418         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1419 }
1420 static int process_cpu_timer_create(struct k_itimer *timer)
1421 {
1422         timer->it_clock = PROCESS_CLOCK;
1423         return posix_cpu_timer_create(timer);
1424 }
1425 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1426                               struct timespec *rqtp,
1427                               struct timespec __user *rmtp)
1428 {
1429         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1430 }
1431 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1432 {
1433         return -EINVAL;
1434 }
1435 static int thread_cpu_clock_getres(const clockid_t which_clock,
1436                                    struct timespec *tp)
1437 {
1438         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1439 }
1440 static int thread_cpu_clock_get(const clockid_t which_clock,
1441                                 struct timespec *tp)
1442 {
1443         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1444 }
1445 static int thread_cpu_timer_create(struct k_itimer *timer)
1446 {
1447         timer->it_clock = THREAD_CLOCK;
1448         return posix_cpu_timer_create(timer);
1449 }
1450
1451 struct k_clock clock_posix_cpu = {
1452         .clock_getres   = posix_cpu_clock_getres,
1453         .clock_set      = posix_cpu_clock_set,
1454         .clock_get      = posix_cpu_clock_get,
1455         .timer_create   = posix_cpu_timer_create,
1456         .nsleep         = posix_cpu_nsleep,
1457         .nsleep_restart = posix_cpu_nsleep_restart,
1458         .timer_set      = posix_cpu_timer_set,
1459         .timer_del      = posix_cpu_timer_del,
1460         .timer_get      = posix_cpu_timer_get,
1461 };
1462
1463 static __init int init_posix_cpu_timers(void)
1464 {
1465         struct k_clock process = {
1466                 .clock_getres   = process_cpu_clock_getres,
1467                 .clock_get      = process_cpu_clock_get,
1468                 .timer_create   = process_cpu_timer_create,
1469                 .nsleep         = process_cpu_nsleep,
1470                 .nsleep_restart = process_cpu_nsleep_restart,
1471         };
1472         struct k_clock thread = {
1473                 .clock_getres   = thread_cpu_clock_getres,
1474                 .clock_get      = thread_cpu_clock_get,
1475                 .timer_create   = thread_cpu_timer_create,
1476         };
1477
1478         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1479         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1480
1481         return 0;
1482 }
1483 __initcall(init_posix_cpu_timers);