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 <asm/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11 #include <trace/events/timer.h>
  12 #include <linux/random.h>
  13 #include <linux/tick.h>
  14 #include <linux/workqueue.h>
  15
  16 /*
  17  * Called after updating RLIMIT_CPU to run cpu timer and update
  18  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  19  * siglock protection since other code may update expiration cache as
  20  * well.
  21  */
  22 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  23 {
  24         cputime_t cputime = secs_to_cputime(rlim_new);
  25
  26         spin_lock_irq(&task->sighand->siglock);
  27         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  28         spin_unlock_irq(&task->sighand->siglock);
  29 }
  30
  31 static int check_clock(const clockid_t which_clock)
  32 {
  33         int error = 0;
  34         struct task_struct *p;
  35         const pid_t pid = CPUCLOCK_PID(which_clock);
  36
  37         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  38                 return -EINVAL;
  39
  40         if (pid == 0)
  41                 return 0;
  42
  43         rcu_read_lock();
  44         p = find_task_by_vpid(pid);
  45         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  46                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  47                 error = -EINVAL;
  48         }
  49         rcu_read_unlock();
  50
  51         return error;
  52 }
  53
  54 static inline unsigned long long
  55 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  56 {
  57         unsigned long long ret;
  58
  59         ret = 0;                /* high half always zero when .cpu used */
  60         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  61                 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  62         } else {
  63                 ret = cputime_to_expires(timespec_to_cputime(tp));
  64         }
  65         return ret;
  66 }
  67
  68 static void sample_to_timespec(const clockid_t which_clock,
  69                                unsigned long long expires,
  70                                struct timespec *tp)
  71 {
  72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  73                 *tp = ns_to_timespec(expires);
  74         else
  75                 cputime_to_timespec((__force cputime_t)expires, tp);
  76 }
  77
  78 /*
  79  * Update expiry time from increment, and increase overrun count,
  80  * given the current clock sample.
  81  */
  82 static void bump_cpu_timer(struct k_itimer *timer,
  83                            unsigned long long now)
  84 {
  85         int i;
  86         unsigned long long delta, incr;
  87
  88         if (timer->it.cpu.incr == 0)
  89                 return;
  90
  91         if (now < timer->it.cpu.expires)
  92                 return;
  93
  94         incr = timer->it.cpu.incr;
  95         delta = now + incr - timer->it.cpu.expires;
  96
  97         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  98         for (i = 0; incr < delta - incr; i++)
  99                 incr = incr << 1;
 100
 101         for (; i >= 0; incr >>= 1, i--) {
 102                 if (delta < incr)
 103                         continue;
 104
 105                 timer->it.cpu.expires += incr;
 106                 timer->it_overrun += 1 << i;
 107                 delta -= incr;
 108         }
 109 }
 110
 111 /**
 112  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 113  *
 114  * @cputime:    The struct to compare.
 115  *
 116  * Checks @cputime to see if all fields are zero.  Returns true if all fields
 117  * are zero, false if any field is nonzero.
 118  */
 119 static inline int task_cputime_zero(const struct task_cputime *cputime)
 120 {
 121         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 static inline unsigned long long prof_ticks(struct task_struct *p)
 127 {
 128         cputime_t utime, stime;
 129
 130         task_cputime(p, &utime, &stime);
 131
 132         return cputime_to_expires(utime + stime);
 133 }
 134 static inline unsigned long long virt_ticks(struct task_struct *p)
 135 {
 136         cputime_t utime;
 137
 138         task_cputime(p, &utime, NULL);
 139
 140         return cputime_to_expires(utime);
 141 }
 142
 143 static int
 144 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 145 {
 146         int error = check_clock(which_clock);
 147         if (!error) {
 148                 tp->tv_sec = 0;
 149                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 150                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 151                         /*
 152                          * If sched_clock is using a cycle counter, we
 153                          * don't have any idea of its true resolution
 154                          * exported, but it is much more than 1s/HZ.
 155                          */
 156                         tp->tv_nsec = 1;
 157                 }
 158         }
 159         return error;
 160 }
 161
 162 static int
 163 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 164 {
 165         /*
 166          * You can never reset a CPU clock, but we check for other errors
 167          * in the call before failing with EPERM.
 168          */
 169         int error = check_clock(which_clock);
 170         if (error == 0) {
 171                 error = -EPERM;
 172         }
 173         return error;
 174 }
 175
 176
 177 /*
 178  * Sample a per-thread clock for the given task.
 179  */
 180 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 181                             unsigned long long *sample)
 182 {
 183         switch (CPUCLOCK_WHICH(which_clock)) {
 184         default:
 185                 return -EINVAL;
 186         case CPUCLOCK_PROF:
 187                 *sample = prof_ticks(p);
 188                 break;
 189         case CPUCLOCK_VIRT:
 190                 *sample = virt_ticks(p);
 191                 break;
 192         case CPUCLOCK_SCHED:
 193                 *sample = task_sched_runtime(p);
 194                 break;
 195         }
 196         return 0;
 197 }
 198
 199 /*
 200  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 201  * to avoid race conditions with concurrent updates to cputime.
 202  */
 203 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 204 {
 205         u64 curr_cputime;
 206 retry:
 207         curr_cputime = atomic64_read(cputime);
 208         if (sum_cputime > curr_cputime) {
 209                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 210                         goto retry;
 211         }
 212 }
 213
 214 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
 215 {
 216         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 217         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 218         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 219 }
 220
 221 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 222 static inline void sample_cputime_atomic(struct task_cputime *times,
 223                                          struct task_cputime_atomic *atomic_times)
 224 {
 225         times->utime = atomic64_read(&atomic_times->utime);
 226         times->stime = atomic64_read(&atomic_times->stime);
 227         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 228 }
 229
 230 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 231 {
 232         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 233         struct task_cputime sum;
 234
 235         /* Check if cputimer isn't running. This is accessed without locking. */
 236         if (!READ_ONCE(cputimer->running)) {
 237                 /*
 238                  * The POSIX timer interface allows for absolute time expiry
 239                  * values through the TIMER_ABSTIME flag, therefore we have
 240                  * to synchronize the timer to the clock every time we start it.
 241                  */
 242                 thread_group_cputime(tsk, &sum);
 243                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 244
 245                 /*
 246                  * We're setting cputimer->running without a lock. Ensure
 247                  * this only gets written to in one operation. We set
 248                  * running after update_gt_cputime() as a small optimization,
 249                  * but barriers are not required because update_gt_cputime()
 250                  * can handle concurrent updates.
 251                  */
 252                 WRITE_ONCE(cputimer->running, true);
 253         }
 254         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 255 }
 256
 257 /*
 258  * Sample a process (thread group) clock for the given group_leader task.
 259  * Must be called with task sighand lock held for safe while_each_thread()
 260  * traversal.
 261  */
 262 static int cpu_clock_sample_group(const clockid_t which_clock,
 263                                   struct task_struct *p,
 264                                   unsigned long long *sample)
 265 {
 266         struct task_cputime cputime;
 267
 268         switch (CPUCLOCK_WHICH(which_clock)) {
 269         default:
 270                 return -EINVAL;
 271         case CPUCLOCK_PROF:
 272                 thread_group_cputime(p, &cputime);
 273                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 274                 break;
 275         case CPUCLOCK_VIRT:
 276                 thread_group_cputime(p, &cputime);
 277                 *sample = cputime_to_expires(cputime.utime);
 278                 break;
 279         case CPUCLOCK_SCHED:
 280                 thread_group_cputime(p, &cputime);
 281                 *sample = cputime.sum_exec_runtime;
 282                 break;
 283         }
 284         return 0;
 285 }
 286
 287 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 288                                     const clockid_t which_clock,
 289                                     struct timespec *tp)
 290 {
 291         int err = -EINVAL;
 292         unsigned long long rtn;
 293
 294         if (CPUCLOCK_PERTHREAD(which_clock)) {
 295                 if (same_thread_group(tsk, current))
 296                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 297         } else {
 298                 if (tsk == current || thread_group_leader(tsk))
 299                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 300         }
 301
 302         if (!err)
 303                 sample_to_timespec(which_clock, rtn, tp);
 304
 305         return err;
 306 }
 307
 308
 309 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 310 {
 311         const pid_t pid = CPUCLOCK_PID(which_clock);
 312         int err = -EINVAL;
 313
 314         if (pid == 0) {
 315                 /*
 316                  * Special case constant value for our own clocks.
 317                  * We don't have to do any lookup to find ourselves.
 318                  */
 319                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 320         } else {
 321                 /*
 322                  * Find the given PID, and validate that the caller
 323                  * should be able to see it.
 324                  */
 325                 struct task_struct *p;
 326                 rcu_read_lock();
 327                 p = find_task_by_vpid(pid);
 328                 if (p)
 329                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 330                 rcu_read_unlock();
 331         }
 332
 333         return err;
 334 }
 335
 336 /*
 337  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 338  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 339  * new timer already all-zeros initialized.
 340  */
 341 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 342 {
 343         int ret = 0;
 344         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 345         struct task_struct *p;
 346
 347         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 348                 return -EINVAL;
 349
 350         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 351
 352         rcu_read_lock();
 353         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 354                 if (pid == 0) {
 355                         p = current;
 356                 } else {
 357                         p = find_task_by_vpid(pid);
 358                         if (p && !same_thread_group(p, current))
 359                                 p = NULL;
 360                 }
 361         } else {
 362                 if (pid == 0) {
 363                         p = current->group_leader;
 364                 } else {
 365                         p = find_task_by_vpid(pid);
 366                         if (p && !has_group_leader_pid(p))
 367                                 p = NULL;
 368                 }
 369         }
 370         new_timer->it.cpu.task = p;
 371         if (p) {
 372                 get_task_struct(p);
 373         } else {
 374                 ret = -EINVAL;
 375         }
 376         rcu_read_unlock();
 377
 378         return ret;
 379 }
 380
 381 /*
 382  * Clean up a CPU-clock timer that is about to be destroyed.
 383  * This is called from timer deletion with the timer already locked.
 384  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 385  * and try again.  (This happens when the timer is in the middle of firing.)
 386  */
 387 static int posix_cpu_timer_del(struct k_itimer *timer)
 388 {
 389         int ret = 0;
 390         unsigned long flags;
 391         struct sighand_struct *sighand;
 392         struct task_struct *p = timer->it.cpu.task;
 393
 394         WARN_ON_ONCE(p == NULL);
 395
 396         /*
 397          * Protect against sighand release/switch in exit/exec and process/
 398          * thread timer list entry concurrent read/writes.
 399          */
 400         sighand = lock_task_sighand(p, &flags);
 401         if (unlikely(sighand == NULL)) {
 402                 /*
 403                  * We raced with the reaping of the task.
 404                  * The deletion should have cleared us off the list.
 405                  */
 406                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 407         } else {
 408                 if (timer->it.cpu.firing)
 409                         ret = TIMER_RETRY;
 410                 else
 411                         list_del(&timer->it.cpu.entry);
 412
 413                 unlock_task_sighand(p, &flags);
 414         }
 415
 416         if (!ret)
 417                 put_task_struct(p);
 418
 419         return ret;
 420 }
 421
 422 static void cleanup_timers_list(struct list_head *head)
 423 {
 424         struct cpu_timer_list *timer, *next;
 425
 426         list_for_each_entry_safe(timer, next, head, entry)
 427                 list_del_init(&timer->entry);
 428 }
 429
 430 /*
 431  * Clean out CPU timers still ticking when a thread exited.  The task
 432  * pointer is cleared, and the expiry time is replaced with the residual
 433  * time for later timer_gettime calls to return.
 434  * This must be called with the siglock held.
 435  */
 436 static void cleanup_timers(struct list_head *head)
 437 {
 438         cleanup_timers_list(head);
 439         cleanup_timers_list(++head);
 440         cleanup_timers_list(++head);
 441 }
 442
 443 /*
 444  * These are both called with the siglock held, when the current thread
 445  * is being reaped.  When the final (leader) thread in the group is reaped,
 446  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 447  */
 448 void posix_cpu_timers_exit(struct task_struct *tsk)
 449 {
 450         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 451                                                 sizeof(unsigned long long));
 452         cleanup_timers(tsk->cpu_timers);
 453
 454 }
 455 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 456 {
 457         cleanup_timers(tsk->signal->cpu_timers);
 458 }
 459
 460 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 461 {
 462         return expires == 0 || expires > new_exp;
 463 }
 464
 465 /*
 466  * Insert the timer on the appropriate list before any timers that
 467  * expire later.  This must be called with the sighand lock held.
 468  */
 469 static void arm_timer(struct k_itimer *timer)
 470 {
 471         struct task_struct *p = timer->it.cpu.task;
 472         struct list_head *head, *listpos;
 473         struct task_cputime *cputime_expires;
 474         struct cpu_timer_list *const nt = &timer->it.cpu;
 475         struct cpu_timer_list *next;
 476
 477         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 478                 head = p->cpu_timers;
 479                 cputime_expires = &p->cputime_expires;
 480         } else {
 481                 head = p->signal->cpu_timers;
 482                 cputime_expires = &p->signal->cputime_expires;
 483         }
 484         head += CPUCLOCK_WHICH(timer->it_clock);
 485
 486         listpos = head;
 487         list_for_each_entry(next, head, entry) {
 488                 if (nt->expires < next->expires)
 489                         break;
 490                 listpos = &next->entry;
 491         }
 492         list_add(&nt->entry, listpos);
 493
 494         if (listpos == head) {
 495                 unsigned long long exp = nt->expires;
 496
 497                 /*
 498                  * We are the new earliest-expiring POSIX 1.b timer, hence
 499                  * need to update expiration cache. Take into account that
 500                  * for process timers we share expiration cache with itimers
 501                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 502                  */
 503
 504                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 505                 case CPUCLOCK_PROF:
 506                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 507                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 508                         break;
 509                 case CPUCLOCK_VIRT:
 510                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 511                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 512                         break;
 513                 case CPUCLOCK_SCHED:
 514                         if (cputime_expires->sched_exp == 0 ||
 515                             cputime_expires->sched_exp > exp)
 516                                 cputime_expires->sched_exp = exp;
 517                         break;
 518                 }
 519                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 520                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 521                 else
 522                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 523         }
 524 }
 525
 526 /*
 527  * The timer is locked, fire it and arrange for its reload.
 528  */
 529 static void cpu_timer_fire(struct k_itimer *timer)
 530 {
 531         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 532                 /*
 533                  * User don't want any signal.
 534                  */
 535                 timer->it.cpu.expires = 0;
 536         } else if (unlikely(timer->sigq == NULL)) {
 537                 /*
 538                  * This a special case for clock_nanosleep,
 539                  * not a normal timer from sys_timer_create.
 540                  */
 541                 wake_up_process(timer->it_process);
 542                 timer->it.cpu.expires = 0;
 543         } else if (timer->it.cpu.incr == 0) {
 544                 /*
 545                  * One-shot timer.  Clear it as soon as it's fired.
 546                  */
 547                 posix_timer_event(timer, 0);
 548                 timer->it.cpu.expires = 0;
 549         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 550                 /*
 551                  * The signal did not get queued because the signal
 552                  * was ignored, so we won't get any callback to
 553                  * reload the timer.  But we need to keep it
 554                  * ticking in case the signal is deliverable next time.
 555                  */
 556                 posix_cpu_timer_schedule(timer);
 557         }
 558 }
 559
 560 /*
 561  * Sample a process (thread group) timer for the given group_leader task.
 562  * Must be called with task sighand lock held for safe while_each_thread()
 563  * traversal.
 564  */
 565 static int cpu_timer_sample_group(const clockid_t which_clock,
 566                                   struct task_struct *p,
 567                                   unsigned long long *sample)
 568 {
 569         struct task_cputime cputime;
 570
 571         thread_group_cputimer(p, &cputime);
 572         switch (CPUCLOCK_WHICH(which_clock)) {
 573         default:
 574                 return -EINVAL;
 575         case CPUCLOCK_PROF:
 576                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 577                 break;
 578         case CPUCLOCK_VIRT:
 579                 *sample = cputime_to_expires(cputime.utime);
 580                 break;
 581         case CPUCLOCK_SCHED:
 582                 *sample = cputime.sum_exec_runtime;
 583                 break;
 584         }
 585         return 0;
 586 }
 587
 588 /*
 589  * Guts of sys_timer_settime for CPU timers.
 590  * This is called with the timer locked and interrupts disabled.
 591  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 592  * and try again.  (This happens when the timer is in the middle of firing.)
 593  */
 594 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 595                                struct itimerspec *new, struct itimerspec *old)
 596 {
 597         unsigned long flags;
 598         struct sighand_struct *sighand;
 599         struct task_struct *p = timer->it.cpu.task;
 600         unsigned long long old_expires, new_expires, old_incr, val;
 601         int ret;
 602
 603         WARN_ON_ONCE(p == NULL);
 604
 605         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 606
 607         /*
 608          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 609          * and p->signal->cpu_timers read/write in arm_timer()
 610          */
 611         sighand = lock_task_sighand(p, &flags);
 612         /*
 613          * If p has just been reaped, we can no
 614          * longer get any information about it at all.
 615          */
 616         if (unlikely(sighand == NULL)) {
 617                 return -ESRCH;
 618         }
 619
 620         /*
 621          * Disarm any old timer after extracting its expiry time.
 622          */
 623         WARN_ON_ONCE(!irqs_disabled());
 624
 625         ret = 0;
 626         old_incr = timer->it.cpu.incr;
 627         old_expires = timer->it.cpu.expires;
 628         if (unlikely(timer->it.cpu.firing)) {
 629                 timer->it.cpu.firing = -1;
 630                 ret = TIMER_RETRY;
 631         } else
 632                 list_del_init(&timer->it.cpu.entry);
 633
 634         /*
 635          * We need to sample the current value to convert the new
 636          * value from to relative and absolute, and to convert the
 637          * old value from absolute to relative.  To set a process
 638          * timer, we need a sample to balance the thread expiry
 639          * times (in arm_timer).  With an absolute time, we must
 640          * check if it's already passed.  In short, we need a sample.
 641          */
 642         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 643                 cpu_clock_sample(timer->it_clock, p, &val);
 644         } else {
 645                 cpu_timer_sample_group(timer->it_clock, p, &val);
 646         }
 647
 648         if (old) {
 649                 if (old_expires == 0) {
 650                         old->it_value.tv_sec = 0;
 651                         old->it_value.tv_nsec = 0;
 652                 } else {
 653                         /*
 654                          * Update the timer in case it has
 655                          * overrun already.  If it has,
 656                          * we'll report it as having overrun
 657                          * and with the next reloaded timer
 658                          * already ticking, though we are
 659                          * swallowing that pending
 660                          * notification here to install the
 661                          * new setting.
 662                          */
 663                         bump_cpu_timer(timer, val);
 664                         if (val < timer->it.cpu.expires) {
 665                                 old_expires = timer->it.cpu.expires - val;
 666                                 sample_to_timespec(timer->it_clock,
 667                                                    old_expires,
 668                                                    &old->it_value);
 669                         } else {
 670                                 old->it_value.tv_nsec = 1;
 671                                 old->it_value.tv_sec = 0;
 672                         }
 673                 }
 674         }
 675
 676         if (unlikely(ret)) {
 677                 /*
 678                  * We are colliding with the timer actually firing.
 679                  * Punt after filling in the timer's old value, and
 680                  * disable this firing since we are already reporting
 681                  * it as an overrun (thanks to bump_cpu_timer above).
 682                  */
 683                 unlock_task_sighand(p, &flags);
 684                 goto out;
 685         }
 686
 687         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 688                 new_expires += val;
 689         }
 690
 691         /*
 692          * Install the new expiry time (or zero).
 693          * For a timer with no notification action, we don't actually
 694          * arm the timer (we'll just fake it for timer_gettime).
 695          */
 696         timer->it.cpu.expires = new_expires;
 697         if (new_expires != 0 && val < new_expires) {
 698                 arm_timer(timer);
 699         }
 700
 701         unlock_task_sighand(p, &flags);
 702         /*
 703          * Install the new reload setting, and
 704          * set up the signal and overrun bookkeeping.
 705          */
 706         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 707                                                 &new->it_interval);
 708
 709         /*
 710          * This acts as a modification timestamp for the timer,
 711          * so any automatic reload attempt will punt on seeing
 712          * that we have reset the timer manually.
 713          */
 714         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 715                 ~REQUEUE_PENDING;
 716         timer->it_overrun_last = 0;
 717         timer->it_overrun = -1;
 718
 719         if (new_expires != 0 && !(val < new_expires)) {
 720                 /*
 721                  * The designated time already passed, so we notify
 722                  * immediately, even if the thread never runs to
 723                  * accumulate more time on this clock.
 724                  */
 725                 cpu_timer_fire(timer);
 726         }
 727
 728         ret = 0;
 729  out:
 730         if (old) {
 731                 sample_to_timespec(timer->it_clock,
 732                                    old_incr, &old->it_interval);
 733         }
 734
 735         return ret;
 736 }
 737
 738 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 739 {
 740         unsigned long long now;
 741         struct task_struct *p = timer->it.cpu.task;
 742
 743         WARN_ON_ONCE(p == NULL);
 744
 745         /*
 746          * Easy part: convert the reload time.
 747          */
 748         sample_to_timespec(timer->it_clock,
 749                            timer->it.cpu.incr, &itp->it_interval);
 750
 751         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 752                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 753                 return;
 754         }
 755
 756         /*
 757          * Sample the clock to take the difference with the expiry time.
 758          */
 759         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 760                 cpu_clock_sample(timer->it_clock, p, &now);
 761         } else {
 762                 struct sighand_struct *sighand;
 763                 unsigned long flags;
 764
 765                 /*
 766                  * Protect against sighand release/switch in exit/exec and
 767                  * also make timer sampling safe if it ends up calling
 768                  * thread_group_cputime().
 769                  */
 770                 sighand = lock_task_sighand(p, &flags);
 771                 if (unlikely(sighand == NULL)) {
 772                         /*
 773                          * The process has been reaped.
 774                          * We can't even collect a sample any more.
 775                          * Call the timer disarmed, nothing else to do.
 776                          */
 777                         timer->it.cpu.expires = 0;
 778                         sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 779                                            &itp->it_value);
 780                         return;
 781                 } else {
 782                         cpu_timer_sample_group(timer->it_clock, p, &now);
 783                         unlock_task_sighand(p, &flags);
 784                 }
 785         }
 786
 787         if (now < timer->it.cpu.expires) {
 788                 sample_to_timespec(timer->it_clock,
 789                                    timer->it.cpu.expires - now,
 790                                    &itp->it_value);
 791         } else {
 792                 /*
 793                  * The timer should have expired already, but the firing
 794                  * hasn't taken place yet.  Say it's just about to expire.
 795                  */
 796                 itp->it_value.tv_nsec = 1;
 797                 itp->it_value.tv_sec = 0;
 798         }
 799 }
 800
 801 static unsigned long long
 802 check_timers_list(struct list_head *timers,
 803                   struct list_head *firing,
 804                   unsigned long long curr)
 805 {
 806         int maxfire = 20;
 807
 808         while (!list_empty(timers)) {
 809                 struct cpu_timer_list *t;
 810
 811                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 812
 813                 if (!--maxfire || curr < t->expires)
 814                         return t->expires;
 815
 816                 t->firing = 1;
 817                 list_move_tail(&t->entry, firing);
 818         }
 819
 820         return 0;
 821 }
 822
 823 /*
 824  * Check for any per-thread CPU timers that have fired and move them off
 825  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 826  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 827  */
 828 static void check_thread_timers(struct task_struct *tsk,
 829                                 struct list_head *firing)
 830 {
 831         struct list_head *timers = tsk->cpu_timers;
 832         struct signal_struct *const sig = tsk->signal;
 833         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 834         unsigned long long expires;
 835         unsigned long soft;
 836
 837         /*
 838          * If cputime_expires is zero, then there are no active
 839          * per thread CPU timers.
 840          */
 841         if (task_cputime_zero(&tsk->cputime_expires))
 842                 return;
 843
 844         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 845         tsk_expires->prof_exp = expires_to_cputime(expires);
 846
 847         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 848         tsk_expires->virt_exp = expires_to_cputime(expires);
 849
 850         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 851                                                    tsk->se.sum_exec_runtime);
 852
 853         /*
 854          * Check for the special case thread timers.
 855          */
 856         soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 857         if (soft != RLIM_INFINITY) {
 858                 unsigned long hard =
 859                         READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 860
 861                 if (hard != RLIM_INFINITY &&
 862                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 863                         /*
 864                          * At the hard limit, we just die.
 865                          * No need to calculate anything else now.
 866                          */
 867                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 868                         return;
 869                 }
 870                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 871                         /*
 872                          * At the soft limit, send a SIGXCPU every second.
 873                          */
 874                         if (soft < hard) {
 875                                 soft += USEC_PER_SEC;
 876                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 877                         }
 878                         printk(KERN_INFO
 879                                 "RT Watchdog Timeout: %s[%d]\n",
 880                                 tsk->comm, task_pid_nr(tsk));
 881                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 882                 }
 883         }
 884         if (task_cputime_zero(tsk_expires))
 885                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 886 }
 887
 888 static inline void stop_process_timers(struct signal_struct *sig)
 889 {
 890         struct thread_group_cputimer *cputimer = &sig->cputimer;
 891
 892         /* Turn off cputimer->running. This is done without locking. */
 893         WRITE_ONCE(cputimer->running, false);
 894         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 895 }
 896
 897 static u32 onecputick;
 898
 899 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 900                              unsigned long long *expires,
 901                              unsigned long long cur_time, int signo)
 902 {
 903         if (!it->expires)
 904                 return;
 905
 906         if (cur_time >= it->expires) {
 907                 if (it->incr) {
 908                         it->expires += it->incr;
 909                         it->error += it->incr_error;
 910                         if (it->error >= onecputick) {
 911                                 it->expires -= cputime_one_jiffy;
 912                                 it->error -= onecputick;
 913                         }
 914                 } else {
 915                         it->expires = 0;
 916                 }
 917
 918                 trace_itimer_expire(signo == SIGPROF ?
 919                                     ITIMER_PROF : ITIMER_VIRTUAL,
 920                                     tsk->signal->leader_pid, cur_time);
 921                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 922         }
 923
 924         if (it->expires && (!*expires || it->expires < *expires)) {
 925                 *expires = it->expires;
 926         }
 927 }
 928
 929 /*
 930  * Check for any per-thread CPU timers that have fired and move them
 931  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 932  * have already been taken off.
 933  */
 934 static void check_process_timers(struct task_struct *tsk,
 935                                  struct list_head *firing)
 936 {
 937         struct signal_struct *const sig = tsk->signal;
 938         unsigned long long utime, ptime, virt_expires, prof_expires;
 939         unsigned long long sum_sched_runtime, sched_expires;
 940         struct list_head *timers = sig->cpu_timers;
 941         struct task_cputime cputime;
 942         unsigned long soft;
 943
 944         /*
 945          * If cputimer is not running, then there are no active
 946          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 947          */
 948         if (!READ_ONCE(tsk->signal->cputimer.running))
 949                 return;
 950
 951         /*
 952          * Signify that a thread is checking for process timers.
 953          * Write access to this field is protected by the sighand lock.
 954          */
 955         sig->cputimer.checking_timer = true;
 956
 957         /*
 958          * Collect the current process totals.
 959          */
 960         thread_group_cputimer(tsk, &cputime);
 961         utime = cputime_to_expires(cputime.utime);
 962         ptime = utime + cputime_to_expires(cputime.stime);
 963         sum_sched_runtime = cputime.sum_exec_runtime;
 964
 965         prof_expires = check_timers_list(timers, firing, ptime);
 966         virt_expires = check_timers_list(++timers, firing, utime);
 967         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 968
 969         /*
 970          * Check for the special case process timers.
 971          */
 972         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 973                          SIGPROF);
 974         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 975                          SIGVTALRM);
 976         soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 977         if (soft != RLIM_INFINITY) {
 978                 unsigned long psecs = cputime_to_secs(ptime);
 979                 unsigned long hard =
 980                         READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 981                 cputime_t x;
 982                 if (psecs >= hard) {
 983                         /*
 984                          * At the hard limit, we just die.
 985                          * No need to calculate anything else now.
 986                          */
 987                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 988                         return;
 989                 }
 990                 if (psecs >= soft) {
 991                         /*
 992                          * At the soft limit, send a SIGXCPU every second.
 993                          */
 994                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 995                         if (soft < hard) {
 996                                 soft++;
 997                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 998                         }
 999                 }
1000                 x = secs_to_cputime(soft);
1001                 if (!prof_expires || x < prof_expires) {
1002                         prof_expires = x;
1003                 }
1004         }
1005
1006         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1007         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1008         sig->cputime_expires.sched_exp = sched_expires;
1009         if (task_cputime_zero(&sig->cputime_expires))
1010                 stop_process_timers(sig);
1011
1012         sig->cputimer.checking_timer = false;
1013 }
1014
1015 /*
1016  * This is called from the signal code (via do_schedule_next_timer)
1017  * when the last timer signal was delivered and we have to reload the timer.
1018  */
1019 void posix_cpu_timer_schedule(struct k_itimer *timer)
1020 {
1021         struct sighand_struct *sighand;
1022         unsigned long flags;
1023         struct task_struct *p = timer->it.cpu.task;
1024         unsigned long long now;
1025
1026         WARN_ON_ONCE(p == NULL);
1027
1028         /*
1029          * Fetch the current sample and update the timer's expiry time.
1030          */
1031         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1032                 cpu_clock_sample(timer->it_clock, p, &now);
1033                 bump_cpu_timer(timer, now);
1034                 if (unlikely(p->exit_state))
1035                         goto out;
1036
1037                 /* Protect timer list r/w in arm_timer() */
1038                 sighand = lock_task_sighand(p, &flags);
1039                 if (!sighand)
1040                         goto out;
1041         } else {
1042                 /*
1043                  * Protect arm_timer() and timer sampling in case of call to
1044                  * thread_group_cputime().
1045                  */
1046                 sighand = lock_task_sighand(p, &flags);
1047                 if (unlikely(sighand == NULL)) {
1048                         /*
1049                          * The process has been reaped.
1050                          * We can't even collect a sample any more.
1051                          */
1052                         timer->it.cpu.expires = 0;
1053                         goto out;
1054                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1055                         unlock_task_sighand(p, &flags);
1056                         /* Optimizations: if the process is dying, no need to rearm */
1057                         goto out;
1058                 }
1059                 cpu_timer_sample_group(timer->it_clock, p, &now);
1060                 bump_cpu_timer(timer, now);
1061                 /* Leave the sighand locked for the call below.  */
1062         }
1063
1064         /*
1065          * Now re-arm for the new expiry time.
1066          */
1067         WARN_ON_ONCE(!irqs_disabled());
1068         arm_timer(timer);
1069         unlock_task_sighand(p, &flags);
1070
1071 out:
1072         timer->it_overrun_last = timer->it_overrun;
1073         timer->it_overrun = -1;
1074         ++timer->it_requeue_pending;
1075 }
1076
1077 /**
1078  * task_cputime_expired - Compare two task_cputime entities.
1079  *
1080  * @sample:     The task_cputime structure to be checked for expiration.
1081  * @expires:    Expiration times, against which @sample will be checked.
1082  *
1083  * Checks @sample against @expires to see if any field of @sample has expired.
1084  * Returns true if any field of the former is greater than the corresponding
1085  * field of the latter if the latter field is set.  Otherwise returns false.
1086  */
1087 static inline int task_cputime_expired(const struct task_cputime *sample,
1088                                         const struct task_cputime *expires)
1089 {
1090         if (expires->utime && sample->utime >= expires->utime)
1091                 return 1;
1092         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1093                 return 1;
1094         if (expires->sum_exec_runtime != 0 &&
1095             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1096                 return 1;
1097         return 0;
1098 }
1099
1100 /**
1101  * fastpath_timer_check - POSIX CPU timers fast path.
1102  *
1103  * @tsk:        The task (thread) being checked.
1104  *
1105  * Check the task and thread group timers.  If both are zero (there are no
1106  * timers set) return false.  Otherwise snapshot the task and thread group
1107  * timers and compare them with the corresponding expiration times.  Return
1108  * true if a timer has expired, else return false.
1109  */
1110 static inline int fastpath_timer_check(struct task_struct *tsk)
1111 {
1112         struct signal_struct *sig;
1113
1114         if (!task_cputime_zero(&tsk->cputime_expires)) {
1115                 struct task_cputime task_sample;
1116
1117                 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1118                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1119                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1120                         return 1;
1121         }
1122
1123         sig = tsk->signal;
1124         /*
1125          * Check if thread group timers expired when the cputimer is
1126          * running and no other thread in the group is already checking
1127          * for thread group cputimers. These fields are read without the
1128          * sighand lock. However, this is fine because this is meant to
1129          * be a fastpath heuristic to determine whether we should try to
1130          * acquire the sighand lock to check/handle timers.
1131          *
1132          * In the worst case scenario, if 'running' or 'checking_timer' gets
1133          * set but the current thread doesn't see the change yet, we'll wait
1134          * until the next thread in the group gets a scheduler interrupt to
1135          * handle the timer. This isn't an issue in practice because these
1136          * types of delays with signals actually getting sent are expected.
1137          */
1138         if (READ_ONCE(sig->cputimer.running) &&
1139             !READ_ONCE(sig->cputimer.checking_timer)) {
1140                 struct task_cputime group_sample;
1141
1142                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1143
1144                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1145                         return 1;
1146         }
1147
1148         return 0;
1149 }
1150
1151 /*
1152  * This is called from the timer interrupt handler.  The irq handler has
1153  * already updated our counts.  We need to check if any timers fire now.
1154  * Interrupts are disabled.
1155  */
1156 void run_posix_cpu_timers(struct task_struct *tsk)
1157 {
1158         LIST_HEAD(firing);
1159         struct k_itimer *timer, *next;
1160         unsigned long flags;
1161
1162         WARN_ON_ONCE(!irqs_disabled());
1163
1164         /*
1165          * The fast path checks that there are no expired thread or thread
1166          * group timers.  If that's so, just return.
1167          */
1168         if (!fastpath_timer_check(tsk))
1169                 return;
1170
1171         if (!lock_task_sighand(tsk, &flags))
1172                 return;
1173         /*
1174          * Here we take off tsk->signal->cpu_timers[N] and
1175          * tsk->cpu_timers[N] all the timers that are firing, and
1176          * put them on the firing list.
1177          */
1178         check_thread_timers(tsk, &firing);
1179
1180         check_process_timers(tsk, &firing);
1181
1182         /*
1183          * We must release these locks before taking any timer's lock.
1184          * There is a potential race with timer deletion here, as the
1185          * siglock now protects our private firing list.  We have set
1186          * the firing flag in each timer, so that a deletion attempt
1187          * that gets the timer lock before we do will give it up and
1188          * spin until we've taken care of that timer below.
1189          */
1190         unlock_task_sighand(tsk, &flags);
1191
1192         /*
1193          * Now that all the timers on our list have the firing flag,
1194          * no one will touch their list entries but us.  We'll take
1195          * each timer's lock before clearing its firing flag, so no
1196          * timer call will interfere.
1197          */
1198         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1199                 int cpu_firing;
1200
1201                 spin_lock(&timer->it_lock);
1202                 list_del_init(&timer->it.cpu.entry);
1203                 cpu_firing = timer->it.cpu.firing;
1204                 timer->it.cpu.firing = 0;
1205                 /*
1206                  * The firing flag is -1 if we collided with a reset
1207                  * of the timer, which already reported this
1208                  * almost-firing as an overrun.  So don't generate an event.
1209                  */
1210                 if (likely(cpu_firing >= 0))
1211                         cpu_timer_fire(timer);
1212                 spin_unlock(&timer->it_lock);
1213         }
1214 }
1215
1216 /*
1217  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1218  * The tsk->sighand->siglock must be held by the caller.
1219  */
1220 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1221                            cputime_t *newval, cputime_t *oldval)
1222 {
1223         unsigned long long now;
1224
1225         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1226         cpu_timer_sample_group(clock_idx, tsk, &now);
1227
1228         if (oldval) {
1229                 /*
1230                  * We are setting itimer. The *oldval is absolute and we update
1231                  * it to be relative, *newval argument is relative and we update
1232                  * it to be absolute.
1233                  */
1234                 if (*oldval) {
1235                         if (*oldval <= now) {
1236                                 /* Just about to fire. */
1237                                 *oldval = cputime_one_jiffy;
1238                         } else {
1239                                 *oldval -= now;
1240                         }
1241                 }
1242
1243                 if (!*newval)
1244                         return;
1245                 *newval += now;
1246         }
1247
1248         /*
1249          * Update expiration cache if we are the earliest timer, or eventually
1250          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1251          */
1252         switch (clock_idx) {
1253         case CPUCLOCK_PROF:
1254                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1255                         tsk->signal->cputime_expires.prof_exp = *newval;
1256                 break;
1257         case CPUCLOCK_VIRT:
1258                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1259                         tsk->signal->cputime_expires.virt_exp = *newval;
1260                 break;
1261         }
1262
1263         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1264 }
1265
1266 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1267                             struct timespec *rqtp, struct itimerspec *it)
1268 {
1269         struct k_itimer timer;
1270         int error;
1271
1272         /*
1273          * Set up a temporary timer and then wait for it to go off.
1274          */
1275         memset(&timer, 0, sizeof timer);
1276         spin_lock_init(&timer.it_lock);
1277         timer.it_clock = which_clock;
1278         timer.it_overrun = -1;
1279         error = posix_cpu_timer_create(&timer);
1280         timer.it_process = current;
1281         if (!error) {
1282                 static struct itimerspec zero_it;
1283
1284                 memset(it, 0, sizeof *it);
1285                 it->it_value = *rqtp;
1286
1287                 spin_lock_irq(&timer.it_lock);
1288                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1289                 if (error) {
1290                         spin_unlock_irq(&timer.it_lock);
1291                         return error;
1292                 }
1293
1294                 while (!signal_pending(current)) {
1295                         if (timer.it.cpu.expires == 0) {
1296                                 /*
1297                                  * Our timer fired and was reset, below
1298                                  * deletion can not fail.
1299                                  */
1300                                 posix_cpu_timer_del(&timer);
1301                                 spin_unlock_irq(&timer.it_lock);
1302                                 return 0;
1303                         }
1304
1305                         /*
1306                          * Block until cpu_timer_fire (or a signal) wakes us.
1307                          */
1308                         __set_current_state(TASK_INTERRUPTIBLE);
1309                         spin_unlock_irq(&timer.it_lock);
1310                         schedule();
1311                         spin_lock_irq(&timer.it_lock);
1312                 }
1313
1314                 /*
1315                  * We were interrupted by a signal.
1316                  */
1317                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1318                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1319                 if (!error) {
1320                         /*
1321                          * Timer is now unarmed, deletion can not fail.
1322                          */
1323                         posix_cpu_timer_del(&timer);
1324                 }
1325                 spin_unlock_irq(&timer.it_lock);
1326
1327                 while (error == TIMER_RETRY) {
1328                         /*
1329                          * We need to handle case when timer was or is in the
1330                          * middle of firing. In other cases we already freed
1331                          * resources.
1332                          */
1333                         spin_lock_irq(&timer.it_lock);
1334                         error = posix_cpu_timer_del(&timer);
1335                         spin_unlock_irq(&timer.it_lock);
1336                 }
1337
1338                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1339                         /*
1340                          * It actually did fire already.
1341                          */
1342                         return 0;
1343                 }
1344
1345                 error = -ERESTART_RESTARTBLOCK;
1346         }
1347
1348         return error;
1349 }
1350
1351 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1352
1353 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1354                             struct timespec *rqtp, struct timespec __user *rmtp)
1355 {
1356         struct restart_block *restart_block = &current->restart_block;
1357         struct itimerspec it;
1358         int error;
1359
1360         /*
1361          * Diagnose required errors first.
1362          */
1363         if (CPUCLOCK_PERTHREAD(which_clock) &&
1364             (CPUCLOCK_PID(which_clock) == 0 ||
1365              CPUCLOCK_PID(which_clock) == current->pid))
1366                 return -EINVAL;
1367
1368         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1369
1370         if (error == -ERESTART_RESTARTBLOCK) {
1371
1372                 if (flags & TIMER_ABSTIME)
1373                         return -ERESTARTNOHAND;
1374                 /*
1375                  * Report back to the user the time still remaining.
1376                  */
1377                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1378                         return -EFAULT;
1379
1380                 restart_block->fn = posix_cpu_nsleep_restart;
1381                 restart_block->nanosleep.clockid = which_clock;
1382                 restart_block->nanosleep.rmtp = rmtp;
1383                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1384         }
1385         return error;
1386 }
1387
1388 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1389 {
1390         clockid_t which_clock = restart_block->nanosleep.clockid;
1391         struct timespec t;
1392         struct itimerspec it;
1393         int error;
1394
1395         t = ns_to_timespec(restart_block->nanosleep.expires);
1396
1397         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1398
1399         if (error == -ERESTART_RESTARTBLOCK) {
1400                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1401                 /*
1402                  * Report back to the user the time still remaining.
1403                  */
1404                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1405                         return -EFAULT;
1406
1407                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1408         }
1409         return error;
1410
1411 }
1412
1413 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1414 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1415
1416 static int process_cpu_clock_getres(const clockid_t which_clock,
1417                                     struct timespec *tp)
1418 {
1419         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1420 }
1421 static int process_cpu_clock_get(const clockid_t which_clock,
1422                                  struct timespec *tp)
1423 {
1424         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1425 }
1426 static int process_cpu_timer_create(struct k_itimer *timer)
1427 {
1428         timer->it_clock = PROCESS_CLOCK;
1429         return posix_cpu_timer_create(timer);
1430 }
1431 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1432                               struct timespec *rqtp,
1433                               struct timespec __user *rmtp)
1434 {
1435         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1436 }
1437 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1438 {
1439         return -EINVAL;
1440 }
1441 static int thread_cpu_clock_getres(const clockid_t which_clock,
1442                                    struct timespec *tp)
1443 {
1444         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1445 }
1446 static int thread_cpu_clock_get(const clockid_t which_clock,
1447                                 struct timespec *tp)
1448 {
1449         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1450 }
1451 static int thread_cpu_timer_create(struct k_itimer *timer)
1452 {
1453         timer->it_clock = THREAD_CLOCK;
1454         return posix_cpu_timer_create(timer);
1455 }
1456
1457 struct k_clock clock_posix_cpu = {
1458         .clock_getres   = posix_cpu_clock_getres,
1459         .clock_set      = posix_cpu_clock_set,
1460         .clock_get      = posix_cpu_clock_get,
1461         .timer_create   = posix_cpu_timer_create,
1462         .nsleep         = posix_cpu_nsleep,
1463         .nsleep_restart = posix_cpu_nsleep_restart,
1464         .timer_set      = posix_cpu_timer_set,
1465         .timer_del      = posix_cpu_timer_del,
1466         .timer_get      = posix_cpu_timer_get,
1467 };
1468
1469 static __init int init_posix_cpu_timers(void)
1470 {
1471         struct k_clock process = {
1472                 .clock_getres   = process_cpu_clock_getres,
1473                 .clock_get      = process_cpu_clock_get,
1474                 .timer_create   = process_cpu_timer_create,
1475                 .nsleep         = process_cpu_nsleep,
1476                 .nsleep_restart = process_cpu_nsleep_restart,
1477         };
1478         struct k_clock thread = {
1479                 .clock_getres   = thread_cpu_clock_getres,
1480                 .clock_get      = thread_cpu_clock_get,
1481                 .timer_create   = thread_cpu_timer_create,
1482         };
1483         struct timespec ts;
1484
1485         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1486         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1487
1488         cputime_to_timespec(cputime_one_jiffy, &ts);
1489         onecputick = ts.tv_nsec;
1490         WARN_ON(ts.tv_sec != 0);
1491
1492         return 0;
1493 }
1494 __initcall(init_posix_cpu_timers);