kernel/time/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched/signal.h>
   6 #include <linux/sched/cputime.h>
   7 #include <linux/posix-timers.h>
   8 #include <linux/errno.h>
   9 #include <linux/math64.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/kernel_stat.h>
  12 #include <trace/events/timer.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         u64 nsecs = rlim_new * NSEC_PER_SEC;
  25
  26         spin_lock_irq(&task->sighand->siglock);
  27         set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, 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 /*
  55  * Update expiry time from increment, and increase overrun count,
  56  * given the current clock sample.
  57  */
  58 static void bump_cpu_timer(struct k_itimer *timer, u64 now)
  59 {
  60         int i;
  61         u64 delta, incr;
  62
  63         if (timer->it.cpu.incr == 0)
  64                 return;
  65
  66         if (now < timer->it.cpu.expires)
  67                 return;
  68
  69         incr = timer->it.cpu.incr;
  70         delta = now + incr - timer->it.cpu.expires;
  71
  72         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  73         for (i = 0; incr < delta - incr; i++)
  74                 incr = incr << 1;
  75
  76         for (; i >= 0; incr >>= 1, i--) {
  77                 if (delta < incr)
  78                         continue;
  79
  80                 timer->it.cpu.expires += incr;
  81                 timer->it_overrun += 1 << i;
  82                 delta -= incr;
  83         }
  84 }
  85
  86 /**
  87  * task_cputime_zero - Check a task_cputime struct for all zero fields.
  88  *
  89  * @cputime:    The struct to compare.
  90  *
  91  * Checks @cputime to see if all fields are zero.  Returns true if all fields
  92  * are zero, false if any field is nonzero.
  93  */
  94 static inline int task_cputime_zero(const struct task_cputime *cputime)
  95 {
  96         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
  97                 return 1;
  98         return 0;
  99 }
 100
 101 static inline u64 prof_ticks(struct task_struct *p)
 102 {
 103         u64 utime, stime;
 104
 105         task_cputime(p, &utime, &stime);
 106
 107         return utime + stime;
 108 }
 109 static inline u64 virt_ticks(struct task_struct *p)
 110 {
 111         u64 utime, stime;
 112
 113         task_cputime(p, &utime, &stime);
 114
 115         return utime;
 116 }
 117
 118 static int
 119 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
 120 {
 121         int error = check_clock(which_clock);
 122         if (!error) {
 123                 tp->tv_sec = 0;
 124                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 125                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 126                         /*
 127                          * If sched_clock is using a cycle counter, we
 128                          * don't have any idea of its true resolution
 129                          * exported, but it is much more than 1s/HZ.
 130                          */
 131                         tp->tv_nsec = 1;
 132                 }
 133         }
 134         return error;
 135 }
 136
 137 static int
 138 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
 139 {
 140         /*
 141          * You can never reset a CPU clock, but we check for other errors
 142          * in the call before failing with EPERM.
 143          */
 144         int error = check_clock(which_clock);
 145         if (error == 0) {
 146                 error = -EPERM;
 147         }
 148         return error;
 149 }
 150
 151
 152 /*
 153  * Sample a per-thread clock for the given task.
 154  */
 155 static int cpu_clock_sample(const clockid_t which_clock,
 156                             struct task_struct *p, u64 *sample)
 157 {
 158         switch (CPUCLOCK_WHICH(which_clock)) {
 159         default:
 160                 return -EINVAL;
 161         case CPUCLOCK_PROF:
 162                 *sample = prof_ticks(p);
 163                 break;
 164         case CPUCLOCK_VIRT:
 165                 *sample = virt_ticks(p);
 166                 break;
 167         case CPUCLOCK_SCHED:
 168                 *sample = task_sched_runtime(p);
 169                 break;
 170         }
 171         return 0;
 172 }
 173
 174 /*
 175  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 176  * to avoid race conditions with concurrent updates to cputime.
 177  */
 178 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 179 {
 180         u64 curr_cputime;
 181 retry:
 182         curr_cputime = atomic64_read(cputime);
 183         if (sum_cputime > curr_cputime) {
 184                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 185                         goto retry;
 186         }
 187 }
 188
 189 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
 190 {
 191         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 192         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 193         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 194 }
 195
 196 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 197 static inline void sample_cputime_atomic(struct task_cputime *times,
 198                                          struct task_cputime_atomic *atomic_times)
 199 {
 200         times->utime = atomic64_read(&atomic_times->utime);
 201         times->stime = atomic64_read(&atomic_times->stime);
 202         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 203 }
 204
 205 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 206 {
 207         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 208         struct task_cputime sum;
 209
 210         /* Check if cputimer isn't running. This is accessed without locking. */
 211         if (!READ_ONCE(cputimer->running)) {
 212                 /*
 213                  * The POSIX timer interface allows for absolute time expiry
 214                  * values through the TIMER_ABSTIME flag, therefore we have
 215                  * to synchronize the timer to the clock every time we start it.
 216                  */
 217                 thread_group_cputime(tsk, &sum);
 218                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 219
 220                 /*
 221                  * We're setting cputimer->running without a lock. Ensure
 222                  * this only gets written to in one operation. We set
 223                  * running after update_gt_cputime() as a small optimization,
 224                  * but barriers are not required because update_gt_cputime()
 225                  * can handle concurrent updates.
 226                  */
 227                 WRITE_ONCE(cputimer->running, true);
 228         }
 229         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 230 }
 231
 232 /*
 233  * Sample a process (thread group) clock for the given group_leader task.
 234  * Must be called with task sighand lock held for safe while_each_thread()
 235  * traversal.
 236  */
 237 static int cpu_clock_sample_group(const clockid_t which_clock,
 238                                   struct task_struct *p,
 239                                   u64 *sample)
 240 {
 241         struct task_cputime cputime;
 242
 243         switch (CPUCLOCK_WHICH(which_clock)) {
 244         default:
 245                 return -EINVAL;
 246         case CPUCLOCK_PROF:
 247                 thread_group_cputime(p, &cputime);
 248                 *sample = cputime.utime + cputime.stime;
 249                 break;
 250         case CPUCLOCK_VIRT:
 251                 thread_group_cputime(p, &cputime);
 252                 *sample = cputime.utime;
 253                 break;
 254         case CPUCLOCK_SCHED:
 255                 thread_group_cputime(p, &cputime);
 256                 *sample = cputime.sum_exec_runtime;
 257                 break;
 258         }
 259         return 0;
 260 }
 261
 262 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 263                                     const clockid_t which_clock,
 264                                     struct timespec64 *tp)
 265 {
 266         int err = -EINVAL;
 267         u64 rtn;
 268
 269         if (CPUCLOCK_PERTHREAD(which_clock)) {
 270                 if (same_thread_group(tsk, current))
 271                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 272         } else {
 273                 if (tsk == current || thread_group_leader(tsk))
 274                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 275         }
 276
 277         if (!err)
 278                 *tp = ns_to_timespec64(rtn);
 279
 280         return err;
 281 }
 282
 283
 284 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
 285 {
 286         const pid_t pid = CPUCLOCK_PID(which_clock);
 287         int err = -EINVAL;
 288
 289         if (pid == 0) {
 290                 /*
 291                  * Special case constant value for our own clocks.
 292                  * We don't have to do any lookup to find ourselves.
 293                  */
 294                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 295         } else {
 296                 /*
 297                  * Find the given PID, and validate that the caller
 298                  * should be able to see it.
 299                  */
 300                 struct task_struct *p;
 301                 rcu_read_lock();
 302                 p = find_task_by_vpid(pid);
 303                 if (p)
 304                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 305                 rcu_read_unlock();
 306         }
 307
 308         return err;
 309 }
 310
 311 /*
 312  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 313  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 314  * new timer already all-zeros initialized.
 315  */
 316 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 317 {
 318         int ret = 0;
 319         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 320         struct task_struct *p;
 321
 322         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 323                 return -EINVAL;
 324
 325         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 326
 327         rcu_read_lock();
 328         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 329                 if (pid == 0) {
 330                         p = current;
 331                 } else {
 332                         p = find_task_by_vpid(pid);
 333                         if (p && !same_thread_group(p, current))
 334                                 p = NULL;
 335                 }
 336         } else {
 337                 if (pid == 0) {
 338                         p = current->group_leader;
 339                 } else {
 340                         p = find_task_by_vpid(pid);
 341                         if (p && !has_group_leader_pid(p))
 342                                 p = NULL;
 343                 }
 344         }
 345         new_timer->it.cpu.task = p;
 346         if (p) {
 347                 get_task_struct(p);
 348         } else {
 349                 ret = -EINVAL;
 350         }
 351         rcu_read_unlock();
 352
 353         return ret;
 354 }
 355
 356 /*
 357  * Clean up a CPU-clock timer that is about to be destroyed.
 358  * This is called from timer deletion with the timer already locked.
 359  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 360  * and try again.  (This happens when the timer is in the middle of firing.)
 361  */
 362 static int posix_cpu_timer_del(struct k_itimer *timer)
 363 {
 364         int ret = 0;
 365         unsigned long flags;
 366         struct sighand_struct *sighand;
 367         struct task_struct *p = timer->it.cpu.task;
 368
 369         WARN_ON_ONCE(p == NULL);
 370
 371         /*
 372          * Protect against sighand release/switch in exit/exec and process/
 373          * thread timer list entry concurrent read/writes.
 374          */
 375         sighand = lock_task_sighand(p, &flags);
 376         if (unlikely(sighand == NULL)) {
 377                 /*
 378                  * We raced with the reaping of the task.
 379                  * The deletion should have cleared us off the list.
 380                  */
 381                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 382         } else {
 383                 if (timer->it.cpu.firing)
 384                         ret = TIMER_RETRY;
 385                 else
 386                         list_del(&timer->it.cpu.entry);
 387
 388                 unlock_task_sighand(p, &flags);
 389         }
 390
 391         if (!ret)
 392                 put_task_struct(p);
 393
 394         return ret;
 395 }
 396
 397 static void cleanup_timers_list(struct list_head *head)
 398 {
 399         struct cpu_timer_list *timer, *next;
 400
 401         list_for_each_entry_safe(timer, next, head, entry)
 402                 list_del_init(&timer->entry);
 403 }
 404
 405 /*
 406  * Clean out CPU timers still ticking when a thread exited.  The task
 407  * pointer is cleared, and the expiry time is replaced with the residual
 408  * time for later timer_gettime calls to return.
 409  * This must be called with the siglock held.
 410  */
 411 static void cleanup_timers(struct list_head *head)
 412 {
 413         cleanup_timers_list(head);
 414         cleanup_timers_list(++head);
 415         cleanup_timers_list(++head);
 416 }
 417
 418 /*
 419  * These are both called with the siglock held, when the current thread
 420  * is being reaped.  When the final (leader) thread in the group is reaped,
 421  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 422  */
 423 void posix_cpu_timers_exit(struct task_struct *tsk)
 424 {
 425         cleanup_timers(tsk->cpu_timers);
 426 }
 427 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 428 {
 429         cleanup_timers(tsk->signal->cpu_timers);
 430 }
 431
 432 static inline int expires_gt(u64 expires, u64 new_exp)
 433 {
 434         return expires == 0 || expires > new_exp;
 435 }
 436
 437 /*
 438  * Insert the timer on the appropriate list before any timers that
 439  * expire later.  This must be called with the sighand lock held.
 440  */
 441 static void arm_timer(struct k_itimer *timer)
 442 {
 443         struct task_struct *p = timer->it.cpu.task;
 444         struct list_head *head, *listpos;
 445         struct task_cputime *cputime_expires;
 446         struct cpu_timer_list *const nt = &timer->it.cpu;
 447         struct cpu_timer_list *next;
 448
 449         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 450                 head = p->cpu_timers;
 451                 cputime_expires = &p->cputime_expires;
 452         } else {
 453                 head = p->signal->cpu_timers;
 454                 cputime_expires = &p->signal->cputime_expires;
 455         }
 456         head += CPUCLOCK_WHICH(timer->it_clock);
 457
 458         listpos = head;
 459         list_for_each_entry(next, head, entry) {
 460                 if (nt->expires < next->expires)
 461                         break;
 462                 listpos = &next->entry;
 463         }
 464         list_add(&nt->entry, listpos);
 465
 466         if (listpos == head) {
 467                 u64 exp = nt->expires;
 468
 469                 /*
 470                  * We are the new earliest-expiring POSIX 1.b timer, hence
 471                  * need to update expiration cache. Take into account that
 472                  * for process timers we share expiration cache with itimers
 473                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 474                  */
 475
 476                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 477                 case CPUCLOCK_PROF:
 478                         if (expires_gt(cputime_expires->prof_exp, exp))
 479                                 cputime_expires->prof_exp = exp;
 480                         break;
 481                 case CPUCLOCK_VIRT:
 482                         if (expires_gt(cputime_expires->virt_exp, exp))
 483                                 cputime_expires->virt_exp = exp;
 484                         break;
 485                 case CPUCLOCK_SCHED:
 486                         if (expires_gt(cputime_expires->sched_exp, exp))
 487                                 cputime_expires->sched_exp = exp;
 488                         break;
 489                 }
 490                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 491                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 492                 else
 493                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 494         }
 495 }
 496
 497 /*
 498  * The timer is locked, fire it and arrange for its reload.
 499  */
 500 static void cpu_timer_fire(struct k_itimer *timer)
 501 {
 502         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 503                 /*
 504                  * User don't want any signal.
 505                  */
 506                 timer->it.cpu.expires = 0;
 507         } else if (unlikely(timer->sigq == NULL)) {
 508                 /*
 509                  * This a special case for clock_nanosleep,
 510                  * not a normal timer from sys_timer_create.
 511                  */
 512                 wake_up_process(timer->it_process);
 513                 timer->it.cpu.expires = 0;
 514         } else if (timer->it.cpu.incr == 0) {
 515                 /*
 516                  * One-shot timer.  Clear it as soon as it's fired.
 517                  */
 518                 posix_timer_event(timer, 0);
 519                 timer->it.cpu.expires = 0;
 520         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 521                 /*
 522                  * The signal did not get queued because the signal
 523                  * was ignored, so we won't get any callback to
 524                  * reload the timer.  But we need to keep it
 525                  * ticking in case the signal is deliverable next time.
 526                  */
 527                 posix_cpu_timer_schedule(timer);
 528         }
 529 }
 530
 531 /*
 532  * Sample a process (thread group) timer for the given group_leader task.
 533  * Must be called with task sighand lock held for safe while_each_thread()
 534  * traversal.
 535  */
 536 static int cpu_timer_sample_group(const clockid_t which_clock,
 537                                   struct task_struct *p, u64 *sample)
 538 {
 539         struct task_cputime cputime;
 540
 541         thread_group_cputimer(p, &cputime);
 542         switch (CPUCLOCK_WHICH(which_clock)) {
 543         default:
 544                 return -EINVAL;
 545         case CPUCLOCK_PROF:
 546                 *sample = cputime.utime + cputime.stime;
 547                 break;
 548         case CPUCLOCK_VIRT:
 549                 *sample = cputime.utime;
 550                 break;
 551         case CPUCLOCK_SCHED:
 552                 *sample = cputime.sum_exec_runtime;
 553                 break;
 554         }
 555         return 0;
 556 }
 557
 558 /*
 559  * Guts of sys_timer_settime for CPU timers.
 560  * This is called with the timer locked and interrupts disabled.
 561  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 562  * and try again.  (This happens when the timer is in the middle of firing.)
 563  */
 564 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 565                                struct itimerspec64 *new, struct itimerspec64 *old)
 566 {
 567         unsigned long flags;
 568         struct sighand_struct *sighand;
 569         struct task_struct *p = timer->it.cpu.task;
 570         u64 old_expires, new_expires, old_incr, val;
 571         int ret;
 572
 573         WARN_ON_ONCE(p == NULL);
 574
 575         new_expires = timespec64_to_ns(&new->it_value);
 576
 577         /*
 578          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 579          * and p->signal->cpu_timers read/write in arm_timer()
 580          */
 581         sighand = lock_task_sighand(p, &flags);
 582         /*
 583          * If p has just been reaped, we can no
 584          * longer get any information about it at all.
 585          */
 586         if (unlikely(sighand == NULL)) {
 587                 return -ESRCH;
 588         }
 589
 590         /*
 591          * Disarm any old timer after extracting its expiry time.
 592          */
 593         WARN_ON_ONCE(!irqs_disabled());
 594
 595         ret = 0;
 596         old_incr = timer->it.cpu.incr;
 597         old_expires = timer->it.cpu.expires;
 598         if (unlikely(timer->it.cpu.firing)) {
 599                 timer->it.cpu.firing = -1;
 600                 ret = TIMER_RETRY;
 601         } else
 602                 list_del_init(&timer->it.cpu.entry);
 603
 604         /*
 605          * We need to sample the current value to convert the new
 606          * value from to relative and absolute, and to convert the
 607          * old value from absolute to relative.  To set a process
 608          * timer, we need a sample to balance the thread expiry
 609          * times (in arm_timer).  With an absolute time, we must
 610          * check if it's already passed.  In short, we need a sample.
 611          */
 612         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 613                 cpu_clock_sample(timer->it_clock, p, &val);
 614         } else {
 615                 cpu_timer_sample_group(timer->it_clock, p, &val);
 616         }
 617
 618         if (old) {
 619                 if (old_expires == 0) {
 620                         old->it_value.tv_sec = 0;
 621                         old->it_value.tv_nsec = 0;
 622                 } else {
 623                         /*
 624                          * Update the timer in case it has
 625                          * overrun already.  If it has,
 626                          * we'll report it as having overrun
 627                          * and with the next reloaded timer
 628                          * already ticking, though we are
 629                          * swallowing that pending
 630                          * notification here to install the
 631                          * new setting.
 632                          */
 633                         bump_cpu_timer(timer, val);
 634                         if (val < timer->it.cpu.expires) {
 635                                 old_expires = timer->it.cpu.expires - val;
 636                                 old->it_value = ns_to_timespec64(old_expires);
 637                         } else {
 638                                 old->it_value.tv_nsec = 1;
 639                                 old->it_value.tv_sec = 0;
 640                         }
 641                 }
 642         }
 643
 644         if (unlikely(ret)) {
 645                 /*
 646                  * We are colliding with the timer actually firing.
 647                  * Punt after filling in the timer's old value, and
 648                  * disable this firing since we are already reporting
 649                  * it as an overrun (thanks to bump_cpu_timer above).
 650                  */
 651                 unlock_task_sighand(p, &flags);
 652                 goto out;
 653         }
 654
 655         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 656                 new_expires += val;
 657         }
 658
 659         /*
 660          * Install the new expiry time (or zero).
 661          * For a timer with no notification action, we don't actually
 662          * arm the timer (we'll just fake it for timer_gettime).
 663          */
 664         timer->it.cpu.expires = new_expires;
 665         if (new_expires != 0 && val < new_expires) {
 666                 arm_timer(timer);
 667         }
 668
 669         unlock_task_sighand(p, &flags);
 670         /*
 671          * Install the new reload setting, and
 672          * set up the signal and overrun bookkeeping.
 673          */
 674         timer->it.cpu.incr = timespec64_to_ns(&new->it_interval);
 675
 676         /*
 677          * This acts as a modification timestamp for the timer,
 678          * so any automatic reload attempt will punt on seeing
 679          * that we have reset the timer manually.
 680          */
 681         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 682                 ~REQUEUE_PENDING;
 683         timer->it_overrun_last = 0;
 684         timer->it_overrun = -1;
 685
 686         if (new_expires != 0 && !(val < new_expires)) {
 687                 /*
 688                  * The designated time already passed, so we notify
 689                  * immediately, even if the thread never runs to
 690                  * accumulate more time on this clock.
 691                  */
 692                 cpu_timer_fire(timer);
 693         }
 694
 695         ret = 0;
 696  out:
 697         if (old)
 698                 old->it_interval = ns_to_timespec64(old_incr);
 699
 700         return ret;
 701 }
 702
 703 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
 704 {
 705         u64 now;
 706         struct task_struct *p = timer->it.cpu.task;
 707
 708         WARN_ON_ONCE(p == NULL);
 709
 710         /*
 711          * Easy part: convert the reload time.
 712          */
 713         itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
 714
 715         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 716                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 717                 return;
 718         }
 719
 720         /*
 721          * Sample the clock to take the difference with the expiry time.
 722          */
 723         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 724                 cpu_clock_sample(timer->it_clock, p, &now);
 725         } else {
 726                 struct sighand_struct *sighand;
 727                 unsigned long flags;
 728
 729                 /*
 730                  * Protect against sighand release/switch in exit/exec and
 731                  * also make timer sampling safe if it ends up calling
 732                  * thread_group_cputime().
 733                  */
 734                 sighand = lock_task_sighand(p, &flags);
 735                 if (unlikely(sighand == NULL)) {
 736                         /*
 737                          * The process has been reaped.
 738                          * We can't even collect a sample any more.
 739                          * Call the timer disarmed, nothing else to do.
 740                          */
 741                         timer->it.cpu.expires = 0;
 742                         itp->it_value = ns_to_timespec64(timer->it.cpu.expires);
 743                         return;
 744                 } else {
 745                         cpu_timer_sample_group(timer->it_clock, p, &now);
 746                         unlock_task_sighand(p, &flags);
 747                 }
 748         }
 749
 750         if (now < timer->it.cpu.expires) {
 751                 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
 752         } else {
 753                 /*
 754                  * The timer should have expired already, but the firing
 755                  * hasn't taken place yet.  Say it's just about to expire.
 756                  */
 757                 itp->it_value.tv_nsec = 1;
 758                 itp->it_value.tv_sec = 0;
 759         }
 760 }
 761
 762 static unsigned long long
 763 check_timers_list(struct list_head *timers,
 764                   struct list_head *firing,
 765                   unsigned long long curr)
 766 {
 767         int maxfire = 20;
 768
 769         while (!list_empty(timers)) {
 770                 struct cpu_timer_list *t;
 771
 772                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 773
 774                 if (!--maxfire || curr < t->expires)
 775                         return t->expires;
 776
 777                 t->firing = 1;
 778                 list_move_tail(&t->entry, firing);
 779         }
 780
 781         return 0;
 782 }
 783
 784 /*
 785  * Check for any per-thread CPU timers that have fired and move them off
 786  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 787  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 788  */
 789 static void check_thread_timers(struct task_struct *tsk,
 790                                 struct list_head *firing)
 791 {
 792         struct list_head *timers = tsk->cpu_timers;
 793         struct signal_struct *const sig = tsk->signal;
 794         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 795         u64 expires;
 796         unsigned long soft;
 797
 798         /*
 799          * If cputime_expires is zero, then there are no active
 800          * per thread CPU timers.
 801          */
 802         if (task_cputime_zero(&tsk->cputime_expires))
 803                 return;
 804
 805         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 806         tsk_expires->prof_exp = expires;
 807
 808         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 809         tsk_expires->virt_exp = expires;
 810
 811         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 812                                                    tsk->se.sum_exec_runtime);
 813
 814         /*
 815          * Check for the special case thread timers.
 816          */
 817         soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 818         if (soft != RLIM_INFINITY) {
 819                 unsigned long hard =
 820                         READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 821
 822                 if (hard != RLIM_INFINITY &&
 823                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 824                         /*
 825                          * At the hard limit, we just die.
 826                          * No need to calculate anything else now.
 827                          */
 828                         pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
 829                                 tsk->comm, task_pid_nr(tsk));
 830                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 831                         return;
 832                 }
 833                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 834                         /*
 835                          * At the soft limit, send a SIGXCPU every second.
 836                          */
 837                         if (soft < hard) {
 838                                 soft += USEC_PER_SEC;
 839                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 840                         }
 841                         pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
 842                                 tsk->comm, task_pid_nr(tsk));
 843                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 844                 }
 845         }
 846         if (task_cputime_zero(tsk_expires))
 847                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 848 }
 849
 850 static inline void stop_process_timers(struct signal_struct *sig)
 851 {
 852         struct thread_group_cputimer *cputimer = &sig->cputimer;
 853
 854         /* Turn off cputimer->running. This is done without locking. */
 855         WRITE_ONCE(cputimer->running, false);
 856         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 857 }
 858
 859 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 860                              u64 *expires, u64 cur_time, int signo)
 861 {
 862         if (!it->expires)
 863                 return;
 864
 865         if (cur_time >= it->expires) {
 866                 if (it->incr)
 867                         it->expires += it->incr;
 868                 else
 869                         it->expires = 0;
 870
 871                 trace_itimer_expire(signo == SIGPROF ?
 872                                     ITIMER_PROF : ITIMER_VIRTUAL,
 873                                     tsk->signal->leader_pid, cur_time);
 874                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 875         }
 876
 877         if (it->expires && (!*expires || it->expires < *expires))
 878                 *expires = it->expires;
 879 }
 880
 881 /*
 882  * Check for any per-thread CPU timers that have fired and move them
 883  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 884  * have already been taken off.
 885  */
 886 static void check_process_timers(struct task_struct *tsk,
 887                                  struct list_head *firing)
 888 {
 889         struct signal_struct *const sig = tsk->signal;
 890         u64 utime, ptime, virt_expires, prof_expires;
 891         u64 sum_sched_runtime, sched_expires;
 892         struct list_head *timers = sig->cpu_timers;
 893         struct task_cputime cputime;
 894         unsigned long soft;
 895
 896         /*
 897          * If cputimer is not running, then there are no active
 898          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 899          */
 900         if (!READ_ONCE(tsk->signal->cputimer.running))
 901                 return;
 902
 903         /*
 904          * Signify that a thread is checking for process timers.
 905          * Write access to this field is protected by the sighand lock.
 906          */
 907         sig->cputimer.checking_timer = true;
 908
 909         /*
 910          * Collect the current process totals.
 911          */
 912         thread_group_cputimer(tsk, &cputime);
 913         utime = cputime.utime;
 914         ptime = utime + cputime.stime;
 915         sum_sched_runtime = cputime.sum_exec_runtime;
 916
 917         prof_expires = check_timers_list(timers, firing, ptime);
 918         virt_expires = check_timers_list(++timers, firing, utime);
 919         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 920
 921         /*
 922          * Check for the special case process timers.
 923          */
 924         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 925                          SIGPROF);
 926         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 927                          SIGVTALRM);
 928         soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 929         if (soft != RLIM_INFINITY) {
 930                 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
 931                 unsigned long hard =
 932                         READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
 933                 u64 x;
 934                 if (psecs >= hard) {
 935                         /*
 936                          * At the hard limit, we just die.
 937                          * No need to calculate anything else now.
 938                          */
 939                         pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
 940                                 tsk->comm, task_pid_nr(tsk));
 941                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 942                         return;
 943                 }
 944                 if (psecs >= soft) {
 945                         /*
 946                          * At the soft limit, send a SIGXCPU every second.
 947                          */
 948                         pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
 949                                 tsk->comm, task_pid_nr(tsk));
 950                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 951                         if (soft < hard) {
 952                                 soft++;
 953                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 954                         }
 955                 }
 956                 x = soft * NSEC_PER_SEC;
 957                 if (!prof_expires || x < prof_expires)
 958                         prof_expires = x;
 959         }
 960
 961         sig->cputime_expires.prof_exp = prof_expires;
 962         sig->cputime_expires.virt_exp = virt_expires;
 963         sig->cputime_expires.sched_exp = sched_expires;
 964         if (task_cputime_zero(&sig->cputime_expires))
 965                 stop_process_timers(sig);
 966
 967         sig->cputimer.checking_timer = false;
 968 }
 969
 970 /*
 971  * This is called from the signal code (via do_schedule_next_timer)
 972  * when the last timer signal was delivered and we have to reload the timer.
 973  */
 974 void posix_cpu_timer_schedule(struct k_itimer *timer)
 975 {
 976         struct sighand_struct *sighand;
 977         unsigned long flags;
 978         struct task_struct *p = timer->it.cpu.task;
 979         u64 now;
 980
 981         WARN_ON_ONCE(p == NULL);
 982
 983         /*
 984          * Fetch the current sample and update the timer's expiry time.
 985          */
 986         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 987                 cpu_clock_sample(timer->it_clock, p, &now);
 988                 bump_cpu_timer(timer, now);
 989                 if (unlikely(p->exit_state))
 990                         goto out;
 991
 992                 /* Protect timer list r/w in arm_timer() */
 993                 sighand = lock_task_sighand(p, &flags);
 994                 if (!sighand)
 995                         goto out;
 996         } else {
 997                 /*
 998                  * Protect arm_timer() and timer sampling in case of call to
 999                  * thread_group_cputime().
1000                  */
1001                 sighand = lock_task_sighand(p, &flags);
1002                 if (unlikely(sighand == NULL)) {
1003                         /*
1004                          * The process has been reaped.
1005                          * We can't even collect a sample any more.
1006                          */
1007                         timer->it.cpu.expires = 0;
1008                         goto out;
1009                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1010                         unlock_task_sighand(p, &flags);
1011                         /* Optimizations: if the process is dying, no need to rearm */
1012                         goto out;
1013                 }
1014                 cpu_timer_sample_group(timer->it_clock, p, &now);
1015                 bump_cpu_timer(timer, now);
1016                 /* Leave the sighand locked for the call below.  */
1017         }
1018
1019         /*
1020          * Now re-arm for the new expiry time.
1021          */
1022         WARN_ON_ONCE(!irqs_disabled());
1023         arm_timer(timer);
1024         unlock_task_sighand(p, &flags);
1025
1026 out:
1027         timer->it_overrun_last = timer->it_overrun;
1028         timer->it_overrun = -1;
1029         ++timer->it_requeue_pending;
1030 }
1031
1032 /**
1033  * task_cputime_expired - Compare two task_cputime entities.
1034  *
1035  * @sample:     The task_cputime structure to be checked for expiration.
1036  * @expires:    Expiration times, against which @sample will be checked.
1037  *
1038  * Checks @sample against @expires to see if any field of @sample has expired.
1039  * Returns true if any field of the former is greater than the corresponding
1040  * field of the latter if the latter field is set.  Otherwise returns false.
1041  */
1042 static inline int task_cputime_expired(const struct task_cputime *sample,
1043                                         const struct task_cputime *expires)
1044 {
1045         if (expires->utime && sample->utime >= expires->utime)
1046                 return 1;
1047         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1048                 return 1;
1049         if (expires->sum_exec_runtime != 0 &&
1050             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1051                 return 1;
1052         return 0;
1053 }
1054
1055 /**
1056  * fastpath_timer_check - POSIX CPU timers fast path.
1057  *
1058  * @tsk:        The task (thread) being checked.
1059  *
1060  * Check the task and thread group timers.  If both are zero (there are no
1061  * timers set) return false.  Otherwise snapshot the task and thread group
1062  * timers and compare them with the corresponding expiration times.  Return
1063  * true if a timer has expired, else return false.
1064  */
1065 static inline int fastpath_timer_check(struct task_struct *tsk)
1066 {
1067         struct signal_struct *sig;
1068
1069         if (!task_cputime_zero(&tsk->cputime_expires)) {
1070                 struct task_cputime task_sample;
1071
1072                 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1073                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1074                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1075                         return 1;
1076         }
1077
1078         sig = tsk->signal;
1079         /*
1080          * Check if thread group timers expired when the cputimer is
1081          * running and no other thread in the group is already checking
1082          * for thread group cputimers. These fields are read without the
1083          * sighand lock. However, this is fine because this is meant to
1084          * be a fastpath heuristic to determine whether we should try to
1085          * acquire the sighand lock to check/handle timers.
1086          *
1087          * In the worst case scenario, if 'running' or 'checking_timer' gets
1088          * set but the current thread doesn't see the change yet, we'll wait
1089          * until the next thread in the group gets a scheduler interrupt to
1090          * handle the timer. This isn't an issue in practice because these
1091          * types of delays with signals actually getting sent are expected.
1092          */
1093         if (READ_ONCE(sig->cputimer.running) &&
1094             !READ_ONCE(sig->cputimer.checking_timer)) {
1095                 struct task_cputime group_sample;
1096
1097                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1098
1099                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1100                         return 1;
1101         }
1102
1103         return 0;
1104 }
1105
1106 /*
1107  * This is called from the timer interrupt handler.  The irq handler has
1108  * already updated our counts.  We need to check if any timers fire now.
1109  * Interrupts are disabled.
1110  */
1111 void run_posix_cpu_timers(struct task_struct *tsk)
1112 {
1113         LIST_HEAD(firing);
1114         struct k_itimer *timer, *next;
1115         unsigned long flags;
1116
1117         WARN_ON_ONCE(!irqs_disabled());
1118
1119         /*
1120          * The fast path checks that there are no expired thread or thread
1121          * group timers.  If that's so, just return.
1122          */
1123         if (!fastpath_timer_check(tsk))
1124                 return;
1125
1126         if (!lock_task_sighand(tsk, &flags))
1127                 return;
1128         /*
1129          * Here we take off tsk->signal->cpu_timers[N] and
1130          * tsk->cpu_timers[N] all the timers that are firing, and
1131          * put them on the firing list.
1132          */
1133         check_thread_timers(tsk, &firing);
1134
1135         check_process_timers(tsk, &firing);
1136
1137         /*
1138          * We must release these locks before taking any timer's lock.
1139          * There is a potential race with timer deletion here, as the
1140          * siglock now protects our private firing list.  We have set
1141          * the firing flag in each timer, so that a deletion attempt
1142          * that gets the timer lock before we do will give it up and
1143          * spin until we've taken care of that timer below.
1144          */
1145         unlock_task_sighand(tsk, &flags);
1146
1147         /*
1148          * Now that all the timers on our list have the firing flag,
1149          * no one will touch their list entries but us.  We'll take
1150          * each timer's lock before clearing its firing flag, so no
1151          * timer call will interfere.
1152          */
1153         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1154                 int cpu_firing;
1155
1156                 spin_lock(&timer->it_lock);
1157                 list_del_init(&timer->it.cpu.entry);
1158                 cpu_firing = timer->it.cpu.firing;
1159                 timer->it.cpu.firing = 0;
1160                 /*
1161                  * The firing flag is -1 if we collided with a reset
1162                  * of the timer, which already reported this
1163                  * almost-firing as an overrun.  So don't generate an event.
1164                  */
1165                 if (likely(cpu_firing >= 0))
1166                         cpu_timer_fire(timer);
1167                 spin_unlock(&timer->it_lock);
1168         }
1169 }
1170
1171 /*
1172  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1173  * The tsk->sighand->siglock must be held by the caller.
1174  */
1175 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1176                            u64 *newval, u64 *oldval)
1177 {
1178         u64 now;
1179
1180         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1181         cpu_timer_sample_group(clock_idx, tsk, &now);
1182
1183         if (oldval) {
1184                 /*
1185                  * We are setting itimer. The *oldval is absolute and we update
1186                  * it to be relative, *newval argument is relative and we update
1187                  * it to be absolute.
1188                  */
1189                 if (*oldval) {
1190                         if (*oldval <= now) {
1191                                 /* Just about to fire. */
1192                                 *oldval = TICK_NSEC;
1193                         } else {
1194                                 *oldval -= now;
1195                         }
1196                 }
1197
1198                 if (!*newval)
1199                         return;
1200                 *newval += now;
1201         }
1202
1203         /*
1204          * Update expiration cache if we are the earliest timer, or eventually
1205          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1206          */
1207         switch (clock_idx) {
1208         case CPUCLOCK_PROF:
1209                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1210                         tsk->signal->cputime_expires.prof_exp = *newval;
1211                 break;
1212         case CPUCLOCK_VIRT:
1213                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1214                         tsk->signal->cputime_expires.virt_exp = *newval;
1215                 break;
1216         }
1217
1218         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1219 }
1220
1221 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1222                             struct timespec64 *rqtp, struct itimerspec64 *it)
1223 {
1224         struct k_itimer timer;
1225         int error;
1226
1227         /*
1228          * Set up a temporary timer and then wait for it to go off.
1229          */
1230         memset(&timer, 0, sizeof timer);
1231         spin_lock_init(&timer.it_lock);
1232         timer.it_clock = which_clock;
1233         timer.it_overrun = -1;
1234         error = posix_cpu_timer_create(&timer);
1235         timer.it_process = current;
1236         if (!error) {
1237                 static struct itimerspec64 zero_it;
1238
1239                 memset(it, 0, sizeof *it);
1240                 it->it_value = *rqtp;
1241
1242                 spin_lock_irq(&timer.it_lock);
1243                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1244                 if (error) {
1245                         spin_unlock_irq(&timer.it_lock);
1246                         return error;
1247                 }
1248
1249                 while (!signal_pending(current)) {
1250                         if (timer.it.cpu.expires == 0) {
1251                                 /*
1252                                  * Our timer fired and was reset, below
1253                                  * deletion can not fail.
1254                                  */
1255                                 posix_cpu_timer_del(&timer);
1256                                 spin_unlock_irq(&timer.it_lock);
1257                                 return 0;
1258                         }
1259
1260                         /*
1261                          * Block until cpu_timer_fire (or a signal) wakes us.
1262                          */
1263                         __set_current_state(TASK_INTERRUPTIBLE);
1264                         spin_unlock_irq(&timer.it_lock);
1265                         schedule();
1266                         spin_lock_irq(&timer.it_lock);
1267                 }
1268
1269                 /*
1270                  * We were interrupted by a signal.
1271                  */
1272                 *rqtp = ns_to_timespec64(timer.it.cpu.expires);
1273                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1274                 if (!error) {
1275                         /*
1276                          * Timer is now unarmed, deletion can not fail.
1277                          */
1278                         posix_cpu_timer_del(&timer);
1279                 }
1280                 spin_unlock_irq(&timer.it_lock);
1281
1282                 while (error == TIMER_RETRY) {
1283                         /*
1284                          * We need to handle case when timer was or is in the
1285                          * middle of firing. In other cases we already freed
1286                          * resources.
1287                          */
1288                         spin_lock_irq(&timer.it_lock);
1289                         error = posix_cpu_timer_del(&timer);
1290                         spin_unlock_irq(&timer.it_lock);
1291                 }
1292
1293                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1294                         /*
1295                          * It actually did fire already.
1296                          */
1297                         return 0;
1298                 }
1299
1300                 error = -ERESTART_RESTARTBLOCK;
1301         }
1302
1303         return error;
1304 }
1305
1306 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1307
1308 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1309                             struct timespec64 *rqtp, struct timespec __user *rmtp)
1310 {
1311         struct restart_block *restart_block = &current->restart_block;
1312         struct itimerspec64 it;
1313         struct timespec ts;
1314         int error;
1315
1316         /*
1317          * Diagnose required errors first.
1318          */
1319         if (CPUCLOCK_PERTHREAD(which_clock) &&
1320             (CPUCLOCK_PID(which_clock) == 0 ||
1321              CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1322                 return -EINVAL;
1323
1324         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1325
1326         if (error == -ERESTART_RESTARTBLOCK) {
1327
1328                 if (flags & TIMER_ABSTIME)
1329                         return -ERESTARTNOHAND;
1330                 /*
1331                  * Report back to the user the time still remaining.
1332                  */
1333                 ts = timespec64_to_timespec(it.it_value);
1334                 if (rmtp && copy_to_user(rmtp, &ts, sizeof(*rmtp)))
1335                         return -EFAULT;
1336
1337                 restart_block->fn = posix_cpu_nsleep_restart;
1338                 restart_block->nanosleep.clockid = which_clock;
1339                 restart_block->nanosleep.rmtp = rmtp;
1340                 restart_block->nanosleep.expires = timespec64_to_ns(rqtp);
1341         }
1342         return error;
1343 }
1344
1345 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1346 {
1347         clockid_t which_clock = restart_block->nanosleep.clockid;
1348         struct itimerspec64 it;
1349         struct timespec64 t;
1350         struct timespec tmp;
1351         int error;
1352
1353         t = ns_to_timespec64(restart_block->nanosleep.expires);
1354
1355         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1356
1357         if (error == -ERESTART_RESTARTBLOCK) {
1358                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1359                 /*
1360                  * Report back to the user the time still remaining.
1361                  */
1362                  tmp = timespec64_to_timespec(it.it_value);
1363                 if (rmtp && copy_to_user(rmtp, &tmp, sizeof(*rmtp)))
1364                         return -EFAULT;
1365
1366                 restart_block->nanosleep.expires = timespec64_to_ns(&t);
1367         }
1368         return error;
1369
1370 }
1371
1372 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1373 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1374
1375 static int process_cpu_clock_getres(const clockid_t which_clock,
1376                                     struct timespec64 *tp)
1377 {
1378         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1379 }
1380 static int process_cpu_clock_get(const clockid_t which_clock,
1381                                  struct timespec64 *tp)
1382 {
1383         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1384 }
1385 static int process_cpu_timer_create(struct k_itimer *timer)
1386 {
1387         timer->it_clock = PROCESS_CLOCK;
1388         return posix_cpu_timer_create(timer);
1389 }
1390 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1391                               struct timespec64 *rqtp,
1392                               struct timespec __user *rmtp)
1393 {
1394         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1395 }
1396 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1397 {
1398         return -EINVAL;
1399 }
1400 static int thread_cpu_clock_getres(const clockid_t which_clock,
1401                                    struct timespec64 *tp)
1402 {
1403         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1404 }
1405 static int thread_cpu_clock_get(const clockid_t which_clock,
1406                                 struct timespec64 *tp)
1407 {
1408         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1409 }
1410 static int thread_cpu_timer_create(struct k_itimer *timer)
1411 {
1412         timer->it_clock = THREAD_CLOCK;
1413         return posix_cpu_timer_create(timer);
1414 }
1415
1416 struct k_clock clock_posix_cpu = {
1417         .clock_getres   = posix_cpu_clock_getres,
1418         .clock_set      = posix_cpu_clock_set,
1419         .clock_get      = posix_cpu_clock_get,
1420         .timer_create   = posix_cpu_timer_create,
1421         .nsleep         = posix_cpu_nsleep,
1422         .nsleep_restart = posix_cpu_nsleep_restart,
1423         .timer_set      = posix_cpu_timer_set,
1424         .timer_del      = posix_cpu_timer_del,
1425         .timer_get      = posix_cpu_timer_get,
1426 };
1427
1428 static __init int init_posix_cpu_timers(void)
1429 {
1430         struct k_clock process = {
1431                 .clock_getres   = process_cpu_clock_getres,
1432                 .clock_get      = process_cpu_clock_get,
1433                 .timer_create   = process_cpu_timer_create,
1434                 .nsleep         = process_cpu_nsleep,
1435                 .nsleep_restart = process_cpu_nsleep_restart,
1436         };
1437         struct k_clock thread = {
1438                 .clock_getres   = thread_cpu_clock_getres,
1439                 .clock_get      = thread_cpu_clock_get,
1440                 .timer_create   = thread_cpu_timer_create,
1441         };
1442
1443         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1444         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1445
1446         return 0;
1447 }
1448 __initcall(init_posix_cpu_timers);