kernel/time/posix-cpu-timers.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Implement CPU time clocks for the POSIX clock interface.
   4  */
   5
   6 #include <linux/sched/signal.h>
   7 #include <linux/sched/cputime.h>
   8 #include <linux/posix-timers.h>
   9 #include <linux/errno.h>
  10 #include <linux/math64.h>
  11 #include <linux/uaccess.h>
  12 #include <linux/kernel_stat.h>
  13 #include <trace/events/timer.h>
  14 #include <linux/tick.h>
  15 #include <linux/workqueue.h>
  16 #include <linux/compat.h>
  17
  18 #include "posix-timers.h"
  19
  20 static void posix_cpu_timer_rearm(struct k_itimer *timer);
  21
  22 /*
  23  * Called after updating RLIMIT_CPU to run cpu timer and update
  24  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  25  * siglock protection since other code may update expiration cache as
  26  * well.
  27  */
  28 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  29 {
  30         u64 nsecs = rlim_new * NSEC_PER_SEC;
  31
  32         spin_lock_irq(&task->sighand->siglock);
  33         set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
  34         spin_unlock_irq(&task->sighand->siglock);
  35 }
  36
  37 static int check_clock(const clockid_t which_clock)
  38 {
  39         int error = 0;
  40         struct task_struct *p;
  41         const pid_t pid = CPUCLOCK_PID(which_clock);
  42
  43         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  44                 return -EINVAL;
  45
  46         if (pid == 0)
  47                 return 0;
  48
  49         rcu_read_lock();
  50         p = find_task_by_vpid(pid);
  51         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  52                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  53                 error = -EINVAL;
  54         }
  55         rcu_read_unlock();
  56
  57         return error;
  58 }
  59
  60 /*
  61  * Update expiry time from increment, and increase overrun count,
  62  * given the current clock sample.
  63  */
  64 static void bump_cpu_timer(struct k_itimer *timer, u64 now)
  65 {
  66         int i;
  67         u64 delta, incr;
  68
  69         if (timer->it.cpu.incr == 0)
  70                 return;
  71
  72         if (now < timer->it.cpu.expires)
  73                 return;
  74
  75         incr = timer->it.cpu.incr;
  76         delta = now + incr - timer->it.cpu.expires;
  77
  78         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  79         for (i = 0; incr < delta - incr; i++)
  80                 incr = incr << 1;
  81
  82         for (; i >= 0; incr >>= 1, i--) {
  83                 if (delta < incr)
  84                         continue;
  85
  86                 timer->it.cpu.expires += incr;
  87                 timer->it_overrun += 1LL << i;
  88                 delta -= incr;
  89         }
  90 }
  91
  92 /**
  93  * task_cputime_zero - Check a task_cputime struct for all zero fields.
  94  *
  95  * @cputime:    The struct to compare.
  96  *
  97  * Checks @cputime to see if all fields are zero.  Returns true if all fields
  98  * are zero, false if any field is nonzero.
  99  */
 100 static inline int task_cputime_zero(const struct task_cputime *cputime)
 101 {
 102         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 103                 return 1;
 104         return 0;
 105 }
 106
 107 static inline u64 prof_ticks(struct task_struct *p)
 108 {
 109         u64 utime, stime;
 110
 111         task_cputime(p, &utime, &stime);
 112
 113         return utime + stime;
 114 }
 115 static inline u64 virt_ticks(struct task_struct *p)
 116 {
 117         u64 utime, stime;
 118
 119         task_cputime(p, &utime, &stime);
 120
 121         return utime;
 122 }
 123
 124 static int
 125 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
 126 {
 127         int error = check_clock(which_clock);
 128         if (!error) {
 129                 tp->tv_sec = 0;
 130                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 131                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 132                         /*
 133                          * If sched_clock is using a cycle counter, we
 134                          * don't have any idea of its true resolution
 135                          * exported, but it is much more than 1s/HZ.
 136                          */
 137                         tp->tv_nsec = 1;
 138                 }
 139         }
 140         return error;
 141 }
 142
 143 static int
 144 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
 145 {
 146         /*
 147          * You can never reset a CPU clock, but we check for other errors
 148          * in the call before failing with EPERM.
 149          */
 150         int error = check_clock(which_clock);
 151         if (error == 0) {
 152                 error = -EPERM;
 153         }
 154         return error;
 155 }
 156
 157
 158 /*
 159  * Sample a per-thread clock for the given task.
 160  */
 161 static int cpu_clock_sample(const clockid_t which_clock,
 162                             struct task_struct *p, u64 *sample)
 163 {
 164         switch (CPUCLOCK_WHICH(which_clock)) {
 165         default:
 166                 return -EINVAL;
 167         case CPUCLOCK_PROF:
 168                 *sample = prof_ticks(p);
 169                 break;
 170         case CPUCLOCK_VIRT:
 171                 *sample = virt_ticks(p);
 172                 break;
 173         case CPUCLOCK_SCHED:
 174                 *sample = task_sched_runtime(p);
 175                 break;
 176         }
 177         return 0;
 178 }
 179
 180 /*
 181  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 182  * to avoid race conditions with concurrent updates to cputime.
 183  */
 184 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 185 {
 186         u64 curr_cputime;
 187 retry:
 188         curr_cputime = atomic64_read(cputime);
 189         if (sum_cputime > curr_cputime) {
 190                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 191                         goto retry;
 192         }
 193 }
 194
 195 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
 196 {
 197         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 198         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 199         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 200 }
 201
 202 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 203 static inline void sample_cputime_atomic(struct task_cputime *times,
 204                                          struct task_cputime_atomic *atomic_times)
 205 {
 206         times->utime = atomic64_read(&atomic_times->utime);
 207         times->stime = atomic64_read(&atomic_times->stime);
 208         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 209 }
 210
 211 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 212 {
 213         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 214         struct task_cputime sum;
 215
 216         /* Check if cputimer isn't running. This is accessed without locking. */
 217         if (!READ_ONCE(cputimer->running)) {
 218                 /*
 219                  * The POSIX timer interface allows for absolute time expiry
 220                  * values through the TIMER_ABSTIME flag, therefore we have
 221                  * to synchronize the timer to the clock every time we start it.
 222                  */
 223                 thread_group_cputime(tsk, &sum);
 224                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 225
 226                 /*
 227                  * We're setting cputimer->running without a lock. Ensure
 228                  * this only gets written to in one operation. We set
 229                  * running after update_gt_cputime() as a small optimization,
 230                  * but barriers are not required because update_gt_cputime()
 231                  * can handle concurrent updates.
 232                  */
 233                 WRITE_ONCE(cputimer->running, true);
 234         }
 235         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 236 }
 237
 238 /*
 239  * Sample a process (thread group) clock for the given group_leader task.
 240  * Must be called with task sighand lock held for safe while_each_thread()
 241  * traversal.
 242  */
 243 static int cpu_clock_sample_group(const clockid_t which_clock,
 244                                   struct task_struct *p,
 245                                   u64 *sample)
 246 {
 247         struct task_cputime cputime;
 248
 249         switch (CPUCLOCK_WHICH(which_clock)) {
 250         default:
 251                 return -EINVAL;
 252         case CPUCLOCK_PROF:
 253                 thread_group_cputime(p, &cputime);
 254                 *sample = cputime.utime + cputime.stime;
 255                 break;
 256         case CPUCLOCK_VIRT:
 257                 thread_group_cputime(p, &cputime);
 258                 *sample = cputime.utime;
 259                 break;
 260         case CPUCLOCK_SCHED:
 261                 thread_group_cputime(p, &cputime);
 262                 *sample = cputime.sum_exec_runtime;
 263                 break;
 264         }
 265         return 0;
 266 }
 267
 268 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 269                                     const clockid_t which_clock,
 270                                     struct timespec64 *tp)
 271 {
 272         int err = -EINVAL;
 273         u64 rtn;
 274
 275         if (CPUCLOCK_PERTHREAD(which_clock)) {
 276                 if (same_thread_group(tsk, current))
 277                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 278         } else {
 279                 if (tsk == current || thread_group_leader(tsk))
 280                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 281         }
 282
 283         if (!err)
 284                 *tp = ns_to_timespec64(rtn);
 285
 286         return err;
 287 }
 288
 289
 290 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
 291 {
 292         const pid_t pid = CPUCLOCK_PID(which_clock);
 293         int err = -EINVAL;
 294
 295         if (pid == 0) {
 296                 /*
 297                  * Special case constant value for our own clocks.
 298                  * We don't have to do any lookup to find ourselves.
 299                  */
 300                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 301         } else {
 302                 /*
 303                  * Find the given PID, and validate that the caller
 304                  * should be able to see it.
 305                  */
 306                 struct task_struct *p;
 307                 rcu_read_lock();
 308                 p = find_task_by_vpid(pid);
 309                 if (p)
 310                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 311                 rcu_read_unlock();
 312         }
 313
 314         return err;
 315 }
 316
 317 /*
 318  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 319  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 320  * new timer already all-zeros initialized.
 321  */
 322 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 323 {
 324         int ret = 0;
 325         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 326         struct task_struct *p;
 327
 328         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 329                 return -EINVAL;
 330
 331         new_timer->kclock = &clock_posix_cpu;
 332
 333         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 334
 335         rcu_read_lock();
 336         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 337                 if (pid == 0) {
 338                         p = current;
 339                 } else {
 340                         p = find_task_by_vpid(pid);
 341                         if (p && !same_thread_group(p, current))
 342                                 p = NULL;
 343                 }
 344         } else {
 345                 if (pid == 0) {
 346                         p = current->group_leader;
 347                 } else {
 348                         p = find_task_by_vpid(pid);
 349                         if (p && !has_group_leader_pid(p))
 350                                 p = NULL;
 351                 }
 352         }
 353         new_timer->it.cpu.task = p;
 354         if (p) {
 355                 get_task_struct(p);
 356         } else {
 357                 ret = -EINVAL;
 358         }
 359         rcu_read_unlock();
 360
 361         return ret;
 362 }
 363
 364 /*
 365  * Clean up a CPU-clock timer that is about to be destroyed.
 366  * This is called from timer deletion with the timer already locked.
 367  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 368  * and try again.  (This happens when the timer is in the middle of firing.)
 369  */
 370 static int posix_cpu_timer_del(struct k_itimer *timer)
 371 {
 372         int ret = 0;
 373         unsigned long flags;
 374         struct sighand_struct *sighand;
 375         struct task_struct *p = timer->it.cpu.task;
 376
 377         WARN_ON_ONCE(p == NULL);
 378
 379         /*
 380          * Protect against sighand release/switch in exit/exec and process/
 381          * thread timer list entry concurrent read/writes.
 382          */
 383         sighand = lock_task_sighand(p, &flags);
 384         if (unlikely(sighand == NULL)) {
 385                 /*
 386                  * We raced with the reaping of the task.
 387                  * The deletion should have cleared us off the list.
 388                  */
 389                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 390         } else {
 391                 if (timer->it.cpu.firing)
 392                         ret = TIMER_RETRY;
 393                 else
 394                         list_del(&timer->it.cpu.entry);
 395
 396                 unlock_task_sighand(p, &flags);
 397         }
 398
 399         if (!ret)
 400                 put_task_struct(p);
 401
 402         return ret;
 403 }
 404
 405 static void cleanup_timers_list(struct list_head *head)
 406 {
 407         struct cpu_timer_list *timer, *next;
 408
 409         list_for_each_entry_safe(timer, next, head, entry)
 410                 list_del_init(&timer->entry);
 411 }
 412
 413 /*
 414  * Clean out CPU timers still ticking when a thread exited.  The task
 415  * pointer is cleared, and the expiry time is replaced with the residual
 416  * time for later timer_gettime calls to return.
 417  * This must be called with the siglock held.
 418  */
 419 static void cleanup_timers(struct list_head *head)
 420 {
 421         cleanup_timers_list(head);
 422         cleanup_timers_list(++head);
 423         cleanup_timers_list(++head);
 424 }
 425
 426 /*
 427  * These are both called with the siglock held, when the current thread
 428  * is being reaped.  When the final (leader) thread in the group is reaped,
 429  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 430  */
 431 void posix_cpu_timers_exit(struct task_struct *tsk)
 432 {
 433         cleanup_timers(tsk->cpu_timers);
 434 }
 435 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 436 {
 437         cleanup_timers(tsk->signal->cpu_timers);
 438 }
 439
 440 static inline int expires_gt(u64 expires, u64 new_exp)
 441 {
 442         return expires == 0 || expires > new_exp;
 443 }
 444
 445 /*
 446  * Insert the timer on the appropriate list before any timers that
 447  * expire later.  This must be called with the sighand lock held.
 448  */
 449 static void arm_timer(struct k_itimer *timer)
 450 {
 451         struct task_struct *p = timer->it.cpu.task;
 452         struct list_head *head, *listpos;
 453         struct task_cputime *cputime_expires;
 454         struct cpu_timer_list *const nt = &timer->it.cpu;
 455         struct cpu_timer_list *next;
 456
 457         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 458                 head = p->cpu_timers;
 459                 cputime_expires = &p->cputime_expires;
 460         } else {
 461                 head = p->signal->cpu_timers;
 462                 cputime_expires = &p->signal->cputime_expires;
 463         }
 464         head += CPUCLOCK_WHICH(timer->it_clock);
 465
 466         listpos = head;
 467         list_for_each_entry(next, head, entry) {
 468                 if (nt->expires < next->expires)
 469                         break;
 470                 listpos = &next->entry;
 471         }
 472         list_add(&nt->entry, listpos);
 473
 474         if (listpos == head) {
 475                 u64 exp = nt->expires;
 476
 477                 /*
 478                  * We are the new earliest-expiring POSIX 1.b timer, hence
 479                  * need to update expiration cache. Take into account that
 480                  * for process timers we share expiration cache with itimers
 481                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 482                  */
 483
 484                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 485                 case CPUCLOCK_PROF:
 486                         if (expires_gt(cputime_expires->prof_exp, exp))
 487                                 cputime_expires->prof_exp = exp;
 488                         break;
 489                 case CPUCLOCK_VIRT:
 490                         if (expires_gt(cputime_expires->virt_exp, exp))
 491                                 cputime_expires->virt_exp = exp;
 492                         break;
 493                 case CPUCLOCK_SCHED:
 494                         if (expires_gt(cputime_expires->sched_exp, exp))
 495                                 cputime_expires->sched_exp = exp;
 496                         break;
 497                 }
 498                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 499                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 500                 else
 501                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 502         }
 503 }
 504
 505 /*
 506  * The timer is locked, fire it and arrange for its reload.
 507  */
 508 static void cpu_timer_fire(struct k_itimer *timer)
 509 {
 510         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 511                 /*
 512                  * User don't want any signal.
 513                  */
 514                 timer->it.cpu.expires = 0;
 515         } else if (unlikely(timer->sigq == NULL)) {
 516                 /*
 517                  * This a special case for clock_nanosleep,
 518                  * not a normal timer from sys_timer_create.
 519                  */
 520                 wake_up_process(timer->it_process);
 521                 timer->it.cpu.expires = 0;
 522         } else if (timer->it.cpu.incr == 0) {
 523                 /*
 524                  * One-shot timer.  Clear it as soon as it's fired.
 525                  */
 526                 posix_timer_event(timer, 0);
 527                 timer->it.cpu.expires = 0;
 528         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 529                 /*
 530                  * The signal did not get queued because the signal
 531                  * was ignored, so we won't get any callback to
 532                  * reload the timer.  But we need to keep it
 533                  * ticking in case the signal is deliverable next time.
 534                  */
 535                 posix_cpu_timer_rearm(timer);
 536                 ++timer->it_requeue_pending;
 537         }
 538 }
 539
 540 /*
 541  * Sample a process (thread group) timer for the given group_leader task.
 542  * Must be called with task sighand lock held for safe while_each_thread()
 543  * traversal.
 544  */
 545 static int cpu_timer_sample_group(const clockid_t which_clock,
 546                                   struct task_struct *p, u64 *sample)
 547 {
 548         struct task_cputime cputime;
 549
 550         thread_group_cputimer(p, &cputime);
 551         switch (CPUCLOCK_WHICH(which_clock)) {
 552         default:
 553                 return -EINVAL;
 554         case CPUCLOCK_PROF:
 555                 *sample = cputime.utime + cputime.stime;
 556                 break;
 557         case CPUCLOCK_VIRT:
 558                 *sample = cputime.utime;
 559                 break;
 560         case CPUCLOCK_SCHED:
 561                 *sample = cputime.sum_exec_runtime;
 562                 break;
 563         }
 564         return 0;
 565 }
 566
 567 /*
 568  * Guts of sys_timer_settime for CPU timers.
 569  * This is called with the timer locked and interrupts disabled.
 570  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 571  * and try again.  (This happens when the timer is in the middle of firing.)
 572  */
 573 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 574                                struct itimerspec64 *new, struct itimerspec64 *old)
 575 {
 576         unsigned long flags;
 577         struct sighand_struct *sighand;
 578         struct task_struct *p = timer->it.cpu.task;
 579         u64 old_expires, new_expires, old_incr, val;
 580         int ret;
 581
 582         WARN_ON_ONCE(p == NULL);
 583
 584         /*
 585          * Use the to_ktime conversion because that clamps the maximum
 586          * value to KTIME_MAX and avoid multiplication overflows.
 587          */
 588         new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
 589
 590         /*
 591          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 592          * and p->signal->cpu_timers read/write in arm_timer()
 593          */
 594         sighand = lock_task_sighand(p, &flags);
 595         /*
 596          * If p has just been reaped, we can no
 597          * longer get any information about it at all.
 598          */
 599         if (unlikely(sighand == NULL)) {
 600                 return -ESRCH;
 601         }
 602
 603         /*
 604          * Disarm any old timer after extracting its expiry time.
 605          */
 606         WARN_ON_ONCE(!irqs_disabled());
 607
 608         ret = 0;
 609         old_incr = timer->it.cpu.incr;
 610         old_expires = timer->it.cpu.expires;
 611         if (unlikely(timer->it.cpu.firing)) {
 612                 timer->it.cpu.firing = -1;
 613                 ret = TIMER_RETRY;
 614         } else
 615                 list_del_init(&timer->it.cpu.entry);
 616
 617         /*
 618          * We need to sample the current value to convert the new
 619          * value from to relative and absolute, and to convert the
 620          * old value from absolute to relative.  To set a process
 621          * timer, we need a sample to balance the thread expiry
 622          * times (in arm_timer).  With an absolute time, we must
 623          * check if it's already passed.  In short, we need a sample.
 624          */
 625         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 626                 cpu_clock_sample(timer->it_clock, p, &val);
 627         } else {
 628                 cpu_timer_sample_group(timer->it_clock, p, &val);
 629         }
 630
 631         if (old) {
 632                 if (old_expires == 0) {
 633                         old->it_value.tv_sec = 0;
 634                         old->it_value.tv_nsec = 0;
 635                 } else {
 636                         /*
 637                          * Update the timer in case it has
 638                          * overrun already.  If it has,
 639                          * we'll report it as having overrun
 640                          * and with the next reloaded timer
 641                          * already ticking, though we are
 642                          * swallowing that pending
 643                          * notification here to install the
 644                          * new setting.
 645                          */
 646                         bump_cpu_timer(timer, val);
 647                         if (val < timer->it.cpu.expires) {
 648                                 old_expires = timer->it.cpu.expires - val;
 649                                 old->it_value = ns_to_timespec64(old_expires);
 650                         } else {
 651                                 old->it_value.tv_nsec = 1;
 652                                 old->it_value.tv_sec = 0;
 653                         }
 654                 }
 655         }
 656
 657         if (unlikely(ret)) {
 658                 /*
 659                  * We are colliding with the timer actually firing.
 660                  * Punt after filling in the timer's old value, and
 661                  * disable this firing since we are already reporting
 662                  * it as an overrun (thanks to bump_cpu_timer above).
 663                  */
 664                 unlock_task_sighand(p, &flags);
 665                 goto out;
 666         }
 667
 668         if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
 669                 new_expires += val;
 670         }
 671
 672         /*
 673          * Install the new expiry time (or zero).
 674          * For a timer with no notification action, we don't actually
 675          * arm the timer (we'll just fake it for timer_gettime).
 676          */
 677         timer->it.cpu.expires = new_expires;
 678         if (new_expires != 0 && val < new_expires) {
 679                 arm_timer(timer);
 680         }
 681
 682         unlock_task_sighand(p, &flags);
 683         /*
 684          * Install the new reload setting, and
 685          * set up the signal and overrun bookkeeping.
 686          */
 687         timer->it.cpu.incr = timespec64_to_ns(&new->it_interval);
 688         timer->it_interval = ns_to_ktime(timer->it.cpu.incr);
 689
 690         /*
 691          * This acts as a modification timestamp for the timer,
 692          * so any automatic reload attempt will punt on seeing
 693          * that we have reset the timer manually.
 694          */
 695         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 696                 ~REQUEUE_PENDING;
 697         timer->it_overrun_last = 0;
 698         timer->it_overrun = -1;
 699
 700         if (new_expires != 0 && !(val < new_expires)) {
 701                 /*
 702                  * The designated time already passed, so we notify
 703                  * immediately, even if the thread never runs to
 704                  * accumulate more time on this clock.
 705                  */
 706                 cpu_timer_fire(timer);
 707         }
 708
 709         ret = 0;
 710  out:
 711         if (old)
 712                 old->it_interval = ns_to_timespec64(old_incr);
 713
 714         return ret;
 715 }
 716
 717 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
 718 {
 719         u64 now;
 720         struct task_struct *p = timer->it.cpu.task;
 721
 722         WARN_ON_ONCE(p == NULL);
 723
 724         /*
 725          * Easy part: convert the reload time.
 726          */
 727         itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
 728
 729         if (!timer->it.cpu.expires)
 730                 return;
 731
 732         /*
 733          * Sample the clock to take the difference with the expiry time.
 734          */
 735         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 736                 cpu_clock_sample(timer->it_clock, p, &now);
 737         } else {
 738                 struct sighand_struct *sighand;
 739                 unsigned long flags;
 740
 741                 /*
 742                  * Protect against sighand release/switch in exit/exec and
 743                  * also make timer sampling safe if it ends up calling
 744                  * thread_group_cputime().
 745                  */
 746                 sighand = lock_task_sighand(p, &flags);
 747                 if (unlikely(sighand == NULL)) {
 748                         /*
 749                          * The process has been reaped.
 750                          * We can't even collect a sample any more.
 751                          * Call the timer disarmed, nothing else to do.
 752                          */
 753                         timer->it.cpu.expires = 0;
 754                         return;
 755                 } else {
 756                         cpu_timer_sample_group(timer->it_clock, p, &now);
 757                         unlock_task_sighand(p, &flags);
 758                 }
 759         }
 760
 761         if (now < timer->it.cpu.expires) {
 762                 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
 763         } else {
 764                 /*
 765                  * The timer should have expired already, but the firing
 766                  * hasn't taken place yet.  Say it's just about to expire.
 767                  */
 768                 itp->it_value.tv_nsec = 1;
 769                 itp->it_value.tv_sec = 0;
 770         }
 771 }
 772
 773 static unsigned long long
 774 check_timers_list(struct list_head *timers,
 775                   struct list_head *firing,
 776                   unsigned long long curr)
 777 {
 778         int maxfire = 20;
 779
 780         while (!list_empty(timers)) {
 781                 struct cpu_timer_list *t;
 782
 783                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 784
 785                 if (!--maxfire || curr < t->expires)
 786                         return t->expires;
 787
 788                 t->firing = 1;
 789                 list_move_tail(&t->entry, firing);
 790         }
 791
 792         return 0;
 793 }
 794
 795 /*
 796  * Check for any per-thread CPU timers that have fired and move them off
 797  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 798  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 799  */
 800 static void check_thread_timers(struct task_struct *tsk,
 801                                 struct list_head *firing)
 802 {
 803         struct list_head *timers = tsk->cpu_timers;
 804         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 805         u64 expires;
 806         unsigned long soft;
 807
 808         /*
 809          * If cputime_expires is zero, then there are no active
 810          * per thread CPU timers.
 811          */
 812         if (task_cputime_zero(&tsk->cputime_expires))
 813                 return;
 814
 815         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 816         tsk_expires->prof_exp = expires;
 817
 818         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 819         tsk_expires->virt_exp = expires;
 820
 821         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 822                                                    tsk->se.sum_exec_runtime);
 823
 824         /*
 825          * Check for the special case thread timers.
 826          */
 827         soft = task_rlimit(tsk, RLIMIT_RTTIME);
 828         if (soft != RLIM_INFINITY) {
 829                 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
 830
 831                 if (hard != RLIM_INFINITY &&
 832                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 833                         /*
 834                          * At the hard limit, we just die.
 835                          * No need to calculate anything else now.
 836                          */
 837                         if (print_fatal_signals) {
 838                                 pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
 839                                         tsk->comm, task_pid_nr(tsk));
 840                         }
 841                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 842                         return;
 843                 }
 844                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 845                         /*
 846                          * At the soft limit, send a SIGXCPU every second.
 847                          */
 848                         if (soft < hard) {
 849                                 soft += USEC_PER_SEC;
 850                                 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
 851                                         soft;
 852                         }
 853                         if (print_fatal_signals) {
 854                                 pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
 855                                         tsk->comm, task_pid_nr(tsk));
 856                         }
 857                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 858                 }
 859         }
 860         if (task_cputime_zero(tsk_expires))
 861                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 862 }
 863
 864 static inline void stop_process_timers(struct signal_struct *sig)
 865 {
 866         struct thread_group_cputimer *cputimer = &sig->cputimer;
 867
 868         /* Turn off cputimer->running. This is done without locking. */
 869         WRITE_ONCE(cputimer->running, false);
 870         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 871 }
 872
 873 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 874                              u64 *expires, u64 cur_time, int signo)
 875 {
 876         if (!it->expires)
 877                 return;
 878
 879         if (cur_time >= it->expires) {
 880                 if (it->incr)
 881                         it->expires += it->incr;
 882                 else
 883                         it->expires = 0;
 884
 885                 trace_itimer_expire(signo == SIGPROF ?
 886                                     ITIMER_PROF : ITIMER_VIRTUAL,
 887                                     tsk->signal->leader_pid, cur_time);
 888                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 889         }
 890
 891         if (it->expires && (!*expires || it->expires < *expires))
 892                 *expires = it->expires;
 893 }
 894
 895 /*
 896  * Check for any per-thread CPU timers that have fired and move them
 897  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 898  * have already been taken off.
 899  */
 900 static void check_process_timers(struct task_struct *tsk,
 901                                  struct list_head *firing)
 902 {
 903         struct signal_struct *const sig = tsk->signal;
 904         u64 utime, ptime, virt_expires, prof_expires;
 905         u64 sum_sched_runtime, sched_expires;
 906         struct list_head *timers = sig->cpu_timers;
 907         struct task_cputime cputime;
 908         unsigned long soft;
 909
 910         /*
 911          * If cputimer is not running, then there are no active
 912          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 913          */
 914         if (!READ_ONCE(tsk->signal->cputimer.running))
 915                 return;
 916
 917         /*
 918          * Signify that a thread is checking for process timers.
 919          * Write access to this field is protected by the sighand lock.
 920          */
 921         sig->cputimer.checking_timer = true;
 922
 923         /*
 924          * Collect the current process totals.
 925          */
 926         thread_group_cputimer(tsk, &cputime);
 927         utime = cputime.utime;
 928         ptime = utime + cputime.stime;
 929         sum_sched_runtime = cputime.sum_exec_runtime;
 930
 931         prof_expires = check_timers_list(timers, firing, ptime);
 932         virt_expires = check_timers_list(++timers, firing, utime);
 933         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 934
 935         /*
 936          * Check for the special case process timers.
 937          */
 938         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 939                          SIGPROF);
 940         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 941                          SIGVTALRM);
 942         soft = task_rlimit(tsk, RLIMIT_CPU);
 943         if (soft != RLIM_INFINITY) {
 944                 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
 945                 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
 946                 u64 x;
 947                 if (psecs >= hard) {
 948                         /*
 949                          * At the hard limit, we just die.
 950                          * No need to calculate anything else now.
 951                          */
 952                         if (print_fatal_signals) {
 953                                 pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
 954                                         tsk->comm, task_pid_nr(tsk));
 955                         }
 956                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 957                         return;
 958                 }
 959                 if (psecs >= soft) {
 960                         /*
 961                          * At the soft limit, send a SIGXCPU every second.
 962                          */
 963                         if (print_fatal_signals) {
 964                                 pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
 965                                         tsk->comm, task_pid_nr(tsk));
 966                         }
 967                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 968                         if (soft < hard) {
 969                                 soft++;
 970                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 971                         }
 972                 }
 973                 x = soft * NSEC_PER_SEC;
 974                 if (!prof_expires || x < prof_expires)
 975                         prof_expires = x;
 976         }
 977
 978         sig->cputime_expires.prof_exp = prof_expires;
 979         sig->cputime_expires.virt_exp = virt_expires;
 980         sig->cputime_expires.sched_exp = sched_expires;
 981         if (task_cputime_zero(&sig->cputime_expires))
 982                 stop_process_timers(sig);
 983
 984         sig->cputimer.checking_timer = false;
 985 }
 986
 987 /*
 988  * This is called from the signal code (via posixtimer_rearm)
 989  * when the last timer signal was delivered and we have to reload the timer.
 990  */
 991 static void posix_cpu_timer_rearm(struct k_itimer *timer)
 992 {
 993         struct sighand_struct *sighand;
 994         unsigned long flags;
 995         struct task_struct *p = timer->it.cpu.task;
 996         u64 now;
 997
 998         WARN_ON_ONCE(p == NULL);
 999
1000         /*
1001          * Fetch the current sample and update the timer's expiry time.
1002          */
1003         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1004                 cpu_clock_sample(timer->it_clock, p, &now);
1005                 bump_cpu_timer(timer, now);
1006                 if (unlikely(p->exit_state))
1007                         return;
1008
1009                 /* Protect timer list r/w in arm_timer() */
1010                 sighand = lock_task_sighand(p, &flags);
1011                 if (!sighand)
1012                         return;
1013         } else {
1014                 /*
1015                  * Protect arm_timer() and timer sampling in case of call to
1016                  * thread_group_cputime().
1017                  */
1018                 sighand = lock_task_sighand(p, &flags);
1019                 if (unlikely(sighand == NULL)) {
1020                         /*
1021                          * The process has been reaped.
1022                          * We can't even collect a sample any more.
1023                          */
1024                         timer->it.cpu.expires = 0;
1025                         return;
1026                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1027                         /* If the process is dying, no need to rearm */
1028                         goto unlock;
1029                 }
1030                 cpu_timer_sample_group(timer->it_clock, p, &now);
1031                 bump_cpu_timer(timer, now);
1032                 /* Leave the sighand locked for the call below.  */
1033         }
1034
1035         /*
1036          * Now re-arm for the new expiry time.
1037          */
1038         WARN_ON_ONCE(!irqs_disabled());
1039         arm_timer(timer);
1040 unlock:
1041         unlock_task_sighand(p, &flags);
1042 }
1043
1044 /**
1045  * task_cputime_expired - Compare two task_cputime entities.
1046  *
1047  * @sample:     The task_cputime structure to be checked for expiration.
1048  * @expires:    Expiration times, against which @sample will be checked.
1049  *
1050  * Checks @sample against @expires to see if any field of @sample has expired.
1051  * Returns true if any field of the former is greater than the corresponding
1052  * field of the latter if the latter field is set.  Otherwise returns false.
1053  */
1054 static inline int task_cputime_expired(const struct task_cputime *sample,
1055                                         const struct task_cputime *expires)
1056 {
1057         if (expires->utime && sample->utime >= expires->utime)
1058                 return 1;
1059         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1060                 return 1;
1061         if (expires->sum_exec_runtime != 0 &&
1062             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1063                 return 1;
1064         return 0;
1065 }
1066
1067 /**
1068  * fastpath_timer_check - POSIX CPU timers fast path.
1069  *
1070  * @tsk:        The task (thread) being checked.
1071  *
1072  * Check the task and thread group timers.  If both are zero (there are no
1073  * timers set) return false.  Otherwise snapshot the task and thread group
1074  * timers and compare them with the corresponding expiration times.  Return
1075  * true if a timer has expired, else return false.
1076  */
1077 static inline int fastpath_timer_check(struct task_struct *tsk)
1078 {
1079         struct signal_struct *sig;
1080
1081         if (!task_cputime_zero(&tsk->cputime_expires)) {
1082                 struct task_cputime task_sample;
1083
1084                 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1085                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1086                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1087                         return 1;
1088         }
1089
1090         sig = tsk->signal;
1091         /*
1092          * Check if thread group timers expired when the cputimer is
1093          * running and no other thread in the group is already checking
1094          * for thread group cputimers. These fields are read without the
1095          * sighand lock. However, this is fine because this is meant to
1096          * be a fastpath heuristic to determine whether we should try to
1097          * acquire the sighand lock to check/handle timers.
1098          *
1099          * In the worst case scenario, if 'running' or 'checking_timer' gets
1100          * set but the current thread doesn't see the change yet, we'll wait
1101          * until the next thread in the group gets a scheduler interrupt to
1102          * handle the timer. This isn't an issue in practice because these
1103          * types of delays with signals actually getting sent are expected.
1104          */
1105         if (READ_ONCE(sig->cputimer.running) &&
1106             !READ_ONCE(sig->cputimer.checking_timer)) {
1107                 struct task_cputime group_sample;
1108
1109                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1110
1111                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1112                         return 1;
1113         }
1114
1115         return 0;
1116 }
1117
1118 /*
1119  * This is called from the timer interrupt handler.  The irq handler has
1120  * already updated our counts.  We need to check if any timers fire now.
1121  * Interrupts are disabled.
1122  */
1123 void run_posix_cpu_timers(struct task_struct *tsk)
1124 {
1125         LIST_HEAD(firing);
1126         struct k_itimer *timer, *next;
1127         unsigned long flags;
1128
1129         WARN_ON_ONCE(!irqs_disabled());
1130
1131         /*
1132          * The fast path checks that there are no expired thread or thread
1133          * group timers.  If that's so, just return.
1134          */
1135         if (!fastpath_timer_check(tsk))
1136                 return;
1137
1138         if (!lock_task_sighand(tsk, &flags))
1139                 return;
1140         /*
1141          * Here we take off tsk->signal->cpu_timers[N] and
1142          * tsk->cpu_timers[N] all the timers that are firing, and
1143          * put them on the firing list.
1144          */
1145         check_thread_timers(tsk, &firing);
1146
1147         check_process_timers(tsk, &firing);
1148
1149         /*
1150          * We must release these locks before taking any timer's lock.
1151          * There is a potential race with timer deletion here, as the
1152          * siglock now protects our private firing list.  We have set
1153          * the firing flag in each timer, so that a deletion attempt
1154          * that gets the timer lock before we do will give it up and
1155          * spin until we've taken care of that timer below.
1156          */
1157         unlock_task_sighand(tsk, &flags);
1158
1159         /*
1160          * Now that all the timers on our list have the firing flag,
1161          * no one will touch their list entries but us.  We'll take
1162          * each timer's lock before clearing its firing flag, so no
1163          * timer call will interfere.
1164          */
1165         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1166                 int cpu_firing;
1167
1168                 spin_lock(&timer->it_lock);
1169                 list_del_init(&timer->it.cpu.entry);
1170                 cpu_firing = timer->it.cpu.firing;
1171                 timer->it.cpu.firing = 0;
1172                 /*
1173                  * The firing flag is -1 if we collided with a reset
1174                  * of the timer, which already reported this
1175                  * almost-firing as an overrun.  So don't generate an event.
1176                  */
1177                 if (likely(cpu_firing >= 0))
1178                         cpu_timer_fire(timer);
1179                 spin_unlock(&timer->it_lock);
1180         }
1181 }
1182
1183 /*
1184  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1185  * The tsk->sighand->siglock must be held by the caller.
1186  */
1187 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1188                            u64 *newval, u64 *oldval)
1189 {
1190         u64 now;
1191
1192         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1193         cpu_timer_sample_group(clock_idx, tsk, &now);
1194
1195         if (oldval) {
1196                 /*
1197                  * We are setting itimer. The *oldval is absolute and we update
1198                  * it to be relative, *newval argument is relative and we update
1199                  * it to be absolute.
1200                  */
1201                 if (*oldval) {
1202                         if (*oldval <= now) {
1203                                 /* Just about to fire. */
1204                                 *oldval = TICK_NSEC;
1205                         } else {
1206                                 *oldval -= now;
1207                         }
1208                 }
1209
1210                 if (!*newval)
1211                         return;
1212                 *newval += now;
1213         }
1214
1215         /*
1216          * Update expiration cache if we are the earliest timer, or eventually
1217          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1218          */
1219         switch (clock_idx) {
1220         case CPUCLOCK_PROF:
1221                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1222                         tsk->signal->cputime_expires.prof_exp = *newval;
1223                 break;
1224         case CPUCLOCK_VIRT:
1225                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1226                         tsk->signal->cputime_expires.virt_exp = *newval;
1227                 break;
1228         }
1229
1230         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1231 }
1232
1233 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1234                             const struct timespec64 *rqtp)
1235 {
1236         struct itimerspec64 it;
1237         struct k_itimer timer;
1238         u64 expires;
1239         int error;
1240
1241         /*
1242          * Set up a temporary timer and then wait for it to go off.
1243          */
1244         memset(&timer, 0, sizeof timer);
1245         spin_lock_init(&timer.it_lock);
1246         timer.it_clock = which_clock;
1247         timer.it_overrun = -1;
1248         error = posix_cpu_timer_create(&timer);
1249         timer.it_process = current;
1250         if (!error) {
1251                 static struct itimerspec64 zero_it;
1252                 struct restart_block *restart;
1253
1254                 memset(&it, 0, sizeof(it));
1255                 it.it_value = *rqtp;
1256
1257                 spin_lock_irq(&timer.it_lock);
1258                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1259                 if (error) {
1260                         spin_unlock_irq(&timer.it_lock);
1261                         return error;
1262                 }
1263
1264                 while (!signal_pending(current)) {
1265                         if (timer.it.cpu.expires == 0) {
1266                                 /*
1267                                  * Our timer fired and was reset, below
1268                                  * deletion can not fail.
1269                                  */
1270                                 posix_cpu_timer_del(&timer);
1271                                 spin_unlock_irq(&timer.it_lock);
1272                                 return 0;
1273                         }
1274
1275                         /*
1276                          * Block until cpu_timer_fire (or a signal) wakes us.
1277                          */
1278                         __set_current_state(TASK_INTERRUPTIBLE);
1279                         spin_unlock_irq(&timer.it_lock);
1280                         schedule();
1281                         spin_lock_irq(&timer.it_lock);
1282                 }
1283
1284                 /*
1285                  * We were interrupted by a signal.
1286                  */
1287                 expires = timer.it.cpu.expires;
1288                 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1289                 if (!error) {
1290                         /*
1291                          * Timer is now unarmed, deletion can not fail.
1292                          */
1293                         posix_cpu_timer_del(&timer);
1294                 }
1295                 spin_unlock_irq(&timer.it_lock);
1296
1297                 while (error == TIMER_RETRY) {
1298                         /*
1299                          * We need to handle case when timer was or is in the
1300                          * middle of firing. In other cases we already freed
1301                          * resources.
1302                          */
1303                         spin_lock_irq(&timer.it_lock);
1304                         error = posix_cpu_timer_del(&timer);
1305                         spin_unlock_irq(&timer.it_lock);
1306                 }
1307
1308                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1309                         /*
1310                          * It actually did fire already.
1311                          */
1312                         return 0;
1313                 }
1314
1315                 error = -ERESTART_RESTARTBLOCK;
1316                 /*
1317                  * Report back to the user the time still remaining.
1318                  */
1319                 restart = &current->restart_block;
1320                 restart->nanosleep.expires = expires;
1321                 if (restart->nanosleep.type != TT_NONE)
1322                         error = nanosleep_copyout(restart, &it.it_value);
1323         }
1324
1325         return error;
1326 }
1327
1328 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1329
1330 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1331                             const struct timespec64 *rqtp)
1332 {
1333         struct restart_block *restart_block = &current->restart_block;
1334         int error;
1335
1336         /*
1337          * Diagnose required errors first.
1338          */
1339         if (CPUCLOCK_PERTHREAD(which_clock) &&
1340             (CPUCLOCK_PID(which_clock) == 0 ||
1341              CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1342                 return -EINVAL;
1343
1344         error = do_cpu_nanosleep(which_clock, flags, rqtp);
1345
1346         if (error == -ERESTART_RESTARTBLOCK) {
1347
1348                 if (flags & TIMER_ABSTIME)
1349                         return -ERESTARTNOHAND;
1350
1351                 restart_block->nanosleep.clockid = which_clock;
1352                 set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1353         }
1354         return error;
1355 }
1356
1357 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1358 {
1359         clockid_t which_clock = restart_block->nanosleep.clockid;
1360         struct timespec64 t;
1361
1362         t = ns_to_timespec64(restart_block->nanosleep.expires);
1363
1364         return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1365 }
1366
1367 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1368 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1369
1370 static int process_cpu_clock_getres(const clockid_t which_clock,
1371                                     struct timespec64 *tp)
1372 {
1373         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1374 }
1375 static int process_cpu_clock_get(const clockid_t which_clock,
1376                                  struct timespec64 *tp)
1377 {
1378         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1379 }
1380 static int process_cpu_timer_create(struct k_itimer *timer)
1381 {
1382         timer->it_clock = PROCESS_CLOCK;
1383         return posix_cpu_timer_create(timer);
1384 }
1385 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1386                               const struct timespec64 *rqtp)
1387 {
1388         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1389 }
1390 static int thread_cpu_clock_getres(const clockid_t which_clock,
1391                                    struct timespec64 *tp)
1392 {
1393         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1394 }
1395 static int thread_cpu_clock_get(const clockid_t which_clock,
1396                                 struct timespec64 *tp)
1397 {
1398         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1399 }
1400 static int thread_cpu_timer_create(struct k_itimer *timer)
1401 {
1402         timer->it_clock = THREAD_CLOCK;
1403         return posix_cpu_timer_create(timer);
1404 }
1405
1406 const struct k_clock clock_posix_cpu = {
1407         .clock_getres   = posix_cpu_clock_getres,
1408         .clock_set      = posix_cpu_clock_set,
1409         .clock_get      = posix_cpu_clock_get,
1410         .timer_create   = posix_cpu_timer_create,
1411         .nsleep         = posix_cpu_nsleep,
1412         .timer_set      = posix_cpu_timer_set,
1413         .timer_del      = posix_cpu_timer_del,
1414         .timer_get      = posix_cpu_timer_get,
1415         .timer_rearm    = posix_cpu_timer_rearm,
1416 };
1417
1418 const struct k_clock clock_process = {
1419         .clock_getres   = process_cpu_clock_getres,
1420         .clock_get      = process_cpu_clock_get,
1421         .timer_create   = process_cpu_timer_create,
1422         .nsleep         = process_cpu_nsleep,
1423 };
1424
1425 const struct k_clock clock_thread = {
1426         .clock_getres   = thread_cpu_clock_getres,
1427         .clock_get      = thread_cpu_clock_get,
1428         .timer_create   = thread_cpu_timer_create,
1429 };