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 #include <linux/sched/deadline.h>
  18
  19 #include "posix-timers.h"
  20
  21 static void posix_cpu_timer_rearm(struct k_itimer *timer);
  22
  23 /*
  24  * Called after updating RLIMIT_CPU to run cpu timer and update
  25  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  26  * siglock protection since other code may update expiration cache as
  27  * well.
  28  */
  29 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  30 {
  31         u64 nsecs = rlim_new * NSEC_PER_SEC;
  32
  33         spin_lock_irq(&task->sighand->siglock);
  34         set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
  35         spin_unlock_irq(&task->sighand->siglock);
  36 }
  37
  38 static int check_clock(const clockid_t which_clock)
  39 {
  40         int error = 0;
  41         struct task_struct *p;
  42         const pid_t pid = CPUCLOCK_PID(which_clock);
  43
  44         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  45                 return -EINVAL;
  46
  47         if (pid == 0)
  48                 return 0;
  49
  50         rcu_read_lock();
  51         p = find_task_by_vpid(pid);
  52         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  53                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  54                 error = -EINVAL;
  55         }
  56         rcu_read_unlock();
  57
  58         return error;
  59 }
  60
  61 /*
  62  * Update expiry time from increment, and increase overrun count,
  63  * given the current clock sample.
  64  */
  65 static void bump_cpu_timer(struct k_itimer *timer, u64 now)
  66 {
  67         int i;
  68         u64 delta, incr;
  69
  70         if (timer->it.cpu.incr == 0)
  71                 return;
  72
  73         if (now < timer->it.cpu.expires)
  74                 return;
  75
  76         incr = timer->it.cpu.incr;
  77         delta = now + incr - timer->it.cpu.expires;
  78
  79         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  80         for (i = 0; incr < delta - incr; i++)
  81                 incr = incr << 1;
  82
  83         for (; i >= 0; incr >>= 1, i--) {
  84                 if (delta < incr)
  85                         continue;
  86
  87                 timer->it.cpu.expires += incr;
  88                 timer->it_overrun += 1LL << i;
  89                 delta -= incr;
  90         }
  91 }
  92
  93 /**
  94  * task_cputime_zero - Check a task_cputime struct for all zero fields.
  95  *
  96  * @cputime:    The struct to compare.
  97  *
  98  * Checks @cputime to see if all fields are zero.  Returns true if all fields
  99  * are zero, false if any field is nonzero.
 100  */
 101 static inline int task_cputime_zero(const struct task_cputime *cputime)
 102 {
 103         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 104                 return 1;
 105         return 0;
 106 }
 107
 108 static inline u64 prof_ticks(struct task_struct *p)
 109 {
 110         u64 utime, stime;
 111
 112         task_cputime(p, &utime, &stime);
 113
 114         return utime + stime;
 115 }
 116 static inline u64 virt_ticks(struct task_struct *p)
 117 {
 118         u64 utime, stime;
 119
 120         task_cputime(p, &utime, &stime);
 121
 122         return utime;
 123 }
 124
 125 static int
 126 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
 127 {
 128         int error = check_clock(which_clock);
 129         if (!error) {
 130                 tp->tv_sec = 0;
 131                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 132                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 133                         /*
 134                          * If sched_clock is using a cycle counter, we
 135                          * don't have any idea of its true resolution
 136                          * exported, but it is much more than 1s/HZ.
 137                          */
 138                         tp->tv_nsec = 1;
 139                 }
 140         }
 141         return error;
 142 }
 143
 144 static int
 145 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
 146 {
 147         /*
 148          * You can never reset a CPU clock, but we check for other errors
 149          * in the call before failing with EPERM.
 150          */
 151         int error = check_clock(which_clock);
 152         if (error == 0) {
 153                 error = -EPERM;
 154         }
 155         return error;
 156 }
 157
 158
 159 /*
 160  * Sample a per-thread clock for the given task.
 161  */
 162 static int cpu_clock_sample(const clockid_t which_clock,
 163                             struct task_struct *p, u64 *sample)
 164 {
 165         switch (CPUCLOCK_WHICH(which_clock)) {
 166         default:
 167                 return -EINVAL;
 168         case CPUCLOCK_PROF:
 169                 *sample = prof_ticks(p);
 170                 break;
 171         case CPUCLOCK_VIRT:
 172                 *sample = virt_ticks(p);
 173                 break;
 174         case CPUCLOCK_SCHED:
 175                 *sample = task_sched_runtime(p);
 176                 break;
 177         }
 178         return 0;
 179 }
 180
 181 /*
 182  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
 183  * to avoid race conditions with concurrent updates to cputime.
 184  */
 185 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
 186 {
 187         u64 curr_cputime;
 188 retry:
 189         curr_cputime = atomic64_read(cputime);
 190         if (sum_cputime > curr_cputime) {
 191                 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
 192                         goto retry;
 193         }
 194 }
 195
 196 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
 197 {
 198         __update_gt_cputime(&cputime_atomic->utime, sum->utime);
 199         __update_gt_cputime(&cputime_atomic->stime, sum->stime);
 200         __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
 201 }
 202
 203 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
 204 static inline void sample_cputime_atomic(struct task_cputime *times,
 205                                          struct task_cputime_atomic *atomic_times)
 206 {
 207         times->utime = atomic64_read(&atomic_times->utime);
 208         times->stime = atomic64_read(&atomic_times->stime);
 209         times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
 210 }
 211
 212 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 213 {
 214         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 215         struct task_cputime sum;
 216
 217         /* Check if cputimer isn't running. This is accessed without locking. */
 218         if (!READ_ONCE(cputimer->running)) {
 219                 /*
 220                  * The POSIX timer interface allows for absolute time expiry
 221                  * values through the TIMER_ABSTIME flag, therefore we have
 222                  * to synchronize the timer to the clock every time we start it.
 223                  */
 224                 thread_group_cputime(tsk, &sum);
 225                 update_gt_cputime(&cputimer->cputime_atomic, &sum);
 226
 227                 /*
 228                  * We're setting cputimer->running without a lock. Ensure
 229                  * this only gets written to in one operation. We set
 230                  * running after update_gt_cputime() as a small optimization,
 231                  * but barriers are not required because update_gt_cputime()
 232                  * can handle concurrent updates.
 233                  */
 234                 WRITE_ONCE(cputimer->running, true);
 235         }
 236         sample_cputime_atomic(times, &cputimer->cputime_atomic);
 237 }
 238
 239 /*
 240  * Sample a process (thread group) clock for the given group_leader task.
 241  * Must be called with task sighand lock held for safe while_each_thread()
 242  * traversal.
 243  */
 244 static int cpu_clock_sample_group(const clockid_t which_clock,
 245                                   struct task_struct *p,
 246                                   u64 *sample)
 247 {
 248         struct task_cputime cputime;
 249
 250         switch (CPUCLOCK_WHICH(which_clock)) {
 251         default:
 252                 return -EINVAL;
 253         case CPUCLOCK_PROF:
 254                 thread_group_cputime(p, &cputime);
 255                 *sample = cputime.utime + cputime.stime;
 256                 break;
 257         case CPUCLOCK_VIRT:
 258                 thread_group_cputime(p, &cputime);
 259                 *sample = cputime.utime;
 260                 break;
 261         case CPUCLOCK_SCHED:
 262                 thread_group_cputime(p, &cputime);
 263                 *sample = cputime.sum_exec_runtime;
 264                 break;
 265         }
 266         return 0;
 267 }
 268
 269 static int posix_cpu_clock_get_task(struct task_struct *tsk,
 270                                     const clockid_t which_clock,
 271                                     struct timespec64 *tp)
 272 {
 273         int err = -EINVAL;
 274         u64 rtn;
 275
 276         if (CPUCLOCK_PERTHREAD(which_clock)) {
 277                 if (same_thread_group(tsk, current))
 278                         err = cpu_clock_sample(which_clock, tsk, &rtn);
 279         } else {
 280                 if (tsk == current || thread_group_leader(tsk))
 281                         err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 282         }
 283
 284         if (!err)
 285                 *tp = ns_to_timespec64(rtn);
 286
 287         return err;
 288 }
 289
 290
 291 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
 292 {
 293         const pid_t pid = CPUCLOCK_PID(which_clock);
 294         int err = -EINVAL;
 295
 296         if (pid == 0) {
 297                 /*
 298                  * Special case constant value for our own clocks.
 299                  * We don't have to do any lookup to find ourselves.
 300                  */
 301                 err = posix_cpu_clock_get_task(current, which_clock, tp);
 302         } else {
 303                 /*
 304                  * Find the given PID, and validate that the caller
 305                  * should be able to see it.
 306                  */
 307                 struct task_struct *p;
 308                 rcu_read_lock();
 309                 p = find_task_by_vpid(pid);
 310                 if (p)
 311                         err = posix_cpu_clock_get_task(p, which_clock, tp);
 312                 rcu_read_unlock();
 313         }
 314
 315         return err;
 316 }
 317
 318 /*
 319  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 320  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 321  * new timer already all-zeros initialized.
 322  */
 323 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 324 {
 325         int ret = 0;
 326         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 327         struct task_struct *p;
 328
 329         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 330                 return -EINVAL;
 331
 332         new_timer->kclock = &clock_posix_cpu;
 333
 334         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 335
 336         rcu_read_lock();
 337         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 338                 if (pid == 0) {
 339                         p = current;
 340                 } else {
 341                         p = find_task_by_vpid(pid);
 342                         if (p && !same_thread_group(p, current))
 343                                 p = NULL;
 344                 }
 345         } else {
 346                 if (pid == 0) {
 347                         p = current->group_leader;
 348                 } else {
 349                         p = find_task_by_vpid(pid);
 350                         if (p && !has_group_leader_pid(p))
 351                                 p = NULL;
 352                 }
 353         }
 354         new_timer->it.cpu.task = p;
 355         if (p) {
 356                 get_task_struct(p);
 357         } else {
 358                 ret = -EINVAL;
 359         }
 360         rcu_read_unlock();
 361
 362         return ret;
 363 }
 364
 365 /*
 366  * Clean up a CPU-clock timer that is about to be destroyed.
 367  * This is called from timer deletion with the timer already locked.
 368  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 369  * and try again.  (This happens when the timer is in the middle of firing.)
 370  */
 371 static int posix_cpu_timer_del(struct k_itimer *timer)
 372 {
 373         int ret = 0;
 374         unsigned long flags;
 375         struct sighand_struct *sighand;
 376         struct task_struct *p = timer->it.cpu.task;
 377
 378         WARN_ON_ONCE(p == NULL);
 379
 380         /*
 381          * Protect against sighand release/switch in exit/exec and process/
 382          * thread timer list entry concurrent read/writes.
 383          */
 384         sighand = lock_task_sighand(p, &flags);
 385         if (unlikely(sighand == NULL)) {
 386                 /*
 387                  * We raced with the reaping of the task.
 388                  * The deletion should have cleared us off the list.
 389                  */
 390                 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
 391         } else {
 392                 if (timer->it.cpu.firing)
 393                         ret = TIMER_RETRY;
 394                 else
 395                         list_del(&timer->it.cpu.entry);
 396
 397                 unlock_task_sighand(p, &flags);
 398         }
 399
 400         if (!ret)
 401                 put_task_struct(p);
 402
 403         return ret;
 404 }
 405
 406 static void cleanup_timers_list(struct list_head *head)
 407 {
 408         struct cpu_timer_list *timer, *next;
 409
 410         list_for_each_entry_safe(timer, next, head, entry)
 411                 list_del_init(&timer->entry);
 412 }
 413
 414 /*
 415  * Clean out CPU timers still ticking when a thread exited.  The task
 416  * pointer is cleared, and the expiry time is replaced with the residual
 417  * time for later timer_gettime calls to return.
 418  * This must be called with the siglock held.
 419  */
 420 static void cleanup_timers(struct list_head *head)
 421 {
 422         cleanup_timers_list(head);
 423         cleanup_timers_list(++head);
 424         cleanup_timers_list(++head);
 425 }
 426
 427 /*
 428  * These are both called with the siglock held, when the current thread
 429  * is being reaped.  When the final (leader) thread in the group is reaped,
 430  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 431  */
 432 void posix_cpu_timers_exit(struct task_struct *tsk)
 433 {
 434         cleanup_timers(tsk->cpu_timers);
 435 }
 436 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 437 {
 438         cleanup_timers(tsk->signal->cpu_timers);
 439 }
 440
 441 static inline int expires_gt(u64 expires, u64 new_exp)
 442 {
 443         return expires == 0 || expires > new_exp;
 444 }
 445
 446 /*
 447  * Insert the timer on the appropriate list before any timers that
 448  * expire later.  This must be called with the sighand lock held.
 449  */
 450 static void arm_timer(struct k_itimer *timer)
 451 {
 452         struct task_struct *p = timer->it.cpu.task;
 453         struct list_head *head, *listpos;
 454         struct task_cputime *cputime_expires;
 455         struct cpu_timer_list *const nt = &timer->it.cpu;
 456         struct cpu_timer_list *next;
 457
 458         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 459                 head = p->cpu_timers;
 460                 cputime_expires = &p->cputime_expires;
 461         } else {
 462                 head = p->signal->cpu_timers;
 463                 cputime_expires = &p->signal->cputime_expires;
 464         }
 465         head += CPUCLOCK_WHICH(timer->it_clock);
 466
 467         listpos = head;
 468         list_for_each_entry(next, head, entry) {
 469                 if (nt->expires < next->expires)
 470                         break;
 471                 listpos = &next->entry;
 472         }
 473         list_add(&nt->entry, listpos);
 474
 475         if (listpos == head) {
 476                 u64 exp = nt->expires;
 477
 478                 /*
 479                  * We are the new earliest-expiring POSIX 1.b timer, hence
 480                  * need to update expiration cache. Take into account that
 481                  * for process timers we share expiration cache with itimers
 482                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 483                  */
 484
 485                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 486                 case CPUCLOCK_PROF:
 487                         if (expires_gt(cputime_expires->prof_exp, exp))
 488                                 cputime_expires->prof_exp = exp;
 489                         break;
 490                 case CPUCLOCK_VIRT:
 491                         if (expires_gt(cputime_expires->virt_exp, exp))
 492                                 cputime_expires->virt_exp = exp;
 493                         break;
 494                 case CPUCLOCK_SCHED:
 495                         if (expires_gt(cputime_expires->sched_exp, exp))
 496                                 cputime_expires->sched_exp = exp;
 497                         break;
 498                 }
 499                 if (CPUCLOCK_PERTHREAD(timer->it_clock))
 500                         tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
 501                 else
 502                         tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
 503         }
 504 }
 505
 506 /*
 507  * The timer is locked, fire it and arrange for its reload.
 508  */
 509 static void cpu_timer_fire(struct k_itimer *timer)
 510 {
 511         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 512                 /*
 513                  * User don't want any signal.
 514                  */
 515                 timer->it.cpu.expires = 0;
 516         } else if (unlikely(timer->sigq == NULL)) {
 517                 /*
 518                  * This a special case for clock_nanosleep,
 519                  * not a normal timer from sys_timer_create.
 520                  */
 521                 wake_up_process(timer->it_process);
 522                 timer->it.cpu.expires = 0;
 523         } else if (timer->it.cpu.incr == 0) {
 524                 /*
 525                  * One-shot timer.  Clear it as soon as it's fired.
 526                  */
 527                 posix_timer_event(timer, 0);
 528                 timer->it.cpu.expires = 0;
 529         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 530                 /*
 531                  * The signal did not get queued because the signal
 532                  * was ignored, so we won't get any callback to
 533                  * reload the timer.  But we need to keep it
 534                  * ticking in case the signal is deliverable next time.
 535                  */
 536                 posix_cpu_timer_rearm(timer);
 537                 ++timer->it_requeue_pending;
 538         }
 539 }
 540
 541 /*
 542  * Sample a process (thread group) timer for the given group_leader task.
 543  * Must be called with task sighand lock held for safe while_each_thread()
 544  * traversal.
 545  */
 546 static int cpu_timer_sample_group(const clockid_t which_clock,
 547                                   struct task_struct *p, u64 *sample)
 548 {
 549         struct task_cputime cputime;
 550
 551         thread_group_cputimer(p, &cputime);
 552         switch (CPUCLOCK_WHICH(which_clock)) {
 553         default:
 554                 return -EINVAL;
 555         case CPUCLOCK_PROF:
 556                 *sample = cputime.utime + cputime.stime;
 557                 break;
 558         case CPUCLOCK_VIRT:
 559                 *sample = cputime.utime;
 560                 break;
 561         case CPUCLOCK_SCHED:
 562                 *sample = cputime.sum_exec_runtime;
 563                 break;
 564         }
 565         return 0;
 566 }
 567
 568 /*
 569  * Guts of sys_timer_settime for CPU timers.
 570  * This is called with the timer locked and interrupts disabled.
 571  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 572  * and try again.  (This happens when the timer is in the middle of firing.)
 573  */
 574 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
 575                                struct itimerspec64 *new, struct itimerspec64 *old)
 576 {
 577         unsigned long flags;
 578         struct sighand_struct *sighand;
 579         struct task_struct *p = timer->it.cpu.task;
 580         u64 old_expires, new_expires, old_incr, val;
 581         int ret;
 582
 583         WARN_ON_ONCE(p == NULL);
 584
 585         /*
 586          * Use the to_ktime conversion because that clamps the maximum
 587          * value to KTIME_MAX and avoid multiplication overflows.
 588          */
 589         new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
 590
 591         /*
 592          * Protect against sighand release/switch in exit/exec and p->cpu_timers
 593          * and p->signal->cpu_timers read/write in arm_timer()
 594          */
 595         sighand = lock_task_sighand(p, &flags);
 596         /*
 597          * If p has just been reaped, we can no
 598          * longer get any information about it at all.
 599          */
 600         if (unlikely(sighand == NULL)) {
 601                 return -ESRCH;
 602         }
 603
 604         /*
 605          * Disarm any old timer after extracting its expiry time.
 606          */
 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
 689         /*
 690          * This acts as a modification timestamp for the timer,
 691          * so any automatic reload attempt will punt on seeing
 692          * that we have reset the timer manually.
 693          */
 694         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 695                 ~REQUEUE_PENDING;
 696         timer->it_overrun_last = 0;
 697         timer->it_overrun = -1;
 698
 699         if (new_expires != 0 && !(val < new_expires)) {
 700                 /*
 701                  * The designated time already passed, so we notify
 702                  * immediately, even if the thread never runs to
 703                  * accumulate more time on this clock.
 704                  */
 705                 cpu_timer_fire(timer);
 706         }
 707
 708         ret = 0;
 709  out:
 710         if (old)
 711                 old->it_interval = ns_to_timespec64(old_incr);
 712
 713         return ret;
 714 }
 715
 716 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
 717 {
 718         u64 now;
 719         struct task_struct *p = timer->it.cpu.task;
 720
 721         WARN_ON_ONCE(p == NULL);
 722
 723         /*
 724          * Easy part: convert the reload time.
 725          */
 726         itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
 727
 728         if (!timer->it.cpu.expires)
 729                 return;
 730
 731         /*
 732          * Sample the clock to take the difference with the expiry time.
 733          */
 734         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 735                 cpu_clock_sample(timer->it_clock, p, &now);
 736         } else {
 737                 struct sighand_struct *sighand;
 738                 unsigned long flags;
 739
 740                 /*
 741                  * Protect against sighand release/switch in exit/exec and
 742                  * also make timer sampling safe if it ends up calling
 743                  * thread_group_cputime().
 744                  */
 745                 sighand = lock_task_sighand(p, &flags);
 746                 if (unlikely(sighand == NULL)) {
 747                         /*
 748                          * The process has been reaped.
 749                          * We can't even collect a sample any more.
 750                          * Call the timer disarmed, nothing else to do.
 751                          */
 752                         timer->it.cpu.expires = 0;
 753                         return;
 754                 } else {
 755                         cpu_timer_sample_group(timer->it_clock, p, &now);
 756                         unlock_task_sighand(p, &flags);
 757                 }
 758         }
 759
 760         if (now < timer->it.cpu.expires) {
 761                 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
 762         } else {
 763                 /*
 764                  * The timer should have expired already, but the firing
 765                  * hasn't taken place yet.  Say it's just about to expire.
 766                  */
 767                 itp->it_value.tv_nsec = 1;
 768                 itp->it_value.tv_sec = 0;
 769         }
 770 }
 771
 772 static unsigned long long
 773 check_timers_list(struct list_head *timers,
 774                   struct list_head *firing,
 775                   unsigned long long curr)
 776 {
 777         int maxfire = 20;
 778
 779         while (!list_empty(timers)) {
 780                 struct cpu_timer_list *t;
 781
 782                 t = list_first_entry(timers, struct cpu_timer_list, entry);
 783
 784                 if (!--maxfire || curr < t->expires)
 785                         return t->expires;
 786
 787                 t->firing = 1;
 788                 list_move_tail(&t->entry, firing);
 789         }
 790
 791         return 0;
 792 }
 793
 794 static inline void check_dl_overrun(struct task_struct *tsk)
 795 {
 796         if (tsk->dl.dl_overrun) {
 797                 tsk->dl.dl_overrun = 0;
 798                 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 799         }
 800 }
 801
 802 /*
 803  * Check for any per-thread CPU timers that have fired and move them off
 804  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 805  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 806  */
 807 static void check_thread_timers(struct task_struct *tsk,
 808                                 struct list_head *firing)
 809 {
 810         struct list_head *timers = tsk->cpu_timers;
 811         struct task_cputime *tsk_expires = &tsk->cputime_expires;
 812         u64 expires;
 813         unsigned long soft;
 814
 815         if (dl_task(tsk))
 816                 check_dl_overrun(tsk);
 817
 818         /*
 819          * If cputime_expires is zero, then there are no active
 820          * per thread CPU timers.
 821          */
 822         if (task_cputime_zero(&tsk->cputime_expires))
 823                 return;
 824
 825         expires = check_timers_list(timers, firing, prof_ticks(tsk));
 826         tsk_expires->prof_exp = expires;
 827
 828         expires = check_timers_list(++timers, firing, virt_ticks(tsk));
 829         tsk_expires->virt_exp = expires;
 830
 831         tsk_expires->sched_exp = check_timers_list(++timers, firing,
 832                                                    tsk->se.sum_exec_runtime);
 833
 834         /*
 835          * Check for the special case thread timers.
 836          */
 837         soft = task_rlimit(tsk, RLIMIT_RTTIME);
 838         if (soft != RLIM_INFINITY) {
 839                 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
 840
 841                 if (hard != RLIM_INFINITY &&
 842                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 843                         /*
 844                          * At the hard limit, we just die.
 845                          * No need to calculate anything else now.
 846                          */
 847                         if (print_fatal_signals) {
 848                                 pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
 849                                         tsk->comm, task_pid_nr(tsk));
 850                         }
 851                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 852                         return;
 853                 }
 854                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 855                         /*
 856                          * At the soft limit, send a SIGXCPU every second.
 857                          */
 858                         if (soft < hard) {
 859                                 soft += USEC_PER_SEC;
 860                                 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
 861                                         soft;
 862                         }
 863                         if (print_fatal_signals) {
 864                                 pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
 865                                         tsk->comm, task_pid_nr(tsk));
 866                         }
 867                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 868                 }
 869         }
 870         if (task_cputime_zero(tsk_expires))
 871                 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
 872 }
 873
 874 static inline void stop_process_timers(struct signal_struct *sig)
 875 {
 876         struct thread_group_cputimer *cputimer = &sig->cputimer;
 877
 878         /* Turn off cputimer->running. This is done without locking. */
 879         WRITE_ONCE(cputimer->running, false);
 880         tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
 881 }
 882
 883 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 884                              u64 *expires, u64 cur_time, int signo)
 885 {
 886         if (!it->expires)
 887                 return;
 888
 889         if (cur_time >= it->expires) {
 890                 if (it->incr)
 891                         it->expires += it->incr;
 892                 else
 893                         it->expires = 0;
 894
 895                 trace_itimer_expire(signo == SIGPROF ?
 896                                     ITIMER_PROF : ITIMER_VIRTUAL,
 897                                     task_tgid(tsk), cur_time);
 898                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 899         }
 900
 901         if (it->expires && (!*expires || it->expires < *expires))
 902                 *expires = it->expires;
 903 }
 904
 905 /*
 906  * Check for any per-thread CPU timers that have fired and move them
 907  * off the tsk->*_timers list onto the firing list.  Per-thread timers
 908  * have already been taken off.
 909  */
 910 static void check_process_timers(struct task_struct *tsk,
 911                                  struct list_head *firing)
 912 {
 913         struct signal_struct *const sig = tsk->signal;
 914         u64 utime, ptime, virt_expires, prof_expires;
 915         u64 sum_sched_runtime, sched_expires;
 916         struct list_head *timers = sig->cpu_timers;
 917         struct task_cputime cputime;
 918         unsigned long soft;
 919
 920         if (dl_task(tsk))
 921                 check_dl_overrun(tsk);
 922
 923         /*
 924          * If cputimer is not running, then there are no active
 925          * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
 926          */
 927         if (!READ_ONCE(tsk->signal->cputimer.running))
 928                 return;
 929
 930         /*
 931          * Signify that a thread is checking for process timers.
 932          * Write access to this field is protected by the sighand lock.
 933          */
 934         sig->cputimer.checking_timer = true;
 935
 936         /*
 937          * Collect the current process totals.
 938          */
 939         thread_group_cputimer(tsk, &cputime);
 940         utime = cputime.utime;
 941         ptime = utime + cputime.stime;
 942         sum_sched_runtime = cputime.sum_exec_runtime;
 943
 944         prof_expires = check_timers_list(timers, firing, ptime);
 945         virt_expires = check_timers_list(++timers, firing, utime);
 946         sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
 947
 948         /*
 949          * Check for the special case process timers.
 950          */
 951         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
 952                          SIGPROF);
 953         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
 954                          SIGVTALRM);
 955         soft = task_rlimit(tsk, RLIMIT_CPU);
 956         if (soft != RLIM_INFINITY) {
 957                 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
 958                 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
 959                 u64 x;
 960                 if (psecs >= hard) {
 961                         /*
 962                          * At the hard limit, we just die.
 963                          * No need to calculate anything else now.
 964                          */
 965                         if (print_fatal_signals) {
 966                                 pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
 967                                         tsk->comm, task_pid_nr(tsk));
 968                         }
 969                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 970                         return;
 971                 }
 972                 if (psecs >= soft) {
 973                         /*
 974                          * At the soft limit, send a SIGXCPU every second.
 975                          */
 976                         if (print_fatal_signals) {
 977                                 pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
 978                                         tsk->comm, task_pid_nr(tsk));
 979                         }
 980                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 981                         if (soft < hard) {
 982                                 soft++;
 983                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
 984                         }
 985                 }
 986                 x = soft * NSEC_PER_SEC;
 987                 if (!prof_expires || x < prof_expires)
 988                         prof_expires = x;
 989         }
 990
 991         sig->cputime_expires.prof_exp = prof_expires;
 992         sig->cputime_expires.virt_exp = virt_expires;
 993         sig->cputime_expires.sched_exp = sched_expires;
 994         if (task_cputime_zero(&sig->cputime_expires))
 995                 stop_process_timers(sig);
 996
 997         sig->cputimer.checking_timer = false;
 998 }
 999
1000 /*
1001  * This is called from the signal code (via posixtimer_rearm)
1002  * when the last timer signal was delivered and we have to reload the timer.
1003  */
1004 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1005 {
1006         struct sighand_struct *sighand;
1007         unsigned long flags;
1008         struct task_struct *p = timer->it.cpu.task;
1009         u64 now;
1010
1011         WARN_ON_ONCE(p == NULL);
1012
1013         /*
1014          * Fetch the current sample and update the timer's expiry time.
1015          */
1016         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1017                 cpu_clock_sample(timer->it_clock, p, &now);
1018                 bump_cpu_timer(timer, now);
1019                 if (unlikely(p->exit_state))
1020                         return;
1021
1022                 /* Protect timer list r/w in arm_timer() */
1023                 sighand = lock_task_sighand(p, &flags);
1024                 if (!sighand)
1025                         return;
1026         } else {
1027                 /*
1028                  * Protect arm_timer() and timer sampling in case of call to
1029                  * thread_group_cputime().
1030                  */
1031                 sighand = lock_task_sighand(p, &flags);
1032                 if (unlikely(sighand == NULL)) {
1033                         /*
1034                          * The process has been reaped.
1035                          * We can't even collect a sample any more.
1036                          */
1037                         timer->it.cpu.expires = 0;
1038                         return;
1039                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1040                         /* If the process is dying, no need to rearm */
1041                         goto unlock;
1042                 }
1043                 cpu_timer_sample_group(timer->it_clock, p, &now);
1044                 bump_cpu_timer(timer, now);
1045                 /* Leave the sighand locked for the call below.  */
1046         }
1047
1048         /*
1049          * Now re-arm for the new expiry time.
1050          */
1051         arm_timer(timer);
1052 unlock:
1053         unlock_task_sighand(p, &flags);
1054 }
1055
1056 /**
1057  * task_cputime_expired - Compare two task_cputime entities.
1058  *
1059  * @sample:     The task_cputime structure to be checked for expiration.
1060  * @expires:    Expiration times, against which @sample will be checked.
1061  *
1062  * Checks @sample against @expires to see if any field of @sample has expired.
1063  * Returns true if any field of the former is greater than the corresponding
1064  * field of the latter if the latter field is set.  Otherwise returns false.
1065  */
1066 static inline int task_cputime_expired(const struct task_cputime *sample,
1067                                         const struct task_cputime *expires)
1068 {
1069         if (expires->utime && sample->utime >= expires->utime)
1070                 return 1;
1071         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1072                 return 1;
1073         if (expires->sum_exec_runtime != 0 &&
1074             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1075                 return 1;
1076         return 0;
1077 }
1078
1079 /**
1080  * fastpath_timer_check - POSIX CPU timers fast path.
1081  *
1082  * @tsk:        The task (thread) being checked.
1083  *
1084  * Check the task and thread group timers.  If both are zero (there are no
1085  * timers set) return false.  Otherwise snapshot the task and thread group
1086  * timers and compare them with the corresponding expiration times.  Return
1087  * true if a timer has expired, else return false.
1088  */
1089 static inline int fastpath_timer_check(struct task_struct *tsk)
1090 {
1091         struct signal_struct *sig;
1092
1093         if (!task_cputime_zero(&tsk->cputime_expires)) {
1094                 struct task_cputime task_sample;
1095
1096                 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1097                 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1098                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1099                         return 1;
1100         }
1101
1102         sig = tsk->signal;
1103         /*
1104          * Check if thread group timers expired when the cputimer is
1105          * running and no other thread in the group is already checking
1106          * for thread group cputimers. These fields are read without the
1107          * sighand lock. However, this is fine because this is meant to
1108          * be a fastpath heuristic to determine whether we should try to
1109          * acquire the sighand lock to check/handle timers.
1110          *
1111          * In the worst case scenario, if 'running' or 'checking_timer' gets
1112          * set but the current thread doesn't see the change yet, we'll wait
1113          * until the next thread in the group gets a scheduler interrupt to
1114          * handle the timer. This isn't an issue in practice because these
1115          * types of delays with signals actually getting sent are expected.
1116          */
1117         if (READ_ONCE(sig->cputimer.running) &&
1118             !READ_ONCE(sig->cputimer.checking_timer)) {
1119                 struct task_cputime group_sample;
1120
1121                 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1122
1123                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1124                         return 1;
1125         }
1126
1127         if (dl_task(tsk) && tsk->dl.dl_overrun)
1128                 return 1;
1129
1130         return 0;
1131 }
1132
1133 /*
1134  * This is called from the timer interrupt handler.  The irq handler has
1135  * already updated our counts.  We need to check if any timers fire now.
1136  * Interrupts are disabled.
1137  */
1138 void run_posix_cpu_timers(struct task_struct *tsk)
1139 {
1140         LIST_HEAD(firing);
1141         struct k_itimer *timer, *next;
1142         unsigned long flags;
1143
1144         lockdep_assert_irqs_disabled();
1145
1146         /*
1147          * The fast path checks that there are no expired thread or thread
1148          * group timers.  If that's so, just return.
1149          */
1150         if (!fastpath_timer_check(tsk))
1151                 return;
1152
1153         if (!lock_task_sighand(tsk, &flags))
1154                 return;
1155         /*
1156          * Here we take off tsk->signal->cpu_timers[N] and
1157          * tsk->cpu_timers[N] all the timers that are firing, and
1158          * put them on the firing list.
1159          */
1160         check_thread_timers(tsk, &firing);
1161
1162         check_process_timers(tsk, &firing);
1163
1164         /*
1165          * We must release these locks before taking any timer's lock.
1166          * There is a potential race with timer deletion here, as the
1167          * siglock now protects our private firing list.  We have set
1168          * the firing flag in each timer, so that a deletion attempt
1169          * that gets the timer lock before we do will give it up and
1170          * spin until we've taken care of that timer below.
1171          */
1172         unlock_task_sighand(tsk, &flags);
1173
1174         /*
1175          * Now that all the timers on our list have the firing flag,
1176          * no one will touch their list entries but us.  We'll take
1177          * each timer's lock before clearing its firing flag, so no
1178          * timer call will interfere.
1179          */
1180         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1181                 int cpu_firing;
1182
1183                 spin_lock(&timer->it_lock);
1184                 list_del_init(&timer->it.cpu.entry);
1185                 cpu_firing = timer->it.cpu.firing;
1186                 timer->it.cpu.firing = 0;
1187                 /*
1188                  * The firing flag is -1 if we collided with a reset
1189                  * of the timer, which already reported this
1190                  * almost-firing as an overrun.  So don't generate an event.
1191                  */
1192                 if (likely(cpu_firing >= 0))
1193                         cpu_timer_fire(timer);
1194                 spin_unlock(&timer->it_lock);
1195         }
1196 }
1197
1198 /*
1199  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1200  * The tsk->sighand->siglock must be held by the caller.
1201  */
1202 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1203                            u64 *newval, u64 *oldval)
1204 {
1205         u64 now;
1206         int ret;
1207
1208         WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1209         ret = cpu_timer_sample_group(clock_idx, tsk, &now);
1210
1211         if (oldval && ret != -EINVAL) {
1212                 /*
1213                  * We are setting itimer. The *oldval is absolute and we update
1214                  * it to be relative, *newval argument is relative and we update
1215                  * it to be absolute.
1216                  */
1217                 if (*oldval) {
1218                         if (*oldval <= now) {
1219                                 /* Just about to fire. */
1220                                 *oldval = TICK_NSEC;
1221                         } else {
1222                                 *oldval -= now;
1223                         }
1224                 }
1225
1226                 if (!*newval)
1227                         return;
1228                 *newval += now;
1229         }
1230
1231         /*
1232          * Update expiration cache if we are the earliest timer, or eventually
1233          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1234          */
1235         switch (clock_idx) {
1236         case CPUCLOCK_PROF:
1237                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1238                         tsk->signal->cputime_expires.prof_exp = *newval;
1239                 break;
1240         case CPUCLOCK_VIRT:
1241                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1242                         tsk->signal->cputime_expires.virt_exp = *newval;
1243                 break;
1244         }
1245
1246         tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1247 }
1248
1249 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1250                             const struct timespec64 *rqtp)
1251 {
1252         struct itimerspec64 it;
1253         struct k_itimer timer;
1254         u64 expires;
1255         int error;
1256
1257         /*
1258          * Set up a temporary timer and then wait for it to go off.
1259          */
1260         memset(&timer, 0, sizeof timer);
1261         spin_lock_init(&timer.it_lock);
1262         timer.it_clock = which_clock;
1263         timer.it_overrun = -1;
1264         error = posix_cpu_timer_create(&timer);
1265         timer.it_process = current;
1266         if (!error) {
1267                 static struct itimerspec64 zero_it;
1268                 struct restart_block *restart;
1269
1270                 memset(&it, 0, sizeof(it));
1271                 it.it_value = *rqtp;
1272
1273                 spin_lock_irq(&timer.it_lock);
1274                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1275                 if (error) {
1276                         spin_unlock_irq(&timer.it_lock);
1277                         return error;
1278                 }
1279
1280                 while (!signal_pending(current)) {
1281                         if (timer.it.cpu.expires == 0) {
1282                                 /*
1283                                  * Our timer fired and was reset, below
1284                                  * deletion can not fail.
1285                                  */
1286                                 posix_cpu_timer_del(&timer);
1287                                 spin_unlock_irq(&timer.it_lock);
1288                                 return 0;
1289                         }
1290
1291                         /*
1292                          * Block until cpu_timer_fire (or a signal) wakes us.
1293                          */
1294                         __set_current_state(TASK_INTERRUPTIBLE);
1295                         spin_unlock_irq(&timer.it_lock);
1296                         schedule();
1297                         spin_lock_irq(&timer.it_lock);
1298                 }
1299
1300                 /*
1301                  * We were interrupted by a signal.
1302                  */
1303                 expires = timer.it.cpu.expires;
1304                 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1305                 if (!error) {
1306                         /*
1307                          * Timer is now unarmed, deletion can not fail.
1308                          */
1309                         posix_cpu_timer_del(&timer);
1310                 }
1311                 spin_unlock_irq(&timer.it_lock);
1312
1313                 while (error == TIMER_RETRY) {
1314                         /*
1315                          * We need to handle case when timer was or is in the
1316                          * middle of firing. In other cases we already freed
1317                          * resources.
1318                          */
1319                         spin_lock_irq(&timer.it_lock);
1320                         error = posix_cpu_timer_del(&timer);
1321                         spin_unlock_irq(&timer.it_lock);
1322                 }
1323
1324                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1325                         /*
1326                          * It actually did fire already.
1327                          */
1328                         return 0;
1329                 }
1330
1331                 error = -ERESTART_RESTARTBLOCK;
1332                 /*
1333                  * Report back to the user the time still remaining.
1334                  */
1335                 restart = &current->restart_block;
1336                 restart->nanosleep.expires = expires;
1337                 if (restart->nanosleep.type != TT_NONE)
1338                         error = nanosleep_copyout(restart, &it.it_value);
1339         }
1340
1341         return error;
1342 }
1343
1344 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1345
1346 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1347                             const struct timespec64 *rqtp)
1348 {
1349         struct restart_block *restart_block = &current->restart_block;
1350         int error;
1351
1352         /*
1353          * Diagnose required errors first.
1354          */
1355         if (CPUCLOCK_PERTHREAD(which_clock) &&
1356             (CPUCLOCK_PID(which_clock) == 0 ||
1357              CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1358                 return -EINVAL;
1359
1360         error = do_cpu_nanosleep(which_clock, flags, rqtp);
1361
1362         if (error == -ERESTART_RESTARTBLOCK) {
1363
1364                 if (flags & TIMER_ABSTIME)
1365                         return -ERESTARTNOHAND;
1366
1367                 restart_block->fn = posix_cpu_nsleep_restart;
1368                 restart_block->nanosleep.clockid = which_clock;
1369         }
1370         return error;
1371 }
1372
1373 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1374 {
1375         clockid_t which_clock = restart_block->nanosleep.clockid;
1376         struct timespec64 t;
1377
1378         t = ns_to_timespec64(restart_block->nanosleep.expires);
1379
1380         return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1381 }
1382
1383 #define PROCESS_CLOCK   make_process_cpuclock(0, CPUCLOCK_SCHED)
1384 #define THREAD_CLOCK    make_thread_cpuclock(0, CPUCLOCK_SCHED)
1385
1386 static int process_cpu_clock_getres(const clockid_t which_clock,
1387                                     struct timespec64 *tp)
1388 {
1389         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1390 }
1391 static int process_cpu_clock_get(const clockid_t which_clock,
1392                                  struct timespec64 *tp)
1393 {
1394         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1395 }
1396 static int process_cpu_timer_create(struct k_itimer *timer)
1397 {
1398         timer->it_clock = PROCESS_CLOCK;
1399         return posix_cpu_timer_create(timer);
1400 }
1401 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1402                               const struct timespec64 *rqtp)
1403 {
1404         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1405 }
1406 static int thread_cpu_clock_getres(const clockid_t which_clock,
1407                                    struct timespec64 *tp)
1408 {
1409         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1410 }
1411 static int thread_cpu_clock_get(const clockid_t which_clock,
1412                                 struct timespec64 *tp)
1413 {
1414         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1415 }
1416 static int thread_cpu_timer_create(struct k_itimer *timer)
1417 {
1418         timer->it_clock = THREAD_CLOCK;
1419         return posix_cpu_timer_create(timer);
1420 }
1421
1422 const struct k_clock clock_posix_cpu = {
1423         .clock_getres   = posix_cpu_clock_getres,
1424         .clock_set      = posix_cpu_clock_set,
1425         .clock_get      = posix_cpu_clock_get,
1426         .timer_create   = posix_cpu_timer_create,
1427         .nsleep         = posix_cpu_nsleep,
1428         .timer_set      = posix_cpu_timer_set,
1429         .timer_del      = posix_cpu_timer_del,
1430         .timer_get      = posix_cpu_timer_get,
1431         .timer_rearm    = posix_cpu_timer_rearm,
1432 };
1433
1434 const struct k_clock clock_process = {
1435         .clock_getres   = process_cpu_clock_getres,
1436         .clock_get      = process_cpu_clock_get,
1437         .timer_create   = process_cpu_timer_create,
1438         .nsleep         = process_cpu_nsleep,
1439 };
1440
1441 const struct k_clock clock_thread = {
1442         .clock_getres   = thread_cpu_clock_getres,
1443         .clock_get      = thread_cpu_clock_get,
1444         .timer_create   = thread_cpu_timer_create,
1445 };