kernel/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/posix-timers.h>
   7 #include <linux/errno.h>
   8 #include <linux/math64.h>
   9 #include <asm/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11 #include <trace/events/timer.h>
  12 #include <linux/random.h>
  13 #include <linux/tick.h>
  14 #include <linux/workqueue.h>
  15
  16 /*
  17  * Called after updating RLIMIT_CPU to run cpu timer and update
  18  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  19  * siglock protection since other code may update expiration cache as
  20  * well.
  21  */
  22 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  23 {
  24         cputime_t cputime = secs_to_cputime(rlim_new);
  25
  26         spin_lock_irq(&task->sighand->siglock);
  27         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  28         spin_unlock_irq(&task->sighand->siglock);
  29 }
  30
  31 static int check_clock(const clockid_t which_clock)
  32 {
  33         int error = 0;
  34         struct task_struct *p;
  35         const pid_t pid = CPUCLOCK_PID(which_clock);
  36
  37         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  38                 return -EINVAL;
  39
  40         if (pid == 0)
  41                 return 0;
  42
  43         rcu_read_lock();
  44         p = find_task_by_vpid(pid);
  45         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  46                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  47                 error = -EINVAL;
  48         }
  49         rcu_read_unlock();
  50
  51         return error;
  52 }
  53
  54 static inline unsigned long long
  55 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  56 {
  57         unsigned long long ret;
  58
  59         ret = 0;                /* high half always zero when .cpu used */
  60         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  61                 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  62         } else {
  63                 ret = cputime_to_expires(timespec_to_cputime(tp));
  64         }
  65         return ret;
  66 }
  67
  68 static void sample_to_timespec(const clockid_t which_clock,
  69                                unsigned long long expires,
  70                                struct timespec *tp)
  71 {
  72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  73                 *tp = ns_to_timespec(expires);
  74         else
  75                 cputime_to_timespec((__force cputime_t)expires, tp);
  76 }
  77
  78 /*
  79  * Update expiry time from increment, and increase overrun count,
  80  * given the current clock sample.
  81  */
  82 static void bump_cpu_timer(struct k_itimer *timer,
  83                            unsigned long long now)
  84 {
  85         int i;
  86         unsigned long long delta, incr;
  87
  88         if (timer->it.cpu.incr == 0)
  89                 return;
  90
  91         if (now < timer->it.cpu.expires)
  92                 return;
  93
  94         incr = timer->it.cpu.incr;
  95         delta = now + incr - timer->it.cpu.expires;
  96
  97         /* Don't use (incr*2 < delta), incr*2 might overflow. */
  98         for (i = 0; incr < delta - incr; i++)
  99                 incr = incr << 1;
 100
 101         for (; i >= 0; incr >>= 1, i--) {
 102                 if (delta < incr)
 103                         continue;
 104
 105                 timer->it.cpu.expires += incr;
 106                 timer->it_overrun += 1 << i;
 107                 delta -= incr;
 108         }
 109 }
 110
 111 /**
 112  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 113  *
 114  * @cputime:    The struct to compare.
 115  *
 116  * Checks @cputime to see if all fields are zero.  Returns true if all fields
 117  * are zero, false if any field is nonzero.
 118  */
 119 static inline int task_cputime_zero(const struct task_cputime *cputime)
 120 {
 121         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 static inline unsigned long long prof_ticks(struct task_struct *p)
 127 {
 128         cputime_t utime, stime;
 129
 130         task_cputime(p, &utime, &stime);
 131
 132         return cputime_to_expires(utime + stime);
 133 }
 134 static inline unsigned long long virt_ticks(struct task_struct *p)
 135 {
 136         cputime_t utime;
 137
 138         task_cputime(p, &utime, NULL);
 139
 140         return cputime_to_expires(utime);
 141 }
 142
 143 static int
 144 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 145 {
 146         int error = check_clock(which_clock);
 147         if (!error) {
 148                 tp->tv_sec = 0;
 149                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 150                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 151                         /*
 152                          * If sched_clock is using a cycle counter, we
 153                          * don't have any idea of its true resolution
 154                          * exported, but it is much more than 1s/HZ.
 155                          */
 156                         tp->tv_nsec = 1;
 157                 }
 158         }
 159         return error;
 160 }
 161
 162 static int
 163 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 164 {
 165         /*
 166          * You can never reset a CPU clock, but we check for other errors
 167          * in the call before failing with EPERM.
 168          */
 169         int error = check_clock(which_clock);
 170         if (error == 0) {
 171                 error = -EPERM;
 172         }
 173         return error;
 174 }
 175
 176
 177 /*
 178  * Sample a per-thread clock for the given task.
 179  */
 180 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 181                             unsigned long long *sample)
 182 {
 183         switch (CPUCLOCK_WHICH(which_clock)) {
 184         default:
 185                 return -EINVAL;
 186         case CPUCLOCK_PROF:
 187                 *sample = prof_ticks(p);
 188                 break;
 189         case CPUCLOCK_VIRT:
 190                 *sample = virt_ticks(p);
 191                 break;
 192         case CPUCLOCK_SCHED:
 193                 *sample = task_sched_runtime(p);
 194                 break;
 195         }
 196         return 0;
 197 }
 198
 199 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 200 {
 201         if (b->utime > a->utime)
 202                 a->utime = b->utime;
 203
 204         if (b->stime > a->stime)
 205                 a->stime = b->stime;
 206
 207         if (b->sum_exec_runtime > a->sum_exec_runtime)
 208                 a->sum_exec_runtime = b->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         unsigned long flags;
 216
 217         if (!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
 222                  * it.
 223                  */
 224                 thread_group_cputime(tsk, &sum);
 225                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 226                 cputimer->running = 1;
 227                 update_gt_cputime(&cputimer->cputime, &sum);
 228         } else
 229                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 230         *times = cputimer->cputime;
 231         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 232 }
 233
 234 /*
 235  * Sample a process (thread group) clock for the given group_leader task.
 236  * Must be called with tasklist_lock held for reading.
 237  */
 238 static int cpu_clock_sample_group(const clockid_t which_clock,
 239                                   struct task_struct *p,
 240                                   unsigned long long *sample)
 241 {
 242         struct task_cputime cputime;
 243
 244         switch (CPUCLOCK_WHICH(which_clock)) {
 245         default:
 246                 return -EINVAL;
 247         case CPUCLOCK_PROF:
 248                 thread_group_cputime(p, &cputime);
 249                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 250                 break;
 251         case CPUCLOCK_VIRT:
 252                 thread_group_cputime(p, &cputime);
 253                 *sample = cputime_to_expires(cputime.utime);
 254                 break;
 255         case CPUCLOCK_SCHED:
 256                 thread_group_cputime(p, &cputime);
 257                 *sample = cputime.sum_exec_runtime;
 258                 break;
 259         }
 260         return 0;
 261 }
 262
 263
 264 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 265 {
 266         const pid_t pid = CPUCLOCK_PID(which_clock);
 267         int error = -EINVAL;
 268         unsigned long long rtn;
 269
 270         if (pid == 0) {
 271                 /*
 272                  * Special case constant value for our own clocks.
 273                  * We don't have to do any lookup to find ourselves.
 274                  */
 275                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 276                         /*
 277                          * Sampling just ourselves we can do with no locking.
 278                          */
 279                         error = cpu_clock_sample(which_clock,
 280                                                  current, &rtn);
 281                 } else {
 282                         read_lock(&tasklist_lock);
 283                         error = cpu_clock_sample_group(which_clock,
 284                                                        current, &rtn);
 285                         read_unlock(&tasklist_lock);
 286                 }
 287         } else {
 288                 /*
 289                  * Find the given PID, and validate that the caller
 290                  * should be able to see it.
 291                  */
 292                 struct task_struct *p;
 293                 rcu_read_lock();
 294                 p = find_task_by_vpid(pid);
 295                 if (p) {
 296                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 297                                 if (same_thread_group(p, current)) {
 298                                         error = cpu_clock_sample(which_clock,
 299                                                                  p, &rtn);
 300                                 }
 301                         } else {
 302                                 read_lock(&tasklist_lock);
 303                                 if (thread_group_leader(p) && p->sighand) {
 304                                         error =
 305                                             cpu_clock_sample_group(which_clock,
 306                                                                    p, &rtn);
 307                                 }
 308                                 read_unlock(&tasklist_lock);
 309                         }
 310                 }
 311                 rcu_read_unlock();
 312         }
 313
 314         if (error)
 315                 return error;
 316         sample_to_timespec(which_clock, rtn, tp);
 317         return 0;
 318 }
 319
 320
 321 /*
 322  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 323  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 324  * new timer already all-zeros initialized.
 325  */
 326 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 327 {
 328         int ret = 0;
 329         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 330         struct task_struct *p;
 331
 332         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 333                 return -EINVAL;
 334
 335         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 336
 337         rcu_read_lock();
 338         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 339                 if (pid == 0) {
 340                         p = current;
 341                 } else {
 342                         p = find_task_by_vpid(pid);
 343                         if (p && !same_thread_group(p, current))
 344                                 p = NULL;
 345                 }
 346         } else {
 347                 if (pid == 0) {
 348                         p = current->group_leader;
 349                 } else {
 350                         p = find_task_by_vpid(pid);
 351                         if (p && !has_group_leader_pid(p))
 352                                 p = NULL;
 353                 }
 354         }
 355         new_timer->it.cpu.task = p;
 356         if (p) {
 357                 get_task_struct(p);
 358         } else {
 359                 ret = -EINVAL;
 360         }
 361         rcu_read_unlock();
 362
 363         return ret;
 364 }
 365
 366 /*
 367  * Clean up a CPU-clock timer that is about to be destroyed.
 368  * This is called from timer deletion with the timer already locked.
 369  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 370  * and try again.  (This happens when the timer is in the middle of firing.)
 371  */
 372 static int posix_cpu_timer_del(struct k_itimer *timer)
 373 {
 374         struct task_struct *p = timer->it.cpu.task;
 375         int ret = 0;
 376
 377         if (likely(p != NULL)) {
 378                 read_lock(&tasklist_lock);
 379                 if (unlikely(p->sighand == NULL)) {
 380                         /*
 381                          * We raced with the reaping of the task.
 382                          * The deletion should have cleared us off the list.
 383                          */
 384                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 385                 } else {
 386                         spin_lock(&p->sighand->siglock);
 387                         if (timer->it.cpu.firing)
 388                                 ret = TIMER_RETRY;
 389                         else
 390                                 list_del(&timer->it.cpu.entry);
 391                         spin_unlock(&p->sighand->siglock);
 392                 }
 393                 read_unlock(&tasklist_lock);
 394
 395                 if (!ret)
 396                         put_task_struct(p);
 397         }
 398
 399         return ret;
 400 }
 401
 402 /*
 403  * Clean out CPU timers still ticking when a thread exited.  The task
 404  * pointer is cleared, and the expiry time is replaced with the residual
 405  * time for later timer_gettime calls to return.
 406  * This must be called with the siglock held.
 407  */
 408 static void cleanup_timers(struct list_head *head,
 409                            cputime_t utime, cputime_t stime,
 410                            unsigned long long sum_exec_runtime)
 411 {
 412         struct cpu_timer_list *timer, *next;
 413         cputime_t ptime = utime + stime;
 414
 415         list_for_each_entry_safe(timer, next, head, entry) {
 416                 list_del_init(&timer->entry);
 417                 if (timer->expires < cputime_to_expires(ptime)) {
 418                         timer->expires = 0;
 419                 } else {
 420                         timer->expires -= cputime_to_expires(ptime);
 421                 }
 422         }
 423
 424         ++head;
 425         list_for_each_entry_safe(timer, next, head, entry) {
 426                 list_del_init(&timer->entry);
 427                 if (timer->expires < cputime_to_expires(utime)) {
 428                         timer->expires = 0;
 429                 } else {
 430                         timer->expires -= cputime_to_expires(utime);
 431                 }
 432         }
 433
 434         ++head;
 435         list_for_each_entry_safe(timer, next, head, entry) {
 436                 list_del_init(&timer->entry);
 437                 if (timer->expires < sum_exec_runtime) {
 438                         timer->expires = 0;
 439                 } else {
 440                         timer->expires -= sum_exec_runtime;
 441                 }
 442         }
 443 }
 444
 445 /*
 446  * These are both called with the siglock held, when the current thread
 447  * is being reaped.  When the final (leader) thread in the group is reaped,
 448  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 449  */
 450 void posix_cpu_timers_exit(struct task_struct *tsk)
 451 {
 452         cputime_t utime, stime;
 453
 454         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 455                                                 sizeof(unsigned long long));
 456         task_cputime(tsk, &utime, &stime);
 457         cleanup_timers(tsk->cpu_timers,
 458                        utime, stime, tsk->se.sum_exec_runtime);
 459
 460 }
 461 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 462 {
 463         struct signal_struct *const sig = tsk->signal;
 464         cputime_t utime, stime;
 465
 466         task_cputime(tsk, &utime, &stime);
 467         cleanup_timers(tsk->signal->cpu_timers,
 468                        utime + sig->utime, stime + sig->stime,
 469                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 470 }
 471
 472 static void clear_dead_task(struct k_itimer *timer, unsigned long long now)
 473 {
 474         /*
 475          * That's all for this thread or process.
 476          * We leave our residual in expires to be reported.
 477          */
 478         put_task_struct(timer->it.cpu.task);
 479         timer->it.cpu.task = NULL;
 480         timer->it.cpu.expires -= now;
 481 }
 482
 483 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 484 {
 485         return expires == 0 || expires > new_exp;
 486 }
 487
 488 /*
 489  * Insert the timer on the appropriate list before any timers that
 490  * expire later.  This must be called with the tasklist_lock held
 491  * for reading, interrupts disabled and p->sighand->siglock taken.
 492  */
 493 static void arm_timer(struct k_itimer *timer)
 494 {
 495         struct task_struct *p = timer->it.cpu.task;
 496         struct list_head *head, *listpos;
 497         struct task_cputime *cputime_expires;
 498         struct cpu_timer_list *const nt = &timer->it.cpu;
 499         struct cpu_timer_list *next;
 500
 501         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 502                 head = p->cpu_timers;
 503                 cputime_expires = &p->cputime_expires;
 504         } else {
 505                 head = p->signal->cpu_timers;
 506                 cputime_expires = &p->signal->cputime_expires;
 507         }
 508         head += CPUCLOCK_WHICH(timer->it_clock);
 509
 510         listpos = head;
 511         list_for_each_entry(next, head, entry) {
 512                 if (nt->expires < next->expires)
 513                         break;
 514                 listpos = &next->entry;
 515         }
 516         list_add(&nt->entry, listpos);
 517
 518         if (listpos == head) {
 519                 unsigned long long exp = nt->expires;
 520
 521                 /*
 522                  * We are the new earliest-expiring POSIX 1.b timer, hence
 523                  * need to update expiration cache. Take into account that
 524                  * for process timers we share expiration cache with itimers
 525                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 526                  */
 527
 528                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 529                 case CPUCLOCK_PROF:
 530                         if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
 531                                 cputime_expires->prof_exp = expires_to_cputime(exp);
 532                         break;
 533                 case CPUCLOCK_VIRT:
 534                         if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
 535                                 cputime_expires->virt_exp = expires_to_cputime(exp);
 536                         break;
 537                 case CPUCLOCK_SCHED:
 538                         if (cputime_expires->sched_exp == 0 ||
 539                             cputime_expires->sched_exp > exp)
 540                                 cputime_expires->sched_exp = exp;
 541                         break;
 542                 }
 543         }
 544 }
 545
 546 /*
 547  * The timer is locked, fire it and arrange for its reload.
 548  */
 549 static void cpu_timer_fire(struct k_itimer *timer)
 550 {
 551         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 552                 /*
 553                  * User don't want any signal.
 554                  */
 555                 timer->it.cpu.expires = 0;
 556         } else if (unlikely(timer->sigq == NULL)) {
 557                 /*
 558                  * This a special case for clock_nanosleep,
 559                  * not a normal timer from sys_timer_create.
 560                  */
 561                 wake_up_process(timer->it_process);
 562                 timer->it.cpu.expires = 0;
 563         } else if (timer->it.cpu.incr == 0) {
 564                 /*
 565                  * One-shot timer.  Clear it as soon as it's fired.
 566                  */
 567                 posix_timer_event(timer, 0);
 568                 timer->it.cpu.expires = 0;
 569         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 570                 /*
 571                  * The signal did not get queued because the signal
 572                  * was ignored, so we won't get any callback to
 573                  * reload the timer.  But we need to keep it
 574                  * ticking in case the signal is deliverable next time.
 575                  */
 576                 posix_cpu_timer_schedule(timer);
 577         }
 578 }
 579
 580 /*
 581  * Sample a process (thread group) timer for the given group_leader task.
 582  * Must be called with tasklist_lock held for reading.
 583  */
 584 static int cpu_timer_sample_group(const clockid_t which_clock,
 585                                   struct task_struct *p,
 586                                   unsigned long long *sample)
 587 {
 588         struct task_cputime cputime;
 589
 590         thread_group_cputimer(p, &cputime);
 591         switch (CPUCLOCK_WHICH(which_clock)) {
 592         default:
 593                 return -EINVAL;
 594         case CPUCLOCK_PROF:
 595                 *sample = cputime_to_expires(cputime.utime + cputime.stime);
 596                 break;
 597         case CPUCLOCK_VIRT:
 598                 *sample = cputime_to_expires(cputime.utime);
 599                 break;
 600         case CPUCLOCK_SCHED:
 601                 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
 602                 break;
 603         }
 604         return 0;
 605 }
 606
 607 #ifdef CONFIG_NO_HZ_FULL
 608 static void nohz_kick_work_fn(struct work_struct *work)
 609 {
 610         tick_nohz_full_kick_all();
 611 }
 612
 613 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
 614
 615 /*
 616  * We need the IPIs to be sent from sane process context.
 617  * The posix cpu timers are always set with irqs disabled.
 618  */
 619 static void posix_cpu_timer_kick_nohz(void)
 620 {
 621         schedule_work(&nohz_kick_work);
 622 }
 623
 624 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
 625 {
 626         if (!task_cputime_zero(&tsk->cputime_expires))
 627                 return false;
 628
 629         if (tsk->signal->cputimer.running)
 630                 return false;
 631
 632         return true;
 633 }
 634 #else
 635 static inline void posix_cpu_timer_kick_nohz(void) { }
 636 #endif
 637
 638 /*
 639  * Guts of sys_timer_settime for CPU timers.
 640  * This is called with the timer locked and interrupts disabled.
 641  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 642  * and try again.  (This happens when the timer is in the middle of firing.)
 643  */
 644 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 645                                struct itimerspec *new, struct itimerspec *old)
 646 {
 647         struct task_struct *p = timer->it.cpu.task;
 648         unsigned long long old_expires, new_expires, old_incr, val;
 649         int ret;
 650
 651         if (unlikely(p == NULL)) {
 652                 /*
 653                  * Timer refers to a dead task's clock.
 654                  */
 655                 return -ESRCH;
 656         }
 657
 658         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 659
 660         read_lock(&tasklist_lock);
 661         /*
 662          * We need the tasklist_lock to protect against reaping that
 663          * clears p->sighand.  If p has just been reaped, we can no
 664          * longer get any information about it at all.
 665          */
 666         if (unlikely(p->sighand == NULL)) {
 667                 read_unlock(&tasklist_lock);
 668                 put_task_struct(p);
 669                 timer->it.cpu.task = NULL;
 670                 return -ESRCH;
 671         }
 672
 673         /*
 674          * Disarm any old timer after extracting its expiry time.
 675          */
 676         BUG_ON(!irqs_disabled());
 677
 678         ret = 0;
 679         old_incr = timer->it.cpu.incr;
 680         spin_lock(&p->sighand->siglock);
 681         old_expires = timer->it.cpu.expires;
 682         if (unlikely(timer->it.cpu.firing)) {
 683                 timer->it.cpu.firing = -1;
 684                 ret = TIMER_RETRY;
 685         } else
 686                 list_del_init(&timer->it.cpu.entry);
 687
 688         /*
 689          * We need to sample the current value to convert the new
 690          * value from to relative and absolute, and to convert the
 691          * old value from absolute to relative.  To set a process
 692          * timer, we need a sample to balance the thread expiry
 693          * times (in arm_timer).  With an absolute time, we must
 694          * check if it's already passed.  In short, we need a sample.
 695          */
 696         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 697                 cpu_clock_sample(timer->it_clock, p, &val);
 698         } else {
 699                 cpu_timer_sample_group(timer->it_clock, p, &val);
 700         }
 701
 702         if (old) {
 703                 if (old_expires == 0) {
 704                         old->it_value.tv_sec = 0;
 705                         old->it_value.tv_nsec = 0;
 706                 } else {
 707                         /*
 708                          * Update the timer in case it has
 709                          * overrun already.  If it has,
 710                          * we'll report it as having overrun
 711                          * and with the next reloaded timer
 712                          * already ticking, though we are
 713                          * swallowing that pending
 714                          * notification here to install the
 715                          * new setting.
 716                          */
 717                         bump_cpu_timer(timer, val);
 718                         if (val < timer->it.cpu.expires) {
 719                                 old_expires = timer->it.cpu.expires - val;
 720                                 sample_to_timespec(timer->it_clock,
 721                                                    old_expires,
 722                                                    &old->it_value);
 723                         } else {
 724                                 old->it_value.tv_nsec = 1;
 725                                 old->it_value.tv_sec = 0;
 726                         }
 727                 }
 728         }
 729
 730         if (unlikely(ret)) {
 731                 /*
 732                  * We are colliding with the timer actually firing.
 733                  * Punt after filling in the timer's old value, and
 734                  * disable this firing since we are already reporting
 735                  * it as an overrun (thanks to bump_cpu_timer above).
 736                  */
 737                 spin_unlock(&p->sighand->siglock);
 738                 read_unlock(&tasklist_lock);
 739                 goto out;
 740         }
 741
 742         if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
 743                 new_expires += val;
 744         }
 745
 746         /*
 747          * Install the new expiry time (or zero).
 748          * For a timer with no notification action, we don't actually
 749          * arm the timer (we'll just fake it for timer_gettime).
 750          */
 751         timer->it.cpu.expires = new_expires;
 752         if (new_expires != 0 && val < new_expires) {
 753                 arm_timer(timer);
 754         }
 755
 756         spin_unlock(&p->sighand->siglock);
 757         read_unlock(&tasklist_lock);
 758
 759         /*
 760          * Install the new reload setting, and
 761          * set up the signal and overrun bookkeeping.
 762          */
 763         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 764                                                 &new->it_interval);
 765
 766         /*
 767          * This acts as a modification timestamp for the timer,
 768          * so any automatic reload attempt will punt on seeing
 769          * that we have reset the timer manually.
 770          */
 771         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 772                 ~REQUEUE_PENDING;
 773         timer->it_overrun_last = 0;
 774         timer->it_overrun = -1;
 775
 776         if (new_expires != 0 && !(val < new_expires)) {
 777                 /*
 778                  * The designated time already passed, so we notify
 779                  * immediately, even if the thread never runs to
 780                  * accumulate more time on this clock.
 781                  */
 782                 cpu_timer_fire(timer);
 783         }
 784
 785         ret = 0;
 786  out:
 787         if (old) {
 788                 sample_to_timespec(timer->it_clock,
 789                                    old_incr, &old->it_interval);
 790         }
 791         if (!ret)
 792                 posix_cpu_timer_kick_nohz();
 793         return ret;
 794 }
 795
 796 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 797 {
 798         unsigned long long now;
 799         struct task_struct *p = timer->it.cpu.task;
 800         int clear_dead;
 801
 802         /*
 803          * Easy part: convert the reload time.
 804          */
 805         sample_to_timespec(timer->it_clock,
 806                            timer->it.cpu.incr, &itp->it_interval);
 807
 808         if (timer->it.cpu.expires == 0) {       /* Timer not armed at all.  */
 809                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 810                 return;
 811         }
 812
 813         if (unlikely(p == NULL)) {
 814                 /*
 815                  * This task already died and the timer will never fire.
 816                  * In this case, expires is actually the dead value.
 817                  */
 818         dead:
 819                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 820                                    &itp->it_value);
 821                 return;
 822         }
 823
 824         /*
 825          * Sample the clock to take the difference with the expiry time.
 826          */
 827         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 828                 cpu_clock_sample(timer->it_clock, p, &now);
 829                 clear_dead = p->exit_state;
 830         } else {
 831                 read_lock(&tasklist_lock);
 832                 if (unlikely(p->sighand == NULL)) {
 833                         /*
 834                          * The process has been reaped.
 835                          * We can't even collect a sample any more.
 836                          * Call the timer disarmed, nothing else to do.
 837                          */
 838                         put_task_struct(p);
 839                         timer->it.cpu.task = NULL;
 840                         timer->it.cpu.expires = 0;
 841                         read_unlock(&tasklist_lock);
 842                         goto dead;
 843                 } else {
 844                         cpu_timer_sample_group(timer->it_clock, p, &now);
 845                         clear_dead = (unlikely(p->exit_state) &&
 846                                       thread_group_empty(p));
 847                 }
 848                 read_unlock(&tasklist_lock);
 849         }
 850
 851         if (unlikely(clear_dead)) {
 852                 /*
 853                  * We've noticed that the thread is dead, but
 854                  * not yet reaped.  Take this opportunity to
 855                  * drop our task ref.
 856                  */
 857                 clear_dead_task(timer, now);
 858                 goto dead;
 859         }
 860
 861         if (now < timer->it.cpu.expires) {
 862                 sample_to_timespec(timer->it_clock,
 863                                    timer->it.cpu.expires - now,
 864                                    &itp->it_value);
 865         } else {
 866                 /*
 867                  * The timer should have expired already, but the firing
 868                  * hasn't taken place yet.  Say it's just about to expire.
 869                  */
 870                 itp->it_value.tv_nsec = 1;
 871                 itp->it_value.tv_sec = 0;
 872         }
 873 }
 874
 875 /*
 876  * Check for any per-thread CPU timers that have fired and move them off
 877  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 878  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 879  */
 880 static void check_thread_timers(struct task_struct *tsk,
 881                                 struct list_head *firing)
 882 {
 883         int maxfire;
 884         struct list_head *timers = tsk->cpu_timers;
 885         struct signal_struct *const sig = tsk->signal;
 886         unsigned long soft;
 887
 888         maxfire = 20;
 889         tsk->cputime_expires.prof_exp = 0;
 890         while (!list_empty(timers)) {
 891                 struct cpu_timer_list *t = list_first_entry(timers,
 892                                                       struct cpu_timer_list,
 893                                                       entry);
 894                 if (!--maxfire || prof_ticks(tsk) < t->expires) {
 895                         tsk->cputime_expires.prof_exp = expires_to_cputime(t->expires);
 896                         break;
 897                 }
 898                 t->firing = 1;
 899                 list_move_tail(&t->entry, firing);
 900         }
 901
 902         ++timers;
 903         maxfire = 20;
 904         tsk->cputime_expires.virt_exp = 0;
 905         while (!list_empty(timers)) {
 906                 struct cpu_timer_list *t = list_first_entry(timers,
 907                                                       struct cpu_timer_list,
 908                                                       entry);
 909                 if (!--maxfire || virt_ticks(tsk) < t->expires) {
 910                         tsk->cputime_expires.virt_exp = expires_to_cputime(t->expires);
 911                         break;
 912                 }
 913                 t->firing = 1;
 914                 list_move_tail(&t->entry, firing);
 915         }
 916
 917         ++timers;
 918         maxfire = 20;
 919         tsk->cputime_expires.sched_exp = 0;
 920         while (!list_empty(timers)) {
 921                 struct cpu_timer_list *t = list_first_entry(timers,
 922                                                       struct cpu_timer_list,
 923                                                       entry);
 924                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires) {
 925                         tsk->cputime_expires.sched_exp = t->expires;
 926                         break;
 927                 }
 928                 t->firing = 1;
 929                 list_move_tail(&t->entry, firing);
 930         }
 931
 932         /*
 933          * Check for the special case thread timers.
 934          */
 935         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 936         if (soft != RLIM_INFINITY) {
 937                 unsigned long hard =
 938                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 939
 940                 if (hard != RLIM_INFINITY &&
 941                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 942                         /*
 943                          * At the hard limit, we just die.
 944                          * No need to calculate anything else now.
 945                          */
 946                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 947                         return;
 948                 }
 949                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 950                         /*
 951                          * At the soft limit, send a SIGXCPU every second.
 952                          */
 953                         if (soft < hard) {
 954                                 soft += USEC_PER_SEC;
 955                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 956                         }
 957                         printk(KERN_INFO
 958                                 "RT Watchdog Timeout: %s[%d]\n",
 959                                 tsk->comm, task_pid_nr(tsk));
 960                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 961                 }
 962         }
 963 }
 964
 965 static void stop_process_timers(struct signal_struct *sig)
 966 {
 967         struct thread_group_cputimer *cputimer = &sig->cputimer;
 968         unsigned long flags;
 969
 970         raw_spin_lock_irqsave(&cputimer->lock, flags);
 971         cputimer->running = 0;
 972         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 973 }
 974
 975 static u32 onecputick;
 976
 977 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 978                              unsigned long long *expires,
 979                              unsigned long long cur_time, int signo)
 980 {
 981         if (!it->expires)
 982                 return;
 983
 984         if (cur_time >= it->expires) {
 985                 if (it->incr) {
 986                         it->expires += it->incr;
 987                         it->error += it->incr_error;
 988                         if (it->error >= onecputick) {
 989                                 it->expires -= cputime_one_jiffy;
 990                                 it->error -= onecputick;
 991                         }
 992                 } else {
 993                         it->expires = 0;
 994                 }
 995
 996                 trace_itimer_expire(signo == SIGPROF ?
 997                                     ITIMER_PROF : ITIMER_VIRTUAL,
 998                                     tsk->signal->leader_pid, cur_time);
 999                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1000         }
1001
1002         if (it->expires && (!*expires || it->expires < *expires)) {
1003                 *expires = it->expires;
1004         }
1005 }
1006
1007 /*
1008  * Check for any per-thread CPU timers that have fired and move them
1009  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1010  * have already been taken off.
1011  */
1012 static void check_process_timers(struct task_struct *tsk,
1013                                  struct list_head *firing)
1014 {
1015         int maxfire;
1016         struct signal_struct *const sig = tsk->signal;
1017         unsigned long long utime, ptime, virt_expires, prof_expires;
1018         unsigned long long sum_sched_runtime, sched_expires;
1019         struct list_head *timers = sig->cpu_timers;
1020         struct task_cputime cputime;
1021         unsigned long soft;
1022
1023         /*
1024          * Collect the current process totals.
1025          */
1026         thread_group_cputimer(tsk, &cputime);
1027         utime = cputime_to_expires(cputime.utime);
1028         ptime = utime + cputime_to_expires(cputime.stime);
1029         sum_sched_runtime = cputime.sum_exec_runtime;
1030         maxfire = 20;
1031         prof_expires = 0;
1032         while (!list_empty(timers)) {
1033                 struct cpu_timer_list *tl = list_first_entry(timers,
1034                                                       struct cpu_timer_list,
1035                                                       entry);
1036                 if (!--maxfire || ptime < tl->expires) {
1037                         prof_expires = tl->expires;
1038                         break;
1039                 }
1040                 tl->firing = 1;
1041                 list_move_tail(&tl->entry, firing);
1042         }
1043
1044         ++timers;
1045         maxfire = 20;
1046         virt_expires = 0;
1047         while (!list_empty(timers)) {
1048                 struct cpu_timer_list *tl = list_first_entry(timers,
1049                                                       struct cpu_timer_list,
1050                                                       entry);
1051                 if (!--maxfire || utime < tl->expires) {
1052                         virt_expires = tl->expires;
1053                         break;
1054                 }
1055                 tl->firing = 1;
1056                 list_move_tail(&tl->entry, firing);
1057         }
1058
1059         ++timers;
1060         maxfire = 20;
1061         sched_expires = 0;
1062         while (!list_empty(timers)) {
1063                 struct cpu_timer_list *tl = list_first_entry(timers,
1064                                                       struct cpu_timer_list,
1065                                                       entry);
1066                 if (!--maxfire || sum_sched_runtime < tl->expires) {
1067                         sched_expires = tl->expires;
1068                         break;
1069                 }
1070                 tl->firing = 1;
1071                 list_move_tail(&tl->entry, firing);
1072         }
1073
1074         /*
1075          * Check for the special case process timers.
1076          */
1077         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1078                          SIGPROF);
1079         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1080                          SIGVTALRM);
1081         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1082         if (soft != RLIM_INFINITY) {
1083                 unsigned long psecs = cputime_to_secs(ptime);
1084                 unsigned long hard =
1085                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1086                 cputime_t x;
1087                 if (psecs >= hard) {
1088                         /*
1089                          * At the hard limit, we just die.
1090                          * No need to calculate anything else now.
1091                          */
1092                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1093                         return;
1094                 }
1095                 if (psecs >= soft) {
1096                         /*
1097                          * At the soft limit, send a SIGXCPU every second.
1098                          */
1099                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1100                         if (soft < hard) {
1101                                 soft++;
1102                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1103                         }
1104                 }
1105                 x = secs_to_cputime(soft);
1106                 if (!prof_expires || x < prof_expires) {
1107                         prof_expires = x;
1108                 }
1109         }
1110
1111         sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1112         sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1113         sig->cputime_expires.sched_exp = sched_expires;
1114         if (task_cputime_zero(&sig->cputime_expires))
1115                 stop_process_timers(sig);
1116 }
1117
1118 /*
1119  * This is called from the signal code (via do_schedule_next_timer)
1120  * when the last timer signal was delivered and we have to reload the timer.
1121  */
1122 void posix_cpu_timer_schedule(struct k_itimer *timer)
1123 {
1124         struct task_struct *p = timer->it.cpu.task;
1125         unsigned long long now;
1126
1127         if (unlikely(p == NULL))
1128                 /*
1129                  * The task was cleaned up already, no future firings.
1130                  */
1131                 goto out;
1132
1133         /*
1134          * Fetch the current sample and update the timer's expiry time.
1135          */
1136         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1137                 cpu_clock_sample(timer->it_clock, p, &now);
1138                 bump_cpu_timer(timer, now);
1139                 if (unlikely(p->exit_state)) {
1140                         clear_dead_task(timer, now);
1141                         goto out;
1142                 }
1143                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1144                 spin_lock(&p->sighand->siglock);
1145         } else {
1146                 read_lock(&tasklist_lock);
1147                 if (unlikely(p->sighand == NULL)) {
1148                         /*
1149                          * The process has been reaped.
1150                          * We can't even collect a sample any more.
1151                          */
1152                         put_task_struct(p);
1153                         timer->it.cpu.task = p = NULL;
1154                         timer->it.cpu.expires = 0;
1155                         goto out_unlock;
1156                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1157                         /*
1158                          * We've noticed that the thread is dead, but
1159                          * not yet reaped.  Take this opportunity to
1160                          * drop our task ref.
1161                          */
1162                         clear_dead_task(timer, now);
1163                         goto out_unlock;
1164                 }
1165                 spin_lock(&p->sighand->siglock);
1166                 cpu_timer_sample_group(timer->it_clock, p, &now);
1167                 bump_cpu_timer(timer, now);
1168                 /* Leave the tasklist_lock locked for the call below.  */
1169         }
1170
1171         /*
1172          * Now re-arm for the new expiry time.
1173          */
1174         BUG_ON(!irqs_disabled());
1175         arm_timer(timer);
1176         spin_unlock(&p->sighand->siglock);
1177
1178 out_unlock:
1179         read_unlock(&tasklist_lock);
1180
1181 out:
1182         timer->it_overrun_last = timer->it_overrun;
1183         timer->it_overrun = -1;
1184         ++timer->it_requeue_pending;
1185 }
1186
1187 /**
1188  * task_cputime_expired - Compare two task_cputime entities.
1189  *
1190  * @sample:     The task_cputime structure to be checked for expiration.
1191  * @expires:    Expiration times, against which @sample will be checked.
1192  *
1193  * Checks @sample against @expires to see if any field of @sample has expired.
1194  * Returns true if any field of the former is greater than the corresponding
1195  * field of the latter if the latter field is set.  Otherwise returns false.
1196  */
1197 static inline int task_cputime_expired(const struct task_cputime *sample,
1198                                         const struct task_cputime *expires)
1199 {
1200         if (expires->utime && sample->utime >= expires->utime)
1201                 return 1;
1202         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1203                 return 1;
1204         if (expires->sum_exec_runtime != 0 &&
1205             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1206                 return 1;
1207         return 0;
1208 }
1209
1210 /**
1211  * fastpath_timer_check - POSIX CPU timers fast path.
1212  *
1213  * @tsk:        The task (thread) being checked.
1214  *
1215  * Check the task and thread group timers.  If both are zero (there are no
1216  * timers set) return false.  Otherwise snapshot the task and thread group
1217  * timers and compare them with the corresponding expiration times.  Return
1218  * true if a timer has expired, else return false.
1219  */
1220 static inline int fastpath_timer_check(struct task_struct *tsk)
1221 {
1222         struct signal_struct *sig;
1223         cputime_t utime, stime;
1224
1225         task_cputime(tsk, &utime, &stime);
1226
1227         if (!task_cputime_zero(&tsk->cputime_expires)) {
1228                 struct task_cputime task_sample = {
1229                         .utime = utime,
1230                         .stime = stime,
1231                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1232                 };
1233
1234                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1235                         return 1;
1236         }
1237
1238         sig = tsk->signal;
1239         if (sig->cputimer.running) {
1240                 struct task_cputime group_sample;
1241
1242                 raw_spin_lock(&sig->cputimer.lock);
1243                 group_sample = sig->cputimer.cputime;
1244                 raw_spin_unlock(&sig->cputimer.lock);
1245
1246                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1247                         return 1;
1248         }
1249
1250         return 0;
1251 }
1252
1253 /*
1254  * This is called from the timer interrupt handler.  The irq handler has
1255  * already updated our counts.  We need to check if any timers fire now.
1256  * Interrupts are disabled.
1257  */
1258 void run_posix_cpu_timers(struct task_struct *tsk)
1259 {
1260         LIST_HEAD(firing);
1261         struct k_itimer *timer, *next;
1262         unsigned long flags;
1263
1264         BUG_ON(!irqs_disabled());
1265
1266         /*
1267          * The fast path checks that there are no expired thread or thread
1268          * group timers.  If that's so, just return.
1269          */
1270         if (!fastpath_timer_check(tsk))
1271                 return;
1272
1273         if (!lock_task_sighand(tsk, &flags))
1274                 return;
1275         /*
1276          * Here we take off tsk->signal->cpu_timers[N] and
1277          * tsk->cpu_timers[N] all the timers that are firing, and
1278          * put them on the firing list.
1279          */
1280         check_thread_timers(tsk, &firing);
1281         /*
1282          * If there are any active process wide timers (POSIX 1.b, itimers,
1283          * RLIMIT_CPU) cputimer must be running.
1284          */
1285         if (tsk->signal->cputimer.running)
1286                 check_process_timers(tsk, &firing);
1287
1288         /*
1289          * We must release these locks before taking any timer's lock.
1290          * There is a potential race with timer deletion here, as the
1291          * siglock now protects our private firing list.  We have set
1292          * the firing flag in each timer, so that a deletion attempt
1293          * that gets the timer lock before we do will give it up and
1294          * spin until we've taken care of that timer below.
1295          */
1296         unlock_task_sighand(tsk, &flags);
1297
1298         /*
1299          * Now that all the timers on our list have the firing flag,
1300          * no one will touch their list entries but us.  We'll take
1301          * each timer's lock before clearing its firing flag, so no
1302          * timer call will interfere.
1303          */
1304         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1305                 int cpu_firing;
1306
1307                 spin_lock(&timer->it_lock);
1308                 list_del_init(&timer->it.cpu.entry);
1309                 cpu_firing = timer->it.cpu.firing;
1310                 timer->it.cpu.firing = 0;
1311                 /*
1312                  * The firing flag is -1 if we collided with a reset
1313                  * of the timer, which already reported this
1314                  * almost-firing as an overrun.  So don't generate an event.
1315                  */
1316                 if (likely(cpu_firing >= 0))
1317                         cpu_timer_fire(timer);
1318                 spin_unlock(&timer->it_lock);
1319         }
1320
1321         /*
1322          * In case some timers were rescheduled after the queue got emptied,
1323          * wake up full dynticks CPUs.
1324          */
1325         if (tsk->signal->cputimer.running)
1326                 posix_cpu_timer_kick_nohz();
1327 }
1328
1329 /*
1330  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1331  * The tsk->sighand->siglock must be held by the caller.
1332  */
1333 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1334                            cputime_t *newval, cputime_t *oldval)
1335 {
1336         unsigned long long now;
1337
1338         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1339         cpu_timer_sample_group(clock_idx, tsk, &now);
1340
1341         if (oldval) {
1342                 /*
1343                  * We are setting itimer. The *oldval is absolute and we update
1344                  * it to be relative, *newval argument is relative and we update
1345                  * it to be absolute.
1346                  */
1347                 if (*oldval) {
1348                         if (*oldval <= now) {
1349                                 /* Just about to fire. */
1350                                 *oldval = cputime_one_jiffy;
1351                         } else {
1352                                 *oldval -= now;
1353                         }
1354                 }
1355
1356                 if (!*newval)
1357                         goto out;
1358                 *newval += now;
1359         }
1360
1361         /*
1362          * Update expiration cache if we are the earliest timer, or eventually
1363          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1364          */
1365         switch (clock_idx) {
1366         case CPUCLOCK_PROF:
1367                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1368                         tsk->signal->cputime_expires.prof_exp = *newval;
1369                 break;
1370         case CPUCLOCK_VIRT:
1371                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1372                         tsk->signal->cputime_expires.virt_exp = *newval;
1373                 break;
1374         }
1375 out:
1376         posix_cpu_timer_kick_nohz();
1377 }
1378
1379 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1380                             struct timespec *rqtp, struct itimerspec *it)
1381 {
1382         struct k_itimer timer;
1383         int error;
1384
1385         /*
1386          * Set up a temporary timer and then wait for it to go off.
1387          */
1388         memset(&timer, 0, sizeof timer);
1389         spin_lock_init(&timer.it_lock);
1390         timer.it_clock = which_clock;
1391         timer.it_overrun = -1;
1392         error = posix_cpu_timer_create(&timer);
1393         timer.it_process = current;
1394         if (!error) {
1395                 static struct itimerspec zero_it;
1396
1397                 memset(it, 0, sizeof *it);
1398                 it->it_value = *rqtp;
1399
1400                 spin_lock_irq(&timer.it_lock);
1401                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1402                 if (error) {
1403                         spin_unlock_irq(&timer.it_lock);
1404                         return error;
1405                 }
1406
1407                 while (!signal_pending(current)) {
1408                         if (timer.it.cpu.expires == 0) {
1409                                 /*
1410                                  * Our timer fired and was reset, below
1411                                  * deletion can not fail.
1412                                  */
1413                                 posix_cpu_timer_del(&timer);
1414                                 spin_unlock_irq(&timer.it_lock);
1415                                 return 0;
1416                         }
1417
1418                         /*
1419                          * Block until cpu_timer_fire (or a signal) wakes us.
1420                          */
1421                         __set_current_state(TASK_INTERRUPTIBLE);
1422                         spin_unlock_irq(&timer.it_lock);
1423                         schedule();
1424                         spin_lock_irq(&timer.it_lock);
1425                 }
1426
1427                 /*
1428                  * We were interrupted by a signal.
1429                  */
1430                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1431                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1432                 if (!error) {
1433                         /*
1434                          * Timer is now unarmed, deletion can not fail.
1435                          */
1436                         posix_cpu_timer_del(&timer);
1437                 }
1438                 spin_unlock_irq(&timer.it_lock);
1439
1440                 while (error == TIMER_RETRY) {
1441                         /*
1442                          * We need to handle case when timer was or is in the
1443                          * middle of firing. In other cases we already freed
1444                          * resources.
1445                          */
1446                         spin_lock_irq(&timer.it_lock);
1447                         error = posix_cpu_timer_del(&timer);
1448                         spin_unlock_irq(&timer.it_lock);
1449                 }
1450
1451                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1452                         /*
1453                          * It actually did fire already.
1454                          */
1455                         return 0;
1456                 }
1457
1458                 error = -ERESTART_RESTARTBLOCK;
1459         }
1460
1461         return error;
1462 }
1463
1464 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1465
1466 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1467                             struct timespec *rqtp, struct timespec __user *rmtp)
1468 {
1469         struct restart_block *restart_block =
1470                 &current_thread_info()->restart_block;
1471         struct itimerspec it;
1472         int error;
1473
1474         /*
1475          * Diagnose required errors first.
1476          */
1477         if (CPUCLOCK_PERTHREAD(which_clock) &&
1478             (CPUCLOCK_PID(which_clock) == 0 ||
1479              CPUCLOCK_PID(which_clock) == current->pid))
1480                 return -EINVAL;
1481
1482         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1483
1484         if (error == -ERESTART_RESTARTBLOCK) {
1485
1486                 if (flags & TIMER_ABSTIME)
1487                         return -ERESTARTNOHAND;
1488                 /*
1489                  * Report back to the user the time still remaining.
1490                  */
1491                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1492                         return -EFAULT;
1493
1494                 restart_block->fn = posix_cpu_nsleep_restart;
1495                 restart_block->nanosleep.clockid = which_clock;
1496                 restart_block->nanosleep.rmtp = rmtp;
1497                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1498         }
1499         return error;
1500 }
1501
1502 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1503 {
1504         clockid_t which_clock = restart_block->nanosleep.clockid;
1505         struct timespec t;
1506         struct itimerspec it;
1507         int error;
1508
1509         t = ns_to_timespec(restart_block->nanosleep.expires);
1510
1511         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1512
1513         if (error == -ERESTART_RESTARTBLOCK) {
1514                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1515                 /*
1516                  * Report back to the user the time still remaining.
1517                  */
1518                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1519                         return -EFAULT;
1520
1521                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1522         }
1523         return error;
1524
1525 }
1526
1527 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1528 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1529
1530 static int process_cpu_clock_getres(const clockid_t which_clock,
1531                                     struct timespec *tp)
1532 {
1533         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1534 }
1535 static int process_cpu_clock_get(const clockid_t which_clock,
1536                                  struct timespec *tp)
1537 {
1538         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1539 }
1540 static int process_cpu_timer_create(struct k_itimer *timer)
1541 {
1542         timer->it_clock = PROCESS_CLOCK;
1543         return posix_cpu_timer_create(timer);
1544 }
1545 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1546                               struct timespec *rqtp,
1547                               struct timespec __user *rmtp)
1548 {
1549         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1550 }
1551 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1552 {
1553         return -EINVAL;
1554 }
1555 static int thread_cpu_clock_getres(const clockid_t which_clock,
1556                                    struct timespec *tp)
1557 {
1558         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1559 }
1560 static int thread_cpu_clock_get(const clockid_t which_clock,
1561                                 struct timespec *tp)
1562 {
1563         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1564 }
1565 static int thread_cpu_timer_create(struct k_itimer *timer)
1566 {
1567         timer->it_clock = THREAD_CLOCK;
1568         return posix_cpu_timer_create(timer);
1569 }
1570
1571 struct k_clock clock_posix_cpu = {
1572         .clock_getres   = posix_cpu_clock_getres,
1573         .clock_set      = posix_cpu_clock_set,
1574         .clock_get      = posix_cpu_clock_get,
1575         .timer_create   = posix_cpu_timer_create,
1576         .nsleep         = posix_cpu_nsleep,
1577         .nsleep_restart = posix_cpu_nsleep_restart,
1578         .timer_set      = posix_cpu_timer_set,
1579         .timer_del      = posix_cpu_timer_del,
1580         .timer_get      = posix_cpu_timer_get,
1581 };
1582
1583 static __init int init_posix_cpu_timers(void)
1584 {
1585         struct k_clock process = {
1586                 .clock_getres   = process_cpu_clock_getres,
1587                 .clock_get      = process_cpu_clock_get,
1588                 .timer_create   = process_cpu_timer_create,
1589                 .nsleep         = process_cpu_nsleep,
1590                 .nsleep_restart = process_cpu_nsleep_restart,
1591         };
1592         struct k_clock thread = {
1593                 .clock_getres   = thread_cpu_clock_getres,
1594                 .clock_get      = thread_cpu_clock_get,
1595                 .timer_create   = thread_cpu_timer_create,
1596         };
1597         struct timespec ts;
1598
1599         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1600         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1601
1602         cputime_to_timespec(cputime_one_jiffy, &ts);
1603         onecputick = ts.tv_nsec;
1604         WARN_ON(ts.tv_sec != 0);
1605
1606         return 0;
1607 }
1608 __initcall(init_posix_cpu_timers);