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
  14 /*
  15  * Called after updating RLIMIT_CPU to run cpu timer and update
  16  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  17  * siglock protection since other code may update expiration cache as
  18  * well.
  19  */
  20 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
  21 {
  22         cputime_t cputime = secs_to_cputime(rlim_new);
  23
  24         spin_lock_irq(&task->sighand->siglock);
  25         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  26         spin_unlock_irq(&task->sighand->siglock);
  27 }
  28
  29 static int check_clock(const clockid_t which_clock)
  30 {
  31         int error = 0;
  32         struct task_struct *p;
  33         const pid_t pid = CPUCLOCK_PID(which_clock);
  34
  35         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  36                 return -EINVAL;
  37
  38         if (pid == 0)
  39                 return 0;
  40
  41         rcu_read_lock();
  42         p = find_task_by_vpid(pid);
  43         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  44                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  45                 error = -EINVAL;
  46         }
  47         rcu_read_unlock();
  48
  49         return error;
  50 }
  51
  52 static inline union cpu_time_count
  53 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  54 {
  55         union cpu_time_count ret;
  56         ret.sched = 0;          /* high half always zero when .cpu used */
  57         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  58                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  59         } else {
  60                 ret.cpu = timespec_to_cputime(tp);
  61         }
  62         return ret;
  63 }
  64
  65 static void sample_to_timespec(const clockid_t which_clock,
  66                                union cpu_time_count cpu,
  67                                struct timespec *tp)
  68 {
  69         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  70                 *tp = ns_to_timespec(cpu.sched);
  71         else
  72                 cputime_to_timespec(cpu.cpu, tp);
  73 }
  74
  75 static inline int cpu_time_before(const clockid_t which_clock,
  76                                   union cpu_time_count now,
  77                                   union cpu_time_count then)
  78 {
  79         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  80                 return now.sched < then.sched;
  81         }  else {
  82                 return now.cpu < then.cpu;
  83         }
  84 }
  85 static inline void cpu_time_add(const clockid_t which_clock,
  86                                 union cpu_time_count *acc,
  87                                 union cpu_time_count val)
  88 {
  89         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  90                 acc->sched += val.sched;
  91         }  else {
  92                 acc->cpu += val.cpu;
  93         }
  94 }
  95 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  96                                                 union cpu_time_count a,
  97                                                 union cpu_time_count b)
  98 {
  99         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 100                 a.sched -= b.sched;
 101         }  else {
 102                 a.cpu -= b.cpu;
 103         }
 104         return a;
 105 }
 106
 107 /*
 108  * Update expiry time from increment, and increase overrun count,
 109  * given the current clock sample.
 110  */
 111 static void bump_cpu_timer(struct k_itimer *timer,
 112                                   union cpu_time_count now)
 113 {
 114         int i;
 115
 116         if (timer->it.cpu.incr.sched == 0)
 117                 return;
 118
 119         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 120                 unsigned long long delta, incr;
 121
 122                 if (now.sched < timer->it.cpu.expires.sched)
 123                         return;
 124                 incr = timer->it.cpu.incr.sched;
 125                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 126                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 127                 for (i = 0; incr < delta - incr; i++)
 128                         incr = incr << 1;
 129                 for (; i >= 0; incr >>= 1, i--) {
 130                         if (delta < incr)
 131                                 continue;
 132                         timer->it.cpu.expires.sched += incr;
 133                         timer->it_overrun += 1 << i;
 134                         delta -= incr;
 135                 }
 136         } else {
 137                 cputime_t delta, incr;
 138
 139                 if (now.cpu < timer->it.cpu.expires.cpu)
 140                         return;
 141                 incr = timer->it.cpu.incr.cpu;
 142                 delta = now.cpu + incr - timer->it.cpu.expires.cpu;
 143                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 144                 for (i = 0; incr < delta - incr; i++)
 145                              incr += incr;
 146                 for (; i >= 0; incr = incr >> 1, i--) {
 147                         if (delta < incr)
 148                                 continue;
 149                         timer->it.cpu.expires.cpu += incr;
 150                         timer->it_overrun += 1 << i;
 151                         delta -= incr;
 152                 }
 153         }
 154 }
 155
 156 static inline cputime_t prof_ticks(struct task_struct *p)
 157 {
 158         return p->utime + p->stime;
 159 }
 160 static inline cputime_t virt_ticks(struct task_struct *p)
 161 {
 162         return p->utime;
 163 }
 164
 165 static int
 166 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 167 {
 168         int error = check_clock(which_clock);
 169         if (!error) {
 170                 tp->tv_sec = 0;
 171                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 172                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 173                         /*
 174                          * If sched_clock is using a cycle counter, we
 175                          * don't have any idea of its true resolution
 176                          * exported, but it is much more than 1s/HZ.
 177                          */
 178                         tp->tv_nsec = 1;
 179                 }
 180         }
 181         return error;
 182 }
 183
 184 static int
 185 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 186 {
 187         /*
 188          * You can never reset a CPU clock, but we check for other errors
 189          * in the call before failing with EPERM.
 190          */
 191         int error = check_clock(which_clock);
 192         if (error == 0) {
 193                 error = -EPERM;
 194         }
 195         return error;
 196 }
 197
 198
 199 /*
 200  * Sample a per-thread clock for the given task.
 201  */
 202 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 203                             union cpu_time_count *cpu)
 204 {
 205         switch (CPUCLOCK_WHICH(which_clock)) {
 206         default:
 207                 return -EINVAL;
 208         case CPUCLOCK_PROF:
 209                 cpu->cpu = prof_ticks(p);
 210                 break;
 211         case CPUCLOCK_VIRT:
 212                 cpu->cpu = virt_ticks(p);
 213                 break;
 214         case CPUCLOCK_SCHED:
 215                 cpu->sched = task_sched_runtime(p);
 216                 break;
 217         }
 218         return 0;
 219 }
 220
 221 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 222 {
 223         if (b->utime > a->utime)
 224                 a->utime = b->utime;
 225
 226         if (b->stime > a->stime)
 227                 a->stime = b->stime;
 228
 229         if (b->sum_exec_runtime > a->sum_exec_runtime)
 230                 a->sum_exec_runtime = b->sum_exec_runtime;
 231 }
 232
 233 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 234 {
 235         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 236         struct task_cputime sum;
 237         unsigned long flags;
 238
 239         if (!cputimer->running) {
 240                 /*
 241                  * The POSIX timer interface allows for absolute time expiry
 242                  * values through the TIMER_ABSTIME flag, therefore we have
 243                  * to synchronize the timer to the clock every time we start
 244                  * it.
 245                  */
 246                 thread_group_cputime(tsk, &sum);
 247                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 248                 cputimer->running = 1;
 249                 update_gt_cputime(&cputimer->cputime, &sum);
 250         } else
 251                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 252         *times = cputimer->cputime;
 253         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 254 }
 255
 256 /*
 257  * Sample a process (thread group) clock for the given group_leader task.
 258  * Must be called with tasklist_lock held for reading.
 259  */
 260 static int cpu_clock_sample_group(const clockid_t which_clock,
 261                                   struct task_struct *p,
 262                                   union cpu_time_count *cpu)
 263 {
 264         struct task_cputime cputime;
 265
 266         switch (CPUCLOCK_WHICH(which_clock)) {
 267         default:
 268                 return -EINVAL;
 269         case CPUCLOCK_PROF:
 270                 thread_group_cputime(p, &cputime);
 271                 cpu->cpu = cputime.utime + cputime.stime;
 272                 break;
 273         case CPUCLOCK_VIRT:
 274                 thread_group_cputime(p, &cputime);
 275                 cpu->cpu = cputime.utime;
 276                 break;
 277         case CPUCLOCK_SCHED:
 278                 thread_group_cputime(p, &cputime);
 279                 cpu->sched = cputime.sum_exec_runtime;
 280                 break;
 281         }
 282         return 0;
 283 }
 284
 285
 286 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 287 {
 288         const pid_t pid = CPUCLOCK_PID(which_clock);
 289         int error = -EINVAL;
 290         union cpu_time_count rtn;
 291
 292         if (pid == 0) {
 293                 /*
 294                  * Special case constant value for our own clocks.
 295                  * We don't have to do any lookup to find ourselves.
 296                  */
 297                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 298                         /*
 299                          * Sampling just ourselves we can do with no locking.
 300                          */
 301                         error = cpu_clock_sample(which_clock,
 302                                                  current, &rtn);
 303                 } else {
 304                         read_lock(&tasklist_lock);
 305                         error = cpu_clock_sample_group(which_clock,
 306                                                        current, &rtn);
 307                         read_unlock(&tasklist_lock);
 308                 }
 309         } else {
 310                 /*
 311                  * Find the given PID, and validate that the caller
 312                  * should be able to see it.
 313                  */
 314                 struct task_struct *p;
 315                 rcu_read_lock();
 316                 p = find_task_by_vpid(pid);
 317                 if (p) {
 318                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 319                                 if (same_thread_group(p, current)) {
 320                                         error = cpu_clock_sample(which_clock,
 321                                                                  p, &rtn);
 322                                 }
 323                         } else {
 324                                 read_lock(&tasklist_lock);
 325                                 if (thread_group_leader(p) && p->sighand) {
 326                                         error =
 327                                             cpu_clock_sample_group(which_clock,
 328                                                                    p, &rtn);
 329                                 }
 330                                 read_unlock(&tasklist_lock);
 331                         }
 332                 }
 333                 rcu_read_unlock();
 334         }
 335
 336         if (error)
 337                 return error;
 338         sample_to_timespec(which_clock, rtn, tp);
 339         return 0;
 340 }
 341
 342
 343 /*
 344  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 345  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 346  * new timer already all-zeros initialized.
 347  */
 348 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 349 {
 350         int ret = 0;
 351         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 352         struct task_struct *p;
 353
 354         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 355                 return -EINVAL;
 356
 357         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 358
 359         rcu_read_lock();
 360         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 361                 if (pid == 0) {
 362                         p = current;
 363                 } else {
 364                         p = find_task_by_vpid(pid);
 365                         if (p && !same_thread_group(p, current))
 366                                 p = NULL;
 367                 }
 368         } else {
 369                 if (pid == 0) {
 370                         p = current->group_leader;
 371                 } else {
 372                         p = find_task_by_vpid(pid);
 373                         if (p && !has_group_leader_pid(p))
 374                                 p = NULL;
 375                 }
 376         }
 377         new_timer->it.cpu.task = p;
 378         if (p) {
 379                 get_task_struct(p);
 380         } else {
 381                 ret = -EINVAL;
 382         }
 383         rcu_read_unlock();
 384
 385         return ret;
 386 }
 387
 388 /*
 389  * Clean up a CPU-clock timer that is about to be destroyed.
 390  * This is called from timer deletion with the timer already locked.
 391  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 392  * and try again.  (This happens when the timer is in the middle of firing.)
 393  */
 394 static int posix_cpu_timer_del(struct k_itimer *timer)
 395 {
 396         struct task_struct *p = timer->it.cpu.task;
 397         int ret = 0;
 398
 399         if (likely(p != NULL)) {
 400                 read_lock(&tasklist_lock);
 401                 if (unlikely(p->sighand == NULL)) {
 402                         /*
 403                          * We raced with the reaping of the task.
 404                          * The deletion should have cleared us off the list.
 405                          */
 406                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 407                 } else {
 408                         spin_lock(&p->sighand->siglock);
 409                         if (timer->it.cpu.firing)
 410                                 ret = TIMER_RETRY;
 411                         else
 412                                 list_del(&timer->it.cpu.entry);
 413                         spin_unlock(&p->sighand->siglock);
 414                 }
 415                 read_unlock(&tasklist_lock);
 416
 417                 if (!ret)
 418                         put_task_struct(p);
 419         }
 420
 421         return ret;
 422 }
 423
 424 /*
 425  * Clean out CPU timers still ticking when a thread exited.  The task
 426  * pointer is cleared, and the expiry time is replaced with the residual
 427  * time for later timer_gettime calls to return.
 428  * This must be called with the siglock held.
 429  */
 430 static void cleanup_timers(struct list_head *head,
 431                            cputime_t utime, cputime_t stime,
 432                            unsigned long long sum_exec_runtime)
 433 {
 434         struct cpu_timer_list *timer, *next;
 435         cputime_t ptime = utime + stime;
 436
 437         list_for_each_entry_safe(timer, next, head, entry) {
 438                 list_del_init(&timer->entry);
 439                 if (timer->expires.cpu < ptime) {
 440                         timer->expires.cpu = 0;
 441                 } else {
 442                         timer->expires.cpu -= ptime;
 443                 }
 444         }
 445
 446         ++head;
 447         list_for_each_entry_safe(timer, next, head, entry) {
 448                 list_del_init(&timer->entry);
 449                 if (timer->expires.cpu < utime) {
 450                         timer->expires.cpu = 0;
 451                 } else {
 452                         timer->expires.cpu -= utime;
 453                 }
 454         }
 455
 456         ++head;
 457         list_for_each_entry_safe(timer, next, head, entry) {
 458                 list_del_init(&timer->entry);
 459                 if (timer->expires.sched < sum_exec_runtime) {
 460                         timer->expires.sched = 0;
 461                 } else {
 462                         timer->expires.sched -= sum_exec_runtime;
 463                 }
 464         }
 465 }
 466
 467 /*
 468  * These are both called with the siglock held, when the current thread
 469  * is being reaped.  When the final (leader) thread in the group is reaped,
 470  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 471  */
 472 void posix_cpu_timers_exit(struct task_struct *tsk)
 473 {
 474         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 475                                                 sizeof(unsigned long long));
 476         cleanup_timers(tsk->cpu_timers,
 477                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 478
 479 }
 480 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 481 {
 482         struct signal_struct *const sig = tsk->signal;
 483
 484         cleanup_timers(tsk->signal->cpu_timers,
 485                        tsk->utime + sig->utime, tsk->stime + sig->stime,
 486                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 487 }
 488
 489 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 490 {
 491         /*
 492          * That's all for this thread or process.
 493          * We leave our residual in expires to be reported.
 494          */
 495         put_task_struct(timer->it.cpu.task);
 496         timer->it.cpu.task = NULL;
 497         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 498                                              timer->it.cpu.expires,
 499                                              now);
 500 }
 501
 502 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 503 {
 504         return expires == 0 || expires > new_exp;
 505 }
 506
 507 /*
 508  * Insert the timer on the appropriate list before any timers that
 509  * expire later.  This must be called with the tasklist_lock held
 510  * for reading, interrupts disabled and p->sighand->siglock taken.
 511  */
 512 static void arm_timer(struct k_itimer *timer)
 513 {
 514         struct task_struct *p = timer->it.cpu.task;
 515         struct list_head *head, *listpos;
 516         struct task_cputime *cputime_expires;
 517         struct cpu_timer_list *const nt = &timer->it.cpu;
 518         struct cpu_timer_list *next;
 519
 520         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 521                 head = p->cpu_timers;
 522                 cputime_expires = &p->cputime_expires;
 523         } else {
 524                 head = p->signal->cpu_timers;
 525                 cputime_expires = &p->signal->cputime_expires;
 526         }
 527         head += CPUCLOCK_WHICH(timer->it_clock);
 528
 529         listpos = head;
 530         list_for_each_entry(next, head, entry) {
 531                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
 532                         break;
 533                 listpos = &next->entry;
 534         }
 535         list_add(&nt->entry, listpos);
 536
 537         if (listpos == head) {
 538                 union cpu_time_count *exp = &nt->expires;
 539
 540                 /*
 541                  * We are the new earliest-expiring POSIX 1.b timer, hence
 542                  * need to update expiration cache. Take into account that
 543                  * for process timers we share expiration cache with itimers
 544                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 545                  */
 546
 547                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 548                 case CPUCLOCK_PROF:
 549                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
 550                                 cputime_expires->prof_exp = exp->cpu;
 551                         break;
 552                 case CPUCLOCK_VIRT:
 553                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
 554                                 cputime_expires->virt_exp = exp->cpu;
 555                         break;
 556                 case CPUCLOCK_SCHED:
 557                         if (cputime_expires->sched_exp == 0 ||
 558                             cputime_expires->sched_exp > exp->sched)
 559                                 cputime_expires->sched_exp = exp->sched;
 560                         break;
 561                 }
 562         }
 563 }
 564
 565 /*
 566  * The timer is locked, fire it and arrange for its reload.
 567  */
 568 static void cpu_timer_fire(struct k_itimer *timer)
 569 {
 570         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 571                 /*
 572                  * User don't want any signal.
 573                  */
 574                 timer->it.cpu.expires.sched = 0;
 575         } else if (unlikely(timer->sigq == NULL)) {
 576                 /*
 577                  * This a special case for clock_nanosleep,
 578                  * not a normal timer from sys_timer_create.
 579                  */
 580                 wake_up_process(timer->it_process);
 581                 timer->it.cpu.expires.sched = 0;
 582         } else if (timer->it.cpu.incr.sched == 0) {
 583                 /*
 584                  * One-shot timer.  Clear it as soon as it's fired.
 585                  */
 586                 posix_timer_event(timer, 0);
 587                 timer->it.cpu.expires.sched = 0;
 588         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 589                 /*
 590                  * The signal did not get queued because the signal
 591                  * was ignored, so we won't get any callback to
 592                  * reload the timer.  But we need to keep it
 593                  * ticking in case the signal is deliverable next time.
 594                  */
 595                 posix_cpu_timer_schedule(timer);
 596         }
 597 }
 598
 599 /*
 600  * Sample a process (thread group) timer for the given group_leader task.
 601  * Must be called with tasklist_lock held for reading.
 602  */
 603 static int cpu_timer_sample_group(const clockid_t which_clock,
 604                                   struct task_struct *p,
 605                                   union cpu_time_count *cpu)
 606 {
 607         struct task_cputime cputime;
 608
 609         thread_group_cputimer(p, &cputime);
 610         switch (CPUCLOCK_WHICH(which_clock)) {
 611         default:
 612                 return -EINVAL;
 613         case CPUCLOCK_PROF:
 614                 cpu->cpu = cputime.utime + cputime.stime;
 615                 break;
 616         case CPUCLOCK_VIRT:
 617                 cpu->cpu = cputime.utime;
 618                 break;
 619         case CPUCLOCK_SCHED:
 620                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 621                 break;
 622         }
 623         return 0;
 624 }
 625
 626 /*
 627  * Guts of sys_timer_settime for CPU timers.
 628  * This is called with the timer locked and interrupts disabled.
 629  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 630  * and try again.  (This happens when the timer is in the middle of firing.)
 631  */
 632 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 633                                struct itimerspec *new, struct itimerspec *old)
 634 {
 635         struct task_struct *p = timer->it.cpu.task;
 636         union cpu_time_count old_expires, new_expires, old_incr, val;
 637         int ret;
 638
 639         if (unlikely(p == NULL)) {
 640                 /*
 641                  * Timer refers to a dead task's clock.
 642                  */
 643                 return -ESRCH;
 644         }
 645
 646         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 647
 648         read_lock(&tasklist_lock);
 649         /*
 650          * We need the tasklist_lock to protect against reaping that
 651          * clears p->sighand.  If p has just been reaped, we can no
 652          * longer get any information about it at all.
 653          */
 654         if (unlikely(p->sighand == NULL)) {
 655                 read_unlock(&tasklist_lock);
 656                 put_task_struct(p);
 657                 timer->it.cpu.task = NULL;
 658                 return -ESRCH;
 659         }
 660
 661         /*
 662          * Disarm any old timer after extracting its expiry time.
 663          */
 664         BUG_ON(!irqs_disabled());
 665
 666         ret = 0;
 667         old_incr = timer->it.cpu.incr;
 668         spin_lock(&p->sighand->siglock);
 669         old_expires = timer->it.cpu.expires;
 670         if (unlikely(timer->it.cpu.firing)) {
 671                 timer->it.cpu.firing = -1;
 672                 ret = TIMER_RETRY;
 673         } else
 674                 list_del_init(&timer->it.cpu.entry);
 675
 676         /*
 677          * We need to sample the current value to convert the new
 678          * value from to relative and absolute, and to convert the
 679          * old value from absolute to relative.  To set a process
 680          * timer, we need a sample to balance the thread expiry
 681          * times (in arm_timer).  With an absolute time, we must
 682          * check if it's already passed.  In short, we need a sample.
 683          */
 684         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 685                 cpu_clock_sample(timer->it_clock, p, &val);
 686         } else {
 687                 cpu_timer_sample_group(timer->it_clock, p, &val);
 688         }
 689
 690         if (old) {
 691                 if (old_expires.sched == 0) {
 692                         old->it_value.tv_sec = 0;
 693                         old->it_value.tv_nsec = 0;
 694                 } else {
 695                         /*
 696                          * Update the timer in case it has
 697                          * overrun already.  If it has,
 698                          * we'll report it as having overrun
 699                          * and with the next reloaded timer
 700                          * already ticking, though we are
 701                          * swallowing that pending
 702                          * notification here to install the
 703                          * new setting.
 704                          */
 705                         bump_cpu_timer(timer, val);
 706                         if (cpu_time_before(timer->it_clock, val,
 707                                             timer->it.cpu.expires)) {
 708                                 old_expires = cpu_time_sub(
 709                                         timer->it_clock,
 710                                         timer->it.cpu.expires, val);
 711                                 sample_to_timespec(timer->it_clock,
 712                                                    old_expires,
 713                                                    &old->it_value);
 714                         } else {
 715                                 old->it_value.tv_nsec = 1;
 716                                 old->it_value.tv_sec = 0;
 717                         }
 718                 }
 719         }
 720
 721         if (unlikely(ret)) {
 722                 /*
 723                  * We are colliding with the timer actually firing.
 724                  * Punt after filling in the timer's old value, and
 725                  * disable this firing since we are already reporting
 726                  * it as an overrun (thanks to bump_cpu_timer above).
 727                  */
 728                 spin_unlock(&p->sighand->siglock);
 729                 read_unlock(&tasklist_lock);
 730                 goto out;
 731         }
 732
 733         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 734                 cpu_time_add(timer->it_clock, &new_expires, val);
 735         }
 736
 737         /*
 738          * Install the new expiry time (or zero).
 739          * For a timer with no notification action, we don't actually
 740          * arm the timer (we'll just fake it for timer_gettime).
 741          */
 742         timer->it.cpu.expires = new_expires;
 743         if (new_expires.sched != 0 &&
 744             cpu_time_before(timer->it_clock, val, new_expires)) {
 745                 arm_timer(timer);
 746         }
 747
 748         spin_unlock(&p->sighand->siglock);
 749         read_unlock(&tasklist_lock);
 750
 751         /*
 752          * Install the new reload setting, and
 753          * set up the signal and overrun bookkeeping.
 754          */
 755         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 756                                                 &new->it_interval);
 757
 758         /*
 759          * This acts as a modification timestamp for the timer,
 760          * so any automatic reload attempt will punt on seeing
 761          * that we have reset the timer manually.
 762          */
 763         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 764                 ~REQUEUE_PENDING;
 765         timer->it_overrun_last = 0;
 766         timer->it_overrun = -1;
 767
 768         if (new_expires.sched != 0 &&
 769             !cpu_time_before(timer->it_clock, val, new_expires)) {
 770                 /*
 771                  * The designated time already passed, so we notify
 772                  * immediately, even if the thread never runs to
 773                  * accumulate more time on this clock.
 774                  */
 775                 cpu_timer_fire(timer);
 776         }
 777
 778         ret = 0;
 779  out:
 780         if (old) {
 781                 sample_to_timespec(timer->it_clock,
 782                                    old_incr, &old->it_interval);
 783         }
 784         return ret;
 785 }
 786
 787 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 788 {
 789         union cpu_time_count now;
 790         struct task_struct *p = timer->it.cpu.task;
 791         int clear_dead;
 792
 793         /*
 794          * Easy part: convert the reload time.
 795          */
 796         sample_to_timespec(timer->it_clock,
 797                            timer->it.cpu.incr, &itp->it_interval);
 798
 799         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 800                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 801                 return;
 802         }
 803
 804         if (unlikely(p == NULL)) {
 805                 /*
 806                  * This task already died and the timer will never fire.
 807                  * In this case, expires is actually the dead value.
 808                  */
 809         dead:
 810                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 811                                    &itp->it_value);
 812                 return;
 813         }
 814
 815         /*
 816          * Sample the clock to take the difference with the expiry time.
 817          */
 818         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 819                 cpu_clock_sample(timer->it_clock, p, &now);
 820                 clear_dead = p->exit_state;
 821         } else {
 822                 read_lock(&tasklist_lock);
 823                 if (unlikely(p->sighand == NULL)) {
 824                         /*
 825                          * The process has been reaped.
 826                          * We can't even collect a sample any more.
 827                          * Call the timer disarmed, nothing else to do.
 828                          */
 829                         put_task_struct(p);
 830                         timer->it.cpu.task = NULL;
 831                         timer->it.cpu.expires.sched = 0;
 832                         read_unlock(&tasklist_lock);
 833                         goto dead;
 834                 } else {
 835                         cpu_timer_sample_group(timer->it_clock, p, &now);
 836                         clear_dead = (unlikely(p->exit_state) &&
 837                                       thread_group_empty(p));
 838                 }
 839                 read_unlock(&tasklist_lock);
 840         }
 841
 842         if (unlikely(clear_dead)) {
 843                 /*
 844                  * We've noticed that the thread is dead, but
 845                  * not yet reaped.  Take this opportunity to
 846                  * drop our task ref.
 847                  */
 848                 clear_dead_task(timer, now);
 849                 goto dead;
 850         }
 851
 852         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 853                 sample_to_timespec(timer->it_clock,
 854                                    cpu_time_sub(timer->it_clock,
 855                                                 timer->it.cpu.expires, now),
 856                                    &itp->it_value);
 857         } else {
 858                 /*
 859                  * The timer should have expired already, but the firing
 860                  * hasn't taken place yet.  Say it's just about to expire.
 861                  */
 862                 itp->it_value.tv_nsec = 1;
 863                 itp->it_value.tv_sec = 0;
 864         }
 865 }
 866
 867 /*
 868  * Check for any per-thread CPU timers that have fired and move them off
 869  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 870  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 871  */
 872 static void check_thread_timers(struct task_struct *tsk,
 873                                 struct list_head *firing)
 874 {
 875         int maxfire;
 876         struct list_head *timers = tsk->cpu_timers;
 877         struct signal_struct *const sig = tsk->signal;
 878         unsigned long soft;
 879
 880         maxfire = 20;
 881         tsk->cputime_expires.prof_exp = 0;
 882         while (!list_empty(timers)) {
 883                 struct cpu_timer_list *t = list_first_entry(timers,
 884                                                       struct cpu_timer_list,
 885                                                       entry);
 886                 if (!--maxfire || prof_ticks(tsk) < t->expires.cpu) {
 887                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 888                         break;
 889                 }
 890                 t->firing = 1;
 891                 list_move_tail(&t->entry, firing);
 892         }
 893
 894         ++timers;
 895         maxfire = 20;
 896         tsk->cputime_expires.virt_exp = 0;
 897         while (!list_empty(timers)) {
 898                 struct cpu_timer_list *t = list_first_entry(timers,
 899                                                       struct cpu_timer_list,
 900                                                       entry);
 901                 if (!--maxfire || virt_ticks(tsk) < t->expires.cpu) {
 902                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 903                         break;
 904                 }
 905                 t->firing = 1;
 906                 list_move_tail(&t->entry, firing);
 907         }
 908
 909         ++timers;
 910         maxfire = 20;
 911         tsk->cputime_expires.sched_exp = 0;
 912         while (!list_empty(timers)) {
 913                 struct cpu_timer_list *t = list_first_entry(timers,
 914                                                       struct cpu_timer_list,
 915                                                       entry);
 916                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
 917                         tsk->cputime_expires.sched_exp = t->expires.sched;
 918                         break;
 919                 }
 920                 t->firing = 1;
 921                 list_move_tail(&t->entry, firing);
 922         }
 923
 924         /*
 925          * Check for the special case thread timers.
 926          */
 927         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 928         if (soft != RLIM_INFINITY) {
 929                 unsigned long hard =
 930                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 931
 932                 if (hard != RLIM_INFINITY &&
 933                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 934                         /*
 935                          * At the hard limit, we just die.
 936                          * No need to calculate anything else now.
 937                          */
 938                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 939                         return;
 940                 }
 941                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 942                         /*
 943                          * At the soft limit, send a SIGXCPU every second.
 944                          */
 945                         if (soft < hard) {
 946                                 soft += USEC_PER_SEC;
 947                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 948                         }
 949                         printk(KERN_INFO
 950                                 "RT Watchdog Timeout: %s[%d]\n",
 951                                 tsk->comm, task_pid_nr(tsk));
 952                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 953                 }
 954         }
 955 }
 956
 957 static void stop_process_timers(struct signal_struct *sig)
 958 {
 959         struct thread_group_cputimer *cputimer = &sig->cputimer;
 960         unsigned long flags;
 961
 962         raw_spin_lock_irqsave(&cputimer->lock, flags);
 963         cputimer->running = 0;
 964         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 965 }
 966
 967 static u32 onecputick;
 968
 969 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 970                              cputime_t *expires, cputime_t cur_time, int signo)
 971 {
 972         if (!it->expires)
 973                 return;
 974
 975         if (cur_time >= it->expires) {
 976                 if (it->incr) {
 977                         it->expires += it->incr;
 978                         it->error += it->incr_error;
 979                         if (it->error >= onecputick) {
 980                                 it->expires -= cputime_one_jiffy;
 981                                 it->error -= onecputick;
 982                         }
 983                 } else {
 984                         it->expires = 0;
 985                 }
 986
 987                 trace_itimer_expire(signo == SIGPROF ?
 988                                     ITIMER_PROF : ITIMER_VIRTUAL,
 989                                     tsk->signal->leader_pid, cur_time);
 990                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 991         }
 992
 993         if (it->expires && (!*expires || it->expires < *expires)) {
 994                 *expires = it->expires;
 995         }
 996 }
 997
 998 /**
 999  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1000  *
1001  * @cputime:    The struct to compare.
1002  *
1003  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1004  * are zero, false if any field is nonzero.
1005  */
1006 static inline int task_cputime_zero(const struct task_cputime *cputime)
1007 {
1008         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
1009                 return 1;
1010         return 0;
1011 }
1012
1013 /*
1014  * Check for any per-thread CPU timers that have fired and move them
1015  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1016  * have already been taken off.
1017  */
1018 static void check_process_timers(struct task_struct *tsk,
1019                                  struct list_head *firing)
1020 {
1021         int maxfire;
1022         struct signal_struct *const sig = tsk->signal;
1023         cputime_t utime, ptime, virt_expires, prof_expires;
1024         unsigned long long sum_sched_runtime, sched_expires;
1025         struct list_head *timers = sig->cpu_timers;
1026         struct task_cputime cputime;
1027         unsigned long soft;
1028
1029         /*
1030          * Collect the current process totals.
1031          */
1032         thread_group_cputimer(tsk, &cputime);
1033         utime = cputime.utime;
1034         ptime = utime + cputime.stime;
1035         sum_sched_runtime = cputime.sum_exec_runtime;
1036         maxfire = 20;
1037         prof_expires = 0;
1038         while (!list_empty(timers)) {
1039                 struct cpu_timer_list *tl = list_first_entry(timers,
1040                                                       struct cpu_timer_list,
1041                                                       entry);
1042                 if (!--maxfire || ptime < tl->expires.cpu) {
1043                         prof_expires = tl->expires.cpu;
1044                         break;
1045                 }
1046                 tl->firing = 1;
1047                 list_move_tail(&tl->entry, firing);
1048         }
1049
1050         ++timers;
1051         maxfire = 20;
1052         virt_expires = 0;
1053         while (!list_empty(timers)) {
1054                 struct cpu_timer_list *tl = list_first_entry(timers,
1055                                                       struct cpu_timer_list,
1056                                                       entry);
1057                 if (!--maxfire || utime < tl->expires.cpu) {
1058                         virt_expires = tl->expires.cpu;
1059                         break;
1060                 }
1061                 tl->firing = 1;
1062                 list_move_tail(&tl->entry, firing);
1063         }
1064
1065         ++timers;
1066         maxfire = 20;
1067         sched_expires = 0;
1068         while (!list_empty(timers)) {
1069                 struct cpu_timer_list *tl = list_first_entry(timers,
1070                                                       struct cpu_timer_list,
1071                                                       entry);
1072                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1073                         sched_expires = tl->expires.sched;
1074                         break;
1075                 }
1076                 tl->firing = 1;
1077                 list_move_tail(&tl->entry, firing);
1078         }
1079
1080         /*
1081          * Check for the special case process timers.
1082          */
1083         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1084                          SIGPROF);
1085         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1086                          SIGVTALRM);
1087         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1088         if (soft != RLIM_INFINITY) {
1089                 unsigned long psecs = cputime_to_secs(ptime);
1090                 unsigned long hard =
1091                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1092                 cputime_t x;
1093                 if (psecs >= hard) {
1094                         /*
1095                          * At the hard limit, we just die.
1096                          * No need to calculate anything else now.
1097                          */
1098                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1099                         return;
1100                 }
1101                 if (psecs >= soft) {
1102                         /*
1103                          * At the soft limit, send a SIGXCPU every second.
1104                          */
1105                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1106                         if (soft < hard) {
1107                                 soft++;
1108                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1109                         }
1110                 }
1111                 x = secs_to_cputime(soft);
1112                 if (!prof_expires || x < prof_expires) {
1113                         prof_expires = x;
1114                 }
1115         }
1116
1117         sig->cputime_expires.prof_exp = prof_expires;
1118         sig->cputime_expires.virt_exp = virt_expires;
1119         sig->cputime_expires.sched_exp = sched_expires;
1120         if (task_cputime_zero(&sig->cputime_expires))
1121                 stop_process_timers(sig);
1122 }
1123
1124 /*
1125  * This is called from the signal code (via do_schedule_next_timer)
1126  * when the last timer signal was delivered and we have to reload the timer.
1127  */
1128 void posix_cpu_timer_schedule(struct k_itimer *timer)
1129 {
1130         struct task_struct *p = timer->it.cpu.task;
1131         union cpu_time_count now;
1132
1133         if (unlikely(p == NULL))
1134                 /*
1135                  * The task was cleaned up already, no future firings.
1136                  */
1137                 goto out;
1138
1139         /*
1140          * Fetch the current sample and update the timer's expiry time.
1141          */
1142         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1143                 cpu_clock_sample(timer->it_clock, p, &now);
1144                 bump_cpu_timer(timer, now);
1145                 if (unlikely(p->exit_state)) {
1146                         clear_dead_task(timer, now);
1147                         goto out;
1148                 }
1149                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1150                 spin_lock(&p->sighand->siglock);
1151         } else {
1152                 read_lock(&tasklist_lock);
1153                 if (unlikely(p->sighand == NULL)) {
1154                         /*
1155                          * The process has been reaped.
1156                          * We can't even collect a sample any more.
1157                          */
1158                         put_task_struct(p);
1159                         timer->it.cpu.task = p = NULL;
1160                         timer->it.cpu.expires.sched = 0;
1161                         goto out_unlock;
1162                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1163                         /*
1164                          * We've noticed that the thread is dead, but
1165                          * not yet reaped.  Take this opportunity to
1166                          * drop our task ref.
1167                          */
1168                         clear_dead_task(timer, now);
1169                         goto out_unlock;
1170                 }
1171                 spin_lock(&p->sighand->siglock);
1172                 cpu_timer_sample_group(timer->it_clock, p, &now);
1173                 bump_cpu_timer(timer, now);
1174                 /* Leave the tasklist_lock locked for the call below.  */
1175         }
1176
1177         /*
1178          * Now re-arm for the new expiry time.
1179          */
1180         BUG_ON(!irqs_disabled());
1181         arm_timer(timer);
1182         spin_unlock(&p->sighand->siglock);
1183
1184 out_unlock:
1185         read_unlock(&tasklist_lock);
1186
1187 out:
1188         timer->it_overrun_last = timer->it_overrun;
1189         timer->it_overrun = -1;
1190         ++timer->it_requeue_pending;
1191 }
1192
1193 /**
1194  * task_cputime_expired - Compare two task_cputime entities.
1195  *
1196  * @sample:     The task_cputime structure to be checked for expiration.
1197  * @expires:    Expiration times, against which @sample will be checked.
1198  *
1199  * Checks @sample against @expires to see if any field of @sample has expired.
1200  * Returns true if any field of the former is greater than the corresponding
1201  * field of the latter if the latter field is set.  Otherwise returns false.
1202  */
1203 static inline int task_cputime_expired(const struct task_cputime *sample,
1204                                         const struct task_cputime *expires)
1205 {
1206         if (expires->utime && sample->utime >= expires->utime)
1207                 return 1;
1208         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1209                 return 1;
1210         if (expires->sum_exec_runtime != 0 &&
1211             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1212                 return 1;
1213         return 0;
1214 }
1215
1216 /**
1217  * fastpath_timer_check - POSIX CPU timers fast path.
1218  *
1219  * @tsk:        The task (thread) being checked.
1220  *
1221  * Check the task and thread group timers.  If both are zero (there are no
1222  * timers set) return false.  Otherwise snapshot the task and thread group
1223  * timers and compare them with the corresponding expiration times.  Return
1224  * true if a timer has expired, else return false.
1225  */
1226 static inline int fastpath_timer_check(struct task_struct *tsk)
1227 {
1228         struct signal_struct *sig;
1229
1230         if (!task_cputime_zero(&tsk->cputime_expires)) {
1231                 struct task_cputime task_sample = {
1232                         .utime = tsk->utime,
1233                         .stime = tsk->stime,
1234                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1235                 };
1236
1237                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1238                         return 1;
1239         }
1240
1241         sig = tsk->signal;
1242         if (sig->cputimer.running) {
1243                 struct task_cputime group_sample;
1244
1245                 raw_spin_lock(&sig->cputimer.lock);
1246                 group_sample = sig->cputimer.cputime;
1247                 raw_spin_unlock(&sig->cputimer.lock);
1248
1249                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1250                         return 1;
1251         }
1252
1253         return 0;
1254 }
1255
1256 /*
1257  * This is called from the timer interrupt handler.  The irq handler has
1258  * already updated our counts.  We need to check if any timers fire now.
1259  * Interrupts are disabled.
1260  */
1261 void run_posix_cpu_timers(struct task_struct *tsk)
1262 {
1263         LIST_HEAD(firing);
1264         struct k_itimer *timer, *next;
1265         unsigned long flags;
1266
1267         BUG_ON(!irqs_disabled());
1268
1269         /*
1270          * The fast path checks that there are no expired thread or thread
1271          * group timers.  If that's so, just return.
1272          */
1273         if (!fastpath_timer_check(tsk))
1274                 return;
1275
1276         if (!lock_task_sighand(tsk, &flags))
1277                 return;
1278         /*
1279          * Here we take off tsk->signal->cpu_timers[N] and
1280          * tsk->cpu_timers[N] all the timers that are firing, and
1281          * put them on the firing list.
1282          */
1283         check_thread_timers(tsk, &firing);
1284         /*
1285          * If there are any active process wide timers (POSIX 1.b, itimers,
1286          * RLIMIT_CPU) cputimer must be running.
1287          */
1288         if (tsk->signal->cputimer.running)
1289                 check_process_timers(tsk, &firing);
1290
1291         /*
1292          * We must release these locks before taking any timer's lock.
1293          * There is a potential race with timer deletion here, as the
1294          * siglock now protects our private firing list.  We have set
1295          * the firing flag in each timer, so that a deletion attempt
1296          * that gets the timer lock before we do will give it up and
1297          * spin until we've taken care of that timer below.
1298          */
1299         unlock_task_sighand(tsk, &flags);
1300
1301         /*
1302          * Now that all the timers on our list have the firing flag,
1303          * no one will touch their list entries but us.  We'll take
1304          * each timer's lock before clearing its firing flag, so no
1305          * timer call will interfere.
1306          */
1307         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1308                 int cpu_firing;
1309
1310                 spin_lock(&timer->it_lock);
1311                 list_del_init(&timer->it.cpu.entry);
1312                 cpu_firing = timer->it.cpu.firing;
1313                 timer->it.cpu.firing = 0;
1314                 /*
1315                  * The firing flag is -1 if we collided with a reset
1316                  * of the timer, which already reported this
1317                  * almost-firing as an overrun.  So don't generate an event.
1318                  */
1319                 if (likely(cpu_firing >= 0))
1320                         cpu_timer_fire(timer);
1321                 spin_unlock(&timer->it_lock);
1322         }
1323 }
1324
1325 /*
1326  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1327  * The tsk->sighand->siglock must be held by the caller.
1328  */
1329 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1330                            cputime_t *newval, cputime_t *oldval)
1331 {
1332         union cpu_time_count now;
1333
1334         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1335         cpu_timer_sample_group(clock_idx, tsk, &now);
1336
1337         if (oldval) {
1338                 /*
1339                  * We are setting itimer. The *oldval is absolute and we update
1340                  * it to be relative, *newval argument is relative and we update
1341                  * it to be absolute.
1342                  */
1343                 if (*oldval) {
1344                         if (*oldval <= now.cpu) {
1345                                 /* Just about to fire. */
1346                                 *oldval = cputime_one_jiffy;
1347                         } else {
1348                                 *oldval -= now.cpu;
1349                         }
1350                 }
1351
1352                 if (!*newval)
1353                         return;
1354                 *newval += now.cpu;
1355         }
1356
1357         /*
1358          * Update expiration cache if we are the earliest timer, or eventually
1359          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1360          */
1361         switch (clock_idx) {
1362         case CPUCLOCK_PROF:
1363                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1364                         tsk->signal->cputime_expires.prof_exp = *newval;
1365                 break;
1366         case CPUCLOCK_VIRT:
1367                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1368                         tsk->signal->cputime_expires.virt_exp = *newval;
1369                 break;
1370         }
1371 }
1372
1373 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1374                             struct timespec *rqtp, struct itimerspec *it)
1375 {
1376         struct k_itimer timer;
1377         int error;
1378
1379         /*
1380          * Set up a temporary timer and then wait for it to go off.
1381          */
1382         memset(&timer, 0, sizeof timer);
1383         spin_lock_init(&timer.it_lock);
1384         timer.it_clock = which_clock;
1385         timer.it_overrun = -1;
1386         error = posix_cpu_timer_create(&timer);
1387         timer.it_process = current;
1388         if (!error) {
1389                 static struct itimerspec zero_it;
1390
1391                 memset(it, 0, sizeof *it);
1392                 it->it_value = *rqtp;
1393
1394                 spin_lock_irq(&timer.it_lock);
1395                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1396                 if (error) {
1397                         spin_unlock_irq(&timer.it_lock);
1398                         return error;
1399                 }
1400
1401                 while (!signal_pending(current)) {
1402                         if (timer.it.cpu.expires.sched == 0) {
1403                                 /*
1404                                  * Our timer fired and was reset.
1405                                  */
1406                                 spin_unlock_irq(&timer.it_lock);
1407                                 return 0;
1408                         }
1409
1410                         /*
1411                          * Block until cpu_timer_fire (or a signal) wakes us.
1412                          */
1413                         __set_current_state(TASK_INTERRUPTIBLE);
1414                         spin_unlock_irq(&timer.it_lock);
1415                         schedule();
1416                         spin_lock_irq(&timer.it_lock);
1417                 }
1418
1419                 /*
1420                  * We were interrupted by a signal.
1421                  */
1422                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1423                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1424                 spin_unlock_irq(&timer.it_lock);
1425
1426                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1427                         /*
1428                          * It actually did fire already.
1429                          */
1430                         return 0;
1431                 }
1432
1433                 error = -ERESTART_RESTARTBLOCK;
1434         }
1435
1436         return error;
1437 }
1438
1439 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1440
1441 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1442                             struct timespec *rqtp, struct timespec __user *rmtp)
1443 {
1444         struct restart_block *restart_block =
1445                 &current_thread_info()->restart_block;
1446         struct itimerspec it;
1447         int error;
1448
1449         /*
1450          * Diagnose required errors first.
1451          */
1452         if (CPUCLOCK_PERTHREAD(which_clock) &&
1453             (CPUCLOCK_PID(which_clock) == 0 ||
1454              CPUCLOCK_PID(which_clock) == current->pid))
1455                 return -EINVAL;
1456
1457         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1458
1459         if (error == -ERESTART_RESTARTBLOCK) {
1460
1461                 if (flags & TIMER_ABSTIME)
1462                         return -ERESTARTNOHAND;
1463                 /*
1464                  * Report back to the user the time still remaining.
1465                  */
1466                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1467                         return -EFAULT;
1468
1469                 restart_block->fn = posix_cpu_nsleep_restart;
1470                 restart_block->nanosleep.clockid = which_clock;
1471                 restart_block->nanosleep.rmtp = rmtp;
1472                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1473         }
1474         return error;
1475 }
1476
1477 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1478 {
1479         clockid_t which_clock = restart_block->nanosleep.clockid;
1480         struct timespec t;
1481         struct itimerspec it;
1482         int error;
1483
1484         t = ns_to_timespec(restart_block->nanosleep.expires);
1485
1486         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1487
1488         if (error == -ERESTART_RESTARTBLOCK) {
1489                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1490                 /*
1491                  * Report back to the user the time still remaining.
1492                  */
1493                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1494                         return -EFAULT;
1495
1496                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1497         }
1498         return error;
1499
1500 }
1501
1502 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1503 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1504
1505 static int process_cpu_clock_getres(const clockid_t which_clock,
1506                                     struct timespec *tp)
1507 {
1508         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1509 }
1510 static int process_cpu_clock_get(const clockid_t which_clock,
1511                                  struct timespec *tp)
1512 {
1513         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1514 }
1515 static int process_cpu_timer_create(struct k_itimer *timer)
1516 {
1517         timer->it_clock = PROCESS_CLOCK;
1518         return posix_cpu_timer_create(timer);
1519 }
1520 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1521                               struct timespec *rqtp,
1522                               struct timespec __user *rmtp)
1523 {
1524         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1525 }
1526 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1527 {
1528         return -EINVAL;
1529 }
1530 static int thread_cpu_clock_getres(const clockid_t which_clock,
1531                                    struct timespec *tp)
1532 {
1533         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1534 }
1535 static int thread_cpu_clock_get(const clockid_t which_clock,
1536                                 struct timespec *tp)
1537 {
1538         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1539 }
1540 static int thread_cpu_timer_create(struct k_itimer *timer)
1541 {
1542         timer->it_clock = THREAD_CLOCK;
1543         return posix_cpu_timer_create(timer);
1544 }
1545
1546 struct k_clock clock_posix_cpu = {
1547         .clock_getres   = posix_cpu_clock_getres,
1548         .clock_set      = posix_cpu_clock_set,
1549         .clock_get      = posix_cpu_clock_get,
1550         .timer_create   = posix_cpu_timer_create,
1551         .nsleep         = posix_cpu_nsleep,
1552         .nsleep_restart = posix_cpu_nsleep_restart,
1553         .timer_set      = posix_cpu_timer_set,
1554         .timer_del      = posix_cpu_timer_del,
1555         .timer_get      = posix_cpu_timer_get,
1556 };
1557
1558 static __init int init_posix_cpu_timers(void)
1559 {
1560         struct k_clock process = {
1561                 .clock_getres   = process_cpu_clock_getres,
1562                 .clock_get      = process_cpu_clock_get,
1563                 .timer_create   = process_cpu_timer_create,
1564                 .nsleep         = process_cpu_nsleep,
1565                 .nsleep_restart = process_cpu_nsleep_restart,
1566         };
1567         struct k_clock thread = {
1568                 .clock_getres   = thread_cpu_clock_getres,
1569                 .clock_get      = thread_cpu_clock_get,
1570                 .timer_create   = thread_cpu_timer_create,
1571         };
1572         struct timespec ts;
1573
1574         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1575         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1576
1577         cputime_to_timespec(cputime_one_jiffy, &ts);
1578         onecputick = ts.tv_nsec;
1579         WARN_ON(ts.tv_sec != 0);
1580
1581         return 0;
1582 }
1583 __initcall(init_posix_cpu_timers);