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