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         cputime_t utime, stime;
 159
 160         task_cputime(p, &utime, &stime);
 161
 162         return utime + stime;
 163 }
 164 static inline cputime_t virt_ticks(struct task_struct *p)
 165 {
 166         cputime_t utime;
 167
 168         task_cputime(p, &utime, NULL);
 169
 170         return utime;
 171 }
 172
 173 static int
 174 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 175 {
 176         int error = check_clock(which_clock);
 177         if (!error) {
 178                 tp->tv_sec = 0;
 179                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 180                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 181                         /*
 182                          * If sched_clock is using a cycle counter, we
 183                          * don't have any idea of its true resolution
 184                          * exported, but it is much more than 1s/HZ.
 185                          */
 186                         tp->tv_nsec = 1;
 187                 }
 188         }
 189         return error;
 190 }
 191
 192 static int
 193 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 194 {
 195         /*
 196          * You can never reset a CPU clock, but we check for other errors
 197          * in the call before failing with EPERM.
 198          */
 199         int error = check_clock(which_clock);
 200         if (error == 0) {
 201                 error = -EPERM;
 202         }
 203         return error;
 204 }
 205
 206
 207 /*
 208  * Sample a per-thread clock for the given task.
 209  */
 210 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 211                             union cpu_time_count *cpu)
 212 {
 213         switch (CPUCLOCK_WHICH(which_clock)) {
 214         default:
 215                 return -EINVAL;
 216         case CPUCLOCK_PROF:
 217                 cpu->cpu = prof_ticks(p);
 218                 break;
 219         case CPUCLOCK_VIRT:
 220                 cpu->cpu = virt_ticks(p);
 221                 break;
 222         case CPUCLOCK_SCHED:
 223                 cpu->sched = task_sched_runtime(p);
 224                 break;
 225         }
 226         return 0;
 227 }
 228
 229 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 230 {
 231         if (b->utime > a->utime)
 232                 a->utime = b->utime;
 233
 234         if (b->stime > a->stime)
 235                 a->stime = b->stime;
 236
 237         if (b->sum_exec_runtime > a->sum_exec_runtime)
 238                 a->sum_exec_runtime = b->sum_exec_runtime;
 239 }
 240
 241 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 242 {
 243         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 244         struct task_cputime sum;
 245         unsigned long flags;
 246
 247         if (!cputimer->running) {
 248                 /*
 249                  * The POSIX timer interface allows for absolute time expiry
 250                  * values through the TIMER_ABSTIME flag, therefore we have
 251                  * to synchronize the timer to the clock every time we start
 252                  * it.
 253                  */
 254                 thread_group_cputime(tsk, &sum);
 255                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 256                 cputimer->running = 1;
 257                 update_gt_cputime(&cputimer->cputime, &sum);
 258         } else
 259                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 260         *times = cputimer->cputime;
 261         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 262 }
 263
 264 /*
 265  * Sample a process (thread group) clock for the given group_leader task.
 266  * Must be called with tasklist_lock held for reading.
 267  */
 268 static int cpu_clock_sample_group(const clockid_t which_clock,
 269                                   struct task_struct *p,
 270                                   union cpu_time_count *cpu)
 271 {
 272         struct task_cputime cputime;
 273
 274         switch (CPUCLOCK_WHICH(which_clock)) {
 275         default:
 276                 return -EINVAL;
 277         case CPUCLOCK_PROF:
 278                 thread_group_cputime(p, &cputime);
 279                 cpu->cpu = cputime.utime + cputime.stime;
 280                 break;
 281         case CPUCLOCK_VIRT:
 282                 thread_group_cputime(p, &cputime);
 283                 cpu->cpu = cputime.utime;
 284                 break;
 285         case CPUCLOCK_SCHED:
 286                 thread_group_cputime(p, &cputime);
 287                 cpu->sched = cputime.sum_exec_runtime;
 288                 break;
 289         }
 290         return 0;
 291 }
 292
 293
 294 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 295 {
 296         const pid_t pid = CPUCLOCK_PID(which_clock);
 297         int error = -EINVAL;
 298         union cpu_time_count rtn;
 299
 300         if (pid == 0) {
 301                 /*
 302                  * Special case constant value for our own clocks.
 303                  * We don't have to do any lookup to find ourselves.
 304                  */
 305                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 306                         /*
 307                          * Sampling just ourselves we can do with no locking.
 308                          */
 309                         error = cpu_clock_sample(which_clock,
 310                                                  current, &rtn);
 311                 } else {
 312                         read_lock(&tasklist_lock);
 313                         error = cpu_clock_sample_group(which_clock,
 314                                                        current, &rtn);
 315                         read_unlock(&tasklist_lock);
 316                 }
 317         } else {
 318                 /*
 319                  * Find the given PID, and validate that the caller
 320                  * should be able to see it.
 321                  */
 322                 struct task_struct *p;
 323                 rcu_read_lock();
 324                 p = find_task_by_vpid(pid);
 325                 if (p) {
 326                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 327                                 if (same_thread_group(p, current)) {
 328                                         error = cpu_clock_sample(which_clock,
 329                                                                  p, &rtn);
 330                                 }
 331                         } else {
 332                                 read_lock(&tasklist_lock);
 333                                 if (thread_group_leader(p) && p->sighand) {
 334                                         error =
 335                                             cpu_clock_sample_group(which_clock,
 336                                                                    p, &rtn);
 337                                 }
 338                                 read_unlock(&tasklist_lock);
 339                         }
 340                 }
 341                 rcu_read_unlock();
 342         }
 343
 344         if (error)
 345                 return error;
 346         sample_to_timespec(which_clock, rtn, tp);
 347         return 0;
 348 }
 349
 350
 351 /*
 352  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 353  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 354  * new timer already all-zeros initialized.
 355  */
 356 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 357 {
 358         int ret = 0;
 359         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 360         struct task_struct *p;
 361
 362         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 363                 return -EINVAL;
 364
 365         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 366
 367         rcu_read_lock();
 368         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 369                 if (pid == 0) {
 370                         p = current;
 371                 } else {
 372                         p = find_task_by_vpid(pid);
 373                         if (p && !same_thread_group(p, current))
 374                                 p = NULL;
 375                 }
 376         } else {
 377                 if (pid == 0) {
 378                         p = current->group_leader;
 379                 } else {
 380                         p = find_task_by_vpid(pid);
 381                         if (p && !has_group_leader_pid(p))
 382                                 p = NULL;
 383                 }
 384         }
 385         new_timer->it.cpu.task = p;
 386         if (p) {
 387                 get_task_struct(p);
 388         } else {
 389                 ret = -EINVAL;
 390         }
 391         rcu_read_unlock();
 392
 393         return ret;
 394 }
 395
 396 /*
 397  * Clean up a CPU-clock timer that is about to be destroyed.
 398  * This is called from timer deletion with the timer already locked.
 399  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 400  * and try again.  (This happens when the timer is in the middle of firing.)
 401  */
 402 static int posix_cpu_timer_del(struct k_itimer *timer)
 403 {
 404         struct task_struct *p = timer->it.cpu.task;
 405         int ret = 0;
 406
 407         if (likely(p != NULL)) {
 408                 read_lock(&tasklist_lock);
 409                 if (unlikely(p->sighand == NULL)) {
 410                         /*
 411                          * We raced with the reaping of the task.
 412                          * The deletion should have cleared us off the list.
 413                          */
 414                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 415                 } else {
 416                         spin_lock(&p->sighand->siglock);
 417                         if (timer->it.cpu.firing)
 418                                 ret = TIMER_RETRY;
 419                         else
 420                                 list_del(&timer->it.cpu.entry);
 421                         spin_unlock(&p->sighand->siglock);
 422                 }
 423                 read_unlock(&tasklist_lock);
 424
 425                 if (!ret)
 426                         put_task_struct(p);
 427         }
 428
 429         return ret;
 430 }
 431
 432 /*
 433  * Clean out CPU timers still ticking when a thread exited.  The task
 434  * pointer is cleared, and the expiry time is replaced with the residual
 435  * time for later timer_gettime calls to return.
 436  * This must be called with the siglock held.
 437  */
 438 static void cleanup_timers(struct list_head *head,
 439                            cputime_t utime, cputime_t stime,
 440                            unsigned long long sum_exec_runtime)
 441 {
 442         struct cpu_timer_list *timer, *next;
 443         cputime_t ptime = utime + stime;
 444
 445         list_for_each_entry_safe(timer, next, head, entry) {
 446                 list_del_init(&timer->entry);
 447                 if (timer->expires.cpu < ptime) {
 448                         timer->expires.cpu = 0;
 449                 } else {
 450                         timer->expires.cpu -= ptime;
 451                 }
 452         }
 453
 454         ++head;
 455         list_for_each_entry_safe(timer, next, head, entry) {
 456                 list_del_init(&timer->entry);
 457                 if (timer->expires.cpu < utime) {
 458                         timer->expires.cpu = 0;
 459                 } else {
 460                         timer->expires.cpu -= utime;
 461                 }
 462         }
 463
 464         ++head;
 465         list_for_each_entry_safe(timer, next, head, entry) {
 466                 list_del_init(&timer->entry);
 467                 if (timer->expires.sched < sum_exec_runtime) {
 468                         timer->expires.sched = 0;
 469                 } else {
 470                         timer->expires.sched -= sum_exec_runtime;
 471                 }
 472         }
 473 }
 474
 475 /*
 476  * These are both called with the siglock held, when the current thread
 477  * is being reaped.  When the final (leader) thread in the group is reaped,
 478  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 479  */
 480 void posix_cpu_timers_exit(struct task_struct *tsk)
 481 {
 482         cputime_t utime, stime;
 483
 484         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 485                                                 sizeof(unsigned long long));
 486         task_cputime(tsk, &utime, &stime);
 487         cleanup_timers(tsk->cpu_timers,
 488                        utime, stime, tsk->se.sum_exec_runtime);
 489
 490 }
 491 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 492 {
 493         struct signal_struct *const sig = tsk->signal;
 494         cputime_t utime, stime;
 495
 496         task_cputime(tsk, &utime, &stime);
 497         cleanup_timers(tsk->signal->cpu_timers,
 498                        utime + sig->utime, stime + sig->stime,
 499                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 500 }
 501
 502 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 503 {
 504         /*
 505          * That's all for this thread or process.
 506          * We leave our residual in expires to be reported.
 507          */
 508         put_task_struct(timer->it.cpu.task);
 509         timer->it.cpu.task = NULL;
 510         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 511                                              timer->it.cpu.expires,
 512                                              now);
 513 }
 514
 515 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 516 {
 517         return expires == 0 || expires > new_exp;
 518 }
 519
 520 /*
 521  * Insert the timer on the appropriate list before any timers that
 522  * expire later.  This must be called with the tasklist_lock held
 523  * for reading, interrupts disabled and p->sighand->siglock taken.
 524  */
 525 static void arm_timer(struct k_itimer *timer)
 526 {
 527         struct task_struct *p = timer->it.cpu.task;
 528         struct list_head *head, *listpos;
 529         struct task_cputime *cputime_expires;
 530         struct cpu_timer_list *const nt = &timer->it.cpu;
 531         struct cpu_timer_list *next;
 532
 533         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 534                 head = p->cpu_timers;
 535                 cputime_expires = &p->cputime_expires;
 536         } else {
 537                 head = p->signal->cpu_timers;
 538                 cputime_expires = &p->signal->cputime_expires;
 539         }
 540         head += CPUCLOCK_WHICH(timer->it_clock);
 541
 542         listpos = head;
 543         list_for_each_entry(next, head, entry) {
 544                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
 545                         break;
 546                 listpos = &next->entry;
 547         }
 548         list_add(&nt->entry, listpos);
 549
 550         if (listpos == head) {
 551                 union cpu_time_count *exp = &nt->expires;
 552
 553                 /*
 554                  * We are the new earliest-expiring POSIX 1.b timer, hence
 555                  * need to update expiration cache. Take into account that
 556                  * for process timers we share expiration cache with itimers
 557                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 558                  */
 559
 560                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 561                 case CPUCLOCK_PROF:
 562                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
 563                                 cputime_expires->prof_exp = exp->cpu;
 564                         break;
 565                 case CPUCLOCK_VIRT:
 566                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
 567                                 cputime_expires->virt_exp = exp->cpu;
 568                         break;
 569                 case CPUCLOCK_SCHED:
 570                         if (cputime_expires->sched_exp == 0 ||
 571                             cputime_expires->sched_exp > exp->sched)
 572                                 cputime_expires->sched_exp = exp->sched;
 573                         break;
 574                 }
 575         }
 576 }
 577
 578 /*
 579  * The timer is locked, fire it and arrange for its reload.
 580  */
 581 static void cpu_timer_fire(struct k_itimer *timer)
 582 {
 583         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 584                 /*
 585                  * User don't want any signal.
 586                  */
 587                 timer->it.cpu.expires.sched = 0;
 588         } else if (unlikely(timer->sigq == NULL)) {
 589                 /*
 590                  * This a special case for clock_nanosleep,
 591                  * not a normal timer from sys_timer_create.
 592                  */
 593                 wake_up_process(timer->it_process);
 594                 timer->it.cpu.expires.sched = 0;
 595         } else if (timer->it.cpu.incr.sched == 0) {
 596                 /*
 597                  * One-shot timer.  Clear it as soon as it's fired.
 598                  */
 599                 posix_timer_event(timer, 0);
 600                 timer->it.cpu.expires.sched = 0;
 601         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 602                 /*
 603                  * The signal did not get queued because the signal
 604                  * was ignored, so we won't get any callback to
 605                  * reload the timer.  But we need to keep it
 606                  * ticking in case the signal is deliverable next time.
 607                  */
 608                 posix_cpu_timer_schedule(timer);
 609         }
 610 }
 611
 612 /*
 613  * Sample a process (thread group) timer for the given group_leader task.
 614  * Must be called with tasklist_lock held for reading.
 615  */
 616 static int cpu_timer_sample_group(const clockid_t which_clock,
 617                                   struct task_struct *p,
 618                                   union cpu_time_count *cpu)
 619 {
 620         struct task_cputime cputime;
 621
 622         thread_group_cputimer(p, &cputime);
 623         switch (CPUCLOCK_WHICH(which_clock)) {
 624         default:
 625                 return -EINVAL;
 626         case CPUCLOCK_PROF:
 627                 cpu->cpu = cputime.utime + cputime.stime;
 628                 break;
 629         case CPUCLOCK_VIRT:
 630                 cpu->cpu = cputime.utime;
 631                 break;
 632         case CPUCLOCK_SCHED:
 633                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 634                 break;
 635         }
 636         return 0;
 637 }
 638
 639 /*
 640  * Guts of sys_timer_settime for CPU timers.
 641  * This is called with the timer locked and interrupts disabled.
 642  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 643  * and try again.  (This happens when the timer is in the middle of firing.)
 644  */
 645 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 646                                struct itimerspec *new, struct itimerspec *old)
 647 {
 648         struct task_struct *p = timer->it.cpu.task;
 649         union cpu_time_count old_expires, new_expires, old_incr, val;
 650         int ret;
 651
 652         if (unlikely(p == NULL)) {
 653                 /*
 654                  * Timer refers to a dead task's clock.
 655                  */
 656                 return -ESRCH;
 657         }
 658
 659         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 660
 661         read_lock(&tasklist_lock);
 662         /*
 663          * We need the tasklist_lock to protect against reaping that
 664          * clears p->sighand.  If p has just been reaped, we can no
 665          * longer get any information about it at all.
 666          */
 667         if (unlikely(p->sighand == NULL)) {
 668                 read_unlock(&tasklist_lock);
 669                 put_task_struct(p);
 670                 timer->it.cpu.task = NULL;
 671                 return -ESRCH;
 672         }
 673
 674         /*
 675          * Disarm any old timer after extracting its expiry time.
 676          */
 677         BUG_ON(!irqs_disabled());
 678
 679         ret = 0;
 680         old_incr = timer->it.cpu.incr;
 681         spin_lock(&p->sighand->siglock);
 682         old_expires = timer->it.cpu.expires;
 683         if (unlikely(timer->it.cpu.firing)) {
 684                 timer->it.cpu.firing = -1;
 685                 ret = TIMER_RETRY;
 686         } else
 687                 list_del_init(&timer->it.cpu.entry);
 688
 689         /*
 690          * We need to sample the current value to convert the new
 691          * value from to relative and absolute, and to convert the
 692          * old value from absolute to relative.  To set a process
 693          * timer, we need a sample to balance the thread expiry
 694          * times (in arm_timer).  With an absolute time, we must
 695          * check if it's already passed.  In short, we need a sample.
 696          */
 697         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 698                 cpu_clock_sample(timer->it_clock, p, &val);
 699         } else {
 700                 cpu_timer_sample_group(timer->it_clock, p, &val);
 701         }
 702
 703         if (old) {
 704                 if (old_expires.sched == 0) {
 705                         old->it_value.tv_sec = 0;
 706                         old->it_value.tv_nsec = 0;
 707                 } else {
 708                         /*
 709                          * Update the timer in case it has
 710                          * overrun already.  If it has,
 711                          * we'll report it as having overrun
 712                          * and with the next reloaded timer
 713                          * already ticking, though we are
 714                          * swallowing that pending
 715                          * notification here to install the
 716                          * new setting.
 717                          */
 718                         bump_cpu_timer(timer, val);
 719                         if (cpu_time_before(timer->it_clock, val,
 720                                             timer->it.cpu.expires)) {
 721                                 old_expires = cpu_time_sub(
 722                                         timer->it_clock,
 723                                         timer->it.cpu.expires, val);
 724                                 sample_to_timespec(timer->it_clock,
 725                                                    old_expires,
 726                                                    &old->it_value);
 727                         } else {
 728                                 old->it_value.tv_nsec = 1;
 729                                 old->it_value.tv_sec = 0;
 730                         }
 731                 }
 732         }
 733
 734         if (unlikely(ret)) {
 735                 /*
 736                  * We are colliding with the timer actually firing.
 737                  * Punt after filling in the timer's old value, and
 738                  * disable this firing since we are already reporting
 739                  * it as an overrun (thanks to bump_cpu_timer above).
 740                  */
 741                 spin_unlock(&p->sighand->siglock);
 742                 read_unlock(&tasklist_lock);
 743                 goto out;
 744         }
 745
 746         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 747                 cpu_time_add(timer->it_clock, &new_expires, val);
 748         }
 749
 750         /*
 751          * Install the new expiry time (or zero).
 752          * For a timer with no notification action, we don't actually
 753          * arm the timer (we'll just fake it for timer_gettime).
 754          */
 755         timer->it.cpu.expires = new_expires;
 756         if (new_expires.sched != 0 &&
 757             cpu_time_before(timer->it_clock, val, new_expires)) {
 758                 arm_timer(timer);
 759         }
 760
 761         spin_unlock(&p->sighand->siglock);
 762         read_unlock(&tasklist_lock);
 763
 764         /*
 765          * Install the new reload setting, and
 766          * set up the signal and overrun bookkeeping.
 767          */
 768         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 769                                                 &new->it_interval);
 770
 771         /*
 772          * This acts as a modification timestamp for the timer,
 773          * so any automatic reload attempt will punt on seeing
 774          * that we have reset the timer manually.
 775          */
 776         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 777                 ~REQUEUE_PENDING;
 778         timer->it_overrun_last = 0;
 779         timer->it_overrun = -1;
 780
 781         if (new_expires.sched != 0 &&
 782             !cpu_time_before(timer->it_clock, val, new_expires)) {
 783                 /*
 784                  * The designated time already passed, so we notify
 785                  * immediately, even if the thread never runs to
 786                  * accumulate more time on this clock.
 787                  */
 788                 cpu_timer_fire(timer);
 789         }
 790
 791         ret = 0;
 792  out:
 793         if (old) {
 794                 sample_to_timespec(timer->it_clock,
 795                                    old_incr, &old->it_interval);
 796         }
 797         return ret;
 798 }
 799
 800 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 801 {
 802         union cpu_time_count now;
 803         struct task_struct *p = timer->it.cpu.task;
 804         int clear_dead;
 805
 806         /*
 807          * Easy part: convert the reload time.
 808          */
 809         sample_to_timespec(timer->it_clock,
 810                            timer->it.cpu.incr, &itp->it_interval);
 811
 812         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 813                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 814                 return;
 815         }
 816
 817         if (unlikely(p == NULL)) {
 818                 /*
 819                  * This task already died and the timer will never fire.
 820                  * In this case, expires is actually the dead value.
 821                  */
 822         dead:
 823                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 824                                    &itp->it_value);
 825                 return;
 826         }
 827
 828         /*
 829          * Sample the clock to take the difference with the expiry time.
 830          */
 831         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 832                 cpu_clock_sample(timer->it_clock, p, &now);
 833                 clear_dead = p->exit_state;
 834         } else {
 835                 read_lock(&tasklist_lock);
 836                 if (unlikely(p->sighand == NULL)) {
 837                         /*
 838                          * The process has been reaped.
 839                          * We can't even collect a sample any more.
 840                          * Call the timer disarmed, nothing else to do.
 841                          */
 842                         put_task_struct(p);
 843                         timer->it.cpu.task = NULL;
 844                         timer->it.cpu.expires.sched = 0;
 845                         read_unlock(&tasklist_lock);
 846                         goto dead;
 847                 } else {
 848                         cpu_timer_sample_group(timer->it_clock, p, &now);
 849                         clear_dead = (unlikely(p->exit_state) &&
 850                                       thread_group_empty(p));
 851                 }
 852                 read_unlock(&tasklist_lock);
 853         }
 854
 855         if (unlikely(clear_dead)) {
 856                 /*
 857                  * We've noticed that the thread is dead, but
 858                  * not yet reaped.  Take this opportunity to
 859                  * drop our task ref.
 860                  */
 861                 clear_dead_task(timer, now);
 862                 goto dead;
 863         }
 864
 865         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 866                 sample_to_timespec(timer->it_clock,
 867                                    cpu_time_sub(timer->it_clock,
 868                                                 timer->it.cpu.expires, now),
 869                                    &itp->it_value);
 870         } else {
 871                 /*
 872                  * The timer should have expired already, but the firing
 873                  * hasn't taken place yet.  Say it's just about to expire.
 874                  */
 875                 itp->it_value.tv_nsec = 1;
 876                 itp->it_value.tv_sec = 0;
 877         }
 878 }
 879
 880 /*
 881  * Check for any per-thread CPU timers that have fired and move them off
 882  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 883  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 884  */
 885 static void check_thread_timers(struct task_struct *tsk,
 886                                 struct list_head *firing)
 887 {
 888         int maxfire;
 889         struct list_head *timers = tsk->cpu_timers;
 890         struct signal_struct *const sig = tsk->signal;
 891         unsigned long soft;
 892
 893         maxfire = 20;
 894         tsk->cputime_expires.prof_exp = 0;
 895         while (!list_empty(timers)) {
 896                 struct cpu_timer_list *t = list_first_entry(timers,
 897                                                       struct cpu_timer_list,
 898                                                       entry);
 899                 if (!--maxfire || prof_ticks(tsk) < t->expires.cpu) {
 900                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 901                         break;
 902                 }
 903                 t->firing = 1;
 904                 list_move_tail(&t->entry, firing);
 905         }
 906
 907         ++timers;
 908         maxfire = 20;
 909         tsk->cputime_expires.virt_exp = 0;
 910         while (!list_empty(timers)) {
 911                 struct cpu_timer_list *t = list_first_entry(timers,
 912                                                       struct cpu_timer_list,
 913                                                       entry);
 914                 if (!--maxfire || virt_ticks(tsk) < t->expires.cpu) {
 915                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 916                         break;
 917                 }
 918                 t->firing = 1;
 919                 list_move_tail(&t->entry, firing);
 920         }
 921
 922         ++timers;
 923         maxfire = 20;
 924         tsk->cputime_expires.sched_exp = 0;
 925         while (!list_empty(timers)) {
 926                 struct cpu_timer_list *t = list_first_entry(timers,
 927                                                       struct cpu_timer_list,
 928                                                       entry);
 929                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
 930                         tsk->cputime_expires.sched_exp = t->expires.sched;
 931                         break;
 932                 }
 933                 t->firing = 1;
 934                 list_move_tail(&t->entry, firing);
 935         }
 936
 937         /*
 938          * Check for the special case thread timers.
 939          */
 940         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 941         if (soft != RLIM_INFINITY) {
 942                 unsigned long hard =
 943                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 944
 945                 if (hard != RLIM_INFINITY &&
 946                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 947                         /*
 948                          * At the hard limit, we just die.
 949                          * No need to calculate anything else now.
 950                          */
 951                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 952                         return;
 953                 }
 954                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 955                         /*
 956                          * At the soft limit, send a SIGXCPU every second.
 957                          */
 958                         if (soft < hard) {
 959                                 soft += USEC_PER_SEC;
 960                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 961                         }
 962                         printk(KERN_INFO
 963                                 "RT Watchdog Timeout: %s[%d]\n",
 964                                 tsk->comm, task_pid_nr(tsk));
 965                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 966                 }
 967         }
 968 }
 969
 970 static void stop_process_timers(struct signal_struct *sig)
 971 {
 972         struct thread_group_cputimer *cputimer = &sig->cputimer;
 973         unsigned long flags;
 974
 975         raw_spin_lock_irqsave(&cputimer->lock, flags);
 976         cputimer->running = 0;
 977         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 978 }
 979
 980 static u32 onecputick;
 981
 982 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 983                              cputime_t *expires, cputime_t cur_time, int signo)
 984 {
 985         if (!it->expires)
 986                 return;
 987
 988         if (cur_time >= it->expires) {
 989                 if (it->incr) {
 990                         it->expires += it->incr;
 991                         it->error += it->incr_error;
 992                         if (it->error >= onecputick) {
 993                                 it->expires -= cputime_one_jiffy;
 994                                 it->error -= onecputick;
 995                         }
 996                 } else {
 997                         it->expires = 0;
 998                 }
 999
1000                 trace_itimer_expire(signo == SIGPROF ?
1001                                     ITIMER_PROF : ITIMER_VIRTUAL,
1002                                     tsk->signal->leader_pid, cur_time);
1003                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1004         }
1005
1006         if (it->expires && (!*expires || it->expires < *expires)) {
1007                 *expires = it->expires;
1008         }
1009 }
1010
1011 /**
1012  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1013  *
1014  * @cputime:    The struct to compare.
1015  *
1016  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1017  * are zero, false if any field is nonzero.
1018  */
1019 static inline int task_cputime_zero(const struct task_cputime *cputime)
1020 {
1021         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
1022                 return 1;
1023         return 0;
1024 }
1025
1026 /*
1027  * Check for any per-thread CPU timers that have fired and move them
1028  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1029  * have already been taken off.
1030  */
1031 static void check_process_timers(struct task_struct *tsk,
1032                                  struct list_head *firing)
1033 {
1034         int maxfire;
1035         struct signal_struct *const sig = tsk->signal;
1036         cputime_t utime, ptime, virt_expires, prof_expires;
1037         unsigned long long sum_sched_runtime, sched_expires;
1038         struct list_head *timers = sig->cpu_timers;
1039         struct task_cputime cputime;
1040         unsigned long soft;
1041
1042         /*
1043          * Collect the current process totals.
1044          */
1045         thread_group_cputimer(tsk, &cputime);
1046         utime = cputime.utime;
1047         ptime = utime + cputime.stime;
1048         sum_sched_runtime = cputime.sum_exec_runtime;
1049         maxfire = 20;
1050         prof_expires = 0;
1051         while (!list_empty(timers)) {
1052                 struct cpu_timer_list *tl = list_first_entry(timers,
1053                                                       struct cpu_timer_list,
1054                                                       entry);
1055                 if (!--maxfire || ptime < tl->expires.cpu) {
1056                         prof_expires = tl->expires.cpu;
1057                         break;
1058                 }
1059                 tl->firing = 1;
1060                 list_move_tail(&tl->entry, firing);
1061         }
1062
1063         ++timers;
1064         maxfire = 20;
1065         virt_expires = 0;
1066         while (!list_empty(timers)) {
1067                 struct cpu_timer_list *tl = list_first_entry(timers,
1068                                                       struct cpu_timer_list,
1069                                                       entry);
1070                 if (!--maxfire || utime < tl->expires.cpu) {
1071                         virt_expires = tl->expires.cpu;
1072                         break;
1073                 }
1074                 tl->firing = 1;
1075                 list_move_tail(&tl->entry, firing);
1076         }
1077
1078         ++timers;
1079         maxfire = 20;
1080         sched_expires = 0;
1081         while (!list_empty(timers)) {
1082                 struct cpu_timer_list *tl = list_first_entry(timers,
1083                                                       struct cpu_timer_list,
1084                                                       entry);
1085                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1086                         sched_expires = tl->expires.sched;
1087                         break;
1088                 }
1089                 tl->firing = 1;
1090                 list_move_tail(&tl->entry, firing);
1091         }
1092
1093         /*
1094          * Check for the special case process timers.
1095          */
1096         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1097                          SIGPROF);
1098         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1099                          SIGVTALRM);
1100         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1101         if (soft != RLIM_INFINITY) {
1102                 unsigned long psecs = cputime_to_secs(ptime);
1103                 unsigned long hard =
1104                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1105                 cputime_t x;
1106                 if (psecs >= hard) {
1107                         /*
1108                          * At the hard limit, we just die.
1109                          * No need to calculate anything else now.
1110                          */
1111                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1112                         return;
1113                 }
1114                 if (psecs >= soft) {
1115                         /*
1116                          * At the soft limit, send a SIGXCPU every second.
1117                          */
1118                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1119                         if (soft < hard) {
1120                                 soft++;
1121                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1122                         }
1123                 }
1124                 x = secs_to_cputime(soft);
1125                 if (!prof_expires || x < prof_expires) {
1126                         prof_expires = x;
1127                 }
1128         }
1129
1130         sig->cputime_expires.prof_exp = prof_expires;
1131         sig->cputime_expires.virt_exp = virt_expires;
1132         sig->cputime_expires.sched_exp = sched_expires;
1133         if (task_cputime_zero(&sig->cputime_expires))
1134                 stop_process_timers(sig);
1135 }
1136
1137 /*
1138  * This is called from the signal code (via do_schedule_next_timer)
1139  * when the last timer signal was delivered and we have to reload the timer.
1140  */
1141 void posix_cpu_timer_schedule(struct k_itimer *timer)
1142 {
1143         struct task_struct *p = timer->it.cpu.task;
1144         union cpu_time_count now;
1145
1146         if (unlikely(p == NULL))
1147                 /*
1148                  * The task was cleaned up already, no future firings.
1149                  */
1150                 goto out;
1151
1152         /*
1153          * Fetch the current sample and update the timer's expiry time.
1154          */
1155         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1156                 cpu_clock_sample(timer->it_clock, p, &now);
1157                 bump_cpu_timer(timer, now);
1158                 if (unlikely(p->exit_state)) {
1159                         clear_dead_task(timer, now);
1160                         goto out;
1161                 }
1162                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1163                 spin_lock(&p->sighand->siglock);
1164         } else {
1165                 read_lock(&tasklist_lock);
1166                 if (unlikely(p->sighand == NULL)) {
1167                         /*
1168                          * The process has been reaped.
1169                          * We can't even collect a sample any more.
1170                          */
1171                         put_task_struct(p);
1172                         timer->it.cpu.task = p = NULL;
1173                         timer->it.cpu.expires.sched = 0;
1174                         goto out_unlock;
1175                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1176                         /*
1177                          * We've noticed that the thread is dead, but
1178                          * not yet reaped.  Take this opportunity to
1179                          * drop our task ref.
1180                          */
1181                         clear_dead_task(timer, now);
1182                         goto out_unlock;
1183                 }
1184                 spin_lock(&p->sighand->siglock);
1185                 cpu_timer_sample_group(timer->it_clock, p, &now);
1186                 bump_cpu_timer(timer, now);
1187                 /* Leave the tasklist_lock locked for the call below.  */
1188         }
1189
1190         /*
1191          * Now re-arm for the new expiry time.
1192          */
1193         BUG_ON(!irqs_disabled());
1194         arm_timer(timer);
1195         spin_unlock(&p->sighand->siglock);
1196
1197 out_unlock:
1198         read_unlock(&tasklist_lock);
1199
1200 out:
1201         timer->it_overrun_last = timer->it_overrun;
1202         timer->it_overrun = -1;
1203         ++timer->it_requeue_pending;
1204 }
1205
1206 /**
1207  * task_cputime_expired - Compare two task_cputime entities.
1208  *
1209  * @sample:     The task_cputime structure to be checked for expiration.
1210  * @expires:    Expiration times, against which @sample will be checked.
1211  *
1212  * Checks @sample against @expires to see if any field of @sample has expired.
1213  * Returns true if any field of the former is greater than the corresponding
1214  * field of the latter if the latter field is set.  Otherwise returns false.
1215  */
1216 static inline int task_cputime_expired(const struct task_cputime *sample,
1217                                         const struct task_cputime *expires)
1218 {
1219         if (expires->utime && sample->utime >= expires->utime)
1220                 return 1;
1221         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1222                 return 1;
1223         if (expires->sum_exec_runtime != 0 &&
1224             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1225                 return 1;
1226         return 0;
1227 }
1228
1229 /**
1230  * fastpath_timer_check - POSIX CPU timers fast path.
1231  *
1232  * @tsk:        The task (thread) being checked.
1233  *
1234  * Check the task and thread group timers.  If both are zero (there are no
1235  * timers set) return false.  Otherwise snapshot the task and thread group
1236  * timers and compare them with the corresponding expiration times.  Return
1237  * true if a timer has expired, else return false.
1238  */
1239 static inline int fastpath_timer_check(struct task_struct *tsk)
1240 {
1241         struct signal_struct *sig;
1242         cputime_t utime, stime;
1243
1244         task_cputime(tsk, &utime, &stime);
1245
1246         if (!task_cputime_zero(&tsk->cputime_expires)) {
1247                 struct task_cputime task_sample = {
1248                         .utime = utime,
1249                         .stime = stime,
1250                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1251                 };
1252
1253                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1254                         return 1;
1255         }
1256
1257         sig = tsk->signal;
1258         if (sig->cputimer.running) {
1259                 struct task_cputime group_sample;
1260
1261                 raw_spin_lock(&sig->cputimer.lock);
1262                 group_sample = sig->cputimer.cputime;
1263                 raw_spin_unlock(&sig->cputimer.lock);
1264
1265                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1266                         return 1;
1267         }
1268
1269         return 0;
1270 }
1271
1272 /*
1273  * This is called from the timer interrupt handler.  The irq handler has
1274  * already updated our counts.  We need to check if any timers fire now.
1275  * Interrupts are disabled.
1276  */
1277 void run_posix_cpu_timers(struct task_struct *tsk)
1278 {
1279         LIST_HEAD(firing);
1280         struct k_itimer *timer, *next;
1281         unsigned long flags;
1282
1283         BUG_ON(!irqs_disabled());
1284
1285         /*
1286          * The fast path checks that there are no expired thread or thread
1287          * group timers.  If that's so, just return.
1288          */
1289         if (!fastpath_timer_check(tsk))
1290                 return;
1291
1292         if (!lock_task_sighand(tsk, &flags))
1293                 return;
1294         /*
1295          * Here we take off tsk->signal->cpu_timers[N] and
1296          * tsk->cpu_timers[N] all the timers that are firing, and
1297          * put them on the firing list.
1298          */
1299         check_thread_timers(tsk, &firing);
1300         /*
1301          * If there are any active process wide timers (POSIX 1.b, itimers,
1302          * RLIMIT_CPU) cputimer must be running.
1303          */
1304         if (tsk->signal->cputimer.running)
1305                 check_process_timers(tsk, &firing);
1306
1307         /*
1308          * We must release these locks before taking any timer's lock.
1309          * There is a potential race with timer deletion here, as the
1310          * siglock now protects our private firing list.  We have set
1311          * the firing flag in each timer, so that a deletion attempt
1312          * that gets the timer lock before we do will give it up and
1313          * spin until we've taken care of that timer below.
1314          */
1315         unlock_task_sighand(tsk, &flags);
1316
1317         /*
1318          * Now that all the timers on our list have the firing flag,
1319          * no one will touch their list entries but us.  We'll take
1320          * each timer's lock before clearing its firing flag, so no
1321          * timer call will interfere.
1322          */
1323         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1324                 int cpu_firing;
1325
1326                 spin_lock(&timer->it_lock);
1327                 list_del_init(&timer->it.cpu.entry);
1328                 cpu_firing = timer->it.cpu.firing;
1329                 timer->it.cpu.firing = 0;
1330                 /*
1331                  * The firing flag is -1 if we collided with a reset
1332                  * of the timer, which already reported this
1333                  * almost-firing as an overrun.  So don't generate an event.
1334                  */
1335                 if (likely(cpu_firing >= 0))
1336                         cpu_timer_fire(timer);
1337                 spin_unlock(&timer->it_lock);
1338         }
1339 }
1340
1341 /*
1342  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1343  * The tsk->sighand->siglock must be held by the caller.
1344  */
1345 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1346                            cputime_t *newval, cputime_t *oldval)
1347 {
1348         union cpu_time_count now;
1349
1350         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1351         cpu_timer_sample_group(clock_idx, tsk, &now);
1352
1353         if (oldval) {
1354                 /*
1355                  * We are setting itimer. The *oldval is absolute and we update
1356                  * it to be relative, *newval argument is relative and we update
1357                  * it to be absolute.
1358                  */
1359                 if (*oldval) {
1360                         if (*oldval <= now.cpu) {
1361                                 /* Just about to fire. */
1362                                 *oldval = cputime_one_jiffy;
1363                         } else {
1364                                 *oldval -= now.cpu;
1365                         }
1366                 }
1367
1368                 if (!*newval)
1369                         return;
1370                 *newval += now.cpu;
1371         }
1372
1373         /*
1374          * Update expiration cache if we are the earliest timer, or eventually
1375          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1376          */
1377         switch (clock_idx) {
1378         case CPUCLOCK_PROF:
1379                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1380                         tsk->signal->cputime_expires.prof_exp = *newval;
1381                 break;
1382         case CPUCLOCK_VIRT:
1383                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1384                         tsk->signal->cputime_expires.virt_exp = *newval;
1385                 break;
1386         }
1387 }
1388
1389 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1390                             struct timespec *rqtp, struct itimerspec *it)
1391 {
1392         struct k_itimer timer;
1393         int error;
1394
1395         /*
1396          * Set up a temporary timer and then wait for it to go off.
1397          */
1398         memset(&timer, 0, sizeof timer);
1399         spin_lock_init(&timer.it_lock);
1400         timer.it_clock = which_clock;
1401         timer.it_overrun = -1;
1402         error = posix_cpu_timer_create(&timer);
1403         timer.it_process = current;
1404         if (!error) {
1405                 static struct itimerspec zero_it;
1406
1407                 memset(it, 0, sizeof *it);
1408                 it->it_value = *rqtp;
1409
1410                 spin_lock_irq(&timer.it_lock);
1411                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1412                 if (error) {
1413                         spin_unlock_irq(&timer.it_lock);
1414                         return error;
1415                 }
1416
1417                 while (!signal_pending(current)) {
1418                         if (timer.it.cpu.expires.sched == 0) {
1419                                 /*
1420                                  * Our timer fired and was reset, below
1421                                  * deletion can not fail.
1422                                  */
1423                                 posix_cpu_timer_del(&timer);
1424                                 spin_unlock_irq(&timer.it_lock);
1425                                 return 0;
1426                         }
1427
1428                         /*
1429                          * Block until cpu_timer_fire (or a signal) wakes us.
1430                          */
1431                         __set_current_state(TASK_INTERRUPTIBLE);
1432                         spin_unlock_irq(&timer.it_lock);
1433                         schedule();
1434                         spin_lock_irq(&timer.it_lock);
1435                 }
1436
1437                 /*
1438                  * We were interrupted by a signal.
1439                  */
1440                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1441                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1442                 if (!error) {
1443                         /*
1444                          * Timer is now unarmed, deletion can not fail.
1445                          */
1446                         posix_cpu_timer_del(&timer);
1447                 }
1448                 spin_unlock_irq(&timer.it_lock);
1449
1450                 while (error == TIMER_RETRY) {
1451                         /*
1452                          * We need to handle case when timer was or is in the
1453                          * middle of firing. In other cases we already freed
1454                          * resources.
1455                          */
1456                         spin_lock_irq(&timer.it_lock);
1457                         error = posix_cpu_timer_del(&timer);
1458                         spin_unlock_irq(&timer.it_lock);
1459                 }
1460
1461                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1462                         /*
1463                          * It actually did fire already.
1464                          */
1465                         return 0;
1466                 }
1467
1468                 error = -ERESTART_RESTARTBLOCK;
1469         }
1470
1471         return error;
1472 }
1473
1474 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1475
1476 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1477                             struct timespec *rqtp, struct timespec __user *rmtp)
1478 {
1479         struct restart_block *restart_block =
1480                 &current_thread_info()->restart_block;
1481         struct itimerspec it;
1482         int error;
1483
1484         /*
1485          * Diagnose required errors first.
1486          */
1487         if (CPUCLOCK_PERTHREAD(which_clock) &&
1488             (CPUCLOCK_PID(which_clock) == 0 ||
1489              CPUCLOCK_PID(which_clock) == current->pid))
1490                 return -EINVAL;
1491
1492         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1493
1494         if (error == -ERESTART_RESTARTBLOCK) {
1495
1496                 if (flags & TIMER_ABSTIME)
1497                         return -ERESTARTNOHAND;
1498                 /*
1499                  * Report back to the user the time still remaining.
1500                  */
1501                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1502                         return -EFAULT;
1503
1504                 restart_block->fn = posix_cpu_nsleep_restart;
1505                 restart_block->nanosleep.clockid = which_clock;
1506                 restart_block->nanosleep.rmtp = rmtp;
1507                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1508         }
1509         return error;
1510 }
1511
1512 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1513 {
1514         clockid_t which_clock = restart_block->nanosleep.clockid;
1515         struct timespec t;
1516         struct itimerspec it;
1517         int error;
1518
1519         t = ns_to_timespec(restart_block->nanosleep.expires);
1520
1521         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1522
1523         if (error == -ERESTART_RESTARTBLOCK) {
1524                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1525                 /*
1526                  * Report back to the user the time still remaining.
1527                  */
1528                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1529                         return -EFAULT;
1530
1531                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1532         }
1533         return error;
1534
1535 }
1536
1537 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1538 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1539
1540 static int process_cpu_clock_getres(const clockid_t which_clock,
1541                                     struct timespec *tp)
1542 {
1543         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1544 }
1545 static int process_cpu_clock_get(const clockid_t which_clock,
1546                                  struct timespec *tp)
1547 {
1548         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1549 }
1550 static int process_cpu_timer_create(struct k_itimer *timer)
1551 {
1552         timer->it_clock = PROCESS_CLOCK;
1553         return posix_cpu_timer_create(timer);
1554 }
1555 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1556                               struct timespec *rqtp,
1557                               struct timespec __user *rmtp)
1558 {
1559         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1560 }
1561 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1562 {
1563         return -EINVAL;
1564 }
1565 static int thread_cpu_clock_getres(const clockid_t which_clock,
1566                                    struct timespec *tp)
1567 {
1568         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1569 }
1570 static int thread_cpu_clock_get(const clockid_t which_clock,
1571                                 struct timespec *tp)
1572 {
1573         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1574 }
1575 static int thread_cpu_timer_create(struct k_itimer *timer)
1576 {
1577         timer->it_clock = THREAD_CLOCK;
1578         return posix_cpu_timer_create(timer);
1579 }
1580
1581 struct k_clock clock_posix_cpu = {
1582         .clock_getres   = posix_cpu_clock_getres,
1583         .clock_set      = posix_cpu_clock_set,
1584         .clock_get      = posix_cpu_clock_get,
1585         .timer_create   = posix_cpu_timer_create,
1586         .nsleep         = posix_cpu_nsleep,
1587         .nsleep_restart = posix_cpu_nsleep_restart,
1588         .timer_set      = posix_cpu_timer_set,
1589         .timer_del      = posix_cpu_timer_del,
1590         .timer_get      = posix_cpu_timer_get,
1591 };
1592
1593 static __init int init_posix_cpu_timers(void)
1594 {
1595         struct k_clock process = {
1596                 .clock_getres   = process_cpu_clock_getres,
1597                 .clock_get      = process_cpu_clock_get,
1598                 .timer_create   = process_cpu_timer_create,
1599                 .nsleep         = process_cpu_nsleep,
1600                 .nsleep_restart = process_cpu_nsleep_restart,
1601         };
1602         struct k_clock thread = {
1603                 .clock_getres   = thread_cpu_clock_getres,
1604                 .clock_get      = thread_cpu_clock_get,
1605                 .timer_create   = thread_cpu_timer_create,
1606         };
1607         struct timespec ts;
1608
1609         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1610         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1611
1612         cputime_to_timespec(cputime_one_jiffy, &ts);
1613         onecputick = ts.tv_nsec;
1614         WARN_ON(ts.tv_sec != 0);
1615
1616         return 0;
1617 }
1618 __initcall(init_posix_cpu_timers);