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
  13 /*
  14  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
  15  */
  16 void update_rlimit_cpu(unsigned long rlim_new)
  17 {
  18         cputime_t cputime = secs_to_cputime(rlim_new);
  19         struct signal_struct *const sig = current->signal;
  20
  21         if (cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) ||
  22             cputime_gt(sig->it[CPUCLOCK_PROF].expires, cputime)) {
  23                 spin_lock_irq(&current->sighand->siglock);
  24                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
  25                 spin_unlock_irq(&current->sighand->siglock);
  26         }
  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         read_lock(&tasklist_lock);
  42         p = find_task_by_vpid(pid);
  43         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  44                    same_thread_group(p, current) : thread_group_leader(p))) {
  45                 error = -EINVAL;
  46         }
  47         read_unlock(&tasklist_lock);
  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 cputime_lt(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 = cputime_add(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 = cputime_sub(a.cpu, b.cpu);
 103         }
 104         return a;
 105 }
 106
 107 /*
 108  * Divide and limit the result to res >= 1
 109  *
 110  * This is necessary to prevent signal delivery starvation, when the result of
 111  * the division would be rounded down to 0.
 112  */
 113 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
 114 {
 115         cputime_t res = cputime_div(time, div);
 116
 117         return max_t(cputime_t, res, 1);
 118 }
 119
 120 /*
 121  * Update expiry time from increment, and increase overrun count,
 122  * given the current clock sample.
 123  */
 124 static void bump_cpu_timer(struct k_itimer *timer,
 125                                   union cpu_time_count now)
 126 {
 127         int i;
 128
 129         if (timer->it.cpu.incr.sched == 0)
 130                 return;
 131
 132         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 133                 unsigned long long delta, incr;
 134
 135                 if (now.sched < timer->it.cpu.expires.sched)
 136                         return;
 137                 incr = timer->it.cpu.incr.sched;
 138                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 139                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 140                 for (i = 0; incr < delta - incr; i++)
 141                         incr = incr << 1;
 142                 for (; i >= 0; incr >>= 1, i--) {
 143                         if (delta < incr)
 144                                 continue;
 145                         timer->it.cpu.expires.sched += incr;
 146                         timer->it_overrun += 1 << i;
 147                         delta -= incr;
 148                 }
 149         } else {
 150                 cputime_t delta, incr;
 151
 152                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 153                         return;
 154                 incr = timer->it.cpu.incr.cpu;
 155                 delta = cputime_sub(cputime_add(now.cpu, incr),
 156                                     timer->it.cpu.expires.cpu);
 157                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 158                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 159                              incr = cputime_add(incr, incr);
 160                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 161                         if (cputime_lt(delta, incr))
 162                                 continue;
 163                         timer->it.cpu.expires.cpu =
 164                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 165                         timer->it_overrun += 1 << i;
 166                         delta = cputime_sub(delta, incr);
 167                 }
 168         }
 169 }
 170
 171 static inline cputime_t prof_ticks(struct task_struct *p)
 172 {
 173         return cputime_add(p->utime, p->stime);
 174 }
 175 static inline cputime_t virt_ticks(struct task_struct *p)
 176 {
 177         return p->utime;
 178 }
 179
 180 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 181 {
 182         int error = check_clock(which_clock);
 183         if (!error) {
 184                 tp->tv_sec = 0;
 185                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 186                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 187                         /*
 188                          * If sched_clock is using a cycle counter, we
 189                          * don't have any idea of its true resolution
 190                          * exported, but it is much more than 1s/HZ.
 191                          */
 192                         tp->tv_nsec = 1;
 193                 }
 194         }
 195         return error;
 196 }
 197
 198 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 199 {
 200         /*
 201          * You can never reset a CPU clock, but we check for other errors
 202          * in the call before failing with EPERM.
 203          */
 204         int error = check_clock(which_clock);
 205         if (error == 0) {
 206                 error = -EPERM;
 207         }
 208         return error;
 209 }
 210
 211
 212 /*
 213  * Sample a per-thread clock for the given task.
 214  */
 215 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 216                             union cpu_time_count *cpu)
 217 {
 218         switch (CPUCLOCK_WHICH(which_clock)) {
 219         default:
 220                 return -EINVAL;
 221         case CPUCLOCK_PROF:
 222                 cpu->cpu = prof_ticks(p);
 223                 break;
 224         case CPUCLOCK_VIRT:
 225                 cpu->cpu = virt_ticks(p);
 226                 break;
 227         case CPUCLOCK_SCHED:
 228                 cpu->sched = task_sched_runtime(p);
 229                 break;
 230         }
 231         return 0;
 232 }
 233
 234 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
 235 {
 236         struct sighand_struct *sighand;
 237         struct signal_struct *sig;
 238         struct task_struct *t;
 239
 240         *times = INIT_CPUTIME;
 241
 242         rcu_read_lock();
 243         sighand = rcu_dereference(tsk->sighand);
 244         if (!sighand)
 245                 goto out;
 246
 247         sig = tsk->signal;
 248
 249         t = tsk;
 250         do {
 251                 times->utime = cputime_add(times->utime, t->utime);
 252                 times->stime = cputime_add(times->stime, t->stime);
 253                 times->sum_exec_runtime += t->se.sum_exec_runtime;
 254
 255                 t = next_thread(t);
 256         } while (t != tsk);
 257
 258         times->utime = cputime_add(times->utime, sig->utime);
 259         times->stime = cputime_add(times->stime, sig->stime);
 260         times->sum_exec_runtime += sig->sum_sched_runtime;
 261 out:
 262         rcu_read_unlock();
 263 }
 264
 265 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 266 {
 267         if (cputime_gt(b->utime, a->utime))
 268                 a->utime = b->utime;
 269
 270         if (cputime_gt(b->stime, a->stime))
 271                 a->stime = b->stime;
 272
 273         if (b->sum_exec_runtime > a->sum_exec_runtime)
 274                 a->sum_exec_runtime = b->sum_exec_runtime;
 275 }
 276
 277 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 278 {
 279         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 280         struct task_cputime sum;
 281         unsigned long flags;
 282
 283         spin_lock_irqsave(&cputimer->lock, flags);
 284         if (!cputimer->running) {
 285                 cputimer->running = 1;
 286                 /*
 287                  * The POSIX timer interface allows for absolute time expiry
 288                  * values through the TIMER_ABSTIME flag, therefore we have
 289                  * to synchronize the timer to the clock every time we start
 290                  * it.
 291                  */
 292                 thread_group_cputime(tsk, &sum);
 293                 update_gt_cputime(&cputimer->cputime, &sum);
 294         }
 295         *times = cputimer->cputime;
 296         spin_unlock_irqrestore(&cputimer->lock, flags);
 297 }
 298
 299 /*
 300  * Sample a process (thread group) clock for the given group_leader task.
 301  * Must be called with tasklist_lock held for reading.
 302  */
 303 static int cpu_clock_sample_group(const clockid_t which_clock,
 304                                   struct task_struct *p,
 305                                   union cpu_time_count *cpu)
 306 {
 307         struct task_cputime cputime;
 308
 309         switch (CPUCLOCK_WHICH(which_clock)) {
 310         default:
 311                 return -EINVAL;
 312         case CPUCLOCK_PROF:
 313                 thread_group_cputime(p, &cputime);
 314                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 315                 break;
 316         case CPUCLOCK_VIRT:
 317                 thread_group_cputime(p, &cputime);
 318                 cpu->cpu = cputime.utime;
 319                 break;
 320         case CPUCLOCK_SCHED:
 321                 cpu->sched = thread_group_sched_runtime(p);
 322                 break;
 323         }
 324         return 0;
 325 }
 326
 327
 328 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 329 {
 330         const pid_t pid = CPUCLOCK_PID(which_clock);
 331         int error = -EINVAL;
 332         union cpu_time_count rtn;
 333
 334         if (pid == 0) {
 335                 /*
 336                  * Special case constant value for our own clocks.
 337                  * We don't have to do any lookup to find ourselves.
 338                  */
 339                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 340                         /*
 341                          * Sampling just ourselves we can do with no locking.
 342                          */
 343                         error = cpu_clock_sample(which_clock,
 344                                                  current, &rtn);
 345                 } else {
 346                         read_lock(&tasklist_lock);
 347                         error = cpu_clock_sample_group(which_clock,
 348                                                        current, &rtn);
 349                         read_unlock(&tasklist_lock);
 350                 }
 351         } else {
 352                 /*
 353                  * Find the given PID, and validate that the caller
 354                  * should be able to see it.
 355                  */
 356                 struct task_struct *p;
 357                 rcu_read_lock();
 358                 p = find_task_by_vpid(pid);
 359                 if (p) {
 360                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 361                                 if (same_thread_group(p, current)) {
 362                                         error = cpu_clock_sample(which_clock,
 363                                                                  p, &rtn);
 364                                 }
 365                         } else {
 366                                 read_lock(&tasklist_lock);
 367                                 if (thread_group_leader(p) && p->signal) {
 368                                         error =
 369                                             cpu_clock_sample_group(which_clock,
 370                                                                    p, &rtn);
 371                                 }
 372                                 read_unlock(&tasklist_lock);
 373                         }
 374                 }
 375                 rcu_read_unlock();
 376         }
 377
 378         if (error)
 379                 return error;
 380         sample_to_timespec(which_clock, rtn, tp);
 381         return 0;
 382 }
 383
 384
 385 /*
 386  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 387  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 388  * new timer already all-zeros initialized.
 389  */
 390 int posix_cpu_timer_create(struct k_itimer *new_timer)
 391 {
 392         int ret = 0;
 393         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 394         struct task_struct *p;
 395
 396         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 397                 return -EINVAL;
 398
 399         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 400
 401         read_lock(&tasklist_lock);
 402         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 403                 if (pid == 0) {
 404                         p = current;
 405                 } else {
 406                         p = find_task_by_vpid(pid);
 407                         if (p && !same_thread_group(p, current))
 408                                 p = NULL;
 409                 }
 410         } else {
 411                 if (pid == 0) {
 412                         p = current->group_leader;
 413                 } else {
 414                         p = find_task_by_vpid(pid);
 415                         if (p && !thread_group_leader(p))
 416                                 p = NULL;
 417                 }
 418         }
 419         new_timer->it.cpu.task = p;
 420         if (p) {
 421                 get_task_struct(p);
 422         } else {
 423                 ret = -EINVAL;
 424         }
 425         read_unlock(&tasklist_lock);
 426
 427         return ret;
 428 }
 429
 430 /*
 431  * Clean up a CPU-clock timer that is about to be destroyed.
 432  * This is called from timer deletion with the timer already locked.
 433  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 434  * and try again.  (This happens when the timer is in the middle of firing.)
 435  */
 436 int posix_cpu_timer_del(struct k_itimer *timer)
 437 {
 438         struct task_struct *p = timer->it.cpu.task;
 439         int ret = 0;
 440
 441         if (likely(p != NULL)) {
 442                 read_lock(&tasklist_lock);
 443                 if (unlikely(p->signal == NULL)) {
 444                         /*
 445                          * We raced with the reaping of the task.
 446                          * The deletion should have cleared us off the list.
 447                          */
 448                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 449                 } else {
 450                         spin_lock(&p->sighand->siglock);
 451                         if (timer->it.cpu.firing)
 452                                 ret = TIMER_RETRY;
 453                         else
 454                                 list_del(&timer->it.cpu.entry);
 455                         spin_unlock(&p->sighand->siglock);
 456                 }
 457                 read_unlock(&tasklist_lock);
 458
 459                 if (!ret)
 460                         put_task_struct(p);
 461         }
 462
 463         return ret;
 464 }
 465
 466 /*
 467  * Clean out CPU timers still ticking when a thread exited.  The task
 468  * pointer is cleared, and the expiry time is replaced with the residual
 469  * time for later timer_gettime calls to return.
 470  * This must be called with the siglock held.
 471  */
 472 static void cleanup_timers(struct list_head *head,
 473                            cputime_t utime, cputime_t stime,
 474                            unsigned long long sum_exec_runtime)
 475 {
 476         struct cpu_timer_list *timer, *next;
 477         cputime_t ptime = cputime_add(utime, stime);
 478
 479         list_for_each_entry_safe(timer, next, head, entry) {
 480                 list_del_init(&timer->entry);
 481                 if (cputime_lt(timer->expires.cpu, ptime)) {
 482                         timer->expires.cpu = cputime_zero;
 483                 } else {
 484                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 485                                                          ptime);
 486                 }
 487         }
 488
 489         ++head;
 490         list_for_each_entry_safe(timer, next, head, entry) {
 491                 list_del_init(&timer->entry);
 492                 if (cputime_lt(timer->expires.cpu, utime)) {
 493                         timer->expires.cpu = cputime_zero;
 494                 } else {
 495                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 496                                                          utime);
 497                 }
 498         }
 499
 500         ++head;
 501         list_for_each_entry_safe(timer, next, head, entry) {
 502                 list_del_init(&timer->entry);
 503                 if (timer->expires.sched < sum_exec_runtime) {
 504                         timer->expires.sched = 0;
 505                 } else {
 506                         timer->expires.sched -= sum_exec_runtime;
 507                 }
 508         }
 509 }
 510
 511 /*
 512  * These are both called with the siglock held, when the current thread
 513  * is being reaped.  When the final (leader) thread in the group is reaped,
 514  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 515  */
 516 void posix_cpu_timers_exit(struct task_struct *tsk)
 517 {
 518         cleanup_timers(tsk->cpu_timers,
 519                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 520
 521 }
 522 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 523 {
 524         struct signal_struct *const sig = tsk->signal;
 525
 526         cleanup_timers(tsk->signal->cpu_timers,
 527                        cputime_add(tsk->utime, sig->utime),
 528                        cputime_add(tsk->stime, sig->stime),
 529                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 530 }
 531
 532 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 533 {
 534         /*
 535          * That's all for this thread or process.
 536          * We leave our residual in expires to be reported.
 537          */
 538         put_task_struct(timer->it.cpu.task);
 539         timer->it.cpu.task = NULL;
 540         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 541                                              timer->it.cpu.expires,
 542                                              now);
 543 }
 544
 545 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 546 {
 547         return cputime_eq(expires, cputime_zero) ||
 548                cputime_gt(expires, new_exp);
 549 }
 550
 551 static inline int expires_le(cputime_t expires, cputime_t new_exp)
 552 {
 553         return !cputime_eq(expires, cputime_zero) &&
 554                cputime_le(expires, new_exp);
 555 }
 556 /*
 557  * Insert the timer on the appropriate list before any timers that
 558  * expire later.  This must be called with the tasklist_lock held
 559  * for reading, and interrupts disabled.
 560  */
 561 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 562 {
 563         struct task_struct *p = timer->it.cpu.task;
 564         struct list_head *head, *listpos;
 565         struct cpu_timer_list *const nt = &timer->it.cpu;
 566         struct cpu_timer_list *next;
 567         unsigned long i;
 568
 569         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 570                 p->cpu_timers : p->signal->cpu_timers);
 571         head += CPUCLOCK_WHICH(timer->it_clock);
 572
 573         BUG_ON(!irqs_disabled());
 574         spin_lock(&p->sighand->siglock);
 575
 576         listpos = head;
 577         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 578                 list_for_each_entry(next, head, entry) {
 579                         if (next->expires.sched > nt->expires.sched)
 580                                 break;
 581                         listpos = &next->entry;
 582                 }
 583         } else {
 584                 list_for_each_entry(next, head, entry) {
 585                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 586                                 break;
 587                         listpos = &next->entry;
 588                 }
 589         }
 590         list_add(&nt->entry, listpos);
 591
 592         if (listpos == head) {
 593                 /*
 594                  * We are the new earliest-expiring timer.
 595                  * If we are a thread timer, there can always
 596                  * be a process timer telling us to stop earlier.
 597                  */
 598
 599                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 600                         union cpu_time_count *exp = &nt->expires;
 601
 602                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 603                         default:
 604                                 BUG();
 605                         case CPUCLOCK_PROF:
 606                                 if (expires_gt(p->cputime_expires.prof_exp,
 607                                                exp->cpu))
 608                                         p->cputime_expires.prof_exp = exp->cpu;
 609                                 break;
 610                         case CPUCLOCK_VIRT:
 611                                 if (expires_gt(p->cputime_expires.virt_exp,
 612                                                exp->cpu))
 613                                         p->cputime_expires.virt_exp = exp->cpu;
 614                                 break;
 615                         case CPUCLOCK_SCHED:
 616                                 if (p->cputime_expires.sched_exp == 0 ||
 617                                     p->cputime_expires.sched_exp > exp->sched)
 618                                         p->cputime_expires.sched_exp =
 619                                                                 exp->sched;
 620                                 break;
 621                         }
 622                 } else {
 623                         struct signal_struct *const sig = p->signal;
 624                         union cpu_time_count *exp = &timer->it.cpu.expires;
 625
 626                         /*
 627                          * For a process timer, set the cached expiration time.
 628                          */
 629                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 630                         default:
 631                                 BUG();
 632                         case CPUCLOCK_VIRT:
 633                                 if (expires_le(sig->it[CPUCLOCK_VIRT].expires,
 634                                                exp->cpu))
 635                                         break;
 636                                 sig->cputime_expires.virt_exp = exp->cpu;
 637                                 break;
 638                         case CPUCLOCK_PROF:
 639                                 if (expires_le(sig->it[CPUCLOCK_PROF].expires,
 640                                                exp->cpu))
 641                                         break;
 642                                 i = sig->rlim[RLIMIT_CPU].rlim_cur;
 643                                 if (i != RLIM_INFINITY &&
 644                                     i <= cputime_to_secs(exp->cpu))
 645                                         break;
 646                                 sig->cputime_expires.prof_exp = exp->cpu;
 647                                 break;
 648                         case CPUCLOCK_SCHED:
 649                                 sig->cputime_expires.sched_exp = exp->sched;
 650                                 break;
 651                         }
 652                 }
 653         }
 654
 655         spin_unlock(&p->sighand->siglock);
 656 }
 657
 658 /*
 659  * The timer is locked, fire it and arrange for its reload.
 660  */
 661 static void cpu_timer_fire(struct k_itimer *timer)
 662 {
 663         if (unlikely(timer->sigq == NULL)) {
 664                 /*
 665                  * This a special case for clock_nanosleep,
 666                  * not a normal timer from sys_timer_create.
 667                  */
 668                 wake_up_process(timer->it_process);
 669                 timer->it.cpu.expires.sched = 0;
 670         } else if (timer->it.cpu.incr.sched == 0) {
 671                 /*
 672                  * One-shot timer.  Clear it as soon as it's fired.
 673                  */
 674                 posix_timer_event(timer, 0);
 675                 timer->it.cpu.expires.sched = 0;
 676         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 677                 /*
 678                  * The signal did not get queued because the signal
 679                  * was ignored, so we won't get any callback to
 680                  * reload the timer.  But we need to keep it
 681                  * ticking in case the signal is deliverable next time.
 682                  */
 683                 posix_cpu_timer_schedule(timer);
 684         }
 685 }
 686
 687 /*
 688  * Sample a process (thread group) timer for the given group_leader task.
 689  * Must be called with tasklist_lock held for reading.
 690  */
 691 static int cpu_timer_sample_group(const clockid_t which_clock,
 692                                   struct task_struct *p,
 693                                   union cpu_time_count *cpu)
 694 {
 695         struct task_cputime cputime;
 696
 697         thread_group_cputimer(p, &cputime);
 698         switch (CPUCLOCK_WHICH(which_clock)) {
 699         default:
 700                 return -EINVAL;
 701         case CPUCLOCK_PROF:
 702                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 703                 break;
 704         case CPUCLOCK_VIRT:
 705                 cpu->cpu = cputime.utime;
 706                 break;
 707         case CPUCLOCK_SCHED:
 708                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 709                 break;
 710         }
 711         return 0;
 712 }
 713
 714 /*
 715  * Guts of sys_timer_settime for CPU timers.
 716  * This is called with the timer locked and interrupts disabled.
 717  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 718  * and try again.  (This happens when the timer is in the middle of firing.)
 719  */
 720 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 721                         struct itimerspec *new, struct itimerspec *old)
 722 {
 723         struct task_struct *p = timer->it.cpu.task;
 724         union cpu_time_count old_expires, new_expires, val;
 725         int ret;
 726
 727         if (unlikely(p == NULL)) {
 728                 /*
 729                  * Timer refers to a dead task's clock.
 730                  */
 731                 return -ESRCH;
 732         }
 733
 734         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 735
 736         read_lock(&tasklist_lock);
 737         /*
 738          * We need the tasklist_lock to protect against reaping that
 739          * clears p->signal.  If p has just been reaped, we can no
 740          * longer get any information about it at all.
 741          */
 742         if (unlikely(p->signal == NULL)) {
 743                 read_unlock(&tasklist_lock);
 744                 put_task_struct(p);
 745                 timer->it.cpu.task = NULL;
 746                 return -ESRCH;
 747         }
 748
 749         /*
 750          * Disarm any old timer after extracting its expiry time.
 751          */
 752         BUG_ON(!irqs_disabled());
 753
 754         ret = 0;
 755         spin_lock(&p->sighand->siglock);
 756         old_expires = timer->it.cpu.expires;
 757         if (unlikely(timer->it.cpu.firing)) {
 758                 timer->it.cpu.firing = -1;
 759                 ret = TIMER_RETRY;
 760         } else
 761                 list_del_init(&timer->it.cpu.entry);
 762         spin_unlock(&p->sighand->siglock);
 763
 764         /*
 765          * We need to sample the current value to convert the new
 766          * value from to relative and absolute, and to convert the
 767          * old value from absolute to relative.  To set a process
 768          * timer, we need a sample to balance the thread expiry
 769          * times (in arm_timer).  With an absolute time, we must
 770          * check if it's already passed.  In short, we need a sample.
 771          */
 772         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 773                 cpu_clock_sample(timer->it_clock, p, &val);
 774         } else {
 775                 cpu_timer_sample_group(timer->it_clock, p, &val);
 776         }
 777
 778         if (old) {
 779                 if (old_expires.sched == 0) {
 780                         old->it_value.tv_sec = 0;
 781                         old->it_value.tv_nsec = 0;
 782                 } else {
 783                         /*
 784                          * Update the timer in case it has
 785                          * overrun already.  If it has,
 786                          * we'll report it as having overrun
 787                          * and with the next reloaded timer
 788                          * already ticking, though we are
 789                          * swallowing that pending
 790                          * notification here to install the
 791                          * new setting.
 792                          */
 793                         bump_cpu_timer(timer, val);
 794                         if (cpu_time_before(timer->it_clock, val,
 795                                             timer->it.cpu.expires)) {
 796                                 old_expires = cpu_time_sub(
 797                                         timer->it_clock,
 798                                         timer->it.cpu.expires, val);
 799                                 sample_to_timespec(timer->it_clock,
 800                                                    old_expires,
 801                                                    &old->it_value);
 802                         } else {
 803                                 old->it_value.tv_nsec = 1;
 804                                 old->it_value.tv_sec = 0;
 805                         }
 806                 }
 807         }
 808
 809         if (unlikely(ret)) {
 810                 /*
 811                  * We are colliding with the timer actually firing.
 812                  * Punt after filling in the timer's old value, and
 813                  * disable this firing since we are already reporting
 814                  * it as an overrun (thanks to bump_cpu_timer above).
 815                  */
 816                 read_unlock(&tasklist_lock);
 817                 goto out;
 818         }
 819
 820         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 821                 cpu_time_add(timer->it_clock, &new_expires, val);
 822         }
 823
 824         /*
 825          * Install the new expiry time (or zero).
 826          * For a timer with no notification action, we don't actually
 827          * arm the timer (we'll just fake it for timer_gettime).
 828          */
 829         timer->it.cpu.expires = new_expires;
 830         if (new_expires.sched != 0 &&
 831             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 832             cpu_time_before(timer->it_clock, val, new_expires)) {
 833                 arm_timer(timer, val);
 834         }
 835
 836         read_unlock(&tasklist_lock);
 837
 838         /*
 839          * Install the new reload setting, and
 840          * set up the signal and overrun bookkeeping.
 841          */
 842         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 843                                                 &new->it_interval);
 844
 845         /*
 846          * This acts as a modification timestamp for the timer,
 847          * so any automatic reload attempt will punt on seeing
 848          * that we have reset the timer manually.
 849          */
 850         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 851                 ~REQUEUE_PENDING;
 852         timer->it_overrun_last = 0;
 853         timer->it_overrun = -1;
 854
 855         if (new_expires.sched != 0 &&
 856             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 857             !cpu_time_before(timer->it_clock, val, new_expires)) {
 858                 /*
 859                  * The designated time already passed, so we notify
 860                  * immediately, even if the thread never runs to
 861                  * accumulate more time on this clock.
 862                  */
 863                 cpu_timer_fire(timer);
 864         }
 865
 866         ret = 0;
 867  out:
 868         if (old) {
 869                 sample_to_timespec(timer->it_clock,
 870                                    timer->it.cpu.incr, &old->it_interval);
 871         }
 872         return ret;
 873 }
 874
 875 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 876 {
 877         union cpu_time_count now;
 878         struct task_struct *p = timer->it.cpu.task;
 879         int clear_dead;
 880
 881         /*
 882          * Easy part: convert the reload time.
 883          */
 884         sample_to_timespec(timer->it_clock,
 885                            timer->it.cpu.incr, &itp->it_interval);
 886
 887         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 888                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 889                 return;
 890         }
 891
 892         if (unlikely(p == NULL)) {
 893                 /*
 894                  * This task already died and the timer will never fire.
 895                  * In this case, expires is actually the dead value.
 896                  */
 897         dead:
 898                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 899                                    &itp->it_value);
 900                 return;
 901         }
 902
 903         /*
 904          * Sample the clock to take the difference with the expiry time.
 905          */
 906         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 907                 cpu_clock_sample(timer->it_clock, p, &now);
 908                 clear_dead = p->exit_state;
 909         } else {
 910                 read_lock(&tasklist_lock);
 911                 if (unlikely(p->signal == NULL)) {
 912                         /*
 913                          * The process has been reaped.
 914                          * We can't even collect a sample any more.
 915                          * Call the timer disarmed, nothing else to do.
 916                          */
 917                         put_task_struct(p);
 918                         timer->it.cpu.task = NULL;
 919                         timer->it.cpu.expires.sched = 0;
 920                         read_unlock(&tasklist_lock);
 921                         goto dead;
 922                 } else {
 923                         cpu_timer_sample_group(timer->it_clock, p, &now);
 924                         clear_dead = (unlikely(p->exit_state) &&
 925                                       thread_group_empty(p));
 926                 }
 927                 read_unlock(&tasklist_lock);
 928         }
 929
 930         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 931                 if (timer->it.cpu.incr.sched == 0 &&
 932                     cpu_time_before(timer->it_clock,
 933                                     timer->it.cpu.expires, now)) {
 934                         /*
 935                          * Do-nothing timer expired and has no reload,
 936                          * so it's as if it was never set.
 937                          */
 938                         timer->it.cpu.expires.sched = 0;
 939                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 940                         return;
 941                 }
 942                 /*
 943                  * Account for any expirations and reloads that should
 944                  * have happened.
 945                  */
 946                 bump_cpu_timer(timer, now);
 947         }
 948
 949         if (unlikely(clear_dead)) {
 950                 /*
 951                  * We've noticed that the thread is dead, but
 952                  * not yet reaped.  Take this opportunity to
 953                  * drop our task ref.
 954                  */
 955                 clear_dead_task(timer, now);
 956                 goto dead;
 957         }
 958
 959         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 960                 sample_to_timespec(timer->it_clock,
 961                                    cpu_time_sub(timer->it_clock,
 962                                                 timer->it.cpu.expires, now),
 963                                    &itp->it_value);
 964         } else {
 965                 /*
 966                  * The timer should have expired already, but the firing
 967                  * hasn't taken place yet.  Say it's just about to expire.
 968                  */
 969                 itp->it_value.tv_nsec = 1;
 970                 itp->it_value.tv_sec = 0;
 971         }
 972 }
 973
 974 /*
 975  * Check for any per-thread CPU timers that have fired and move them off
 976  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 977  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 978  */
 979 static void check_thread_timers(struct task_struct *tsk,
 980                                 struct list_head *firing)
 981 {
 982         int maxfire;
 983         struct list_head *timers = tsk->cpu_timers;
 984         struct signal_struct *const sig = tsk->signal;
 985
 986         maxfire = 20;
 987         tsk->cputime_expires.prof_exp = cputime_zero;
 988         while (!list_empty(timers)) {
 989                 struct cpu_timer_list *t = list_first_entry(timers,
 990                                                       struct cpu_timer_list,
 991                                                       entry);
 992                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 993                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 994                         break;
 995                 }
 996                 t->firing = 1;
 997                 list_move_tail(&t->entry, firing);
 998         }
 999
1000         ++timers;
1001         maxfire = 20;
1002         tsk->cputime_expires.virt_exp = cputime_zero;
1003         while (!list_empty(timers)) {
1004                 struct cpu_timer_list *t = list_first_entry(timers,
1005                                                       struct cpu_timer_list,
1006                                                       entry);
1007                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1008                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1009                         break;
1010                 }
1011                 t->firing = 1;
1012                 list_move_tail(&t->entry, firing);
1013         }
1014
1015         ++timers;
1016         maxfire = 20;
1017         tsk->cputime_expires.sched_exp = 0;
1018         while (!list_empty(timers)) {
1019                 struct cpu_timer_list *t = list_first_entry(timers,
1020                                                       struct cpu_timer_list,
1021                                                       entry);
1022                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1023                         tsk->cputime_expires.sched_exp = t->expires.sched;
1024                         break;
1025                 }
1026                 t->firing = 1;
1027                 list_move_tail(&t->entry, firing);
1028         }
1029
1030         /*
1031          * Check for the special case thread timers.
1032          */
1033         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1034                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1035                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1036
1037                 if (hard != RLIM_INFINITY &&
1038                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1039                         /*
1040                          * At the hard limit, we just die.
1041                          * No need to calculate anything else now.
1042                          */
1043                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1044                         return;
1045                 }
1046                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1047                         /*
1048                          * At the soft limit, send a SIGXCPU every second.
1049                          */
1050                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1051                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1052                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1053                                                                 USEC_PER_SEC;
1054                         }
1055                         printk(KERN_INFO
1056                                 "RT Watchdog Timeout: %s[%d]\n",
1057                                 tsk->comm, task_pid_nr(tsk));
1058                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1059                 }
1060         }
1061 }
1062
1063 static void stop_process_timers(struct task_struct *tsk)
1064 {
1065         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1066         unsigned long flags;
1067
1068         if (!cputimer->running)
1069                 return;
1070
1071         spin_lock_irqsave(&cputimer->lock, flags);
1072         cputimer->running = 0;
1073         spin_unlock_irqrestore(&cputimer->lock, flags);
1074 }
1075
1076 static u32 onecputick;
1077
1078 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1079                              cputime_t *expires, cputime_t cur_time, int signo)
1080 {
1081         if (cputime_eq(it->expires, cputime_zero))
1082                 return;
1083
1084         if (cputime_ge(cur_time, it->expires)) {
1085                 if (!cputime_eq(it->incr, cputime_zero)) {
1086                         it->expires = cputime_add(it->expires, it->incr);
1087                         it->error += it->incr_error;
1088                         if (it->error >= onecputick) {
1089                                 it->expires = cputime_sub(it->expires,
1090                                                           cputime_one_jiffy);
1091                                 it->error -= onecputick;
1092                         }
1093                 } else {
1094                         it->expires = cputime_zero;
1095                 }
1096
1097                 trace_itimer_expire(signo == SIGPROF ?
1098                                     ITIMER_PROF : ITIMER_VIRTUAL,
1099                                     tsk->signal->leader_pid, cur_time);
1100                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1101         }
1102
1103         if (!cputime_eq(it->expires, cputime_zero) &&
1104             (cputime_eq(*expires, cputime_zero) ||
1105              cputime_lt(it->expires, *expires))) {
1106                 *expires = it->expires;
1107         }
1108 }
1109
1110 /*
1111  * Check for any per-thread CPU timers that have fired and move them
1112  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1113  * have already been taken off.
1114  */
1115 static void check_process_timers(struct task_struct *tsk,
1116                                  struct list_head *firing)
1117 {
1118         int maxfire;
1119         struct signal_struct *const sig = tsk->signal;
1120         cputime_t utime, ptime, virt_expires, prof_expires;
1121         unsigned long long sum_sched_runtime, sched_expires;
1122         struct list_head *timers = sig->cpu_timers;
1123         struct task_cputime cputime;
1124
1125         /*
1126          * Don't sample the current process CPU clocks if there are no timers.
1127          */
1128         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1129             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1130             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1131             list_empty(&timers[CPUCLOCK_VIRT]) &&
1132             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1133             list_empty(&timers[CPUCLOCK_SCHED])) {
1134                 stop_process_timers(tsk);
1135                 return;
1136         }
1137
1138         /*
1139          * Collect the current process totals.
1140          */
1141         thread_group_cputimer(tsk, &cputime);
1142         utime = cputime.utime;
1143         ptime = cputime_add(utime, cputime.stime);
1144         sum_sched_runtime = cputime.sum_exec_runtime;
1145         maxfire = 20;
1146         prof_expires = cputime_zero;
1147         while (!list_empty(timers)) {
1148                 struct cpu_timer_list *tl = list_first_entry(timers,
1149                                                       struct cpu_timer_list,
1150                                                       entry);
1151                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1152                         prof_expires = tl->expires.cpu;
1153                         break;
1154                 }
1155                 tl->firing = 1;
1156                 list_move_tail(&tl->entry, firing);
1157         }
1158
1159         ++timers;
1160         maxfire = 20;
1161         virt_expires = cputime_zero;
1162         while (!list_empty(timers)) {
1163                 struct cpu_timer_list *tl = list_first_entry(timers,
1164                                                       struct cpu_timer_list,
1165                                                       entry);
1166                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1167                         virt_expires = tl->expires.cpu;
1168                         break;
1169                 }
1170                 tl->firing = 1;
1171                 list_move_tail(&tl->entry, firing);
1172         }
1173
1174         ++timers;
1175         maxfire = 20;
1176         sched_expires = 0;
1177         while (!list_empty(timers)) {
1178                 struct cpu_timer_list *tl = list_first_entry(timers,
1179                                                       struct cpu_timer_list,
1180                                                       entry);
1181                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1182                         sched_expires = tl->expires.sched;
1183                         break;
1184                 }
1185                 tl->firing = 1;
1186                 list_move_tail(&tl->entry, firing);
1187         }
1188
1189         /*
1190          * Check for the special case process timers.
1191          */
1192         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1193                          SIGPROF);
1194         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1195                          SIGVTALRM);
1196
1197         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1198                 unsigned long psecs = cputime_to_secs(ptime);
1199                 cputime_t x;
1200                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1201                         /*
1202                          * At the hard limit, we just die.
1203                          * No need to calculate anything else now.
1204                          */
1205                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1206                         return;
1207                 }
1208                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1209                         /*
1210                          * At the soft limit, send a SIGXCPU every second.
1211                          */
1212                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1213                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1214                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1215                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1216                         }
1217                 }
1218                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1219                 if (cputime_eq(prof_expires, cputime_zero) ||
1220                     cputime_lt(x, prof_expires)) {
1221                         prof_expires = x;
1222                 }
1223         }
1224
1225         if (!cputime_eq(prof_expires, cputime_zero) &&
1226             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1227              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1228                 sig->cputime_expires.prof_exp = prof_expires;
1229         if (!cputime_eq(virt_expires, cputime_zero) &&
1230             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1231              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1232                 sig->cputime_expires.virt_exp = virt_expires;
1233         if (sched_expires != 0 &&
1234             (sig->cputime_expires.sched_exp == 0 ||
1235              sig->cputime_expires.sched_exp > sched_expires))
1236                 sig->cputime_expires.sched_exp = sched_expires;
1237 }
1238
1239 /*
1240  * This is called from the signal code (via do_schedule_next_timer)
1241  * when the last timer signal was delivered and we have to reload the timer.
1242  */
1243 void posix_cpu_timer_schedule(struct k_itimer *timer)
1244 {
1245         struct task_struct *p = timer->it.cpu.task;
1246         union cpu_time_count now;
1247
1248         if (unlikely(p == NULL))
1249                 /*
1250                  * The task was cleaned up already, no future firings.
1251                  */
1252                 goto out;
1253
1254         /*
1255          * Fetch the current sample and update the timer's expiry time.
1256          */
1257         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1258                 cpu_clock_sample(timer->it_clock, p, &now);
1259                 bump_cpu_timer(timer, now);
1260                 if (unlikely(p->exit_state)) {
1261                         clear_dead_task(timer, now);
1262                         goto out;
1263                 }
1264                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1265         } else {
1266                 read_lock(&tasklist_lock);
1267                 if (unlikely(p->signal == NULL)) {
1268                         /*
1269                          * The process has been reaped.
1270                          * We can't even collect a sample any more.
1271                          */
1272                         put_task_struct(p);
1273                         timer->it.cpu.task = p = NULL;
1274                         timer->it.cpu.expires.sched = 0;
1275                         goto out_unlock;
1276                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1277                         /*
1278                          * We've noticed that the thread is dead, but
1279                          * not yet reaped.  Take this opportunity to
1280                          * drop our task ref.
1281                          */
1282                         clear_dead_task(timer, now);
1283                         goto out_unlock;
1284                 }
1285                 cpu_timer_sample_group(timer->it_clock, p, &now);
1286                 bump_cpu_timer(timer, now);
1287                 /* Leave the tasklist_lock locked for the call below.  */
1288         }
1289
1290         /*
1291          * Now re-arm for the new expiry time.
1292          */
1293         arm_timer(timer, now);
1294
1295 out_unlock:
1296         read_unlock(&tasklist_lock);
1297
1298 out:
1299         timer->it_overrun_last = timer->it_overrun;
1300         timer->it_overrun = -1;
1301         ++timer->it_requeue_pending;
1302 }
1303
1304 /**
1305  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1306  *
1307  * @cputime:    The struct to compare.
1308  *
1309  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1310  * are zero, false if any field is nonzero.
1311  */
1312 static inline int task_cputime_zero(const struct task_cputime *cputime)
1313 {
1314         if (cputime_eq(cputime->utime, cputime_zero) &&
1315             cputime_eq(cputime->stime, cputime_zero) &&
1316             cputime->sum_exec_runtime == 0)
1317                 return 1;
1318         return 0;
1319 }
1320
1321 /**
1322  * task_cputime_expired - Compare two task_cputime entities.
1323  *
1324  * @sample:     The task_cputime structure to be checked for expiration.
1325  * @expires:    Expiration times, against which @sample will be checked.
1326  *
1327  * Checks @sample against @expires to see if any field of @sample has expired.
1328  * Returns true if any field of the former is greater than the corresponding
1329  * field of the latter if the latter field is set.  Otherwise returns false.
1330  */
1331 static inline int task_cputime_expired(const struct task_cputime *sample,
1332                                         const struct task_cputime *expires)
1333 {
1334         if (!cputime_eq(expires->utime, cputime_zero) &&
1335             cputime_ge(sample->utime, expires->utime))
1336                 return 1;
1337         if (!cputime_eq(expires->stime, cputime_zero) &&
1338             cputime_ge(cputime_add(sample->utime, sample->stime),
1339                        expires->stime))
1340                 return 1;
1341         if (expires->sum_exec_runtime != 0 &&
1342             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1343                 return 1;
1344         return 0;
1345 }
1346
1347 /**
1348  * fastpath_timer_check - POSIX CPU timers fast path.
1349  *
1350  * @tsk:        The task (thread) being checked.
1351  *
1352  * Check the task and thread group timers.  If both are zero (there are no
1353  * timers set) return false.  Otherwise snapshot the task and thread group
1354  * timers and compare them with the corresponding expiration times.  Return
1355  * true if a timer has expired, else return false.
1356  */
1357 static inline int fastpath_timer_check(struct task_struct *tsk)
1358 {
1359         struct signal_struct *sig;
1360
1361         /* tsk == current, ensure it is safe to use ->signal/sighand */
1362         if (unlikely(tsk->exit_state))
1363                 return 0;
1364
1365         if (!task_cputime_zero(&tsk->cputime_expires)) {
1366                 struct task_cputime task_sample = {
1367                         .utime = tsk->utime,
1368                         .stime = tsk->stime,
1369                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1370                 };
1371
1372                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1373                         return 1;
1374         }
1375
1376         sig = tsk->signal;
1377         if (!task_cputime_zero(&sig->cputime_expires)) {
1378                 struct task_cputime group_sample;
1379
1380                 thread_group_cputimer(tsk, &group_sample);
1381                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1382                         return 1;
1383         }
1384
1385         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1386 }
1387
1388 /*
1389  * This is called from the timer interrupt handler.  The irq handler has
1390  * already updated our counts.  We need to check if any timers fire now.
1391  * Interrupts are disabled.
1392  */
1393 void run_posix_cpu_timers(struct task_struct *tsk)
1394 {
1395         LIST_HEAD(firing);
1396         struct k_itimer *timer, *next;
1397
1398         BUG_ON(!irqs_disabled());
1399
1400         /*
1401          * The fast path checks that there are no expired thread or thread
1402          * group timers.  If that's so, just return.
1403          */
1404         if (!fastpath_timer_check(tsk))
1405                 return;
1406
1407         spin_lock(&tsk->sighand->siglock);
1408         /*
1409          * Here we take off tsk->signal->cpu_timers[N] and
1410          * tsk->cpu_timers[N] all the timers that are firing, and
1411          * put them on the firing list.
1412          */
1413         check_thread_timers(tsk, &firing);
1414         check_process_timers(tsk, &firing);
1415
1416         /*
1417          * We must release these locks before taking any timer's lock.
1418          * There is a potential race with timer deletion here, as the
1419          * siglock now protects our private firing list.  We have set
1420          * the firing flag in each timer, so that a deletion attempt
1421          * that gets the timer lock before we do will give it up and
1422          * spin until we've taken care of that timer below.
1423          */
1424         spin_unlock(&tsk->sighand->siglock);
1425
1426         /*
1427          * Now that all the timers on our list have the firing flag,
1428          * noone will touch their list entries but us.  We'll take
1429          * each timer's lock before clearing its firing flag, so no
1430          * timer call will interfere.
1431          */
1432         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1433                 int cpu_firing;
1434
1435                 spin_lock(&timer->it_lock);
1436                 list_del_init(&timer->it.cpu.entry);
1437                 cpu_firing = timer->it.cpu.firing;
1438                 timer->it.cpu.firing = 0;
1439                 /*
1440                  * The firing flag is -1 if we collided with a reset
1441                  * of the timer, which already reported this
1442                  * almost-firing as an overrun.  So don't generate an event.
1443                  */
1444                 if (likely(cpu_firing >= 0))
1445                         cpu_timer_fire(timer);
1446                 spin_unlock(&timer->it_lock);
1447         }
1448 }
1449
1450 /*
1451  * Set one of the process-wide special case CPU timers.
1452  * The tsk->sighand->siglock must be held by the caller.
1453  * The *newval argument is relative and we update it to be absolute, *oldval
1454  * is absolute and we update it to be relative.
1455  */
1456 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1457                            cputime_t *newval, cputime_t *oldval)
1458 {
1459         union cpu_time_count now;
1460         struct list_head *head;
1461
1462         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1463         cpu_timer_sample_group(clock_idx, tsk, &now);
1464
1465         if (oldval) {
1466                 if (!cputime_eq(*oldval, cputime_zero)) {
1467                         if (cputime_le(*oldval, now.cpu)) {
1468                                 /* Just about to fire. */
1469                                 *oldval = cputime_one_jiffy;
1470                         } else {
1471                                 *oldval = cputime_sub(*oldval, now.cpu);
1472                         }
1473                 }
1474
1475                 if (cputime_eq(*newval, cputime_zero))
1476                         return;
1477                 *newval = cputime_add(*newval, now.cpu);
1478
1479                 /*
1480                  * If the RLIMIT_CPU timer will expire before the
1481                  * ITIMER_PROF timer, we have nothing else to do.
1482                  */
1483                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1484                     < cputime_to_secs(*newval))
1485                         return;
1486         }
1487
1488         /*
1489          * Check whether there are any process timers already set to fire
1490          * before this one.  If so, we don't have anything more to do.
1491          */
1492         head = &tsk->signal->cpu_timers[clock_idx];
1493         if (list_empty(head) ||
1494             cputime_ge(list_first_entry(head,
1495                                   struct cpu_timer_list, entry)->expires.cpu,
1496                        *newval)) {
1497                 switch (clock_idx) {
1498                 case CPUCLOCK_PROF:
1499                         tsk->signal->cputime_expires.prof_exp = *newval;
1500                         break;
1501                 case CPUCLOCK_VIRT:
1502                         tsk->signal->cputime_expires.virt_exp = *newval;
1503                         break;
1504                 }
1505         }
1506 }
1507
1508 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1509                             struct timespec *rqtp, struct itimerspec *it)
1510 {
1511         struct k_itimer timer;
1512         int error;
1513
1514         /*
1515          * Set up a temporary timer and then wait for it to go off.
1516          */
1517         memset(&timer, 0, sizeof timer);
1518         spin_lock_init(&timer.it_lock);
1519         timer.it_clock = which_clock;
1520         timer.it_overrun = -1;
1521         error = posix_cpu_timer_create(&timer);
1522         timer.it_process = current;
1523         if (!error) {
1524                 static struct itimerspec zero_it;
1525
1526                 memset(it, 0, sizeof *it);
1527                 it->it_value = *rqtp;
1528
1529                 spin_lock_irq(&timer.it_lock);
1530                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1531                 if (error) {
1532                         spin_unlock_irq(&timer.it_lock);
1533                         return error;
1534                 }
1535
1536                 while (!signal_pending(current)) {
1537                         if (timer.it.cpu.expires.sched == 0) {
1538                                 /*
1539                                  * Our timer fired and was reset.
1540                                  */
1541                                 spin_unlock_irq(&timer.it_lock);
1542                                 return 0;
1543                         }
1544
1545                         /*
1546                          * Block until cpu_timer_fire (or a signal) wakes us.
1547                          */
1548                         __set_current_state(TASK_INTERRUPTIBLE);
1549                         spin_unlock_irq(&timer.it_lock);
1550                         schedule();
1551                         spin_lock_irq(&timer.it_lock);
1552                 }
1553
1554                 /*
1555                  * We were interrupted by a signal.
1556                  */
1557                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1558                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1559                 spin_unlock_irq(&timer.it_lock);
1560
1561                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1562                         /*
1563                          * It actually did fire already.
1564                          */
1565                         return 0;
1566                 }
1567
1568                 error = -ERESTART_RESTARTBLOCK;
1569         }
1570
1571         return error;
1572 }
1573
1574 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1575                      struct timespec *rqtp, struct timespec __user *rmtp)
1576 {
1577         struct restart_block *restart_block =
1578             &current_thread_info()->restart_block;
1579         struct itimerspec it;
1580         int error;
1581
1582         /*
1583          * Diagnose required errors first.
1584          */
1585         if (CPUCLOCK_PERTHREAD(which_clock) &&
1586             (CPUCLOCK_PID(which_clock) == 0 ||
1587              CPUCLOCK_PID(which_clock) == current->pid))
1588                 return -EINVAL;
1589
1590         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1591
1592         if (error == -ERESTART_RESTARTBLOCK) {
1593
1594                 if (flags & TIMER_ABSTIME)
1595                         return -ERESTARTNOHAND;
1596                 /*
1597                  * Report back to the user the time still remaining.
1598                  */
1599                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1600                         return -EFAULT;
1601
1602                 restart_block->fn = posix_cpu_nsleep_restart;
1603                 restart_block->arg0 = which_clock;
1604                 restart_block->arg1 = (unsigned long) rmtp;
1605                 restart_block->arg2 = rqtp->tv_sec;
1606                 restart_block->arg3 = rqtp->tv_nsec;
1607         }
1608         return error;
1609 }
1610
1611 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1612 {
1613         clockid_t which_clock = restart_block->arg0;
1614         struct timespec __user *rmtp;
1615         struct timespec t;
1616         struct itimerspec it;
1617         int error;
1618
1619         rmtp = (struct timespec __user *) restart_block->arg1;
1620         t.tv_sec = restart_block->arg2;
1621         t.tv_nsec = restart_block->arg3;
1622
1623         restart_block->fn = do_no_restart_syscall;
1624         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1625
1626         if (error == -ERESTART_RESTARTBLOCK) {
1627                 /*
1628                  * Report back to the user the time still remaining.
1629                  */
1630                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1631                         return -EFAULT;
1632
1633                 restart_block->fn = posix_cpu_nsleep_restart;
1634                 restart_block->arg0 = which_clock;
1635                 restart_block->arg1 = (unsigned long) rmtp;
1636                 restart_block->arg2 = t.tv_sec;
1637                 restart_block->arg3 = t.tv_nsec;
1638         }
1639         return error;
1640
1641 }
1642
1643
1644 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1645 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1646
1647 static int process_cpu_clock_getres(const clockid_t which_clock,
1648                                     struct timespec *tp)
1649 {
1650         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1651 }
1652 static int process_cpu_clock_get(const clockid_t which_clock,
1653                                  struct timespec *tp)
1654 {
1655         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1656 }
1657 static int process_cpu_timer_create(struct k_itimer *timer)
1658 {
1659         timer->it_clock = PROCESS_CLOCK;
1660         return posix_cpu_timer_create(timer);
1661 }
1662 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1663                               struct timespec *rqtp,
1664                               struct timespec __user *rmtp)
1665 {
1666         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1667 }
1668 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1669 {
1670         return -EINVAL;
1671 }
1672 static int thread_cpu_clock_getres(const clockid_t which_clock,
1673                                    struct timespec *tp)
1674 {
1675         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1676 }
1677 static int thread_cpu_clock_get(const clockid_t which_clock,
1678                                 struct timespec *tp)
1679 {
1680         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1681 }
1682 static int thread_cpu_timer_create(struct k_itimer *timer)
1683 {
1684         timer->it_clock = THREAD_CLOCK;
1685         return posix_cpu_timer_create(timer);
1686 }
1687 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1688                               struct timespec *rqtp, struct timespec __user *rmtp)
1689 {
1690         return -EINVAL;
1691 }
1692 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1693 {
1694         return -EINVAL;
1695 }
1696
1697 static __init int init_posix_cpu_timers(void)
1698 {
1699         struct k_clock process = {
1700                 .clock_getres = process_cpu_clock_getres,
1701                 .clock_get = process_cpu_clock_get,
1702                 .clock_set = do_posix_clock_nosettime,
1703                 .timer_create = process_cpu_timer_create,
1704                 .nsleep = process_cpu_nsleep,
1705                 .nsleep_restart = process_cpu_nsleep_restart,
1706         };
1707         struct k_clock thread = {
1708                 .clock_getres = thread_cpu_clock_getres,
1709                 .clock_get = thread_cpu_clock_get,
1710                 .clock_set = do_posix_clock_nosettime,
1711                 .timer_create = thread_cpu_timer_create,
1712                 .nsleep = thread_cpu_nsleep,
1713                 .nsleep_restart = thread_cpu_nsleep_restart,
1714         };
1715         struct timespec ts;
1716
1717         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1718         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1719
1720         cputime_to_timespec(cputime_one_jiffy, &ts);
1721         onecputick = ts.tv_nsec;
1722         WARN_ON(ts.tv_sec != 0);
1723
1724         return 0;
1725 }
1726 __initcall(init_posix_cpu_timers);