SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  12 /*
  13  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
  14  */
  15 void update_rlimit_cpu(unsigned long rlim_new)
  16 {
  17         cputime_t cputime = secs_to_cputime(rlim_new);
  18         struct signal_struct *const sig = current->signal;
  19
  20         if (cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) ||
  21             cputime_gt(sig->it[CPUCLOCK_PROF].expires, cputime)) {
  22                 spin_lock_irq(&current->sighand->siglock);
  23                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
  24                 spin_unlock_irq(&current->sighand->siglock);
  25         }
  26 }
  27
  28 static int check_clock(const clockid_t which_clock)
  29 {
  30         int error = 0;
  31         struct task_struct *p;
  32         const pid_t pid = CPUCLOCK_PID(which_clock);
  33
  34         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  35                 return -EINVAL;
  36
  37         if (pid == 0)
  38                 return 0;
  39
  40         read_lock(&tasklist_lock);
  41         p = find_task_by_vpid(pid);
  42         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  43                    same_thread_group(p, current) : thread_group_leader(p))) {
  44                 error = -EINVAL;
  45         }
  46         read_unlock(&tasklist_lock);
  47
  48         return error;
  49 }
  50
  51 static inline union cpu_time_count
  52 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  53 {
  54         union cpu_time_count ret;
  55         ret.sched = 0;          /* high half always zero when .cpu used */
  56         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  57                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  58         } else {
  59                 ret.cpu = timespec_to_cputime(tp);
  60         }
  61         return ret;
  62 }
  63
  64 static void sample_to_timespec(const clockid_t which_clock,
  65                                union cpu_time_count cpu,
  66                                struct timespec *tp)
  67 {
  68         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  69                 *tp = ns_to_timespec(cpu.sched);
  70         else
  71                 cputime_to_timespec(cpu.cpu, tp);
  72 }
  73
  74 static inline int cpu_time_before(const clockid_t which_clock,
  75                                   union cpu_time_count now,
  76                                   union cpu_time_count then)
  77 {
  78         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  79                 return now.sched < then.sched;
  80         }  else {
  81                 return cputime_lt(now.cpu, then.cpu);
  82         }
  83 }
  84 static inline void cpu_time_add(const clockid_t which_clock,
  85                                 union cpu_time_count *acc,
  86                                 union cpu_time_count val)
  87 {
  88         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  89                 acc->sched += val.sched;
  90         }  else {
  91                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  92         }
  93 }
  94 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  95                                                 union cpu_time_count a,
  96                                                 union cpu_time_count b)
  97 {
  98         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  99                 a.sched -= b.sched;
 100         }  else {
 101                 a.cpu = cputime_sub(a.cpu, b.cpu);
 102         }
 103         return a;
 104 }
 105
 106 /*
 107  * Divide and limit the result to res >= 1
 108  *
 109  * This is necessary to prevent signal delivery starvation, when the result of
 110  * the division would be rounded down to 0.
 111  */
 112 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
 113 {
 114         cputime_t res = cputime_div(time, div);
 115
 116         return max_t(cputime_t, res, 1);
 117 }
 118
 119 /*
 120  * Update expiry time from increment, and increase overrun count,
 121  * given the current clock sample.
 122  */
 123 static void bump_cpu_timer(struct k_itimer *timer,
 124                                   union cpu_time_count now)
 125 {
 126         int i;
 127
 128         if (timer->it.cpu.incr.sched == 0)
 129                 return;
 130
 131         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 132                 unsigned long long delta, incr;
 133
 134                 if (now.sched < timer->it.cpu.expires.sched)
 135                         return;
 136                 incr = timer->it.cpu.incr.sched;
 137                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 138                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 139                 for (i = 0; incr < delta - incr; i++)
 140                         incr = incr << 1;
 141                 for (; i >= 0; incr >>= 1, i--) {
 142                         if (delta < incr)
 143                                 continue;
 144                         timer->it.cpu.expires.sched += incr;
 145                         timer->it_overrun += 1 << i;
 146                         delta -= incr;
 147                 }
 148         } else {
 149                 cputime_t delta, incr;
 150
 151                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 152                         return;
 153                 incr = timer->it.cpu.incr.cpu;
 154                 delta = cputime_sub(cputime_add(now.cpu, incr),
 155                                     timer->it.cpu.expires.cpu);
 156                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 157                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 158                              incr = cputime_add(incr, incr);
 159                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 160                         if (cputime_lt(delta, incr))
 161                                 continue;
 162                         timer->it.cpu.expires.cpu =
 163                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 164                         timer->it_overrun += 1 << i;
 165                         delta = cputime_sub(delta, incr);
 166                 }
 167         }
 168 }
 169
 170 static inline cputime_t prof_ticks(struct task_struct *p)
 171 {
 172         return cputime_add(p->utime, p->stime);
 173 }
 174 static inline cputime_t virt_ticks(struct task_struct *p)
 175 {
 176         return p->utime;
 177 }
 178
 179 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 180 {
 181         int error = check_clock(which_clock);
 182         if (!error) {
 183                 tp->tv_sec = 0;
 184                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 185                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 186                         /*
 187                          * If sched_clock is using a cycle counter, we
 188                          * don't have any idea of its true resolution
 189                          * exported, but it is much more than 1s/HZ.
 190                          */
 191                         tp->tv_nsec = 1;
 192                 }
 193         }
 194         return error;
 195 }
 196
 197 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 198 {
 199         /*
 200          * You can never reset a CPU clock, but we check for other errors
 201          * in the call before failing with EPERM.
 202          */
 203         int error = check_clock(which_clock);
 204         if (error == 0) {
 205                 error = -EPERM;
 206         }
 207         return error;
 208 }
 209
 210
 211 /*
 212  * Sample a per-thread clock for the given task.
 213  */
 214 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 215                             union cpu_time_count *cpu)
 216 {
 217         switch (CPUCLOCK_WHICH(which_clock)) {
 218         default:
 219                 return -EINVAL;
 220         case CPUCLOCK_PROF:
 221                 cpu->cpu = prof_ticks(p);
 222                 break;
 223         case CPUCLOCK_VIRT:
 224                 cpu->cpu = virt_ticks(p);
 225                 break;
 226         case CPUCLOCK_SCHED:
 227                 cpu->sched = task_sched_runtime(p);
 228                 break;
 229         }
 230         return 0;
 231 }
 232
 233 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
 234 {
 235         struct sighand_struct *sighand;
 236         struct signal_struct *sig;
 237         struct task_struct *t;
 238
 239         *times = INIT_CPUTIME;
 240
 241         rcu_read_lock();
 242         sighand = rcu_dereference(tsk->sighand);
 243         if (!sighand)
 244                 goto out;
 245
 246         sig = tsk->signal;
 247
 248         t = tsk;
 249         do {
 250                 times->utime = cputime_add(times->utime, t->utime);
 251                 times->stime = cputime_add(times->stime, t->stime);
 252                 times->sum_exec_runtime += t->se.sum_exec_runtime;
 253
 254                 t = next_thread(t);
 255         } while (t != tsk);
 256
 257         times->utime = cputime_add(times->utime, sig->utime);
 258         times->stime = cputime_add(times->stime, sig->stime);
 259         times->sum_exec_runtime += sig->sum_sched_runtime;
 260 out:
 261         rcu_read_unlock();
 262 }
 263
 264 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 265 {
 266         if (cputime_gt(b->utime, a->utime))
 267                 a->utime = b->utime;
 268
 269         if (cputime_gt(b->stime, a->stime))
 270                 a->stime = b->stime;
 271
 272         if (b->sum_exec_runtime > a->sum_exec_runtime)
 273                 a->sum_exec_runtime = b->sum_exec_runtime;
 274 }
 275
 276 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 277 {
 278         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 279         struct task_cputime sum;
 280         unsigned long flags;
 281
 282         spin_lock_irqsave(&cputimer->lock, flags);
 283         if (!cputimer->running) {
 284                 cputimer->running = 1;
 285                 /*
 286                  * The POSIX timer interface allows for absolute time expiry
 287                  * values through the TIMER_ABSTIME flag, therefore we have
 288                  * to synchronize the timer to the clock every time we start
 289                  * it.
 290                  */
 291                 thread_group_cputime(tsk, &sum);
 292                 update_gt_cputime(&cputimer->cputime, &sum);
 293         }
 294         *times = cputimer->cputime;
 295         spin_unlock_irqrestore(&cputimer->lock, flags);
 296 }
 297
 298 /*
 299  * Sample a process (thread group) clock for the given group_leader task.
 300  * Must be called with tasklist_lock held for reading.
 301  */
 302 static int cpu_clock_sample_group(const clockid_t which_clock,
 303                                   struct task_struct *p,
 304                                   union cpu_time_count *cpu)
 305 {
 306         struct task_cputime cputime;
 307
 308         switch (CPUCLOCK_WHICH(which_clock)) {
 309         default:
 310                 return -EINVAL;
 311         case CPUCLOCK_PROF:
 312                 thread_group_cputime(p, &cputime);
 313                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 314                 break;
 315         case CPUCLOCK_VIRT:
 316                 thread_group_cputime(p, &cputime);
 317                 cpu->cpu = cputime.utime;
 318                 break;
 319         case CPUCLOCK_SCHED:
 320                 cpu->sched = thread_group_sched_runtime(p);
 321                 break;
 322         }
 323         return 0;
 324 }
 325
 326
 327 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 328 {
 329         const pid_t pid = CPUCLOCK_PID(which_clock);
 330         int error = -EINVAL;
 331         union cpu_time_count rtn;
 332
 333         if (pid == 0) {
 334                 /*
 335                  * Special case constant value for our own clocks.
 336                  * We don't have to do any lookup to find ourselves.
 337                  */
 338                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 339                         /*
 340                          * Sampling just ourselves we can do with no locking.
 341                          */
 342                         error = cpu_clock_sample(which_clock,
 343                                                  current, &rtn);
 344                 } else {
 345                         read_lock(&tasklist_lock);
 346                         error = cpu_clock_sample_group(which_clock,
 347                                                        current, &rtn);
 348                         read_unlock(&tasklist_lock);
 349                 }
 350         } else {
 351                 /*
 352                  * Find the given PID, and validate that the caller
 353                  * should be able to see it.
 354                  */
 355                 struct task_struct *p;
 356                 rcu_read_lock();
 357                 p = find_task_by_vpid(pid);
 358                 if (p) {
 359                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 360                                 if (same_thread_group(p, current)) {
 361                                         error = cpu_clock_sample(which_clock,
 362                                                                  p, &rtn);
 363                                 }
 364                         } else {
 365                                 read_lock(&tasklist_lock);
 366                                 if (thread_group_leader(p) && p->signal) {
 367                                         error =
 368                                             cpu_clock_sample_group(which_clock,
 369                                                                    p, &rtn);
 370                                 }
 371                                 read_unlock(&tasklist_lock);
 372                         }
 373                 }
 374                 rcu_read_unlock();
 375         }
 376
 377         if (error)
 378                 return error;
 379         sample_to_timespec(which_clock, rtn, tp);
 380         return 0;
 381 }
 382
 383
 384 /*
 385  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 386  * This is called from sys_timer_create with the new timer already locked.
 387  */
 388 int posix_cpu_timer_create(struct k_itimer *new_timer)
 389 {
 390         int ret = 0;
 391         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 392         struct task_struct *p;
 393
 394         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 395                 return -EINVAL;
 396
 397         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 398         new_timer->it.cpu.incr.sched = 0;
 399         new_timer->it.cpu.expires.sched = 0;
 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                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1097         }
1098
1099         if (!cputime_eq(it->expires, cputime_zero) &&
1100             (cputime_eq(*expires, cputime_zero) ||
1101              cputime_lt(it->expires, *expires))) {
1102                 *expires = it->expires;
1103         }
1104 }
1105
1106 /*
1107  * Check for any per-thread CPU timers that have fired and move them
1108  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1109  * have already been taken off.
1110  */
1111 static void check_process_timers(struct task_struct *tsk,
1112                                  struct list_head *firing)
1113 {
1114         int maxfire;
1115         struct signal_struct *const sig = tsk->signal;
1116         cputime_t utime, ptime, virt_expires, prof_expires;
1117         unsigned long long sum_sched_runtime, sched_expires;
1118         struct list_head *timers = sig->cpu_timers;
1119         struct task_cputime cputime;
1120
1121         /*
1122          * Don't sample the current process CPU clocks if there are no timers.
1123          */
1124         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1125             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1126             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1127             list_empty(&timers[CPUCLOCK_VIRT]) &&
1128             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1129             list_empty(&timers[CPUCLOCK_SCHED])) {
1130                 stop_process_timers(tsk);
1131                 return;
1132         }
1133
1134         /*
1135          * Collect the current process totals.
1136          */
1137         thread_group_cputimer(tsk, &cputime);
1138         utime = cputime.utime;
1139         ptime = cputime_add(utime, cputime.stime);
1140         sum_sched_runtime = cputime.sum_exec_runtime;
1141         maxfire = 20;
1142         prof_expires = cputime_zero;
1143         while (!list_empty(timers)) {
1144                 struct cpu_timer_list *tl = list_first_entry(timers,
1145                                                       struct cpu_timer_list,
1146                                                       entry);
1147                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1148                         prof_expires = tl->expires.cpu;
1149                         break;
1150                 }
1151                 tl->firing = 1;
1152                 list_move_tail(&tl->entry, firing);
1153         }
1154
1155         ++timers;
1156         maxfire = 20;
1157         virt_expires = cputime_zero;
1158         while (!list_empty(timers)) {
1159                 struct cpu_timer_list *tl = list_first_entry(timers,
1160                                                       struct cpu_timer_list,
1161                                                       entry);
1162                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1163                         virt_expires = tl->expires.cpu;
1164                         break;
1165                 }
1166                 tl->firing = 1;
1167                 list_move_tail(&tl->entry, firing);
1168         }
1169
1170         ++timers;
1171         maxfire = 20;
1172         sched_expires = 0;
1173         while (!list_empty(timers)) {
1174                 struct cpu_timer_list *tl = list_first_entry(timers,
1175                                                       struct cpu_timer_list,
1176                                                       entry);
1177                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1178                         sched_expires = tl->expires.sched;
1179                         break;
1180                 }
1181                 tl->firing = 1;
1182                 list_move_tail(&tl->entry, firing);
1183         }
1184
1185         /*
1186          * Check for the special case process timers.
1187          */
1188         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1189                          SIGPROF);
1190         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1191                          SIGVTALRM);
1192
1193         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1194                 unsigned long psecs = cputime_to_secs(ptime);
1195                 cputime_t x;
1196                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1197                         /*
1198                          * At the hard limit, we just die.
1199                          * No need to calculate anything else now.
1200                          */
1201                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1202                         return;
1203                 }
1204                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1205                         /*
1206                          * At the soft limit, send a SIGXCPU every second.
1207                          */
1208                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1209                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1210                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1211                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1212                         }
1213                 }
1214                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1215                 if (cputime_eq(prof_expires, cputime_zero) ||
1216                     cputime_lt(x, prof_expires)) {
1217                         prof_expires = x;
1218                 }
1219         }
1220
1221         if (!cputime_eq(prof_expires, cputime_zero) &&
1222             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1223              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1224                 sig->cputime_expires.prof_exp = prof_expires;
1225         if (!cputime_eq(virt_expires, cputime_zero) &&
1226             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1227              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1228                 sig->cputime_expires.virt_exp = virt_expires;
1229         if (sched_expires != 0 &&
1230             (sig->cputime_expires.sched_exp == 0 ||
1231              sig->cputime_expires.sched_exp > sched_expires))
1232                 sig->cputime_expires.sched_exp = sched_expires;
1233 }
1234
1235 /*
1236  * This is called from the signal code (via do_schedule_next_timer)
1237  * when the last timer signal was delivered and we have to reload the timer.
1238  */
1239 void posix_cpu_timer_schedule(struct k_itimer *timer)
1240 {
1241         struct task_struct *p = timer->it.cpu.task;
1242         union cpu_time_count now;
1243
1244         if (unlikely(p == NULL))
1245                 /*
1246                  * The task was cleaned up already, no future firings.
1247                  */
1248                 goto out;
1249
1250         /*
1251          * Fetch the current sample and update the timer's expiry time.
1252          */
1253         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1254                 cpu_clock_sample(timer->it_clock, p, &now);
1255                 bump_cpu_timer(timer, now);
1256                 if (unlikely(p->exit_state)) {
1257                         clear_dead_task(timer, now);
1258                         goto out;
1259                 }
1260                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1261         } else {
1262                 read_lock(&tasklist_lock);
1263                 if (unlikely(p->signal == NULL)) {
1264                         /*
1265                          * The process has been reaped.
1266                          * We can't even collect a sample any more.
1267                          */
1268                         put_task_struct(p);
1269                         timer->it.cpu.task = p = NULL;
1270                         timer->it.cpu.expires.sched = 0;
1271                         goto out_unlock;
1272                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1273                         /*
1274                          * We've noticed that the thread is dead, but
1275                          * not yet reaped.  Take this opportunity to
1276                          * drop our task ref.
1277                          */
1278                         clear_dead_task(timer, now);
1279                         goto out_unlock;
1280                 }
1281                 cpu_timer_sample_group(timer->it_clock, p, &now);
1282                 bump_cpu_timer(timer, now);
1283                 /* Leave the tasklist_lock locked for the call below.  */
1284         }
1285
1286         /*
1287          * Now re-arm for the new expiry time.
1288          */
1289         arm_timer(timer, now);
1290
1291 out_unlock:
1292         read_unlock(&tasklist_lock);
1293
1294 out:
1295         timer->it_overrun_last = timer->it_overrun;
1296         timer->it_overrun = -1;
1297         ++timer->it_requeue_pending;
1298 }
1299
1300 /**
1301  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1302  *
1303  * @cputime:    The struct to compare.
1304  *
1305  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1306  * are zero, false if any field is nonzero.
1307  */
1308 static inline int task_cputime_zero(const struct task_cputime *cputime)
1309 {
1310         if (cputime_eq(cputime->utime, cputime_zero) &&
1311             cputime_eq(cputime->stime, cputime_zero) &&
1312             cputime->sum_exec_runtime == 0)
1313                 return 1;
1314         return 0;
1315 }
1316
1317 /**
1318  * task_cputime_expired - Compare two task_cputime entities.
1319  *
1320  * @sample:     The task_cputime structure to be checked for expiration.
1321  * @expires:    Expiration times, against which @sample will be checked.
1322  *
1323  * Checks @sample against @expires to see if any field of @sample has expired.
1324  * Returns true if any field of the former is greater than the corresponding
1325  * field of the latter if the latter field is set.  Otherwise returns false.
1326  */
1327 static inline int task_cputime_expired(const struct task_cputime *sample,
1328                                         const struct task_cputime *expires)
1329 {
1330         if (!cputime_eq(expires->utime, cputime_zero) &&
1331             cputime_ge(sample->utime, expires->utime))
1332                 return 1;
1333         if (!cputime_eq(expires->stime, cputime_zero) &&
1334             cputime_ge(cputime_add(sample->utime, sample->stime),
1335                        expires->stime))
1336                 return 1;
1337         if (expires->sum_exec_runtime != 0 &&
1338             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1339                 return 1;
1340         return 0;
1341 }
1342
1343 /**
1344  * fastpath_timer_check - POSIX CPU timers fast path.
1345  *
1346  * @tsk:        The task (thread) being checked.
1347  *
1348  * Check the task and thread group timers.  If both are zero (there are no
1349  * timers set) return false.  Otherwise snapshot the task and thread group
1350  * timers and compare them with the corresponding expiration times.  Return
1351  * true if a timer has expired, else return false.
1352  */
1353 static inline int fastpath_timer_check(struct task_struct *tsk)
1354 {
1355         struct signal_struct *sig;
1356
1357         /* tsk == current, ensure it is safe to use ->signal/sighand */
1358         if (unlikely(tsk->exit_state))
1359                 return 0;
1360
1361         if (!task_cputime_zero(&tsk->cputime_expires)) {
1362                 struct task_cputime task_sample = {
1363                         .utime = tsk->utime,
1364                         .stime = tsk->stime,
1365                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1366                 };
1367
1368                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1369                         return 1;
1370         }
1371
1372         sig = tsk->signal;
1373         if (!task_cputime_zero(&sig->cputime_expires)) {
1374                 struct task_cputime group_sample;
1375
1376                 thread_group_cputimer(tsk, &group_sample);
1377                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1378                         return 1;
1379         }
1380
1381         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1382 }
1383
1384 /*
1385  * This is called from the timer interrupt handler.  The irq handler has
1386  * already updated our counts.  We need to check if any timers fire now.
1387  * Interrupts are disabled.
1388  */
1389 void run_posix_cpu_timers(struct task_struct *tsk)
1390 {
1391         LIST_HEAD(firing);
1392         struct k_itimer *timer, *next;
1393
1394         BUG_ON(!irqs_disabled());
1395
1396         /*
1397          * The fast path checks that there are no expired thread or thread
1398          * group timers.  If that's so, just return.
1399          */
1400         if (!fastpath_timer_check(tsk))
1401                 return;
1402
1403         spin_lock(&tsk->sighand->siglock);
1404         /*
1405          * Here we take off tsk->signal->cpu_timers[N] and
1406          * tsk->cpu_timers[N] all the timers that are firing, and
1407          * put them on the firing list.
1408          */
1409         check_thread_timers(tsk, &firing);
1410         check_process_timers(tsk, &firing);
1411
1412         /*
1413          * We must release these locks before taking any timer's lock.
1414          * There is a potential race with timer deletion here, as the
1415          * siglock now protects our private firing list.  We have set
1416          * the firing flag in each timer, so that a deletion attempt
1417          * that gets the timer lock before we do will give it up and
1418          * spin until we've taken care of that timer below.
1419          */
1420         spin_unlock(&tsk->sighand->siglock);
1421
1422         /*
1423          * Now that all the timers on our list have the firing flag,
1424          * noone will touch their list entries but us.  We'll take
1425          * each timer's lock before clearing its firing flag, so no
1426          * timer call will interfere.
1427          */
1428         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1429                 int cpu_firing;
1430
1431                 spin_lock(&timer->it_lock);
1432                 list_del_init(&timer->it.cpu.entry);
1433                 cpu_firing = timer->it.cpu.firing;
1434                 timer->it.cpu.firing = 0;
1435                 /*
1436                  * The firing flag is -1 if we collided with a reset
1437                  * of the timer, which already reported this
1438                  * almost-firing as an overrun.  So don't generate an event.
1439                  */
1440                 if (likely(cpu_firing >= 0))
1441                         cpu_timer_fire(timer);
1442                 spin_unlock(&timer->it_lock);
1443         }
1444 }
1445
1446 /*
1447  * Set one of the process-wide special case CPU timers.
1448  * The tsk->sighand->siglock must be held by the caller.
1449  * The *newval argument is relative and we update it to be absolute, *oldval
1450  * is absolute and we update it to be relative.
1451  */
1452 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1453                            cputime_t *newval, cputime_t *oldval)
1454 {
1455         union cpu_time_count now;
1456         struct list_head *head;
1457
1458         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1459         cpu_timer_sample_group(clock_idx, tsk, &now);
1460
1461         if (oldval) {
1462                 if (!cputime_eq(*oldval, cputime_zero)) {
1463                         if (cputime_le(*oldval, now.cpu)) {
1464                                 /* Just about to fire. */
1465                                 *oldval = cputime_one_jiffy;
1466                         } else {
1467                                 *oldval = cputime_sub(*oldval, now.cpu);
1468                         }
1469                 }
1470
1471                 if (cputime_eq(*newval, cputime_zero))
1472                         return;
1473                 *newval = cputime_add(*newval, now.cpu);
1474
1475                 /*
1476                  * If the RLIMIT_CPU timer will expire before the
1477                  * ITIMER_PROF timer, we have nothing else to do.
1478                  */
1479                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1480                     < cputime_to_secs(*newval))
1481                         return;
1482         }
1483
1484         /*
1485          * Check whether there are any process timers already set to fire
1486          * before this one.  If so, we don't have anything more to do.
1487          */
1488         head = &tsk->signal->cpu_timers[clock_idx];
1489         if (list_empty(head) ||
1490             cputime_ge(list_first_entry(head,
1491                                   struct cpu_timer_list, entry)->expires.cpu,
1492                        *newval)) {
1493                 switch (clock_idx) {
1494                 case CPUCLOCK_PROF:
1495                         tsk->signal->cputime_expires.prof_exp = *newval;
1496                         break;
1497                 case CPUCLOCK_VIRT:
1498                         tsk->signal->cputime_expires.virt_exp = *newval;
1499                         break;
1500                 }
1501         }
1502 }
1503
1504 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1505                             struct timespec *rqtp, struct itimerspec *it)
1506 {
1507         struct k_itimer timer;
1508         int error;
1509
1510         /*
1511          * Set up a temporary timer and then wait for it to go off.
1512          */
1513         memset(&timer, 0, sizeof timer);
1514         spin_lock_init(&timer.it_lock);
1515         timer.it_clock = which_clock;
1516         timer.it_overrun = -1;
1517         error = posix_cpu_timer_create(&timer);
1518         timer.it_process = current;
1519         if (!error) {
1520                 static struct itimerspec zero_it;
1521
1522                 memset(it, 0, sizeof *it);
1523                 it->it_value = *rqtp;
1524
1525                 spin_lock_irq(&timer.it_lock);
1526                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1527                 if (error) {
1528                         spin_unlock_irq(&timer.it_lock);
1529                         return error;
1530                 }
1531
1532                 while (!signal_pending(current)) {
1533                         if (timer.it.cpu.expires.sched == 0) {
1534                                 /*
1535                                  * Our timer fired and was reset.
1536                                  */
1537                                 spin_unlock_irq(&timer.it_lock);
1538                                 return 0;
1539                         }
1540
1541                         /*
1542                          * Block until cpu_timer_fire (or a signal) wakes us.
1543                          */
1544                         __set_current_state(TASK_INTERRUPTIBLE);
1545                         spin_unlock_irq(&timer.it_lock);
1546                         schedule();
1547                         spin_lock_irq(&timer.it_lock);
1548                 }
1549
1550                 /*
1551                  * We were interrupted by a signal.
1552                  */
1553                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1554                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1555                 spin_unlock_irq(&timer.it_lock);
1556
1557                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1558                         /*
1559                          * It actually did fire already.
1560                          */
1561                         return 0;
1562                 }
1563
1564                 error = -ERESTART_RESTARTBLOCK;
1565         }
1566
1567         return error;
1568 }
1569
1570 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1571                      struct timespec *rqtp, struct timespec __user *rmtp)
1572 {
1573         struct restart_block *restart_block =
1574             &current_thread_info()->restart_block;
1575         struct itimerspec it;
1576         int error;
1577
1578         /*
1579          * Diagnose required errors first.
1580          */
1581         if (CPUCLOCK_PERTHREAD(which_clock) &&
1582             (CPUCLOCK_PID(which_clock) == 0 ||
1583              CPUCLOCK_PID(which_clock) == current->pid))
1584                 return -EINVAL;
1585
1586         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1587
1588         if (error == -ERESTART_RESTARTBLOCK) {
1589
1590                 if (flags & TIMER_ABSTIME)
1591                         return -ERESTARTNOHAND;
1592                 /*
1593                  * Report back to the user the time still remaining.
1594                  */
1595                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1596                         return -EFAULT;
1597
1598                 restart_block->fn = posix_cpu_nsleep_restart;
1599                 restart_block->arg0 = which_clock;
1600                 restart_block->arg1 = (unsigned long) rmtp;
1601                 restart_block->arg2 = rqtp->tv_sec;
1602                 restart_block->arg3 = rqtp->tv_nsec;
1603         }
1604         return error;
1605 }
1606
1607 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1608 {
1609         clockid_t which_clock = restart_block->arg0;
1610         struct timespec __user *rmtp;
1611         struct timespec t;
1612         struct itimerspec it;
1613         int error;
1614
1615         rmtp = (struct timespec __user *) restart_block->arg1;
1616         t.tv_sec = restart_block->arg2;
1617         t.tv_nsec = restart_block->arg3;
1618
1619         restart_block->fn = do_no_restart_syscall;
1620         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1621
1622         if (error == -ERESTART_RESTARTBLOCK) {
1623                 /*
1624                  * Report back to the user the time still remaining.
1625                  */
1626                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1627                         return -EFAULT;
1628
1629                 restart_block->fn = posix_cpu_nsleep_restart;
1630                 restart_block->arg0 = which_clock;
1631                 restart_block->arg1 = (unsigned long) rmtp;
1632                 restart_block->arg2 = t.tv_sec;
1633                 restart_block->arg3 = t.tv_nsec;
1634         }
1635         return error;
1636
1637 }
1638
1639
1640 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1641 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1642
1643 static int process_cpu_clock_getres(const clockid_t which_clock,
1644                                     struct timespec *tp)
1645 {
1646         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1647 }
1648 static int process_cpu_clock_get(const clockid_t which_clock,
1649                                  struct timespec *tp)
1650 {
1651         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1652 }
1653 static int process_cpu_timer_create(struct k_itimer *timer)
1654 {
1655         timer->it_clock = PROCESS_CLOCK;
1656         return posix_cpu_timer_create(timer);
1657 }
1658 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1659                               struct timespec *rqtp,
1660                               struct timespec __user *rmtp)
1661 {
1662         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1663 }
1664 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1665 {
1666         return -EINVAL;
1667 }
1668 static int thread_cpu_clock_getres(const clockid_t which_clock,
1669                                    struct timespec *tp)
1670 {
1671         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1672 }
1673 static int thread_cpu_clock_get(const clockid_t which_clock,
1674                                 struct timespec *tp)
1675 {
1676         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1677 }
1678 static int thread_cpu_timer_create(struct k_itimer *timer)
1679 {
1680         timer->it_clock = THREAD_CLOCK;
1681         return posix_cpu_timer_create(timer);
1682 }
1683 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1684                               struct timespec *rqtp, struct timespec __user *rmtp)
1685 {
1686         return -EINVAL;
1687 }
1688 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1689 {
1690         return -EINVAL;
1691 }
1692
1693 static __init int init_posix_cpu_timers(void)
1694 {
1695         struct k_clock process = {
1696                 .clock_getres = process_cpu_clock_getres,
1697                 .clock_get = process_cpu_clock_get,
1698                 .clock_set = do_posix_clock_nosettime,
1699                 .timer_create = process_cpu_timer_create,
1700                 .nsleep = process_cpu_nsleep,
1701                 .nsleep_restart = process_cpu_nsleep_restart,
1702         };
1703         struct k_clock thread = {
1704                 .clock_getres = thread_cpu_clock_getres,
1705                 .clock_get = thread_cpu_clock_get,
1706                 .clock_set = do_posix_clock_nosettime,
1707                 .timer_create = thread_cpu_timer_create,
1708                 .nsleep = thread_cpu_nsleep,
1709                 .nsleep_restart = thread_cpu_nsleep_restart,
1710         };
1711         struct timespec ts;
1712
1713         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1714         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1715
1716         cputime_to_timespec(cputime_one_jiffy, &ts);
1717         onecputick = ts.tv_nsec;
1718         WARN_ON(ts.tv_sec != 0);
1719
1720         return 0;
1721 }
1722 __initcall(init_posix_cpu_timers);