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 task_cputime cputime;
 525
 526         thread_group_cputimer(tsk, &cputime);
 527         cleanup_timers(tsk->signal->cpu_timers,
 528                        cputime.utime, cputime.stime, cputime.sum_exec_runtime);
 529 }
 530
 531 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 532 {
 533         /*
 534          * That's all for this thread or process.
 535          * We leave our residual in expires to be reported.
 536          */
 537         put_task_struct(timer->it.cpu.task);
 538         timer->it.cpu.task = NULL;
 539         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 540                                              timer->it.cpu.expires,
 541                                              now);
 542 }
 543
 544 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 545 {
 546         return cputime_eq(expires, cputime_zero) ||
 547                cputime_gt(expires, new_exp);
 548 }
 549
 550 static inline int expires_le(cputime_t expires, cputime_t new_exp)
 551 {
 552         return !cputime_eq(expires, cputime_zero) &&
 553                cputime_le(expires, new_exp);
 554 }
 555 /*
 556  * Insert the timer on the appropriate list before any timers that
 557  * expire later.  This must be called with the tasklist_lock held
 558  * for reading, and interrupts disabled.
 559  */
 560 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 561 {
 562         struct task_struct *p = timer->it.cpu.task;
 563         struct list_head *head, *listpos;
 564         struct cpu_timer_list *const nt = &timer->it.cpu;
 565         struct cpu_timer_list *next;
 566         unsigned long i;
 567
 568         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 569                 p->cpu_timers : p->signal->cpu_timers);
 570         head += CPUCLOCK_WHICH(timer->it_clock);
 571
 572         BUG_ON(!irqs_disabled());
 573         spin_lock(&p->sighand->siglock);
 574
 575         listpos = head;
 576         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 577                 list_for_each_entry(next, head, entry) {
 578                         if (next->expires.sched > nt->expires.sched)
 579                                 break;
 580                         listpos = &next->entry;
 581                 }
 582         } else {
 583                 list_for_each_entry(next, head, entry) {
 584                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 585                                 break;
 586                         listpos = &next->entry;
 587                 }
 588         }
 589         list_add(&nt->entry, listpos);
 590
 591         if (listpos == head) {
 592                 /*
 593                  * We are the new earliest-expiring timer.
 594                  * If we are a thread timer, there can always
 595                  * be a process timer telling us to stop earlier.
 596                  */
 597
 598                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 599                         union cpu_time_count *exp = &nt->expires;
 600
 601                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 602                         default:
 603                                 BUG();
 604                         case CPUCLOCK_PROF:
 605                                 if (expires_gt(p->cputime_expires.prof_exp,
 606                                                exp->cpu))
 607                                         p->cputime_expires.prof_exp = exp->cpu;
 608                                 break;
 609                         case CPUCLOCK_VIRT:
 610                                 if (expires_gt(p->cputime_expires.virt_exp,
 611                                                exp->cpu))
 612                                         p->cputime_expires.virt_exp = exp->cpu;
 613                                 break;
 614                         case CPUCLOCK_SCHED:
 615                                 if (p->cputime_expires.sched_exp == 0 ||
 616                                     p->cputime_expires.sched_exp > exp->sched)
 617                                         p->cputime_expires.sched_exp =
 618                                                                 exp->sched;
 619                                 break;
 620                         }
 621                 } else {
 622                         struct signal_struct *const sig = p->signal;
 623                         union cpu_time_count *exp = &timer->it.cpu.expires;
 624
 625                         /*
 626                          * For a process timer, set the cached expiration time.
 627                          */
 628                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 629                         default:
 630                                 BUG();
 631                         case CPUCLOCK_VIRT:
 632                                 if (expires_le(sig->it[CPUCLOCK_VIRT].expires,
 633                                                exp->cpu))
 634                                         break;
 635                                 sig->cputime_expires.virt_exp = exp->cpu;
 636                                 break;
 637                         case CPUCLOCK_PROF:
 638                                 if (expires_le(sig->it[CPUCLOCK_PROF].expires,
 639                                                exp->cpu))
 640                                         break;
 641                                 i = sig->rlim[RLIMIT_CPU].rlim_cur;
 642                                 if (i != RLIM_INFINITY &&
 643                                     i <= cputime_to_secs(exp->cpu))
 644                                         break;
 645                                 sig->cputime_expires.prof_exp = exp->cpu;
 646                                 break;
 647                         case CPUCLOCK_SCHED:
 648                                 sig->cputime_expires.sched_exp = exp->sched;
 649                                 break;
 650                         }
 651                 }
 652         }
 653
 654         spin_unlock(&p->sighand->siglock);
 655 }
 656
 657 /*
 658  * The timer is locked, fire it and arrange for its reload.
 659  */
 660 static void cpu_timer_fire(struct k_itimer *timer)
 661 {
 662         if (unlikely(timer->sigq == NULL)) {
 663                 /*
 664                  * This a special case for clock_nanosleep,
 665                  * not a normal timer from sys_timer_create.
 666                  */
 667                 wake_up_process(timer->it_process);
 668                 timer->it.cpu.expires.sched = 0;
 669         } else if (timer->it.cpu.incr.sched == 0) {
 670                 /*
 671                  * One-shot timer.  Clear it as soon as it's fired.
 672                  */
 673                 posix_timer_event(timer, 0);
 674                 timer->it.cpu.expires.sched = 0;
 675         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 676                 /*
 677                  * The signal did not get queued because the signal
 678                  * was ignored, so we won't get any callback to
 679                  * reload the timer.  But we need to keep it
 680                  * ticking in case the signal is deliverable next time.
 681                  */
 682                 posix_cpu_timer_schedule(timer);
 683         }
 684 }
 685
 686 /*
 687  * Sample a process (thread group) timer for the given group_leader task.
 688  * Must be called with tasklist_lock held for reading.
 689  */
 690 static int cpu_timer_sample_group(const clockid_t which_clock,
 691                                   struct task_struct *p,
 692                                   union cpu_time_count *cpu)
 693 {
 694         struct task_cputime cputime;
 695
 696         thread_group_cputimer(p, &cputime);
 697         switch (CPUCLOCK_WHICH(which_clock)) {
 698         default:
 699                 return -EINVAL;
 700         case CPUCLOCK_PROF:
 701                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 702                 break;
 703         case CPUCLOCK_VIRT:
 704                 cpu->cpu = cputime.utime;
 705                 break;
 706         case CPUCLOCK_SCHED:
 707                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 708                 break;
 709         }
 710         return 0;
 711 }
 712
 713 /*
 714  * Guts of sys_timer_settime for CPU timers.
 715  * This is called with the timer locked and interrupts disabled.
 716  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 717  * and try again.  (This happens when the timer is in the middle of firing.)
 718  */
 719 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 720                         struct itimerspec *new, struct itimerspec *old)
 721 {
 722         struct task_struct *p = timer->it.cpu.task;
 723         union cpu_time_count old_expires, new_expires, val;
 724         int ret;
 725
 726         if (unlikely(p == NULL)) {
 727                 /*
 728                  * Timer refers to a dead task's clock.
 729                  */
 730                 return -ESRCH;
 731         }
 732
 733         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 734
 735         read_lock(&tasklist_lock);
 736         /*
 737          * We need the tasklist_lock to protect against reaping that
 738          * clears p->signal.  If p has just been reaped, we can no
 739          * longer get any information about it at all.
 740          */
 741         if (unlikely(p->signal == NULL)) {
 742                 read_unlock(&tasklist_lock);
 743                 put_task_struct(p);
 744                 timer->it.cpu.task = NULL;
 745                 return -ESRCH;
 746         }
 747
 748         /*
 749          * Disarm any old timer after extracting its expiry time.
 750          */
 751         BUG_ON(!irqs_disabled());
 752
 753         ret = 0;
 754         spin_lock(&p->sighand->siglock);
 755         old_expires = timer->it.cpu.expires;
 756         if (unlikely(timer->it.cpu.firing)) {
 757                 timer->it.cpu.firing = -1;
 758                 ret = TIMER_RETRY;
 759         } else
 760                 list_del_init(&timer->it.cpu.entry);
 761         spin_unlock(&p->sighand->siglock);
 762
 763         /*
 764          * We need to sample the current value to convert the new
 765          * value from to relative and absolute, and to convert the
 766          * old value from absolute to relative.  To set a process
 767          * timer, we need a sample to balance the thread expiry
 768          * times (in arm_timer).  With an absolute time, we must
 769          * check if it's already passed.  In short, we need a sample.
 770          */
 771         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 772                 cpu_clock_sample(timer->it_clock, p, &val);
 773         } else {
 774                 cpu_timer_sample_group(timer->it_clock, p, &val);
 775         }
 776
 777         if (old) {
 778                 if (old_expires.sched == 0) {
 779                         old->it_value.tv_sec = 0;
 780                         old->it_value.tv_nsec = 0;
 781                 } else {
 782                         /*
 783                          * Update the timer in case it has
 784                          * overrun already.  If it has,
 785                          * we'll report it as having overrun
 786                          * and with the next reloaded timer
 787                          * already ticking, though we are
 788                          * swallowing that pending
 789                          * notification here to install the
 790                          * new setting.
 791                          */
 792                         bump_cpu_timer(timer, val);
 793                         if (cpu_time_before(timer->it_clock, val,
 794                                             timer->it.cpu.expires)) {
 795                                 old_expires = cpu_time_sub(
 796                                         timer->it_clock,
 797                                         timer->it.cpu.expires, val);
 798                                 sample_to_timespec(timer->it_clock,
 799                                                    old_expires,
 800                                                    &old->it_value);
 801                         } else {
 802                                 old->it_value.tv_nsec = 1;
 803                                 old->it_value.tv_sec = 0;
 804                         }
 805                 }
 806         }
 807
 808         if (unlikely(ret)) {
 809                 /*
 810                  * We are colliding with the timer actually firing.
 811                  * Punt after filling in the timer's old value, and
 812                  * disable this firing since we are already reporting
 813                  * it as an overrun (thanks to bump_cpu_timer above).
 814                  */
 815                 read_unlock(&tasklist_lock);
 816                 goto out;
 817         }
 818
 819         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 820                 cpu_time_add(timer->it_clock, &new_expires, val);
 821         }
 822
 823         /*
 824          * Install the new expiry time (or zero).
 825          * For a timer with no notification action, we don't actually
 826          * arm the timer (we'll just fake it for timer_gettime).
 827          */
 828         timer->it.cpu.expires = new_expires;
 829         if (new_expires.sched != 0 &&
 830             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 831             cpu_time_before(timer->it_clock, val, new_expires)) {
 832                 arm_timer(timer, val);
 833         }
 834
 835         read_unlock(&tasklist_lock);
 836
 837         /*
 838          * Install the new reload setting, and
 839          * set up the signal and overrun bookkeeping.
 840          */
 841         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 842                                                 &new->it_interval);
 843
 844         /*
 845          * This acts as a modification timestamp for the timer,
 846          * so any automatic reload attempt will punt on seeing
 847          * that we have reset the timer manually.
 848          */
 849         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 850                 ~REQUEUE_PENDING;
 851         timer->it_overrun_last = 0;
 852         timer->it_overrun = -1;
 853
 854         if (new_expires.sched != 0 &&
 855             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 856             !cpu_time_before(timer->it_clock, val, new_expires)) {
 857                 /*
 858                  * The designated time already passed, so we notify
 859                  * immediately, even if the thread never runs to
 860                  * accumulate more time on this clock.
 861                  */
 862                 cpu_timer_fire(timer);
 863         }
 864
 865         ret = 0;
 866  out:
 867         if (old) {
 868                 sample_to_timespec(timer->it_clock,
 869                                    timer->it.cpu.incr, &old->it_interval);
 870         }
 871         return ret;
 872 }
 873
 874 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 875 {
 876         union cpu_time_count now;
 877         struct task_struct *p = timer->it.cpu.task;
 878         int clear_dead;
 879
 880         /*
 881          * Easy part: convert the reload time.
 882          */
 883         sample_to_timespec(timer->it_clock,
 884                            timer->it.cpu.incr, &itp->it_interval);
 885
 886         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 887                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 888                 return;
 889         }
 890
 891         if (unlikely(p == NULL)) {
 892                 /*
 893                  * This task already died and the timer will never fire.
 894                  * In this case, expires is actually the dead value.
 895                  */
 896         dead:
 897                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 898                                    &itp->it_value);
 899                 return;
 900         }
 901
 902         /*
 903          * Sample the clock to take the difference with the expiry time.
 904          */
 905         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 906                 cpu_clock_sample(timer->it_clock, p, &now);
 907                 clear_dead = p->exit_state;
 908         } else {
 909                 read_lock(&tasklist_lock);
 910                 if (unlikely(p->signal == NULL)) {
 911                         /*
 912                          * The process has been reaped.
 913                          * We can't even collect a sample any more.
 914                          * Call the timer disarmed, nothing else to do.
 915                          */
 916                         put_task_struct(p);
 917                         timer->it.cpu.task = NULL;
 918                         timer->it.cpu.expires.sched = 0;
 919                         read_unlock(&tasklist_lock);
 920                         goto dead;
 921                 } else {
 922                         cpu_timer_sample_group(timer->it_clock, p, &now);
 923                         clear_dead = (unlikely(p->exit_state) &&
 924                                       thread_group_empty(p));
 925                 }
 926                 read_unlock(&tasklist_lock);
 927         }
 928
 929         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 930                 if (timer->it.cpu.incr.sched == 0 &&
 931                     cpu_time_before(timer->it_clock,
 932                                     timer->it.cpu.expires, now)) {
 933                         /*
 934                          * Do-nothing timer expired and has no reload,
 935                          * so it's as if it was never set.
 936                          */
 937                         timer->it.cpu.expires.sched = 0;
 938                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 939                         return;
 940                 }
 941                 /*
 942                  * Account for any expirations and reloads that should
 943                  * have happened.
 944                  */
 945                 bump_cpu_timer(timer, now);
 946         }
 947
 948         if (unlikely(clear_dead)) {
 949                 /*
 950                  * We've noticed that the thread is dead, but
 951                  * not yet reaped.  Take this opportunity to
 952                  * drop our task ref.
 953                  */
 954                 clear_dead_task(timer, now);
 955                 goto dead;
 956         }
 957
 958         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 959                 sample_to_timespec(timer->it_clock,
 960                                    cpu_time_sub(timer->it_clock,
 961                                                 timer->it.cpu.expires, now),
 962                                    &itp->it_value);
 963         } else {
 964                 /*
 965                  * The timer should have expired already, but the firing
 966                  * hasn't taken place yet.  Say it's just about to expire.
 967                  */
 968                 itp->it_value.tv_nsec = 1;
 969                 itp->it_value.tv_sec = 0;
 970         }
 971 }
 972
 973 /*
 974  * Check for any per-thread CPU timers that have fired and move them off
 975  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 976  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 977  */
 978 static void check_thread_timers(struct task_struct *tsk,
 979                                 struct list_head *firing)
 980 {
 981         int maxfire;
 982         struct list_head *timers = tsk->cpu_timers;
 983         struct signal_struct *const sig = tsk->signal;
 984
 985         maxfire = 20;
 986         tsk->cputime_expires.prof_exp = cputime_zero;
 987         while (!list_empty(timers)) {
 988                 struct cpu_timer_list *t = list_first_entry(timers,
 989                                                       struct cpu_timer_list,
 990                                                       entry);
 991                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 992                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 993                         break;
 994                 }
 995                 t->firing = 1;
 996                 list_move_tail(&t->entry, firing);
 997         }
 998
 999         ++timers;
1000         maxfire = 20;
1001         tsk->cputime_expires.virt_exp = cputime_zero;
1002         while (!list_empty(timers)) {
1003                 struct cpu_timer_list *t = list_first_entry(timers,
1004                                                       struct cpu_timer_list,
1005                                                       entry);
1006                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1007                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1008                         break;
1009                 }
1010                 t->firing = 1;
1011                 list_move_tail(&t->entry, firing);
1012         }
1013
1014         ++timers;
1015         maxfire = 20;
1016         tsk->cputime_expires.sched_exp = 0;
1017         while (!list_empty(timers)) {
1018                 struct cpu_timer_list *t = list_first_entry(timers,
1019                                                       struct cpu_timer_list,
1020                                                       entry);
1021                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1022                         tsk->cputime_expires.sched_exp = t->expires.sched;
1023                         break;
1024                 }
1025                 t->firing = 1;
1026                 list_move_tail(&t->entry, firing);
1027         }
1028
1029         /*
1030          * Check for the special case thread timers.
1031          */
1032         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1033                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1034                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1035
1036                 if (hard != RLIM_INFINITY &&
1037                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1038                         /*
1039                          * At the hard limit, we just die.
1040                          * No need to calculate anything else now.
1041                          */
1042                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1043                         return;
1044                 }
1045                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1046                         /*
1047                          * At the soft limit, send a SIGXCPU every second.
1048                          */
1049                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1050                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1051                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1052                                                                 USEC_PER_SEC;
1053                         }
1054                         printk(KERN_INFO
1055                                 "RT Watchdog Timeout: %s[%d]\n",
1056                                 tsk->comm, task_pid_nr(tsk));
1057                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1058                 }
1059         }
1060 }
1061
1062 static void stop_process_timers(struct task_struct *tsk)
1063 {
1064         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1065         unsigned long flags;
1066
1067         if (!cputimer->running)
1068                 return;
1069
1070         spin_lock_irqsave(&cputimer->lock, flags);
1071         cputimer->running = 0;
1072         spin_unlock_irqrestore(&cputimer->lock, flags);
1073 }
1074
1075 static u32 onecputick;
1076
1077 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1078                              cputime_t *expires, cputime_t cur_time, int signo)
1079 {
1080         if (cputime_eq(it->expires, cputime_zero))
1081                 return;
1082
1083         if (cputime_ge(cur_time, it->expires)) {
1084                 if (!cputime_eq(it->incr, cputime_zero)) {
1085                         it->expires = cputime_add(it->expires, it->incr);
1086                         it->error += it->incr_error;
1087                         if (it->error >= onecputick) {
1088                                 it->expires = cputime_sub(it->expires,
1089                                                           cputime_one_jiffy);
1090                                 it->error -= onecputick;
1091                         }
1092                 } else
1093                         it->expires = cputime_zero;
1094
1095                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1096         }
1097
1098         if (!cputime_eq(it->expires, cputime_zero) &&
1099             (cputime_eq(*expires, cputime_zero) ||
1100              cputime_lt(it->expires, *expires))) {
1101                 *expires = it->expires;
1102         }
1103 }
1104
1105 /*
1106  * Check for any per-thread CPU timers that have fired and move them
1107  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1108  * have already been taken off.
1109  */
1110 static void check_process_timers(struct task_struct *tsk,
1111                                  struct list_head *firing)
1112 {
1113         int maxfire;
1114         struct signal_struct *const sig = tsk->signal;
1115         cputime_t utime, ptime, virt_expires, prof_expires;
1116         unsigned long long sum_sched_runtime, sched_expires;
1117         struct list_head *timers = sig->cpu_timers;
1118         struct task_cputime cputime;
1119
1120         /*
1121          * Don't sample the current process CPU clocks if there are no timers.
1122          */
1123         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1124             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1125             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1126             list_empty(&timers[CPUCLOCK_VIRT]) &&
1127             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1128             list_empty(&timers[CPUCLOCK_SCHED])) {
1129                 stop_process_timers(tsk);
1130                 return;
1131         }
1132
1133         /*
1134          * Collect the current process totals.
1135          */
1136         thread_group_cputimer(tsk, &cputime);
1137         utime = cputime.utime;
1138         ptime = cputime_add(utime, cputime.stime);
1139         sum_sched_runtime = cputime.sum_exec_runtime;
1140         maxfire = 20;
1141         prof_expires = cputime_zero;
1142         while (!list_empty(timers)) {
1143                 struct cpu_timer_list *tl = list_first_entry(timers,
1144                                                       struct cpu_timer_list,
1145                                                       entry);
1146                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1147                         prof_expires = tl->expires.cpu;
1148                         break;
1149                 }
1150                 tl->firing = 1;
1151                 list_move_tail(&tl->entry, firing);
1152         }
1153
1154         ++timers;
1155         maxfire = 20;
1156         virt_expires = cputime_zero;
1157         while (!list_empty(timers)) {
1158                 struct cpu_timer_list *tl = list_first_entry(timers,
1159                                                       struct cpu_timer_list,
1160                                                       entry);
1161                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1162                         virt_expires = tl->expires.cpu;
1163                         break;
1164                 }
1165                 tl->firing = 1;
1166                 list_move_tail(&tl->entry, firing);
1167         }
1168
1169         ++timers;
1170         maxfire = 20;
1171         sched_expires = 0;
1172         while (!list_empty(timers)) {
1173                 struct cpu_timer_list *tl = list_first_entry(timers,
1174                                                       struct cpu_timer_list,
1175                                                       entry);
1176                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1177                         sched_expires = tl->expires.sched;
1178                         break;
1179                 }
1180                 tl->firing = 1;
1181                 list_move_tail(&tl->entry, firing);
1182         }
1183
1184         /*
1185          * Check for the special case process timers.
1186          */
1187         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1188                          SIGPROF);
1189         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1190                          SIGVTALRM);
1191
1192         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1193                 unsigned long psecs = cputime_to_secs(ptime);
1194                 cputime_t x;
1195                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1196                         /*
1197                          * At the hard limit, we just die.
1198                          * No need to calculate anything else now.
1199                          */
1200                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1201                         return;
1202                 }
1203                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1204                         /*
1205                          * At the soft limit, send a SIGXCPU every second.
1206                          */
1207                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1208                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1209                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1210                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1211                         }
1212                 }
1213                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1214                 if (cputime_eq(prof_expires, cputime_zero) ||
1215                     cputime_lt(x, prof_expires)) {
1216                         prof_expires = x;
1217                 }
1218         }
1219
1220         if (!cputime_eq(prof_expires, cputime_zero) &&
1221             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1222              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1223                 sig->cputime_expires.prof_exp = prof_expires;
1224         if (!cputime_eq(virt_expires, cputime_zero) &&
1225             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1226              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1227                 sig->cputime_expires.virt_exp = virt_expires;
1228         if (sched_expires != 0 &&
1229             (sig->cputime_expires.sched_exp == 0 ||
1230              sig->cputime_expires.sched_exp > sched_expires))
1231                 sig->cputime_expires.sched_exp = sched_expires;
1232 }
1233
1234 /*
1235  * This is called from the signal code (via do_schedule_next_timer)
1236  * when the last timer signal was delivered and we have to reload the timer.
1237  */
1238 void posix_cpu_timer_schedule(struct k_itimer *timer)
1239 {
1240         struct task_struct *p = timer->it.cpu.task;
1241         union cpu_time_count now;
1242
1243         if (unlikely(p == NULL))
1244                 /*
1245                  * The task was cleaned up already, no future firings.
1246                  */
1247                 goto out;
1248
1249         /*
1250          * Fetch the current sample and update the timer's expiry time.
1251          */
1252         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1253                 cpu_clock_sample(timer->it_clock, p, &now);
1254                 bump_cpu_timer(timer, now);
1255                 if (unlikely(p->exit_state)) {
1256                         clear_dead_task(timer, now);
1257                         goto out;
1258                 }
1259                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1260         } else {
1261                 read_lock(&tasklist_lock);
1262                 if (unlikely(p->signal == NULL)) {
1263                         /*
1264                          * The process has been reaped.
1265                          * We can't even collect a sample any more.
1266                          */
1267                         put_task_struct(p);
1268                         timer->it.cpu.task = p = NULL;
1269                         timer->it.cpu.expires.sched = 0;
1270                         goto out_unlock;
1271                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1272                         /*
1273                          * We've noticed that the thread is dead, but
1274                          * not yet reaped.  Take this opportunity to
1275                          * drop our task ref.
1276                          */
1277                         clear_dead_task(timer, now);
1278                         goto out_unlock;
1279                 }
1280                 cpu_timer_sample_group(timer->it_clock, p, &now);
1281                 bump_cpu_timer(timer, now);
1282                 /* Leave the tasklist_lock locked for the call below.  */
1283         }
1284
1285         /*
1286          * Now re-arm for the new expiry time.
1287          */
1288         arm_timer(timer, now);
1289
1290 out_unlock:
1291         read_unlock(&tasklist_lock);
1292
1293 out:
1294         timer->it_overrun_last = timer->it_overrun;
1295         timer->it_overrun = -1;
1296         ++timer->it_requeue_pending;
1297 }
1298
1299 /**
1300  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1301  *
1302  * @cputime:    The struct to compare.
1303  *
1304  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1305  * are zero, false if any field is nonzero.
1306  */
1307 static inline int task_cputime_zero(const struct task_cputime *cputime)
1308 {
1309         if (cputime_eq(cputime->utime, cputime_zero) &&
1310             cputime_eq(cputime->stime, cputime_zero) &&
1311             cputime->sum_exec_runtime == 0)
1312                 return 1;
1313         return 0;
1314 }
1315
1316 /**
1317  * task_cputime_expired - Compare two task_cputime entities.
1318  *
1319  * @sample:     The task_cputime structure to be checked for expiration.
1320  * @expires:    Expiration times, against which @sample will be checked.
1321  *
1322  * Checks @sample against @expires to see if any field of @sample has expired.
1323  * Returns true if any field of the former is greater than the corresponding
1324  * field of the latter if the latter field is set.  Otherwise returns false.
1325  */
1326 static inline int task_cputime_expired(const struct task_cputime *sample,
1327                                         const struct task_cputime *expires)
1328 {
1329         if (!cputime_eq(expires->utime, cputime_zero) &&
1330             cputime_ge(sample->utime, expires->utime))
1331                 return 1;
1332         if (!cputime_eq(expires->stime, cputime_zero) &&
1333             cputime_ge(cputime_add(sample->utime, sample->stime),
1334                        expires->stime))
1335                 return 1;
1336         if (expires->sum_exec_runtime != 0 &&
1337             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1338                 return 1;
1339         return 0;
1340 }
1341
1342 /**
1343  * fastpath_timer_check - POSIX CPU timers fast path.
1344  *
1345  * @tsk:        The task (thread) being checked.
1346  *
1347  * Check the task and thread group timers.  If both are zero (there are no
1348  * timers set) return false.  Otherwise snapshot the task and thread group
1349  * timers and compare them with the corresponding expiration times.  Return
1350  * true if a timer has expired, else return false.
1351  */
1352 static inline int fastpath_timer_check(struct task_struct *tsk)
1353 {
1354         struct signal_struct *sig;
1355
1356         /* tsk == current, ensure it is safe to use ->signal/sighand */
1357         if (unlikely(tsk->exit_state))
1358                 return 0;
1359
1360         if (!task_cputime_zero(&tsk->cputime_expires)) {
1361                 struct task_cputime task_sample = {
1362                         .utime = tsk->utime,
1363                         .stime = tsk->stime,
1364                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1365                 };
1366
1367                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1368                         return 1;
1369         }
1370
1371         sig = tsk->signal;
1372         if (!task_cputime_zero(&sig->cputime_expires)) {
1373                 struct task_cputime group_sample;
1374
1375                 thread_group_cputimer(tsk, &group_sample);
1376                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1377                         return 1;
1378         }
1379
1380         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1381 }
1382
1383 /*
1384  * This is called from the timer interrupt handler.  The irq handler has
1385  * already updated our counts.  We need to check if any timers fire now.
1386  * Interrupts are disabled.
1387  */
1388 void run_posix_cpu_timers(struct task_struct *tsk)
1389 {
1390         LIST_HEAD(firing);
1391         struct k_itimer *timer, *next;
1392
1393         BUG_ON(!irqs_disabled());
1394
1395         /*
1396          * The fast path checks that there are no expired thread or thread
1397          * group timers.  If that's so, just return.
1398          */
1399         if (!fastpath_timer_check(tsk))
1400                 return;
1401
1402         spin_lock(&tsk->sighand->siglock);
1403         /*
1404          * Here we take off tsk->signal->cpu_timers[N] and
1405          * tsk->cpu_timers[N] all the timers that are firing, and
1406          * put them on the firing list.
1407          */
1408         check_thread_timers(tsk, &firing);
1409         check_process_timers(tsk, &firing);
1410
1411         /*
1412          * We must release these locks before taking any timer's lock.
1413          * There is a potential race with timer deletion here, as the
1414          * siglock now protects our private firing list.  We have set
1415          * the firing flag in each timer, so that a deletion attempt
1416          * that gets the timer lock before we do will give it up and
1417          * spin until we've taken care of that timer below.
1418          */
1419         spin_unlock(&tsk->sighand->siglock);
1420
1421         /*
1422          * Now that all the timers on our list have the firing flag,
1423          * noone will touch their list entries but us.  We'll take
1424          * each timer's lock before clearing its firing flag, so no
1425          * timer call will interfere.
1426          */
1427         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1428                 int cpu_firing;
1429
1430                 spin_lock(&timer->it_lock);
1431                 list_del_init(&timer->it.cpu.entry);
1432                 cpu_firing = timer->it.cpu.firing;
1433                 timer->it.cpu.firing = 0;
1434                 /*
1435                  * The firing flag is -1 if we collided with a reset
1436                  * of the timer, which already reported this
1437                  * almost-firing as an overrun.  So don't generate an event.
1438                  */
1439                 if (likely(cpu_firing >= 0))
1440                         cpu_timer_fire(timer);
1441                 spin_unlock(&timer->it_lock);
1442         }
1443 }
1444
1445 /*
1446  * Set one of the process-wide special case CPU timers.
1447  * The tsk->sighand->siglock must be held by the caller.
1448  * The *newval argument is relative and we update it to be absolute, *oldval
1449  * is absolute and we update it to be relative.
1450  */
1451 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1452                            cputime_t *newval, cputime_t *oldval)
1453 {
1454         union cpu_time_count now;
1455         struct list_head *head;
1456
1457         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1458         cpu_timer_sample_group(clock_idx, tsk, &now);
1459
1460         if (oldval) {
1461                 if (!cputime_eq(*oldval, cputime_zero)) {
1462                         if (cputime_le(*oldval, now.cpu)) {
1463                                 /* Just about to fire. */
1464                                 *oldval = cputime_one_jiffy;
1465                         } else {
1466                                 *oldval = cputime_sub(*oldval, now.cpu);
1467                         }
1468                 }
1469
1470                 if (cputime_eq(*newval, cputime_zero))
1471                         return;
1472                 *newval = cputime_add(*newval, now.cpu);
1473
1474                 /*
1475                  * If the RLIMIT_CPU timer will expire before the
1476                  * ITIMER_PROF timer, we have nothing else to do.
1477                  */
1478                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1479                     < cputime_to_secs(*newval))
1480                         return;
1481         }
1482
1483         /*
1484          * Check whether there are any process timers already set to fire
1485          * before this one.  If so, we don't have anything more to do.
1486          */
1487         head = &tsk->signal->cpu_timers[clock_idx];
1488         if (list_empty(head) ||
1489             cputime_ge(list_first_entry(head,
1490                                   struct cpu_timer_list, entry)->expires.cpu,
1491                        *newval)) {
1492                 switch (clock_idx) {
1493                 case CPUCLOCK_PROF:
1494                         tsk->signal->cputime_expires.prof_exp = *newval;
1495                         break;
1496                 case CPUCLOCK_VIRT:
1497                         tsk->signal->cputime_expires.virt_exp = *newval;
1498                         break;
1499                 }
1500         }
1501 }
1502
1503 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1504                             struct timespec *rqtp, struct itimerspec *it)
1505 {
1506         struct k_itimer timer;
1507         int error;
1508
1509         /*
1510          * Set up a temporary timer and then wait for it to go off.
1511          */
1512         memset(&timer, 0, sizeof timer);
1513         spin_lock_init(&timer.it_lock);
1514         timer.it_clock = which_clock;
1515         timer.it_overrun = -1;
1516         error = posix_cpu_timer_create(&timer);
1517         timer.it_process = current;
1518         if (!error) {
1519                 static struct itimerspec zero_it;
1520
1521                 memset(it, 0, sizeof *it);
1522                 it->it_value = *rqtp;
1523
1524                 spin_lock_irq(&timer.it_lock);
1525                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1526                 if (error) {
1527                         spin_unlock_irq(&timer.it_lock);
1528                         return error;
1529                 }
1530
1531                 while (!signal_pending(current)) {
1532                         if (timer.it.cpu.expires.sched == 0) {
1533                                 /*
1534                                  * Our timer fired and was reset.
1535                                  */
1536                                 spin_unlock_irq(&timer.it_lock);
1537                                 return 0;
1538                         }
1539
1540                         /*
1541                          * Block until cpu_timer_fire (or a signal) wakes us.
1542                          */
1543                         __set_current_state(TASK_INTERRUPTIBLE);
1544                         spin_unlock_irq(&timer.it_lock);
1545                         schedule();
1546                         spin_lock_irq(&timer.it_lock);
1547                 }
1548
1549                 /*
1550                  * We were interrupted by a signal.
1551                  */
1552                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1553                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1554                 spin_unlock_irq(&timer.it_lock);
1555
1556                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1557                         /*
1558                          * It actually did fire already.
1559                          */
1560                         return 0;
1561                 }
1562
1563                 error = -ERESTART_RESTARTBLOCK;
1564         }
1565
1566         return error;
1567 }
1568
1569 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1570                      struct timespec *rqtp, struct timespec __user *rmtp)
1571 {
1572         struct restart_block *restart_block =
1573             &current_thread_info()->restart_block;
1574         struct itimerspec it;
1575         int error;
1576
1577         /*
1578          * Diagnose required errors first.
1579          */
1580         if (CPUCLOCK_PERTHREAD(which_clock) &&
1581             (CPUCLOCK_PID(which_clock) == 0 ||
1582              CPUCLOCK_PID(which_clock) == current->pid))
1583                 return -EINVAL;
1584
1585         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1586
1587         if (error == -ERESTART_RESTARTBLOCK) {
1588
1589                 if (flags & TIMER_ABSTIME)
1590                         return -ERESTARTNOHAND;
1591                 /*
1592                  * Report back to the user the time still remaining.
1593                  */
1594                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1595                         return -EFAULT;
1596
1597                 restart_block->fn = posix_cpu_nsleep_restart;
1598                 restart_block->arg0 = which_clock;
1599                 restart_block->arg1 = (unsigned long) rmtp;
1600                 restart_block->arg2 = rqtp->tv_sec;
1601                 restart_block->arg3 = rqtp->tv_nsec;
1602         }
1603         return error;
1604 }
1605
1606 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1607 {
1608         clockid_t which_clock = restart_block->arg0;
1609         struct timespec __user *rmtp;
1610         struct timespec t;
1611         struct itimerspec it;
1612         int error;
1613
1614         rmtp = (struct timespec __user *) restart_block->arg1;
1615         t.tv_sec = restart_block->arg2;
1616         t.tv_nsec = restart_block->arg3;
1617
1618         restart_block->fn = do_no_restart_syscall;
1619         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1620
1621         if (error == -ERESTART_RESTARTBLOCK) {
1622                 /*
1623                  * Report back to the user the time still remaining.
1624                  */
1625                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1626                         return -EFAULT;
1627
1628                 restart_block->fn = posix_cpu_nsleep_restart;
1629                 restart_block->arg0 = which_clock;
1630                 restart_block->arg1 = (unsigned long) rmtp;
1631                 restart_block->arg2 = t.tv_sec;
1632                 restart_block->arg3 = t.tv_nsec;
1633         }
1634         return error;
1635
1636 }
1637
1638
1639 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1640 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1641
1642 static int process_cpu_clock_getres(const clockid_t which_clock,
1643                                     struct timespec *tp)
1644 {
1645         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1646 }
1647 static int process_cpu_clock_get(const clockid_t which_clock,
1648                                  struct timespec *tp)
1649 {
1650         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1651 }
1652 static int process_cpu_timer_create(struct k_itimer *timer)
1653 {
1654         timer->it_clock = PROCESS_CLOCK;
1655         return posix_cpu_timer_create(timer);
1656 }
1657 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1658                               struct timespec *rqtp,
1659                               struct timespec __user *rmtp)
1660 {
1661         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1662 }
1663 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1664 {
1665         return -EINVAL;
1666 }
1667 static int thread_cpu_clock_getres(const clockid_t which_clock,
1668                                    struct timespec *tp)
1669 {
1670         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1671 }
1672 static int thread_cpu_clock_get(const clockid_t which_clock,
1673                                 struct timespec *tp)
1674 {
1675         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1676 }
1677 static int thread_cpu_timer_create(struct k_itimer *timer)
1678 {
1679         timer->it_clock = THREAD_CLOCK;
1680         return posix_cpu_timer_create(timer);
1681 }
1682 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1683                               struct timespec *rqtp, struct timespec __user *rmtp)
1684 {
1685         return -EINVAL;
1686 }
1687 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1688 {
1689         return -EINVAL;
1690 }
1691
1692 static __init int init_posix_cpu_timers(void)
1693 {
1694         struct k_clock process = {
1695                 .clock_getres = process_cpu_clock_getres,
1696                 .clock_get = process_cpu_clock_get,
1697                 .clock_set = do_posix_clock_nosettime,
1698                 .timer_create = process_cpu_timer_create,
1699                 .nsleep = process_cpu_nsleep,
1700                 .nsleep_restart = process_cpu_nsleep_restart,
1701         };
1702         struct k_clock thread = {
1703                 .clock_getres = thread_cpu_clock_getres,
1704                 .clock_get = thread_cpu_clock_get,
1705                 .clock_set = do_posix_clock_nosettime,
1706                 .timer_create = thread_cpu_timer_create,
1707                 .nsleep = thread_cpu_nsleep,
1708                 .nsleep_restart = thread_cpu_nsleep_restart,
1709         };
1710         struct timespec ts;
1711
1712         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1713         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1714
1715         cputime_to_timespec(cputime_one_jiffy, &ts);
1716         onecputick = ts.tv_nsec;
1717         WARN_ON(ts.tv_sec != 0);
1718
1719         return 0;
1720 }
1721 __initcall(init_posix_cpu_timers);