git.oblomov.eu Git - 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;
  18
  19         cputime = secs_to_cputime(rlim_new);
  20         if (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
  21             cputime_gt(current->signal->it_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 /*
 545  * Insert the timer on the appropriate list before any timers that
 546  * expire later.  This must be called with the tasklist_lock held
 547  * for reading, and interrupts disabled.
 548  */
 549 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 550 {
 551         struct task_struct *p = timer->it.cpu.task;
 552         struct list_head *head, *listpos;
 553         struct cpu_timer_list *const nt = &timer->it.cpu;
 554         struct cpu_timer_list *next;
 555         unsigned long i;
 556
 557         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 558                 p->cpu_timers : p->signal->cpu_timers);
 559         head += CPUCLOCK_WHICH(timer->it_clock);
 560
 561         BUG_ON(!irqs_disabled());
 562         spin_lock(&p->sighand->siglock);
 563
 564         listpos = head;
 565         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 566                 list_for_each_entry(next, head, entry) {
 567                         if (next->expires.sched > nt->expires.sched)
 568                                 break;
 569                         listpos = &next->entry;
 570                 }
 571         } else {
 572                 list_for_each_entry(next, head, entry) {
 573                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 574                                 break;
 575                         listpos = &next->entry;
 576                 }
 577         }
 578         list_add(&nt->entry, listpos);
 579
 580         if (listpos == head) {
 581                 /*
 582                  * We are the new earliest-expiring timer.
 583                  * If we are a thread timer, there can always
 584                  * be a process timer telling us to stop earlier.
 585                  */
 586
 587                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 588                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 589                         default:
 590                                 BUG();
 591                         case CPUCLOCK_PROF:
 592                                 if (cputime_eq(p->cputime_expires.prof_exp,
 593                                                cputime_zero) ||
 594                                     cputime_gt(p->cputime_expires.prof_exp,
 595                                                nt->expires.cpu))
 596                                         p->cputime_expires.prof_exp =
 597                                                 nt->expires.cpu;
 598                                 break;
 599                         case CPUCLOCK_VIRT:
 600                                 if (cputime_eq(p->cputime_expires.virt_exp,
 601                                                cputime_zero) ||
 602                                     cputime_gt(p->cputime_expires.virt_exp,
 603                                                nt->expires.cpu))
 604                                         p->cputime_expires.virt_exp =
 605                                                 nt->expires.cpu;
 606                                 break;
 607                         case CPUCLOCK_SCHED:
 608                                 if (p->cputime_expires.sched_exp == 0 ||
 609                                     p->cputime_expires.sched_exp >
 610                                                         nt->expires.sched)
 611                                         p->cputime_expires.sched_exp =
 612                                                 nt->expires.sched;
 613                                 break;
 614                         }
 615                 } else {
 616                         /*
 617                          * For a process timer, set the cached expiration time.
 618                          */
 619                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 620                         default:
 621                                 BUG();
 622                         case CPUCLOCK_VIRT:
 623                                 if (!cputime_eq(p->signal->it_virt_expires,
 624                                                 cputime_zero) &&
 625                                     cputime_lt(p->signal->it_virt_expires,
 626                                                timer->it.cpu.expires.cpu))
 627                                         break;
 628                                 p->signal->cputime_expires.virt_exp =
 629                                         timer->it.cpu.expires.cpu;
 630                                 break;
 631                         case CPUCLOCK_PROF:
 632                                 if (!cputime_eq(p->signal->it_prof_expires,
 633                                                 cputime_zero) &&
 634                                     cputime_lt(p->signal->it_prof_expires,
 635                                                timer->it.cpu.expires.cpu))
 636                                         break;
 637                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 638                                 if (i != RLIM_INFINITY &&
 639                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 640                                         break;
 641                                 p->signal->cputime_expires.prof_exp =
 642                                         timer->it.cpu.expires.cpu;
 643                                 break;
 644                         case CPUCLOCK_SCHED:
 645                                 p->signal->cputime_expires.sched_exp =
 646                                         timer->it.cpu.expires.sched;
 647                                 break;
 648                         }
 649                 }
 650         }
 651
 652         spin_unlock(&p->sighand->siglock);
 653 }
 654
 655 /*
 656  * The timer is locked, fire it and arrange for its reload.
 657  */
 658 static void cpu_timer_fire(struct k_itimer *timer)
 659 {
 660         if (unlikely(timer->sigq == NULL)) {
 661                 /*
 662                  * This a special case for clock_nanosleep,
 663                  * not a normal timer from sys_timer_create.
 664                  */
 665                 wake_up_process(timer->it_process);
 666                 timer->it.cpu.expires.sched = 0;
 667         } else if (timer->it.cpu.incr.sched == 0) {
 668                 /*
 669                  * One-shot timer.  Clear it as soon as it's fired.
 670                  */
 671                 posix_timer_event(timer, 0);
 672                 timer->it.cpu.expires.sched = 0;
 673         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 674                 /*
 675                  * The signal did not get queued because the signal
 676                  * was ignored, so we won't get any callback to
 677                  * reload the timer.  But we need to keep it
 678                  * ticking in case the signal is deliverable next time.
 679                  */
 680                 posix_cpu_timer_schedule(timer);
 681         }
 682 }
 683
 684 /*
 685  * Sample a process (thread group) timer for the given group_leader task.
 686  * Must be called with tasklist_lock held for reading.
 687  */
 688 static int cpu_timer_sample_group(const clockid_t which_clock,
 689                                   struct task_struct *p,
 690                                   union cpu_time_count *cpu)
 691 {
 692         struct task_cputime cputime;
 693
 694         thread_group_cputimer(p, &cputime);
 695         switch (CPUCLOCK_WHICH(which_clock)) {
 696         default:
 697                 return -EINVAL;
 698         case CPUCLOCK_PROF:
 699                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 700                 break;
 701         case CPUCLOCK_VIRT:
 702                 cpu->cpu = cputime.utime;
 703                 break;
 704         case CPUCLOCK_SCHED:
 705                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 706                 break;
 707         }
 708         return 0;
 709 }
 710
 711 /*
 712  * Guts of sys_timer_settime for CPU timers.
 713  * This is called with the timer locked and interrupts disabled.
 714  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 715  * and try again.  (This happens when the timer is in the middle of firing.)
 716  */
 717 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 718                         struct itimerspec *new, struct itimerspec *old)
 719 {
 720         struct task_struct *p = timer->it.cpu.task;
 721         union cpu_time_count old_expires, new_expires, val;
 722         int ret;
 723
 724         if (unlikely(p == NULL)) {
 725                 /*
 726                  * Timer refers to a dead task's clock.
 727                  */
 728                 return -ESRCH;
 729         }
 730
 731         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 732
 733         read_lock(&tasklist_lock);
 734         /*
 735          * We need the tasklist_lock to protect against reaping that
 736          * clears p->signal.  If p has just been reaped, we can no
 737          * longer get any information about it at all.
 738          */
 739         if (unlikely(p->signal == NULL)) {
 740                 read_unlock(&tasklist_lock);
 741                 put_task_struct(p);
 742                 timer->it.cpu.task = NULL;
 743                 return -ESRCH;
 744         }
 745
 746         /*
 747          * Disarm any old timer after extracting its expiry time.
 748          */
 749         BUG_ON(!irqs_disabled());
 750
 751         ret = 0;
 752         spin_lock(&p->sighand->siglock);
 753         old_expires = timer->it.cpu.expires;
 754         if (unlikely(timer->it.cpu.firing)) {
 755                 timer->it.cpu.firing = -1;
 756                 ret = TIMER_RETRY;
 757         } else
 758                 list_del_init(&timer->it.cpu.entry);
 759         spin_unlock(&p->sighand->siglock);
 760
 761         /*
 762          * We need to sample the current value to convert the new
 763          * value from to relative and absolute, and to convert the
 764          * old value from absolute to relative.  To set a process
 765          * timer, we need a sample to balance the thread expiry
 766          * times (in arm_timer).  With an absolute time, we must
 767          * check if it's already passed.  In short, we need a sample.
 768          */
 769         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 770                 cpu_clock_sample(timer->it_clock, p, &val);
 771         } else {
 772                 cpu_timer_sample_group(timer->it_clock, p, &val);
 773         }
 774
 775         if (old) {
 776                 if (old_expires.sched == 0) {
 777                         old->it_value.tv_sec = 0;
 778                         old->it_value.tv_nsec = 0;
 779                 } else {
 780                         /*
 781                          * Update the timer in case it has
 782                          * overrun already.  If it has,
 783                          * we'll report it as having overrun
 784                          * and with the next reloaded timer
 785                          * already ticking, though we are
 786                          * swallowing that pending
 787                          * notification here to install the
 788                          * new setting.
 789                          */
 790                         bump_cpu_timer(timer, val);
 791                         if (cpu_time_before(timer->it_clock, val,
 792                                             timer->it.cpu.expires)) {
 793                                 old_expires = cpu_time_sub(
 794                                         timer->it_clock,
 795                                         timer->it.cpu.expires, val);
 796                                 sample_to_timespec(timer->it_clock,
 797                                                    old_expires,
 798                                                    &old->it_value);
 799                         } else {
 800                                 old->it_value.tv_nsec = 1;
 801                                 old->it_value.tv_sec = 0;
 802                         }
 803                 }
 804         }
 805
 806         if (unlikely(ret)) {
 807                 /*
 808                  * We are colliding with the timer actually firing.
 809                  * Punt after filling in the timer's old value, and
 810                  * disable this firing since we are already reporting
 811                  * it as an overrun (thanks to bump_cpu_timer above).
 812                  */
 813                 read_unlock(&tasklist_lock);
 814                 goto out;
 815         }
 816
 817         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 818                 cpu_time_add(timer->it_clock, &new_expires, val);
 819         }
 820
 821         /*
 822          * Install the new expiry time (or zero).
 823          * For a timer with no notification action, we don't actually
 824          * arm the timer (we'll just fake it for timer_gettime).
 825          */
 826         timer->it.cpu.expires = new_expires;
 827         if (new_expires.sched != 0 &&
 828             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 829             cpu_time_before(timer->it_clock, val, new_expires)) {
 830                 arm_timer(timer, val);
 831         }
 832
 833         read_unlock(&tasklist_lock);
 834
 835         /*
 836          * Install the new reload setting, and
 837          * set up the signal and overrun bookkeeping.
 838          */
 839         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 840                                                 &new->it_interval);
 841
 842         /*
 843          * This acts as a modification timestamp for the timer,
 844          * so any automatic reload attempt will punt on seeing
 845          * that we have reset the timer manually.
 846          */
 847         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 848                 ~REQUEUE_PENDING;
 849         timer->it_overrun_last = 0;
 850         timer->it_overrun = -1;
 851
 852         if (new_expires.sched != 0 &&
 853             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 854             !cpu_time_before(timer->it_clock, val, new_expires)) {
 855                 /*
 856                  * The designated time already passed, so we notify
 857                  * immediately, even if the thread never runs to
 858                  * accumulate more time on this clock.
 859                  */
 860                 cpu_timer_fire(timer);
 861         }
 862
 863         ret = 0;
 864  out:
 865         if (old) {
 866                 sample_to_timespec(timer->it_clock,
 867                                    timer->it.cpu.incr, &old->it_interval);
 868         }
 869         return ret;
 870 }
 871
 872 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 873 {
 874         union cpu_time_count now;
 875         struct task_struct *p = timer->it.cpu.task;
 876         int clear_dead;
 877
 878         /*
 879          * Easy part: convert the reload time.
 880          */
 881         sample_to_timespec(timer->it_clock,
 882                            timer->it.cpu.incr, &itp->it_interval);
 883
 884         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 885                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 886                 return;
 887         }
 888
 889         if (unlikely(p == NULL)) {
 890                 /*
 891                  * This task already died and the timer will never fire.
 892                  * In this case, expires is actually the dead value.
 893                  */
 894         dead:
 895                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 896                                    &itp->it_value);
 897                 return;
 898         }
 899
 900         /*
 901          * Sample the clock to take the difference with the expiry time.
 902          */
 903         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 904                 cpu_clock_sample(timer->it_clock, p, &now);
 905                 clear_dead = p->exit_state;
 906         } else {
 907                 read_lock(&tasklist_lock);
 908                 if (unlikely(p->signal == NULL)) {
 909                         /*
 910                          * The process has been reaped.
 911                          * We can't even collect a sample any more.
 912                          * Call the timer disarmed, nothing else to do.
 913                          */
 914                         put_task_struct(p);
 915                         timer->it.cpu.task = NULL;
 916                         timer->it.cpu.expires.sched = 0;
 917                         read_unlock(&tasklist_lock);
 918                         goto dead;
 919                 } else {
 920                         cpu_timer_sample_group(timer->it_clock, p, &now);
 921                         clear_dead = (unlikely(p->exit_state) &&
 922                                       thread_group_empty(p));
 923                 }
 924                 read_unlock(&tasklist_lock);
 925         }
 926
 927         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 928                 if (timer->it.cpu.incr.sched == 0 &&
 929                     cpu_time_before(timer->it_clock,
 930                                     timer->it.cpu.expires, now)) {
 931                         /*
 932                          * Do-nothing timer expired and has no reload,
 933                          * so it's as if it was never set.
 934                          */
 935                         timer->it.cpu.expires.sched = 0;
 936                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 937                         return;
 938                 }
 939                 /*
 940                  * Account for any expirations and reloads that should
 941                  * have happened.
 942                  */
 943                 bump_cpu_timer(timer, now);
 944         }
 945
 946         if (unlikely(clear_dead)) {
 947                 /*
 948                  * We've noticed that the thread is dead, but
 949                  * not yet reaped.  Take this opportunity to
 950                  * drop our task ref.
 951                  */
 952                 clear_dead_task(timer, now);
 953                 goto dead;
 954         }
 955
 956         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 957                 sample_to_timespec(timer->it_clock,
 958                                    cpu_time_sub(timer->it_clock,
 959                                                 timer->it.cpu.expires, now),
 960                                    &itp->it_value);
 961         } else {
 962                 /*
 963                  * The timer should have expired already, but the firing
 964                  * hasn't taken place yet.  Say it's just about to expire.
 965                  */
 966                 itp->it_value.tv_nsec = 1;
 967                 itp->it_value.tv_sec = 0;
 968         }
 969 }
 970
 971 /*
 972  * Check for any per-thread CPU timers that have fired and move them off
 973  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 974  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 975  */
 976 static void check_thread_timers(struct task_struct *tsk,
 977                                 struct list_head *firing)
 978 {
 979         int maxfire;
 980         struct list_head *timers = tsk->cpu_timers;
 981         struct signal_struct *const sig = tsk->signal;
 982
 983         maxfire = 20;
 984         tsk->cputime_expires.prof_exp = cputime_zero;
 985         while (!list_empty(timers)) {
 986                 struct cpu_timer_list *t = list_first_entry(timers,
 987                                                       struct cpu_timer_list,
 988                                                       entry);
 989                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 990                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 991                         break;
 992                 }
 993                 t->firing = 1;
 994                 list_move_tail(&t->entry, firing);
 995         }
 996
 997         ++timers;
 998         maxfire = 20;
 999         tsk->cputime_expires.virt_exp = cputime_zero;
1000         while (!list_empty(timers)) {
1001                 struct cpu_timer_list *t = list_first_entry(timers,
1002                                                       struct cpu_timer_list,
1003                                                       entry);
1004                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1005                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1006                         break;
1007                 }
1008                 t->firing = 1;
1009                 list_move_tail(&t->entry, firing);
1010         }
1011
1012         ++timers;
1013         maxfire = 20;
1014         tsk->cputime_expires.sched_exp = 0;
1015         while (!list_empty(timers)) {
1016                 struct cpu_timer_list *t = list_first_entry(timers,
1017                                                       struct cpu_timer_list,
1018                                                       entry);
1019                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1020                         tsk->cputime_expires.sched_exp = t->expires.sched;
1021                         break;
1022                 }
1023                 t->firing = 1;
1024                 list_move_tail(&t->entry, firing);
1025         }
1026
1027         /*
1028          * Check for the special case thread timers.
1029          */
1030         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1031                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1032                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1033
1034                 if (hard != RLIM_INFINITY &&
1035                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1036                         /*
1037                          * At the hard limit, we just die.
1038                          * No need to calculate anything else now.
1039                          */
1040                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1041                         return;
1042                 }
1043                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1044                         /*
1045                          * At the soft limit, send a SIGXCPU every second.
1046                          */
1047                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1048                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1049                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1050                                                                 USEC_PER_SEC;
1051                         }
1052                         printk(KERN_INFO
1053                                 "RT Watchdog Timeout: %s[%d]\n",
1054                                 tsk->comm, task_pid_nr(tsk));
1055                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1056                 }
1057         }
1058 }
1059
1060 static void stop_process_timers(struct task_struct *tsk)
1061 {
1062         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1063         unsigned long flags;
1064
1065         if (!cputimer->running)
1066                 return;
1067
1068         spin_lock_irqsave(&cputimer->lock, flags);
1069         cputimer->running = 0;
1070         spin_unlock_irqrestore(&cputimer->lock, flags);
1071 }
1072
1073 /*
1074  * Check for any per-thread CPU timers that have fired and move them
1075  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1076  * have already been taken off.
1077  */
1078 static void check_process_timers(struct task_struct *tsk,
1079                                  struct list_head *firing)
1080 {
1081         int maxfire;
1082         struct signal_struct *const sig = tsk->signal;
1083         cputime_t utime, ptime, virt_expires, prof_expires;
1084         unsigned long long sum_sched_runtime, sched_expires;
1085         struct list_head *timers = sig->cpu_timers;
1086         struct task_cputime cputime;
1087
1088         /*
1089          * Don't sample the current process CPU clocks if there are no timers.
1090          */
1091         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1092             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1093             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1094             list_empty(&timers[CPUCLOCK_VIRT]) &&
1095             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1096             list_empty(&timers[CPUCLOCK_SCHED])) {
1097                 stop_process_timers(tsk);
1098                 return;
1099         }
1100
1101         /*
1102          * Collect the current process totals.
1103          */
1104         thread_group_cputimer(tsk, &cputime);
1105         utime = cputime.utime;
1106         ptime = cputime_add(utime, cputime.stime);
1107         sum_sched_runtime = cputime.sum_exec_runtime;
1108         maxfire = 20;
1109         prof_expires = cputime_zero;
1110         while (!list_empty(timers)) {
1111                 struct cpu_timer_list *tl = list_first_entry(timers,
1112                                                       struct cpu_timer_list,
1113                                                       entry);
1114                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1115                         prof_expires = tl->expires.cpu;
1116                         break;
1117                 }
1118                 tl->firing = 1;
1119                 list_move_tail(&tl->entry, firing);
1120         }
1121
1122         ++timers;
1123         maxfire = 20;
1124         virt_expires = cputime_zero;
1125         while (!list_empty(timers)) {
1126                 struct cpu_timer_list *tl = list_first_entry(timers,
1127                                                       struct cpu_timer_list,
1128                                                       entry);
1129                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1130                         virt_expires = tl->expires.cpu;
1131                         break;
1132                 }
1133                 tl->firing = 1;
1134                 list_move_tail(&tl->entry, firing);
1135         }
1136
1137         ++timers;
1138         maxfire = 20;
1139         sched_expires = 0;
1140         while (!list_empty(timers)) {
1141                 struct cpu_timer_list *tl = list_first_entry(timers,
1142                                                       struct cpu_timer_list,
1143                                                       entry);
1144                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1145                         sched_expires = tl->expires.sched;
1146                         break;
1147                 }
1148                 tl->firing = 1;
1149                 list_move_tail(&tl->entry, firing);
1150         }
1151
1152         /*
1153          * Check for the special case process timers.
1154          */
1155         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1156                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1157                         /* ITIMER_PROF fires and reloads.  */
1158                         sig->it_prof_expires = sig->it_prof_incr;
1159                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1160                                 sig->it_prof_expires = cputime_add(
1161                                         sig->it_prof_expires, ptime);
1162                         }
1163                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1164                 }
1165                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1166                     (cputime_eq(prof_expires, cputime_zero) ||
1167                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1168                         prof_expires = sig->it_prof_expires;
1169                 }
1170         }
1171         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1172                 if (cputime_ge(utime, sig->it_virt_expires)) {
1173                         /* ITIMER_VIRTUAL fires and reloads.  */
1174                         sig->it_virt_expires = sig->it_virt_incr;
1175                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1176                                 sig->it_virt_expires = cputime_add(
1177                                         sig->it_virt_expires, utime);
1178                         }
1179                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1180                 }
1181                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1182                     (cputime_eq(virt_expires, cputime_zero) ||
1183                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1184                         virt_expires = sig->it_virt_expires;
1185                 }
1186         }
1187         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1188                 unsigned long psecs = cputime_to_secs(ptime);
1189                 cputime_t x;
1190                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1191                         /*
1192                          * At the hard limit, we just die.
1193                          * No need to calculate anything else now.
1194                          */
1195                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1196                         return;
1197                 }
1198                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1199                         /*
1200                          * At the soft limit, send a SIGXCPU every second.
1201                          */
1202                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1203                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1204                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1205                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1206                         }
1207                 }
1208                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1209                 if (cputime_eq(prof_expires, cputime_zero) ||
1210                     cputime_lt(x, prof_expires)) {
1211                         prof_expires = x;
1212                 }
1213         }
1214
1215         if (!cputime_eq(prof_expires, cputime_zero) &&
1216             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1217              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1218                 sig->cputime_expires.prof_exp = prof_expires;
1219         if (!cputime_eq(virt_expires, cputime_zero) &&
1220             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1221              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1222                 sig->cputime_expires.virt_exp = virt_expires;
1223         if (sched_expires != 0 &&
1224             (sig->cputime_expires.sched_exp == 0 ||
1225              sig->cputime_expires.sched_exp > sched_expires))
1226                 sig->cputime_expires.sched_exp = sched_expires;
1227 }
1228
1229 /*
1230  * This is called from the signal code (via do_schedule_next_timer)
1231  * when the last timer signal was delivered and we have to reload the timer.
1232  */
1233 void posix_cpu_timer_schedule(struct k_itimer *timer)
1234 {
1235         struct task_struct *p = timer->it.cpu.task;
1236         union cpu_time_count now;
1237
1238         if (unlikely(p == NULL))
1239                 /*
1240                  * The task was cleaned up already, no future firings.
1241                  */
1242                 goto out;
1243
1244         /*
1245          * Fetch the current sample and update the timer's expiry time.
1246          */
1247         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1248                 cpu_clock_sample(timer->it_clock, p, &now);
1249                 bump_cpu_timer(timer, now);
1250                 if (unlikely(p->exit_state)) {
1251                         clear_dead_task(timer, now);
1252                         goto out;
1253                 }
1254                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1255         } else {
1256                 read_lock(&tasklist_lock);
1257                 if (unlikely(p->signal == NULL)) {
1258                         /*
1259                          * The process has been reaped.
1260                          * We can't even collect a sample any more.
1261                          */
1262                         put_task_struct(p);
1263                         timer->it.cpu.task = p = NULL;
1264                         timer->it.cpu.expires.sched = 0;
1265                         goto out_unlock;
1266                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1267                         /*
1268                          * We've noticed that the thread is dead, but
1269                          * not yet reaped.  Take this opportunity to
1270                          * drop our task ref.
1271                          */
1272                         clear_dead_task(timer, now);
1273                         goto out_unlock;
1274                 }
1275                 cpu_timer_sample_group(timer->it_clock, p, &now);
1276                 bump_cpu_timer(timer, now);
1277                 /* Leave the tasklist_lock locked for the call below.  */
1278         }
1279
1280         /*
1281          * Now re-arm for the new expiry time.
1282          */
1283         arm_timer(timer, now);
1284
1285 out_unlock:
1286         read_unlock(&tasklist_lock);
1287
1288 out:
1289         timer->it_overrun_last = timer->it_overrun;
1290         timer->it_overrun = -1;
1291         ++timer->it_requeue_pending;
1292 }
1293
1294 /**
1295  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1296  *
1297  * @cputime:    The struct to compare.
1298  *
1299  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1300  * are zero, false if any field is nonzero.
1301  */
1302 static inline int task_cputime_zero(const struct task_cputime *cputime)
1303 {
1304         if (cputime_eq(cputime->utime, cputime_zero) &&
1305             cputime_eq(cputime->stime, cputime_zero) &&
1306             cputime->sum_exec_runtime == 0)
1307                 return 1;
1308         return 0;
1309 }
1310
1311 /**
1312  * task_cputime_expired - Compare two task_cputime entities.
1313  *
1314  * @sample:     The task_cputime structure to be checked for expiration.
1315  * @expires:    Expiration times, against which @sample will be checked.
1316  *
1317  * Checks @sample against @expires to see if any field of @sample has expired.
1318  * Returns true if any field of the former is greater than the corresponding
1319  * field of the latter if the latter field is set.  Otherwise returns false.
1320  */
1321 static inline int task_cputime_expired(const struct task_cputime *sample,
1322                                         const struct task_cputime *expires)
1323 {
1324         if (!cputime_eq(expires->utime, cputime_zero) &&
1325             cputime_ge(sample->utime, expires->utime))
1326                 return 1;
1327         if (!cputime_eq(expires->stime, cputime_zero) &&
1328             cputime_ge(cputime_add(sample->utime, sample->stime),
1329                        expires->stime))
1330                 return 1;
1331         if (expires->sum_exec_runtime != 0 &&
1332             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1333                 return 1;
1334         return 0;
1335 }
1336
1337 /**
1338  * fastpath_timer_check - POSIX CPU timers fast path.
1339  *
1340  * @tsk:        The task (thread) being checked.
1341  *
1342  * Check the task and thread group timers.  If both are zero (there are no
1343  * timers set) return false.  Otherwise snapshot the task and thread group
1344  * timers and compare them with the corresponding expiration times.  Return
1345  * true if a timer has expired, else return false.
1346  */
1347 static inline int fastpath_timer_check(struct task_struct *tsk)
1348 {
1349         struct signal_struct *sig;
1350
1351         /* tsk == current, ensure it is safe to use ->signal/sighand */
1352         if (unlikely(tsk->exit_state))
1353                 return 0;
1354
1355         if (!task_cputime_zero(&tsk->cputime_expires)) {
1356                 struct task_cputime task_sample = {
1357                         .utime = tsk->utime,
1358                         .stime = tsk->stime,
1359                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1360                 };
1361
1362                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1363                         return 1;
1364         }
1365
1366         sig = tsk->signal;
1367         if (!task_cputime_zero(&sig->cputime_expires)) {
1368                 struct task_cputime group_sample;
1369
1370                 thread_group_cputimer(tsk, &group_sample);
1371                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1372                         return 1;
1373         }
1374
1375         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1376 }
1377
1378 /*
1379  * This is called from the timer interrupt handler.  The irq handler has
1380  * already updated our counts.  We need to check if any timers fire now.
1381  * Interrupts are disabled.
1382  */
1383 void run_posix_cpu_timers(struct task_struct *tsk)
1384 {
1385         LIST_HEAD(firing);
1386         struct k_itimer *timer, *next;
1387
1388         BUG_ON(!irqs_disabled());
1389
1390         /*
1391          * The fast path checks that there are no expired thread or thread
1392          * group timers.  If that's so, just return.
1393          */
1394         if (!fastpath_timer_check(tsk))
1395                 return;
1396
1397         spin_lock(&tsk->sighand->siglock);
1398         /*
1399          * Here we take off tsk->signal->cpu_timers[N] and
1400          * tsk->cpu_timers[N] all the timers that are firing, and
1401          * put them on the firing list.
1402          */
1403         check_thread_timers(tsk, &firing);
1404         check_process_timers(tsk, &firing);
1405
1406         /*
1407          * We must release these locks before taking any timer's lock.
1408          * There is a potential race with timer deletion here, as the
1409          * siglock now protects our private firing list.  We have set
1410          * the firing flag in each timer, so that a deletion attempt
1411          * that gets the timer lock before we do will give it up and
1412          * spin until we've taken care of that timer below.
1413          */
1414         spin_unlock(&tsk->sighand->siglock);
1415
1416         /*
1417          * Now that all the timers on our list have the firing flag,
1418          * noone will touch their list entries but us.  We'll take
1419          * each timer's lock before clearing its firing flag, so no
1420          * timer call will interfere.
1421          */
1422         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1423                 int firing;
1424                 spin_lock(&timer->it_lock);
1425                 list_del_init(&timer->it.cpu.entry);
1426                 firing = timer->it.cpu.firing;
1427                 timer->it.cpu.firing = 0;
1428                 /*
1429                  * The firing flag is -1 if we collided with a reset
1430                  * of the timer, which already reported this
1431                  * almost-firing as an overrun.  So don't generate an event.
1432                  */
1433                 if (likely(firing >= 0)) {
1434                         cpu_timer_fire(timer);
1435                 }
1436                 spin_unlock(&timer->it_lock);
1437         }
1438 }
1439
1440 /*
1441  * Set one of the process-wide special case CPU timers.
1442  * The tsk->sighand->siglock must be held by the caller.
1443  * The *newval argument is relative and we update it to be absolute, *oldval
1444  * is absolute and we update it to be relative.
1445  */
1446 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1447                            cputime_t *newval, cputime_t *oldval)
1448 {
1449         union cpu_time_count now;
1450         struct list_head *head;
1451
1452         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1453         cpu_timer_sample_group(clock_idx, tsk, &now);
1454
1455         if (oldval) {
1456                 if (!cputime_eq(*oldval, cputime_zero)) {
1457                         if (cputime_le(*oldval, now.cpu)) {
1458                                 /* Just about to fire. */
1459                                 *oldval = jiffies_to_cputime(1);
1460                         } else {
1461                                 *oldval = cputime_sub(*oldval, now.cpu);
1462                         }
1463                 }
1464
1465                 if (cputime_eq(*newval, cputime_zero))
1466                         return;
1467                 *newval = cputime_add(*newval, now.cpu);
1468
1469                 /*
1470                  * If the RLIMIT_CPU timer will expire before the
1471                  * ITIMER_PROF timer, we have nothing else to do.
1472                  */
1473                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1474                     < cputime_to_secs(*newval))
1475                         return;
1476         }
1477
1478         /*
1479          * Check whether there are any process timers already set to fire
1480          * before this one.  If so, we don't have anything more to do.
1481          */
1482         head = &tsk->signal->cpu_timers[clock_idx];
1483         if (list_empty(head) ||
1484             cputime_ge(list_first_entry(head,
1485                                   struct cpu_timer_list, entry)->expires.cpu,
1486                        *newval)) {
1487                 switch (clock_idx) {
1488                 case CPUCLOCK_PROF:
1489                         tsk->signal->cputime_expires.prof_exp = *newval;
1490                         break;
1491                 case CPUCLOCK_VIRT:
1492                         tsk->signal->cputime_expires.virt_exp = *newval;
1493                         break;
1494                 }
1495         }
1496 }
1497
1498 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1499                             struct timespec *rqtp, struct itimerspec *it)
1500 {
1501         struct k_itimer timer;
1502         int error;
1503
1504         /*
1505          * Set up a temporary timer and then wait for it to go off.
1506          */
1507         memset(&timer, 0, sizeof timer);
1508         spin_lock_init(&timer.it_lock);
1509         timer.it_clock = which_clock;
1510         timer.it_overrun = -1;
1511         error = posix_cpu_timer_create(&timer);
1512         timer.it_process = current;
1513         if (!error) {
1514                 static struct itimerspec zero_it;
1515
1516                 memset(it, 0, sizeof *it);
1517                 it->it_value = *rqtp;
1518
1519                 spin_lock_irq(&timer.it_lock);
1520                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1521                 if (error) {
1522                         spin_unlock_irq(&timer.it_lock);
1523                         return error;
1524                 }
1525
1526                 while (!signal_pending(current)) {
1527                         if (timer.it.cpu.expires.sched == 0) {
1528                                 /*
1529                                  * Our timer fired and was reset.
1530                                  */
1531                                 spin_unlock_irq(&timer.it_lock);
1532                                 return 0;
1533                         }
1534
1535                         /*
1536                          * Block until cpu_timer_fire (or a signal) wakes us.
1537                          */
1538                         __set_current_state(TASK_INTERRUPTIBLE);
1539                         spin_unlock_irq(&timer.it_lock);
1540                         schedule();
1541                         spin_lock_irq(&timer.it_lock);
1542                 }
1543
1544                 /*
1545                  * We were interrupted by a signal.
1546                  */
1547                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1548                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1549                 spin_unlock_irq(&timer.it_lock);
1550
1551                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1552                         /*
1553                          * It actually did fire already.
1554                          */
1555                         return 0;
1556                 }
1557
1558                 error = -ERESTART_RESTARTBLOCK;
1559         }
1560
1561         return error;
1562 }
1563
1564 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1565                      struct timespec *rqtp, struct timespec __user *rmtp)
1566 {
1567         struct restart_block *restart_block =
1568             &current_thread_info()->restart_block;
1569         struct itimerspec it;
1570         int error;
1571
1572         /*
1573          * Diagnose required errors first.
1574          */
1575         if (CPUCLOCK_PERTHREAD(which_clock) &&
1576             (CPUCLOCK_PID(which_clock) == 0 ||
1577              CPUCLOCK_PID(which_clock) == current->pid))
1578                 return -EINVAL;
1579
1580         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1581
1582         if (error == -ERESTART_RESTARTBLOCK) {
1583
1584                 if (flags & TIMER_ABSTIME)
1585                         return -ERESTARTNOHAND;
1586                 /*
1587                  * Report back to the user the time still remaining.
1588                  */
1589                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1590                         return -EFAULT;
1591
1592                 restart_block->fn = posix_cpu_nsleep_restart;
1593                 restart_block->arg0 = which_clock;
1594                 restart_block->arg1 = (unsigned long) rmtp;
1595                 restart_block->arg2 = rqtp->tv_sec;
1596                 restart_block->arg3 = rqtp->tv_nsec;
1597         }
1598         return error;
1599 }
1600
1601 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1602 {
1603         clockid_t which_clock = restart_block->arg0;
1604         struct timespec __user *rmtp;
1605         struct timespec t;
1606         struct itimerspec it;
1607         int error;
1608
1609         rmtp = (struct timespec __user *) restart_block->arg1;
1610         t.tv_sec = restart_block->arg2;
1611         t.tv_nsec = restart_block->arg3;
1612
1613         restart_block->fn = do_no_restart_syscall;
1614         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1615
1616         if (error == -ERESTART_RESTARTBLOCK) {
1617                 /*
1618                  * Report back to the user the time still remaining.
1619                  */
1620                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1621                         return -EFAULT;
1622
1623                 restart_block->fn = posix_cpu_nsleep_restart;
1624                 restart_block->arg0 = which_clock;
1625                 restart_block->arg1 = (unsigned long) rmtp;
1626                 restart_block->arg2 = t.tv_sec;
1627                 restart_block->arg3 = t.tv_nsec;
1628         }
1629         return error;
1630
1631 }
1632
1633
1634 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1635 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1636
1637 static int process_cpu_clock_getres(const clockid_t which_clock,
1638                                     struct timespec *tp)
1639 {
1640         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1641 }
1642 static int process_cpu_clock_get(const clockid_t which_clock,
1643                                  struct timespec *tp)
1644 {
1645         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1646 }
1647 static int process_cpu_timer_create(struct k_itimer *timer)
1648 {
1649         timer->it_clock = PROCESS_CLOCK;
1650         return posix_cpu_timer_create(timer);
1651 }
1652 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1653                               struct timespec *rqtp,
1654                               struct timespec __user *rmtp)
1655 {
1656         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1657 }
1658 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1659 {
1660         return -EINVAL;
1661 }
1662 static int thread_cpu_clock_getres(const clockid_t which_clock,
1663                                    struct timespec *tp)
1664 {
1665         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1666 }
1667 static int thread_cpu_clock_get(const clockid_t which_clock,
1668                                 struct timespec *tp)
1669 {
1670         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1671 }
1672 static int thread_cpu_timer_create(struct k_itimer *timer)
1673 {
1674         timer->it_clock = THREAD_CLOCK;
1675         return posix_cpu_timer_create(timer);
1676 }
1677 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1678                               struct timespec *rqtp, struct timespec __user *rmtp)
1679 {
1680         return -EINVAL;
1681 }
1682 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1683 {
1684         return -EINVAL;
1685 }
1686
1687 static __init int init_posix_cpu_timers(void)
1688 {
1689         struct k_clock process = {
1690                 .clock_getres = process_cpu_clock_getres,
1691                 .clock_get = process_cpu_clock_get,
1692                 .clock_set = do_posix_clock_nosettime,
1693                 .timer_create = process_cpu_timer_create,
1694                 .nsleep = process_cpu_nsleep,
1695                 .nsleep_restart = process_cpu_nsleep_restart,
1696         };
1697         struct k_clock thread = {
1698                 .clock_getres = thread_cpu_clock_getres,
1699                 .clock_get = thread_cpu_clock_get,
1700                 .clock_set = do_posix_clock_nosettime,
1701                 .timer_create = thread_cpu_timer_create,
1702                 .nsleep = thread_cpu_nsleep,
1703                 .nsleep_restart = thread_cpu_nsleep_restart,
1704         };
1705
1706         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1707         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1708
1709         return 0;
1710 }
1711 __initcall(init_posix_cpu_timers);