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 #include <trace/events/timer.h>
  12
  13 /*
  14  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
  15  */
  16 void update_rlimit_cpu(unsigned long rlim_new)
  17 {
  18         cputime_t cputime = secs_to_cputime(rlim_new);
  19         struct signal_struct *const sig = current->signal;
  20
  21         if (cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) ||
  22             cputime_gt(sig->it[CPUCLOCK_PROF].expires, cputime)) {
  23                 spin_lock_irq(&current->sighand->siglock);
  24                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
  25                 spin_unlock_irq(&current->sighand->siglock);
  26         }
  27 }
  28
  29 static int check_clock(const clockid_t which_clock)
  30 {
  31         int error = 0;
  32         struct task_struct *p;
  33         const pid_t pid = CPUCLOCK_PID(which_clock);
  34
  35         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  36                 return -EINVAL;
  37
  38         if (pid == 0)
  39                 return 0;
  40
  41         read_lock(&tasklist_lock);
  42         p = find_task_by_vpid(pid);
  43         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  44                    same_thread_group(p, current) : thread_group_leader(p))) {
  45                 error = -EINVAL;
  46         }
  47         read_unlock(&tasklist_lock);
  48
  49         return error;
  50 }
  51
  52 static inline union cpu_time_count
  53 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  54 {
  55         union cpu_time_count ret;
  56         ret.sched = 0;          /* high half always zero when .cpu used */
  57         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  58                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  59         } else {
  60                 ret.cpu = timespec_to_cputime(tp);
  61         }
  62         return ret;
  63 }
  64
  65 static void sample_to_timespec(const clockid_t which_clock,
  66                                union cpu_time_count cpu,
  67                                struct timespec *tp)
  68 {
  69         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  70                 *tp = ns_to_timespec(cpu.sched);
  71         else
  72                 cputime_to_timespec(cpu.cpu, tp);
  73 }
  74
  75 static inline int cpu_time_before(const clockid_t which_clock,
  76                                   union cpu_time_count now,
  77                                   union cpu_time_count then)
  78 {
  79         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  80                 return now.sched < then.sched;
  81         }  else {
  82                 return cputime_lt(now.cpu, then.cpu);
  83         }
  84 }
  85 static inline void cpu_time_add(const clockid_t which_clock,
  86                                 union cpu_time_count *acc,
  87                                 union cpu_time_count val)
  88 {
  89         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  90                 acc->sched += val.sched;
  91         }  else {
  92                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  93         }
  94 }
  95 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  96                                                 union cpu_time_count a,
  97                                                 union cpu_time_count b)
  98 {
  99         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 100                 a.sched -= b.sched;
 101         }  else {
 102                 a.cpu = cputime_sub(a.cpu, b.cpu);
 103         }
 104         return a;
 105 }
 106
 107 /*
 108  * Divide and limit the result to res >= 1
 109  *
 110  * This is necessary to prevent signal delivery starvation, when the result of
 111  * the division would be rounded down to 0.
 112  */
 113 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
 114 {
 115         cputime_t res = cputime_div(time, div);
 116
 117         return max_t(cputime_t, res, 1);
 118 }
 119
 120 /*
 121  * Update expiry time from increment, and increase overrun count,
 122  * given the current clock sample.
 123  */
 124 static void bump_cpu_timer(struct k_itimer *timer,
 125                                   union cpu_time_count now)
 126 {
 127         int i;
 128
 129         if (timer->it.cpu.incr.sched == 0)
 130                 return;
 131
 132         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 133                 unsigned long long delta, incr;
 134
 135                 if (now.sched < timer->it.cpu.expires.sched)
 136                         return;
 137                 incr = timer->it.cpu.incr.sched;
 138                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 139                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 140                 for (i = 0; incr < delta - incr; i++)
 141                         incr = incr << 1;
 142                 for (; i >= 0; incr >>= 1, i--) {
 143                         if (delta < incr)
 144                                 continue;
 145                         timer->it.cpu.expires.sched += incr;
 146                         timer->it_overrun += 1 << i;
 147                         delta -= incr;
 148                 }
 149         } else {
 150                 cputime_t delta, incr;
 151
 152                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 153                         return;
 154                 incr = timer->it.cpu.incr.cpu;
 155                 delta = cputime_sub(cputime_add(now.cpu, incr),
 156                                     timer->it.cpu.expires.cpu);
 157                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 158                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 159                              incr = cputime_add(incr, incr);
 160                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 161                         if (cputime_lt(delta, incr))
 162                                 continue;
 163                         timer->it.cpu.expires.cpu =
 164                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 165                         timer->it_overrun += 1 << i;
 166                         delta = cputime_sub(delta, incr);
 167                 }
 168         }
 169 }
 170
 171 static inline cputime_t prof_ticks(struct task_struct *p)
 172 {
 173         return cputime_add(p->utime, p->stime);
 174 }
 175 static inline cputime_t virt_ticks(struct task_struct *p)
 176 {
 177         return p->utime;
 178 }
 179
 180 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 181 {
 182         int error = check_clock(which_clock);
 183         if (!error) {
 184                 tp->tv_sec = 0;
 185                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 186                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 187                         /*
 188                          * If sched_clock is using a cycle counter, we
 189                          * don't have any idea of its true resolution
 190                          * exported, but it is much more than 1s/HZ.
 191                          */
 192                         tp->tv_nsec = 1;
 193                 }
 194         }
 195         return error;
 196 }
 197
 198 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 199 {
 200         /*
 201          * You can never reset a CPU clock, but we check for other errors
 202          * in the call before failing with EPERM.
 203          */
 204         int error = check_clock(which_clock);
 205         if (error == 0) {
 206                 error = -EPERM;
 207         }
 208         return error;
 209 }
 210
 211
 212 /*
 213  * Sample a per-thread clock for the given task.
 214  */
 215 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 216                             union cpu_time_count *cpu)
 217 {
 218         switch (CPUCLOCK_WHICH(which_clock)) {
 219         default:
 220                 return -EINVAL;
 221         case CPUCLOCK_PROF:
 222                 cpu->cpu = prof_ticks(p);
 223                 break;
 224         case CPUCLOCK_VIRT:
 225                 cpu->cpu = virt_ticks(p);
 226                 break;
 227         case CPUCLOCK_SCHED:
 228                 cpu->sched = task_sched_runtime(p);
 229                 break;
 230         }
 231         return 0;
 232 }
 233
 234 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
 235 {
 236         struct sighand_struct *sighand;
 237         struct signal_struct *sig;
 238         struct task_struct *t;
 239
 240         *times = INIT_CPUTIME;
 241
 242         rcu_read_lock();
 243         sighand = rcu_dereference(tsk->sighand);
 244         if (!sighand)
 245                 goto out;
 246
 247         sig = tsk->signal;
 248
 249         t = tsk;
 250         do {
 251                 times->utime = cputime_add(times->utime, t->utime);
 252                 times->stime = cputime_add(times->stime, t->stime);
 253                 times->sum_exec_runtime += t->se.sum_exec_runtime;
 254
 255                 t = next_thread(t);
 256         } while (t != tsk);
 257
 258         times->utime = cputime_add(times->utime, sig->utime);
 259         times->stime = cputime_add(times->stime, sig->stime);
 260         times->sum_exec_runtime += sig->sum_sched_runtime;
 261 out:
 262         rcu_read_unlock();
 263 }
 264
 265 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 266 {
 267         if (cputime_gt(b->utime, a->utime))
 268                 a->utime = b->utime;
 269
 270         if (cputime_gt(b->stime, a->stime))
 271                 a->stime = b->stime;
 272
 273         if (b->sum_exec_runtime > a->sum_exec_runtime)
 274                 a->sum_exec_runtime = b->sum_exec_runtime;
 275 }
 276
 277 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 278 {
 279         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 280         struct task_cputime sum;
 281         unsigned long flags;
 282
 283         spin_lock_irqsave(&cputimer->lock, flags);
 284         if (!cputimer->running) {
 285                 cputimer->running = 1;
 286                 /*
 287                  * The POSIX timer interface allows for absolute time expiry
 288                  * values through the TIMER_ABSTIME flag, therefore we have
 289                  * to synchronize the timer to the clock every time we start
 290                  * it.
 291                  */
 292                 thread_group_cputime(tsk, &sum);
 293                 update_gt_cputime(&cputimer->cputime, &sum);
 294         }
 295         *times = cputimer->cputime;
 296         spin_unlock_irqrestore(&cputimer->lock, flags);
 297 }
 298
 299 /*
 300  * Sample a process (thread group) clock for the given group_leader task.
 301  * Must be called with tasklist_lock held for reading.
 302  */
 303 static int cpu_clock_sample_group(const clockid_t which_clock,
 304                                   struct task_struct *p,
 305                                   union cpu_time_count *cpu)
 306 {
 307         struct task_cputime cputime;
 308
 309         switch (CPUCLOCK_WHICH(which_clock)) {
 310         default:
 311                 return -EINVAL;
 312         case CPUCLOCK_PROF:
 313                 thread_group_cputime(p, &cputime);
 314                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 315                 break;
 316         case CPUCLOCK_VIRT:
 317                 thread_group_cputime(p, &cputime);
 318                 cpu->cpu = cputime.utime;
 319                 break;
 320         case CPUCLOCK_SCHED:
 321                 cpu->sched = thread_group_sched_runtime(p);
 322                 break;
 323         }
 324         return 0;
 325 }
 326
 327
 328 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 329 {
 330         const pid_t pid = CPUCLOCK_PID(which_clock);
 331         int error = -EINVAL;
 332         union cpu_time_count rtn;
 333
 334         if (pid == 0) {
 335                 /*
 336                  * Special case constant value for our own clocks.
 337                  * We don't have to do any lookup to find ourselves.
 338                  */
 339                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 340                         /*
 341                          * Sampling just ourselves we can do with no locking.
 342                          */
 343                         error = cpu_clock_sample(which_clock,
 344                                                  current, &rtn);
 345                 } else {
 346                         read_lock(&tasklist_lock);
 347                         error = cpu_clock_sample_group(which_clock,
 348                                                        current, &rtn);
 349                         read_unlock(&tasklist_lock);
 350                 }
 351         } else {
 352                 /*
 353                  * Find the given PID, and validate that the caller
 354                  * should be able to see it.
 355                  */
 356                 struct task_struct *p;
 357                 rcu_read_lock();
 358                 p = find_task_by_vpid(pid);
 359                 if (p) {
 360                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 361                                 if (same_thread_group(p, current)) {
 362                                         error = cpu_clock_sample(which_clock,
 363                                                                  p, &rtn);
 364                                 }
 365                         } else {
 366                                 read_lock(&tasklist_lock);
 367                                 if (thread_group_leader(p) && p->signal) {
 368                                         error =
 369                                             cpu_clock_sample_group(which_clock,
 370                                                                    p, &rtn);
 371                                 }
 372                                 read_unlock(&tasklist_lock);
 373                         }
 374                 }
 375                 rcu_read_unlock();
 376         }
 377
 378         if (error)
 379                 return error;
 380         sample_to_timespec(which_clock, rtn, tp);
 381         return 0;
 382 }
 383
 384
 385 /*
 386  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 387  * This is called from sys_timer_create with the new timer already locked.
 388  */
 389 int posix_cpu_timer_create(struct k_itimer *new_timer)
 390 {
 391         int ret = 0;
 392         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 393         struct task_struct *p;
 394
 395         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 396                 return -EINVAL;
 397
 398         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 399         new_timer->it.cpu.incr.sched = 0;
 400         new_timer->it.cpu.expires.sched = 0;
 401
 402         read_lock(&tasklist_lock);
 403         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 404                 if (pid == 0) {
 405                         p = current;
 406                 } else {
 407                         p = find_task_by_vpid(pid);
 408                         if (p && !same_thread_group(p, current))
 409                                 p = NULL;
 410                 }
 411         } else {
 412                 if (pid == 0) {
 413                         p = current->group_leader;
 414                 } else {
 415                         p = find_task_by_vpid(pid);
 416                         if (p && !thread_group_leader(p))
 417                                 p = NULL;
 418                 }
 419         }
 420         new_timer->it.cpu.task = p;
 421         if (p) {
 422                 get_task_struct(p);
 423         } else {
 424                 ret = -EINVAL;
 425         }
 426         read_unlock(&tasklist_lock);
 427
 428         return ret;
 429 }
 430
 431 /*
 432  * Clean up a CPU-clock timer that is about to be destroyed.
 433  * This is called from timer deletion with the timer already locked.
 434  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 435  * and try again.  (This happens when the timer is in the middle of firing.)
 436  */
 437 int posix_cpu_timer_del(struct k_itimer *timer)
 438 {
 439         struct task_struct *p = timer->it.cpu.task;
 440         int ret = 0;
 441
 442         if (likely(p != NULL)) {
 443                 read_lock(&tasklist_lock);
 444                 if (unlikely(p->signal == NULL)) {
 445                         /*
 446                          * We raced with the reaping of the task.
 447                          * The deletion should have cleared us off the list.
 448                          */
 449                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 450                 } else {
 451                         spin_lock(&p->sighand->siglock);
 452                         if (timer->it.cpu.firing)
 453                                 ret = TIMER_RETRY;
 454                         else
 455                                 list_del(&timer->it.cpu.entry);
 456                         spin_unlock(&p->sighand->siglock);
 457                 }
 458                 read_unlock(&tasklist_lock);
 459
 460                 if (!ret)
 461                         put_task_struct(p);
 462         }
 463
 464         return ret;
 465 }
 466
 467 /*
 468  * Clean out CPU timers still ticking when a thread exited.  The task
 469  * pointer is cleared, and the expiry time is replaced with the residual
 470  * time for later timer_gettime calls to return.
 471  * This must be called with the siglock held.
 472  */
 473 static void cleanup_timers(struct list_head *head,
 474                            cputime_t utime, cputime_t stime,
 475                            unsigned long long sum_exec_runtime)
 476 {
 477         struct cpu_timer_list *timer, *next;
 478         cputime_t ptime = cputime_add(utime, stime);
 479
 480         list_for_each_entry_safe(timer, next, head, entry) {
 481                 list_del_init(&timer->entry);
 482                 if (cputime_lt(timer->expires.cpu, ptime)) {
 483                         timer->expires.cpu = cputime_zero;
 484                 } else {
 485                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 486                                                          ptime);
 487                 }
 488         }
 489
 490         ++head;
 491         list_for_each_entry_safe(timer, next, head, entry) {
 492                 list_del_init(&timer->entry);
 493                 if (cputime_lt(timer->expires.cpu, utime)) {
 494                         timer->expires.cpu = cputime_zero;
 495                 } else {
 496                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 497                                                          utime);
 498                 }
 499         }
 500
 501         ++head;
 502         list_for_each_entry_safe(timer, next, head, entry) {
 503                 list_del_init(&timer->entry);
 504                 if (timer->expires.sched < sum_exec_runtime) {
 505                         timer->expires.sched = 0;
 506                 } else {
 507                         timer->expires.sched -= sum_exec_runtime;
 508                 }
 509         }
 510 }
 511
 512 /*
 513  * These are both called with the siglock held, when the current thread
 514  * is being reaped.  When the final (leader) thread in the group is reaped,
 515  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 516  */
 517 void posix_cpu_timers_exit(struct task_struct *tsk)
 518 {
 519         cleanup_timers(tsk->cpu_timers,
 520                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 521
 522 }
 523 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 524 {
 525         struct signal_struct *const sig = tsk->signal;
 526
 527         cleanup_timers(tsk->signal->cpu_timers,
 528                        cputime_add(tsk->utime, sig->utime),
 529                        cputime_add(tsk->stime, sig->stime),
 530                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 531 }
 532
 533 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 534 {
 535         /*
 536          * That's all for this thread or process.
 537          * We leave our residual in expires to be reported.
 538          */
 539         put_task_struct(timer->it.cpu.task);
 540         timer->it.cpu.task = NULL;
 541         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 542                                              timer->it.cpu.expires,
 543                                              now);
 544 }
 545
 546 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 547 {
 548         return cputime_eq(expires, cputime_zero) ||
 549                cputime_gt(expires, new_exp);
 550 }
 551
 552 static inline int expires_le(cputime_t expires, cputime_t new_exp)
 553 {
 554         return !cputime_eq(expires, cputime_zero) &&
 555                cputime_le(expires, new_exp);
 556 }
 557 /*
 558  * Insert the timer on the appropriate list before any timers that
 559  * expire later.  This must be called with the tasklist_lock held
 560  * for reading, and interrupts disabled.
 561  */
 562 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 563 {
 564         struct task_struct *p = timer->it.cpu.task;
 565         struct list_head *head, *listpos;
 566         struct cpu_timer_list *const nt = &timer->it.cpu;
 567         struct cpu_timer_list *next;
 568         unsigned long i;
 569
 570         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 571                 p->cpu_timers : p->signal->cpu_timers);
 572         head += CPUCLOCK_WHICH(timer->it_clock);
 573
 574         BUG_ON(!irqs_disabled());
 575         spin_lock(&p->sighand->siglock);
 576
 577         listpos = head;
 578         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 579                 list_for_each_entry(next, head, entry) {
 580                         if (next->expires.sched > nt->expires.sched)
 581                                 break;
 582                         listpos = &next->entry;
 583                 }
 584         } else {
 585                 list_for_each_entry(next, head, entry) {
 586                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 587                                 break;
 588                         listpos = &next->entry;
 589                 }
 590         }
 591         list_add(&nt->entry, listpos);
 592
 593         if (listpos == head) {
 594                 /*
 595                  * We are the new earliest-expiring timer.
 596                  * If we are a thread timer, there can always
 597                  * be a process timer telling us to stop earlier.
 598                  */
 599
 600                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 601                         union cpu_time_count *exp = &nt->expires;
 602
 603                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 604                         default:
 605                                 BUG();
 606                         case CPUCLOCK_PROF:
 607                                 if (expires_gt(p->cputime_expires.prof_exp,
 608                                                exp->cpu))
 609                                         p->cputime_expires.prof_exp = exp->cpu;
 610                                 break;
 611                         case CPUCLOCK_VIRT:
 612                                 if (expires_gt(p->cputime_expires.virt_exp,
 613                                                exp->cpu))
 614                                         p->cputime_expires.virt_exp = exp->cpu;
 615                                 break;
 616                         case CPUCLOCK_SCHED:
 617                                 if (p->cputime_expires.sched_exp == 0 ||
 618                                     p->cputime_expires.sched_exp > exp->sched)
 619                                         p->cputime_expires.sched_exp =
 620                                                                 exp->sched;
 621                                 break;
 622                         }
 623                 } else {
 624                         struct signal_struct *const sig = p->signal;
 625                         union cpu_time_count *exp = &timer->it.cpu.expires;
 626
 627                         /*
 628                          * For a process timer, set the cached expiration time.
 629                          */
 630                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 631                         default:
 632                                 BUG();
 633                         case CPUCLOCK_VIRT:
 634                                 if (expires_le(sig->it[CPUCLOCK_VIRT].expires,
 635                                                exp->cpu))
 636                                         break;
 637                                 sig->cputime_expires.virt_exp = exp->cpu;
 638                                 break;
 639                         case CPUCLOCK_PROF:
 640                                 if (expires_le(sig->it[CPUCLOCK_PROF].expires,
 641                                                exp->cpu))
 642                                         break;
 643                                 i = sig->rlim[RLIMIT_CPU].rlim_cur;
 644                                 if (i != RLIM_INFINITY &&
 645                                     i <= cputime_to_secs(exp->cpu))
 646                                         break;
 647                                 sig->cputime_expires.prof_exp = exp->cpu;
 648                                 break;
 649                         case CPUCLOCK_SCHED:
 650                                 sig->cputime_expires.sched_exp = exp->sched;
 651                                 break;
 652                         }
 653                 }
 654         }
 655
 656         spin_unlock(&p->sighand->siglock);
 657 }
 658
 659 /*
 660  * The timer is locked, fire it and arrange for its reload.
 661  */
 662 static void cpu_timer_fire(struct k_itimer *timer)
 663 {
 664         if (unlikely(timer->sigq == NULL)) {
 665                 /*
 666                  * This a special case for clock_nanosleep,
 667                  * not a normal timer from sys_timer_create.
 668                  */
 669                 wake_up_process(timer->it_process);
 670                 timer->it.cpu.expires.sched = 0;
 671         } else if (timer->it.cpu.incr.sched == 0) {
 672                 /*
 673                  * One-shot timer.  Clear it as soon as it's fired.
 674                  */
 675                 posix_timer_event(timer, 0);
 676                 timer->it.cpu.expires.sched = 0;
 677         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 678                 /*
 679                  * The signal did not get queued because the signal
 680                  * was ignored, so we won't get any callback to
 681                  * reload the timer.  But we need to keep it
 682                  * ticking in case the signal is deliverable next time.
 683                  */
 684                 posix_cpu_timer_schedule(timer);
 685         }
 686 }
 687
 688 /*
 689  * Sample a process (thread group) timer for the given group_leader task.
 690  * Must be called with tasklist_lock held for reading.
 691  */
 692 static int cpu_timer_sample_group(const clockid_t which_clock,
 693                                   struct task_struct *p,
 694                                   union cpu_time_count *cpu)
 695 {
 696         struct task_cputime cputime;
 697
 698         thread_group_cputimer(p, &cputime);
 699         switch (CPUCLOCK_WHICH(which_clock)) {
 700         default:
 701                 return -EINVAL;
 702         case CPUCLOCK_PROF:
 703                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 704                 break;
 705         case CPUCLOCK_VIRT:
 706                 cpu->cpu = cputime.utime;
 707                 break;
 708         case CPUCLOCK_SCHED:
 709                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 710                 break;
 711         }
 712         return 0;
 713 }
 714
 715 /*
 716  * Guts of sys_timer_settime for CPU timers.
 717  * This is called with the timer locked and interrupts disabled.
 718  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 719  * and try again.  (This happens when the timer is in the middle of firing.)
 720  */
 721 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 722                         struct itimerspec *new, struct itimerspec *old)
 723 {
 724         struct task_struct *p = timer->it.cpu.task;
 725         union cpu_time_count old_expires, new_expires, val;
 726         int ret;
 727
 728         if (unlikely(p == NULL)) {
 729                 /*
 730                  * Timer refers to a dead task's clock.
 731                  */
 732                 return -ESRCH;
 733         }
 734
 735         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 736
 737         read_lock(&tasklist_lock);
 738         /*
 739          * We need the tasklist_lock to protect against reaping that
 740          * clears p->signal.  If p has just been reaped, we can no
 741          * longer get any information about it at all.
 742          */
 743         if (unlikely(p->signal == NULL)) {
 744                 read_unlock(&tasklist_lock);
 745                 put_task_struct(p);
 746                 timer->it.cpu.task = NULL;
 747                 return -ESRCH;
 748         }
 749
 750         /*
 751          * Disarm any old timer after extracting its expiry time.
 752          */
 753         BUG_ON(!irqs_disabled());
 754
 755         ret = 0;
 756         spin_lock(&p->sighand->siglock);
 757         old_expires = timer->it.cpu.expires;
 758         if (unlikely(timer->it.cpu.firing)) {
 759                 timer->it.cpu.firing = -1;
 760                 ret = TIMER_RETRY;
 761         } else
 762                 list_del_init(&timer->it.cpu.entry);
 763         spin_unlock(&p->sighand->siglock);
 764
 765         /*
 766          * We need to sample the current value to convert the new
 767          * value from to relative and absolute, and to convert the
 768          * old value from absolute to relative.  To set a process
 769          * timer, we need a sample to balance the thread expiry
 770          * times (in arm_timer).  With an absolute time, we must
 771          * check if it's already passed.  In short, we need a sample.
 772          */
 773         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 774                 cpu_clock_sample(timer->it_clock, p, &val);
 775         } else {
 776                 cpu_timer_sample_group(timer->it_clock, p, &val);
 777         }
 778
 779         if (old) {
 780                 if (old_expires.sched == 0) {
 781                         old->it_value.tv_sec = 0;
 782                         old->it_value.tv_nsec = 0;
 783                 } else {
 784                         /*
 785                          * Update the timer in case it has
 786                          * overrun already.  If it has,
 787                          * we'll report it as having overrun
 788                          * and with the next reloaded timer
 789                          * already ticking, though we are
 790                          * swallowing that pending
 791                          * notification here to install the
 792                          * new setting.
 793                          */
 794                         bump_cpu_timer(timer, val);
 795                         if (cpu_time_before(timer->it_clock, val,
 796                                             timer->it.cpu.expires)) {
 797                                 old_expires = cpu_time_sub(
 798                                         timer->it_clock,
 799                                         timer->it.cpu.expires, val);
 800                                 sample_to_timespec(timer->it_clock,
 801                                                    old_expires,
 802                                                    &old->it_value);
 803                         } else {
 804                                 old->it_value.tv_nsec = 1;
 805                                 old->it_value.tv_sec = 0;
 806                         }
 807                 }
 808         }
 809
 810         if (unlikely(ret)) {
 811                 /*
 812                  * We are colliding with the timer actually firing.
 813                  * Punt after filling in the timer's old value, and
 814                  * disable this firing since we are already reporting
 815                  * it as an overrun (thanks to bump_cpu_timer above).
 816                  */
 817                 read_unlock(&tasklist_lock);
 818                 goto out;
 819         }
 820
 821         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 822                 cpu_time_add(timer->it_clock, &new_expires, val);
 823         }
 824
 825         /*
 826          * Install the new expiry time (or zero).
 827          * For a timer with no notification action, we don't actually
 828          * arm the timer (we'll just fake it for timer_gettime).
 829          */
 830         timer->it.cpu.expires = new_expires;
 831         if (new_expires.sched != 0 &&
 832             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 833             cpu_time_before(timer->it_clock, val, new_expires)) {
 834                 arm_timer(timer, val);
 835         }
 836
 837         read_unlock(&tasklist_lock);
 838
 839         /*
 840          * Install the new reload setting, and
 841          * set up the signal and overrun bookkeeping.
 842          */
 843         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 844                                                 &new->it_interval);
 845
 846         /*
 847          * This acts as a modification timestamp for the timer,
 848          * so any automatic reload attempt will punt on seeing
 849          * that we have reset the timer manually.
 850          */
 851         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 852                 ~REQUEUE_PENDING;
 853         timer->it_overrun_last = 0;
 854         timer->it_overrun = -1;
 855
 856         if (new_expires.sched != 0 &&
 857             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 858             !cpu_time_before(timer->it_clock, val, new_expires)) {
 859                 /*
 860                  * The designated time already passed, so we notify
 861                  * immediately, even if the thread never runs to
 862                  * accumulate more time on this clock.
 863                  */
 864                 cpu_timer_fire(timer);
 865         }
 866
 867         ret = 0;
 868  out:
 869         if (old) {
 870                 sample_to_timespec(timer->it_clock,
 871                                    timer->it.cpu.incr, &old->it_interval);
 872         }
 873         return ret;
 874 }
 875
 876 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 877 {
 878         union cpu_time_count now;
 879         struct task_struct *p = timer->it.cpu.task;
 880         int clear_dead;
 881
 882         /*
 883          * Easy part: convert the reload time.
 884          */
 885         sample_to_timespec(timer->it_clock,
 886                            timer->it.cpu.incr, &itp->it_interval);
 887
 888         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 889                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 890                 return;
 891         }
 892
 893         if (unlikely(p == NULL)) {
 894                 /*
 895                  * This task already died and the timer will never fire.
 896                  * In this case, expires is actually the dead value.
 897                  */
 898         dead:
 899                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 900                                    &itp->it_value);
 901                 return;
 902         }
 903
 904         /*
 905          * Sample the clock to take the difference with the expiry time.
 906          */
 907         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 908                 cpu_clock_sample(timer->it_clock, p, &now);
 909                 clear_dead = p->exit_state;
 910         } else {
 911                 read_lock(&tasklist_lock);
 912                 if (unlikely(p->signal == NULL)) {
 913                         /*
 914                          * The process has been reaped.
 915                          * We can't even collect a sample any more.
 916                          * Call the timer disarmed, nothing else to do.
 917                          */
 918                         put_task_struct(p);
 919                         timer->it.cpu.task = NULL;
 920                         timer->it.cpu.expires.sched = 0;
 921                         read_unlock(&tasklist_lock);
 922                         goto dead;
 923                 } else {
 924                         cpu_timer_sample_group(timer->it_clock, p, &now);
 925                         clear_dead = (unlikely(p->exit_state) &&
 926                                       thread_group_empty(p));
 927                 }
 928                 read_unlock(&tasklist_lock);
 929         }
 930
 931         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 932                 if (timer->it.cpu.incr.sched == 0 &&
 933                     cpu_time_before(timer->it_clock,
 934                                     timer->it.cpu.expires, now)) {
 935                         /*
 936                          * Do-nothing timer expired and has no reload,
 937                          * so it's as if it was never set.
 938                          */
 939                         timer->it.cpu.expires.sched = 0;
 940                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 941                         return;
 942                 }
 943                 /*
 944                  * Account for any expirations and reloads that should
 945                  * have happened.
 946                  */
 947                 bump_cpu_timer(timer, now);
 948         }
 949
 950         if (unlikely(clear_dead)) {
 951                 /*
 952                  * We've noticed that the thread is dead, but
 953                  * not yet reaped.  Take this opportunity to
 954                  * drop our task ref.
 955                  */
 956                 clear_dead_task(timer, now);
 957                 goto dead;
 958         }
 959
 960         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 961                 sample_to_timespec(timer->it_clock,
 962                                    cpu_time_sub(timer->it_clock,
 963                                                 timer->it.cpu.expires, now),
 964                                    &itp->it_value);
 965         } else {
 966                 /*
 967                  * The timer should have expired already, but the firing
 968                  * hasn't taken place yet.  Say it's just about to expire.
 969                  */
 970                 itp->it_value.tv_nsec = 1;
 971                 itp->it_value.tv_sec = 0;
 972         }
 973 }
 974
 975 /*
 976  * Check for any per-thread CPU timers that have fired and move them off
 977  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 978  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 979  */
 980 static void check_thread_timers(struct task_struct *tsk,
 981                                 struct list_head *firing)
 982 {
 983         int maxfire;
 984         struct list_head *timers = tsk->cpu_timers;
 985         struct signal_struct *const sig = tsk->signal;
 986
 987         maxfire = 20;
 988         tsk->cputime_expires.prof_exp = cputime_zero;
 989         while (!list_empty(timers)) {
 990                 struct cpu_timer_list *t = list_first_entry(timers,
 991                                                       struct cpu_timer_list,
 992                                                       entry);
 993                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 994                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 995                         break;
 996                 }
 997                 t->firing = 1;
 998                 list_move_tail(&t->entry, firing);
 999         }
1000
1001         ++timers;
1002         maxfire = 20;
1003         tsk->cputime_expires.virt_exp = cputime_zero;
1004         while (!list_empty(timers)) {
1005                 struct cpu_timer_list *t = list_first_entry(timers,
1006                                                       struct cpu_timer_list,
1007                                                       entry);
1008                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1009                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1010                         break;
1011                 }
1012                 t->firing = 1;
1013                 list_move_tail(&t->entry, firing);
1014         }
1015
1016         ++timers;
1017         maxfire = 20;
1018         tsk->cputime_expires.sched_exp = 0;
1019         while (!list_empty(timers)) {
1020                 struct cpu_timer_list *t = list_first_entry(timers,
1021                                                       struct cpu_timer_list,
1022                                                       entry);
1023                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1024                         tsk->cputime_expires.sched_exp = t->expires.sched;
1025                         break;
1026                 }
1027                 t->firing = 1;
1028                 list_move_tail(&t->entry, firing);
1029         }
1030
1031         /*
1032          * Check for the special case thread timers.
1033          */
1034         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1035                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1036                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1037
1038                 if (hard != RLIM_INFINITY &&
1039                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1040                         /*
1041                          * At the hard limit, we just die.
1042                          * No need to calculate anything else now.
1043                          */
1044                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1045                         return;
1046                 }
1047                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1048                         /*
1049                          * At the soft limit, send a SIGXCPU every second.
1050                          */
1051                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1052                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1053                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1054                                                                 USEC_PER_SEC;
1055                         }
1056                         printk(KERN_INFO
1057                                 "RT Watchdog Timeout: %s[%d]\n",
1058                                 tsk->comm, task_pid_nr(tsk));
1059                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1060                 }
1061         }
1062 }
1063
1064 static void stop_process_timers(struct task_struct *tsk)
1065 {
1066         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1067         unsigned long flags;
1068
1069         if (!cputimer->running)
1070                 return;
1071
1072         spin_lock_irqsave(&cputimer->lock, flags);
1073         cputimer->running = 0;
1074         spin_unlock_irqrestore(&cputimer->lock, flags);
1075 }
1076
1077 static u32 onecputick;
1078
1079 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1080                              cputime_t *expires, cputime_t cur_time, int signo)
1081 {
1082         if (cputime_eq(it->expires, cputime_zero))
1083                 return;
1084
1085         if (cputime_ge(cur_time, it->expires)) {
1086                 if (!cputime_eq(it->incr, cputime_zero)) {
1087                         it->expires = cputime_add(it->expires, it->incr);
1088                         it->error += it->incr_error;
1089                         if (it->error >= onecputick) {
1090                                 it->expires = cputime_sub(it->expires,
1091                                                           cputime_one_jiffy);
1092                                 it->error -= onecputick;
1093                         }
1094                 } else {
1095                         it->expires = cputime_zero;
1096                 }
1097
1098                 trace_itimer_expire(signo == SIGPROF ?
1099                                     ITIMER_PROF : ITIMER_VIRTUAL,
1100                                     tsk->signal->leader_pid, cur_time);
1101                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1102         }
1103
1104         if (!cputime_eq(it->expires, cputime_zero) &&
1105             (cputime_eq(*expires, cputime_zero) ||
1106              cputime_lt(it->expires, *expires))) {
1107                 *expires = it->expires;
1108         }
1109 }
1110
1111 /*
1112  * Check for any per-thread CPU timers that have fired and move them
1113  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1114  * have already been taken off.
1115  */
1116 static void check_process_timers(struct task_struct *tsk,
1117                                  struct list_head *firing)
1118 {
1119         int maxfire;
1120         struct signal_struct *const sig = tsk->signal;
1121         cputime_t utime, ptime, virt_expires, prof_expires;
1122         unsigned long long sum_sched_runtime, sched_expires;
1123         struct list_head *timers = sig->cpu_timers;
1124         struct task_cputime cputime;
1125
1126         /*
1127          * Don't sample the current process CPU clocks if there are no timers.
1128          */
1129         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1130             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1131             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1132             list_empty(&timers[CPUCLOCK_VIRT]) &&
1133             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1134             list_empty(&timers[CPUCLOCK_SCHED])) {
1135                 stop_process_timers(tsk);
1136                 return;
1137         }
1138
1139         /*
1140          * Collect the current process totals.
1141          */
1142         thread_group_cputimer(tsk, &cputime);
1143         utime = cputime.utime;
1144         ptime = cputime_add(utime, cputime.stime);
1145         sum_sched_runtime = cputime.sum_exec_runtime;
1146         maxfire = 20;
1147         prof_expires = cputime_zero;
1148         while (!list_empty(timers)) {
1149                 struct cpu_timer_list *tl = list_first_entry(timers,
1150                                                       struct cpu_timer_list,
1151                                                       entry);
1152                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1153                         prof_expires = tl->expires.cpu;
1154                         break;
1155                 }
1156                 tl->firing = 1;
1157                 list_move_tail(&tl->entry, firing);
1158         }
1159
1160         ++timers;
1161         maxfire = 20;
1162         virt_expires = cputime_zero;
1163         while (!list_empty(timers)) {
1164                 struct cpu_timer_list *tl = list_first_entry(timers,
1165                                                       struct cpu_timer_list,
1166                                                       entry);
1167                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1168                         virt_expires = tl->expires.cpu;
1169                         break;
1170                 }
1171                 tl->firing = 1;
1172                 list_move_tail(&tl->entry, firing);
1173         }
1174
1175         ++timers;
1176         maxfire = 20;
1177         sched_expires = 0;
1178         while (!list_empty(timers)) {
1179                 struct cpu_timer_list *tl = list_first_entry(timers,
1180                                                       struct cpu_timer_list,
1181                                                       entry);
1182                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1183                         sched_expires = tl->expires.sched;
1184                         break;
1185                 }
1186                 tl->firing = 1;
1187                 list_move_tail(&tl->entry, firing);
1188         }
1189
1190         /*
1191          * Check for the special case process timers.
1192          */
1193         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1194                          SIGPROF);
1195         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1196                          SIGVTALRM);
1197
1198         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1199                 unsigned long psecs = cputime_to_secs(ptime);
1200                 cputime_t x;
1201                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1202                         /*
1203                          * At the hard limit, we just die.
1204                          * No need to calculate anything else now.
1205                          */
1206                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1207                         return;
1208                 }
1209                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1210                         /*
1211                          * At the soft limit, send a SIGXCPU every second.
1212                          */
1213                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1214                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1215                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1216                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1217                         }
1218                 }
1219                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1220                 if (cputime_eq(prof_expires, cputime_zero) ||
1221                     cputime_lt(x, prof_expires)) {
1222                         prof_expires = x;
1223                 }
1224         }
1225
1226         if (!cputime_eq(prof_expires, cputime_zero) &&
1227             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1228              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1229                 sig->cputime_expires.prof_exp = prof_expires;
1230         if (!cputime_eq(virt_expires, cputime_zero) &&
1231             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1232              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1233                 sig->cputime_expires.virt_exp = virt_expires;
1234         if (sched_expires != 0 &&
1235             (sig->cputime_expires.sched_exp == 0 ||
1236              sig->cputime_expires.sched_exp > sched_expires))
1237                 sig->cputime_expires.sched_exp = sched_expires;
1238 }
1239
1240 /*
1241  * This is called from the signal code (via do_schedule_next_timer)
1242  * when the last timer signal was delivered and we have to reload the timer.
1243  */
1244 void posix_cpu_timer_schedule(struct k_itimer *timer)
1245 {
1246         struct task_struct *p = timer->it.cpu.task;
1247         union cpu_time_count now;
1248
1249         if (unlikely(p == NULL))
1250                 /*
1251                  * The task was cleaned up already, no future firings.
1252                  */
1253                 goto out;
1254
1255         /*
1256          * Fetch the current sample and update the timer's expiry time.
1257          */
1258         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1259                 cpu_clock_sample(timer->it_clock, p, &now);
1260                 bump_cpu_timer(timer, now);
1261                 if (unlikely(p->exit_state)) {
1262                         clear_dead_task(timer, now);
1263                         goto out;
1264                 }
1265                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1266         } else {
1267                 read_lock(&tasklist_lock);
1268                 if (unlikely(p->signal == NULL)) {
1269                         /*
1270                          * The process has been reaped.
1271                          * We can't even collect a sample any more.
1272                          */
1273                         put_task_struct(p);
1274                         timer->it.cpu.task = p = NULL;
1275                         timer->it.cpu.expires.sched = 0;
1276                         goto out_unlock;
1277                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1278                         /*
1279                          * We've noticed that the thread is dead, but
1280                          * not yet reaped.  Take this opportunity to
1281                          * drop our task ref.
1282                          */
1283                         clear_dead_task(timer, now);
1284                         goto out_unlock;
1285                 }
1286                 cpu_timer_sample_group(timer->it_clock, p, &now);
1287                 bump_cpu_timer(timer, now);
1288                 /* Leave the tasklist_lock locked for the call below.  */
1289         }
1290
1291         /*
1292          * Now re-arm for the new expiry time.
1293          */
1294         arm_timer(timer, now);
1295
1296 out_unlock:
1297         read_unlock(&tasklist_lock);
1298
1299 out:
1300         timer->it_overrun_last = timer->it_overrun;
1301         timer->it_overrun = -1;
1302         ++timer->it_requeue_pending;
1303 }
1304
1305 /**
1306  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1307  *
1308  * @cputime:    The struct to compare.
1309  *
1310  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1311  * are zero, false if any field is nonzero.
1312  */
1313 static inline int task_cputime_zero(const struct task_cputime *cputime)
1314 {
1315         if (cputime_eq(cputime->utime, cputime_zero) &&
1316             cputime_eq(cputime->stime, cputime_zero) &&
1317             cputime->sum_exec_runtime == 0)
1318                 return 1;
1319         return 0;
1320 }
1321
1322 /**
1323  * task_cputime_expired - Compare two task_cputime entities.
1324  *
1325  * @sample:     The task_cputime structure to be checked for expiration.
1326  * @expires:    Expiration times, against which @sample will be checked.
1327  *
1328  * Checks @sample against @expires to see if any field of @sample has expired.
1329  * Returns true if any field of the former is greater than the corresponding
1330  * field of the latter if the latter field is set.  Otherwise returns false.
1331  */
1332 static inline int task_cputime_expired(const struct task_cputime *sample,
1333                                         const struct task_cputime *expires)
1334 {
1335         if (!cputime_eq(expires->utime, cputime_zero) &&
1336             cputime_ge(sample->utime, expires->utime))
1337                 return 1;
1338         if (!cputime_eq(expires->stime, cputime_zero) &&
1339             cputime_ge(cputime_add(sample->utime, sample->stime),
1340                        expires->stime))
1341                 return 1;
1342         if (expires->sum_exec_runtime != 0 &&
1343             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1344                 return 1;
1345         return 0;
1346 }
1347
1348 /**
1349  * fastpath_timer_check - POSIX CPU timers fast path.
1350  *
1351  * @tsk:        The task (thread) being checked.
1352  *
1353  * Check the task and thread group timers.  If both are zero (there are no
1354  * timers set) return false.  Otherwise snapshot the task and thread group
1355  * timers and compare them with the corresponding expiration times.  Return
1356  * true if a timer has expired, else return false.
1357  */
1358 static inline int fastpath_timer_check(struct task_struct *tsk)
1359 {
1360         struct signal_struct *sig;
1361
1362         /* tsk == current, ensure it is safe to use ->signal/sighand */
1363         if (unlikely(tsk->exit_state))
1364                 return 0;
1365
1366         if (!task_cputime_zero(&tsk->cputime_expires)) {
1367                 struct task_cputime task_sample = {
1368                         .utime = tsk->utime,
1369                         .stime = tsk->stime,
1370                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1371                 };
1372
1373                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1374                         return 1;
1375         }
1376
1377         sig = tsk->signal;
1378         if (!task_cputime_zero(&sig->cputime_expires)) {
1379                 struct task_cputime group_sample;
1380
1381                 thread_group_cputimer(tsk, &group_sample);
1382                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1383                         return 1;
1384         }
1385
1386         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1387 }
1388
1389 /*
1390  * This is called from the timer interrupt handler.  The irq handler has
1391  * already updated our counts.  We need to check if any timers fire now.
1392  * Interrupts are disabled.
1393  */
1394 void run_posix_cpu_timers(struct task_struct *tsk)
1395 {
1396         LIST_HEAD(firing);
1397         struct k_itimer *timer, *next;
1398
1399         BUG_ON(!irqs_disabled());
1400
1401         /*
1402          * The fast path checks that there are no expired thread or thread
1403          * group timers.  If that's so, just return.
1404          */
1405         if (!fastpath_timer_check(tsk))
1406                 return;
1407
1408         spin_lock(&tsk->sighand->siglock);
1409         /*
1410          * Here we take off tsk->signal->cpu_timers[N] and
1411          * tsk->cpu_timers[N] all the timers that are firing, and
1412          * put them on the firing list.
1413          */
1414         check_thread_timers(tsk, &firing);
1415         check_process_timers(tsk, &firing);
1416
1417         /*
1418          * We must release these locks before taking any timer's lock.
1419          * There is a potential race with timer deletion here, as the
1420          * siglock now protects our private firing list.  We have set
1421          * the firing flag in each timer, so that a deletion attempt
1422          * that gets the timer lock before we do will give it up and
1423          * spin until we've taken care of that timer below.
1424          */
1425         spin_unlock(&tsk->sighand->siglock);
1426
1427         /*
1428          * Now that all the timers on our list have the firing flag,
1429          * noone will touch their list entries but us.  We'll take
1430          * each timer's lock before clearing its firing flag, so no
1431          * timer call will interfere.
1432          */
1433         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1434                 int cpu_firing;
1435
1436                 spin_lock(&timer->it_lock);
1437                 list_del_init(&timer->it.cpu.entry);
1438                 cpu_firing = timer->it.cpu.firing;
1439                 timer->it.cpu.firing = 0;
1440                 /*
1441                  * The firing flag is -1 if we collided with a reset
1442                  * of the timer, which already reported this
1443                  * almost-firing as an overrun.  So don't generate an event.
1444                  */
1445                 if (likely(cpu_firing >= 0))
1446                         cpu_timer_fire(timer);
1447                 spin_unlock(&timer->it_lock);
1448         }
1449 }
1450
1451 /*
1452  * Set one of the process-wide special case CPU timers.
1453  * The tsk->sighand->siglock must be held by the caller.
1454  * The *newval argument is relative and we update it to be absolute, *oldval
1455  * is absolute and we update it to be relative.
1456  */
1457 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1458                            cputime_t *newval, cputime_t *oldval)
1459 {
1460         union cpu_time_count now;
1461         struct list_head *head;
1462
1463         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1464         cpu_timer_sample_group(clock_idx, tsk, &now);
1465
1466         if (oldval) {
1467                 if (!cputime_eq(*oldval, cputime_zero)) {
1468                         if (cputime_le(*oldval, now.cpu)) {
1469                                 /* Just about to fire. */
1470                                 *oldval = cputime_one_jiffy;
1471                         } else {
1472                                 *oldval = cputime_sub(*oldval, now.cpu);
1473                         }
1474                 }
1475
1476                 if (cputime_eq(*newval, cputime_zero))
1477                         return;
1478                 *newval = cputime_add(*newval, now.cpu);
1479
1480                 /*
1481                  * If the RLIMIT_CPU timer will expire before the
1482                  * ITIMER_PROF timer, we have nothing else to do.
1483                  */
1484                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1485                     < cputime_to_secs(*newval))
1486                         return;
1487         }
1488
1489         /*
1490          * Check whether there are any process timers already set to fire
1491          * before this one.  If so, we don't have anything more to do.
1492          */
1493         head = &tsk->signal->cpu_timers[clock_idx];
1494         if (list_empty(head) ||
1495             cputime_ge(list_first_entry(head,
1496                                   struct cpu_timer_list, entry)->expires.cpu,
1497                        *newval)) {
1498                 switch (clock_idx) {
1499                 case CPUCLOCK_PROF:
1500                         tsk->signal->cputime_expires.prof_exp = *newval;
1501                         break;
1502                 case CPUCLOCK_VIRT:
1503                         tsk->signal->cputime_expires.virt_exp = *newval;
1504                         break;
1505                 }
1506         }
1507 }
1508
1509 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1510                             struct timespec *rqtp, struct itimerspec *it)
1511 {
1512         struct k_itimer timer;
1513         int error;
1514
1515         /*
1516          * Set up a temporary timer and then wait for it to go off.
1517          */
1518         memset(&timer, 0, sizeof timer);
1519         spin_lock_init(&timer.it_lock);
1520         timer.it_clock = which_clock;
1521         timer.it_overrun = -1;
1522         error = posix_cpu_timer_create(&timer);
1523         timer.it_process = current;
1524         if (!error) {
1525                 static struct itimerspec zero_it;
1526
1527                 memset(it, 0, sizeof *it);
1528                 it->it_value = *rqtp;
1529
1530                 spin_lock_irq(&timer.it_lock);
1531                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1532                 if (error) {
1533                         spin_unlock_irq(&timer.it_lock);
1534                         return error;
1535                 }
1536
1537                 while (!signal_pending(current)) {
1538                         if (timer.it.cpu.expires.sched == 0) {
1539                                 /*
1540                                  * Our timer fired and was reset.
1541                                  */
1542                                 spin_unlock_irq(&timer.it_lock);
1543                                 return 0;
1544                         }
1545
1546                         /*
1547                          * Block until cpu_timer_fire (or a signal) wakes us.
1548                          */
1549                         __set_current_state(TASK_INTERRUPTIBLE);
1550                         spin_unlock_irq(&timer.it_lock);
1551                         schedule();
1552                         spin_lock_irq(&timer.it_lock);
1553                 }
1554
1555                 /*
1556                  * We were interrupted by a signal.
1557                  */
1558                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1559                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1560                 spin_unlock_irq(&timer.it_lock);
1561
1562                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1563                         /*
1564                          * It actually did fire already.
1565                          */
1566                         return 0;
1567                 }
1568
1569                 error = -ERESTART_RESTARTBLOCK;
1570         }
1571
1572         return error;
1573 }
1574
1575 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1576                      struct timespec *rqtp, struct timespec __user *rmtp)
1577 {
1578         struct restart_block *restart_block =
1579             &current_thread_info()->restart_block;
1580         struct itimerspec it;
1581         int error;
1582
1583         /*
1584          * Diagnose required errors first.
1585          */
1586         if (CPUCLOCK_PERTHREAD(which_clock) &&
1587             (CPUCLOCK_PID(which_clock) == 0 ||
1588              CPUCLOCK_PID(which_clock) == current->pid))
1589                 return -EINVAL;
1590
1591         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1592
1593         if (error == -ERESTART_RESTARTBLOCK) {
1594
1595                 if (flags & TIMER_ABSTIME)
1596                         return -ERESTARTNOHAND;
1597                 /*
1598                  * Report back to the user the time still remaining.
1599                  */
1600                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1601                         return -EFAULT;
1602
1603                 restart_block->fn = posix_cpu_nsleep_restart;
1604                 restart_block->arg0 = which_clock;
1605                 restart_block->arg1 = (unsigned long) rmtp;
1606                 restart_block->arg2 = rqtp->tv_sec;
1607                 restart_block->arg3 = rqtp->tv_nsec;
1608         }
1609         return error;
1610 }
1611
1612 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1613 {
1614         clockid_t which_clock = restart_block->arg0;
1615         struct timespec __user *rmtp;
1616         struct timespec t;
1617         struct itimerspec it;
1618         int error;
1619
1620         rmtp = (struct timespec __user *) restart_block->arg1;
1621         t.tv_sec = restart_block->arg2;
1622         t.tv_nsec = restart_block->arg3;
1623
1624         restart_block->fn = do_no_restart_syscall;
1625         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1626
1627         if (error == -ERESTART_RESTARTBLOCK) {
1628                 /*
1629                  * Report back to the user the time still remaining.
1630                  */
1631                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1632                         return -EFAULT;
1633
1634                 restart_block->fn = posix_cpu_nsleep_restart;
1635                 restart_block->arg0 = which_clock;
1636                 restart_block->arg1 = (unsigned long) rmtp;
1637                 restart_block->arg2 = t.tv_sec;
1638                 restart_block->arg3 = t.tv_nsec;
1639         }
1640         return error;
1641
1642 }
1643
1644
1645 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1646 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1647
1648 static int process_cpu_clock_getres(const clockid_t which_clock,
1649                                     struct timespec *tp)
1650 {
1651         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1652 }
1653 static int process_cpu_clock_get(const clockid_t which_clock,
1654                                  struct timespec *tp)
1655 {
1656         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1657 }
1658 static int process_cpu_timer_create(struct k_itimer *timer)
1659 {
1660         timer->it_clock = PROCESS_CLOCK;
1661         return posix_cpu_timer_create(timer);
1662 }
1663 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1664                               struct timespec *rqtp,
1665                               struct timespec __user *rmtp)
1666 {
1667         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1668 }
1669 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1670 {
1671         return -EINVAL;
1672 }
1673 static int thread_cpu_clock_getres(const clockid_t which_clock,
1674                                    struct timespec *tp)
1675 {
1676         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1677 }
1678 static int thread_cpu_clock_get(const clockid_t which_clock,
1679                                 struct timespec *tp)
1680 {
1681         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1682 }
1683 static int thread_cpu_timer_create(struct k_itimer *timer)
1684 {
1685         timer->it_clock = THREAD_CLOCK;
1686         return posix_cpu_timer_create(timer);
1687 }
1688 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1689                               struct timespec *rqtp, struct timespec __user *rmtp)
1690 {
1691         return -EINVAL;
1692 }
1693 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1694 {
1695         return -EINVAL;
1696 }
1697
1698 static __init int init_posix_cpu_timers(void)
1699 {
1700         struct k_clock process = {
1701                 .clock_getres = process_cpu_clock_getres,
1702                 .clock_get = process_cpu_clock_get,
1703                 .clock_set = do_posix_clock_nosettime,
1704                 .timer_create = process_cpu_timer_create,
1705                 .nsleep = process_cpu_nsleep,
1706                 .nsleep_restart = process_cpu_nsleep_restart,
1707         };
1708         struct k_clock thread = {
1709                 .clock_getres = thread_cpu_clock_getres,
1710                 .clock_get = thread_cpu_clock_get,
1711                 .clock_set = do_posix_clock_nosettime,
1712                 .timer_create = thread_cpu_timer_create,
1713                 .nsleep = thread_cpu_nsleep,
1714                 .nsleep_restart = thread_cpu_nsleep_restart,
1715         };
1716         struct timespec ts;
1717
1718         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1719         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1720
1721         cputime_to_timespec(cputime_one_jiffy, &ts);
1722         onecputick = ts.tv_nsec;
1723         WARN_ON(ts.tv_sec != 0);
1724
1725         return 0;
1726 }
1727 __initcall(init_posix_cpu_timers);