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 <asm/uaccess.h>
   8 #include <linux/errno.h>
   9
  10 static int check_clock(clockid_t which_clock)
  11 {
  12         int error = 0;
  13         struct task_struct *p;
  14         const pid_t pid = CPUCLOCK_PID(which_clock);
  15
  16         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  17                 return -EINVAL;
  18
  19         if (pid == 0)
  20                 return 0;
  21
  22         read_lock(&tasklist_lock);
  23         p = find_task_by_pid(pid);
  24         if (!p || (CPUCLOCK_PERTHREAD(which_clock) ?
  25                    p->tgid != current->tgid : p->tgid != pid)) {
  26                 error = -EINVAL;
  27         }
  28         read_unlock(&tasklist_lock);
  29
  30         return error;
  31 }
  32
  33 static inline union cpu_time_count
  34 timespec_to_sample(clockid_t which_clock, const struct timespec *tp)
  35 {
  36         union cpu_time_count ret;
  37         ret.sched = 0;          /* high half always zero when .cpu used */
  38         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  39                 ret.sched = tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  40         } else {
  41                 ret.cpu = timespec_to_cputime(tp);
  42         }
  43         return ret;
  44 }
  45
  46 static void sample_to_timespec(clockid_t which_clock,
  47                                union cpu_time_count cpu,
  48                                struct timespec *tp)
  49 {
  50         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  51                 tp->tv_sec = div_long_long_rem(cpu.sched,
  52                                                NSEC_PER_SEC, &tp->tv_nsec);
  53         } else {
  54                 cputime_to_timespec(cpu.cpu, tp);
  55         }
  56 }
  57
  58 static inline int cpu_time_before(clockid_t which_clock,
  59                                   union cpu_time_count now,
  60                                   union cpu_time_count then)
  61 {
  62         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  63                 return now.sched < then.sched;
  64         }  else {
  65                 return cputime_lt(now.cpu, then.cpu);
  66         }
  67 }
  68 static inline void cpu_time_add(clockid_t which_clock,
  69                                 union cpu_time_count *acc,
  70                                 union cpu_time_count val)
  71 {
  72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  73                 acc->sched += val.sched;
  74         }  else {
  75                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  76         }
  77 }
  78 static inline union cpu_time_count cpu_time_sub(clockid_t which_clock,
  79                                                 union cpu_time_count a,
  80                                                 union cpu_time_count b)
  81 {
  82         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  83                 a.sched -= b.sched;
  84         }  else {
  85                 a.cpu = cputime_sub(a.cpu, b.cpu);
  86         }
  87         return a;
  88 }
  89
  90 /*
  91  * Update expiry time from increment, and increase overrun count,
  92  * given the current clock sample.
  93  */
  94 static inline void bump_cpu_timer(struct k_itimer *timer,
  95                                   union cpu_time_count now)
  96 {
  97         int i;
  98
  99         if (timer->it.cpu.incr.sched == 0)
 100                 return;
 101
 102         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 103                 unsigned long long delta, incr;
 104
 105                 if (now.sched < timer->it.cpu.expires.sched)
 106                         return;
 107                 incr = timer->it.cpu.incr.sched;
 108                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 109                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 110                 for (i = 0; incr < delta - incr; i++)
 111                         incr = incr << 1;
 112                 for (; i >= 0; incr >>= 1, i--) {
 113                         if (delta <= incr)
 114                                 continue;
 115                         timer->it.cpu.expires.sched += incr;
 116                         timer->it_overrun += 1 << i;
 117                         delta -= incr;
 118                 }
 119         } else {
 120                 cputime_t delta, incr;
 121
 122                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 123                         return;
 124                 incr = timer->it.cpu.incr.cpu;
 125                 delta = cputime_sub(cputime_add(now.cpu, incr),
 126                                     timer->it.cpu.expires.cpu);
 127                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 128                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 129                              incr = cputime_add(incr, incr);
 130                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 131                         if (cputime_le(delta, incr))
 132                                 continue;
 133                         timer->it.cpu.expires.cpu =
 134                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 135                         timer->it_overrun += 1 << i;
 136                         delta = cputime_sub(delta, incr);
 137                 }
 138         }
 139 }
 140
 141 static inline cputime_t prof_ticks(struct task_struct *p)
 142 {
 143         return cputime_add(p->utime, p->stime);
 144 }
 145 static inline cputime_t virt_ticks(struct task_struct *p)
 146 {
 147         return p->utime;
 148 }
 149 static inline unsigned long long sched_ns(struct task_struct *p)
 150 {
 151         return (p == current) ? current_sched_time(p) : p->sched_time;
 152 }
 153
 154 int posix_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
 155 {
 156         int error = check_clock(which_clock);
 157         if (!error) {
 158                 tp->tv_sec = 0;
 159                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 160                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 161                         /*
 162                          * If sched_clock is using a cycle counter, we
 163                          * don't have any idea of its true resolution
 164                          * exported, but it is much more than 1s/HZ.
 165                          */
 166                         tp->tv_nsec = 1;
 167                 }
 168         }
 169         return error;
 170 }
 171
 172 int posix_cpu_clock_set(clockid_t which_clock, const struct timespec *tp)
 173 {
 174         /*
 175          * You can never reset a CPU clock, but we check for other errors
 176          * in the call before failing with EPERM.
 177          */
 178         int error = check_clock(which_clock);
 179         if (error == 0) {
 180                 error = -EPERM;
 181         }
 182         return error;
 183 }
 184
 185
 186 /*
 187  * Sample a per-thread clock for the given task.
 188  */
 189 static int cpu_clock_sample(clockid_t which_clock, struct task_struct *p,
 190                             union cpu_time_count *cpu)
 191 {
 192         switch (CPUCLOCK_WHICH(which_clock)) {
 193         default:
 194                 return -EINVAL;
 195         case CPUCLOCK_PROF:
 196                 cpu->cpu = prof_ticks(p);
 197                 break;
 198         case CPUCLOCK_VIRT:
 199                 cpu->cpu = virt_ticks(p);
 200                 break;
 201         case CPUCLOCK_SCHED:
 202                 cpu->sched = sched_ns(p);
 203                 break;
 204         }
 205         return 0;
 206 }
 207
 208 /*
 209  * Sample a process (thread group) clock for the given group_leader task.
 210  * Must be called with tasklist_lock held for reading.
 211  * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
 212  */
 213 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
 214                                          struct task_struct *p,
 215                                          union cpu_time_count *cpu)
 216 {
 217         struct task_struct *t = p;
 218         switch (clock_idx) {
 219         default:
 220                 return -EINVAL;
 221         case CPUCLOCK_PROF:
 222                 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
 223                 do {
 224                         cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
 225                         t = next_thread(t);
 226                 } while (t != p);
 227                 break;
 228         case CPUCLOCK_VIRT:
 229                 cpu->cpu = p->signal->utime;
 230                 do {
 231                         cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
 232                         t = next_thread(t);
 233                 } while (t != p);
 234                 break;
 235         case CPUCLOCK_SCHED:
 236                 cpu->sched = p->signal->sched_time;
 237                 /* Add in each other live thread.  */
 238                 while ((t = next_thread(t)) != p) {
 239                         cpu->sched += t->sched_time;
 240                 }
 241                 if (p->tgid == current->tgid) {
 242                         /*
 243                          * We're sampling ourselves, so include the
 244                          * cycles not yet banked.  We still omit
 245                          * other threads running on other CPUs,
 246                          * so the total can always be behind as
 247                          * much as max(nthreads-1,ncpus) * (NSEC_PER_SEC/HZ).
 248                          */
 249                         cpu->sched += current_sched_time(current);
 250                 } else {
 251                         cpu->sched += p->sched_time;
 252                 }
 253                 break;
 254         }
 255         return 0;
 256 }
 257
 258 /*
 259  * Sample a process (thread group) clock for the given group_leader task.
 260  * Must be called with tasklist_lock held for reading.
 261  */
 262 static int cpu_clock_sample_group(clockid_t which_clock,
 263                                   struct task_struct *p,
 264                                   union cpu_time_count *cpu)
 265 {
 266         int ret;
 267         unsigned long flags;
 268         spin_lock_irqsave(&p->sighand->siglock, flags);
 269         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
 270                                             cpu);
 271         spin_unlock_irqrestore(&p->sighand->siglock, flags);
 272         return ret;
 273 }
 274
 275
 276 int posix_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
 277 {
 278         const pid_t pid = CPUCLOCK_PID(which_clock);
 279         int error = -EINVAL;
 280         union cpu_time_count rtn;
 281
 282         if (pid == 0) {
 283                 /*
 284                  * Special case constant value for our own clocks.
 285                  * We don't have to do any lookup to find ourselves.
 286                  */
 287                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 288                         /*
 289                          * Sampling just ourselves we can do with no locking.
 290                          */
 291                         error = cpu_clock_sample(which_clock,
 292                                                  current, &rtn);
 293                 } else {
 294                         read_lock(&tasklist_lock);
 295                         error = cpu_clock_sample_group(which_clock,
 296                                                        current, &rtn);
 297                         read_unlock(&tasklist_lock);
 298                 }
 299         } else {
 300                 /*
 301                  * Find the given PID, and validate that the caller
 302                  * should be able to see it.
 303                  */
 304                 struct task_struct *p;
 305                 read_lock(&tasklist_lock);
 306                 p = find_task_by_pid(pid);
 307                 if (p) {
 308                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 309                                 if (p->tgid == current->tgid) {
 310                                         error = cpu_clock_sample(which_clock,
 311                                                                  p, &rtn);
 312                                 }
 313                         } else if (p->tgid == pid && p->signal) {
 314                                 error = cpu_clock_sample_group(which_clock,
 315                                                                p, &rtn);
 316                         }
 317                 }
 318                 read_unlock(&tasklist_lock);
 319         }
 320
 321         if (error)
 322                 return error;
 323         sample_to_timespec(which_clock, rtn, tp);
 324         return 0;
 325 }
 326
 327
 328 /*
 329  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 330  * This is called from sys_timer_create with the new timer already locked.
 331  */
 332 int posix_cpu_timer_create(struct k_itimer *new_timer)
 333 {
 334         int ret = 0;
 335         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 336         struct task_struct *p;
 337
 338         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 339                 return -EINVAL;
 340
 341         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 342         new_timer->it.cpu.incr.sched = 0;
 343         new_timer->it.cpu.expires.sched = 0;
 344
 345         read_lock(&tasklist_lock);
 346         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 347                 if (pid == 0) {
 348                         p = current;
 349                 } else {
 350                         p = find_task_by_pid(pid);
 351                         if (p && p->tgid != current->tgid)
 352                                 p = NULL;
 353                 }
 354         } else {
 355                 if (pid == 0) {
 356                         p = current->group_leader;
 357                 } else {
 358                         p = find_task_by_pid(pid);
 359                         if (p && p->tgid != pid)
 360                                 p = NULL;
 361                 }
 362         }
 363         new_timer->it.cpu.task = p;
 364         if (p) {
 365                 get_task_struct(p);
 366         } else {
 367                 ret = -EINVAL;
 368         }
 369         read_unlock(&tasklist_lock);
 370
 371         return ret;
 372 }
 373
 374 /*
 375  * Clean up a CPU-clock timer that is about to be destroyed.
 376  * This is called from timer deletion with the timer already locked.
 377  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 378  * and try again.  (This happens when the timer is in the middle of firing.)
 379  */
 380 int posix_cpu_timer_del(struct k_itimer *timer)
 381 {
 382         struct task_struct *p = timer->it.cpu.task;
 383         int ret = 0;
 384
 385         if (likely(p != NULL)) {
 386                 read_lock(&tasklist_lock);
 387                 if (unlikely(p->signal == NULL)) {
 388                         /*
 389                          * We raced with the reaping of the task.
 390                          * The deletion should have cleared us off the list.
 391                          */
 392                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 393                 } else {
 394                         spin_lock(&p->sighand->siglock);
 395                         if (timer->it.cpu.firing)
 396                                 ret = TIMER_RETRY;
 397                         else
 398                                 list_del(&timer->it.cpu.entry);
 399                         spin_unlock(&p->sighand->siglock);
 400                 }
 401                 read_unlock(&tasklist_lock);
 402
 403                 if (!ret)
 404                         put_task_struct(p);
 405         }
 406
 407         return ret;
 408 }
 409
 410 /*
 411  * Clean out CPU timers still ticking when a thread exited.  The task
 412  * pointer is cleared, and the expiry time is replaced with the residual
 413  * time for later timer_gettime calls to return.
 414  * This must be called with the siglock held.
 415  */
 416 static void cleanup_timers(struct list_head *head,
 417                            cputime_t utime, cputime_t stime,
 418                            unsigned long long sched_time)
 419 {
 420         struct cpu_timer_list *timer, *next;
 421         cputime_t ptime = cputime_add(utime, stime);
 422
 423         list_for_each_entry_safe(timer, next, head, entry) {
 424                 list_del_init(&timer->entry);
 425                 if (cputime_lt(timer->expires.cpu, ptime)) {
 426                         timer->expires.cpu = cputime_zero;
 427                 } else {
 428                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 429                                                          ptime);
 430                 }
 431         }
 432
 433         ++head;
 434         list_for_each_entry_safe(timer, next, head, entry) {
 435                 list_del_init(&timer->entry);
 436                 if (cputime_lt(timer->expires.cpu, utime)) {
 437                         timer->expires.cpu = cputime_zero;
 438                 } else {
 439                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 440                                                          utime);
 441                 }
 442         }
 443
 444         ++head;
 445         list_for_each_entry_safe(timer, next, head, entry) {
 446                 list_del_init(&timer->entry);
 447                 if (timer->expires.sched < sched_time) {
 448                         timer->expires.sched = 0;
 449                 } else {
 450                         timer->expires.sched -= sched_time;
 451                 }
 452         }
 453 }
 454
 455 /*
 456  * These are both called with the siglock held, when the current thread
 457  * is being reaped.  When the final (leader) thread in the group is reaped,
 458  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 459  */
 460 void posix_cpu_timers_exit(struct task_struct *tsk)
 461 {
 462         cleanup_timers(tsk->cpu_timers,
 463                        tsk->utime, tsk->stime, tsk->sched_time);
 464
 465 }
 466 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 467 {
 468         cleanup_timers(tsk->signal->cpu_timers,
 469                        cputime_add(tsk->utime, tsk->signal->utime),
 470                        cputime_add(tsk->stime, tsk->signal->stime),
 471                        tsk->sched_time + tsk->signal->sched_time);
 472 }
 473
 474
 475 /*
 476  * Set the expiry times of all the threads in the process so one of them
 477  * will go off before the process cumulative expiry total is reached.
 478  */
 479 static void process_timer_rebalance(struct task_struct *p,
 480                                     unsigned int clock_idx,
 481                                     union cpu_time_count expires,
 482                                     union cpu_time_count val)
 483 {
 484         cputime_t ticks, left;
 485         unsigned long long ns, nsleft;
 486         struct task_struct *t = p;
 487         unsigned int nthreads = atomic_read(&p->signal->live);
 488
 489         if (!nthreads)
 490                 return;
 491
 492         switch (clock_idx) {
 493         default:
 494                 BUG();
 495                 break;
 496         case CPUCLOCK_PROF:
 497                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 498                                    nthreads);
 499                 do {
 500                         if (!unlikely(t->exit_state)) {
 501                                 ticks = cputime_add(prof_ticks(t), left);
 502                                 if (cputime_eq(t->it_prof_expires,
 503                                                cputime_zero) ||
 504                                     cputime_gt(t->it_prof_expires, ticks)) {
 505                                         t->it_prof_expires = ticks;
 506                                 }
 507                         }
 508                         t = next_thread(t);
 509                 } while (t != p);
 510                 break;
 511         case CPUCLOCK_VIRT:
 512                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 513                                    nthreads);
 514                 do {
 515                         if (!unlikely(t->exit_state)) {
 516                                 ticks = cputime_add(virt_ticks(t), left);
 517                                 if (cputime_eq(t->it_virt_expires,
 518                                                cputime_zero) ||
 519                                     cputime_gt(t->it_virt_expires, ticks)) {
 520                                         t->it_virt_expires = ticks;
 521                                 }
 522                         }
 523                         t = next_thread(t);
 524                 } while (t != p);
 525                 break;
 526         case CPUCLOCK_SCHED:
 527                 nsleft = expires.sched - val.sched;
 528                 do_div(nsleft, nthreads);
 529                 do {
 530                         if (!unlikely(t->exit_state)) {
 531                                 ns = t->sched_time + nsleft;
 532                                 if (t->it_sched_expires == 0 ||
 533                                     t->it_sched_expires > ns) {
 534                                         t->it_sched_expires = ns;
 535                                 }
 536                         }
 537                         t = next_thread(t);
 538                 } while (t != p);
 539                 break;
 540         }
 541 }
 542
 543 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 544 {
 545         /*
 546          * That's all for this thread or process.
 547          * We leave our residual in expires to be reported.
 548          */
 549         put_task_struct(timer->it.cpu.task);
 550         timer->it.cpu.task = NULL;
 551         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 552                                              timer->it.cpu.expires,
 553                                              now);
 554 }
 555
 556 /*
 557  * Insert the timer on the appropriate list before any timers that
 558  * expire later.  This must be called with the tasklist_lock held
 559  * for reading, and interrupts disabled.
 560  */
 561 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 562 {
 563         struct task_struct *p = timer->it.cpu.task;
 564         struct list_head *head, *listpos;
 565         struct cpu_timer_list *const nt = &timer->it.cpu;
 566         struct cpu_timer_list *next;
 567         unsigned long i;
 568
 569         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 570                 p->cpu_timers : p->signal->cpu_timers);
 571         head += CPUCLOCK_WHICH(timer->it_clock);
 572
 573         BUG_ON(!irqs_disabled());
 574         spin_lock(&p->sighand->siglock);
 575
 576         listpos = head;
 577         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 578                 list_for_each_entry(next, head, entry) {
 579                         if (next->expires.sched > nt->expires.sched) {
 580                                 listpos = &next->entry;
 581                                 break;
 582                         }
 583                 }
 584         } else {
 585                 list_for_each_entry(next, head, entry) {
 586                         if (cputime_gt(next->expires.cpu, nt->expires.cpu)) {
 587                                 listpos = &next->entry;
 588                                 break;
 589                         }
 590                 }
 591         }
 592         list_add(&nt->entry, listpos);
 593
 594         if (listpos == head) {
 595                 /*
 596                  * We are the new earliest-expiring timer.
 597                  * If we are a thread timer, there can always
 598                  * be a process timer telling us to stop earlier.
 599                  */
 600
 601                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 602                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 603                         default:
 604                                 BUG();
 605                         case CPUCLOCK_PROF:
 606                                 if (cputime_eq(p->it_prof_expires,
 607                                                cputime_zero) ||
 608                                     cputime_gt(p->it_prof_expires,
 609                                                nt->expires.cpu))
 610                                         p->it_prof_expires = nt->expires.cpu;
 611                                 break;
 612                         case CPUCLOCK_VIRT:
 613                                 if (cputime_eq(p->it_virt_expires,
 614                                                cputime_zero) ||
 615                                     cputime_gt(p->it_virt_expires,
 616                                                nt->expires.cpu))
 617                                         p->it_virt_expires = nt->expires.cpu;
 618                                 break;
 619                         case CPUCLOCK_SCHED:
 620                                 if (p->it_sched_expires == 0 ||
 621                                     p->it_sched_expires > nt->expires.sched)
 622                                         p->it_sched_expires = nt->expires.sched;
 623                                 break;
 624                         }
 625                 } else {
 626                         /*
 627                          * For a process timer, we must balance
 628                          * all the live threads' expirations.
 629                          */
 630                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 631                         default:
 632                                 BUG();
 633                         case CPUCLOCK_VIRT:
 634                                 if (!cputime_eq(p->signal->it_virt_expires,
 635                                                 cputime_zero) &&
 636                                     cputime_lt(p->signal->it_virt_expires,
 637                                                timer->it.cpu.expires.cpu))
 638                                         break;
 639                                 goto rebalance;
 640                         case CPUCLOCK_PROF:
 641                                 if (!cputime_eq(p->signal->it_prof_expires,
 642                                                 cputime_zero) &&
 643                                     cputime_lt(p->signal->it_prof_expires,
 644                                                timer->it.cpu.expires.cpu))
 645                                         break;
 646                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 647                                 if (i != RLIM_INFINITY &&
 648                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 649                                         break;
 650                                 goto rebalance;
 651                         case CPUCLOCK_SCHED:
 652                         rebalance:
 653                                 process_timer_rebalance(
 654                                         timer->it.cpu.task,
 655                                         CPUCLOCK_WHICH(timer->it_clock),
 656                                         timer->it.cpu.expires, now);
 657                                 break;
 658                         }
 659                 }
 660         }
 661
 662         spin_unlock(&p->sighand->siglock);
 663 }
 664
 665 /*
 666  * The timer is locked, fire it and arrange for its reload.
 667  */
 668 static void cpu_timer_fire(struct k_itimer *timer)
 669 {
 670         if (unlikely(timer->sigq == NULL)) {
 671                 /*
 672                  * This a special case for clock_nanosleep,
 673                  * not a normal timer from sys_timer_create.
 674                  */
 675                 wake_up_process(timer->it_process);
 676                 timer->it.cpu.expires.sched = 0;
 677         } else if (timer->it.cpu.incr.sched == 0) {
 678                 /*
 679                  * One-shot timer.  Clear it as soon as it's fired.
 680                  */
 681                 posix_timer_event(timer, 0);
 682                 timer->it.cpu.expires.sched = 0;
 683         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 684                 /*
 685                  * The signal did not get queued because the signal
 686                  * was ignored, so we won't get any callback to
 687                  * reload the timer.  But we need to keep it
 688                  * ticking in case the signal is deliverable next time.
 689                  */
 690                 posix_cpu_timer_schedule(timer);
 691         }
 692 }
 693
 694 /*
 695  * Guts of sys_timer_settime for CPU timers.
 696  * This is called with the timer locked and interrupts disabled.
 697  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 698  * and try again.  (This happens when the timer is in the middle of firing.)
 699  */
 700 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 701                         struct itimerspec *new, struct itimerspec *old)
 702 {
 703         struct task_struct *p = timer->it.cpu.task;
 704         union cpu_time_count old_expires, new_expires, val;
 705         int ret;
 706
 707         if (unlikely(p == NULL)) {
 708                 /*
 709                  * Timer refers to a dead task's clock.
 710                  */
 711                 return -ESRCH;
 712         }
 713
 714         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 715
 716         read_lock(&tasklist_lock);
 717         /*
 718          * We need the tasklist_lock to protect against reaping that
 719          * clears p->signal.  If p has just been reaped, we can no
 720          * longer get any information about it at all.
 721          */
 722         if (unlikely(p->signal == NULL)) {
 723                 read_unlock(&tasklist_lock);
 724                 put_task_struct(p);
 725                 timer->it.cpu.task = NULL;
 726                 return -ESRCH;
 727         }
 728
 729         /*
 730          * Disarm any old timer after extracting its expiry time.
 731          */
 732         BUG_ON(!irqs_disabled());
 733         spin_lock(&p->sighand->siglock);
 734         old_expires = timer->it.cpu.expires;
 735         list_del_init(&timer->it.cpu.entry);
 736         spin_unlock(&p->sighand->siglock);
 737
 738         /*
 739          * We need to sample the current value to convert the new
 740          * value from to relative and absolute, and to convert the
 741          * old value from absolute to relative.  To set a process
 742          * timer, we need a sample to balance the thread expiry
 743          * times (in arm_timer).  With an absolute time, we must
 744          * check if it's already passed.  In short, we need a sample.
 745          */
 746         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 747                 cpu_clock_sample(timer->it_clock, p, &val);
 748         } else {
 749                 cpu_clock_sample_group(timer->it_clock, p, &val);
 750         }
 751
 752         if (old) {
 753                 if (old_expires.sched == 0) {
 754                         old->it_value.tv_sec = 0;
 755                         old->it_value.tv_nsec = 0;
 756                 } else {
 757                         /*
 758                          * Update the timer in case it has
 759                          * overrun already.  If it has,
 760                          * we'll report it as having overrun
 761                          * and with the next reloaded timer
 762                          * already ticking, though we are
 763                          * swallowing that pending
 764                          * notification here to install the
 765                          * new setting.
 766                          */
 767                         bump_cpu_timer(timer, val);
 768                         if (cpu_time_before(timer->it_clock, val,
 769                                             timer->it.cpu.expires)) {
 770                                 old_expires = cpu_time_sub(
 771                                         timer->it_clock,
 772                                         timer->it.cpu.expires, val);
 773                                 sample_to_timespec(timer->it_clock,
 774                                                    old_expires,
 775                                                    &old->it_value);
 776                         } else {
 777                                 old->it_value.tv_nsec = 1;
 778                                 old->it_value.tv_sec = 0;
 779                         }
 780                 }
 781         }
 782
 783         if (unlikely(timer->it.cpu.firing)) {
 784                 /*
 785                  * We are colliding with the timer actually firing.
 786                  * Punt after filling in the timer's old value, and
 787                  * disable this firing since we are already reporting
 788                  * it as an overrun (thanks to bump_cpu_timer above).
 789                  */
 790                 read_unlock(&tasklist_lock);
 791                 timer->it.cpu.firing = -1;
 792                 ret = TIMER_RETRY;
 793                 goto out;
 794         }
 795
 796         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 797                 cpu_time_add(timer->it_clock, &new_expires, val);
 798         }
 799
 800         /*
 801          * Install the new expiry time (or zero).
 802          * For a timer with no notification action, we don't actually
 803          * arm the timer (we'll just fake it for timer_gettime).
 804          */
 805         timer->it.cpu.expires = new_expires;
 806         if (new_expires.sched != 0 &&
 807             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 808             cpu_time_before(timer->it_clock, val, new_expires)) {
 809                 arm_timer(timer, val);
 810         }
 811
 812         read_unlock(&tasklist_lock);
 813
 814         /*
 815          * Install the new reload setting, and
 816          * set up the signal and overrun bookkeeping.
 817          */
 818         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 819                                                 &new->it_interval);
 820
 821         /*
 822          * This acts as a modification timestamp for the timer,
 823          * so any automatic reload attempt will punt on seeing
 824          * that we have reset the timer manually.
 825          */
 826         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 827                 ~REQUEUE_PENDING;
 828         timer->it_overrun_last = 0;
 829         timer->it_overrun = -1;
 830
 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                 /*
 835                  * The designated time already passed, so we notify
 836                  * immediately, even if the thread never runs to
 837                  * accumulate more time on this clock.
 838                  */
 839                 cpu_timer_fire(timer);
 840         }
 841
 842         ret = 0;
 843  out:
 844         if (old) {
 845                 sample_to_timespec(timer->it_clock,
 846                                    timer->it.cpu.incr, &old->it_interval);
 847         }
 848         return ret;
 849 }
 850
 851 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 852 {
 853         union cpu_time_count now;
 854         struct task_struct *p = timer->it.cpu.task;
 855         int clear_dead;
 856
 857         /*
 858          * Easy part: convert the reload time.
 859          */
 860         sample_to_timespec(timer->it_clock,
 861                            timer->it.cpu.incr, &itp->it_interval);
 862
 863         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 864                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 865                 return;
 866         }
 867
 868         if (unlikely(p == NULL)) {
 869                 /*
 870                  * This task already died and the timer will never fire.
 871                  * In this case, expires is actually the dead value.
 872                  */
 873         dead:
 874                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 875                                    &itp->it_value);
 876                 return;
 877         }
 878
 879         /*
 880          * Sample the clock to take the difference with the expiry time.
 881          */
 882         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 883                 cpu_clock_sample(timer->it_clock, p, &now);
 884                 clear_dead = p->exit_state;
 885         } else {
 886                 read_lock(&tasklist_lock);
 887                 if (unlikely(p->signal == NULL)) {
 888                         /*
 889                          * The process has been reaped.
 890                          * We can't even collect a sample any more.
 891                          * Call the timer disarmed, nothing else to do.
 892                          */
 893                         put_task_struct(p);
 894                         timer->it.cpu.task = NULL;
 895                         timer->it.cpu.expires.sched = 0;
 896                         read_unlock(&tasklist_lock);
 897                         goto dead;
 898                 } else {
 899                         cpu_clock_sample_group(timer->it_clock, p, &now);
 900                         clear_dead = (unlikely(p->exit_state) &&
 901                                       thread_group_empty(p));
 902                 }
 903                 read_unlock(&tasklist_lock);
 904         }
 905
 906         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 907                 if (timer->it.cpu.incr.sched == 0 &&
 908                     cpu_time_before(timer->it_clock,
 909                                     timer->it.cpu.expires, now)) {
 910                         /*
 911                          * Do-nothing timer expired and has no reload,
 912                          * so it's as if it was never set.
 913                          */
 914                         timer->it.cpu.expires.sched = 0;
 915                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 916                         return;
 917                 }
 918                 /*
 919                  * Account for any expirations and reloads that should
 920                  * have happened.
 921                  */
 922                 bump_cpu_timer(timer, now);
 923         }
 924
 925         if (unlikely(clear_dead)) {
 926                 /*
 927                  * We've noticed that the thread is dead, but
 928                  * not yet reaped.  Take this opportunity to
 929                  * drop our task ref.
 930                  */
 931                 clear_dead_task(timer, now);
 932                 goto dead;
 933         }
 934
 935         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 936                 sample_to_timespec(timer->it_clock,
 937                                    cpu_time_sub(timer->it_clock,
 938                                                 timer->it.cpu.expires, now),
 939                                    &itp->it_value);
 940         } else {
 941                 /*
 942                  * The timer should have expired already, but the firing
 943                  * hasn't taken place yet.  Say it's just about to expire.
 944                  */
 945                 itp->it_value.tv_nsec = 1;
 946                 itp->it_value.tv_sec = 0;
 947         }
 948 }
 949
 950 /*
 951  * Check for any per-thread CPU timers that have fired and move them off
 952  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 953  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 954  */
 955 static void check_thread_timers(struct task_struct *tsk,
 956                                 struct list_head *firing)
 957 {
 958         int maxfire;
 959         struct list_head *timers = tsk->cpu_timers;
 960
 961         maxfire = 20;
 962         tsk->it_prof_expires = cputime_zero;
 963         while (!list_empty(timers)) {
 964                 struct cpu_timer_list *t = list_entry(timers->next,
 965                                                       struct cpu_timer_list,
 966                                                       entry);
 967                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 968                         tsk->it_prof_expires = t->expires.cpu;
 969                         break;
 970                 }
 971                 t->firing = 1;
 972                 list_move_tail(&t->entry, firing);
 973         }
 974
 975         ++timers;
 976         maxfire = 20;
 977         tsk->it_virt_expires = cputime_zero;
 978         while (!list_empty(timers)) {
 979                 struct cpu_timer_list *t = list_entry(timers->next,
 980                                                       struct cpu_timer_list,
 981                                                       entry);
 982                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 983                         tsk->it_virt_expires = t->expires.cpu;
 984                         break;
 985                 }
 986                 t->firing = 1;
 987                 list_move_tail(&t->entry, firing);
 988         }
 989
 990         ++timers;
 991         maxfire = 20;
 992         tsk->it_sched_expires = 0;
 993         while (!list_empty(timers)) {
 994                 struct cpu_timer_list *t = list_entry(timers->next,
 995                                                       struct cpu_timer_list,
 996                                                       entry);
 997                 if (!--maxfire || tsk->sched_time < t->expires.sched) {
 998                         tsk->it_sched_expires = t->expires.sched;
 999                         break;
1000                 }
1001                 t->firing = 1;
1002                 list_move_tail(&t->entry, firing);
1003         }
1004 }
1005
1006 /*
1007  * Check for any per-thread CPU timers that have fired and move them
1008  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1009  * have already been taken off.
1010  */
1011 static void check_process_timers(struct task_struct *tsk,
1012                                  struct list_head *firing)
1013 {
1014         int maxfire;
1015         struct signal_struct *const sig = tsk->signal;
1016         cputime_t utime, stime, ptime, virt_expires, prof_expires;
1017         unsigned long long sched_time, sched_expires;
1018         struct task_struct *t;
1019         struct list_head *timers = sig->cpu_timers;
1020
1021         /*
1022          * Don't sample the current process CPU clocks if there are no timers.
1023          */
1024         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1025             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1026             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1027             list_empty(&timers[CPUCLOCK_VIRT]) &&
1028             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1029             list_empty(&timers[CPUCLOCK_SCHED]))
1030                 return;
1031
1032         /*
1033          * Collect the current process totals.
1034          */
1035         utime = sig->utime;
1036         stime = sig->stime;
1037         sched_time = sig->sched_time;
1038         t = tsk;
1039         do {
1040                 utime = cputime_add(utime, t->utime);
1041                 stime = cputime_add(stime, t->stime);
1042                 sched_time += t->sched_time;
1043                 t = next_thread(t);
1044         } while (t != tsk);
1045         ptime = cputime_add(utime, stime);
1046
1047         maxfire = 20;
1048         prof_expires = cputime_zero;
1049         while (!list_empty(timers)) {
1050                 struct cpu_timer_list *t = list_entry(timers->next,
1051                                                       struct cpu_timer_list,
1052                                                       entry);
1053                 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
1054                         prof_expires = t->expires.cpu;
1055                         break;
1056                 }
1057                 t->firing = 1;
1058                 list_move_tail(&t->entry, firing);
1059         }
1060
1061         ++timers;
1062         maxfire = 20;
1063         virt_expires = cputime_zero;
1064         while (!list_empty(timers)) {
1065                 struct cpu_timer_list *t = list_entry(timers->next,
1066                                                       struct cpu_timer_list,
1067                                                       entry);
1068                 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
1069                         virt_expires = t->expires.cpu;
1070                         break;
1071                 }
1072                 t->firing = 1;
1073                 list_move_tail(&t->entry, firing);
1074         }
1075
1076         ++timers;
1077         maxfire = 20;
1078         sched_expires = 0;
1079         while (!list_empty(timers)) {
1080                 struct cpu_timer_list *t = list_entry(timers->next,
1081                                                       struct cpu_timer_list,
1082                                                       entry);
1083                 if (!--maxfire || sched_time < t->expires.sched) {
1084                         sched_expires = t->expires.sched;
1085                         break;
1086                 }
1087                 t->firing = 1;
1088                 list_move_tail(&t->entry, firing);
1089         }
1090
1091         /*
1092          * Check for the special case process timers.
1093          */
1094         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1095                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1096                         /* ITIMER_PROF fires and reloads.  */
1097                         sig->it_prof_expires = sig->it_prof_incr;
1098                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1099                                 sig->it_prof_expires = cputime_add(
1100                                         sig->it_prof_expires, ptime);
1101                         }
1102                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1103                 }
1104                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1105                     (cputime_eq(prof_expires, cputime_zero) ||
1106                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1107                         prof_expires = sig->it_prof_expires;
1108                 }
1109         }
1110         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1111                 if (cputime_ge(utime, sig->it_virt_expires)) {
1112                         /* ITIMER_VIRTUAL fires and reloads.  */
1113                         sig->it_virt_expires = sig->it_virt_incr;
1114                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1115                                 sig->it_virt_expires = cputime_add(
1116                                         sig->it_virt_expires, utime);
1117                         }
1118                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1119                 }
1120                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1121                     (cputime_eq(virt_expires, cputime_zero) ||
1122                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1123                         virt_expires = sig->it_virt_expires;
1124                 }
1125         }
1126         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1127                 unsigned long psecs = cputime_to_secs(ptime);
1128                 cputime_t x;
1129                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1130                         /*
1131                          * At the hard limit, we just die.
1132                          * No need to calculate anything else now.
1133                          */
1134                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1135                         return;
1136                 }
1137                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1138                         /*
1139                          * At the soft limit, send a SIGXCPU every second.
1140                          */
1141                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1142                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1143                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1144                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1145                         }
1146                 }
1147                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1148                 if (cputime_eq(prof_expires, cputime_zero) ||
1149                     cputime_lt(x, prof_expires)) {
1150                         prof_expires = x;
1151                 }
1152         }
1153
1154         if (!cputime_eq(prof_expires, cputime_zero) ||
1155             !cputime_eq(virt_expires, cputime_zero) ||
1156             sched_expires != 0) {
1157                 /*
1158                  * Rebalance the threads' expiry times for the remaining
1159                  * process CPU timers.
1160                  */
1161
1162                 cputime_t prof_left, virt_left, ticks;
1163                 unsigned long long sched_left, sched;
1164                 const unsigned int nthreads = atomic_read(&sig->live);
1165
1166                 if (!nthreads)
1167                         return;
1168
1169                 prof_left = cputime_sub(prof_expires, utime);
1170                 prof_left = cputime_sub(prof_left, stime);
1171                 prof_left = cputime_div(prof_left, nthreads);
1172                 virt_left = cputime_sub(virt_expires, utime);
1173                 virt_left = cputime_div(virt_left, nthreads);
1174                 if (sched_expires) {
1175                         sched_left = sched_expires - sched_time;
1176                         do_div(sched_left, nthreads);
1177                 } else {
1178                         sched_left = 0;
1179                 }
1180                 t = tsk;
1181                 do {
1182                         ticks = cputime_add(cputime_add(t->utime, t->stime),
1183                                             prof_left);
1184                         if (!cputime_eq(prof_expires, cputime_zero) &&
1185                             (cputime_eq(t->it_prof_expires, cputime_zero) ||
1186                              cputime_gt(t->it_prof_expires, ticks))) {
1187                                 t->it_prof_expires = ticks;
1188                         }
1189
1190                         ticks = cputime_add(t->utime, virt_left);
1191                         if (!cputime_eq(virt_expires, cputime_zero) &&
1192                             (cputime_eq(t->it_virt_expires, cputime_zero) ||
1193                              cputime_gt(t->it_virt_expires, ticks))) {
1194                                 t->it_virt_expires = ticks;
1195                         }
1196
1197                         sched = t->sched_time + sched_left;
1198                         if (sched_expires && (t->it_sched_expires == 0 ||
1199                                               t->it_sched_expires > sched)) {
1200                                 t->it_sched_expires = sched;
1201                         }
1202
1203                         do {
1204                                 t = next_thread(t);
1205                         } while (unlikely(t->exit_state));
1206                 } while (t != tsk);
1207         }
1208 }
1209
1210 /*
1211  * This is called from the signal code (via do_schedule_next_timer)
1212  * when the last timer signal was delivered and we have to reload the timer.
1213  */
1214 void posix_cpu_timer_schedule(struct k_itimer *timer)
1215 {
1216         struct task_struct *p = timer->it.cpu.task;
1217         union cpu_time_count now;
1218
1219         if (unlikely(p == NULL))
1220                 /*
1221                  * The task was cleaned up already, no future firings.
1222                  */
1223                 return;
1224
1225         /*
1226          * Fetch the current sample and update the timer's expiry time.
1227          */
1228         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1229                 cpu_clock_sample(timer->it_clock, p, &now);
1230                 bump_cpu_timer(timer, now);
1231                 if (unlikely(p->exit_state)) {
1232                         clear_dead_task(timer, now);
1233                         return;
1234                 }
1235                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1236         } else {
1237                 read_lock(&tasklist_lock);
1238                 if (unlikely(p->signal == NULL)) {
1239                         /*
1240                          * The process has been reaped.
1241                          * We can't even collect a sample any more.
1242                          */
1243                         put_task_struct(p);
1244                         timer->it.cpu.task = p = NULL;
1245                         timer->it.cpu.expires.sched = 0;
1246                         read_unlock(&tasklist_lock);
1247                         return;
1248                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1249                         /*
1250                          * We've noticed that the thread is dead, but
1251                          * not yet reaped.  Take this opportunity to
1252                          * drop our task ref.
1253                          */
1254                         clear_dead_task(timer, now);
1255                         read_unlock(&tasklist_lock);
1256                         return;
1257                 }
1258                 cpu_clock_sample_group(timer->it_clock, p, &now);
1259                 bump_cpu_timer(timer, now);
1260                 /* Leave the tasklist_lock locked for the call below.  */
1261         }
1262
1263         /*
1264          * Now re-arm for the new expiry time.
1265          */
1266         arm_timer(timer, now);
1267
1268         read_unlock(&tasklist_lock);
1269 }
1270
1271 /*
1272  * This is called from the timer interrupt handler.  The irq handler has
1273  * already updated our counts.  We need to check if any timers fire now.
1274  * Interrupts are disabled.
1275  */
1276 void run_posix_cpu_timers(struct task_struct *tsk)
1277 {
1278         LIST_HEAD(firing);
1279         struct k_itimer *timer, *next;
1280
1281         BUG_ON(!irqs_disabled());
1282
1283 #define UNEXPIRED(clock) \
1284                 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1285                  cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1286
1287         if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1288             (tsk->it_sched_expires == 0 ||
1289              tsk->sched_time < tsk->it_sched_expires))
1290                 return;
1291
1292 #undef  UNEXPIRED
1293
1294         /*
1295          * Double-check with locks held.
1296          */
1297         read_lock(&tasklist_lock);
1298         if (likely(tsk->signal != NULL)) {
1299                 spin_lock(&tsk->sighand->siglock);
1300
1301                 /*
1302                  * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1303                  * all the timers that are firing, and put them on the firing list.
1304                  */
1305                 check_thread_timers(tsk, &firing);
1306                 check_process_timers(tsk, &firing);
1307
1308                 /*
1309                  * We must release these locks before taking any timer's lock.
1310                  * There is a potential race with timer deletion here, as the
1311                  * siglock now protects our private firing list.  We have set
1312                  * the firing flag in each timer, so that a deletion attempt
1313                  * that gets the timer lock before we do will give it up and
1314                  * spin until we've taken care of that timer below.
1315                  */
1316                 spin_unlock(&tsk->sighand->siglock);
1317         }
1318         read_unlock(&tasklist_lock);
1319
1320         /*
1321          * Now that all the timers on our list have the firing flag,
1322          * noone will touch their list entries but us.  We'll take
1323          * each timer's lock before clearing its firing flag, so no
1324          * timer call will interfere.
1325          */
1326         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1327                 int firing;
1328                 spin_lock(&timer->it_lock);
1329                 list_del_init(&timer->it.cpu.entry);
1330                 firing = timer->it.cpu.firing;
1331                 timer->it.cpu.firing = 0;
1332                 /*
1333                  * The firing flag is -1 if we collided with a reset
1334                  * of the timer, which already reported this
1335                  * almost-firing as an overrun.  So don't generate an event.
1336                  */
1337                 if (likely(firing >= 0)) {
1338                         cpu_timer_fire(timer);
1339                 }
1340                 spin_unlock(&timer->it_lock);
1341         }
1342 }
1343
1344 /*
1345  * Set one of the process-wide special case CPU timers.
1346  * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1347  * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1348  * absolute; non-null for ITIMER_*, where *newval is relative and we update
1349  * it to be absolute, *oldval is absolute and we update it to be relative.
1350  */
1351 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1352                            cputime_t *newval, cputime_t *oldval)
1353 {
1354         union cpu_time_count now;
1355         struct list_head *head;
1356
1357         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1358         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1359
1360         if (oldval) {
1361                 if (!cputime_eq(*oldval, cputime_zero)) {
1362                         if (cputime_le(*oldval, now.cpu)) {
1363                                 /* Just about to fire. */
1364                                 *oldval = jiffies_to_cputime(1);
1365                         } else {
1366                                 *oldval = cputime_sub(*oldval, now.cpu);
1367                         }
1368                 }
1369
1370                 if (cputime_eq(*newval, cputime_zero))
1371                         return;
1372                 *newval = cputime_add(*newval, now.cpu);
1373
1374                 /*
1375                  * If the RLIMIT_CPU timer will expire before the
1376                  * ITIMER_PROF timer, we have nothing else to do.
1377                  */
1378                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1379                     < cputime_to_secs(*newval))
1380                         return;
1381         }
1382
1383         /*
1384          * Check whether there are any process timers already set to fire
1385          * before this one.  If so, we don't have anything more to do.
1386          */
1387         head = &tsk->signal->cpu_timers[clock_idx];
1388         if (list_empty(head) ||
1389             cputime_ge(list_entry(head->next,
1390                                   struct cpu_timer_list, entry)->expires.cpu,
1391                        *newval)) {
1392                 /*
1393                  * Rejigger each thread's expiry time so that one will
1394                  * notice before we hit the process-cumulative expiry time.
1395                  */
1396                 union cpu_time_count expires = { .sched = 0 };
1397                 expires.cpu = *newval;
1398                 process_timer_rebalance(tsk, clock_idx, expires, now);
1399         }
1400 }
1401
1402 static long posix_cpu_clock_nanosleep_restart(struct restart_block *);
1403
1404 int posix_cpu_nsleep(clockid_t which_clock, int flags,
1405                      struct timespec *rqtp)
1406 {
1407         struct restart_block *restart_block =
1408             &current_thread_info()->restart_block;
1409         struct k_itimer timer;
1410         int error;
1411
1412         /*
1413          * Diagnose required errors first.
1414          */
1415         if (CPUCLOCK_PERTHREAD(which_clock) &&
1416             (CPUCLOCK_PID(which_clock) == 0 ||
1417              CPUCLOCK_PID(which_clock) == current->pid))
1418                 return -EINVAL;
1419
1420         /*
1421          * Set up a temporary timer and then wait for it to go off.
1422          */
1423         memset(&timer, 0, sizeof timer);
1424         spin_lock_init(&timer.it_lock);
1425         timer.it_clock = which_clock;
1426         timer.it_overrun = -1;
1427         error = posix_cpu_timer_create(&timer);
1428         timer.it_process = current;
1429         if (!error) {
1430                 struct timespec __user *rmtp;
1431                 static struct itimerspec zero_it;
1432                 struct itimerspec it = { .it_value = *rqtp,
1433                                          .it_interval = {} };
1434
1435                 spin_lock_irq(&timer.it_lock);
1436                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1437                 if (error) {
1438                         spin_unlock_irq(&timer.it_lock);
1439                         return error;
1440                 }
1441
1442                 while (!signal_pending(current)) {
1443                         if (timer.it.cpu.expires.sched == 0) {
1444                                 /*
1445                                  * Our timer fired and was reset.
1446                                  */
1447                                 spin_unlock_irq(&timer.it_lock);
1448                                 return 0;
1449                         }
1450
1451                         /*
1452                          * Block until cpu_timer_fire (or a signal) wakes us.
1453                          */
1454                         __set_current_state(TASK_INTERRUPTIBLE);
1455                         spin_unlock_irq(&timer.it_lock);
1456                         schedule();
1457                         spin_lock_irq(&timer.it_lock);
1458                 }
1459
1460                 /*
1461                  * We were interrupted by a signal.
1462                  */
1463                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1464                 posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1465                 spin_unlock_irq(&timer.it_lock);
1466
1467                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1468                         /*
1469                          * It actually did fire already.
1470                          */
1471                         return 0;
1472                 }
1473
1474                 /*
1475                  * Report back to the user the time still remaining.
1476                  */
1477                 rmtp = (struct timespec __user *) restart_block->arg1;
1478                 if (rmtp != NULL && !(flags & TIMER_ABSTIME) &&
1479                     copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1480                         return -EFAULT;
1481
1482                 restart_block->fn = posix_cpu_clock_nanosleep_restart;
1483                 /* Caller already set restart_block->arg1 */
1484                 restart_block->arg0 = which_clock;
1485                 restart_block->arg2 = rqtp->tv_sec;
1486                 restart_block->arg3 = rqtp->tv_nsec;
1487
1488                 error = -ERESTART_RESTARTBLOCK;
1489         }
1490
1491         return error;
1492 }
1493
1494 static long
1495 posix_cpu_clock_nanosleep_restart(struct restart_block *restart_block)
1496 {
1497         clockid_t which_clock = restart_block->arg0;
1498         struct timespec t = { .tv_sec = restart_block->arg2,
1499                               .tv_nsec = restart_block->arg3 };
1500         restart_block->fn = do_no_restart_syscall;
1501         return posix_cpu_nsleep(which_clock, TIMER_ABSTIME, &t);
1502 }
1503
1504
1505 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1506 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1507
1508 static int process_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1509 {
1510         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1511 }
1512 static int process_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1513 {
1514         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1515 }
1516 static int process_cpu_timer_create(struct k_itimer *timer)
1517 {
1518         timer->it_clock = PROCESS_CLOCK;
1519         return posix_cpu_timer_create(timer);
1520 }
1521 static int process_cpu_nsleep(clockid_t which_clock, int flags,
1522                               struct timespec *rqtp)
1523 {
1524         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1525 }
1526 static int thread_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1527 {
1528         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1529 }
1530 static int thread_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1531 {
1532         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1533 }
1534 static int thread_cpu_timer_create(struct k_itimer *timer)
1535 {
1536         timer->it_clock = THREAD_CLOCK;
1537         return posix_cpu_timer_create(timer);
1538 }
1539 static int thread_cpu_nsleep(clockid_t which_clock, int flags,
1540                               struct timespec *rqtp)
1541 {
1542         return -EINVAL;
1543 }
1544
1545 static __init int init_posix_cpu_timers(void)
1546 {
1547         struct k_clock process = {
1548                 .clock_getres = process_cpu_clock_getres,
1549                 .clock_get = process_cpu_clock_get,
1550                 .clock_set = do_posix_clock_nosettime,
1551                 .timer_create = process_cpu_timer_create,
1552                 .nsleep = process_cpu_nsleep,
1553         };
1554         struct k_clock thread = {
1555                 .clock_getres = thread_cpu_clock_getres,
1556                 .clock_get = thread_cpu_clock_get,
1557                 .clock_set = do_posix_clock_nosettime,
1558                 .timer_create = thread_cpu_timer_create,
1559                 .nsleep = thread_cpu_nsleep,
1560         };
1561
1562         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1563         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1564
1565         return 0;
1566 }
1567 __initcall(init_posix_cpu_timers);