git.oblomov.eu Git - linux-2.6/blob - kernel/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/posix-timers.h>
   7 #include <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 = (unsigned long long)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 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_lt(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                 cpu->sched += sched_ns(p);
 242                 break;
 243         }
 244         return 0;
 245 }
 246
 247 /*
 248  * Sample a process (thread group) clock for the given group_leader task.
 249  * Must be called with tasklist_lock held for reading.
 250  */
 251 static int cpu_clock_sample_group(clockid_t which_clock,
 252                                   struct task_struct *p,
 253                                   union cpu_time_count *cpu)
 254 {
 255         int ret;
 256         unsigned long flags;
 257         spin_lock_irqsave(&p->sighand->siglock, flags);
 258         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
 259                                             cpu);
 260         spin_unlock_irqrestore(&p->sighand->siglock, flags);
 261         return ret;
 262 }
 263
 264
 265 int posix_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
 266 {
 267         const pid_t pid = CPUCLOCK_PID(which_clock);
 268         int error = -EINVAL;
 269         union cpu_time_count rtn;
 270
 271         if (pid == 0) {
 272                 /*
 273                  * Special case constant value for our own clocks.
 274                  * We don't have to do any lookup to find ourselves.
 275                  */
 276                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 277                         /*
 278                          * Sampling just ourselves we can do with no locking.
 279                          */
 280                         error = cpu_clock_sample(which_clock,
 281                                                  current, &rtn);
 282                 } else {
 283                         read_lock(&tasklist_lock);
 284                         error = cpu_clock_sample_group(which_clock,
 285                                                        current, &rtn);
 286                         read_unlock(&tasklist_lock);
 287                 }
 288         } else {
 289                 /*
 290                  * Find the given PID, and validate that the caller
 291                  * should be able to see it.
 292                  */
 293                 struct task_struct *p;
 294                 read_lock(&tasklist_lock);
 295                 p = find_task_by_pid(pid);
 296                 if (p) {
 297                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 298                                 if (p->tgid == current->tgid) {
 299                                         error = cpu_clock_sample(which_clock,
 300                                                                  p, &rtn);
 301                                 }
 302                         } else if (p->tgid == pid && p->signal) {
 303                                 error = cpu_clock_sample_group(which_clock,
 304                                                                p, &rtn);
 305                         }
 306                 }
 307                 read_unlock(&tasklist_lock);
 308         }
 309
 310         if (error)
 311                 return error;
 312         sample_to_timespec(which_clock, rtn, tp);
 313         return 0;
 314 }
 315
 316
 317 /*
 318  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 319  * This is called from sys_timer_create with the new timer already locked.
 320  */
 321 int posix_cpu_timer_create(struct k_itimer *new_timer)
 322 {
 323         int ret = 0;
 324         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 325         struct task_struct *p;
 326
 327         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 328                 return -EINVAL;
 329
 330         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 331         new_timer->it.cpu.incr.sched = 0;
 332         new_timer->it.cpu.expires.sched = 0;
 333
 334         read_lock(&tasklist_lock);
 335         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 336                 if (pid == 0) {
 337                         p = current;
 338                 } else {
 339                         p = find_task_by_pid(pid);
 340                         if (p && p->tgid != current->tgid)
 341                                 p = NULL;
 342                 }
 343         } else {
 344                 if (pid == 0) {
 345                         p = current->group_leader;
 346                 } else {
 347                         p = find_task_by_pid(pid);
 348                         if (p && p->tgid != pid)
 349                                 p = NULL;
 350                 }
 351         }
 352         new_timer->it.cpu.task = p;
 353         if (p) {
 354                 get_task_struct(p);
 355         } else {
 356                 ret = -EINVAL;
 357         }
 358         read_unlock(&tasklist_lock);
 359
 360         return ret;
 361 }
 362
 363 /*
 364  * Clean up a CPU-clock timer that is about to be destroyed.
 365  * This is called from timer deletion with the timer already locked.
 366  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 367  * and try again.  (This happens when the timer is in the middle of firing.)
 368  */
 369 int posix_cpu_timer_del(struct k_itimer *timer)
 370 {
 371         struct task_struct *p = timer->it.cpu.task;
 372         int ret = 0;
 373
 374         if (likely(p != NULL)) {
 375                 read_lock(&tasklist_lock);
 376                 if (unlikely(p->signal == NULL)) {
 377                         /*
 378                          * We raced with the reaping of the task.
 379                          * The deletion should have cleared us off the list.
 380                          */
 381                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 382                 } else {
 383                         spin_lock(&p->sighand->siglock);
 384                         if (timer->it.cpu.firing)
 385                                 ret = TIMER_RETRY;
 386                         else
 387                                 list_del(&timer->it.cpu.entry);
 388                         spin_unlock(&p->sighand->siglock);
 389                 }
 390                 read_unlock(&tasklist_lock);
 391
 392                 if (!ret)
 393                         put_task_struct(p);
 394         }
 395
 396         return ret;
 397 }
 398
 399 /*
 400  * Clean out CPU timers still ticking when a thread exited.  The task
 401  * pointer is cleared, and the expiry time is replaced with the residual
 402  * time for later timer_gettime calls to return.
 403  * This must be called with the siglock held.
 404  */
 405 static void cleanup_timers(struct list_head *head,
 406                            cputime_t utime, cputime_t stime,
 407                            unsigned long long sched_time)
 408 {
 409         struct cpu_timer_list *timer, *next;
 410         cputime_t ptime = cputime_add(utime, stime);
 411
 412         list_for_each_entry_safe(timer, next, head, entry) {
 413                 list_del_init(&timer->entry);
 414                 if (cputime_lt(timer->expires.cpu, ptime)) {
 415                         timer->expires.cpu = cputime_zero;
 416                 } else {
 417                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 418                                                          ptime);
 419                 }
 420         }
 421
 422         ++head;
 423         list_for_each_entry_safe(timer, next, head, entry) {
 424                 list_del_init(&timer->entry);
 425                 if (cputime_lt(timer->expires.cpu, utime)) {
 426                         timer->expires.cpu = cputime_zero;
 427                 } else {
 428                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 429                                                          utime);
 430                 }
 431         }
 432
 433         ++head;
 434         list_for_each_entry_safe(timer, next, head, entry) {
 435                 list_del_init(&timer->entry);
 436                 if (timer->expires.sched < sched_time) {
 437                         timer->expires.sched = 0;
 438                 } else {
 439                         timer->expires.sched -= sched_time;
 440                 }
 441         }
 442 }
 443
 444 /*
 445  * These are both called with the siglock held, when the current thread
 446  * is being reaped.  When the final (leader) thread in the group is reaped,
 447  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 448  */
 449 void posix_cpu_timers_exit(struct task_struct *tsk)
 450 {
 451         cleanup_timers(tsk->cpu_timers,
 452                        tsk->utime, tsk->stime, tsk->sched_time);
 453
 454 }
 455 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 456 {
 457         cleanup_timers(tsk->signal->cpu_timers,
 458                        cputime_add(tsk->utime, tsk->signal->utime),
 459                        cputime_add(tsk->stime, tsk->signal->stime),
 460                        tsk->sched_time + tsk->signal->sched_time);
 461 }
 462
 463
 464 /*
 465  * Set the expiry times of all the threads in the process so one of them
 466  * will go off before the process cumulative expiry total is reached.
 467  */
 468 static void process_timer_rebalance(struct task_struct *p,
 469                                     unsigned int clock_idx,
 470                                     union cpu_time_count expires,
 471                                     union cpu_time_count val)
 472 {
 473         cputime_t ticks, left;
 474         unsigned long long ns, nsleft;
 475         struct task_struct *t = p;
 476         unsigned int nthreads = atomic_read(&p->signal->live);
 477
 478         if (!nthreads)
 479                 return;
 480
 481         switch (clock_idx) {
 482         default:
 483                 BUG();
 484                 break;
 485         case CPUCLOCK_PROF:
 486                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 487                                    nthreads);
 488                 do {
 489                         if (likely(!(t->flags & PF_EXITING))) {
 490                                 ticks = cputime_add(prof_ticks(t), left);
 491                                 if (cputime_eq(t->it_prof_expires,
 492                                                cputime_zero) ||
 493                                     cputime_gt(t->it_prof_expires, ticks)) {
 494                                         t->it_prof_expires = ticks;
 495                                 }
 496                         }
 497                         t = next_thread(t);
 498                 } while (t != p);
 499                 break;
 500         case CPUCLOCK_VIRT:
 501                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 502                                    nthreads);
 503                 do {
 504                         if (likely(!(t->flags & PF_EXITING))) {
 505                                 ticks = cputime_add(virt_ticks(t), left);
 506                                 if (cputime_eq(t->it_virt_expires,
 507                                                cputime_zero) ||
 508                                     cputime_gt(t->it_virt_expires, ticks)) {
 509                                         t->it_virt_expires = ticks;
 510                                 }
 511                         }
 512                         t = next_thread(t);
 513                 } while (t != p);
 514                 break;
 515         case CPUCLOCK_SCHED:
 516                 nsleft = expires.sched - val.sched;
 517                 do_div(nsleft, nthreads);
 518                 do {
 519                         if (likely(!(t->flags & PF_EXITING))) {
 520                                 ns = t->sched_time + nsleft;
 521                                 if (t->it_sched_expires == 0 ||
 522                                     t->it_sched_expires > ns) {
 523                                         t->it_sched_expires = ns;
 524                                 }
 525                         }
 526                         t = next_thread(t);
 527                 } while (t != p);
 528                 break;
 529         }
 530 }
 531
 532 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 533 {
 534         /*
 535          * That's all for this thread or process.
 536          * We leave our residual in expires to be reported.
 537          */
 538         put_task_struct(timer->it.cpu.task);
 539         timer->it.cpu.task = NULL;
 540         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 541                                              timer->it.cpu.expires,
 542                                              now);
 543 }
 544
 545 /*
 546  * Insert the timer on the appropriate list before any timers that
 547  * expire later.  This must be called with the tasklist_lock held
 548  * for reading, and interrupts disabled.
 549  */
 550 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 551 {
 552         struct task_struct *p = timer->it.cpu.task;
 553         struct list_head *head, *listpos;
 554         struct cpu_timer_list *const nt = &timer->it.cpu;
 555         struct cpu_timer_list *next;
 556         unsigned long i;
 557
 558         if (CPUCLOCK_PERTHREAD(timer->it_clock) && (p->flags & PF_EXITING))
 559                 return;
 560
 561         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 562                 p->cpu_timers : p->signal->cpu_timers);
 563         head += CPUCLOCK_WHICH(timer->it_clock);
 564
 565         BUG_ON(!irqs_disabled());
 566         spin_lock(&p->sighand->siglock);
 567
 568         listpos = head;
 569         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 570                 list_for_each_entry(next, head, entry) {
 571                         if (next->expires.sched > nt->expires.sched)
 572                                 break;
 573                         listpos = &next->entry;
 574                 }
 575         } else {
 576                 list_for_each_entry(next, head, entry) {
 577                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 578                                 break;
 579                         listpos = &next->entry;
 580                 }
 581         }
 582         list_add(&nt->entry, listpos);
 583
 584         if (listpos == head) {
 585                 /*
 586                  * We are the new earliest-expiring timer.
 587                  * If we are a thread timer, there can always
 588                  * be a process timer telling us to stop earlier.
 589                  */
 590
 591                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 592                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 593                         default:
 594                                 BUG();
 595                         case CPUCLOCK_PROF:
 596                                 if (cputime_eq(p->it_prof_expires,
 597                                                cputime_zero) ||
 598                                     cputime_gt(p->it_prof_expires,
 599                                                nt->expires.cpu))
 600                                         p->it_prof_expires = nt->expires.cpu;
 601                                 break;
 602                         case CPUCLOCK_VIRT:
 603                                 if (cputime_eq(p->it_virt_expires,
 604                                                cputime_zero) ||
 605                                     cputime_gt(p->it_virt_expires,
 606                                                nt->expires.cpu))
 607                                         p->it_virt_expires = nt->expires.cpu;
 608                                 break;
 609                         case CPUCLOCK_SCHED:
 610                                 if (p->it_sched_expires == 0 ||
 611                                     p->it_sched_expires > nt->expires.sched)
 612                                         p->it_sched_expires = nt->expires.sched;
 613                                 break;
 614                         }
 615                 } else {
 616                         /*
 617                          * For a process timer, we must balance
 618                          * all the live threads' expirations.
 619                          */
 620                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 621                         default:
 622                                 BUG();
 623                         case CPUCLOCK_VIRT:
 624                                 if (!cputime_eq(p->signal->it_virt_expires,
 625                                                 cputime_zero) &&
 626                                     cputime_lt(p->signal->it_virt_expires,
 627                                                timer->it.cpu.expires.cpu))
 628                                         break;
 629                                 goto rebalance;
 630                         case CPUCLOCK_PROF:
 631                                 if (!cputime_eq(p->signal->it_prof_expires,
 632                                                 cputime_zero) &&
 633                                     cputime_lt(p->signal->it_prof_expires,
 634                                                timer->it.cpu.expires.cpu))
 635                                         break;
 636                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 637                                 if (i != RLIM_INFINITY &&
 638                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 639                                         break;
 640                                 goto rebalance;
 641                         case CPUCLOCK_SCHED:
 642                         rebalance:
 643                                 process_timer_rebalance(
 644                                         timer->it.cpu.task,
 645                                         CPUCLOCK_WHICH(timer->it_clock),
 646                                         timer->it.cpu.expires, now);
 647                                 break;
 648                         }
 649                 }
 650         }
 651
 652         spin_unlock(&p->sighand->siglock);
 653 }
 654
 655 /*
 656  * The timer is locked, fire it and arrange for its reload.
 657  */
 658 static void cpu_timer_fire(struct k_itimer *timer)
 659 {
 660         if (unlikely(timer->sigq == NULL)) {
 661                 /*
 662                  * This a special case for clock_nanosleep,
 663                  * not a normal timer from sys_timer_create.
 664                  */
 665                 wake_up_process(timer->it_process);
 666                 timer->it.cpu.expires.sched = 0;
 667         } else if (timer->it.cpu.incr.sched == 0) {
 668                 /*
 669                  * One-shot timer.  Clear it as soon as it's fired.
 670                  */
 671                 posix_timer_event(timer, 0);
 672                 timer->it.cpu.expires.sched = 0;
 673         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 674                 /*
 675                  * The signal did not get queued because the signal
 676                  * was ignored, so we won't get any callback to
 677                  * reload the timer.  But we need to keep it
 678                  * ticking in case the signal is deliverable next time.
 679                  */
 680                 posix_cpu_timer_schedule(timer);
 681         }
 682 }
 683
 684 /*
 685  * Guts of sys_timer_settime for CPU timers.
 686  * This is called with the timer locked and interrupts disabled.
 687  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 688  * and try again.  (This happens when the timer is in the middle of firing.)
 689  */
 690 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 691                         struct itimerspec *new, struct itimerspec *old)
 692 {
 693         struct task_struct *p = timer->it.cpu.task;
 694         union cpu_time_count old_expires, new_expires, val;
 695         int ret;
 696
 697         if (unlikely(p == NULL)) {
 698                 /*
 699                  * Timer refers to a dead task's clock.
 700                  */
 701                 return -ESRCH;
 702         }
 703
 704         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 705
 706         read_lock(&tasklist_lock);
 707         /*
 708          * We need the tasklist_lock to protect against reaping that
 709          * clears p->signal.  If p has just been reaped, we can no
 710          * longer get any information about it at all.
 711          */
 712         if (unlikely(p->signal == NULL)) {
 713                 read_unlock(&tasklist_lock);
 714                 put_task_struct(p);
 715                 timer->it.cpu.task = NULL;
 716                 return -ESRCH;
 717         }
 718
 719         /*
 720          * Disarm any old timer after extracting its expiry time.
 721          */
 722         BUG_ON(!irqs_disabled());
 723
 724         ret = 0;
 725         spin_lock(&p->sighand->siglock);
 726         old_expires = timer->it.cpu.expires;
 727         if (unlikely(timer->it.cpu.firing)) {
 728                 timer->it.cpu.firing = -1;
 729                 ret = TIMER_RETRY;
 730         } else
 731                 list_del_init(&timer->it.cpu.entry);
 732         spin_unlock(&p->sighand->siglock);
 733
 734         /*
 735          * We need to sample the current value to convert the new
 736          * value from to relative and absolute, and to convert the
 737          * old value from absolute to relative.  To set a process
 738          * timer, we need a sample to balance the thread expiry
 739          * times (in arm_timer).  With an absolute time, we must
 740          * check if it's already passed.  In short, we need a sample.
 741          */
 742         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 743                 cpu_clock_sample(timer->it_clock, p, &val);
 744         } else {
 745                 cpu_clock_sample_group(timer->it_clock, p, &val);
 746         }
 747
 748         if (old) {
 749                 if (old_expires.sched == 0) {
 750                         old->it_value.tv_sec = 0;
 751                         old->it_value.tv_nsec = 0;
 752                 } else {
 753                         /*
 754                          * Update the timer in case it has
 755                          * overrun already.  If it has,
 756                          * we'll report it as having overrun
 757                          * and with the next reloaded timer
 758                          * already ticking, though we are
 759                          * swallowing that pending
 760                          * notification here to install the
 761                          * new setting.
 762                          */
 763                         bump_cpu_timer(timer, val);
 764                         if (cpu_time_before(timer->it_clock, val,
 765                                             timer->it.cpu.expires)) {
 766                                 old_expires = cpu_time_sub(
 767                                         timer->it_clock,
 768                                         timer->it.cpu.expires, val);
 769                                 sample_to_timespec(timer->it_clock,
 770                                                    old_expires,
 771                                                    &old->it_value);
 772                         } else {
 773                                 old->it_value.tv_nsec = 1;
 774                                 old->it_value.tv_sec = 0;
 775                         }
 776                 }
 777         }
 778
 779         if (unlikely(ret)) {
 780                 /*
 781                  * We are colliding with the timer actually firing.
 782                  * Punt after filling in the timer's old value, and
 783                  * disable this firing since we are already reporting
 784                  * it as an overrun (thanks to bump_cpu_timer above).
 785                  */
 786                 read_unlock(&tasklist_lock);
 787                 goto out;
 788         }
 789
 790         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 791                 cpu_time_add(timer->it_clock, &new_expires, val);
 792         }
 793
 794         /*
 795          * Install the new expiry time (or zero).
 796          * For a timer with no notification action, we don't actually
 797          * arm the timer (we'll just fake it for timer_gettime).
 798          */
 799         timer->it.cpu.expires = new_expires;
 800         if (new_expires.sched != 0 &&
 801             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 802             cpu_time_before(timer->it_clock, val, new_expires)) {
 803                 arm_timer(timer, val);
 804         }
 805
 806         read_unlock(&tasklist_lock);
 807
 808         /*
 809          * Install the new reload setting, and
 810          * set up the signal and overrun bookkeeping.
 811          */
 812         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 813                                                 &new->it_interval);
 814
 815         /*
 816          * This acts as a modification timestamp for the timer,
 817          * so any automatic reload attempt will punt on seeing
 818          * that we have reset the timer manually.
 819          */
 820         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 821                 ~REQUEUE_PENDING;
 822         timer->it_overrun_last = 0;
 823         timer->it_overrun = -1;
 824
 825         if (new_expires.sched != 0 &&
 826             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 827             !cpu_time_before(timer->it_clock, val, new_expires)) {
 828                 /*
 829                  * The designated time already passed, so we notify
 830                  * immediately, even if the thread never runs to
 831                  * accumulate more time on this clock.
 832                  */
 833                 cpu_timer_fire(timer);
 834         }
 835
 836         ret = 0;
 837  out:
 838         if (old) {
 839                 sample_to_timespec(timer->it_clock,
 840                                    timer->it.cpu.incr, &old->it_interval);
 841         }
 842         return ret;
 843 }
 844
 845 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 846 {
 847         union cpu_time_count now;
 848         struct task_struct *p = timer->it.cpu.task;
 849         int clear_dead;
 850
 851         /*
 852          * Easy part: convert the reload time.
 853          */
 854         sample_to_timespec(timer->it_clock,
 855                            timer->it.cpu.incr, &itp->it_interval);
 856
 857         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 858                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 859                 return;
 860         }
 861
 862         if (unlikely(p == NULL)) {
 863                 /*
 864                  * This task already died and the timer will never fire.
 865                  * In this case, expires is actually the dead value.
 866                  */
 867         dead:
 868                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 869                                    &itp->it_value);
 870                 return;
 871         }
 872
 873         /*
 874          * Sample the clock to take the difference with the expiry time.
 875          */
 876         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 877                 cpu_clock_sample(timer->it_clock, p, &now);
 878                 clear_dead = p->exit_state;
 879         } else {
 880                 read_lock(&tasklist_lock);
 881                 if (unlikely(p->signal == NULL)) {
 882                         /*
 883                          * The process has been reaped.
 884                          * We can't even collect a sample any more.
 885                          * Call the timer disarmed, nothing else to do.
 886                          */
 887                         put_task_struct(p);
 888                         timer->it.cpu.task = NULL;
 889                         timer->it.cpu.expires.sched = 0;
 890                         read_unlock(&tasklist_lock);
 891                         goto dead;
 892                 } else {
 893                         cpu_clock_sample_group(timer->it_clock, p, &now);
 894                         clear_dead = (unlikely(p->exit_state) &&
 895                                       thread_group_empty(p));
 896                 }
 897                 read_unlock(&tasklist_lock);
 898         }
 899
 900         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 901                 if (timer->it.cpu.incr.sched == 0 &&
 902                     cpu_time_before(timer->it_clock,
 903                                     timer->it.cpu.expires, now)) {
 904                         /*
 905                          * Do-nothing timer expired and has no reload,
 906                          * so it's as if it was never set.
 907                          */
 908                         timer->it.cpu.expires.sched = 0;
 909                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 910                         return;
 911                 }
 912                 /*
 913                  * Account for any expirations and reloads that should
 914                  * have happened.
 915                  */
 916                 bump_cpu_timer(timer, now);
 917         }
 918
 919         if (unlikely(clear_dead)) {
 920                 /*
 921                  * We've noticed that the thread is dead, but
 922                  * not yet reaped.  Take this opportunity to
 923                  * drop our task ref.
 924                  */
 925                 clear_dead_task(timer, now);
 926                 goto dead;
 927         }
 928
 929         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 930                 sample_to_timespec(timer->it_clock,
 931                                    cpu_time_sub(timer->it_clock,
 932                                                 timer->it.cpu.expires, now),
 933                                    &itp->it_value);
 934         } else {
 935                 /*
 936                  * The timer should have expired already, but the firing
 937                  * hasn't taken place yet.  Say it's just about to expire.
 938                  */
 939                 itp->it_value.tv_nsec = 1;
 940                 itp->it_value.tv_sec = 0;
 941         }
 942 }
 943
 944 /*
 945  * Check for any per-thread CPU timers that have fired and move them off
 946  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 947  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 948  */
 949 static void check_thread_timers(struct task_struct *tsk,
 950                                 struct list_head *firing)
 951 {
 952         int maxfire;
 953         struct list_head *timers = tsk->cpu_timers;
 954
 955         maxfire = 20;
 956         tsk->it_prof_expires = cputime_zero;
 957         while (!list_empty(timers)) {
 958                 struct cpu_timer_list *t = list_entry(timers->next,
 959                                                       struct cpu_timer_list,
 960                                                       entry);
 961                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 962                         tsk->it_prof_expires = t->expires.cpu;
 963                         break;
 964                 }
 965                 t->firing = 1;
 966                 list_move_tail(&t->entry, firing);
 967         }
 968
 969         ++timers;
 970         maxfire = 20;
 971         tsk->it_virt_expires = cputime_zero;
 972         while (!list_empty(timers)) {
 973                 struct cpu_timer_list *t = list_entry(timers->next,
 974                                                       struct cpu_timer_list,
 975                                                       entry);
 976                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 977                         tsk->it_virt_expires = t->expires.cpu;
 978                         break;
 979                 }
 980                 t->firing = 1;
 981                 list_move_tail(&t->entry, firing);
 982         }
 983
 984         ++timers;
 985         maxfire = 20;
 986         tsk->it_sched_expires = 0;
 987         while (!list_empty(timers)) {
 988                 struct cpu_timer_list *t = list_entry(timers->next,
 989                                                       struct cpu_timer_list,
 990                                                       entry);
 991                 if (!--maxfire || tsk->sched_time < t->expires.sched) {
 992                         tsk->it_sched_expires = t->expires.sched;
 993                         break;
 994                 }
 995                 t->firing = 1;
 996                 list_move_tail(&t->entry, firing);
 997         }
 998 }
 999
1000 /*
1001  * Check for any per-thread CPU timers that have fired and move them
1002  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1003  * have already been taken off.
1004  */
1005 static void check_process_timers(struct task_struct *tsk,
1006                                  struct list_head *firing)
1007 {
1008         int maxfire;
1009         struct signal_struct *const sig = tsk->signal;
1010         cputime_t utime, stime, ptime, virt_expires, prof_expires;
1011         unsigned long long sched_time, sched_expires;
1012         struct task_struct *t;
1013         struct list_head *timers = sig->cpu_timers;
1014
1015         /*
1016          * Don't sample the current process CPU clocks if there are no timers.
1017          */
1018         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1019             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1020             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1021             list_empty(&timers[CPUCLOCK_VIRT]) &&
1022             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1023             list_empty(&timers[CPUCLOCK_SCHED]))
1024                 return;
1025
1026         /*
1027          * Collect the current process totals.
1028          */
1029         utime = sig->utime;
1030         stime = sig->stime;
1031         sched_time = sig->sched_time;
1032         t = tsk;
1033         do {
1034                 utime = cputime_add(utime, t->utime);
1035                 stime = cputime_add(stime, t->stime);
1036                 sched_time += t->sched_time;
1037                 t = next_thread(t);
1038         } while (t != tsk);
1039         ptime = cputime_add(utime, stime);
1040
1041         maxfire = 20;
1042         prof_expires = cputime_zero;
1043         while (!list_empty(timers)) {
1044                 struct cpu_timer_list *t = list_entry(timers->next,
1045                                                       struct cpu_timer_list,
1046                                                       entry);
1047                 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
1048                         prof_expires = t->expires.cpu;
1049                         break;
1050                 }
1051                 t->firing = 1;
1052                 list_move_tail(&t->entry, firing);
1053         }
1054
1055         ++timers;
1056         maxfire = 20;
1057         virt_expires = cputime_zero;
1058         while (!list_empty(timers)) {
1059                 struct cpu_timer_list *t = list_entry(timers->next,
1060                                                       struct cpu_timer_list,
1061                                                       entry);
1062                 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
1063                         virt_expires = t->expires.cpu;
1064                         break;
1065                 }
1066                 t->firing = 1;
1067                 list_move_tail(&t->entry, firing);
1068         }
1069
1070         ++timers;
1071         maxfire = 20;
1072         sched_expires = 0;
1073         while (!list_empty(timers)) {
1074                 struct cpu_timer_list *t = list_entry(timers->next,
1075                                                       struct cpu_timer_list,
1076                                                       entry);
1077                 if (!--maxfire || sched_time < t->expires.sched) {
1078                         sched_expires = t->expires.sched;
1079                         break;
1080                 }
1081                 t->firing = 1;
1082                 list_move_tail(&t->entry, firing);
1083         }
1084
1085         /*
1086          * Check for the special case process timers.
1087          */
1088         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1089                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1090                         /* ITIMER_PROF fires and reloads.  */
1091                         sig->it_prof_expires = sig->it_prof_incr;
1092                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1093                                 sig->it_prof_expires = cputime_add(
1094                                         sig->it_prof_expires, ptime);
1095                         }
1096                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1097                 }
1098                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1099                     (cputime_eq(prof_expires, cputime_zero) ||
1100                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1101                         prof_expires = sig->it_prof_expires;
1102                 }
1103         }
1104         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1105                 if (cputime_ge(utime, sig->it_virt_expires)) {
1106                         /* ITIMER_VIRTUAL fires and reloads.  */
1107                         sig->it_virt_expires = sig->it_virt_incr;
1108                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1109                                 sig->it_virt_expires = cputime_add(
1110                                         sig->it_virt_expires, utime);
1111                         }
1112                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1113                 }
1114                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1115                     (cputime_eq(virt_expires, cputime_zero) ||
1116                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1117                         virt_expires = sig->it_virt_expires;
1118                 }
1119         }
1120         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1121                 unsigned long psecs = cputime_to_secs(ptime);
1122                 cputime_t x;
1123                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1124                         /*
1125                          * At the hard limit, we just die.
1126                          * No need to calculate anything else now.
1127                          */
1128                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1129                         return;
1130                 }
1131                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1132                         /*
1133                          * At the soft limit, send a SIGXCPU every second.
1134                          */
1135                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1136                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1137                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1138                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1139                         }
1140                 }
1141                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1142                 if (cputime_eq(prof_expires, cputime_zero) ||
1143                     cputime_lt(x, prof_expires)) {
1144                         prof_expires = x;
1145                 }
1146         }
1147
1148         if (!cputime_eq(prof_expires, cputime_zero) ||
1149             !cputime_eq(virt_expires, cputime_zero) ||
1150             sched_expires != 0) {
1151                 /*
1152                  * Rebalance the threads' expiry times for the remaining
1153                  * process CPU timers.
1154                  */
1155
1156                 cputime_t prof_left, virt_left, ticks;
1157                 unsigned long long sched_left, sched;
1158                 const unsigned int nthreads = atomic_read(&sig->live);
1159
1160                 if (!nthreads)
1161                         return;
1162
1163                 prof_left = cputime_sub(prof_expires, utime);
1164                 prof_left = cputime_sub(prof_left, stime);
1165                 prof_left = cputime_div(prof_left, nthreads);
1166                 virt_left = cputime_sub(virt_expires, utime);
1167                 virt_left = cputime_div(virt_left, nthreads);
1168                 if (sched_expires) {
1169                         sched_left = sched_expires - sched_time;
1170                         do_div(sched_left, nthreads);
1171                 } else {
1172                         sched_left = 0;
1173                 }
1174                 t = tsk;
1175                 do {
1176                         ticks = cputime_add(cputime_add(t->utime, t->stime),
1177                                             prof_left);
1178                         if (!cputime_eq(prof_expires, cputime_zero) &&
1179                             (cputime_eq(t->it_prof_expires, cputime_zero) ||
1180                              cputime_gt(t->it_prof_expires, ticks))) {
1181                                 t->it_prof_expires = ticks;
1182                         }
1183
1184                         ticks = cputime_add(t->utime, virt_left);
1185                         if (!cputime_eq(virt_expires, cputime_zero) &&
1186                             (cputime_eq(t->it_virt_expires, cputime_zero) ||
1187                              cputime_gt(t->it_virt_expires, ticks))) {
1188                                 t->it_virt_expires = ticks;
1189                         }
1190
1191                         sched = t->sched_time + sched_left;
1192                         if (sched_expires && (t->it_sched_expires == 0 ||
1193                                               t->it_sched_expires > sched)) {
1194                                 t->it_sched_expires = sched;
1195                         }
1196
1197                         do {
1198                                 t = next_thread(t);
1199                         } while (unlikely(t->flags & PF_EXITING));
1200                 } while (t != tsk);
1201         }
1202 }
1203
1204 /*
1205  * This is called from the signal code (via do_schedule_next_timer)
1206  * when the last timer signal was delivered and we have to reload the timer.
1207  */
1208 void posix_cpu_timer_schedule(struct k_itimer *timer)
1209 {
1210         struct task_struct *p = timer->it.cpu.task;
1211         union cpu_time_count now;
1212
1213         if (unlikely(p == NULL))
1214                 /*
1215                  * The task was cleaned up already, no future firings.
1216                  */
1217                 goto out;
1218
1219         /*
1220          * Fetch the current sample and update the timer's expiry time.
1221          */
1222         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1223                 cpu_clock_sample(timer->it_clock, p, &now);
1224                 bump_cpu_timer(timer, now);
1225                 if (unlikely(p->exit_state)) {
1226                         clear_dead_task(timer, now);
1227                         goto out;
1228                 }
1229                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1230         } else {
1231                 read_lock(&tasklist_lock);
1232                 if (unlikely(p->signal == NULL)) {
1233                         /*
1234                          * The process has been reaped.
1235                          * We can't even collect a sample any more.
1236                          */
1237                         put_task_struct(p);
1238                         timer->it.cpu.task = p = NULL;
1239                         timer->it.cpu.expires.sched = 0;
1240                         goto out_unlock;
1241                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1242                         /*
1243                          * We've noticed that the thread is dead, but
1244                          * not yet reaped.  Take this opportunity to
1245                          * drop our task ref.
1246                          */
1247                         clear_dead_task(timer, now);
1248                         goto out_unlock;
1249                 }
1250                 cpu_clock_sample_group(timer->it_clock, p, &now);
1251                 bump_cpu_timer(timer, now);
1252                 /* Leave the tasklist_lock locked for the call below.  */
1253         }
1254
1255         /*
1256          * Now re-arm for the new expiry time.
1257          */
1258         arm_timer(timer, now);
1259
1260 out_unlock:
1261         read_unlock(&tasklist_lock);
1262
1263 out:
1264         timer->it_overrun_last = timer->it_overrun;
1265         timer->it_overrun = -1;
1266         ++timer->it_requeue_pending;
1267 }
1268
1269 /*
1270  * This is called from the timer interrupt handler.  The irq handler has
1271  * already updated our counts.  We need to check if any timers fire now.
1272  * Interrupts are disabled.
1273  */
1274 void run_posix_cpu_timers(struct task_struct *tsk)
1275 {
1276         LIST_HEAD(firing);
1277         struct k_itimer *timer, *next;
1278
1279         BUG_ON(!irqs_disabled());
1280
1281 #define UNEXPIRED(clock) \
1282                 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1283                  cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1284
1285         if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1286             (tsk->it_sched_expires == 0 ||
1287              tsk->sched_time < tsk->it_sched_expires))
1288                 return;
1289
1290 #undef  UNEXPIRED
1291
1292         BUG_ON(tsk->exit_state);
1293
1294         /*
1295          * Double-check with locks held.
1296          */
1297         read_lock(&tasklist_lock);
1298         spin_lock(&tsk->sighand->siglock);
1299
1300         /*
1301          * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1302          * all the timers that are firing, and put them on the firing list.
1303          */
1304         check_thread_timers(tsk, &firing);
1305         check_process_timers(tsk, &firing);
1306
1307         /*
1308          * We must release these locks before taking any timer's lock.
1309          * There is a potential race with timer deletion here, as the
1310          * siglock now protects our private firing list.  We have set
1311          * the firing flag in each timer, so that a deletion attempt
1312          * that gets the timer lock before we do will give it up and
1313          * spin until we've taken care of that timer below.
1314          */
1315         spin_unlock(&tsk->sighand->siglock);
1316         read_unlock(&tasklist_lock);
1317
1318         /*
1319          * Now that all the timers on our list have the firing flag,
1320          * noone will touch their list entries but us.  We'll take
1321          * each timer's lock before clearing its firing flag, so no
1322          * timer call will interfere.
1323          */
1324         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1325                 int firing;
1326                 spin_lock(&timer->it_lock);
1327                 list_del_init(&timer->it.cpu.entry);
1328                 firing = timer->it.cpu.firing;
1329                 timer->it.cpu.firing = 0;
1330                 /*
1331                  * The firing flag is -1 if we collided with a reset
1332                  * of the timer, which already reported this
1333                  * almost-firing as an overrun.  So don't generate an event.
1334                  */
1335                 if (likely(firing >= 0)) {
1336                         cpu_timer_fire(timer);
1337                 }
1338                 spin_unlock(&timer->it_lock);
1339         }
1340 }
1341
1342 /*
1343  * Set one of the process-wide special case CPU timers.
1344  * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1345  * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1346  * absolute; non-null for ITIMER_*, where *newval is relative and we update
1347  * it to be absolute, *oldval is absolute and we update it to be relative.
1348  */
1349 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1350                            cputime_t *newval, cputime_t *oldval)
1351 {
1352         union cpu_time_count now;
1353         struct list_head *head;
1354
1355         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1356         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1357
1358         if (oldval) {
1359                 if (!cputime_eq(*oldval, cputime_zero)) {
1360                         if (cputime_le(*oldval, now.cpu)) {
1361                                 /* Just about to fire. */
1362                                 *oldval = jiffies_to_cputime(1);
1363                         } else {
1364                                 *oldval = cputime_sub(*oldval, now.cpu);
1365                         }
1366                 }
1367
1368                 if (cputime_eq(*newval, cputime_zero))
1369                         return;
1370                 *newval = cputime_add(*newval, now.cpu);
1371
1372                 /*
1373                  * If the RLIMIT_CPU timer will expire before the
1374                  * ITIMER_PROF timer, we have nothing else to do.
1375                  */
1376                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1377                     < cputime_to_secs(*newval))
1378                         return;
1379         }
1380
1381         /*
1382          * Check whether there are any process timers already set to fire
1383          * before this one.  If so, we don't have anything more to do.
1384          */
1385         head = &tsk->signal->cpu_timers[clock_idx];
1386         if (list_empty(head) ||
1387             cputime_ge(list_entry(head->next,
1388                                   struct cpu_timer_list, entry)->expires.cpu,
1389                        *newval)) {
1390                 /*
1391                  * Rejigger each thread's expiry time so that one will
1392                  * notice before we hit the process-cumulative expiry time.
1393                  */
1394                 union cpu_time_count expires = { .sched = 0 };
1395                 expires.cpu = *newval;
1396                 process_timer_rebalance(tsk, clock_idx, expires, now);
1397         }
1398 }
1399
1400 static long posix_cpu_clock_nanosleep_restart(struct restart_block *);
1401
1402 int posix_cpu_nsleep(clockid_t which_clock, int flags,
1403                      struct timespec *rqtp)
1404 {
1405         struct restart_block *restart_block =
1406             &current_thread_info()->restart_block;
1407         struct k_itimer timer;
1408         int error;
1409
1410         /*
1411          * Diagnose required errors first.
1412          */
1413         if (CPUCLOCK_PERTHREAD(which_clock) &&
1414             (CPUCLOCK_PID(which_clock) == 0 ||
1415              CPUCLOCK_PID(which_clock) == current->pid))
1416                 return -EINVAL;
1417
1418         /*
1419          * Set up a temporary timer and then wait for it to go off.
1420          */
1421         memset(&timer, 0, sizeof timer);
1422         spin_lock_init(&timer.it_lock);
1423         timer.it_clock = which_clock;
1424         timer.it_overrun = -1;
1425         error = posix_cpu_timer_create(&timer);
1426         timer.it_process = current;
1427         if (!error) {
1428                 struct timespec __user *rmtp;
1429                 static struct itimerspec zero_it;
1430                 struct itimerspec it = { .it_value = *rqtp,
1431                                          .it_interval = {} };
1432
1433                 spin_lock_irq(&timer.it_lock);
1434                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1435                 if (error) {
1436                         spin_unlock_irq(&timer.it_lock);
1437                         return error;
1438                 }
1439
1440                 while (!signal_pending(current)) {
1441                         if (timer.it.cpu.expires.sched == 0) {
1442                                 /*
1443                                  * Our timer fired and was reset.
1444                                  */
1445                                 spin_unlock_irq(&timer.it_lock);
1446                                 return 0;
1447                         }
1448
1449                         /*
1450                          * Block until cpu_timer_fire (or a signal) wakes us.
1451                          */
1452                         __set_current_state(TASK_INTERRUPTIBLE);
1453                         spin_unlock_irq(&timer.it_lock);
1454                         schedule();
1455                         spin_lock_irq(&timer.it_lock);
1456                 }
1457
1458                 /*
1459                  * We were interrupted by a signal.
1460                  */
1461                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1462                 posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1463                 spin_unlock_irq(&timer.it_lock);
1464
1465                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1466                         /*
1467                          * It actually did fire already.
1468                          */
1469                         return 0;
1470                 }
1471
1472                 /*
1473                  * Report back to the user the time still remaining.
1474                  */
1475                 rmtp = (struct timespec __user *) restart_block->arg1;
1476                 if (rmtp != NULL && !(flags & TIMER_ABSTIME) &&
1477                     copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1478                         return -EFAULT;
1479
1480                 restart_block->fn = posix_cpu_clock_nanosleep_restart;
1481                 /* Caller already set restart_block->arg1 */
1482                 restart_block->arg0 = which_clock;
1483                 restart_block->arg2 = rqtp->tv_sec;
1484                 restart_block->arg3 = rqtp->tv_nsec;
1485
1486                 error = -ERESTART_RESTARTBLOCK;
1487         }
1488
1489         return error;
1490 }
1491
1492 static long
1493 posix_cpu_clock_nanosleep_restart(struct restart_block *restart_block)
1494 {
1495         clockid_t which_clock = restart_block->arg0;
1496         struct timespec t = { .tv_sec = restart_block->arg2,
1497                               .tv_nsec = restart_block->arg3 };
1498         restart_block->fn = do_no_restart_syscall;
1499         return posix_cpu_nsleep(which_clock, TIMER_ABSTIME, &t);
1500 }
1501
1502
1503 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1504 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1505
1506 static int process_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1507 {
1508         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1509 }
1510 static int process_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1511 {
1512         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1513 }
1514 static int process_cpu_timer_create(struct k_itimer *timer)
1515 {
1516         timer->it_clock = PROCESS_CLOCK;
1517         return posix_cpu_timer_create(timer);
1518 }
1519 static int process_cpu_nsleep(clockid_t which_clock, int flags,
1520                               struct timespec *rqtp)
1521 {
1522         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1523 }
1524 static int thread_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1525 {
1526         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1527 }
1528 static int thread_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1529 {
1530         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1531 }
1532 static int thread_cpu_timer_create(struct k_itimer *timer)
1533 {
1534         timer->it_clock = THREAD_CLOCK;
1535         return posix_cpu_timer_create(timer);
1536 }
1537 static int thread_cpu_nsleep(clockid_t which_clock, int flags,
1538                               struct timespec *rqtp)
1539 {
1540         return -EINVAL;
1541 }
1542
1543 static __init int init_posix_cpu_timers(void)
1544 {
1545         struct k_clock process = {
1546                 .clock_getres = process_cpu_clock_getres,
1547                 .clock_get = process_cpu_clock_get,
1548                 .clock_set = do_posix_clock_nosettime,
1549                 .timer_create = process_cpu_timer_create,
1550                 .nsleep = process_cpu_nsleep,
1551         };
1552         struct k_clock thread = {
1553                 .clock_getres = thread_cpu_clock_getres,
1554                 .clock_get = thread_cpu_clock_get,
1555                 .clock_set = do_posix_clock_nosettime,
1556                 .timer_create = thread_cpu_timer_create,
1557                 .nsleep = thread_cpu_nsleep,
1558         };
1559
1560         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1561         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1562
1563         return 0;
1564 }
1565 __initcall(init_posix_cpu_timers);