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