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 #include <linux/vs_cvirt.h>
  10
  11 static int check_clock(const clockid_t which_clock)
  12 {
  13         int error = 0;
  14         struct task_struct *p;
  15         const pid_t pid = CPUCLOCK_PID(which_clock);
  16
  17         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  18                 return -EINVAL;
  19
  20         if (pid == 0)
  21                 return 0;
  22
  23         read_lock(&tasklist_lock);
  24         p = find_task_by_pid(pid);
  25         if (!p || (CPUCLOCK_PERTHREAD(which_clock) ?
  26                    p->tgid != current->tgid : p->tgid != pid)) {
  27                 error = -EINVAL;
  28         }
  29         read_unlock(&tasklist_lock);
  30
  31         return error;
  32 }
  33
  34 static inline union cpu_time_count
  35 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  36 {
  37         union cpu_time_count ret;
  38         ret.sched = 0;          /* high half always zero when .cpu used */
  39         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  40                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  41         } else {
  42                 ret.cpu = timespec_to_cputime(tp);
  43         }
  44         return ret;
  45 }
  46
  47 static void sample_to_timespec(const clockid_t which_clock,
  48                                union cpu_time_count cpu,
  49                                struct timespec *tp)
  50 {
  51         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  52                 tp->tv_sec = div_long_long_rem(cpu.sched,
  53                                                NSEC_PER_SEC, &tp->tv_nsec);
  54         } else {
  55                 cputime_to_timespec(cpu.cpu, tp);
  56         }
  57 }
  58
  59 static inline int cpu_time_before(const clockid_t which_clock,
  60                                   union cpu_time_count now,
  61                                   union cpu_time_count then)
  62 {
  63         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  64                 return now.sched < then.sched;
  65         }  else {
  66                 return cputime_lt(now.cpu, then.cpu);
  67         }
  68 }
  69 static inline void cpu_time_add(const clockid_t which_clock,
  70                                 union cpu_time_count *acc,
  71                                 union cpu_time_count val)
  72 {
  73         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  74                 acc->sched += val.sched;
  75         }  else {
  76                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  77         }
  78 }
  79 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  80                                                 union cpu_time_count a,
  81                                                 union cpu_time_count b)
  82 {
  83         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  84                 a.sched -= b.sched;
  85         }  else {
  86                 a.cpu = cputime_sub(a.cpu, b.cpu);
  87         }
  88         return a;
  89 }
  90
  91 /*
  92  * Divide and limit the result to res >= 1
  93  *
  94  * This is necessary to prevent signal delivery starvation, when the result of
  95  * the division would be rounded down to 0.
  96  */
  97 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
  98 {
  99         cputime_t res = cputime_div(time, div);
 100
 101         return max_t(cputime_t, res, 1);
 102 }
 103
 104 /*
 105  * Update expiry time from increment, and increase overrun count,
 106  * given the current clock sample.
 107  */
 108 static void bump_cpu_timer(struct k_itimer *timer,
 109                                   union cpu_time_count now)
 110 {
 111         int i;
 112
 113         if (timer->it.cpu.incr.sched == 0)
 114                 return;
 115
 116         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 117                 unsigned long long delta, incr;
 118
 119                 if (now.sched < timer->it.cpu.expires.sched)
 120                         return;
 121                 incr = timer->it.cpu.incr.sched;
 122                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 123                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 124                 for (i = 0; incr < delta - incr; i++)
 125                         incr = incr << 1;
 126                 for (; i >= 0; incr >>= 1, i--) {
 127                         if (delta < incr)
 128                                 continue;
 129                         timer->it.cpu.expires.sched += incr;
 130                         timer->it_overrun += 1 << i;
 131                         delta -= incr;
 132                 }
 133         } else {
 134                 cputime_t delta, incr;
 135
 136                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 137                         return;
 138                 incr = timer->it.cpu.incr.cpu;
 139                 delta = cputime_sub(cputime_add(now.cpu, incr),
 140                                     timer->it.cpu.expires.cpu);
 141                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 142                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 143                              incr = cputime_add(incr, incr);
 144                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 145                         if (cputime_lt(delta, incr))
 146                                 continue;
 147                         timer->it.cpu.expires.cpu =
 148                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 149                         timer->it_overrun += 1 << i;
 150                         delta = cputime_sub(delta, incr);
 151                 }
 152         }
 153 }
 154
 155 static inline cputime_t prof_ticks(struct task_struct *p)
 156 {
 157         return cputime_add(p->utime, p->stime);
 158 }
 159 static inline cputime_t virt_ticks(struct task_struct *p)
 160 {
 161         return p->utime;
 162 }
 163 static inline unsigned long long sched_ns(struct task_struct *p)
 164 {
 165         return (p == current) ? current_sched_time(p) : p->sched_time;
 166 }
 167
 168 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 169 {
 170         int error = check_clock(which_clock);
 171         if (!error) {
 172                 tp->tv_sec = 0;
 173                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 174                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 175                         /*
 176                          * If sched_clock is using a cycle counter, we
 177                          * don't have any idea of its true resolution
 178                          * exported, but it is much more than 1s/HZ.
 179                          */
 180                         tp->tv_nsec = 1;
 181                 }
 182         }
 183         return error;
 184 }
 185
 186 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 187 {
 188         /*
 189          * You can never reset a CPU clock, but we check for other errors
 190          * in the call before failing with EPERM.
 191          */
 192         int error = check_clock(which_clock);
 193         if (error == 0) {
 194                 error = -EPERM;
 195         }
 196         return error;
 197 }
 198
 199
 200 /*
 201  * Sample a per-thread clock for the given task.
 202  */
 203 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 204                             union cpu_time_count *cpu)
 205 {
 206         switch (CPUCLOCK_WHICH(which_clock)) {
 207         default:
 208                 return -EINVAL;
 209         case CPUCLOCK_PROF:
 210                 cpu->cpu = prof_ticks(p);
 211                 break;
 212         case CPUCLOCK_VIRT:
 213                 cpu->cpu = virt_ticks(p);
 214                 break;
 215         case CPUCLOCK_SCHED:
 216                 cpu->sched = sched_ns(p);
 217                 break;
 218         }
 219         return 0;
 220 }
 221
 222 /*
 223  * Sample a process (thread group) clock for the given group_leader task.
 224  * Must be called with tasklist_lock held for reading.
 225  * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
 226  */
 227 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
 228                                          struct task_struct *p,
 229                                          union cpu_time_count *cpu)
 230 {
 231         struct task_struct *t = p;
 232         switch (clock_idx) {
 233         default:
 234                 return -EINVAL;
 235         case CPUCLOCK_PROF:
 236                 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
 237                 do {
 238                         cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
 239                         t = next_thread(t);
 240                 } while (t != p);
 241                 break;
 242         case CPUCLOCK_VIRT:
 243                 cpu->cpu = p->signal->utime;
 244                 do {
 245                         cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
 246                         t = next_thread(t);
 247                 } while (t != p);
 248                 break;
 249         case CPUCLOCK_SCHED:
 250                 cpu->sched = p->signal->sched_time;
 251                 /* Add in each other live thread.  */
 252                 while ((t = next_thread(t)) != p) {
 253                         cpu->sched += t->sched_time;
 254                 }
 255                 cpu->sched += sched_ns(p);
 256                 break;
 257         }
 258         return 0;
 259 }
 260
 261 /*
 262  * Sample a process (thread group) clock for the given group_leader task.
 263  * Must be called with tasklist_lock held for reading.
 264  */
 265 static int cpu_clock_sample_group(const clockid_t which_clock,
 266                                   struct task_struct *p,
 267                                   union cpu_time_count *cpu)
 268 {
 269         int ret;
 270         unsigned long flags;
 271         spin_lock_irqsave(&p->sighand->siglock, flags);
 272         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
 273                                             cpu);
 274         spin_unlock_irqrestore(&p->sighand->siglock, flags);
 275         return ret;
 276 }
 277
 278
 279 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 280 {
 281         const pid_t pid = CPUCLOCK_PID(which_clock);
 282         int error = -EINVAL;
 283         union cpu_time_count rtn;
 284
 285         if (pid == 0) {
 286                 /*
 287                  * Special case constant value for our own clocks.
 288                  * We don't have to do any lookup to find ourselves.
 289                  */
 290                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 291                         /*
 292                          * Sampling just ourselves we can do with no locking.
 293                          */
 294                         error = cpu_clock_sample(which_clock,
 295                                                  current, &rtn);
 296                 } else {
 297                         read_lock(&tasklist_lock);
 298                         error = cpu_clock_sample_group(which_clock,
 299                                                        current, &rtn);
 300                         read_unlock(&tasklist_lock);
 301                 }
 302         } else {
 303                 /*
 304                  * Find the given PID, and validate that the caller
 305                  * should be able to see it.
 306                  */
 307                 struct task_struct *p;
 308                 read_lock(&tasklist_lock);
 309                 p = find_task_by_pid(pid);
 310                 if (p) {
 311                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 312                                 if (p->tgid == current->tgid) {
 313                                         error = cpu_clock_sample(which_clock,
 314                                                                  p, &rtn);
 315                                 }
 316                         } else if (p->tgid == pid && p->signal) {
 317                                 error = cpu_clock_sample_group(which_clock,
 318                                                                p, &rtn);
 319                         }
 320                 }
 321                 read_unlock(&tasklist_lock);
 322         }
 323
 324         if (error)
 325                 return error;
 326         sample_to_timespec(which_clock, rtn, tp);
 327         return 0;
 328 }
 329
 330
 331 /*
 332  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 333  * This is called from sys_timer_create with the new timer already locked.
 334  */
 335 int posix_cpu_timer_create(struct k_itimer *new_timer)
 336 {
 337         int ret = 0;
 338         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 339         struct task_struct *p;
 340
 341         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 342                 return -EINVAL;
 343
 344         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 345         new_timer->it.cpu.incr.sched = 0;
 346         new_timer->it.cpu.expires.sched = 0;
 347
 348         read_lock(&tasklist_lock);
 349         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 350                 if (pid == 0) {
 351                         p = current;
 352                 } else {
 353                         p = find_task_by_pid(pid);
 354                         if (p && p->tgid != current->tgid)
 355                                 p = NULL;
 356                 }
 357         } else {
 358                 if (pid == 0) {
 359                         p = current->group_leader;
 360                 } else {
 361                         p = find_task_by_pid(pid);
 362                         if (p && p->tgid != pid)
 363                                 p = NULL;
 364                 }
 365         }
 366         new_timer->it.cpu.task = p;
 367         if (p) {
 368                 get_task_struct(p);
 369         } else {
 370                 ret = -EINVAL;
 371         }
 372         read_unlock(&tasklist_lock);
 373
 374         return ret;
 375 }
 376
 377 /*
 378  * Clean up a CPU-clock timer that is about to be destroyed.
 379  * This is called from timer deletion with the timer already locked.
 380  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 381  * and try again.  (This happens when the timer is in the middle of firing.)
 382  */
 383 int posix_cpu_timer_del(struct k_itimer *timer)
 384 {
 385         struct task_struct *p = timer->it.cpu.task;
 386         int ret = 0;
 387
 388         if (likely(p != NULL)) {
 389                 read_lock(&tasklist_lock);
 390                 if (unlikely(p->signal == NULL)) {
 391                         /*
 392                          * We raced with the reaping of the task.
 393                          * The deletion should have cleared us off the list.
 394                          */
 395                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 396                 } else {
 397                         spin_lock(&p->sighand->siglock);
 398                         if (timer->it.cpu.firing)
 399                                 ret = TIMER_RETRY;
 400                         else
 401                                 list_del(&timer->it.cpu.entry);
 402                         spin_unlock(&p->sighand->siglock);
 403                 }
 404                 read_unlock(&tasklist_lock);
 405
 406                 if (!ret)
 407                         put_task_struct(p);
 408         }
 409
 410         return ret;
 411 }
 412
 413 /*
 414  * Clean out CPU timers still ticking when a thread exited.  The task
 415  * pointer is cleared, and the expiry time is replaced with the residual
 416  * time for later timer_gettime calls to return.
 417  * This must be called with the siglock held.
 418  */
 419 static void cleanup_timers(struct list_head *head,
 420                            cputime_t utime, cputime_t stime,
 421                            unsigned long long sched_time)
 422 {
 423         struct cpu_timer_list *timer, *next;
 424         cputime_t ptime = cputime_add(utime, stime);
 425
 426         list_for_each_entry_safe(timer, next, head, entry) {
 427                 list_del_init(&timer->entry);
 428                 if (cputime_lt(timer->expires.cpu, ptime)) {
 429                         timer->expires.cpu = cputime_zero;
 430                 } else {
 431                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 432                                                          ptime);
 433                 }
 434         }
 435
 436         ++head;
 437         list_for_each_entry_safe(timer, next, head, entry) {
 438                 list_del_init(&timer->entry);
 439                 if (cputime_lt(timer->expires.cpu, utime)) {
 440                         timer->expires.cpu = cputime_zero;
 441                 } else {
 442                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 443                                                          utime);
 444                 }
 445         }
 446
 447         ++head;
 448         list_for_each_entry_safe(timer, next, head, entry) {
 449                 list_del_init(&timer->entry);
 450                 if (timer->expires.sched < sched_time) {
 451                         timer->expires.sched = 0;
 452                 } else {
 453                         timer->expires.sched -= sched_time;
 454                 }
 455         }
 456 }
 457
 458 /*
 459  * These are both called with the siglock held, when the current thread
 460  * is being reaped.  When the final (leader) thread in the group is reaped,
 461  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 462  */
 463 void posix_cpu_timers_exit(struct task_struct *tsk)
 464 {
 465         cleanup_timers(tsk->cpu_timers,
 466                        tsk->utime, tsk->stime, tsk->sched_time);
 467
 468 }
 469 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 470 {
 471         cleanup_timers(tsk->signal->cpu_timers,
 472                        cputime_add(tsk->utime, tsk->signal->utime),
 473                        cputime_add(tsk->stime, tsk->signal->stime),
 474                        tsk->sched_time + tsk->signal->sched_time);
 475 }
 476
 477
 478 /*
 479  * Set the expiry times of all the threads in the process so one of them
 480  * will go off before the process cumulative expiry total is reached.
 481  */
 482 static void process_timer_rebalance(struct task_struct *p,
 483                                     unsigned int clock_idx,
 484                                     union cpu_time_count expires,
 485                                     union cpu_time_count val)
 486 {
 487         cputime_t ticks, left;
 488         unsigned long long ns, nsleft;
 489         struct task_struct *t = p;
 490         unsigned int nthreads = atomic_read(&p->signal->live);
 491
 492         if (!nthreads)
 493                 return;
 494
 495         switch (clock_idx) {
 496         default:
 497                 BUG();
 498                 break;
 499         case CPUCLOCK_PROF:
 500                 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
 501                                        nthreads);
 502                 do {
 503                         if (likely(!(t->flags & PF_EXITING))) {
 504                                 ticks = cputime_add(prof_ticks(t), left);
 505                                 if (cputime_eq(t->it_prof_expires,
 506                                                cputime_zero) ||
 507                                     cputime_gt(t->it_prof_expires, ticks)) {
 508                                         t->it_prof_expires = ticks;
 509                                 }
 510                         }
 511                         t = next_thread(t);
 512                 } while (t != p);
 513                 break;
 514         case CPUCLOCK_VIRT:
 515                 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
 516                                        nthreads);
 517                 do {
 518                         if (likely(!(t->flags & PF_EXITING))) {
 519                                 ticks = cputime_add(virt_ticks(t), left);
 520                                 if (cputime_eq(t->it_virt_expires,
 521                                                cputime_zero) ||
 522                                     cputime_gt(t->it_virt_expires, ticks)) {
 523                                         t->it_virt_expires = ticks;
 524                                 }
 525                         }
 526                         t = next_thread(t);
 527                 } while (t != p);
 528                 break;
 529         case CPUCLOCK_SCHED:
 530                 nsleft = expires.sched - val.sched;
 531                 do_div(nsleft, nthreads);
 532                 nsleft = max_t(unsigned long long, nsleft, 1);
 533                 do {
 534                         if (likely(!(t->flags & PF_EXITING))) {
 535                                 ns = t->sched_time + nsleft;
 536                                 if (t->it_sched_expires == 0 ||
 537                                     t->it_sched_expires > ns) {
 538                                         t->it_sched_expires = ns;
 539                                 }
 540                         }
 541                         t = next_thread(t);
 542                 } while (t != p);
 543                 break;
 544         }
 545 }
 546
 547 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 548 {
 549         /*
 550          * That's all for this thread or process.
 551          * We leave our residual in expires to be reported.
 552          */
 553         put_task_struct(timer->it.cpu.task);
 554         timer->it.cpu.task = NULL;
 555         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 556                                              timer->it.cpu.expires,
 557                                              now);
 558 }
 559
 560 /*
 561  * Insert the timer on the appropriate list before any timers that
 562  * expire later.  This must be called with the tasklist_lock held
 563  * for reading, and interrupts disabled.
 564  */
 565 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 566 {
 567         struct task_struct *p = timer->it.cpu.task;
 568         struct list_head *head, *listpos;
 569         struct cpu_timer_list *const nt = &timer->it.cpu;
 570         struct cpu_timer_list *next;
 571         unsigned long i;
 572
 573         if (CPUCLOCK_PERTHREAD(timer->it_clock) && (p->flags & PF_EXITING))
 574                 return;
 575
 576         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 577                 p->cpu_timers : p->signal->cpu_timers);
 578         head += CPUCLOCK_WHICH(timer->it_clock);
 579
 580         BUG_ON(!irqs_disabled());
 581         spin_lock(&p->sighand->siglock);
 582
 583         listpos = head;
 584         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 585                 list_for_each_entry(next, head, entry) {
 586                         if (next->expires.sched > nt->expires.sched)
 587                                 break;
 588                         listpos = &next->entry;
 589                 }
 590         } else {
 591                 list_for_each_entry(next, head, entry) {
 592                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 593                                 break;
 594                         listpos = &next->entry;
 595                 }
 596         }
 597         list_add(&nt->entry, listpos);
 598
 599         if (listpos == head) {
 600                 /*
 601                  * We are the new earliest-expiring timer.
 602                  * If we are a thread timer, there can always
 603                  * be a process timer telling us to stop earlier.
 604                  */
 605
 606                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 607                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 608                         default:
 609                                 BUG();
 610                         case CPUCLOCK_PROF:
 611                                 if (cputime_eq(p->it_prof_expires,
 612                                                cputime_zero) ||
 613                                     cputime_gt(p->it_prof_expires,
 614                                                nt->expires.cpu))
 615                                         p->it_prof_expires = nt->expires.cpu;
 616                                 break;
 617                         case CPUCLOCK_VIRT:
 618                                 if (cputime_eq(p->it_virt_expires,
 619                                                cputime_zero) ||
 620                                     cputime_gt(p->it_virt_expires,
 621                                                nt->expires.cpu))
 622                                         p->it_virt_expires = nt->expires.cpu;
 623                                 break;
 624                         case CPUCLOCK_SCHED:
 625                                 if (p->it_sched_expires == 0 ||
 626                                     p->it_sched_expires > nt->expires.sched)
 627                                         p->it_sched_expires = nt->expires.sched;
 628                                 break;
 629                         }
 630                 } else {
 631                         /*
 632                          * For a process timer, we must balance
 633                          * all the live threads' expirations.
 634                          */
 635                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 636                         default:
 637                                 BUG();
 638                         case CPUCLOCK_VIRT:
 639                                 if (!cputime_eq(p->signal->it_virt_expires,
 640                                                 cputime_zero) &&
 641                                     cputime_lt(p->signal->it_virt_expires,
 642                                                timer->it.cpu.expires.cpu))
 643                                         break;
 644                                 goto rebalance;
 645                         case CPUCLOCK_PROF:
 646                                 if (!cputime_eq(p->signal->it_prof_expires,
 647                                                 cputime_zero) &&
 648                                     cputime_lt(p->signal->it_prof_expires,
 649                                                timer->it.cpu.expires.cpu))
 650                                         break;
 651                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 652                                 if (i != RLIM_INFINITY &&
 653                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 654                                         break;
 655                                 goto rebalance;
 656                         case CPUCLOCK_SCHED:
 657                         rebalance:
 658                                 process_timer_rebalance(
 659                                         timer->it.cpu.task,
 660                                         CPUCLOCK_WHICH(timer->it_clock),
 661                                         timer->it.cpu.expires, now);
 662                                 break;
 663                         }
 664                 }
 665         }
 666
 667         spin_unlock(&p->sighand->siglock);
 668 }
 669
 670 /*
 671  * The timer is locked, fire it and arrange for its reload.
 672  */
 673 static void cpu_timer_fire(struct k_itimer *timer)
 674 {
 675         if (unlikely(timer->sigq == NULL)) {
 676                 /*
 677                  * This a special case for clock_nanosleep,
 678                  * not a normal timer from sys_timer_create.
 679                  */
 680                 wake_up_process(timer->it_process);
 681                 timer->it.cpu.expires.sched = 0;
 682         } else if (timer->it.cpu.incr.sched == 0) {
 683                 /*
 684                  * One-shot timer.  Clear it as soon as it's fired.
 685                  */
 686                 posix_timer_event(timer, 0);
 687                 timer->it.cpu.expires.sched = 0;
 688         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 689                 /*
 690                  * The signal did not get queued because the signal
 691                  * was ignored, so we won't get any callback to
 692                  * reload the timer.  But we need to keep it
 693                  * ticking in case the signal is deliverable next time.
 694                  */
 695                 posix_cpu_timer_schedule(timer);
 696         }
 697 }
 698
 699 /*
 700  * Guts of sys_timer_settime for CPU timers.
 701  * This is called with the timer locked and interrupts disabled.
 702  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 703  * and try again.  (This happens when the timer is in the middle of firing.)
 704  */
 705 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 706                         struct itimerspec *new, struct itimerspec *old)
 707 {
 708         struct task_struct *p = timer->it.cpu.task;
 709         union cpu_time_count old_expires, new_expires, val;
 710         int ret;
 711
 712         if (unlikely(p == NULL)) {
 713                 /*
 714                  * Timer refers to a dead task's clock.
 715                  */
 716                 return -ESRCH;
 717         }
 718
 719         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 720
 721         read_lock(&tasklist_lock);
 722         /*
 723          * We need the tasklist_lock to protect against reaping that
 724          * clears p->signal.  If p has just been reaped, we can no
 725          * longer get any information about it at all.
 726          */
 727         if (unlikely(p->signal == NULL)) {
 728                 read_unlock(&tasklist_lock);
 729                 put_task_struct(p);
 730                 timer->it.cpu.task = NULL;
 731                 return -ESRCH;
 732         }
 733
 734         /*
 735          * Disarm any old timer after extracting its expiry time.
 736          */
 737         BUG_ON(!irqs_disabled());
 738
 739         ret = 0;
 740         spin_lock(&p->sighand->siglock);
 741         old_expires = timer->it.cpu.expires;
 742         if (unlikely(timer->it.cpu.firing)) {
 743                 timer->it.cpu.firing = -1;
 744                 ret = TIMER_RETRY;
 745         } else
 746                 list_del_init(&timer->it.cpu.entry);
 747         spin_unlock(&p->sighand->siglock);
 748
 749         /*
 750          * We need to sample the current value to convert the new
 751          * value from to relative and absolute, and to convert the
 752          * old value from absolute to relative.  To set a process
 753          * timer, we need a sample to balance the thread expiry
 754          * times (in arm_timer).  With an absolute time, we must
 755          * check if it's already passed.  In short, we need a sample.
 756          */
 757         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 758                 cpu_clock_sample(timer->it_clock, p, &val);
 759         } else {
 760                 cpu_clock_sample_group(timer->it_clock, p, &val);
 761         }
 762
 763         if (old) {
 764                 if (old_expires.sched == 0) {
 765                         old->it_value.tv_sec = 0;
 766                         old->it_value.tv_nsec = 0;
 767                 } else {
 768                         /*
 769                          * Update the timer in case it has
 770                          * overrun already.  If it has,
 771                          * we'll report it as having overrun
 772                          * and with the next reloaded timer
 773                          * already ticking, though we are
 774                          * swallowing that pending
 775                          * notification here to install the
 776                          * new setting.
 777                          */
 778                         bump_cpu_timer(timer, val);
 779                         if (cpu_time_before(timer->it_clock, val,
 780                                             timer->it.cpu.expires)) {
 781                                 old_expires = cpu_time_sub(
 782                                         timer->it_clock,
 783                                         timer->it.cpu.expires, val);
 784                                 sample_to_timespec(timer->it_clock,
 785                                                    old_expires,
 786                                                    &old->it_value);
 787                         } else {
 788                                 old->it_value.tv_nsec = 1;
 789                                 old->it_value.tv_sec = 0;
 790                         }
 791                 }
 792         }
 793
 794         if (unlikely(ret)) {
 795                 /*
 796                  * We are colliding with the timer actually firing.
 797                  * Punt after filling in the timer's old value, and
 798                  * disable this firing since we are already reporting
 799                  * it as an overrun (thanks to bump_cpu_timer above).
 800                  */
 801                 read_unlock(&tasklist_lock);
 802                 goto out;
 803         }
 804
 805         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 806                 cpu_time_add(timer->it_clock, &new_expires, val);
 807         }
 808
 809         /*
 810          * Install the new expiry time (or zero).
 811          * For a timer with no notification action, we don't actually
 812          * arm the timer (we'll just fake it for timer_gettime).
 813          */
 814         timer->it.cpu.expires = new_expires;
 815         if (new_expires.sched != 0 &&
 816             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 817             cpu_time_before(timer->it_clock, val, new_expires)) {
 818                 arm_timer(timer, val);
 819         }
 820
 821         read_unlock(&tasklist_lock);
 822
 823         /*
 824          * Install the new reload setting, and
 825          * set up the signal and overrun bookkeeping.
 826          */
 827         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 828                                                 &new->it_interval);
 829
 830         /*
 831          * This acts as a modification timestamp for the timer,
 832          * so any automatic reload attempt will punt on seeing
 833          * that we have reset the timer manually.
 834          */
 835         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 836                 ~REQUEUE_PENDING;
 837         timer->it_overrun_last = 0;
 838         timer->it_overrun = -1;
 839
 840         if (new_expires.sched != 0 &&
 841             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 842             !cpu_time_before(timer->it_clock, val, new_expires)) {
 843                 /*
 844                  * The designated time already passed, so we notify
 845                  * immediately, even if the thread never runs to
 846                  * accumulate more time on this clock.
 847                  */
 848                 cpu_timer_fire(timer);
 849         }
 850
 851         ret = 0;
 852  out:
 853         if (old) {
 854                 sample_to_timespec(timer->it_clock,
 855                                    timer->it.cpu.incr, &old->it_interval);
 856         }
 857         return ret;
 858 }
 859
 860 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 861 {
 862         union cpu_time_count now;
 863         struct task_struct *p = timer->it.cpu.task;
 864         int clear_dead;
 865
 866         /*
 867          * Easy part: convert the reload time.
 868          */
 869         sample_to_timespec(timer->it_clock,
 870                            timer->it.cpu.incr, &itp->it_interval);
 871
 872         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 873                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 874                 return;
 875         }
 876
 877         if (unlikely(p == NULL)) {
 878                 /*
 879                  * This task already died and the timer will never fire.
 880                  * In this case, expires is actually the dead value.
 881                  */
 882         dead:
 883                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 884                                    &itp->it_value);
 885                 return;
 886         }
 887
 888         /*
 889          * Sample the clock to take the difference with the expiry time.
 890          */
 891         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 892                 cpu_clock_sample(timer->it_clock, p, &now);
 893                 clear_dead = p->exit_state;
 894         } else {
 895                 read_lock(&tasklist_lock);
 896                 if (unlikely(p->signal == NULL)) {
 897                         /*
 898                          * The process has been reaped.
 899                          * We can't even collect a sample any more.
 900                          * Call the timer disarmed, nothing else to do.
 901                          */
 902                         put_task_struct(p);
 903                         timer->it.cpu.task = NULL;
 904                         timer->it.cpu.expires.sched = 0;
 905                         read_unlock(&tasklist_lock);
 906                         goto dead;
 907                 } else {
 908                         cpu_clock_sample_group(timer->it_clock, p, &now);
 909                         clear_dead = (unlikely(p->exit_state) &&
 910                                       thread_group_empty(p));
 911                 }
 912                 read_unlock(&tasklist_lock);
 913         }
 914
 915         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 916                 if (timer->it.cpu.incr.sched == 0 &&
 917                     cpu_time_before(timer->it_clock,
 918                                     timer->it.cpu.expires, now)) {
 919                         /*
 920                          * Do-nothing timer expired and has no reload,
 921                          * so it's as if it was never set.
 922                          */
 923                         timer->it.cpu.expires.sched = 0;
 924                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 925                         return;
 926                 }
 927                 /*
 928                  * Account for any expirations and reloads that should
 929                  * have happened.
 930                  */
 931                 bump_cpu_timer(timer, now);
 932         }
 933
 934         if (unlikely(clear_dead)) {
 935                 /*
 936                  * We've noticed that the thread is dead, but
 937                  * not yet reaped.  Take this opportunity to
 938                  * drop our task ref.
 939                  */
 940                 clear_dead_task(timer, now);
 941                 goto dead;
 942         }
 943
 944         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 945                 sample_to_timespec(timer->it_clock,
 946                                    cpu_time_sub(timer->it_clock,
 947                                                 timer->it.cpu.expires, now),
 948                                    &itp->it_value);
 949         } else {
 950                 /*
 951                  * The timer should have expired already, but the firing
 952                  * hasn't taken place yet.  Say it's just about to expire.
 953                  */
 954                 itp->it_value.tv_nsec = 1;
 955                 itp->it_value.tv_sec = 0;
 956         }
 957 }
 958
 959 /*
 960  * Check for any per-thread CPU timers that have fired and move them off
 961  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 962  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 963  */
 964 static void check_thread_timers(struct task_struct *tsk,
 965                                 struct list_head *firing)
 966 {
 967         int maxfire;
 968         struct list_head *timers = tsk->cpu_timers;
 969
 970         maxfire = 20;
 971         tsk->it_prof_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(prof_ticks(tsk), t->expires.cpu)) {
 977                         tsk->it_prof_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_virt_expires = cputime_zero;
 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 || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 992                         tsk->it_virt_expires = t->expires.cpu;
 993                         break;
 994                 }
 995                 t->firing = 1;
 996                 list_move_tail(&t->entry, firing);
 997         }
 998
 999         ++timers;
1000         maxfire = 20;
1001         tsk->it_sched_expires = 0;
1002         while (!list_empty(timers)) {
1003                 struct cpu_timer_list *t = list_entry(timers->next,
1004                                                       struct cpu_timer_list,
1005                                                       entry);
1006                 if (!--maxfire || tsk->sched_time < t->expires.sched) {
1007                         tsk->it_sched_expires = t->expires.sched;
1008                         break;
1009                 }
1010                 t->firing = 1;
1011                 list_move_tail(&t->entry, firing);
1012         }
1013 }
1014
1015 /*
1016  * Check for any per-thread CPU timers that have fired and move them
1017  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1018  * have already been taken off.
1019  */
1020 static void check_process_timers(struct task_struct *tsk,
1021                                  struct list_head *firing)
1022 {
1023         int maxfire;
1024         struct signal_struct *const sig = tsk->signal;
1025         cputime_t utime, stime, ptime, virt_expires, prof_expires;
1026         unsigned long long sched_time, sched_expires;
1027         struct task_struct *t;
1028         struct list_head *timers = sig->cpu_timers;
1029
1030         /*
1031          * Don't sample the current process CPU clocks if there are no timers.
1032          */
1033         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1034             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1035             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1036             list_empty(&timers[CPUCLOCK_VIRT]) &&
1037             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1038             list_empty(&timers[CPUCLOCK_SCHED]))
1039                 return;
1040
1041         /*
1042          * Collect the current process totals.
1043          */
1044         utime = sig->utime;
1045         stime = sig->stime;
1046         sched_time = sig->sched_time;
1047         t = tsk;
1048         do {
1049                 utime = cputime_add(utime, t->utime);
1050                 stime = cputime_add(stime, t->stime);
1051                 sched_time += t->sched_time;
1052                 t = next_thread(t);
1053         } while (t != tsk);
1054         ptime = cputime_add(utime, stime);
1055
1056         maxfire = 20;
1057         prof_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(ptime, t->expires.cpu)) {
1063                         prof_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         virt_expires = cputime_zero;
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 || cputime_lt(utime, t->expires.cpu)) {
1078                         virt_expires = t->expires.cpu;
1079                         break;
1080                 }
1081                 t->firing = 1;
1082                 list_move_tail(&t->entry, firing);
1083         }
1084
1085         ++timers;
1086         maxfire = 20;
1087         sched_expires = 0;
1088         while (!list_empty(timers)) {
1089                 struct cpu_timer_list *t = list_entry(timers->next,
1090                                                       struct cpu_timer_list,
1091                                                       entry);
1092                 if (!--maxfire || sched_time < t->expires.sched) {
1093                         sched_expires = t->expires.sched;
1094                         break;
1095                 }
1096                 t->firing = 1;
1097                 list_move_tail(&t->entry, firing);
1098         }
1099
1100         /*
1101          * Check for the special case process timers.
1102          */
1103         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1104                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1105                         /* ITIMER_PROF fires and reloads.  */
1106                         sig->it_prof_expires = sig->it_prof_incr;
1107                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1108                                 sig->it_prof_expires = cputime_add(
1109                                         sig->it_prof_expires, ptime);
1110                         }
1111                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1112                 }
1113                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1114                     (cputime_eq(prof_expires, cputime_zero) ||
1115                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1116                         prof_expires = sig->it_prof_expires;
1117                 }
1118         }
1119         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1120                 if (cputime_ge(utime, sig->it_virt_expires)) {
1121                         /* ITIMER_VIRTUAL fires and reloads.  */
1122                         sig->it_virt_expires = sig->it_virt_incr;
1123                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1124                                 sig->it_virt_expires = cputime_add(
1125                                         sig->it_virt_expires, utime);
1126                         }
1127                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1128                 }
1129                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1130                     (cputime_eq(virt_expires, cputime_zero) ||
1131                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1132                         virt_expires = sig->it_virt_expires;
1133                 }
1134         }
1135         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1136                 unsigned long psecs = cputime_to_secs(ptime);
1137                 cputime_t x;
1138                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1139                         /*
1140                          * At the hard limit, we just die.
1141                          * No need to calculate anything else now.
1142                          */
1143                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1144                         return;
1145                 }
1146                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1147                         /*
1148                          * At the soft limit, send a SIGXCPU every second.
1149                          */
1150                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1151                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1152                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1153                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1154                         }
1155                 }
1156                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1157                 if (cputime_eq(prof_expires, cputime_zero) ||
1158                     cputime_lt(x, prof_expires)) {
1159                         prof_expires = x;
1160                 }
1161         }
1162
1163         if (!cputime_eq(prof_expires, cputime_zero) ||
1164             !cputime_eq(virt_expires, cputime_zero) ||
1165             sched_expires != 0) {
1166                 /*
1167                  * Rebalance the threads' expiry times for the remaining
1168                  * process CPU timers.
1169                  */
1170
1171                 cputime_t prof_left, virt_left, ticks;
1172                 unsigned long long sched_left, sched;
1173                 const unsigned int nthreads = atomic_read(&sig->live);
1174
1175                 if (!nthreads)
1176                         return;
1177
1178                 prof_left = cputime_sub(prof_expires, utime);
1179                 prof_left = cputime_sub(prof_left, stime);
1180                 prof_left = cputime_div_non_zero(prof_left, nthreads);
1181                 virt_left = cputime_sub(virt_expires, utime);
1182                 virt_left = cputime_div_non_zero(virt_left, nthreads);
1183                 if (sched_expires) {
1184                         sched_left = sched_expires - sched_time;
1185                         do_div(sched_left, nthreads);
1186                         sched_left = max_t(unsigned long long, sched_left, 1);
1187                 } else {
1188                         sched_left = 0;
1189                 }
1190                 t = tsk;
1191                 do {
1192                         if (unlikely(t->flags & PF_EXITING))
1193                                 continue;
1194
1195                         ticks = cputime_add(cputime_add(t->utime, t->stime),
1196                                             prof_left);
1197                         if (!cputime_eq(prof_expires, cputime_zero) &&
1198                             (cputime_eq(t->it_prof_expires, cputime_zero) ||
1199                              cputime_gt(t->it_prof_expires, ticks))) {
1200                                 t->it_prof_expires = ticks;
1201                         }
1202
1203                         ticks = cputime_add(t->utime, virt_left);
1204                         if (!cputime_eq(virt_expires, cputime_zero) &&
1205                             (cputime_eq(t->it_virt_expires, cputime_zero) ||
1206                              cputime_gt(t->it_virt_expires, ticks))) {
1207                                 t->it_virt_expires = ticks;
1208                         }
1209
1210                         sched = t->sched_time + sched_left;
1211                         if (sched_expires && (t->it_sched_expires == 0 ||
1212                                               t->it_sched_expires > sched)) {
1213                                 t->it_sched_expires = sched;
1214                         }
1215                 } while ((t = next_thread(t)) != tsk);
1216         }
1217 }
1218
1219 /*
1220  * This is called from the signal code (via do_schedule_next_timer)
1221  * when the last timer signal was delivered and we have to reload the timer.
1222  */
1223 void posix_cpu_timer_schedule(struct k_itimer *timer)
1224 {
1225         struct task_struct *p = timer->it.cpu.task;
1226         union cpu_time_count now;
1227
1228         if (unlikely(p == NULL))
1229                 /*
1230                  * The task was cleaned up already, no future firings.
1231                  */
1232                 goto out;
1233
1234         /*
1235          * Fetch the current sample and update the timer's expiry time.
1236          */
1237         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1238                 cpu_clock_sample(timer->it_clock, p, &now);
1239                 bump_cpu_timer(timer, now);
1240                 if (unlikely(p->exit_state)) {
1241                         clear_dead_task(timer, now);
1242                         goto out;
1243                 }
1244                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1245         } else {
1246                 read_lock(&tasklist_lock);
1247                 if (unlikely(p->signal == NULL)) {
1248                         /*
1249                          * The process has been reaped.
1250                          * We can't even collect a sample any more.
1251                          */
1252                         put_task_struct(p);
1253                         timer->it.cpu.task = p = NULL;
1254                         timer->it.cpu.expires.sched = 0;
1255                         goto out_unlock;
1256                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1257                         /*
1258                          * We've noticed that the thread is dead, but
1259                          * not yet reaped.  Take this opportunity to
1260                          * drop our task ref.
1261                          */
1262                         clear_dead_task(timer, now);
1263                         goto out_unlock;
1264                 }
1265                 cpu_clock_sample_group(timer->it_clock, p, &now);
1266                 bump_cpu_timer(timer, now);
1267                 /* Leave the tasklist_lock locked for the call below.  */
1268         }
1269
1270         /*
1271          * Now re-arm for the new expiry time.
1272          */
1273         arm_timer(timer, now);
1274
1275 out_unlock:
1276         read_unlock(&tasklist_lock);
1277
1278 out:
1279         timer->it_overrun_last = timer->it_overrun;
1280         timer->it_overrun = -1;
1281         ++timer->it_requeue_pending;
1282 }
1283
1284 /*
1285  * This is called from the timer interrupt handler.  The irq handler has
1286  * already updated our counts.  We need to check if any timers fire now.
1287  * Interrupts are disabled.
1288  */
1289 void run_posix_cpu_timers(struct task_struct *tsk)
1290 {
1291         LIST_HEAD(firing);
1292         struct k_itimer *timer, *next;
1293
1294         BUG_ON(!irqs_disabled());
1295
1296 #define UNEXPIRED(clock) \
1297                 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1298                  cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1299
1300         if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1301             (tsk->it_sched_expires == 0 ||
1302              tsk->sched_time < tsk->it_sched_expires))
1303                 return;
1304
1305 #undef  UNEXPIRED
1306
1307         /*
1308          * Double-check with locks held.
1309          */
1310         read_lock(&tasklist_lock);
1311         if (likely(tsk->signal != NULL)) {
1312                 spin_lock(&tsk->sighand->siglock);
1313
1314                 /*
1315                  * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1316                  * all the timers that are firing, and put them on the firing list.
1317                  */
1318                 check_thread_timers(tsk, &firing);
1319                 check_process_timers(tsk, &firing);
1320
1321                 /*
1322                  * We must release these locks before taking any timer's lock.
1323                  * There is a potential race with timer deletion here, as the
1324                  * siglock now protects our private firing list.  We have set
1325                  * the firing flag in each timer, so that a deletion attempt
1326                  * that gets the timer lock before we do will give it up and
1327                  * spin until we've taken care of that timer below.
1328                  */
1329                 spin_unlock(&tsk->sighand->siglock);
1330         }
1331         read_unlock(&tasklist_lock);
1332
1333         /*
1334          * Now that all the timers on our list have the firing flag,
1335          * noone will touch their list entries but us.  We'll take
1336          * each timer's lock before clearing its firing flag, so no
1337          * timer call will interfere.
1338          */
1339         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1340                 int firing;
1341                 spin_lock(&timer->it_lock);
1342                 list_del_init(&timer->it.cpu.entry);
1343                 firing = timer->it.cpu.firing;
1344                 timer->it.cpu.firing = 0;
1345                 /*
1346                  * The firing flag is -1 if we collided with a reset
1347                  * of the timer, which already reported this
1348                  * almost-firing as an overrun.  So don't generate an event.
1349                  */
1350                 if (likely(firing >= 0)) {
1351                         cpu_timer_fire(timer);
1352                 }
1353                 spin_unlock(&timer->it_lock);
1354         }
1355 }
1356
1357 /*
1358  * Set one of the process-wide special case CPU timers.
1359  * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1360  * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1361  * absolute; non-null for ITIMER_*, where *newval is relative and we update
1362  * it to be absolute, *oldval is absolute and we update it to be relative.
1363  */
1364 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1365                            cputime_t *newval, cputime_t *oldval)
1366 {
1367         union cpu_time_count now;
1368         struct list_head *head;
1369
1370         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1371         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1372
1373         if (oldval) {
1374                 if (!cputime_eq(*oldval, cputime_zero)) {
1375                         if (cputime_le(*oldval, now.cpu)) {
1376                                 /* Just about to fire. */
1377                                 *oldval = jiffies_to_cputime(1);
1378                         } else {
1379                                 *oldval = cputime_sub(*oldval, now.cpu);
1380                         }
1381                 }
1382
1383                 if (cputime_eq(*newval, cputime_zero))
1384                         return;
1385                 *newval = cputime_add(*newval, now.cpu);
1386
1387                 /*
1388                  * If the RLIMIT_CPU timer will expire before the
1389                  * ITIMER_PROF timer, we have nothing else to do.
1390                  */
1391                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1392                     < cputime_to_secs(*newval))
1393                         return;
1394         }
1395
1396         /*
1397          * Check whether there are any process timers already set to fire
1398          * before this one.  If so, we don't have anything more to do.
1399          */
1400         head = &tsk->signal->cpu_timers[clock_idx];
1401         if (list_empty(head) ||
1402             cputime_ge(list_entry(head->next,
1403                                   struct cpu_timer_list, entry)->expires.cpu,
1404                        *newval)) {
1405                 /*
1406                  * Rejigger each thread's expiry time so that one will
1407                  * notice before we hit the process-cumulative expiry time.
1408                  */
1409                 union cpu_time_count expires = { .sched = 0 };
1410                 expires.cpu = *newval;
1411                 process_timer_rebalance(tsk, clock_idx, expires, now);
1412         }
1413 }
1414
1415 static long posix_cpu_clock_nanosleep_restart(struct restart_block *);
1416
1417 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1418                      struct timespec *rqtp, struct timespec __user *rmtp)
1419 {
1420         struct restart_block *restart_block =
1421             &current_thread_info()->restart_block;
1422         struct k_itimer timer;
1423         int error;
1424
1425         /*
1426          * Diagnose required errors first.
1427          */
1428         if (CPUCLOCK_PERTHREAD(which_clock) &&
1429             (CPUCLOCK_PID(which_clock) == 0 ||
1430              CPUCLOCK_PID(which_clock) == current->pid))
1431                 return -EINVAL;
1432
1433         /*
1434          * Set up a temporary timer and then wait for it to go off.
1435          */
1436         memset(&timer, 0, sizeof timer);
1437         spin_lock_init(&timer.it_lock);
1438         timer.it_clock = which_clock;
1439         timer.it_overrun = -1;
1440         error = posix_cpu_timer_create(&timer);
1441         timer.it_process = current;
1442         if (!error) {
1443                 static struct itimerspec zero_it;
1444                 struct itimerspec it = { .it_value = *rqtp,
1445                                          .it_interval = {} };
1446
1447                 spin_lock_irq(&timer.it_lock);
1448                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1449                 if (error) {
1450                         spin_unlock_irq(&timer.it_lock);
1451                         return error;
1452                 }
1453
1454                 while (!signal_pending(current)) {
1455                         if (timer.it.cpu.expires.sched == 0) {
1456                                 /*
1457                                  * Our timer fired and was reset.
1458                                  */
1459                                 spin_unlock_irq(&timer.it_lock);
1460                                 return 0;
1461                         }
1462
1463                         /*
1464                          * Block until cpu_timer_fire (or a signal) wakes us.
1465                          */
1466                         __set_current_state(TASK_INTERRUPTIBLE);
1467                         spin_unlock_irq(&timer.it_lock);
1468                         schedule();
1469                         spin_lock_irq(&timer.it_lock);
1470                 }
1471
1472                 /*
1473                  * We were interrupted by a signal.
1474                  */
1475                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1476                 posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1477                 spin_unlock_irq(&timer.it_lock);
1478
1479                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1480                         /*
1481                          * It actually did fire already.
1482                          */
1483                         return 0;
1484                 }
1485
1486                 /*
1487                  * Report back to the user the time still remaining.
1488                  */
1489                 if (rmtp != NULL && !(flags & TIMER_ABSTIME) &&
1490                     copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1491                         return -EFAULT;
1492
1493                 restart_block->fn = posix_cpu_clock_nanosleep_restart;
1494                 /* Caller already set restart_block->arg1 */
1495                 restart_block->arg0 = which_clock;
1496                 restart_block->arg1 = (unsigned long) rmtp;
1497                 restart_block->arg2 = rqtp->tv_sec;
1498                 restart_block->arg3 = rqtp->tv_nsec;
1499
1500                 error = -ERESTART_RESTARTBLOCK;
1501         }
1502
1503         return error;
1504 }
1505
1506 static long
1507 posix_cpu_clock_nanosleep_restart(struct restart_block *restart_block)
1508 {
1509         clockid_t which_clock = restart_block->arg0;
1510         struct timespec __user *rmtp;
1511         struct timespec t;
1512
1513         rmtp = (struct timespec __user *) restart_block->arg1;
1514         t.tv_sec = restart_block->arg2;
1515         t.tv_nsec = restart_block->arg3;
1516
1517         restart_block->fn = do_no_restart_syscall;
1518         return posix_cpu_nsleep(which_clock, TIMER_ABSTIME, &t, rmtp);
1519 }
1520
1521
1522 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1523 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1524
1525 static int process_cpu_clock_getres(const clockid_t which_clock,
1526                                     struct timespec *tp)
1527 {
1528         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1529 }
1530 static int process_cpu_clock_get(const clockid_t which_clock,
1531                                  struct timespec *tp)
1532 {
1533         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1534 }
1535 static int process_cpu_timer_create(struct k_itimer *timer)
1536 {
1537         timer->it_clock = PROCESS_CLOCK;
1538         return posix_cpu_timer_create(timer);
1539 }
1540 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1541                               struct timespec *rqtp,
1542                               struct timespec __user *rmtp)
1543 {
1544         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1545 }
1546 static int thread_cpu_clock_getres(const clockid_t which_clock,
1547                                    struct timespec *tp)
1548 {
1549         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1550 }
1551 static int thread_cpu_clock_get(const clockid_t which_clock,
1552                                 struct timespec *tp)
1553 {
1554         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1555 }
1556 static int thread_cpu_timer_create(struct k_itimer *timer)
1557 {
1558         timer->it_clock = THREAD_CLOCK;
1559         return posix_cpu_timer_create(timer);
1560 }
1561 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1562                               struct timespec *rqtp, struct timespec __user *rmtp)
1563 {
1564         return -EINVAL;
1565 }
1566
1567 static __init int init_posix_cpu_timers(void)
1568 {
1569         struct k_clock process = {
1570                 .clock_getres = process_cpu_clock_getres,
1571                 .clock_get = process_cpu_clock_get,
1572                 .clock_set = do_posix_clock_nosettime,
1573                 .timer_create = process_cpu_timer_create,
1574                 .nsleep = process_cpu_nsleep,
1575         };
1576         struct k_clock thread = {
1577                 .clock_getres = thread_cpu_clock_getres,
1578                 .clock_get = thread_cpu_clock_get,
1579                 .clock_set = do_posix_clock_nosettime,
1580                 .timer_create = thread_cpu_timer_create,
1581                 .nsleep = thread_cpu_nsleep,
1582         };
1583
1584         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1585         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1586
1587         return 0;
1588 }
1589 __initcall(init_posix_cpu_timers);