kernel/posix-cpu-timers.c

   1 /*
   2  * Implement CPU time clocks for the POSIX clock interface.
   3  */
   4
   5 #include <linux/sched.h>
   6 #include <linux/posix-timers.h>
   7 #include <linux/errno.h>
   8 #include <linux/math64.h>
   9 #include <asm/uaccess.h>
  10 #include <linux/kernel_stat.h>
  11
  12 /*
  13  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
  14  */
  15 void update_rlimit_cpu(unsigned long rlim_new)
  16 {
  17         cputime_t cputime;
  18
  19         cputime = secs_to_cputime(rlim_new);
  20         if (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
  21             cputime_gt(current->signal->it_prof_expires, cputime)) {
  22                 spin_lock_irq(&current->sighand->siglock);
  23                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
  24                 spin_unlock_irq(&current->sighand->siglock);
  25         }
  26 }
  27
  28 static int check_clock(const clockid_t which_clock)
  29 {
  30         int error = 0;
  31         struct task_struct *p;
  32         const pid_t pid = CPUCLOCK_PID(which_clock);
  33
  34         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  35                 return -EINVAL;
  36
  37         if (pid == 0)
  38                 return 0;
  39
  40         read_lock(&tasklist_lock);
  41         p = find_task_by_vpid(pid);
  42         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  43                    same_thread_group(p, current) : thread_group_leader(p))) {
  44                 error = -EINVAL;
  45         }
  46         read_unlock(&tasklist_lock);
  47
  48         return error;
  49 }
  50
  51 static inline union cpu_time_count
  52 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  53 {
  54         union cpu_time_count ret;
  55         ret.sched = 0;          /* high half always zero when .cpu used */
  56         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  57                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  58         } else {
  59                 ret.cpu = timespec_to_cputime(tp);
  60         }
  61         return ret;
  62 }
  63
  64 static void sample_to_timespec(const clockid_t which_clock,
  65                                union cpu_time_count cpu,
  66                                struct timespec *tp)
  67 {
  68         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
  69                 *tp = ns_to_timespec(cpu.sched);
  70         else
  71                 cputime_to_timespec(cpu.cpu, tp);
  72 }
  73
  74 static inline int cpu_time_before(const clockid_t which_clock,
  75                                   union cpu_time_count now,
  76                                   union cpu_time_count then)
  77 {
  78         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  79                 return now.sched < then.sched;
  80         }  else {
  81                 return cputime_lt(now.cpu, then.cpu);
  82         }
  83 }
  84 static inline void cpu_time_add(const clockid_t which_clock,
  85                                 union cpu_time_count *acc,
  86                                 union cpu_time_count val)
  87 {
  88         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  89                 acc->sched += val.sched;
  90         }  else {
  91                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  92         }
  93 }
  94 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  95                                                 union cpu_time_count a,
  96                                                 union cpu_time_count b)
  97 {
  98         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  99                 a.sched -= b.sched;
 100         }  else {
 101                 a.cpu = cputime_sub(a.cpu, b.cpu);
 102         }
 103         return a;
 104 }
 105
 106 /*
 107  * Divide and limit the result to res >= 1
 108  *
 109  * This is necessary to prevent signal delivery starvation, when the result of
 110  * the division would be rounded down to 0.
 111  */
 112 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
 113 {
 114         cputime_t res = cputime_div(time, div);
 115
 116         return max_t(cputime_t, res, 1);
 117 }
 118
 119 /*
 120  * Update expiry time from increment, and increase overrun count,
 121  * given the current clock sample.
 122  */
 123 static void bump_cpu_timer(struct k_itimer *timer,
 124                                   union cpu_time_count now)
 125 {
 126         int i;
 127
 128         if (timer->it.cpu.incr.sched == 0)
 129                 return;
 130
 131         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 132                 unsigned long long delta, incr;
 133
 134                 if (now.sched < timer->it.cpu.expires.sched)
 135                         return;
 136                 incr = timer->it.cpu.incr.sched;
 137                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 138                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 139                 for (i = 0; incr < delta - incr; i++)
 140                         incr = incr << 1;
 141                 for (; i >= 0; incr >>= 1, i--) {
 142                         if (delta < incr)
 143                                 continue;
 144                         timer->it.cpu.expires.sched += incr;
 145                         timer->it_overrun += 1 << i;
 146                         delta -= incr;
 147                 }
 148         } else {
 149                 cputime_t delta, incr;
 150
 151                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 152                         return;
 153                 incr = timer->it.cpu.incr.cpu;
 154                 delta = cputime_sub(cputime_add(now.cpu, incr),
 155                                     timer->it.cpu.expires.cpu);
 156                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 157                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 158                              incr = cputime_add(incr, incr);
 159                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 160                         if (cputime_lt(delta, incr))
 161                                 continue;
 162                         timer->it.cpu.expires.cpu =
 163                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 164                         timer->it_overrun += 1 << i;
 165                         delta = cputime_sub(delta, incr);
 166                 }
 167         }
 168 }
 169
 170 static inline cputime_t prof_ticks(struct task_struct *p)
 171 {
 172         return cputime_add(p->utime, p->stime);
 173 }
 174 static inline cputime_t virt_ticks(struct task_struct *p)
 175 {
 176         return p->utime;
 177 }
 178
 179 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 180 {
 181         int error = check_clock(which_clock);
 182         if (!error) {
 183                 tp->tv_sec = 0;
 184                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 185                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 186                         /*
 187                          * If sched_clock is using a cycle counter, we
 188                          * don't have any idea of its true resolution
 189                          * exported, but it is much more than 1s/HZ.
 190                          */
 191                         tp->tv_nsec = 1;
 192                 }
 193         }
 194         return error;
 195 }
 196
 197 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 198 {
 199         /*
 200          * You can never reset a CPU clock, but we check for other errors
 201          * in the call before failing with EPERM.
 202          */
 203         int error = check_clock(which_clock);
 204         if (error == 0) {
 205                 error = -EPERM;
 206         }
 207         return error;
 208 }
 209
 210
 211 /*
 212  * Sample a per-thread clock for the given task.
 213  */
 214 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 215                             union cpu_time_count *cpu)
 216 {
 217         switch (CPUCLOCK_WHICH(which_clock)) {
 218         default:
 219                 return -EINVAL;
 220         case CPUCLOCK_PROF:
 221                 cpu->cpu = prof_ticks(p);
 222                 break;
 223         case CPUCLOCK_VIRT:
 224                 cpu->cpu = virt_ticks(p);
 225                 break;
 226         case CPUCLOCK_SCHED:
 227                 cpu->sched = task_sched_runtime(p);
 228                 break;
 229         }
 230         return 0;
 231 }
 232
 233 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
 234 {
 235         struct sighand_struct *sighand;
 236         struct signal_struct *sig;
 237         struct task_struct *t;
 238
 239         *times = INIT_CPUTIME;
 240
 241         rcu_read_lock();
 242         sighand = rcu_dereference(tsk->sighand);
 243         if (!sighand)
 244                 goto out;
 245
 246         sig = tsk->signal;
 247
 248         t = tsk;
 249         do {
 250                 times->utime = cputime_add(times->utime, t->utime);
 251                 times->stime = cputime_add(times->stime, t->stime);
 252                 times->sum_exec_runtime += t->se.sum_exec_runtime;
 253
 254                 t = next_thread(t);
 255         } while (t != tsk);
 256
 257         times->utime = cputime_add(times->utime, sig->utime);
 258         times->stime = cputime_add(times->stime, sig->stime);
 259         times->sum_exec_runtime += sig->sum_sched_runtime;
 260 out:
 261         rcu_read_unlock();
 262 }
 263
 264 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 265 {
 266         if (cputime_gt(b->utime, a->utime))
 267                 a->utime = b->utime;
 268
 269         if (cputime_gt(b->stime, a->stime))
 270                 a->stime = b->stime;
 271
 272         if (b->sum_exec_runtime > a->sum_exec_runtime)
 273                 a->sum_exec_runtime = b->sum_exec_runtime;
 274 }
 275
 276 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 277 {
 278         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 279         struct task_cputime sum;
 280         unsigned long flags;
 281
 282         spin_lock_irqsave(&cputimer->lock, flags);
 283         if (!cputimer->running) {
 284                 cputimer->running = 1;
 285                 /*
 286                  * The POSIX timer interface allows for absolute time expiry
 287                  * values through the TIMER_ABSTIME flag, therefore we have
 288                  * to synchronize the timer to the clock every time we start
 289                  * it.
 290                  */
 291                 thread_group_cputime(tsk, &sum);
 292                 update_gt_cputime(&cputimer->cputime, &sum);
 293         }
 294         *times = cputimer->cputime;
 295         spin_unlock_irqrestore(&cputimer->lock, flags);
 296 }
 297
 298 /*
 299  * Sample a process (thread group) clock for the given group_leader task.
 300  * Must be called with tasklist_lock held for reading.
 301  */
 302 static int cpu_clock_sample_group(const clockid_t which_clock,
 303                                   struct task_struct *p,
 304                                   union cpu_time_count *cpu)
 305 {
 306         struct task_cputime cputime;
 307
 308         switch (CPUCLOCK_WHICH(which_clock)) {
 309         default:
 310                 return -EINVAL;
 311         case CPUCLOCK_PROF:
 312                 thread_group_cputime(p, &cputime);
 313                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 314                 break;
 315         case CPUCLOCK_VIRT:
 316                 thread_group_cputime(p, &cputime);
 317                 cpu->cpu = cputime.utime;
 318                 break;
 319         case CPUCLOCK_SCHED:
 320                 cpu->sched = thread_group_sched_runtime(p);
 321                 break;
 322         }
 323         return 0;
 324 }
 325
 326
 327 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 328 {
 329         const pid_t pid = CPUCLOCK_PID(which_clock);
 330         int error = -EINVAL;
 331         union cpu_time_count rtn;
 332
 333         if (pid == 0) {
 334                 /*
 335                  * Special case constant value for our own clocks.
 336                  * We don't have to do any lookup to find ourselves.
 337                  */
 338                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 339                         /*
 340                          * Sampling just ourselves we can do with no locking.
 341                          */
 342                         error = cpu_clock_sample(which_clock,
 343                                                  current, &rtn);
 344                 } else {
 345                         read_lock(&tasklist_lock);
 346                         error = cpu_clock_sample_group(which_clock,
 347                                                        current, &rtn);
 348                         read_unlock(&tasklist_lock);
 349                 }
 350         } else {
 351                 /*
 352                  * Find the given PID, and validate that the caller
 353                  * should be able to see it.
 354                  */
 355                 struct task_struct *p;
 356                 rcu_read_lock();
 357                 p = find_task_by_vpid(pid);
 358                 if (p) {
 359                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 360                                 if (same_thread_group(p, current)) {
 361                                         error = cpu_clock_sample(which_clock,
 362                                                                  p, &rtn);
 363                                 }
 364                         } else {
 365                                 read_lock(&tasklist_lock);
 366                                 if (thread_group_leader(p) && p->signal) {
 367                                         error =
 368                                             cpu_clock_sample_group(which_clock,
 369                                                                    p, &rtn);
 370                                 }
 371                                 read_unlock(&tasklist_lock);
 372                         }
 373                 }
 374                 rcu_read_unlock();
 375         }
 376
 377         if (error)
 378                 return error;
 379         sample_to_timespec(which_clock, rtn, tp);
 380         return 0;
 381 }
 382
 383
 384 /*
 385  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 386  * This is called from sys_timer_create with the new timer already locked.
 387  */
 388 int posix_cpu_timer_create(struct k_itimer *new_timer)
 389 {
 390         int ret = 0;
 391         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 392         struct task_struct *p;
 393
 394         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 395                 return -EINVAL;
 396
 397         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 398         new_timer->it.cpu.incr.sched = 0;
 399         new_timer->it.cpu.expires.sched = 0;
 400
 401         read_lock(&tasklist_lock);
 402         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 403                 if (pid == 0) {
 404                         p = current;
 405                 } else {
 406                         p = find_task_by_vpid(pid);
 407                         if (p && !same_thread_group(p, current))
 408                                 p = NULL;
 409                 }
 410         } else {
 411                 if (pid == 0) {
 412                         p = current->group_leader;
 413                 } else {
 414                         p = find_task_by_vpid(pid);
 415                         if (p && !thread_group_leader(p))
 416                                 p = NULL;
 417                 }
 418         }
 419         new_timer->it.cpu.task = p;
 420         if (p) {
 421                 get_task_struct(p);
 422         } else {
 423                 ret = -EINVAL;
 424         }
 425         read_unlock(&tasklist_lock);
 426
 427         return ret;
 428 }
 429
 430 /*
 431  * Clean up a CPU-clock timer that is about to be destroyed.
 432  * This is called from timer deletion with the timer already locked.
 433  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 434  * and try again.  (This happens when the timer is in the middle of firing.)
 435  */
 436 int posix_cpu_timer_del(struct k_itimer *timer)
 437 {
 438         struct task_struct *p = timer->it.cpu.task;
 439         int ret = 0;
 440
 441         if (likely(p != NULL)) {
 442                 read_lock(&tasklist_lock);
 443                 if (unlikely(p->signal == NULL)) {
 444                         /*
 445                          * We raced with the reaping of the task.
 446                          * The deletion should have cleared us off the list.
 447                          */
 448                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 449                 } else {
 450                         spin_lock(&p->sighand->siglock);
 451                         if (timer->it.cpu.firing)
 452                                 ret = TIMER_RETRY;
 453                         else
 454                                 list_del(&timer->it.cpu.entry);
 455                         spin_unlock(&p->sighand->siglock);
 456                 }
 457                 read_unlock(&tasklist_lock);
 458
 459                 if (!ret)
 460                         put_task_struct(p);
 461         }
 462
 463         return ret;
 464 }
 465
 466 /*
 467  * Clean out CPU timers still ticking when a thread exited.  The task
 468  * pointer is cleared, and the expiry time is replaced with the residual
 469  * time for later timer_gettime calls to return.
 470  * This must be called with the siglock held.
 471  */
 472 static void cleanup_timers(struct list_head *head,
 473                            cputime_t utime, cputime_t stime,
 474                            unsigned long long sum_exec_runtime)
 475 {
 476         struct cpu_timer_list *timer, *next;
 477         cputime_t ptime = cputime_add(utime, stime);
 478
 479         list_for_each_entry_safe(timer, next, head, entry) {
 480                 list_del_init(&timer->entry);
 481                 if (cputime_lt(timer->expires.cpu, ptime)) {
 482                         timer->expires.cpu = cputime_zero;
 483                 } else {
 484                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 485                                                          ptime);
 486                 }
 487         }
 488
 489         ++head;
 490         list_for_each_entry_safe(timer, next, head, entry) {
 491                 list_del_init(&timer->entry);
 492                 if (cputime_lt(timer->expires.cpu, utime)) {
 493                         timer->expires.cpu = cputime_zero;
 494                 } else {
 495                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 496                                                          utime);
 497                 }
 498         }
 499
 500         ++head;
 501         list_for_each_entry_safe(timer, next, head, entry) {
 502                 list_del_init(&timer->entry);
 503                 if (timer->expires.sched < sum_exec_runtime) {
 504                         timer->expires.sched = 0;
 505                 } else {
 506                         timer->expires.sched -= sum_exec_runtime;
 507                 }
 508         }
 509 }
 510
 511 /*
 512  * These are both called with the siglock held, when the current thread
 513  * is being reaped.  When the final (leader) thread in the group is reaped,
 514  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 515  */
 516 void posix_cpu_timers_exit(struct task_struct *tsk)
 517 {
 518         cleanup_timers(tsk->cpu_timers,
 519                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 520
 521 }
 522 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 523 {
 524         struct signal_struct *const sig = tsk->signal;
 525
 526         cleanup_timers(tsk->signal->cpu_timers,
 527                        cputime_add(tsk->utime, sig->utime),
 528                        cputime_add(tsk->stime, sig->stime),
 529                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 530 }
 531
 532 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 533 {
 534         /*
 535          * That's all for this thread or process.
 536          * We leave our residual in expires to be reported.
 537          */
 538         put_task_struct(timer->it.cpu.task);
 539         timer->it.cpu.task = NULL;
 540         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 541                                              timer->it.cpu.expires,
 542                                              now);
 543 }
 544
 545 /*
 546  * Insert the timer on the appropriate list before any timers that
 547  * expire later.  This must be called with the tasklist_lock held
 548  * for reading, and interrupts disabled.
 549  */
 550 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 551 {
 552         struct task_struct *p = timer->it.cpu.task;
 553         struct list_head *head, *listpos;
 554         struct cpu_timer_list *const nt = &timer->it.cpu;
 555         struct cpu_timer_list *next;
 556         unsigned long i;
 557
 558         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 559                 p->cpu_timers : p->signal->cpu_timers);
 560         head += CPUCLOCK_WHICH(timer->it_clock);
 561
 562         BUG_ON(!irqs_disabled());
 563         spin_lock(&p->sighand->siglock);
 564
 565         listpos = head;
 566         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 567                 list_for_each_entry(next, head, entry) {
 568                         if (next->expires.sched > nt->expires.sched)
 569                                 break;
 570                         listpos = &next->entry;
 571                 }
 572         } else {
 573                 list_for_each_entry(next, head, entry) {
 574                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 575                                 break;
 576                         listpos = &next->entry;
 577                 }
 578         }
 579         list_add(&nt->entry, listpos);
 580
 581         if (listpos == head) {
 582                 /*
 583                  * We are the new earliest-expiring timer.
 584                  * If we are a thread timer, there can always
 585                  * be a process timer telling us to stop earlier.
 586                  */
 587
 588                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 589                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 590                         default:
 591                                 BUG();
 592                         case CPUCLOCK_PROF:
 593                                 if (cputime_eq(p->cputime_expires.prof_exp,
 594                                                cputime_zero) ||
 595                                     cputime_gt(p->cputime_expires.prof_exp,
 596                                                nt->expires.cpu))
 597                                         p->cputime_expires.prof_exp =
 598                                                 nt->expires.cpu;
 599                                 break;
 600                         case CPUCLOCK_VIRT:
 601                                 if (cputime_eq(p->cputime_expires.virt_exp,
 602                                                cputime_zero) ||
 603                                     cputime_gt(p->cputime_expires.virt_exp,
 604                                                nt->expires.cpu))
 605                                         p->cputime_expires.virt_exp =
 606                                                 nt->expires.cpu;
 607                                 break;
 608                         case CPUCLOCK_SCHED:
 609                                 if (p->cputime_expires.sched_exp == 0 ||
 610                                     p->cputime_expires.sched_exp >
 611                                                         nt->expires.sched)
 612                                         p->cputime_expires.sched_exp =
 613                                                 nt->expires.sched;
 614                                 break;
 615                         }
 616                 } else {
 617                         /*
 618                          * For a process timer, set the cached expiration time.
 619                          */
 620                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 621                         default:
 622                                 BUG();
 623                         case CPUCLOCK_VIRT:
 624                                 if (!cputime_eq(p->signal->it_virt_expires,
 625                                                 cputime_zero) &&
 626                                     cputime_lt(p->signal->it_virt_expires,
 627                                                timer->it.cpu.expires.cpu))
 628                                         break;
 629                                 p->signal->cputime_expires.virt_exp =
 630                                         timer->it.cpu.expires.cpu;
 631                                 break;
 632                         case CPUCLOCK_PROF:
 633                                 if (!cputime_eq(p->signal->it_prof_expires,
 634                                                 cputime_zero) &&
 635                                     cputime_lt(p->signal->it_prof_expires,
 636                                                timer->it.cpu.expires.cpu))
 637                                         break;
 638                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 639                                 if (i != RLIM_INFINITY &&
 640                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 641                                         break;
 642                                 p->signal->cputime_expires.prof_exp =
 643                                         timer->it.cpu.expires.cpu;
 644                                 break;
 645                         case CPUCLOCK_SCHED:
 646                                 p->signal->cputime_expires.sched_exp =
 647                                         timer->it.cpu.expires.sched;
 648                                 break;
 649                         }
 650                 }
 651         }
 652
 653         spin_unlock(&p->sighand->siglock);
 654 }
 655
 656 /*
 657  * The timer is locked, fire it and arrange for its reload.
 658  */
 659 static void cpu_timer_fire(struct k_itimer *timer)
 660 {
 661         if (unlikely(timer->sigq == NULL)) {
 662                 /*
 663                  * This a special case for clock_nanosleep,
 664                  * not a normal timer from sys_timer_create.
 665                  */
 666                 wake_up_process(timer->it_process);
 667                 timer->it.cpu.expires.sched = 0;
 668         } else if (timer->it.cpu.incr.sched == 0) {
 669                 /*
 670                  * One-shot timer.  Clear it as soon as it's fired.
 671                  */
 672                 posix_timer_event(timer, 0);
 673                 timer->it.cpu.expires.sched = 0;
 674         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 675                 /*
 676                  * The signal did not get queued because the signal
 677                  * was ignored, so we won't get any callback to
 678                  * reload the timer.  But we need to keep it
 679                  * ticking in case the signal is deliverable next time.
 680                  */
 681                 posix_cpu_timer_schedule(timer);
 682         }
 683 }
 684
 685 /*
 686  * Sample a process (thread group) timer for the given group_leader task.
 687  * Must be called with tasklist_lock held for reading.
 688  */
 689 static int cpu_timer_sample_group(const clockid_t which_clock,
 690                                   struct task_struct *p,
 691                                   union cpu_time_count *cpu)
 692 {
 693         struct task_cputime cputime;
 694
 695         thread_group_cputimer(p, &cputime);
 696         switch (CPUCLOCK_WHICH(which_clock)) {
 697         default:
 698                 return -EINVAL;
 699         case CPUCLOCK_PROF:
 700                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 701                 break;
 702         case CPUCLOCK_VIRT:
 703                 cpu->cpu = cputime.utime;
 704                 break;
 705         case CPUCLOCK_SCHED:
 706                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 707                 break;
 708         }
 709         return 0;
 710 }
 711
 712 /*
 713  * Guts of sys_timer_settime for CPU timers.
 714  * This is called with the timer locked and interrupts disabled.
 715  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 716  * and try again.  (This happens when the timer is in the middle of firing.)
 717  */
 718 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 719                         struct itimerspec *new, struct itimerspec *old)
 720 {
 721         struct task_struct *p = timer->it.cpu.task;
 722         union cpu_time_count old_expires, new_expires, val;
 723         int ret;
 724
 725         if (unlikely(p == NULL)) {
 726                 /*
 727                  * Timer refers to a dead task's clock.
 728                  */
 729                 return -ESRCH;
 730         }
 731
 732         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 733
 734         read_lock(&tasklist_lock);
 735         /*
 736          * We need the tasklist_lock to protect against reaping that
 737          * clears p->signal.  If p has just been reaped, we can no
 738          * longer get any information about it at all.
 739          */
 740         if (unlikely(p->signal == NULL)) {
 741                 read_unlock(&tasklist_lock);
 742                 put_task_struct(p);
 743                 timer->it.cpu.task = NULL;
 744                 return -ESRCH;
 745         }
 746
 747         /*
 748          * Disarm any old timer after extracting its expiry time.
 749          */
 750         BUG_ON(!irqs_disabled());
 751
 752         ret = 0;
 753         spin_lock(&p->sighand->siglock);
 754         old_expires = timer->it.cpu.expires;
 755         if (unlikely(timer->it.cpu.firing)) {
 756                 timer->it.cpu.firing = -1;
 757                 ret = TIMER_RETRY;
 758         } else
 759                 list_del_init(&timer->it.cpu.entry);
 760         spin_unlock(&p->sighand->siglock);
 761
 762         /*
 763          * We need to sample the current value to convert the new
 764          * value from to relative and absolute, and to convert the
 765          * old value from absolute to relative.  To set a process
 766          * timer, we need a sample to balance the thread expiry
 767          * times (in arm_timer).  With an absolute time, we must
 768          * check if it's already passed.  In short, we need a sample.
 769          */
 770         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 771                 cpu_clock_sample(timer->it_clock, p, &val);
 772         } else {
 773                 cpu_timer_sample_group(timer->it_clock, p, &val);
 774         }
 775
 776         if (old) {
 777                 if (old_expires.sched == 0) {
 778                         old->it_value.tv_sec = 0;
 779                         old->it_value.tv_nsec = 0;
 780                 } else {
 781                         /*
 782                          * Update the timer in case it has
 783                          * overrun already.  If it has,
 784                          * we'll report it as having overrun
 785                          * and with the next reloaded timer
 786                          * already ticking, though we are
 787                          * swallowing that pending
 788                          * notification here to install the
 789                          * new setting.
 790                          */
 791                         bump_cpu_timer(timer, val);
 792                         if (cpu_time_before(timer->it_clock, val,
 793                                             timer->it.cpu.expires)) {
 794                                 old_expires = cpu_time_sub(
 795                                         timer->it_clock,
 796                                         timer->it.cpu.expires, val);
 797                                 sample_to_timespec(timer->it_clock,
 798                                                    old_expires,
 799                                                    &old->it_value);
 800                         } else {
 801                                 old->it_value.tv_nsec = 1;
 802                                 old->it_value.tv_sec = 0;
 803                         }
 804                 }
 805         }
 806
 807         if (unlikely(ret)) {
 808                 /*
 809                  * We are colliding with the timer actually firing.
 810                  * Punt after filling in the timer's old value, and
 811                  * disable this firing since we are already reporting
 812                  * it as an overrun (thanks to bump_cpu_timer above).
 813                  */
 814                 read_unlock(&tasklist_lock);
 815                 goto out;
 816         }
 817
 818         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 819                 cpu_time_add(timer->it_clock, &new_expires, val);
 820         }
 821
 822         /*
 823          * Install the new expiry time (or zero).
 824          * For a timer with no notification action, we don't actually
 825          * arm the timer (we'll just fake it for timer_gettime).
 826          */
 827         timer->it.cpu.expires = new_expires;
 828         if (new_expires.sched != 0 &&
 829             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 830             cpu_time_before(timer->it_clock, val, new_expires)) {
 831                 arm_timer(timer, val);
 832         }
 833
 834         read_unlock(&tasklist_lock);
 835
 836         /*
 837          * Install the new reload setting, and
 838          * set up the signal and overrun bookkeeping.
 839          */
 840         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 841                                                 &new->it_interval);
 842
 843         /*
 844          * This acts as a modification timestamp for the timer,
 845          * so any automatic reload attempt will punt on seeing
 846          * that we have reset the timer manually.
 847          */
 848         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 849                 ~REQUEUE_PENDING;
 850         timer->it_overrun_last = 0;
 851         timer->it_overrun = -1;
 852
 853         if (new_expires.sched != 0 &&
 854             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 855             !cpu_time_before(timer->it_clock, val, new_expires)) {
 856                 /*
 857                  * The designated time already passed, so we notify
 858                  * immediately, even if the thread never runs to
 859                  * accumulate more time on this clock.
 860                  */
 861                 cpu_timer_fire(timer);
 862         }
 863
 864         ret = 0;
 865  out:
 866         if (old) {
 867                 sample_to_timespec(timer->it_clock,
 868                                    timer->it.cpu.incr, &old->it_interval);
 869         }
 870         return ret;
 871 }
 872
 873 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 874 {
 875         union cpu_time_count now;
 876         struct task_struct *p = timer->it.cpu.task;
 877         int clear_dead;
 878
 879         /*
 880          * Easy part: convert the reload time.
 881          */
 882         sample_to_timespec(timer->it_clock,
 883                            timer->it.cpu.incr, &itp->it_interval);
 884
 885         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 886                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 887                 return;
 888         }
 889
 890         if (unlikely(p == NULL)) {
 891                 /*
 892                  * This task already died and the timer will never fire.
 893                  * In this case, expires is actually the dead value.
 894                  */
 895         dead:
 896                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 897                                    &itp->it_value);
 898                 return;
 899         }
 900
 901         /*
 902          * Sample the clock to take the difference with the expiry time.
 903          */
 904         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 905                 cpu_clock_sample(timer->it_clock, p, &now);
 906                 clear_dead = p->exit_state;
 907         } else {
 908                 read_lock(&tasklist_lock);
 909                 if (unlikely(p->signal == NULL)) {
 910                         /*
 911                          * The process has been reaped.
 912                          * We can't even collect a sample any more.
 913                          * Call the timer disarmed, nothing else to do.
 914                          */
 915                         put_task_struct(p);
 916                         timer->it.cpu.task = NULL;
 917                         timer->it.cpu.expires.sched = 0;
 918                         read_unlock(&tasklist_lock);
 919                         goto dead;
 920                 } else {
 921                         cpu_timer_sample_group(timer->it_clock, p, &now);
 922                         clear_dead = (unlikely(p->exit_state) &&
 923                                       thread_group_empty(p));
 924                 }
 925                 read_unlock(&tasklist_lock);
 926         }
 927
 928         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 929                 if (timer->it.cpu.incr.sched == 0 &&
 930                     cpu_time_before(timer->it_clock,
 931                                     timer->it.cpu.expires, now)) {
 932                         /*
 933                          * Do-nothing timer expired and has no reload,
 934                          * so it's as if it was never set.
 935                          */
 936                         timer->it.cpu.expires.sched = 0;
 937                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 938                         return;
 939                 }
 940                 /*
 941                  * Account for any expirations and reloads that should
 942                  * have happened.
 943                  */
 944                 bump_cpu_timer(timer, now);
 945         }
 946
 947         if (unlikely(clear_dead)) {
 948                 /*
 949                  * We've noticed that the thread is dead, but
 950                  * not yet reaped.  Take this opportunity to
 951                  * drop our task ref.
 952                  */
 953                 clear_dead_task(timer, now);
 954                 goto dead;
 955         }
 956
 957         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 958                 sample_to_timespec(timer->it_clock,
 959                                    cpu_time_sub(timer->it_clock,
 960                                                 timer->it.cpu.expires, now),
 961                                    &itp->it_value);
 962         } else {
 963                 /*
 964                  * The timer should have expired already, but the firing
 965                  * hasn't taken place yet.  Say it's just about to expire.
 966                  */
 967                 itp->it_value.tv_nsec = 1;
 968                 itp->it_value.tv_sec = 0;
 969         }
 970 }
 971
 972 /*
 973  * Check for any per-thread CPU timers that have fired and move them off
 974  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 975  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 976  */
 977 static void check_thread_timers(struct task_struct *tsk,
 978                                 struct list_head *firing)
 979 {
 980         int maxfire;
 981         struct list_head *timers = tsk->cpu_timers;
 982         struct signal_struct *const sig = tsk->signal;
 983
 984         maxfire = 20;
 985         tsk->cputime_expires.prof_exp = cputime_zero;
 986         while (!list_empty(timers)) {
 987                 struct cpu_timer_list *t = list_first_entry(timers,
 988                                                       struct cpu_timer_list,
 989                                                       entry);
 990                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 991                         tsk->cputime_expires.prof_exp = 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->cputime_expires.virt_exp = cputime_zero;
1001         while (!list_empty(timers)) {
1002                 struct cpu_timer_list *t = list_first_entry(timers,
1003                                                       struct cpu_timer_list,
1004                                                       entry);
1005                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1006                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1007                         break;
1008                 }
1009                 t->firing = 1;
1010                 list_move_tail(&t->entry, firing);
1011         }
1012
1013         ++timers;
1014         maxfire = 20;
1015         tsk->cputime_expires.sched_exp = 0;
1016         while (!list_empty(timers)) {
1017                 struct cpu_timer_list *t = list_first_entry(timers,
1018                                                       struct cpu_timer_list,
1019                                                       entry);
1020                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1021                         tsk->cputime_expires.sched_exp = t->expires.sched;
1022                         break;
1023                 }
1024                 t->firing = 1;
1025                 list_move_tail(&t->entry, firing);
1026         }
1027
1028         /*
1029          * Check for the special case thread timers.
1030          */
1031         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1032                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1033                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1034
1035                 if (hard != RLIM_INFINITY &&
1036                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1037                         /*
1038                          * At the hard limit, we just die.
1039                          * No need to calculate anything else now.
1040                          */
1041                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1042                         return;
1043                 }
1044                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1045                         /*
1046                          * At the soft limit, send a SIGXCPU every second.
1047                          */
1048                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1049                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1050                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1051                                                                 USEC_PER_SEC;
1052                         }
1053                         printk(KERN_INFO
1054                                 "RT Watchdog Timeout: %s[%d]\n",
1055                                 tsk->comm, task_pid_nr(tsk));
1056                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1057                 }
1058         }
1059 }
1060
1061 static void stop_process_timers(struct task_struct *tsk)
1062 {
1063         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1064         unsigned long flags;
1065
1066         if (!cputimer->running)
1067                 return;
1068
1069         spin_lock_irqsave(&cputimer->lock, flags);
1070         cputimer->running = 0;
1071         spin_unlock_irqrestore(&cputimer->lock, flags);
1072 }
1073
1074 /*
1075  * Check for any per-thread CPU timers that have fired and move them
1076  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1077  * have already been taken off.
1078  */
1079 static void check_process_timers(struct task_struct *tsk,
1080                                  struct list_head *firing)
1081 {
1082         int maxfire;
1083         struct signal_struct *const sig = tsk->signal;
1084         cputime_t utime, ptime, virt_expires, prof_expires;
1085         unsigned long long sum_sched_runtime, sched_expires;
1086         struct list_head *timers = sig->cpu_timers;
1087         struct task_cputime cputime;
1088
1089         /*
1090          * Don't sample the current process CPU clocks if there are no timers.
1091          */
1092         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1093             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1094             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1095             list_empty(&timers[CPUCLOCK_VIRT]) &&
1096             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1097             list_empty(&timers[CPUCLOCK_SCHED])) {
1098                 stop_process_timers(tsk);
1099                 return;
1100         }
1101
1102         /*
1103          * Collect the current process totals.
1104          */
1105         thread_group_cputimer(tsk, &cputime);
1106         utime = cputime.utime;
1107         ptime = cputime_add(utime, cputime.stime);
1108         sum_sched_runtime = cputime.sum_exec_runtime;
1109         maxfire = 20;
1110         prof_expires = cputime_zero;
1111         while (!list_empty(timers)) {
1112                 struct cpu_timer_list *tl = list_first_entry(timers,
1113                                                       struct cpu_timer_list,
1114                                                       entry);
1115                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1116                         prof_expires = tl->expires.cpu;
1117                         break;
1118                 }
1119                 tl->firing = 1;
1120                 list_move_tail(&tl->entry, firing);
1121         }
1122
1123         ++timers;
1124         maxfire = 20;
1125         virt_expires = cputime_zero;
1126         while (!list_empty(timers)) {
1127                 struct cpu_timer_list *tl = list_first_entry(timers,
1128                                                       struct cpu_timer_list,
1129                                                       entry);
1130                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1131                         virt_expires = tl->expires.cpu;
1132                         break;
1133                 }
1134                 tl->firing = 1;
1135                 list_move_tail(&tl->entry, firing);
1136         }
1137
1138         ++timers;
1139         maxfire = 20;
1140         sched_expires = 0;
1141         while (!list_empty(timers)) {
1142                 struct cpu_timer_list *tl = list_first_entry(timers,
1143                                                       struct cpu_timer_list,
1144                                                       entry);
1145                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1146                         sched_expires = tl->expires.sched;
1147                         break;
1148                 }
1149                 tl->firing = 1;
1150                 list_move_tail(&tl->entry, firing);
1151         }
1152
1153         /*
1154          * Check for the special case process timers.
1155          */
1156         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1157                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1158                         /* ITIMER_PROF fires and reloads.  */
1159                         sig->it_prof_expires = sig->it_prof_incr;
1160                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1161                                 sig->it_prof_expires = cputime_add(
1162                                         sig->it_prof_expires, ptime);
1163                         }
1164                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1165                 }
1166                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1167                     (cputime_eq(prof_expires, cputime_zero) ||
1168                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1169                         prof_expires = sig->it_prof_expires;
1170                 }
1171         }
1172         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1173                 if (cputime_ge(utime, sig->it_virt_expires)) {
1174                         /* ITIMER_VIRTUAL fires and reloads.  */
1175                         sig->it_virt_expires = sig->it_virt_incr;
1176                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1177                                 sig->it_virt_expires = cputime_add(
1178                                         sig->it_virt_expires, utime);
1179                         }
1180                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1181                 }
1182                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1183                     (cputime_eq(virt_expires, cputime_zero) ||
1184                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1185                         virt_expires = sig->it_virt_expires;
1186                 }
1187         }
1188         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1189                 unsigned long psecs = cputime_to_secs(ptime);
1190                 cputime_t x;
1191                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1192                         /*
1193                          * At the hard limit, we just die.
1194                          * No need to calculate anything else now.
1195                          */
1196                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1197                         return;
1198                 }
1199                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1200                         /*
1201                          * At the soft limit, send a SIGXCPU every second.
1202                          */
1203                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1204                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1205                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1206                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1207                         }
1208                 }
1209                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1210                 if (cputime_eq(prof_expires, cputime_zero) ||
1211                     cputime_lt(x, prof_expires)) {
1212                         prof_expires = x;
1213                 }
1214         }
1215
1216         if (!cputime_eq(prof_expires, cputime_zero) &&
1217             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1218              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1219                 sig->cputime_expires.prof_exp = prof_expires;
1220         if (!cputime_eq(virt_expires, cputime_zero) &&
1221             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1222              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1223                 sig->cputime_expires.virt_exp = virt_expires;
1224         if (sched_expires != 0 &&
1225             (sig->cputime_expires.sched_exp == 0 ||
1226              sig->cputime_expires.sched_exp > sched_expires))
1227                 sig->cputime_expires.sched_exp = sched_expires;
1228 }
1229
1230 /*
1231  * This is called from the signal code (via do_schedule_next_timer)
1232  * when the last timer signal was delivered and we have to reload the timer.
1233  */
1234 void posix_cpu_timer_schedule(struct k_itimer *timer)
1235 {
1236         struct task_struct *p = timer->it.cpu.task;
1237         union cpu_time_count now;
1238
1239         if (unlikely(p == NULL))
1240                 /*
1241                  * The task was cleaned up already, no future firings.
1242                  */
1243                 goto out;
1244
1245         /*
1246          * Fetch the current sample and update the timer's expiry time.
1247          */
1248         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1249                 cpu_clock_sample(timer->it_clock, p, &now);
1250                 bump_cpu_timer(timer, now);
1251                 if (unlikely(p->exit_state)) {
1252                         clear_dead_task(timer, now);
1253                         goto out;
1254                 }
1255                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1256         } else {
1257                 read_lock(&tasklist_lock);
1258                 if (unlikely(p->signal == NULL)) {
1259                         /*
1260                          * The process has been reaped.
1261                          * We can't even collect a sample any more.
1262                          */
1263                         put_task_struct(p);
1264                         timer->it.cpu.task = p = NULL;
1265                         timer->it.cpu.expires.sched = 0;
1266                         goto out_unlock;
1267                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1268                         /*
1269                          * We've noticed that the thread is dead, but
1270                          * not yet reaped.  Take this opportunity to
1271                          * drop our task ref.
1272                          */
1273                         clear_dead_task(timer, now);
1274                         goto out_unlock;
1275                 }
1276                 cpu_timer_sample_group(timer->it_clock, p, &now);
1277                 bump_cpu_timer(timer, now);
1278                 /* Leave the tasklist_lock locked for the call below.  */
1279         }
1280
1281         /*
1282          * Now re-arm for the new expiry time.
1283          */
1284         arm_timer(timer, now);
1285
1286 out_unlock:
1287         read_unlock(&tasklist_lock);
1288
1289 out:
1290         timer->it_overrun_last = timer->it_overrun;
1291         timer->it_overrun = -1;
1292         ++timer->it_requeue_pending;
1293 }
1294
1295 /**
1296  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1297  *
1298  * @cputime:    The struct to compare.
1299  *
1300  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1301  * are zero, false if any field is nonzero.
1302  */
1303 static inline int task_cputime_zero(const struct task_cputime *cputime)
1304 {
1305         if (cputime_eq(cputime->utime, cputime_zero) &&
1306             cputime_eq(cputime->stime, cputime_zero) &&
1307             cputime->sum_exec_runtime == 0)
1308                 return 1;
1309         return 0;
1310 }
1311
1312 /**
1313  * task_cputime_expired - Compare two task_cputime entities.
1314  *
1315  * @sample:     The task_cputime structure to be checked for expiration.
1316  * @expires:    Expiration times, against which @sample will be checked.
1317  *
1318  * Checks @sample against @expires to see if any field of @sample has expired.
1319  * Returns true if any field of the former is greater than the corresponding
1320  * field of the latter if the latter field is set.  Otherwise returns false.
1321  */
1322 static inline int task_cputime_expired(const struct task_cputime *sample,
1323                                         const struct task_cputime *expires)
1324 {
1325         if (!cputime_eq(expires->utime, cputime_zero) &&
1326             cputime_ge(sample->utime, expires->utime))
1327                 return 1;
1328         if (!cputime_eq(expires->stime, cputime_zero) &&
1329             cputime_ge(cputime_add(sample->utime, sample->stime),
1330                        expires->stime))
1331                 return 1;
1332         if (expires->sum_exec_runtime != 0 &&
1333             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1334                 return 1;
1335         return 0;
1336 }
1337
1338 /**
1339  * fastpath_timer_check - POSIX CPU timers fast path.
1340  *
1341  * @tsk:        The task (thread) being checked.
1342  *
1343  * Check the task and thread group timers.  If both are zero (there are no
1344  * timers set) return false.  Otherwise snapshot the task and thread group
1345  * timers and compare them with the corresponding expiration times.  Return
1346  * true if a timer has expired, else return false.
1347  */
1348 static inline int fastpath_timer_check(struct task_struct *tsk)
1349 {
1350         struct signal_struct *sig;
1351
1352         /* tsk == current, ensure it is safe to use ->signal/sighand */
1353         if (unlikely(tsk->exit_state))
1354                 return 0;
1355
1356         if (!task_cputime_zero(&tsk->cputime_expires)) {
1357                 struct task_cputime task_sample = {
1358                         .utime = tsk->utime,
1359                         .stime = tsk->stime,
1360                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1361                 };
1362
1363                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1364                         return 1;
1365         }
1366
1367         sig = tsk->signal;
1368         if (!task_cputime_zero(&sig->cputime_expires)) {
1369                 struct task_cputime group_sample;
1370
1371                 thread_group_cputimer(tsk, &group_sample);
1372                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1373                         return 1;
1374         }
1375
1376         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1377 }
1378
1379 /*
1380  * This is called from the timer interrupt handler.  The irq handler has
1381  * already updated our counts.  We need to check if any timers fire now.
1382  * Interrupts are disabled.
1383  */
1384 void run_posix_cpu_timers(struct task_struct *tsk)
1385 {
1386         LIST_HEAD(firing);
1387         struct k_itimer *timer, *next;
1388
1389         BUG_ON(!irqs_disabled());
1390
1391         /*
1392          * The fast path checks that there are no expired thread or thread
1393          * group timers.  If that's so, just return.
1394          */
1395         if (!fastpath_timer_check(tsk))
1396                 return;
1397
1398         spin_lock(&tsk->sighand->siglock);
1399         /*
1400          * Here we take off tsk->signal->cpu_timers[N] and
1401          * tsk->cpu_timers[N] all the timers that are firing, and
1402          * put them on the firing list.
1403          */
1404         check_thread_timers(tsk, &firing);
1405         check_process_timers(tsk, &firing);
1406
1407         /*
1408          * We must release these locks before taking any timer's lock.
1409          * There is a potential race with timer deletion here, as the
1410          * siglock now protects our private firing list.  We have set
1411          * the firing flag in each timer, so that a deletion attempt
1412          * that gets the timer lock before we do will give it up and
1413          * spin until we've taken care of that timer below.
1414          */
1415         spin_unlock(&tsk->sighand->siglock);
1416
1417         /*
1418          * Now that all the timers on our list have the firing flag,
1419          * noone will touch their list entries but us.  We'll take
1420          * each timer's lock before clearing its firing flag, so no
1421          * timer call will interfere.
1422          */
1423         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1424                 int cpu_firing;
1425
1426                 spin_lock(&timer->it_lock);
1427                 list_del_init(&timer->it.cpu.entry);
1428                 cpu_firing = timer->it.cpu.firing;
1429                 timer->it.cpu.firing = 0;
1430                 /*
1431                  * The firing flag is -1 if we collided with a reset
1432                  * of the timer, which already reported this
1433                  * almost-firing as an overrun.  So don't generate an event.
1434                  */
1435                 if (likely(cpu_firing >= 0))
1436                         cpu_timer_fire(timer);
1437                 spin_unlock(&timer->it_lock);
1438         }
1439 }
1440
1441 /*
1442  * Set one of the process-wide special case CPU timers.
1443  * The tsk->sighand->siglock must be held by the caller.
1444  * The *newval argument is relative and we update it to be absolute, *oldval
1445  * is absolute and we update it to be relative.
1446  */
1447 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1448                            cputime_t *newval, cputime_t *oldval)
1449 {
1450         union cpu_time_count now;
1451         struct list_head *head;
1452
1453         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1454         cpu_timer_sample_group(clock_idx, tsk, &now);
1455
1456         if (oldval) {
1457                 if (!cputime_eq(*oldval, cputime_zero)) {
1458                         if (cputime_le(*oldval, now.cpu)) {
1459                                 /* Just about to fire. */
1460                                 *oldval = jiffies_to_cputime(1);
1461                         } else {
1462                                 *oldval = cputime_sub(*oldval, now.cpu);
1463                         }
1464                 }
1465
1466                 if (cputime_eq(*newval, cputime_zero))
1467                         return;
1468                 *newval = cputime_add(*newval, now.cpu);
1469
1470                 /*
1471                  * If the RLIMIT_CPU timer will expire before the
1472                  * ITIMER_PROF timer, we have nothing else to do.
1473                  */
1474                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1475                     < cputime_to_secs(*newval))
1476                         return;
1477         }
1478
1479         /*
1480          * Check whether there are any process timers already set to fire
1481          * before this one.  If so, we don't have anything more to do.
1482          */
1483         head = &tsk->signal->cpu_timers[clock_idx];
1484         if (list_empty(head) ||
1485             cputime_ge(list_first_entry(head,
1486                                   struct cpu_timer_list, entry)->expires.cpu,
1487                        *newval)) {
1488                 switch (clock_idx) {
1489                 case CPUCLOCK_PROF:
1490                         tsk->signal->cputime_expires.prof_exp = *newval;
1491                         break;
1492                 case CPUCLOCK_VIRT:
1493                         tsk->signal->cputime_expires.virt_exp = *newval;
1494                         break;
1495                 }
1496         }
1497 }
1498
1499 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1500                             struct timespec *rqtp, struct itimerspec *it)
1501 {
1502         struct k_itimer timer;
1503         int error;
1504
1505         /*
1506          * Set up a temporary timer and then wait for it to go off.
1507          */
1508         memset(&timer, 0, sizeof timer);
1509         spin_lock_init(&timer.it_lock);
1510         timer.it_clock = which_clock;
1511         timer.it_overrun = -1;
1512         error = posix_cpu_timer_create(&timer);
1513         timer.it_process = current;
1514         if (!error) {
1515                 static struct itimerspec zero_it;
1516
1517                 memset(it, 0, sizeof *it);
1518                 it->it_value = *rqtp;
1519
1520                 spin_lock_irq(&timer.it_lock);
1521                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1522                 if (error) {
1523                         spin_unlock_irq(&timer.it_lock);
1524                         return error;
1525                 }
1526
1527                 while (!signal_pending(current)) {
1528                         if (timer.it.cpu.expires.sched == 0) {
1529                                 /*
1530                                  * Our timer fired and was reset.
1531                                  */
1532                                 spin_unlock_irq(&timer.it_lock);
1533                                 return 0;
1534                         }
1535
1536                         /*
1537                          * Block until cpu_timer_fire (or a signal) wakes us.
1538                          */
1539                         __set_current_state(TASK_INTERRUPTIBLE);
1540                         spin_unlock_irq(&timer.it_lock);
1541                         schedule();
1542                         spin_lock_irq(&timer.it_lock);
1543                 }
1544
1545                 /*
1546                  * We were interrupted by a signal.
1547                  */
1548                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1549                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1550                 spin_unlock_irq(&timer.it_lock);
1551
1552                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1553                         /*
1554                          * It actually did fire already.
1555                          */
1556                         return 0;
1557                 }
1558
1559                 error = -ERESTART_RESTARTBLOCK;
1560         }
1561
1562         return error;
1563 }
1564
1565 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1566                      struct timespec *rqtp, struct timespec __user *rmtp)
1567 {
1568         struct restart_block *restart_block =
1569             &current_thread_info()->restart_block;
1570         struct itimerspec it;
1571         int error;
1572
1573         /*
1574          * Diagnose required errors first.
1575          */
1576         if (CPUCLOCK_PERTHREAD(which_clock) &&
1577             (CPUCLOCK_PID(which_clock) == 0 ||
1578              CPUCLOCK_PID(which_clock) == current->pid))
1579                 return -EINVAL;
1580
1581         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1582
1583         if (error == -ERESTART_RESTARTBLOCK) {
1584
1585                 if (flags & TIMER_ABSTIME)
1586                         return -ERESTARTNOHAND;
1587                 /*
1588                  * Report back to the user the time still remaining.
1589                  */
1590                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1591                         return -EFAULT;
1592
1593                 restart_block->fn = posix_cpu_nsleep_restart;
1594                 restart_block->arg0 = which_clock;
1595                 restart_block->arg1 = (unsigned long) rmtp;
1596                 restart_block->arg2 = rqtp->tv_sec;
1597                 restart_block->arg3 = rqtp->tv_nsec;
1598         }
1599         return error;
1600 }
1601
1602 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1603 {
1604         clockid_t which_clock = restart_block->arg0;
1605         struct timespec __user *rmtp;
1606         struct timespec t;
1607         struct itimerspec it;
1608         int error;
1609
1610         rmtp = (struct timespec __user *) restart_block->arg1;
1611         t.tv_sec = restart_block->arg2;
1612         t.tv_nsec = restart_block->arg3;
1613
1614         restart_block->fn = do_no_restart_syscall;
1615         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1616
1617         if (error == -ERESTART_RESTARTBLOCK) {
1618                 /*
1619                  * Report back to the user the time still remaining.
1620                  */
1621                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1622                         return -EFAULT;
1623
1624                 restart_block->fn = posix_cpu_nsleep_restart;
1625                 restart_block->arg0 = which_clock;
1626                 restart_block->arg1 = (unsigned long) rmtp;
1627                 restart_block->arg2 = t.tv_sec;
1628                 restart_block->arg3 = t.tv_nsec;
1629         }
1630         return error;
1631
1632 }
1633
1634
1635 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1636 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1637
1638 static int process_cpu_clock_getres(const clockid_t which_clock,
1639                                     struct timespec *tp)
1640 {
1641         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1642 }
1643 static int process_cpu_clock_get(const clockid_t which_clock,
1644                                  struct timespec *tp)
1645 {
1646         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1647 }
1648 static int process_cpu_timer_create(struct k_itimer *timer)
1649 {
1650         timer->it_clock = PROCESS_CLOCK;
1651         return posix_cpu_timer_create(timer);
1652 }
1653 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1654                               struct timespec *rqtp,
1655                               struct timespec __user *rmtp)
1656 {
1657         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1658 }
1659 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1660 {
1661         return -EINVAL;
1662 }
1663 static int thread_cpu_clock_getres(const clockid_t which_clock,
1664                                    struct timespec *tp)
1665 {
1666         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1667 }
1668 static int thread_cpu_clock_get(const clockid_t which_clock,
1669                                 struct timespec *tp)
1670 {
1671         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1672 }
1673 static int thread_cpu_timer_create(struct k_itimer *timer)
1674 {
1675         timer->it_clock = THREAD_CLOCK;
1676         return posix_cpu_timer_create(timer);
1677 }
1678 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1679                               struct timespec *rqtp, struct timespec __user *rmtp)
1680 {
1681         return -EINVAL;
1682 }
1683 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1684 {
1685         return -EINVAL;
1686 }
1687
1688 static __init int init_posix_cpu_timers(void)
1689 {
1690         struct k_clock process = {
1691                 .clock_getres = process_cpu_clock_getres,
1692                 .clock_get = process_cpu_clock_get,
1693                 .clock_set = do_posix_clock_nosettime,
1694                 .timer_create = process_cpu_timer_create,
1695                 .nsleep = process_cpu_nsleep,
1696                 .nsleep_restart = process_cpu_nsleep_restart,
1697         };
1698         struct k_clock thread = {
1699                 .clock_getres = thread_cpu_clock_getres,
1700                 .clock_get = thread_cpu_clock_get,
1701                 .clock_set = do_posix_clock_nosettime,
1702                 .timer_create = thread_cpu_timer_create,
1703                 .nsleep = thread_cpu_nsleep,
1704                 .nsleep_restart = thread_cpu_nsleep_restart,
1705         };
1706
1707         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1708         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1709
1710         return 0;
1711 }
1712 __initcall(init_posix_cpu_timers);