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 #include <trace/events/timer.h>
  12
  13 /*
  14  * Called after updating RLIMIT_CPU to run cpu timer and update
  15  * tsk->signal->cputime_expires expiration cache if necessary. Needs
  16  * siglock protection since other code may update expiration cache as
  17  * well.
  18  */
  19 void update_rlimit_cpu(unsigned long rlim_new)
  20 {
  21         cputime_t cputime = secs_to_cputime(rlim_new);
  22
  23         spin_lock_irq(&current->sighand->siglock);
  24         set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
  25         spin_unlock_irq(&current->sighand->siglock);
  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() and do_cpu_nanosleep() with the
 387  * new timer already all-zeros initialized.
 388  */
 389 int posix_cpu_timer_create(struct k_itimer *new_timer)
 390 {
 391         int ret = 0;
 392         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 393         struct task_struct *p;
 394
 395         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 396                 return -EINVAL;
 397
 398         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 399
 400         read_lock(&tasklist_lock);
 401         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 402                 if (pid == 0) {
 403                         p = current;
 404                 } else {
 405                         p = find_task_by_vpid(pid);
 406                         if (p && !same_thread_group(p, current))
 407                                 p = NULL;
 408                 }
 409         } else {
 410                 if (pid == 0) {
 411                         p = current->group_leader;
 412                 } else {
 413                         p = find_task_by_vpid(pid);
 414                         if (p && !thread_group_leader(p))
 415                                 p = NULL;
 416                 }
 417         }
 418         new_timer->it.cpu.task = p;
 419         if (p) {
 420                 get_task_struct(p);
 421         } else {
 422                 ret = -EINVAL;
 423         }
 424         read_unlock(&tasklist_lock);
 425
 426         return ret;
 427 }
 428
 429 /*
 430  * Clean up a CPU-clock timer that is about to be destroyed.
 431  * This is called from timer deletion with the timer already locked.
 432  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 433  * and try again.  (This happens when the timer is in the middle of firing.)
 434  */
 435 int posix_cpu_timer_del(struct k_itimer *timer)
 436 {
 437         struct task_struct *p = timer->it.cpu.task;
 438         int ret = 0;
 439
 440         if (likely(p != NULL)) {
 441                 read_lock(&tasklist_lock);
 442                 if (unlikely(p->signal == NULL)) {
 443                         /*
 444                          * We raced with the reaping of the task.
 445                          * The deletion should have cleared us off the list.
 446                          */
 447                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 448                 } else {
 449                         spin_lock(&p->sighand->siglock);
 450                         if (timer->it.cpu.firing)
 451                                 ret = TIMER_RETRY;
 452                         else
 453                                 list_del(&timer->it.cpu.entry);
 454                         spin_unlock(&p->sighand->siglock);
 455                 }
 456                 read_unlock(&tasklist_lock);
 457
 458                 if (!ret)
 459                         put_task_struct(p);
 460         }
 461
 462         return ret;
 463 }
 464
 465 /*
 466  * Clean out CPU timers still ticking when a thread exited.  The task
 467  * pointer is cleared, and the expiry time is replaced with the residual
 468  * time for later timer_gettime calls to return.
 469  * This must be called with the siglock held.
 470  */
 471 static void cleanup_timers(struct list_head *head,
 472                            cputime_t utime, cputime_t stime,
 473                            unsigned long long sum_exec_runtime)
 474 {
 475         struct cpu_timer_list *timer, *next;
 476         cputime_t ptime = cputime_add(utime, stime);
 477
 478         list_for_each_entry_safe(timer, next, head, entry) {
 479                 list_del_init(&timer->entry);
 480                 if (cputime_lt(timer->expires.cpu, ptime)) {
 481                         timer->expires.cpu = cputime_zero;
 482                 } else {
 483                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 484                                                          ptime);
 485                 }
 486         }
 487
 488         ++head;
 489         list_for_each_entry_safe(timer, next, head, entry) {
 490                 list_del_init(&timer->entry);
 491                 if (cputime_lt(timer->expires.cpu, utime)) {
 492                         timer->expires.cpu = cputime_zero;
 493                 } else {
 494                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 495                                                          utime);
 496                 }
 497         }
 498
 499         ++head;
 500         list_for_each_entry_safe(timer, next, head, entry) {
 501                 list_del_init(&timer->entry);
 502                 if (timer->expires.sched < sum_exec_runtime) {
 503                         timer->expires.sched = 0;
 504                 } else {
 505                         timer->expires.sched -= sum_exec_runtime;
 506                 }
 507         }
 508 }
 509
 510 /*
 511  * These are both called with the siglock held, when the current thread
 512  * is being reaped.  When the final (leader) thread in the group is reaped,
 513  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 514  */
 515 void posix_cpu_timers_exit(struct task_struct *tsk)
 516 {
 517         cleanup_timers(tsk->cpu_timers,
 518                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 519
 520 }
 521 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 522 {
 523         struct signal_struct *const sig = tsk->signal;
 524
 525         cleanup_timers(tsk->signal->cpu_timers,
 526                        cputime_add(tsk->utime, sig->utime),
 527                        cputime_add(tsk->stime, sig->stime),
 528                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 529 }
 530
 531 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 532 {
 533         /*
 534          * That's all for this thread or process.
 535          * We leave our residual in expires to be reported.
 536          */
 537         put_task_struct(timer->it.cpu.task);
 538         timer->it.cpu.task = NULL;
 539         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 540                                              timer->it.cpu.expires,
 541                                              now);
 542 }
 543
 544 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 545 {
 546         return cputime_eq(expires, cputime_zero) ||
 547                cputime_gt(expires, new_exp);
 548 }
 549
 550 static inline int expires_le(cputime_t expires, cputime_t new_exp)
 551 {
 552         return !cputime_eq(expires, cputime_zero) &&
 553                cputime_le(expires, new_exp);
 554 }
 555 /*
 556  * Insert the timer on the appropriate list before any timers that
 557  * expire later.  This must be called with the tasklist_lock held
 558  * for reading, and interrupts disabled.
 559  */
 560 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 561 {
 562         struct task_struct *p = timer->it.cpu.task;
 563         struct list_head *head, *listpos;
 564         struct cpu_timer_list *const nt = &timer->it.cpu;
 565         struct cpu_timer_list *next;
 566
 567         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 568                 p->cpu_timers : p->signal->cpu_timers);
 569         head += CPUCLOCK_WHICH(timer->it_clock);
 570
 571         BUG_ON(!irqs_disabled());
 572         spin_lock(&p->sighand->siglock);
 573
 574         listpos = head;
 575         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 576                 list_for_each_entry(next, head, entry) {
 577                         if (next->expires.sched > nt->expires.sched)
 578                                 break;
 579                         listpos = &next->entry;
 580                 }
 581         } else {
 582                 list_for_each_entry(next, head, entry) {
 583                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 584                                 break;
 585                         listpos = &next->entry;
 586                 }
 587         }
 588         list_add(&nt->entry, listpos);
 589
 590         if (listpos == head) {
 591                 /*
 592                  * We are the new earliest-expiring timer.
 593                  * If we are a thread timer, there can always
 594                  * be a process timer telling us to stop earlier.
 595                  */
 596
 597                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 598                         union cpu_time_count *exp = &nt->expires;
 599
 600                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 601                         default:
 602                                 BUG();
 603                         case CPUCLOCK_PROF:
 604                                 if (expires_gt(p->cputime_expires.prof_exp,
 605                                                exp->cpu))
 606                                         p->cputime_expires.prof_exp = exp->cpu;
 607                                 break;
 608                         case CPUCLOCK_VIRT:
 609                                 if (expires_gt(p->cputime_expires.virt_exp,
 610                                                exp->cpu))
 611                                         p->cputime_expires.virt_exp = exp->cpu;
 612                                 break;
 613                         case CPUCLOCK_SCHED:
 614                                 if (p->cputime_expires.sched_exp == 0 ||
 615                                     p->cputime_expires.sched_exp > exp->sched)
 616                                         p->cputime_expires.sched_exp =
 617                                                                 exp->sched;
 618                                 break;
 619                         }
 620                 } else {
 621                         struct signal_struct *const sig = p->signal;
 622                         union cpu_time_count *exp = &timer->it.cpu.expires;
 623
 624                         /*
 625                          * For a process timer, set the cached expiration time.
 626                          */
 627                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 628                         default:
 629                                 BUG();
 630                         case CPUCLOCK_VIRT:
 631                                 if (expires_gt(sig->cputime_expires.virt_exp, exp->cpu))
 632                                         sig->cputime_expires.virt_exp = exp->cpu;
 633                         case CPUCLOCK_PROF:
 634                                 if (expires_gt(sig->cputime_expires.prof_exp, exp->cpu))
 635                                         sig->cputime_expires.prof_exp = exp->cpu;
 636                                 break;
 637                         case CPUCLOCK_SCHED:
 638                                 sig->cputime_expires.sched_exp = exp->sched;
 639                                 break;
 640                         }
 641                 }
 642         }
 643
 644         spin_unlock(&p->sighand->siglock);
 645 }
 646
 647 /*
 648  * The timer is locked, fire it and arrange for its reload.
 649  */
 650 static void cpu_timer_fire(struct k_itimer *timer)
 651 {
 652         if (unlikely(timer->sigq == NULL)) {
 653                 /*
 654                  * This a special case for clock_nanosleep,
 655                  * not a normal timer from sys_timer_create.
 656                  */
 657                 wake_up_process(timer->it_process);
 658                 timer->it.cpu.expires.sched = 0;
 659         } else if (timer->it.cpu.incr.sched == 0) {
 660                 /*
 661                  * One-shot timer.  Clear it as soon as it's fired.
 662                  */
 663                 posix_timer_event(timer, 0);
 664                 timer->it.cpu.expires.sched = 0;
 665         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 666                 /*
 667                  * The signal did not get queued because the signal
 668                  * was ignored, so we won't get any callback to
 669                  * reload the timer.  But we need to keep it
 670                  * ticking in case the signal is deliverable next time.
 671                  */
 672                 posix_cpu_timer_schedule(timer);
 673         }
 674 }
 675
 676 /*
 677  * Sample a process (thread group) timer for the given group_leader task.
 678  * Must be called with tasklist_lock held for reading.
 679  */
 680 static int cpu_timer_sample_group(const clockid_t which_clock,
 681                                   struct task_struct *p,
 682                                   union cpu_time_count *cpu)
 683 {
 684         struct task_cputime cputime;
 685
 686         thread_group_cputimer(p, &cputime);
 687         switch (CPUCLOCK_WHICH(which_clock)) {
 688         default:
 689                 return -EINVAL;
 690         case CPUCLOCK_PROF:
 691                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 692                 break;
 693         case CPUCLOCK_VIRT:
 694                 cpu->cpu = cputime.utime;
 695                 break;
 696         case CPUCLOCK_SCHED:
 697                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 698                 break;
 699         }
 700         return 0;
 701 }
 702
 703 /*
 704  * Guts of sys_timer_settime for CPU timers.
 705  * This is called with the timer locked and interrupts disabled.
 706  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 707  * and try again.  (This happens when the timer is in the middle of firing.)
 708  */
 709 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 710                         struct itimerspec *new, struct itimerspec *old)
 711 {
 712         struct task_struct *p = timer->it.cpu.task;
 713         union cpu_time_count old_expires, new_expires, val;
 714         int ret;
 715
 716         if (unlikely(p == NULL)) {
 717                 /*
 718                  * Timer refers to a dead task's clock.
 719                  */
 720                 return -ESRCH;
 721         }
 722
 723         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 724
 725         read_lock(&tasklist_lock);
 726         /*
 727          * We need the tasklist_lock to protect against reaping that
 728          * clears p->signal.  If p has just been reaped, we can no
 729          * longer get any information about it at all.
 730          */
 731         if (unlikely(p->signal == NULL)) {
 732                 read_unlock(&tasklist_lock);
 733                 put_task_struct(p);
 734                 timer->it.cpu.task = NULL;
 735                 return -ESRCH;
 736         }
 737
 738         /*
 739          * Disarm any old timer after extracting its expiry time.
 740          */
 741         BUG_ON(!irqs_disabled());
 742
 743         ret = 0;
 744         spin_lock(&p->sighand->siglock);
 745         old_expires = timer->it.cpu.expires;
 746         if (unlikely(timer->it.cpu.firing)) {
 747                 timer->it.cpu.firing = -1;
 748                 ret = TIMER_RETRY;
 749         } else
 750                 list_del_init(&timer->it.cpu.entry);
 751         spin_unlock(&p->sighand->siglock);
 752
 753         /*
 754          * We need to sample the current value to convert the new
 755          * value from to relative and absolute, and to convert the
 756          * old value from absolute to relative.  To set a process
 757          * timer, we need a sample to balance the thread expiry
 758          * times (in arm_timer).  With an absolute time, we must
 759          * check if it's already passed.  In short, we need a sample.
 760          */
 761         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 762                 cpu_clock_sample(timer->it_clock, p, &val);
 763         } else {
 764                 cpu_timer_sample_group(timer->it_clock, p, &val);
 765         }
 766
 767         if (old) {
 768                 if (old_expires.sched == 0) {
 769                         old->it_value.tv_sec = 0;
 770                         old->it_value.tv_nsec = 0;
 771                 } else {
 772                         /*
 773                          * Update the timer in case it has
 774                          * overrun already.  If it has,
 775                          * we'll report it as having overrun
 776                          * and with the next reloaded timer
 777                          * already ticking, though we are
 778                          * swallowing that pending
 779                          * notification here to install the
 780                          * new setting.
 781                          */
 782                         bump_cpu_timer(timer, val);
 783                         if (cpu_time_before(timer->it_clock, val,
 784                                             timer->it.cpu.expires)) {
 785                                 old_expires = cpu_time_sub(
 786                                         timer->it_clock,
 787                                         timer->it.cpu.expires, val);
 788                                 sample_to_timespec(timer->it_clock,
 789                                                    old_expires,
 790                                                    &old->it_value);
 791                         } else {
 792                                 old->it_value.tv_nsec = 1;
 793                                 old->it_value.tv_sec = 0;
 794                         }
 795                 }
 796         }
 797
 798         if (unlikely(ret)) {
 799                 /*
 800                  * We are colliding with the timer actually firing.
 801                  * Punt after filling in the timer's old value, and
 802                  * disable this firing since we are already reporting
 803                  * it as an overrun (thanks to bump_cpu_timer above).
 804                  */
 805                 read_unlock(&tasklist_lock);
 806                 goto out;
 807         }
 808
 809         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 810                 cpu_time_add(timer->it_clock, &new_expires, val);
 811         }
 812
 813         /*
 814          * Install the new expiry time (or zero).
 815          * For a timer with no notification action, we don't actually
 816          * arm the timer (we'll just fake it for timer_gettime).
 817          */
 818         timer->it.cpu.expires = new_expires;
 819         if (new_expires.sched != 0 &&
 820             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 821             cpu_time_before(timer->it_clock, val, new_expires)) {
 822                 arm_timer(timer, val);
 823         }
 824
 825         read_unlock(&tasklist_lock);
 826
 827         /*
 828          * Install the new reload setting, and
 829          * set up the signal and overrun bookkeeping.
 830          */
 831         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 832                                                 &new->it_interval);
 833
 834         /*
 835          * This acts as a modification timestamp for the timer,
 836          * so any automatic reload attempt will punt on seeing
 837          * that we have reset the timer manually.
 838          */
 839         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 840                 ~REQUEUE_PENDING;
 841         timer->it_overrun_last = 0;
 842         timer->it_overrun = -1;
 843
 844         if (new_expires.sched != 0 &&
 845             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 846             !cpu_time_before(timer->it_clock, val, new_expires)) {
 847                 /*
 848                  * The designated time already passed, so we notify
 849                  * immediately, even if the thread never runs to
 850                  * accumulate more time on this clock.
 851                  */
 852                 cpu_timer_fire(timer);
 853         }
 854
 855         ret = 0;
 856  out:
 857         if (old) {
 858                 sample_to_timespec(timer->it_clock,
 859                                    timer->it.cpu.incr, &old->it_interval);
 860         }
 861         return ret;
 862 }
 863
 864 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 865 {
 866         union cpu_time_count now;
 867         struct task_struct *p = timer->it.cpu.task;
 868         int clear_dead;
 869
 870         /*
 871          * Easy part: convert the reload time.
 872          */
 873         sample_to_timespec(timer->it_clock,
 874                            timer->it.cpu.incr, &itp->it_interval);
 875
 876         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 877                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 878                 return;
 879         }
 880
 881         if (unlikely(p == NULL)) {
 882                 /*
 883                  * This task already died and the timer will never fire.
 884                  * In this case, expires is actually the dead value.
 885                  */
 886         dead:
 887                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 888                                    &itp->it_value);
 889                 return;
 890         }
 891
 892         /*
 893          * Sample the clock to take the difference with the expiry time.
 894          */
 895         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 896                 cpu_clock_sample(timer->it_clock, p, &now);
 897                 clear_dead = p->exit_state;
 898         } else {
 899                 read_lock(&tasklist_lock);
 900                 if (unlikely(p->signal == NULL)) {
 901                         /*
 902                          * The process has been reaped.
 903                          * We can't even collect a sample any more.
 904                          * Call the timer disarmed, nothing else to do.
 905                          */
 906                         put_task_struct(p);
 907                         timer->it.cpu.task = NULL;
 908                         timer->it.cpu.expires.sched = 0;
 909                         read_unlock(&tasklist_lock);
 910                         goto dead;
 911                 } else {
 912                         cpu_timer_sample_group(timer->it_clock, p, &now);
 913                         clear_dead = (unlikely(p->exit_state) &&
 914                                       thread_group_empty(p));
 915                 }
 916                 read_unlock(&tasklist_lock);
 917         }
 918
 919         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 920                 if (timer->it.cpu.incr.sched == 0 &&
 921                     cpu_time_before(timer->it_clock,
 922                                     timer->it.cpu.expires, now)) {
 923                         /*
 924                          * Do-nothing timer expired and has no reload,
 925                          * so it's as if it was never set.
 926                          */
 927                         timer->it.cpu.expires.sched = 0;
 928                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 929                         return;
 930                 }
 931                 /*
 932                  * Account for any expirations and reloads that should
 933                  * have happened.
 934                  */
 935                 bump_cpu_timer(timer, now);
 936         }
 937
 938         if (unlikely(clear_dead)) {
 939                 /*
 940                  * We've noticed that the thread is dead, but
 941                  * not yet reaped.  Take this opportunity to
 942                  * drop our task ref.
 943                  */
 944                 clear_dead_task(timer, now);
 945                 goto dead;
 946         }
 947
 948         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 949                 sample_to_timespec(timer->it_clock,
 950                                    cpu_time_sub(timer->it_clock,
 951                                                 timer->it.cpu.expires, now),
 952                                    &itp->it_value);
 953         } else {
 954                 /*
 955                  * The timer should have expired already, but the firing
 956                  * hasn't taken place yet.  Say it's just about to expire.
 957                  */
 958                 itp->it_value.tv_nsec = 1;
 959                 itp->it_value.tv_sec = 0;
 960         }
 961 }
 962
 963 /*
 964  * Check for any per-thread CPU timers that have fired and move them off
 965  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 966  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 967  */
 968 static void check_thread_timers(struct task_struct *tsk,
 969                                 struct list_head *firing)
 970 {
 971         int maxfire;
 972         struct list_head *timers = tsk->cpu_timers;
 973         struct signal_struct *const sig = tsk->signal;
 974         unsigned long soft;
 975
 976         maxfire = 20;
 977         tsk->cputime_expires.prof_exp = cputime_zero;
 978         while (!list_empty(timers)) {
 979                 struct cpu_timer_list *t = list_first_entry(timers,
 980                                                       struct cpu_timer_list,
 981                                                       entry);
 982                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 983                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 984                         break;
 985                 }
 986                 t->firing = 1;
 987                 list_move_tail(&t->entry, firing);
 988         }
 989
 990         ++timers;
 991         maxfire = 20;
 992         tsk->cputime_expires.virt_exp = cputime_zero;
 993         while (!list_empty(timers)) {
 994                 struct cpu_timer_list *t = list_first_entry(timers,
 995                                                       struct cpu_timer_list,
 996                                                       entry);
 997                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 998                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 999                         break;
1000                 }
1001                 t->firing = 1;
1002                 list_move_tail(&t->entry, firing);
1003         }
1004
1005         ++timers;
1006         maxfire = 20;
1007         tsk->cputime_expires.sched_exp = 0;
1008         while (!list_empty(timers)) {
1009                 struct cpu_timer_list *t = list_first_entry(timers,
1010                                                       struct cpu_timer_list,
1011                                                       entry);
1012                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1013                         tsk->cputime_expires.sched_exp = t->expires.sched;
1014                         break;
1015                 }
1016                 t->firing = 1;
1017                 list_move_tail(&t->entry, firing);
1018         }
1019
1020         /*
1021          * Check for the special case thread timers.
1022          */
1023         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
1024         if (soft != RLIM_INFINITY) {
1025                 unsigned long hard =
1026                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
1027
1028                 if (hard != RLIM_INFINITY &&
1029                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1030                         /*
1031                          * At the hard limit, we just die.
1032                          * No need to calculate anything else now.
1033                          */
1034                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1035                         return;
1036                 }
1037                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
1038                         /*
1039                          * At the soft limit, send a SIGXCPU every second.
1040                          */
1041                         if (soft < hard) {
1042                                 soft += USEC_PER_SEC;
1043                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
1044                         }
1045                         printk(KERN_INFO
1046                                 "RT Watchdog Timeout: %s[%d]\n",
1047                                 tsk->comm, task_pid_nr(tsk));
1048                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1049                 }
1050         }
1051 }
1052
1053 static void stop_process_timers(struct task_struct *tsk)
1054 {
1055         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1056         unsigned long flags;
1057
1058         if (!cputimer->running)
1059                 return;
1060
1061         spin_lock_irqsave(&cputimer->lock, flags);
1062         cputimer->running = 0;
1063         spin_unlock_irqrestore(&cputimer->lock, flags);
1064 }
1065
1066 static u32 onecputick;
1067
1068 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1069                              cputime_t *expires, cputime_t cur_time, int signo)
1070 {
1071         if (cputime_eq(it->expires, cputime_zero))
1072                 return;
1073
1074         if (cputime_ge(cur_time, it->expires)) {
1075                 if (!cputime_eq(it->incr, cputime_zero)) {
1076                         it->expires = cputime_add(it->expires, it->incr);
1077                         it->error += it->incr_error;
1078                         if (it->error >= onecputick) {
1079                                 it->expires = cputime_sub(it->expires,
1080                                                           cputime_one_jiffy);
1081                                 it->error -= onecputick;
1082                         }
1083                 } else {
1084                         it->expires = cputime_zero;
1085                 }
1086
1087                 trace_itimer_expire(signo == SIGPROF ?
1088                                     ITIMER_PROF : ITIMER_VIRTUAL,
1089                                     tsk->signal->leader_pid, cur_time);
1090                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1091         }
1092
1093         if (!cputime_eq(it->expires, cputime_zero) &&
1094             (cputime_eq(*expires, cputime_zero) ||
1095              cputime_lt(it->expires, *expires))) {
1096                 *expires = it->expires;
1097         }
1098 }
1099
1100 /*
1101  * Check for any per-thread CPU timers that have fired and move them
1102  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1103  * have already been taken off.
1104  */
1105 static void check_process_timers(struct task_struct *tsk,
1106                                  struct list_head *firing)
1107 {
1108         int maxfire;
1109         struct signal_struct *const sig = tsk->signal;
1110         cputime_t utime, ptime, virt_expires, prof_expires;
1111         unsigned long long sum_sched_runtime, sched_expires;
1112         struct list_head *timers = sig->cpu_timers;
1113         struct task_cputime cputime;
1114         unsigned long soft;
1115
1116         /*
1117          * Don't sample the current process CPU clocks if there are no timers.
1118          */
1119         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1120             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1121             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1122             list_empty(&timers[CPUCLOCK_VIRT]) &&
1123             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1124             list_empty(&timers[CPUCLOCK_SCHED])) {
1125                 stop_process_timers(tsk);
1126                 return;
1127         }
1128
1129         /*
1130          * Collect the current process totals.
1131          */
1132         thread_group_cputimer(tsk, &cputime);
1133         utime = cputime.utime;
1134         ptime = cputime_add(utime, cputime.stime);
1135         sum_sched_runtime = cputime.sum_exec_runtime;
1136         maxfire = 20;
1137         prof_expires = cputime_zero;
1138         while (!list_empty(timers)) {
1139                 struct cpu_timer_list *tl = list_first_entry(timers,
1140                                                       struct cpu_timer_list,
1141                                                       entry);
1142                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1143                         prof_expires = tl->expires.cpu;
1144                         break;
1145                 }
1146                 tl->firing = 1;
1147                 list_move_tail(&tl->entry, firing);
1148         }
1149
1150         ++timers;
1151         maxfire = 20;
1152         virt_expires = cputime_zero;
1153         while (!list_empty(timers)) {
1154                 struct cpu_timer_list *tl = list_first_entry(timers,
1155                                                       struct cpu_timer_list,
1156                                                       entry);
1157                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1158                         virt_expires = tl->expires.cpu;
1159                         break;
1160                 }
1161                 tl->firing = 1;
1162                 list_move_tail(&tl->entry, firing);
1163         }
1164
1165         ++timers;
1166         maxfire = 20;
1167         sched_expires = 0;
1168         while (!list_empty(timers)) {
1169                 struct cpu_timer_list *tl = list_first_entry(timers,
1170                                                       struct cpu_timer_list,
1171                                                       entry);
1172                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1173                         sched_expires = tl->expires.sched;
1174                         break;
1175                 }
1176                 tl->firing = 1;
1177                 list_move_tail(&tl->entry, firing);
1178         }
1179
1180         /*
1181          * Check for the special case process timers.
1182          */
1183         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1184                          SIGPROF);
1185         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1186                          SIGVTALRM);
1187         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1188         if (soft != RLIM_INFINITY) {
1189                 unsigned long psecs = cputime_to_secs(ptime);
1190                 unsigned long hard =
1191                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1192                 cputime_t x;
1193                 if (psecs >= hard) {
1194                         /*
1195                          * At the hard limit, we just die.
1196                          * No need to calculate anything else now.
1197                          */
1198                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1199                         return;
1200                 }
1201                 if (psecs >= soft) {
1202                         /*
1203                          * At the soft limit, send a SIGXCPU every second.
1204                          */
1205                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1206                         if (soft < hard) {
1207                                 soft++;
1208                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1209                         }
1210                 }
1211                 x = secs_to_cputime(soft);
1212                 if (cputime_eq(prof_expires, cputime_zero) ||
1213                     cputime_lt(x, prof_expires)) {
1214                         prof_expires = x;
1215                 }
1216         }
1217
1218         if (!cputime_eq(prof_expires, cputime_zero) &&
1219             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1220              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1221                 sig->cputime_expires.prof_exp = prof_expires;
1222         if (!cputime_eq(virt_expires, cputime_zero) &&
1223             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1224              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1225                 sig->cputime_expires.virt_exp = virt_expires;
1226         if (sched_expires != 0 &&
1227             (sig->cputime_expires.sched_exp == 0 ||
1228              sig->cputime_expires.sched_exp > sched_expires))
1229                 sig->cputime_expires.sched_exp = sched_expires;
1230 }
1231
1232 /*
1233  * This is called from the signal code (via do_schedule_next_timer)
1234  * when the last timer signal was delivered and we have to reload the timer.
1235  */
1236 void posix_cpu_timer_schedule(struct k_itimer *timer)
1237 {
1238         struct task_struct *p = timer->it.cpu.task;
1239         union cpu_time_count now;
1240
1241         if (unlikely(p == NULL))
1242                 /*
1243                  * The task was cleaned up already, no future firings.
1244                  */
1245                 goto out;
1246
1247         /*
1248          * Fetch the current sample and update the timer's expiry time.
1249          */
1250         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1251                 cpu_clock_sample(timer->it_clock, p, &now);
1252                 bump_cpu_timer(timer, now);
1253                 if (unlikely(p->exit_state)) {
1254                         clear_dead_task(timer, now);
1255                         goto out;
1256                 }
1257                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1258         } else {
1259                 read_lock(&tasklist_lock);
1260                 if (unlikely(p->signal == NULL)) {
1261                         /*
1262                          * The process has been reaped.
1263                          * We can't even collect a sample any more.
1264                          */
1265                         put_task_struct(p);
1266                         timer->it.cpu.task = p = NULL;
1267                         timer->it.cpu.expires.sched = 0;
1268                         goto out_unlock;
1269                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1270                         /*
1271                          * We've noticed that the thread is dead, but
1272                          * not yet reaped.  Take this opportunity to
1273                          * drop our task ref.
1274                          */
1275                         clear_dead_task(timer, now);
1276                         goto out_unlock;
1277                 }
1278                 cpu_timer_sample_group(timer->it_clock, p, &now);
1279                 bump_cpu_timer(timer, now);
1280                 /* Leave the tasklist_lock locked for the call below.  */
1281         }
1282
1283         /*
1284          * Now re-arm for the new expiry time.
1285          */
1286         arm_timer(timer, now);
1287
1288 out_unlock:
1289         read_unlock(&tasklist_lock);
1290
1291 out:
1292         timer->it_overrun_last = timer->it_overrun;
1293         timer->it_overrun = -1;
1294         ++timer->it_requeue_pending;
1295 }
1296
1297 /**
1298  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1299  *
1300  * @cputime:    The struct to compare.
1301  *
1302  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1303  * are zero, false if any field is nonzero.
1304  */
1305 static inline int task_cputime_zero(const struct task_cputime *cputime)
1306 {
1307         if (cputime_eq(cputime->utime, cputime_zero) &&
1308             cputime_eq(cputime->stime, cputime_zero) &&
1309             cputime->sum_exec_runtime == 0)
1310                 return 1;
1311         return 0;
1312 }
1313
1314 /**
1315  * task_cputime_expired - Compare two task_cputime entities.
1316  *
1317  * @sample:     The task_cputime structure to be checked for expiration.
1318  * @expires:    Expiration times, against which @sample will be checked.
1319  *
1320  * Checks @sample against @expires to see if any field of @sample has expired.
1321  * Returns true if any field of the former is greater than the corresponding
1322  * field of the latter if the latter field is set.  Otherwise returns false.
1323  */
1324 static inline int task_cputime_expired(const struct task_cputime *sample,
1325                                         const struct task_cputime *expires)
1326 {
1327         if (!cputime_eq(expires->utime, cputime_zero) &&
1328             cputime_ge(sample->utime, expires->utime))
1329                 return 1;
1330         if (!cputime_eq(expires->stime, cputime_zero) &&
1331             cputime_ge(cputime_add(sample->utime, sample->stime),
1332                        expires->stime))
1333                 return 1;
1334         if (expires->sum_exec_runtime != 0 &&
1335             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1336                 return 1;
1337         return 0;
1338 }
1339
1340 /**
1341  * fastpath_timer_check - POSIX CPU timers fast path.
1342  *
1343  * @tsk:        The task (thread) being checked.
1344  *
1345  * Check the task and thread group timers.  If both are zero (there are no
1346  * timers set) return false.  Otherwise snapshot the task and thread group
1347  * timers and compare them with the corresponding expiration times.  Return
1348  * true if a timer has expired, else return false.
1349  */
1350 static inline int fastpath_timer_check(struct task_struct *tsk)
1351 {
1352         struct signal_struct *sig;
1353
1354         /* tsk == current, ensure it is safe to use ->signal/sighand */
1355         if (unlikely(tsk->exit_state))
1356                 return 0;
1357
1358         if (!task_cputime_zero(&tsk->cputime_expires)) {
1359                 struct task_cputime task_sample = {
1360                         .utime = tsk->utime,
1361                         .stime = tsk->stime,
1362                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1363                 };
1364
1365                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1366                         return 1;
1367         }
1368
1369         sig = tsk->signal;
1370         if (!task_cputime_zero(&sig->cputime_expires)) {
1371                 struct task_cputime group_sample;
1372
1373                 thread_group_cputimer(tsk, &group_sample);
1374                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1375                         return 1;
1376         }
1377
1378         return 0;
1379 }
1380
1381 /*
1382  * This is called from the timer interrupt handler.  The irq handler has
1383  * already updated our counts.  We need to check if any timers fire now.
1384  * Interrupts are disabled.
1385  */
1386 void run_posix_cpu_timers(struct task_struct *tsk)
1387 {
1388         LIST_HEAD(firing);
1389         struct k_itimer *timer, *next;
1390
1391         BUG_ON(!irqs_disabled());
1392
1393         /*
1394          * The fast path checks that there are no expired thread or thread
1395          * group timers.  If that's so, just return.
1396          */
1397         if (!fastpath_timer_check(tsk))
1398                 return;
1399
1400         spin_lock(&tsk->sighand->siglock);
1401         /*
1402          * Here we take off tsk->signal->cpu_timers[N] and
1403          * tsk->cpu_timers[N] all the timers that are firing, and
1404          * put them on the firing list.
1405          */
1406         check_thread_timers(tsk, &firing);
1407         check_process_timers(tsk, &firing);
1408
1409         /*
1410          * We must release these locks before taking any timer's lock.
1411          * There is a potential race with timer deletion here, as the
1412          * siglock now protects our private firing list.  We have set
1413          * the firing flag in each timer, so that a deletion attempt
1414          * that gets the timer lock before we do will give it up and
1415          * spin until we've taken care of that timer below.
1416          */
1417         spin_unlock(&tsk->sighand->siglock);
1418
1419         /*
1420          * Now that all the timers on our list have the firing flag,
1421          * noone will touch their list entries but us.  We'll take
1422          * each timer's lock before clearing its firing flag, so no
1423          * timer call will interfere.
1424          */
1425         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1426                 int cpu_firing;
1427
1428                 spin_lock(&timer->it_lock);
1429                 list_del_init(&timer->it.cpu.entry);
1430                 cpu_firing = timer->it.cpu.firing;
1431                 timer->it.cpu.firing = 0;
1432                 /*
1433                  * The firing flag is -1 if we collided with a reset
1434                  * of the timer, which already reported this
1435                  * almost-firing as an overrun.  So don't generate an event.
1436                  */
1437                 if (likely(cpu_firing >= 0))
1438                         cpu_timer_fire(timer);
1439                 spin_unlock(&timer->it_lock);
1440         }
1441 }
1442
1443 /*
1444  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1445  * The tsk->sighand->siglock must be held by the caller.
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
1452         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1453         cpu_timer_sample_group(clock_idx, tsk, &now);
1454
1455         if (oldval) {
1456                 /*
1457                  * We are setting itimer. The *oldval is absolute and we update
1458                  * it to be relative, *newval argument is relative and we update
1459                  * it to be absolute.
1460                  */
1461                 if (!cputime_eq(*oldval, cputime_zero)) {
1462                         if (cputime_le(*oldval, now.cpu)) {
1463                                 /* Just about to fire. */
1464                                 *oldval = cputime_one_jiffy;
1465                         } else {
1466                                 *oldval = cputime_sub(*oldval, now.cpu);
1467                         }
1468                 }
1469
1470                 if (cputime_eq(*newval, cputime_zero))
1471                         return;
1472                 *newval = cputime_add(*newval, now.cpu);
1473         }
1474
1475         /*
1476          * Update expiration cache if we are the earliest timer, or eventually
1477          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1478          */
1479         switch (clock_idx) {
1480         case CPUCLOCK_PROF:
1481                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1482                         tsk->signal->cputime_expires.prof_exp = *newval;
1483                 break;
1484         case CPUCLOCK_VIRT:
1485                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1486                         tsk->signal->cputime_expires.virt_exp = *newval;
1487                 break;
1488         }
1489 }
1490
1491 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1492                             struct timespec *rqtp, struct itimerspec *it)
1493 {
1494         struct k_itimer timer;
1495         int error;
1496
1497         /*
1498          * Set up a temporary timer and then wait for it to go off.
1499          */
1500         memset(&timer, 0, sizeof timer);
1501         spin_lock_init(&timer.it_lock);
1502         timer.it_clock = which_clock;
1503         timer.it_overrun = -1;
1504         error = posix_cpu_timer_create(&timer);
1505         timer.it_process = current;
1506         if (!error) {
1507                 static struct itimerspec zero_it;
1508
1509                 memset(it, 0, sizeof *it);
1510                 it->it_value = *rqtp;
1511
1512                 spin_lock_irq(&timer.it_lock);
1513                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1514                 if (error) {
1515                         spin_unlock_irq(&timer.it_lock);
1516                         return error;
1517                 }
1518
1519                 while (!signal_pending(current)) {
1520                         if (timer.it.cpu.expires.sched == 0) {
1521                                 /*
1522                                  * Our timer fired and was reset.
1523                                  */
1524                                 spin_unlock_irq(&timer.it_lock);
1525                                 return 0;
1526                         }
1527
1528                         /*
1529                          * Block until cpu_timer_fire (or a signal) wakes us.
1530                          */
1531                         __set_current_state(TASK_INTERRUPTIBLE);
1532                         spin_unlock_irq(&timer.it_lock);
1533                         schedule();
1534                         spin_lock_irq(&timer.it_lock);
1535                 }
1536
1537                 /*
1538                  * We were interrupted by a signal.
1539                  */
1540                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1541                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1542                 spin_unlock_irq(&timer.it_lock);
1543
1544                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1545                         /*
1546                          * It actually did fire already.
1547                          */
1548                         return 0;
1549                 }
1550
1551                 error = -ERESTART_RESTARTBLOCK;
1552         }
1553
1554         return error;
1555 }
1556
1557 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1558                      struct timespec *rqtp, struct timespec __user *rmtp)
1559 {
1560         struct restart_block *restart_block =
1561             &current_thread_info()->restart_block;
1562         struct itimerspec it;
1563         int error;
1564
1565         /*
1566          * Diagnose required errors first.
1567          */
1568         if (CPUCLOCK_PERTHREAD(which_clock) &&
1569             (CPUCLOCK_PID(which_clock) == 0 ||
1570              CPUCLOCK_PID(which_clock) == current->pid))
1571                 return -EINVAL;
1572
1573         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1574
1575         if (error == -ERESTART_RESTARTBLOCK) {
1576
1577                 if (flags & TIMER_ABSTIME)
1578                         return -ERESTARTNOHAND;
1579                 /*
1580                  * Report back to the user the time still remaining.
1581                  */
1582                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1583                         return -EFAULT;
1584
1585                 restart_block->fn = posix_cpu_nsleep_restart;
1586                 restart_block->arg0 = which_clock;
1587                 restart_block->arg1 = (unsigned long) rmtp;
1588                 restart_block->arg2 = rqtp->tv_sec;
1589                 restart_block->arg3 = rqtp->tv_nsec;
1590         }
1591         return error;
1592 }
1593
1594 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1595 {
1596         clockid_t which_clock = restart_block->arg0;
1597         struct timespec __user *rmtp;
1598         struct timespec t;
1599         struct itimerspec it;
1600         int error;
1601
1602         rmtp = (struct timespec __user *) restart_block->arg1;
1603         t.tv_sec = restart_block->arg2;
1604         t.tv_nsec = restart_block->arg3;
1605
1606         restart_block->fn = do_no_restart_syscall;
1607         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1608
1609         if (error == -ERESTART_RESTARTBLOCK) {
1610                 /*
1611                  * Report back to the user the time still remaining.
1612                  */
1613                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1614                         return -EFAULT;
1615
1616                 restart_block->fn = posix_cpu_nsleep_restart;
1617                 restart_block->arg0 = which_clock;
1618                 restart_block->arg1 = (unsigned long) rmtp;
1619                 restart_block->arg2 = t.tv_sec;
1620                 restart_block->arg3 = t.tv_nsec;
1621         }
1622         return error;
1623
1624 }
1625
1626
1627 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1628 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1629
1630 static int process_cpu_clock_getres(const clockid_t which_clock,
1631                                     struct timespec *tp)
1632 {
1633         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1634 }
1635 static int process_cpu_clock_get(const clockid_t which_clock,
1636                                  struct timespec *tp)
1637 {
1638         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1639 }
1640 static int process_cpu_timer_create(struct k_itimer *timer)
1641 {
1642         timer->it_clock = PROCESS_CLOCK;
1643         return posix_cpu_timer_create(timer);
1644 }
1645 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1646                               struct timespec *rqtp,
1647                               struct timespec __user *rmtp)
1648 {
1649         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1650 }
1651 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1652 {
1653         return -EINVAL;
1654 }
1655 static int thread_cpu_clock_getres(const clockid_t which_clock,
1656                                    struct timespec *tp)
1657 {
1658         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1659 }
1660 static int thread_cpu_clock_get(const clockid_t which_clock,
1661                                 struct timespec *tp)
1662 {
1663         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1664 }
1665 static int thread_cpu_timer_create(struct k_itimer *timer)
1666 {
1667         timer->it_clock = THREAD_CLOCK;
1668         return posix_cpu_timer_create(timer);
1669 }
1670 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1671                               struct timespec *rqtp, struct timespec __user *rmtp)
1672 {
1673         return -EINVAL;
1674 }
1675 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1676 {
1677         return -EINVAL;
1678 }
1679
1680 static __init int init_posix_cpu_timers(void)
1681 {
1682         struct k_clock process = {
1683                 .clock_getres = process_cpu_clock_getres,
1684                 .clock_get = process_cpu_clock_get,
1685                 .clock_set = do_posix_clock_nosettime,
1686                 .timer_create = process_cpu_timer_create,
1687                 .nsleep = process_cpu_nsleep,
1688                 .nsleep_restart = process_cpu_nsleep_restart,
1689         };
1690         struct k_clock thread = {
1691                 .clock_getres = thread_cpu_clock_getres,
1692                 .clock_get = thread_cpu_clock_get,
1693                 .clock_set = do_posix_clock_nosettime,
1694                 .timer_create = thread_cpu_timer_create,
1695                 .nsleep = thread_cpu_nsleep,
1696                 .nsleep_restart = thread_cpu_nsleep_restart,
1697         };
1698         struct timespec ts;
1699
1700         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1701         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1702
1703         cputime_to_timespec(cputime_one_jiffy, &ts);
1704         onecputick = ts.tv_nsec;
1705         WARN_ON(ts.tv_sec != 0);
1706
1707         return 0;
1708 }
1709 __initcall(init_posix_cpu_timers);