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 /*
 551  * Insert the timer on the appropriate list before any timers that
 552  * expire later.  This must be called with the tasklist_lock held
 553  * for reading, and interrupts disabled.
 554  */
 555 static void arm_timer(struct k_itimer *timer)
 556 {
 557         struct task_struct *p = timer->it.cpu.task;
 558         struct list_head *head, *listpos;
 559         struct task_cputime *cputime_expires;
 560         struct cpu_timer_list *const nt = &timer->it.cpu;
 561         struct cpu_timer_list *next;
 562
 563         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 564                 head = p->cpu_timers;
 565                 cputime_expires = &p->cputime_expires;
 566         } else {
 567                 head = p->signal->cpu_timers;
 568                 cputime_expires = &p->signal->cputime_expires;
 569         }
 570         head += CPUCLOCK_WHICH(timer->it_clock);
 571
 572         BUG_ON(!irqs_disabled());
 573         spin_lock(&p->sighand->siglock);
 574
 575         listpos = head;
 576         list_for_each_entry(next, head, entry) {
 577                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
 578                         break;
 579                 listpos = &next->entry;
 580         }
 581         list_add(&nt->entry, listpos);
 582
 583         if (listpos == head) {
 584                 union cpu_time_count *exp = &nt->expires;
 585
 586                 /*
 587                  * We are the new earliest-expiring POSIX 1.b timer, hence
 588                  * need to update expiration cache. Take into account that
 589                  * for process timers we share expiration cache with itimers
 590                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 591                  */
 592
 593                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 594                 case CPUCLOCK_PROF:
 595                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
 596                                 cputime_expires->prof_exp = exp->cpu;
 597                         break;
 598                 case CPUCLOCK_VIRT:
 599                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
 600                                 cputime_expires->virt_exp = exp->cpu;
 601                         break;
 602                 case CPUCLOCK_SCHED:
 603                         if (cputime_expires->sched_exp == 0 ||
 604                             cputime_expires->sched_exp > exp->sched)
 605                                 cputime_expires->sched_exp = exp->sched;
 606                         break;
 607                 }
 608         }
 609
 610         spin_unlock(&p->sighand->siglock);
 611 }
 612
 613 /*
 614  * The timer is locked, fire it and arrange for its reload.
 615  */
 616 static void cpu_timer_fire(struct k_itimer *timer)
 617 {
 618         if (unlikely(timer->sigq == NULL)) {
 619                 /*
 620                  * This a special case for clock_nanosleep,
 621                  * not a normal timer from sys_timer_create.
 622                  */
 623                 wake_up_process(timer->it_process);
 624                 timer->it.cpu.expires.sched = 0;
 625         } else if (timer->it.cpu.incr.sched == 0) {
 626                 /*
 627                  * One-shot timer.  Clear it as soon as it's fired.
 628                  */
 629                 posix_timer_event(timer, 0);
 630                 timer->it.cpu.expires.sched = 0;
 631         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 632                 /*
 633                  * The signal did not get queued because the signal
 634                  * was ignored, so we won't get any callback to
 635                  * reload the timer.  But we need to keep it
 636                  * ticking in case the signal is deliverable next time.
 637                  */
 638                 posix_cpu_timer_schedule(timer);
 639         }
 640 }
 641
 642 /*
 643  * Sample a process (thread group) timer for the given group_leader task.
 644  * Must be called with tasklist_lock held for reading.
 645  */
 646 static int cpu_timer_sample_group(const clockid_t which_clock,
 647                                   struct task_struct *p,
 648                                   union cpu_time_count *cpu)
 649 {
 650         struct task_cputime cputime;
 651
 652         thread_group_cputimer(p, &cputime);
 653         switch (CPUCLOCK_WHICH(which_clock)) {
 654         default:
 655                 return -EINVAL;
 656         case CPUCLOCK_PROF:
 657                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 658                 break;
 659         case CPUCLOCK_VIRT:
 660                 cpu->cpu = cputime.utime;
 661                 break;
 662         case CPUCLOCK_SCHED:
 663                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 664                 break;
 665         }
 666         return 0;
 667 }
 668
 669 /*
 670  * Guts of sys_timer_settime for CPU timers.
 671  * This is called with the timer locked and interrupts disabled.
 672  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 673  * and try again.  (This happens when the timer is in the middle of firing.)
 674  */
 675 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 676                         struct itimerspec *new, struct itimerspec *old)
 677 {
 678         struct task_struct *p = timer->it.cpu.task;
 679         union cpu_time_count old_expires, new_expires, old_incr, val;
 680         int ret;
 681
 682         if (unlikely(p == NULL)) {
 683                 /*
 684                  * Timer refers to a dead task's clock.
 685                  */
 686                 return -ESRCH;
 687         }
 688
 689         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 690
 691         read_lock(&tasklist_lock);
 692         /*
 693          * We need the tasklist_lock to protect against reaping that
 694          * clears p->signal.  If p has just been reaped, we can no
 695          * longer get any information about it at all.
 696          */
 697         if (unlikely(p->signal == NULL)) {
 698                 read_unlock(&tasklist_lock);
 699                 put_task_struct(p);
 700                 timer->it.cpu.task = NULL;
 701                 return -ESRCH;
 702         }
 703
 704         /*
 705          * Disarm any old timer after extracting its expiry time.
 706          */
 707         BUG_ON(!irqs_disabled());
 708
 709         ret = 0;
 710         old_incr = timer->it.cpu.incr;
 711         spin_lock(&p->sighand->siglock);
 712         old_expires = timer->it.cpu.expires;
 713         if (unlikely(timer->it.cpu.firing)) {
 714                 timer->it.cpu.firing = -1;
 715                 ret = TIMER_RETRY;
 716         } else
 717                 list_del_init(&timer->it.cpu.entry);
 718         spin_unlock(&p->sighand->siglock);
 719
 720         /*
 721          * We need to sample the current value to convert the new
 722          * value from to relative and absolute, and to convert the
 723          * old value from absolute to relative.  To set a process
 724          * timer, we need a sample to balance the thread expiry
 725          * times (in arm_timer).  With an absolute time, we must
 726          * check if it's already passed.  In short, we need a sample.
 727          */
 728         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 729                 cpu_clock_sample(timer->it_clock, p, &val);
 730         } else {
 731                 cpu_timer_sample_group(timer->it_clock, p, &val);
 732         }
 733
 734         if (old) {
 735                 if (old_expires.sched == 0) {
 736                         old->it_value.tv_sec = 0;
 737                         old->it_value.tv_nsec = 0;
 738                 } else {
 739                         /*
 740                          * Update the timer in case it has
 741                          * overrun already.  If it has,
 742                          * we'll report it as having overrun
 743                          * and with the next reloaded timer
 744                          * already ticking, though we are
 745                          * swallowing that pending
 746                          * notification here to install the
 747                          * new setting.
 748                          */
 749                         bump_cpu_timer(timer, val);
 750                         if (cpu_time_before(timer->it_clock, val,
 751                                             timer->it.cpu.expires)) {
 752                                 old_expires = cpu_time_sub(
 753                                         timer->it_clock,
 754                                         timer->it.cpu.expires, val);
 755                                 sample_to_timespec(timer->it_clock,
 756                                                    old_expires,
 757                                                    &old->it_value);
 758                         } else {
 759                                 old->it_value.tv_nsec = 1;
 760                                 old->it_value.tv_sec = 0;
 761                         }
 762                 }
 763         }
 764
 765         if (unlikely(ret)) {
 766                 /*
 767                  * We are colliding with the timer actually firing.
 768                  * Punt after filling in the timer's old value, and
 769                  * disable this firing since we are already reporting
 770                  * it as an overrun (thanks to bump_cpu_timer above).
 771                  */
 772                 read_unlock(&tasklist_lock);
 773                 goto out;
 774         }
 775
 776         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 777                 cpu_time_add(timer->it_clock, &new_expires, val);
 778         }
 779
 780         /*
 781          * Install the new expiry time (or zero).
 782          * For a timer with no notification action, we don't actually
 783          * arm the timer (we'll just fake it for timer_gettime).
 784          */
 785         timer->it.cpu.expires = new_expires;
 786         if (new_expires.sched != 0 &&
 787             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 788             cpu_time_before(timer->it_clock, val, new_expires)) {
 789                 arm_timer(timer);
 790         }
 791
 792         read_unlock(&tasklist_lock);
 793
 794         /*
 795          * Install the new reload setting, and
 796          * set up the signal and overrun bookkeeping.
 797          */
 798         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 799                                                 &new->it_interval);
 800
 801         /*
 802          * This acts as a modification timestamp for the timer,
 803          * so any automatic reload attempt will punt on seeing
 804          * that we have reset the timer manually.
 805          */
 806         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 807                 ~REQUEUE_PENDING;
 808         timer->it_overrun_last = 0;
 809         timer->it_overrun = -1;
 810
 811         if (new_expires.sched != 0 &&
 812             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 813             !cpu_time_before(timer->it_clock, val, new_expires)) {
 814                 /*
 815                  * The designated time already passed, so we notify
 816                  * immediately, even if the thread never runs to
 817                  * accumulate more time on this clock.
 818                  */
 819                 cpu_timer_fire(timer);
 820         }
 821
 822         ret = 0;
 823  out:
 824         if (old) {
 825                 sample_to_timespec(timer->it_clock,
 826                                    old_incr, &old->it_interval);
 827         }
 828         return ret;
 829 }
 830
 831 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 832 {
 833         union cpu_time_count now;
 834         struct task_struct *p = timer->it.cpu.task;
 835         int clear_dead;
 836
 837         /*
 838          * Easy part: convert the reload time.
 839          */
 840         sample_to_timespec(timer->it_clock,
 841                            timer->it.cpu.incr, &itp->it_interval);
 842
 843         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 844                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 845                 return;
 846         }
 847
 848         if (unlikely(p == NULL)) {
 849                 /*
 850                  * This task already died and the timer will never fire.
 851                  * In this case, expires is actually the dead value.
 852                  */
 853         dead:
 854                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 855                                    &itp->it_value);
 856                 return;
 857         }
 858
 859         /*
 860          * Sample the clock to take the difference with the expiry time.
 861          */
 862         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 863                 cpu_clock_sample(timer->it_clock, p, &now);
 864                 clear_dead = p->exit_state;
 865         } else {
 866                 read_lock(&tasklist_lock);
 867                 if (unlikely(p->signal == NULL)) {
 868                         /*
 869                          * The process has been reaped.
 870                          * We can't even collect a sample any more.
 871                          * Call the timer disarmed, nothing else to do.
 872                          */
 873                         put_task_struct(p);
 874                         timer->it.cpu.task = NULL;
 875                         timer->it.cpu.expires.sched = 0;
 876                         read_unlock(&tasklist_lock);
 877                         goto dead;
 878                 } else {
 879                         cpu_timer_sample_group(timer->it_clock, p, &now);
 880                         clear_dead = (unlikely(p->exit_state) &&
 881                                       thread_group_empty(p));
 882                 }
 883                 read_unlock(&tasklist_lock);
 884         }
 885
 886         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 887                 if (timer->it.cpu.incr.sched == 0 &&
 888                     cpu_time_before(timer->it_clock,
 889                                     timer->it.cpu.expires, now)) {
 890                         /*
 891                          * Do-nothing timer expired and has no reload,
 892                          * so it's as if it was never set.
 893                          */
 894                         timer->it.cpu.expires.sched = 0;
 895                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 896                         return;
 897                 }
 898                 /*
 899                  * Account for any expirations and reloads that should
 900                  * have happened.
 901                  */
 902                 bump_cpu_timer(timer, now);
 903         }
 904
 905         if (unlikely(clear_dead)) {
 906                 /*
 907                  * We've noticed that the thread is dead, but
 908                  * not yet reaped.  Take this opportunity to
 909                  * drop our task ref.
 910                  */
 911                 clear_dead_task(timer, now);
 912                 goto dead;
 913         }
 914
 915         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 916                 sample_to_timespec(timer->it_clock,
 917                                    cpu_time_sub(timer->it_clock,
 918                                                 timer->it.cpu.expires, now),
 919                                    &itp->it_value);
 920         } else {
 921                 /*
 922                  * The timer should have expired already, but the firing
 923                  * hasn't taken place yet.  Say it's just about to expire.
 924                  */
 925                 itp->it_value.tv_nsec = 1;
 926                 itp->it_value.tv_sec = 0;
 927         }
 928 }
 929
 930 /*
 931  * Check for any per-thread CPU timers that have fired and move them off
 932  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 933  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 934  */
 935 static void check_thread_timers(struct task_struct *tsk,
 936                                 struct list_head *firing)
 937 {
 938         int maxfire;
 939         struct list_head *timers = tsk->cpu_timers;
 940         struct signal_struct *const sig = tsk->signal;
 941         unsigned long soft;
 942
 943         maxfire = 20;
 944         tsk->cputime_expires.prof_exp = cputime_zero;
 945         while (!list_empty(timers)) {
 946                 struct cpu_timer_list *t = list_first_entry(timers,
 947                                                       struct cpu_timer_list,
 948                                                       entry);
 949                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 950                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 951                         break;
 952                 }
 953                 t->firing = 1;
 954                 list_move_tail(&t->entry, firing);
 955         }
 956
 957         ++timers;
 958         maxfire = 20;
 959         tsk->cputime_expires.virt_exp = cputime_zero;
 960         while (!list_empty(timers)) {
 961                 struct cpu_timer_list *t = list_first_entry(timers,
 962                                                       struct cpu_timer_list,
 963                                                       entry);
 964                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 965                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 966                         break;
 967                 }
 968                 t->firing = 1;
 969                 list_move_tail(&t->entry, firing);
 970         }
 971
 972         ++timers;
 973         maxfire = 20;
 974         tsk->cputime_expires.sched_exp = 0;
 975         while (!list_empty(timers)) {
 976                 struct cpu_timer_list *t = list_first_entry(timers,
 977                                                       struct cpu_timer_list,
 978                                                       entry);
 979                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
 980                         tsk->cputime_expires.sched_exp = t->expires.sched;
 981                         break;
 982                 }
 983                 t->firing = 1;
 984                 list_move_tail(&t->entry, firing);
 985         }
 986
 987         /*
 988          * Check for the special case thread timers.
 989          */
 990         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 991         if (soft != RLIM_INFINITY) {
 992                 unsigned long hard =
 993                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 994
 995                 if (hard != RLIM_INFINITY &&
 996                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 997                         /*
 998                          * At the hard limit, we just die.
 999                          * No need to calculate anything else now.
1000                          */
1001                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1002                         return;
1003                 }
1004                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
1005                         /*
1006                          * At the soft limit, send a SIGXCPU every second.
1007                          */
1008                         if (soft < hard) {
1009                                 soft += USEC_PER_SEC;
1010                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
1011                         }
1012                         printk(KERN_INFO
1013                                 "RT Watchdog Timeout: %s[%d]\n",
1014                                 tsk->comm, task_pid_nr(tsk));
1015                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1016                 }
1017         }
1018 }
1019
1020 static void stop_process_timers(struct task_struct *tsk)
1021 {
1022         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1023         unsigned long flags;
1024
1025         if (!cputimer->running)
1026                 return;
1027
1028         spin_lock_irqsave(&cputimer->lock, flags);
1029         cputimer->running = 0;
1030         spin_unlock_irqrestore(&cputimer->lock, flags);
1031 }
1032
1033 static u32 onecputick;
1034
1035 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1036                              cputime_t *expires, cputime_t cur_time, int signo)
1037 {
1038         if (cputime_eq(it->expires, cputime_zero))
1039                 return;
1040
1041         if (cputime_ge(cur_time, it->expires)) {
1042                 if (!cputime_eq(it->incr, cputime_zero)) {
1043                         it->expires = cputime_add(it->expires, it->incr);
1044                         it->error += it->incr_error;
1045                         if (it->error >= onecputick) {
1046                                 it->expires = cputime_sub(it->expires,
1047                                                           cputime_one_jiffy);
1048                                 it->error -= onecputick;
1049                         }
1050                 } else {
1051                         it->expires = cputime_zero;
1052                 }
1053
1054                 trace_itimer_expire(signo == SIGPROF ?
1055                                     ITIMER_PROF : ITIMER_VIRTUAL,
1056                                     tsk->signal->leader_pid, cur_time);
1057                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1058         }
1059
1060         if (!cputime_eq(it->expires, cputime_zero) &&
1061             (cputime_eq(*expires, cputime_zero) ||
1062              cputime_lt(it->expires, *expires))) {
1063                 *expires = it->expires;
1064         }
1065 }
1066
1067 /*
1068  * Check for any per-thread CPU timers that have fired and move them
1069  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1070  * have already been taken off.
1071  */
1072 static void check_process_timers(struct task_struct *tsk,
1073                                  struct list_head *firing)
1074 {
1075         int maxfire;
1076         struct signal_struct *const sig = tsk->signal;
1077         cputime_t utime, ptime, virt_expires, prof_expires;
1078         unsigned long long sum_sched_runtime, sched_expires;
1079         struct list_head *timers = sig->cpu_timers;
1080         struct task_cputime cputime;
1081         unsigned long soft;
1082
1083         /*
1084          * Don't sample the current process CPU clocks if there are no timers.
1085          */
1086         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1087             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1088             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1089             list_empty(&timers[CPUCLOCK_VIRT]) &&
1090             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1091             list_empty(&timers[CPUCLOCK_SCHED])) {
1092                 stop_process_timers(tsk);
1093                 return;
1094         }
1095
1096         /*
1097          * Collect the current process totals.
1098          */
1099         thread_group_cputimer(tsk, &cputime);
1100         utime = cputime.utime;
1101         ptime = cputime_add(utime, cputime.stime);
1102         sum_sched_runtime = cputime.sum_exec_runtime;
1103         maxfire = 20;
1104         prof_expires = cputime_zero;
1105         while (!list_empty(timers)) {
1106                 struct cpu_timer_list *tl = list_first_entry(timers,
1107                                                       struct cpu_timer_list,
1108                                                       entry);
1109                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1110                         prof_expires = tl->expires.cpu;
1111                         break;
1112                 }
1113                 tl->firing = 1;
1114                 list_move_tail(&tl->entry, firing);
1115         }
1116
1117         ++timers;
1118         maxfire = 20;
1119         virt_expires = cputime_zero;
1120         while (!list_empty(timers)) {
1121                 struct cpu_timer_list *tl = list_first_entry(timers,
1122                                                       struct cpu_timer_list,
1123                                                       entry);
1124                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1125                         virt_expires = tl->expires.cpu;
1126                         break;
1127                 }
1128                 tl->firing = 1;
1129                 list_move_tail(&tl->entry, firing);
1130         }
1131
1132         ++timers;
1133         maxfire = 20;
1134         sched_expires = 0;
1135         while (!list_empty(timers)) {
1136                 struct cpu_timer_list *tl = list_first_entry(timers,
1137                                                       struct cpu_timer_list,
1138                                                       entry);
1139                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1140                         sched_expires = tl->expires.sched;
1141                         break;
1142                 }
1143                 tl->firing = 1;
1144                 list_move_tail(&tl->entry, firing);
1145         }
1146
1147         /*
1148          * Check for the special case process timers.
1149          */
1150         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1151                          SIGPROF);
1152         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1153                          SIGVTALRM);
1154         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1155         if (soft != RLIM_INFINITY) {
1156                 unsigned long psecs = cputime_to_secs(ptime);
1157                 unsigned long hard =
1158                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1159                 cputime_t x;
1160                 if (psecs >= hard) {
1161                         /*
1162                          * At the hard limit, we just die.
1163                          * No need to calculate anything else now.
1164                          */
1165                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1166                         return;
1167                 }
1168                 if (psecs >= soft) {
1169                         /*
1170                          * At the soft limit, send a SIGXCPU every second.
1171                          */
1172                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1173                         if (soft < hard) {
1174                                 soft++;
1175                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1176                         }
1177                 }
1178                 x = secs_to_cputime(soft);
1179                 if (cputime_eq(prof_expires, cputime_zero) ||
1180                     cputime_lt(x, prof_expires)) {
1181                         prof_expires = x;
1182                 }
1183         }
1184
1185         if (!cputime_eq(prof_expires, cputime_zero) &&
1186             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1187              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1188                 sig->cputime_expires.prof_exp = prof_expires;
1189         if (!cputime_eq(virt_expires, cputime_zero) &&
1190             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1191              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1192                 sig->cputime_expires.virt_exp = virt_expires;
1193         if (sched_expires != 0 &&
1194             (sig->cputime_expires.sched_exp == 0 ||
1195              sig->cputime_expires.sched_exp > sched_expires))
1196                 sig->cputime_expires.sched_exp = sched_expires;
1197 }
1198
1199 /*
1200  * This is called from the signal code (via do_schedule_next_timer)
1201  * when the last timer signal was delivered and we have to reload the timer.
1202  */
1203 void posix_cpu_timer_schedule(struct k_itimer *timer)
1204 {
1205         struct task_struct *p = timer->it.cpu.task;
1206         union cpu_time_count now;
1207
1208         if (unlikely(p == NULL))
1209                 /*
1210                  * The task was cleaned up already, no future firings.
1211                  */
1212                 goto out;
1213
1214         /*
1215          * Fetch the current sample and update the timer's expiry time.
1216          */
1217         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1218                 cpu_clock_sample(timer->it_clock, p, &now);
1219                 bump_cpu_timer(timer, now);
1220                 if (unlikely(p->exit_state)) {
1221                         clear_dead_task(timer, now);
1222                         goto out;
1223                 }
1224                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1225         } else {
1226                 read_lock(&tasklist_lock);
1227                 if (unlikely(p->signal == NULL)) {
1228                         /*
1229                          * The process has been reaped.
1230                          * We can't even collect a sample any more.
1231                          */
1232                         put_task_struct(p);
1233                         timer->it.cpu.task = p = NULL;
1234                         timer->it.cpu.expires.sched = 0;
1235                         goto out_unlock;
1236                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1237                         /*
1238                          * We've noticed that the thread is dead, but
1239                          * not yet reaped.  Take this opportunity to
1240                          * drop our task ref.
1241                          */
1242                         clear_dead_task(timer, now);
1243                         goto out_unlock;
1244                 }
1245                 cpu_timer_sample_group(timer->it_clock, p, &now);
1246                 bump_cpu_timer(timer, now);
1247                 /* Leave the tasklist_lock locked for the call below.  */
1248         }
1249
1250         /*
1251          * Now re-arm for the new expiry time.
1252          */
1253         arm_timer(timer);
1254
1255 out_unlock:
1256         read_unlock(&tasklist_lock);
1257
1258 out:
1259         timer->it_overrun_last = timer->it_overrun;
1260         timer->it_overrun = -1;
1261         ++timer->it_requeue_pending;
1262 }
1263
1264 /**
1265  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1266  *
1267  * @cputime:    The struct to compare.
1268  *
1269  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1270  * are zero, false if any field is nonzero.
1271  */
1272 static inline int task_cputime_zero(const struct task_cputime *cputime)
1273 {
1274         if (cputime_eq(cputime->utime, cputime_zero) &&
1275             cputime_eq(cputime->stime, cputime_zero) &&
1276             cputime->sum_exec_runtime == 0)
1277                 return 1;
1278         return 0;
1279 }
1280
1281 /**
1282  * task_cputime_expired - Compare two task_cputime entities.
1283  *
1284  * @sample:     The task_cputime structure to be checked for expiration.
1285  * @expires:    Expiration times, against which @sample will be checked.
1286  *
1287  * Checks @sample against @expires to see if any field of @sample has expired.
1288  * Returns true if any field of the former is greater than the corresponding
1289  * field of the latter if the latter field is set.  Otherwise returns false.
1290  */
1291 static inline int task_cputime_expired(const struct task_cputime *sample,
1292                                         const struct task_cputime *expires)
1293 {
1294         if (!cputime_eq(expires->utime, cputime_zero) &&
1295             cputime_ge(sample->utime, expires->utime))
1296                 return 1;
1297         if (!cputime_eq(expires->stime, cputime_zero) &&
1298             cputime_ge(cputime_add(sample->utime, sample->stime),
1299                        expires->stime))
1300                 return 1;
1301         if (expires->sum_exec_runtime != 0 &&
1302             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1303                 return 1;
1304         return 0;
1305 }
1306
1307 /**
1308  * fastpath_timer_check - POSIX CPU timers fast path.
1309  *
1310  * @tsk:        The task (thread) being checked.
1311  *
1312  * Check the task and thread group timers.  If both are zero (there are no
1313  * timers set) return false.  Otherwise snapshot the task and thread group
1314  * timers and compare them with the corresponding expiration times.  Return
1315  * true if a timer has expired, else return false.
1316  */
1317 static inline int fastpath_timer_check(struct task_struct *tsk)
1318 {
1319         struct signal_struct *sig;
1320
1321         /* tsk == current, ensure it is safe to use ->signal/sighand */
1322         if (unlikely(tsk->exit_state))
1323                 return 0;
1324
1325         if (!task_cputime_zero(&tsk->cputime_expires)) {
1326                 struct task_cputime task_sample = {
1327                         .utime = tsk->utime,
1328                         .stime = tsk->stime,
1329                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1330                 };
1331
1332                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1333                         return 1;
1334         }
1335
1336         sig = tsk->signal;
1337         if (!task_cputime_zero(&sig->cputime_expires)) {
1338                 struct task_cputime group_sample;
1339
1340                 thread_group_cputimer(tsk, &group_sample);
1341                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1342                         return 1;
1343         }
1344
1345         return 0;
1346 }
1347
1348 /*
1349  * This is called from the timer interrupt handler.  The irq handler has
1350  * already updated our counts.  We need to check if any timers fire now.
1351  * Interrupts are disabled.
1352  */
1353 void run_posix_cpu_timers(struct task_struct *tsk)
1354 {
1355         LIST_HEAD(firing);
1356         struct k_itimer *timer, *next;
1357
1358         BUG_ON(!irqs_disabled());
1359
1360         /*
1361          * The fast path checks that there are no expired thread or thread
1362          * group timers.  If that's so, just return.
1363          */
1364         if (!fastpath_timer_check(tsk))
1365                 return;
1366
1367         spin_lock(&tsk->sighand->siglock);
1368         /*
1369          * Here we take off tsk->signal->cpu_timers[N] and
1370          * tsk->cpu_timers[N] all the timers that are firing, and
1371          * put them on the firing list.
1372          */
1373         check_thread_timers(tsk, &firing);
1374         check_process_timers(tsk, &firing);
1375
1376         /*
1377          * We must release these locks before taking any timer's lock.
1378          * There is a potential race with timer deletion here, as the
1379          * siglock now protects our private firing list.  We have set
1380          * the firing flag in each timer, so that a deletion attempt
1381          * that gets the timer lock before we do will give it up and
1382          * spin until we've taken care of that timer below.
1383          */
1384         spin_unlock(&tsk->sighand->siglock);
1385
1386         /*
1387          * Now that all the timers on our list have the firing flag,
1388          * noone will touch their list entries but us.  We'll take
1389          * each timer's lock before clearing its firing flag, so no
1390          * timer call will interfere.
1391          */
1392         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1393                 int cpu_firing;
1394
1395                 spin_lock(&timer->it_lock);
1396                 list_del_init(&timer->it.cpu.entry);
1397                 cpu_firing = timer->it.cpu.firing;
1398                 timer->it.cpu.firing = 0;
1399                 /*
1400                  * The firing flag is -1 if we collided with a reset
1401                  * of the timer, which already reported this
1402                  * almost-firing as an overrun.  So don't generate an event.
1403                  */
1404                 if (likely(cpu_firing >= 0))
1405                         cpu_timer_fire(timer);
1406                 spin_unlock(&timer->it_lock);
1407         }
1408 }
1409
1410 /*
1411  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1412  * The tsk->sighand->siglock must be held by the caller.
1413  */
1414 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1415                            cputime_t *newval, cputime_t *oldval)
1416 {
1417         union cpu_time_count now;
1418
1419         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1420         cpu_timer_sample_group(clock_idx, tsk, &now);
1421
1422         if (oldval) {
1423                 /*
1424                  * We are setting itimer. The *oldval is absolute and we update
1425                  * it to be relative, *newval argument is relative and we update
1426                  * it to be absolute.
1427                  */
1428                 if (!cputime_eq(*oldval, cputime_zero)) {
1429                         if (cputime_le(*oldval, now.cpu)) {
1430                                 /* Just about to fire. */
1431                                 *oldval = cputime_one_jiffy;
1432                         } else {
1433                                 *oldval = cputime_sub(*oldval, now.cpu);
1434                         }
1435                 }
1436
1437                 if (cputime_eq(*newval, cputime_zero))
1438                         return;
1439                 *newval = cputime_add(*newval, now.cpu);
1440         }
1441
1442         /*
1443          * Update expiration cache if we are the earliest timer, or eventually
1444          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1445          */
1446         switch (clock_idx) {
1447         case CPUCLOCK_PROF:
1448                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1449                         tsk->signal->cputime_expires.prof_exp = *newval;
1450                 break;
1451         case CPUCLOCK_VIRT:
1452                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1453                         tsk->signal->cputime_expires.virt_exp = *newval;
1454                 break;
1455         }
1456 }
1457
1458 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1459                             struct timespec *rqtp, struct itimerspec *it)
1460 {
1461         struct k_itimer timer;
1462         int error;
1463
1464         /*
1465          * Set up a temporary timer and then wait for it to go off.
1466          */
1467         memset(&timer, 0, sizeof timer);
1468         spin_lock_init(&timer.it_lock);
1469         timer.it_clock = which_clock;
1470         timer.it_overrun = -1;
1471         error = posix_cpu_timer_create(&timer);
1472         timer.it_process = current;
1473         if (!error) {
1474                 static struct itimerspec zero_it;
1475
1476                 memset(it, 0, sizeof *it);
1477                 it->it_value = *rqtp;
1478
1479                 spin_lock_irq(&timer.it_lock);
1480                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1481                 if (error) {
1482                         spin_unlock_irq(&timer.it_lock);
1483                         return error;
1484                 }
1485
1486                 while (!signal_pending(current)) {
1487                         if (timer.it.cpu.expires.sched == 0) {
1488                                 /*
1489                                  * Our timer fired and was reset.
1490                                  */
1491                                 spin_unlock_irq(&timer.it_lock);
1492                                 return 0;
1493                         }
1494
1495                         /*
1496                          * Block until cpu_timer_fire (or a signal) wakes us.
1497                          */
1498                         __set_current_state(TASK_INTERRUPTIBLE);
1499                         spin_unlock_irq(&timer.it_lock);
1500                         schedule();
1501                         spin_lock_irq(&timer.it_lock);
1502                 }
1503
1504                 /*
1505                  * We were interrupted by a signal.
1506                  */
1507                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1508                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1509                 spin_unlock_irq(&timer.it_lock);
1510
1511                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1512                         /*
1513                          * It actually did fire already.
1514                          */
1515                         return 0;
1516                 }
1517
1518                 error = -ERESTART_RESTARTBLOCK;
1519         }
1520
1521         return error;
1522 }
1523
1524 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1525                      struct timespec *rqtp, struct timespec __user *rmtp)
1526 {
1527         struct restart_block *restart_block =
1528             &current_thread_info()->restart_block;
1529         struct itimerspec it;
1530         int error;
1531
1532         /*
1533          * Diagnose required errors first.
1534          */
1535         if (CPUCLOCK_PERTHREAD(which_clock) &&
1536             (CPUCLOCK_PID(which_clock) == 0 ||
1537              CPUCLOCK_PID(which_clock) == current->pid))
1538                 return -EINVAL;
1539
1540         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1541
1542         if (error == -ERESTART_RESTARTBLOCK) {
1543
1544                 if (flags & TIMER_ABSTIME)
1545                         return -ERESTARTNOHAND;
1546                 /*
1547                  * Report back to the user the time still remaining.
1548                  */
1549                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1550                         return -EFAULT;
1551
1552                 restart_block->fn = posix_cpu_nsleep_restart;
1553                 restart_block->arg0 = which_clock;
1554                 restart_block->arg1 = (unsigned long) rmtp;
1555                 restart_block->arg2 = rqtp->tv_sec;
1556                 restart_block->arg3 = rqtp->tv_nsec;
1557         }
1558         return error;
1559 }
1560
1561 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1562 {
1563         clockid_t which_clock = restart_block->arg0;
1564         struct timespec __user *rmtp;
1565         struct timespec t;
1566         struct itimerspec it;
1567         int error;
1568
1569         rmtp = (struct timespec __user *) restart_block->arg1;
1570         t.tv_sec = restart_block->arg2;
1571         t.tv_nsec = restart_block->arg3;
1572
1573         restart_block->fn = do_no_restart_syscall;
1574         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1575
1576         if (error == -ERESTART_RESTARTBLOCK) {
1577                 /*
1578                  * Report back to the user the time still remaining.
1579                  */
1580                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1581                         return -EFAULT;
1582
1583                 restart_block->fn = posix_cpu_nsleep_restart;
1584                 restart_block->arg0 = which_clock;
1585                 restart_block->arg1 = (unsigned long) rmtp;
1586                 restart_block->arg2 = t.tv_sec;
1587                 restart_block->arg3 = t.tv_nsec;
1588         }
1589         return error;
1590
1591 }
1592
1593
1594 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1595 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1596
1597 static int process_cpu_clock_getres(const clockid_t which_clock,
1598                                     struct timespec *tp)
1599 {
1600         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1601 }
1602 static int process_cpu_clock_get(const clockid_t which_clock,
1603                                  struct timespec *tp)
1604 {
1605         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1606 }
1607 static int process_cpu_timer_create(struct k_itimer *timer)
1608 {
1609         timer->it_clock = PROCESS_CLOCK;
1610         return posix_cpu_timer_create(timer);
1611 }
1612 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1613                               struct timespec *rqtp,
1614                               struct timespec __user *rmtp)
1615 {
1616         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1617 }
1618 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1619 {
1620         return -EINVAL;
1621 }
1622 static int thread_cpu_clock_getres(const clockid_t which_clock,
1623                                    struct timespec *tp)
1624 {
1625         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1626 }
1627 static int thread_cpu_clock_get(const clockid_t which_clock,
1628                                 struct timespec *tp)
1629 {
1630         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1631 }
1632 static int thread_cpu_timer_create(struct k_itimer *timer)
1633 {
1634         timer->it_clock = THREAD_CLOCK;
1635         return posix_cpu_timer_create(timer);
1636 }
1637 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1638                               struct timespec *rqtp, struct timespec __user *rmtp)
1639 {
1640         return -EINVAL;
1641 }
1642 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1643 {
1644         return -EINVAL;
1645 }
1646
1647 static __init int init_posix_cpu_timers(void)
1648 {
1649         struct k_clock process = {
1650                 .clock_getres = process_cpu_clock_getres,
1651                 .clock_get = process_cpu_clock_get,
1652                 .clock_set = do_posix_clock_nosettime,
1653                 .timer_create = process_cpu_timer_create,
1654                 .nsleep = process_cpu_nsleep,
1655                 .nsleep_restart = process_cpu_nsleep_restart,
1656         };
1657         struct k_clock thread = {
1658                 .clock_getres = thread_cpu_clock_getres,
1659                 .clock_get = thread_cpu_clock_get,
1660                 .clock_set = do_posix_clock_nosettime,
1661                 .timer_create = thread_cpu_timer_create,
1662                 .nsleep = thread_cpu_nsleep,
1663                 .nsleep_restart = thread_cpu_nsleep_restart,
1664         };
1665         struct timespec ts;
1666
1667         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1668         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1669
1670         cputime_to_timespec(cputime_one_jiffy, &ts);
1671         onecputick = ts.tv_nsec;
1672         WARN_ON(ts.tv_sec != 0);
1673
1674         return 0;
1675 }
1676 __initcall(init_posix_cpu_timers);