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(struct task_struct *task, unsigned long rlim_new)
  20 {
  21         cputime_t cputime = secs_to_cputime(rlim_new);
  22
  23         spin_lock_irq(&task->sighand->siglock);
  24         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
  25         spin_unlock_irq(&task->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         rcu_read_lock();
  41         p = find_task_by_vpid(pid);
  42         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
  43                    same_thread_group(p, current) : has_group_leader_pid(p))) {
  44                 error = -EINVAL;
  45         }
  46         rcu_read_unlock();
  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 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 += 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 -= b.cpu;
 102         }
 103         return a;
 104 }
 105
 106 /*
 107  * Update expiry time from increment, and increase overrun count,
 108  * given the current clock sample.
 109  */
 110 static void bump_cpu_timer(struct k_itimer *timer,
 111                                   union cpu_time_count now)
 112 {
 113         int i;
 114
 115         if (timer->it.cpu.incr.sched == 0)
 116                 return;
 117
 118         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 119                 unsigned long long delta, incr;
 120
 121                 if (now.sched < timer->it.cpu.expires.sched)
 122                         return;
 123                 incr = timer->it.cpu.incr.sched;
 124                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 125                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 126                 for (i = 0; incr < delta - incr; i++)
 127                         incr = incr << 1;
 128                 for (; i >= 0; incr >>= 1, i--) {
 129                         if (delta < incr)
 130                                 continue;
 131                         timer->it.cpu.expires.sched += incr;
 132                         timer->it_overrun += 1 << i;
 133                         delta -= incr;
 134                 }
 135         } else {
 136                 cputime_t delta, incr;
 137
 138                 if (now.cpu < timer->it.cpu.expires.cpu)
 139                         return;
 140                 incr = timer->it.cpu.incr.cpu;
 141                 delta = now.cpu + incr - timer->it.cpu.expires.cpu;
 142                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 143                 for (i = 0; incr < delta - incr; i++)
 144                              incr += incr;
 145                 for (; i >= 0; incr = incr >> 1, i--) {
 146                         if (delta < incr)
 147                                 continue;
 148                         timer->it.cpu.expires.cpu += incr;
 149                         timer->it_overrun += 1 << i;
 150                         delta -= incr;
 151                 }
 152         }
 153 }
 154
 155 static inline cputime_t prof_ticks(struct task_struct *p)
 156 {
 157         return p->utime + p->stime;
 158 }
 159 static inline cputime_t virt_ticks(struct task_struct *p)
 160 {
 161         return p->utime;
 162 }
 163
 164 static int
 165 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 166 {
 167         int error = check_clock(which_clock);
 168         if (!error) {
 169                 tp->tv_sec = 0;
 170                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 171                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 172                         /*
 173                          * If sched_clock is using a cycle counter, we
 174                          * don't have any idea of its true resolution
 175                          * exported, but it is much more than 1s/HZ.
 176                          */
 177                         tp->tv_nsec = 1;
 178                 }
 179         }
 180         return error;
 181 }
 182
 183 static int
 184 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 185 {
 186         /*
 187          * You can never reset a CPU clock, but we check for other errors
 188          * in the call before failing with EPERM.
 189          */
 190         int error = check_clock(which_clock);
 191         if (error == 0) {
 192                 error = -EPERM;
 193         }
 194         return error;
 195 }
 196
 197
 198 /*
 199  * Sample a per-thread clock for the given task.
 200  */
 201 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 202                             union cpu_time_count *cpu)
 203 {
 204         switch (CPUCLOCK_WHICH(which_clock)) {
 205         default:
 206                 return -EINVAL;
 207         case CPUCLOCK_PROF:
 208                 cpu->cpu = prof_ticks(p);
 209                 break;
 210         case CPUCLOCK_VIRT:
 211                 cpu->cpu = virt_ticks(p);
 212                 break;
 213         case CPUCLOCK_SCHED:
 214                 cpu->sched = task_sched_runtime(p);
 215                 break;
 216         }
 217         return 0;
 218 }
 219
 220 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 221 {
 222         if (b->utime > a->utime)
 223                 a->utime = b->utime;
 224
 225         if (b->stime > a->stime)
 226                 a->stime = b->stime;
 227
 228         if (b->sum_exec_runtime > a->sum_exec_runtime)
 229                 a->sum_exec_runtime = b->sum_exec_runtime;
 230 }
 231
 232 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 233 {
 234         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 235         struct task_cputime sum;
 236         unsigned long flags;
 237
 238         if (!cputimer->running) {
 239                 /*
 240                  * The POSIX timer interface allows for absolute time expiry
 241                  * values through the TIMER_ABSTIME flag, therefore we have
 242                  * to synchronize the timer to the clock every time we start
 243                  * it.
 244                  */
 245                 thread_group_cputime(tsk, &sum);
 246                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 247                 cputimer->running = 1;
 248                 update_gt_cputime(&cputimer->cputime, &sum);
 249         } else
 250                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 251         *times = cputimer->cputime;
 252         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 253 }
 254
 255 /*
 256  * Sample a process (thread group) clock for the given group_leader task.
 257  * Must be called with tasklist_lock held for reading.
 258  */
 259 static int cpu_clock_sample_group(const clockid_t which_clock,
 260                                   struct task_struct *p,
 261                                   union cpu_time_count *cpu)
 262 {
 263         struct task_cputime cputime;
 264
 265         switch (CPUCLOCK_WHICH(which_clock)) {
 266         default:
 267                 return -EINVAL;
 268         case CPUCLOCK_PROF:
 269                 thread_group_cputime(p, &cputime);
 270                 cpu->cpu = cputime.utime + cputime.stime;
 271                 break;
 272         case CPUCLOCK_VIRT:
 273                 thread_group_cputime(p, &cputime);
 274                 cpu->cpu = cputime.utime;
 275                 break;
 276         case CPUCLOCK_SCHED:
 277                 thread_group_cputime(p, &cputime);
 278                 cpu->sched = cputime.sum_exec_runtime;
 279                 break;
 280         }
 281         return 0;
 282 }
 283
 284
 285 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 286 {
 287         const pid_t pid = CPUCLOCK_PID(which_clock);
 288         int error = -EINVAL;
 289         union cpu_time_count rtn;
 290
 291         if (pid == 0) {
 292                 /*
 293                  * Special case constant value for our own clocks.
 294                  * We don't have to do any lookup to find ourselves.
 295                  */
 296                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 297                         /*
 298                          * Sampling just ourselves we can do with no locking.
 299                          */
 300                         error = cpu_clock_sample(which_clock,
 301                                                  current, &rtn);
 302                 } else {
 303                         read_lock(&tasklist_lock);
 304                         error = cpu_clock_sample_group(which_clock,
 305                                                        current, &rtn);
 306                         read_unlock(&tasklist_lock);
 307                 }
 308         } else {
 309                 /*
 310                  * Find the given PID, and validate that the caller
 311                  * should be able to see it.
 312                  */
 313                 struct task_struct *p;
 314                 rcu_read_lock();
 315                 p = find_task_by_vpid(pid);
 316                 if (p) {
 317                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 318                                 if (same_thread_group(p, current)) {
 319                                         error = cpu_clock_sample(which_clock,
 320                                                                  p, &rtn);
 321                                 }
 322                         } else {
 323                                 read_lock(&tasklist_lock);
 324                                 if (thread_group_leader(p) && p->sighand) {
 325                                         error =
 326                                             cpu_clock_sample_group(which_clock,
 327                                                                    p, &rtn);
 328                                 }
 329                                 read_unlock(&tasklist_lock);
 330                         }
 331                 }
 332                 rcu_read_unlock();
 333         }
 334
 335         if (error)
 336                 return error;
 337         sample_to_timespec(which_clock, rtn, tp);
 338         return 0;
 339 }
 340
 341
 342 /*
 343  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 344  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 345  * new timer already all-zeros initialized.
 346  */
 347 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 348 {
 349         int ret = 0;
 350         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 351         struct task_struct *p;
 352
 353         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 354                 return -EINVAL;
 355
 356         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 357
 358         rcu_read_lock();
 359         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 360                 if (pid == 0) {
 361                         p = current;
 362                 } else {
 363                         p = find_task_by_vpid(pid);
 364                         if (p && !same_thread_group(p, current))
 365                                 p = NULL;
 366                 }
 367         } else {
 368                 if (pid == 0) {
 369                         p = current->group_leader;
 370                 } else {
 371                         p = find_task_by_vpid(pid);
 372                         if (p && !has_group_leader_pid(p))
 373                                 p = NULL;
 374                 }
 375         }
 376         new_timer->it.cpu.task = p;
 377         if (p) {
 378                 get_task_struct(p);
 379         } else {
 380                 ret = -EINVAL;
 381         }
 382         rcu_read_unlock();
 383
 384         return ret;
 385 }
 386
 387 /*
 388  * Clean up a CPU-clock timer that is about to be destroyed.
 389  * This is called from timer deletion with the timer already locked.
 390  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 391  * and try again.  (This happens when the timer is in the middle of firing.)
 392  */
 393 static int posix_cpu_timer_del(struct k_itimer *timer)
 394 {
 395         struct task_struct *p = timer->it.cpu.task;
 396         int ret = 0;
 397
 398         if (likely(p != NULL)) {
 399                 read_lock(&tasklist_lock);
 400                 if (unlikely(p->sighand == NULL)) {
 401                         /*
 402                          * We raced with the reaping of the task.
 403                          * The deletion should have cleared us off the list.
 404                          */
 405                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 406                 } else {
 407                         spin_lock(&p->sighand->siglock);
 408                         if (timer->it.cpu.firing)
 409                                 ret = TIMER_RETRY;
 410                         else
 411                                 list_del(&timer->it.cpu.entry);
 412                         spin_unlock(&p->sighand->siglock);
 413                 }
 414                 read_unlock(&tasklist_lock);
 415
 416                 if (!ret)
 417                         put_task_struct(p);
 418         }
 419
 420         return ret;
 421 }
 422
 423 /*
 424  * Clean out CPU timers still ticking when a thread exited.  The task
 425  * pointer is cleared, and the expiry time is replaced with the residual
 426  * time for later timer_gettime calls to return.
 427  * This must be called with the siglock held.
 428  */
 429 static void cleanup_timers(struct list_head *head,
 430                            cputime_t utime, cputime_t stime,
 431                            unsigned long long sum_exec_runtime)
 432 {
 433         struct cpu_timer_list *timer, *next;
 434         cputime_t ptime = utime + stime;
 435
 436         list_for_each_entry_safe(timer, next, head, entry) {
 437                 list_del_init(&timer->entry);
 438                 if (timer->expires.cpu < ptime) {
 439                         timer->expires.cpu = 0;
 440                 } else {
 441                         timer->expires.cpu -= ptime;
 442                 }
 443         }
 444
 445         ++head;
 446         list_for_each_entry_safe(timer, next, head, entry) {
 447                 list_del_init(&timer->entry);
 448                 if (timer->expires.cpu < utime) {
 449                         timer->expires.cpu = 0;
 450                 } else {
 451                         timer->expires.cpu -= utime;
 452                 }
 453         }
 454
 455         ++head;
 456         list_for_each_entry_safe(timer, next, head, entry) {
 457                 list_del_init(&timer->entry);
 458                 if (timer->expires.sched < sum_exec_runtime) {
 459                         timer->expires.sched = 0;
 460                 } else {
 461                         timer->expires.sched -= sum_exec_runtime;
 462                 }
 463         }
 464 }
 465
 466 /*
 467  * These are both called with the siglock held, when the current thread
 468  * is being reaped.  When the final (leader) thread in the group is reaped,
 469  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 470  */
 471 void posix_cpu_timers_exit(struct task_struct *tsk)
 472 {
 473         cleanup_timers(tsk->cpu_timers,
 474                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 475
 476 }
 477 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 478 {
 479         struct signal_struct *const sig = tsk->signal;
 480
 481         cleanup_timers(tsk->signal->cpu_timers,
 482                        tsk->utime + sig->utime, tsk->stime + sig->stime,
 483                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 484 }
 485
 486 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 487 {
 488         /*
 489          * That's all for this thread or process.
 490          * We leave our residual in expires to be reported.
 491          */
 492         put_task_struct(timer->it.cpu.task);
 493         timer->it.cpu.task = NULL;
 494         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 495                                              timer->it.cpu.expires,
 496                                              now);
 497 }
 498
 499 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 500 {
 501         return expires == 0 || expires > new_exp;
 502 }
 503
 504 /*
 505  * Insert the timer on the appropriate list before any timers that
 506  * expire later.  This must be called with the tasklist_lock held
 507  * for reading, interrupts disabled and p->sighand->siglock taken.
 508  */
 509 static void arm_timer(struct k_itimer *timer)
 510 {
 511         struct task_struct *p = timer->it.cpu.task;
 512         struct list_head *head, *listpos;
 513         struct task_cputime *cputime_expires;
 514         struct cpu_timer_list *const nt = &timer->it.cpu;
 515         struct cpu_timer_list *next;
 516
 517         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 518                 head = p->cpu_timers;
 519                 cputime_expires = &p->cputime_expires;
 520         } else {
 521                 head = p->signal->cpu_timers;
 522                 cputime_expires = &p->signal->cputime_expires;
 523         }
 524         head += CPUCLOCK_WHICH(timer->it_clock);
 525
 526         listpos = head;
 527         list_for_each_entry(next, head, entry) {
 528                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
 529                         break;
 530                 listpos = &next->entry;
 531         }
 532         list_add(&nt->entry, listpos);
 533
 534         if (listpos == head) {
 535                 union cpu_time_count *exp = &nt->expires;
 536
 537                 /*
 538                  * We are the new earliest-expiring POSIX 1.b timer, hence
 539                  * need to update expiration cache. Take into account that
 540                  * for process timers we share expiration cache with itimers
 541                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 542                  */
 543
 544                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 545                 case CPUCLOCK_PROF:
 546                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
 547                                 cputime_expires->prof_exp = exp->cpu;
 548                         break;
 549                 case CPUCLOCK_VIRT:
 550                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
 551                                 cputime_expires->virt_exp = exp->cpu;
 552                         break;
 553                 case CPUCLOCK_SCHED:
 554                         if (cputime_expires->sched_exp == 0 ||
 555                             cputime_expires->sched_exp > exp->sched)
 556                                 cputime_expires->sched_exp = exp->sched;
 557                         break;
 558                 }
 559         }
 560 }
 561
 562 /*
 563  * The timer is locked, fire it and arrange for its reload.
 564  */
 565 static void cpu_timer_fire(struct k_itimer *timer)
 566 {
 567         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 568                 /*
 569                  * User don't want any signal.
 570                  */
 571                 timer->it.cpu.expires.sched = 0;
 572         } else if (unlikely(timer->sigq == NULL)) {
 573                 /*
 574                  * This a special case for clock_nanosleep,
 575                  * not a normal timer from sys_timer_create.
 576                  */
 577                 wake_up_process(timer->it_process);
 578                 timer->it.cpu.expires.sched = 0;
 579         } else if (timer->it.cpu.incr.sched == 0) {
 580                 /*
 581                  * One-shot timer.  Clear it as soon as it's fired.
 582                  */
 583                 posix_timer_event(timer, 0);
 584                 timer->it.cpu.expires.sched = 0;
 585         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 586                 /*
 587                  * The signal did not get queued because the signal
 588                  * was ignored, so we won't get any callback to
 589                  * reload the timer.  But we need to keep it
 590                  * ticking in case the signal is deliverable next time.
 591                  */
 592                 posix_cpu_timer_schedule(timer);
 593         }
 594 }
 595
 596 /*
 597  * Sample a process (thread group) timer for the given group_leader task.
 598  * Must be called with tasklist_lock held for reading.
 599  */
 600 static int cpu_timer_sample_group(const clockid_t which_clock,
 601                                   struct task_struct *p,
 602                                   union cpu_time_count *cpu)
 603 {
 604         struct task_cputime cputime;
 605
 606         thread_group_cputimer(p, &cputime);
 607         switch (CPUCLOCK_WHICH(which_clock)) {
 608         default:
 609                 return -EINVAL;
 610         case CPUCLOCK_PROF:
 611                 cpu->cpu = cputime.utime + cputime.stime;
 612                 break;
 613         case CPUCLOCK_VIRT:
 614                 cpu->cpu = cputime.utime;
 615                 break;
 616         case CPUCLOCK_SCHED:
 617                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 618                 break;
 619         }
 620         return 0;
 621 }
 622
 623 /*
 624  * Guts of sys_timer_settime for CPU timers.
 625  * This is called with the timer locked and interrupts disabled.
 626  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 627  * and try again.  (This happens when the timer is in the middle of firing.)
 628  */
 629 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 630                                struct itimerspec *new, struct itimerspec *old)
 631 {
 632         struct task_struct *p = timer->it.cpu.task;
 633         union cpu_time_count old_expires, new_expires, old_incr, val;
 634         int ret;
 635
 636         if (unlikely(p == NULL)) {
 637                 /*
 638                  * Timer refers to a dead task's clock.
 639                  */
 640                 return -ESRCH;
 641         }
 642
 643         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 644
 645         read_lock(&tasklist_lock);
 646         /*
 647          * We need the tasklist_lock to protect against reaping that
 648          * clears p->sighand.  If p has just been reaped, we can no
 649          * longer get any information about it at all.
 650          */
 651         if (unlikely(p->sighand == NULL)) {
 652                 read_unlock(&tasklist_lock);
 653                 put_task_struct(p);
 654                 timer->it.cpu.task = NULL;
 655                 return -ESRCH;
 656         }
 657
 658         /*
 659          * Disarm any old timer after extracting its expiry time.
 660          */
 661         BUG_ON(!irqs_disabled());
 662
 663         ret = 0;
 664         old_incr = timer->it.cpu.incr;
 665         spin_lock(&p->sighand->siglock);
 666         old_expires = timer->it.cpu.expires;
 667         if (unlikely(timer->it.cpu.firing)) {
 668                 timer->it.cpu.firing = -1;
 669                 ret = TIMER_RETRY;
 670         } else
 671                 list_del_init(&timer->it.cpu.entry);
 672
 673         /*
 674          * We need to sample the current value to convert the new
 675          * value from to relative and absolute, and to convert the
 676          * old value from absolute to relative.  To set a process
 677          * timer, we need a sample to balance the thread expiry
 678          * times (in arm_timer).  With an absolute time, we must
 679          * check if it's already passed.  In short, we need a sample.
 680          */
 681         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 682                 cpu_clock_sample(timer->it_clock, p, &val);
 683         } else {
 684                 cpu_timer_sample_group(timer->it_clock, p, &val);
 685         }
 686
 687         if (old) {
 688                 if (old_expires.sched == 0) {
 689                         old->it_value.tv_sec = 0;
 690                         old->it_value.tv_nsec = 0;
 691                 } else {
 692                         /*
 693                          * Update the timer in case it has
 694                          * overrun already.  If it has,
 695                          * we'll report it as having overrun
 696                          * and with the next reloaded timer
 697                          * already ticking, though we are
 698                          * swallowing that pending
 699                          * notification here to install the
 700                          * new setting.
 701                          */
 702                         bump_cpu_timer(timer, val);
 703                         if (cpu_time_before(timer->it_clock, val,
 704                                             timer->it.cpu.expires)) {
 705                                 old_expires = cpu_time_sub(
 706                                         timer->it_clock,
 707                                         timer->it.cpu.expires, val);
 708                                 sample_to_timespec(timer->it_clock,
 709                                                    old_expires,
 710                                                    &old->it_value);
 711                         } else {
 712                                 old->it_value.tv_nsec = 1;
 713                                 old->it_value.tv_sec = 0;
 714                         }
 715                 }
 716         }
 717
 718         if (unlikely(ret)) {
 719                 /*
 720                  * We are colliding with the timer actually firing.
 721                  * Punt after filling in the timer's old value, and
 722                  * disable this firing since we are already reporting
 723                  * it as an overrun (thanks to bump_cpu_timer above).
 724                  */
 725                 spin_unlock(&p->sighand->siglock);
 726                 read_unlock(&tasklist_lock);
 727                 goto out;
 728         }
 729
 730         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 731                 cpu_time_add(timer->it_clock, &new_expires, val);
 732         }
 733
 734         /*
 735          * Install the new expiry time (or zero).
 736          * For a timer with no notification action, we don't actually
 737          * arm the timer (we'll just fake it for timer_gettime).
 738          */
 739         timer->it.cpu.expires = new_expires;
 740         if (new_expires.sched != 0 &&
 741             cpu_time_before(timer->it_clock, val, new_expires)) {
 742                 arm_timer(timer);
 743         }
 744
 745         spin_unlock(&p->sighand->siglock);
 746         read_unlock(&tasklist_lock);
 747
 748         /*
 749          * Install the new reload setting, and
 750          * set up the signal and overrun bookkeeping.
 751          */
 752         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 753                                                 &new->it_interval);
 754
 755         /*
 756          * This acts as a modification timestamp for the timer,
 757          * so any automatic reload attempt will punt on seeing
 758          * that we have reset the timer manually.
 759          */
 760         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 761                 ~REQUEUE_PENDING;
 762         timer->it_overrun_last = 0;
 763         timer->it_overrun = -1;
 764
 765         if (new_expires.sched != 0 &&
 766             !cpu_time_before(timer->it_clock, val, new_expires)) {
 767                 /*
 768                  * The designated time already passed, so we notify
 769                  * immediately, even if the thread never runs to
 770                  * accumulate more time on this clock.
 771                  */
 772                 cpu_timer_fire(timer);
 773         }
 774
 775         ret = 0;
 776  out:
 777         if (old) {
 778                 sample_to_timespec(timer->it_clock,
 779                                    old_incr, &old->it_interval);
 780         }
 781         return ret;
 782 }
 783
 784 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 785 {
 786         union cpu_time_count now;
 787         struct task_struct *p = timer->it.cpu.task;
 788         int clear_dead;
 789
 790         /*
 791          * Easy part: convert the reload time.
 792          */
 793         sample_to_timespec(timer->it_clock,
 794                            timer->it.cpu.incr, &itp->it_interval);
 795
 796         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 797                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 798                 return;
 799         }
 800
 801         if (unlikely(p == NULL)) {
 802                 /*
 803                  * This task already died and the timer will never fire.
 804                  * In this case, expires is actually the dead value.
 805                  */
 806         dead:
 807                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 808                                    &itp->it_value);
 809                 return;
 810         }
 811
 812         /*
 813          * Sample the clock to take the difference with the expiry time.
 814          */
 815         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 816                 cpu_clock_sample(timer->it_clock, p, &now);
 817                 clear_dead = p->exit_state;
 818         } else {
 819                 read_lock(&tasklist_lock);
 820                 if (unlikely(p->sighand == NULL)) {
 821                         /*
 822                          * The process has been reaped.
 823                          * We can't even collect a sample any more.
 824                          * Call the timer disarmed, nothing else to do.
 825                          */
 826                         put_task_struct(p);
 827                         timer->it.cpu.task = NULL;
 828                         timer->it.cpu.expires.sched = 0;
 829                         read_unlock(&tasklist_lock);
 830                         goto dead;
 831                 } else {
 832                         cpu_timer_sample_group(timer->it_clock, p, &now);
 833                         clear_dead = (unlikely(p->exit_state) &&
 834                                       thread_group_empty(p));
 835                 }
 836                 read_unlock(&tasklist_lock);
 837         }
 838
 839         if (unlikely(clear_dead)) {
 840                 /*
 841                  * We've noticed that the thread is dead, but
 842                  * not yet reaped.  Take this opportunity to
 843                  * drop our task ref.
 844                  */
 845                 clear_dead_task(timer, now);
 846                 goto dead;
 847         }
 848
 849         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 850                 sample_to_timespec(timer->it_clock,
 851                                    cpu_time_sub(timer->it_clock,
 852                                                 timer->it.cpu.expires, now),
 853                                    &itp->it_value);
 854         } else {
 855                 /*
 856                  * The timer should have expired already, but the firing
 857                  * hasn't taken place yet.  Say it's just about to expire.
 858                  */
 859                 itp->it_value.tv_nsec = 1;
 860                 itp->it_value.tv_sec = 0;
 861         }
 862 }
 863
 864 /*
 865  * Check for any per-thread CPU timers that have fired and move them off
 866  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 867  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 868  */
 869 static void check_thread_timers(struct task_struct *tsk,
 870                                 struct list_head *firing)
 871 {
 872         int maxfire;
 873         struct list_head *timers = tsk->cpu_timers;
 874         struct signal_struct *const sig = tsk->signal;
 875         unsigned long soft;
 876
 877         maxfire = 20;
 878         tsk->cputime_expires.prof_exp = 0;
 879         while (!list_empty(timers)) {
 880                 struct cpu_timer_list *t = list_first_entry(timers,
 881                                                       struct cpu_timer_list,
 882                                                       entry);
 883                 if (!--maxfire || prof_ticks(tsk) < t->expires.cpu) {
 884                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 885                         break;
 886                 }
 887                 t->firing = 1;
 888                 list_move_tail(&t->entry, firing);
 889         }
 890
 891         ++timers;
 892         maxfire = 20;
 893         tsk->cputime_expires.virt_exp = 0;
 894         while (!list_empty(timers)) {
 895                 struct cpu_timer_list *t = list_first_entry(timers,
 896                                                       struct cpu_timer_list,
 897                                                       entry);
 898                 if (!--maxfire || virt_ticks(tsk) < t->expires.cpu) {
 899                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 900                         break;
 901                 }
 902                 t->firing = 1;
 903                 list_move_tail(&t->entry, firing);
 904         }
 905
 906         ++timers;
 907         maxfire = 20;
 908         tsk->cputime_expires.sched_exp = 0;
 909         while (!list_empty(timers)) {
 910                 struct cpu_timer_list *t = list_first_entry(timers,
 911                                                       struct cpu_timer_list,
 912                                                       entry);
 913                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
 914                         tsk->cputime_expires.sched_exp = t->expires.sched;
 915                         break;
 916                 }
 917                 t->firing = 1;
 918                 list_move_tail(&t->entry, firing);
 919         }
 920
 921         /*
 922          * Check for the special case thread timers.
 923          */
 924         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 925         if (soft != RLIM_INFINITY) {
 926                 unsigned long hard =
 927                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 928
 929                 if (hard != RLIM_INFINITY &&
 930                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 931                         /*
 932                          * At the hard limit, we just die.
 933                          * No need to calculate anything else now.
 934                          */
 935                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 936                         return;
 937                 }
 938                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 939                         /*
 940                          * At the soft limit, send a SIGXCPU every second.
 941                          */
 942                         if (soft < hard) {
 943                                 soft += USEC_PER_SEC;
 944                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 945                         }
 946                         printk(KERN_INFO
 947                                 "RT Watchdog Timeout: %s[%d]\n",
 948                                 tsk->comm, task_pid_nr(tsk));
 949                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 950                 }
 951         }
 952 }
 953
 954 static void stop_process_timers(struct signal_struct *sig)
 955 {
 956         struct thread_group_cputimer *cputimer = &sig->cputimer;
 957         unsigned long flags;
 958
 959         raw_spin_lock_irqsave(&cputimer->lock, flags);
 960         cputimer->running = 0;
 961         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 962 }
 963
 964 static u32 onecputick;
 965
 966 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
 967                              cputime_t *expires, cputime_t cur_time, int signo)
 968 {
 969         if (!it->expires)
 970                 return;
 971
 972         if (cur_time >= it->expires) {
 973                 if (it->incr) {
 974                         it->expires += it->incr;
 975                         it->error += it->incr_error;
 976                         if (it->error >= onecputick) {
 977                                 it->expires -= cputime_one_jiffy;
 978                                 it->error -= onecputick;
 979                         }
 980                 } else {
 981                         it->expires = 0;
 982                 }
 983
 984                 trace_itimer_expire(signo == SIGPROF ?
 985                                     ITIMER_PROF : ITIMER_VIRTUAL,
 986                                     tsk->signal->leader_pid, cur_time);
 987                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
 988         }
 989
 990         if (it->expires && (!*expires || it->expires < *expires)) {
 991                 *expires = it->expires;
 992         }
 993 }
 994
 995 /**
 996  * task_cputime_zero - Check a task_cputime struct for all zero fields.
 997  *
 998  * @cputime:    The struct to compare.
 999  *
1000  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1001  * are zero, false if any field is nonzero.
1002  */
1003 static inline int task_cputime_zero(const struct task_cputime *cputime)
1004 {
1005         if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
1006                 return 1;
1007         return 0;
1008 }
1009
1010 /*
1011  * Check for any per-thread CPU timers that have fired and move them
1012  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1013  * have already been taken off.
1014  */
1015 static void check_process_timers(struct task_struct *tsk,
1016                                  struct list_head *firing)
1017 {
1018         int maxfire;
1019         struct signal_struct *const sig = tsk->signal;
1020         cputime_t utime, ptime, virt_expires, prof_expires;
1021         unsigned long long sum_sched_runtime, sched_expires;
1022         struct list_head *timers = sig->cpu_timers;
1023         struct task_cputime cputime;
1024         unsigned long soft;
1025
1026         /*
1027          * Collect the current process totals.
1028          */
1029         thread_group_cputimer(tsk, &cputime);
1030         utime = cputime.utime;
1031         ptime = utime + cputime.stime;
1032         sum_sched_runtime = cputime.sum_exec_runtime;
1033         maxfire = 20;
1034         prof_expires = 0;
1035         while (!list_empty(timers)) {
1036                 struct cpu_timer_list *tl = list_first_entry(timers,
1037                                                       struct cpu_timer_list,
1038                                                       entry);
1039                 if (!--maxfire || ptime < tl->expires.cpu) {
1040                         prof_expires = tl->expires.cpu;
1041                         break;
1042                 }
1043                 tl->firing = 1;
1044                 list_move_tail(&tl->entry, firing);
1045         }
1046
1047         ++timers;
1048         maxfire = 20;
1049         virt_expires = 0;
1050         while (!list_empty(timers)) {
1051                 struct cpu_timer_list *tl = list_first_entry(timers,
1052                                                       struct cpu_timer_list,
1053                                                       entry);
1054                 if (!--maxfire || utime < tl->expires.cpu) {
1055                         virt_expires = tl->expires.cpu;
1056                         break;
1057                 }
1058                 tl->firing = 1;
1059                 list_move_tail(&tl->entry, firing);
1060         }
1061
1062         ++timers;
1063         maxfire = 20;
1064         sched_expires = 0;
1065         while (!list_empty(timers)) {
1066                 struct cpu_timer_list *tl = list_first_entry(timers,
1067                                                       struct cpu_timer_list,
1068                                                       entry);
1069                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1070                         sched_expires = tl->expires.sched;
1071                         break;
1072                 }
1073                 tl->firing = 1;
1074                 list_move_tail(&tl->entry, firing);
1075         }
1076
1077         /*
1078          * Check for the special case process timers.
1079          */
1080         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1081                          SIGPROF);
1082         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1083                          SIGVTALRM);
1084         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1085         if (soft != RLIM_INFINITY) {
1086                 unsigned long psecs = cputime_to_secs(ptime);
1087                 unsigned long hard =
1088                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1089                 cputime_t x;
1090                 if (psecs >= hard) {
1091                         /*
1092                          * At the hard limit, we just die.
1093                          * No need to calculate anything else now.
1094                          */
1095                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1096                         return;
1097                 }
1098                 if (psecs >= soft) {
1099                         /*
1100                          * At the soft limit, send a SIGXCPU every second.
1101                          */
1102                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1103                         if (soft < hard) {
1104                                 soft++;
1105                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1106                         }
1107                 }
1108                 x = secs_to_cputime(soft);
1109                 if (!prof_expires || x < prof_expires) {
1110                         prof_expires = x;
1111                 }
1112         }
1113
1114         sig->cputime_expires.prof_exp = prof_expires;
1115         sig->cputime_expires.virt_exp = virt_expires;
1116         sig->cputime_expires.sched_exp = sched_expires;
1117         if (task_cputime_zero(&sig->cputime_expires))
1118                 stop_process_timers(sig);
1119 }
1120
1121 /*
1122  * This is called from the signal code (via do_schedule_next_timer)
1123  * when the last timer signal was delivered and we have to reload the timer.
1124  */
1125 void posix_cpu_timer_schedule(struct k_itimer *timer)
1126 {
1127         struct task_struct *p = timer->it.cpu.task;
1128         union cpu_time_count now;
1129
1130         if (unlikely(p == NULL))
1131                 /*
1132                  * The task was cleaned up already, no future firings.
1133                  */
1134                 goto out;
1135
1136         /*
1137          * Fetch the current sample and update the timer's expiry time.
1138          */
1139         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1140                 cpu_clock_sample(timer->it_clock, p, &now);
1141                 bump_cpu_timer(timer, now);
1142                 if (unlikely(p->exit_state)) {
1143                         clear_dead_task(timer, now);
1144                         goto out;
1145                 }
1146                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1147                 spin_lock(&p->sighand->siglock);
1148         } else {
1149                 read_lock(&tasklist_lock);
1150                 if (unlikely(p->sighand == NULL)) {
1151                         /*
1152                          * The process has been reaped.
1153                          * We can't even collect a sample any more.
1154                          */
1155                         put_task_struct(p);
1156                         timer->it.cpu.task = p = NULL;
1157                         timer->it.cpu.expires.sched = 0;
1158                         goto out_unlock;
1159                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1160                         /*
1161                          * We've noticed that the thread is dead, but
1162                          * not yet reaped.  Take this opportunity to
1163                          * drop our task ref.
1164                          */
1165                         clear_dead_task(timer, now);
1166                         goto out_unlock;
1167                 }
1168                 spin_lock(&p->sighand->siglock);
1169                 cpu_timer_sample_group(timer->it_clock, p, &now);
1170                 bump_cpu_timer(timer, now);
1171                 /* Leave the tasklist_lock locked for the call below.  */
1172         }
1173
1174         /*
1175          * Now re-arm for the new expiry time.
1176          */
1177         BUG_ON(!irqs_disabled());
1178         arm_timer(timer);
1179         spin_unlock(&p->sighand->siglock);
1180
1181 out_unlock:
1182         read_unlock(&tasklist_lock);
1183
1184 out:
1185         timer->it_overrun_last = timer->it_overrun;
1186         timer->it_overrun = -1;
1187         ++timer->it_requeue_pending;
1188 }
1189
1190 /**
1191  * task_cputime_expired - Compare two task_cputime entities.
1192  *
1193  * @sample:     The task_cputime structure to be checked for expiration.
1194  * @expires:    Expiration times, against which @sample will be checked.
1195  *
1196  * Checks @sample against @expires to see if any field of @sample has expired.
1197  * Returns true if any field of the former is greater than the corresponding
1198  * field of the latter if the latter field is set.  Otherwise returns false.
1199  */
1200 static inline int task_cputime_expired(const struct task_cputime *sample,
1201                                         const struct task_cputime *expires)
1202 {
1203         if (expires->utime && sample->utime >= expires->utime)
1204                 return 1;
1205         if (expires->stime && sample->utime + sample->stime >= expires->stime)
1206                 return 1;
1207         if (expires->sum_exec_runtime != 0 &&
1208             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1209                 return 1;
1210         return 0;
1211 }
1212
1213 /**
1214  * fastpath_timer_check - POSIX CPU timers fast path.
1215  *
1216  * @tsk:        The task (thread) being checked.
1217  *
1218  * Check the task and thread group timers.  If both are zero (there are no
1219  * timers set) return false.  Otherwise snapshot the task and thread group
1220  * timers and compare them with the corresponding expiration times.  Return
1221  * true if a timer has expired, else return false.
1222  */
1223 static inline int fastpath_timer_check(struct task_struct *tsk)
1224 {
1225         struct signal_struct *sig;
1226
1227         if (!task_cputime_zero(&tsk->cputime_expires)) {
1228                 struct task_cputime task_sample = {
1229                         .utime = tsk->utime,
1230                         .stime = tsk->stime,
1231                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1232                 };
1233
1234                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1235                         return 1;
1236         }
1237
1238         sig = tsk->signal;
1239         if (sig->cputimer.running) {
1240                 struct task_cputime group_sample;
1241
1242                 raw_spin_lock(&sig->cputimer.lock);
1243                 group_sample = sig->cputimer.cputime;
1244                 raw_spin_unlock(&sig->cputimer.lock);
1245
1246                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1247                         return 1;
1248         }
1249
1250         return 0;
1251 }
1252
1253 /*
1254  * This is called from the timer interrupt handler.  The irq handler has
1255  * already updated our counts.  We need to check if any timers fire now.
1256  * Interrupts are disabled.
1257  */
1258 void run_posix_cpu_timers(struct task_struct *tsk)
1259 {
1260         LIST_HEAD(firing);
1261         struct k_itimer *timer, *next;
1262         unsigned long flags;
1263
1264         BUG_ON(!irqs_disabled());
1265
1266         /*
1267          * The fast path checks that there are no expired thread or thread
1268          * group timers.  If that's so, just return.
1269          */
1270         if (!fastpath_timer_check(tsk))
1271                 return;
1272
1273         if (!lock_task_sighand(tsk, &flags))
1274                 return;
1275         /*
1276          * Here we take off tsk->signal->cpu_timers[N] and
1277          * tsk->cpu_timers[N] all the timers that are firing, and
1278          * put them on the firing list.
1279          */
1280         check_thread_timers(tsk, &firing);
1281         /*
1282          * If there are any active process wide timers (POSIX 1.b, itimers,
1283          * RLIMIT_CPU) cputimer must be running.
1284          */
1285         if (tsk->signal->cputimer.running)
1286                 check_process_timers(tsk, &firing);
1287
1288         /*
1289          * We must release these locks before taking any timer's lock.
1290          * There is a potential race with timer deletion here, as the
1291          * siglock now protects our private firing list.  We have set
1292          * the firing flag in each timer, so that a deletion attempt
1293          * that gets the timer lock before we do will give it up and
1294          * spin until we've taken care of that timer below.
1295          */
1296         unlock_task_sighand(tsk, &flags);
1297
1298         /*
1299          * Now that all the timers on our list have the firing flag,
1300          * no one will touch their list entries but us.  We'll take
1301          * each timer's lock before clearing its firing flag, so no
1302          * timer call will interfere.
1303          */
1304         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1305                 int cpu_firing;
1306
1307                 spin_lock(&timer->it_lock);
1308                 list_del_init(&timer->it.cpu.entry);
1309                 cpu_firing = timer->it.cpu.firing;
1310                 timer->it.cpu.firing = 0;
1311                 /*
1312                  * The firing flag is -1 if we collided with a reset
1313                  * of the timer, which already reported this
1314                  * almost-firing as an overrun.  So don't generate an event.
1315                  */
1316                 if (likely(cpu_firing >= 0))
1317                         cpu_timer_fire(timer);
1318                 spin_unlock(&timer->it_lock);
1319         }
1320 }
1321
1322 /*
1323  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1324  * The tsk->sighand->siglock must be held by the caller.
1325  */
1326 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1327                            cputime_t *newval, cputime_t *oldval)
1328 {
1329         union cpu_time_count now;
1330
1331         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1332         cpu_timer_sample_group(clock_idx, tsk, &now);
1333
1334         if (oldval) {
1335                 /*
1336                  * We are setting itimer. The *oldval is absolute and we update
1337                  * it to be relative, *newval argument is relative and we update
1338                  * it to be absolute.
1339                  */
1340                 if (*oldval) {
1341                         if (*oldval <= now.cpu) {
1342                                 /* Just about to fire. */
1343                                 *oldval = cputime_one_jiffy;
1344                         } else {
1345                                 *oldval -= now.cpu;
1346                         }
1347                 }
1348
1349                 if (!*newval)
1350                         return;
1351                 *newval += now.cpu;
1352         }
1353
1354         /*
1355          * Update expiration cache if we are the earliest timer, or eventually
1356          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1357          */
1358         switch (clock_idx) {
1359         case CPUCLOCK_PROF:
1360                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1361                         tsk->signal->cputime_expires.prof_exp = *newval;
1362                 break;
1363         case CPUCLOCK_VIRT:
1364                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1365                         tsk->signal->cputime_expires.virt_exp = *newval;
1366                 break;
1367         }
1368 }
1369
1370 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1371                             struct timespec *rqtp, struct itimerspec *it)
1372 {
1373         struct k_itimer timer;
1374         int error;
1375
1376         /*
1377          * Set up a temporary timer and then wait for it to go off.
1378          */
1379         memset(&timer, 0, sizeof timer);
1380         spin_lock_init(&timer.it_lock);
1381         timer.it_clock = which_clock;
1382         timer.it_overrun = -1;
1383         error = posix_cpu_timer_create(&timer);
1384         timer.it_process = current;
1385         if (!error) {
1386                 static struct itimerspec zero_it;
1387
1388                 memset(it, 0, sizeof *it);
1389                 it->it_value = *rqtp;
1390
1391                 spin_lock_irq(&timer.it_lock);
1392                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1393                 if (error) {
1394                         spin_unlock_irq(&timer.it_lock);
1395                         return error;
1396                 }
1397
1398                 while (!signal_pending(current)) {
1399                         if (timer.it.cpu.expires.sched == 0) {
1400                                 /*
1401                                  * Our timer fired and was reset.
1402                                  */
1403                                 spin_unlock_irq(&timer.it_lock);
1404                                 return 0;
1405                         }
1406
1407                         /*
1408                          * Block until cpu_timer_fire (or a signal) wakes us.
1409                          */
1410                         __set_current_state(TASK_INTERRUPTIBLE);
1411                         spin_unlock_irq(&timer.it_lock);
1412                         schedule();
1413                         spin_lock_irq(&timer.it_lock);
1414                 }
1415
1416                 /*
1417                  * We were interrupted by a signal.
1418                  */
1419                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1420                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1421                 spin_unlock_irq(&timer.it_lock);
1422
1423                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1424                         /*
1425                          * It actually did fire already.
1426                          */
1427                         return 0;
1428                 }
1429
1430                 error = -ERESTART_RESTARTBLOCK;
1431         }
1432
1433         return error;
1434 }
1435
1436 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1437
1438 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1439                             struct timespec *rqtp, struct timespec __user *rmtp)
1440 {
1441         struct restart_block *restart_block =
1442                 &current_thread_info()->restart_block;
1443         struct itimerspec it;
1444         int error;
1445
1446         /*
1447          * Diagnose required errors first.
1448          */
1449         if (CPUCLOCK_PERTHREAD(which_clock) &&
1450             (CPUCLOCK_PID(which_clock) == 0 ||
1451              CPUCLOCK_PID(which_clock) == current->pid))
1452                 return -EINVAL;
1453
1454         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1455
1456         if (error == -ERESTART_RESTARTBLOCK) {
1457
1458                 if (flags & TIMER_ABSTIME)
1459                         return -ERESTARTNOHAND;
1460                 /*
1461                  * Report back to the user the time still remaining.
1462                  */
1463                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1464                         return -EFAULT;
1465
1466                 restart_block->fn = posix_cpu_nsleep_restart;
1467                 restart_block->nanosleep.clockid = which_clock;
1468                 restart_block->nanosleep.rmtp = rmtp;
1469                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1470         }
1471         return error;
1472 }
1473
1474 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1475 {
1476         clockid_t which_clock = restart_block->nanosleep.clockid;
1477         struct timespec t;
1478         struct itimerspec it;
1479         int error;
1480
1481         t = ns_to_timespec(restart_block->nanosleep.expires);
1482
1483         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1484
1485         if (error == -ERESTART_RESTARTBLOCK) {
1486                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1487                 /*
1488                  * Report back to the user the time still remaining.
1489                  */
1490                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1491                         return -EFAULT;
1492
1493                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1494         }
1495         return error;
1496
1497 }
1498
1499 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1500 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1501
1502 static int process_cpu_clock_getres(const clockid_t which_clock,
1503                                     struct timespec *tp)
1504 {
1505         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1506 }
1507 static int process_cpu_clock_get(const clockid_t which_clock,
1508                                  struct timespec *tp)
1509 {
1510         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1511 }
1512 static int process_cpu_timer_create(struct k_itimer *timer)
1513 {
1514         timer->it_clock = PROCESS_CLOCK;
1515         return posix_cpu_timer_create(timer);
1516 }
1517 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1518                               struct timespec *rqtp,
1519                               struct timespec __user *rmtp)
1520 {
1521         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1522 }
1523 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1524 {
1525         return -EINVAL;
1526 }
1527 static int thread_cpu_clock_getres(const clockid_t which_clock,
1528                                    struct timespec *tp)
1529 {
1530         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1531 }
1532 static int thread_cpu_clock_get(const clockid_t which_clock,
1533                                 struct timespec *tp)
1534 {
1535         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1536 }
1537 static int thread_cpu_timer_create(struct k_itimer *timer)
1538 {
1539         timer->it_clock = THREAD_CLOCK;
1540         return posix_cpu_timer_create(timer);
1541 }
1542
1543 struct k_clock clock_posix_cpu = {
1544         .clock_getres   = posix_cpu_clock_getres,
1545         .clock_set      = posix_cpu_clock_set,
1546         .clock_get      = posix_cpu_clock_get,
1547         .timer_create   = posix_cpu_timer_create,
1548         .nsleep         = posix_cpu_nsleep,
1549         .nsleep_restart = posix_cpu_nsleep_restart,
1550         .timer_set      = posix_cpu_timer_set,
1551         .timer_del      = posix_cpu_timer_del,
1552         .timer_get      = posix_cpu_timer_get,
1553 };
1554
1555 static __init int init_posix_cpu_timers(void)
1556 {
1557         struct k_clock process = {
1558                 .clock_getres   = process_cpu_clock_getres,
1559                 .clock_get      = process_cpu_clock_get,
1560                 .timer_create   = process_cpu_timer_create,
1561                 .nsleep         = process_cpu_nsleep,
1562                 .nsleep_restart = process_cpu_nsleep_restart,
1563         };
1564         struct k_clock thread = {
1565                 .clock_getres   = thread_cpu_clock_getres,
1566                 .clock_get      = thread_cpu_clock_get,
1567                 .timer_create   = thread_cpu_timer_create,
1568         };
1569         struct timespec ts;
1570
1571         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1572         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1573
1574         cputime_to_timespec(cputime_one_jiffy, &ts);
1575         onecputick = ts.tv_nsec;
1576         WARN_ON(ts.tv_sec != 0);
1577
1578         return 0;
1579 }
1580 __initcall(init_posix_cpu_timers);