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 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 static int
 180 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 181 {
 182         int error = check_clock(which_clock);
 183         if (!error) {
 184                 tp->tv_sec = 0;
 185                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 186                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 187                         /*
 188                          * If sched_clock is using a cycle counter, we
 189                          * don't have any idea of its true resolution
 190                          * exported, but it is much more than 1s/HZ.
 191                          */
 192                         tp->tv_nsec = 1;
 193                 }
 194         }
 195         return error;
 196 }
 197
 198 static int
 199 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 200 {
 201         /*
 202          * You can never reset a CPU clock, but we check for other errors
 203          * in the call before failing with EPERM.
 204          */
 205         int error = check_clock(which_clock);
 206         if (error == 0) {
 207                 error = -EPERM;
 208         }
 209         return error;
 210 }
 211
 212
 213 /*
 214  * Sample a per-thread clock for the given task.
 215  */
 216 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 217                             union cpu_time_count *cpu)
 218 {
 219         switch (CPUCLOCK_WHICH(which_clock)) {
 220         default:
 221                 return -EINVAL;
 222         case CPUCLOCK_PROF:
 223                 cpu->cpu = prof_ticks(p);
 224                 break;
 225         case CPUCLOCK_VIRT:
 226                 cpu->cpu = virt_ticks(p);
 227                 break;
 228         case CPUCLOCK_SCHED:
 229                 cpu->sched = task_sched_runtime(p);
 230                 break;
 231         }
 232         return 0;
 233 }
 234
 235 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
 236 {
 237         struct signal_struct *sig = tsk->signal;
 238         struct task_struct *t;
 239
 240         times->utime = sig->utime;
 241         times->stime = sig->stime;
 242         times->sum_exec_runtime = sig->sum_sched_runtime;
 243
 244         rcu_read_lock();
 245         /* make sure we can trust tsk->thread_group list */
 246         if (!likely(pid_alive(tsk)))
 247                 goto out;
 248
 249         t = tsk;
 250         do {
 251                 times->utime = cputime_add(times->utime, t->utime);
 252                 times->stime = cputime_add(times->stime, t->stime);
 253                 times->sum_exec_runtime += task_sched_runtime(t);
 254         } while_each_thread(tsk, t);
 255 out:
 256         rcu_read_unlock();
 257 }
 258
 259 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
 260 {
 261         if (cputime_gt(b->utime, a->utime))
 262                 a->utime = b->utime;
 263
 264         if (cputime_gt(b->stime, a->stime))
 265                 a->stime = b->stime;
 266
 267         if (b->sum_exec_runtime > a->sum_exec_runtime)
 268                 a->sum_exec_runtime = b->sum_exec_runtime;
 269 }
 270
 271 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
 272 {
 273         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
 274         struct task_cputime sum;
 275         unsigned long flags;
 276
 277         if (!cputimer->running) {
 278                 /*
 279                  * The POSIX timer interface allows for absolute time expiry
 280                  * values through the TIMER_ABSTIME flag, therefore we have
 281                  * to synchronize the timer to the clock every time we start
 282                  * it.
 283                  */
 284                 thread_group_cputime(tsk, &sum);
 285                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 286                 cputimer->running = 1;
 287                 update_gt_cputime(&cputimer->cputime, &sum);
 288         } else
 289                 raw_spin_lock_irqsave(&cputimer->lock, flags);
 290         *times = cputimer->cputime;
 291         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
 292 }
 293
 294 /*
 295  * Sample a process (thread group) clock for the given group_leader task.
 296  * Must be called with tasklist_lock held for reading.
 297  */
 298 static int cpu_clock_sample_group(const clockid_t which_clock,
 299                                   struct task_struct *p,
 300                                   union cpu_time_count *cpu)
 301 {
 302         struct task_cputime cputime;
 303
 304         switch (CPUCLOCK_WHICH(which_clock)) {
 305         default:
 306                 return -EINVAL;
 307         case CPUCLOCK_PROF:
 308                 thread_group_cputime(p, &cputime);
 309                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 310                 break;
 311         case CPUCLOCK_VIRT:
 312                 thread_group_cputime(p, &cputime);
 313                 cpu->cpu = cputime.utime;
 314                 break;
 315         case CPUCLOCK_SCHED:
 316                 thread_group_cputime(p, &cputime);
 317                 cpu->sched = cputime.sum_exec_runtime;
 318                 break;
 319         }
 320         return 0;
 321 }
 322
 323
 324 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 325 {
 326         const pid_t pid = CPUCLOCK_PID(which_clock);
 327         int error = -EINVAL;
 328         union cpu_time_count rtn;
 329
 330         if (pid == 0) {
 331                 /*
 332                  * Special case constant value for our own clocks.
 333                  * We don't have to do any lookup to find ourselves.
 334                  */
 335                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 336                         /*
 337                          * Sampling just ourselves we can do with no locking.
 338                          */
 339                         error = cpu_clock_sample(which_clock,
 340                                                  current, &rtn);
 341                 } else {
 342                         read_lock(&tasklist_lock);
 343                         error = cpu_clock_sample_group(which_clock,
 344                                                        current, &rtn);
 345                         read_unlock(&tasklist_lock);
 346                 }
 347         } else {
 348                 /*
 349                  * Find the given PID, and validate that the caller
 350                  * should be able to see it.
 351                  */
 352                 struct task_struct *p;
 353                 rcu_read_lock();
 354                 p = find_task_by_vpid(pid);
 355                 if (p) {
 356                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 357                                 if (same_thread_group(p, current)) {
 358                                         error = cpu_clock_sample(which_clock,
 359                                                                  p, &rtn);
 360                                 }
 361                         } else {
 362                                 read_lock(&tasklist_lock);
 363                                 if (thread_group_leader(p) && p->sighand) {
 364                                         error =
 365                                             cpu_clock_sample_group(which_clock,
 366                                                                    p, &rtn);
 367                                 }
 368                                 read_unlock(&tasklist_lock);
 369                         }
 370                 }
 371                 rcu_read_unlock();
 372         }
 373
 374         if (error)
 375                 return error;
 376         sample_to_timespec(which_clock, rtn, tp);
 377         return 0;
 378 }
 379
 380
 381 /*
 382  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 383  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
 384  * new timer already all-zeros initialized.
 385  */
 386 static int posix_cpu_timer_create(struct k_itimer *new_timer)
 387 {
 388         int ret = 0;
 389         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 390         struct task_struct *p;
 391
 392         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 393                 return -EINVAL;
 394
 395         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 396
 397         rcu_read_lock();
 398         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 399                 if (pid == 0) {
 400                         p = current;
 401                 } else {
 402                         p = find_task_by_vpid(pid);
 403                         if (p && !same_thread_group(p, current))
 404                                 p = NULL;
 405                 }
 406         } else {
 407                 if (pid == 0) {
 408                         p = current->group_leader;
 409                 } else {
 410                         p = find_task_by_vpid(pid);
 411                         if (p && !has_group_leader_pid(p))
 412                                 p = NULL;
 413                 }
 414         }
 415         new_timer->it.cpu.task = p;
 416         if (p) {
 417                 get_task_struct(p);
 418         } else {
 419                 ret = -EINVAL;
 420         }
 421         rcu_read_unlock();
 422
 423         return ret;
 424 }
 425
 426 /*
 427  * Clean up a CPU-clock timer that is about to be destroyed.
 428  * This is called from timer deletion with the timer already locked.
 429  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 430  * and try again.  (This happens when the timer is in the middle of firing.)
 431  */
 432 static int posix_cpu_timer_del(struct k_itimer *timer)
 433 {
 434         struct task_struct *p = timer->it.cpu.task;
 435         int ret = 0;
 436
 437         if (likely(p != NULL)) {
 438                 read_lock(&tasklist_lock);
 439                 if (unlikely(p->sighand == NULL)) {
 440                         /*
 441                          * We raced with the reaping of the task.
 442                          * The deletion should have cleared us off the list.
 443                          */
 444                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 445                 } else {
 446                         spin_lock(&p->sighand->siglock);
 447                         if (timer->it.cpu.firing)
 448                                 ret = TIMER_RETRY;
 449                         else
 450                                 list_del(&timer->it.cpu.entry);
 451                         spin_unlock(&p->sighand->siglock);
 452                 }
 453                 read_unlock(&tasklist_lock);
 454
 455                 if (!ret)
 456                         put_task_struct(p);
 457         }
 458
 459         return ret;
 460 }
 461
 462 /*
 463  * Clean out CPU timers still ticking when a thread exited.  The task
 464  * pointer is cleared, and the expiry time is replaced with the residual
 465  * time for later timer_gettime calls to return.
 466  * This must be called with the siglock held.
 467  */
 468 static void cleanup_timers(struct list_head *head,
 469                            cputime_t utime, cputime_t stime,
 470                            unsigned long long sum_exec_runtime)
 471 {
 472         struct cpu_timer_list *timer, *next;
 473         cputime_t ptime = cputime_add(utime, stime);
 474
 475         list_for_each_entry_safe(timer, next, head, entry) {
 476                 list_del_init(&timer->entry);
 477                 if (cputime_lt(timer->expires.cpu, ptime)) {
 478                         timer->expires.cpu = cputime_zero;
 479                 } else {
 480                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 481                                                          ptime);
 482                 }
 483         }
 484
 485         ++head;
 486         list_for_each_entry_safe(timer, next, head, entry) {
 487                 list_del_init(&timer->entry);
 488                 if (cputime_lt(timer->expires.cpu, utime)) {
 489                         timer->expires.cpu = cputime_zero;
 490                 } else {
 491                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 492                                                          utime);
 493                 }
 494         }
 495
 496         ++head;
 497         list_for_each_entry_safe(timer, next, head, entry) {
 498                 list_del_init(&timer->entry);
 499                 if (timer->expires.sched < sum_exec_runtime) {
 500                         timer->expires.sched = 0;
 501                 } else {
 502                         timer->expires.sched -= sum_exec_runtime;
 503                 }
 504         }
 505 }
 506
 507 /*
 508  * These are both called with the siglock held, when the current thread
 509  * is being reaped.  When the final (leader) thread in the group is reaped,
 510  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 511  */
 512 void posix_cpu_timers_exit(struct task_struct *tsk)
 513 {
 514         cleanup_timers(tsk->cpu_timers,
 515                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
 516
 517 }
 518 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 519 {
 520         struct signal_struct *const sig = tsk->signal;
 521
 522         cleanup_timers(tsk->signal->cpu_timers,
 523                        cputime_add(tsk->utime, sig->utime),
 524                        cputime_add(tsk->stime, sig->stime),
 525                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
 526 }
 527
 528 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 529 {
 530         /*
 531          * That's all for this thread or process.
 532          * We leave our residual in expires to be reported.
 533          */
 534         put_task_struct(timer->it.cpu.task);
 535         timer->it.cpu.task = NULL;
 536         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 537                                              timer->it.cpu.expires,
 538                                              now);
 539 }
 540
 541 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
 542 {
 543         return cputime_eq(expires, cputime_zero) ||
 544                cputime_gt(expires, new_exp);
 545 }
 546
 547 /*
 548  * Insert the timer on the appropriate list before any timers that
 549  * expire later.  This must be called with the tasklist_lock held
 550  * for reading, interrupts disabled and p->sighand->siglock taken.
 551  */
 552 static void arm_timer(struct k_itimer *timer)
 553 {
 554         struct task_struct *p = timer->it.cpu.task;
 555         struct list_head *head, *listpos;
 556         struct task_cputime *cputime_expires;
 557         struct cpu_timer_list *const nt = &timer->it.cpu;
 558         struct cpu_timer_list *next;
 559
 560         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 561                 head = p->cpu_timers;
 562                 cputime_expires = &p->cputime_expires;
 563         } else {
 564                 head = p->signal->cpu_timers;
 565                 cputime_expires = &p->signal->cputime_expires;
 566         }
 567         head += CPUCLOCK_WHICH(timer->it_clock);
 568
 569         listpos = head;
 570         list_for_each_entry(next, head, entry) {
 571                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
 572                         break;
 573                 listpos = &next->entry;
 574         }
 575         list_add(&nt->entry, listpos);
 576
 577         if (listpos == head) {
 578                 union cpu_time_count *exp = &nt->expires;
 579
 580                 /*
 581                  * We are the new earliest-expiring POSIX 1.b timer, hence
 582                  * need to update expiration cache. Take into account that
 583                  * for process timers we share expiration cache with itimers
 584                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
 585                  */
 586
 587                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
 588                 case CPUCLOCK_PROF:
 589                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
 590                                 cputime_expires->prof_exp = exp->cpu;
 591                         break;
 592                 case CPUCLOCK_VIRT:
 593                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
 594                                 cputime_expires->virt_exp = exp->cpu;
 595                         break;
 596                 case CPUCLOCK_SCHED:
 597                         if (cputime_expires->sched_exp == 0 ||
 598                             cputime_expires->sched_exp > exp->sched)
 599                                 cputime_expires->sched_exp = exp->sched;
 600                         break;
 601                 }
 602         }
 603 }
 604
 605 /*
 606  * The timer is locked, fire it and arrange for its reload.
 607  */
 608 static void cpu_timer_fire(struct k_itimer *timer)
 609 {
 610         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 611                 /*
 612                  * User don't want any signal.
 613                  */
 614                 timer->it.cpu.expires.sched = 0;
 615         } else if (unlikely(timer->sigq == NULL)) {
 616                 /*
 617                  * This a special case for clock_nanosleep,
 618                  * not a normal timer from sys_timer_create.
 619                  */
 620                 wake_up_process(timer->it_process);
 621                 timer->it.cpu.expires.sched = 0;
 622         } else if (timer->it.cpu.incr.sched == 0) {
 623                 /*
 624                  * One-shot timer.  Clear it as soon as it's fired.
 625                  */
 626                 posix_timer_event(timer, 0);
 627                 timer->it.cpu.expires.sched = 0;
 628         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 629                 /*
 630                  * The signal did not get queued because the signal
 631                  * was ignored, so we won't get any callback to
 632                  * reload the timer.  But we need to keep it
 633                  * ticking in case the signal is deliverable next time.
 634                  */
 635                 posix_cpu_timer_schedule(timer);
 636         }
 637 }
 638
 639 /*
 640  * Sample a process (thread group) timer for the given group_leader task.
 641  * Must be called with tasklist_lock held for reading.
 642  */
 643 static int cpu_timer_sample_group(const clockid_t which_clock,
 644                                   struct task_struct *p,
 645                                   union cpu_time_count *cpu)
 646 {
 647         struct task_cputime cputime;
 648
 649         thread_group_cputimer(p, &cputime);
 650         switch (CPUCLOCK_WHICH(which_clock)) {
 651         default:
 652                 return -EINVAL;
 653         case CPUCLOCK_PROF:
 654                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
 655                 break;
 656         case CPUCLOCK_VIRT:
 657                 cpu->cpu = cputime.utime;
 658                 break;
 659         case CPUCLOCK_SCHED:
 660                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
 661                 break;
 662         }
 663         return 0;
 664 }
 665
 666 /*
 667  * Guts of sys_timer_settime for CPU timers.
 668  * This is called with the timer locked and interrupts disabled.
 669  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 670  * and try again.  (This happens when the timer is in the middle of firing.)
 671  */
 672 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 673                                struct itimerspec *new, struct itimerspec *old)
 674 {
 675         struct task_struct *p = timer->it.cpu.task;
 676         union cpu_time_count old_expires, new_expires, old_incr, val;
 677         int ret;
 678
 679         if (unlikely(p == NULL)) {
 680                 /*
 681                  * Timer refers to a dead task's clock.
 682                  */
 683                 return -ESRCH;
 684         }
 685
 686         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 687
 688         read_lock(&tasklist_lock);
 689         /*
 690          * We need the tasklist_lock to protect against reaping that
 691          * clears p->sighand.  If p has just been reaped, we can no
 692          * longer get any information about it at all.
 693          */
 694         if (unlikely(p->sighand == NULL)) {
 695                 read_unlock(&tasklist_lock);
 696                 put_task_struct(p);
 697                 timer->it.cpu.task = NULL;
 698                 return -ESRCH;
 699         }
 700
 701         /*
 702          * Disarm any old timer after extracting its expiry time.
 703          */
 704         BUG_ON(!irqs_disabled());
 705
 706         ret = 0;
 707         old_incr = timer->it.cpu.incr;
 708         spin_lock(&p->sighand->siglock);
 709         old_expires = timer->it.cpu.expires;
 710         if (unlikely(timer->it.cpu.firing)) {
 711                 timer->it.cpu.firing = -1;
 712                 ret = TIMER_RETRY;
 713         } else
 714                 list_del_init(&timer->it.cpu.entry);
 715
 716         /*
 717          * We need to sample the current value to convert the new
 718          * value from to relative and absolute, and to convert the
 719          * old value from absolute to relative.  To set a process
 720          * timer, we need a sample to balance the thread expiry
 721          * times (in arm_timer).  With an absolute time, we must
 722          * check if it's already passed.  In short, we need a sample.
 723          */
 724         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 725                 cpu_clock_sample(timer->it_clock, p, &val);
 726         } else {
 727                 cpu_timer_sample_group(timer->it_clock, p, &val);
 728         }
 729
 730         if (old) {
 731                 if (old_expires.sched == 0) {
 732                         old->it_value.tv_sec = 0;
 733                         old->it_value.tv_nsec = 0;
 734                 } else {
 735                         /*
 736                          * Update the timer in case it has
 737                          * overrun already.  If it has,
 738                          * we'll report it as having overrun
 739                          * and with the next reloaded timer
 740                          * already ticking, though we are
 741                          * swallowing that pending
 742                          * notification here to install the
 743                          * new setting.
 744                          */
 745                         bump_cpu_timer(timer, val);
 746                         if (cpu_time_before(timer->it_clock, val,
 747                                             timer->it.cpu.expires)) {
 748                                 old_expires = cpu_time_sub(
 749                                         timer->it_clock,
 750                                         timer->it.cpu.expires, val);
 751                                 sample_to_timespec(timer->it_clock,
 752                                                    old_expires,
 753                                                    &old->it_value);
 754                         } else {
 755                                 old->it_value.tv_nsec = 1;
 756                                 old->it_value.tv_sec = 0;
 757                         }
 758                 }
 759         }
 760
 761         if (unlikely(ret)) {
 762                 /*
 763                  * We are colliding with the timer actually firing.
 764                  * Punt after filling in the timer's old value, and
 765                  * disable this firing since we are already reporting
 766                  * it as an overrun (thanks to bump_cpu_timer above).
 767                  */
 768                 spin_unlock(&p->sighand->siglock);
 769                 read_unlock(&tasklist_lock);
 770                 goto out;
 771         }
 772
 773         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 774                 cpu_time_add(timer->it_clock, &new_expires, val);
 775         }
 776
 777         /*
 778          * Install the new expiry time (or zero).
 779          * For a timer with no notification action, we don't actually
 780          * arm the timer (we'll just fake it for timer_gettime).
 781          */
 782         timer->it.cpu.expires = new_expires;
 783         if (new_expires.sched != 0 &&
 784             cpu_time_before(timer->it_clock, val, new_expires)) {
 785                 arm_timer(timer);
 786         }
 787
 788         spin_unlock(&p->sighand->siglock);
 789         read_unlock(&tasklist_lock);
 790
 791         /*
 792          * Install the new reload setting, and
 793          * set up the signal and overrun bookkeeping.
 794          */
 795         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 796                                                 &new->it_interval);
 797
 798         /*
 799          * This acts as a modification timestamp for the timer,
 800          * so any automatic reload attempt will punt on seeing
 801          * that we have reset the timer manually.
 802          */
 803         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 804                 ~REQUEUE_PENDING;
 805         timer->it_overrun_last = 0;
 806         timer->it_overrun = -1;
 807
 808         if (new_expires.sched != 0 &&
 809             !cpu_time_before(timer->it_clock, val, new_expires)) {
 810                 /*
 811                  * The designated time already passed, so we notify
 812                  * immediately, even if the thread never runs to
 813                  * accumulate more time on this clock.
 814                  */
 815                 cpu_timer_fire(timer);
 816         }
 817
 818         ret = 0;
 819  out:
 820         if (old) {
 821                 sample_to_timespec(timer->it_clock,
 822                                    old_incr, &old->it_interval);
 823         }
 824         return ret;
 825 }
 826
 827 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 828 {
 829         union cpu_time_count now;
 830         struct task_struct *p = timer->it.cpu.task;
 831         int clear_dead;
 832
 833         /*
 834          * Easy part: convert the reload time.
 835          */
 836         sample_to_timespec(timer->it_clock,
 837                            timer->it.cpu.incr, &itp->it_interval);
 838
 839         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 840                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 841                 return;
 842         }
 843
 844         if (unlikely(p == NULL)) {
 845                 /*
 846                  * This task already died and the timer will never fire.
 847                  * In this case, expires is actually the dead value.
 848                  */
 849         dead:
 850                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 851                                    &itp->it_value);
 852                 return;
 853         }
 854
 855         /*
 856          * Sample the clock to take the difference with the expiry time.
 857          */
 858         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 859                 cpu_clock_sample(timer->it_clock, p, &now);
 860                 clear_dead = p->exit_state;
 861         } else {
 862                 read_lock(&tasklist_lock);
 863                 if (unlikely(p->sighand == NULL)) {
 864                         /*
 865                          * The process has been reaped.
 866                          * We can't even collect a sample any more.
 867                          * Call the timer disarmed, nothing else to do.
 868                          */
 869                         put_task_struct(p);
 870                         timer->it.cpu.task = NULL;
 871                         timer->it.cpu.expires.sched = 0;
 872                         read_unlock(&tasklist_lock);
 873                         goto dead;
 874                 } else {
 875                         cpu_timer_sample_group(timer->it_clock, p, &now);
 876                         clear_dead = (unlikely(p->exit_state) &&
 877                                       thread_group_empty(p));
 878                 }
 879                 read_unlock(&tasklist_lock);
 880         }
 881
 882         if (unlikely(clear_dead)) {
 883                 /*
 884                  * We've noticed that the thread is dead, but
 885                  * not yet reaped.  Take this opportunity to
 886                  * drop our task ref.
 887                  */
 888                 clear_dead_task(timer, now);
 889                 goto dead;
 890         }
 891
 892         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 893                 sample_to_timespec(timer->it_clock,
 894                                    cpu_time_sub(timer->it_clock,
 895                                                 timer->it.cpu.expires, now),
 896                                    &itp->it_value);
 897         } else {
 898                 /*
 899                  * The timer should have expired already, but the firing
 900                  * hasn't taken place yet.  Say it's just about to expire.
 901                  */
 902                 itp->it_value.tv_nsec = 1;
 903                 itp->it_value.tv_sec = 0;
 904         }
 905 }
 906
 907 /*
 908  * Check for any per-thread CPU timers that have fired and move them off
 909  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 910  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 911  */
 912 static void check_thread_timers(struct task_struct *tsk,
 913                                 struct list_head *firing)
 914 {
 915         int maxfire;
 916         struct list_head *timers = tsk->cpu_timers;
 917         struct signal_struct *const sig = tsk->signal;
 918         unsigned long soft;
 919
 920         maxfire = 20;
 921         tsk->cputime_expires.prof_exp = cputime_zero;
 922         while (!list_empty(timers)) {
 923                 struct cpu_timer_list *t = list_first_entry(timers,
 924                                                       struct cpu_timer_list,
 925                                                       entry);
 926                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 927                         tsk->cputime_expires.prof_exp = t->expires.cpu;
 928                         break;
 929                 }
 930                 t->firing = 1;
 931                 list_move_tail(&t->entry, firing);
 932         }
 933
 934         ++timers;
 935         maxfire = 20;
 936         tsk->cputime_expires.virt_exp = cputime_zero;
 937         while (!list_empty(timers)) {
 938                 struct cpu_timer_list *t = list_first_entry(timers,
 939                                                       struct cpu_timer_list,
 940                                                       entry);
 941                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 942                         tsk->cputime_expires.virt_exp = t->expires.cpu;
 943                         break;
 944                 }
 945                 t->firing = 1;
 946                 list_move_tail(&t->entry, firing);
 947         }
 948
 949         ++timers;
 950         maxfire = 20;
 951         tsk->cputime_expires.sched_exp = 0;
 952         while (!list_empty(timers)) {
 953                 struct cpu_timer_list *t = list_first_entry(timers,
 954                                                       struct cpu_timer_list,
 955                                                       entry);
 956                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
 957                         tsk->cputime_expires.sched_exp = t->expires.sched;
 958                         break;
 959                 }
 960                 t->firing = 1;
 961                 list_move_tail(&t->entry, firing);
 962         }
 963
 964         /*
 965          * Check for the special case thread timers.
 966          */
 967         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
 968         if (soft != RLIM_INFINITY) {
 969                 unsigned long hard =
 970                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
 971
 972                 if (hard != RLIM_INFINITY &&
 973                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
 974                         /*
 975                          * At the hard limit, we just die.
 976                          * No need to calculate anything else now.
 977                          */
 978                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
 979                         return;
 980                 }
 981                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
 982                         /*
 983                          * At the soft limit, send a SIGXCPU every second.
 984                          */
 985                         if (soft < hard) {
 986                                 soft += USEC_PER_SEC;
 987                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
 988                         }
 989                         printk(KERN_INFO
 990                                 "RT Watchdog Timeout: %s[%d]\n",
 991                                 tsk->comm, task_pid_nr(tsk));
 992                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
 993                 }
 994         }
 995 }
 996
 997 static void stop_process_timers(struct signal_struct *sig)
 998 {
 999         struct thread_group_cputimer *cputimer = &sig->cputimer;
1000         unsigned long flags;
1001
1002         raw_spin_lock_irqsave(&cputimer->lock, flags);
1003         cputimer->running = 0;
1004         raw_spin_unlock_irqrestore(&cputimer->lock, flags);
1005 }
1006
1007 static u32 onecputick;
1008
1009 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1010                              cputime_t *expires, cputime_t cur_time, int signo)
1011 {
1012         if (cputime_eq(it->expires, cputime_zero))
1013                 return;
1014
1015         if (cputime_ge(cur_time, it->expires)) {
1016                 if (!cputime_eq(it->incr, cputime_zero)) {
1017                         it->expires = cputime_add(it->expires, it->incr);
1018                         it->error += it->incr_error;
1019                         if (it->error >= onecputick) {
1020                                 it->expires = cputime_sub(it->expires,
1021                                                           cputime_one_jiffy);
1022                                 it->error -= onecputick;
1023                         }
1024                 } else {
1025                         it->expires = cputime_zero;
1026                 }
1027
1028                 trace_itimer_expire(signo == SIGPROF ?
1029                                     ITIMER_PROF : ITIMER_VIRTUAL,
1030                                     tsk->signal->leader_pid, cur_time);
1031                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1032         }
1033
1034         if (!cputime_eq(it->expires, cputime_zero) &&
1035             (cputime_eq(*expires, cputime_zero) ||
1036              cputime_lt(it->expires, *expires))) {
1037                 *expires = it->expires;
1038         }
1039 }
1040
1041 /**
1042  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1043  *
1044  * @cputime:    The struct to compare.
1045  *
1046  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1047  * are zero, false if any field is nonzero.
1048  */
1049 static inline int task_cputime_zero(const struct task_cputime *cputime)
1050 {
1051         if (cputime_eq(cputime->utime, cputime_zero) &&
1052             cputime_eq(cputime->stime, cputime_zero) &&
1053             cputime->sum_exec_runtime == 0)
1054                 return 1;
1055         return 0;
1056 }
1057
1058 /*
1059  * Check for any per-thread CPU timers that have fired and move them
1060  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1061  * have already been taken off.
1062  */
1063 static void check_process_timers(struct task_struct *tsk,
1064                                  struct list_head *firing)
1065 {
1066         int maxfire;
1067         struct signal_struct *const sig = tsk->signal;
1068         cputime_t utime, ptime, virt_expires, prof_expires;
1069         unsigned long long sum_sched_runtime, sched_expires;
1070         struct list_head *timers = sig->cpu_timers;
1071         struct task_cputime cputime;
1072         unsigned long soft;
1073
1074         /*
1075          * Collect the current process totals.
1076          */
1077         thread_group_cputimer(tsk, &cputime);
1078         utime = cputime.utime;
1079         ptime = cputime_add(utime, cputime.stime);
1080         sum_sched_runtime = cputime.sum_exec_runtime;
1081         maxfire = 20;
1082         prof_expires = cputime_zero;
1083         while (!list_empty(timers)) {
1084                 struct cpu_timer_list *tl = list_first_entry(timers,
1085                                                       struct cpu_timer_list,
1086                                                       entry);
1087                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1088                         prof_expires = tl->expires.cpu;
1089                         break;
1090                 }
1091                 tl->firing = 1;
1092                 list_move_tail(&tl->entry, firing);
1093         }
1094
1095         ++timers;
1096         maxfire = 20;
1097         virt_expires = cputime_zero;
1098         while (!list_empty(timers)) {
1099                 struct cpu_timer_list *tl = list_first_entry(timers,
1100                                                       struct cpu_timer_list,
1101                                                       entry);
1102                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1103                         virt_expires = tl->expires.cpu;
1104                         break;
1105                 }
1106                 tl->firing = 1;
1107                 list_move_tail(&tl->entry, firing);
1108         }
1109
1110         ++timers;
1111         maxfire = 20;
1112         sched_expires = 0;
1113         while (!list_empty(timers)) {
1114                 struct cpu_timer_list *tl = list_first_entry(timers,
1115                                                       struct cpu_timer_list,
1116                                                       entry);
1117                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1118                         sched_expires = tl->expires.sched;
1119                         break;
1120                 }
1121                 tl->firing = 1;
1122                 list_move_tail(&tl->entry, firing);
1123         }
1124
1125         /*
1126          * Check for the special case process timers.
1127          */
1128         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1129                          SIGPROF);
1130         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1131                          SIGVTALRM);
1132         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1133         if (soft != RLIM_INFINITY) {
1134                 unsigned long psecs = cputime_to_secs(ptime);
1135                 unsigned long hard =
1136                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1137                 cputime_t x;
1138                 if (psecs >= hard) {
1139                         /*
1140                          * At the hard limit, we just die.
1141                          * No need to calculate anything else now.
1142                          */
1143                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1144                         return;
1145                 }
1146                 if (psecs >= soft) {
1147                         /*
1148                          * At the soft limit, send a SIGXCPU every second.
1149                          */
1150                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1151                         if (soft < hard) {
1152                                 soft++;
1153                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1154                         }
1155                 }
1156                 x = secs_to_cputime(soft);
1157                 if (cputime_eq(prof_expires, cputime_zero) ||
1158                     cputime_lt(x, prof_expires)) {
1159                         prof_expires = x;
1160                 }
1161         }
1162
1163         sig->cputime_expires.prof_exp = prof_expires;
1164         sig->cputime_expires.virt_exp = virt_expires;
1165         sig->cputime_expires.sched_exp = sched_expires;
1166         if (task_cputime_zero(&sig->cputime_expires))
1167                 stop_process_timers(sig);
1168 }
1169
1170 /*
1171  * This is called from the signal code (via do_schedule_next_timer)
1172  * when the last timer signal was delivered and we have to reload the timer.
1173  */
1174 void posix_cpu_timer_schedule(struct k_itimer *timer)
1175 {
1176         struct task_struct *p = timer->it.cpu.task;
1177         union cpu_time_count now;
1178
1179         if (unlikely(p == NULL))
1180                 /*
1181                  * The task was cleaned up already, no future firings.
1182                  */
1183                 goto out;
1184
1185         /*
1186          * Fetch the current sample and update the timer's expiry time.
1187          */
1188         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1189                 cpu_clock_sample(timer->it_clock, p, &now);
1190                 bump_cpu_timer(timer, now);
1191                 if (unlikely(p->exit_state)) {
1192                         clear_dead_task(timer, now);
1193                         goto out;
1194                 }
1195                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1196                 spin_lock(&p->sighand->siglock);
1197         } else {
1198                 read_lock(&tasklist_lock);
1199                 if (unlikely(p->sighand == NULL)) {
1200                         /*
1201                          * The process has been reaped.
1202                          * We can't even collect a sample any more.
1203                          */
1204                         put_task_struct(p);
1205                         timer->it.cpu.task = p = NULL;
1206                         timer->it.cpu.expires.sched = 0;
1207                         goto out_unlock;
1208                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1209                         /*
1210                          * We've noticed that the thread is dead, but
1211                          * not yet reaped.  Take this opportunity to
1212                          * drop our task ref.
1213                          */
1214                         clear_dead_task(timer, now);
1215                         goto out_unlock;
1216                 }
1217                 spin_lock(&p->sighand->siglock);
1218                 cpu_timer_sample_group(timer->it_clock, p, &now);
1219                 bump_cpu_timer(timer, now);
1220                 /* Leave the tasklist_lock locked for the call below.  */
1221         }
1222
1223         /*
1224          * Now re-arm for the new expiry time.
1225          */
1226         BUG_ON(!irqs_disabled());
1227         arm_timer(timer);
1228         spin_unlock(&p->sighand->siglock);
1229
1230 out_unlock:
1231         read_unlock(&tasklist_lock);
1232
1233 out:
1234         timer->it_overrun_last = timer->it_overrun;
1235         timer->it_overrun = -1;
1236         ++timer->it_requeue_pending;
1237 }
1238
1239 /**
1240  * task_cputime_expired - Compare two task_cputime entities.
1241  *
1242  * @sample:     The task_cputime structure to be checked for expiration.
1243  * @expires:    Expiration times, against which @sample will be checked.
1244  *
1245  * Checks @sample against @expires to see if any field of @sample has expired.
1246  * Returns true if any field of the former is greater than the corresponding
1247  * field of the latter if the latter field is set.  Otherwise returns false.
1248  */
1249 static inline int task_cputime_expired(const struct task_cputime *sample,
1250                                         const struct task_cputime *expires)
1251 {
1252         if (!cputime_eq(expires->utime, cputime_zero) &&
1253             cputime_ge(sample->utime, expires->utime))
1254                 return 1;
1255         if (!cputime_eq(expires->stime, cputime_zero) &&
1256             cputime_ge(cputime_add(sample->utime, sample->stime),
1257                        expires->stime))
1258                 return 1;
1259         if (expires->sum_exec_runtime != 0 &&
1260             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1261                 return 1;
1262         return 0;
1263 }
1264
1265 /**
1266  * fastpath_timer_check - POSIX CPU timers fast path.
1267  *
1268  * @tsk:        The task (thread) being checked.
1269  *
1270  * Check the task and thread group timers.  If both are zero (there are no
1271  * timers set) return false.  Otherwise snapshot the task and thread group
1272  * timers and compare them with the corresponding expiration times.  Return
1273  * true if a timer has expired, else return false.
1274  */
1275 static inline int fastpath_timer_check(struct task_struct *tsk)
1276 {
1277         struct signal_struct *sig;
1278
1279         if (!task_cputime_zero(&tsk->cputime_expires)) {
1280                 struct task_cputime task_sample = {
1281                         .utime = tsk->utime,
1282                         .stime = tsk->stime,
1283                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1284                 };
1285
1286                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1287                         return 1;
1288         }
1289
1290         sig = tsk->signal;
1291         if (sig->cputimer.running) {
1292                 struct task_cputime group_sample;
1293
1294                 raw_spin_lock(&sig->cputimer.lock);
1295                 group_sample = sig->cputimer.cputime;
1296                 raw_spin_unlock(&sig->cputimer.lock);
1297
1298                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1299                         return 1;
1300         }
1301
1302         return 0;
1303 }
1304
1305 /*
1306  * This is called from the timer interrupt handler.  The irq handler has
1307  * already updated our counts.  We need to check if any timers fire now.
1308  * Interrupts are disabled.
1309  */
1310 void run_posix_cpu_timers(struct task_struct *tsk)
1311 {
1312         LIST_HEAD(firing);
1313         struct k_itimer *timer, *next;
1314         unsigned long flags;
1315
1316         BUG_ON(!irqs_disabled());
1317
1318         /*
1319          * The fast path checks that there are no expired thread or thread
1320          * group timers.  If that's so, just return.
1321          */
1322         if (!fastpath_timer_check(tsk))
1323                 return;
1324
1325         if (!lock_task_sighand(tsk, &flags))
1326                 return;
1327         /*
1328          * Here we take off tsk->signal->cpu_timers[N] and
1329          * tsk->cpu_timers[N] all the timers that are firing, and
1330          * put them on the firing list.
1331          */
1332         check_thread_timers(tsk, &firing);
1333         /*
1334          * If there are any active process wide timers (POSIX 1.b, itimers,
1335          * RLIMIT_CPU) cputimer must be running.
1336          */
1337         if (tsk->signal->cputimer.running)
1338                 check_process_timers(tsk, &firing);
1339
1340         /*
1341          * We must release these locks before taking any timer's lock.
1342          * There is a potential race with timer deletion here, as the
1343          * siglock now protects our private firing list.  We have set
1344          * the firing flag in each timer, so that a deletion attempt
1345          * that gets the timer lock before we do will give it up and
1346          * spin until we've taken care of that timer below.
1347          */
1348         unlock_task_sighand(tsk, &flags);
1349
1350         /*
1351          * Now that all the timers on our list have the firing flag,
1352          * no one will touch their list entries but us.  We'll take
1353          * each timer's lock before clearing its firing flag, so no
1354          * timer call will interfere.
1355          */
1356         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1357                 int cpu_firing;
1358
1359                 spin_lock(&timer->it_lock);
1360                 list_del_init(&timer->it.cpu.entry);
1361                 cpu_firing = timer->it.cpu.firing;
1362                 timer->it.cpu.firing = 0;
1363                 /*
1364                  * The firing flag is -1 if we collided with a reset
1365                  * of the timer, which already reported this
1366                  * almost-firing as an overrun.  So don't generate an event.
1367                  */
1368                 if (likely(cpu_firing >= 0))
1369                         cpu_timer_fire(timer);
1370                 spin_unlock(&timer->it_lock);
1371         }
1372 }
1373
1374 /*
1375  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1376  * The tsk->sighand->siglock must be held by the caller.
1377  */
1378 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1379                            cputime_t *newval, cputime_t *oldval)
1380 {
1381         union cpu_time_count now;
1382
1383         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1384         cpu_timer_sample_group(clock_idx, tsk, &now);
1385
1386         if (oldval) {
1387                 /*
1388                  * We are setting itimer. The *oldval is absolute and we update
1389                  * it to be relative, *newval argument is relative and we update
1390                  * it to be absolute.
1391                  */
1392                 if (!cputime_eq(*oldval, cputime_zero)) {
1393                         if (cputime_le(*oldval, now.cpu)) {
1394                                 /* Just about to fire. */
1395                                 *oldval = cputime_one_jiffy;
1396                         } else {
1397                                 *oldval = cputime_sub(*oldval, now.cpu);
1398                         }
1399                 }
1400
1401                 if (cputime_eq(*newval, cputime_zero))
1402                         return;
1403                 *newval = cputime_add(*newval, now.cpu);
1404         }
1405
1406         /*
1407          * Update expiration cache if we are the earliest timer, or eventually
1408          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1409          */
1410         switch (clock_idx) {
1411         case CPUCLOCK_PROF:
1412                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1413                         tsk->signal->cputime_expires.prof_exp = *newval;
1414                 break;
1415         case CPUCLOCK_VIRT:
1416                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1417                         tsk->signal->cputime_expires.virt_exp = *newval;
1418                 break;
1419         }
1420 }
1421
1422 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1423                             struct timespec *rqtp, struct itimerspec *it)
1424 {
1425         struct k_itimer timer;
1426         int error;
1427
1428         /*
1429          * Set up a temporary timer and then wait for it to go off.
1430          */
1431         memset(&timer, 0, sizeof timer);
1432         spin_lock_init(&timer.it_lock);
1433         timer.it_clock = which_clock;
1434         timer.it_overrun = -1;
1435         error = posix_cpu_timer_create(&timer);
1436         timer.it_process = current;
1437         if (!error) {
1438                 static struct itimerspec zero_it;
1439
1440                 memset(it, 0, sizeof *it);
1441                 it->it_value = *rqtp;
1442
1443                 spin_lock_irq(&timer.it_lock);
1444                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1445                 if (error) {
1446                         spin_unlock_irq(&timer.it_lock);
1447                         return error;
1448                 }
1449
1450                 while (!signal_pending(current)) {
1451                         if (timer.it.cpu.expires.sched == 0) {
1452                                 /*
1453                                  * Our timer fired and was reset, below
1454                                  * deletion can not fail.
1455                                  */
1456                                 posix_cpu_timer_del(&timer);
1457                                 spin_unlock_irq(&timer.it_lock);
1458                                 return 0;
1459                         }
1460
1461                         /*
1462                          * Block until cpu_timer_fire (or a signal) wakes us.
1463                          */
1464                         __set_current_state(TASK_INTERRUPTIBLE);
1465                         spin_unlock_irq(&timer.it_lock);
1466                         schedule();
1467                         spin_lock_irq(&timer.it_lock);
1468                 }
1469
1470                 /*
1471                  * We were interrupted by a signal.
1472                  */
1473                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1474                 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1475                 if (!error) {
1476                         /*
1477                          * Timer is now unarmed, deletion can not fail.
1478                          */
1479                         posix_cpu_timer_del(&timer);
1480                 }
1481                 spin_unlock_irq(&timer.it_lock);
1482
1483                 while (error == TIMER_RETRY) {
1484                         /*
1485                          * We need to handle case when timer was or is in the
1486                          * middle of firing. In other cases we already freed
1487                          * resources.
1488                          */
1489                         spin_lock_irq(&timer.it_lock);
1490                         error = posix_cpu_timer_del(&timer);
1491                         spin_unlock_irq(&timer.it_lock);
1492                 }
1493
1494                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1495                         /*
1496                          * It actually did fire already.
1497                          */
1498                         return 0;
1499                 }
1500
1501                 error = -ERESTART_RESTARTBLOCK;
1502         }
1503
1504         return error;
1505 }
1506
1507 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1508
1509 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1510                             struct timespec *rqtp, struct timespec __user *rmtp)
1511 {
1512         struct restart_block *restart_block =
1513                 &current_thread_info()->restart_block;
1514         struct itimerspec it;
1515         int error;
1516
1517         /*
1518          * Diagnose required errors first.
1519          */
1520         if (CPUCLOCK_PERTHREAD(which_clock) &&
1521             (CPUCLOCK_PID(which_clock) == 0 ||
1522              CPUCLOCK_PID(which_clock) == current->pid))
1523                 return -EINVAL;
1524
1525         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1526
1527         if (error == -ERESTART_RESTARTBLOCK) {
1528
1529                 if (flags & TIMER_ABSTIME)
1530                         return -ERESTARTNOHAND;
1531                 /*
1532                  * Report back to the user the time still remaining.
1533                  */
1534                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1535                         return -EFAULT;
1536
1537                 restart_block->fn = posix_cpu_nsleep_restart;
1538                 restart_block->nanosleep.clockid = which_clock;
1539                 restart_block->nanosleep.rmtp = rmtp;
1540                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1541         }
1542         return error;
1543 }
1544
1545 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1546 {
1547         clockid_t which_clock = restart_block->nanosleep.clockid;
1548         struct timespec t;
1549         struct itimerspec it;
1550         int error;
1551
1552         t = ns_to_timespec(restart_block->nanosleep.expires);
1553
1554         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1555
1556         if (error == -ERESTART_RESTARTBLOCK) {
1557                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1558                 /*
1559                  * Report back to the user the time still remaining.
1560                  */
1561                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1562                         return -EFAULT;
1563
1564                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1565         }
1566         return error;
1567
1568 }
1569
1570 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1571 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1572
1573 static int process_cpu_clock_getres(const clockid_t which_clock,
1574                                     struct timespec *tp)
1575 {
1576         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1577 }
1578 static int process_cpu_clock_get(const clockid_t which_clock,
1579                                  struct timespec *tp)
1580 {
1581         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1582 }
1583 static int process_cpu_timer_create(struct k_itimer *timer)
1584 {
1585         timer->it_clock = PROCESS_CLOCK;
1586         return posix_cpu_timer_create(timer);
1587 }
1588 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1589                               struct timespec *rqtp,
1590                               struct timespec __user *rmtp)
1591 {
1592         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1593 }
1594 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1595 {
1596         return -EINVAL;
1597 }
1598 static int thread_cpu_clock_getres(const clockid_t which_clock,
1599                                    struct timespec *tp)
1600 {
1601         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1602 }
1603 static int thread_cpu_clock_get(const clockid_t which_clock,
1604                                 struct timespec *tp)
1605 {
1606         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1607 }
1608 static int thread_cpu_timer_create(struct k_itimer *timer)
1609 {
1610         timer->it_clock = THREAD_CLOCK;
1611         return posix_cpu_timer_create(timer);
1612 }
1613
1614 struct k_clock clock_posix_cpu = {
1615         .clock_getres   = posix_cpu_clock_getres,
1616         .clock_set      = posix_cpu_clock_set,
1617         .clock_get      = posix_cpu_clock_get,
1618         .timer_create   = posix_cpu_timer_create,
1619         .nsleep         = posix_cpu_nsleep,
1620         .nsleep_restart = posix_cpu_nsleep_restart,
1621         .timer_set      = posix_cpu_timer_set,
1622         .timer_del      = posix_cpu_timer_del,
1623         .timer_get      = posix_cpu_timer_get,
1624 };
1625
1626 static __init int init_posix_cpu_timers(void)
1627 {
1628         struct k_clock process = {
1629                 .clock_getres   = process_cpu_clock_getres,
1630                 .clock_get      = process_cpu_clock_get,
1631                 .timer_create   = process_cpu_timer_create,
1632                 .nsleep         = process_cpu_nsleep,
1633                 .nsleep_restart = process_cpu_nsleep_restart,
1634         };
1635         struct k_clock thread = {
1636                 .clock_getres   = thread_cpu_clock_getres,
1637                 .clock_get      = thread_cpu_clock_get,
1638                 .timer_create   = thread_cpu_timer_create,
1639         };
1640         struct timespec ts;
1641
1642         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1643         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1644
1645         cputime_to_timespec(cputime_one_jiffy, &ts);
1646         onecputick = ts.tv_nsec;
1647         WARN_ON(ts.tv_sec != 0);
1648
1649         return 0;
1650 }
1651 __initcall(init_posix_cpu_timers);