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 <asm/uaccess.h>
   8 #include <linux/errno.h>
   9 #include <linux/vs_cvirt.h>
  10
  11 static int check_clock(const clockid_t which_clock)
  12 {
  13         int error = 0;
  14         struct task_struct *p;
  15         const pid_t pid = CPUCLOCK_PID(which_clock);
  16
  17         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
  18                 return -EINVAL;
  19
  20         if (pid == 0)
  21                 return 0;
  22
  23         read_lock(&tasklist_lock);
  24         p = find_task_by_pid(pid);
  25         if (!p || (CPUCLOCK_PERTHREAD(which_clock) ?
  26                    p->tgid != current->tgid : p->tgid != pid)) {
  27                 error = -EINVAL;
  28         }
  29         read_unlock(&tasklist_lock);
  30
  31         return error;
  32 }
  33
  34 static inline union cpu_time_count
  35 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
  36 {
  37         union cpu_time_count ret;
  38         ret.sched = 0;          /* high half always zero when .cpu used */
  39         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  40                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
  41         } else {
  42                 ret.cpu = timespec_to_cputime(tp);
  43         }
  44         return ret;
  45 }
  46
  47 static void sample_to_timespec(const clockid_t which_clock,
  48                                union cpu_time_count cpu,
  49                                struct timespec *tp)
  50 {
  51         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  52                 tp->tv_sec = div_long_long_rem(cpu.sched,
  53                                                NSEC_PER_SEC, &tp->tv_nsec);
  54         } else {
  55                 cputime_to_timespec(cpu.cpu, tp);
  56         }
  57 }
  58
  59 static inline int cpu_time_before(const clockid_t which_clock,
  60                                   union cpu_time_count now,
  61                                   union cpu_time_count then)
  62 {
  63         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  64                 return now.sched < then.sched;
  65         }  else {
  66                 return cputime_lt(now.cpu, then.cpu);
  67         }
  68 }
  69 static inline void cpu_time_add(const clockid_t which_clock,
  70                                 union cpu_time_count *acc,
  71                                 union cpu_time_count val)
  72 {
  73         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  74                 acc->sched += val.sched;
  75         }  else {
  76                 acc->cpu = cputime_add(acc->cpu, val.cpu);
  77         }
  78 }
  79 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
  80                                                 union cpu_time_count a,
  81                                                 union cpu_time_count b)
  82 {
  83         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
  84                 a.sched -= b.sched;
  85         }  else {
  86                 a.cpu = cputime_sub(a.cpu, b.cpu);
  87         }
  88         return a;
  89 }
  90
  91 /*
  92  * Update expiry time from increment, and increase overrun count,
  93  * given the current clock sample.
  94  */
  95 static void bump_cpu_timer(struct k_itimer *timer,
  96                                   union cpu_time_count now)
  97 {
  98         int i;
  99
 100         if (timer->it.cpu.incr.sched == 0)
 101                 return;
 102
 103         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 104                 unsigned long long delta, incr;
 105
 106                 if (now.sched < timer->it.cpu.expires.sched)
 107                         return;
 108                 incr = timer->it.cpu.incr.sched;
 109                 delta = now.sched + incr - timer->it.cpu.expires.sched;
 110                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 111                 for (i = 0; incr < delta - incr; i++)
 112                         incr = incr << 1;
 113                 for (; i >= 0; incr >>= 1, i--) {
 114                         if (delta < incr)
 115                                 continue;
 116                         timer->it.cpu.expires.sched += incr;
 117                         timer->it_overrun += 1 << i;
 118                         delta -= incr;
 119                 }
 120         } else {
 121                 cputime_t delta, incr;
 122
 123                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
 124                         return;
 125                 incr = timer->it.cpu.incr.cpu;
 126                 delta = cputime_sub(cputime_add(now.cpu, incr),
 127                                     timer->it.cpu.expires.cpu);
 128                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
 129                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
 130                              incr = cputime_add(incr, incr);
 131                 for (; i >= 0; incr = cputime_halve(incr), i--) {
 132                         if (cputime_lt(delta, incr))
 133                                 continue;
 134                         timer->it.cpu.expires.cpu =
 135                                 cputime_add(timer->it.cpu.expires.cpu, incr);
 136                         timer->it_overrun += 1 << i;
 137                         delta = cputime_sub(delta, incr);
 138                 }
 139         }
 140 }
 141
 142 static inline cputime_t prof_ticks(struct task_struct *p)
 143 {
 144         return cputime_add(p->utime, p->stime);
 145 }
 146 static inline cputime_t virt_ticks(struct task_struct *p)
 147 {
 148         return p->utime;
 149 }
 150 static inline unsigned long long sched_ns(struct task_struct *p)
 151 {
 152         return (p == current) ? current_sched_time(p) : p->sched_time;
 153 }
 154
 155 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
 156 {
 157         int error = check_clock(which_clock);
 158         if (!error) {
 159                 tp->tv_sec = 0;
 160                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
 161                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
 162                         /*
 163                          * If sched_clock is using a cycle counter, we
 164                          * don't have any idea of its true resolution
 165                          * exported, but it is much more than 1s/HZ.
 166                          */
 167                         tp->tv_nsec = 1;
 168                 }
 169         }
 170         return error;
 171 }
 172
 173 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
 174 {
 175         /*
 176          * You can never reset a CPU clock, but we check for other errors
 177          * in the call before failing with EPERM.
 178          */
 179         int error = check_clock(which_clock);
 180         if (error == 0) {
 181                 error = -EPERM;
 182         }
 183         return error;
 184 }
 185
 186
 187 /*
 188  * Sample a per-thread clock for the given task.
 189  */
 190 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
 191                             union cpu_time_count *cpu)
 192 {
 193         switch (CPUCLOCK_WHICH(which_clock)) {
 194         default:
 195                 return -EINVAL;
 196         case CPUCLOCK_PROF:
 197                 cpu->cpu = prof_ticks(p);
 198                 break;
 199         case CPUCLOCK_VIRT:
 200                 cpu->cpu = virt_ticks(p);
 201                 break;
 202         case CPUCLOCK_SCHED:
 203                 cpu->sched = sched_ns(p);
 204                 break;
 205         }
 206         return 0;
 207 }
 208
 209 /*
 210  * Sample a process (thread group) clock for the given group_leader task.
 211  * Must be called with tasklist_lock held for reading.
 212  * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
 213  */
 214 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
 215                                          struct task_struct *p,
 216                                          union cpu_time_count *cpu)
 217 {
 218         struct task_struct *t = p;
 219         switch (clock_idx) {
 220         default:
 221                 return -EINVAL;
 222         case CPUCLOCK_PROF:
 223                 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
 224                 do {
 225                         cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
 226                         t = next_thread(t);
 227                 } while (t != p);
 228                 break;
 229         case CPUCLOCK_VIRT:
 230                 cpu->cpu = p->signal->utime;
 231                 do {
 232                         cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
 233                         t = next_thread(t);
 234                 } while (t != p);
 235                 break;
 236         case CPUCLOCK_SCHED:
 237                 cpu->sched = p->signal->sched_time;
 238                 /* Add in each other live thread.  */
 239                 while ((t = next_thread(t)) != p) {
 240                         cpu->sched += t->sched_time;
 241                 }
 242                 cpu->sched += sched_ns(p);
 243                 break;
 244         }
 245         return 0;
 246 }
 247
 248 /*
 249  * Sample a process (thread group) clock for the given group_leader task.
 250  * Must be called with tasklist_lock held for reading.
 251  */
 252 static int cpu_clock_sample_group(const clockid_t which_clock,
 253                                   struct task_struct *p,
 254                                   union cpu_time_count *cpu)
 255 {
 256         int ret;
 257         unsigned long flags;
 258         spin_lock_irqsave(&p->sighand->siglock, flags);
 259         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
 260                                             cpu);
 261         spin_unlock_irqrestore(&p->sighand->siglock, flags);
 262         return ret;
 263 }
 264
 265
 266 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 267 {
 268         const pid_t pid = CPUCLOCK_PID(which_clock);
 269         int error = -EINVAL;
 270         union cpu_time_count rtn;
 271
 272         if (pid == 0) {
 273                 /*
 274                  * Special case constant value for our own clocks.
 275                  * We don't have to do any lookup to find ourselves.
 276                  */
 277                 if (CPUCLOCK_PERTHREAD(which_clock)) {
 278                         /*
 279                          * Sampling just ourselves we can do with no locking.
 280                          */
 281                         error = cpu_clock_sample(which_clock,
 282                                                  current, &rtn);
 283                 } else {
 284                         read_lock(&tasklist_lock);
 285                         error = cpu_clock_sample_group(which_clock,
 286                                                        current, &rtn);
 287                         read_unlock(&tasklist_lock);
 288                 }
 289         } else {
 290                 /*
 291                  * Find the given PID, and validate that the caller
 292                  * should be able to see it.
 293                  */
 294                 struct task_struct *p;
 295                 read_lock(&tasklist_lock);
 296                 p = find_task_by_pid(pid);
 297                 if (p) {
 298                         if (CPUCLOCK_PERTHREAD(which_clock)) {
 299                                 if (p->tgid == current->tgid) {
 300                                         error = cpu_clock_sample(which_clock,
 301                                                                  p, &rtn);
 302                                 }
 303                         } else if (p->tgid == pid && p->signal) {
 304                                 error = cpu_clock_sample_group(which_clock,
 305                                                                p, &rtn);
 306                         }
 307                 }
 308                 read_unlock(&tasklist_lock);
 309         }
 310
 311         if (error)
 312                 return error;
 313         sample_to_timespec(which_clock, rtn, tp);
 314         return 0;
 315 }
 316
 317
 318 /*
 319  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
 320  * This is called from sys_timer_create with the new timer already locked.
 321  */
 322 int posix_cpu_timer_create(struct k_itimer *new_timer)
 323 {
 324         int ret = 0;
 325         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
 326         struct task_struct *p;
 327
 328         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
 329                 return -EINVAL;
 330
 331         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
 332         new_timer->it.cpu.incr.sched = 0;
 333         new_timer->it.cpu.expires.sched = 0;
 334
 335         read_lock(&tasklist_lock);
 336         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
 337                 if (pid == 0) {
 338                         p = current;
 339                 } else {
 340                         p = find_task_by_pid(pid);
 341                         if (p && p->tgid != current->tgid)
 342                                 p = NULL;
 343                 }
 344         } else {
 345                 if (pid == 0) {
 346                         p = current->group_leader;
 347                 } else {
 348                         p = find_task_by_pid(pid);
 349                         if (p && p->tgid != pid)
 350                                 p = NULL;
 351                 }
 352         }
 353         new_timer->it.cpu.task = p;
 354         if (p) {
 355                 get_task_struct(p);
 356         } else {
 357                 ret = -EINVAL;
 358         }
 359         read_unlock(&tasklist_lock);
 360
 361         return ret;
 362 }
 363
 364 /*
 365  * Clean up a CPU-clock timer that is about to be destroyed.
 366  * This is called from timer deletion with the timer already locked.
 367  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 368  * and try again.  (This happens when the timer is in the middle of firing.)
 369  */
 370 int posix_cpu_timer_del(struct k_itimer *timer)
 371 {
 372         struct task_struct *p = timer->it.cpu.task;
 373         int ret = 0;
 374
 375         if (likely(p != NULL)) {
 376                 read_lock(&tasklist_lock);
 377                 if (unlikely(p->signal == NULL)) {
 378                         /*
 379                          * We raced with the reaping of the task.
 380                          * The deletion should have cleared us off the list.
 381                          */
 382                         BUG_ON(!list_empty(&timer->it.cpu.entry));
 383                 } else {
 384                         spin_lock(&p->sighand->siglock);
 385                         if (timer->it.cpu.firing)
 386                                 ret = TIMER_RETRY;
 387                         else
 388                                 list_del(&timer->it.cpu.entry);
 389                         spin_unlock(&p->sighand->siglock);
 390                 }
 391                 read_unlock(&tasklist_lock);
 392
 393                 if (!ret)
 394                         put_task_struct(p);
 395         }
 396
 397         return ret;
 398 }
 399
 400 /*
 401  * Clean out CPU timers still ticking when a thread exited.  The task
 402  * pointer is cleared, and the expiry time is replaced with the residual
 403  * time for later timer_gettime calls to return.
 404  * This must be called with the siglock held.
 405  */
 406 static void cleanup_timers(struct list_head *head,
 407                            cputime_t utime, cputime_t stime,
 408                            unsigned long long sched_time)
 409 {
 410         struct cpu_timer_list *timer, *next;
 411         cputime_t ptime = cputime_add(utime, stime);
 412
 413         list_for_each_entry_safe(timer, next, head, entry) {
 414                 list_del_init(&timer->entry);
 415                 if (cputime_lt(timer->expires.cpu, ptime)) {
 416                         timer->expires.cpu = cputime_zero;
 417                 } else {
 418                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 419                                                          ptime);
 420                 }
 421         }
 422
 423         ++head;
 424         list_for_each_entry_safe(timer, next, head, entry) {
 425                 list_del_init(&timer->entry);
 426                 if (cputime_lt(timer->expires.cpu, utime)) {
 427                         timer->expires.cpu = cputime_zero;
 428                 } else {
 429                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
 430                                                          utime);
 431                 }
 432         }
 433
 434         ++head;
 435         list_for_each_entry_safe(timer, next, head, entry) {
 436                 list_del_init(&timer->entry);
 437                 if (timer->expires.sched < sched_time) {
 438                         timer->expires.sched = 0;
 439                 } else {
 440                         timer->expires.sched -= sched_time;
 441                 }
 442         }
 443 }
 444
 445 /*
 446  * These are both called with the siglock held, when the current thread
 447  * is being reaped.  When the final (leader) thread in the group is reaped,
 448  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
 449  */
 450 void posix_cpu_timers_exit(struct task_struct *tsk)
 451 {
 452         cleanup_timers(tsk->cpu_timers,
 453                        tsk->utime, tsk->stime, tsk->sched_time);
 454
 455 }
 456 void posix_cpu_timers_exit_group(struct task_struct *tsk)
 457 {
 458         cleanup_timers(tsk->signal->cpu_timers,
 459                        cputime_add(tsk->utime, tsk->signal->utime),
 460                        cputime_add(tsk->stime, tsk->signal->stime),
 461                        tsk->sched_time + tsk->signal->sched_time);
 462 }
 463
 464
 465 /*
 466  * Set the expiry times of all the threads in the process so one of them
 467  * will go off before the process cumulative expiry total is reached.
 468  */
 469 static void process_timer_rebalance(struct task_struct *p,
 470                                     unsigned int clock_idx,
 471                                     union cpu_time_count expires,
 472                                     union cpu_time_count val)
 473 {
 474         cputime_t ticks, left;
 475         unsigned long long ns, nsleft;
 476         struct task_struct *t = p;
 477         unsigned int nthreads = atomic_read(&p->signal->live);
 478
 479         if (!nthreads)
 480                 return;
 481
 482         switch (clock_idx) {
 483         default:
 484                 BUG();
 485                 break;
 486         case CPUCLOCK_PROF:
 487                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 488                                    nthreads);
 489                 do {
 490                         if (likely(!(t->flags & PF_EXITING))) {
 491                                 ticks = cputime_add(prof_ticks(t), left);
 492                                 if (cputime_eq(t->it_prof_expires,
 493                                                cputime_zero) ||
 494                                     cputime_gt(t->it_prof_expires, ticks)) {
 495                                         t->it_prof_expires = ticks;
 496                                 }
 497                         }
 498                         t = next_thread(t);
 499                 } while (t != p);
 500                 break;
 501         case CPUCLOCK_VIRT:
 502                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
 503                                    nthreads);
 504                 do {
 505                         if (likely(!(t->flags & PF_EXITING))) {
 506                                 ticks = cputime_add(virt_ticks(t), left);
 507                                 if (cputime_eq(t->it_virt_expires,
 508                                                cputime_zero) ||
 509                                     cputime_gt(t->it_virt_expires, ticks)) {
 510                                         t->it_virt_expires = ticks;
 511                                 }
 512                         }
 513                         t = next_thread(t);
 514                 } while (t != p);
 515                 break;
 516         case CPUCLOCK_SCHED:
 517                 nsleft = expires.sched - val.sched;
 518                 do_div(nsleft, nthreads);
 519                 do {
 520                         if (likely(!(t->flags & PF_EXITING))) {
 521                                 ns = t->sched_time + nsleft;
 522                                 if (t->it_sched_expires == 0 ||
 523                                     t->it_sched_expires > ns) {
 524                                         t->it_sched_expires = ns;
 525                                 }
 526                         }
 527                         t = next_thread(t);
 528                 } while (t != p);
 529                 break;
 530         }
 531 }
 532
 533 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
 534 {
 535         /*
 536          * That's all for this thread or process.
 537          * We leave our residual in expires to be reported.
 538          */
 539         put_task_struct(timer->it.cpu.task);
 540         timer->it.cpu.task = NULL;
 541         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
 542                                              timer->it.cpu.expires,
 543                                              now);
 544 }
 545
 546 /*
 547  * Insert the timer on the appropriate list before any timers that
 548  * expire later.  This must be called with the tasklist_lock held
 549  * for reading, and interrupts disabled.
 550  */
 551 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
 552 {
 553         struct task_struct *p = timer->it.cpu.task;
 554         struct list_head *head, *listpos;
 555         struct cpu_timer_list *const nt = &timer->it.cpu;
 556         struct cpu_timer_list *next;
 557         unsigned long i;
 558
 559         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
 560                 p->cpu_timers : p->signal->cpu_timers);
 561         head += CPUCLOCK_WHICH(timer->it_clock);
 562
 563         BUG_ON(!irqs_disabled());
 564         spin_lock(&p->sighand->siglock);
 565
 566         listpos = head;
 567         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
 568                 list_for_each_entry(next, head, entry) {
 569                         if (next->expires.sched > nt->expires.sched)
 570                                 break;
 571                         listpos = &next->entry;
 572                 }
 573         } else {
 574                 list_for_each_entry(next, head, entry) {
 575                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
 576                                 break;
 577                         listpos = &next->entry;
 578                 }
 579         }
 580         list_add(&nt->entry, listpos);
 581
 582         if (listpos == head) {
 583                 /*
 584                  * We are the new earliest-expiring timer.
 585                  * If we are a thread timer, there can always
 586                  * be a process timer telling us to stop earlier.
 587                  */
 588
 589                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 590                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 591                         default:
 592                                 BUG();
 593                         case CPUCLOCK_PROF:
 594                                 if (cputime_eq(p->it_prof_expires,
 595                                                cputime_zero) ||
 596                                     cputime_gt(p->it_prof_expires,
 597                                                nt->expires.cpu))
 598                                         p->it_prof_expires = nt->expires.cpu;
 599                                 break;
 600                         case CPUCLOCK_VIRT:
 601                                 if (cputime_eq(p->it_virt_expires,
 602                                                cputime_zero) ||
 603                                     cputime_gt(p->it_virt_expires,
 604                                                nt->expires.cpu))
 605                                         p->it_virt_expires = nt->expires.cpu;
 606                                 break;
 607                         case CPUCLOCK_SCHED:
 608                                 if (p->it_sched_expires == 0 ||
 609                                     p->it_sched_expires > nt->expires.sched)
 610                                         p->it_sched_expires = nt->expires.sched;
 611                                 break;
 612                         }
 613                 } else {
 614                         /*
 615                          * For a process timer, we must balance
 616                          * all the live threads' expirations.
 617                          */
 618                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
 619                         default:
 620                                 BUG();
 621                         case CPUCLOCK_VIRT:
 622                                 if (!cputime_eq(p->signal->it_virt_expires,
 623                                                 cputime_zero) &&
 624                                     cputime_lt(p->signal->it_virt_expires,
 625                                                timer->it.cpu.expires.cpu))
 626                                         break;
 627                                 goto rebalance;
 628                         case CPUCLOCK_PROF:
 629                                 if (!cputime_eq(p->signal->it_prof_expires,
 630                                                 cputime_zero) &&
 631                                     cputime_lt(p->signal->it_prof_expires,
 632                                                timer->it.cpu.expires.cpu))
 633                                         break;
 634                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
 635                                 if (i != RLIM_INFINITY &&
 636                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
 637                                         break;
 638                                 goto rebalance;
 639                         case CPUCLOCK_SCHED:
 640                         rebalance:
 641                                 process_timer_rebalance(
 642                                         timer->it.cpu.task,
 643                                         CPUCLOCK_WHICH(timer->it_clock),
 644                                         timer->it.cpu.expires, now);
 645                                 break;
 646                         }
 647                 }
 648         }
 649
 650         spin_unlock(&p->sighand->siglock);
 651 }
 652
 653 /*
 654  * The timer is locked, fire it and arrange for its reload.
 655  */
 656 static void cpu_timer_fire(struct k_itimer *timer)
 657 {
 658         if (unlikely(timer->sigq == NULL)) {
 659                 /*
 660                  * This a special case for clock_nanosleep,
 661                  * not a normal timer from sys_timer_create.
 662                  */
 663                 wake_up_process(timer->it_process);
 664                 timer->it.cpu.expires.sched = 0;
 665         } else if (timer->it.cpu.incr.sched == 0) {
 666                 /*
 667                  * One-shot timer.  Clear it as soon as it's fired.
 668                  */
 669                 posix_timer_event(timer, 0);
 670                 timer->it.cpu.expires.sched = 0;
 671         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
 672                 /*
 673                  * The signal did not get queued because the signal
 674                  * was ignored, so we won't get any callback to
 675                  * reload the timer.  But we need to keep it
 676                  * ticking in case the signal is deliverable next time.
 677                  */
 678                 posix_cpu_timer_schedule(timer);
 679         }
 680 }
 681
 682 /*
 683  * Guts of sys_timer_settime for CPU timers.
 684  * This is called with the timer locked and interrupts disabled.
 685  * If we return TIMER_RETRY, it's necessary to release the timer's lock
 686  * and try again.  (This happens when the timer is in the middle of firing.)
 687  */
 688 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
 689                         struct itimerspec *new, struct itimerspec *old)
 690 {
 691         struct task_struct *p = timer->it.cpu.task;
 692         union cpu_time_count old_expires, new_expires, val;
 693         int ret;
 694
 695         if (unlikely(p == NULL)) {
 696                 /*
 697                  * Timer refers to a dead task's clock.
 698                  */
 699                 return -ESRCH;
 700         }
 701
 702         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
 703
 704         read_lock(&tasklist_lock);
 705         /*
 706          * We need the tasklist_lock to protect against reaping that
 707          * clears p->signal.  If p has just been reaped, we can no
 708          * longer get any information about it at all.
 709          */
 710         if (unlikely(p->signal == NULL)) {
 711                 read_unlock(&tasklist_lock);
 712                 put_task_struct(p);
 713                 timer->it.cpu.task = NULL;
 714                 return -ESRCH;
 715         }
 716
 717         /*
 718          * Disarm any old timer after extracting its expiry time.
 719          */
 720         BUG_ON(!irqs_disabled());
 721
 722         ret = 0;
 723         spin_lock(&p->sighand->siglock);
 724         old_expires = timer->it.cpu.expires;
 725         if (unlikely(timer->it.cpu.firing)) {
 726                 timer->it.cpu.firing = -1;
 727                 ret = TIMER_RETRY;
 728         } else
 729                 list_del_init(&timer->it.cpu.entry);
 730         spin_unlock(&p->sighand->siglock);
 731
 732         /*
 733          * We need to sample the current value to convert the new
 734          * value from to relative and absolute, and to convert the
 735          * old value from absolute to relative.  To set a process
 736          * timer, we need a sample to balance the thread expiry
 737          * times (in arm_timer).  With an absolute time, we must
 738          * check if it's already passed.  In short, we need a sample.
 739          */
 740         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 741                 cpu_clock_sample(timer->it_clock, p, &val);
 742         } else {
 743                 cpu_clock_sample_group(timer->it_clock, p, &val);
 744         }
 745
 746         if (old) {
 747                 if (old_expires.sched == 0) {
 748                         old->it_value.tv_sec = 0;
 749                         old->it_value.tv_nsec = 0;
 750                 } else {
 751                         /*
 752                          * Update the timer in case it has
 753                          * overrun already.  If it has,
 754                          * we'll report it as having overrun
 755                          * and with the next reloaded timer
 756                          * already ticking, though we are
 757                          * swallowing that pending
 758                          * notification here to install the
 759                          * new setting.
 760                          */
 761                         bump_cpu_timer(timer, val);
 762                         if (cpu_time_before(timer->it_clock, val,
 763                                             timer->it.cpu.expires)) {
 764                                 old_expires = cpu_time_sub(
 765                                         timer->it_clock,
 766                                         timer->it.cpu.expires, val);
 767                                 sample_to_timespec(timer->it_clock,
 768                                                    old_expires,
 769                                                    &old->it_value);
 770                         } else {
 771                                 old->it_value.tv_nsec = 1;
 772                                 old->it_value.tv_sec = 0;
 773                         }
 774                 }
 775         }
 776
 777         if (unlikely(ret)) {
 778                 /*
 779                  * We are colliding with the timer actually firing.
 780                  * Punt after filling in the timer's old value, and
 781                  * disable this firing since we are already reporting
 782                  * it as an overrun (thanks to bump_cpu_timer above).
 783                  */
 784                 read_unlock(&tasklist_lock);
 785                 goto out;
 786         }
 787
 788         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
 789                 cpu_time_add(timer->it_clock, &new_expires, val);
 790         }
 791
 792         /*
 793          * Install the new expiry time (or zero).
 794          * For a timer with no notification action, we don't actually
 795          * arm the timer (we'll just fake it for timer_gettime).
 796          */
 797         timer->it.cpu.expires = new_expires;
 798         if (new_expires.sched != 0 &&
 799             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 800             cpu_time_before(timer->it_clock, val, new_expires)) {
 801                 arm_timer(timer, val);
 802         }
 803
 804         read_unlock(&tasklist_lock);
 805
 806         /*
 807          * Install the new reload setting, and
 808          * set up the signal and overrun bookkeeping.
 809          */
 810         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
 811                                                 &new->it_interval);
 812
 813         /*
 814          * This acts as a modification timestamp for the timer,
 815          * so any automatic reload attempt will punt on seeing
 816          * that we have reset the timer manually.
 817          */
 818         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
 819                 ~REQUEUE_PENDING;
 820         timer->it_overrun_last = 0;
 821         timer->it_overrun = -1;
 822
 823         if (new_expires.sched != 0 &&
 824             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
 825             !cpu_time_before(timer->it_clock, val, new_expires)) {
 826                 /*
 827                  * The designated time already passed, so we notify
 828                  * immediately, even if the thread never runs to
 829                  * accumulate more time on this clock.
 830                  */
 831                 cpu_timer_fire(timer);
 832         }
 833
 834         ret = 0;
 835  out:
 836         if (old) {
 837                 sample_to_timespec(timer->it_clock,
 838                                    timer->it.cpu.incr, &old->it_interval);
 839         }
 840         return ret;
 841 }
 842
 843 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
 844 {
 845         union cpu_time_count now;
 846         struct task_struct *p = timer->it.cpu.task;
 847         int clear_dead;
 848
 849         /*
 850          * Easy part: convert the reload time.
 851          */
 852         sample_to_timespec(timer->it_clock,
 853                            timer->it.cpu.incr, &itp->it_interval);
 854
 855         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
 856                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 857                 return;
 858         }
 859
 860         if (unlikely(p == NULL)) {
 861                 /*
 862                  * This task already died and the timer will never fire.
 863                  * In this case, expires is actually the dead value.
 864                  */
 865         dead:
 866                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
 867                                    &itp->it_value);
 868                 return;
 869         }
 870
 871         /*
 872          * Sample the clock to take the difference with the expiry time.
 873          */
 874         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
 875                 cpu_clock_sample(timer->it_clock, p, &now);
 876                 clear_dead = p->exit_state;
 877         } else {
 878                 read_lock(&tasklist_lock);
 879                 if (unlikely(p->signal == NULL)) {
 880                         /*
 881                          * The process has been reaped.
 882                          * We can't even collect a sample any more.
 883                          * Call the timer disarmed, nothing else to do.
 884                          */
 885                         put_task_struct(p);
 886                         timer->it.cpu.task = NULL;
 887                         timer->it.cpu.expires.sched = 0;
 888                         read_unlock(&tasklist_lock);
 889                         goto dead;
 890                 } else {
 891                         cpu_clock_sample_group(timer->it_clock, p, &now);
 892                         clear_dead = (unlikely(p->exit_state) &&
 893                                       thread_group_empty(p));
 894                 }
 895                 read_unlock(&tasklist_lock);
 896         }
 897
 898         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
 899                 if (timer->it.cpu.incr.sched == 0 &&
 900                     cpu_time_before(timer->it_clock,
 901                                     timer->it.cpu.expires, now)) {
 902                         /*
 903                          * Do-nothing timer expired and has no reload,
 904                          * so it's as if it was never set.
 905                          */
 906                         timer->it.cpu.expires.sched = 0;
 907                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
 908                         return;
 909                 }
 910                 /*
 911                  * Account for any expirations and reloads that should
 912                  * have happened.
 913                  */
 914                 bump_cpu_timer(timer, now);
 915         }
 916
 917         if (unlikely(clear_dead)) {
 918                 /*
 919                  * We've noticed that the thread is dead, but
 920                  * not yet reaped.  Take this opportunity to
 921                  * drop our task ref.
 922                  */
 923                 clear_dead_task(timer, now);
 924                 goto dead;
 925         }
 926
 927         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
 928                 sample_to_timespec(timer->it_clock,
 929                                    cpu_time_sub(timer->it_clock,
 930                                                 timer->it.cpu.expires, now),
 931                                    &itp->it_value);
 932         } else {
 933                 /*
 934                  * The timer should have expired already, but the firing
 935                  * hasn't taken place yet.  Say it's just about to expire.
 936                  */
 937                 itp->it_value.tv_nsec = 1;
 938                 itp->it_value.tv_sec = 0;
 939         }
 940 }
 941
 942 /*
 943  * Check for any per-thread CPU timers that have fired and move them off
 944  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
 945  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
 946  */
 947 static void check_thread_timers(struct task_struct *tsk,
 948                                 struct list_head *firing)
 949 {
 950         int maxfire;
 951         struct list_head *timers = tsk->cpu_timers;
 952
 953         maxfire = 20;
 954         tsk->it_prof_expires = cputime_zero;
 955         while (!list_empty(timers)) {
 956                 struct cpu_timer_list *t = list_entry(timers->next,
 957                                                       struct cpu_timer_list,
 958                                                       entry);
 959                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
 960                         tsk->it_prof_expires = t->expires.cpu;
 961                         break;
 962                 }
 963                 t->firing = 1;
 964                 list_move_tail(&t->entry, firing);
 965         }
 966
 967         ++timers;
 968         maxfire = 20;
 969         tsk->it_virt_expires = cputime_zero;
 970         while (!list_empty(timers)) {
 971                 struct cpu_timer_list *t = list_entry(timers->next,
 972                                                       struct cpu_timer_list,
 973                                                       entry);
 974                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
 975                         tsk->it_virt_expires = t->expires.cpu;
 976                         break;
 977                 }
 978                 t->firing = 1;
 979                 list_move_tail(&t->entry, firing);
 980         }
 981
 982         ++timers;
 983         maxfire = 20;
 984         tsk->it_sched_expires = 0;
 985         while (!list_empty(timers)) {
 986                 struct cpu_timer_list *t = list_entry(timers->next,
 987                                                       struct cpu_timer_list,
 988                                                       entry);
 989                 if (!--maxfire || tsk->sched_time < t->expires.sched) {
 990                         tsk->it_sched_expires = t->expires.sched;
 991                         break;
 992                 }
 993                 t->firing = 1;
 994                 list_move_tail(&t->entry, firing);
 995         }
 996 }
 997
 998 /*
 999  * Check for any per-thread CPU timers that have fired and move them
1000  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1001  * have already been taken off.
1002  */
1003 static void check_process_timers(struct task_struct *tsk,
1004                                  struct list_head *firing)
1005 {
1006         int maxfire;
1007         struct signal_struct *const sig = tsk->signal;
1008         cputime_t utime, stime, ptime, virt_expires, prof_expires;
1009         unsigned long long sched_time, sched_expires;
1010         struct task_struct *t;
1011         struct list_head *timers = sig->cpu_timers;
1012
1013         /*
1014          * Don't sample the current process CPU clocks if there are no timers.
1015          */
1016         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1017             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1018             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1019             list_empty(&timers[CPUCLOCK_VIRT]) &&
1020             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1021             list_empty(&timers[CPUCLOCK_SCHED]))
1022                 return;
1023
1024         /*
1025          * Collect the current process totals.
1026          */
1027         utime = sig->utime;
1028         stime = sig->stime;
1029         sched_time = sig->sched_time;
1030         t = tsk;
1031         do {
1032                 utime = cputime_add(utime, t->utime);
1033                 stime = cputime_add(stime, t->stime);
1034                 sched_time += t->sched_time;
1035                 t = next_thread(t);
1036         } while (t != tsk);
1037         ptime = cputime_add(utime, stime);
1038
1039         maxfire = 20;
1040         prof_expires = cputime_zero;
1041         while (!list_empty(timers)) {
1042                 struct cpu_timer_list *t = list_entry(timers->next,
1043                                                       struct cpu_timer_list,
1044                                                       entry);
1045                 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
1046                         prof_expires = t->expires.cpu;
1047                         break;
1048                 }
1049                 t->firing = 1;
1050                 list_move_tail(&t->entry, firing);
1051         }
1052
1053         ++timers;
1054         maxfire = 20;
1055         virt_expires = cputime_zero;
1056         while (!list_empty(timers)) {
1057                 struct cpu_timer_list *t = list_entry(timers->next,
1058                                                       struct cpu_timer_list,
1059                                                       entry);
1060                 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
1061                         virt_expires = t->expires.cpu;
1062                         break;
1063                 }
1064                 t->firing = 1;
1065                 list_move_tail(&t->entry, firing);
1066         }
1067
1068         ++timers;
1069         maxfire = 20;
1070         sched_expires = 0;
1071         while (!list_empty(timers)) {
1072                 struct cpu_timer_list *t = list_entry(timers->next,
1073                                                       struct cpu_timer_list,
1074                                                       entry);
1075                 if (!--maxfire || sched_time < t->expires.sched) {
1076                         sched_expires = t->expires.sched;
1077                         break;
1078                 }
1079                 t->firing = 1;
1080                 list_move_tail(&t->entry, firing);
1081         }
1082
1083         /*
1084          * Check for the special case process timers.
1085          */
1086         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1087                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1088                         /* ITIMER_PROF fires and reloads.  */
1089                         sig->it_prof_expires = sig->it_prof_incr;
1090                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1091                                 sig->it_prof_expires = cputime_add(
1092                                         sig->it_prof_expires, ptime);
1093                         }
1094                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1095                 }
1096                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1097                     (cputime_eq(prof_expires, cputime_zero) ||
1098                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1099                         prof_expires = sig->it_prof_expires;
1100                 }
1101         }
1102         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1103                 if (cputime_ge(utime, sig->it_virt_expires)) {
1104                         /* ITIMER_VIRTUAL fires and reloads.  */
1105                         sig->it_virt_expires = sig->it_virt_incr;
1106                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1107                                 sig->it_virt_expires = cputime_add(
1108                                         sig->it_virt_expires, utime);
1109                         }
1110                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1111                 }
1112                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1113                     (cputime_eq(virt_expires, cputime_zero) ||
1114                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1115                         virt_expires = sig->it_virt_expires;
1116                 }
1117         }
1118         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1119                 unsigned long psecs = cputime_to_secs(ptime);
1120                 cputime_t x;
1121                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1122                         /*
1123                          * At the hard limit, we just die.
1124                          * No need to calculate anything else now.
1125                          */
1126                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1127                         return;
1128                 }
1129                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1130                         /*
1131                          * At the soft limit, send a SIGXCPU every second.
1132                          */
1133                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1134                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1135                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1136                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1137                         }
1138                 }
1139                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1140                 if (cputime_eq(prof_expires, cputime_zero) ||
1141                     cputime_lt(x, prof_expires)) {
1142                         prof_expires = x;
1143                 }
1144         }
1145
1146         if (!cputime_eq(prof_expires, cputime_zero) ||
1147             !cputime_eq(virt_expires, cputime_zero) ||
1148             sched_expires != 0) {
1149                 /*
1150                  * Rebalance the threads' expiry times for the remaining
1151                  * process CPU timers.
1152                  */
1153
1154                 cputime_t prof_left, virt_left, ticks;
1155                 unsigned long long sched_left, sched;
1156                 const unsigned int nthreads = atomic_read(&sig->live);
1157
1158                 if (!nthreads)
1159                         return;
1160
1161                 prof_left = cputime_sub(prof_expires, utime);
1162                 prof_left = cputime_sub(prof_left, stime);
1163                 prof_left = cputime_div(prof_left, nthreads);
1164                 virt_left = cputime_sub(virt_expires, utime);
1165                 virt_left = cputime_div(virt_left, nthreads);
1166                 if (sched_expires) {
1167                         sched_left = sched_expires - sched_time;
1168                         do_div(sched_left, nthreads);
1169                 } else {
1170                         sched_left = 0;
1171                 }
1172                 t = tsk;
1173                 do {
1174                         if (unlikely(t->flags & PF_EXITING))
1175                                 continue;
1176
1177                         ticks = cputime_add(cputime_add(t->utime, t->stime),
1178                                             prof_left);
1179                         if (!cputime_eq(prof_expires, cputime_zero) &&
1180                             (cputime_eq(t->it_prof_expires, cputime_zero) ||
1181                              cputime_gt(t->it_prof_expires, ticks))) {
1182                                 t->it_prof_expires = ticks;
1183                         }
1184
1185                         ticks = cputime_add(t->utime, virt_left);
1186                         if (!cputime_eq(virt_expires, cputime_zero) &&
1187                             (cputime_eq(t->it_virt_expires, cputime_zero) ||
1188                              cputime_gt(t->it_virt_expires, ticks))) {
1189                                 t->it_virt_expires = ticks;
1190                         }
1191
1192                         sched = t->sched_time + sched_left;
1193                         if (sched_expires && (t->it_sched_expires == 0 ||
1194                                               t->it_sched_expires > sched)) {
1195                                 t->it_sched_expires = sched;
1196                         }
1197                 } while ((t = next_thread(t)) != tsk);
1198         }
1199 }
1200
1201 /*
1202  * This is called from the signal code (via do_schedule_next_timer)
1203  * when the last timer signal was delivered and we have to reload the timer.
1204  */
1205 void posix_cpu_timer_schedule(struct k_itimer *timer)
1206 {
1207         struct task_struct *p = timer->it.cpu.task;
1208         union cpu_time_count now;
1209
1210         if (unlikely(p == NULL))
1211                 /*
1212                  * The task was cleaned up already, no future firings.
1213                  */
1214                 goto out;
1215
1216         /*
1217          * Fetch the current sample and update the timer's expiry time.
1218          */
1219         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1220                 cpu_clock_sample(timer->it_clock, p, &now);
1221                 bump_cpu_timer(timer, now);
1222                 if (unlikely(p->exit_state)) {
1223                         clear_dead_task(timer, now);
1224                         goto out;
1225                 }
1226                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1227         } else {
1228                 read_lock(&tasklist_lock);
1229                 if (unlikely(p->signal == NULL)) {
1230                         /*
1231                          * The process has been reaped.
1232                          * We can't even collect a sample any more.
1233                          */
1234                         put_task_struct(p);
1235                         timer->it.cpu.task = p = NULL;
1236                         timer->it.cpu.expires.sched = 0;
1237                         goto out_unlock;
1238                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1239                         /*
1240                          * We've noticed that the thread is dead, but
1241                          * not yet reaped.  Take this opportunity to
1242                          * drop our task ref.
1243                          */
1244                         clear_dead_task(timer, now);
1245                         goto out_unlock;
1246                 }
1247                 cpu_clock_sample_group(timer->it_clock, p, &now);
1248                 bump_cpu_timer(timer, now);
1249                 /* Leave the tasklist_lock locked for the call below.  */
1250         }
1251
1252         /*
1253          * Now re-arm for the new expiry time.
1254          */
1255         arm_timer(timer, now);
1256
1257 out_unlock:
1258         read_unlock(&tasklist_lock);
1259
1260 out:
1261         timer->it_overrun_last = timer->it_overrun;
1262         timer->it_overrun = -1;
1263         ++timer->it_requeue_pending;
1264 }
1265
1266 /*
1267  * This is called from the timer interrupt handler.  The irq handler has
1268  * already updated our counts.  We need to check if any timers fire now.
1269  * Interrupts are disabled.
1270  */
1271 void run_posix_cpu_timers(struct task_struct *tsk)
1272 {
1273         LIST_HEAD(firing);
1274         struct k_itimer *timer, *next;
1275
1276         BUG_ON(!irqs_disabled());
1277
1278 #define UNEXPIRED(clock) \
1279                 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1280                  cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1281
1282         if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1283             (tsk->it_sched_expires == 0 ||
1284              tsk->sched_time < tsk->it_sched_expires))
1285                 return;
1286
1287 #undef  UNEXPIRED
1288
1289         /*
1290          * Double-check with locks held.
1291          */
1292         read_lock(&tasklist_lock);
1293         if (likely(tsk->signal != NULL)) {
1294                 spin_lock(&tsk->sighand->siglock);
1295
1296                 /*
1297                  * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1298                  * all the timers that are firing, and put them on the firing list.
1299                  */
1300                 check_thread_timers(tsk, &firing);
1301                 check_process_timers(tsk, &firing);
1302
1303                 /*
1304                  * We must release these locks before taking any timer's lock.
1305                  * There is a potential race with timer deletion here, as the
1306                  * siglock now protects our private firing list.  We have set
1307                  * the firing flag in each timer, so that a deletion attempt
1308                  * that gets the timer lock before we do will give it up and
1309                  * spin until we've taken care of that timer below.
1310                  */
1311                 spin_unlock(&tsk->sighand->siglock);
1312         }
1313         read_unlock(&tasklist_lock);
1314
1315         /*
1316          * Now that all the timers on our list have the firing flag,
1317          * noone will touch their list entries but us.  We'll take
1318          * each timer's lock before clearing its firing flag, so no
1319          * timer call will interfere.
1320          */
1321         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1322                 int firing;
1323                 spin_lock(&timer->it_lock);
1324                 list_del_init(&timer->it.cpu.entry);
1325                 firing = timer->it.cpu.firing;
1326                 timer->it.cpu.firing = 0;
1327                 /*
1328                  * The firing flag is -1 if we collided with a reset
1329                  * of the timer, which already reported this
1330                  * almost-firing as an overrun.  So don't generate an event.
1331                  */
1332                 if (likely(firing >= 0)) {
1333                         cpu_timer_fire(timer);
1334                 }
1335                 spin_unlock(&timer->it_lock);
1336         }
1337 }
1338
1339 /*
1340  * Set one of the process-wide special case CPU timers.
1341  * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1342  * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1343  * absolute; non-null for ITIMER_*, where *newval is relative and we update
1344  * it to be absolute, *oldval is absolute and we update it to be relative.
1345  */
1346 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1347                            cputime_t *newval, cputime_t *oldval)
1348 {
1349         union cpu_time_count now;
1350         struct list_head *head;
1351
1352         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1353         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1354
1355         if (oldval) {
1356                 if (!cputime_eq(*oldval, cputime_zero)) {
1357                         if (cputime_le(*oldval, now.cpu)) {
1358                                 /* Just about to fire. */
1359                                 *oldval = jiffies_to_cputime(1);
1360                         } else {
1361                                 *oldval = cputime_sub(*oldval, now.cpu);
1362                         }
1363                 }
1364
1365                 if (cputime_eq(*newval, cputime_zero))
1366                         return;
1367                 *newval = cputime_add(*newval, now.cpu);
1368
1369                 /*
1370                  * If the RLIMIT_CPU timer will expire before the
1371                  * ITIMER_PROF timer, we have nothing else to do.
1372                  */
1373                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1374                     < cputime_to_secs(*newval))
1375                         return;
1376         }
1377
1378         /*
1379          * Check whether there are any process timers already set to fire
1380          * before this one.  If so, we don't have anything more to do.
1381          */
1382         head = &tsk->signal->cpu_timers[clock_idx];
1383         if (list_empty(head) ||
1384             cputime_ge(list_entry(head->next,
1385                                   struct cpu_timer_list, entry)->expires.cpu,
1386                        *newval)) {
1387                 /*
1388                  * Rejigger each thread's expiry time so that one will
1389                  * notice before we hit the process-cumulative expiry time.
1390                  */
1391                 union cpu_time_count expires = { .sched = 0 };
1392                 expires.cpu = *newval;
1393                 process_timer_rebalance(tsk, clock_idx, expires, now);
1394         }
1395 }
1396
1397 static long posix_cpu_clock_nanosleep_restart(struct restart_block *);
1398
1399 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1400                      struct timespec *rqtp, struct timespec __user *rmtp)
1401 {
1402         struct restart_block *restart_block =
1403             &current_thread_info()->restart_block;
1404         struct k_itimer timer;
1405         int error;
1406
1407         /*
1408          * Diagnose required errors first.
1409          */
1410         if (CPUCLOCK_PERTHREAD(which_clock) &&
1411             (CPUCLOCK_PID(which_clock) == 0 ||
1412              CPUCLOCK_PID(which_clock) == current->pid))
1413                 return -EINVAL;
1414
1415         /*
1416          * Set up a temporary timer and then wait for it to go off.
1417          */
1418         memset(&timer, 0, sizeof timer);
1419         spin_lock_init(&timer.it_lock);
1420         timer.it_clock = which_clock;
1421         timer.it_overrun = -1;
1422         error = posix_cpu_timer_create(&timer);
1423         timer.it_process = current;
1424         if (!error) {
1425                 static struct itimerspec zero_it;
1426                 struct itimerspec it = { .it_value = *rqtp,
1427                                          .it_interval = {} };
1428
1429                 spin_lock_irq(&timer.it_lock);
1430                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1431                 if (error) {
1432                         spin_unlock_irq(&timer.it_lock);
1433                         return error;
1434                 }
1435
1436                 while (!signal_pending(current)) {
1437                         if (timer.it.cpu.expires.sched == 0) {
1438                                 /*
1439                                  * Our timer fired and was reset.
1440                                  */
1441                                 spin_unlock_irq(&timer.it_lock);
1442                                 return 0;
1443                         }
1444
1445                         /*
1446                          * Block until cpu_timer_fire (or a signal) wakes us.
1447                          */
1448                         __set_current_state(TASK_INTERRUPTIBLE);
1449                         spin_unlock_irq(&timer.it_lock);
1450                         schedule();
1451                         spin_lock_irq(&timer.it_lock);
1452                 }
1453
1454                 /*
1455                  * We were interrupted by a signal.
1456                  */
1457                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1458                 posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1459                 spin_unlock_irq(&timer.it_lock);
1460
1461                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1462                         /*
1463                          * It actually did fire already.
1464                          */
1465                         return 0;
1466                 }
1467
1468                 /*
1469                  * Report back to the user the time still remaining.
1470                  */
1471                 if (rmtp != NULL && !(flags & TIMER_ABSTIME) &&
1472                     copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1473                         return -EFAULT;
1474
1475                 restart_block->fn = posix_cpu_clock_nanosleep_restart;
1476                 /* Caller already set restart_block->arg1 */
1477                 restart_block->arg0 = which_clock;
1478                 restart_block->arg1 = (unsigned long) rmtp;
1479                 restart_block->arg2 = rqtp->tv_sec;
1480                 restart_block->arg3 = rqtp->tv_nsec;
1481
1482                 error = -ERESTART_RESTARTBLOCK;
1483         }
1484
1485         return error;
1486 }
1487
1488 static long
1489 posix_cpu_clock_nanosleep_restart(struct restart_block *restart_block)
1490 {
1491         clockid_t which_clock = restart_block->arg0;
1492         struct timespec __user *rmtp;
1493         struct timespec t;
1494
1495         rmtp = (struct timespec __user *) restart_block->arg1;
1496         t.tv_sec = restart_block->arg2;
1497         t.tv_nsec = restart_block->arg3;
1498
1499         restart_block->fn = do_no_restart_syscall;
1500         return posix_cpu_nsleep(which_clock, TIMER_ABSTIME, &t, rmtp);
1501 }
1502
1503
1504 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1505 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1506
1507 static int process_cpu_clock_getres(const clockid_t which_clock,
1508                                     struct timespec *tp)
1509 {
1510         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1511 }
1512 static int process_cpu_clock_get(const clockid_t which_clock,
1513                                  struct timespec *tp)
1514 {
1515         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1516 }
1517 static int process_cpu_timer_create(struct k_itimer *timer)
1518 {
1519         timer->it_clock = PROCESS_CLOCK;
1520         return posix_cpu_timer_create(timer);
1521 }
1522 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1523                               struct timespec *rqtp,
1524                               struct timespec __user *rmtp)
1525 {
1526         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1527 }
1528 static int thread_cpu_clock_getres(const clockid_t which_clock,
1529                                    struct timespec *tp)
1530 {
1531         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1532 }
1533 static int thread_cpu_clock_get(const clockid_t which_clock,
1534                                 struct timespec *tp)
1535 {
1536         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1537 }
1538 static int thread_cpu_timer_create(struct k_itimer *timer)
1539 {
1540         timer->it_clock = THREAD_CLOCK;
1541         return posix_cpu_timer_create(timer);
1542 }
1543 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1544                               struct timespec *rqtp, struct timespec __user *rmtp)
1545 {
1546         return -EINVAL;
1547 }
1548
1549 static __init int init_posix_cpu_timers(void)
1550 {
1551         struct k_clock process = {
1552                 .clock_getres = process_cpu_clock_getres,
1553                 .clock_get = process_cpu_clock_get,
1554                 .clock_set = do_posix_clock_nosettime,
1555                 .timer_create = process_cpu_timer_create,
1556                 .nsleep = process_cpu_nsleep,
1557         };
1558         struct k_clock thread = {
1559                 .clock_getres = thread_cpu_clock_getres,
1560                 .clock_get = thread_cpu_clock_get,
1561                 .clock_set = do_posix_clock_nosettime,
1562                 .timer_create = thread_cpu_timer_create,
1563                 .nsleep = thread_cpu_nsleep,
1564         };
1565
1566         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1567         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1568
1569         return 0;
1570 }
1571 __initcall(init_posix_cpu_timers);