Revert "remove false BUG_ON() from run_posix_cpu_timers()"
[linux-2.6.git] / 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
10 static int check_clock(clockid_t which_clock)
11 {
12         int error = 0;
13         struct task_struct *p;
14         const pid_t pid = CPUCLOCK_PID(which_clock);
15
16         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
17                 return -EINVAL;
18
19         if (pid == 0)
20                 return 0;
21
22         read_lock(&tasklist_lock);
23         p = find_task_by_pid(pid);
24         if (!p || (CPUCLOCK_PERTHREAD(which_clock) ?
25                    p->tgid != current->tgid : p->tgid != pid)) {
26                 error = -EINVAL;
27         }
28         read_unlock(&tasklist_lock);
29
30         return error;
31 }
32
33 static inline union cpu_time_count
34 timespec_to_sample(clockid_t which_clock, const struct timespec *tp)
35 {
36         union cpu_time_count ret;
37         ret.sched = 0;          /* high half always zero when .cpu used */
38         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
39                 ret.sched = tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
40         } else {
41                 ret.cpu = timespec_to_cputime(tp);
42         }
43         return ret;
44 }
45
46 static void sample_to_timespec(clockid_t which_clock,
47                                union cpu_time_count cpu,
48                                struct timespec *tp)
49 {
50         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
51                 tp->tv_sec = div_long_long_rem(cpu.sched,
52                                                NSEC_PER_SEC, &tp->tv_nsec);
53         } else {
54                 cputime_to_timespec(cpu.cpu, tp);
55         }
56 }
57
58 static inline int cpu_time_before(clockid_t which_clock,
59                                   union cpu_time_count now,
60                                   union cpu_time_count then)
61 {
62         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
63                 return now.sched < then.sched;
64         }  else {
65                 return cputime_lt(now.cpu, then.cpu);
66         }
67 }
68 static inline void cpu_time_add(clockid_t which_clock,
69                                 union cpu_time_count *acc,
70                                 union cpu_time_count val)
71 {
72         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
73                 acc->sched += val.sched;
74         }  else {
75                 acc->cpu = cputime_add(acc->cpu, val.cpu);
76         }
77 }
78 static inline union cpu_time_count cpu_time_sub(clockid_t which_clock,
79                                                 union cpu_time_count a,
80                                                 union cpu_time_count b)
81 {
82         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
83                 a.sched -= b.sched;
84         }  else {
85                 a.cpu = cputime_sub(a.cpu, b.cpu);
86         }
87         return a;
88 }
89
90 /*
91  * Update expiry time from increment, and increase overrun count,
92  * given the current clock sample.
93  */
94 static void bump_cpu_timer(struct k_itimer *timer,
95                                   union cpu_time_count now)
96 {
97         int i;
98
99         if (timer->it.cpu.incr.sched == 0)
100                 return;
101
102         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
103                 unsigned long long delta, incr;
104
105                 if (now.sched < timer->it.cpu.expires.sched)
106                         return;
107                 incr = timer->it.cpu.incr.sched;
108                 delta = now.sched + incr - timer->it.cpu.expires.sched;
109                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
110                 for (i = 0; incr < delta - incr; i++)
111                         incr = incr << 1;
112                 for (; i >= 0; incr >>= 1, i--) {
113                         if (delta < incr)
114                                 continue;
115                         timer->it.cpu.expires.sched += incr;
116                         timer->it_overrun += 1 << i;
117                         delta -= incr;
118                 }
119         } else {
120                 cputime_t delta, incr;
121
122                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
123                         return;
124                 incr = timer->it.cpu.incr.cpu;
125                 delta = cputime_sub(cputime_add(now.cpu, incr),
126                                     timer->it.cpu.expires.cpu);
127                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
128                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
129                              incr = cputime_add(incr, incr);
130                 for (; i >= 0; incr = cputime_halve(incr), i--) {
131                         if (cputime_lt(delta, incr))
132                                 continue;
133                         timer->it.cpu.expires.cpu =
134                                 cputime_add(timer->it.cpu.expires.cpu, incr);
135                         timer->it_overrun += 1 << i;
136                         delta = cputime_sub(delta, incr);
137                 }
138         }
139 }
140
141 static inline cputime_t prof_ticks(struct task_struct *p)
142 {
143         return cputime_add(p->utime, p->stime);
144 }
145 static inline cputime_t virt_ticks(struct task_struct *p)
146 {
147         return p->utime;
148 }
149 static inline unsigned long long sched_ns(struct task_struct *p)
150 {
151         return (p == current) ? current_sched_time(p) : p->sched_time;
152 }
153
154 int posix_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
155 {
156         int error = check_clock(which_clock);
157         if (!error) {
158                 tp->tv_sec = 0;
159                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
160                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
161                         /*
162                          * If sched_clock is using a cycle counter, we
163                          * don't have any idea of its true resolution
164                          * exported, but it is much more than 1s/HZ.
165                          */
166                         tp->tv_nsec = 1;
167                 }
168         }
169         return error;
170 }
171
172 int posix_cpu_clock_set(clockid_t which_clock, const struct timespec *tp)
173 {
174         /*
175          * You can never reset a CPU clock, but we check for other errors
176          * in the call before failing with EPERM.
177          */
178         int error = check_clock(which_clock);
179         if (error == 0) {
180                 error = -EPERM;
181         }
182         return error;
183 }
184
185
186 /*
187  * Sample a per-thread clock for the given task.
188  */
189 static int cpu_clock_sample(clockid_t which_clock, struct task_struct *p,
190                             union cpu_time_count *cpu)
191 {
192         switch (CPUCLOCK_WHICH(which_clock)) {
193         default:
194                 return -EINVAL;
195         case CPUCLOCK_PROF:
196                 cpu->cpu = prof_ticks(p);
197                 break;
198         case CPUCLOCK_VIRT:
199                 cpu->cpu = virt_ticks(p);
200                 break;
201         case CPUCLOCK_SCHED:
202                 cpu->sched = sched_ns(p);
203                 break;
204         }
205         return 0;
206 }
207
208 /*
209  * Sample a process (thread group) clock for the given group_leader task.
210  * Must be called with tasklist_lock held for reading.
211  * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
212  */
213 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
214                                          struct task_struct *p,
215                                          union cpu_time_count *cpu)
216 {
217         struct task_struct *t = p;
218         switch (clock_idx) {
219         default:
220                 return -EINVAL;
221         case CPUCLOCK_PROF:
222                 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
223                 do {
224                         cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
225                         t = next_thread(t);
226                 } while (t != p);
227                 break;
228         case CPUCLOCK_VIRT:
229                 cpu->cpu = p->signal->utime;
230                 do {
231                         cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
232                         t = next_thread(t);
233                 } while (t != p);
234                 break;
235         case CPUCLOCK_SCHED:
236                 cpu->sched = p->signal->sched_time;
237                 /* Add in each other live thread.  */
238                 while ((t = next_thread(t)) != p) {
239                         cpu->sched += t->sched_time;
240                 }
241                 if (p->tgid == current->tgid) {
242                         /*
243                          * We're sampling ourselves, so include the
244                          * cycles not yet banked.  We still omit
245                          * other threads running on other CPUs,
246                          * so the total can always be behind as
247                          * much as max(nthreads-1,ncpus) * (NSEC_PER_SEC/HZ).
248                          */
249                         cpu->sched += current_sched_time(current);
250                 } else {
251                         cpu->sched += p->sched_time;
252                 }
253                 break;
254         }
255         return 0;
256 }
257
258 /*
259  * Sample a process (thread group) clock for the given group_leader task.
260  * Must be called with tasklist_lock held for reading.
261  */
262 static int cpu_clock_sample_group(clockid_t which_clock,
263                                   struct task_struct *p,
264                                   union cpu_time_count *cpu)
265 {
266         int ret;
267         unsigned long flags;
268         spin_lock_irqsave(&p->sighand->siglock, flags);
269         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
270                                             cpu);
271         spin_unlock_irqrestore(&p->sighand->siglock, flags);
272         return ret;
273 }
274
275
276 int posix_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
277 {
278         const pid_t pid = CPUCLOCK_PID(which_clock);
279         int error = -EINVAL;
280         union cpu_time_count rtn;
281
282         if (pid == 0) {
283                 /*
284                  * Special case constant value for our own clocks.
285                  * We don't have to do any lookup to find ourselves.
286                  */
287                 if (CPUCLOCK_PERTHREAD(which_clock)) {
288                         /*
289                          * Sampling just ourselves we can do with no locking.
290                          */
291                         error = cpu_clock_sample(which_clock,
292                                                  current, &rtn);
293                 } else {
294                         read_lock(&tasklist_lock);
295                         error = cpu_clock_sample_group(which_clock,
296                                                        current, &rtn);
297                         read_unlock(&tasklist_lock);
298                 }
299         } else {
300                 /*
301                  * Find the given PID, and validate that the caller
302                  * should be able to see it.
303                  */
304                 struct task_struct *p;
305                 read_lock(&tasklist_lock);
306                 p = find_task_by_pid(pid);
307                 if (p) {
308                         if (CPUCLOCK_PERTHREAD(which_clock)) {
309                                 if (p->tgid == current->tgid) {
310                                         error = cpu_clock_sample(which_clock,
311                                                                  p, &rtn);
312                                 }
313                         } else if (p->tgid == pid && p->signal) {
314                                 error = cpu_clock_sample_group(which_clock,
315                                                                p, &rtn);
316                         }
317                 }
318                 read_unlock(&tasklist_lock);
319         }
320
321         if (error)
322                 return error;
323         sample_to_timespec(which_clock, rtn, tp);
324         return 0;
325 }
326
327
328 /*
329  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
330  * This is called from sys_timer_create with the new timer already locked.
331  */
332 int posix_cpu_timer_create(struct k_itimer *new_timer)
333 {
334         int ret = 0;
335         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
336         struct task_struct *p;
337
338         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
339                 return -EINVAL;
340
341         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
342         new_timer->it.cpu.incr.sched = 0;
343         new_timer->it.cpu.expires.sched = 0;
344
345         read_lock(&tasklist_lock);
346         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
347                 if (pid == 0) {
348                         p = current;
349                 } else {
350                         p = find_task_by_pid(pid);
351                         if (p && p->tgid != current->tgid)
352                                 p = NULL;
353                 }
354         } else {
355                 if (pid == 0) {
356                         p = current->group_leader;
357                 } else {
358                         p = find_task_by_pid(pid);
359                         if (p && p->tgid != pid)
360                                 p = NULL;
361                 }
362         }
363         new_timer->it.cpu.task = p;
364         if (p) {
365                 get_task_struct(p);
366         } else {
367                 ret = -EINVAL;
368         }
369         read_unlock(&tasklist_lock);
370
371         return ret;
372 }
373
374 /*
375  * Clean up a CPU-clock timer that is about to be destroyed.
376  * This is called from timer deletion with the timer already locked.
377  * If we return TIMER_RETRY, it's necessary to release the timer's lock
378  * and try again.  (This happens when the timer is in the middle of firing.)
379  */
380 int posix_cpu_timer_del(struct k_itimer *timer)
381 {
382         struct task_struct *p = timer->it.cpu.task;
383         int ret = 0;
384
385         if (likely(p != NULL)) {
386                 read_lock(&tasklist_lock);
387                 if (unlikely(p->signal == NULL)) {
388                         /*
389                          * We raced with the reaping of the task.
390                          * The deletion should have cleared us off the list.
391                          */
392                         BUG_ON(!list_empty(&timer->it.cpu.entry));
393                 } else {
394                         spin_lock(&p->sighand->siglock);
395                         if (timer->it.cpu.firing)
396                                 ret = TIMER_RETRY;
397                         else
398                                 list_del(&timer->it.cpu.entry);
399                         spin_unlock(&p->sighand->siglock);
400                 }
401                 read_unlock(&tasklist_lock);
402
403                 if (!ret)
404                         put_task_struct(p);
405         }
406
407         return ret;
408 }
409
410 /*
411  * Clean out CPU timers still ticking when a thread exited.  The task
412  * pointer is cleared, and the expiry time is replaced with the residual
413  * time for later timer_gettime calls to return.
414  * This must be called with the siglock held.
415  */
416 static void cleanup_timers(struct list_head *head,
417                            cputime_t utime, cputime_t stime,
418                            unsigned long long sched_time)
419 {
420         struct cpu_timer_list *timer, *next;
421         cputime_t ptime = cputime_add(utime, stime);
422
423         list_for_each_entry_safe(timer, next, head, entry) {
424                 list_del_init(&timer->entry);
425                 if (cputime_lt(timer->expires.cpu, ptime)) {
426                         timer->expires.cpu = cputime_zero;
427                 } else {
428                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
429                                                          ptime);
430                 }
431         }
432
433         ++head;
434         list_for_each_entry_safe(timer, next, head, entry) {
435                 list_del_init(&timer->entry);
436                 if (cputime_lt(timer->expires.cpu, utime)) {
437                         timer->expires.cpu = cputime_zero;
438                 } else {
439                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
440                                                          utime);
441                 }
442         }
443
444         ++head;
445         list_for_each_entry_safe(timer, next, head, entry) {
446                 list_del_init(&timer->entry);
447                 if (timer->expires.sched < sched_time) {
448                         timer->expires.sched = 0;
449                 } else {
450                         timer->expires.sched -= sched_time;
451                 }
452         }
453 }
454
455 /*
456  * These are both called with the siglock held, when the current thread
457  * is being reaped.  When the final (leader) thread in the group is reaped,
458  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
459  */
460 void posix_cpu_timers_exit(struct task_struct *tsk)
461 {
462         cleanup_timers(tsk->cpu_timers,
463                        tsk->utime, tsk->stime, tsk->sched_time);
464
465 }
466 void posix_cpu_timers_exit_group(struct task_struct *tsk)
467 {
468         cleanup_timers(tsk->signal->cpu_timers,
469                        cputime_add(tsk->utime, tsk->signal->utime),
470                        cputime_add(tsk->stime, tsk->signal->stime),
471                        tsk->sched_time + tsk->signal->sched_time);
472 }
473
474
475 /*
476  * Set the expiry times of all the threads in the process so one of them
477  * will go off before the process cumulative expiry total is reached.
478  */
479 static void process_timer_rebalance(struct task_struct *p,
480                                     unsigned int clock_idx,
481                                     union cpu_time_count expires,
482                                     union cpu_time_count val)
483 {
484         cputime_t ticks, left;
485         unsigned long long ns, nsleft;
486         struct task_struct *t = p;
487         unsigned int nthreads = atomic_read(&p->signal->live);
488
489         if (!nthreads)
490                 return;
491
492         switch (clock_idx) {
493         default:
494                 BUG();
495                 break;
496         case CPUCLOCK_PROF:
497                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
498                                    nthreads);
499                 do {
500                         if (!unlikely(t->exit_state)) {
501                                 ticks = cputime_add(prof_ticks(t), left);
502                                 if (cputime_eq(t->it_prof_expires,
503                                                cputime_zero) ||
504                                     cputime_gt(t->it_prof_expires, ticks)) {
505                                         t->it_prof_expires = ticks;
506                                 }
507                         }
508                         t = next_thread(t);
509                 } while (t != p);
510                 break;
511         case CPUCLOCK_VIRT:
512                 left = cputime_div(cputime_sub(expires.cpu, val.cpu),
513                                    nthreads);
514                 do {
515                         if (!unlikely(t->exit_state)) {
516                                 ticks = cputime_add(virt_ticks(t), left);
517                                 if (cputime_eq(t->it_virt_expires,
518                                                cputime_zero) ||
519                                     cputime_gt(t->it_virt_expires, ticks)) {
520                                         t->it_virt_expires = ticks;
521                                 }
522                         }
523                         t = next_thread(t);
524                 } while (t != p);
525                 break;
526         case CPUCLOCK_SCHED:
527                 nsleft = expires.sched - val.sched;
528                 do_div(nsleft, nthreads);
529                 do {
530                         if (!unlikely(t->exit_state)) {
531                                 ns = t->sched_time + nsleft;
532                                 if (t->it_sched_expires == 0 ||
533                                     t->it_sched_expires > ns) {
534                                         t->it_sched_expires = ns;
535                                 }
536                         }
537                         t = next_thread(t);
538                 } while (t != p);
539                 break;
540         }
541 }
542
543 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
544 {
545         /*
546          * That's all for this thread or process.
547          * We leave our residual in expires to be reported.
548          */
549         put_task_struct(timer->it.cpu.task);
550         timer->it.cpu.task = NULL;
551         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
552                                              timer->it.cpu.expires,
553                                              now);
554 }
555
556 /*
557  * Insert the timer on the appropriate list before any timers that
558  * expire later.  This must be called with the tasklist_lock held
559  * for reading, and interrupts disabled.
560  */
561 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
562 {
563         struct task_struct *p = timer->it.cpu.task;
564         struct list_head *head, *listpos;
565         struct cpu_timer_list *const nt = &timer->it.cpu;
566         struct cpu_timer_list *next;
567         unsigned long i;
568
569         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
570                 p->cpu_timers : p->signal->cpu_timers);
571         head += CPUCLOCK_WHICH(timer->it_clock);
572
573         BUG_ON(!irqs_disabled());
574         spin_lock(&p->sighand->siglock);
575
576         listpos = head;
577         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
578                 list_for_each_entry(next, head, entry) {
579                         if (next->expires.sched > nt->expires.sched)
580                                 break;
581                         listpos = &next->entry;
582                 }
583         } else {
584                 list_for_each_entry(next, head, entry) {
585                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
586                                 break;
587                         listpos = &next->entry;
588                 }
589         }
590         list_add(&nt->entry, listpos);
591
592         if (listpos == head) {
593                 /*
594                  * We are the new earliest-expiring timer.
595                  * If we are a thread timer, there can always
596                  * be a process timer telling us to stop earlier.
597                  */
598
599                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
600                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
601                         default:
602                                 BUG();
603                         case CPUCLOCK_PROF:
604                                 if (cputime_eq(p->it_prof_expires,
605                                                cputime_zero) ||
606                                     cputime_gt(p->it_prof_expires,
607                                                nt->expires.cpu))
608                                         p->it_prof_expires = nt->expires.cpu;
609                                 break;
610                         case CPUCLOCK_VIRT:
611                                 if (cputime_eq(p->it_virt_expires,
612                                                cputime_zero) ||
613                                     cputime_gt(p->it_virt_expires,
614                                                nt->expires.cpu))
615                                         p->it_virt_expires = nt->expires.cpu;
616                                 break;
617                         case CPUCLOCK_SCHED:
618                                 if (p->it_sched_expires == 0 ||
619                                     p->it_sched_expires > nt->expires.sched)
620                                         p->it_sched_expires = nt->expires.sched;
621                                 break;
622                         }
623                 } else {
624                         /*
625                          * For a process timer, we must balance
626                          * all the live threads' expirations.
627                          */
628                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
629                         default:
630                                 BUG();
631                         case CPUCLOCK_VIRT:
632                                 if (!cputime_eq(p->signal->it_virt_expires,
633                                                 cputime_zero) &&
634                                     cputime_lt(p->signal->it_virt_expires,
635                                                timer->it.cpu.expires.cpu))
636                                         break;
637                                 goto rebalance;
638                         case CPUCLOCK_PROF:
639                                 if (!cputime_eq(p->signal->it_prof_expires,
640                                                 cputime_zero) &&
641                                     cputime_lt(p->signal->it_prof_expires,
642                                                timer->it.cpu.expires.cpu))
643                                         break;
644                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
645                                 if (i != RLIM_INFINITY &&
646                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
647                                         break;
648                                 goto rebalance;
649                         case CPUCLOCK_SCHED:
650                         rebalance:
651                                 process_timer_rebalance(
652                                         timer->it.cpu.task,
653                                         CPUCLOCK_WHICH(timer->it_clock),
654                                         timer->it.cpu.expires, now);
655                                 break;
656                         }
657                 }
658         }
659
660         spin_unlock(&p->sighand->siglock);
661 }
662
663 /*
664  * The timer is locked, fire it and arrange for its reload.
665  */
666 static void cpu_timer_fire(struct k_itimer *timer)
667 {
668         if (unlikely(timer->sigq == NULL)) {
669                 /*
670                  * This a special case for clock_nanosleep,
671                  * not a normal timer from sys_timer_create.
672                  */
673                 wake_up_process(timer->it_process);
674                 timer->it.cpu.expires.sched = 0;
675         } else if (timer->it.cpu.incr.sched == 0) {
676                 /*
677                  * One-shot timer.  Clear it as soon as it's fired.
678                  */
679                 posix_timer_event(timer, 0);
680                 timer->it.cpu.expires.sched = 0;
681         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
682                 /*
683                  * The signal did not get queued because the signal
684                  * was ignored, so we won't get any callback to
685                  * reload the timer.  But we need to keep it
686                  * ticking in case the signal is deliverable next time.
687                  */
688                 posix_cpu_timer_schedule(timer);
689         }
690 }
691
692 /*
693  * Guts of sys_timer_settime for CPU timers.
694  * This is called with the timer locked and interrupts disabled.
695  * If we return TIMER_RETRY, it's necessary to release the timer's lock
696  * and try again.  (This happens when the timer is in the middle of firing.)
697  */
698 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
699                         struct itimerspec *new, struct itimerspec *old)
700 {
701         struct task_struct *p = timer->it.cpu.task;
702         union cpu_time_count old_expires, new_expires, val;
703         int ret;
704
705         if (unlikely(p == NULL)) {
706                 /*
707                  * Timer refers to a dead task's clock.
708                  */
709                 return -ESRCH;
710         }
711
712         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
713
714         read_lock(&tasklist_lock);
715         /*
716          * We need the tasklist_lock to protect against reaping that
717          * clears p->signal.  If p has just been reaped, we can no
718          * longer get any information about it at all.
719          */
720         if (unlikely(p->signal == NULL)) {
721                 read_unlock(&tasklist_lock);
722                 put_task_struct(p);
723                 timer->it.cpu.task = NULL;
724                 return -ESRCH;
725         }
726
727         /*
728          * Disarm any old timer after extracting its expiry time.
729          */
730         BUG_ON(!irqs_disabled());
731
732         ret = 0;
733         spin_lock(&p->sighand->siglock);
734         old_expires = timer->it.cpu.expires;
735         if (unlikely(timer->it.cpu.firing)) {
736                 timer->it.cpu.firing = -1;
737                 ret = TIMER_RETRY;
738         } else
739                 list_del_init(&timer->it.cpu.entry);
740         spin_unlock(&p->sighand->siglock);
741
742         /*
743          * We need to sample the current value to convert the new
744          * value from to relative and absolute, and to convert the
745          * old value from absolute to relative.  To set a process
746          * timer, we need a sample to balance the thread expiry
747          * times (in arm_timer).  With an absolute time, we must
748          * check if it's already passed.  In short, we need a sample.
749          */
750         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
751                 cpu_clock_sample(timer->it_clock, p, &val);
752         } else {
753                 cpu_clock_sample_group(timer->it_clock, p, &val);
754         }
755
756         if (old) {
757                 if (old_expires.sched == 0) {
758                         old->it_value.tv_sec = 0;
759                         old->it_value.tv_nsec = 0;
760                 } else {
761                         /*
762                          * Update the timer in case it has
763                          * overrun already.  If it has,
764                          * we'll report it as having overrun
765                          * and with the next reloaded timer
766                          * already ticking, though we are
767                          * swallowing that pending
768                          * notification here to install the
769                          * new setting.
770                          */
771                         bump_cpu_timer(timer, val);
772                         if (cpu_time_before(timer->it_clock, val,
773                                             timer->it.cpu.expires)) {
774                                 old_expires = cpu_time_sub(
775                                         timer->it_clock,
776                                         timer->it.cpu.expires, val);
777                                 sample_to_timespec(timer->it_clock,
778                                                    old_expires,
779                                                    &old->it_value);
780                         } else {
781                                 old->it_value.tv_nsec = 1;
782                                 old->it_value.tv_sec = 0;
783                         }
784                 }
785         }
786
787         if (unlikely(ret)) {
788                 /*
789                  * We are colliding with the timer actually firing.
790                  * Punt after filling in the timer's old value, and
791                  * disable this firing since we are already reporting
792                  * it as an overrun (thanks to bump_cpu_timer above).
793                  */
794                 read_unlock(&tasklist_lock);
795                 goto out;
796         }
797
798         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
799                 cpu_time_add(timer->it_clock, &new_expires, val);
800         }
801
802         /*
803          * Install the new expiry time (or zero).
804          * For a timer with no notification action, we don't actually
805          * arm the timer (we'll just fake it for timer_gettime).
806          */
807         timer->it.cpu.expires = new_expires;
808         if (new_expires.sched != 0 &&
809             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
810             cpu_time_before(timer->it_clock, val, new_expires)) {
811                 arm_timer(timer, val);
812         }
813
814         read_unlock(&tasklist_lock);
815
816         /*
817          * Install the new reload setting, and
818          * set up the signal and overrun bookkeeping.
819          */
820         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
821                                                 &new->it_interval);
822
823         /*
824          * This acts as a modification timestamp for the timer,
825          * so any automatic reload attempt will punt on seeing
826          * that we have reset the timer manually.
827          */
828         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
829                 ~REQUEUE_PENDING;
830         timer->it_overrun_last = 0;
831         timer->it_overrun = -1;
832
833         if (new_expires.sched != 0 &&
834             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
835             !cpu_time_before(timer->it_clock, val, new_expires)) {
836                 /*
837                  * The designated time already passed, so we notify
838                  * immediately, even if the thread never runs to
839                  * accumulate more time on this clock.
840                  */
841                 cpu_timer_fire(timer);
842         }
843
844         ret = 0;
845  out:
846         if (old) {
847                 sample_to_timespec(timer->it_clock,
848                                    timer->it.cpu.incr, &old->it_interval);
849         }
850         return ret;
851 }
852
853 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
854 {
855         union cpu_time_count now;
856         struct task_struct *p = timer->it.cpu.task;
857         int clear_dead;
858
859         /*
860          * Easy part: convert the reload time.
861          */
862         sample_to_timespec(timer->it_clock,
863                            timer->it.cpu.incr, &itp->it_interval);
864
865         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
866                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
867                 return;
868         }
869
870         if (unlikely(p == NULL)) {
871                 /*
872                  * This task already died and the timer will never fire.
873                  * In this case, expires is actually the dead value.
874                  */
875         dead:
876                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
877                                    &itp->it_value);
878                 return;
879         }
880
881         /*
882          * Sample the clock to take the difference with the expiry time.
883          */
884         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
885                 cpu_clock_sample(timer->it_clock, p, &now);
886                 clear_dead = p->exit_state;
887         } else {
888                 read_lock(&tasklist_lock);
889                 if (unlikely(p->signal == NULL)) {
890                         /*
891                          * The process has been reaped.
892                          * We can't even collect a sample any more.
893                          * Call the timer disarmed, nothing else to do.
894                          */
895                         put_task_struct(p);
896                         timer->it.cpu.task = NULL;
897                         timer->it.cpu.expires.sched = 0;
898                         read_unlock(&tasklist_lock);
899                         goto dead;
900                 } else {
901                         cpu_clock_sample_group(timer->it_clock, p, &now);
902                         clear_dead = (unlikely(p->exit_state) &&
903                                       thread_group_empty(p));
904                 }
905                 read_unlock(&tasklist_lock);
906         }
907
908         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
909                 if (timer->it.cpu.incr.sched == 0 &&
910                     cpu_time_before(timer->it_clock,
911                                     timer->it.cpu.expires, now)) {
912                         /*
913                          * Do-nothing timer expired and has no reload,
914                          * so it's as if it was never set.
915                          */
916                         timer->it.cpu.expires.sched = 0;
917                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
918                         return;
919                 }
920                 /*
921                  * Account for any expirations and reloads that should
922                  * have happened.
923                  */
924                 bump_cpu_timer(timer, now);
925         }
926
927         if (unlikely(clear_dead)) {
928                 /*
929                  * We've noticed that the thread is dead, but
930                  * not yet reaped.  Take this opportunity to
931                  * drop our task ref.
932                  */
933                 clear_dead_task(timer, now);
934                 goto dead;
935         }
936
937         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
938                 sample_to_timespec(timer->it_clock,
939                                    cpu_time_sub(timer->it_clock,
940                                                 timer->it.cpu.expires, now),
941                                    &itp->it_value);
942         } else {
943                 /*
944                  * The timer should have expired already, but the firing
945                  * hasn't taken place yet.  Say it's just about to expire.
946                  */
947                 itp->it_value.tv_nsec = 1;
948                 itp->it_value.tv_sec = 0;
949         }
950 }
951
952 /*
953  * Check for any per-thread CPU timers that have fired and move them off
954  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
955  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
956  */
957 static void check_thread_timers(struct task_struct *tsk,
958                                 struct list_head *firing)
959 {
960         int maxfire;
961         struct list_head *timers = tsk->cpu_timers;
962
963         maxfire = 20;
964         tsk->it_prof_expires = cputime_zero;
965         while (!list_empty(timers)) {
966                 struct cpu_timer_list *t = list_entry(timers->next,
967                                                       struct cpu_timer_list,
968                                                       entry);
969                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
970                         tsk->it_prof_expires = t->expires.cpu;
971                         break;
972                 }
973                 t->firing = 1;
974                 list_move_tail(&t->entry, firing);
975         }
976
977         ++timers;
978         maxfire = 20;
979         tsk->it_virt_expires = cputime_zero;
980         while (!list_empty(timers)) {
981                 struct cpu_timer_list *t = list_entry(timers->next,
982                                                       struct cpu_timer_list,
983                                                       entry);
984                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
985                         tsk->it_virt_expires = t->expires.cpu;
986                         break;
987                 }
988                 t->firing = 1;
989                 list_move_tail(&t->entry, firing);
990         }
991
992         ++timers;
993         maxfire = 20;
994         tsk->it_sched_expires = 0;
995         while (!list_empty(timers)) {
996                 struct cpu_timer_list *t = list_entry(timers->next,
997                                                       struct cpu_timer_list,
998                                                       entry);
999                 if (!--maxfire || tsk->sched_time < t->expires.sched) {
1000                         tsk->it_sched_expires = t->expires.sched;
1001                         break;
1002                 }
1003                 t->firing = 1;
1004                 list_move_tail(&t->entry, firing);
1005         }
1006 }
1007
1008 /*
1009  * Check for any per-thread CPU timers that have fired and move them
1010  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1011  * have already been taken off.
1012  */
1013 static void check_process_timers(struct task_struct *tsk,
1014                                  struct list_head *firing)
1015 {
1016         int maxfire;
1017         struct signal_struct *const sig = tsk->signal;
1018         cputime_t utime, stime, ptime, virt_expires, prof_expires;
1019         unsigned long long sched_time, sched_expires;
1020         struct task_struct *t;
1021         struct list_head *timers = sig->cpu_timers;
1022
1023         /*
1024          * Don't sample the current process CPU clocks if there are no timers.
1025          */
1026         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1027             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1028             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1029             list_empty(&timers[CPUCLOCK_VIRT]) &&
1030             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1031             list_empty(&timers[CPUCLOCK_SCHED]))
1032                 return;
1033
1034         /*
1035          * Collect the current process totals.
1036          */
1037         utime = sig->utime;
1038         stime = sig->stime;
1039         sched_time = sig->sched_time;
1040         t = tsk;
1041         do {
1042                 utime = cputime_add(utime, t->utime);
1043                 stime = cputime_add(stime, t->stime);
1044                 sched_time += t->sched_time;
1045                 t = next_thread(t);
1046         } while (t != tsk);
1047         ptime = cputime_add(utime, stime);
1048
1049         maxfire = 20;
1050         prof_expires = cputime_zero;
1051         while (!list_empty(timers)) {
1052                 struct cpu_timer_list *t = list_entry(timers->next,
1053                                                       struct cpu_timer_list,
1054                                                       entry);
1055                 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
1056                         prof_expires = t->expires.cpu;
1057                         break;
1058                 }
1059                 t->firing = 1;
1060                 list_move_tail(&t->entry, firing);
1061         }
1062
1063         ++timers;
1064         maxfire = 20;
1065         virt_expires = cputime_zero;
1066         while (!list_empty(timers)) {
1067                 struct cpu_timer_list *t = list_entry(timers->next,
1068                                                       struct cpu_timer_list,
1069                                                       entry);
1070                 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
1071                         virt_expires = t->expires.cpu;
1072                         break;
1073                 }
1074                 t->firing = 1;
1075                 list_move_tail(&t->entry, firing);
1076         }
1077
1078         ++timers;
1079         maxfire = 20;
1080         sched_expires = 0;
1081         while (!list_empty(timers)) {
1082                 struct cpu_timer_list *t = list_entry(timers->next,
1083                                                       struct cpu_timer_list,
1084                                                       entry);
1085                 if (!--maxfire || sched_time < t->expires.sched) {
1086                         sched_expires = t->expires.sched;
1087                         break;
1088                 }
1089                 t->firing = 1;
1090                 list_move_tail(&t->entry, firing);
1091         }
1092
1093         /*
1094          * Check for the special case process timers.
1095          */
1096         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1097                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1098                         /* ITIMER_PROF fires and reloads.  */
1099                         sig->it_prof_expires = sig->it_prof_incr;
1100                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1101                                 sig->it_prof_expires = cputime_add(
1102                                         sig->it_prof_expires, ptime);
1103                         }
1104                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1105                 }
1106                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1107                     (cputime_eq(prof_expires, cputime_zero) ||
1108                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1109                         prof_expires = sig->it_prof_expires;
1110                 }
1111         }
1112         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1113                 if (cputime_ge(utime, sig->it_virt_expires)) {
1114                         /* ITIMER_VIRTUAL fires and reloads.  */
1115                         sig->it_virt_expires = sig->it_virt_incr;
1116                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1117                                 sig->it_virt_expires = cputime_add(
1118                                         sig->it_virt_expires, utime);
1119                         }
1120                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1121                 }
1122                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1123                     (cputime_eq(virt_expires, cputime_zero) ||
1124                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1125                         virt_expires = sig->it_virt_expires;
1126                 }
1127         }
1128         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1129                 unsigned long psecs = cputime_to_secs(ptime);
1130                 cputime_t x;
1131                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1132                         /*
1133                          * At the hard limit, we just die.
1134                          * No need to calculate anything else now.
1135                          */
1136                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1137                         return;
1138                 }
1139                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1140                         /*
1141                          * At the soft limit, send a SIGXCPU every second.
1142                          */
1143                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1144                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1145                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1146                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1147                         }
1148                 }
1149                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1150                 if (cputime_eq(prof_expires, cputime_zero) ||
1151                     cputime_lt(x, prof_expires)) {
1152                         prof_expires = x;
1153                 }
1154         }
1155
1156         if (!cputime_eq(prof_expires, cputime_zero) ||
1157             !cputime_eq(virt_expires, cputime_zero) ||
1158             sched_expires != 0) {
1159                 /*
1160                  * Rebalance the threads' expiry times for the remaining
1161                  * process CPU timers.
1162                  */
1163
1164                 cputime_t prof_left, virt_left, ticks;
1165                 unsigned long long sched_left, sched;
1166                 const unsigned int nthreads = atomic_read(&sig->live);
1167
1168                 if (!nthreads)
1169                         return;
1170
1171                 prof_left = cputime_sub(prof_expires, utime);
1172                 prof_left = cputime_sub(prof_left, stime);
1173                 prof_left = cputime_div(prof_left, nthreads);
1174                 virt_left = cputime_sub(virt_expires, utime);
1175                 virt_left = cputime_div(virt_left, nthreads);
1176                 if (sched_expires) {
1177                         sched_left = sched_expires - sched_time;
1178                         do_div(sched_left, nthreads);
1179                 } else {
1180                         sched_left = 0;
1181                 }
1182                 t = tsk;
1183                 do {
1184                         ticks = cputime_add(cputime_add(t->utime, t->stime),
1185                                             prof_left);
1186                         if (!cputime_eq(prof_expires, cputime_zero) &&
1187                             (cputime_eq(t->it_prof_expires, cputime_zero) ||
1188                              cputime_gt(t->it_prof_expires, ticks))) {
1189                                 t->it_prof_expires = ticks;
1190                         }
1191
1192                         ticks = cputime_add(t->utime, virt_left);
1193                         if (!cputime_eq(virt_expires, cputime_zero) &&
1194                             (cputime_eq(t->it_virt_expires, cputime_zero) ||
1195                              cputime_gt(t->it_virt_expires, ticks))) {
1196                                 t->it_virt_expires = ticks;
1197                         }
1198
1199                         sched = t->sched_time + sched_left;
1200                         if (sched_expires && (t->it_sched_expires == 0 ||
1201                                               t->it_sched_expires > sched)) {
1202                                 t->it_sched_expires = sched;
1203                         }
1204
1205                         do {
1206                                 t = next_thread(t);
1207                         } while (unlikely(t->exit_state));
1208                 } while (t != tsk);
1209         }
1210 }
1211
1212 /*
1213  * This is called from the signal code (via do_schedule_next_timer)
1214  * when the last timer signal was delivered and we have to reload the timer.
1215  */
1216 void posix_cpu_timer_schedule(struct k_itimer *timer)
1217 {
1218         struct task_struct *p = timer->it.cpu.task;
1219         union cpu_time_count now;
1220
1221         if (unlikely(p == NULL))
1222                 /*
1223                  * The task was cleaned up already, no future firings.
1224                  */
1225                 return;
1226
1227         /*
1228          * Fetch the current sample and update the timer's expiry time.
1229          */
1230         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1231                 cpu_clock_sample(timer->it_clock, p, &now);
1232                 bump_cpu_timer(timer, now);
1233                 if (unlikely(p->exit_state)) {
1234                         clear_dead_task(timer, now);
1235                         return;
1236                 }
1237                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1238         } else {
1239                 read_lock(&tasklist_lock);
1240                 if (unlikely(p->signal == NULL)) {
1241                         /*
1242                          * The process has been reaped.
1243                          * We can't even collect a sample any more.
1244                          */
1245                         put_task_struct(p);
1246                         timer->it.cpu.task = p = NULL;
1247                         timer->it.cpu.expires.sched = 0;
1248                         read_unlock(&tasklist_lock);
1249                         return;
1250                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1251                         /*
1252                          * We've noticed that the thread is dead, but
1253                          * not yet reaped.  Take this opportunity to
1254                          * drop our task ref.
1255                          */
1256                         clear_dead_task(timer, now);
1257                         read_unlock(&tasklist_lock);
1258                         return;
1259                 }
1260                 cpu_clock_sample_group(timer->it_clock, p, &now);
1261                 bump_cpu_timer(timer, now);
1262                 /* Leave the tasklist_lock locked for the call below.  */
1263         }
1264
1265         /*
1266          * Now re-arm for the new expiry time.
1267          */
1268         arm_timer(timer, now);
1269
1270         read_unlock(&tasklist_lock);
1271 }
1272
1273 /*
1274  * This is called from the timer interrupt handler.  The irq handler has
1275  * already updated our counts.  We need to check if any timers fire now.
1276  * Interrupts are disabled.
1277  */
1278 void run_posix_cpu_timers(struct task_struct *tsk)
1279 {
1280         LIST_HEAD(firing);
1281         struct k_itimer *timer, *next;
1282
1283         BUG_ON(!irqs_disabled());
1284
1285 #define UNEXPIRED(clock) \
1286                 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1287                  cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1288
1289         if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1290             (tsk->it_sched_expires == 0 ||
1291              tsk->sched_time < tsk->it_sched_expires))
1292                 return;
1293
1294 #undef  UNEXPIRED
1295
1296         BUG_ON(tsk->exit_state);
1297
1298         /*
1299          * Double-check with locks held.
1300          */
1301         read_lock(&tasklist_lock);
1302         spin_lock(&tsk->sighand->siglock);
1303
1304         /*
1305          * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1306          * all the timers that are firing, and put them on the firing list.
1307          */
1308         check_thread_timers(tsk, &firing);
1309         check_process_timers(tsk, &firing);
1310
1311         /*
1312          * We must release these locks before taking any timer's lock.
1313          * There is a potential race with timer deletion here, as the
1314          * siglock now protects our private firing list.  We have set
1315          * the firing flag in each timer, so that a deletion attempt
1316          * that gets the timer lock before we do will give it up and
1317          * spin until we've taken care of that timer below.
1318          */
1319         spin_unlock(&tsk->sighand->siglock);
1320         read_unlock(&tasklist_lock);
1321
1322         /*
1323          * Now that all the timers on our list have the firing flag,
1324          * noone will touch their list entries but us.  We'll take
1325          * each timer's lock before clearing its firing flag, so no
1326          * timer call will interfere.
1327          */
1328         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1329                 int firing;
1330                 spin_lock(&timer->it_lock);
1331                 list_del_init(&timer->it.cpu.entry);
1332                 firing = timer->it.cpu.firing;
1333                 timer->it.cpu.firing = 0;
1334                 /*
1335                  * The firing flag is -1 if we collided with a reset
1336                  * of the timer, which already reported this
1337                  * almost-firing as an overrun.  So don't generate an event.
1338                  */
1339                 if (likely(firing >= 0)) {
1340                         cpu_timer_fire(timer);
1341                 }
1342                 spin_unlock(&timer->it_lock);
1343         }
1344 }
1345
1346 /*
1347  * Set one of the process-wide special case CPU timers.
1348  * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1349  * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1350  * absolute; non-null for ITIMER_*, where *newval is relative and we update
1351  * it to be absolute, *oldval is absolute and we update it to be relative.
1352  */
1353 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1354                            cputime_t *newval, cputime_t *oldval)
1355 {
1356         union cpu_time_count now;
1357         struct list_head *head;
1358
1359         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1360         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1361
1362         if (oldval) {
1363                 if (!cputime_eq(*oldval, cputime_zero)) {
1364                         if (cputime_le(*oldval, now.cpu)) {
1365                                 /* Just about to fire. */
1366                                 *oldval = jiffies_to_cputime(1);
1367                         } else {
1368                                 *oldval = cputime_sub(*oldval, now.cpu);
1369                         }
1370                 }
1371
1372                 if (cputime_eq(*newval, cputime_zero))
1373                         return;
1374                 *newval = cputime_add(*newval, now.cpu);
1375
1376                 /*
1377                  * If the RLIMIT_CPU timer will expire before the
1378                  * ITIMER_PROF timer, we have nothing else to do.
1379                  */
1380                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1381                     < cputime_to_secs(*newval))
1382                         return;
1383         }
1384
1385         /*
1386          * Check whether there are any process timers already set to fire
1387          * before this one.  If so, we don't have anything more to do.
1388          */
1389         head = &tsk->signal->cpu_timers[clock_idx];
1390         if (list_empty(head) ||
1391             cputime_ge(list_entry(head->next,
1392                                   struct cpu_timer_list, entry)->expires.cpu,
1393                        *newval)) {
1394                 /*
1395                  * Rejigger each thread's expiry time so that one will
1396                  * notice before we hit the process-cumulative expiry time.
1397                  */
1398                 union cpu_time_count expires = { .sched = 0 };
1399                 expires.cpu = *newval;
1400                 process_timer_rebalance(tsk, clock_idx, expires, now);
1401         }
1402 }
1403
1404 static long posix_cpu_clock_nanosleep_restart(struct restart_block *);
1405
1406 int posix_cpu_nsleep(clockid_t which_clock, int flags,
1407                      struct timespec *rqtp)
1408 {
1409         struct restart_block *restart_block =
1410             &current_thread_info()->restart_block;
1411         struct k_itimer timer;
1412         int error;
1413
1414         /*
1415          * Diagnose required errors first.
1416          */
1417         if (CPUCLOCK_PERTHREAD(which_clock) &&
1418             (CPUCLOCK_PID(which_clock) == 0 ||
1419              CPUCLOCK_PID(which_clock) == current->pid))
1420                 return -EINVAL;
1421
1422         /*
1423          * Set up a temporary timer and then wait for it to go off.
1424          */
1425         memset(&timer, 0, sizeof timer);
1426         spin_lock_init(&timer.it_lock);
1427         timer.it_clock = which_clock;
1428         timer.it_overrun = -1;
1429         error = posix_cpu_timer_create(&timer);
1430         timer.it_process = current;
1431         if (!error) {
1432                 struct timespec __user *rmtp;
1433                 static struct itimerspec zero_it;
1434                 struct itimerspec it = { .it_value = *rqtp,
1435                                          .it_interval = {} };
1436
1437                 spin_lock_irq(&timer.it_lock);
1438                 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1439                 if (error) {
1440                         spin_unlock_irq(&timer.it_lock);
1441                         return error;
1442                 }
1443
1444                 while (!signal_pending(current)) {
1445                         if (timer.it.cpu.expires.sched == 0) {
1446                                 /*
1447                                  * Our timer fired and was reset.
1448                                  */
1449                                 spin_unlock_irq(&timer.it_lock);
1450                                 return 0;
1451                         }
1452
1453                         /*
1454                          * Block until cpu_timer_fire (or a signal) wakes us.
1455                          */
1456                         __set_current_state(TASK_INTERRUPTIBLE);
1457                         spin_unlock_irq(&timer.it_lock);
1458                         schedule();
1459                         spin_lock_irq(&timer.it_lock);
1460                 }
1461
1462                 /*
1463                  * We were interrupted by a signal.
1464                  */
1465                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1466                 posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1467                 spin_unlock_irq(&timer.it_lock);
1468
1469                 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1470                         /*
1471                          * It actually did fire already.
1472                          */
1473                         return 0;
1474                 }
1475
1476                 /*
1477                  * Report back to the user the time still remaining.
1478                  */
1479                 rmtp = (struct timespec __user *) restart_block->arg1;
1480                 if (rmtp != NULL && !(flags & TIMER_ABSTIME) &&
1481                     copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1482                         return -EFAULT;
1483
1484                 restart_block->fn = posix_cpu_clock_nanosleep_restart;
1485                 /* Caller already set restart_block->arg1 */
1486                 restart_block->arg0 = which_clock;
1487                 restart_block->arg2 = rqtp->tv_sec;
1488                 restart_block->arg3 = rqtp->tv_nsec;
1489
1490                 error = -ERESTART_RESTARTBLOCK;
1491         }
1492
1493         return error;
1494 }
1495
1496 static long
1497 posix_cpu_clock_nanosleep_restart(struct restart_block *restart_block)
1498 {
1499         clockid_t which_clock = restart_block->arg0;
1500         struct timespec t = { .tv_sec = restart_block->arg2,
1501                               .tv_nsec = restart_block->arg3 };
1502         restart_block->fn = do_no_restart_syscall;
1503         return posix_cpu_nsleep(which_clock, TIMER_ABSTIME, &t);
1504 }
1505
1506
1507 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1508 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1509
1510 static int process_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1511 {
1512         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1513 }
1514 static int process_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1515 {
1516         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1517 }
1518 static int process_cpu_timer_create(struct k_itimer *timer)
1519 {
1520         timer->it_clock = PROCESS_CLOCK;
1521         return posix_cpu_timer_create(timer);
1522 }
1523 static int process_cpu_nsleep(clockid_t which_clock, int flags,
1524                               struct timespec *rqtp)
1525 {
1526         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1527 }
1528 static int thread_cpu_clock_getres(clockid_t which_clock, struct timespec *tp)
1529 {
1530         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1531 }
1532 static int thread_cpu_clock_get(clockid_t which_clock, struct timespec *tp)
1533 {
1534         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1535 }
1536 static int thread_cpu_timer_create(struct k_itimer *timer)
1537 {
1538         timer->it_clock = THREAD_CLOCK;
1539         return posix_cpu_timer_create(timer);
1540 }
1541 static int thread_cpu_nsleep(clockid_t which_clock, int flags,
1542                               struct timespec *rqtp)
1543 {
1544         return -EINVAL;
1545 }
1546
1547 static __init int init_posix_cpu_timers(void)
1548 {
1549         struct k_clock process = {
1550                 .clock_getres = process_cpu_clock_getres,
1551                 .clock_get = process_cpu_clock_get,
1552                 .clock_set = do_posix_clock_nosettime,
1553                 .timer_create = process_cpu_timer_create,
1554                 .nsleep = process_cpu_nsleep,
1555         };
1556         struct k_clock thread = {
1557                 .clock_getres = thread_cpu_clock_getres,
1558                 .clock_get = thread_cpu_clock_get,
1559                 .clock_set = do_posix_clock_nosettime,
1560                 .timer_create = thread_cpu_timer_create,
1561                 .nsleep = thread_cpu_nsleep,
1562         };
1563
1564         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1565         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1566
1567         return 0;
1568 }
1569 __initcall(init_posix_cpu_timers);