itimers: Fix periodic tics precision
[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 <linux/errno.h>
8 #include <linux/math64.h>
9 #include <asm/uaccess.h>
10 #include <linux/kernel_stat.h>
11
12 /*
13  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
14  */
15 void update_rlimit_cpu(unsigned long rlim_new)
16 {
17         cputime_t cputime = secs_to_cputime(rlim_new);
18         struct signal_struct *const sig = current->signal;
19
20         if (cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) ||
21             cputime_gt(sig->it[CPUCLOCK_PROF].expires, cputime)) {
22                 spin_lock_irq(&current->sighand->siglock);
23                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
24                 spin_unlock_irq(&current->sighand->siglock);
25         }
26 }
27
28 static int check_clock(const clockid_t which_clock)
29 {
30         int error = 0;
31         struct task_struct *p;
32         const pid_t pid = CPUCLOCK_PID(which_clock);
33
34         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
35                 return -EINVAL;
36
37         if (pid == 0)
38                 return 0;
39
40         read_lock(&tasklist_lock);
41         p = find_task_by_vpid(pid);
42         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
43                    same_thread_group(p, current) : thread_group_leader(p))) {
44                 error = -EINVAL;
45         }
46         read_unlock(&tasklist_lock);
47
48         return error;
49 }
50
51 static inline union cpu_time_count
52 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
53 {
54         union cpu_time_count ret;
55         ret.sched = 0;          /* high half always zero when .cpu used */
56         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
57                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
58         } else {
59                 ret.cpu = timespec_to_cputime(tp);
60         }
61         return ret;
62 }
63
64 static void sample_to_timespec(const clockid_t which_clock,
65                                union cpu_time_count cpu,
66                                struct timespec *tp)
67 {
68         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
69                 *tp = ns_to_timespec(cpu.sched);
70         else
71                 cputime_to_timespec(cpu.cpu, tp);
72 }
73
74 static inline int cpu_time_before(const clockid_t which_clock,
75                                   union cpu_time_count now,
76                                   union cpu_time_count then)
77 {
78         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
79                 return now.sched < then.sched;
80         }  else {
81                 return cputime_lt(now.cpu, then.cpu);
82         }
83 }
84 static inline void cpu_time_add(const clockid_t which_clock,
85                                 union cpu_time_count *acc,
86                                 union cpu_time_count val)
87 {
88         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
89                 acc->sched += val.sched;
90         }  else {
91                 acc->cpu = cputime_add(acc->cpu, val.cpu);
92         }
93 }
94 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
95                                                 union cpu_time_count a,
96                                                 union cpu_time_count b)
97 {
98         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
99                 a.sched -= b.sched;
100         }  else {
101                 a.cpu = cputime_sub(a.cpu, b.cpu);
102         }
103         return a;
104 }
105
106 /*
107  * Divide and limit the result to res >= 1
108  *
109  * This is necessary to prevent signal delivery starvation, when the result of
110  * the division would be rounded down to 0.
111  */
112 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
113 {
114         cputime_t res = cputime_div(time, div);
115
116         return max_t(cputime_t, res, 1);
117 }
118
119 /*
120  * Update expiry time from increment, and increase overrun count,
121  * given the current clock sample.
122  */
123 static void bump_cpu_timer(struct k_itimer *timer,
124                                   union cpu_time_count now)
125 {
126         int i;
127
128         if (timer->it.cpu.incr.sched == 0)
129                 return;
130
131         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
132                 unsigned long long delta, incr;
133
134                 if (now.sched < timer->it.cpu.expires.sched)
135                         return;
136                 incr = timer->it.cpu.incr.sched;
137                 delta = now.sched + incr - timer->it.cpu.expires.sched;
138                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
139                 for (i = 0; incr < delta - incr; i++)
140                         incr = incr << 1;
141                 for (; i >= 0; incr >>= 1, i--) {
142                         if (delta < incr)
143                                 continue;
144                         timer->it.cpu.expires.sched += incr;
145                         timer->it_overrun += 1 << i;
146                         delta -= incr;
147                 }
148         } else {
149                 cputime_t delta, incr;
150
151                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
152                         return;
153                 incr = timer->it.cpu.incr.cpu;
154                 delta = cputime_sub(cputime_add(now.cpu, incr),
155                                     timer->it.cpu.expires.cpu);
156                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
157                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
158                              incr = cputime_add(incr, incr);
159                 for (; i >= 0; incr = cputime_halve(incr), i--) {
160                         if (cputime_lt(delta, incr))
161                                 continue;
162                         timer->it.cpu.expires.cpu =
163                                 cputime_add(timer->it.cpu.expires.cpu, incr);
164                         timer->it_overrun += 1 << i;
165                         delta = cputime_sub(delta, incr);
166                 }
167         }
168 }
169
170 static inline cputime_t prof_ticks(struct task_struct *p)
171 {
172         return cputime_add(p->utime, p->stime);
173 }
174 static inline cputime_t virt_ticks(struct task_struct *p)
175 {
176         return p->utime;
177 }
178
179 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
180 {
181         int error = check_clock(which_clock);
182         if (!error) {
183                 tp->tv_sec = 0;
184                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
185                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
186                         /*
187                          * If sched_clock is using a cycle counter, we
188                          * don't have any idea of its true resolution
189                          * exported, but it is much more than 1s/HZ.
190                          */
191                         tp->tv_nsec = 1;
192                 }
193         }
194         return error;
195 }
196
197 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
198 {
199         /*
200          * You can never reset a CPU clock, but we check for other errors
201          * in the call before failing with EPERM.
202          */
203         int error = check_clock(which_clock);
204         if (error == 0) {
205                 error = -EPERM;
206         }
207         return error;
208 }
209
210
211 /*
212  * Sample a per-thread clock for the given task.
213  */
214 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
215                             union cpu_time_count *cpu)
216 {
217         switch (CPUCLOCK_WHICH(which_clock)) {
218         default:
219                 return -EINVAL;
220         case CPUCLOCK_PROF:
221                 cpu->cpu = prof_ticks(p);
222                 break;
223         case CPUCLOCK_VIRT:
224                 cpu->cpu = virt_ticks(p);
225                 break;
226         case CPUCLOCK_SCHED:
227                 cpu->sched = task_sched_runtime(p);
228                 break;
229         }
230         return 0;
231 }
232
233 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
234 {
235         struct sighand_struct *sighand;
236         struct signal_struct *sig;
237         struct task_struct *t;
238
239         *times = INIT_CPUTIME;
240
241         rcu_read_lock();
242         sighand = rcu_dereference(tsk->sighand);
243         if (!sighand)
244                 goto out;
245
246         sig = tsk->signal;
247
248         t = tsk;
249         do {
250                 times->utime = cputime_add(times->utime, t->utime);
251                 times->stime = cputime_add(times->stime, t->stime);
252                 times->sum_exec_runtime += t->se.sum_exec_runtime;
253
254                 t = next_thread(t);
255         } while (t != tsk);
256
257         times->utime = cputime_add(times->utime, sig->utime);
258         times->stime = cputime_add(times->stime, sig->stime);
259         times->sum_exec_runtime += sig->sum_sched_runtime;
260 out:
261         rcu_read_unlock();
262 }
263
264 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
265 {
266         if (cputime_gt(b->utime, a->utime))
267                 a->utime = b->utime;
268
269         if (cputime_gt(b->stime, a->stime))
270                 a->stime = b->stime;
271
272         if (b->sum_exec_runtime > a->sum_exec_runtime)
273                 a->sum_exec_runtime = b->sum_exec_runtime;
274 }
275
276 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
277 {
278         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
279         struct task_cputime sum;
280         unsigned long flags;
281
282         spin_lock_irqsave(&cputimer->lock, flags);
283         if (!cputimer->running) {
284                 cputimer->running = 1;
285                 /*
286                  * The POSIX timer interface allows for absolute time expiry
287                  * values through the TIMER_ABSTIME flag, therefore we have
288                  * to synchronize the timer to the clock every time we start
289                  * it.
290                  */
291                 thread_group_cputime(tsk, &sum);
292                 update_gt_cputime(&cputimer->cputime, &sum);
293         }
294         *times = cputimer->cputime;
295         spin_unlock_irqrestore(&cputimer->lock, flags);
296 }
297
298 /*
299  * Sample a process (thread group) clock for the given group_leader task.
300  * Must be called with tasklist_lock held for reading.
301  */
302 static int cpu_clock_sample_group(const clockid_t which_clock,
303                                   struct task_struct *p,
304                                   union cpu_time_count *cpu)
305 {
306         struct task_cputime cputime;
307
308         switch (CPUCLOCK_WHICH(which_clock)) {
309         default:
310                 return -EINVAL;
311         case CPUCLOCK_PROF:
312                 thread_group_cputime(p, &cputime);
313                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
314                 break;
315         case CPUCLOCK_VIRT:
316                 thread_group_cputime(p, &cputime);
317                 cpu->cpu = cputime.utime;
318                 break;
319         case CPUCLOCK_SCHED:
320                 cpu->sched = thread_group_sched_runtime(p);
321                 break;
322         }
323         return 0;
324 }
325
326
327 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
328 {
329         const pid_t pid = CPUCLOCK_PID(which_clock);
330         int error = -EINVAL;
331         union cpu_time_count rtn;
332
333         if (pid == 0) {
334                 /*
335                  * Special case constant value for our own clocks.
336                  * We don't have to do any lookup to find ourselves.
337                  */
338                 if (CPUCLOCK_PERTHREAD(which_clock)) {
339                         /*
340                          * Sampling just ourselves we can do with no locking.
341                          */
342                         error = cpu_clock_sample(which_clock,
343                                                  current, &rtn);
344                 } else {
345                         read_lock(&tasklist_lock);
346                         error = cpu_clock_sample_group(which_clock,
347                                                        current, &rtn);
348                         read_unlock(&tasklist_lock);
349                 }
350         } else {
351                 /*
352                  * Find the given PID, and validate that the caller
353                  * should be able to see it.
354                  */
355                 struct task_struct *p;
356                 rcu_read_lock();
357                 p = find_task_by_vpid(pid);
358                 if (p) {
359                         if (CPUCLOCK_PERTHREAD(which_clock)) {
360                                 if (same_thread_group(p, current)) {
361                                         error = cpu_clock_sample(which_clock,
362                                                                  p, &rtn);
363                                 }
364                         } else {
365                                 read_lock(&tasklist_lock);
366                                 if (thread_group_leader(p) && p->signal) {
367                                         error =
368                                             cpu_clock_sample_group(which_clock,
369                                                                    p, &rtn);
370                                 }
371                                 read_unlock(&tasklist_lock);
372                         }
373                 }
374                 rcu_read_unlock();
375         }
376
377         if (error)
378                 return error;
379         sample_to_timespec(which_clock, rtn, tp);
380         return 0;
381 }
382
383
384 /*
385  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
386  * This is called from sys_timer_create with the new timer already locked.
387  */
388 int posix_cpu_timer_create(struct k_itimer *new_timer)
389 {
390         int ret = 0;
391         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
392         struct task_struct *p;
393
394         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
395                 return -EINVAL;
396
397         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
398         new_timer->it.cpu.incr.sched = 0;
399         new_timer->it.cpu.expires.sched = 0;
400
401         read_lock(&tasklist_lock);
402         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
403                 if (pid == 0) {
404                         p = current;
405                 } else {
406                         p = find_task_by_vpid(pid);
407                         if (p && !same_thread_group(p, current))
408                                 p = NULL;
409                 }
410         } else {
411                 if (pid == 0) {
412                         p = current->group_leader;
413                 } else {
414                         p = find_task_by_vpid(pid);
415                         if (p && !thread_group_leader(p))
416                                 p = NULL;
417                 }
418         }
419         new_timer->it.cpu.task = p;
420         if (p) {
421                 get_task_struct(p);
422         } else {
423                 ret = -EINVAL;
424         }
425         read_unlock(&tasklist_lock);
426
427         return ret;
428 }
429
430 /*
431  * Clean up a CPU-clock timer that is about to be destroyed.
432  * This is called from timer deletion with the timer already locked.
433  * If we return TIMER_RETRY, it's necessary to release the timer's lock
434  * and try again.  (This happens when the timer is in the middle of firing.)
435  */
436 int posix_cpu_timer_del(struct k_itimer *timer)
437 {
438         struct task_struct *p = timer->it.cpu.task;
439         int ret = 0;
440
441         if (likely(p != NULL)) {
442                 read_lock(&tasklist_lock);
443                 if (unlikely(p->signal == NULL)) {
444                         /*
445                          * We raced with the reaping of the task.
446                          * The deletion should have cleared us off the list.
447                          */
448                         BUG_ON(!list_empty(&timer->it.cpu.entry));
449                 } else {
450                         spin_lock(&p->sighand->siglock);
451                         if (timer->it.cpu.firing)
452                                 ret = TIMER_RETRY;
453                         else
454                                 list_del(&timer->it.cpu.entry);
455                         spin_unlock(&p->sighand->siglock);
456                 }
457                 read_unlock(&tasklist_lock);
458
459                 if (!ret)
460                         put_task_struct(p);
461         }
462
463         return ret;
464 }
465
466 /*
467  * Clean out CPU timers still ticking when a thread exited.  The task
468  * pointer is cleared, and the expiry time is replaced with the residual
469  * time for later timer_gettime calls to return.
470  * This must be called with the siglock held.
471  */
472 static void cleanup_timers(struct list_head *head,
473                            cputime_t utime, cputime_t stime,
474                            unsigned long long sum_exec_runtime)
475 {
476         struct cpu_timer_list *timer, *next;
477         cputime_t ptime = cputime_add(utime, stime);
478
479         list_for_each_entry_safe(timer, next, head, entry) {
480                 list_del_init(&timer->entry);
481                 if (cputime_lt(timer->expires.cpu, ptime)) {
482                         timer->expires.cpu = cputime_zero;
483                 } else {
484                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
485                                                          ptime);
486                 }
487         }
488
489         ++head;
490         list_for_each_entry_safe(timer, next, head, entry) {
491                 list_del_init(&timer->entry);
492                 if (cputime_lt(timer->expires.cpu, utime)) {
493                         timer->expires.cpu = cputime_zero;
494                 } else {
495                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
496                                                          utime);
497                 }
498         }
499
500         ++head;
501         list_for_each_entry_safe(timer, next, head, entry) {
502                 list_del_init(&timer->entry);
503                 if (timer->expires.sched < sum_exec_runtime) {
504                         timer->expires.sched = 0;
505                 } else {
506                         timer->expires.sched -= sum_exec_runtime;
507                 }
508         }
509 }
510
511 /*
512  * These are both called with the siglock held, when the current thread
513  * is being reaped.  When the final (leader) thread in the group is reaped,
514  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
515  */
516 void posix_cpu_timers_exit(struct task_struct *tsk)
517 {
518         cleanup_timers(tsk->cpu_timers,
519                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
520
521 }
522 void posix_cpu_timers_exit_group(struct task_struct *tsk)
523 {
524         struct task_cputime cputime;
525
526         thread_group_cputimer(tsk, &cputime);
527         cleanup_timers(tsk->signal->cpu_timers,
528                        cputime.utime, cputime.stime, cputime.sum_exec_runtime);
529 }
530
531 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
532 {
533         /*
534          * That's all for this thread or process.
535          * We leave our residual in expires to be reported.
536          */
537         put_task_struct(timer->it.cpu.task);
538         timer->it.cpu.task = NULL;
539         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
540                                              timer->it.cpu.expires,
541                                              now);
542 }
543
544 /*
545  * Insert the timer on the appropriate list before any timers that
546  * expire later.  This must be called with the tasklist_lock held
547  * for reading, and interrupts disabled.
548  */
549 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
550 {
551         struct task_struct *p = timer->it.cpu.task;
552         struct list_head *head, *listpos;
553         struct cpu_timer_list *const nt = &timer->it.cpu;
554         struct cpu_timer_list *next;
555         unsigned long i;
556
557         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
558                 p->cpu_timers : p->signal->cpu_timers);
559         head += CPUCLOCK_WHICH(timer->it_clock);
560
561         BUG_ON(!irqs_disabled());
562         spin_lock(&p->sighand->siglock);
563
564         listpos = head;
565         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
566                 list_for_each_entry(next, head, entry) {
567                         if (next->expires.sched > nt->expires.sched)
568                                 break;
569                         listpos = &next->entry;
570                 }
571         } else {
572                 list_for_each_entry(next, head, entry) {
573                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
574                                 break;
575                         listpos = &next->entry;
576                 }
577         }
578         list_add(&nt->entry, listpos);
579
580         if (listpos == head) {
581                 /*
582                  * We are the new earliest-expiring timer.
583                  * If we are a thread timer, there can always
584                  * be a process timer telling us to stop earlier.
585                  */
586
587                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
588                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
589                         default:
590                                 BUG();
591                         case CPUCLOCK_PROF:
592                                 if (cputime_eq(p->cputime_expires.prof_exp,
593                                                cputime_zero) ||
594                                     cputime_gt(p->cputime_expires.prof_exp,
595                                                nt->expires.cpu))
596                                         p->cputime_expires.prof_exp =
597                                                 nt->expires.cpu;
598                                 break;
599                         case CPUCLOCK_VIRT:
600                                 if (cputime_eq(p->cputime_expires.virt_exp,
601                                                cputime_zero) ||
602                                     cputime_gt(p->cputime_expires.virt_exp,
603                                                nt->expires.cpu))
604                                         p->cputime_expires.virt_exp =
605                                                 nt->expires.cpu;
606                                 break;
607                         case CPUCLOCK_SCHED:
608                                 if (p->cputime_expires.sched_exp == 0 ||
609                                     p->cputime_expires.sched_exp >
610                                                         nt->expires.sched)
611                                         p->cputime_expires.sched_exp =
612                                                 nt->expires.sched;
613                                 break;
614                         }
615                 } else {
616                         struct signal_struct *const sig = p->signal;
617                         union cpu_time_count *exp = &timer->it.cpu.expires;
618
619                         /*
620                          * For a process timer, set the cached expiration time.
621                          */
622                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
623                         default:
624                                 BUG();
625                         case CPUCLOCK_VIRT:
626                                 if (!cputime_eq(sig->it[CPUCLOCK_VIRT].expires,
627                                                 cputime_zero) &&
628                                     cputime_lt(sig->it[CPUCLOCK_VIRT].expires,
629                                                exp->cpu))
630                                         break;
631                                 sig->cputime_expires.virt_exp = exp->cpu;
632                                 break;
633                         case CPUCLOCK_PROF:
634                                 if (!cputime_eq(sig->it[CPUCLOCK_PROF].expires,
635                                                 cputime_zero) &&
636                                     cputime_lt(sig->it[CPUCLOCK_PROF].expires,
637                                                exp->cpu))
638                                         break;
639                                 i = sig->rlim[RLIMIT_CPU].rlim_cur;
640                                 if (i != RLIM_INFINITY &&
641                                     i <= cputime_to_secs(exp->cpu))
642                                         break;
643                                 sig->cputime_expires.prof_exp = exp->cpu;
644                                 break;
645                         case CPUCLOCK_SCHED:
646                                 sig->cputime_expires.sched_exp = exp->sched;
647                                 break;
648                         }
649                 }
650         }
651
652         spin_unlock(&p->sighand->siglock);
653 }
654
655 /*
656  * The timer is locked, fire it and arrange for its reload.
657  */
658 static void cpu_timer_fire(struct k_itimer *timer)
659 {
660         if (unlikely(timer->sigq == NULL)) {
661                 /*
662                  * This a special case for clock_nanosleep,
663                  * not a normal timer from sys_timer_create.
664                  */
665                 wake_up_process(timer->it_process);
666                 timer->it.cpu.expires.sched = 0;
667         } else if (timer->it.cpu.incr.sched == 0) {
668                 /*
669                  * One-shot timer.  Clear it as soon as it's fired.
670                  */
671                 posix_timer_event(timer, 0);
672                 timer->it.cpu.expires.sched = 0;
673         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
674                 /*
675                  * The signal did not get queued because the signal
676                  * was ignored, so we won't get any callback to
677                  * reload the timer.  But we need to keep it
678                  * ticking in case the signal is deliverable next time.
679                  */
680                 posix_cpu_timer_schedule(timer);
681         }
682 }
683
684 /*
685  * Sample a process (thread group) timer for the given group_leader task.
686  * Must be called with tasklist_lock held for reading.
687  */
688 static int cpu_timer_sample_group(const clockid_t which_clock,
689                                   struct task_struct *p,
690                                   union cpu_time_count *cpu)
691 {
692         struct task_cputime cputime;
693
694         thread_group_cputimer(p, &cputime);
695         switch (CPUCLOCK_WHICH(which_clock)) {
696         default:
697                 return -EINVAL;
698         case CPUCLOCK_PROF:
699                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
700                 break;
701         case CPUCLOCK_VIRT:
702                 cpu->cpu = cputime.utime;
703                 break;
704         case CPUCLOCK_SCHED:
705                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
706                 break;
707         }
708         return 0;
709 }
710
711 /*
712  * Guts of sys_timer_settime for CPU timers.
713  * This is called with the timer locked and interrupts disabled.
714  * If we return TIMER_RETRY, it's necessary to release the timer's lock
715  * and try again.  (This happens when the timer is in the middle of firing.)
716  */
717 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
718                         struct itimerspec *new, struct itimerspec *old)
719 {
720         struct task_struct *p = timer->it.cpu.task;
721         union cpu_time_count old_expires, new_expires, val;
722         int ret;
723
724         if (unlikely(p == NULL)) {
725                 /*
726                  * Timer refers to a dead task's clock.
727                  */
728                 return -ESRCH;
729         }
730
731         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
732
733         read_lock(&tasklist_lock);
734         /*
735          * We need the tasklist_lock to protect against reaping that
736          * clears p->signal.  If p has just been reaped, we can no
737          * longer get any information about it at all.
738          */
739         if (unlikely(p->signal == NULL)) {
740                 read_unlock(&tasklist_lock);
741                 put_task_struct(p);
742                 timer->it.cpu.task = NULL;
743                 return -ESRCH;
744         }
745
746         /*
747          * Disarm any old timer after extracting its expiry time.
748          */
749         BUG_ON(!irqs_disabled());
750
751         ret = 0;
752         spin_lock(&p->sighand->siglock);
753         old_expires = timer->it.cpu.expires;
754         if (unlikely(timer->it.cpu.firing)) {
755                 timer->it.cpu.firing = -1;
756                 ret = TIMER_RETRY;
757         } else
758                 list_del_init(&timer->it.cpu.entry);
759         spin_unlock(&p->sighand->siglock);
760
761         /*
762          * We need to sample the current value to convert the new
763          * value from to relative and absolute, and to convert the
764          * old value from absolute to relative.  To set a process
765          * timer, we need a sample to balance the thread expiry
766          * times (in arm_timer).  With an absolute time, we must
767          * check if it's already passed.  In short, we need a sample.
768          */
769         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
770                 cpu_clock_sample(timer->it_clock, p, &val);
771         } else {
772                 cpu_timer_sample_group(timer->it_clock, p, &val);
773         }
774
775         if (old) {
776                 if (old_expires.sched == 0) {
777                         old->it_value.tv_sec = 0;
778                         old->it_value.tv_nsec = 0;
779                 } else {
780                         /*
781                          * Update the timer in case it has
782                          * overrun already.  If it has,
783                          * we'll report it as having overrun
784                          * and with the next reloaded timer
785                          * already ticking, though we are
786                          * swallowing that pending
787                          * notification here to install the
788                          * new setting.
789                          */
790                         bump_cpu_timer(timer, val);
791                         if (cpu_time_before(timer->it_clock, val,
792                                             timer->it.cpu.expires)) {
793                                 old_expires = cpu_time_sub(
794                                         timer->it_clock,
795                                         timer->it.cpu.expires, val);
796                                 sample_to_timespec(timer->it_clock,
797                                                    old_expires,
798                                                    &old->it_value);
799                         } else {
800                                 old->it_value.tv_nsec = 1;
801                                 old->it_value.tv_sec = 0;
802                         }
803                 }
804         }
805
806         if (unlikely(ret)) {
807                 /*
808                  * We are colliding with the timer actually firing.
809                  * Punt after filling in the timer's old value, and
810                  * disable this firing since we are already reporting
811                  * it as an overrun (thanks to bump_cpu_timer above).
812                  */
813                 read_unlock(&tasklist_lock);
814                 goto out;
815         }
816
817         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
818                 cpu_time_add(timer->it_clock, &new_expires, val);
819         }
820
821         /*
822          * Install the new expiry time (or zero).
823          * For a timer with no notification action, we don't actually
824          * arm the timer (we'll just fake it for timer_gettime).
825          */
826         timer->it.cpu.expires = new_expires;
827         if (new_expires.sched != 0 &&
828             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
829             cpu_time_before(timer->it_clock, val, new_expires)) {
830                 arm_timer(timer, val);
831         }
832
833         read_unlock(&tasklist_lock);
834
835         /*
836          * Install the new reload setting, and
837          * set up the signal and overrun bookkeeping.
838          */
839         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
840                                                 &new->it_interval);
841
842         /*
843          * This acts as a modification timestamp for the timer,
844          * so any automatic reload attempt will punt on seeing
845          * that we have reset the timer manually.
846          */
847         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
848                 ~REQUEUE_PENDING;
849         timer->it_overrun_last = 0;
850         timer->it_overrun = -1;
851
852         if (new_expires.sched != 0 &&
853             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
854             !cpu_time_before(timer->it_clock, val, new_expires)) {
855                 /*
856                  * The designated time already passed, so we notify
857                  * immediately, even if the thread never runs to
858                  * accumulate more time on this clock.
859                  */
860                 cpu_timer_fire(timer);
861         }
862
863         ret = 0;
864  out:
865         if (old) {
866                 sample_to_timespec(timer->it_clock,
867                                    timer->it.cpu.incr, &old->it_interval);
868         }
869         return ret;
870 }
871
872 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
873 {
874         union cpu_time_count now;
875         struct task_struct *p = timer->it.cpu.task;
876         int clear_dead;
877
878         /*
879          * Easy part: convert the reload time.
880          */
881         sample_to_timespec(timer->it_clock,
882                            timer->it.cpu.incr, &itp->it_interval);
883
884         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
885                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
886                 return;
887         }
888
889         if (unlikely(p == NULL)) {
890                 /*
891                  * This task already died and the timer will never fire.
892                  * In this case, expires is actually the dead value.
893                  */
894         dead:
895                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
896                                    &itp->it_value);
897                 return;
898         }
899
900         /*
901          * Sample the clock to take the difference with the expiry time.
902          */
903         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
904                 cpu_clock_sample(timer->it_clock, p, &now);
905                 clear_dead = p->exit_state;
906         } else {
907                 read_lock(&tasklist_lock);
908                 if (unlikely(p->signal == NULL)) {
909                         /*
910                          * The process has been reaped.
911                          * We can't even collect a sample any more.
912                          * Call the timer disarmed, nothing else to do.
913                          */
914                         put_task_struct(p);
915                         timer->it.cpu.task = NULL;
916                         timer->it.cpu.expires.sched = 0;
917                         read_unlock(&tasklist_lock);
918                         goto dead;
919                 } else {
920                         cpu_timer_sample_group(timer->it_clock, p, &now);
921                         clear_dead = (unlikely(p->exit_state) &&
922                                       thread_group_empty(p));
923                 }
924                 read_unlock(&tasklist_lock);
925         }
926
927         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
928                 if (timer->it.cpu.incr.sched == 0 &&
929                     cpu_time_before(timer->it_clock,
930                                     timer->it.cpu.expires, now)) {
931                         /*
932                          * Do-nothing timer expired and has no reload,
933                          * so it's as if it was never set.
934                          */
935                         timer->it.cpu.expires.sched = 0;
936                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
937                         return;
938                 }
939                 /*
940                  * Account for any expirations and reloads that should
941                  * have happened.
942                  */
943                 bump_cpu_timer(timer, now);
944         }
945
946         if (unlikely(clear_dead)) {
947                 /*
948                  * We've noticed that the thread is dead, but
949                  * not yet reaped.  Take this opportunity to
950                  * drop our task ref.
951                  */
952                 clear_dead_task(timer, now);
953                 goto dead;
954         }
955
956         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
957                 sample_to_timespec(timer->it_clock,
958                                    cpu_time_sub(timer->it_clock,
959                                                 timer->it.cpu.expires, now),
960                                    &itp->it_value);
961         } else {
962                 /*
963                  * The timer should have expired already, but the firing
964                  * hasn't taken place yet.  Say it's just about to expire.
965                  */
966                 itp->it_value.tv_nsec = 1;
967                 itp->it_value.tv_sec = 0;
968         }
969 }
970
971 /*
972  * Check for any per-thread CPU timers that have fired and move them off
973  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
974  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
975  */
976 static void check_thread_timers(struct task_struct *tsk,
977                                 struct list_head *firing)
978 {
979         int maxfire;
980         struct list_head *timers = tsk->cpu_timers;
981         struct signal_struct *const sig = tsk->signal;
982
983         maxfire = 20;
984         tsk->cputime_expires.prof_exp = cputime_zero;
985         while (!list_empty(timers)) {
986                 struct cpu_timer_list *t = list_first_entry(timers,
987                                                       struct cpu_timer_list,
988                                                       entry);
989                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
990                         tsk->cputime_expires.prof_exp = t->expires.cpu;
991                         break;
992                 }
993                 t->firing = 1;
994                 list_move_tail(&t->entry, firing);
995         }
996
997         ++timers;
998         maxfire = 20;
999         tsk->cputime_expires.virt_exp = cputime_zero;
1000         while (!list_empty(timers)) {
1001                 struct cpu_timer_list *t = list_first_entry(timers,
1002                                                       struct cpu_timer_list,
1003                                                       entry);
1004                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1005                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1006                         break;
1007                 }
1008                 t->firing = 1;
1009                 list_move_tail(&t->entry, firing);
1010         }
1011
1012         ++timers;
1013         maxfire = 20;
1014         tsk->cputime_expires.sched_exp = 0;
1015         while (!list_empty(timers)) {
1016                 struct cpu_timer_list *t = list_first_entry(timers,
1017                                                       struct cpu_timer_list,
1018                                                       entry);
1019                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1020                         tsk->cputime_expires.sched_exp = t->expires.sched;
1021                         break;
1022                 }
1023                 t->firing = 1;
1024                 list_move_tail(&t->entry, firing);
1025         }
1026
1027         /*
1028          * Check for the special case thread timers.
1029          */
1030         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1031                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1032                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1033
1034                 if (hard != RLIM_INFINITY &&
1035                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1036                         /*
1037                          * At the hard limit, we just die.
1038                          * No need to calculate anything else now.
1039                          */
1040                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1041                         return;
1042                 }
1043                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1044                         /*
1045                          * At the soft limit, send a SIGXCPU every second.
1046                          */
1047                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1048                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1049                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1050                                                                 USEC_PER_SEC;
1051                         }
1052                         printk(KERN_INFO
1053                                 "RT Watchdog Timeout: %s[%d]\n",
1054                                 tsk->comm, task_pid_nr(tsk));
1055                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1056                 }
1057         }
1058 }
1059
1060 static void stop_process_timers(struct task_struct *tsk)
1061 {
1062         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
1063         unsigned long flags;
1064
1065         if (!cputimer->running)
1066                 return;
1067
1068         spin_lock_irqsave(&cputimer->lock, flags);
1069         cputimer->running = 0;
1070         spin_unlock_irqrestore(&cputimer->lock, flags);
1071 }
1072
1073 static u32 onecputick;
1074
1075 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1076                              cputime_t *expires, cputime_t cur_time, int signo)
1077 {
1078         if (cputime_eq(it->expires, cputime_zero))
1079                 return;
1080
1081         if (cputime_ge(cur_time, it->expires)) {
1082                 if (!cputime_eq(it->incr, cputime_zero)) {
1083                         it->expires = cputime_add(it->expires, it->incr);
1084                         it->error += it->incr_error;
1085                         if (it->error >= onecputick) {
1086                                 it->expires = cputime_sub(it->expires,
1087                                                         jiffies_to_cputime(1));
1088                                 it->error -= onecputick;
1089                         }
1090                 } else
1091                         it->expires = cputime_zero;
1092
1093                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1094         }
1095
1096         if (!cputime_eq(it->expires, cputime_zero) &&
1097             (cputime_eq(*expires, cputime_zero) ||
1098              cputime_lt(it->expires, *expires))) {
1099                 *expires = it->expires;
1100         }
1101 }
1102
1103 /*
1104  * Check for any per-thread CPU timers that have fired and move them
1105  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1106  * have already been taken off.
1107  */
1108 static void check_process_timers(struct task_struct *tsk,
1109                                  struct list_head *firing)
1110 {
1111         int maxfire;
1112         struct signal_struct *const sig = tsk->signal;
1113         cputime_t utime, ptime, virt_expires, prof_expires;
1114         unsigned long long sum_sched_runtime, sched_expires;
1115         struct list_head *timers = sig->cpu_timers;
1116         struct task_cputime cputime;
1117
1118         /*
1119          * Don't sample the current process CPU clocks if there are no timers.
1120          */
1121         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1122             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1123             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1124             list_empty(&timers[CPUCLOCK_VIRT]) &&
1125             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1126             list_empty(&timers[CPUCLOCK_SCHED])) {
1127                 stop_process_timers(tsk);
1128                 return;
1129         }
1130
1131         /*
1132          * Collect the current process totals.
1133          */
1134         thread_group_cputimer(tsk, &cputime);
1135         utime = cputime.utime;
1136         ptime = cputime_add(utime, cputime.stime);
1137         sum_sched_runtime = cputime.sum_exec_runtime;
1138         maxfire = 20;
1139         prof_expires = cputime_zero;
1140         while (!list_empty(timers)) {
1141                 struct cpu_timer_list *tl = list_first_entry(timers,
1142                                                       struct cpu_timer_list,
1143                                                       entry);
1144                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1145                         prof_expires = tl->expires.cpu;
1146                         break;
1147                 }
1148                 tl->firing = 1;
1149                 list_move_tail(&tl->entry, firing);
1150         }
1151
1152         ++timers;
1153         maxfire = 20;
1154         virt_expires = cputime_zero;
1155         while (!list_empty(timers)) {
1156                 struct cpu_timer_list *tl = list_first_entry(timers,
1157                                                       struct cpu_timer_list,
1158                                                       entry);
1159                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1160                         virt_expires = tl->expires.cpu;
1161                         break;
1162                 }
1163                 tl->firing = 1;
1164                 list_move_tail(&tl->entry, firing);
1165         }
1166
1167         ++timers;
1168         maxfire = 20;
1169         sched_expires = 0;
1170         while (!list_empty(timers)) {
1171                 struct cpu_timer_list *tl = list_first_entry(timers,
1172                                                       struct cpu_timer_list,
1173                                                       entry);
1174                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1175                         sched_expires = tl->expires.sched;
1176                         break;
1177                 }
1178                 tl->firing = 1;
1179                 list_move_tail(&tl->entry, firing);
1180         }
1181
1182         /*
1183          * Check for the special case process timers.
1184          */
1185         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1186                          SIGPROF);
1187         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1188                          SIGVTALRM);
1189
1190         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1191                 unsigned long psecs = cputime_to_secs(ptime);
1192                 cputime_t x;
1193                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1194                         /*
1195                          * At the hard limit, we just die.
1196                          * No need to calculate anything else now.
1197                          */
1198                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1199                         return;
1200                 }
1201                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1202                         /*
1203                          * At the soft limit, send a SIGXCPU every second.
1204                          */
1205                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1206                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1207                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1208                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1209                         }
1210                 }
1211                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1212                 if (cputime_eq(prof_expires, cputime_zero) ||
1213                     cputime_lt(x, prof_expires)) {
1214                         prof_expires = x;
1215                 }
1216         }
1217
1218         if (!cputime_eq(prof_expires, cputime_zero) &&
1219             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1220              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1221                 sig->cputime_expires.prof_exp = prof_expires;
1222         if (!cputime_eq(virt_expires, cputime_zero) &&
1223             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1224              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1225                 sig->cputime_expires.virt_exp = virt_expires;
1226         if (sched_expires != 0 &&
1227             (sig->cputime_expires.sched_exp == 0 ||
1228              sig->cputime_expires.sched_exp > sched_expires))
1229                 sig->cputime_expires.sched_exp = sched_expires;
1230 }
1231
1232 /*
1233  * This is called from the signal code (via do_schedule_next_timer)
1234  * when the last timer signal was delivered and we have to reload the timer.
1235  */
1236 void posix_cpu_timer_schedule(struct k_itimer *timer)
1237 {
1238         struct task_struct *p = timer->it.cpu.task;
1239         union cpu_time_count now;
1240
1241         if (unlikely(p == NULL))
1242                 /*
1243                  * The task was cleaned up already, no future firings.
1244                  */
1245                 goto out;
1246
1247         /*
1248          * Fetch the current sample and update the timer's expiry time.
1249          */
1250         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1251                 cpu_clock_sample(timer->it_clock, p, &now);
1252                 bump_cpu_timer(timer, now);
1253                 if (unlikely(p->exit_state)) {
1254                         clear_dead_task(timer, now);
1255                         goto out;
1256                 }
1257                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1258         } else {
1259                 read_lock(&tasklist_lock);
1260                 if (unlikely(p->signal == NULL)) {
1261                         /*
1262                          * The process has been reaped.
1263                          * We can't even collect a sample any more.
1264                          */
1265                         put_task_struct(p);
1266                         timer->it.cpu.task = p = NULL;
1267                         timer->it.cpu.expires.sched = 0;
1268                         goto out_unlock;
1269                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1270                         /*
1271                          * We've noticed that the thread is dead, but
1272                          * not yet reaped.  Take this opportunity to
1273                          * drop our task ref.
1274                          */
1275                         clear_dead_task(timer, now);
1276                         goto out_unlock;
1277                 }
1278                 cpu_timer_sample_group(timer->it_clock, p, &now);
1279                 bump_cpu_timer(timer, now);
1280                 /* Leave the tasklist_lock locked for the call below.  */
1281         }
1282
1283         /*
1284          * Now re-arm for the new expiry time.
1285          */
1286         arm_timer(timer, now);
1287
1288 out_unlock:
1289         read_unlock(&tasklist_lock);
1290
1291 out:
1292         timer->it_overrun_last = timer->it_overrun;
1293         timer->it_overrun = -1;
1294         ++timer->it_requeue_pending;
1295 }
1296
1297 /**
1298  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1299  *
1300  * @cputime:    The struct to compare.
1301  *
1302  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1303  * are zero, false if any field is nonzero.
1304  */
1305 static inline int task_cputime_zero(const struct task_cputime *cputime)
1306 {
1307         if (cputime_eq(cputime->utime, cputime_zero) &&
1308             cputime_eq(cputime->stime, cputime_zero) &&
1309             cputime->sum_exec_runtime == 0)
1310                 return 1;
1311         return 0;
1312 }
1313
1314 /**
1315  * task_cputime_expired - Compare two task_cputime entities.
1316  *
1317  * @sample:     The task_cputime structure to be checked for expiration.
1318  * @expires:    Expiration times, against which @sample will be checked.
1319  *
1320  * Checks @sample against @expires to see if any field of @sample has expired.
1321  * Returns true if any field of the former is greater than the corresponding
1322  * field of the latter if the latter field is set.  Otherwise returns false.
1323  */
1324 static inline int task_cputime_expired(const struct task_cputime *sample,
1325                                         const struct task_cputime *expires)
1326 {
1327         if (!cputime_eq(expires->utime, cputime_zero) &&
1328             cputime_ge(sample->utime, expires->utime))
1329                 return 1;
1330         if (!cputime_eq(expires->stime, cputime_zero) &&
1331             cputime_ge(cputime_add(sample->utime, sample->stime),
1332                        expires->stime))
1333                 return 1;
1334         if (expires->sum_exec_runtime != 0 &&
1335             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1336                 return 1;
1337         return 0;
1338 }
1339
1340 /**
1341  * fastpath_timer_check - POSIX CPU timers fast path.
1342  *
1343  * @tsk:        The task (thread) being checked.
1344  *
1345  * Check the task and thread group timers.  If both are zero (there are no
1346  * timers set) return false.  Otherwise snapshot the task and thread group
1347  * timers and compare them with the corresponding expiration times.  Return
1348  * true if a timer has expired, else return false.
1349  */
1350 static inline int fastpath_timer_check(struct task_struct *tsk)
1351 {
1352         struct signal_struct *sig;
1353
1354         /* tsk == current, ensure it is safe to use ->signal/sighand */
1355         if (unlikely(tsk->exit_state))
1356                 return 0;
1357
1358         if (!task_cputime_zero(&tsk->cputime_expires)) {
1359                 struct task_cputime task_sample = {
1360                         .utime = tsk->utime,
1361                         .stime = tsk->stime,
1362                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1363                 };
1364
1365                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1366                         return 1;
1367         }
1368
1369         sig = tsk->signal;
1370         if (!task_cputime_zero(&sig->cputime_expires)) {
1371                 struct task_cputime group_sample;
1372
1373                 thread_group_cputimer(tsk, &group_sample);
1374                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1375                         return 1;
1376         }
1377
1378         return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1379 }
1380
1381 /*
1382  * This is called from the timer interrupt handler.  The irq handler has
1383  * already updated our counts.  We need to check if any timers fire now.
1384  * Interrupts are disabled.
1385  */
1386 void run_posix_cpu_timers(struct task_struct *tsk)
1387 {
1388         LIST_HEAD(firing);
1389         struct k_itimer *timer, *next;
1390
1391         BUG_ON(!irqs_disabled());
1392
1393         /*
1394          * The fast path checks that there are no expired thread or thread
1395          * group timers.  If that's so, just return.
1396          */
1397         if (!fastpath_timer_check(tsk))
1398                 return;
1399
1400         spin_lock(&tsk->sighand->siglock);
1401         /*
1402          * Here we take off tsk->signal->cpu_timers[N] and
1403          * tsk->cpu_timers[N] all the timers that are firing, and
1404          * put them on the firing list.
1405          */
1406         check_thread_timers(tsk, &firing);
1407         check_process_timers(tsk, &firing);
1408
1409         /*
1410          * We must release these locks before taking any timer's lock.
1411          * There is a potential race with timer deletion here, as the
1412          * siglock now protects our private firing list.  We have set
1413          * the firing flag in each timer, so that a deletion attempt
1414          * that gets the timer lock before we do will give it up and
1415          * spin until we've taken care of that timer below.
1416          */
1417         spin_unlock(&tsk->sighand->siglock);
1418
1419         /*
1420          * Now that all the timers on our list have the firing flag,
1421          * noone will touch their list entries but us.  We'll take
1422          * each timer's lock before clearing its firing flag, so no
1423          * timer call will interfere.
1424          */
1425         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1426                 int cpu_firing;
1427
1428                 spin_lock(&timer->it_lock);
1429                 list_del_init(&timer->it.cpu.entry);
1430                 cpu_firing = timer->it.cpu.firing;
1431                 timer->it.cpu.firing = 0;
1432                 /*
1433                  * The firing flag is -1 if we collided with a reset
1434                  * of the timer, which already reported this
1435                  * almost-firing as an overrun.  So don't generate an event.
1436                  */
1437                 if (likely(cpu_firing >= 0))
1438                         cpu_timer_fire(timer);
1439                 spin_unlock(&timer->it_lock);
1440         }
1441 }
1442
1443 /*
1444  * Set one of the process-wide special case CPU timers.
1445  * The tsk->sighand->siglock must be held by the caller.
1446  * The *newval argument is relative and we update it to be absolute, *oldval
1447  * is absolute and we update it to be relative.
1448  */
1449 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1450                            cputime_t *newval, cputime_t *oldval)
1451 {
1452         union cpu_time_count now;
1453         struct list_head *head;
1454
1455         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1456         cpu_timer_sample_group(clock_idx, tsk, &now);
1457
1458         if (oldval) {
1459                 if (!cputime_eq(*oldval, cputime_zero)) {
1460                         if (cputime_le(*oldval, now.cpu)) {
1461                                 /* Just about to fire. */
1462                                 *oldval = jiffies_to_cputime(1);
1463                         } else {
1464                                 *oldval = cputime_sub(*oldval, now.cpu);
1465                         }
1466                 }
1467
1468                 if (cputime_eq(*newval, cputime_zero))
1469                         return;
1470                 *newval = cputime_add(*newval, now.cpu);
1471
1472                 /*
1473                  * If the RLIMIT_CPU timer will expire before the
1474                  * ITIMER_PROF timer, we have nothing else to do.
1475                  */
1476                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1477                     < cputime_to_secs(*newval))
1478                         return;
1479         }
1480
1481         /*
1482          * Check whether there are any process timers already set to fire
1483          * before this one.  If so, we don't have anything more to do.
1484          */
1485         head = &tsk->signal->cpu_timers[clock_idx];
1486         if (list_empty(head) ||
1487             cputime_ge(list_first_entry(head,
1488                                   struct cpu_timer_list, entry)->expires.cpu,
1489                        *newval)) {
1490                 switch (clock_idx) {
1491                 case CPUCLOCK_PROF:
1492                         tsk->signal->cputime_expires.prof_exp = *newval;
1493                         break;
1494                 case CPUCLOCK_VIRT:
1495                         tsk->signal->cputime_expires.virt_exp = *newval;
1496                         break;
1497                 }
1498         }
1499 }
1500
1501 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1502                             struct timespec *rqtp, struct itimerspec *it)
1503 {
1504         struct k_itimer timer;
1505         int error;
1506
1507         /*
1508          * Set up a temporary timer and then wait for it to go off.
1509          */
1510         memset(&timer, 0, sizeof timer);
1511         spin_lock_init(&timer.it_lock);
1512         timer.it_clock = which_clock;
1513         timer.it_overrun = -1;
1514         error = posix_cpu_timer_create(&timer);
1515         timer.it_process = current;
1516         if (!error) {
1517                 static struct itimerspec zero_it;
1518
1519                 memset(it, 0, sizeof *it);
1520                 it->it_value = *rqtp;
1521
1522                 spin_lock_irq(&timer.it_lock);
1523                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1524                 if (error) {
1525                         spin_unlock_irq(&timer.it_lock);
1526                         return error;
1527                 }
1528
1529                 while (!signal_pending(current)) {
1530                         if (timer.it.cpu.expires.sched == 0) {
1531                                 /*
1532                                  * Our timer fired and was reset.
1533                                  */
1534                                 spin_unlock_irq(&timer.it_lock);
1535                                 return 0;
1536                         }
1537
1538                         /*
1539                          * Block until cpu_timer_fire (or a signal) wakes us.
1540                          */
1541                         __set_current_state(TASK_INTERRUPTIBLE);
1542                         spin_unlock_irq(&timer.it_lock);
1543                         schedule();
1544                         spin_lock_irq(&timer.it_lock);
1545                 }
1546
1547                 /*
1548                  * We were interrupted by a signal.
1549                  */
1550                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1551                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1552                 spin_unlock_irq(&timer.it_lock);
1553
1554                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1555                         /*
1556                          * It actually did fire already.
1557                          */
1558                         return 0;
1559                 }
1560
1561                 error = -ERESTART_RESTARTBLOCK;
1562         }
1563
1564         return error;
1565 }
1566
1567 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1568                      struct timespec *rqtp, struct timespec __user *rmtp)
1569 {
1570         struct restart_block *restart_block =
1571             &current_thread_info()->restart_block;
1572         struct itimerspec it;
1573         int error;
1574
1575         /*
1576          * Diagnose required errors first.
1577          */
1578         if (CPUCLOCK_PERTHREAD(which_clock) &&
1579             (CPUCLOCK_PID(which_clock) == 0 ||
1580              CPUCLOCK_PID(which_clock) == current->pid))
1581                 return -EINVAL;
1582
1583         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1584
1585         if (error == -ERESTART_RESTARTBLOCK) {
1586
1587                 if (flags & TIMER_ABSTIME)
1588                         return -ERESTARTNOHAND;
1589                 /*
1590                  * Report back to the user the time still remaining.
1591                  */
1592                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1593                         return -EFAULT;
1594
1595                 restart_block->fn = posix_cpu_nsleep_restart;
1596                 restart_block->arg0 = which_clock;
1597                 restart_block->arg1 = (unsigned long) rmtp;
1598                 restart_block->arg2 = rqtp->tv_sec;
1599                 restart_block->arg3 = rqtp->tv_nsec;
1600         }
1601         return error;
1602 }
1603
1604 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1605 {
1606         clockid_t which_clock = restart_block->arg0;
1607         struct timespec __user *rmtp;
1608         struct timespec t;
1609         struct itimerspec it;
1610         int error;
1611
1612         rmtp = (struct timespec __user *) restart_block->arg1;
1613         t.tv_sec = restart_block->arg2;
1614         t.tv_nsec = restart_block->arg3;
1615
1616         restart_block->fn = do_no_restart_syscall;
1617         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1618
1619         if (error == -ERESTART_RESTARTBLOCK) {
1620                 /*
1621                  * Report back to the user the time still remaining.
1622                  */
1623                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1624                         return -EFAULT;
1625
1626                 restart_block->fn = posix_cpu_nsleep_restart;
1627                 restart_block->arg0 = which_clock;
1628                 restart_block->arg1 = (unsigned long) rmtp;
1629                 restart_block->arg2 = t.tv_sec;
1630                 restart_block->arg3 = t.tv_nsec;
1631         }
1632         return error;
1633
1634 }
1635
1636
1637 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1638 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1639
1640 static int process_cpu_clock_getres(const clockid_t which_clock,
1641                                     struct timespec *tp)
1642 {
1643         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1644 }
1645 static int process_cpu_clock_get(const clockid_t which_clock,
1646                                  struct timespec *tp)
1647 {
1648         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1649 }
1650 static int process_cpu_timer_create(struct k_itimer *timer)
1651 {
1652         timer->it_clock = PROCESS_CLOCK;
1653         return posix_cpu_timer_create(timer);
1654 }
1655 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1656                               struct timespec *rqtp,
1657                               struct timespec __user *rmtp)
1658 {
1659         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1660 }
1661 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1662 {
1663         return -EINVAL;
1664 }
1665 static int thread_cpu_clock_getres(const clockid_t which_clock,
1666                                    struct timespec *tp)
1667 {
1668         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1669 }
1670 static int thread_cpu_clock_get(const clockid_t which_clock,
1671                                 struct timespec *tp)
1672 {
1673         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1674 }
1675 static int thread_cpu_timer_create(struct k_itimer *timer)
1676 {
1677         timer->it_clock = THREAD_CLOCK;
1678         return posix_cpu_timer_create(timer);
1679 }
1680 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1681                               struct timespec *rqtp, struct timespec __user *rmtp)
1682 {
1683         return -EINVAL;
1684 }
1685 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1686 {
1687         return -EINVAL;
1688 }
1689
1690 static __init int init_posix_cpu_timers(void)
1691 {
1692         struct k_clock process = {
1693                 .clock_getres = process_cpu_clock_getres,
1694                 .clock_get = process_cpu_clock_get,
1695                 .clock_set = do_posix_clock_nosettime,
1696                 .timer_create = process_cpu_timer_create,
1697                 .nsleep = process_cpu_nsleep,
1698                 .nsleep_restart = process_cpu_nsleep_restart,
1699         };
1700         struct k_clock thread = {
1701                 .clock_getres = thread_cpu_clock_getres,
1702                 .clock_get = thread_cpu_clock_get,
1703                 .clock_set = do_posix_clock_nosettime,
1704                 .timer_create = thread_cpu_timer_create,
1705                 .nsleep = thread_cpu_nsleep,
1706                 .nsleep_restart = thread_cpu_nsleep_restart,
1707         };
1708         struct timespec ts;
1709
1710         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1711         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1712
1713         cputime_to_timespec(jiffies_to_cputime(1), &ts);
1714         onecputick = ts.tv_nsec;
1715         WARN_ON(ts.tv_sec != 0);
1716
1717         return 0;
1718 }
1719 __initcall(init_posix_cpu_timers);