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