Merge branch 'linus' into timers/core
[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 #include <trace/events/timer.h>
12
13 /*
14  * Called after updating RLIMIT_CPU to run cpu timer and update
15  * tsk->signal->cputime_expires expiration cache if necessary. Needs
16  * siglock protection since other code may update expiration cache as
17  * well.
18  */
19 void update_rlimit_cpu(unsigned long rlim_new)
20 {
21         cputime_t cputime = secs_to_cputime(rlim_new);
22
23         spin_lock_irq(&current->sighand->siglock);
24         set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
25         spin_unlock_irq(&current->sighand->siglock);
26 }
27
28 static int check_clock(const clockid_t which_clock)
29 {
30         int error = 0;
31         struct task_struct *p;
32         const pid_t pid = CPUCLOCK_PID(which_clock);
33
34         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
35                 return -EINVAL;
36
37         if (pid == 0)
38                 return 0;
39
40         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() and do_cpu_nanosleep() with the
387  * new timer already all-zeros initialized.
388  */
389 int posix_cpu_timer_create(struct k_itimer *new_timer)
390 {
391         int ret = 0;
392         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
393         struct task_struct *p;
394
395         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
396                 return -EINVAL;
397
398         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
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 signal_struct *const sig = tsk->signal;
524
525         cleanup_timers(tsk->signal->cpu_timers,
526                        cputime_add(tsk->utime, sig->utime),
527                        cputime_add(tsk->stime, sig->stime),
528                        tsk->se.sum_exec_runtime + sig->sum_sched_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 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
545 {
546         return cputime_eq(expires, cputime_zero) ||
547                cputime_gt(expires, new_exp);
548 }
549
550 /*
551  * Insert the timer on the appropriate list before any timers that
552  * expire later.  This must be called with the tasklist_lock held
553  * for reading, interrupts disabled and p->sighand->siglock taken.
554  */
555 static void arm_timer(struct k_itimer *timer)
556 {
557         struct task_struct *p = timer->it.cpu.task;
558         struct list_head *head, *listpos;
559         struct task_cputime *cputime_expires;
560         struct cpu_timer_list *const nt = &timer->it.cpu;
561         struct cpu_timer_list *next;
562
563         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
564                 head = p->cpu_timers;
565                 cputime_expires = &p->cputime_expires;
566         } else {
567                 head = p->signal->cpu_timers;
568                 cputime_expires = &p->signal->cputime_expires;
569         }
570         head += CPUCLOCK_WHICH(timer->it_clock);
571
572         listpos = head;
573         list_for_each_entry(next, head, entry) {
574                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
575                         break;
576                 listpos = &next->entry;
577         }
578         list_add(&nt->entry, listpos);
579
580         if (listpos == head) {
581                 union cpu_time_count *exp = &nt->expires;
582
583                 /*
584                  * We are the new earliest-expiring POSIX 1.b timer, hence
585                  * need to update expiration cache. Take into account that
586                  * for process timers we share expiration cache with itimers
587                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
588                  */
589
590                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
591                 case CPUCLOCK_PROF:
592                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
593                                 cputime_expires->prof_exp = exp->cpu;
594                         break;
595                 case CPUCLOCK_VIRT:
596                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
597                                 cputime_expires->virt_exp = exp->cpu;
598                         break;
599                 case CPUCLOCK_SCHED:
600                         if (cputime_expires->sched_exp == 0 ||
601                             cputime_expires->sched_exp > exp->sched)
602                                 cputime_expires->sched_exp = exp->sched;
603                         break;
604                 }
605         }
606 }
607
608 /*
609  * The timer is locked, fire it and arrange for its reload.
610  */
611 static void cpu_timer_fire(struct k_itimer *timer)
612 {
613         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
614                 /*
615                  * User don't want any signal.
616                  */
617                 timer->it.cpu.expires.sched = 0;
618         } else if (unlikely(timer->sigq == NULL)) {
619                 /*
620                  * This a special case for clock_nanosleep,
621                  * not a normal timer from sys_timer_create.
622                  */
623                 wake_up_process(timer->it_process);
624                 timer->it.cpu.expires.sched = 0;
625         } else if (timer->it.cpu.incr.sched == 0) {
626                 /*
627                  * One-shot timer.  Clear it as soon as it's fired.
628                  */
629                 posix_timer_event(timer, 0);
630                 timer->it.cpu.expires.sched = 0;
631         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
632                 /*
633                  * The signal did not get queued because the signal
634                  * was ignored, so we won't get any callback to
635                  * reload the timer.  But we need to keep it
636                  * ticking in case the signal is deliverable next time.
637                  */
638                 posix_cpu_timer_schedule(timer);
639         }
640 }
641
642 /*
643  * Sample a process (thread group) timer for the given group_leader task.
644  * Must be called with tasklist_lock held for reading.
645  */
646 static int cpu_timer_sample_group(const clockid_t which_clock,
647                                   struct task_struct *p,
648                                   union cpu_time_count *cpu)
649 {
650         struct task_cputime cputime;
651
652         thread_group_cputimer(p, &cputime);
653         switch (CPUCLOCK_WHICH(which_clock)) {
654         default:
655                 return -EINVAL;
656         case CPUCLOCK_PROF:
657                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
658                 break;
659         case CPUCLOCK_VIRT:
660                 cpu->cpu = cputime.utime;
661                 break;
662         case CPUCLOCK_SCHED:
663                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
664                 break;
665         }
666         return 0;
667 }
668
669 /*
670  * Guts of sys_timer_settime for CPU timers.
671  * This is called with the timer locked and interrupts disabled.
672  * If we return TIMER_RETRY, it's necessary to release the timer's lock
673  * and try again.  (This happens when the timer is in the middle of firing.)
674  */
675 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
676                         struct itimerspec *new, struct itimerspec *old)
677 {
678         struct task_struct *p = timer->it.cpu.task;
679         union cpu_time_count old_expires, new_expires, old_incr, val;
680         int ret;
681
682         if (unlikely(p == NULL)) {
683                 /*
684                  * Timer refers to a dead task's clock.
685                  */
686                 return -ESRCH;
687         }
688
689         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
690
691         read_lock(&tasklist_lock);
692         /*
693          * We need the tasklist_lock to protect against reaping that
694          * clears p->signal.  If p has just been reaped, we can no
695          * longer get any information about it at all.
696          */
697         if (unlikely(p->signal == NULL)) {
698                 read_unlock(&tasklist_lock);
699                 put_task_struct(p);
700                 timer->it.cpu.task = NULL;
701                 return -ESRCH;
702         }
703
704         /*
705          * Disarm any old timer after extracting its expiry time.
706          */
707         BUG_ON(!irqs_disabled());
708
709         ret = 0;
710         old_incr = timer->it.cpu.incr;
711         spin_lock(&p->sighand->siglock);
712         old_expires = timer->it.cpu.expires;
713         if (unlikely(timer->it.cpu.firing)) {
714                 timer->it.cpu.firing = -1;
715                 ret = TIMER_RETRY;
716         } else
717                 list_del_init(&timer->it.cpu.entry);
718
719         /*
720          * We need to sample the current value to convert the new
721          * value from to relative and absolute, and to convert the
722          * old value from absolute to relative.  To set a process
723          * timer, we need a sample to balance the thread expiry
724          * times (in arm_timer).  With an absolute time, we must
725          * check if it's already passed.  In short, we need a sample.
726          */
727         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
728                 cpu_clock_sample(timer->it_clock, p, &val);
729         } else {
730                 cpu_timer_sample_group(timer->it_clock, p, &val);
731         }
732
733         if (old) {
734                 if (old_expires.sched == 0) {
735                         old->it_value.tv_sec = 0;
736                         old->it_value.tv_nsec = 0;
737                 } else {
738                         /*
739                          * Update the timer in case it has
740                          * overrun already.  If it has,
741                          * we'll report it as having overrun
742                          * and with the next reloaded timer
743                          * already ticking, though we are
744                          * swallowing that pending
745                          * notification here to install the
746                          * new setting.
747                          */
748                         bump_cpu_timer(timer, val);
749                         if (cpu_time_before(timer->it_clock, val,
750                                             timer->it.cpu.expires)) {
751                                 old_expires = cpu_time_sub(
752                                         timer->it_clock,
753                                         timer->it.cpu.expires, val);
754                                 sample_to_timespec(timer->it_clock,
755                                                    old_expires,
756                                                    &old->it_value);
757                         } else {
758                                 old->it_value.tv_nsec = 1;
759                                 old->it_value.tv_sec = 0;
760                         }
761                 }
762         }
763
764         if (unlikely(ret)) {
765                 /*
766                  * We are colliding with the timer actually firing.
767                  * Punt after filling in the timer's old value, and
768                  * disable this firing since we are already reporting
769                  * it as an overrun (thanks to bump_cpu_timer above).
770                  */
771                 spin_unlock(&p->sighand->siglock);
772                 read_unlock(&tasklist_lock);
773                 goto out;
774         }
775
776         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
777                 cpu_time_add(timer->it_clock, &new_expires, val);
778         }
779
780         /*
781          * Install the new expiry time (or zero).
782          * For a timer with no notification action, we don't actually
783          * arm the timer (we'll just fake it for timer_gettime).
784          */
785         timer->it.cpu.expires = new_expires;
786         if (new_expires.sched != 0 &&
787             cpu_time_before(timer->it_clock, val, new_expires)) {
788                 arm_timer(timer);
789         }
790
791         spin_unlock(&p->sighand->siglock);
792         read_unlock(&tasklist_lock);
793
794         /*
795          * Install the new reload setting, and
796          * set up the signal and overrun bookkeeping.
797          */
798         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
799                                                 &new->it_interval);
800
801         /*
802          * This acts as a modification timestamp for the timer,
803          * so any automatic reload attempt will punt on seeing
804          * that we have reset the timer manually.
805          */
806         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
807                 ~REQUEUE_PENDING;
808         timer->it_overrun_last = 0;
809         timer->it_overrun = -1;
810
811         if (new_expires.sched != 0 &&
812             !cpu_time_before(timer->it_clock, val, new_expires)) {
813                 /*
814                  * The designated time already passed, so we notify
815                  * immediately, even if the thread never runs to
816                  * accumulate more time on this clock.
817                  */
818                 cpu_timer_fire(timer);
819         }
820
821         ret = 0;
822  out:
823         if (old) {
824                 sample_to_timespec(timer->it_clock,
825                                    old_incr, &old->it_interval);
826         }
827         return ret;
828 }
829
830 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
831 {
832         union cpu_time_count now;
833         struct task_struct *p = timer->it.cpu.task;
834         int clear_dead;
835
836         /*
837          * Easy part: convert the reload time.
838          */
839         sample_to_timespec(timer->it_clock,
840                            timer->it.cpu.incr, &itp->it_interval);
841
842         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
843                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
844                 return;
845         }
846
847         if (unlikely(p == NULL)) {
848                 /*
849                  * This task already died and the timer will never fire.
850                  * In this case, expires is actually the dead value.
851                  */
852         dead:
853                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
854                                    &itp->it_value);
855                 return;
856         }
857
858         /*
859          * Sample the clock to take the difference with the expiry time.
860          */
861         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
862                 cpu_clock_sample(timer->it_clock, p, &now);
863                 clear_dead = p->exit_state;
864         } else {
865                 read_lock(&tasklist_lock);
866                 if (unlikely(p->signal == NULL)) {
867                         /*
868                          * The process has been reaped.
869                          * We can't even collect a sample any more.
870                          * Call the timer disarmed, nothing else to do.
871                          */
872                         put_task_struct(p);
873                         timer->it.cpu.task = NULL;
874                         timer->it.cpu.expires.sched = 0;
875                         read_unlock(&tasklist_lock);
876                         goto dead;
877                 } else {
878                         cpu_timer_sample_group(timer->it_clock, p, &now);
879                         clear_dead = (unlikely(p->exit_state) &&
880                                       thread_group_empty(p));
881                 }
882                 read_unlock(&tasklist_lock);
883         }
884
885         if (unlikely(clear_dead)) {
886                 /*
887                  * We've noticed that the thread is dead, but
888                  * not yet reaped.  Take this opportunity to
889                  * drop our task ref.
890                  */
891                 clear_dead_task(timer, now);
892                 goto dead;
893         }
894
895         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
896                 sample_to_timespec(timer->it_clock,
897                                    cpu_time_sub(timer->it_clock,
898                                                 timer->it.cpu.expires, now),
899                                    &itp->it_value);
900         } else {
901                 /*
902                  * The timer should have expired already, but the firing
903                  * hasn't taken place yet.  Say it's just about to expire.
904                  */
905                 itp->it_value.tv_nsec = 1;
906                 itp->it_value.tv_sec = 0;
907         }
908 }
909
910 /*
911  * Check for any per-thread CPU timers that have fired and move them off
912  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
913  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
914  */
915 static void check_thread_timers(struct task_struct *tsk,
916                                 struct list_head *firing)
917 {
918         int maxfire;
919         struct list_head *timers = tsk->cpu_timers;
920         struct signal_struct *const sig = tsk->signal;
921         unsigned long soft;
922
923         maxfire = 20;
924         tsk->cputime_expires.prof_exp = cputime_zero;
925         while (!list_empty(timers)) {
926                 struct cpu_timer_list *t = list_first_entry(timers,
927                                                       struct cpu_timer_list,
928                                                       entry);
929                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
930                         tsk->cputime_expires.prof_exp = t->expires.cpu;
931                         break;
932                 }
933                 t->firing = 1;
934                 list_move_tail(&t->entry, firing);
935         }
936
937         ++timers;
938         maxfire = 20;
939         tsk->cputime_expires.virt_exp = cputime_zero;
940         while (!list_empty(timers)) {
941                 struct cpu_timer_list *t = list_first_entry(timers,
942                                                       struct cpu_timer_list,
943                                                       entry);
944                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
945                         tsk->cputime_expires.virt_exp = t->expires.cpu;
946                         break;
947                 }
948                 t->firing = 1;
949                 list_move_tail(&t->entry, firing);
950         }
951
952         ++timers;
953         maxfire = 20;
954         tsk->cputime_expires.sched_exp = 0;
955         while (!list_empty(timers)) {
956                 struct cpu_timer_list *t = list_first_entry(timers,
957                                                       struct cpu_timer_list,
958                                                       entry);
959                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
960                         tsk->cputime_expires.sched_exp = t->expires.sched;
961                         break;
962                 }
963                 t->firing = 1;
964                 list_move_tail(&t->entry, firing);
965         }
966
967         /*
968          * Check for the special case thread timers.
969          */
970         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
971         if (soft != RLIM_INFINITY) {
972                 unsigned long hard =
973                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
974
975                 if (hard != RLIM_INFINITY &&
976                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
977                         /*
978                          * At the hard limit, we just die.
979                          * No need to calculate anything else now.
980                          */
981                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
982                         return;
983                 }
984                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
985                         /*
986                          * At the soft limit, send a SIGXCPU every second.
987                          */
988                         if (soft < hard) {
989                                 soft += USEC_PER_SEC;
990                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
991                         }
992                         printk(KERN_INFO
993                                 "RT Watchdog Timeout: %s[%d]\n",
994                                 tsk->comm, task_pid_nr(tsk));
995                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
996                 }
997         }
998 }
999
1000 static void stop_process_timers(struct signal_struct *sig)
1001 {
1002         struct thread_group_cputimer *cputimer = &sig->cputimer;
1003         unsigned long flags;
1004
1005         if (!cputimer->running)
1006                 return;
1007
1008         spin_lock_irqsave(&cputimer->lock, flags);
1009         cputimer->running = 0;
1010         spin_unlock_irqrestore(&cputimer->lock, flags);
1011
1012         sig->cputime_expires.prof_exp = cputime_zero;
1013         sig->cputime_expires.virt_exp = cputime_zero;
1014         sig->cputime_expires.sched_exp = 0;
1015 }
1016
1017 static u32 onecputick;
1018
1019 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1020                              cputime_t *expires, cputime_t cur_time, int signo)
1021 {
1022         if (cputime_eq(it->expires, cputime_zero))
1023                 return;
1024
1025         if (cputime_ge(cur_time, it->expires)) {
1026                 if (!cputime_eq(it->incr, cputime_zero)) {
1027                         it->expires = cputime_add(it->expires, it->incr);
1028                         it->error += it->incr_error;
1029                         if (it->error >= onecputick) {
1030                                 it->expires = cputime_sub(it->expires,
1031                                                           cputime_one_jiffy);
1032                                 it->error -= onecputick;
1033                         }
1034                 } else {
1035                         it->expires = cputime_zero;
1036                 }
1037
1038                 trace_itimer_expire(signo == SIGPROF ?
1039                                     ITIMER_PROF : ITIMER_VIRTUAL,
1040                                     tsk->signal->leader_pid, cur_time);
1041                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1042         }
1043
1044         if (!cputime_eq(it->expires, cputime_zero) &&
1045             (cputime_eq(*expires, cputime_zero) ||
1046              cputime_lt(it->expires, *expires))) {
1047                 *expires = it->expires;
1048         }
1049 }
1050
1051 /*
1052  * Check for any per-thread CPU timers that have fired and move them
1053  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1054  * have already been taken off.
1055  */
1056 static void check_process_timers(struct task_struct *tsk,
1057                                  struct list_head *firing)
1058 {
1059         int maxfire;
1060         struct signal_struct *const sig = tsk->signal;
1061         cputime_t utime, ptime, virt_expires, prof_expires;
1062         unsigned long long sum_sched_runtime, sched_expires;
1063         struct list_head *timers = sig->cpu_timers;
1064         struct task_cputime cputime;
1065         unsigned long soft;
1066
1067         /*
1068          * Don't sample the current process CPU clocks if there are no timers.
1069          */
1070         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1071             cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) &&
1072             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1073             list_empty(&timers[CPUCLOCK_VIRT]) &&
1074             cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) &&
1075             list_empty(&timers[CPUCLOCK_SCHED])) {
1076                 stop_process_timers(sig);
1077                 return;
1078         }
1079
1080         /*
1081          * Collect the current process totals.
1082          */
1083         thread_group_cputimer(tsk, &cputime);
1084         utime = cputime.utime;
1085         ptime = cputime_add(utime, cputime.stime);
1086         sum_sched_runtime = cputime.sum_exec_runtime;
1087         maxfire = 20;
1088         prof_expires = cputime_zero;
1089         while (!list_empty(timers)) {
1090                 struct cpu_timer_list *tl = list_first_entry(timers,
1091                                                       struct cpu_timer_list,
1092                                                       entry);
1093                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1094                         prof_expires = tl->expires.cpu;
1095                         break;
1096                 }
1097                 tl->firing = 1;
1098                 list_move_tail(&tl->entry, firing);
1099         }
1100
1101         ++timers;
1102         maxfire = 20;
1103         virt_expires = cputime_zero;
1104         while (!list_empty(timers)) {
1105                 struct cpu_timer_list *tl = list_first_entry(timers,
1106                                                       struct cpu_timer_list,
1107                                                       entry);
1108                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1109                         virt_expires = tl->expires.cpu;
1110                         break;
1111                 }
1112                 tl->firing = 1;
1113                 list_move_tail(&tl->entry, firing);
1114         }
1115
1116         ++timers;
1117         maxfire = 20;
1118         sched_expires = 0;
1119         while (!list_empty(timers)) {
1120                 struct cpu_timer_list *tl = list_first_entry(timers,
1121                                                       struct cpu_timer_list,
1122                                                       entry);
1123                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1124                         sched_expires = tl->expires.sched;
1125                         break;
1126                 }
1127                 tl->firing = 1;
1128                 list_move_tail(&tl->entry, firing);
1129         }
1130
1131         /*
1132          * Check for the special case process timers.
1133          */
1134         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1135                          SIGPROF);
1136         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1137                          SIGVTALRM);
1138         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1139         if (soft != RLIM_INFINITY) {
1140                 unsigned long psecs = cputime_to_secs(ptime);
1141                 unsigned long hard =
1142                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1143                 cputime_t x;
1144                 if (psecs >= hard) {
1145                         /*
1146                          * At the hard limit, we just die.
1147                          * No need to calculate anything else now.
1148                          */
1149                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1150                         return;
1151                 }
1152                 if (psecs >= soft) {
1153                         /*
1154                          * At the soft limit, send a SIGXCPU every second.
1155                          */
1156                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1157                         if (soft < hard) {
1158                                 soft++;
1159                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1160                         }
1161                 }
1162                 x = secs_to_cputime(soft);
1163                 if (cputime_eq(prof_expires, cputime_zero) ||
1164                     cputime_lt(x, prof_expires)) {
1165                         prof_expires = x;
1166                 }
1167         }
1168
1169         if (!cputime_eq(prof_expires, cputime_zero) &&
1170             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1171              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1172                 sig->cputime_expires.prof_exp = prof_expires;
1173         if (!cputime_eq(virt_expires, cputime_zero) &&
1174             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1175              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1176                 sig->cputime_expires.virt_exp = virt_expires;
1177         if (sched_expires != 0 &&
1178             (sig->cputime_expires.sched_exp == 0 ||
1179              sig->cputime_expires.sched_exp > sched_expires))
1180                 sig->cputime_expires.sched_exp = sched_expires;
1181 }
1182
1183 /*
1184  * This is called from the signal code (via do_schedule_next_timer)
1185  * when the last timer signal was delivered and we have to reload the timer.
1186  */
1187 void posix_cpu_timer_schedule(struct k_itimer *timer)
1188 {
1189         struct task_struct *p = timer->it.cpu.task;
1190         union cpu_time_count now;
1191
1192         if (unlikely(p == NULL))
1193                 /*
1194                  * The task was cleaned up already, no future firings.
1195                  */
1196                 goto out;
1197
1198         /*
1199          * Fetch the current sample and update the timer's expiry time.
1200          */
1201         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1202                 cpu_clock_sample(timer->it_clock, p, &now);
1203                 bump_cpu_timer(timer, now);
1204                 if (unlikely(p->exit_state)) {
1205                         clear_dead_task(timer, now);
1206                         goto out;
1207                 }
1208                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1209                 spin_lock(&p->sighand->siglock);
1210         } else {
1211                 read_lock(&tasklist_lock);
1212                 if (unlikely(p->signal == NULL)) {
1213                         /*
1214                          * The process has been reaped.
1215                          * We can't even collect a sample any more.
1216                          */
1217                         put_task_struct(p);
1218                         timer->it.cpu.task = p = NULL;
1219                         timer->it.cpu.expires.sched = 0;
1220                         goto out_unlock;
1221                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1222                         /*
1223                          * We've noticed that the thread is dead, but
1224                          * not yet reaped.  Take this opportunity to
1225                          * drop our task ref.
1226                          */
1227                         clear_dead_task(timer, now);
1228                         goto out_unlock;
1229                 }
1230                 spin_lock(&p->sighand->siglock);
1231                 cpu_timer_sample_group(timer->it_clock, p, &now);
1232                 bump_cpu_timer(timer, now);
1233                 /* Leave the tasklist_lock locked for the call below.  */
1234         }
1235
1236         /*
1237          * Now re-arm for the new expiry time.
1238          */
1239         BUG_ON(!irqs_disabled());
1240         arm_timer(timer);
1241         spin_unlock(&p->sighand->siglock);
1242
1243 out_unlock:
1244         read_unlock(&tasklist_lock);
1245
1246 out:
1247         timer->it_overrun_last = timer->it_overrun;
1248         timer->it_overrun = -1;
1249         ++timer->it_requeue_pending;
1250 }
1251
1252 /**
1253  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1254  *
1255  * @cputime:    The struct to compare.
1256  *
1257  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1258  * are zero, false if any field is nonzero.
1259  */
1260 static inline int task_cputime_zero(const struct task_cputime *cputime)
1261 {
1262         if (cputime_eq(cputime->utime, cputime_zero) &&
1263             cputime_eq(cputime->stime, cputime_zero) &&
1264             cputime->sum_exec_runtime == 0)
1265                 return 1;
1266         return 0;
1267 }
1268
1269 /**
1270  * task_cputime_expired - Compare two task_cputime entities.
1271  *
1272  * @sample:     The task_cputime structure to be checked for expiration.
1273  * @expires:    Expiration times, against which @sample will be checked.
1274  *
1275  * Checks @sample against @expires to see if any field of @sample has expired.
1276  * Returns true if any field of the former is greater than the corresponding
1277  * field of the latter if the latter field is set.  Otherwise returns false.
1278  */
1279 static inline int task_cputime_expired(const struct task_cputime *sample,
1280                                         const struct task_cputime *expires)
1281 {
1282         if (!cputime_eq(expires->utime, cputime_zero) &&
1283             cputime_ge(sample->utime, expires->utime))
1284                 return 1;
1285         if (!cputime_eq(expires->stime, cputime_zero) &&
1286             cputime_ge(cputime_add(sample->utime, sample->stime),
1287                        expires->stime))
1288                 return 1;
1289         if (expires->sum_exec_runtime != 0 &&
1290             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1291                 return 1;
1292         return 0;
1293 }
1294
1295 /**
1296  * fastpath_timer_check - POSIX CPU timers fast path.
1297  *
1298  * @tsk:        The task (thread) being checked.
1299  *
1300  * Check the task and thread group timers.  If both are zero (there are no
1301  * timers set) return false.  Otherwise snapshot the task and thread group
1302  * timers and compare them with the corresponding expiration times.  Return
1303  * true if a timer has expired, else return false.
1304  */
1305 static inline int fastpath_timer_check(struct task_struct *tsk)
1306 {
1307         struct signal_struct *sig;
1308
1309         /* tsk == current, ensure it is safe to use ->signal/sighand */
1310         if (unlikely(tsk->exit_state))
1311                 return 0;
1312
1313         if (!task_cputime_zero(&tsk->cputime_expires)) {
1314                 struct task_cputime task_sample = {
1315                         .utime = tsk->utime,
1316                         .stime = tsk->stime,
1317                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1318                 };
1319
1320                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1321                         return 1;
1322         }
1323
1324         sig = tsk->signal;
1325         if (!task_cputime_zero(&sig->cputime_expires)) {
1326                 struct task_cputime group_sample;
1327
1328                 thread_group_cputimer(tsk, &group_sample);
1329                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1330                         return 1;
1331         }
1332
1333         return 0;
1334 }
1335
1336 /*
1337  * This is called from the timer interrupt handler.  The irq handler has
1338  * already updated our counts.  We need to check if any timers fire now.
1339  * Interrupts are disabled.
1340  */
1341 void run_posix_cpu_timers(struct task_struct *tsk)
1342 {
1343         LIST_HEAD(firing);
1344         struct k_itimer *timer, *next;
1345
1346         BUG_ON(!irqs_disabled());
1347
1348         /*
1349          * The fast path checks that there are no expired thread or thread
1350          * group timers.  If that's so, just return.
1351          */
1352         if (!fastpath_timer_check(tsk))
1353                 return;
1354
1355         spin_lock(&tsk->sighand->siglock);
1356         /*
1357          * Here we take off tsk->signal->cpu_timers[N] and
1358          * tsk->cpu_timers[N] all the timers that are firing, and
1359          * put them on the firing list.
1360          */
1361         check_thread_timers(tsk, &firing);
1362         check_process_timers(tsk, &firing);
1363
1364         /*
1365          * We must release these locks before taking any timer's lock.
1366          * There is a potential race with timer deletion here, as the
1367          * siglock now protects our private firing list.  We have set
1368          * the firing flag in each timer, so that a deletion attempt
1369          * that gets the timer lock before we do will give it up and
1370          * spin until we've taken care of that timer below.
1371          */
1372         spin_unlock(&tsk->sighand->siglock);
1373
1374         /*
1375          * Now that all the timers on our list have the firing flag,
1376          * noone will touch their list entries but us.  We'll take
1377          * each timer's lock before clearing its firing flag, so no
1378          * timer call will interfere.
1379          */
1380         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1381                 int cpu_firing;
1382
1383                 spin_lock(&timer->it_lock);
1384                 list_del_init(&timer->it.cpu.entry);
1385                 cpu_firing = timer->it.cpu.firing;
1386                 timer->it.cpu.firing = 0;
1387                 /*
1388                  * The firing flag is -1 if we collided with a reset
1389                  * of the timer, which already reported this
1390                  * almost-firing as an overrun.  So don't generate an event.
1391                  */
1392                 if (likely(cpu_firing >= 0))
1393                         cpu_timer_fire(timer);
1394                 spin_unlock(&timer->it_lock);
1395         }
1396 }
1397
1398 /*
1399  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1400  * The tsk->sighand->siglock must be held by the caller.
1401  */
1402 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1403                            cputime_t *newval, cputime_t *oldval)
1404 {
1405         union cpu_time_count now;
1406
1407         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1408         cpu_timer_sample_group(clock_idx, tsk, &now);
1409
1410         if (oldval) {
1411                 /*
1412                  * We are setting itimer. The *oldval is absolute and we update
1413                  * it to be relative, *newval argument is relative and we update
1414                  * it to be absolute.
1415                  */
1416                 if (!cputime_eq(*oldval, cputime_zero)) {
1417                         if (cputime_le(*oldval, now.cpu)) {
1418                                 /* Just about to fire. */
1419                                 *oldval = cputime_one_jiffy;
1420                         } else {
1421                                 *oldval = cputime_sub(*oldval, now.cpu);
1422                         }
1423                 }
1424
1425                 if (cputime_eq(*newval, cputime_zero))
1426                         return;
1427                 *newval = cputime_add(*newval, now.cpu);
1428         }
1429
1430         /*
1431          * Update expiration cache if we are the earliest timer, or eventually
1432          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1433          */
1434         switch (clock_idx) {
1435         case CPUCLOCK_PROF:
1436                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1437                         tsk->signal->cputime_expires.prof_exp = *newval;
1438                 break;
1439         case CPUCLOCK_VIRT:
1440                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1441                         tsk->signal->cputime_expires.virt_exp = *newval;
1442                 break;
1443         }
1444 }
1445
1446 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1447                             struct timespec *rqtp, struct itimerspec *it)
1448 {
1449         struct k_itimer timer;
1450         int error;
1451
1452         /*
1453          * Set up a temporary timer and then wait for it to go off.
1454          */
1455         memset(&timer, 0, sizeof timer);
1456         spin_lock_init(&timer.it_lock);
1457         timer.it_clock = which_clock;
1458         timer.it_overrun = -1;
1459         error = posix_cpu_timer_create(&timer);
1460         timer.it_process = current;
1461         if (!error) {
1462                 static struct itimerspec zero_it;
1463
1464                 memset(it, 0, sizeof *it);
1465                 it->it_value = *rqtp;
1466
1467                 spin_lock_irq(&timer.it_lock);
1468                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1469                 if (error) {
1470                         spin_unlock_irq(&timer.it_lock);
1471                         return error;
1472                 }
1473
1474                 while (!signal_pending(current)) {
1475                         if (timer.it.cpu.expires.sched == 0) {
1476                                 /*
1477                                  * Our timer fired and was reset.
1478                                  */
1479                                 spin_unlock_irq(&timer.it_lock);
1480                                 return 0;
1481                         }
1482
1483                         /*
1484                          * Block until cpu_timer_fire (or a signal) wakes us.
1485                          */
1486                         __set_current_state(TASK_INTERRUPTIBLE);
1487                         spin_unlock_irq(&timer.it_lock);
1488                         schedule();
1489                         spin_lock_irq(&timer.it_lock);
1490                 }
1491
1492                 /*
1493                  * We were interrupted by a signal.
1494                  */
1495                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1496                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1497                 spin_unlock_irq(&timer.it_lock);
1498
1499                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1500                         /*
1501                          * It actually did fire already.
1502                          */
1503                         return 0;
1504                 }
1505
1506                 error = -ERESTART_RESTARTBLOCK;
1507         }
1508
1509         return error;
1510 }
1511
1512 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1513                      struct timespec *rqtp, struct timespec __user *rmtp)
1514 {
1515         struct restart_block *restart_block =
1516             &current_thread_info()->restart_block;
1517         struct itimerspec it;
1518         int error;
1519
1520         /*
1521          * Diagnose required errors first.
1522          */
1523         if (CPUCLOCK_PERTHREAD(which_clock) &&
1524             (CPUCLOCK_PID(which_clock) == 0 ||
1525              CPUCLOCK_PID(which_clock) == current->pid))
1526                 return -EINVAL;
1527
1528         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1529
1530         if (error == -ERESTART_RESTARTBLOCK) {
1531
1532                 if (flags & TIMER_ABSTIME)
1533                         return -ERESTARTNOHAND;
1534                 /*
1535                  * Report back to the user the time still remaining.
1536                  */
1537                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1538                         return -EFAULT;
1539
1540                 restart_block->fn = posix_cpu_nsleep_restart;
1541                 restart_block->arg0 = which_clock;
1542                 restart_block->arg1 = (unsigned long) rmtp;
1543                 restart_block->arg2 = rqtp->tv_sec;
1544                 restart_block->arg3 = rqtp->tv_nsec;
1545         }
1546         return error;
1547 }
1548
1549 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1550 {
1551         clockid_t which_clock = restart_block->arg0;
1552         struct timespec __user *rmtp;
1553         struct timespec t;
1554         struct itimerspec it;
1555         int error;
1556
1557         rmtp = (struct timespec __user *) restart_block->arg1;
1558         t.tv_sec = restart_block->arg2;
1559         t.tv_nsec = restart_block->arg3;
1560
1561         restart_block->fn = do_no_restart_syscall;
1562         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1563
1564         if (error == -ERESTART_RESTARTBLOCK) {
1565                 /*
1566                  * Report back to the user the time still remaining.
1567                  */
1568                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1569                         return -EFAULT;
1570
1571                 restart_block->fn = posix_cpu_nsleep_restart;
1572                 restart_block->arg0 = which_clock;
1573                 restart_block->arg1 = (unsigned long) rmtp;
1574                 restart_block->arg2 = t.tv_sec;
1575                 restart_block->arg3 = t.tv_nsec;
1576         }
1577         return error;
1578
1579 }
1580
1581
1582 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1583 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1584
1585 static int process_cpu_clock_getres(const clockid_t which_clock,
1586                                     struct timespec *tp)
1587 {
1588         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1589 }
1590 static int process_cpu_clock_get(const clockid_t which_clock,
1591                                  struct timespec *tp)
1592 {
1593         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1594 }
1595 static int process_cpu_timer_create(struct k_itimer *timer)
1596 {
1597         timer->it_clock = PROCESS_CLOCK;
1598         return posix_cpu_timer_create(timer);
1599 }
1600 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1601                               struct timespec *rqtp,
1602                               struct timespec __user *rmtp)
1603 {
1604         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1605 }
1606 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1607 {
1608         return -EINVAL;
1609 }
1610 static int thread_cpu_clock_getres(const clockid_t which_clock,
1611                                    struct timespec *tp)
1612 {
1613         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1614 }
1615 static int thread_cpu_clock_get(const clockid_t which_clock,
1616                                 struct timespec *tp)
1617 {
1618         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1619 }
1620 static int thread_cpu_timer_create(struct k_itimer *timer)
1621 {
1622         timer->it_clock = THREAD_CLOCK;
1623         return posix_cpu_timer_create(timer);
1624 }
1625 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1626                               struct timespec *rqtp, struct timespec __user *rmtp)
1627 {
1628         return -EINVAL;
1629 }
1630 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1631 {
1632         return -EINVAL;
1633 }
1634
1635 static __init int init_posix_cpu_timers(void)
1636 {
1637         struct k_clock process = {
1638                 .clock_getres = process_cpu_clock_getres,
1639                 .clock_get = process_cpu_clock_get,
1640                 .clock_set = do_posix_clock_nosettime,
1641                 .timer_create = process_cpu_timer_create,
1642                 .nsleep = process_cpu_nsleep,
1643                 .nsleep_restart = process_cpu_nsleep_restart,
1644         };
1645         struct k_clock thread = {
1646                 .clock_getres = thread_cpu_clock_getres,
1647                 .clock_get = thread_cpu_clock_get,
1648                 .clock_set = do_posix_clock_nosettime,
1649                 .timer_create = thread_cpu_timer_create,
1650                 .nsleep = thread_cpu_nsleep,
1651                 .nsleep_restart = thread_cpu_nsleep_restart,
1652         };
1653         struct timespec ts;
1654
1655         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1656         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1657
1658         cputime_to_timespec(cputime_one_jiffy, &ts);
1659         onecputick = ts.tv_nsec;
1660         WARN_ON(ts.tv_sec != 0);
1661
1662         return 0;
1663 }
1664 __initcall(init_posix_cpu_timers);