timers: split process wide cpu clocks/timers
[linux-2.6.git] / kernel / posix-cpu-timers.c
1 /*
2  * Implement CPU time clocks for the POSIX clock interface.
3  */
4
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <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 /*
265  * Sample a process (thread group) clock for the given group_leader task.
266  * Must be called with tasklist_lock held for reading.
267  */
268 static int cpu_clock_sample_group(const clockid_t which_clock,
269                                   struct task_struct *p,
270                                   union cpu_time_count *cpu)
271 {
272         struct task_cputime cputime;
273
274         thread_group_cputime(p, &cputime);
275         switch (CPUCLOCK_WHICH(which_clock)) {
276         default:
277                 return -EINVAL;
278         case CPUCLOCK_PROF:
279                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
280                 break;
281         case CPUCLOCK_VIRT:
282                 cpu->cpu = cputime.utime;
283                 break;
284         case CPUCLOCK_SCHED:
285                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
286                 break;
287         }
288         return 0;
289 }
290
291
292 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
293 {
294         const pid_t pid = CPUCLOCK_PID(which_clock);
295         int error = -EINVAL;
296         union cpu_time_count rtn;
297
298         if (pid == 0) {
299                 /*
300                  * Special case constant value for our own clocks.
301                  * We don't have to do any lookup to find ourselves.
302                  */
303                 if (CPUCLOCK_PERTHREAD(which_clock)) {
304                         /*
305                          * Sampling just ourselves we can do with no locking.
306                          */
307                         error = cpu_clock_sample(which_clock,
308                                                  current, &rtn);
309                 } else {
310                         read_lock(&tasklist_lock);
311                         error = cpu_clock_sample_group(which_clock,
312                                                        current, &rtn);
313                         read_unlock(&tasklist_lock);
314                 }
315         } else {
316                 /*
317                  * Find the given PID, and validate that the caller
318                  * should be able to see it.
319                  */
320                 struct task_struct *p;
321                 rcu_read_lock();
322                 p = find_task_by_vpid(pid);
323                 if (p) {
324                         if (CPUCLOCK_PERTHREAD(which_clock)) {
325                                 if (same_thread_group(p, current)) {
326                                         error = cpu_clock_sample(which_clock,
327                                                                  p, &rtn);
328                                 }
329                         } else {
330                                 read_lock(&tasklist_lock);
331                                 if (thread_group_leader(p) && p->signal) {
332                                         error =
333                                             cpu_clock_sample_group(which_clock,
334                                                                    p, &rtn);
335                                 }
336                                 read_unlock(&tasklist_lock);
337                         }
338                 }
339                 rcu_read_unlock();
340         }
341
342         if (error)
343                 return error;
344         sample_to_timespec(which_clock, rtn, tp);
345         return 0;
346 }
347
348
349 /*
350  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
351  * This is called from sys_timer_create with the new timer already locked.
352  */
353 int posix_cpu_timer_create(struct k_itimer *new_timer)
354 {
355         int ret = 0;
356         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
357         struct task_struct *p;
358
359         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
360                 return -EINVAL;
361
362         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
363         new_timer->it.cpu.incr.sched = 0;
364         new_timer->it.cpu.expires.sched = 0;
365
366         read_lock(&tasklist_lock);
367         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
368                 if (pid == 0) {
369                         p = current;
370                 } else {
371                         p = find_task_by_vpid(pid);
372                         if (p && !same_thread_group(p, current))
373                                 p = NULL;
374                 }
375         } else {
376                 if (pid == 0) {
377                         p = current->group_leader;
378                 } else {
379                         p = find_task_by_vpid(pid);
380                         if (p && !thread_group_leader(p))
381                                 p = NULL;
382                 }
383         }
384         new_timer->it.cpu.task = p;
385         if (p) {
386                 get_task_struct(p);
387         } else {
388                 ret = -EINVAL;
389         }
390         read_unlock(&tasklist_lock);
391
392         return ret;
393 }
394
395 /*
396  * Clean up a CPU-clock timer that is about to be destroyed.
397  * This is called from timer deletion with the timer already locked.
398  * If we return TIMER_RETRY, it's necessary to release the timer's lock
399  * and try again.  (This happens when the timer is in the middle of firing.)
400  */
401 int posix_cpu_timer_del(struct k_itimer *timer)
402 {
403         struct task_struct *p = timer->it.cpu.task;
404         int ret = 0;
405
406         if (likely(p != NULL)) {
407                 read_lock(&tasklist_lock);
408                 if (unlikely(p->signal == NULL)) {
409                         /*
410                          * We raced with the reaping of the task.
411                          * The deletion should have cleared us off the list.
412                          */
413                         BUG_ON(!list_empty(&timer->it.cpu.entry));
414                 } else {
415                         spin_lock(&p->sighand->siglock);
416                         if (timer->it.cpu.firing)
417                                 ret = TIMER_RETRY;
418                         else
419                                 list_del(&timer->it.cpu.entry);
420                         spin_unlock(&p->sighand->siglock);
421                 }
422                 read_unlock(&tasklist_lock);
423
424                 if (!ret)
425                         put_task_struct(p);
426         }
427
428         return ret;
429 }
430
431 /*
432  * Clean out CPU timers still ticking when a thread exited.  The task
433  * pointer is cleared, and the expiry time is replaced with the residual
434  * time for later timer_gettime calls to return.
435  * This must be called with the siglock held.
436  */
437 static void cleanup_timers(struct list_head *head,
438                            cputime_t utime, cputime_t stime,
439                            unsigned long long sum_exec_runtime)
440 {
441         struct cpu_timer_list *timer, *next;
442         cputime_t ptime = cputime_add(utime, stime);
443
444         list_for_each_entry_safe(timer, next, head, entry) {
445                 list_del_init(&timer->entry);
446                 if (cputime_lt(timer->expires.cpu, ptime)) {
447                         timer->expires.cpu = cputime_zero;
448                 } else {
449                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
450                                                          ptime);
451                 }
452         }
453
454         ++head;
455         list_for_each_entry_safe(timer, next, head, entry) {
456                 list_del_init(&timer->entry);
457                 if (cputime_lt(timer->expires.cpu, utime)) {
458                         timer->expires.cpu = cputime_zero;
459                 } else {
460                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
461                                                          utime);
462                 }
463         }
464
465         ++head;
466         list_for_each_entry_safe(timer, next, head, entry) {
467                 list_del_init(&timer->entry);
468                 if (timer->expires.sched < sum_exec_runtime) {
469                         timer->expires.sched = 0;
470                 } else {
471                         timer->expires.sched -= sum_exec_runtime;
472                 }
473         }
474 }
475
476 /*
477  * These are both called with the siglock held, when the current thread
478  * is being reaped.  When the final (leader) thread in the group is reaped,
479  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
480  */
481 void posix_cpu_timers_exit(struct task_struct *tsk)
482 {
483         cleanup_timers(tsk->cpu_timers,
484                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
485
486 }
487 void posix_cpu_timers_exit_group(struct task_struct *tsk)
488 {
489         struct task_cputime cputime;
490
491         thread_group_cputime(tsk, &cputime);
492         cleanup_timers(tsk->signal->cpu_timers,
493                        cputime.utime, cputime.stime, cputime.sum_exec_runtime);
494 }
495
496 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
497 {
498         /*
499          * That's all for this thread or process.
500          * We leave our residual in expires to be reported.
501          */
502         put_task_struct(timer->it.cpu.task);
503         timer->it.cpu.task = NULL;
504         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
505                                              timer->it.cpu.expires,
506                                              now);
507 }
508
509 /*
510  * Enable the process wide cpu timer accounting.
511  *
512  * serialized using ->sighand->siglock
513  */
514 static void start_process_timers(struct task_struct *tsk)
515 {
516         tsk->signal->cputimer.running = 1;
517         barrier();
518 }
519
520 /*
521  * Release the process wide timer accounting -- timer stops ticking when
522  * nobody cares about it.
523  *
524  * serialized using ->sighand->siglock
525  */
526 static void stop_process_timers(struct task_struct *tsk)
527 {
528         tsk->signal->cputimer.running = 0;
529         barrier();
530 }
531
532 /*
533  * Insert the timer on the appropriate list before any timers that
534  * expire later.  This must be called with the tasklist_lock held
535  * for reading, and interrupts disabled.
536  */
537 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
538 {
539         struct task_struct *p = timer->it.cpu.task;
540         struct list_head *head, *listpos;
541         struct cpu_timer_list *const nt = &timer->it.cpu;
542         struct cpu_timer_list *next;
543         unsigned long i;
544
545         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
546                 p->cpu_timers : p->signal->cpu_timers);
547         head += CPUCLOCK_WHICH(timer->it_clock);
548
549         BUG_ON(!irqs_disabled());
550         spin_lock(&p->sighand->siglock);
551
552         if (!CPUCLOCK_PERTHREAD(timer->it_clock))
553                 start_process_timers(p);
554
555         listpos = head;
556         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
557                 list_for_each_entry(next, head, entry) {
558                         if (next->expires.sched > nt->expires.sched)
559                                 break;
560                         listpos = &next->entry;
561                 }
562         } else {
563                 list_for_each_entry(next, head, entry) {
564                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
565                                 break;
566                         listpos = &next->entry;
567                 }
568         }
569         list_add(&nt->entry, listpos);
570
571         if (listpos == head) {
572                 /*
573                  * We are the new earliest-expiring timer.
574                  * If we are a thread timer, there can always
575                  * be a process timer telling us to stop earlier.
576                  */
577
578                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
579                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
580                         default:
581                                 BUG();
582                         case CPUCLOCK_PROF:
583                                 if (cputime_eq(p->cputime_expires.prof_exp,
584                                                cputime_zero) ||
585                                     cputime_gt(p->cputime_expires.prof_exp,
586                                                nt->expires.cpu))
587                                         p->cputime_expires.prof_exp =
588                                                 nt->expires.cpu;
589                                 break;
590                         case CPUCLOCK_VIRT:
591                                 if (cputime_eq(p->cputime_expires.virt_exp,
592                                                cputime_zero) ||
593                                     cputime_gt(p->cputime_expires.virt_exp,
594                                                nt->expires.cpu))
595                                         p->cputime_expires.virt_exp =
596                                                 nt->expires.cpu;
597                                 break;
598                         case CPUCLOCK_SCHED:
599                                 if (p->cputime_expires.sched_exp == 0 ||
600                                     p->cputime_expires.sched_exp >
601                                                         nt->expires.sched)
602                                         p->cputime_expires.sched_exp =
603                                                 nt->expires.sched;
604                                 break;
605                         }
606                 } else {
607                         /*
608                          * For a process timer, set the cached expiration time.
609                          */
610                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
611                         default:
612                                 BUG();
613                         case CPUCLOCK_VIRT:
614                                 if (!cputime_eq(p->signal->it_virt_expires,
615                                                 cputime_zero) &&
616                                     cputime_lt(p->signal->it_virt_expires,
617                                                timer->it.cpu.expires.cpu))
618                                         break;
619                                 p->signal->cputime_expires.virt_exp =
620                                         timer->it.cpu.expires.cpu;
621                                 break;
622                         case CPUCLOCK_PROF:
623                                 if (!cputime_eq(p->signal->it_prof_expires,
624                                                 cputime_zero) &&
625                                     cputime_lt(p->signal->it_prof_expires,
626                                                timer->it.cpu.expires.cpu))
627                                         break;
628                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
629                                 if (i != RLIM_INFINITY &&
630                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
631                                         break;
632                                 p->signal->cputime_expires.prof_exp =
633                                         timer->it.cpu.expires.cpu;
634                                 break;
635                         case CPUCLOCK_SCHED:
636                                 p->signal->cputime_expires.sched_exp =
637                                         timer->it.cpu.expires.sched;
638                                 break;
639                         }
640                 }
641         }
642
643         spin_unlock(&p->sighand->siglock);
644 }
645
646 /*
647  * The timer is locked, fire it and arrange for its reload.
648  */
649 static void cpu_timer_fire(struct k_itimer *timer)
650 {
651         if (unlikely(timer->sigq == NULL)) {
652                 /*
653                  * This a special case for clock_nanosleep,
654                  * not a normal timer from sys_timer_create.
655                  */
656                 wake_up_process(timer->it_process);
657                 timer->it.cpu.expires.sched = 0;
658         } else if (timer->it.cpu.incr.sched == 0) {
659                 /*
660                  * One-shot timer.  Clear it as soon as it's fired.
661                  */
662                 posix_timer_event(timer, 0);
663                 timer->it.cpu.expires.sched = 0;
664         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
665                 /*
666                  * The signal did not get queued because the signal
667                  * was ignored, so we won't get any callback to
668                  * reload the timer.  But we need to keep it
669                  * ticking in case the signal is deliverable next time.
670                  */
671                 posix_cpu_timer_schedule(timer);
672         }
673 }
674
675 /*
676  * Guts of sys_timer_settime for CPU timers.
677  * This is called with the timer locked and interrupts disabled.
678  * If we return TIMER_RETRY, it's necessary to release the timer's lock
679  * and try again.  (This happens when the timer is in the middle of firing.)
680  */
681 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
682                         struct itimerspec *new, struct itimerspec *old)
683 {
684         struct task_struct *p = timer->it.cpu.task;
685         union cpu_time_count old_expires, new_expires, val;
686         int ret;
687
688         if (unlikely(p == NULL)) {
689                 /*
690                  * Timer refers to a dead task's clock.
691                  */
692                 return -ESRCH;
693         }
694
695         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
696
697         read_lock(&tasklist_lock);
698         /*
699          * We need the tasklist_lock to protect against reaping that
700          * clears p->signal.  If p has just been reaped, we can no
701          * longer get any information about it at all.
702          */
703         if (unlikely(p->signal == NULL)) {
704                 read_unlock(&tasklist_lock);
705                 put_task_struct(p);
706                 timer->it.cpu.task = NULL;
707                 return -ESRCH;
708         }
709
710         /*
711          * Disarm any old timer after extracting its expiry time.
712          */
713         BUG_ON(!irqs_disabled());
714
715         ret = 0;
716         spin_lock(&p->sighand->siglock);
717         old_expires = timer->it.cpu.expires;
718         if (unlikely(timer->it.cpu.firing)) {
719                 timer->it.cpu.firing = -1;
720                 ret = TIMER_RETRY;
721         } else
722                 list_del_init(&timer->it.cpu.entry);
723         spin_unlock(&p->sighand->siglock);
724
725         /*
726          * We need to sample the current value to convert the new
727          * value from to relative and absolute, and to convert the
728          * old value from absolute to relative.  To set a process
729          * timer, we need a sample to balance the thread expiry
730          * times (in arm_timer).  With an absolute time, we must
731          * check if it's already passed.  In short, we need a sample.
732          */
733         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
734                 cpu_clock_sample(timer->it_clock, p, &val);
735         } else {
736                 cpu_clock_sample_group(timer->it_clock, p, &val);
737         }
738
739         if (old) {
740                 if (old_expires.sched == 0) {
741                         old->it_value.tv_sec = 0;
742                         old->it_value.tv_nsec = 0;
743                 } else {
744                         /*
745                          * Update the timer in case it has
746                          * overrun already.  If it has,
747                          * we'll report it as having overrun
748                          * and with the next reloaded timer
749                          * already ticking, though we are
750                          * swallowing that pending
751                          * notification here to install the
752                          * new setting.
753                          */
754                         bump_cpu_timer(timer, val);
755                         if (cpu_time_before(timer->it_clock, val,
756                                             timer->it.cpu.expires)) {
757                                 old_expires = cpu_time_sub(
758                                         timer->it_clock,
759                                         timer->it.cpu.expires, val);
760                                 sample_to_timespec(timer->it_clock,
761                                                    old_expires,
762                                                    &old->it_value);
763                         } else {
764                                 old->it_value.tv_nsec = 1;
765                                 old->it_value.tv_sec = 0;
766                         }
767                 }
768         }
769
770         if (unlikely(ret)) {
771                 /*
772                  * We are colliding with the timer actually firing.
773                  * Punt after filling in the timer's old value, and
774                  * disable this firing since we are already reporting
775                  * it as an overrun (thanks to bump_cpu_timer above).
776                  */
777                 read_unlock(&tasklist_lock);
778                 goto out;
779         }
780
781         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
782                 cpu_time_add(timer->it_clock, &new_expires, val);
783         }
784
785         /*
786          * Install the new expiry time (or zero).
787          * For a timer with no notification action, we don't actually
788          * arm the timer (we'll just fake it for timer_gettime).
789          */
790         timer->it.cpu.expires = new_expires;
791         if (new_expires.sched != 0 &&
792             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
793             cpu_time_before(timer->it_clock, val, new_expires)) {
794                 arm_timer(timer, val);
795         }
796
797         read_unlock(&tasklist_lock);
798
799         /*
800          * Install the new reload setting, and
801          * set up the signal and overrun bookkeeping.
802          */
803         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
804                                                 &new->it_interval);
805
806         /*
807          * This acts as a modification timestamp for the timer,
808          * so any automatic reload attempt will punt on seeing
809          * that we have reset the timer manually.
810          */
811         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
812                 ~REQUEUE_PENDING;
813         timer->it_overrun_last = 0;
814         timer->it_overrun = -1;
815
816         if (new_expires.sched != 0 &&
817             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
818             !cpu_time_before(timer->it_clock, val, new_expires)) {
819                 /*
820                  * The designated time already passed, so we notify
821                  * immediately, even if the thread never runs to
822                  * accumulate more time on this clock.
823                  */
824                 cpu_timer_fire(timer);
825         }
826
827         ret = 0;
828  out:
829         if (old) {
830                 sample_to_timespec(timer->it_clock,
831                                    timer->it.cpu.incr, &old->it_interval);
832         }
833         return ret;
834 }
835
836 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
837 {
838         union cpu_time_count now;
839         struct task_struct *p = timer->it.cpu.task;
840         int clear_dead;
841
842         /*
843          * Easy part: convert the reload time.
844          */
845         sample_to_timespec(timer->it_clock,
846                            timer->it.cpu.incr, &itp->it_interval);
847
848         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
849                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
850                 return;
851         }
852
853         if (unlikely(p == NULL)) {
854                 /*
855                  * This task already died and the timer will never fire.
856                  * In this case, expires is actually the dead value.
857                  */
858         dead:
859                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
860                                    &itp->it_value);
861                 return;
862         }
863
864         /*
865          * Sample the clock to take the difference with the expiry time.
866          */
867         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
868                 cpu_clock_sample(timer->it_clock, p, &now);
869                 clear_dead = p->exit_state;
870         } else {
871                 read_lock(&tasklist_lock);
872                 if (unlikely(p->signal == NULL)) {
873                         /*
874                          * The process has been reaped.
875                          * We can't even collect a sample any more.
876                          * Call the timer disarmed, nothing else to do.
877                          */
878                         put_task_struct(p);
879                         timer->it.cpu.task = NULL;
880                         timer->it.cpu.expires.sched = 0;
881                         read_unlock(&tasklist_lock);
882                         goto dead;
883                 } else {
884                         cpu_clock_sample_group(timer->it_clock, p, &now);
885                         clear_dead = (unlikely(p->exit_state) &&
886                                       thread_group_empty(p));
887                 }
888                 read_unlock(&tasklist_lock);
889         }
890
891         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
892                 if (timer->it.cpu.incr.sched == 0 &&
893                     cpu_time_before(timer->it_clock,
894                                     timer->it.cpu.expires, now)) {
895                         /*
896                          * Do-nothing timer expired and has no reload,
897                          * so it's as if it was never set.
898                          */
899                         timer->it.cpu.expires.sched = 0;
900                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
901                         return;
902                 }
903                 /*
904                  * Account for any expirations and reloads that should
905                  * have happened.
906                  */
907                 bump_cpu_timer(timer, now);
908         }
909
910         if (unlikely(clear_dead)) {
911                 /*
912                  * We've noticed that the thread is dead, but
913                  * not yet reaped.  Take this opportunity to
914                  * drop our task ref.
915                  */
916                 clear_dead_task(timer, now);
917                 goto dead;
918         }
919
920         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
921                 sample_to_timespec(timer->it_clock,
922                                    cpu_time_sub(timer->it_clock,
923                                                 timer->it.cpu.expires, now),
924                                    &itp->it_value);
925         } else {
926                 /*
927                  * The timer should have expired already, but the firing
928                  * hasn't taken place yet.  Say it's just about to expire.
929                  */
930                 itp->it_value.tv_nsec = 1;
931                 itp->it_value.tv_sec = 0;
932         }
933 }
934
935 /*
936  * Check for any per-thread CPU timers that have fired and move them off
937  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
938  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
939  */
940 static void check_thread_timers(struct task_struct *tsk,
941                                 struct list_head *firing)
942 {
943         int maxfire;
944         struct list_head *timers = tsk->cpu_timers;
945         struct signal_struct *const sig = tsk->signal;
946
947         maxfire = 20;
948         tsk->cputime_expires.prof_exp = cputime_zero;
949         while (!list_empty(timers)) {
950                 struct cpu_timer_list *t = list_first_entry(timers,
951                                                       struct cpu_timer_list,
952                                                       entry);
953                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
954                         tsk->cputime_expires.prof_exp = t->expires.cpu;
955                         break;
956                 }
957                 t->firing = 1;
958                 list_move_tail(&t->entry, firing);
959         }
960
961         ++timers;
962         maxfire = 20;
963         tsk->cputime_expires.virt_exp = cputime_zero;
964         while (!list_empty(timers)) {
965                 struct cpu_timer_list *t = list_first_entry(timers,
966                                                       struct cpu_timer_list,
967                                                       entry);
968                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
969                         tsk->cputime_expires.virt_exp = t->expires.cpu;
970                         break;
971                 }
972                 t->firing = 1;
973                 list_move_tail(&t->entry, firing);
974         }
975
976         ++timers;
977         maxfire = 20;
978         tsk->cputime_expires.sched_exp = 0;
979         while (!list_empty(timers)) {
980                 struct cpu_timer_list *t = list_first_entry(timers,
981                                                       struct cpu_timer_list,
982                                                       entry);
983                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
984                         tsk->cputime_expires.sched_exp = t->expires.sched;
985                         break;
986                 }
987                 t->firing = 1;
988                 list_move_tail(&t->entry, firing);
989         }
990
991         /*
992          * Check for the special case thread timers.
993          */
994         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
995                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
996                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
997
998                 if (hard != RLIM_INFINITY &&
999                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1000                         /*
1001                          * At the hard limit, we just die.
1002                          * No need to calculate anything else now.
1003                          */
1004                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1005                         return;
1006                 }
1007                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1008                         /*
1009                          * At the soft limit, send a SIGXCPU every second.
1010                          */
1011                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1012                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1013                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1014                                                                 USEC_PER_SEC;
1015                         }
1016                         printk(KERN_INFO
1017                                 "RT Watchdog Timeout: %s[%d]\n",
1018                                 tsk->comm, task_pid_nr(tsk));
1019                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1020                 }
1021         }
1022 }
1023
1024 /*
1025  * Check for any per-thread CPU timers that have fired and move them
1026  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1027  * have already been taken off.
1028  */
1029 static void check_process_timers(struct task_struct *tsk,
1030                                  struct list_head *firing)
1031 {
1032         int maxfire;
1033         struct signal_struct *const sig = tsk->signal;
1034         cputime_t utime, ptime, virt_expires, prof_expires;
1035         unsigned long long sum_sched_runtime, sched_expires;
1036         struct list_head *timers = sig->cpu_timers;
1037         struct task_cputime cputime;
1038
1039         /*
1040          * Don't sample the current process CPU clocks if there are no timers.
1041          */
1042         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1043             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1044             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1045             list_empty(&timers[CPUCLOCK_VIRT]) &&
1046             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1047             list_empty(&timers[CPUCLOCK_SCHED])) {
1048                 stop_process_timers(tsk);
1049                 return;
1050         }
1051
1052         /*
1053          * Collect the current process totals.
1054          */
1055         thread_group_cputimer(tsk, &cputime);
1056         utime = cputime.utime;
1057         ptime = cputime_add(utime, cputime.stime);
1058         sum_sched_runtime = cputime.sum_exec_runtime;
1059         maxfire = 20;
1060         prof_expires = cputime_zero;
1061         while (!list_empty(timers)) {
1062                 struct cpu_timer_list *tl = list_first_entry(timers,
1063                                                       struct cpu_timer_list,
1064                                                       entry);
1065                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1066                         prof_expires = tl->expires.cpu;
1067                         break;
1068                 }
1069                 tl->firing = 1;
1070                 list_move_tail(&tl->entry, firing);
1071         }
1072
1073         ++timers;
1074         maxfire = 20;
1075         virt_expires = cputime_zero;
1076         while (!list_empty(timers)) {
1077                 struct cpu_timer_list *tl = list_first_entry(timers,
1078                                                       struct cpu_timer_list,
1079                                                       entry);
1080                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1081                         virt_expires = tl->expires.cpu;
1082                         break;
1083                 }
1084                 tl->firing = 1;
1085                 list_move_tail(&tl->entry, firing);
1086         }
1087
1088         ++timers;
1089         maxfire = 20;
1090         sched_expires = 0;
1091         while (!list_empty(timers)) {
1092                 struct cpu_timer_list *tl = list_first_entry(timers,
1093                                                       struct cpu_timer_list,
1094                                                       entry);
1095                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1096                         sched_expires = tl->expires.sched;
1097                         break;
1098                 }
1099                 tl->firing = 1;
1100                 list_move_tail(&tl->entry, firing);
1101         }
1102
1103         /*
1104          * Check for the special case process timers.
1105          */
1106         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1107                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1108                         /* ITIMER_PROF fires and reloads.  */
1109                         sig->it_prof_expires = sig->it_prof_incr;
1110                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1111                                 sig->it_prof_expires = cputime_add(
1112                                         sig->it_prof_expires, ptime);
1113                         }
1114                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1115                 }
1116                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1117                     (cputime_eq(prof_expires, cputime_zero) ||
1118                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1119                         prof_expires = sig->it_prof_expires;
1120                 }
1121         }
1122         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1123                 if (cputime_ge(utime, sig->it_virt_expires)) {
1124                         /* ITIMER_VIRTUAL fires and reloads.  */
1125                         sig->it_virt_expires = sig->it_virt_incr;
1126                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1127                                 sig->it_virt_expires = cputime_add(
1128                                         sig->it_virt_expires, utime);
1129                         }
1130                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1131                 }
1132                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1133                     (cputime_eq(virt_expires, cputime_zero) ||
1134                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1135                         virt_expires = sig->it_virt_expires;
1136                 }
1137         }
1138         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1139                 unsigned long psecs = cputime_to_secs(ptime);
1140                 cputime_t x;
1141                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1142                         /*
1143                          * At the hard limit, we just die.
1144                          * No need to calculate anything else now.
1145                          */
1146                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1147                         return;
1148                 }
1149                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1150                         /*
1151                          * At the soft limit, send a SIGXCPU every second.
1152                          */
1153                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1154                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1155                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1156                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1157                         }
1158                 }
1159                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1160                 if (cputime_eq(prof_expires, cputime_zero) ||
1161                     cputime_lt(x, prof_expires)) {
1162                         prof_expires = x;
1163                 }
1164         }
1165
1166         if (!cputime_eq(prof_expires, cputime_zero) &&
1167             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1168              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1169                 sig->cputime_expires.prof_exp = prof_expires;
1170         if (!cputime_eq(virt_expires, cputime_zero) &&
1171             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1172              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1173                 sig->cputime_expires.virt_exp = virt_expires;
1174         if (sched_expires != 0 &&
1175             (sig->cputime_expires.sched_exp == 0 ||
1176              sig->cputime_expires.sched_exp > sched_expires))
1177                 sig->cputime_expires.sched_exp = sched_expires;
1178 }
1179
1180 /*
1181  * This is called from the signal code (via do_schedule_next_timer)
1182  * when the last timer signal was delivered and we have to reload the timer.
1183  */
1184 void posix_cpu_timer_schedule(struct k_itimer *timer)
1185 {
1186         struct task_struct *p = timer->it.cpu.task;
1187         union cpu_time_count now;
1188
1189         if (unlikely(p == NULL))
1190                 /*
1191                  * The task was cleaned up already, no future firings.
1192                  */
1193                 goto out;
1194
1195         /*
1196          * Fetch the current sample and update the timer's expiry time.
1197          */
1198         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1199                 cpu_clock_sample(timer->it_clock, p, &now);
1200                 bump_cpu_timer(timer, now);
1201                 if (unlikely(p->exit_state)) {
1202                         clear_dead_task(timer, now);
1203                         goto out;
1204                 }
1205                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1206         } else {
1207                 read_lock(&tasklist_lock);
1208                 if (unlikely(p->signal == NULL)) {
1209                         /*
1210                          * The process has been reaped.
1211                          * We can't even collect a sample any more.
1212                          */
1213                         put_task_struct(p);
1214                         timer->it.cpu.task = p = NULL;
1215                         timer->it.cpu.expires.sched = 0;
1216                         goto out_unlock;
1217                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1218                         /*
1219                          * We've noticed that the thread is dead, but
1220                          * not yet reaped.  Take this opportunity to
1221                          * drop our task ref.
1222                          */
1223                         clear_dead_task(timer, now);
1224                         goto out_unlock;
1225                 }
1226                 cpu_clock_sample_group(timer->it_clock, p, &now);
1227                 bump_cpu_timer(timer, now);
1228                 /* Leave the tasklist_lock locked for the call below.  */
1229         }
1230
1231         /*
1232          * Now re-arm for the new expiry time.
1233          */
1234         arm_timer(timer, now);
1235
1236 out_unlock:
1237         read_unlock(&tasklist_lock);
1238
1239 out:
1240         timer->it_overrun_last = timer->it_overrun;
1241         timer->it_overrun = -1;
1242         ++timer->it_requeue_pending;
1243 }
1244
1245 /**
1246  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1247  *
1248  * @cputime:    The struct to compare.
1249  *
1250  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1251  * are zero, false if any field is nonzero.
1252  */
1253 static inline int task_cputime_zero(const struct task_cputime *cputime)
1254 {
1255         if (cputime_eq(cputime->utime, cputime_zero) &&
1256             cputime_eq(cputime->stime, cputime_zero) &&
1257             cputime->sum_exec_runtime == 0)
1258                 return 1;
1259         return 0;
1260 }
1261
1262 /**
1263  * task_cputime_expired - Compare two task_cputime entities.
1264  *
1265  * @sample:     The task_cputime structure to be checked for expiration.
1266  * @expires:    Expiration times, against which @sample will be checked.
1267  *
1268  * Checks @sample against @expires to see if any field of @sample has expired.
1269  * Returns true if any field of the former is greater than the corresponding
1270  * field of the latter if the latter field is set.  Otherwise returns false.
1271  */
1272 static inline int task_cputime_expired(const struct task_cputime *sample,
1273                                         const struct task_cputime *expires)
1274 {
1275         if (!cputime_eq(expires->utime, cputime_zero) &&
1276             cputime_ge(sample->utime, expires->utime))
1277                 return 1;
1278         if (!cputime_eq(expires->stime, cputime_zero) &&
1279             cputime_ge(cputime_add(sample->utime, sample->stime),
1280                        expires->stime))
1281                 return 1;
1282         if (expires->sum_exec_runtime != 0 &&
1283             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1284                 return 1;
1285         return 0;
1286 }
1287
1288 /**
1289  * fastpath_timer_check - POSIX CPU timers fast path.
1290  *
1291  * @tsk:        The task (thread) being checked.
1292  *
1293  * Check the task and thread group timers.  If both are zero (there are no
1294  * timers set) return false.  Otherwise snapshot the task and thread group
1295  * timers and compare them with the corresponding expiration times.  Return
1296  * true if a timer has expired, else return false.
1297  */
1298 static inline int fastpath_timer_check(struct task_struct *tsk)
1299 {
1300         struct signal_struct *sig;
1301
1302         /* tsk == current, ensure it is safe to use ->signal/sighand */
1303         if (unlikely(tsk->exit_state))
1304                 return 0;
1305
1306         if (!task_cputime_zero(&tsk->cputime_expires)) {
1307                 struct task_cputime task_sample = {
1308                         .utime = tsk->utime,
1309                         .stime = tsk->stime,
1310                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1311                 };
1312
1313                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1314                         return 1;
1315         }
1316
1317         sig = tsk->signal;
1318         if (!task_cputime_zero(&sig->cputime_expires)) {
1319                 struct task_cputime group_sample;
1320
1321                 thread_group_cputimer(tsk, &group_sample);
1322                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1323                         return 1;
1324         }
1325         return 0;
1326 }
1327
1328 /*
1329  * This is called from the timer interrupt handler.  The irq handler has
1330  * already updated our counts.  We need to check if any timers fire now.
1331  * Interrupts are disabled.
1332  */
1333 void run_posix_cpu_timers(struct task_struct *tsk)
1334 {
1335         LIST_HEAD(firing);
1336         struct k_itimer *timer, *next;
1337
1338         BUG_ON(!irqs_disabled());
1339
1340         /*
1341          * The fast path checks that there are no expired thread or thread
1342          * group timers.  If that's so, just return.
1343          */
1344         if (!fastpath_timer_check(tsk))
1345                 return;
1346
1347         spin_lock(&tsk->sighand->siglock);
1348         /*
1349          * Here we take off tsk->signal->cpu_timers[N] and
1350          * tsk->cpu_timers[N] all the timers that are firing, and
1351          * put them on the firing list.
1352          */
1353         check_thread_timers(tsk, &firing);
1354         check_process_timers(tsk, &firing);
1355
1356         /*
1357          * We must release these locks before taking any timer's lock.
1358          * There is a potential race with timer deletion here, as the
1359          * siglock now protects our private firing list.  We have set
1360          * the firing flag in each timer, so that a deletion attempt
1361          * that gets the timer lock before we do will give it up and
1362          * spin until we've taken care of that timer below.
1363          */
1364         spin_unlock(&tsk->sighand->siglock);
1365
1366         /*
1367          * Now that all the timers on our list have the firing flag,
1368          * noone will touch their list entries but us.  We'll take
1369          * each timer's lock before clearing its firing flag, so no
1370          * timer call will interfere.
1371          */
1372         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1373                 int firing;
1374                 spin_lock(&timer->it_lock);
1375                 list_del_init(&timer->it.cpu.entry);
1376                 firing = timer->it.cpu.firing;
1377                 timer->it.cpu.firing = 0;
1378                 /*
1379                  * The firing flag is -1 if we collided with a reset
1380                  * of the timer, which already reported this
1381                  * almost-firing as an overrun.  So don't generate an event.
1382                  */
1383                 if (likely(firing >= 0)) {
1384                         cpu_timer_fire(timer);
1385                 }
1386                 spin_unlock(&timer->it_lock);
1387         }
1388 }
1389
1390 /*
1391  * Sample a process (thread group) timer for the given group_leader task.
1392  * Must be called with tasklist_lock held for reading.
1393  */
1394 static int cpu_timer_sample_group(const clockid_t which_clock,
1395                                   struct task_struct *p,
1396                                   union cpu_time_count *cpu)
1397 {
1398         struct task_cputime cputime;
1399
1400         thread_group_cputimer(p, &cputime);
1401         switch (CPUCLOCK_WHICH(which_clock)) {
1402         default:
1403                 return -EINVAL;
1404         case CPUCLOCK_PROF:
1405                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
1406                 break;
1407         case CPUCLOCK_VIRT:
1408                 cpu->cpu = cputime.utime;
1409                 break;
1410         case CPUCLOCK_SCHED:
1411                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
1412                 break;
1413         }
1414         return 0;
1415 }
1416
1417 /*
1418  * Set one of the process-wide special case CPU timers.
1419  * The tsk->sighand->siglock must be held by the caller.
1420  * The *newval argument is relative and we update it to be absolute, *oldval
1421  * is absolute and we update it to be relative.
1422  */
1423 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1424                            cputime_t *newval, cputime_t *oldval)
1425 {
1426         union cpu_time_count now;
1427         struct list_head *head;
1428
1429         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1430         start_process_timers(tsk);
1431         cpu_timer_sample_group(clock_idx, tsk, &now);
1432
1433         if (oldval) {
1434                 if (!cputime_eq(*oldval, cputime_zero)) {
1435                         if (cputime_le(*oldval, now.cpu)) {
1436                                 /* Just about to fire. */
1437                                 *oldval = jiffies_to_cputime(1);
1438                         } else {
1439                                 *oldval = cputime_sub(*oldval, now.cpu);
1440                         }
1441                 }
1442
1443                 if (cputime_eq(*newval, cputime_zero))
1444                         return;
1445                 *newval = cputime_add(*newval, now.cpu);
1446
1447                 /*
1448                  * If the RLIMIT_CPU timer will expire before the
1449                  * ITIMER_PROF timer, we have nothing else to do.
1450                  */
1451                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1452                     < cputime_to_secs(*newval))
1453                         return;
1454         }
1455
1456         /*
1457          * Check whether there are any process timers already set to fire
1458          * before this one.  If so, we don't have anything more to do.
1459          */
1460         head = &tsk->signal->cpu_timers[clock_idx];
1461         if (list_empty(head) ||
1462             cputime_ge(list_first_entry(head,
1463                                   struct cpu_timer_list, entry)->expires.cpu,
1464                        *newval)) {
1465                 switch (clock_idx) {
1466                 case CPUCLOCK_PROF:
1467                         tsk->signal->cputime_expires.prof_exp = *newval;
1468                         break;
1469                 case CPUCLOCK_VIRT:
1470                         tsk->signal->cputime_expires.virt_exp = *newval;
1471                         break;
1472                 }
1473         }
1474 }
1475
1476 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1477                             struct timespec *rqtp, struct itimerspec *it)
1478 {
1479         struct k_itimer timer;
1480         int error;
1481
1482         /*
1483          * Set up a temporary timer and then wait for it to go off.
1484          */
1485         memset(&timer, 0, sizeof timer);
1486         spin_lock_init(&timer.it_lock);
1487         timer.it_clock = which_clock;
1488         timer.it_overrun = -1;
1489         error = posix_cpu_timer_create(&timer);
1490         timer.it_process = current;
1491         if (!error) {
1492                 static struct itimerspec zero_it;
1493
1494                 memset(it, 0, sizeof *it);
1495                 it->it_value = *rqtp;
1496
1497                 spin_lock_irq(&timer.it_lock);
1498                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1499                 if (error) {
1500                         spin_unlock_irq(&timer.it_lock);
1501                         return error;
1502                 }
1503
1504                 while (!signal_pending(current)) {
1505                         if (timer.it.cpu.expires.sched == 0) {
1506                                 /*
1507                                  * Our timer fired and was reset.
1508                                  */
1509                                 spin_unlock_irq(&timer.it_lock);
1510                                 return 0;
1511                         }
1512
1513                         /*
1514                          * Block until cpu_timer_fire (or a signal) wakes us.
1515                          */
1516                         __set_current_state(TASK_INTERRUPTIBLE);
1517                         spin_unlock_irq(&timer.it_lock);
1518                         schedule();
1519                         spin_lock_irq(&timer.it_lock);
1520                 }
1521
1522                 /*
1523                  * We were interrupted by a signal.
1524                  */
1525                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1526                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1527                 spin_unlock_irq(&timer.it_lock);
1528
1529                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1530                         /*
1531                          * It actually did fire already.
1532                          */
1533                         return 0;
1534                 }
1535
1536                 error = -ERESTART_RESTARTBLOCK;
1537         }
1538
1539         return error;
1540 }
1541
1542 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1543                      struct timespec *rqtp, struct timespec __user *rmtp)
1544 {
1545         struct restart_block *restart_block =
1546             &current_thread_info()->restart_block;
1547         struct itimerspec it;
1548         int error;
1549
1550         /*
1551          * Diagnose required errors first.
1552          */
1553         if (CPUCLOCK_PERTHREAD(which_clock) &&
1554             (CPUCLOCK_PID(which_clock) == 0 ||
1555              CPUCLOCK_PID(which_clock) == current->pid))
1556                 return -EINVAL;
1557
1558         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1559
1560         if (error == -ERESTART_RESTARTBLOCK) {
1561
1562                 if (flags & TIMER_ABSTIME)
1563                         return -ERESTARTNOHAND;
1564                 /*
1565                  * Report back to the user the time still remaining.
1566                  */
1567                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1568                         return -EFAULT;
1569
1570                 restart_block->fn = posix_cpu_nsleep_restart;
1571                 restart_block->arg0 = which_clock;
1572                 restart_block->arg1 = (unsigned long) rmtp;
1573                 restart_block->arg2 = rqtp->tv_sec;
1574                 restart_block->arg3 = rqtp->tv_nsec;
1575         }
1576         return error;
1577 }
1578
1579 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1580 {
1581         clockid_t which_clock = restart_block->arg0;
1582         struct timespec __user *rmtp;
1583         struct timespec t;
1584         struct itimerspec it;
1585         int error;
1586
1587         rmtp = (struct timespec __user *) restart_block->arg1;
1588         t.tv_sec = restart_block->arg2;
1589         t.tv_nsec = restart_block->arg3;
1590
1591         restart_block->fn = do_no_restart_syscall;
1592         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1593
1594         if (error == -ERESTART_RESTARTBLOCK) {
1595                 /*
1596                  * Report back to the user the time still remaining.
1597                  */
1598                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1599                         return -EFAULT;
1600
1601                 restart_block->fn = posix_cpu_nsleep_restart;
1602                 restart_block->arg0 = which_clock;
1603                 restart_block->arg1 = (unsigned long) rmtp;
1604                 restart_block->arg2 = t.tv_sec;
1605                 restart_block->arg3 = t.tv_nsec;
1606         }
1607         return error;
1608
1609 }
1610
1611
1612 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1613 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1614
1615 static int process_cpu_clock_getres(const clockid_t which_clock,
1616                                     struct timespec *tp)
1617 {
1618         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1619 }
1620 static int process_cpu_clock_get(const clockid_t which_clock,
1621                                  struct timespec *tp)
1622 {
1623         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1624 }
1625 static int process_cpu_timer_create(struct k_itimer *timer)
1626 {
1627         timer->it_clock = PROCESS_CLOCK;
1628         return posix_cpu_timer_create(timer);
1629 }
1630 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1631                               struct timespec *rqtp,
1632                               struct timespec __user *rmtp)
1633 {
1634         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1635 }
1636 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1637 {
1638         return -EINVAL;
1639 }
1640 static int thread_cpu_clock_getres(const clockid_t which_clock,
1641                                    struct timespec *tp)
1642 {
1643         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1644 }
1645 static int thread_cpu_clock_get(const clockid_t which_clock,
1646                                 struct timespec *tp)
1647 {
1648         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1649 }
1650 static int thread_cpu_timer_create(struct k_itimer *timer)
1651 {
1652         timer->it_clock = THREAD_CLOCK;
1653         return posix_cpu_timer_create(timer);
1654 }
1655 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1656                               struct timespec *rqtp, struct timespec __user *rmtp)
1657 {
1658         return -EINVAL;
1659 }
1660 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1661 {
1662         return -EINVAL;
1663 }
1664
1665 static __init int init_posix_cpu_timers(void)
1666 {
1667         struct k_clock process = {
1668                 .clock_getres = process_cpu_clock_getres,
1669                 .clock_get = process_cpu_clock_get,
1670                 .clock_set = do_posix_clock_nosettime,
1671                 .timer_create = process_cpu_timer_create,
1672                 .nsleep = process_cpu_nsleep,
1673                 .nsleep_restart = process_cpu_nsleep_restart,
1674         };
1675         struct k_clock thread = {
1676                 .clock_getres = thread_cpu_clock_getres,
1677                 .clock_get = thread_cpu_clock_get,
1678                 .clock_set = do_posix_clock_nosettime,
1679                 .timer_create = thread_cpu_timer_create,
1680                 .nsleep = thread_cpu_nsleep,
1681                 .nsleep_restart = thread_cpu_nsleep_restart,
1682         };
1683
1684         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1685         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1686
1687         return 0;
1688 }
1689 __initcall(init_posix_cpu_timers);