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