posix-cpu-timers: Cure SMP wobbles
[linux-2.6.git] / kernel / posix-cpu-timers.c
1 /*
2  * Implement CPU time clocks for the POSIX clock interface.
3  */
4
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <linux/errno.h>
8 #include <linux/math64.h>
9 #include <asm/uaccess.h>
10 #include <linux/kernel_stat.h>
11 #include <trace/events/timer.h>
12
13 /*
14  * Called after updating RLIMIT_CPU to run cpu timer and update
15  * tsk->signal->cputime_expires expiration cache if necessary. Needs
16  * siglock protection since other code may update expiration cache as
17  * well.
18  */
19 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
20 {
21         cputime_t cputime = secs_to_cputime(rlim_new);
22
23         spin_lock_irq(&task->sighand->siglock);
24         set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
25         spin_unlock_irq(&task->sighand->siglock);
26 }
27
28 static int check_clock(const clockid_t which_clock)
29 {
30         int error = 0;
31         struct task_struct *p;
32         const pid_t pid = CPUCLOCK_PID(which_clock);
33
34         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
35                 return -EINVAL;
36
37         if (pid == 0)
38                 return 0;
39
40         rcu_read_lock();
41         p = find_task_by_vpid(pid);
42         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
43                    same_thread_group(p, current) : has_group_leader_pid(p))) {
44                 error = -EINVAL;
45         }
46         rcu_read_unlock();
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 static int
180 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
181 {
182         int error = check_clock(which_clock);
183         if (!error) {
184                 tp->tv_sec = 0;
185                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
186                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
187                         /*
188                          * If sched_clock is using a cycle counter, we
189                          * don't have any idea of its true resolution
190                          * exported, but it is much more than 1s/HZ.
191                          */
192                         tp->tv_nsec = 1;
193                 }
194         }
195         return error;
196 }
197
198 static int
199 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
200 {
201         /*
202          * You can never reset a CPU clock, but we check for other errors
203          * in the call before failing with EPERM.
204          */
205         int error = check_clock(which_clock);
206         if (error == 0) {
207                 error = -EPERM;
208         }
209         return error;
210 }
211
212
213 /*
214  * Sample a per-thread clock for the given task.
215  */
216 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
217                             union cpu_time_count *cpu)
218 {
219         switch (CPUCLOCK_WHICH(which_clock)) {
220         default:
221                 return -EINVAL;
222         case CPUCLOCK_PROF:
223                 cpu->cpu = prof_ticks(p);
224                 break;
225         case CPUCLOCK_VIRT:
226                 cpu->cpu = virt_ticks(p);
227                 break;
228         case CPUCLOCK_SCHED:
229                 cpu->sched = task_sched_runtime(p);
230                 break;
231         }
232         return 0;
233 }
234
235 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
236 {
237         struct signal_struct *sig = tsk->signal;
238         struct task_struct *t;
239
240         times->utime = sig->utime;
241         times->stime = sig->stime;
242         times->sum_exec_runtime = sig->sum_sched_runtime;
243
244         rcu_read_lock();
245         /* make sure we can trust tsk->thread_group list */
246         if (!likely(pid_alive(tsk)))
247                 goto out;
248
249         t = tsk;
250         do {
251                 times->utime = cputime_add(times->utime, t->utime);
252                 times->stime = cputime_add(times->stime, t->stime);
253                 times->sum_exec_runtime += task_sched_runtime(t);
254         } while_each_thread(tsk, t);
255 out:
256         rcu_read_unlock();
257 }
258
259 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
260 {
261         if (cputime_gt(b->utime, a->utime))
262                 a->utime = b->utime;
263
264         if (cputime_gt(b->stime, a->stime))
265                 a->stime = b->stime;
266
267         if (b->sum_exec_runtime > a->sum_exec_runtime)
268                 a->sum_exec_runtime = b->sum_exec_runtime;
269 }
270
271 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
272 {
273         struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
274         struct task_cputime sum;
275         unsigned long flags;
276
277         spin_lock_irqsave(&cputimer->lock, flags);
278         if (!cputimer->running) {
279                 cputimer->running = 1;
280                 /*
281                  * The POSIX timer interface allows for absolute time expiry
282                  * values through the TIMER_ABSTIME flag, therefore we have
283                  * to synchronize the timer to the clock every time we start
284                  * it.
285                  */
286                 thread_group_cputime(tsk, &sum);
287                 update_gt_cputime(&cputimer->cputime, &sum);
288         }
289         *times = cputimer->cputime;
290         spin_unlock_irqrestore(&cputimer->lock, flags);
291 }
292
293 /*
294  * Sample a process (thread group) clock for the given group_leader task.
295  * Must be called with tasklist_lock held for reading.
296  */
297 static int cpu_clock_sample_group(const clockid_t which_clock,
298                                   struct task_struct *p,
299                                   union cpu_time_count *cpu)
300 {
301         struct task_cputime cputime;
302
303         switch (CPUCLOCK_WHICH(which_clock)) {
304         default:
305                 return -EINVAL;
306         case CPUCLOCK_PROF:
307                 thread_group_cputime(p, &cputime);
308                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
309                 break;
310         case CPUCLOCK_VIRT:
311                 thread_group_cputime(p, &cputime);
312                 cpu->cpu = cputime.utime;
313                 break;
314         case CPUCLOCK_SCHED:
315                 thread_group_cputime(p, &cputime);
316                 cpu->sched = cputime.sum_exec_runtime;
317                 break;
318         }
319         return 0;
320 }
321
322
323 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
324 {
325         const pid_t pid = CPUCLOCK_PID(which_clock);
326         int error = -EINVAL;
327         union cpu_time_count rtn;
328
329         if (pid == 0) {
330                 /*
331                  * Special case constant value for our own clocks.
332                  * We don't have to do any lookup to find ourselves.
333                  */
334                 if (CPUCLOCK_PERTHREAD(which_clock)) {
335                         /*
336                          * Sampling just ourselves we can do with no locking.
337                          */
338                         error = cpu_clock_sample(which_clock,
339                                                  current, &rtn);
340                 } else {
341                         read_lock(&tasklist_lock);
342                         error = cpu_clock_sample_group(which_clock,
343                                                        current, &rtn);
344                         read_unlock(&tasklist_lock);
345                 }
346         } else {
347                 /*
348                  * Find the given PID, and validate that the caller
349                  * should be able to see it.
350                  */
351                 struct task_struct *p;
352                 rcu_read_lock();
353                 p = find_task_by_vpid(pid);
354                 if (p) {
355                         if (CPUCLOCK_PERTHREAD(which_clock)) {
356                                 if (same_thread_group(p, current)) {
357                                         error = cpu_clock_sample(which_clock,
358                                                                  p, &rtn);
359                                 }
360                         } else {
361                                 read_lock(&tasklist_lock);
362                                 if (thread_group_leader(p) && p->sighand) {
363                                         error =
364                                             cpu_clock_sample_group(which_clock,
365                                                                    p, &rtn);
366                                 }
367                                 read_unlock(&tasklist_lock);
368                         }
369                 }
370                 rcu_read_unlock();
371         }
372
373         if (error)
374                 return error;
375         sample_to_timespec(which_clock, rtn, tp);
376         return 0;
377 }
378
379
380 /*
381  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
382  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
383  * new timer already all-zeros initialized.
384  */
385 static int posix_cpu_timer_create(struct k_itimer *new_timer)
386 {
387         int ret = 0;
388         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
389         struct task_struct *p;
390
391         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
392                 return -EINVAL;
393
394         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
395
396         rcu_read_lock();
397         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
398                 if (pid == 0) {
399                         p = current;
400                 } else {
401                         p = find_task_by_vpid(pid);
402                         if (p && !same_thread_group(p, current))
403                                 p = NULL;
404                 }
405         } else {
406                 if (pid == 0) {
407                         p = current->group_leader;
408                 } else {
409                         p = find_task_by_vpid(pid);
410                         if (p && !has_group_leader_pid(p))
411                                 p = NULL;
412                 }
413         }
414         new_timer->it.cpu.task = p;
415         if (p) {
416                 get_task_struct(p);
417         } else {
418                 ret = -EINVAL;
419         }
420         rcu_read_unlock();
421
422         return ret;
423 }
424
425 /*
426  * Clean up a CPU-clock timer that is about to be destroyed.
427  * This is called from timer deletion with the timer already locked.
428  * If we return TIMER_RETRY, it's necessary to release the timer's lock
429  * and try again.  (This happens when the timer is in the middle of firing.)
430  */
431 static int posix_cpu_timer_del(struct k_itimer *timer)
432 {
433         struct task_struct *p = timer->it.cpu.task;
434         int ret = 0;
435
436         if (likely(p != NULL)) {
437                 read_lock(&tasklist_lock);
438                 if (unlikely(p->sighand == NULL)) {
439                         /*
440                          * We raced with the reaping of the task.
441                          * The deletion should have cleared us off the list.
442                          */
443                         BUG_ON(!list_empty(&timer->it.cpu.entry));
444                 } else {
445                         spin_lock(&p->sighand->siglock);
446                         if (timer->it.cpu.firing)
447                                 ret = TIMER_RETRY;
448                         else
449                                 list_del(&timer->it.cpu.entry);
450                         spin_unlock(&p->sighand->siglock);
451                 }
452                 read_unlock(&tasklist_lock);
453
454                 if (!ret)
455                         put_task_struct(p);
456         }
457
458         return ret;
459 }
460
461 /*
462  * Clean out CPU timers still ticking when a thread exited.  The task
463  * pointer is cleared, and the expiry time is replaced with the residual
464  * time for later timer_gettime calls to return.
465  * This must be called with the siglock held.
466  */
467 static void cleanup_timers(struct list_head *head,
468                            cputime_t utime, cputime_t stime,
469                            unsigned long long sum_exec_runtime)
470 {
471         struct cpu_timer_list *timer, *next;
472         cputime_t ptime = cputime_add(utime, stime);
473
474         list_for_each_entry_safe(timer, next, head, entry) {
475                 list_del_init(&timer->entry);
476                 if (cputime_lt(timer->expires.cpu, ptime)) {
477                         timer->expires.cpu = cputime_zero;
478                 } else {
479                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
480                                                          ptime);
481                 }
482         }
483
484         ++head;
485         list_for_each_entry_safe(timer, next, head, entry) {
486                 list_del_init(&timer->entry);
487                 if (cputime_lt(timer->expires.cpu, utime)) {
488                         timer->expires.cpu = cputime_zero;
489                 } else {
490                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
491                                                          utime);
492                 }
493         }
494
495         ++head;
496         list_for_each_entry_safe(timer, next, head, entry) {
497                 list_del_init(&timer->entry);
498                 if (timer->expires.sched < sum_exec_runtime) {
499                         timer->expires.sched = 0;
500                 } else {
501                         timer->expires.sched -= sum_exec_runtime;
502                 }
503         }
504 }
505
506 /*
507  * These are both called with the siglock held, when the current thread
508  * is being reaped.  When the final (leader) thread in the group is reaped,
509  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
510  */
511 void posix_cpu_timers_exit(struct task_struct *tsk)
512 {
513         cleanup_timers(tsk->cpu_timers,
514                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
515
516 }
517 void posix_cpu_timers_exit_group(struct task_struct *tsk)
518 {
519         struct signal_struct *const sig = tsk->signal;
520
521         cleanup_timers(tsk->signal->cpu_timers,
522                        cputime_add(tsk->utime, sig->utime),
523                        cputime_add(tsk->stime, sig->stime),
524                        tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
525 }
526
527 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
528 {
529         /*
530          * That's all for this thread or process.
531          * We leave our residual in expires to be reported.
532          */
533         put_task_struct(timer->it.cpu.task);
534         timer->it.cpu.task = NULL;
535         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
536                                              timer->it.cpu.expires,
537                                              now);
538 }
539
540 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
541 {
542         return cputime_eq(expires, cputime_zero) ||
543                cputime_gt(expires, new_exp);
544 }
545
546 /*
547  * Insert the timer on the appropriate list before any timers that
548  * expire later.  This must be called with the tasklist_lock held
549  * for reading, interrupts disabled and p->sighand->siglock taken.
550  */
551 static void arm_timer(struct k_itimer *timer)
552 {
553         struct task_struct *p = timer->it.cpu.task;
554         struct list_head *head, *listpos;
555         struct task_cputime *cputime_expires;
556         struct cpu_timer_list *const nt = &timer->it.cpu;
557         struct cpu_timer_list *next;
558
559         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
560                 head = p->cpu_timers;
561                 cputime_expires = &p->cputime_expires;
562         } else {
563                 head = p->signal->cpu_timers;
564                 cputime_expires = &p->signal->cputime_expires;
565         }
566         head += CPUCLOCK_WHICH(timer->it_clock);
567
568         listpos = head;
569         list_for_each_entry(next, head, entry) {
570                 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
571                         break;
572                 listpos = &next->entry;
573         }
574         list_add(&nt->entry, listpos);
575
576         if (listpos == head) {
577                 union cpu_time_count *exp = &nt->expires;
578
579                 /*
580                  * We are the new earliest-expiring POSIX 1.b timer, hence
581                  * need to update expiration cache. Take into account that
582                  * for process timers we share expiration cache with itimers
583                  * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
584                  */
585
586                 switch (CPUCLOCK_WHICH(timer->it_clock)) {
587                 case CPUCLOCK_PROF:
588                         if (expires_gt(cputime_expires->prof_exp, exp->cpu))
589                                 cputime_expires->prof_exp = exp->cpu;
590                         break;
591                 case CPUCLOCK_VIRT:
592                         if (expires_gt(cputime_expires->virt_exp, exp->cpu))
593                                 cputime_expires->virt_exp = exp->cpu;
594                         break;
595                 case CPUCLOCK_SCHED:
596                         if (cputime_expires->sched_exp == 0 ||
597                             cputime_expires->sched_exp > exp->sched)
598                                 cputime_expires->sched_exp = exp->sched;
599                         break;
600                 }
601         }
602 }
603
604 /*
605  * The timer is locked, fire it and arrange for its reload.
606  */
607 static void cpu_timer_fire(struct k_itimer *timer)
608 {
609         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
610                 /*
611                  * User don't want any signal.
612                  */
613                 timer->it.cpu.expires.sched = 0;
614         } else if (unlikely(timer->sigq == NULL)) {
615                 /*
616                  * This a special case for clock_nanosleep,
617                  * not a normal timer from sys_timer_create.
618                  */
619                 wake_up_process(timer->it_process);
620                 timer->it.cpu.expires.sched = 0;
621         } else if (timer->it.cpu.incr.sched == 0) {
622                 /*
623                  * One-shot timer.  Clear it as soon as it's fired.
624                  */
625                 posix_timer_event(timer, 0);
626                 timer->it.cpu.expires.sched = 0;
627         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
628                 /*
629                  * The signal did not get queued because the signal
630                  * was ignored, so we won't get any callback to
631                  * reload the timer.  But we need to keep it
632                  * ticking in case the signal is deliverable next time.
633                  */
634                 posix_cpu_timer_schedule(timer);
635         }
636 }
637
638 /*
639  * Sample a process (thread group) timer for the given group_leader task.
640  * Must be called with tasklist_lock held for reading.
641  */
642 static int cpu_timer_sample_group(const clockid_t which_clock,
643                                   struct task_struct *p,
644                                   union cpu_time_count *cpu)
645 {
646         struct task_cputime cputime;
647
648         thread_group_cputimer(p, &cputime);
649         switch (CPUCLOCK_WHICH(which_clock)) {
650         default:
651                 return -EINVAL;
652         case CPUCLOCK_PROF:
653                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
654                 break;
655         case CPUCLOCK_VIRT:
656                 cpu->cpu = cputime.utime;
657                 break;
658         case CPUCLOCK_SCHED:
659                 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
660                 break;
661         }
662         return 0;
663 }
664
665 /*
666  * Guts of sys_timer_settime for CPU timers.
667  * This is called with the timer locked and interrupts disabled.
668  * If we return TIMER_RETRY, it's necessary to release the timer's lock
669  * and try again.  (This happens when the timer is in the middle of firing.)
670  */
671 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
672                                struct itimerspec *new, struct itimerspec *old)
673 {
674         struct task_struct *p = timer->it.cpu.task;
675         union cpu_time_count old_expires, new_expires, old_incr, val;
676         int ret;
677
678         if (unlikely(p == NULL)) {
679                 /*
680                  * Timer refers to a dead task's clock.
681                  */
682                 return -ESRCH;
683         }
684
685         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
686
687         read_lock(&tasklist_lock);
688         /*
689          * We need the tasklist_lock to protect against reaping that
690          * clears p->sighand.  If p has just been reaped, we can no
691          * longer get any information about it at all.
692          */
693         if (unlikely(p->sighand == NULL)) {
694                 read_unlock(&tasklist_lock);
695                 put_task_struct(p);
696                 timer->it.cpu.task = NULL;
697                 return -ESRCH;
698         }
699
700         /*
701          * Disarm any old timer after extracting its expiry time.
702          */
703         BUG_ON(!irqs_disabled());
704
705         ret = 0;
706         old_incr = timer->it.cpu.incr;
707         spin_lock(&p->sighand->siglock);
708         old_expires = timer->it.cpu.expires;
709         if (unlikely(timer->it.cpu.firing)) {
710                 timer->it.cpu.firing = -1;
711                 ret = TIMER_RETRY;
712         } else
713                 list_del_init(&timer->it.cpu.entry);
714
715         /*
716          * We need to sample the current value to convert the new
717          * value from to relative and absolute, and to convert the
718          * old value from absolute to relative.  To set a process
719          * timer, we need a sample to balance the thread expiry
720          * times (in arm_timer).  With an absolute time, we must
721          * check if it's already passed.  In short, we need a sample.
722          */
723         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
724                 cpu_clock_sample(timer->it_clock, p, &val);
725         } else {
726                 cpu_timer_sample_group(timer->it_clock, p, &val);
727         }
728
729         if (old) {
730                 if (old_expires.sched == 0) {
731                         old->it_value.tv_sec = 0;
732                         old->it_value.tv_nsec = 0;
733                 } else {
734                         /*
735                          * Update the timer in case it has
736                          * overrun already.  If it has,
737                          * we'll report it as having overrun
738                          * and with the next reloaded timer
739                          * already ticking, though we are
740                          * swallowing that pending
741                          * notification here to install the
742                          * new setting.
743                          */
744                         bump_cpu_timer(timer, val);
745                         if (cpu_time_before(timer->it_clock, val,
746                                             timer->it.cpu.expires)) {
747                                 old_expires = cpu_time_sub(
748                                         timer->it_clock,
749                                         timer->it.cpu.expires, val);
750                                 sample_to_timespec(timer->it_clock,
751                                                    old_expires,
752                                                    &old->it_value);
753                         } else {
754                                 old->it_value.tv_nsec = 1;
755                                 old->it_value.tv_sec = 0;
756                         }
757                 }
758         }
759
760         if (unlikely(ret)) {
761                 /*
762                  * We are colliding with the timer actually firing.
763                  * Punt after filling in the timer's old value, and
764                  * disable this firing since we are already reporting
765                  * it as an overrun (thanks to bump_cpu_timer above).
766                  */
767                 spin_unlock(&p->sighand->siglock);
768                 read_unlock(&tasklist_lock);
769                 goto out;
770         }
771
772         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
773                 cpu_time_add(timer->it_clock, &new_expires, val);
774         }
775
776         /*
777          * Install the new expiry time (or zero).
778          * For a timer with no notification action, we don't actually
779          * arm the timer (we'll just fake it for timer_gettime).
780          */
781         timer->it.cpu.expires = new_expires;
782         if (new_expires.sched != 0 &&
783             cpu_time_before(timer->it_clock, val, new_expires)) {
784                 arm_timer(timer);
785         }
786
787         spin_unlock(&p->sighand->siglock);
788         read_unlock(&tasklist_lock);
789
790         /*
791          * Install the new reload setting, and
792          * set up the signal and overrun bookkeeping.
793          */
794         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
795                                                 &new->it_interval);
796
797         /*
798          * This acts as a modification timestamp for the timer,
799          * so any automatic reload attempt will punt on seeing
800          * that we have reset the timer manually.
801          */
802         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
803                 ~REQUEUE_PENDING;
804         timer->it_overrun_last = 0;
805         timer->it_overrun = -1;
806
807         if (new_expires.sched != 0 &&
808             !cpu_time_before(timer->it_clock, val, new_expires)) {
809                 /*
810                  * The designated time already passed, so we notify
811                  * immediately, even if the thread never runs to
812                  * accumulate more time on this clock.
813                  */
814                 cpu_timer_fire(timer);
815         }
816
817         ret = 0;
818  out:
819         if (old) {
820                 sample_to_timespec(timer->it_clock,
821                                    old_incr, &old->it_interval);
822         }
823         return ret;
824 }
825
826 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
827 {
828         union cpu_time_count now;
829         struct task_struct *p = timer->it.cpu.task;
830         int clear_dead;
831
832         /*
833          * Easy part: convert the reload time.
834          */
835         sample_to_timespec(timer->it_clock,
836                            timer->it.cpu.incr, &itp->it_interval);
837
838         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
839                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
840                 return;
841         }
842
843         if (unlikely(p == NULL)) {
844                 /*
845                  * This task already died and the timer will never fire.
846                  * In this case, expires is actually the dead value.
847                  */
848         dead:
849                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
850                                    &itp->it_value);
851                 return;
852         }
853
854         /*
855          * Sample the clock to take the difference with the expiry time.
856          */
857         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
858                 cpu_clock_sample(timer->it_clock, p, &now);
859                 clear_dead = p->exit_state;
860         } else {
861                 read_lock(&tasklist_lock);
862                 if (unlikely(p->sighand == NULL)) {
863                         /*
864                          * The process has been reaped.
865                          * We can't even collect a sample any more.
866                          * Call the timer disarmed, nothing else to do.
867                          */
868                         put_task_struct(p);
869                         timer->it.cpu.task = NULL;
870                         timer->it.cpu.expires.sched = 0;
871                         read_unlock(&tasklist_lock);
872                         goto dead;
873                 } else {
874                         cpu_timer_sample_group(timer->it_clock, p, &now);
875                         clear_dead = (unlikely(p->exit_state) &&
876                                       thread_group_empty(p));
877                 }
878                 read_unlock(&tasklist_lock);
879         }
880
881         if (unlikely(clear_dead)) {
882                 /*
883                  * We've noticed that the thread is dead, but
884                  * not yet reaped.  Take this opportunity to
885                  * drop our task ref.
886                  */
887                 clear_dead_task(timer, now);
888                 goto dead;
889         }
890
891         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
892                 sample_to_timespec(timer->it_clock,
893                                    cpu_time_sub(timer->it_clock,
894                                                 timer->it.cpu.expires, now),
895                                    &itp->it_value);
896         } else {
897                 /*
898                  * The timer should have expired already, but the firing
899                  * hasn't taken place yet.  Say it's just about to expire.
900                  */
901                 itp->it_value.tv_nsec = 1;
902                 itp->it_value.tv_sec = 0;
903         }
904 }
905
906 /*
907  * Check for any per-thread CPU timers that have fired and move them off
908  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
909  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
910  */
911 static void check_thread_timers(struct task_struct *tsk,
912                                 struct list_head *firing)
913 {
914         int maxfire;
915         struct list_head *timers = tsk->cpu_timers;
916         struct signal_struct *const sig = tsk->signal;
917         unsigned long soft;
918
919         maxfire = 20;
920         tsk->cputime_expires.prof_exp = cputime_zero;
921         while (!list_empty(timers)) {
922                 struct cpu_timer_list *t = list_first_entry(timers,
923                                                       struct cpu_timer_list,
924                                                       entry);
925                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
926                         tsk->cputime_expires.prof_exp = t->expires.cpu;
927                         break;
928                 }
929                 t->firing = 1;
930                 list_move_tail(&t->entry, firing);
931         }
932
933         ++timers;
934         maxfire = 20;
935         tsk->cputime_expires.virt_exp = cputime_zero;
936         while (!list_empty(timers)) {
937                 struct cpu_timer_list *t = list_first_entry(timers,
938                                                       struct cpu_timer_list,
939                                                       entry);
940                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
941                         tsk->cputime_expires.virt_exp = t->expires.cpu;
942                         break;
943                 }
944                 t->firing = 1;
945                 list_move_tail(&t->entry, firing);
946         }
947
948         ++timers;
949         maxfire = 20;
950         tsk->cputime_expires.sched_exp = 0;
951         while (!list_empty(timers)) {
952                 struct cpu_timer_list *t = list_first_entry(timers,
953                                                       struct cpu_timer_list,
954                                                       entry);
955                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
956                         tsk->cputime_expires.sched_exp = t->expires.sched;
957                         break;
958                 }
959                 t->firing = 1;
960                 list_move_tail(&t->entry, firing);
961         }
962
963         /*
964          * Check for the special case thread timers.
965          */
966         soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
967         if (soft != RLIM_INFINITY) {
968                 unsigned long hard =
969                         ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
970
971                 if (hard != RLIM_INFINITY &&
972                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
973                         /*
974                          * At the hard limit, we just die.
975                          * No need to calculate anything else now.
976                          */
977                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
978                         return;
979                 }
980                 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
981                         /*
982                          * At the soft limit, send a SIGXCPU every second.
983                          */
984                         if (soft < hard) {
985                                 soft += USEC_PER_SEC;
986                                 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
987                         }
988                         printk(KERN_INFO
989                                 "RT Watchdog Timeout: %s[%d]\n",
990                                 tsk->comm, task_pid_nr(tsk));
991                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
992                 }
993         }
994 }
995
996 static void stop_process_timers(struct signal_struct *sig)
997 {
998         struct thread_group_cputimer *cputimer = &sig->cputimer;
999         unsigned long flags;
1000
1001         spin_lock_irqsave(&cputimer->lock, flags);
1002         cputimer->running = 0;
1003         spin_unlock_irqrestore(&cputimer->lock, flags);
1004 }
1005
1006 static u32 onecputick;
1007
1008 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1009                              cputime_t *expires, cputime_t cur_time, int signo)
1010 {
1011         if (cputime_eq(it->expires, cputime_zero))
1012                 return;
1013
1014         if (cputime_ge(cur_time, it->expires)) {
1015                 if (!cputime_eq(it->incr, cputime_zero)) {
1016                         it->expires = cputime_add(it->expires, it->incr);
1017                         it->error += it->incr_error;
1018                         if (it->error >= onecputick) {
1019                                 it->expires = cputime_sub(it->expires,
1020                                                           cputime_one_jiffy);
1021                                 it->error -= onecputick;
1022                         }
1023                 } else {
1024                         it->expires = cputime_zero;
1025                 }
1026
1027                 trace_itimer_expire(signo == SIGPROF ?
1028                                     ITIMER_PROF : ITIMER_VIRTUAL,
1029                                     tsk->signal->leader_pid, cur_time);
1030                 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1031         }
1032
1033         if (!cputime_eq(it->expires, cputime_zero) &&
1034             (cputime_eq(*expires, cputime_zero) ||
1035              cputime_lt(it->expires, *expires))) {
1036                 *expires = it->expires;
1037         }
1038 }
1039
1040 /**
1041  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1042  *
1043  * @cputime:    The struct to compare.
1044  *
1045  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1046  * are zero, false if any field is nonzero.
1047  */
1048 static inline int task_cputime_zero(const struct task_cputime *cputime)
1049 {
1050         if (cputime_eq(cputime->utime, cputime_zero) &&
1051             cputime_eq(cputime->stime, cputime_zero) &&
1052             cputime->sum_exec_runtime == 0)
1053                 return 1;
1054         return 0;
1055 }
1056
1057 /*
1058  * Check for any per-thread CPU timers that have fired and move them
1059  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1060  * have already been taken off.
1061  */
1062 static void check_process_timers(struct task_struct *tsk,
1063                                  struct list_head *firing)
1064 {
1065         int maxfire;
1066         struct signal_struct *const sig = tsk->signal;
1067         cputime_t utime, ptime, virt_expires, prof_expires;
1068         unsigned long long sum_sched_runtime, sched_expires;
1069         struct list_head *timers = sig->cpu_timers;
1070         struct task_cputime cputime;
1071         unsigned long soft;
1072
1073         /*
1074          * Collect the current process totals.
1075          */
1076         thread_group_cputimer(tsk, &cputime);
1077         utime = cputime.utime;
1078         ptime = cputime_add(utime, cputime.stime);
1079         sum_sched_runtime = cputime.sum_exec_runtime;
1080         maxfire = 20;
1081         prof_expires = cputime_zero;
1082         while (!list_empty(timers)) {
1083                 struct cpu_timer_list *tl = list_first_entry(timers,
1084                                                       struct cpu_timer_list,
1085                                                       entry);
1086                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1087                         prof_expires = tl->expires.cpu;
1088                         break;
1089                 }
1090                 tl->firing = 1;
1091                 list_move_tail(&tl->entry, firing);
1092         }
1093
1094         ++timers;
1095         maxfire = 20;
1096         virt_expires = cputime_zero;
1097         while (!list_empty(timers)) {
1098                 struct cpu_timer_list *tl = list_first_entry(timers,
1099                                                       struct cpu_timer_list,
1100                                                       entry);
1101                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1102                         virt_expires = tl->expires.cpu;
1103                         break;
1104                 }
1105                 tl->firing = 1;
1106                 list_move_tail(&tl->entry, firing);
1107         }
1108
1109         ++timers;
1110         maxfire = 20;
1111         sched_expires = 0;
1112         while (!list_empty(timers)) {
1113                 struct cpu_timer_list *tl = list_first_entry(timers,
1114                                                       struct cpu_timer_list,
1115                                                       entry);
1116                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1117                         sched_expires = tl->expires.sched;
1118                         break;
1119                 }
1120                 tl->firing = 1;
1121                 list_move_tail(&tl->entry, firing);
1122         }
1123
1124         /*
1125          * Check for the special case process timers.
1126          */
1127         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1128                          SIGPROF);
1129         check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1130                          SIGVTALRM);
1131         soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1132         if (soft != RLIM_INFINITY) {
1133                 unsigned long psecs = cputime_to_secs(ptime);
1134                 unsigned long hard =
1135                         ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1136                 cputime_t x;
1137                 if (psecs >= hard) {
1138                         /*
1139                          * At the hard limit, we just die.
1140                          * No need to calculate anything else now.
1141                          */
1142                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1143                         return;
1144                 }
1145                 if (psecs >= soft) {
1146                         /*
1147                          * At the soft limit, send a SIGXCPU every second.
1148                          */
1149                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1150                         if (soft < hard) {
1151                                 soft++;
1152                                 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1153                         }
1154                 }
1155                 x = secs_to_cputime(soft);
1156                 if (cputime_eq(prof_expires, cputime_zero) ||
1157                     cputime_lt(x, prof_expires)) {
1158                         prof_expires = x;
1159                 }
1160         }
1161
1162         sig->cputime_expires.prof_exp = prof_expires;
1163         sig->cputime_expires.virt_exp = virt_expires;
1164         sig->cputime_expires.sched_exp = sched_expires;
1165         if (task_cputime_zero(&sig->cputime_expires))
1166                 stop_process_timers(sig);
1167 }
1168
1169 /*
1170  * This is called from the signal code (via do_schedule_next_timer)
1171  * when the last timer signal was delivered and we have to reload the timer.
1172  */
1173 void posix_cpu_timer_schedule(struct k_itimer *timer)
1174 {
1175         struct task_struct *p = timer->it.cpu.task;
1176         union cpu_time_count now;
1177
1178         if (unlikely(p == NULL))
1179                 /*
1180                  * The task was cleaned up already, no future firings.
1181                  */
1182                 goto out;
1183
1184         /*
1185          * Fetch the current sample and update the timer's expiry time.
1186          */
1187         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1188                 cpu_clock_sample(timer->it_clock, p, &now);
1189                 bump_cpu_timer(timer, now);
1190                 if (unlikely(p->exit_state)) {
1191                         clear_dead_task(timer, now);
1192                         goto out;
1193                 }
1194                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1195                 spin_lock(&p->sighand->siglock);
1196         } else {
1197                 read_lock(&tasklist_lock);
1198                 if (unlikely(p->sighand == NULL)) {
1199                         /*
1200                          * The process has been reaped.
1201                          * We can't even collect a sample any more.
1202                          */
1203                         put_task_struct(p);
1204                         timer->it.cpu.task = p = NULL;
1205                         timer->it.cpu.expires.sched = 0;
1206                         goto out_unlock;
1207                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1208                         /*
1209                          * We've noticed that the thread is dead, but
1210                          * not yet reaped.  Take this opportunity to
1211                          * drop our task ref.
1212                          */
1213                         clear_dead_task(timer, now);
1214                         goto out_unlock;
1215                 }
1216                 spin_lock(&p->sighand->siglock);
1217                 cpu_timer_sample_group(timer->it_clock, p, &now);
1218                 bump_cpu_timer(timer, now);
1219                 /* Leave the tasklist_lock locked for the call below.  */
1220         }
1221
1222         /*
1223          * Now re-arm for the new expiry time.
1224          */
1225         BUG_ON(!irqs_disabled());
1226         arm_timer(timer);
1227         spin_unlock(&p->sighand->siglock);
1228
1229 out_unlock:
1230         read_unlock(&tasklist_lock);
1231
1232 out:
1233         timer->it_overrun_last = timer->it_overrun;
1234         timer->it_overrun = -1;
1235         ++timer->it_requeue_pending;
1236 }
1237
1238 /**
1239  * task_cputime_expired - Compare two task_cputime entities.
1240  *
1241  * @sample:     The task_cputime structure to be checked for expiration.
1242  * @expires:    Expiration times, against which @sample will be checked.
1243  *
1244  * Checks @sample against @expires to see if any field of @sample has expired.
1245  * Returns true if any field of the former is greater than the corresponding
1246  * field of the latter if the latter field is set.  Otherwise returns false.
1247  */
1248 static inline int task_cputime_expired(const struct task_cputime *sample,
1249                                         const struct task_cputime *expires)
1250 {
1251         if (!cputime_eq(expires->utime, cputime_zero) &&
1252             cputime_ge(sample->utime, expires->utime))
1253                 return 1;
1254         if (!cputime_eq(expires->stime, cputime_zero) &&
1255             cputime_ge(cputime_add(sample->utime, sample->stime),
1256                        expires->stime))
1257                 return 1;
1258         if (expires->sum_exec_runtime != 0 &&
1259             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1260                 return 1;
1261         return 0;
1262 }
1263
1264 /**
1265  * fastpath_timer_check - POSIX CPU timers fast path.
1266  *
1267  * @tsk:        The task (thread) being checked.
1268  *
1269  * Check the task and thread group timers.  If both are zero (there are no
1270  * timers set) return false.  Otherwise snapshot the task and thread group
1271  * timers and compare them with the corresponding expiration times.  Return
1272  * true if a timer has expired, else return false.
1273  */
1274 static inline int fastpath_timer_check(struct task_struct *tsk)
1275 {
1276         struct signal_struct *sig;
1277
1278         if (!task_cputime_zero(&tsk->cputime_expires)) {
1279                 struct task_cputime task_sample = {
1280                         .utime = tsk->utime,
1281                         .stime = tsk->stime,
1282                         .sum_exec_runtime = tsk->se.sum_exec_runtime
1283                 };
1284
1285                 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1286                         return 1;
1287         }
1288
1289         sig = tsk->signal;
1290         if (sig->cputimer.running) {
1291                 struct task_cputime group_sample;
1292
1293                 spin_lock(&sig->cputimer.lock);
1294                 group_sample = sig->cputimer.cputime;
1295                 spin_unlock(&sig->cputimer.lock);
1296
1297                 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1298                         return 1;
1299         }
1300
1301         return 0;
1302 }
1303
1304 /*
1305  * This is called from the timer interrupt handler.  The irq handler has
1306  * already updated our counts.  We need to check if any timers fire now.
1307  * Interrupts are disabled.
1308  */
1309 void run_posix_cpu_timers(struct task_struct *tsk)
1310 {
1311         LIST_HEAD(firing);
1312         struct k_itimer *timer, *next;
1313         unsigned long flags;
1314
1315         BUG_ON(!irqs_disabled());
1316
1317         /*
1318          * The fast path checks that there are no expired thread or thread
1319          * group timers.  If that's so, just return.
1320          */
1321         if (!fastpath_timer_check(tsk))
1322                 return;
1323
1324         if (!lock_task_sighand(tsk, &flags))
1325                 return;
1326         /*
1327          * Here we take off tsk->signal->cpu_timers[N] and
1328          * tsk->cpu_timers[N] all the timers that are firing, and
1329          * put them on the firing list.
1330          */
1331         check_thread_timers(tsk, &firing);
1332         /*
1333          * If there are any active process wide timers (POSIX 1.b, itimers,
1334          * RLIMIT_CPU) cputimer must be running.
1335          */
1336         if (tsk->signal->cputimer.running)
1337                 check_process_timers(tsk, &firing);
1338
1339         /*
1340          * We must release these locks before taking any timer's lock.
1341          * There is a potential race with timer deletion here, as the
1342          * siglock now protects our private firing list.  We have set
1343          * the firing flag in each timer, so that a deletion attempt
1344          * that gets the timer lock before we do will give it up and
1345          * spin until we've taken care of that timer below.
1346          */
1347         unlock_task_sighand(tsk, &flags);
1348
1349         /*
1350          * Now that all the timers on our list have the firing flag,
1351          * no one will touch their list entries but us.  We'll take
1352          * each timer's lock before clearing its firing flag, so no
1353          * timer call will interfere.
1354          */
1355         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1356                 int cpu_firing;
1357
1358                 spin_lock(&timer->it_lock);
1359                 list_del_init(&timer->it.cpu.entry);
1360                 cpu_firing = timer->it.cpu.firing;
1361                 timer->it.cpu.firing = 0;
1362                 /*
1363                  * The firing flag is -1 if we collided with a reset
1364                  * of the timer, which already reported this
1365                  * almost-firing as an overrun.  So don't generate an event.
1366                  */
1367                 if (likely(cpu_firing >= 0))
1368                         cpu_timer_fire(timer);
1369                 spin_unlock(&timer->it_lock);
1370         }
1371 }
1372
1373 /*
1374  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1375  * The tsk->sighand->siglock must be held by the caller.
1376  */
1377 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1378                            cputime_t *newval, cputime_t *oldval)
1379 {
1380         union cpu_time_count now;
1381
1382         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1383         cpu_timer_sample_group(clock_idx, tsk, &now);
1384
1385         if (oldval) {
1386                 /*
1387                  * We are setting itimer. The *oldval is absolute and we update
1388                  * it to be relative, *newval argument is relative and we update
1389                  * it to be absolute.
1390                  */
1391                 if (!cputime_eq(*oldval, cputime_zero)) {
1392                         if (cputime_le(*oldval, now.cpu)) {
1393                                 /* Just about to fire. */
1394                                 *oldval = cputime_one_jiffy;
1395                         } else {
1396                                 *oldval = cputime_sub(*oldval, now.cpu);
1397                         }
1398                 }
1399
1400                 if (cputime_eq(*newval, cputime_zero))
1401                         return;
1402                 *newval = cputime_add(*newval, now.cpu);
1403         }
1404
1405         /*
1406          * Update expiration cache if we are the earliest timer, or eventually
1407          * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1408          */
1409         switch (clock_idx) {
1410         case CPUCLOCK_PROF:
1411                 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1412                         tsk->signal->cputime_expires.prof_exp = *newval;
1413                 break;
1414         case CPUCLOCK_VIRT:
1415                 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1416                         tsk->signal->cputime_expires.virt_exp = *newval;
1417                 break;
1418         }
1419 }
1420
1421 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1422                             struct timespec *rqtp, struct itimerspec *it)
1423 {
1424         struct k_itimer timer;
1425         int error;
1426
1427         /*
1428          * Set up a temporary timer and then wait for it to go off.
1429          */
1430         memset(&timer, 0, sizeof timer);
1431         spin_lock_init(&timer.it_lock);
1432         timer.it_clock = which_clock;
1433         timer.it_overrun = -1;
1434         error = posix_cpu_timer_create(&timer);
1435         timer.it_process = current;
1436         if (!error) {
1437                 static struct itimerspec zero_it;
1438
1439                 memset(it, 0, sizeof *it);
1440                 it->it_value = *rqtp;
1441
1442                 spin_lock_irq(&timer.it_lock);
1443                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1444                 if (error) {
1445                         spin_unlock_irq(&timer.it_lock);
1446                         return error;
1447                 }
1448
1449                 while (!signal_pending(current)) {
1450                         if (timer.it.cpu.expires.sched == 0) {
1451                                 /*
1452                                  * Our timer fired and was reset.
1453                                  */
1454                                 spin_unlock_irq(&timer.it_lock);
1455                                 return 0;
1456                         }
1457
1458                         /*
1459                          * Block until cpu_timer_fire (or a signal) wakes us.
1460                          */
1461                         __set_current_state(TASK_INTERRUPTIBLE);
1462                         spin_unlock_irq(&timer.it_lock);
1463                         schedule();
1464                         spin_lock_irq(&timer.it_lock);
1465                 }
1466
1467                 /*
1468                  * We were interrupted by a signal.
1469                  */
1470                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1471                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1472                 spin_unlock_irq(&timer.it_lock);
1473
1474                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1475                         /*
1476                          * It actually did fire already.
1477                          */
1478                         return 0;
1479                 }
1480
1481                 error = -ERESTART_RESTARTBLOCK;
1482         }
1483
1484         return error;
1485 }
1486
1487 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1488
1489 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1490                             struct timespec *rqtp, struct timespec __user *rmtp)
1491 {
1492         struct restart_block *restart_block =
1493                 &current_thread_info()->restart_block;
1494         struct itimerspec it;
1495         int error;
1496
1497         /*
1498          * Diagnose required errors first.
1499          */
1500         if (CPUCLOCK_PERTHREAD(which_clock) &&
1501             (CPUCLOCK_PID(which_clock) == 0 ||
1502              CPUCLOCK_PID(which_clock) == current->pid))
1503                 return -EINVAL;
1504
1505         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1506
1507         if (error == -ERESTART_RESTARTBLOCK) {
1508
1509                 if (flags & TIMER_ABSTIME)
1510                         return -ERESTARTNOHAND;
1511                 /*
1512                  * Report back to the user the time still remaining.
1513                  */
1514                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1515                         return -EFAULT;
1516
1517                 restart_block->fn = posix_cpu_nsleep_restart;
1518                 restart_block->nanosleep.clockid = which_clock;
1519                 restart_block->nanosleep.rmtp = rmtp;
1520                 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1521         }
1522         return error;
1523 }
1524
1525 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1526 {
1527         clockid_t which_clock = restart_block->nanosleep.clockid;
1528         struct timespec t;
1529         struct itimerspec it;
1530         int error;
1531
1532         t = ns_to_timespec(restart_block->nanosleep.expires);
1533
1534         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1535
1536         if (error == -ERESTART_RESTARTBLOCK) {
1537                 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1538                 /*
1539                  * Report back to the user the time still remaining.
1540                  */
1541                 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1542                         return -EFAULT;
1543
1544                 restart_block->nanosleep.expires = timespec_to_ns(&t);
1545         }
1546         return error;
1547
1548 }
1549
1550 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1551 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1552
1553 static int process_cpu_clock_getres(const clockid_t which_clock,
1554                                     struct timespec *tp)
1555 {
1556         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1557 }
1558 static int process_cpu_clock_get(const clockid_t which_clock,
1559                                  struct timespec *tp)
1560 {
1561         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1562 }
1563 static int process_cpu_timer_create(struct k_itimer *timer)
1564 {
1565         timer->it_clock = PROCESS_CLOCK;
1566         return posix_cpu_timer_create(timer);
1567 }
1568 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1569                               struct timespec *rqtp,
1570                               struct timespec __user *rmtp)
1571 {
1572         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1573 }
1574 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1575 {
1576         return -EINVAL;
1577 }
1578 static int thread_cpu_clock_getres(const clockid_t which_clock,
1579                                    struct timespec *tp)
1580 {
1581         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1582 }
1583 static int thread_cpu_clock_get(const clockid_t which_clock,
1584                                 struct timespec *tp)
1585 {
1586         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1587 }
1588 static int thread_cpu_timer_create(struct k_itimer *timer)
1589 {
1590         timer->it_clock = THREAD_CLOCK;
1591         return posix_cpu_timer_create(timer);
1592 }
1593
1594 struct k_clock clock_posix_cpu = {
1595         .clock_getres   = posix_cpu_clock_getres,
1596         .clock_set      = posix_cpu_clock_set,
1597         .clock_get      = posix_cpu_clock_get,
1598         .timer_create   = posix_cpu_timer_create,
1599         .nsleep         = posix_cpu_nsleep,
1600         .nsleep_restart = posix_cpu_nsleep_restart,
1601         .timer_set      = posix_cpu_timer_set,
1602         .timer_del      = posix_cpu_timer_del,
1603         .timer_get      = posix_cpu_timer_get,
1604 };
1605
1606 static __init int init_posix_cpu_timers(void)
1607 {
1608         struct k_clock process = {
1609                 .clock_getres   = process_cpu_clock_getres,
1610                 .clock_get      = process_cpu_clock_get,
1611                 .timer_create   = process_cpu_timer_create,
1612                 .nsleep         = process_cpu_nsleep,
1613                 .nsleep_restart = process_cpu_nsleep_restart,
1614         };
1615         struct k_clock thread = {
1616                 .clock_getres   = thread_cpu_clock_getres,
1617                 .clock_get      = thread_cpu_clock_get,
1618                 .timer_create   = thread_cpu_timer_create,
1619         };
1620         struct timespec ts;
1621
1622         posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1623         posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1624
1625         cputime_to_timespec(cputime_one_jiffy, &ts);
1626         onecputick = ts.tv_nsec;
1627         WARN_ON(ts.tv_sec != 0);
1628
1629         return 0;
1630 }
1631 __initcall(init_posix_cpu_timers);