timers: fix itimer/many thread hang
[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
11 #ifdef CONFIG_SMP
12 /*
13  * Allocate the thread_group_cputime structure appropriately for SMP kernels
14  * and fill in the current values of the fields.  Called from copy_signal()
15  * via thread_group_cputime_clone_thread() when adding a second or subsequent
16  * thread to a thread group.  Assumes interrupts are enabled when called.
17  */
18 int thread_group_cputime_alloc_smp(struct task_struct *tsk)
19 {
20         struct signal_struct *sig = tsk->signal;
21         struct task_cputime *cputime;
22
23         /*
24          * If we have multiple threads and we don't already have a
25          * per-CPU task_cputime struct, allocate one and fill it in with
26          * the times accumulated so far.
27          */
28         if (sig->cputime.totals)
29                 return 0;
30         cputime = alloc_percpu(struct task_cputime);
31         if (cputime == NULL)
32                 return -ENOMEM;
33         read_lock(&tasklist_lock);
34         spin_lock_irq(&tsk->sighand->siglock);
35         if (sig->cputime.totals) {
36                 spin_unlock_irq(&tsk->sighand->siglock);
37                 read_unlock(&tasklist_lock);
38                 free_percpu(cputime);
39                 return 0;
40         }
41         sig->cputime.totals = cputime;
42         cputime = per_cpu_ptr(sig->cputime.totals, get_cpu());
43         cputime->utime = tsk->utime;
44         cputime->stime = tsk->stime;
45         cputime->sum_exec_runtime = tsk->se.sum_exec_runtime;
46         put_cpu_no_resched();
47         spin_unlock_irq(&tsk->sighand->siglock);
48         read_unlock(&tasklist_lock);
49         return 0;
50 }
51
52 /**
53  * thread_group_cputime_smp - Sum the thread group time fields across all CPUs.
54  *
55  * @tsk:        The task we use to identify the thread group.
56  * @times:      task_cputime structure in which we return the summed fields.
57  *
58  * Walk the list of CPUs to sum the per-CPU time fields in the thread group
59  * time structure.
60  */
61 void thread_group_cputime_smp(
62         struct task_struct *tsk,
63         struct task_cputime *times)
64 {
65         struct signal_struct *sig;
66         int i;
67         struct task_cputime *tot;
68
69         sig = tsk->signal;
70         if (unlikely(!sig) || !sig->cputime.totals) {
71                 times->utime = tsk->utime;
72                 times->stime = tsk->stime;
73                 times->sum_exec_runtime = tsk->se.sum_exec_runtime;
74                 return;
75         }
76         times->stime = times->utime = cputime_zero;
77         times->sum_exec_runtime = 0;
78         for_each_possible_cpu(i) {
79                 tot = per_cpu_ptr(tsk->signal->cputime.totals, i);
80                 times->utime = cputime_add(times->utime, tot->utime);
81                 times->stime = cputime_add(times->stime, tot->stime);
82                 times->sum_exec_runtime += tot->sum_exec_runtime;
83         }
84 }
85
86 #endif /* CONFIG_SMP */
87
88 /*
89  * Called after updating RLIMIT_CPU to set timer expiration if necessary.
90  */
91 void update_rlimit_cpu(unsigned long rlim_new)
92 {
93         cputime_t cputime;
94
95         cputime = secs_to_cputime(rlim_new);
96         if (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
97             cputime_lt(current->signal->it_prof_expires, cputime)) {
98                 spin_lock_irq(&current->sighand->siglock);
99                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
100                 spin_unlock_irq(&current->sighand->siglock);
101         }
102 }
103
104 static int check_clock(const clockid_t which_clock)
105 {
106         int error = 0;
107         struct task_struct *p;
108         const pid_t pid = CPUCLOCK_PID(which_clock);
109
110         if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
111                 return -EINVAL;
112
113         if (pid == 0)
114                 return 0;
115
116         read_lock(&tasklist_lock);
117         p = find_task_by_vpid(pid);
118         if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
119                    same_thread_group(p, current) : thread_group_leader(p))) {
120                 error = -EINVAL;
121         }
122         read_unlock(&tasklist_lock);
123
124         return error;
125 }
126
127 static inline union cpu_time_count
128 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
129 {
130         union cpu_time_count ret;
131         ret.sched = 0;          /* high half always zero when .cpu used */
132         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
133                 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
134         } else {
135                 ret.cpu = timespec_to_cputime(tp);
136         }
137         return ret;
138 }
139
140 static void sample_to_timespec(const clockid_t which_clock,
141                                union cpu_time_count cpu,
142                                struct timespec *tp)
143 {
144         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
145                 *tp = ns_to_timespec(cpu.sched);
146         else
147                 cputime_to_timespec(cpu.cpu, tp);
148 }
149
150 static inline int cpu_time_before(const clockid_t which_clock,
151                                   union cpu_time_count now,
152                                   union cpu_time_count then)
153 {
154         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
155                 return now.sched < then.sched;
156         }  else {
157                 return cputime_lt(now.cpu, then.cpu);
158         }
159 }
160 static inline void cpu_time_add(const clockid_t which_clock,
161                                 union cpu_time_count *acc,
162                                 union cpu_time_count val)
163 {
164         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
165                 acc->sched += val.sched;
166         }  else {
167                 acc->cpu = cputime_add(acc->cpu, val.cpu);
168         }
169 }
170 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
171                                                 union cpu_time_count a,
172                                                 union cpu_time_count b)
173 {
174         if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
175                 a.sched -= b.sched;
176         }  else {
177                 a.cpu = cputime_sub(a.cpu, b.cpu);
178         }
179         return a;
180 }
181
182 /*
183  * Divide and limit the result to res >= 1
184  *
185  * This is necessary to prevent signal delivery starvation, when the result of
186  * the division would be rounded down to 0.
187  */
188 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
189 {
190         cputime_t res = cputime_div(time, div);
191
192         return max_t(cputime_t, res, 1);
193 }
194
195 /*
196  * Update expiry time from increment, and increase overrun count,
197  * given the current clock sample.
198  */
199 static void bump_cpu_timer(struct k_itimer *timer,
200                                   union cpu_time_count now)
201 {
202         int i;
203
204         if (timer->it.cpu.incr.sched == 0)
205                 return;
206
207         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
208                 unsigned long long delta, incr;
209
210                 if (now.sched < timer->it.cpu.expires.sched)
211                         return;
212                 incr = timer->it.cpu.incr.sched;
213                 delta = now.sched + incr - timer->it.cpu.expires.sched;
214                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
215                 for (i = 0; incr < delta - incr; i++)
216                         incr = incr << 1;
217                 for (; i >= 0; incr >>= 1, i--) {
218                         if (delta < incr)
219                                 continue;
220                         timer->it.cpu.expires.sched += incr;
221                         timer->it_overrun += 1 << i;
222                         delta -= incr;
223                 }
224         } else {
225                 cputime_t delta, incr;
226
227                 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
228                         return;
229                 incr = timer->it.cpu.incr.cpu;
230                 delta = cputime_sub(cputime_add(now.cpu, incr),
231                                     timer->it.cpu.expires.cpu);
232                 /* Don't use (incr*2 < delta), incr*2 might overflow. */
233                 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
234                              incr = cputime_add(incr, incr);
235                 for (; i >= 0; incr = cputime_halve(incr), i--) {
236                         if (cputime_lt(delta, incr))
237                                 continue;
238                         timer->it.cpu.expires.cpu =
239                                 cputime_add(timer->it.cpu.expires.cpu, incr);
240                         timer->it_overrun += 1 << i;
241                         delta = cputime_sub(delta, incr);
242                 }
243         }
244 }
245
246 static inline cputime_t prof_ticks(struct task_struct *p)
247 {
248         return cputime_add(p->utime, p->stime);
249 }
250 static inline cputime_t virt_ticks(struct task_struct *p)
251 {
252         return p->utime;
253 }
254
255 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
256 {
257         int error = check_clock(which_clock);
258         if (!error) {
259                 tp->tv_sec = 0;
260                 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
261                 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
262                         /*
263                          * If sched_clock is using a cycle counter, we
264                          * don't have any idea of its true resolution
265                          * exported, but it is much more than 1s/HZ.
266                          */
267                         tp->tv_nsec = 1;
268                 }
269         }
270         return error;
271 }
272
273 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
274 {
275         /*
276          * You can never reset a CPU clock, but we check for other errors
277          * in the call before failing with EPERM.
278          */
279         int error = check_clock(which_clock);
280         if (error == 0) {
281                 error = -EPERM;
282         }
283         return error;
284 }
285
286
287 /*
288  * Sample a per-thread clock for the given task.
289  */
290 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
291                             union cpu_time_count *cpu)
292 {
293         switch (CPUCLOCK_WHICH(which_clock)) {
294         default:
295                 return -EINVAL;
296         case CPUCLOCK_PROF:
297                 cpu->cpu = prof_ticks(p);
298                 break;
299         case CPUCLOCK_VIRT:
300                 cpu->cpu = virt_ticks(p);
301                 break;
302         case CPUCLOCK_SCHED:
303                 cpu->sched = task_sched_runtime(p);
304                 break;
305         }
306         return 0;
307 }
308
309 /*
310  * Sample a process (thread group) clock for the given group_leader task.
311  * Must be called with tasklist_lock held for reading.
312  * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
313  */
314 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
315                                          struct task_struct *p,
316                                          union cpu_time_count *cpu)
317 {
318         struct task_cputime cputime;
319
320         thread_group_cputime(p, &cputime);
321         switch (clock_idx) {
322         default:
323                 return -EINVAL;
324         case CPUCLOCK_PROF:
325                 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
326                 break;
327         case CPUCLOCK_VIRT:
328                 cpu->cpu = cputime.utime;
329                 break;
330         case CPUCLOCK_SCHED:
331                 cpu->sched = thread_group_sched_runtime(p);
332                 break;
333         }
334         return 0;
335 }
336
337 /*
338  * Sample a process (thread group) clock for the given group_leader task.
339  * Must be called with tasklist_lock held for reading.
340  */
341 static int cpu_clock_sample_group(const clockid_t which_clock,
342                                   struct task_struct *p,
343                                   union cpu_time_count *cpu)
344 {
345         int ret;
346         unsigned long flags;
347         spin_lock_irqsave(&p->sighand->siglock, flags);
348         ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
349                                             cpu);
350         spin_unlock_irqrestore(&p->sighand->siglock, flags);
351         return ret;
352 }
353
354
355 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
356 {
357         const pid_t pid = CPUCLOCK_PID(which_clock);
358         int error = -EINVAL;
359         union cpu_time_count rtn;
360
361         if (pid == 0) {
362                 /*
363                  * Special case constant value for our own clocks.
364                  * We don't have to do any lookup to find ourselves.
365                  */
366                 if (CPUCLOCK_PERTHREAD(which_clock)) {
367                         /*
368                          * Sampling just ourselves we can do with no locking.
369                          */
370                         error = cpu_clock_sample(which_clock,
371                                                  current, &rtn);
372                 } else {
373                         read_lock(&tasklist_lock);
374                         error = cpu_clock_sample_group(which_clock,
375                                                        current, &rtn);
376                         read_unlock(&tasklist_lock);
377                 }
378         } else {
379                 /*
380                  * Find the given PID, and validate that the caller
381                  * should be able to see it.
382                  */
383                 struct task_struct *p;
384                 rcu_read_lock();
385                 p = find_task_by_vpid(pid);
386                 if (p) {
387                         if (CPUCLOCK_PERTHREAD(which_clock)) {
388                                 if (same_thread_group(p, current)) {
389                                         error = cpu_clock_sample(which_clock,
390                                                                  p, &rtn);
391                                 }
392                         } else {
393                                 read_lock(&tasklist_lock);
394                                 if (thread_group_leader(p) && p->signal) {
395                                         error =
396                                             cpu_clock_sample_group(which_clock,
397                                                                    p, &rtn);
398                                 }
399                                 read_unlock(&tasklist_lock);
400                         }
401                 }
402                 rcu_read_unlock();
403         }
404
405         if (error)
406                 return error;
407         sample_to_timespec(which_clock, rtn, tp);
408         return 0;
409 }
410
411
412 /*
413  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
414  * This is called from sys_timer_create with the new timer already locked.
415  */
416 int posix_cpu_timer_create(struct k_itimer *new_timer)
417 {
418         int ret = 0;
419         const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
420         struct task_struct *p;
421
422         if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
423                 return -EINVAL;
424
425         INIT_LIST_HEAD(&new_timer->it.cpu.entry);
426         new_timer->it.cpu.incr.sched = 0;
427         new_timer->it.cpu.expires.sched = 0;
428
429         read_lock(&tasklist_lock);
430         if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
431                 if (pid == 0) {
432                         p = current;
433                 } else {
434                         p = find_task_by_vpid(pid);
435                         if (p && !same_thread_group(p, current))
436                                 p = NULL;
437                 }
438         } else {
439                 if (pid == 0) {
440                         p = current->group_leader;
441                 } else {
442                         p = find_task_by_vpid(pid);
443                         if (p && !thread_group_leader(p))
444                                 p = NULL;
445                 }
446         }
447         new_timer->it.cpu.task = p;
448         if (p) {
449                 get_task_struct(p);
450         } else {
451                 ret = -EINVAL;
452         }
453         read_unlock(&tasklist_lock);
454
455         return ret;
456 }
457
458 /*
459  * Clean up a CPU-clock timer that is about to be destroyed.
460  * This is called from timer deletion with the timer already locked.
461  * If we return TIMER_RETRY, it's necessary to release the timer's lock
462  * and try again.  (This happens when the timer is in the middle of firing.)
463  */
464 int posix_cpu_timer_del(struct k_itimer *timer)
465 {
466         struct task_struct *p = timer->it.cpu.task;
467         int ret = 0;
468
469         if (likely(p != NULL)) {
470                 read_lock(&tasklist_lock);
471                 if (unlikely(p->signal == NULL)) {
472                         /*
473                          * We raced with the reaping of the task.
474                          * The deletion should have cleared us off the list.
475                          */
476                         BUG_ON(!list_empty(&timer->it.cpu.entry));
477                 } else {
478                         spin_lock(&p->sighand->siglock);
479                         if (timer->it.cpu.firing)
480                                 ret = TIMER_RETRY;
481                         else
482                                 list_del(&timer->it.cpu.entry);
483                         spin_unlock(&p->sighand->siglock);
484                 }
485                 read_unlock(&tasklist_lock);
486
487                 if (!ret)
488                         put_task_struct(p);
489         }
490
491         return ret;
492 }
493
494 /*
495  * Clean out CPU timers still ticking when a thread exited.  The task
496  * pointer is cleared, and the expiry time is replaced with the residual
497  * time for later timer_gettime calls to return.
498  * This must be called with the siglock held.
499  */
500 static void cleanup_timers(struct list_head *head,
501                            cputime_t utime, cputime_t stime,
502                            unsigned long long sum_exec_runtime)
503 {
504         struct cpu_timer_list *timer, *next;
505         cputime_t ptime = cputime_add(utime, stime);
506
507         list_for_each_entry_safe(timer, next, head, entry) {
508                 list_del_init(&timer->entry);
509                 if (cputime_lt(timer->expires.cpu, ptime)) {
510                         timer->expires.cpu = cputime_zero;
511                 } else {
512                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
513                                                          ptime);
514                 }
515         }
516
517         ++head;
518         list_for_each_entry_safe(timer, next, head, entry) {
519                 list_del_init(&timer->entry);
520                 if (cputime_lt(timer->expires.cpu, utime)) {
521                         timer->expires.cpu = cputime_zero;
522                 } else {
523                         timer->expires.cpu = cputime_sub(timer->expires.cpu,
524                                                          utime);
525                 }
526         }
527
528         ++head;
529         list_for_each_entry_safe(timer, next, head, entry) {
530                 list_del_init(&timer->entry);
531                 if (timer->expires.sched < sum_exec_runtime) {
532                         timer->expires.sched = 0;
533                 } else {
534                         timer->expires.sched -= sum_exec_runtime;
535                 }
536         }
537 }
538
539 /*
540  * These are both called with the siglock held, when the current thread
541  * is being reaped.  When the final (leader) thread in the group is reaped,
542  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
543  */
544 void posix_cpu_timers_exit(struct task_struct *tsk)
545 {
546         cleanup_timers(tsk->cpu_timers,
547                        tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
548
549 }
550 void posix_cpu_timers_exit_group(struct task_struct *tsk)
551 {
552         struct task_cputime cputime;
553
554         thread_group_cputime(tsk, &cputime);
555         cleanup_timers(tsk->signal->cpu_timers,
556                        cputime.utime, cputime.stime, cputime.sum_exec_runtime);
557 }
558
559 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
560 {
561         /*
562          * That's all for this thread or process.
563          * We leave our residual in expires to be reported.
564          */
565         put_task_struct(timer->it.cpu.task);
566         timer->it.cpu.task = NULL;
567         timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
568                                              timer->it.cpu.expires,
569                                              now);
570 }
571
572 /*
573  * Insert the timer on the appropriate list before any timers that
574  * expire later.  This must be called with the tasklist_lock held
575  * for reading, and interrupts disabled.
576  */
577 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
578 {
579         struct task_struct *p = timer->it.cpu.task;
580         struct list_head *head, *listpos;
581         struct cpu_timer_list *const nt = &timer->it.cpu;
582         struct cpu_timer_list *next;
583         unsigned long i;
584
585         head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
586                 p->cpu_timers : p->signal->cpu_timers);
587         head += CPUCLOCK_WHICH(timer->it_clock);
588
589         BUG_ON(!irqs_disabled());
590         spin_lock(&p->sighand->siglock);
591
592         listpos = head;
593         if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
594                 list_for_each_entry(next, head, entry) {
595                         if (next->expires.sched > nt->expires.sched)
596                                 break;
597                         listpos = &next->entry;
598                 }
599         } else {
600                 list_for_each_entry(next, head, entry) {
601                         if (cputime_gt(next->expires.cpu, nt->expires.cpu))
602                                 break;
603                         listpos = &next->entry;
604                 }
605         }
606         list_add(&nt->entry, listpos);
607
608         if (listpos == head) {
609                 /*
610                  * We are the new earliest-expiring timer.
611                  * If we are a thread timer, there can always
612                  * be a process timer telling us to stop earlier.
613                  */
614
615                 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
616                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
617                         default:
618                                 BUG();
619                         case CPUCLOCK_PROF:
620                                 if (cputime_eq(p->cputime_expires.prof_exp,
621                                                cputime_zero) ||
622                                     cputime_gt(p->cputime_expires.prof_exp,
623                                                nt->expires.cpu))
624                                         p->cputime_expires.prof_exp =
625                                                 nt->expires.cpu;
626                                 break;
627                         case CPUCLOCK_VIRT:
628                                 if (cputime_eq(p->cputime_expires.virt_exp,
629                                                cputime_zero) ||
630                                     cputime_gt(p->cputime_expires.virt_exp,
631                                                nt->expires.cpu))
632                                         p->cputime_expires.virt_exp =
633                                                 nt->expires.cpu;
634                                 break;
635                         case CPUCLOCK_SCHED:
636                                 if (p->cputime_expires.sched_exp == 0 ||
637                                     p->cputime_expires.sched_exp >
638                                                         nt->expires.sched)
639                                         p->cputime_expires.sched_exp =
640                                                 nt->expires.sched;
641                                 break;
642                         }
643                 } else {
644                         /*
645                          * For a process timer, set the cached expiration time.
646                          */
647                         switch (CPUCLOCK_WHICH(timer->it_clock)) {
648                         default:
649                                 BUG();
650                         case CPUCLOCK_VIRT:
651                                 if (!cputime_eq(p->signal->it_virt_expires,
652                                                 cputime_zero) &&
653                                     cputime_lt(p->signal->it_virt_expires,
654                                                timer->it.cpu.expires.cpu))
655                                         break;
656                                 p->signal->cputime_expires.virt_exp =
657                                         timer->it.cpu.expires.cpu;
658                                 break;
659                         case CPUCLOCK_PROF:
660                                 if (!cputime_eq(p->signal->it_prof_expires,
661                                                 cputime_zero) &&
662                                     cputime_lt(p->signal->it_prof_expires,
663                                                timer->it.cpu.expires.cpu))
664                                         break;
665                                 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
666                                 if (i != RLIM_INFINITY &&
667                                     i <= cputime_to_secs(timer->it.cpu.expires.cpu))
668                                         break;
669                                 p->signal->cputime_expires.prof_exp =
670                                         timer->it.cpu.expires.cpu;
671                                 break;
672                         case CPUCLOCK_SCHED:
673                                 p->signal->cputime_expires.sched_exp =
674                                         timer->it.cpu.expires.sched;
675                                 break;
676                         }
677                 }
678         }
679
680         spin_unlock(&p->sighand->siglock);
681 }
682
683 /*
684  * The timer is locked, fire it and arrange for its reload.
685  */
686 static void cpu_timer_fire(struct k_itimer *timer)
687 {
688         if (unlikely(timer->sigq == NULL)) {
689                 /*
690                  * This a special case for clock_nanosleep,
691                  * not a normal timer from sys_timer_create.
692                  */
693                 wake_up_process(timer->it_process);
694                 timer->it.cpu.expires.sched = 0;
695         } else if (timer->it.cpu.incr.sched == 0) {
696                 /*
697                  * One-shot timer.  Clear it as soon as it's fired.
698                  */
699                 posix_timer_event(timer, 0);
700                 timer->it.cpu.expires.sched = 0;
701         } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
702                 /*
703                  * The signal did not get queued because the signal
704                  * was ignored, so we won't get any callback to
705                  * reload the timer.  But we need to keep it
706                  * ticking in case the signal is deliverable next time.
707                  */
708                 posix_cpu_timer_schedule(timer);
709         }
710 }
711
712 /*
713  * Guts of sys_timer_settime for CPU timers.
714  * This is called with the timer locked and interrupts disabled.
715  * If we return TIMER_RETRY, it's necessary to release the timer's lock
716  * and try again.  (This happens when the timer is in the middle of firing.)
717  */
718 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
719                         struct itimerspec *new, struct itimerspec *old)
720 {
721         struct task_struct *p = timer->it.cpu.task;
722         union cpu_time_count old_expires, new_expires, val;
723         int ret;
724
725         if (unlikely(p == NULL)) {
726                 /*
727                  * Timer refers to a dead task's clock.
728                  */
729                 return -ESRCH;
730         }
731
732         new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
733
734         read_lock(&tasklist_lock);
735         /*
736          * We need the tasklist_lock to protect against reaping that
737          * clears p->signal.  If p has just been reaped, we can no
738          * longer get any information about it at all.
739          */
740         if (unlikely(p->signal == NULL)) {
741                 read_unlock(&tasklist_lock);
742                 put_task_struct(p);
743                 timer->it.cpu.task = NULL;
744                 return -ESRCH;
745         }
746
747         /*
748          * Disarm any old timer after extracting its expiry time.
749          */
750         BUG_ON(!irqs_disabled());
751
752         ret = 0;
753         spin_lock(&p->sighand->siglock);
754         old_expires = timer->it.cpu.expires;
755         if (unlikely(timer->it.cpu.firing)) {
756                 timer->it.cpu.firing = -1;
757                 ret = TIMER_RETRY;
758         } else
759                 list_del_init(&timer->it.cpu.entry);
760         spin_unlock(&p->sighand->siglock);
761
762         /*
763          * We need to sample the current value to convert the new
764          * value from to relative and absolute, and to convert the
765          * old value from absolute to relative.  To set a process
766          * timer, we need a sample to balance the thread expiry
767          * times (in arm_timer).  With an absolute time, we must
768          * check if it's already passed.  In short, we need a sample.
769          */
770         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
771                 cpu_clock_sample(timer->it_clock, p, &val);
772         } else {
773                 cpu_clock_sample_group(timer->it_clock, p, &val);
774         }
775
776         if (old) {
777                 if (old_expires.sched == 0) {
778                         old->it_value.tv_sec = 0;
779                         old->it_value.tv_nsec = 0;
780                 } else {
781                         /*
782                          * Update the timer in case it has
783                          * overrun already.  If it has,
784                          * we'll report it as having overrun
785                          * and with the next reloaded timer
786                          * already ticking, though we are
787                          * swallowing that pending
788                          * notification here to install the
789                          * new setting.
790                          */
791                         bump_cpu_timer(timer, val);
792                         if (cpu_time_before(timer->it_clock, val,
793                                             timer->it.cpu.expires)) {
794                                 old_expires = cpu_time_sub(
795                                         timer->it_clock,
796                                         timer->it.cpu.expires, val);
797                                 sample_to_timespec(timer->it_clock,
798                                                    old_expires,
799                                                    &old->it_value);
800                         } else {
801                                 old->it_value.tv_nsec = 1;
802                                 old->it_value.tv_sec = 0;
803                         }
804                 }
805         }
806
807         if (unlikely(ret)) {
808                 /*
809                  * We are colliding with the timer actually firing.
810                  * Punt after filling in the timer's old value, and
811                  * disable this firing since we are already reporting
812                  * it as an overrun (thanks to bump_cpu_timer above).
813                  */
814                 read_unlock(&tasklist_lock);
815                 goto out;
816         }
817
818         if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
819                 cpu_time_add(timer->it_clock, &new_expires, val);
820         }
821
822         /*
823          * Install the new expiry time (or zero).
824          * For a timer with no notification action, we don't actually
825          * arm the timer (we'll just fake it for timer_gettime).
826          */
827         timer->it.cpu.expires = new_expires;
828         if (new_expires.sched != 0 &&
829             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
830             cpu_time_before(timer->it_clock, val, new_expires)) {
831                 arm_timer(timer, val);
832         }
833
834         read_unlock(&tasklist_lock);
835
836         /*
837          * Install the new reload setting, and
838          * set up the signal and overrun bookkeeping.
839          */
840         timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
841                                                 &new->it_interval);
842
843         /*
844          * This acts as a modification timestamp for the timer,
845          * so any automatic reload attempt will punt on seeing
846          * that we have reset the timer manually.
847          */
848         timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
849                 ~REQUEUE_PENDING;
850         timer->it_overrun_last = 0;
851         timer->it_overrun = -1;
852
853         if (new_expires.sched != 0 &&
854             (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
855             !cpu_time_before(timer->it_clock, val, new_expires)) {
856                 /*
857                  * The designated time already passed, so we notify
858                  * immediately, even if the thread never runs to
859                  * accumulate more time on this clock.
860                  */
861                 cpu_timer_fire(timer);
862         }
863
864         ret = 0;
865  out:
866         if (old) {
867                 sample_to_timespec(timer->it_clock,
868                                    timer->it.cpu.incr, &old->it_interval);
869         }
870         return ret;
871 }
872
873 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
874 {
875         union cpu_time_count now;
876         struct task_struct *p = timer->it.cpu.task;
877         int clear_dead;
878
879         /*
880          * Easy part: convert the reload time.
881          */
882         sample_to_timespec(timer->it_clock,
883                            timer->it.cpu.incr, &itp->it_interval);
884
885         if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all.  */
886                 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
887                 return;
888         }
889
890         if (unlikely(p == NULL)) {
891                 /*
892                  * This task already died and the timer will never fire.
893                  * In this case, expires is actually the dead value.
894                  */
895         dead:
896                 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
897                                    &itp->it_value);
898                 return;
899         }
900
901         /*
902          * Sample the clock to take the difference with the expiry time.
903          */
904         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
905                 cpu_clock_sample(timer->it_clock, p, &now);
906                 clear_dead = p->exit_state;
907         } else {
908                 read_lock(&tasklist_lock);
909                 if (unlikely(p->signal == NULL)) {
910                         /*
911                          * The process has been reaped.
912                          * We can't even collect a sample any more.
913                          * Call the timer disarmed, nothing else to do.
914                          */
915                         put_task_struct(p);
916                         timer->it.cpu.task = NULL;
917                         timer->it.cpu.expires.sched = 0;
918                         read_unlock(&tasklist_lock);
919                         goto dead;
920                 } else {
921                         cpu_clock_sample_group(timer->it_clock, p, &now);
922                         clear_dead = (unlikely(p->exit_state) &&
923                                       thread_group_empty(p));
924                 }
925                 read_unlock(&tasklist_lock);
926         }
927
928         if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
929                 if (timer->it.cpu.incr.sched == 0 &&
930                     cpu_time_before(timer->it_clock,
931                                     timer->it.cpu.expires, now)) {
932                         /*
933                          * Do-nothing timer expired and has no reload,
934                          * so it's as if it was never set.
935                          */
936                         timer->it.cpu.expires.sched = 0;
937                         itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
938                         return;
939                 }
940                 /*
941                  * Account for any expirations and reloads that should
942                  * have happened.
943                  */
944                 bump_cpu_timer(timer, now);
945         }
946
947         if (unlikely(clear_dead)) {
948                 /*
949                  * We've noticed that the thread is dead, but
950                  * not yet reaped.  Take this opportunity to
951                  * drop our task ref.
952                  */
953                 clear_dead_task(timer, now);
954                 goto dead;
955         }
956
957         if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
958                 sample_to_timespec(timer->it_clock,
959                                    cpu_time_sub(timer->it_clock,
960                                                 timer->it.cpu.expires, now),
961                                    &itp->it_value);
962         } else {
963                 /*
964                  * The timer should have expired already, but the firing
965                  * hasn't taken place yet.  Say it's just about to expire.
966                  */
967                 itp->it_value.tv_nsec = 1;
968                 itp->it_value.tv_sec = 0;
969         }
970 }
971
972 /*
973  * Check for any per-thread CPU timers that have fired and move them off
974  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
975  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
976  */
977 static void check_thread_timers(struct task_struct *tsk,
978                                 struct list_head *firing)
979 {
980         int maxfire;
981         struct list_head *timers = tsk->cpu_timers;
982         struct signal_struct *const sig = tsk->signal;
983
984         maxfire = 20;
985         tsk->cputime_expires.prof_exp = cputime_zero;
986         while (!list_empty(timers)) {
987                 struct cpu_timer_list *t = list_first_entry(timers,
988                                                       struct cpu_timer_list,
989                                                       entry);
990                 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
991                         tsk->cputime_expires.prof_exp = t->expires.cpu;
992                         break;
993                 }
994                 t->firing = 1;
995                 list_move_tail(&t->entry, firing);
996         }
997
998         ++timers;
999         maxfire = 20;
1000         tsk->cputime_expires.virt_exp = cputime_zero;
1001         while (!list_empty(timers)) {
1002                 struct cpu_timer_list *t = list_first_entry(timers,
1003                                                       struct cpu_timer_list,
1004                                                       entry);
1005                 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
1006                         tsk->cputime_expires.virt_exp = t->expires.cpu;
1007                         break;
1008                 }
1009                 t->firing = 1;
1010                 list_move_tail(&t->entry, firing);
1011         }
1012
1013         ++timers;
1014         maxfire = 20;
1015         tsk->cputime_expires.sched_exp = 0;
1016         while (!list_empty(timers)) {
1017                 struct cpu_timer_list *t = list_first_entry(timers,
1018                                                       struct cpu_timer_list,
1019                                                       entry);
1020                 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1021                         tsk->cputime_expires.sched_exp = t->expires.sched;
1022                         break;
1023                 }
1024                 t->firing = 1;
1025                 list_move_tail(&t->entry, firing);
1026         }
1027
1028         /*
1029          * Check for the special case thread timers.
1030          */
1031         if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1032                 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1033                 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1034
1035                 if (hard != RLIM_INFINITY &&
1036                     tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1037                         /*
1038                          * At the hard limit, we just die.
1039                          * No need to calculate anything else now.
1040                          */
1041                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1042                         return;
1043                 }
1044                 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1045                         /*
1046                          * At the soft limit, send a SIGXCPU every second.
1047                          */
1048                         if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1049                             < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1050                                 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1051                                                                 USEC_PER_SEC;
1052                         }
1053                         printk(KERN_INFO
1054                                 "RT Watchdog Timeout: %s[%d]\n",
1055                                 tsk->comm, task_pid_nr(tsk));
1056                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1057                 }
1058         }
1059 }
1060
1061 /*
1062  * Check for any per-thread CPU timers that have fired and move them
1063  * off the tsk->*_timers list onto the firing list.  Per-thread timers
1064  * have already been taken off.
1065  */
1066 static void check_process_timers(struct task_struct *tsk,
1067                                  struct list_head *firing)
1068 {
1069         int maxfire;
1070         struct signal_struct *const sig = tsk->signal;
1071         cputime_t utime, ptime, virt_expires, prof_expires;
1072         unsigned long long sum_sched_runtime, sched_expires;
1073         struct list_head *timers = sig->cpu_timers;
1074         struct task_cputime cputime;
1075
1076         /*
1077          * Don't sample the current process CPU clocks if there are no timers.
1078          */
1079         if (list_empty(&timers[CPUCLOCK_PROF]) &&
1080             cputime_eq(sig->it_prof_expires, cputime_zero) &&
1081             sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1082             list_empty(&timers[CPUCLOCK_VIRT]) &&
1083             cputime_eq(sig->it_virt_expires, cputime_zero) &&
1084             list_empty(&timers[CPUCLOCK_SCHED]))
1085                 return;
1086
1087         /*
1088          * Collect the current process totals.
1089          */
1090         thread_group_cputime(tsk, &cputime);
1091         utime = cputime.utime;
1092         ptime = cputime_add(utime, cputime.stime);
1093         sum_sched_runtime = cputime.sum_exec_runtime;
1094         maxfire = 20;
1095         prof_expires = cputime_zero;
1096         while (!list_empty(timers)) {
1097                 struct cpu_timer_list *tl = list_first_entry(timers,
1098                                                       struct cpu_timer_list,
1099                                                       entry);
1100                 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1101                         prof_expires = tl->expires.cpu;
1102                         break;
1103                 }
1104                 tl->firing = 1;
1105                 list_move_tail(&tl->entry, firing);
1106         }
1107
1108         ++timers;
1109         maxfire = 20;
1110         virt_expires = cputime_zero;
1111         while (!list_empty(timers)) {
1112                 struct cpu_timer_list *tl = list_first_entry(timers,
1113                                                       struct cpu_timer_list,
1114                                                       entry);
1115                 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1116                         virt_expires = tl->expires.cpu;
1117                         break;
1118                 }
1119                 tl->firing = 1;
1120                 list_move_tail(&tl->entry, firing);
1121         }
1122
1123         ++timers;
1124         maxfire = 20;
1125         sched_expires = 0;
1126         while (!list_empty(timers)) {
1127                 struct cpu_timer_list *tl = list_first_entry(timers,
1128                                                       struct cpu_timer_list,
1129                                                       entry);
1130                 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1131                         sched_expires = tl->expires.sched;
1132                         break;
1133                 }
1134                 tl->firing = 1;
1135                 list_move_tail(&tl->entry, firing);
1136         }
1137
1138         /*
1139          * Check for the special case process timers.
1140          */
1141         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1142                 if (cputime_ge(ptime, sig->it_prof_expires)) {
1143                         /* ITIMER_PROF fires and reloads.  */
1144                         sig->it_prof_expires = sig->it_prof_incr;
1145                         if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1146                                 sig->it_prof_expires = cputime_add(
1147                                         sig->it_prof_expires, ptime);
1148                         }
1149                         __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1150                 }
1151                 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1152                     (cputime_eq(prof_expires, cputime_zero) ||
1153                      cputime_lt(sig->it_prof_expires, prof_expires))) {
1154                         prof_expires = sig->it_prof_expires;
1155                 }
1156         }
1157         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1158                 if (cputime_ge(utime, sig->it_virt_expires)) {
1159                         /* ITIMER_VIRTUAL fires and reloads.  */
1160                         sig->it_virt_expires = sig->it_virt_incr;
1161                         if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1162                                 sig->it_virt_expires = cputime_add(
1163                                         sig->it_virt_expires, utime);
1164                         }
1165                         __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1166                 }
1167                 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1168                     (cputime_eq(virt_expires, cputime_zero) ||
1169                      cputime_lt(sig->it_virt_expires, virt_expires))) {
1170                         virt_expires = sig->it_virt_expires;
1171                 }
1172         }
1173         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1174                 unsigned long psecs = cputime_to_secs(ptime);
1175                 cputime_t x;
1176                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1177                         /*
1178                          * At the hard limit, we just die.
1179                          * No need to calculate anything else now.
1180                          */
1181                         __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1182                         return;
1183                 }
1184                 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1185                         /*
1186                          * At the soft limit, send a SIGXCPU every second.
1187                          */
1188                         __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1189                         if (sig->rlim[RLIMIT_CPU].rlim_cur
1190                             < sig->rlim[RLIMIT_CPU].rlim_max) {
1191                                 sig->rlim[RLIMIT_CPU].rlim_cur++;
1192                         }
1193                 }
1194                 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1195                 if (cputime_eq(prof_expires, cputime_zero) ||
1196                     cputime_lt(x, prof_expires)) {
1197                         prof_expires = x;
1198                 }
1199         }
1200
1201         if (!cputime_eq(prof_expires, cputime_zero) &&
1202             (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) ||
1203              cputime_gt(sig->cputime_expires.prof_exp, prof_expires)))
1204                 sig->cputime_expires.prof_exp = prof_expires;
1205         if (!cputime_eq(virt_expires, cputime_zero) &&
1206             (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) ||
1207              cputime_gt(sig->cputime_expires.virt_exp, virt_expires)))
1208                 sig->cputime_expires.virt_exp = virt_expires;
1209         if (sched_expires != 0 &&
1210             (sig->cputime_expires.sched_exp == 0 ||
1211              sig->cputime_expires.sched_exp > sched_expires))
1212                 sig->cputime_expires.sched_exp = sched_expires;
1213 }
1214
1215 /*
1216  * This is called from the signal code (via do_schedule_next_timer)
1217  * when the last timer signal was delivered and we have to reload the timer.
1218  */
1219 void posix_cpu_timer_schedule(struct k_itimer *timer)
1220 {
1221         struct task_struct *p = timer->it.cpu.task;
1222         union cpu_time_count now;
1223
1224         if (unlikely(p == NULL))
1225                 /*
1226                  * The task was cleaned up already, no future firings.
1227                  */
1228                 goto out;
1229
1230         /*
1231          * Fetch the current sample and update the timer's expiry time.
1232          */
1233         if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1234                 cpu_clock_sample(timer->it_clock, p, &now);
1235                 bump_cpu_timer(timer, now);
1236                 if (unlikely(p->exit_state)) {
1237                         clear_dead_task(timer, now);
1238                         goto out;
1239                 }
1240                 read_lock(&tasklist_lock); /* arm_timer needs it.  */
1241         } else {
1242                 read_lock(&tasklist_lock);
1243                 if (unlikely(p->signal == NULL)) {
1244                         /*
1245                          * The process has been reaped.
1246                          * We can't even collect a sample any more.
1247                          */
1248                         put_task_struct(p);
1249                         timer->it.cpu.task = p = NULL;
1250                         timer->it.cpu.expires.sched = 0;
1251                         goto out_unlock;
1252                 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1253                         /*
1254                          * We've noticed that the thread is dead, but
1255                          * not yet reaped.  Take this opportunity to
1256                          * drop our task ref.
1257                          */
1258                         clear_dead_task(timer, now);
1259                         goto out_unlock;
1260                 }
1261                 cpu_clock_sample_group(timer->it_clock, p, &now);
1262                 bump_cpu_timer(timer, now);
1263                 /* Leave the tasklist_lock locked for the call below.  */
1264         }
1265
1266         /*
1267          * Now re-arm for the new expiry time.
1268          */
1269         arm_timer(timer, now);
1270
1271 out_unlock:
1272         read_unlock(&tasklist_lock);
1273
1274 out:
1275         timer->it_overrun_last = timer->it_overrun;
1276         timer->it_overrun = -1;
1277         ++timer->it_requeue_pending;
1278 }
1279
1280 /**
1281  * task_cputime_zero - Check a task_cputime struct for all zero fields.
1282  *
1283  * @cputime:    The struct to compare.
1284  *
1285  * Checks @cputime to see if all fields are zero.  Returns true if all fields
1286  * are zero, false if any field is nonzero.
1287  */
1288 static inline int task_cputime_zero(const struct task_cputime *cputime)
1289 {
1290         if (cputime_eq(cputime->utime, cputime_zero) &&
1291             cputime_eq(cputime->stime, cputime_zero) &&
1292             cputime->sum_exec_runtime == 0)
1293                 return 1;
1294         return 0;
1295 }
1296
1297 /**
1298  * task_cputime_expired - Compare two task_cputime entities.
1299  *
1300  * @sample:     The task_cputime structure to be checked for expiration.
1301  * @expires:    Expiration times, against which @sample will be checked.
1302  *
1303  * Checks @sample against @expires to see if any field of @sample has expired.
1304  * Returns true if any field of the former is greater than the corresponding
1305  * field of the latter if the latter field is set.  Otherwise returns false.
1306  */
1307 static inline int task_cputime_expired(const struct task_cputime *sample,
1308                                         const struct task_cputime *expires)
1309 {
1310         if (!cputime_eq(expires->utime, cputime_zero) &&
1311             cputime_ge(sample->utime, expires->utime))
1312                 return 1;
1313         if (!cputime_eq(expires->stime, cputime_zero) &&
1314             cputime_ge(cputime_add(sample->utime, sample->stime),
1315                        expires->stime))
1316                 return 1;
1317         if (expires->sum_exec_runtime != 0 &&
1318             sample->sum_exec_runtime >= expires->sum_exec_runtime)
1319                 return 1;
1320         return 0;
1321 }
1322
1323 /**
1324  * fastpath_timer_check - POSIX CPU timers fast path.
1325  *
1326  * @tsk:        The task (thread) being checked.
1327  * @sig:        The signal pointer for that task.
1328  *
1329  * If there are no timers set return false.  Otherwise snapshot the task and
1330  * thread group timers, then compare them with the corresponding expiration
1331  # times.  Returns true if a timer has expired, else returns false.
1332  */
1333 static inline int fastpath_timer_check(struct task_struct *tsk,
1334                                         struct signal_struct *sig)
1335 {
1336         struct task_cputime task_sample = {
1337                 .utime = tsk->utime,
1338                 .stime = tsk->stime,
1339                 .sum_exec_runtime = tsk->se.sum_exec_runtime
1340         };
1341         struct task_cputime group_sample;
1342
1343         if (task_cputime_zero(&tsk->cputime_expires) &&
1344             task_cputime_zero(&sig->cputime_expires))
1345                 return 0;
1346         if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1347                 return 1;
1348         thread_group_cputime(tsk, &group_sample);
1349         return task_cputime_expired(&group_sample, &sig->cputime_expires);
1350 }
1351
1352 /*
1353  * This is called from the timer interrupt handler.  The irq handler has
1354  * already updated our counts.  We need to check if any timers fire now.
1355  * Interrupts are disabled.
1356  */
1357 void run_posix_cpu_timers(struct task_struct *tsk)
1358 {
1359         LIST_HEAD(firing);
1360         struct k_itimer *timer, *next;
1361         struct signal_struct *sig;
1362         struct sighand_struct *sighand;
1363         unsigned long flags;
1364
1365         BUG_ON(!irqs_disabled());
1366
1367         /* Pick up tsk->signal and make sure it's valid. */
1368         sig = tsk->signal;
1369         /*
1370          * The fast path checks that there are no expired thread or thread
1371          * group timers.  If that's so, just return.  Also check that
1372          * tsk->signal is non-NULL; this probably can't happen but cover the
1373          * possibility anyway.
1374          */
1375         if (unlikely(!sig) || !fastpath_timer_check(tsk, sig)) {
1376                 return;
1377         }
1378         sighand = lock_task_sighand(tsk, &flags);
1379         if (likely(sighand)) {
1380                 /*
1381                  * Here we take off tsk->signal->cpu_timers[N] and
1382                  * tsk->cpu_timers[N] all the timers that are firing, and
1383                  * put them on the firing list.
1384                  */
1385                 check_thread_timers(tsk, &firing);
1386                 check_process_timers(tsk, &firing);
1387
1388                 /*
1389                  * We must release these locks before taking any timer's lock.
1390                  * There is a potential race with timer deletion here, as the
1391                  * siglock now protects our private firing list.  We have set
1392                  * the firing flag in each timer, so that a deletion attempt
1393                  * that gets the timer lock before we do will give it up and
1394                  * spin until we've taken care of that timer below.
1395                  */
1396         }
1397         unlock_task_sighand(tsk, &flags);
1398
1399         /*
1400          * Now that all the timers on our list have the firing flag,
1401          * noone will touch their list entries but us.  We'll take
1402          * each timer's lock before clearing its firing flag, so no
1403          * timer call will interfere.
1404          */
1405         list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1406                 int firing;
1407                 spin_lock(&timer->it_lock);
1408                 list_del_init(&timer->it.cpu.entry);
1409                 firing = timer->it.cpu.firing;
1410                 timer->it.cpu.firing = 0;
1411                 /*
1412                  * The firing flag is -1 if we collided with a reset
1413                  * of the timer, which already reported this
1414                  * almost-firing as an overrun.  So don't generate an event.
1415                  */
1416                 if (likely(firing >= 0)) {
1417                         cpu_timer_fire(timer);
1418                 }
1419                 spin_unlock(&timer->it_lock);
1420         }
1421 }
1422
1423 /*
1424  * Set one of the process-wide special case CPU timers.
1425  * The tsk->sighand->siglock must be held by the caller.
1426  * The *newval argument is relative and we update it to be absolute, *oldval
1427  * is absolute and we update it to be relative.
1428  */
1429 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1430                            cputime_t *newval, cputime_t *oldval)
1431 {
1432         union cpu_time_count now;
1433         struct list_head *head;
1434
1435         BUG_ON(clock_idx == CPUCLOCK_SCHED);
1436         cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1437
1438         if (oldval) {
1439                 if (!cputime_eq(*oldval, cputime_zero)) {
1440                         if (cputime_le(*oldval, now.cpu)) {
1441                                 /* Just about to fire. */
1442                                 *oldval = jiffies_to_cputime(1);
1443                         } else {
1444                                 *oldval = cputime_sub(*oldval, now.cpu);
1445                         }
1446                 }
1447
1448                 if (cputime_eq(*newval, cputime_zero))
1449                         return;
1450                 *newval = cputime_add(*newval, now.cpu);
1451
1452                 /*
1453                  * If the RLIMIT_CPU timer will expire before the
1454                  * ITIMER_PROF timer, we have nothing else to do.
1455                  */
1456                 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1457                     < cputime_to_secs(*newval))
1458                         return;
1459         }
1460
1461         /*
1462          * Check whether there are any process timers already set to fire
1463          * before this one.  If so, we don't have anything more to do.
1464          */
1465         head = &tsk->signal->cpu_timers[clock_idx];
1466         if (list_empty(head) ||
1467             cputime_ge(list_first_entry(head,
1468                                   struct cpu_timer_list, entry)->expires.cpu,
1469                        *newval)) {
1470                 switch (clock_idx) {
1471                 case CPUCLOCK_PROF:
1472                         tsk->signal->cputime_expires.prof_exp = *newval;
1473                         break;
1474                 case CPUCLOCK_VIRT:
1475                         tsk->signal->cputime_expires.virt_exp = *newval;
1476                         break;
1477                 }
1478         }
1479 }
1480
1481 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1482                             struct timespec *rqtp, struct itimerspec *it)
1483 {
1484         struct k_itimer timer;
1485         int error;
1486
1487         /*
1488          * Set up a temporary timer and then wait for it to go off.
1489          */
1490         memset(&timer, 0, sizeof timer);
1491         spin_lock_init(&timer.it_lock);
1492         timer.it_clock = which_clock;
1493         timer.it_overrun = -1;
1494         error = posix_cpu_timer_create(&timer);
1495         timer.it_process = current;
1496         if (!error) {
1497                 static struct itimerspec zero_it;
1498
1499                 memset(it, 0, sizeof *it);
1500                 it->it_value = *rqtp;
1501
1502                 spin_lock_irq(&timer.it_lock);
1503                 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1504                 if (error) {
1505                         spin_unlock_irq(&timer.it_lock);
1506                         return error;
1507                 }
1508
1509                 while (!signal_pending(current)) {
1510                         if (timer.it.cpu.expires.sched == 0) {
1511                                 /*
1512                                  * Our timer fired and was reset.
1513                                  */
1514                                 spin_unlock_irq(&timer.it_lock);
1515                                 return 0;
1516                         }
1517
1518                         /*
1519                          * Block until cpu_timer_fire (or a signal) wakes us.
1520                          */
1521                         __set_current_state(TASK_INTERRUPTIBLE);
1522                         spin_unlock_irq(&timer.it_lock);
1523                         schedule();
1524                         spin_lock_irq(&timer.it_lock);
1525                 }
1526
1527                 /*
1528                  * We were interrupted by a signal.
1529                  */
1530                 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1531                 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1532                 spin_unlock_irq(&timer.it_lock);
1533
1534                 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1535                         /*
1536                          * It actually did fire already.
1537                          */
1538                         return 0;
1539                 }
1540
1541                 error = -ERESTART_RESTARTBLOCK;
1542         }
1543
1544         return error;
1545 }
1546
1547 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1548                      struct timespec *rqtp, struct timespec __user *rmtp)
1549 {
1550         struct restart_block *restart_block =
1551             &current_thread_info()->restart_block;
1552         struct itimerspec it;
1553         int error;
1554
1555         /*
1556          * Diagnose required errors first.
1557          */
1558         if (CPUCLOCK_PERTHREAD(which_clock) &&
1559             (CPUCLOCK_PID(which_clock) == 0 ||
1560              CPUCLOCK_PID(which_clock) == current->pid))
1561                 return -EINVAL;
1562
1563         error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1564
1565         if (error == -ERESTART_RESTARTBLOCK) {
1566
1567                 if (flags & TIMER_ABSTIME)
1568                         return -ERESTARTNOHAND;
1569                 /*
1570                  * Report back to the user the time still remaining.
1571                  */
1572                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1573                         return -EFAULT;
1574
1575                 restart_block->fn = posix_cpu_nsleep_restart;
1576                 restart_block->arg0 = which_clock;
1577                 restart_block->arg1 = (unsigned long) rmtp;
1578                 restart_block->arg2 = rqtp->tv_sec;
1579                 restart_block->arg3 = rqtp->tv_nsec;
1580         }
1581         return error;
1582 }
1583
1584 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1585 {
1586         clockid_t which_clock = restart_block->arg0;
1587         struct timespec __user *rmtp;
1588         struct timespec t;
1589         struct itimerspec it;
1590         int error;
1591
1592         rmtp = (struct timespec __user *) restart_block->arg1;
1593         t.tv_sec = restart_block->arg2;
1594         t.tv_nsec = restart_block->arg3;
1595
1596         restart_block->fn = do_no_restart_syscall;
1597         error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1598
1599         if (error == -ERESTART_RESTARTBLOCK) {
1600                 /*
1601                  * Report back to the user the time still remaining.
1602                  */
1603                 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1604                         return -EFAULT;
1605
1606                 restart_block->fn = posix_cpu_nsleep_restart;
1607                 restart_block->arg0 = which_clock;
1608                 restart_block->arg1 = (unsigned long) rmtp;
1609                 restart_block->arg2 = t.tv_sec;
1610                 restart_block->arg3 = t.tv_nsec;
1611         }
1612         return error;
1613
1614 }
1615
1616
1617 #define PROCESS_CLOCK   MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1618 #define THREAD_CLOCK    MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1619
1620 static int process_cpu_clock_getres(const clockid_t which_clock,
1621                                     struct timespec *tp)
1622 {
1623         return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1624 }
1625 static int process_cpu_clock_get(const clockid_t which_clock,
1626                                  struct timespec *tp)
1627 {
1628         return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1629 }
1630 static int process_cpu_timer_create(struct k_itimer *timer)
1631 {
1632         timer->it_clock = PROCESS_CLOCK;
1633         return posix_cpu_timer_create(timer);
1634 }
1635 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1636                               struct timespec *rqtp,
1637                               struct timespec __user *rmtp)
1638 {
1639         return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1640 }
1641 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1642 {
1643         return -EINVAL;
1644 }
1645 static int thread_cpu_clock_getres(const clockid_t which_clock,
1646                                    struct timespec *tp)
1647 {
1648         return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1649 }
1650 static int thread_cpu_clock_get(const clockid_t which_clock,
1651                                 struct timespec *tp)
1652 {
1653         return posix_cpu_clock_get(THREAD_CLOCK, tp);
1654 }
1655 static int thread_cpu_timer_create(struct k_itimer *timer)
1656 {
1657         timer->it_clock = THREAD_CLOCK;
1658         return posix_cpu_timer_create(timer);
1659 }
1660 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1661                               struct timespec *rqtp, struct timespec __user *rmtp)
1662 {
1663         return -EINVAL;
1664 }
1665 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1666 {
1667         return -EINVAL;
1668 }
1669
1670 static __init int init_posix_cpu_timers(void)
1671 {
1672         struct k_clock process = {
1673                 .clock_getres = process_cpu_clock_getres,
1674                 .clock_get = process_cpu_clock_get,
1675                 .clock_set = do_posix_clock_nosettime,
1676                 .timer_create = process_cpu_timer_create,
1677                 .nsleep = process_cpu_nsleep,
1678                 .nsleep_restart = process_cpu_nsleep_restart,
1679         };
1680         struct k_clock thread = {
1681                 .clock_getres = thread_cpu_clock_getres,
1682                 .clock_get = thread_cpu_clock_get,
1683                 .clock_set = do_posix_clock_nosettime,
1684                 .timer_create = thread_cpu_timer_create,
1685                 .nsleep = thread_cpu_nsleep,
1686                 .nsleep_restart = thread_cpu_nsleep_restart,
1687         };
1688
1689         register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1690         register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1691
1692         return 0;
1693 }
1694 __initcall(init_posix_cpu_timers);