1
2
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#include <linux/random.h>
13#include <linux/tick.h>
14#include <linux/workqueue.h>
15
16
17
18
19
20
21
22void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
23{
24 cputime_t cputime = secs_to_cputime(rlim_new);
25
26 spin_lock_irq(&task->sighand->siglock);
27 set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
28 spin_unlock_irq(&task->sighand->siglock);
29}
30
31static int check_clock(const clockid_t which_clock)
32{
33 int error = 0;
34 struct task_struct *p;
35 const pid_t pid = CPUCLOCK_PID(which_clock);
36
37 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
38 return -EINVAL;
39
40 if (pid == 0)
41 return 0;
42
43 rcu_read_lock();
44 p = find_task_by_vpid(pid);
45 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
46 same_thread_group(p, current) : has_group_leader_pid(p))) {
47 error = -EINVAL;
48 }
49 rcu_read_unlock();
50
51 return error;
52}
53
54static inline unsigned long long
55timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
56{
57 unsigned long long ret;
58
59 ret = 0;
60 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
61 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
62 } else {
63 ret = cputime_to_expires(timespec_to_cputime(tp));
64 }
65 return ret;
66}
67
68static void sample_to_timespec(const clockid_t which_clock,
69 unsigned long long expires,
70 struct timespec *tp)
71{
72 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
73 *tp = ns_to_timespec(expires);
74 else
75 cputime_to_timespec((__force cputime_t)expires, tp);
76}
77
78
79
80
81
82static void bump_cpu_timer(struct k_itimer *timer,
83 unsigned long long now)
84{
85 int i;
86 unsigned long long delta, incr;
87
88 if (timer->it.cpu.incr == 0)
89 return;
90
91 if (now < timer->it.cpu.expires)
92 return;
93
94 incr = timer->it.cpu.incr;
95 delta = now + incr - timer->it.cpu.expires;
96
97
98 for (i = 0; incr < delta - incr; i++)
99 incr = incr << 1;
100
101 for (; i >= 0; incr >>= 1, i--) {
102 if (delta < incr)
103 continue;
104
105 timer->it.cpu.expires += incr;
106 timer->it_overrun += 1 << i;
107 delta -= incr;
108 }
109}
110
111
112
113
114
115
116
117
118
119static inline int task_cputime_zero(const struct task_cputime *cputime)
120{
121 if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
122 return 1;
123 return 0;
124}
125
126static inline unsigned long long prof_ticks(struct task_struct *p)
127{
128 cputime_t utime, stime;
129
130 task_cputime(p, &utime, &stime);
131
132 return cputime_to_expires(utime + stime);
133}
134static inline unsigned long long virt_ticks(struct task_struct *p)
135{
136 cputime_t utime;
137
138 task_cputime(p, &utime, NULL);
139
140 return cputime_to_expires(utime);
141}
142
143static int
144posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
145{
146 int error = check_clock(which_clock);
147 if (!error) {
148 tp->tv_sec = 0;
149 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
150 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
151
152
153
154
155
156 tp->tv_nsec = 1;
157 }
158 }
159 return error;
160}
161
162static int
163posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
164{
165
166
167
168
169 int error = check_clock(which_clock);
170 if (error == 0) {
171 error = -EPERM;
172 }
173 return error;
174}
175
176
177
178
179
180static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
181 unsigned long long *sample)
182{
183 switch (CPUCLOCK_WHICH(which_clock)) {
184 default:
185 return -EINVAL;
186 case CPUCLOCK_PROF:
187 *sample = prof_ticks(p);
188 break;
189 case CPUCLOCK_VIRT:
190 *sample = virt_ticks(p);
191 break;
192 case CPUCLOCK_SCHED:
193 *sample = task_sched_runtime(p);
194 break;
195 }
196 return 0;
197}
198
199
200
201
202
203static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
204{
205 u64 curr_cputime;
206retry:
207 curr_cputime = atomic64_read(cputime);
208 if (sum_cputime > curr_cputime) {
209 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
210 goto retry;
211 }
212}
213
214static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
215{
216 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
217 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
218 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
219}
220
221
222static inline void sample_cputime_atomic(struct task_cputime *times,
223 struct task_cputime_atomic *atomic_times)
224{
225 times->utime = atomic64_read(&atomic_times->utime);
226 times->stime = atomic64_read(&atomic_times->stime);
227 times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
228}
229
230void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
231{
232 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
233 struct task_cputime sum;
234
235
236 if (!READ_ONCE(cputimer->running)) {
237
238
239
240
241
242 thread_group_cputime(tsk, &sum);
243 update_gt_cputime(&cputimer->cputime_atomic, &sum);
244
245
246
247
248
249
250
251
252 WRITE_ONCE(cputimer->running, 1);
253 }
254 sample_cputime_atomic(times, &cputimer->cputime_atomic);
255}
256
257
258
259
260
261
262static int cpu_clock_sample_group(const clockid_t which_clock,
263 struct task_struct *p,
264 unsigned long long *sample)
265{
266 struct task_cputime cputime;
267
268 switch (CPUCLOCK_WHICH(which_clock)) {
269 default:
270 return -EINVAL;
271 case CPUCLOCK_PROF:
272 thread_group_cputime(p, &cputime);
273 *sample = cputime_to_expires(cputime.utime + cputime.stime);
274 break;
275 case CPUCLOCK_VIRT:
276 thread_group_cputime(p, &cputime);
277 *sample = cputime_to_expires(cputime.utime);
278 break;
279 case CPUCLOCK_SCHED:
280 thread_group_cputime(p, &cputime);
281 *sample = cputime.sum_exec_runtime;
282 break;
283 }
284 return 0;
285}
286
287static int posix_cpu_clock_get_task(struct task_struct *tsk,
288 const clockid_t which_clock,
289 struct timespec *tp)
290{
291 int err = -EINVAL;
292 unsigned long long rtn;
293
294 if (CPUCLOCK_PERTHREAD(which_clock)) {
295 if (same_thread_group(tsk, current))
296 err = cpu_clock_sample(which_clock, tsk, &rtn);
297 } else {
298 if (tsk == current || thread_group_leader(tsk))
299 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
300 }
301
302 if (!err)
303 sample_to_timespec(which_clock, rtn, tp);
304
305 return err;
306}
307
308
309static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
310{
311 const pid_t pid = CPUCLOCK_PID(which_clock);
312 int err = -EINVAL;
313
314 if (pid == 0) {
315
316
317
318
319 err = posix_cpu_clock_get_task(current, which_clock, tp);
320 } else {
321
322
323
324
325 struct task_struct *p;
326 rcu_read_lock();
327 p = find_task_by_vpid(pid);
328 if (p)
329 err = posix_cpu_clock_get_task(p, which_clock, tp);
330 rcu_read_unlock();
331 }
332
333 return err;
334}
335
336
337
338
339
340
341
342static int posix_cpu_timer_create(struct k_itimer *new_timer)
343{
344 int ret = 0;
345 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
346 struct task_struct *p;
347
348 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
349 return -EINVAL;
350
351 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
352
353 rcu_read_lock();
354 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
355 if (pid == 0) {
356 p = current;
357 } else {
358 p = find_task_by_vpid(pid);
359 if (p && !same_thread_group(p, current))
360 p = NULL;
361 }
362 } else {
363 if (pid == 0) {
364 p = current->group_leader;
365 } else {
366 p = find_task_by_vpid(pid);
367 if (p && !has_group_leader_pid(p))
368 p = NULL;
369 }
370 }
371 new_timer->it.cpu.task = p;
372 if (p) {
373 get_task_struct(p);
374 } else {
375 ret = -EINVAL;
376 }
377 rcu_read_unlock();
378
379 return ret;
380}
381
382
383
384
385
386
387
388static int posix_cpu_timer_del(struct k_itimer *timer)
389{
390 int ret = 0;
391 unsigned long flags;
392 struct sighand_struct *sighand;
393 struct task_struct *p = timer->it.cpu.task;
394
395 WARN_ON_ONCE(p == NULL);
396
397
398
399
400
401 sighand = lock_task_sighand(p, &flags);
402 if (unlikely(sighand == NULL)) {
403
404
405
406
407 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
408 } else {
409 if (timer->it.cpu.firing)
410 ret = TIMER_RETRY;
411 else
412 list_del(&timer->it.cpu.entry);
413
414 unlock_task_sighand(p, &flags);
415 }
416
417 if (!ret)
418 put_task_struct(p);
419
420 return ret;
421}
422
423static void cleanup_timers_list(struct list_head *head)
424{
425 struct cpu_timer_list *timer, *next;
426
427 list_for_each_entry_safe(timer, next, head, entry)
428 list_del_init(&timer->entry);
429}
430
431
432
433
434
435
436
437static void cleanup_timers(struct list_head *head)
438{
439 cleanup_timers_list(head);
440 cleanup_timers_list(++head);
441 cleanup_timers_list(++head);
442}
443
444
445
446
447
448
449void posix_cpu_timers_exit(struct task_struct *tsk)
450{
451 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
452 sizeof(unsigned long long));
453 cleanup_timers(tsk->cpu_timers);
454
455}
456void posix_cpu_timers_exit_group(struct task_struct *tsk)
457{
458 cleanup_timers(tsk->signal->cpu_timers);
459}
460
461static inline int expires_gt(cputime_t expires, cputime_t new_exp)
462{
463 return expires == 0 || expires > new_exp;
464}
465
466
467
468
469
470static void arm_timer(struct k_itimer *timer)
471{
472 struct task_struct *p = timer->it.cpu.task;
473 struct list_head *head, *listpos;
474 struct task_cputime *cputime_expires;
475 struct cpu_timer_list *const nt = &timer->it.cpu;
476 struct cpu_timer_list *next;
477
478 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
479 head = p->cpu_timers;
480 cputime_expires = &p->cputime_expires;
481 } else {
482 head = p->signal->cpu_timers;
483 cputime_expires = &p->signal->cputime_expires;
484 }
485 head += CPUCLOCK_WHICH(timer->it_clock);
486
487 listpos = head;
488 list_for_each_entry(next, head, entry) {
489 if (nt->expires < next->expires)
490 break;
491 listpos = &next->entry;
492 }
493 list_add(&nt->entry, listpos);
494
495 if (listpos == head) {
496 unsigned long long exp = nt->expires;
497
498
499
500
501
502
503
504
505 switch (CPUCLOCK_WHICH(timer->it_clock)) {
506 case CPUCLOCK_PROF:
507 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
508 cputime_expires->prof_exp = expires_to_cputime(exp);
509 break;
510 case CPUCLOCK_VIRT:
511 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
512 cputime_expires->virt_exp = expires_to_cputime(exp);
513 break;
514 case CPUCLOCK_SCHED:
515 if (cputime_expires->sched_exp == 0 ||
516 cputime_expires->sched_exp > exp)
517 cputime_expires->sched_exp = exp;
518 break;
519 }
520 }
521}
522
523
524
525
526static void cpu_timer_fire(struct k_itimer *timer)
527{
528 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
529
530
531
532 timer->it.cpu.expires = 0;
533 } else if (unlikely(timer->sigq == NULL)) {
534
535
536
537
538 wake_up_process(timer->it_process);
539 timer->it.cpu.expires = 0;
540 } else if (timer->it.cpu.incr == 0) {
541
542
543
544 posix_timer_event(timer, 0);
545 timer->it.cpu.expires = 0;
546 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
547
548
549
550
551
552
553 posix_cpu_timer_schedule(timer);
554 }
555}
556
557
558
559
560
561
562static int cpu_timer_sample_group(const clockid_t which_clock,
563 struct task_struct *p,
564 unsigned long long *sample)
565{
566 struct task_cputime cputime;
567
568 thread_group_cputimer(p, &cputime);
569 switch (CPUCLOCK_WHICH(which_clock)) {
570 default:
571 return -EINVAL;
572 case CPUCLOCK_PROF:
573 *sample = cputime_to_expires(cputime.utime + cputime.stime);
574 break;
575 case CPUCLOCK_VIRT:
576 *sample = cputime_to_expires(cputime.utime);
577 break;
578 case CPUCLOCK_SCHED:
579 *sample = cputime.sum_exec_runtime;
580 break;
581 }
582 return 0;
583}
584
585#ifdef CONFIG_NO_HZ_FULL
586static void nohz_kick_work_fn(struct work_struct *work)
587{
588 tick_nohz_full_kick_all();
589}
590
591static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
592
593
594
595
596
597static void posix_cpu_timer_kick_nohz(void)
598{
599 if (context_tracking_is_enabled())
600 schedule_work(&nohz_kick_work);
601}
602
603bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
604{
605 if (!task_cputime_zero(&tsk->cputime_expires))
606 return false;
607
608
609 if (READ_ONCE(tsk->signal->cputimer.running))
610 return false;
611
612 return true;
613}
614#else
615static inline void posix_cpu_timer_kick_nohz(void) { }
616#endif
617
618
619
620
621
622
623
624static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
625 struct itimerspec *new, struct itimerspec *old)
626{
627 unsigned long flags;
628 struct sighand_struct *sighand;
629 struct task_struct *p = timer->it.cpu.task;
630 unsigned long long old_expires, new_expires, old_incr, val;
631 int ret;
632
633 WARN_ON_ONCE(p == NULL);
634
635 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
636
637
638
639
640
641 sighand = lock_task_sighand(p, &flags);
642
643
644
645
646 if (unlikely(sighand == NULL)) {
647 return -ESRCH;
648 }
649
650
651
652
653 WARN_ON_ONCE(!irqs_disabled());
654
655 ret = 0;
656 old_incr = timer->it.cpu.incr;
657 old_expires = timer->it.cpu.expires;
658 if (unlikely(timer->it.cpu.firing)) {
659 timer->it.cpu.firing = -1;
660 ret = TIMER_RETRY;
661 } else
662 list_del_init(&timer->it.cpu.entry);
663
664
665
666
667
668
669
670
671
672 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
673 cpu_clock_sample(timer->it_clock, p, &val);
674 } else {
675 cpu_timer_sample_group(timer->it_clock, p, &val);
676 }
677
678 if (old) {
679 if (old_expires == 0) {
680 old->it_value.tv_sec = 0;
681 old->it_value.tv_nsec = 0;
682 } else {
683
684
685
686
687
688
689
690
691
692
693 bump_cpu_timer(timer, val);
694 if (val < timer->it.cpu.expires) {
695 old_expires = timer->it.cpu.expires - val;
696 sample_to_timespec(timer->it_clock,
697 old_expires,
698 &old->it_value);
699 } else {
700 old->it_value.tv_nsec = 1;
701 old->it_value.tv_sec = 0;
702 }
703 }
704 }
705
706 if (unlikely(ret)) {
707
708
709
710
711
712
713 unlock_task_sighand(p, &flags);
714 goto out;
715 }
716
717 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
718 new_expires += val;
719 }
720
721
722
723
724
725
726 timer->it.cpu.expires = new_expires;
727 if (new_expires != 0 && val < new_expires) {
728 arm_timer(timer);
729 }
730
731 unlock_task_sighand(p, &flags);
732
733
734
735
736 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
737 &new->it_interval);
738
739
740
741
742
743
744 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
745 ~REQUEUE_PENDING;
746 timer->it_overrun_last = 0;
747 timer->it_overrun = -1;
748
749 if (new_expires != 0 && !(val < new_expires)) {
750
751
752
753
754
755 cpu_timer_fire(timer);
756 }
757
758 ret = 0;
759 out:
760 if (old) {
761 sample_to_timespec(timer->it_clock,
762 old_incr, &old->it_interval);
763 }
764 if (!ret)
765 posix_cpu_timer_kick_nohz();
766 return ret;
767}
768
769static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
770{
771 unsigned long long now;
772 struct task_struct *p = timer->it.cpu.task;
773
774 WARN_ON_ONCE(p == NULL);
775
776
777
778
779 sample_to_timespec(timer->it_clock,
780 timer->it.cpu.incr, &itp->it_interval);
781
782 if (timer->it.cpu.expires == 0) {
783 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
784 return;
785 }
786
787
788
789
790 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
791 cpu_clock_sample(timer->it_clock, p, &now);
792 } else {
793 struct sighand_struct *sighand;
794 unsigned long flags;
795
796
797
798
799
800
801 sighand = lock_task_sighand(p, &flags);
802 if (unlikely(sighand == NULL)) {
803
804
805
806
807
808 timer->it.cpu.expires = 0;
809 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
810 &itp->it_value);
811 } else {
812 cpu_timer_sample_group(timer->it_clock, p, &now);
813 unlock_task_sighand(p, &flags);
814 }
815 }
816
817 if (now < timer->it.cpu.expires) {
818 sample_to_timespec(timer->it_clock,
819 timer->it.cpu.expires - now,
820 &itp->it_value);
821 } else {
822
823
824
825
826 itp->it_value.tv_nsec = 1;
827 itp->it_value.tv_sec = 0;
828 }
829}
830
831static unsigned long long
832check_timers_list(struct list_head *timers,
833 struct list_head *firing,
834 unsigned long long curr)
835{
836 int maxfire = 20;
837
838 while (!list_empty(timers)) {
839 struct cpu_timer_list *t;
840
841 t = list_first_entry(timers, struct cpu_timer_list, entry);
842
843 if (!--maxfire || curr < t->expires)
844 return t->expires;
845
846 t->firing = 1;
847 list_move_tail(&t->entry, firing);
848 }
849
850 return 0;
851}
852
853
854
855
856
857
858static void check_thread_timers(struct task_struct *tsk,
859 struct list_head *firing)
860{
861 struct list_head *timers = tsk->cpu_timers;
862 struct signal_struct *const sig = tsk->signal;
863 struct task_cputime *tsk_expires = &tsk->cputime_expires;
864 unsigned long long expires;
865 unsigned long soft;
866
867 expires = check_timers_list(timers, firing, prof_ticks(tsk));
868 tsk_expires->prof_exp = expires_to_cputime(expires);
869
870 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
871 tsk_expires->virt_exp = expires_to_cputime(expires);
872
873 tsk_expires->sched_exp = check_timers_list(++timers, firing,
874 tsk->se.sum_exec_runtime);
875
876
877
878
879 soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
880 if (soft != RLIM_INFINITY) {
881 unsigned long hard =
882 READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
883
884 if (hard != RLIM_INFINITY &&
885 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
886
887
888
889
890 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
891 return;
892 }
893 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
894
895
896
897 if (soft < hard) {
898 soft += USEC_PER_SEC;
899 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
900 }
901 printk(KERN_INFO
902 "RT Watchdog Timeout: %s[%d]\n",
903 tsk->comm, task_pid_nr(tsk));
904 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
905 }
906 }
907}
908
909static inline void stop_process_timers(struct signal_struct *sig)
910{
911 struct thread_group_cputimer *cputimer = &sig->cputimer;
912
913
914 WRITE_ONCE(cputimer->running, 0);
915}
916
917static u32 onecputick;
918
919static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
920 unsigned long long *expires,
921 unsigned long long cur_time, int signo)
922{
923 if (!it->expires)
924 return;
925
926 if (cur_time >= it->expires) {
927 if (it->incr) {
928 it->expires += it->incr;
929 it->error += it->incr_error;
930 if (it->error >= onecputick) {
931 it->expires -= cputime_one_jiffy;
932 it->error -= onecputick;
933 }
934 } else {
935 it->expires = 0;
936 }
937
938 trace_itimer_expire(signo == SIGPROF ?
939 ITIMER_PROF : ITIMER_VIRTUAL,
940 tsk->signal->leader_pid, cur_time);
941 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
942 }
943
944 if (it->expires && (!*expires || it->expires < *expires)) {
945 *expires = it->expires;
946 }
947}
948
949
950
951
952
953
954static void check_process_timers(struct task_struct *tsk,
955 struct list_head *firing)
956{
957 struct signal_struct *const sig = tsk->signal;
958 unsigned long long utime, ptime, virt_expires, prof_expires;
959 unsigned long long sum_sched_runtime, sched_expires;
960 struct list_head *timers = sig->cpu_timers;
961 struct task_cputime cputime;
962 unsigned long soft;
963
964
965
966
967 thread_group_cputimer(tsk, &cputime);
968 utime = cputime_to_expires(cputime.utime);
969 ptime = utime + cputime_to_expires(cputime.stime);
970 sum_sched_runtime = cputime.sum_exec_runtime;
971
972 prof_expires = check_timers_list(timers, firing, ptime);
973 virt_expires = check_timers_list(++timers, firing, utime);
974 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
975
976
977
978
979 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
980 SIGPROF);
981 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
982 SIGVTALRM);
983 soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
984 if (soft != RLIM_INFINITY) {
985 unsigned long psecs = cputime_to_secs(ptime);
986 unsigned long hard =
987 READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
988 cputime_t x;
989 if (psecs >= hard) {
990
991
992
993
994 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
995 return;
996 }
997 if (psecs >= soft) {
998
999
1000
1001 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1002 if (soft < hard) {
1003 soft++;
1004 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1005 }
1006 }
1007 x = secs_to_cputime(soft);
1008 if (!prof_expires || x < prof_expires) {
1009 prof_expires = x;
1010 }
1011 }
1012
1013 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1014 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
1015 sig->cputime_expires.sched_exp = sched_expires;
1016 if (task_cputime_zero(&sig->cputime_expires))
1017 stop_process_timers(sig);
1018}
1019
1020
1021
1022
1023
1024void posix_cpu_timer_schedule(struct k_itimer *timer)
1025{
1026 struct sighand_struct *sighand;
1027 unsigned long flags;
1028 struct task_struct *p = timer->it.cpu.task;
1029 unsigned long long now;
1030
1031 WARN_ON_ONCE(p == NULL);
1032
1033
1034
1035
1036 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1037 cpu_clock_sample(timer->it_clock, p, &now);
1038 bump_cpu_timer(timer, now);
1039 if (unlikely(p->exit_state))
1040 goto out;
1041
1042
1043 sighand = lock_task_sighand(p, &flags);
1044 if (!sighand)
1045 goto out;
1046 } else {
1047
1048
1049
1050
1051 sighand = lock_task_sighand(p, &flags);
1052 if (unlikely(sighand == NULL)) {
1053
1054
1055
1056
1057 timer->it.cpu.expires = 0;
1058 goto out;
1059 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1060 unlock_task_sighand(p, &flags);
1061
1062 goto out;
1063 }
1064 cpu_timer_sample_group(timer->it_clock, p, &now);
1065 bump_cpu_timer(timer, now);
1066
1067 }
1068
1069
1070
1071
1072 WARN_ON_ONCE(!irqs_disabled());
1073 arm_timer(timer);
1074 unlock_task_sighand(p, &flags);
1075
1076
1077 posix_cpu_timer_kick_nohz();
1078out:
1079 timer->it_overrun_last = timer->it_overrun;
1080 timer->it_overrun = -1;
1081 ++timer->it_requeue_pending;
1082}
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094static inline int task_cputime_expired(const struct task_cputime *sample,
1095 const struct task_cputime *expires)
1096{
1097 if (expires->utime && sample->utime >= expires->utime)
1098 return 1;
1099 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1100 return 1;
1101 if (expires->sum_exec_runtime != 0 &&
1102 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1103 return 1;
1104 return 0;
1105}
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117static inline int fastpath_timer_check(struct task_struct *tsk)
1118{
1119 struct signal_struct *sig;
1120 cputime_t utime, stime;
1121
1122 task_cputime(tsk, &utime, &stime);
1123
1124 if (!task_cputime_zero(&tsk->cputime_expires)) {
1125 struct task_cputime task_sample = {
1126 .utime = utime,
1127 .stime = stime,
1128 .sum_exec_runtime = tsk->se.sum_exec_runtime
1129 };
1130
1131 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1132 return 1;
1133 }
1134
1135 sig = tsk->signal;
1136
1137 if (READ_ONCE(sig->cputimer.running)) {
1138 struct task_cputime group_sample;
1139
1140 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1141
1142 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1143 return 1;
1144 }
1145
1146 return 0;
1147}
1148
1149
1150
1151
1152
1153
1154void run_posix_cpu_timers(struct task_struct *tsk)
1155{
1156 LIST_HEAD(firing);
1157 struct k_itimer *timer, *next;
1158 unsigned long flags;
1159
1160 WARN_ON_ONCE(!irqs_disabled());
1161
1162
1163
1164
1165
1166 if (!fastpath_timer_check(tsk))
1167 return;
1168
1169 if (!lock_task_sighand(tsk, &flags))
1170 return;
1171
1172
1173
1174
1175
1176 check_thread_timers(tsk, &firing);
1177
1178
1179
1180
1181 if (READ_ONCE(tsk->signal->cputimer.running))
1182 check_process_timers(tsk, &firing);
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192 unlock_task_sighand(tsk, &flags);
1193
1194
1195
1196
1197
1198
1199
1200 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1201 int cpu_firing;
1202
1203 spin_lock(&timer->it_lock);
1204 list_del_init(&timer->it.cpu.entry);
1205 cpu_firing = timer->it.cpu.firing;
1206 timer->it.cpu.firing = 0;
1207
1208
1209
1210
1211
1212 if (likely(cpu_firing >= 0))
1213 cpu_timer_fire(timer);
1214 spin_unlock(&timer->it_lock);
1215 }
1216}
1217
1218
1219
1220
1221
1222void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1223 cputime_t *newval, cputime_t *oldval)
1224{
1225 unsigned long long now;
1226
1227 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1228 cpu_timer_sample_group(clock_idx, tsk, &now);
1229
1230 if (oldval) {
1231
1232
1233
1234
1235
1236 if (*oldval) {
1237 if (*oldval <= now) {
1238
1239 *oldval = cputime_one_jiffy;
1240 } else {
1241 *oldval -= now;
1242 }
1243 }
1244
1245 if (!*newval)
1246 goto out;
1247 *newval += now;
1248 }
1249
1250
1251
1252
1253
1254 switch (clock_idx) {
1255 case CPUCLOCK_PROF:
1256 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1257 tsk->signal->cputime_expires.prof_exp = *newval;
1258 break;
1259 case CPUCLOCK_VIRT:
1260 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1261 tsk->signal->cputime_expires.virt_exp = *newval;
1262 break;
1263 }
1264out:
1265 posix_cpu_timer_kick_nohz();
1266}
1267
1268static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1269 struct timespec *rqtp, struct itimerspec *it)
1270{
1271 struct k_itimer timer;
1272 int error;
1273
1274
1275
1276
1277 memset(&timer, 0, sizeof timer);
1278 spin_lock_init(&timer.it_lock);
1279 timer.it_clock = which_clock;
1280 timer.it_overrun = -1;
1281 error = posix_cpu_timer_create(&timer);
1282 timer.it_process = current;
1283 if (!error) {
1284 static struct itimerspec zero_it;
1285
1286 memset(it, 0, sizeof *it);
1287 it->it_value = *rqtp;
1288
1289 spin_lock_irq(&timer.it_lock);
1290 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1291 if (error) {
1292 spin_unlock_irq(&timer.it_lock);
1293 return error;
1294 }
1295
1296 while (!signal_pending(current)) {
1297 if (timer.it.cpu.expires == 0) {
1298
1299
1300
1301
1302 posix_cpu_timer_del(&timer);
1303 spin_unlock_irq(&timer.it_lock);
1304 return 0;
1305 }
1306
1307
1308
1309
1310 __set_current_state(TASK_INTERRUPTIBLE);
1311 spin_unlock_irq(&timer.it_lock);
1312 schedule();
1313 spin_lock_irq(&timer.it_lock);
1314 }
1315
1316
1317
1318
1319 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1320 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1321 if (!error) {
1322
1323
1324
1325 posix_cpu_timer_del(&timer);
1326 }
1327 spin_unlock_irq(&timer.it_lock);
1328
1329 while (error == TIMER_RETRY) {
1330
1331
1332
1333
1334
1335 spin_lock_irq(&timer.it_lock);
1336 error = posix_cpu_timer_del(&timer);
1337 spin_unlock_irq(&timer.it_lock);
1338 }
1339
1340 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1341
1342
1343
1344 return 0;
1345 }
1346
1347 error = -ERESTART_RESTARTBLOCK;
1348 }
1349
1350 return error;
1351}
1352
1353static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1354
1355static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1356 struct timespec *rqtp, struct timespec __user *rmtp)
1357{
1358 struct restart_block *restart_block = ¤t->restart_block;
1359 struct itimerspec it;
1360 int error;
1361
1362
1363
1364
1365 if (CPUCLOCK_PERTHREAD(which_clock) &&
1366 (CPUCLOCK_PID(which_clock) == 0 ||
1367 CPUCLOCK_PID(which_clock) == current->pid))
1368 return -EINVAL;
1369
1370 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1371
1372 if (error == -ERESTART_RESTARTBLOCK) {
1373
1374 if (flags & TIMER_ABSTIME)
1375 return -ERESTARTNOHAND;
1376
1377
1378
1379 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1380 return -EFAULT;
1381
1382 restart_block->fn = posix_cpu_nsleep_restart;
1383 restart_block->nanosleep.clockid = which_clock;
1384 restart_block->nanosleep.rmtp = rmtp;
1385 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1386 }
1387 return error;
1388}
1389
1390static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1391{
1392 clockid_t which_clock = restart_block->nanosleep.clockid;
1393 struct timespec t;
1394 struct itimerspec it;
1395 int error;
1396
1397 t = ns_to_timespec(restart_block->nanosleep.expires);
1398
1399 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1400
1401 if (error == -ERESTART_RESTARTBLOCK) {
1402 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1403
1404
1405
1406 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1407 return -EFAULT;
1408
1409 restart_block->nanosleep.expires = timespec_to_ns(&t);
1410 }
1411 return error;
1412
1413}
1414
1415#define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1416#define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1417
1418static int process_cpu_clock_getres(const clockid_t which_clock,
1419 struct timespec *tp)
1420{
1421 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1422}
1423static int process_cpu_clock_get(const clockid_t which_clock,
1424 struct timespec *tp)
1425{
1426 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1427}
1428static int process_cpu_timer_create(struct k_itimer *timer)
1429{
1430 timer->it_clock = PROCESS_CLOCK;
1431 return posix_cpu_timer_create(timer);
1432}
1433static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1434 struct timespec *rqtp,
1435 struct timespec __user *rmtp)
1436{
1437 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1438}
1439static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1440{
1441 return -EINVAL;
1442}
1443static int thread_cpu_clock_getres(const clockid_t which_clock,
1444 struct timespec *tp)
1445{
1446 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1447}
1448static int thread_cpu_clock_get(const clockid_t which_clock,
1449 struct timespec *tp)
1450{
1451 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1452}
1453static int thread_cpu_timer_create(struct k_itimer *timer)
1454{
1455 timer->it_clock = THREAD_CLOCK;
1456 return posix_cpu_timer_create(timer);
1457}
1458
1459struct k_clock clock_posix_cpu = {
1460 .clock_getres = posix_cpu_clock_getres,
1461 .clock_set = posix_cpu_clock_set,
1462 .clock_get = posix_cpu_clock_get,
1463 .timer_create = posix_cpu_timer_create,
1464 .nsleep = posix_cpu_nsleep,
1465 .nsleep_restart = posix_cpu_nsleep_restart,
1466 .timer_set = posix_cpu_timer_set,
1467 .timer_del = posix_cpu_timer_del,
1468 .timer_get = posix_cpu_timer_get,
1469};
1470
1471static __init int init_posix_cpu_timers(void)
1472{
1473 struct k_clock process = {
1474 .clock_getres = process_cpu_clock_getres,
1475 .clock_get = process_cpu_clock_get,
1476 .timer_create = process_cpu_timer_create,
1477 .nsleep = process_cpu_nsleep,
1478 .nsleep_restart = process_cpu_nsleep_restart,
1479 };
1480 struct k_clock thread = {
1481 .clock_getres = thread_cpu_clock_getres,
1482 .clock_get = thread_cpu_clock_get,
1483 .timer_create = thread_cpu_timer_create,
1484 };
1485 struct timespec ts;
1486
1487 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1488 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1489
1490 cputime_to_timespec(cputime_one_jiffy, &ts);
1491 onecputick = ts.tv_nsec;
1492 WARN_ON(ts.tv_sec != 0);
1493
1494 return 0;
1495}
1496__initcall(init_posix_cpu_timers);
1497