]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - kernel/time/posix-cpu-timers.c
posix_cpu_timer: Optimize fastpath_timer_check()
[mirror_ubuntu-jammy-kernel.git] / kernel / time / posix-cpu-timers.c
CommitLineData
1da177e4
LT
1/*
2 * Implement CPU time clocks for the POSIX clock interface.
3 */
4
5#include <linux/sched.h>
6#include <linux/posix-timers.h>
1da177e4 7#include <linux/errno.h>
f8bd2258
RZ
8#include <linux/math64.h>
9#include <asm/uaccess.h>
bb34d92f 10#include <linux/kernel_stat.h>
3f0a525e 11#include <trace/events/timer.h>
61337054 12#include <linux/random.h>
a8572160
FW
13#include <linux/tick.h>
14#include <linux/workqueue.h>
1da177e4 15
f06febc9 16/*
f55db609
SG
17 * Called after updating RLIMIT_CPU to run cpu timer and update
18 * tsk->signal->cputime_expires expiration cache if necessary. Needs
19 * siglock protection since other code may update expiration cache as
20 * well.
f06febc9 21 */
5ab46b34 22void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
f06febc9 23{
42c4ab41 24 cputime_t cputime = secs_to_cputime(rlim_new);
f06febc9 25
5ab46b34
JS
26 spin_lock_irq(&task->sighand->siglock);
27 set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
28 spin_unlock_irq(&task->sighand->siglock);
f06febc9
FM
29}
30
a924b04d 31static int check_clock(const clockid_t which_clock)
1da177e4
LT
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
c0deae8c 43 rcu_read_lock();
8dc86af0 44 p = find_task_by_vpid(pid);
bac0abd6 45 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
c0deae8c 46 same_thread_group(p, current) : has_group_leader_pid(p))) {
1da177e4
LT
47 error = -EINVAL;
48 }
c0deae8c 49 rcu_read_unlock();
1da177e4
LT
50
51 return error;
52}
53
55ccb616 54static inline unsigned long long
a924b04d 55timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
1da177e4 56{
55ccb616
FW
57 unsigned long long ret;
58
59 ret = 0; /* high half always zero when .cpu used */
1da177e4 60 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
55ccb616 61 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
1da177e4 62 } else {
55ccb616 63 ret = cputime_to_expires(timespec_to_cputime(tp));
1da177e4
LT
64 }
65 return ret;
66}
67
a924b04d 68static void sample_to_timespec(const clockid_t which_clock,
55ccb616 69 unsigned long long expires,
1da177e4
LT
70 struct timespec *tp)
71{
f8bd2258 72 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
55ccb616 73 *tp = ns_to_timespec(expires);
f8bd2258 74 else
55ccb616 75 cputime_to_timespec((__force cputime_t)expires, tp);
1da177e4
LT
76}
77
78/*
79 * Update expiry time from increment, and increase overrun count,
80 * given the current clock sample.
81 */
7a4ed937 82static void bump_cpu_timer(struct k_itimer *timer,
55ccb616 83 unsigned long long now)
1da177e4
LT
84{
85 int i;
55ccb616 86 unsigned long long delta, incr;
1da177e4 87
55ccb616 88 if (timer->it.cpu.incr == 0)
1da177e4
LT
89 return;
90
55ccb616
FW
91 if (now < timer->it.cpu.expires)
92 return;
1da177e4 93
55ccb616
FW
94 incr = timer->it.cpu.incr;
95 delta = now + incr - timer->it.cpu.expires;
1da177e4 96
55ccb616
FW
97 /* Don't use (incr*2 < delta), incr*2 might overflow. */
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;
1da177e4
LT
108 }
109}
110
555347f6
FW
111/**
112 * task_cputime_zero - Check a task_cputime struct for all zero fields.
113 *
114 * @cputime: The struct to compare.
115 *
116 * Checks @cputime to see if all fields are zero. Returns true if all fields
117 * are zero, false if any field is nonzero.
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
55ccb616 126static inline unsigned long long prof_ticks(struct task_struct *p)
1da177e4 127{
6fac4829
FW
128 cputime_t utime, stime;
129
130 task_cputime(p, &utime, &stime);
131
55ccb616 132 return cputime_to_expires(utime + stime);
1da177e4 133}
55ccb616 134static inline unsigned long long virt_ticks(struct task_struct *p)
1da177e4 135{
6fac4829
FW
136 cputime_t utime;
137
138 task_cputime(p, &utime, NULL);
139
55ccb616 140 return cputime_to_expires(utime);
1da177e4 141}
1da177e4 142
bc2c8ea4
TG
143static int
144posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
1da177e4
LT
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 * If sched_clock is using a cycle counter, we
153 * don't have any idea of its true resolution
154 * exported, but it is much more than 1s/HZ.
155 */
156 tp->tv_nsec = 1;
157 }
158 }
159 return error;
160}
161
bc2c8ea4
TG
162static int
163posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
1da177e4
LT
164{
165 /*
166 * You can never reset a CPU clock, but we check for other errors
167 * in the call before failing with EPERM.
168 */
169 int error = check_clock(which_clock);
170 if (error == 0) {
171 error = -EPERM;
172 }
173 return error;
174}
175
176
177/*
178 * Sample a per-thread clock for the given task.
179 */
a924b04d 180static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
55ccb616 181 unsigned long long *sample)
1da177e4
LT
182{
183 switch (CPUCLOCK_WHICH(which_clock)) {
184 default:
185 return -EINVAL;
186 case CPUCLOCK_PROF:
55ccb616 187 *sample = prof_ticks(p);
1da177e4
LT
188 break;
189 case CPUCLOCK_VIRT:
55ccb616 190 *sample = virt_ticks(p);
1da177e4
LT
191 break;
192 case CPUCLOCK_SCHED:
55ccb616 193 *sample = task_sched_runtime(p);
1da177e4
LT
194 break;
195 }
196 return 0;
197}
198
1018016c
JL
199/*
200 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
201 * to avoid race conditions with concurrent updates to cputime.
202 */
203static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
4da94d49 204{
1018016c
JL
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}
4da94d49 213
71107445 214static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
1018016c 215{
71107445
JL
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);
1018016c 219}
4da94d49 220
71107445
JL
221/* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
222static inline void sample_cputime_atomic(struct task_cputime *times,
223 struct task_cputime_atomic *atomic_times)
1018016c 224{
71107445
JL
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);
4da94d49
PZ
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;
4da94d49 234
1018016c
JL
235 /* Check if cputimer isn't running. This is accessed without locking. */
236 if (!READ_ONCE(cputimer->running)) {
4da94d49
PZ
237 /*
238 * The POSIX timer interface allows for absolute time expiry
239 * values through the TIMER_ABSTIME flag, therefore we have
1018016c 240 * to synchronize the timer to the clock every time we start it.
4da94d49
PZ
241 */
242 thread_group_cputime(tsk, &sum);
71107445 243 update_gt_cputime(&cputimer->cputime_atomic, &sum);
1018016c
JL
244
245 /*
246 * We're setting cputimer->running without a lock. Ensure
247 * this only gets written to in one operation. We set
248 * running after update_gt_cputime() as a small optimization,
249 * but barriers are not required because update_gt_cputime()
250 * can handle concurrent updates.
251 */
252 WRITE_ONCE(cputimer->running, 1);
253 }
71107445 254 sample_cputime_atomic(times, &cputimer->cputime_atomic);
4da94d49
PZ
255}
256
1da177e4
LT
257/*
258 * Sample a process (thread group) clock for the given group_leader task.
e73d84e3
FW
259 * Must be called with task sighand lock held for safe while_each_thread()
260 * traversal.
1da177e4 261 */
bb34d92f
FM
262static int cpu_clock_sample_group(const clockid_t which_clock,
263 struct task_struct *p,
55ccb616 264 unsigned long long *sample)
1da177e4 265{
f06febc9
FM
266 struct task_cputime cputime;
267
eccdaeaf 268 switch (CPUCLOCK_WHICH(which_clock)) {
1da177e4
LT
269 default:
270 return -EINVAL;
271 case CPUCLOCK_PROF:
c5f8d995 272 thread_group_cputime(p, &cputime);
55ccb616 273 *sample = cputime_to_expires(cputime.utime + cputime.stime);
1da177e4
LT
274 break;
275 case CPUCLOCK_VIRT:
c5f8d995 276 thread_group_cputime(p, &cputime);
55ccb616 277 *sample = cputime_to_expires(cputime.utime);
1da177e4
LT
278 break;
279 case CPUCLOCK_SCHED:
d670ec13 280 thread_group_cputime(p, &cputime);
55ccb616 281 *sample = cputime.sum_exec_runtime;
1da177e4
LT
282 break;
283 }
284 return 0;
285}
286
33ab0fec
FW
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 {
50875788 298 if (tsk == current || thread_group_leader(tsk))
33ab0fec 299 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
33ab0fec
FW
300 }
301
302 if (!err)
303 sample_to_timespec(which_clock, rtn, tp);
304
305 return err;
306}
307
1da177e4 308
bc2c8ea4 309static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
1da177e4
LT
310{
311 const pid_t pid = CPUCLOCK_PID(which_clock);
33ab0fec 312 int err = -EINVAL;
1da177e4
LT
313
314 if (pid == 0) {
315 /*
316 * Special case constant value for our own clocks.
317 * We don't have to do any lookup to find ourselves.
318 */
33ab0fec 319 err = posix_cpu_clock_get_task(current, which_clock, tp);
1da177e4
LT
320 } else {
321 /*
322 * Find the given PID, and validate that the caller
323 * should be able to see it.
324 */
325 struct task_struct *p;
1f2ea083 326 rcu_read_lock();
8dc86af0 327 p = find_task_by_vpid(pid);
33ab0fec
FW
328 if (p)
329 err = posix_cpu_clock_get_task(p, which_clock, tp);
1f2ea083 330 rcu_read_unlock();
1da177e4
LT
331 }
332
33ab0fec 333 return err;
1da177e4
LT
334}
335
336
337/*
338 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
ba5ea951
SG
339 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
340 * new timer already all-zeros initialized.
1da177e4 341 */
bc2c8ea4 342static int posix_cpu_timer_create(struct k_itimer *new_timer)
1da177e4
LT
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);
1da177e4 352
c0deae8c 353 rcu_read_lock();
1da177e4
LT
354 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
355 if (pid == 0) {
356 p = current;
357 } else {
8dc86af0 358 p = find_task_by_vpid(pid);
bac0abd6 359 if (p && !same_thread_group(p, current))
1da177e4
LT
360 p = NULL;
361 }
362 } else {
363 if (pid == 0) {
364 p = current->group_leader;
365 } else {
8dc86af0 366 p = find_task_by_vpid(pid);
c0deae8c 367 if (p && !has_group_leader_pid(p))
1da177e4
LT
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 }
c0deae8c 377 rcu_read_unlock();
1da177e4
LT
378
379 return ret;
380}
381
382/*
383 * Clean up a CPU-clock timer that is about to be destroyed.
384 * This is called from timer deletion with the timer already locked.
385 * If we return TIMER_RETRY, it's necessary to release the timer's lock
386 * and try again. (This happens when the timer is in the middle of firing.)
387 */
bc2c8ea4 388static int posix_cpu_timer_del(struct k_itimer *timer)
1da177e4 389{
108150ea 390 int ret = 0;
3d7a1427
FW
391 unsigned long flags;
392 struct sighand_struct *sighand;
393 struct task_struct *p = timer->it.cpu.task;
1da177e4 394
a3222f88 395 WARN_ON_ONCE(p == NULL);
108150ea 396
3d7a1427
FW
397 /*
398 * Protect against sighand release/switch in exit/exec and process/
399 * thread timer list entry concurrent read/writes.
400 */
401 sighand = lock_task_sighand(p, &flags);
402 if (unlikely(sighand == NULL)) {
a3222f88
FW
403 /*
404 * We raced with the reaping of the task.
405 * The deletion should have cleared us off the list.
406 */
531f64fd 407 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
a3222f88 408 } else {
a3222f88
FW
409 if (timer->it.cpu.firing)
410 ret = TIMER_RETRY;
411 else
412 list_del(&timer->it.cpu.entry);
3d7a1427
FW
413
414 unlock_task_sighand(p, &flags);
1da177e4 415 }
a3222f88
FW
416
417 if (!ret)
418 put_task_struct(p);
1da177e4 419
108150ea 420 return ret;
1da177e4
LT
421}
422
af82eb3c 423static void cleanup_timers_list(struct list_head *head)
1a7fa510
FW
424{
425 struct cpu_timer_list *timer, *next;
426
a0b2062b 427 list_for_each_entry_safe(timer, next, head, entry)
1a7fa510 428 list_del_init(&timer->entry);
1a7fa510
FW
429}
430
1da177e4
LT
431/*
432 * Clean out CPU timers still ticking when a thread exited. The task
433 * pointer is cleared, and the expiry time is replaced with the residual
434 * time for later timer_gettime calls to return.
435 * This must be called with the siglock held.
436 */
af82eb3c 437static void cleanup_timers(struct list_head *head)
1da177e4 438{
af82eb3c
FW
439 cleanup_timers_list(head);
440 cleanup_timers_list(++head);
441 cleanup_timers_list(++head);
1da177e4
LT
442}
443
444/*
445 * These are both called with the siglock held, when the current thread
446 * is being reaped. When the final (leader) thread in the group is reaped,
447 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
448 */
449void posix_cpu_timers_exit(struct task_struct *tsk)
450{
61337054
NK
451 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
452 sizeof(unsigned long long));
af82eb3c 453 cleanup_timers(tsk->cpu_timers);
1da177e4
LT
454
455}
456void posix_cpu_timers_exit_group(struct task_struct *tsk)
457{
af82eb3c 458 cleanup_timers(tsk->signal->cpu_timers);
1da177e4
LT
459}
460
d1e3b6d1
SG
461static inline int expires_gt(cputime_t expires, cputime_t new_exp)
462{
64861634 463 return expires == 0 || expires > new_exp;
d1e3b6d1
SG
464}
465
1da177e4
LT
466/*
467 * Insert the timer on the appropriate list before any timers that
e73d84e3 468 * expire later. This must be called with the sighand lock held.
1da177e4 469 */
5eb9aa64 470static void arm_timer(struct k_itimer *timer)
1da177e4
LT
471{
472 struct task_struct *p = timer->it.cpu.task;
473 struct list_head *head, *listpos;
5eb9aa64 474 struct task_cputime *cputime_expires;
1da177e4
LT
475 struct cpu_timer_list *const nt = &timer->it.cpu;
476 struct cpu_timer_list *next;
1da177e4 477
5eb9aa64
SG
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 }
1da177e4
LT
485 head += CPUCLOCK_WHICH(timer->it_clock);
486
1da177e4 487 listpos = head;
5eb9aa64 488 list_for_each_entry(next, head, entry) {
55ccb616 489 if (nt->expires < next->expires)
5eb9aa64
SG
490 break;
491 listpos = &next->entry;
1da177e4
LT
492 }
493 list_add(&nt->entry, listpos);
494
495 if (listpos == head) {
55ccb616 496 unsigned long long exp = nt->expires;
5eb9aa64 497
1da177e4 498 /*
5eb9aa64
SG
499 * We are the new earliest-expiring POSIX 1.b timer, hence
500 * need to update expiration cache. Take into account that
501 * for process timers we share expiration cache with itimers
502 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
1da177e4
LT
503 */
504
5eb9aa64
SG
505 switch (CPUCLOCK_WHICH(timer->it_clock)) {
506 case CPUCLOCK_PROF:
55ccb616
FW
507 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
508 cputime_expires->prof_exp = expires_to_cputime(exp);
5eb9aa64
SG
509 break;
510 case CPUCLOCK_VIRT:
55ccb616
FW
511 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
512 cputime_expires->virt_exp = expires_to_cputime(exp);
5eb9aa64
SG
513 break;
514 case CPUCLOCK_SCHED:
515 if (cputime_expires->sched_exp == 0 ||
55ccb616
FW
516 cputime_expires->sched_exp > exp)
517 cputime_expires->sched_exp = exp;
5eb9aa64 518 break;
1da177e4
LT
519 }
520 }
1da177e4
LT
521}
522
523/*
524 * The timer is locked, fire it and arrange for its reload.
525 */
526static void cpu_timer_fire(struct k_itimer *timer)
527{
1f169f84
SG
528 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
529 /*
530 * User don't want any signal.
531 */
55ccb616 532 timer->it.cpu.expires = 0;
1f169f84 533 } else if (unlikely(timer->sigq == NULL)) {
1da177e4
LT
534 /*
535 * This a special case for clock_nanosleep,
536 * not a normal timer from sys_timer_create.
537 */
538 wake_up_process(timer->it_process);
55ccb616
FW
539 timer->it.cpu.expires = 0;
540 } else if (timer->it.cpu.incr == 0) {
1da177e4
LT
541 /*
542 * One-shot timer. Clear it as soon as it's fired.
543 */
544 posix_timer_event(timer, 0);
55ccb616 545 timer->it.cpu.expires = 0;
1da177e4
LT
546 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
547 /*
548 * The signal did not get queued because the signal
549 * was ignored, so we won't get any callback to
550 * reload the timer. But we need to keep it
551 * ticking in case the signal is deliverable next time.
552 */
553 posix_cpu_timer_schedule(timer);
554 }
555}
556
3997ad31
PZ
557/*
558 * Sample a process (thread group) timer for the given group_leader task.
e73d84e3
FW
559 * Must be called with task sighand lock held for safe while_each_thread()
560 * traversal.
3997ad31
PZ
561 */
562static int cpu_timer_sample_group(const clockid_t which_clock,
563 struct task_struct *p,
55ccb616 564 unsigned long long *sample)
3997ad31
PZ
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:
55ccb616 573 *sample = cputime_to_expires(cputime.utime + cputime.stime);
3997ad31
PZ
574 break;
575 case CPUCLOCK_VIRT:
55ccb616 576 *sample = cputime_to_expires(cputime.utime);
3997ad31
PZ
577 break;
578 case CPUCLOCK_SCHED:
23cfa361 579 *sample = cputime.sum_exec_runtime;
3997ad31
PZ
580 break;
581 }
582 return 0;
583}
584
a8572160
FW
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 * We need the IPIs to be sent from sane process context.
595 * The posix cpu timers are always set with irqs disabled.
596 */
597static void posix_cpu_timer_kick_nohz(void)
598{
d4283c65
FW
599 if (context_tracking_is_enabled())
600 schedule_work(&nohz_kick_work);
a8572160 601}
555347f6
FW
602
603bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
604{
605 if (!task_cputime_zero(&tsk->cputime_expires))
6ac29178 606 return false;
555347f6 607
1018016c
JL
608 /* Check if cputimer is running. This is accessed without locking. */
609 if (READ_ONCE(tsk->signal->cputimer.running))
6ac29178 610 return false;
555347f6 611
6ac29178 612 return true;
555347f6 613}
a8572160
FW
614#else
615static inline void posix_cpu_timer_kick_nohz(void) { }
616#endif
617
1da177e4
LT
618/*
619 * Guts of sys_timer_settime for CPU timers.
620 * This is called with the timer locked and interrupts disabled.
621 * If we return TIMER_RETRY, it's necessary to release the timer's lock
622 * and try again. (This happens when the timer is in the middle of firing.)
623 */
e73d84e3 624static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
bc2c8ea4 625 struct itimerspec *new, struct itimerspec *old)
1da177e4 626{
e73d84e3
FW
627 unsigned long flags;
628 struct sighand_struct *sighand;
1da177e4 629 struct task_struct *p = timer->it.cpu.task;
55ccb616 630 unsigned long long old_expires, new_expires, old_incr, val;
1da177e4
LT
631 int ret;
632
a3222f88 633 WARN_ON_ONCE(p == NULL);
1da177e4
LT
634
635 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
636
1da177e4 637 /*
e73d84e3
FW
638 * Protect against sighand release/switch in exit/exec and p->cpu_timers
639 * and p->signal->cpu_timers read/write in arm_timer()
640 */
641 sighand = lock_task_sighand(p, &flags);
642 /*
643 * If p has just been reaped, we can no
1da177e4
LT
644 * longer get any information about it at all.
645 */
e73d84e3 646 if (unlikely(sighand == NULL)) {
1da177e4
LT
647 return -ESRCH;
648 }
649
650 /*
651 * Disarm any old timer after extracting its expiry time.
652 */
531f64fd 653 WARN_ON_ONCE(!irqs_disabled());
a69ac4a7
ON
654
655 ret = 0;
ae1a78ee 656 old_incr = timer->it.cpu.incr;
1da177e4 657 old_expires = timer->it.cpu.expires;
a69ac4a7
ON
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);
1da177e4
LT
663
664 /*
665 * We need to sample the current value to convert the new
666 * value from to relative and absolute, and to convert the
667 * old value from absolute to relative. To set a process
668 * timer, we need a sample to balance the thread expiry
669 * times (in arm_timer). With an absolute time, we must
670 * check if it's already passed. In short, we need a sample.
671 */
672 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
673 cpu_clock_sample(timer->it_clock, p, &val);
674 } else {
3997ad31 675 cpu_timer_sample_group(timer->it_clock, p, &val);
1da177e4
LT
676 }
677
678 if (old) {
55ccb616 679 if (old_expires == 0) {
1da177e4
LT
680 old->it_value.tv_sec = 0;
681 old->it_value.tv_nsec = 0;
682 } else {
683 /*
684 * Update the timer in case it has
685 * overrun already. If it has,
686 * we'll report it as having overrun
687 * and with the next reloaded timer
688 * already ticking, though we are
689 * swallowing that pending
690 * notification here to install the
691 * new setting.
692 */
693 bump_cpu_timer(timer, val);
55ccb616
FW
694 if (val < timer->it.cpu.expires) {
695 old_expires = timer->it.cpu.expires - val;
1da177e4
LT
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
a69ac4a7 706 if (unlikely(ret)) {
1da177e4
LT
707 /*
708 * We are colliding with the timer actually firing.
709 * Punt after filling in the timer's old value, and
710 * disable this firing since we are already reporting
711 * it as an overrun (thanks to bump_cpu_timer above).
712 */
e73d84e3 713 unlock_task_sighand(p, &flags);
1da177e4
LT
714 goto out;
715 }
716
e73d84e3 717 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
55ccb616 718 new_expires += val;
1da177e4
LT
719 }
720
721 /*
722 * Install the new expiry time (or zero).
723 * For a timer with no notification action, we don't actually
724 * arm the timer (we'll just fake it for timer_gettime).
725 */
726 timer->it.cpu.expires = new_expires;
55ccb616 727 if (new_expires != 0 && val < new_expires) {
5eb9aa64 728 arm_timer(timer);
1da177e4
LT
729 }
730
e73d84e3 731 unlock_task_sighand(p, &flags);
1da177e4
LT
732 /*
733 * Install the new reload setting, and
734 * set up the signal and overrun bookkeeping.
735 */
736 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
737 &new->it_interval);
738
739 /*
740 * This acts as a modification timestamp for the timer,
741 * so any automatic reload attempt will punt on seeing
742 * that we have reset the timer manually.
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
55ccb616 749 if (new_expires != 0 && !(val < new_expires)) {
1da177e4
LT
750 /*
751 * The designated time already passed, so we notify
752 * immediately, even if the thread never runs to
753 * accumulate more time on this clock.
754 */
755 cpu_timer_fire(timer);
756 }
757
758 ret = 0;
759 out:
760 if (old) {
761 sample_to_timespec(timer->it_clock,
ae1a78ee 762 old_incr, &old->it_interval);
1da177e4 763 }
a8572160
FW
764 if (!ret)
765 posix_cpu_timer_kick_nohz();
1da177e4
LT
766 return ret;
767}
768
bc2c8ea4 769static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
1da177e4 770{
55ccb616 771 unsigned long long now;
1da177e4 772 struct task_struct *p = timer->it.cpu.task;
1da177e4 773
a3222f88
FW
774 WARN_ON_ONCE(p == NULL);
775
1da177e4
LT
776 /*
777 * Easy part: convert the reload time.
778 */
779 sample_to_timespec(timer->it_clock,
780 timer->it.cpu.incr, &itp->it_interval);
781
55ccb616 782 if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */
1da177e4
LT
783 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
784 return;
785 }
786
1da177e4
LT
787 /*
788 * Sample the clock to take the difference with the expiry time.
789 */
790 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
791 cpu_clock_sample(timer->it_clock, p, &now);
1da177e4 792 } else {
e73d84e3
FW
793 struct sighand_struct *sighand;
794 unsigned long flags;
795
796 /*
797 * Protect against sighand release/switch in exit/exec and
798 * also make timer sampling safe if it ends up calling
799 * thread_group_cputime().
800 */
801 sighand = lock_task_sighand(p, &flags);
802 if (unlikely(sighand == NULL)) {
1da177e4
LT
803 /*
804 * The process has been reaped.
805 * We can't even collect a sample any more.
806 * Call the timer disarmed, nothing else to do.
807 */
55ccb616 808 timer->it.cpu.expires = 0;
a3222f88
FW
809 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
810 &itp->it_value);
1da177e4 811 } else {
3997ad31 812 cpu_timer_sample_group(timer->it_clock, p, &now);
e73d84e3 813 unlock_task_sighand(p, &flags);
1da177e4 814 }
1da177e4
LT
815 }
816
55ccb616 817 if (now < timer->it.cpu.expires) {
1da177e4 818 sample_to_timespec(timer->it_clock,
55ccb616 819 timer->it.cpu.expires - now,
1da177e4
LT
820 &itp->it_value);
821 } else {
822 /*
823 * The timer should have expired already, but the firing
824 * hasn't taken place yet. Say it's just about to expire.
825 */
826 itp->it_value.tv_nsec = 1;
827 itp->it_value.tv_sec = 0;
828 }
829}
830
2473f3e7
FW
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
1da177e4
LT
853/*
854 * Check for any per-thread CPU timers that have fired and move them off
855 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
856 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
857 */
858static void check_thread_timers(struct task_struct *tsk,
859 struct list_head *firing)
860{
861 struct list_head *timers = tsk->cpu_timers;
78f2c7db 862 struct signal_struct *const sig = tsk->signal;
2473f3e7
FW
863 struct task_cputime *tsk_expires = &tsk->cputime_expires;
864 unsigned long long expires;
d4bb5274 865 unsigned long soft;
1da177e4 866
2473f3e7
FW
867 expires = check_timers_list(timers, firing, prof_ticks(tsk));
868 tsk_expires->prof_exp = expires_to_cputime(expires);
1da177e4 869
2473f3e7
FW
870 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
871 tsk_expires->virt_exp = expires_to_cputime(expires);
1da177e4 872
2473f3e7
FW
873 tsk_expires->sched_exp = check_timers_list(++timers, firing,
874 tsk->se.sum_exec_runtime);
78f2c7db
PZ
875
876 /*
877 * Check for the special case thread timers.
878 */
316c1608 879 soft = READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
d4bb5274 880 if (soft != RLIM_INFINITY) {
78d7d407 881 unsigned long hard =
316c1608 882 READ_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
78f2c7db 883
5a52dd50
PZ
884 if (hard != RLIM_INFINITY &&
885 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
78f2c7db
PZ
886 /*
887 * At the hard limit, we just die.
888 * No need to calculate anything else now.
889 */
890 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
891 return;
892 }
d4bb5274 893 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
78f2c7db
PZ
894 /*
895 * At the soft limit, send a SIGXCPU every second.
896 */
d4bb5274
JS
897 if (soft < hard) {
898 soft += USEC_PER_SEC;
899 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
78f2c7db 900 }
81d50bb2
HS
901 printk(KERN_INFO
902 "RT Watchdog Timeout: %s[%d]\n",
903 tsk->comm, task_pid_nr(tsk));
78f2c7db
PZ
904 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
905 }
906 }
1da177e4
LT
907}
908
1018016c 909static inline void stop_process_timers(struct signal_struct *sig)
3fccfd67 910{
15365c10 911 struct thread_group_cputimer *cputimer = &sig->cputimer;
3fccfd67 912
1018016c
JL
913 /* Turn off cputimer->running. This is done without locking. */
914 WRITE_ONCE(cputimer->running, 0);
3fccfd67
PZ
915}
916
8356b5f9
SG
917static u32 onecputick;
918
42c4ab41 919static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
55ccb616
FW
920 unsigned long long *expires,
921 unsigned long long cur_time, int signo)
42c4ab41 922{
64861634 923 if (!it->expires)
42c4ab41
SG
924 return;
925
64861634
MS
926 if (cur_time >= it->expires) {
927 if (it->incr) {
928 it->expires += it->incr;
8356b5f9
SG
929 it->error += it->incr_error;
930 if (it->error >= onecputick) {
64861634 931 it->expires -= cputime_one_jiffy;
8356b5f9
SG
932 it->error -= onecputick;
933 }
3f0a525e 934 } else {
64861634 935 it->expires = 0;
3f0a525e 936 }
42c4ab41 937
3f0a525e
XG
938 trace_itimer_expire(signo == SIGPROF ?
939 ITIMER_PROF : ITIMER_VIRTUAL,
940 tsk->signal->leader_pid, cur_time);
42c4ab41
SG
941 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
942 }
943
64861634 944 if (it->expires && (!*expires || it->expires < *expires)) {
42c4ab41
SG
945 *expires = it->expires;
946 }
947}
948
1da177e4
LT
949/*
950 * Check for any per-thread CPU timers that have fired and move them
951 * off the tsk->*_timers list onto the firing list. Per-thread timers
952 * have already been taken off.
953 */
954static void check_process_timers(struct task_struct *tsk,
955 struct list_head *firing)
956{
957 struct signal_struct *const sig = tsk->signal;
55ccb616 958 unsigned long long utime, ptime, virt_expires, prof_expires;
41b86e9c 959 unsigned long long sum_sched_runtime, sched_expires;
1da177e4 960 struct list_head *timers = sig->cpu_timers;
f06febc9 961 struct task_cputime cputime;
d4bb5274 962 unsigned long soft;
1da177e4 963
1da177e4
LT
964 /*
965 * Collect the current process totals.
966 */
4cd4c1b4 967 thread_group_cputimer(tsk, &cputime);
55ccb616
FW
968 utime = cputime_to_expires(cputime.utime);
969 ptime = utime + cputime_to_expires(cputime.stime);
f06febc9 970 sum_sched_runtime = cputime.sum_exec_runtime;
1da177e4 971
2473f3e7
FW
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);
1da177e4
LT
975
976 /*
977 * Check for the special case process timers.
978 */
42c4ab41
SG
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);
316c1608 983 soft = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
d4bb5274 984 if (soft != RLIM_INFINITY) {
1da177e4 985 unsigned long psecs = cputime_to_secs(ptime);
78d7d407 986 unsigned long hard =
316c1608 987 READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1da177e4 988 cputime_t x;
d4bb5274 989 if (psecs >= hard) {
1da177e4
LT
990 /*
991 * At the hard limit, we just die.
992 * No need to calculate anything else now.
993 */
994 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
995 return;
996 }
d4bb5274 997 if (psecs >= soft) {
1da177e4
LT
998 /*
999 * At the soft limit, send a SIGXCPU every second.
1000 */
1001 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
d4bb5274
JS
1002 if (soft < hard) {
1003 soft++;
1004 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1da177e4
LT
1005 }
1006 }
d4bb5274 1007 x = secs_to_cputime(soft);
64861634 1008 if (!prof_expires || x < prof_expires) {
1da177e4
LT
1009 prof_expires = x;
1010 }
1011 }
1012
55ccb616
FW
1013 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1014 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
29f87b79
SG
1015 sig->cputime_expires.sched_exp = sched_expires;
1016 if (task_cputime_zero(&sig->cputime_expires))
1017 stop_process_timers(sig);
1da177e4
LT
1018}
1019
1020/*
1021 * This is called from the signal code (via do_schedule_next_timer)
1022 * when the last timer signal was delivered and we have to reload the timer.
1023 */
1024void posix_cpu_timer_schedule(struct k_itimer *timer)
1025{
e73d84e3
FW
1026 struct sighand_struct *sighand;
1027 unsigned long flags;
1da177e4 1028 struct task_struct *p = timer->it.cpu.task;
55ccb616 1029 unsigned long long now;
1da177e4 1030
a3222f88 1031 WARN_ON_ONCE(p == NULL);
1da177e4
LT
1032
1033 /*
1034 * Fetch the current sample and update the timer's expiry time.
1035 */
1036 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1037 cpu_clock_sample(timer->it_clock, p, &now);
1038 bump_cpu_timer(timer, now);
724a3713 1039 if (unlikely(p->exit_state))
708f430d 1040 goto out;
724a3713 1041
e73d84e3
FW
1042 /* Protect timer list r/w in arm_timer() */
1043 sighand = lock_task_sighand(p, &flags);
1044 if (!sighand)
1045 goto out;
1da177e4 1046 } else {
e73d84e3
FW
1047 /*
1048 * Protect arm_timer() and timer sampling in case of call to
1049 * thread_group_cputime().
1050 */
1051 sighand = lock_task_sighand(p, &flags);
1052 if (unlikely(sighand == NULL)) {
1da177e4
LT
1053 /*
1054 * The process has been reaped.
1055 * We can't even collect a sample any more.
1056 */
55ccb616 1057 timer->it.cpu.expires = 0;
c925077c 1058 goto out;
1da177e4 1059 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
e73d84e3 1060 unlock_task_sighand(p, &flags);
d430b917 1061 /* Optimizations: if the process is dying, no need to rearm */
c925077c 1062 goto out;
1da177e4 1063 }
3997ad31 1064 cpu_timer_sample_group(timer->it_clock, p, &now);
1da177e4 1065 bump_cpu_timer(timer, now);
e73d84e3 1066 /* Leave the sighand locked for the call below. */
1da177e4
LT
1067 }
1068
1069 /*
1070 * Now re-arm for the new expiry time.
1071 */
531f64fd 1072 WARN_ON_ONCE(!irqs_disabled());
5eb9aa64 1073 arm_timer(timer);
e73d84e3 1074 unlock_task_sighand(p, &flags);
708f430d 1075
c925077c
FW
1076 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1077 posix_cpu_timer_kick_nohz();
708f430d
RM
1078out:
1079 timer->it_overrun_last = timer->it_overrun;
1080 timer->it_overrun = -1;
1081 ++timer->it_requeue_pending;
1da177e4
LT
1082}
1083
f06febc9
FM
1084/**
1085 * task_cputime_expired - Compare two task_cputime entities.
1086 *
1087 * @sample: The task_cputime structure to be checked for expiration.
1088 * @expires: Expiration times, against which @sample will be checked.
1089 *
1090 * Checks @sample against @expires to see if any field of @sample has expired.
1091 * Returns true if any field of the former is greater than the corresponding
1092 * field of the latter if the latter field is set. Otherwise returns false.
1093 */
1094static inline int task_cputime_expired(const struct task_cputime *sample,
1095 const struct task_cputime *expires)
1096{
64861634 1097 if (expires->utime && sample->utime >= expires->utime)
f06febc9 1098 return 1;
64861634 1099 if (expires->stime && sample->utime + sample->stime >= expires->stime)
f06febc9
FM
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 * fastpath_timer_check - POSIX CPU timers fast path.
1109 *
1110 * @tsk: The task (thread) being checked.
f06febc9 1111 *
bb34d92f
FM
1112 * Check the task and thread group timers. If both are zero (there are no
1113 * timers set) return false. Otherwise snapshot the task and thread group
1114 * timers and compare them with the corresponding expiration times. Return
1115 * true if a timer has expired, else return false.
f06febc9 1116 */
bb34d92f 1117static inline int fastpath_timer_check(struct task_struct *tsk)
f06febc9 1118{
ad133ba3 1119 struct signal_struct *sig;
bb34d92f 1120
bb34d92f 1121 if (!task_cputime_zero(&tsk->cputime_expires)) {
7c177d99 1122 struct task_cputime task_sample;
bb34d92f 1123
7c177d99
JL
1124 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1125 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
bb34d92f
FM
1126 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1127 return 1;
1128 }
ad133ba3
ON
1129
1130 sig = tsk->signal;
1018016c
JL
1131 /* Check if cputimer is running. This is accessed without locking. */
1132 if (READ_ONCE(sig->cputimer.running)) {
bb34d92f
FM
1133 struct task_cputime group_sample;
1134
71107445 1135 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
8d1f431c 1136
bb34d92f
FM
1137 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1138 return 1;
1139 }
37bebc70 1140
f55db609 1141 return 0;
f06febc9
FM
1142}
1143
1da177e4
LT
1144/*
1145 * This is called from the timer interrupt handler. The irq handler has
1146 * already updated our counts. We need to check if any timers fire now.
1147 * Interrupts are disabled.
1148 */
1149void run_posix_cpu_timers(struct task_struct *tsk)
1150{
1151 LIST_HEAD(firing);
1152 struct k_itimer *timer, *next;
0bdd2ed4 1153 unsigned long flags;
1da177e4 1154
531f64fd 1155 WARN_ON_ONCE(!irqs_disabled());
1da177e4 1156
1da177e4 1157 /*
f06febc9 1158 * The fast path checks that there are no expired thread or thread
bb34d92f 1159 * group timers. If that's so, just return.
1da177e4 1160 */
bb34d92f 1161 if (!fastpath_timer_check(tsk))
f06febc9 1162 return;
5ce73a4a 1163
0bdd2ed4
ON
1164 if (!lock_task_sighand(tsk, &flags))
1165 return;
bb34d92f
FM
1166 /*
1167 * Here we take off tsk->signal->cpu_timers[N] and
1168 * tsk->cpu_timers[N] all the timers that are firing, and
1169 * put them on the firing list.
1170 */
1171 check_thread_timers(tsk, &firing);
29f87b79
SG
1172 /*
1173 * If there are any active process wide timers (POSIX 1.b, itimers,
1174 * RLIMIT_CPU) cputimer must be running.
1175 */
1018016c 1176 if (READ_ONCE(tsk->signal->cputimer.running))
29f87b79 1177 check_process_timers(tsk, &firing);
1da177e4 1178
bb34d92f
FM
1179 /*
1180 * We must release these locks before taking any timer's lock.
1181 * There is a potential race with timer deletion here, as the
1182 * siglock now protects our private firing list. We have set
1183 * the firing flag in each timer, so that a deletion attempt
1184 * that gets the timer lock before we do will give it up and
1185 * spin until we've taken care of that timer below.
1186 */
0bdd2ed4 1187 unlock_task_sighand(tsk, &flags);
1da177e4
LT
1188
1189 /*
1190 * Now that all the timers on our list have the firing flag,
25985edc 1191 * no one will touch their list entries but us. We'll take
1da177e4
LT
1192 * each timer's lock before clearing its firing flag, so no
1193 * timer call will interfere.
1194 */
1195 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
6e85c5ba
HS
1196 int cpu_firing;
1197
1da177e4
LT
1198 spin_lock(&timer->it_lock);
1199 list_del_init(&timer->it.cpu.entry);
6e85c5ba 1200 cpu_firing = timer->it.cpu.firing;
1da177e4
LT
1201 timer->it.cpu.firing = 0;
1202 /*
1203 * The firing flag is -1 if we collided with a reset
1204 * of the timer, which already reported this
1205 * almost-firing as an overrun. So don't generate an event.
1206 */
6e85c5ba 1207 if (likely(cpu_firing >= 0))
1da177e4 1208 cpu_timer_fire(timer);
1da177e4
LT
1209 spin_unlock(&timer->it_lock);
1210 }
1211}
1212
1213/*
f55db609 1214 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
f06febc9 1215 * The tsk->sighand->siglock must be held by the caller.
1da177e4
LT
1216 */
1217void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1218 cputime_t *newval, cputime_t *oldval)
1219{
55ccb616 1220 unsigned long long now;
1da177e4 1221
531f64fd 1222 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
4cd4c1b4 1223 cpu_timer_sample_group(clock_idx, tsk, &now);
1da177e4
LT
1224
1225 if (oldval) {
f55db609
SG
1226 /*
1227 * We are setting itimer. The *oldval is absolute and we update
1228 * it to be relative, *newval argument is relative and we update
1229 * it to be absolute.
1230 */
64861634 1231 if (*oldval) {
55ccb616 1232 if (*oldval <= now) {
1da177e4 1233 /* Just about to fire. */
a42548a1 1234 *oldval = cputime_one_jiffy;
1da177e4 1235 } else {
55ccb616 1236 *oldval -= now;
1da177e4
LT
1237 }
1238 }
1239
64861634 1240 if (!*newval)
a8572160 1241 goto out;
55ccb616 1242 *newval += now;
1da177e4
LT
1243 }
1244
1245 /*
f55db609
SG
1246 * Update expiration cache if we are the earliest timer, or eventually
1247 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1da177e4 1248 */
f55db609
SG
1249 switch (clock_idx) {
1250 case CPUCLOCK_PROF:
1251 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
f06febc9 1252 tsk->signal->cputime_expires.prof_exp = *newval;
f55db609
SG
1253 break;
1254 case CPUCLOCK_VIRT:
1255 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
f06febc9 1256 tsk->signal->cputime_expires.virt_exp = *newval;
f55db609 1257 break;
1da177e4 1258 }
a8572160
FW
1259out:
1260 posix_cpu_timer_kick_nohz();
1da177e4
LT
1261}
1262
e4b76555
TA
1263static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1264 struct timespec *rqtp, struct itimerspec *it)
1da177e4 1265{
1da177e4
LT
1266 struct k_itimer timer;
1267 int error;
1268
1da177e4
LT
1269 /*
1270 * Set up a temporary timer and then wait for it to go off.
1271 */
1272 memset(&timer, 0, sizeof timer);
1273 spin_lock_init(&timer.it_lock);
1274 timer.it_clock = which_clock;
1275 timer.it_overrun = -1;
1276 error = posix_cpu_timer_create(&timer);
1277 timer.it_process = current;
1278 if (!error) {
1da177e4 1279 static struct itimerspec zero_it;
e4b76555
TA
1280
1281 memset(it, 0, sizeof *it);
1282 it->it_value = *rqtp;
1da177e4
LT
1283
1284 spin_lock_irq(&timer.it_lock);
e4b76555 1285 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1da177e4
LT
1286 if (error) {
1287 spin_unlock_irq(&timer.it_lock);
1288 return error;
1289 }
1290
1291 while (!signal_pending(current)) {
55ccb616 1292 if (timer.it.cpu.expires == 0) {
1da177e4 1293 /*
e6c42c29
SG
1294 * Our timer fired and was reset, below
1295 * deletion can not fail.
1da177e4 1296 */
e6c42c29 1297 posix_cpu_timer_del(&timer);
1da177e4
LT
1298 spin_unlock_irq(&timer.it_lock);
1299 return 0;
1300 }
1301
1302 /*
1303 * Block until cpu_timer_fire (or a signal) wakes us.
1304 */
1305 __set_current_state(TASK_INTERRUPTIBLE);
1306 spin_unlock_irq(&timer.it_lock);
1307 schedule();
1308 spin_lock_irq(&timer.it_lock);
1309 }
1310
1311 /*
1312 * We were interrupted by a signal.
1313 */
1314 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
e6c42c29
SG
1315 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1316 if (!error) {
1317 /*
1318 * Timer is now unarmed, deletion can not fail.
1319 */
1320 posix_cpu_timer_del(&timer);
1321 }
1da177e4
LT
1322 spin_unlock_irq(&timer.it_lock);
1323
e6c42c29
SG
1324 while (error == TIMER_RETRY) {
1325 /*
1326 * We need to handle case when timer was or is in the
1327 * middle of firing. In other cases we already freed
1328 * resources.
1329 */
1330 spin_lock_irq(&timer.it_lock);
1331 error = posix_cpu_timer_del(&timer);
1332 spin_unlock_irq(&timer.it_lock);
1333 }
1334
e4b76555 1335 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1da177e4
LT
1336 /*
1337 * It actually did fire already.
1338 */
1339 return 0;
1340 }
1341
e4b76555
TA
1342 error = -ERESTART_RESTARTBLOCK;
1343 }
1344
1345 return error;
1346}
1347
bc2c8ea4
TG
1348static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1349
1350static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1351 struct timespec *rqtp, struct timespec __user *rmtp)
e4b76555 1352{
f56141e3 1353 struct restart_block *restart_block = &current->restart_block;
e4b76555
TA
1354 struct itimerspec it;
1355 int error;
1356
1357 /*
1358 * Diagnose required errors first.
1359 */
1360 if (CPUCLOCK_PERTHREAD(which_clock) &&
1361 (CPUCLOCK_PID(which_clock) == 0 ||
1362 CPUCLOCK_PID(which_clock) == current->pid))
1363 return -EINVAL;
1364
1365 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1366
1367 if (error == -ERESTART_RESTARTBLOCK) {
1368
3751f9f2 1369 if (flags & TIMER_ABSTIME)
e4b76555 1370 return -ERESTARTNOHAND;
1da177e4 1371 /*
3751f9f2
TG
1372 * Report back to the user the time still remaining.
1373 */
1374 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1da177e4
LT
1375 return -EFAULT;
1376
1711ef38 1377 restart_block->fn = posix_cpu_nsleep_restart;
ab8177bc 1378 restart_block->nanosleep.clockid = which_clock;
3751f9f2
TG
1379 restart_block->nanosleep.rmtp = rmtp;
1380 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1da177e4 1381 }
1da177e4
LT
1382 return error;
1383}
1384
bc2c8ea4 1385static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1da177e4 1386{
ab8177bc 1387 clockid_t which_clock = restart_block->nanosleep.clockid;
97735f25 1388 struct timespec t;
e4b76555
TA
1389 struct itimerspec it;
1390 int error;
97735f25 1391
3751f9f2 1392 t = ns_to_timespec(restart_block->nanosleep.expires);
97735f25 1393
e4b76555
TA
1394 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1395
1396 if (error == -ERESTART_RESTARTBLOCK) {
3751f9f2 1397 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
e4b76555 1398 /*
3751f9f2
TG
1399 * Report back to the user the time still remaining.
1400 */
1401 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
e4b76555
TA
1402 return -EFAULT;
1403
3751f9f2 1404 restart_block->nanosleep.expires = timespec_to_ns(&t);
e4b76555
TA
1405 }
1406 return error;
1407
1da177e4
LT
1408}
1409
1da177e4
LT
1410#define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1411#define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1412
a924b04d
TG
1413static int process_cpu_clock_getres(const clockid_t which_clock,
1414 struct timespec *tp)
1da177e4
LT
1415{
1416 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1417}
a924b04d
TG
1418static int process_cpu_clock_get(const clockid_t which_clock,
1419 struct timespec *tp)
1da177e4
LT
1420{
1421 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1422}
1423static int process_cpu_timer_create(struct k_itimer *timer)
1424{
1425 timer->it_clock = PROCESS_CLOCK;
1426 return posix_cpu_timer_create(timer);
1427}
a924b04d 1428static int process_cpu_nsleep(const clockid_t which_clock, int flags,
97735f25
TG
1429 struct timespec *rqtp,
1430 struct timespec __user *rmtp)
1da177e4 1431{
97735f25 1432 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1da177e4 1433}
1711ef38
TA
1434static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1435{
1436 return -EINVAL;
1437}
a924b04d
TG
1438static int thread_cpu_clock_getres(const clockid_t which_clock,
1439 struct timespec *tp)
1da177e4
LT
1440{
1441 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1442}
a924b04d
TG
1443static int thread_cpu_clock_get(const clockid_t which_clock,
1444 struct timespec *tp)
1da177e4
LT
1445{
1446 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1447}
1448static int thread_cpu_timer_create(struct k_itimer *timer)
1449{
1450 timer->it_clock = THREAD_CLOCK;
1451 return posix_cpu_timer_create(timer);
1452}
1da177e4 1453
1976945e
TG
1454struct k_clock clock_posix_cpu = {
1455 .clock_getres = posix_cpu_clock_getres,
1456 .clock_set = posix_cpu_clock_set,
1457 .clock_get = posix_cpu_clock_get,
1458 .timer_create = posix_cpu_timer_create,
1459 .nsleep = posix_cpu_nsleep,
1460 .nsleep_restart = posix_cpu_nsleep_restart,
1461 .timer_set = posix_cpu_timer_set,
1462 .timer_del = posix_cpu_timer_del,
1463 .timer_get = posix_cpu_timer_get,
1464};
1465
1da177e4
LT
1466static __init int init_posix_cpu_timers(void)
1467{
1468 struct k_clock process = {
2fd1f040
TG
1469 .clock_getres = process_cpu_clock_getres,
1470 .clock_get = process_cpu_clock_get,
2fd1f040
TG
1471 .timer_create = process_cpu_timer_create,
1472 .nsleep = process_cpu_nsleep,
1473 .nsleep_restart = process_cpu_nsleep_restart,
1da177e4
LT
1474 };
1475 struct k_clock thread = {
2fd1f040
TG
1476 .clock_getres = thread_cpu_clock_getres,
1477 .clock_get = thread_cpu_clock_get,
2fd1f040 1478 .timer_create = thread_cpu_timer_create,
1da177e4 1479 };
8356b5f9 1480 struct timespec ts;
1da177e4 1481
52708737
TG
1482 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1483 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1da177e4 1484
a42548a1 1485 cputime_to_timespec(cputime_one_jiffy, &ts);
8356b5f9
SG
1486 onecputick = ts.tv_nsec;
1487 WARN_ON(ts.tv_sec != 0);
1488
1da177e4
LT
1489 return 0;
1490}
1491__initcall(init_posix_cpu_timers);