]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - kernel/posix-cpu-timers.c
x86/speculation: Provide IBPB always command line options
[thirdparty/kernel/stable.git] / kernel / 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;
f795b11f 106 timer->it_overrun += 1LL << i;
55ccb616 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
4da94d49
PZ
199static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
200{
64861634 201 if (b->utime > a->utime)
4da94d49
PZ
202 a->utime = b->utime;
203
64861634 204 if (b->stime > a->stime)
4da94d49
PZ
205 a->stime = b->stime;
206
207 if (b->sum_exec_runtime > a->sum_exec_runtime)
208 a->sum_exec_runtime = b->sum_exec_runtime;
209}
210
211void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
212{
213 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
214 struct task_cputime sum;
215 unsigned long flags;
216
4da94d49 217 if (!cputimer->running) {
4da94d49
PZ
218 /*
219 * The POSIX timer interface allows for absolute time expiry
220 * values through the TIMER_ABSTIME flag, therefore we have
221 * to synchronize the timer to the clock every time we start
222 * it.
223 */
224 thread_group_cputime(tsk, &sum);
3cfef952 225 raw_spin_lock_irqsave(&cputimer->lock, flags);
bcd5cff7 226 cputimer->running = 1;
4da94d49 227 update_gt_cputime(&cputimer->cputime, &sum);
bcd5cff7 228 } else
3cfef952 229 raw_spin_lock_irqsave(&cputimer->lock, flags);
4da94d49 230 *times = cputimer->cputime;
ee30a7b2 231 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
4da94d49
PZ
232}
233
1da177e4
LT
234/*
235 * Sample a process (thread group) clock for the given group_leader task.
e73d84e3
FW
236 * Must be called with task sighand lock held for safe while_each_thread()
237 * traversal.
1da177e4 238 */
bb34d92f
FM
239static int cpu_clock_sample_group(const clockid_t which_clock,
240 struct task_struct *p,
55ccb616 241 unsigned long long *sample)
1da177e4 242{
f06febc9
FM
243 struct task_cputime cputime;
244
eccdaeaf 245 switch (CPUCLOCK_WHICH(which_clock)) {
1da177e4
LT
246 default:
247 return -EINVAL;
248 case CPUCLOCK_PROF:
c5f8d995 249 thread_group_cputime(p, &cputime);
55ccb616 250 *sample = cputime_to_expires(cputime.utime + cputime.stime);
1da177e4
LT
251 break;
252 case CPUCLOCK_VIRT:
c5f8d995 253 thread_group_cputime(p, &cputime);
55ccb616 254 *sample = cputime_to_expires(cputime.utime);
1da177e4
LT
255 break;
256 case CPUCLOCK_SCHED:
d670ec13 257 thread_group_cputime(p, &cputime);
55ccb616 258 *sample = cputime.sum_exec_runtime;
1da177e4
LT
259 break;
260 }
261 return 0;
262}
263
33ab0fec
FW
264static int posix_cpu_clock_get_task(struct task_struct *tsk,
265 const clockid_t which_clock,
266 struct timespec *tp)
267{
268 int err = -EINVAL;
269 unsigned long long rtn;
270
271 if (CPUCLOCK_PERTHREAD(which_clock)) {
272 if (same_thread_group(tsk, current))
273 err = cpu_clock_sample(which_clock, tsk, &rtn);
274 } else {
50875788
FW
275 unsigned long flags;
276 struct sighand_struct *sighand;
33ab0fec 277
50875788
FW
278 /*
279 * while_each_thread() is not yet entirely RCU safe,
280 * keep locking the group while sampling process
281 * clock for now.
282 */
283 sighand = lock_task_sighand(tsk, &flags);
284 if (!sighand)
285 return err;
286
287 if (tsk == current || thread_group_leader(tsk))
33ab0fec
FW
288 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
289
50875788 290 unlock_task_sighand(tsk, &flags);
33ab0fec
FW
291 }
292
293 if (!err)
294 sample_to_timespec(which_clock, rtn, tp);
295
296 return err;
297}
298
1da177e4 299
bc2c8ea4 300static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
1da177e4
LT
301{
302 const pid_t pid = CPUCLOCK_PID(which_clock);
33ab0fec 303 int err = -EINVAL;
1da177e4
LT
304
305 if (pid == 0) {
306 /*
307 * Special case constant value for our own clocks.
308 * We don't have to do any lookup to find ourselves.
309 */
33ab0fec 310 err = posix_cpu_clock_get_task(current, which_clock, tp);
1da177e4
LT
311 } else {
312 /*
313 * Find the given PID, and validate that the caller
314 * should be able to see it.
315 */
316 struct task_struct *p;
1f2ea083 317 rcu_read_lock();
8dc86af0 318 p = find_task_by_vpid(pid);
33ab0fec
FW
319 if (p)
320 err = posix_cpu_clock_get_task(p, which_clock, tp);
1f2ea083 321 rcu_read_unlock();
1da177e4
LT
322 }
323
33ab0fec 324 return err;
1da177e4
LT
325}
326
327
328/*
329 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
ba5ea951
SG
330 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
331 * new timer already all-zeros initialized.
1da177e4 332 */
bc2c8ea4 333static int posix_cpu_timer_create(struct k_itimer *new_timer)
1da177e4
LT
334{
335 int ret = 0;
336 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
337 struct task_struct *p;
338
339 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
340 return -EINVAL;
341
342 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
1da177e4 343
c0deae8c 344 rcu_read_lock();
1da177e4
LT
345 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
346 if (pid == 0) {
347 p = current;
348 } else {
8dc86af0 349 p = find_task_by_vpid(pid);
bac0abd6 350 if (p && !same_thread_group(p, current))
1da177e4
LT
351 p = NULL;
352 }
353 } else {
354 if (pid == 0) {
355 p = current->group_leader;
356 } else {
8dc86af0 357 p = find_task_by_vpid(pid);
c0deae8c 358 if (p && !has_group_leader_pid(p))
1da177e4
LT
359 p = NULL;
360 }
361 }
362 new_timer->it.cpu.task = p;
363 if (p) {
364 get_task_struct(p);
365 } else {
366 ret = -EINVAL;
367 }
c0deae8c 368 rcu_read_unlock();
1da177e4
LT
369
370 return ret;
371}
372
373/*
374 * Clean up a CPU-clock timer that is about to be destroyed.
375 * This is called from timer deletion with the timer already locked.
376 * If we return TIMER_RETRY, it's necessary to release the timer's lock
377 * and try again. (This happens when the timer is in the middle of firing.)
378 */
bc2c8ea4 379static int posix_cpu_timer_del(struct k_itimer *timer)
1da177e4 380{
108150ea 381 int ret = 0;
3d7a1427
FW
382 unsigned long flags;
383 struct sighand_struct *sighand;
384 struct task_struct *p = timer->it.cpu.task;
1da177e4 385
a3222f88 386 WARN_ON_ONCE(p == NULL);
108150ea 387
3d7a1427
FW
388 /*
389 * Protect against sighand release/switch in exit/exec and process/
390 * thread timer list entry concurrent read/writes.
391 */
392 sighand = lock_task_sighand(p, &flags);
393 if (unlikely(sighand == NULL)) {
a3222f88
FW
394 /*
395 * We raced with the reaping of the task.
396 * The deletion should have cleared us off the list.
397 */
531f64fd 398 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
a3222f88 399 } else {
a3222f88
FW
400 if (timer->it.cpu.firing)
401 ret = TIMER_RETRY;
402 else
403 list_del(&timer->it.cpu.entry);
3d7a1427
FW
404
405 unlock_task_sighand(p, &flags);
1da177e4 406 }
a3222f88
FW
407
408 if (!ret)
409 put_task_struct(p);
1da177e4 410
108150ea 411 return ret;
1da177e4
LT
412}
413
af82eb3c 414static void cleanup_timers_list(struct list_head *head)
1a7fa510
FW
415{
416 struct cpu_timer_list *timer, *next;
417
a0b2062b 418 list_for_each_entry_safe(timer, next, head, entry)
1a7fa510 419 list_del_init(&timer->entry);
1a7fa510
FW
420}
421
1da177e4
LT
422/*
423 * Clean out CPU timers still ticking when a thread exited. The task
424 * pointer is cleared, and the expiry time is replaced with the residual
425 * time for later timer_gettime calls to return.
426 * This must be called with the siglock held.
427 */
af82eb3c 428static void cleanup_timers(struct list_head *head)
1da177e4 429{
af82eb3c
FW
430 cleanup_timers_list(head);
431 cleanup_timers_list(++head);
432 cleanup_timers_list(++head);
1da177e4
LT
433}
434
435/*
436 * These are both called with the siglock held, when the current thread
437 * is being reaped. When the final (leader) thread in the group is reaped,
438 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
439 */
440void posix_cpu_timers_exit(struct task_struct *tsk)
441{
61337054
NK
442 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
443 sizeof(unsigned long long));
af82eb3c 444 cleanup_timers(tsk->cpu_timers);
1da177e4
LT
445
446}
447void posix_cpu_timers_exit_group(struct task_struct *tsk)
448{
af82eb3c 449 cleanup_timers(tsk->signal->cpu_timers);
1da177e4
LT
450}
451
d1e3b6d1
SG
452static inline int expires_gt(cputime_t expires, cputime_t new_exp)
453{
64861634 454 return expires == 0 || expires > new_exp;
d1e3b6d1
SG
455}
456
1da177e4
LT
457/*
458 * Insert the timer on the appropriate list before any timers that
e73d84e3 459 * expire later. This must be called with the sighand lock held.
1da177e4 460 */
5eb9aa64 461static void arm_timer(struct k_itimer *timer)
1da177e4
LT
462{
463 struct task_struct *p = timer->it.cpu.task;
464 struct list_head *head, *listpos;
5eb9aa64 465 struct task_cputime *cputime_expires;
1da177e4
LT
466 struct cpu_timer_list *const nt = &timer->it.cpu;
467 struct cpu_timer_list *next;
1da177e4 468
5eb9aa64
SG
469 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
470 head = p->cpu_timers;
471 cputime_expires = &p->cputime_expires;
472 } else {
473 head = p->signal->cpu_timers;
474 cputime_expires = &p->signal->cputime_expires;
475 }
1da177e4
LT
476 head += CPUCLOCK_WHICH(timer->it_clock);
477
1da177e4 478 listpos = head;
5eb9aa64 479 list_for_each_entry(next, head, entry) {
55ccb616 480 if (nt->expires < next->expires)
5eb9aa64
SG
481 break;
482 listpos = &next->entry;
1da177e4
LT
483 }
484 list_add(&nt->entry, listpos);
485
486 if (listpos == head) {
55ccb616 487 unsigned long long exp = nt->expires;
5eb9aa64 488
1da177e4 489 /*
5eb9aa64
SG
490 * We are the new earliest-expiring POSIX 1.b timer, hence
491 * need to update expiration cache. Take into account that
492 * for process timers we share expiration cache with itimers
493 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
1da177e4
LT
494 */
495
5eb9aa64
SG
496 switch (CPUCLOCK_WHICH(timer->it_clock)) {
497 case CPUCLOCK_PROF:
55ccb616
FW
498 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
499 cputime_expires->prof_exp = expires_to_cputime(exp);
5eb9aa64
SG
500 break;
501 case CPUCLOCK_VIRT:
55ccb616
FW
502 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
503 cputime_expires->virt_exp = expires_to_cputime(exp);
5eb9aa64
SG
504 break;
505 case CPUCLOCK_SCHED:
506 if (cputime_expires->sched_exp == 0 ||
55ccb616
FW
507 cputime_expires->sched_exp > exp)
508 cputime_expires->sched_exp = exp;
5eb9aa64 509 break;
1da177e4
LT
510 }
511 }
1da177e4
LT
512}
513
514/*
515 * The timer is locked, fire it and arrange for its reload.
516 */
517static void cpu_timer_fire(struct k_itimer *timer)
518{
1f169f84
SG
519 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
520 /*
521 * User don't want any signal.
522 */
55ccb616 523 timer->it.cpu.expires = 0;
1f169f84 524 } else if (unlikely(timer->sigq == NULL)) {
1da177e4
LT
525 /*
526 * This a special case for clock_nanosleep,
527 * not a normal timer from sys_timer_create.
528 */
529 wake_up_process(timer->it_process);
55ccb616
FW
530 timer->it.cpu.expires = 0;
531 } else if (timer->it.cpu.incr == 0) {
1da177e4
LT
532 /*
533 * One-shot timer. Clear it as soon as it's fired.
534 */
535 posix_timer_event(timer, 0);
55ccb616 536 timer->it.cpu.expires = 0;
1da177e4
LT
537 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
538 /*
539 * The signal did not get queued because the signal
540 * was ignored, so we won't get any callback to
541 * reload the timer. But we need to keep it
542 * ticking in case the signal is deliverable next time.
543 */
544 posix_cpu_timer_schedule(timer);
545 }
546}
547
3997ad31
PZ
548/*
549 * Sample a process (thread group) timer for the given group_leader task.
e73d84e3
FW
550 * Must be called with task sighand lock held for safe while_each_thread()
551 * traversal.
3997ad31
PZ
552 */
553static int cpu_timer_sample_group(const clockid_t which_clock,
554 struct task_struct *p,
55ccb616 555 unsigned long long *sample)
3997ad31
PZ
556{
557 struct task_cputime cputime;
558
559 thread_group_cputimer(p, &cputime);
560 switch (CPUCLOCK_WHICH(which_clock)) {
561 default:
562 return -EINVAL;
563 case CPUCLOCK_PROF:
55ccb616 564 *sample = cputime_to_expires(cputime.utime + cputime.stime);
3997ad31
PZ
565 break;
566 case CPUCLOCK_VIRT:
55ccb616 567 *sample = cputime_to_expires(cputime.utime);
3997ad31
PZ
568 break;
569 case CPUCLOCK_SCHED:
55ccb616 570 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
3997ad31
PZ
571 break;
572 }
573 return 0;
574}
575
a8572160
FW
576#ifdef CONFIG_NO_HZ_FULL
577static void nohz_kick_work_fn(struct work_struct *work)
578{
579 tick_nohz_full_kick_all();
580}
581
582static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
583
584/*
585 * We need the IPIs to be sent from sane process context.
586 * The posix cpu timers are always set with irqs disabled.
587 */
588static void posix_cpu_timer_kick_nohz(void)
589{
d4283c65
FW
590 if (context_tracking_is_enabled())
591 schedule_work(&nohz_kick_work);
a8572160 592}
555347f6
FW
593
594bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
595{
596 if (!task_cputime_zero(&tsk->cputime_expires))
6ac29178 597 return false;
555347f6
FW
598
599 if (tsk->signal->cputimer.running)
6ac29178 600 return false;
555347f6 601
6ac29178 602 return true;
555347f6 603}
a8572160
FW
604#else
605static inline void posix_cpu_timer_kick_nohz(void) { }
606#endif
607
1da177e4
LT
608/*
609 * Guts of sys_timer_settime for CPU timers.
610 * This is called with the timer locked and interrupts disabled.
611 * If we return TIMER_RETRY, it's necessary to release the timer's lock
612 * and try again. (This happens when the timer is in the middle of firing.)
613 */
e73d84e3 614static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
bc2c8ea4 615 struct itimerspec *new, struct itimerspec *old)
1da177e4 616{
e73d84e3
FW
617 unsigned long flags;
618 struct sighand_struct *sighand;
1da177e4 619 struct task_struct *p = timer->it.cpu.task;
55ccb616 620 unsigned long long old_expires, new_expires, old_incr, val;
1da177e4
LT
621 int ret;
622
a3222f88 623 WARN_ON_ONCE(p == NULL);
1da177e4
LT
624
625 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
626
1da177e4 627 /*
e73d84e3
FW
628 * Protect against sighand release/switch in exit/exec and p->cpu_timers
629 * and p->signal->cpu_timers read/write in arm_timer()
630 */
631 sighand = lock_task_sighand(p, &flags);
632 /*
633 * If p has just been reaped, we can no
1da177e4
LT
634 * longer get any information about it at all.
635 */
e73d84e3 636 if (unlikely(sighand == NULL)) {
1da177e4
LT
637 return -ESRCH;
638 }
639
640 /*
641 * Disarm any old timer after extracting its expiry time.
642 */
531f64fd 643 WARN_ON_ONCE(!irqs_disabled());
a69ac4a7
ON
644
645 ret = 0;
ae1a78ee 646 old_incr = timer->it.cpu.incr;
1da177e4 647 old_expires = timer->it.cpu.expires;
a69ac4a7
ON
648 if (unlikely(timer->it.cpu.firing)) {
649 timer->it.cpu.firing = -1;
650 ret = TIMER_RETRY;
651 } else
652 list_del_init(&timer->it.cpu.entry);
1da177e4
LT
653
654 /*
655 * We need to sample the current value to convert the new
656 * value from to relative and absolute, and to convert the
657 * old value from absolute to relative. To set a process
658 * timer, we need a sample to balance the thread expiry
659 * times (in arm_timer). With an absolute time, we must
660 * check if it's already passed. In short, we need a sample.
661 */
662 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
663 cpu_clock_sample(timer->it_clock, p, &val);
664 } else {
3997ad31 665 cpu_timer_sample_group(timer->it_clock, p, &val);
1da177e4
LT
666 }
667
668 if (old) {
55ccb616 669 if (old_expires == 0) {
1da177e4
LT
670 old->it_value.tv_sec = 0;
671 old->it_value.tv_nsec = 0;
672 } else {
673 /*
674 * Update the timer in case it has
675 * overrun already. If it has,
676 * we'll report it as having overrun
677 * and with the next reloaded timer
678 * already ticking, though we are
679 * swallowing that pending
680 * notification here to install the
681 * new setting.
682 */
683 bump_cpu_timer(timer, val);
55ccb616
FW
684 if (val < timer->it.cpu.expires) {
685 old_expires = timer->it.cpu.expires - val;
1da177e4
LT
686 sample_to_timespec(timer->it_clock,
687 old_expires,
688 &old->it_value);
689 } else {
690 old->it_value.tv_nsec = 1;
691 old->it_value.tv_sec = 0;
692 }
693 }
694 }
695
a69ac4a7 696 if (unlikely(ret)) {
1da177e4
LT
697 /*
698 * We are colliding with the timer actually firing.
699 * Punt after filling in the timer's old value, and
700 * disable this firing since we are already reporting
701 * it as an overrun (thanks to bump_cpu_timer above).
702 */
e73d84e3 703 unlock_task_sighand(p, &flags);
1da177e4
LT
704 goto out;
705 }
706
e73d84e3 707 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
55ccb616 708 new_expires += val;
1da177e4
LT
709 }
710
711 /*
712 * Install the new expiry time (or zero).
713 * For a timer with no notification action, we don't actually
714 * arm the timer (we'll just fake it for timer_gettime).
715 */
716 timer->it.cpu.expires = new_expires;
55ccb616 717 if (new_expires != 0 && val < new_expires) {
5eb9aa64 718 arm_timer(timer);
1da177e4
LT
719 }
720
e73d84e3 721 unlock_task_sighand(p, &flags);
1da177e4
LT
722 /*
723 * Install the new reload setting, and
724 * set up the signal and overrun bookkeeping.
725 */
726 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
727 &new->it_interval);
728
729 /*
730 * This acts as a modification timestamp for the timer,
731 * so any automatic reload attempt will punt on seeing
732 * that we have reset the timer manually.
733 */
734 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
735 ~REQUEUE_PENDING;
736 timer->it_overrun_last = 0;
737 timer->it_overrun = -1;
738
55ccb616 739 if (new_expires != 0 && !(val < new_expires)) {
1da177e4
LT
740 /*
741 * The designated time already passed, so we notify
742 * immediately, even if the thread never runs to
743 * accumulate more time on this clock.
744 */
745 cpu_timer_fire(timer);
746 }
747
748 ret = 0;
749 out:
750 if (old) {
751 sample_to_timespec(timer->it_clock,
ae1a78ee 752 old_incr, &old->it_interval);
1da177e4 753 }
a8572160
FW
754 if (!ret)
755 posix_cpu_timer_kick_nohz();
1da177e4
LT
756 return ret;
757}
758
bc2c8ea4 759static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
1da177e4 760{
55ccb616 761 unsigned long long now;
1da177e4 762 struct task_struct *p = timer->it.cpu.task;
1da177e4 763
a3222f88
FW
764 WARN_ON_ONCE(p == NULL);
765
1da177e4
LT
766 /*
767 * Easy part: convert the reload time.
768 */
769 sample_to_timespec(timer->it_clock,
770 timer->it.cpu.incr, &itp->it_interval);
771
55ccb616 772 if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */
1da177e4
LT
773 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
774 return;
775 }
776
1da177e4
LT
777 /*
778 * Sample the clock to take the difference with the expiry time.
779 */
780 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
781 cpu_clock_sample(timer->it_clock, p, &now);
1da177e4 782 } else {
e73d84e3
FW
783 struct sighand_struct *sighand;
784 unsigned long flags;
785
786 /*
787 * Protect against sighand release/switch in exit/exec and
788 * also make timer sampling safe if it ends up calling
789 * thread_group_cputime().
790 */
791 sighand = lock_task_sighand(p, &flags);
792 if (unlikely(sighand == NULL)) {
1da177e4
LT
793 /*
794 * The process has been reaped.
795 * We can't even collect a sample any more.
796 * Call the timer disarmed, nothing else to do.
797 */
55ccb616 798 timer->it.cpu.expires = 0;
a3222f88
FW
799 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
800 &itp->it_value);
b829502b 801 return;
1da177e4 802 } else {
3997ad31 803 cpu_timer_sample_group(timer->it_clock, p, &now);
e73d84e3 804 unlock_task_sighand(p, &flags);
1da177e4 805 }
1da177e4
LT
806 }
807
55ccb616 808 if (now < timer->it.cpu.expires) {
1da177e4 809 sample_to_timespec(timer->it_clock,
55ccb616 810 timer->it.cpu.expires - now,
1da177e4
LT
811 &itp->it_value);
812 } else {
813 /*
814 * The timer should have expired already, but the firing
815 * hasn't taken place yet. Say it's just about to expire.
816 */
817 itp->it_value.tv_nsec = 1;
818 itp->it_value.tv_sec = 0;
819 }
820}
821
2473f3e7
FW
822static unsigned long long
823check_timers_list(struct list_head *timers,
824 struct list_head *firing,
825 unsigned long long curr)
826{
827 int maxfire = 20;
828
829 while (!list_empty(timers)) {
830 struct cpu_timer_list *t;
831
832 t = list_first_entry(timers, struct cpu_timer_list, entry);
833
834 if (!--maxfire || curr < t->expires)
835 return t->expires;
836
837 t->firing = 1;
838 list_move_tail(&t->entry, firing);
839 }
840
841 return 0;
842}
843
1da177e4
LT
844/*
845 * Check for any per-thread CPU timers that have fired and move them off
846 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
847 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
848 */
849static void check_thread_timers(struct task_struct *tsk,
850 struct list_head *firing)
851{
852 struct list_head *timers = tsk->cpu_timers;
78f2c7db 853 struct signal_struct *const sig = tsk->signal;
2473f3e7
FW
854 struct task_cputime *tsk_expires = &tsk->cputime_expires;
855 unsigned long long expires;
d4bb5274 856 unsigned long soft;
1da177e4 857
2473f3e7
FW
858 expires = check_timers_list(timers, firing, prof_ticks(tsk));
859 tsk_expires->prof_exp = expires_to_cputime(expires);
1da177e4 860
2473f3e7
FW
861 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
862 tsk_expires->virt_exp = expires_to_cputime(expires);
1da177e4 863
2473f3e7
FW
864 tsk_expires->sched_exp = check_timers_list(++timers, firing,
865 tsk->se.sum_exec_runtime);
78f2c7db
PZ
866
867 /*
868 * Check for the special case thread timers.
869 */
78d7d407 870 soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
d4bb5274 871 if (soft != RLIM_INFINITY) {
78d7d407
JS
872 unsigned long hard =
873 ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
78f2c7db 874
5a52dd50
PZ
875 if (hard != RLIM_INFINITY &&
876 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
78f2c7db
PZ
877 /*
878 * At the hard limit, we just die.
879 * No need to calculate anything else now.
880 */
881 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
882 return;
883 }
d4bb5274 884 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
78f2c7db
PZ
885 /*
886 * At the soft limit, send a SIGXCPU every second.
887 */
d4bb5274
JS
888 if (soft < hard) {
889 soft += USEC_PER_SEC;
890 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
78f2c7db 891 }
81d50bb2
HS
892 printk(KERN_INFO
893 "RT Watchdog Timeout: %s[%d]\n",
894 tsk->comm, task_pid_nr(tsk));
78f2c7db
PZ
895 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
896 }
897 }
1da177e4
LT
898}
899
15365c10 900static void stop_process_timers(struct signal_struct *sig)
3fccfd67 901{
15365c10 902 struct thread_group_cputimer *cputimer = &sig->cputimer;
3fccfd67
PZ
903 unsigned long flags;
904
ee30a7b2 905 raw_spin_lock_irqsave(&cputimer->lock, flags);
3fccfd67 906 cputimer->running = 0;
ee30a7b2 907 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
3fccfd67
PZ
908}
909
8356b5f9
SG
910static u32 onecputick;
911
42c4ab41 912static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
55ccb616
FW
913 unsigned long long *expires,
914 unsigned long long cur_time, int signo)
42c4ab41 915{
64861634 916 if (!it->expires)
42c4ab41
SG
917 return;
918
64861634
MS
919 if (cur_time >= it->expires) {
920 if (it->incr) {
921 it->expires += it->incr;
8356b5f9
SG
922 it->error += it->incr_error;
923 if (it->error >= onecputick) {
64861634 924 it->expires -= cputime_one_jiffy;
8356b5f9
SG
925 it->error -= onecputick;
926 }
3f0a525e 927 } else {
64861634 928 it->expires = 0;
3f0a525e 929 }
42c4ab41 930
3f0a525e
XG
931 trace_itimer_expire(signo == SIGPROF ?
932 ITIMER_PROF : ITIMER_VIRTUAL,
933 tsk->signal->leader_pid, cur_time);
42c4ab41
SG
934 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
935 }
936
64861634 937 if (it->expires && (!*expires || it->expires < *expires)) {
42c4ab41
SG
938 *expires = it->expires;
939 }
940}
941
1da177e4
LT
942/*
943 * Check for any per-thread CPU timers that have fired and move them
944 * off the tsk->*_timers list onto the firing list. Per-thread timers
945 * have already been taken off.
946 */
947static void check_process_timers(struct task_struct *tsk,
948 struct list_head *firing)
949{
950 struct signal_struct *const sig = tsk->signal;
55ccb616 951 unsigned long long utime, ptime, virt_expires, prof_expires;
41b86e9c 952 unsigned long long sum_sched_runtime, sched_expires;
1da177e4 953 struct list_head *timers = sig->cpu_timers;
f06febc9 954 struct task_cputime cputime;
d4bb5274 955 unsigned long soft;
1da177e4 956
1da177e4
LT
957 /*
958 * Collect the current process totals.
959 */
4cd4c1b4 960 thread_group_cputimer(tsk, &cputime);
55ccb616
FW
961 utime = cputime_to_expires(cputime.utime);
962 ptime = utime + cputime_to_expires(cputime.stime);
f06febc9 963 sum_sched_runtime = cputime.sum_exec_runtime;
1da177e4 964
2473f3e7
FW
965 prof_expires = check_timers_list(timers, firing, ptime);
966 virt_expires = check_timers_list(++timers, firing, utime);
967 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
1da177e4
LT
968
969 /*
970 * Check for the special case process timers.
971 */
42c4ab41
SG
972 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
973 SIGPROF);
974 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
975 SIGVTALRM);
78d7d407 976 soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
d4bb5274 977 if (soft != RLIM_INFINITY) {
1da177e4 978 unsigned long psecs = cputime_to_secs(ptime);
78d7d407
JS
979 unsigned long hard =
980 ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
1da177e4 981 cputime_t x;
d4bb5274 982 if (psecs >= hard) {
1da177e4
LT
983 /*
984 * At the hard limit, we just die.
985 * No need to calculate anything else now.
986 */
987 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
988 return;
989 }
d4bb5274 990 if (psecs >= soft) {
1da177e4
LT
991 /*
992 * At the soft limit, send a SIGXCPU every second.
993 */
994 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
d4bb5274
JS
995 if (soft < hard) {
996 soft++;
997 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
1da177e4
LT
998 }
999 }
d4bb5274 1000 x = secs_to_cputime(soft);
64861634 1001 if (!prof_expires || x < prof_expires) {
1da177e4
LT
1002 prof_expires = x;
1003 }
1004 }
1005
55ccb616
FW
1006 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
1007 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
29f87b79
SG
1008 sig->cputime_expires.sched_exp = sched_expires;
1009 if (task_cputime_zero(&sig->cputime_expires))
1010 stop_process_timers(sig);
1da177e4
LT
1011}
1012
1013/*
1014 * This is called from the signal code (via do_schedule_next_timer)
1015 * when the last timer signal was delivered and we have to reload the timer.
1016 */
1017void posix_cpu_timer_schedule(struct k_itimer *timer)
1018{
e73d84e3
FW
1019 struct sighand_struct *sighand;
1020 unsigned long flags;
1da177e4 1021 struct task_struct *p = timer->it.cpu.task;
55ccb616 1022 unsigned long long now;
1da177e4 1023
a3222f88 1024 WARN_ON_ONCE(p == NULL);
1da177e4
LT
1025
1026 /*
1027 * Fetch the current sample and update the timer's expiry time.
1028 */
1029 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1030 cpu_clock_sample(timer->it_clock, p, &now);
1031 bump_cpu_timer(timer, now);
724a3713 1032 if (unlikely(p->exit_state))
708f430d 1033 goto out;
724a3713 1034
e73d84e3
FW
1035 /* Protect timer list r/w in arm_timer() */
1036 sighand = lock_task_sighand(p, &flags);
1037 if (!sighand)
1038 goto out;
1da177e4 1039 } else {
e73d84e3
FW
1040 /*
1041 * Protect arm_timer() and timer sampling in case of call to
1042 * thread_group_cputime().
1043 */
1044 sighand = lock_task_sighand(p, &flags);
1045 if (unlikely(sighand == NULL)) {
1da177e4
LT
1046 /*
1047 * The process has been reaped.
1048 * We can't even collect a sample any more.
1049 */
55ccb616 1050 timer->it.cpu.expires = 0;
c925077c 1051 goto out;
1da177e4 1052 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
e73d84e3 1053 unlock_task_sighand(p, &flags);
d430b917 1054 /* Optimizations: if the process is dying, no need to rearm */
c925077c 1055 goto out;
1da177e4 1056 }
3997ad31 1057 cpu_timer_sample_group(timer->it_clock, p, &now);
1da177e4 1058 bump_cpu_timer(timer, now);
e73d84e3 1059 /* Leave the sighand locked for the call below. */
1da177e4
LT
1060 }
1061
1062 /*
1063 * Now re-arm for the new expiry time.
1064 */
531f64fd 1065 WARN_ON_ONCE(!irqs_disabled());
5eb9aa64 1066 arm_timer(timer);
e73d84e3 1067 unlock_task_sighand(p, &flags);
708f430d 1068
c925077c
FW
1069 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1070 posix_cpu_timer_kick_nohz();
708f430d
RM
1071out:
1072 timer->it_overrun_last = timer->it_overrun;
1073 timer->it_overrun = -1;
1074 ++timer->it_requeue_pending;
1da177e4
LT
1075}
1076
f06febc9
FM
1077/**
1078 * task_cputime_expired - Compare two task_cputime entities.
1079 *
1080 * @sample: The task_cputime structure to be checked for expiration.
1081 * @expires: Expiration times, against which @sample will be checked.
1082 *
1083 * Checks @sample against @expires to see if any field of @sample has expired.
1084 * Returns true if any field of the former is greater than the corresponding
1085 * field of the latter if the latter field is set. Otherwise returns false.
1086 */
1087static inline int task_cputime_expired(const struct task_cputime *sample,
1088 const struct task_cputime *expires)
1089{
64861634 1090 if (expires->utime && sample->utime >= expires->utime)
f06febc9 1091 return 1;
64861634 1092 if (expires->stime && sample->utime + sample->stime >= expires->stime)
f06febc9
FM
1093 return 1;
1094 if (expires->sum_exec_runtime != 0 &&
1095 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1096 return 1;
1097 return 0;
1098}
1099
1100/**
1101 * fastpath_timer_check - POSIX CPU timers fast path.
1102 *
1103 * @tsk: The task (thread) being checked.
f06febc9 1104 *
bb34d92f
FM
1105 * Check the task and thread group timers. If both are zero (there are no
1106 * timers set) return false. Otherwise snapshot the task and thread group
1107 * timers and compare them with the corresponding expiration times. Return
1108 * true if a timer has expired, else return false.
f06febc9 1109 */
bb34d92f 1110static inline int fastpath_timer_check(struct task_struct *tsk)
f06febc9 1111{
ad133ba3 1112 struct signal_struct *sig;
6fac4829
FW
1113 cputime_t utime, stime;
1114
1115 task_cputime(tsk, &utime, &stime);
bb34d92f 1116
bb34d92f
FM
1117 if (!task_cputime_zero(&tsk->cputime_expires)) {
1118 struct task_cputime task_sample = {
6fac4829
FW
1119 .utime = utime,
1120 .stime = stime,
bb34d92f
FM
1121 .sum_exec_runtime = tsk->se.sum_exec_runtime
1122 };
1123
1124 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1125 return 1;
1126 }
ad133ba3
ON
1127
1128 sig = tsk->signal;
29f87b79 1129 if (sig->cputimer.running) {
bb34d92f
FM
1130 struct task_cputime group_sample;
1131
ee30a7b2 1132 raw_spin_lock(&sig->cputimer.lock);
8d1f431c 1133 group_sample = sig->cputimer.cputime;
ee30a7b2 1134 raw_spin_unlock(&sig->cputimer.lock);
8d1f431c 1135
bb34d92f
FM
1136 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1137 return 1;
1138 }
37bebc70 1139
f55db609 1140 return 0;
f06febc9
FM
1141}
1142
1da177e4
LT
1143/*
1144 * This is called from the timer interrupt handler. The irq handler has
1145 * already updated our counts. We need to check if any timers fire now.
1146 * Interrupts are disabled.
1147 */
1148void run_posix_cpu_timers(struct task_struct *tsk)
1149{
1150 LIST_HEAD(firing);
1151 struct k_itimer *timer, *next;
0bdd2ed4 1152 unsigned long flags;
1da177e4 1153
531f64fd 1154 WARN_ON_ONCE(!irqs_disabled());
1da177e4 1155
1da177e4 1156 /*
f06febc9 1157 * The fast path checks that there are no expired thread or thread
bb34d92f 1158 * group timers. If that's so, just return.
1da177e4 1159 */
bb34d92f 1160 if (!fastpath_timer_check(tsk))
f06febc9 1161 return;
5ce73a4a 1162
0bdd2ed4
ON
1163 if (!lock_task_sighand(tsk, &flags))
1164 return;
bb34d92f
FM
1165 /*
1166 * Here we take off tsk->signal->cpu_timers[N] and
1167 * tsk->cpu_timers[N] all the timers that are firing, and
1168 * put them on the firing list.
1169 */
1170 check_thread_timers(tsk, &firing);
29f87b79
SG
1171 /*
1172 * If there are any active process wide timers (POSIX 1.b, itimers,
1173 * RLIMIT_CPU) cputimer must be running.
1174 */
1175 if (tsk->signal->cputimer.running)
1176 check_process_timers(tsk, &firing);
1da177e4 1177
bb34d92f
FM
1178 /*
1179 * We must release these locks before taking any timer's lock.
1180 * There is a potential race with timer deletion here, as the
1181 * siglock now protects our private firing list. We have set
1182 * the firing flag in each timer, so that a deletion attempt
1183 * that gets the timer lock before we do will give it up and
1184 * spin until we've taken care of that timer below.
1185 */
0bdd2ed4 1186 unlock_task_sighand(tsk, &flags);
1da177e4
LT
1187
1188 /*
1189 * Now that all the timers on our list have the firing flag,
25985edc 1190 * no one will touch their list entries but us. We'll take
1da177e4
LT
1191 * each timer's lock before clearing its firing flag, so no
1192 * timer call will interfere.
1193 */
1194 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
6e85c5ba
HS
1195 int cpu_firing;
1196
1da177e4
LT
1197 spin_lock(&timer->it_lock);
1198 list_del_init(&timer->it.cpu.entry);
6e85c5ba 1199 cpu_firing = timer->it.cpu.firing;
1da177e4
LT
1200 timer->it.cpu.firing = 0;
1201 /*
1202 * The firing flag is -1 if we collided with a reset
1203 * of the timer, which already reported this
1204 * almost-firing as an overrun. So don't generate an event.
1205 */
6e85c5ba 1206 if (likely(cpu_firing >= 0))
1da177e4 1207 cpu_timer_fire(timer);
1da177e4
LT
1208 spin_unlock(&timer->it_lock);
1209 }
1210}
1211
1212/*
f55db609 1213 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
f06febc9 1214 * The tsk->sighand->siglock must be held by the caller.
1da177e4
LT
1215 */
1216void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1217 cputime_t *newval, cputime_t *oldval)
1218{
55ccb616 1219 unsigned long long now;
1da177e4 1220
531f64fd 1221 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
4cd4c1b4 1222 cpu_timer_sample_group(clock_idx, tsk, &now);
1da177e4
LT
1223
1224 if (oldval) {
f55db609
SG
1225 /*
1226 * We are setting itimer. The *oldval is absolute and we update
1227 * it to be relative, *newval argument is relative and we update
1228 * it to be absolute.
1229 */
64861634 1230 if (*oldval) {
55ccb616 1231 if (*oldval <= now) {
1da177e4 1232 /* Just about to fire. */
a42548a1 1233 *oldval = cputime_one_jiffy;
1da177e4 1234 } else {
55ccb616 1235 *oldval -= now;
1da177e4
LT
1236 }
1237 }
1238
64861634 1239 if (!*newval)
a8572160 1240 goto out;
55ccb616 1241 *newval += now;
1da177e4
LT
1242 }
1243
1244 /*
f55db609
SG
1245 * Update expiration cache if we are the earliest timer, or eventually
1246 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1da177e4 1247 */
f55db609
SG
1248 switch (clock_idx) {
1249 case CPUCLOCK_PROF:
1250 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
f06febc9 1251 tsk->signal->cputime_expires.prof_exp = *newval;
f55db609
SG
1252 break;
1253 case CPUCLOCK_VIRT:
1254 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
f06febc9 1255 tsk->signal->cputime_expires.virt_exp = *newval;
f55db609 1256 break;
1da177e4 1257 }
a8572160
FW
1258out:
1259 posix_cpu_timer_kick_nohz();
1da177e4
LT
1260}
1261
e4b76555
TA
1262static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1263 struct timespec *rqtp, struct itimerspec *it)
1da177e4 1264{
1da177e4
LT
1265 struct k_itimer timer;
1266 int error;
1267
1da177e4
LT
1268 /*
1269 * Set up a temporary timer and then wait for it to go off.
1270 */
1271 memset(&timer, 0, sizeof timer);
1272 spin_lock_init(&timer.it_lock);
1273 timer.it_clock = which_clock;
1274 timer.it_overrun = -1;
1275 error = posix_cpu_timer_create(&timer);
1276 timer.it_process = current;
1277 if (!error) {
1da177e4 1278 static struct itimerspec zero_it;
e4b76555
TA
1279
1280 memset(it, 0, sizeof *it);
1281 it->it_value = *rqtp;
1da177e4
LT
1282
1283 spin_lock_irq(&timer.it_lock);
e4b76555 1284 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1da177e4
LT
1285 if (error) {
1286 spin_unlock_irq(&timer.it_lock);
1287 return error;
1288 }
1289
1290 while (!signal_pending(current)) {
55ccb616 1291 if (timer.it.cpu.expires == 0) {
1da177e4 1292 /*
e6c42c29
SG
1293 * Our timer fired and was reset, below
1294 * deletion can not fail.
1da177e4 1295 */
e6c42c29 1296 posix_cpu_timer_del(&timer);
1da177e4
LT
1297 spin_unlock_irq(&timer.it_lock);
1298 return 0;
1299 }
1300
1301 /*
1302 * Block until cpu_timer_fire (or a signal) wakes us.
1303 */
1304 __set_current_state(TASK_INTERRUPTIBLE);
1305 spin_unlock_irq(&timer.it_lock);
1306 schedule();
1307 spin_lock_irq(&timer.it_lock);
1308 }
1309
1310 /*
1311 * We were interrupted by a signal.
1312 */
1313 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
e6c42c29
SG
1314 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1315 if (!error) {
1316 /*
1317 * Timer is now unarmed, deletion can not fail.
1318 */
1319 posix_cpu_timer_del(&timer);
1320 }
1da177e4
LT
1321 spin_unlock_irq(&timer.it_lock);
1322
e6c42c29
SG
1323 while (error == TIMER_RETRY) {
1324 /*
1325 * We need to handle case when timer was or is in the
1326 * middle of firing. In other cases we already freed
1327 * resources.
1328 */
1329 spin_lock_irq(&timer.it_lock);
1330 error = posix_cpu_timer_del(&timer);
1331 spin_unlock_irq(&timer.it_lock);
1332 }
1333
e4b76555 1334 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1da177e4
LT
1335 /*
1336 * It actually did fire already.
1337 */
1338 return 0;
1339 }
1340
e4b76555
TA
1341 error = -ERESTART_RESTARTBLOCK;
1342 }
1343
1344 return error;
1345}
1346
bc2c8ea4
TG
1347static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1348
1349static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1350 struct timespec *rqtp, struct timespec __user *rmtp)
e4b76555
TA
1351{
1352 struct restart_block *restart_block =
3751f9f2 1353 &current_thread_info()->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);