Root/
1 | /* |
2 | * Implement CPU time clocks for the POSIX clock interface. |
3 | */ |
4 | |
5 | #include <linux/sched.h> |
6 | #include <linux/posix-timers.h> |
7 | #include <linux/errno.h> |
8 | #include <linux/math64.h> |
9 | #include <asm/uaccess.h> |
10 | #include <linux/kernel_stat.h> |
11 | #include <trace/events/timer.h> |
12 | |
13 | /* |
14 | * Called after updating RLIMIT_CPU to set timer expiration if necessary. |
15 | */ |
16 | void update_rlimit_cpu(unsigned long rlim_new) |
17 | { |
18 | cputime_t cputime = secs_to_cputime(rlim_new); |
19 | struct signal_struct *const sig = current->signal; |
20 | |
21 | if (cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) || |
22 | cputime_gt(sig->it[CPUCLOCK_PROF].expires, cputime)) { |
23 | spin_lock_irq(¤t->sighand->siglock); |
24 | set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL); |
25 | spin_unlock_irq(¤t->sighand->siglock); |
26 | } |
27 | } |
28 | |
29 | static int check_clock(const clockid_t which_clock) |
30 | { |
31 | int error = 0; |
32 | struct task_struct *p; |
33 | const pid_t pid = CPUCLOCK_PID(which_clock); |
34 | |
35 | if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX) |
36 | return -EINVAL; |
37 | |
38 | if (pid == 0) |
39 | return 0; |
40 | |
41 | read_lock(&tasklist_lock); |
42 | p = find_task_by_vpid(pid); |
43 | if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ? |
44 | same_thread_group(p, current) : thread_group_leader(p))) { |
45 | error = -EINVAL; |
46 | } |
47 | read_unlock(&tasklist_lock); |
48 | |
49 | return error; |
50 | } |
51 | |
52 | static inline union cpu_time_count |
53 | timespec_to_sample(const clockid_t which_clock, const struct timespec *tp) |
54 | { |
55 | union cpu_time_count ret; |
56 | ret.sched = 0; /* high half always zero when .cpu used */ |
57 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
58 | ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec; |
59 | } else { |
60 | ret.cpu = timespec_to_cputime(tp); |
61 | } |
62 | return ret; |
63 | } |
64 | |
65 | static void sample_to_timespec(const clockid_t which_clock, |
66 | union cpu_time_count cpu, |
67 | struct timespec *tp) |
68 | { |
69 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) |
70 | *tp = ns_to_timespec(cpu.sched); |
71 | else |
72 | cputime_to_timespec(cpu.cpu, tp); |
73 | } |
74 | |
75 | static inline int cpu_time_before(const clockid_t which_clock, |
76 | union cpu_time_count now, |
77 | union cpu_time_count then) |
78 | { |
79 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
80 | return now.sched < then.sched; |
81 | } else { |
82 | return cputime_lt(now.cpu, then.cpu); |
83 | } |
84 | } |
85 | static inline void cpu_time_add(const clockid_t which_clock, |
86 | union cpu_time_count *acc, |
87 | union cpu_time_count val) |
88 | { |
89 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
90 | acc->sched += val.sched; |
91 | } else { |
92 | acc->cpu = cputime_add(acc->cpu, val.cpu); |
93 | } |
94 | } |
95 | static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock, |
96 | union cpu_time_count a, |
97 | union cpu_time_count b) |
98 | { |
99 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
100 | a.sched -= b.sched; |
101 | } else { |
102 | a.cpu = cputime_sub(a.cpu, b.cpu); |
103 | } |
104 | return a; |
105 | } |
106 | |
107 | /* |
108 | * Divide and limit the result to res >= 1 |
109 | * |
110 | * This is necessary to prevent signal delivery starvation, when the result of |
111 | * the division would be rounded down to 0. |
112 | */ |
113 | static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div) |
114 | { |
115 | cputime_t res = cputime_div(time, div); |
116 | |
117 | return max_t(cputime_t, res, 1); |
118 | } |
119 | |
120 | /* |
121 | * Update expiry time from increment, and increase overrun count, |
122 | * given the current clock sample. |
123 | */ |
124 | static void bump_cpu_timer(struct k_itimer *timer, |
125 | union cpu_time_count now) |
126 | { |
127 | int i; |
128 | |
129 | if (timer->it.cpu.incr.sched == 0) |
130 | return; |
131 | |
132 | if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) { |
133 | unsigned long long delta, incr; |
134 | |
135 | if (now.sched < timer->it.cpu.expires.sched) |
136 | return; |
137 | incr = timer->it.cpu.incr.sched; |
138 | delta = now.sched + incr - timer->it.cpu.expires.sched; |
139 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
140 | for (i = 0; incr < delta - incr; i++) |
141 | incr = incr << 1; |
142 | for (; i >= 0; incr >>= 1, i--) { |
143 | if (delta < incr) |
144 | continue; |
145 | timer->it.cpu.expires.sched += incr; |
146 | timer->it_overrun += 1 << i; |
147 | delta -= incr; |
148 | } |
149 | } else { |
150 | cputime_t delta, incr; |
151 | |
152 | if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu)) |
153 | return; |
154 | incr = timer->it.cpu.incr.cpu; |
155 | delta = cputime_sub(cputime_add(now.cpu, incr), |
156 | timer->it.cpu.expires.cpu); |
157 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
158 | for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++) |
159 | incr = cputime_add(incr, incr); |
160 | for (; i >= 0; incr = cputime_halve(incr), i--) { |
161 | if (cputime_lt(delta, incr)) |
162 | continue; |
163 | timer->it.cpu.expires.cpu = |
164 | cputime_add(timer->it.cpu.expires.cpu, incr); |
165 | timer->it_overrun += 1 << i; |
166 | delta = cputime_sub(delta, incr); |
167 | } |
168 | } |
169 | } |
170 | |
171 | static inline cputime_t prof_ticks(struct task_struct *p) |
172 | { |
173 | return cputime_add(p->utime, p->stime); |
174 | } |
175 | static inline cputime_t virt_ticks(struct task_struct *p) |
176 | { |
177 | return p->utime; |
178 | } |
179 | |
180 | int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp) |
181 | { |
182 | int error = check_clock(which_clock); |
183 | if (!error) { |
184 | tp->tv_sec = 0; |
185 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); |
186 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
187 | /* |
188 | * If sched_clock is using a cycle counter, we |
189 | * don't have any idea of its true resolution |
190 | * exported, but it is much more than 1s/HZ. |
191 | */ |
192 | tp->tv_nsec = 1; |
193 | } |
194 | } |
195 | return error; |
196 | } |
197 | |
198 | int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp) |
199 | { |
200 | /* |
201 | * You can never reset a CPU clock, but we check for other errors |
202 | * in the call before failing with EPERM. |
203 | */ |
204 | int error = check_clock(which_clock); |
205 | if (error == 0) { |
206 | error = -EPERM; |
207 | } |
208 | return error; |
209 | } |
210 | |
211 | |
212 | /* |
213 | * Sample a per-thread clock for the given task. |
214 | */ |
215 | static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p, |
216 | union cpu_time_count *cpu) |
217 | { |
218 | switch (CPUCLOCK_WHICH(which_clock)) { |
219 | default: |
220 | return -EINVAL; |
221 | case CPUCLOCK_PROF: |
222 | cpu->cpu = prof_ticks(p); |
223 | break; |
224 | case CPUCLOCK_VIRT: |
225 | cpu->cpu = virt_ticks(p); |
226 | break; |
227 | case CPUCLOCK_SCHED: |
228 | cpu->sched = task_sched_runtime(p); |
229 | break; |
230 | } |
231 | return 0; |
232 | } |
233 | |
234 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) |
235 | { |
236 | struct sighand_struct *sighand; |
237 | struct signal_struct *sig; |
238 | struct task_struct *t; |
239 | |
240 | *times = INIT_CPUTIME; |
241 | |
242 | rcu_read_lock(); |
243 | sighand = rcu_dereference(tsk->sighand); |
244 | if (!sighand) |
245 | goto out; |
246 | |
247 | sig = tsk->signal; |
248 | |
249 | t = tsk; |
250 | do { |
251 | times->utime = cputime_add(times->utime, t->utime); |
252 | times->stime = cputime_add(times->stime, t->stime); |
253 | times->sum_exec_runtime += t->se.sum_exec_runtime; |
254 | |
255 | t = next_thread(t); |
256 | } while (t != tsk); |
257 | |
258 | times->utime = cputime_add(times->utime, sig->utime); |
259 | times->stime = cputime_add(times->stime, sig->stime); |
260 | times->sum_exec_runtime += sig->sum_sched_runtime; |
261 | out: |
262 | rcu_read_unlock(); |
263 | } |
264 | |
265 | static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b) |
266 | { |
267 | if (cputime_gt(b->utime, a->utime)) |
268 | a->utime = b->utime; |
269 | |
270 | if (cputime_gt(b->stime, a->stime)) |
271 | a->stime = b->stime; |
272 | |
273 | if (b->sum_exec_runtime > a->sum_exec_runtime) |
274 | a->sum_exec_runtime = b->sum_exec_runtime; |
275 | } |
276 | |
277 | void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times) |
278 | { |
279 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; |
280 | struct task_cputime sum; |
281 | unsigned long flags; |
282 | |
283 | spin_lock_irqsave(&cputimer->lock, flags); |
284 | if (!cputimer->running) { |
285 | cputimer->running = 1; |
286 | /* |
287 | * The POSIX timer interface allows for absolute time expiry |
288 | * values through the TIMER_ABSTIME flag, therefore we have |
289 | * to synchronize the timer to the clock every time we start |
290 | * it. |
291 | */ |
292 | thread_group_cputime(tsk, &sum); |
293 | update_gt_cputime(&cputimer->cputime, &sum); |
294 | } |
295 | *times = cputimer->cputime; |
296 | spin_unlock_irqrestore(&cputimer->lock, flags); |
297 | } |
298 | |
299 | /* |
300 | * Sample a process (thread group) clock for the given group_leader task. |
301 | * Must be called with tasklist_lock held for reading. |
302 | */ |
303 | static int cpu_clock_sample_group(const clockid_t which_clock, |
304 | struct task_struct *p, |
305 | union cpu_time_count *cpu) |
306 | { |
307 | struct task_cputime cputime; |
308 | |
309 | switch (CPUCLOCK_WHICH(which_clock)) { |
310 | default: |
311 | return -EINVAL; |
312 | case CPUCLOCK_PROF: |
313 | thread_group_cputime(p, &cputime); |
314 | cpu->cpu = cputime_add(cputime.utime, cputime.stime); |
315 | break; |
316 | case CPUCLOCK_VIRT: |
317 | thread_group_cputime(p, &cputime); |
318 | cpu->cpu = cputime.utime; |
319 | break; |
320 | case CPUCLOCK_SCHED: |
321 | cpu->sched = thread_group_sched_runtime(p); |
322 | break; |
323 | } |
324 | return 0; |
325 | } |
326 | |
327 | |
328 | int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp) |
329 | { |
330 | const pid_t pid = CPUCLOCK_PID(which_clock); |
331 | int error = -EINVAL; |
332 | union cpu_time_count rtn; |
333 | |
334 | if (pid == 0) { |
335 | /* |
336 | * Special case constant value for our own clocks. |
337 | * We don't have to do any lookup to find ourselves. |
338 | */ |
339 | if (CPUCLOCK_PERTHREAD(which_clock)) { |
340 | /* |
341 | * Sampling just ourselves we can do with no locking. |
342 | */ |
343 | error = cpu_clock_sample(which_clock, |
344 | current, &rtn); |
345 | } else { |
346 | read_lock(&tasklist_lock); |
347 | error = cpu_clock_sample_group(which_clock, |
348 | current, &rtn); |
349 | read_unlock(&tasklist_lock); |
350 | } |
351 | } else { |
352 | /* |
353 | * Find the given PID, and validate that the caller |
354 | * should be able to see it. |
355 | */ |
356 | struct task_struct *p; |
357 | rcu_read_lock(); |
358 | p = find_task_by_vpid(pid); |
359 | if (p) { |
360 | if (CPUCLOCK_PERTHREAD(which_clock)) { |
361 | if (same_thread_group(p, current)) { |
362 | error = cpu_clock_sample(which_clock, |
363 | p, &rtn); |
364 | } |
365 | } else { |
366 | read_lock(&tasklist_lock); |
367 | if (thread_group_leader(p) && p->signal) { |
368 | error = |
369 | cpu_clock_sample_group(which_clock, |
370 | p, &rtn); |
371 | } |
372 | read_unlock(&tasklist_lock); |
373 | } |
374 | } |
375 | rcu_read_unlock(); |
376 | } |
377 | |
378 | if (error) |
379 | return error; |
380 | sample_to_timespec(which_clock, rtn, tp); |
381 | return 0; |
382 | } |
383 | |
384 | |
385 | /* |
386 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. |
387 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
388 | * new timer already all-zeros initialized. |
389 | */ |
390 | int posix_cpu_timer_create(struct k_itimer *new_timer) |
391 | { |
392 | int ret = 0; |
393 | const pid_t pid = CPUCLOCK_PID(new_timer->it_clock); |
394 | struct task_struct *p; |
395 | |
396 | if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX) |
397 | return -EINVAL; |
398 | |
399 | INIT_LIST_HEAD(&new_timer->it.cpu.entry); |
400 | |
401 | read_lock(&tasklist_lock); |
402 | if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) { |
403 | if (pid == 0) { |
404 | p = current; |
405 | } else { |
406 | p = find_task_by_vpid(pid); |
407 | if (p && !same_thread_group(p, current)) |
408 | p = NULL; |
409 | } |
410 | } else { |
411 | if (pid == 0) { |
412 | p = current->group_leader; |
413 | } else { |
414 | p = find_task_by_vpid(pid); |
415 | if (p && !thread_group_leader(p)) |
416 | p = NULL; |
417 | } |
418 | } |
419 | new_timer->it.cpu.task = p; |
420 | if (p) { |
421 | get_task_struct(p); |
422 | } else { |
423 | ret = -EINVAL; |
424 | } |
425 | read_unlock(&tasklist_lock); |
426 | |
427 | return ret; |
428 | } |
429 | |
430 | /* |
431 | * Clean up a CPU-clock timer that is about to be destroyed. |
432 | * This is called from timer deletion with the timer already locked. |
433 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
434 | * and try again. (This happens when the timer is in the middle of firing.) |
435 | */ |
436 | int posix_cpu_timer_del(struct k_itimer *timer) |
437 | { |
438 | struct task_struct *p = timer->it.cpu.task; |
439 | int ret = 0; |
440 | |
441 | if (likely(p != NULL)) { |
442 | read_lock(&tasklist_lock); |
443 | if (unlikely(p->signal == NULL)) { |
444 | /* |
445 | * We raced with the reaping of the task. |
446 | * The deletion should have cleared us off the list. |
447 | */ |
448 | BUG_ON(!list_empty(&timer->it.cpu.entry)); |
449 | } else { |
450 | spin_lock(&p->sighand->siglock); |
451 | if (timer->it.cpu.firing) |
452 | ret = TIMER_RETRY; |
453 | else |
454 | list_del(&timer->it.cpu.entry); |
455 | spin_unlock(&p->sighand->siglock); |
456 | } |
457 | read_unlock(&tasklist_lock); |
458 | |
459 | if (!ret) |
460 | put_task_struct(p); |
461 | } |
462 | |
463 | return ret; |
464 | } |
465 | |
466 | /* |
467 | * Clean out CPU timers still ticking when a thread exited. The task |
468 | * pointer is cleared, and the expiry time is replaced with the residual |
469 | * time for later timer_gettime calls to return. |
470 | * This must be called with the siglock held. |
471 | */ |
472 | static void cleanup_timers(struct list_head *head, |
473 | cputime_t utime, cputime_t stime, |
474 | unsigned long long sum_exec_runtime) |
475 | { |
476 | struct cpu_timer_list *timer, *next; |
477 | cputime_t ptime = cputime_add(utime, stime); |
478 | |
479 | list_for_each_entry_safe(timer, next, head, entry) { |
480 | list_del_init(&timer->entry); |
481 | if (cputime_lt(timer->expires.cpu, ptime)) { |
482 | timer->expires.cpu = cputime_zero; |
483 | } else { |
484 | timer->expires.cpu = cputime_sub(timer->expires.cpu, |
485 | ptime); |
486 | } |
487 | } |
488 | |
489 | ++head; |
490 | list_for_each_entry_safe(timer, next, head, entry) { |
491 | list_del_init(&timer->entry); |
492 | if (cputime_lt(timer->expires.cpu, utime)) { |
493 | timer->expires.cpu = cputime_zero; |
494 | } else { |
495 | timer->expires.cpu = cputime_sub(timer->expires.cpu, |
496 | utime); |
497 | } |
498 | } |
499 | |
500 | ++head; |
501 | list_for_each_entry_safe(timer, next, head, entry) { |
502 | list_del_init(&timer->entry); |
503 | if (timer->expires.sched < sum_exec_runtime) { |
504 | timer->expires.sched = 0; |
505 | } else { |
506 | timer->expires.sched -= sum_exec_runtime; |
507 | } |
508 | } |
509 | } |
510 | |
511 | /* |
512 | * These are both called with the siglock held, when the current thread |
513 | * is being reaped. When the final (leader) thread in the group is reaped, |
514 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. |
515 | */ |
516 | void posix_cpu_timers_exit(struct task_struct *tsk) |
517 | { |
518 | cleanup_timers(tsk->cpu_timers, |
519 | tsk->utime, tsk->stime, tsk->se.sum_exec_runtime); |
520 | |
521 | } |
522 | void posix_cpu_timers_exit_group(struct task_struct *tsk) |
523 | { |
524 | struct signal_struct *const sig = tsk->signal; |
525 | |
526 | cleanup_timers(tsk->signal->cpu_timers, |
527 | cputime_add(tsk->utime, sig->utime), |
528 | cputime_add(tsk->stime, sig->stime), |
529 | tsk->se.sum_exec_runtime + sig->sum_sched_runtime); |
530 | } |
531 | |
532 | static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now) |
533 | { |
534 | /* |
535 | * That's all for this thread or process. |
536 | * We leave our residual in expires to be reported. |
537 | */ |
538 | put_task_struct(timer->it.cpu.task); |
539 | timer->it.cpu.task = NULL; |
540 | timer->it.cpu.expires = cpu_time_sub(timer->it_clock, |
541 | timer->it.cpu.expires, |
542 | now); |
543 | } |
544 | |
545 | static inline int expires_gt(cputime_t expires, cputime_t new_exp) |
546 | { |
547 | return cputime_eq(expires, cputime_zero) || |
548 | cputime_gt(expires, new_exp); |
549 | } |
550 | |
551 | static inline int expires_le(cputime_t expires, cputime_t new_exp) |
552 | { |
553 | return !cputime_eq(expires, cputime_zero) && |
554 | cputime_le(expires, new_exp); |
555 | } |
556 | /* |
557 | * Insert the timer on the appropriate list before any timers that |
558 | * expire later. This must be called with the tasklist_lock held |
559 | * for reading, and interrupts disabled. |
560 | */ |
561 | static void arm_timer(struct k_itimer *timer, union cpu_time_count now) |
562 | { |
563 | struct task_struct *p = timer->it.cpu.task; |
564 | struct list_head *head, *listpos; |
565 | struct cpu_timer_list *const nt = &timer->it.cpu; |
566 | struct cpu_timer_list *next; |
567 | unsigned long i; |
568 | |
569 | head = (CPUCLOCK_PERTHREAD(timer->it_clock) ? |
570 | p->cpu_timers : p->signal->cpu_timers); |
571 | head += CPUCLOCK_WHICH(timer->it_clock); |
572 | |
573 | BUG_ON(!irqs_disabled()); |
574 | spin_lock(&p->sighand->siglock); |
575 | |
576 | listpos = head; |
577 | if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) { |
578 | list_for_each_entry(next, head, entry) { |
579 | if (next->expires.sched > nt->expires.sched) |
580 | break; |
581 | listpos = &next->entry; |
582 | } |
583 | } else { |
584 | list_for_each_entry(next, head, entry) { |
585 | if (cputime_gt(next->expires.cpu, nt->expires.cpu)) |
586 | break; |
587 | listpos = &next->entry; |
588 | } |
589 | } |
590 | list_add(&nt->entry, listpos); |
591 | |
592 | if (listpos == head) { |
593 | /* |
594 | * We are the new earliest-expiring timer. |
595 | * If we are a thread timer, there can always |
596 | * be a process timer telling us to stop earlier. |
597 | */ |
598 | |
599 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
600 | union cpu_time_count *exp = &nt->expires; |
601 | |
602 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
603 | default: |
604 | BUG(); |
605 | case CPUCLOCK_PROF: |
606 | if (expires_gt(p->cputime_expires.prof_exp, |
607 | exp->cpu)) |
608 | p->cputime_expires.prof_exp = exp->cpu; |
609 | break; |
610 | case CPUCLOCK_VIRT: |
611 | if (expires_gt(p->cputime_expires.virt_exp, |
612 | exp->cpu)) |
613 | p->cputime_expires.virt_exp = exp->cpu; |
614 | break; |
615 | case CPUCLOCK_SCHED: |
616 | if (p->cputime_expires.sched_exp == 0 || |
617 | p->cputime_expires.sched_exp > exp->sched) |
618 | p->cputime_expires.sched_exp = |
619 | exp->sched; |
620 | break; |
621 | } |
622 | } else { |
623 | struct signal_struct *const sig = p->signal; |
624 | union cpu_time_count *exp = &timer->it.cpu.expires; |
625 | |
626 | /* |
627 | * For a process timer, set the cached expiration time. |
628 | */ |
629 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
630 | default: |
631 | BUG(); |
632 | case CPUCLOCK_VIRT: |
633 | if (expires_le(sig->it[CPUCLOCK_VIRT].expires, |
634 | exp->cpu)) |
635 | break; |
636 | sig->cputime_expires.virt_exp = exp->cpu; |
637 | break; |
638 | case CPUCLOCK_PROF: |
639 | if (expires_le(sig->it[CPUCLOCK_PROF].expires, |
640 | exp->cpu)) |
641 | break; |
642 | i = sig->rlim[RLIMIT_CPU].rlim_cur; |
643 | if (i != RLIM_INFINITY && |
644 | i <= cputime_to_secs(exp->cpu)) |
645 | break; |
646 | sig->cputime_expires.prof_exp = exp->cpu; |
647 | break; |
648 | case CPUCLOCK_SCHED: |
649 | sig->cputime_expires.sched_exp = exp->sched; |
650 | break; |
651 | } |
652 | } |
653 | } |
654 | |
655 | spin_unlock(&p->sighand->siglock); |
656 | } |
657 | |
658 | /* |
659 | * The timer is locked, fire it and arrange for its reload. |
660 | */ |
661 | static void cpu_timer_fire(struct k_itimer *timer) |
662 | { |
663 | if (unlikely(timer->sigq == NULL)) { |
664 | /* |
665 | * This a special case for clock_nanosleep, |
666 | * not a normal timer from sys_timer_create. |
667 | */ |
668 | wake_up_process(timer->it_process); |
669 | timer->it.cpu.expires.sched = 0; |
670 | } else if (timer->it.cpu.incr.sched == 0) { |
671 | /* |
672 | * One-shot timer. Clear it as soon as it's fired. |
673 | */ |
674 | posix_timer_event(timer, 0); |
675 | timer->it.cpu.expires.sched = 0; |
676 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
677 | /* |
678 | * The signal did not get queued because the signal |
679 | * was ignored, so we won't get any callback to |
680 | * reload the timer. But we need to keep it |
681 | * ticking in case the signal is deliverable next time. |
682 | */ |
683 | posix_cpu_timer_schedule(timer); |
684 | } |
685 | } |
686 | |
687 | /* |
688 | * Sample a process (thread group) timer for the given group_leader task. |
689 | * Must be called with tasklist_lock held for reading. |
690 | */ |
691 | static int cpu_timer_sample_group(const clockid_t which_clock, |
692 | struct task_struct *p, |
693 | union cpu_time_count *cpu) |
694 | { |
695 | struct task_cputime cputime; |
696 | |
697 | thread_group_cputimer(p, &cputime); |
698 | switch (CPUCLOCK_WHICH(which_clock)) { |
699 | default: |
700 | return -EINVAL; |
701 | case CPUCLOCK_PROF: |
702 | cpu->cpu = cputime_add(cputime.utime, cputime.stime); |
703 | break; |
704 | case CPUCLOCK_VIRT: |
705 | cpu->cpu = cputime.utime; |
706 | break; |
707 | case CPUCLOCK_SCHED: |
708 | cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p); |
709 | break; |
710 | } |
711 | return 0; |
712 | } |
713 | |
714 | /* |
715 | * Guts of sys_timer_settime for CPU timers. |
716 | * This is called with the timer locked and interrupts disabled. |
717 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
718 | * and try again. (This happens when the timer is in the middle of firing.) |
719 | */ |
720 | int posix_cpu_timer_set(struct k_itimer *timer, int flags, |
721 | struct itimerspec *new, struct itimerspec *old) |
722 | { |
723 | struct task_struct *p = timer->it.cpu.task; |
724 | union cpu_time_count old_expires, new_expires, val; |
725 | int ret; |
726 | |
727 | if (unlikely(p == NULL)) { |
728 | /* |
729 | * Timer refers to a dead task's clock. |
730 | */ |
731 | return -ESRCH; |
732 | } |
733 | |
734 | new_expires = timespec_to_sample(timer->it_clock, &new->it_value); |
735 | |
736 | read_lock(&tasklist_lock); |
737 | /* |
738 | * We need the tasklist_lock to protect against reaping that |
739 | * clears p->signal. If p has just been reaped, we can no |
740 | * longer get any information about it at all. |
741 | */ |
742 | if (unlikely(p->signal == NULL)) { |
743 | read_unlock(&tasklist_lock); |
744 | put_task_struct(p); |
745 | timer->it.cpu.task = NULL; |
746 | return -ESRCH; |
747 | } |
748 | |
749 | /* |
750 | * Disarm any old timer after extracting its expiry time. |
751 | */ |
752 | BUG_ON(!irqs_disabled()); |
753 | |
754 | ret = 0; |
755 | spin_lock(&p->sighand->siglock); |
756 | old_expires = timer->it.cpu.expires; |
757 | if (unlikely(timer->it.cpu.firing)) { |
758 | timer->it.cpu.firing = -1; |
759 | ret = TIMER_RETRY; |
760 | } else |
761 | list_del_init(&timer->it.cpu.entry); |
762 | spin_unlock(&p->sighand->siglock); |
763 | |
764 | /* |
765 | * We need to sample the current value to convert the new |
766 | * value from to relative and absolute, and to convert the |
767 | * old value from absolute to relative. To set a process |
768 | * timer, we need a sample to balance the thread expiry |
769 | * times (in arm_timer). With an absolute time, we must |
770 | * check if it's already passed. In short, we need a sample. |
771 | */ |
772 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
773 | cpu_clock_sample(timer->it_clock, p, &val); |
774 | } else { |
775 | cpu_timer_sample_group(timer->it_clock, p, &val); |
776 | } |
777 | |
778 | if (old) { |
779 | if (old_expires.sched == 0) { |
780 | old->it_value.tv_sec = 0; |
781 | old->it_value.tv_nsec = 0; |
782 | } else { |
783 | /* |
784 | * Update the timer in case it has |
785 | * overrun already. If it has, |
786 | * we'll report it as having overrun |
787 | * and with the next reloaded timer |
788 | * already ticking, though we are |
789 | * swallowing that pending |
790 | * notification here to install the |
791 | * new setting. |
792 | */ |
793 | bump_cpu_timer(timer, val); |
794 | if (cpu_time_before(timer->it_clock, val, |
795 | timer->it.cpu.expires)) { |
796 | old_expires = cpu_time_sub( |
797 | timer->it_clock, |
798 | timer->it.cpu.expires, val); |
799 | sample_to_timespec(timer->it_clock, |
800 | old_expires, |
801 | &old->it_value); |
802 | } else { |
803 | old->it_value.tv_nsec = 1; |
804 | old->it_value.tv_sec = 0; |
805 | } |
806 | } |
807 | } |
808 | |
809 | if (unlikely(ret)) { |
810 | /* |
811 | * We are colliding with the timer actually firing. |
812 | * Punt after filling in the timer's old value, and |
813 | * disable this firing since we are already reporting |
814 | * it as an overrun (thanks to bump_cpu_timer above). |
815 | */ |
816 | read_unlock(&tasklist_lock); |
817 | goto out; |
818 | } |
819 | |
820 | if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) { |
821 | cpu_time_add(timer->it_clock, &new_expires, val); |
822 | } |
823 | |
824 | /* |
825 | * Install the new expiry time (or zero). |
826 | * For a timer with no notification action, we don't actually |
827 | * arm the timer (we'll just fake it for timer_gettime). |
828 | */ |
829 | timer->it.cpu.expires = new_expires; |
830 | if (new_expires.sched != 0 && |
831 | (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE && |
832 | cpu_time_before(timer->it_clock, val, new_expires)) { |
833 | arm_timer(timer, val); |
834 | } |
835 | |
836 | read_unlock(&tasklist_lock); |
837 | |
838 | /* |
839 | * Install the new reload setting, and |
840 | * set up the signal and overrun bookkeeping. |
841 | */ |
842 | timer->it.cpu.incr = timespec_to_sample(timer->it_clock, |
843 | &new->it_interval); |
844 | |
845 | /* |
846 | * This acts as a modification timestamp for the timer, |
847 | * so any automatic reload attempt will punt on seeing |
848 | * that we have reset the timer manually. |
849 | */ |
850 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & |
851 | ~REQUEUE_PENDING; |
852 | timer->it_overrun_last = 0; |
853 | timer->it_overrun = -1; |
854 | |
855 | if (new_expires.sched != 0 && |
856 | (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE && |
857 | !cpu_time_before(timer->it_clock, val, new_expires)) { |
858 | /* |
859 | * The designated time already passed, so we notify |
860 | * immediately, even if the thread never runs to |
861 | * accumulate more time on this clock. |
862 | */ |
863 | cpu_timer_fire(timer); |
864 | } |
865 | |
866 | ret = 0; |
867 | out: |
868 | if (old) { |
869 | sample_to_timespec(timer->it_clock, |
870 | timer->it.cpu.incr, &old->it_interval); |
871 | } |
872 | return ret; |
873 | } |
874 | |
875 | void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp) |
876 | { |
877 | union cpu_time_count now; |
878 | struct task_struct *p = timer->it.cpu.task; |
879 | int clear_dead; |
880 | |
881 | /* |
882 | * Easy part: convert the reload time. |
883 | */ |
884 | sample_to_timespec(timer->it_clock, |
885 | timer->it.cpu.incr, &itp->it_interval); |
886 | |
887 | if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all. */ |
888 | itp->it_value.tv_sec = itp->it_value.tv_nsec = 0; |
889 | return; |
890 | } |
891 | |
892 | if (unlikely(p == NULL)) { |
893 | /* |
894 | * This task already died and the timer will never fire. |
895 | * In this case, expires is actually the dead value. |
896 | */ |
897 | dead: |
898 | sample_to_timespec(timer->it_clock, timer->it.cpu.expires, |
899 | &itp->it_value); |
900 | return; |
901 | } |
902 | |
903 | /* |
904 | * Sample the clock to take the difference with the expiry time. |
905 | */ |
906 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
907 | cpu_clock_sample(timer->it_clock, p, &now); |
908 | clear_dead = p->exit_state; |
909 | } else { |
910 | read_lock(&tasklist_lock); |
911 | if (unlikely(p->signal == NULL)) { |
912 | /* |
913 | * The process has been reaped. |
914 | * We can't even collect a sample any more. |
915 | * Call the timer disarmed, nothing else to do. |
916 | */ |
917 | put_task_struct(p); |
918 | timer->it.cpu.task = NULL; |
919 | timer->it.cpu.expires.sched = 0; |
920 | read_unlock(&tasklist_lock); |
921 | goto dead; |
922 | } else { |
923 | cpu_timer_sample_group(timer->it_clock, p, &now); |
924 | clear_dead = (unlikely(p->exit_state) && |
925 | thread_group_empty(p)); |
926 | } |
927 | read_unlock(&tasklist_lock); |
928 | } |
929 | |
930 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
931 | if (timer->it.cpu.incr.sched == 0 && |
932 | cpu_time_before(timer->it_clock, |
933 | timer->it.cpu.expires, now)) { |
934 | /* |
935 | * Do-nothing timer expired and has no reload, |
936 | * so it's as if it was never set. |
937 | */ |
938 | timer->it.cpu.expires.sched = 0; |
939 | itp->it_value.tv_sec = itp->it_value.tv_nsec = 0; |
940 | return; |
941 | } |
942 | /* |
943 | * Account for any expirations and reloads that should |
944 | * have happened. |
945 | */ |
946 | bump_cpu_timer(timer, now); |
947 | } |
948 | |
949 | if (unlikely(clear_dead)) { |
950 | /* |
951 | * We've noticed that the thread is dead, but |
952 | * not yet reaped. Take this opportunity to |
953 | * drop our task ref. |
954 | */ |
955 | clear_dead_task(timer, now); |
956 | goto dead; |
957 | } |
958 | |
959 | if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) { |
960 | sample_to_timespec(timer->it_clock, |
961 | cpu_time_sub(timer->it_clock, |
962 | timer->it.cpu.expires, now), |
963 | &itp->it_value); |
964 | } else { |
965 | /* |
966 | * The timer should have expired already, but the firing |
967 | * hasn't taken place yet. Say it's just about to expire. |
968 | */ |
969 | itp->it_value.tv_nsec = 1; |
970 | itp->it_value.tv_sec = 0; |
971 | } |
972 | } |
973 | |
974 | /* |
975 | * Check for any per-thread CPU timers that have fired and move them off |
976 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the |
977 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. |
978 | */ |
979 | static void check_thread_timers(struct task_struct *tsk, |
980 | struct list_head *firing) |
981 | { |
982 | int maxfire; |
983 | struct list_head *timers = tsk->cpu_timers; |
984 | struct signal_struct *const sig = tsk->signal; |
985 | unsigned long soft; |
986 | |
987 | maxfire = 20; |
988 | tsk->cputime_expires.prof_exp = cputime_zero; |
989 | while (!list_empty(timers)) { |
990 | struct cpu_timer_list *t = list_first_entry(timers, |
991 | struct cpu_timer_list, |
992 | entry); |
993 | if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) { |
994 | tsk->cputime_expires.prof_exp = t->expires.cpu; |
995 | break; |
996 | } |
997 | t->firing = 1; |
998 | list_move_tail(&t->entry, firing); |
999 | } |
1000 | |
1001 | ++timers; |
1002 | maxfire = 20; |
1003 | tsk->cputime_expires.virt_exp = cputime_zero; |
1004 | while (!list_empty(timers)) { |
1005 | struct cpu_timer_list *t = list_first_entry(timers, |
1006 | struct cpu_timer_list, |
1007 | entry); |
1008 | if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) { |
1009 | tsk->cputime_expires.virt_exp = t->expires.cpu; |
1010 | break; |
1011 | } |
1012 | t->firing = 1; |
1013 | list_move_tail(&t->entry, firing); |
1014 | } |
1015 | |
1016 | ++timers; |
1017 | maxfire = 20; |
1018 | tsk->cputime_expires.sched_exp = 0; |
1019 | while (!list_empty(timers)) { |
1020 | struct cpu_timer_list *t = list_first_entry(timers, |
1021 | struct cpu_timer_list, |
1022 | entry); |
1023 | if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) { |
1024 | tsk->cputime_expires.sched_exp = t->expires.sched; |
1025 | break; |
1026 | } |
1027 | t->firing = 1; |
1028 | list_move_tail(&t->entry, firing); |
1029 | } |
1030 | |
1031 | /* |
1032 | * Check for the special case thread timers. |
1033 | */ |
1034 | soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur); |
1035 | if (soft != RLIM_INFINITY) { |
1036 | unsigned long hard = |
1037 | ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max); |
1038 | |
1039 | if (hard != RLIM_INFINITY && |
1040 | tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) { |
1041 | /* |
1042 | * At the hard limit, we just die. |
1043 | * No need to calculate anything else now. |
1044 | */ |
1045 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
1046 | return; |
1047 | } |
1048 | if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) { |
1049 | /* |
1050 | * At the soft limit, send a SIGXCPU every second. |
1051 | */ |
1052 | if (soft < hard) { |
1053 | soft += USEC_PER_SEC; |
1054 | sig->rlim[RLIMIT_RTTIME].rlim_cur = soft; |
1055 | } |
1056 | printk(KERN_INFO |
1057 | "RT Watchdog Timeout: %s[%d]\n", |
1058 | tsk->comm, task_pid_nr(tsk)); |
1059 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
1060 | } |
1061 | } |
1062 | } |
1063 | |
1064 | static void stop_process_timers(struct signal_struct *sig) |
1065 | { |
1066 | struct thread_group_cputimer *cputimer = &sig->cputimer; |
1067 | unsigned long flags; |
1068 | |
1069 | if (!cputimer->running) |
1070 | return; |
1071 | |
1072 | spin_lock_irqsave(&cputimer->lock, flags); |
1073 | cputimer->running = 0; |
1074 | spin_unlock_irqrestore(&cputimer->lock, flags); |
1075 | |
1076 | sig->cputime_expires.prof_exp = cputime_zero; |
1077 | sig->cputime_expires.virt_exp = cputime_zero; |
1078 | sig->cputime_expires.sched_exp = 0; |
1079 | } |
1080 | |
1081 | static u32 onecputick; |
1082 | |
1083 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
1084 | cputime_t *expires, cputime_t cur_time, int signo) |
1085 | { |
1086 | if (cputime_eq(it->expires, cputime_zero)) |
1087 | return; |
1088 | |
1089 | if (cputime_ge(cur_time, it->expires)) { |
1090 | if (!cputime_eq(it->incr, cputime_zero)) { |
1091 | it->expires = cputime_add(it->expires, it->incr); |
1092 | it->error += it->incr_error; |
1093 | if (it->error >= onecputick) { |
1094 | it->expires = cputime_sub(it->expires, |
1095 | cputime_one_jiffy); |
1096 | it->error -= onecputick; |
1097 | } |
1098 | } else { |
1099 | it->expires = cputime_zero; |
1100 | } |
1101 | |
1102 | trace_itimer_expire(signo == SIGPROF ? |
1103 | ITIMER_PROF : ITIMER_VIRTUAL, |
1104 | tsk->signal->leader_pid, cur_time); |
1105 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
1106 | } |
1107 | |
1108 | if (!cputime_eq(it->expires, cputime_zero) && |
1109 | (cputime_eq(*expires, cputime_zero) || |
1110 | cputime_lt(it->expires, *expires))) { |
1111 | *expires = it->expires; |
1112 | } |
1113 | } |
1114 | |
1115 | /* |
1116 | * Check for any per-thread CPU timers that have fired and move them |
1117 | * off the tsk->*_timers list onto the firing list. Per-thread timers |
1118 | * have already been taken off. |
1119 | */ |
1120 | static void check_process_timers(struct task_struct *tsk, |
1121 | struct list_head *firing) |
1122 | { |
1123 | int maxfire; |
1124 | struct signal_struct *const sig = tsk->signal; |
1125 | cputime_t utime, ptime, virt_expires, prof_expires; |
1126 | unsigned long long sum_sched_runtime, sched_expires; |
1127 | struct list_head *timers = sig->cpu_timers; |
1128 | struct task_cputime cputime; |
1129 | unsigned long soft; |
1130 | |
1131 | /* |
1132 | * Don't sample the current process CPU clocks if there are no timers. |
1133 | */ |
1134 | if (list_empty(&timers[CPUCLOCK_PROF]) && |
1135 | cputime_eq(sig->it[CPUCLOCK_PROF].expires, cputime_zero) && |
1136 | sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY && |
1137 | list_empty(&timers[CPUCLOCK_VIRT]) && |
1138 | cputime_eq(sig->it[CPUCLOCK_VIRT].expires, cputime_zero) && |
1139 | list_empty(&timers[CPUCLOCK_SCHED])) { |
1140 | stop_process_timers(sig); |
1141 | return; |
1142 | } |
1143 | |
1144 | /* |
1145 | * Collect the current process totals. |
1146 | */ |
1147 | thread_group_cputimer(tsk, &cputime); |
1148 | utime = cputime.utime; |
1149 | ptime = cputime_add(utime, cputime.stime); |
1150 | sum_sched_runtime = cputime.sum_exec_runtime; |
1151 | maxfire = 20; |
1152 | prof_expires = cputime_zero; |
1153 | while (!list_empty(timers)) { |
1154 | struct cpu_timer_list *tl = list_first_entry(timers, |
1155 | struct cpu_timer_list, |
1156 | entry); |
1157 | if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) { |
1158 | prof_expires = tl->expires.cpu; |
1159 | break; |
1160 | } |
1161 | tl->firing = 1; |
1162 | list_move_tail(&tl->entry, firing); |
1163 | } |
1164 | |
1165 | ++timers; |
1166 | maxfire = 20; |
1167 | virt_expires = cputime_zero; |
1168 | while (!list_empty(timers)) { |
1169 | struct cpu_timer_list *tl = list_first_entry(timers, |
1170 | struct cpu_timer_list, |
1171 | entry); |
1172 | if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) { |
1173 | virt_expires = tl->expires.cpu; |
1174 | break; |
1175 | } |
1176 | tl->firing = 1; |
1177 | list_move_tail(&tl->entry, firing); |
1178 | } |
1179 | |
1180 | ++timers; |
1181 | maxfire = 20; |
1182 | sched_expires = 0; |
1183 | while (!list_empty(timers)) { |
1184 | struct cpu_timer_list *tl = list_first_entry(timers, |
1185 | struct cpu_timer_list, |
1186 | entry); |
1187 | if (!--maxfire || sum_sched_runtime < tl->expires.sched) { |
1188 | sched_expires = tl->expires.sched; |
1189 | break; |
1190 | } |
1191 | tl->firing = 1; |
1192 | list_move_tail(&tl->entry, firing); |
1193 | } |
1194 | |
1195 | /* |
1196 | * Check for the special case process timers. |
1197 | */ |
1198 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime, |
1199 | SIGPROF); |
1200 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime, |
1201 | SIGVTALRM); |
1202 | soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); |
1203 | if (soft != RLIM_INFINITY) { |
1204 | unsigned long psecs = cputime_to_secs(ptime); |
1205 | unsigned long hard = |
1206 | ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max); |
1207 | cputime_t x; |
1208 | if (psecs >= hard) { |
1209 | /* |
1210 | * At the hard limit, we just die. |
1211 | * No need to calculate anything else now. |
1212 | */ |
1213 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
1214 | return; |
1215 | } |
1216 | if (psecs >= soft) { |
1217 | /* |
1218 | * At the soft limit, send a SIGXCPU every second. |
1219 | */ |
1220 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
1221 | if (soft < hard) { |
1222 | soft++; |
1223 | sig->rlim[RLIMIT_CPU].rlim_cur = soft; |
1224 | } |
1225 | } |
1226 | x = secs_to_cputime(soft); |
1227 | if (cputime_eq(prof_expires, cputime_zero) || |
1228 | cputime_lt(x, prof_expires)) { |
1229 | prof_expires = x; |
1230 | } |
1231 | } |
1232 | |
1233 | if (!cputime_eq(prof_expires, cputime_zero) && |
1234 | (cputime_eq(sig->cputime_expires.prof_exp, cputime_zero) || |
1235 | cputime_gt(sig->cputime_expires.prof_exp, prof_expires))) |
1236 | sig->cputime_expires.prof_exp = prof_expires; |
1237 | if (!cputime_eq(virt_expires, cputime_zero) && |
1238 | (cputime_eq(sig->cputime_expires.virt_exp, cputime_zero) || |
1239 | cputime_gt(sig->cputime_expires.virt_exp, virt_expires))) |
1240 | sig->cputime_expires.virt_exp = virt_expires; |
1241 | if (sched_expires != 0 && |
1242 | (sig->cputime_expires.sched_exp == 0 || |
1243 | sig->cputime_expires.sched_exp > sched_expires)) |
1244 | sig->cputime_expires.sched_exp = sched_expires; |
1245 | } |
1246 | |
1247 | /* |
1248 | * This is called from the signal code (via do_schedule_next_timer) |
1249 | * when the last timer signal was delivered and we have to reload the timer. |
1250 | */ |
1251 | void posix_cpu_timer_schedule(struct k_itimer *timer) |
1252 | { |
1253 | struct task_struct *p = timer->it.cpu.task; |
1254 | union cpu_time_count now; |
1255 | |
1256 | if (unlikely(p == NULL)) |
1257 | /* |
1258 | * The task was cleaned up already, no future firings. |
1259 | */ |
1260 | goto out; |
1261 | |
1262 | /* |
1263 | * Fetch the current sample and update the timer's expiry time. |
1264 | */ |
1265 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
1266 | cpu_clock_sample(timer->it_clock, p, &now); |
1267 | bump_cpu_timer(timer, now); |
1268 | if (unlikely(p->exit_state)) { |
1269 | clear_dead_task(timer, now); |
1270 | goto out; |
1271 | } |
1272 | read_lock(&tasklist_lock); /* arm_timer needs it. */ |
1273 | } else { |
1274 | read_lock(&tasklist_lock); |
1275 | if (unlikely(p->signal == NULL)) { |
1276 | /* |
1277 | * The process has been reaped. |
1278 | * We can't even collect a sample any more. |
1279 | */ |
1280 | put_task_struct(p); |
1281 | timer->it.cpu.task = p = NULL; |
1282 | timer->it.cpu.expires.sched = 0; |
1283 | goto out_unlock; |
1284 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
1285 | /* |
1286 | * We've noticed that the thread is dead, but |
1287 | * not yet reaped. Take this opportunity to |
1288 | * drop our task ref. |
1289 | */ |
1290 | clear_dead_task(timer, now); |
1291 | goto out_unlock; |
1292 | } |
1293 | cpu_timer_sample_group(timer->it_clock, p, &now); |
1294 | bump_cpu_timer(timer, now); |
1295 | /* Leave the tasklist_lock locked for the call below. */ |
1296 | } |
1297 | |
1298 | /* |
1299 | * Now re-arm for the new expiry time. |
1300 | */ |
1301 | arm_timer(timer, now); |
1302 | |
1303 | out_unlock: |
1304 | read_unlock(&tasklist_lock); |
1305 | |
1306 | out: |
1307 | timer->it_overrun_last = timer->it_overrun; |
1308 | timer->it_overrun = -1; |
1309 | ++timer->it_requeue_pending; |
1310 | } |
1311 | |
1312 | /** |
1313 | * task_cputime_zero - Check a task_cputime struct for all zero fields. |
1314 | * |
1315 | * @cputime: The struct to compare. |
1316 | * |
1317 | * Checks @cputime to see if all fields are zero. Returns true if all fields |
1318 | * are zero, false if any field is nonzero. |
1319 | */ |
1320 | static inline int task_cputime_zero(const struct task_cputime *cputime) |
1321 | { |
1322 | if (cputime_eq(cputime->utime, cputime_zero) && |
1323 | cputime_eq(cputime->stime, cputime_zero) && |
1324 | cputime->sum_exec_runtime == 0) |
1325 | return 1; |
1326 | return 0; |
1327 | } |
1328 | |
1329 | /** |
1330 | * task_cputime_expired - Compare two task_cputime entities. |
1331 | * |
1332 | * @sample: The task_cputime structure to be checked for expiration. |
1333 | * @expires: Expiration times, against which @sample will be checked. |
1334 | * |
1335 | * Checks @sample against @expires to see if any field of @sample has expired. |
1336 | * Returns true if any field of the former is greater than the corresponding |
1337 | * field of the latter if the latter field is set. Otherwise returns false. |
1338 | */ |
1339 | static inline int task_cputime_expired(const struct task_cputime *sample, |
1340 | const struct task_cputime *expires) |
1341 | { |
1342 | if (!cputime_eq(expires->utime, cputime_zero) && |
1343 | cputime_ge(sample->utime, expires->utime)) |
1344 | return 1; |
1345 | if (!cputime_eq(expires->stime, cputime_zero) && |
1346 | cputime_ge(cputime_add(sample->utime, sample->stime), |
1347 | expires->stime)) |
1348 | return 1; |
1349 | if (expires->sum_exec_runtime != 0 && |
1350 | sample->sum_exec_runtime >= expires->sum_exec_runtime) |
1351 | return 1; |
1352 | return 0; |
1353 | } |
1354 | |
1355 | /** |
1356 | * fastpath_timer_check - POSIX CPU timers fast path. |
1357 | * |
1358 | * @tsk: The task (thread) being checked. |
1359 | * |
1360 | * Check the task and thread group timers. If both are zero (there are no |
1361 | * timers set) return false. Otherwise snapshot the task and thread group |
1362 | * timers and compare them with the corresponding expiration times. Return |
1363 | * true if a timer has expired, else return false. |
1364 | */ |
1365 | static inline int fastpath_timer_check(struct task_struct *tsk) |
1366 | { |
1367 | struct signal_struct *sig; |
1368 | |
1369 | /* tsk == current, ensure it is safe to use ->signal/sighand */ |
1370 | if (unlikely(tsk->exit_state)) |
1371 | return 0; |
1372 | |
1373 | if (!task_cputime_zero(&tsk->cputime_expires)) { |
1374 | struct task_cputime task_sample = { |
1375 | .utime = tsk->utime, |
1376 | .stime = tsk->stime, |
1377 | .sum_exec_runtime = tsk->se.sum_exec_runtime |
1378 | }; |
1379 | |
1380 | if (task_cputime_expired(&task_sample, &tsk->cputime_expires)) |
1381 | return 1; |
1382 | } |
1383 | |
1384 | sig = tsk->signal; |
1385 | if (!task_cputime_zero(&sig->cputime_expires)) { |
1386 | struct task_cputime group_sample; |
1387 | |
1388 | thread_group_cputimer(tsk, &group_sample); |
1389 | if (task_cputime_expired(&group_sample, &sig->cputime_expires)) |
1390 | return 1; |
1391 | } |
1392 | |
1393 | return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY; |
1394 | } |
1395 | |
1396 | /* |
1397 | * This is called from the timer interrupt handler. The irq handler has |
1398 | * already updated our counts. We need to check if any timers fire now. |
1399 | * Interrupts are disabled. |
1400 | */ |
1401 | void run_posix_cpu_timers(struct task_struct *tsk) |
1402 | { |
1403 | LIST_HEAD(firing); |
1404 | struct k_itimer *timer, *next; |
1405 | |
1406 | BUG_ON(!irqs_disabled()); |
1407 | |
1408 | /* |
1409 | * The fast path checks that there are no expired thread or thread |
1410 | * group timers. If that's so, just return. |
1411 | */ |
1412 | if (!fastpath_timer_check(tsk)) |
1413 | return; |
1414 | |
1415 | spin_lock(&tsk->sighand->siglock); |
1416 | /* |
1417 | * Here we take off tsk->signal->cpu_timers[N] and |
1418 | * tsk->cpu_timers[N] all the timers that are firing, and |
1419 | * put them on the firing list. |
1420 | */ |
1421 | check_thread_timers(tsk, &firing); |
1422 | check_process_timers(tsk, &firing); |
1423 | |
1424 | /* |
1425 | * We must release these locks before taking any timer's lock. |
1426 | * There is a potential race with timer deletion here, as the |
1427 | * siglock now protects our private firing list. We have set |
1428 | * the firing flag in each timer, so that a deletion attempt |
1429 | * that gets the timer lock before we do will give it up and |
1430 | * spin until we've taken care of that timer below. |
1431 | */ |
1432 | spin_unlock(&tsk->sighand->siglock); |
1433 | |
1434 | /* |
1435 | * Now that all the timers on our list have the firing flag, |
1436 | * noone will touch their list entries but us. We'll take |
1437 | * each timer's lock before clearing its firing flag, so no |
1438 | * timer call will interfere. |
1439 | */ |
1440 | list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) { |
1441 | int cpu_firing; |
1442 | |
1443 | spin_lock(&timer->it_lock); |
1444 | list_del_init(&timer->it.cpu.entry); |
1445 | cpu_firing = timer->it.cpu.firing; |
1446 | timer->it.cpu.firing = 0; |
1447 | /* |
1448 | * The firing flag is -1 if we collided with a reset |
1449 | * of the timer, which already reported this |
1450 | * almost-firing as an overrun. So don't generate an event. |
1451 | */ |
1452 | if (likely(cpu_firing >= 0)) |
1453 | cpu_timer_fire(timer); |
1454 | spin_unlock(&timer->it_lock); |
1455 | } |
1456 | } |
1457 | |
1458 | /* |
1459 | * Set one of the process-wide special case CPU timers. |
1460 | * The tsk->sighand->siglock must be held by the caller. |
1461 | * The *newval argument is relative and we update it to be absolute, *oldval |
1462 | * is absolute and we update it to be relative. |
1463 | */ |
1464 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx, |
1465 | cputime_t *newval, cputime_t *oldval) |
1466 | { |
1467 | union cpu_time_count now; |
1468 | struct list_head *head; |
1469 | |
1470 | BUG_ON(clock_idx == CPUCLOCK_SCHED); |
1471 | cpu_timer_sample_group(clock_idx, tsk, &now); |
1472 | |
1473 | if (oldval) { |
1474 | if (!cputime_eq(*oldval, cputime_zero)) { |
1475 | if (cputime_le(*oldval, now.cpu)) { |
1476 | /* Just about to fire. */ |
1477 | *oldval = cputime_one_jiffy; |
1478 | } else { |
1479 | *oldval = cputime_sub(*oldval, now.cpu); |
1480 | } |
1481 | } |
1482 | |
1483 | if (cputime_eq(*newval, cputime_zero)) |
1484 | return; |
1485 | *newval = cputime_add(*newval, now.cpu); |
1486 | |
1487 | /* |
1488 | * If the RLIMIT_CPU timer will expire before the |
1489 | * ITIMER_PROF timer, we have nothing else to do. |
1490 | */ |
1491 | if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur |
1492 | < cputime_to_secs(*newval)) |
1493 | return; |
1494 | } |
1495 | |
1496 | /* |
1497 | * Check whether there are any process timers already set to fire |
1498 | * before this one. If so, we don't have anything more to do. |
1499 | */ |
1500 | head = &tsk->signal->cpu_timers[clock_idx]; |
1501 | if (list_empty(head) || |
1502 | cputime_ge(list_first_entry(head, |
1503 | struct cpu_timer_list, entry)->expires.cpu, |
1504 | *newval)) { |
1505 | switch (clock_idx) { |
1506 | case CPUCLOCK_PROF: |
1507 | tsk->signal->cputime_expires.prof_exp = *newval; |
1508 | break; |
1509 | case CPUCLOCK_VIRT: |
1510 | tsk->signal->cputime_expires.virt_exp = *newval; |
1511 | break; |
1512 | } |
1513 | } |
1514 | } |
1515 | |
1516 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
1517 | struct timespec *rqtp, struct itimerspec *it) |
1518 | { |
1519 | struct k_itimer timer; |
1520 | int error; |
1521 | |
1522 | /* |
1523 | * Set up a temporary timer and then wait for it to go off. |
1524 | */ |
1525 | memset(&timer, 0, sizeof timer); |
1526 | spin_lock_init(&timer.it_lock); |
1527 | timer.it_clock = which_clock; |
1528 | timer.it_overrun = -1; |
1529 | error = posix_cpu_timer_create(&timer); |
1530 | timer.it_process = current; |
1531 | if (!error) { |
1532 | static struct itimerspec zero_it; |
1533 | |
1534 | memset(it, 0, sizeof *it); |
1535 | it->it_value = *rqtp; |
1536 | |
1537 | spin_lock_irq(&timer.it_lock); |
1538 | error = posix_cpu_timer_set(&timer, flags, it, NULL); |
1539 | if (error) { |
1540 | spin_unlock_irq(&timer.it_lock); |
1541 | return error; |
1542 | } |
1543 | |
1544 | while (!signal_pending(current)) { |
1545 | if (timer.it.cpu.expires.sched == 0) { |
1546 | /* |
1547 | * Our timer fired and was reset. |
1548 | */ |
1549 | spin_unlock_irq(&timer.it_lock); |
1550 | return 0; |
1551 | } |
1552 | |
1553 | /* |
1554 | * Block until cpu_timer_fire (or a signal) wakes us. |
1555 | */ |
1556 | __set_current_state(TASK_INTERRUPTIBLE); |
1557 | spin_unlock_irq(&timer.it_lock); |
1558 | schedule(); |
1559 | spin_lock_irq(&timer.it_lock); |
1560 | } |
1561 | |
1562 | /* |
1563 | * We were interrupted by a signal. |
1564 | */ |
1565 | sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp); |
1566 | posix_cpu_timer_set(&timer, 0, &zero_it, it); |
1567 | spin_unlock_irq(&timer.it_lock); |
1568 | |
1569 | if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) { |
1570 | /* |
1571 | * It actually did fire already. |
1572 | */ |
1573 | return 0; |
1574 | } |
1575 | |
1576 | error = -ERESTART_RESTARTBLOCK; |
1577 | } |
1578 | |
1579 | return error; |
1580 | } |
1581 | |
1582 | int posix_cpu_nsleep(const clockid_t which_clock, int flags, |
1583 | struct timespec *rqtp, struct timespec __user *rmtp) |
1584 | { |
1585 | struct restart_block *restart_block = |
1586 | ¤t_thread_info()->restart_block; |
1587 | struct itimerspec it; |
1588 | int error; |
1589 | |
1590 | /* |
1591 | * Diagnose required errors first. |
1592 | */ |
1593 | if (CPUCLOCK_PERTHREAD(which_clock) && |
1594 | (CPUCLOCK_PID(which_clock) == 0 || |
1595 | CPUCLOCK_PID(which_clock) == current->pid)) |
1596 | return -EINVAL; |
1597 | |
1598 | error = do_cpu_nanosleep(which_clock, flags, rqtp, &it); |
1599 | |
1600 | if (error == -ERESTART_RESTARTBLOCK) { |
1601 | |
1602 | if (flags & TIMER_ABSTIME) |
1603 | return -ERESTARTNOHAND; |
1604 | /* |
1605 | * Report back to the user the time still remaining. |
1606 | */ |
1607 | if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp)) |
1608 | return -EFAULT; |
1609 | |
1610 | restart_block->fn = posix_cpu_nsleep_restart; |
1611 | restart_block->arg0 = which_clock; |
1612 | restart_block->arg1 = (unsigned long) rmtp; |
1613 | restart_block->arg2 = rqtp->tv_sec; |
1614 | restart_block->arg3 = rqtp->tv_nsec; |
1615 | } |
1616 | return error; |
1617 | } |
1618 | |
1619 | long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1620 | { |
1621 | clockid_t which_clock = restart_block->arg0; |
1622 | struct timespec __user *rmtp; |
1623 | struct timespec t; |
1624 | struct itimerspec it; |
1625 | int error; |
1626 | |
1627 | rmtp = (struct timespec __user *) restart_block->arg1; |
1628 | t.tv_sec = restart_block->arg2; |
1629 | t.tv_nsec = restart_block->arg3; |
1630 | |
1631 | restart_block->fn = do_no_restart_syscall; |
1632 | error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it); |
1633 | |
1634 | if (error == -ERESTART_RESTARTBLOCK) { |
1635 | /* |
1636 | * Report back to the user the time still remaining. |
1637 | */ |
1638 | if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp)) |
1639 | return -EFAULT; |
1640 | |
1641 | restart_block->fn = posix_cpu_nsleep_restart; |
1642 | restart_block->arg0 = which_clock; |
1643 | restart_block->arg1 = (unsigned long) rmtp; |
1644 | restart_block->arg2 = t.tv_sec; |
1645 | restart_block->arg3 = t.tv_nsec; |
1646 | } |
1647 | return error; |
1648 | |
1649 | } |
1650 | |
1651 | |
1652 | #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED) |
1653 | #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED) |
1654 | |
1655 | static int process_cpu_clock_getres(const clockid_t which_clock, |
1656 | struct timespec *tp) |
1657 | { |
1658 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); |
1659 | } |
1660 | static int process_cpu_clock_get(const clockid_t which_clock, |
1661 | struct timespec *tp) |
1662 | { |
1663 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); |
1664 | } |
1665 | static int process_cpu_timer_create(struct k_itimer *timer) |
1666 | { |
1667 | timer->it_clock = PROCESS_CLOCK; |
1668 | return posix_cpu_timer_create(timer); |
1669 | } |
1670 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
1671 | struct timespec *rqtp, |
1672 | struct timespec __user *rmtp) |
1673 | { |
1674 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp); |
1675 | } |
1676 | static long process_cpu_nsleep_restart(struct restart_block *restart_block) |
1677 | { |
1678 | return -EINVAL; |
1679 | } |
1680 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
1681 | struct timespec *tp) |
1682 | { |
1683 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); |
1684 | } |
1685 | static int thread_cpu_clock_get(const clockid_t which_clock, |
1686 | struct timespec *tp) |
1687 | { |
1688 | return posix_cpu_clock_get(THREAD_CLOCK, tp); |
1689 | } |
1690 | static int thread_cpu_timer_create(struct k_itimer *timer) |
1691 | { |
1692 | timer->it_clock = THREAD_CLOCK; |
1693 | return posix_cpu_timer_create(timer); |
1694 | } |
1695 | static int thread_cpu_nsleep(const clockid_t which_clock, int flags, |
1696 | struct timespec *rqtp, struct timespec __user *rmtp) |
1697 | { |
1698 | return -EINVAL; |
1699 | } |
1700 | static long thread_cpu_nsleep_restart(struct restart_block *restart_block) |
1701 | { |
1702 | return -EINVAL; |
1703 | } |
1704 | |
1705 | static __init int init_posix_cpu_timers(void) |
1706 | { |
1707 | struct k_clock process = { |
1708 | .clock_getres = process_cpu_clock_getres, |
1709 | .clock_get = process_cpu_clock_get, |
1710 | .clock_set = do_posix_clock_nosettime, |
1711 | .timer_create = process_cpu_timer_create, |
1712 | .nsleep = process_cpu_nsleep, |
1713 | .nsleep_restart = process_cpu_nsleep_restart, |
1714 | }; |
1715 | struct k_clock thread = { |
1716 | .clock_getres = thread_cpu_clock_getres, |
1717 | .clock_get = thread_cpu_clock_get, |
1718 | .clock_set = do_posix_clock_nosettime, |
1719 | .timer_create = thread_cpu_timer_create, |
1720 | .nsleep = thread_cpu_nsleep, |
1721 | .nsleep_restart = thread_cpu_nsleep_restart, |
1722 | }; |
1723 | struct timespec ts; |
1724 | |
1725 | register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process); |
1726 | register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread); |
1727 | |
1728 | cputime_to_timespec(cputime_one_jiffy, &ts); |
1729 | onecputick = ts.tv_nsec; |
1730 | WARN_ON(ts.tv_sec != 0); |
1731 | |
1732 | return 0; |
1733 | } |
1734 | __initcall(init_posix_cpu_timers); |
1735 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9