blob: 05bb7173850e065a0899821d8e41c922d3548c07 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Implement CPU time clocks for the POSIX clock interface.
3 */
4
5#include <linux/sched.h>
6#include <linux/posix-timers.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007#include <linux/errno.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -07008#include <linux/math64.h>
9#include <asm/uaccess.h>
Frank Mayharbb34d922008-09-12 09:54:39 -070010#include <linux/kernel_stat.h>
Xiao Guangrong3f0a5252009-08-10 10:52:30 +080011#include <trace/events/timer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012
Frank Mayharf06febc2008-09-12 09:54:39 -070013/*
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -080014 * Called after updating RLIMIT_CPU to run cpu timer and update
15 * tsk->signal->cputime_expires expiration cache if necessary. Needs
16 * siglock protection since other code may update expiration cache as
17 * well.
Frank Mayharf06febc2008-09-12 09:54:39 -070018 */
Jiri Slaby5ab46b32009-08-28 14:05:12 +020019void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
Frank Mayharf06febc2008-09-12 09:54:39 -070020{
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +020021 cputime_t cputime = secs_to_cputime(rlim_new);
Frank Mayharf06febc2008-09-12 09:54:39 -070022
Jiri Slaby5ab46b32009-08-28 14:05:12 +020023 spin_lock_irq(&task->sighand->siglock);
24 set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
25 spin_unlock_irq(&task->sighand->siglock);
Frank Mayharf06febc2008-09-12 09:54:39 -070026}
27
Thomas Gleixnera924b042006-01-09 20:52:27 -080028static int check_clock(const clockid_t which_clock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070029{
30 int error = 0;
31 struct task_struct *p;
32 const pid_t pid = CPUCLOCK_PID(which_clock);
33
34 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
35 return -EINVAL;
36
37 if (pid == 0)
38 return 0;
39
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +020040 rcu_read_lock();
Pavel Emelyanov8dc86af2008-02-08 04:21:52 -080041 p = find_task_by_vpid(pid);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -070042 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +020043 same_thread_group(p, current) : has_group_leader_pid(p))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 error = -EINVAL;
45 }
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +020046 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48 return error;
49}
50
51static inline union cpu_time_count
Thomas Gleixnera924b042006-01-09 20:52:27 -080052timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
54 union cpu_time_count ret;
55 ret.sched = 0; /* high half always zero when .cpu used */
56 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
Oleg Nesterovee500f22005-11-28 13:43:55 -080057 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 } else {
59 ret.cpu = timespec_to_cputime(tp);
60 }
61 return ret;
62}
63
Thomas Gleixnera924b042006-01-09 20:52:27 -080064static void sample_to_timespec(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 union cpu_time_count cpu,
66 struct timespec *tp)
67{
Roman Zippelf8bd2252008-05-01 04:34:31 -070068 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
69 *tp = ns_to_timespec(cpu.sched);
70 else
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 cputime_to_timespec(cpu.cpu, tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072}
73
Thomas Gleixnera924b042006-01-09 20:52:27 -080074static inline int cpu_time_before(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 union cpu_time_count now,
76 union cpu_time_count then)
77{
78 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
79 return now.sched < then.sched;
80 } else {
81 return cputime_lt(now.cpu, then.cpu);
82 }
83}
Thomas Gleixnera924b042006-01-09 20:52:27 -080084static inline void cpu_time_add(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 union cpu_time_count *acc,
86 union cpu_time_count val)
87{
88 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
89 acc->sched += val.sched;
90 } else {
91 acc->cpu = cputime_add(acc->cpu, val.cpu);
92 }
93}
Thomas Gleixnera924b042006-01-09 20:52:27 -080094static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 union cpu_time_count a,
96 union cpu_time_count b)
97{
98 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
99 a.sched -= b.sched;
100 } else {
101 a.cpu = cputime_sub(a.cpu, b.cpu);
102 }
103 return a;
104}
105
106/*
Thomas Gleixnerac08c262006-10-17 00:09:39 -0700107 * Divide and limit the result to res >= 1
108 *
109 * This is necessary to prevent signal delivery starvation, when the result of
110 * the division would be rounded down to 0.
111 */
112static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
113{
114 cputime_t res = cputime_div(time, div);
115
116 return max_t(cputime_t, res, 1);
117}
118
119/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 * Update expiry time from increment, and increase overrun count,
121 * given the current clock sample.
122 */
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400123static void bump_cpu_timer(struct k_itimer *timer,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 union cpu_time_count now)
125{
126 int i;
127
128 if (timer->it.cpu.incr.sched == 0)
129 return;
130
131 if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
132 unsigned long long delta, incr;
133
134 if (now.sched < timer->it.cpu.expires.sched)
135 return;
136 incr = timer->it.cpu.incr.sched;
137 delta = now.sched + incr - timer->it.cpu.expires.sched;
138 /* Don't use (incr*2 < delta), incr*2 might overflow. */
139 for (i = 0; incr < delta - incr; i++)
140 incr = incr << 1;
141 for (; i >= 0; incr >>= 1, i--) {
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400142 if (delta < incr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 continue;
144 timer->it.cpu.expires.sched += incr;
145 timer->it_overrun += 1 << i;
146 delta -= incr;
147 }
148 } else {
149 cputime_t delta, incr;
150
151 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
152 return;
153 incr = timer->it.cpu.incr.cpu;
154 delta = cputime_sub(cputime_add(now.cpu, incr),
155 timer->it.cpu.expires.cpu);
156 /* Don't use (incr*2 < delta), incr*2 might overflow. */
157 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
158 incr = cputime_add(incr, incr);
159 for (; i >= 0; incr = cputime_halve(incr), i--) {
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400160 if (cputime_lt(delta, incr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 continue;
162 timer->it.cpu.expires.cpu =
163 cputime_add(timer->it.cpu.expires.cpu, incr);
164 timer->it_overrun += 1 << i;
165 delta = cputime_sub(delta, incr);
166 }
167 }
168}
169
170static inline cputime_t prof_ticks(struct task_struct *p)
171{
172 return cputime_add(p->utime, p->stime);
173}
174static inline cputime_t virt_ticks(struct task_struct *p)
175{
176 return p->utime;
177}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Thomas Gleixnera924b042006-01-09 20:52:27 -0800179int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
181 int error = check_clock(which_clock);
182 if (!error) {
183 tp->tv_sec = 0;
184 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
185 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
186 /*
187 * If sched_clock is using a cycle counter, we
188 * don't have any idea of its true resolution
189 * exported, but it is much more than 1s/HZ.
190 */
191 tp->tv_nsec = 1;
192 }
193 }
194 return error;
195}
196
Thomas Gleixnera924b042006-01-09 20:52:27 -0800197int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
199 /*
200 * You can never reset a CPU clock, but we check for other errors
201 * in the call before failing with EPERM.
202 */
203 int error = check_clock(which_clock);
204 if (error == 0) {
205 error = -EPERM;
206 }
207 return error;
208}
209
210
211/*
212 * Sample a per-thread clock for the given task.
213 */
Thomas Gleixnera924b042006-01-09 20:52:27 -0800214static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 union cpu_time_count *cpu)
216{
217 switch (CPUCLOCK_WHICH(which_clock)) {
218 default:
219 return -EINVAL;
220 case CPUCLOCK_PROF:
221 cpu->cpu = prof_ticks(p);
222 break;
223 case CPUCLOCK_VIRT:
224 cpu->cpu = virt_ticks(p);
225 break;
226 case CPUCLOCK_SCHED:
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +0900227 cpu->sched = task_sched_runtime(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 break;
229 }
230 return 0;
231}
232
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100233void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
234{
Oleg Nesterovbfac7002010-06-11 01:09:56 +0200235 struct signal_struct *sig = tsk->signal;
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100236 struct task_struct *t;
237
Oleg Nesterovbfac7002010-06-11 01:09:56 +0200238 times->utime = sig->utime;
239 times->stime = sig->stime;
240 times->sum_exec_runtime = sig->sum_sched_runtime;
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100241
242 rcu_read_lock();
Oleg Nesterovbfac7002010-06-11 01:09:56 +0200243 /* make sure we can trust tsk->thread_group list */
244 if (!likely(pid_alive(tsk)))
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100245 goto out;
246
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100247 t = tsk;
248 do {
249 times->utime = cputime_add(times->utime, t->utime);
250 times->stime = cputime_add(times->stime, t->stime);
251 times->sum_exec_runtime += t->se.sum_exec_runtime;
Oleg Nesterovbfac7002010-06-11 01:09:56 +0200252 } while_each_thread(tsk, t);
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +0100253out:
254 rcu_read_unlock();
255}
256
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100257static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
258{
259 if (cputime_gt(b->utime, a->utime))
260 a->utime = b->utime;
261
262 if (cputime_gt(b->stime, a->stime))
263 a->stime = b->stime;
264
265 if (b->sum_exec_runtime > a->sum_exec_runtime)
266 a->sum_exec_runtime = b->sum_exec_runtime;
267}
268
269void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
270{
271 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
272 struct task_cputime sum;
273 unsigned long flags;
274
275 spin_lock_irqsave(&cputimer->lock, flags);
276 if (!cputimer->running) {
277 cputimer->running = 1;
278 /*
279 * The POSIX timer interface allows for absolute time expiry
280 * values through the TIMER_ABSTIME flag, therefore we have
281 * to synchronize the timer to the clock every time we start
282 * it.
283 */
284 thread_group_cputime(tsk, &sum);
285 update_gt_cputime(&cputimer->cputime, &sum);
286 }
287 *times = cputimer->cputime;
288 spin_unlock_irqrestore(&cputimer->lock, flags);
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291/*
292 * Sample a process (thread group) clock for the given group_leader task.
293 * Must be called with tasklist_lock held for reading.
294 */
Thomas Gleixnera924b042006-01-09 20:52:27 -0800295static int cpu_clock_sample_group(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 struct task_struct *p,
297 union cpu_time_count *cpu)
298{
Frank Mayharbb34d922008-09-12 09:54:39 -0700299 struct task_cputime cputime;
300
Petr Tesarikeccdaea2008-11-24 15:46:31 +0100301 switch (CPUCLOCK_WHICH(which_clock)) {
Frank Mayharbb34d922008-09-12 09:54:39 -0700302 default:
303 return -EINVAL;
304 case CPUCLOCK_PROF:
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +0900305 thread_group_cputime(p, &cputime);
Frank Mayharbb34d922008-09-12 09:54:39 -0700306 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
307 break;
308 case CPUCLOCK_VIRT:
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +0900309 thread_group_cputime(p, &cputime);
Frank Mayharbb34d922008-09-12 09:54:39 -0700310 cpu->cpu = cputime.utime;
311 break;
312 case CPUCLOCK_SCHED:
Hidetoshi Setoc5f8d992009-03-31 16:56:03 +0900313 cpu->sched = thread_group_sched_runtime(p);
Frank Mayharbb34d922008-09-12 09:54:39 -0700314 break;
315 }
316 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317}
318
319
Thomas Gleixnera924b042006-01-09 20:52:27 -0800320int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 const pid_t pid = CPUCLOCK_PID(which_clock);
323 int error = -EINVAL;
324 union cpu_time_count rtn;
325
326 if (pid == 0) {
327 /*
328 * Special case constant value for our own clocks.
329 * We don't have to do any lookup to find ourselves.
330 */
331 if (CPUCLOCK_PERTHREAD(which_clock)) {
332 /*
333 * Sampling just ourselves we can do with no locking.
334 */
335 error = cpu_clock_sample(which_clock,
336 current, &rtn);
337 } else {
338 read_lock(&tasklist_lock);
339 error = cpu_clock_sample_group(which_clock,
340 current, &rtn);
341 read_unlock(&tasklist_lock);
342 }
343 } else {
344 /*
345 * Find the given PID, and validate that the caller
346 * should be able to see it.
347 */
348 struct task_struct *p;
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800349 rcu_read_lock();
Pavel Emelyanov8dc86af2008-02-08 04:21:52 -0800350 p = find_task_by_vpid(pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (p) {
352 if (CPUCLOCK_PERTHREAD(which_clock)) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700353 if (same_thread_group(p, current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 error = cpu_clock_sample(which_clock,
355 p, &rtn);
356 }
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800357 } else {
358 read_lock(&tasklist_lock);
Oleg Nesterovd30fda32010-05-26 14:43:13 -0700359 if (thread_group_leader(p) && p->sighand) {
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800360 error =
361 cpu_clock_sample_group(which_clock,
362 p, &rtn);
363 }
364 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
366 }
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800367 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 }
369
370 if (error)
371 return error;
372 sample_to_timespec(which_clock, rtn, tp);
373 return 0;
374}
375
376
377/*
378 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
Stanislaw Gruszkaba5ea952009-11-17 14:14:13 -0800379 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
380 * new timer already all-zeros initialized.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
382int posix_cpu_timer_create(struct k_itimer *new_timer)
383{
384 int ret = 0;
385 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
386 struct task_struct *p;
387
388 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
389 return -EINVAL;
390
391 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +0200393 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
395 if (pid == 0) {
396 p = current;
397 } else {
Pavel Emelyanov8dc86af2008-02-08 04:21:52 -0800398 p = find_task_by_vpid(pid);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700399 if (p && !same_thread_group(p, current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 p = NULL;
401 }
402 } else {
403 if (pid == 0) {
404 p = current->group_leader;
405 } else {
Pavel Emelyanov8dc86af2008-02-08 04:21:52 -0800406 p = find_task_by_vpid(pid);
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +0200407 if (p && !has_group_leader_pid(p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 p = NULL;
409 }
410 }
411 new_timer->it.cpu.task = p;
412 if (p) {
413 get_task_struct(p);
414 } else {
415 ret = -EINVAL;
416 }
Sergey Senozhatskyc0deae82010-11-03 18:52:56 +0200417 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 return ret;
420}
421
422/*
423 * Clean up a CPU-clock timer that is about to be destroyed.
424 * This is called from timer deletion with the timer already locked.
425 * If we return TIMER_RETRY, it's necessary to release the timer's lock
426 * and try again. (This happens when the timer is in the middle of firing.)
427 */
428int posix_cpu_timer_del(struct k_itimer *timer)
429{
430 struct task_struct *p = timer->it.cpu.task;
Oleg Nesterov108150e2005-10-23 20:25:39 +0400431 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Oleg Nesterov108150e2005-10-23 20:25:39 +0400433 if (likely(p != NULL)) {
Linus Torvalds9465bee2005-10-21 15:36:00 -0700434 read_lock(&tasklist_lock);
Oleg Nesterovd30fda32010-05-26 14:43:13 -0700435 if (unlikely(p->sighand == NULL)) {
Linus Torvalds9465bee2005-10-21 15:36:00 -0700436 /*
437 * We raced with the reaping of the task.
438 * The deletion should have cleared us off the list.
439 */
440 BUG_ON(!list_empty(&timer->it.cpu.entry));
441 } else {
Linus Torvalds9465bee2005-10-21 15:36:00 -0700442 spin_lock(&p->sighand->siglock);
Oleg Nesterov108150e2005-10-23 20:25:39 +0400443 if (timer->it.cpu.firing)
444 ret = TIMER_RETRY;
445 else
446 list_del(&timer->it.cpu.entry);
Linus Torvalds9465bee2005-10-21 15:36:00 -0700447 spin_unlock(&p->sighand->siglock);
448 }
449 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
Oleg Nesterov108150e2005-10-23 20:25:39 +0400451 if (!ret)
452 put_task_struct(p);
453 }
454
455 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
458/*
459 * Clean out CPU timers still ticking when a thread exited. The task
460 * pointer is cleared, and the expiry time is replaced with the residual
461 * time for later timer_gettime calls to return.
462 * This must be called with the siglock held.
463 */
464static void cleanup_timers(struct list_head *head,
465 cputime_t utime, cputime_t stime,
Ingo Molnar41b86e92007-07-09 18:51:58 +0200466 unsigned long long sum_exec_runtime)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct cpu_timer_list *timer, *next;
469 cputime_t ptime = cputime_add(utime, stime);
470
471 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 list_del_init(&timer->entry);
473 if (cputime_lt(timer->expires.cpu, ptime)) {
474 timer->expires.cpu = cputime_zero;
475 } else {
476 timer->expires.cpu = cputime_sub(timer->expires.cpu,
477 ptime);
478 }
479 }
480
481 ++head;
482 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 list_del_init(&timer->entry);
484 if (cputime_lt(timer->expires.cpu, utime)) {
485 timer->expires.cpu = cputime_zero;
486 } else {
487 timer->expires.cpu = cputime_sub(timer->expires.cpu,
488 utime);
489 }
490 }
491
492 ++head;
493 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 list_del_init(&timer->entry);
Ingo Molnar41b86e92007-07-09 18:51:58 +0200495 if (timer->expires.sched < sum_exec_runtime) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 timer->expires.sched = 0;
497 } else {
Ingo Molnar41b86e92007-07-09 18:51:58 +0200498 timer->expires.sched -= sum_exec_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500 }
501}
502
503/*
504 * These are both called with the siglock held, when the current thread
505 * is being reaped. When the final (leader) thread in the group is reaped,
506 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
507 */
508void posix_cpu_timers_exit(struct task_struct *tsk)
509{
510 cleanup_timers(tsk->cpu_timers,
Ingo Molnar41b86e92007-07-09 18:51:58 +0200511 tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513}
514void posix_cpu_timers_exit_group(struct task_struct *tsk)
515{
Stanislaw Gruszka17d42c12009-08-06 16:03:30 -0700516 struct signal_struct *const sig = tsk->signal;
Frank Mayharf06febc2008-09-12 09:54:39 -0700517
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 cleanup_timers(tsk->signal->cpu_timers,
Stanislaw Gruszka17d42c12009-08-06 16:03:30 -0700519 cputime_add(tsk->utime, sig->utime),
520 cputime_add(tsk->stime, sig->stime),
521 tsk->se.sum_exec_runtime + sig->sum_sched_runtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
525{
526 /*
527 * That's all for this thread or process.
528 * We leave our residual in expires to be reported.
529 */
530 put_task_struct(timer->it.cpu.task);
531 timer->it.cpu.task = NULL;
532 timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
533 timer->it.cpu.expires,
534 now);
535}
536
Stanislaw Gruszkad1e3b6d2009-07-29 12:15:28 +0200537static inline int expires_gt(cputime_t expires, cputime_t new_exp)
538{
539 return cputime_eq(expires, cputime_zero) ||
540 cputime_gt(expires, new_exp);
541}
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543/*
544 * Insert the timer on the appropriate list before any timers that
545 * expire later. This must be called with the tasklist_lock held
Stanislaw Gruszkac2873932010-03-11 14:04:42 -0800546 * for reading, interrupts disabled and p->sighand->siglock taken.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800548static void arm_timer(struct k_itimer *timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
550 struct task_struct *p = timer->it.cpu.task;
551 struct list_head *head, *listpos;
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800552 struct task_cputime *cputime_expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 struct cpu_timer_list *const nt = &timer->it.cpu;
554 struct cpu_timer_list *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800556 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
557 head = p->cpu_timers;
558 cputime_expires = &p->cputime_expires;
559 } else {
560 head = p->signal->cpu_timers;
561 cputime_expires = &p->signal->cputime_expires;
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 head += CPUCLOCK_WHICH(timer->it_clock);
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 listpos = head;
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800566 list_for_each_entry(next, head, entry) {
567 if (cpu_time_before(timer->it_clock, nt->expires, next->expires))
568 break;
569 listpos = &next->entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
571 list_add(&nt->entry, listpos);
572
573 if (listpos == head) {
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800574 union cpu_time_count *exp = &nt->expires;
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 /*
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800577 * We are the new earliest-expiring POSIX 1.b timer, hence
578 * need to update expiration cache. Take into account that
579 * for process timers we share expiration cache with itimers
580 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 */
582
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800583 switch (CPUCLOCK_WHICH(timer->it_clock)) {
584 case CPUCLOCK_PROF:
585 if (expires_gt(cputime_expires->prof_exp, exp->cpu))
586 cputime_expires->prof_exp = exp->cpu;
587 break;
588 case CPUCLOCK_VIRT:
589 if (expires_gt(cputime_expires->virt_exp, exp->cpu))
590 cputime_expires->virt_exp = exp->cpu;
591 break;
592 case CPUCLOCK_SCHED:
593 if (cputime_expires->sched_exp == 0 ||
594 cputime_expires->sched_exp > exp->sched)
595 cputime_expires->sched_exp = exp->sched;
596 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
600
601/*
602 * The timer is locked, fire it and arrange for its reload.
603 */
604static void cpu_timer_fire(struct k_itimer *timer)
605{
Stanislaw Gruszka1f169f82010-03-11 14:04:41 -0800606 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
607 /*
608 * User don't want any signal.
609 */
610 timer->it.cpu.expires.sched = 0;
611 } else if (unlikely(timer->sigq == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 /*
613 * This a special case for clock_nanosleep,
614 * not a normal timer from sys_timer_create.
615 */
616 wake_up_process(timer->it_process);
617 timer->it.cpu.expires.sched = 0;
618 } else if (timer->it.cpu.incr.sched == 0) {
619 /*
620 * One-shot timer. Clear it as soon as it's fired.
621 */
622 posix_timer_event(timer, 0);
623 timer->it.cpu.expires.sched = 0;
624 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
625 /*
626 * The signal did not get queued because the signal
627 * was ignored, so we won't get any callback to
628 * reload the timer. But we need to keep it
629 * ticking in case the signal is deliverable next time.
630 */
631 posix_cpu_timer_schedule(timer);
632 }
633}
634
635/*
Peter Zijlstra3997ad32009-02-12 15:00:52 +0100636 * Sample a process (thread group) timer for the given group_leader task.
637 * Must be called with tasklist_lock held for reading.
638 */
639static int cpu_timer_sample_group(const clockid_t which_clock,
640 struct task_struct *p,
641 union cpu_time_count *cpu)
642{
643 struct task_cputime cputime;
644
645 thread_group_cputimer(p, &cputime);
646 switch (CPUCLOCK_WHICH(which_clock)) {
647 default:
648 return -EINVAL;
649 case CPUCLOCK_PROF:
650 cpu->cpu = cputime_add(cputime.utime, cputime.stime);
651 break;
652 case CPUCLOCK_VIRT:
653 cpu->cpu = cputime.utime;
654 break;
655 case CPUCLOCK_SCHED:
656 cpu->sched = cputime.sum_exec_runtime + task_delta_exec(p);
657 break;
658 }
659 return 0;
660}
661
662/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 * Guts of sys_timer_settime for CPU timers.
664 * This is called with the timer locked and interrupts disabled.
665 * If we return TIMER_RETRY, it's necessary to release the timer's lock
666 * and try again. (This happens when the timer is in the middle of firing.)
667 */
668int posix_cpu_timer_set(struct k_itimer *timer, int flags,
669 struct itimerspec *new, struct itimerspec *old)
670{
671 struct task_struct *p = timer->it.cpu.task;
Stanislaw Gruszkaae1a78e2010-03-11 14:04:39 -0800672 union cpu_time_count old_expires, new_expires, old_incr, val;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 int ret;
674
675 if (unlikely(p == NULL)) {
676 /*
677 * Timer refers to a dead task's clock.
678 */
679 return -ESRCH;
680 }
681
682 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
683
684 read_lock(&tasklist_lock);
685 /*
686 * We need the tasklist_lock to protect against reaping that
Oleg Nesterovd30fda32010-05-26 14:43:13 -0700687 * clears p->sighand. If p has just been reaped, we can no
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 * longer get any information about it at all.
689 */
Oleg Nesterovd30fda32010-05-26 14:43:13 -0700690 if (unlikely(p->sighand == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 read_unlock(&tasklist_lock);
692 put_task_struct(p);
693 timer->it.cpu.task = NULL;
694 return -ESRCH;
695 }
696
697 /*
698 * Disarm any old timer after extracting its expiry time.
699 */
700 BUG_ON(!irqs_disabled());
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400701
702 ret = 0;
Stanislaw Gruszkaae1a78e2010-03-11 14:04:39 -0800703 old_incr = timer->it.cpu.incr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 spin_lock(&p->sighand->siglock);
705 old_expires = timer->it.cpu.expires;
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400706 if (unlikely(timer->it.cpu.firing)) {
707 timer->it.cpu.firing = -1;
708 ret = TIMER_RETRY;
709 } else
710 list_del_init(&timer->it.cpu.entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 /*
713 * We need to sample the current value to convert the new
714 * value from to relative and absolute, and to convert the
715 * old value from absolute to relative. To set a process
716 * timer, we need a sample to balance the thread expiry
717 * times (in arm_timer). With an absolute time, we must
718 * check if it's already passed. In short, we need a sample.
719 */
720 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
721 cpu_clock_sample(timer->it_clock, p, &val);
722 } else {
Peter Zijlstra3997ad32009-02-12 15:00:52 +0100723 cpu_timer_sample_group(timer->it_clock, p, &val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 }
725
726 if (old) {
727 if (old_expires.sched == 0) {
728 old->it_value.tv_sec = 0;
729 old->it_value.tv_nsec = 0;
730 } else {
731 /*
732 * Update the timer in case it has
733 * overrun already. If it has,
734 * we'll report it as having overrun
735 * and with the next reloaded timer
736 * already ticking, though we are
737 * swallowing that pending
738 * notification here to install the
739 * new setting.
740 */
741 bump_cpu_timer(timer, val);
742 if (cpu_time_before(timer->it_clock, val,
743 timer->it.cpu.expires)) {
744 old_expires = cpu_time_sub(
745 timer->it_clock,
746 timer->it.cpu.expires, val);
747 sample_to_timespec(timer->it_clock,
748 old_expires,
749 &old->it_value);
750 } else {
751 old->it_value.tv_nsec = 1;
752 old->it_value.tv_sec = 0;
753 }
754 }
755 }
756
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400757 if (unlikely(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 /*
759 * We are colliding with the timer actually firing.
760 * Punt after filling in the timer's old value, and
761 * disable this firing since we are already reporting
762 * it as an overrun (thanks to bump_cpu_timer above).
763 */
Stanislaw Gruszkac2873932010-03-11 14:04:42 -0800764 spin_unlock(&p->sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 goto out;
767 }
768
769 if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
770 cpu_time_add(timer->it_clock, &new_expires, val);
771 }
772
773 /*
774 * Install the new expiry time (or zero).
775 * For a timer with no notification action, we don't actually
776 * arm the timer (we'll just fake it for timer_gettime).
777 */
778 timer->it.cpu.expires = new_expires;
779 if (new_expires.sched != 0 &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 cpu_time_before(timer->it_clock, val, new_expires)) {
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800781 arm_timer(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
783
Stanislaw Gruszkac2873932010-03-11 14:04:42 -0800784 spin_unlock(&p->sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 read_unlock(&tasklist_lock);
786
787 /*
788 * Install the new reload setting, and
789 * set up the signal and overrun bookkeeping.
790 */
791 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
792 &new->it_interval);
793
794 /*
795 * This acts as a modification timestamp for the timer,
796 * so any automatic reload attempt will punt on seeing
797 * that we have reset the timer manually.
798 */
799 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
800 ~REQUEUE_PENDING;
801 timer->it_overrun_last = 0;
802 timer->it_overrun = -1;
803
804 if (new_expires.sched != 0 &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 !cpu_time_before(timer->it_clock, val, new_expires)) {
806 /*
807 * The designated time already passed, so we notify
808 * immediately, even if the thread never runs to
809 * accumulate more time on this clock.
810 */
811 cpu_timer_fire(timer);
812 }
813
814 ret = 0;
815 out:
816 if (old) {
817 sample_to_timespec(timer->it_clock,
Stanislaw Gruszkaae1a78e2010-03-11 14:04:39 -0800818 old_incr, &old->it_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 }
820 return ret;
821}
822
823void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
824{
825 union cpu_time_count now;
826 struct task_struct *p = timer->it.cpu.task;
827 int clear_dead;
828
829 /*
830 * Easy part: convert the reload time.
831 */
832 sample_to_timespec(timer->it_clock,
833 timer->it.cpu.incr, &itp->it_interval);
834
835 if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all. */
836 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
837 return;
838 }
839
840 if (unlikely(p == NULL)) {
841 /*
842 * This task already died and the timer will never fire.
843 * In this case, expires is actually the dead value.
844 */
845 dead:
846 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
847 &itp->it_value);
848 return;
849 }
850
851 /*
852 * Sample the clock to take the difference with the expiry time.
853 */
854 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
855 cpu_clock_sample(timer->it_clock, p, &now);
856 clear_dead = p->exit_state;
857 } else {
858 read_lock(&tasklist_lock);
Oleg Nesterovd30fda32010-05-26 14:43:13 -0700859 if (unlikely(p->sighand == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 /*
861 * The process has been reaped.
862 * We can't even collect a sample any more.
863 * Call the timer disarmed, nothing else to do.
864 */
865 put_task_struct(p);
866 timer->it.cpu.task = NULL;
867 timer->it.cpu.expires.sched = 0;
868 read_unlock(&tasklist_lock);
869 goto dead;
870 } else {
Peter Zijlstra3997ad32009-02-12 15:00:52 +0100871 cpu_timer_sample_group(timer->it_clock, p, &now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 clear_dead = (unlikely(p->exit_state) &&
873 thread_group_empty(p));
874 }
875 read_unlock(&tasklist_lock);
876 }
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (unlikely(clear_dead)) {
879 /*
880 * We've noticed that the thread is dead, but
881 * not yet reaped. Take this opportunity to
882 * drop our task ref.
883 */
884 clear_dead_task(timer, now);
885 goto dead;
886 }
887
888 if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
889 sample_to_timespec(timer->it_clock,
890 cpu_time_sub(timer->it_clock,
891 timer->it.cpu.expires, now),
892 &itp->it_value);
893 } else {
894 /*
895 * The timer should have expired already, but the firing
896 * hasn't taken place yet. Say it's just about to expire.
897 */
898 itp->it_value.tv_nsec = 1;
899 itp->it_value.tv_sec = 0;
900 }
901}
902
903/*
904 * Check for any per-thread CPU timers that have fired and move them off
905 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
906 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
907 */
908static void check_thread_timers(struct task_struct *tsk,
909 struct list_head *firing)
910{
Linus Torvaldse80eda92005-10-23 10:02:50 -0700911 int maxfire;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct list_head *timers = tsk->cpu_timers;
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100913 struct signal_struct *const sig = tsk->signal;
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800914 unsigned long soft;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Linus Torvaldse80eda92005-10-23 10:02:50 -0700916 maxfire = 20;
Frank Mayharf06febc2008-09-12 09:54:39 -0700917 tsk->cputime_expires.prof_exp = cputime_zero;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700919 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 struct cpu_timer_list,
921 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -0700922 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
Frank Mayharf06febc2008-09-12 09:54:39 -0700923 tsk->cputime_expires.prof_exp = t->expires.cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 break;
925 }
926 t->firing = 1;
927 list_move_tail(&t->entry, firing);
928 }
929
930 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -0700931 maxfire = 20;
Frank Mayharf06febc2008-09-12 09:54:39 -0700932 tsk->cputime_expires.virt_exp = cputime_zero;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700934 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 struct cpu_timer_list,
936 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -0700937 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
Frank Mayharf06febc2008-09-12 09:54:39 -0700938 tsk->cputime_expires.virt_exp = t->expires.cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 break;
940 }
941 t->firing = 1;
942 list_move_tail(&t->entry, firing);
943 }
944
945 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -0700946 maxfire = 20;
Frank Mayharf06febc2008-09-12 09:54:39 -0700947 tsk->cputime_expires.sched_exp = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700949 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 struct cpu_timer_list,
951 entry);
Ingo Molnar41b86e92007-07-09 18:51:58 +0200952 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
Frank Mayharf06febc2008-09-12 09:54:39 -0700953 tsk->cputime_expires.sched_exp = t->expires.sched;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 break;
955 }
956 t->firing = 1;
957 list_move_tail(&t->entry, firing);
958 }
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100959
960 /*
961 * Check for the special case thread timers.
962 */
Jiri Slaby78d7d402010-03-05 13:42:54 -0800963 soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800964 if (soft != RLIM_INFINITY) {
Jiri Slaby78d7d402010-03-05 13:42:54 -0800965 unsigned long hard =
966 ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100967
Peter Zijlstra5a52dd52008-01-25 21:08:32 +0100968 if (hard != RLIM_INFINITY &&
969 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100970 /*
971 * At the hard limit, we just die.
972 * No need to calculate anything else now.
973 */
974 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
975 return;
976 }
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800977 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100978 /*
979 * At the soft limit, send a SIGXCPU every second.
980 */
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800981 if (soft < hard) {
982 soft += USEC_PER_SEC;
983 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100984 }
Hiroshi Shimamoto81d50bb2008-05-15 19:42:49 -0700985 printk(KERN_INFO
986 "RT Watchdog Timeout: %s[%d]\n",
987 tsk->comm, task_pid_nr(tsk));
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100988 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
989 }
990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
992
Stanislaw Gruszka15365c12010-03-11 14:04:31 -0800993static void stop_process_timers(struct signal_struct *sig)
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100994{
Stanislaw Gruszka15365c12010-03-11 14:04:31 -0800995 struct thread_group_cputimer *cputimer = &sig->cputimer;
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100996 unsigned long flags;
997
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100998 spin_lock_irqsave(&cputimer->lock, flags);
999 cputimer->running = 0;
1000 spin_unlock_irqrestore(&cputimer->lock, flags);
1001}
1002
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001003static u32 onecputick;
1004
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +02001005static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
1006 cputime_t *expires, cputime_t cur_time, int signo)
1007{
1008 if (cputime_eq(it->expires, cputime_zero))
1009 return;
1010
1011 if (cputime_ge(cur_time, it->expires)) {
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001012 if (!cputime_eq(it->incr, cputime_zero)) {
1013 it->expires = cputime_add(it->expires, it->incr);
1014 it->error += it->incr_error;
1015 if (it->error >= onecputick) {
1016 it->expires = cputime_sub(it->expires,
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02001017 cputime_one_jiffy);
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001018 it->error -= onecputick;
1019 }
Xiao Guangrong3f0a5252009-08-10 10:52:30 +08001020 } else {
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001021 it->expires = cputime_zero;
Xiao Guangrong3f0a5252009-08-10 10:52:30 +08001022 }
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +02001023
Xiao Guangrong3f0a5252009-08-10 10:52:30 +08001024 trace_itimer_expire(signo == SIGPROF ?
1025 ITIMER_PROF : ITIMER_VIRTUAL,
1026 tsk->signal->leader_pid, cur_time);
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +02001027 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
1028 }
1029
1030 if (!cputime_eq(it->expires, cputime_zero) &&
1031 (cputime_eq(*expires, cputime_zero) ||
1032 cputime_lt(it->expires, *expires))) {
1033 *expires = it->expires;
1034 }
1035}
1036
Stanislaw Gruszka29f87b72010-04-27 14:12:15 -07001037/**
1038 * task_cputime_zero - Check a task_cputime struct for all zero fields.
1039 *
1040 * @cputime: The struct to compare.
1041 *
1042 * Checks @cputime to see if all fields are zero. Returns true if all fields
1043 * are zero, false if any field is nonzero.
1044 */
1045static inline int task_cputime_zero(const struct task_cputime *cputime)
1046{
1047 if (cputime_eq(cputime->utime, cputime_zero) &&
1048 cputime_eq(cputime->stime, cputime_zero) &&
1049 cputime->sum_exec_runtime == 0)
1050 return 1;
1051 return 0;
1052}
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054/*
1055 * Check for any per-thread CPU timers that have fired and move them
1056 * off the tsk->*_timers list onto the firing list. Per-thread timers
1057 * have already been taken off.
1058 */
1059static void check_process_timers(struct task_struct *tsk,
1060 struct list_head *firing)
1061{
Linus Torvaldse80eda92005-10-23 10:02:50 -07001062 int maxfire;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 struct signal_struct *const sig = tsk->signal;
Frank Mayharf06febc2008-09-12 09:54:39 -07001064 cputime_t utime, ptime, virt_expires, prof_expires;
Ingo Molnar41b86e92007-07-09 18:51:58 +02001065 unsigned long long sum_sched_runtime, sched_expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 struct list_head *timers = sig->cpu_timers;
Frank Mayharf06febc2008-09-12 09:54:39 -07001067 struct task_cputime cputime;
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001068 unsigned long soft;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 * Collect the current process totals.
1072 */
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +01001073 thread_group_cputimer(tsk, &cputime);
Frank Mayharf06febc2008-09-12 09:54:39 -07001074 utime = cputime.utime;
1075 ptime = cputime_add(utime, cputime.stime);
1076 sum_sched_runtime = cputime.sum_exec_runtime;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001077 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 prof_expires = cputime_zero;
1079 while (!list_empty(timers)) {
WANG Congee7dd202008-04-04 20:54:10 +02001080 struct cpu_timer_list *tl = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 struct cpu_timer_list,
1082 entry);
WANG Congee7dd202008-04-04 20:54:10 +02001083 if (!--maxfire || cputime_lt(ptime, tl->expires.cpu)) {
1084 prof_expires = tl->expires.cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 break;
1086 }
WANG Congee7dd202008-04-04 20:54:10 +02001087 tl->firing = 1;
1088 list_move_tail(&tl->entry, firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 }
1090
1091 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001092 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 virt_expires = cputime_zero;
1094 while (!list_empty(timers)) {
WANG Congee7dd202008-04-04 20:54:10 +02001095 struct cpu_timer_list *tl = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 struct cpu_timer_list,
1097 entry);
WANG Congee7dd202008-04-04 20:54:10 +02001098 if (!--maxfire || cputime_lt(utime, tl->expires.cpu)) {
1099 virt_expires = tl->expires.cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 break;
1101 }
WANG Congee7dd202008-04-04 20:54:10 +02001102 tl->firing = 1;
1103 list_move_tail(&tl->entry, firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
1106 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001107 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 sched_expires = 0;
1109 while (!list_empty(timers)) {
WANG Congee7dd202008-04-04 20:54:10 +02001110 struct cpu_timer_list *tl = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 struct cpu_timer_list,
1112 entry);
WANG Congee7dd202008-04-04 20:54:10 +02001113 if (!--maxfire || sum_sched_runtime < tl->expires.sched) {
1114 sched_expires = tl->expires.sched;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 break;
1116 }
WANG Congee7dd202008-04-04 20:54:10 +02001117 tl->firing = 1;
1118 list_move_tail(&tl->entry, firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120
1121 /*
1122 * Check for the special case process timers.
1123 */
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +02001124 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
1125 SIGPROF);
1126 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
1127 SIGVTALRM);
Jiri Slaby78d7d402010-03-05 13:42:54 -08001128 soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001129 if (soft != RLIM_INFINITY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 unsigned long psecs = cputime_to_secs(ptime);
Jiri Slaby78d7d402010-03-05 13:42:54 -08001131 unsigned long hard =
1132 ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 cputime_t x;
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001134 if (psecs >= hard) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 /*
1136 * At the hard limit, we just die.
1137 * No need to calculate anything else now.
1138 */
1139 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1140 return;
1141 }
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001142 if (psecs >= soft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 /*
1144 * At the soft limit, send a SIGXCPU every second.
1145 */
1146 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001147 if (soft < hard) {
1148 soft++;
1149 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151 }
Jiri Slabyd4bb52742010-03-05 13:42:53 -08001152 x = secs_to_cputime(soft);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (cputime_eq(prof_expires, cputime_zero) ||
1154 cputime_lt(x, prof_expires)) {
1155 prof_expires = x;
1156 }
1157 }
1158
Stanislaw Gruszka29f87b72010-04-27 14:12:15 -07001159 sig->cputime_expires.prof_exp = prof_expires;
1160 sig->cputime_expires.virt_exp = virt_expires;
1161 sig->cputime_expires.sched_exp = sched_expires;
1162 if (task_cputime_zero(&sig->cputime_expires))
1163 stop_process_timers(sig);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164}
1165
1166/*
1167 * This is called from the signal code (via do_schedule_next_timer)
1168 * when the last timer signal was delivered and we have to reload the timer.
1169 */
1170void posix_cpu_timer_schedule(struct k_itimer *timer)
1171{
1172 struct task_struct *p = timer->it.cpu.task;
1173 union cpu_time_count now;
1174
1175 if (unlikely(p == NULL))
1176 /*
1177 * The task was cleaned up already, no future firings.
1178 */
Roland McGrath708f4302005-10-30 15:03:13 -08001179 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181 /*
1182 * Fetch the current sample and update the timer's expiry time.
1183 */
1184 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1185 cpu_clock_sample(timer->it_clock, p, &now);
1186 bump_cpu_timer(timer, now);
1187 if (unlikely(p->exit_state)) {
1188 clear_dead_task(timer, now);
Roland McGrath708f4302005-10-30 15:03:13 -08001189 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
1191 read_lock(&tasklist_lock); /* arm_timer needs it. */
Stanislaw Gruszkac2873932010-03-11 14:04:42 -08001192 spin_lock(&p->sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 } else {
1194 read_lock(&tasklist_lock);
Oleg Nesterovd30fda32010-05-26 14:43:13 -07001195 if (unlikely(p->sighand == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 /*
1197 * The process has been reaped.
1198 * We can't even collect a sample any more.
1199 */
1200 put_task_struct(p);
1201 timer->it.cpu.task = p = NULL;
1202 timer->it.cpu.expires.sched = 0;
Roland McGrath708f4302005-10-30 15:03:13 -08001203 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1205 /*
1206 * We've noticed that the thread is dead, but
1207 * not yet reaped. Take this opportunity to
1208 * drop our task ref.
1209 */
1210 clear_dead_task(timer, now);
Roland McGrath708f4302005-10-30 15:03:13 -08001211 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 }
Stanislaw Gruszkac2873932010-03-11 14:04:42 -08001213 spin_lock(&p->sighand->siglock);
Peter Zijlstra3997ad32009-02-12 15:00:52 +01001214 cpu_timer_sample_group(timer->it_clock, p, &now);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 bump_cpu_timer(timer, now);
1216 /* Leave the tasklist_lock locked for the call below. */
1217 }
1218
1219 /*
1220 * Now re-arm for the new expiry time.
1221 */
Stanislaw Gruszkac2873932010-03-11 14:04:42 -08001222 BUG_ON(!irqs_disabled());
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -08001223 arm_timer(timer);
Stanislaw Gruszkac2873932010-03-11 14:04:42 -08001224 spin_unlock(&p->sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Roland McGrath708f4302005-10-30 15:03:13 -08001226out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 read_unlock(&tasklist_lock);
Roland McGrath708f4302005-10-30 15:03:13 -08001228
1229out:
1230 timer->it_overrun_last = timer->it_overrun;
1231 timer->it_overrun = -1;
1232 ++timer->it_requeue_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233}
1234
Frank Mayharf06febc2008-09-12 09:54:39 -07001235/**
Frank Mayharf06febc2008-09-12 09:54:39 -07001236 * task_cputime_expired - Compare two task_cputime entities.
1237 *
1238 * @sample: The task_cputime structure to be checked for expiration.
1239 * @expires: Expiration times, against which @sample will be checked.
1240 *
1241 * Checks @sample against @expires to see if any field of @sample has expired.
1242 * Returns true if any field of the former is greater than the corresponding
1243 * field of the latter if the latter field is set. Otherwise returns false.
1244 */
1245static inline int task_cputime_expired(const struct task_cputime *sample,
1246 const struct task_cputime *expires)
1247{
1248 if (!cputime_eq(expires->utime, cputime_zero) &&
1249 cputime_ge(sample->utime, expires->utime))
1250 return 1;
1251 if (!cputime_eq(expires->stime, cputime_zero) &&
1252 cputime_ge(cputime_add(sample->utime, sample->stime),
1253 expires->stime))
1254 return 1;
1255 if (expires->sum_exec_runtime != 0 &&
1256 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1257 return 1;
1258 return 0;
1259}
1260
1261/**
1262 * fastpath_timer_check - POSIX CPU timers fast path.
1263 *
1264 * @tsk: The task (thread) being checked.
Frank Mayharf06febc2008-09-12 09:54:39 -07001265 *
Frank Mayharbb34d922008-09-12 09:54:39 -07001266 * Check the task and thread group timers. If both are zero (there are no
1267 * timers set) return false. Otherwise snapshot the task and thread group
1268 * timers and compare them with the corresponding expiration times. Return
1269 * true if a timer has expired, else return false.
Frank Mayharf06febc2008-09-12 09:54:39 -07001270 */
Frank Mayharbb34d922008-09-12 09:54:39 -07001271static inline int fastpath_timer_check(struct task_struct *tsk)
Frank Mayharf06febc2008-09-12 09:54:39 -07001272{
Oleg Nesterovad133ba2008-11-17 15:39:47 +01001273 struct signal_struct *sig;
Frank Mayharf06febc2008-09-12 09:54:39 -07001274
Frank Mayharbb34d922008-09-12 09:54:39 -07001275 if (!task_cputime_zero(&tsk->cputime_expires)) {
1276 struct task_cputime task_sample = {
1277 .utime = tsk->utime,
1278 .stime = tsk->stime,
1279 .sum_exec_runtime = tsk->se.sum_exec_runtime
1280 };
1281
1282 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1283 return 1;
1284 }
Oleg Nesterovad133ba2008-11-17 15:39:47 +01001285
1286 sig = tsk->signal;
Stanislaw Gruszka29f87b72010-04-27 14:12:15 -07001287 if (sig->cputimer.running) {
Frank Mayharbb34d922008-09-12 09:54:39 -07001288 struct task_cputime group_sample;
1289
Oleg Nesterov8d1f4312010-06-11 20:04:46 +02001290 spin_lock(&sig->cputimer.lock);
1291 group_sample = sig->cputimer.cputime;
1292 spin_unlock(&sig->cputimer.lock);
1293
Frank Mayharbb34d922008-09-12 09:54:39 -07001294 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1295 return 1;
1296 }
Oleg Nesterov37bebc72009-03-23 20:34:11 +01001297
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001298 return 0;
Frank Mayharf06febc2008-09-12 09:54:39 -07001299}
1300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301/*
1302 * This is called from the timer interrupt handler. The irq handler has
1303 * already updated our counts. We need to check if any timers fire now.
1304 * Interrupts are disabled.
1305 */
1306void run_posix_cpu_timers(struct task_struct *tsk)
1307{
1308 LIST_HEAD(firing);
1309 struct k_itimer *timer, *next;
Oleg Nesterov0bdd2ed2010-06-11 01:10:18 +02001310 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 BUG_ON(!irqs_disabled());
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 /*
Frank Mayharf06febc2008-09-12 09:54:39 -07001315 * The fast path checks that there are no expired thread or thread
Frank Mayharbb34d922008-09-12 09:54:39 -07001316 * group timers. If that's so, just return.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 */
Frank Mayharbb34d922008-09-12 09:54:39 -07001318 if (!fastpath_timer_check(tsk))
Frank Mayharf06febc2008-09-12 09:54:39 -07001319 return;
Ingo Molnar5ce73a42008-09-14 17:11:46 +02001320
Oleg Nesterov0bdd2ed2010-06-11 01:10:18 +02001321 if (!lock_task_sighand(tsk, &flags))
1322 return;
Frank Mayharbb34d922008-09-12 09:54:39 -07001323 /*
1324 * Here we take off tsk->signal->cpu_timers[N] and
1325 * tsk->cpu_timers[N] all the timers that are firing, and
1326 * put them on the firing list.
1327 */
1328 check_thread_timers(tsk, &firing);
Stanislaw Gruszka29f87b72010-04-27 14:12:15 -07001329 /*
1330 * If there are any active process wide timers (POSIX 1.b, itimers,
1331 * RLIMIT_CPU) cputimer must be running.
1332 */
1333 if (tsk->signal->cputimer.running)
1334 check_process_timers(tsk, &firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Frank Mayharbb34d922008-09-12 09:54:39 -07001336 /*
1337 * We must release these locks before taking any timer's lock.
1338 * There is a potential race with timer deletion here, as the
1339 * siglock now protects our private firing list. We have set
1340 * the firing flag in each timer, so that a deletion attempt
1341 * that gets the timer lock before we do will give it up and
1342 * spin until we've taken care of that timer below.
1343 */
Oleg Nesterov0bdd2ed2010-06-11 01:10:18 +02001344 unlock_task_sighand(tsk, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 /*
1347 * Now that all the timers on our list have the firing flag,
1348 * noone will touch their list entries but us. We'll take
1349 * each timer's lock before clearing its firing flag, so no
1350 * timer call will interfere.
1351 */
1352 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
H Hartley Sweeten6e85c5b2009-04-29 19:14:32 -04001353 int cpu_firing;
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 spin_lock(&timer->it_lock);
1356 list_del_init(&timer->it.cpu.entry);
H Hartley Sweeten6e85c5b2009-04-29 19:14:32 -04001357 cpu_firing = timer->it.cpu.firing;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 timer->it.cpu.firing = 0;
1359 /*
1360 * The firing flag is -1 if we collided with a reset
1361 * of the timer, which already reported this
1362 * almost-firing as an overrun. So don't generate an event.
1363 */
H Hartley Sweeten6e85c5b2009-04-29 19:14:32 -04001364 if (likely(cpu_firing >= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 cpu_timer_fire(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 spin_unlock(&timer->it_lock);
1367 }
1368}
1369
1370/*
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001371 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
Frank Mayharf06febc2008-09-12 09:54:39 -07001372 * The tsk->sighand->siglock must be held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 */
1374void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1375 cputime_t *newval, cputime_t *oldval)
1376{
1377 union cpu_time_count now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379 BUG_ON(clock_idx == CPUCLOCK_SCHED);
Peter Zijlstra4cd4c1b2009-02-05 12:24:16 +01001380 cpu_timer_sample_group(clock_idx, tsk, &now);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
1382 if (oldval) {
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001383 /*
1384 * We are setting itimer. The *oldval is absolute and we update
1385 * it to be relative, *newval argument is relative and we update
1386 * it to be absolute.
1387 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (!cputime_eq(*oldval, cputime_zero)) {
1389 if (cputime_le(*oldval, now.cpu)) {
1390 /* Just about to fire. */
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02001391 *oldval = cputime_one_jiffy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 } else {
1393 *oldval = cputime_sub(*oldval, now.cpu);
1394 }
1395 }
1396
1397 if (cputime_eq(*newval, cputime_zero))
1398 return;
1399 *newval = cputime_add(*newval, now.cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
1401
1402 /*
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001403 * Update expiration cache if we are the earliest timer, or eventually
1404 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 */
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001406 switch (clock_idx) {
1407 case CPUCLOCK_PROF:
1408 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
Frank Mayharf06febc2008-09-12 09:54:39 -07001409 tsk->signal->cputime_expires.prof_exp = *newval;
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001410 break;
1411 case CPUCLOCK_VIRT:
1412 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
Frank Mayharf06febc2008-09-12 09:54:39 -07001413 tsk->signal->cputime_expires.virt_exp = *newval;
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001414 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
1416}
1417
Toyo Abee4b76552006-09-29 02:00:29 -07001418static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1419 struct timespec *rqtp, struct itimerspec *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 struct k_itimer timer;
1422 int error;
1423
1424 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 * Set up a temporary timer and then wait for it to go off.
1426 */
1427 memset(&timer, 0, sizeof timer);
1428 spin_lock_init(&timer.it_lock);
1429 timer.it_clock = which_clock;
1430 timer.it_overrun = -1;
1431 error = posix_cpu_timer_create(&timer);
1432 timer.it_process = current;
1433 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 static struct itimerspec zero_it;
Toyo Abee4b76552006-09-29 02:00:29 -07001435
1436 memset(it, 0, sizeof *it);
1437 it->it_value = *rqtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 spin_lock_irq(&timer.it_lock);
Toyo Abee4b76552006-09-29 02:00:29 -07001440 error = posix_cpu_timer_set(&timer, flags, it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (error) {
1442 spin_unlock_irq(&timer.it_lock);
1443 return error;
1444 }
1445
1446 while (!signal_pending(current)) {
1447 if (timer.it.cpu.expires.sched == 0) {
1448 /*
1449 * Our timer fired and was reset.
1450 */
1451 spin_unlock_irq(&timer.it_lock);
1452 return 0;
1453 }
1454
1455 /*
1456 * Block until cpu_timer_fire (or a signal) wakes us.
1457 */
1458 __set_current_state(TASK_INTERRUPTIBLE);
1459 spin_unlock_irq(&timer.it_lock);
1460 schedule();
1461 spin_lock_irq(&timer.it_lock);
1462 }
1463
1464 /*
1465 * We were interrupted by a signal.
1466 */
1467 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
Toyo Abee4b76552006-09-29 02:00:29 -07001468 posix_cpu_timer_set(&timer, 0, &zero_it, it);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 spin_unlock_irq(&timer.it_lock);
1470
Toyo Abee4b76552006-09-29 02:00:29 -07001471 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 /*
1473 * It actually did fire already.
1474 */
1475 return 0;
1476 }
1477
Toyo Abee4b76552006-09-29 02:00:29 -07001478 error = -ERESTART_RESTARTBLOCK;
1479 }
1480
1481 return error;
1482}
1483
1484int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1485 struct timespec *rqtp, struct timespec __user *rmtp)
1486{
1487 struct restart_block *restart_block =
1488 &current_thread_info()->restart_block;
1489 struct itimerspec it;
1490 int error;
1491
1492 /*
1493 * Diagnose required errors first.
1494 */
1495 if (CPUCLOCK_PERTHREAD(which_clock) &&
1496 (CPUCLOCK_PID(which_clock) == 0 ||
1497 CPUCLOCK_PID(which_clock) == current->pid))
1498 return -EINVAL;
1499
1500 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1501
1502 if (error == -ERESTART_RESTARTBLOCK) {
1503
1504 if (flags & TIMER_ABSTIME)
1505 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 /*
Toyo Abee4b76552006-09-29 02:00:29 -07001507 * Report back to the user the time still remaining.
1508 */
1509 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 return -EFAULT;
1511
Toyo Abe1711ef32006-09-29 02:00:28 -07001512 restart_block->fn = posix_cpu_nsleep_restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 restart_block->arg0 = which_clock;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001514 restart_block->arg1 = (unsigned long) rmtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 restart_block->arg2 = rqtp->tv_sec;
1516 restart_block->arg3 = rqtp->tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return error;
1519}
1520
Toyo Abe1711ef32006-09-29 02:00:28 -07001521long posix_cpu_nsleep_restart(struct restart_block *restart_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
1523 clockid_t which_clock = restart_block->arg0;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001524 struct timespec __user *rmtp;
1525 struct timespec t;
Toyo Abee4b76552006-09-29 02:00:29 -07001526 struct itimerspec it;
1527 int error;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001528
1529 rmtp = (struct timespec __user *) restart_block->arg1;
1530 t.tv_sec = restart_block->arg2;
1531 t.tv_nsec = restart_block->arg3;
1532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 restart_block->fn = do_no_restart_syscall;
Toyo Abee4b76552006-09-29 02:00:29 -07001534 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1535
1536 if (error == -ERESTART_RESTARTBLOCK) {
1537 /*
1538 * Report back to the user the time still remaining.
1539 */
1540 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1541 return -EFAULT;
1542
1543 restart_block->fn = posix_cpu_nsleep_restart;
1544 restart_block->arg0 = which_clock;
1545 restart_block->arg1 = (unsigned long) rmtp;
1546 restart_block->arg2 = t.tv_sec;
1547 restart_block->arg3 = t.tv_nsec;
1548 }
1549 return error;
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
1553
1554#define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1555#define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1556
Thomas Gleixnera924b042006-01-09 20:52:27 -08001557static int process_cpu_clock_getres(const clockid_t which_clock,
1558 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
1560 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1561}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001562static int process_cpu_clock_get(const clockid_t which_clock,
1563 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564{
1565 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1566}
1567static int process_cpu_timer_create(struct k_itimer *timer)
1568{
1569 timer->it_clock = PROCESS_CLOCK;
1570 return posix_cpu_timer_create(timer);
1571}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001572static int process_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner97735f22006-01-09 20:52:37 -08001573 struct timespec *rqtp,
1574 struct timespec __user *rmtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
Thomas Gleixner97735f22006-01-09 20:52:37 -08001576 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577}
Toyo Abe1711ef32006-09-29 02:00:28 -07001578static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1579{
1580 return -EINVAL;
1581}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001582static int thread_cpu_clock_getres(const clockid_t which_clock,
1583 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
1585 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1586}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001587static int thread_cpu_clock_get(const clockid_t which_clock,
1588 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589{
1590 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1591}
1592static int thread_cpu_timer_create(struct k_itimer *timer)
1593{
1594 timer->it_clock = THREAD_CLOCK;
1595 return posix_cpu_timer_create(timer);
1596}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001597static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner97735f22006-01-09 20:52:37 -08001598 struct timespec *rqtp, struct timespec __user *rmtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599{
1600 return -EINVAL;
1601}
Toyo Abe1711ef32006-09-29 02:00:28 -07001602static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1603{
1604 return -EINVAL;
1605}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
1607static __init int init_posix_cpu_timers(void)
1608{
1609 struct k_clock process = {
1610 .clock_getres = process_cpu_clock_getres,
1611 .clock_get = process_cpu_clock_get,
1612 .clock_set = do_posix_clock_nosettime,
1613 .timer_create = process_cpu_timer_create,
1614 .nsleep = process_cpu_nsleep,
Toyo Abe1711ef32006-09-29 02:00:28 -07001615 .nsleep_restart = process_cpu_nsleep_restart,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 };
1617 struct k_clock thread = {
1618 .clock_getres = thread_cpu_clock_getres,
1619 .clock_get = thread_cpu_clock_get,
1620 .clock_set = do_posix_clock_nosettime,
1621 .timer_create = thread_cpu_timer_create,
1622 .nsleep = thread_cpu_nsleep,
Toyo Abe1711ef32006-09-29 02:00:28 -07001623 .nsleep_restart = thread_cpu_nsleep_restart,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 };
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001625 struct timespec ts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
1627 register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1628 register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1629
Stanislaw Gruszkaa42548a2009-07-29 12:15:29 +02001630 cputime_to_timespec(cputime_one_jiffy, &ts);
Stanislaw Gruszka8356b5f2009-07-29 12:15:27 +02001631 onecputick = ts.tv_nsec;
1632 WARN_ON(ts.tv_sec != 0);
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 return 0;
1635}
1636__initcall(init_posix_cpu_timers);