blob: 68c96376e84a6078da9c07d6ebfc2b1c64a3e2a7 [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>
7#include <asm/uaccess.h>
8#include <linux/errno.h>
9
Thomas Gleixnera924b042006-01-09 20:52:27 -080010static int check_clock(const clockid_t which_clock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070011{
12 int error = 0;
13 struct task_struct *p;
14 const pid_t pid = CPUCLOCK_PID(which_clock);
15
16 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
17 return -EINVAL;
18
19 if (pid == 0)
20 return 0;
21
22 read_lock(&tasklist_lock);
23 p = find_task_by_pid(pid);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -070024 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
25 same_thread_group(p, current) : thread_group_leader(p))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 error = -EINVAL;
27 }
28 read_unlock(&tasklist_lock);
29
30 return error;
31}
32
33static inline union cpu_time_count
Thomas Gleixnera924b042006-01-09 20:52:27 -080034timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070035{
36 union cpu_time_count ret;
37 ret.sched = 0; /* high half always zero when .cpu used */
38 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
Oleg Nesterovee500f22005-11-28 13:43:55 -080039 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 } else {
41 ret.cpu = timespec_to_cputime(tp);
42 }
43 return ret;
44}
45
Thomas Gleixnera924b042006-01-09 20:52:27 -080046static void sample_to_timespec(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 union cpu_time_count cpu,
48 struct timespec *tp)
49{
50 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
51 tp->tv_sec = div_long_long_rem(cpu.sched,
52 NSEC_PER_SEC, &tp->tv_nsec);
53 } else {
54 cputime_to_timespec(cpu.cpu, tp);
55 }
56}
57
Thomas Gleixnera924b042006-01-09 20:52:27 -080058static inline int cpu_time_before(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 union cpu_time_count now,
60 union cpu_time_count then)
61{
62 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
63 return now.sched < then.sched;
64 } else {
65 return cputime_lt(now.cpu, then.cpu);
66 }
67}
Thomas Gleixnera924b042006-01-09 20:52:27 -080068static inline void cpu_time_add(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 union cpu_time_count *acc,
70 union cpu_time_count val)
71{
72 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
73 acc->sched += val.sched;
74 } else {
75 acc->cpu = cputime_add(acc->cpu, val.cpu);
76 }
77}
Thomas Gleixnera924b042006-01-09 20:52:27 -080078static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 union cpu_time_count a,
80 union cpu_time_count b)
81{
82 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
83 a.sched -= b.sched;
84 } else {
85 a.cpu = cputime_sub(a.cpu, b.cpu);
86 }
87 return a;
88}
89
90/*
Thomas Gleixnerac08c262006-10-17 00:09:39 -070091 * Divide and limit the result to res >= 1
92 *
93 * This is necessary to prevent signal delivery starvation, when the result of
94 * the division would be rounded down to 0.
95 */
96static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
97{
98 cputime_t res = cputime_div(time, div);
99
100 return max_t(cputime_t, res, 1);
101}
102
103/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * Update expiry time from increment, and increase overrun count,
105 * given the current clock sample.
106 */
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400107static void bump_cpu_timer(struct k_itimer *timer,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 union cpu_time_count now)
109{
110 int i;
111
112 if (timer->it.cpu.incr.sched == 0)
113 return;
114
115 if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
116 unsigned long long delta, incr;
117
118 if (now.sched < timer->it.cpu.expires.sched)
119 return;
120 incr = timer->it.cpu.incr.sched;
121 delta = now.sched + incr - timer->it.cpu.expires.sched;
122 /* Don't use (incr*2 < delta), incr*2 might overflow. */
123 for (i = 0; incr < delta - incr; i++)
124 incr = incr << 1;
125 for (; i >= 0; incr >>= 1, i--) {
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400126 if (delta < incr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 continue;
128 timer->it.cpu.expires.sched += incr;
129 timer->it_overrun += 1 << i;
130 delta -= incr;
131 }
132 } else {
133 cputime_t delta, incr;
134
135 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
136 return;
137 incr = timer->it.cpu.incr.cpu;
138 delta = cputime_sub(cputime_add(now.cpu, incr),
139 timer->it.cpu.expires.cpu);
140 /* Don't use (incr*2 < delta), incr*2 might overflow. */
141 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
142 incr = cputime_add(incr, incr);
143 for (; i >= 0; incr = cputime_halve(incr), i--) {
Oleg Nesterov7a4ed932005-10-26 20:26:53 +0400144 if (cputime_lt(delta, incr))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 continue;
146 timer->it.cpu.expires.cpu =
147 cputime_add(timer->it.cpu.expires.cpu, incr);
148 timer->it_overrun += 1 << i;
149 delta = cputime_sub(delta, incr);
150 }
151 }
152}
153
154static inline cputime_t prof_ticks(struct task_struct *p)
155{
156 return cputime_add(p->utime, p->stime);
157}
158static inline cputime_t virt_ticks(struct task_struct *p)
159{
160 return p->utime;
161}
162static inline unsigned long long sched_ns(struct task_struct *p)
163{
Ingo Molnar41b86e92007-07-09 18:51:58 +0200164 return task_sched_runtime(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
Thomas Gleixnera924b042006-01-09 20:52:27 -0800167int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
169 int error = check_clock(which_clock);
170 if (!error) {
171 tp->tv_sec = 0;
172 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
173 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
174 /*
175 * If sched_clock is using a cycle counter, we
176 * don't have any idea of its true resolution
177 * exported, but it is much more than 1s/HZ.
178 */
179 tp->tv_nsec = 1;
180 }
181 }
182 return error;
183}
184
Thomas Gleixnera924b042006-01-09 20:52:27 -0800185int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
187 /*
188 * You can never reset a CPU clock, but we check for other errors
189 * in the call before failing with EPERM.
190 */
191 int error = check_clock(which_clock);
192 if (error == 0) {
193 error = -EPERM;
194 }
195 return error;
196}
197
198
199/*
200 * Sample a per-thread clock for the given task.
201 */
Thomas Gleixnera924b042006-01-09 20:52:27 -0800202static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 union cpu_time_count *cpu)
204{
205 switch (CPUCLOCK_WHICH(which_clock)) {
206 default:
207 return -EINVAL;
208 case CPUCLOCK_PROF:
209 cpu->cpu = prof_ticks(p);
210 break;
211 case CPUCLOCK_VIRT:
212 cpu->cpu = virt_ticks(p);
213 break;
214 case CPUCLOCK_SCHED:
215 cpu->sched = sched_ns(p);
216 break;
217 }
218 return 0;
219}
220
221/*
222 * Sample a process (thread group) clock for the given group_leader task.
223 * Must be called with tasklist_lock held for reading.
224 * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
225 */
226static int cpu_clock_sample_group_locked(unsigned int clock_idx,
227 struct task_struct *p,
228 union cpu_time_count *cpu)
229{
230 struct task_struct *t = p;
231 switch (clock_idx) {
232 default:
233 return -EINVAL;
234 case CPUCLOCK_PROF:
235 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
236 do {
237 cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
238 t = next_thread(t);
239 } while (t != p);
240 break;
241 case CPUCLOCK_VIRT:
242 cpu->cpu = p->signal->utime;
243 do {
244 cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
245 t = next_thread(t);
246 } while (t != p);
247 break;
248 case CPUCLOCK_SCHED:
Ingo Molnar41b86e92007-07-09 18:51:58 +0200249 cpu->sched = p->signal->sum_sched_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 /* Add in each other live thread. */
251 while ((t = next_thread(t)) != p) {
Ingo Molnar41b86e92007-07-09 18:51:58 +0200252 cpu->sched += t->se.sum_exec_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
David S. Miller0aec63e2006-01-06 15:36:48 -0800254 cpu->sched += sched_ns(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 break;
256 }
257 return 0;
258}
259
260/*
261 * Sample a process (thread group) clock for the given group_leader task.
262 * Must be called with tasklist_lock held for reading.
263 */
Thomas Gleixnera924b042006-01-09 20:52:27 -0800264static int cpu_clock_sample_group(const clockid_t which_clock,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 struct task_struct *p,
266 union cpu_time_count *cpu)
267{
268 int ret;
269 unsigned long flags;
270 spin_lock_irqsave(&p->sighand->siglock, flags);
271 ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
272 cpu);
273 spin_unlock_irqrestore(&p->sighand->siglock, flags);
274 return ret;
275}
276
277
Thomas Gleixnera924b042006-01-09 20:52:27 -0800278int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 const pid_t pid = CPUCLOCK_PID(which_clock);
281 int error = -EINVAL;
282 union cpu_time_count rtn;
283
284 if (pid == 0) {
285 /*
286 * Special case constant value for our own clocks.
287 * We don't have to do any lookup to find ourselves.
288 */
289 if (CPUCLOCK_PERTHREAD(which_clock)) {
290 /*
291 * Sampling just ourselves we can do with no locking.
292 */
293 error = cpu_clock_sample(which_clock,
294 current, &rtn);
295 } else {
296 read_lock(&tasklist_lock);
297 error = cpu_clock_sample_group(which_clock,
298 current, &rtn);
299 read_unlock(&tasklist_lock);
300 }
301 } else {
302 /*
303 * Find the given PID, and validate that the caller
304 * should be able to see it.
305 */
306 struct task_struct *p;
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800307 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 p = find_task_by_pid(pid);
309 if (p) {
310 if (CPUCLOCK_PERTHREAD(which_clock)) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700311 if (same_thread_group(p, current)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 error = cpu_clock_sample(which_clock,
313 p, &rtn);
314 }
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800315 } else {
316 read_lock(&tasklist_lock);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700317 if (thread_group_leader(p) && p->signal) {
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800318 error =
319 cpu_clock_sample_group(which_clock,
320 p, &rtn);
321 }
322 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 }
324 }
Paul E. McKenney1f2ea082007-02-16 01:28:22 -0800325 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327
328 if (error)
329 return error;
330 sample_to_timespec(which_clock, rtn, tp);
331 return 0;
332}
333
334
335/*
336 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
337 * This is called from sys_timer_create with the new timer already locked.
338 */
339int posix_cpu_timer_create(struct k_itimer *new_timer)
340{
341 int ret = 0;
342 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
343 struct task_struct *p;
344
345 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
346 return -EINVAL;
347
348 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
349 new_timer->it.cpu.incr.sched = 0;
350 new_timer->it.cpu.expires.sched = 0;
351
352 read_lock(&tasklist_lock);
353 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
354 if (pid == 0) {
355 p = current;
356 } else {
357 p = find_task_by_pid(pid);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700358 if (p && !same_thread_group(p, current))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 p = NULL;
360 }
361 } else {
362 if (pid == 0) {
363 p = current->group_leader;
364 } else {
365 p = find_task_by_pid(pid);
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700366 if (p && !thread_group_leader(p))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 p = NULL;
368 }
369 }
370 new_timer->it.cpu.task = p;
371 if (p) {
372 get_task_struct(p);
373 } else {
374 ret = -EINVAL;
375 }
376 read_unlock(&tasklist_lock);
377
378 return ret;
379}
380
381/*
382 * Clean up a CPU-clock timer that is about to be destroyed.
383 * This is called from timer deletion with the timer already locked.
384 * If we return TIMER_RETRY, it's necessary to release the timer's lock
385 * and try again. (This happens when the timer is in the middle of firing.)
386 */
387int posix_cpu_timer_del(struct k_itimer *timer)
388{
389 struct task_struct *p = timer->it.cpu.task;
Oleg Nesterov108150e2005-10-23 20:25:39 +0400390 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Oleg Nesterov108150e2005-10-23 20:25:39 +0400392 if (likely(p != NULL)) {
Linus Torvalds9465bee2005-10-21 15:36:00 -0700393 read_lock(&tasklist_lock);
394 if (unlikely(p->signal == NULL)) {
395 /*
396 * We raced with the reaping of the task.
397 * The deletion should have cleared us off the list.
398 */
399 BUG_ON(!list_empty(&timer->it.cpu.entry));
400 } else {
Linus Torvalds9465bee2005-10-21 15:36:00 -0700401 spin_lock(&p->sighand->siglock);
Oleg Nesterov108150e2005-10-23 20:25:39 +0400402 if (timer->it.cpu.firing)
403 ret = TIMER_RETRY;
404 else
405 list_del(&timer->it.cpu.entry);
Linus Torvalds9465bee2005-10-21 15:36:00 -0700406 spin_unlock(&p->sighand->siglock);
407 }
408 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Oleg Nesterov108150e2005-10-23 20:25:39 +0400410 if (!ret)
411 put_task_struct(p);
412 }
413
414 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
417/*
418 * Clean out CPU timers still ticking when a thread exited. The task
419 * pointer is cleared, and the expiry time is replaced with the residual
420 * time for later timer_gettime calls to return.
421 * This must be called with the siglock held.
422 */
423static void cleanup_timers(struct list_head *head,
424 cputime_t utime, cputime_t stime,
Ingo Molnar41b86e92007-07-09 18:51:58 +0200425 unsigned long long sum_exec_runtime)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426{
427 struct cpu_timer_list *timer, *next;
428 cputime_t ptime = cputime_add(utime, stime);
429
430 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 list_del_init(&timer->entry);
432 if (cputime_lt(timer->expires.cpu, ptime)) {
433 timer->expires.cpu = cputime_zero;
434 } else {
435 timer->expires.cpu = cputime_sub(timer->expires.cpu,
436 ptime);
437 }
438 }
439
440 ++head;
441 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 list_del_init(&timer->entry);
443 if (cputime_lt(timer->expires.cpu, utime)) {
444 timer->expires.cpu = cputime_zero;
445 } else {
446 timer->expires.cpu = cputime_sub(timer->expires.cpu,
447 utime);
448 }
449 }
450
451 ++head;
452 list_for_each_entry_safe(timer, next, head, entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 list_del_init(&timer->entry);
Ingo Molnar41b86e92007-07-09 18:51:58 +0200454 if (timer->expires.sched < sum_exec_runtime) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 timer->expires.sched = 0;
456 } else {
Ingo Molnar41b86e92007-07-09 18:51:58 +0200457 timer->expires.sched -= sum_exec_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
459 }
460}
461
462/*
463 * These are both called with the siglock held, when the current thread
464 * is being reaped. When the final (leader) thread in the group is reaped,
465 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
466 */
467void posix_cpu_timers_exit(struct task_struct *tsk)
468{
469 cleanup_timers(tsk->cpu_timers,
Ingo Molnar41b86e92007-07-09 18:51:58 +0200470 tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472}
473void posix_cpu_timers_exit_group(struct task_struct *tsk)
474{
475 cleanup_timers(tsk->signal->cpu_timers,
476 cputime_add(tsk->utime, tsk->signal->utime),
477 cputime_add(tsk->stime, tsk->signal->stime),
Ingo Molnar41b86e92007-07-09 18:51:58 +0200478 tsk->se.sum_exec_runtime + tsk->signal->sum_sched_runtime);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
481
482/*
483 * Set the expiry times of all the threads in the process so one of them
484 * will go off before the process cumulative expiry total is reached.
485 */
486static void process_timer_rebalance(struct task_struct *p,
487 unsigned int clock_idx,
488 union cpu_time_count expires,
489 union cpu_time_count val)
490{
491 cputime_t ticks, left;
492 unsigned long long ns, nsleft;
493 struct task_struct *t = p;
494 unsigned int nthreads = atomic_read(&p->signal->live);
495
Oleg Nesterovca531a02005-10-24 14:36:28 +0400496 if (!nthreads)
497 return;
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 switch (clock_idx) {
500 default:
501 BUG();
502 break;
503 case CPUCLOCK_PROF:
Thomas Gleixnerac08c262006-10-17 00:09:39 -0700504 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
505 nthreads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 do {
Andrew Morton7fd93cf2005-11-07 00:57:59 -0800507 if (likely(!(t->flags & PF_EXITING))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 ticks = cputime_add(prof_ticks(t), left);
509 if (cputime_eq(t->it_prof_expires,
510 cputime_zero) ||
511 cputime_gt(t->it_prof_expires, ticks)) {
512 t->it_prof_expires = ticks;
513 }
514 }
515 t = next_thread(t);
516 } while (t != p);
517 break;
518 case CPUCLOCK_VIRT:
Thomas Gleixnerac08c262006-10-17 00:09:39 -0700519 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
520 nthreads);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 do {
Andrew Morton7fd93cf2005-11-07 00:57:59 -0800522 if (likely(!(t->flags & PF_EXITING))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 ticks = cputime_add(virt_ticks(t), left);
524 if (cputime_eq(t->it_virt_expires,
525 cputime_zero) ||
526 cputime_gt(t->it_virt_expires, ticks)) {
527 t->it_virt_expires = ticks;
528 }
529 }
530 t = next_thread(t);
531 } while (t != p);
532 break;
533 case CPUCLOCK_SCHED:
534 nsleft = expires.sched - val.sched;
535 do_div(nsleft, nthreads);
Thomas Gleixnerac08c262006-10-17 00:09:39 -0700536 nsleft = max_t(unsigned long long, nsleft, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 do {
Andrew Morton7fd93cf2005-11-07 00:57:59 -0800538 if (likely(!(t->flags & PF_EXITING))) {
Ingo Molnar41b86e92007-07-09 18:51:58 +0200539 ns = t->se.sum_exec_runtime + nsleft;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (t->it_sched_expires == 0 ||
541 t->it_sched_expires > ns) {
542 t->it_sched_expires = ns;
543 }
544 }
545 t = next_thread(t);
546 } while (t != p);
547 break;
548 }
549}
550
551static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
552{
553 /*
554 * That's all for this thread or process.
555 * We leave our residual in expires to be reported.
556 */
557 put_task_struct(timer->it.cpu.task);
558 timer->it.cpu.task = NULL;
559 timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
560 timer->it.cpu.expires,
561 now);
562}
563
564/*
565 * Insert the timer on the appropriate list before any timers that
566 * expire later. This must be called with the tasklist_lock held
567 * for reading, and interrupts disabled.
568 */
569static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
570{
571 struct task_struct *p = timer->it.cpu.task;
572 struct list_head *head, *listpos;
573 struct cpu_timer_list *const nt = &timer->it.cpu;
574 struct cpu_timer_list *next;
575 unsigned long i;
576
577 head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
578 p->cpu_timers : p->signal->cpu_timers);
579 head += CPUCLOCK_WHICH(timer->it_clock);
580
581 BUG_ON(!irqs_disabled());
582 spin_lock(&p->sighand->siglock);
583
584 listpos = head;
585 if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
586 list_for_each_entry(next, head, entry) {
Linus Torvalds70ab81c2005-10-26 11:23:06 -0700587 if (next->expires.sched > nt->expires.sched)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 break;
Linus Torvalds70ab81c2005-10-26 11:23:06 -0700589 listpos = &next->entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
591 } else {
592 list_for_each_entry(next, head, entry) {
Linus Torvalds70ab81c2005-10-26 11:23:06 -0700593 if (cputime_gt(next->expires.cpu, nt->expires.cpu))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 break;
Linus Torvalds70ab81c2005-10-26 11:23:06 -0700595 listpos = &next->entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 }
597 }
598 list_add(&nt->entry, listpos);
599
600 if (listpos == head) {
601 /*
602 * We are the new earliest-expiring timer.
603 * If we are a thread timer, there can always
604 * be a process timer telling us to stop earlier.
605 */
606
607 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
608 switch (CPUCLOCK_WHICH(timer->it_clock)) {
609 default:
610 BUG();
611 case CPUCLOCK_PROF:
612 if (cputime_eq(p->it_prof_expires,
613 cputime_zero) ||
614 cputime_gt(p->it_prof_expires,
615 nt->expires.cpu))
616 p->it_prof_expires = nt->expires.cpu;
617 break;
618 case CPUCLOCK_VIRT:
619 if (cputime_eq(p->it_virt_expires,
620 cputime_zero) ||
621 cputime_gt(p->it_virt_expires,
622 nt->expires.cpu))
623 p->it_virt_expires = nt->expires.cpu;
624 break;
625 case CPUCLOCK_SCHED:
626 if (p->it_sched_expires == 0 ||
627 p->it_sched_expires > nt->expires.sched)
628 p->it_sched_expires = nt->expires.sched;
629 break;
630 }
631 } else {
632 /*
633 * For a process timer, we must balance
634 * all the live threads' expirations.
635 */
636 switch (CPUCLOCK_WHICH(timer->it_clock)) {
637 default:
638 BUG();
639 case CPUCLOCK_VIRT:
640 if (!cputime_eq(p->signal->it_virt_expires,
641 cputime_zero) &&
642 cputime_lt(p->signal->it_virt_expires,
643 timer->it.cpu.expires.cpu))
644 break;
645 goto rebalance;
646 case CPUCLOCK_PROF:
647 if (!cputime_eq(p->signal->it_prof_expires,
648 cputime_zero) &&
649 cputime_lt(p->signal->it_prof_expires,
650 timer->it.cpu.expires.cpu))
651 break;
652 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
653 if (i != RLIM_INFINITY &&
654 i <= cputime_to_secs(timer->it.cpu.expires.cpu))
655 break;
656 goto rebalance;
657 case CPUCLOCK_SCHED:
658 rebalance:
659 process_timer_rebalance(
660 timer->it.cpu.task,
661 CPUCLOCK_WHICH(timer->it_clock),
662 timer->it.cpu.expires, now);
663 break;
664 }
665 }
666 }
667
668 spin_unlock(&p->sighand->siglock);
669}
670
671/*
672 * The timer is locked, fire it and arrange for its reload.
673 */
674static void cpu_timer_fire(struct k_itimer *timer)
675{
676 if (unlikely(timer->sigq == NULL)) {
677 /*
678 * This a special case for clock_nanosleep,
679 * not a normal timer from sys_timer_create.
680 */
681 wake_up_process(timer->it_process);
682 timer->it.cpu.expires.sched = 0;
683 } else if (timer->it.cpu.incr.sched == 0) {
684 /*
685 * One-shot timer. Clear it as soon as it's fired.
686 */
687 posix_timer_event(timer, 0);
688 timer->it.cpu.expires.sched = 0;
689 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
690 /*
691 * The signal did not get queued because the signal
692 * was ignored, so we won't get any callback to
693 * reload the timer. But we need to keep it
694 * ticking in case the signal is deliverable next time.
695 */
696 posix_cpu_timer_schedule(timer);
697 }
698}
699
700/*
701 * Guts of sys_timer_settime for CPU timers.
702 * This is called with the timer locked and interrupts disabled.
703 * If we return TIMER_RETRY, it's necessary to release the timer's lock
704 * and try again. (This happens when the timer is in the middle of firing.)
705 */
706int posix_cpu_timer_set(struct k_itimer *timer, int flags,
707 struct itimerspec *new, struct itimerspec *old)
708{
709 struct task_struct *p = timer->it.cpu.task;
710 union cpu_time_count old_expires, new_expires, val;
711 int ret;
712
713 if (unlikely(p == NULL)) {
714 /*
715 * Timer refers to a dead task's clock.
716 */
717 return -ESRCH;
718 }
719
720 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
721
722 read_lock(&tasklist_lock);
723 /*
724 * We need the tasklist_lock to protect against reaping that
725 * clears p->signal. If p has just been reaped, we can no
726 * longer get any information about it at all.
727 */
728 if (unlikely(p->signal == NULL)) {
729 read_unlock(&tasklist_lock);
730 put_task_struct(p);
731 timer->it.cpu.task = NULL;
732 return -ESRCH;
733 }
734
735 /*
736 * Disarm any old timer after extracting its expiry time.
737 */
738 BUG_ON(!irqs_disabled());
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400739
740 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 spin_lock(&p->sighand->siglock);
742 old_expires = timer->it.cpu.expires;
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400743 if (unlikely(timer->it.cpu.firing)) {
744 timer->it.cpu.firing = -1;
745 ret = TIMER_RETRY;
746 } else
747 list_del_init(&timer->it.cpu.entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 spin_unlock(&p->sighand->siglock);
749
750 /*
751 * We need to sample the current value to convert the new
752 * value from to relative and absolute, and to convert the
753 * old value from absolute to relative. To set a process
754 * timer, we need a sample to balance the thread expiry
755 * times (in arm_timer). With an absolute time, we must
756 * check if it's already passed. In short, we need a sample.
757 */
758 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
759 cpu_clock_sample(timer->it_clock, p, &val);
760 } else {
761 cpu_clock_sample_group(timer->it_clock, p, &val);
762 }
763
764 if (old) {
765 if (old_expires.sched == 0) {
766 old->it_value.tv_sec = 0;
767 old->it_value.tv_nsec = 0;
768 } else {
769 /*
770 * Update the timer in case it has
771 * overrun already. If it has,
772 * we'll report it as having overrun
773 * and with the next reloaded timer
774 * already ticking, though we are
775 * swallowing that pending
776 * notification here to install the
777 * new setting.
778 */
779 bump_cpu_timer(timer, val);
780 if (cpu_time_before(timer->it_clock, val,
781 timer->it.cpu.expires)) {
782 old_expires = cpu_time_sub(
783 timer->it_clock,
784 timer->it.cpu.expires, val);
785 sample_to_timespec(timer->it_clock,
786 old_expires,
787 &old->it_value);
788 } else {
789 old->it_value.tv_nsec = 1;
790 old->it_value.tv_sec = 0;
791 }
792 }
793 }
794
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400795 if (unlikely(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 /*
797 * We are colliding with the timer actually firing.
798 * Punt after filling in the timer's old value, and
799 * disable this firing since we are already reporting
800 * it as an overrun (thanks to bump_cpu_timer above).
801 */
802 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 goto out;
804 }
805
806 if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
807 cpu_time_add(timer->it_clock, &new_expires, val);
808 }
809
810 /*
811 * Install the new expiry time (or zero).
812 * For a timer with no notification action, we don't actually
813 * arm the timer (we'll just fake it for timer_gettime).
814 */
815 timer->it.cpu.expires = new_expires;
816 if (new_expires.sched != 0 &&
817 (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
818 cpu_time_before(timer->it_clock, val, new_expires)) {
819 arm_timer(timer, val);
820 }
821
822 read_unlock(&tasklist_lock);
823
824 /*
825 * Install the new reload setting, and
826 * set up the signal and overrun bookkeeping.
827 */
828 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
829 &new->it_interval);
830
831 /*
832 * This acts as a modification timestamp for the timer,
833 * so any automatic reload attempt will punt on seeing
834 * that we have reset the timer manually.
835 */
836 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
837 ~REQUEUE_PENDING;
838 timer->it_overrun_last = 0;
839 timer->it_overrun = -1;
840
841 if (new_expires.sched != 0 &&
842 (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
843 !cpu_time_before(timer->it_clock, val, new_expires)) {
844 /*
845 * The designated time already passed, so we notify
846 * immediately, even if the thread never runs to
847 * accumulate more time on this clock.
848 */
849 cpu_timer_fire(timer);
850 }
851
852 ret = 0;
853 out:
854 if (old) {
855 sample_to_timespec(timer->it_clock,
856 timer->it.cpu.incr, &old->it_interval);
857 }
858 return ret;
859}
860
861void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
862{
863 union cpu_time_count now;
864 struct task_struct *p = timer->it.cpu.task;
865 int clear_dead;
866
867 /*
868 * Easy part: convert the reload time.
869 */
870 sample_to_timespec(timer->it_clock,
871 timer->it.cpu.incr, &itp->it_interval);
872
873 if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all. */
874 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
875 return;
876 }
877
878 if (unlikely(p == NULL)) {
879 /*
880 * This task already died and the timer will never fire.
881 * In this case, expires is actually the dead value.
882 */
883 dead:
884 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
885 &itp->it_value);
886 return;
887 }
888
889 /*
890 * Sample the clock to take the difference with the expiry time.
891 */
892 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
893 cpu_clock_sample(timer->it_clock, p, &now);
894 clear_dead = p->exit_state;
895 } else {
896 read_lock(&tasklist_lock);
897 if (unlikely(p->signal == NULL)) {
898 /*
899 * The process has been reaped.
900 * We can't even collect a sample any more.
901 * Call the timer disarmed, nothing else to do.
902 */
903 put_task_struct(p);
904 timer->it.cpu.task = NULL;
905 timer->it.cpu.expires.sched = 0;
906 read_unlock(&tasklist_lock);
907 goto dead;
908 } else {
909 cpu_clock_sample_group(timer->it_clock, p, &now);
910 clear_dead = (unlikely(p->exit_state) &&
911 thread_group_empty(p));
912 }
913 read_unlock(&tasklist_lock);
914 }
915
916 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
917 if (timer->it.cpu.incr.sched == 0 &&
918 cpu_time_before(timer->it_clock,
919 timer->it.cpu.expires, now)) {
920 /*
921 * Do-nothing timer expired and has no reload,
922 * so it's as if it was never set.
923 */
924 timer->it.cpu.expires.sched = 0;
925 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
926 return;
927 }
928 /*
929 * Account for any expirations and reloads that should
930 * have happened.
931 */
932 bump_cpu_timer(timer, now);
933 }
934
935 if (unlikely(clear_dead)) {
936 /*
937 * We've noticed that the thread is dead, but
938 * not yet reaped. Take this opportunity to
939 * drop our task ref.
940 */
941 clear_dead_task(timer, now);
942 goto dead;
943 }
944
945 if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
946 sample_to_timespec(timer->it_clock,
947 cpu_time_sub(timer->it_clock,
948 timer->it.cpu.expires, now),
949 &itp->it_value);
950 } else {
951 /*
952 * The timer should have expired already, but the firing
953 * hasn't taken place yet. Say it's just about to expire.
954 */
955 itp->it_value.tv_nsec = 1;
956 itp->it_value.tv_sec = 0;
957 }
958}
959
960/*
961 * Check for any per-thread CPU timers that have fired and move them off
962 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
963 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
964 */
965static void check_thread_timers(struct task_struct *tsk,
966 struct list_head *firing)
967{
Linus Torvaldse80eda92005-10-23 10:02:50 -0700968 int maxfire;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 struct list_head *timers = tsk->cpu_timers;
970
Linus Torvaldse80eda92005-10-23 10:02:50 -0700971 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 tsk->it_prof_expires = cputime_zero;
973 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700974 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 struct cpu_timer_list,
976 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -0700977 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 tsk->it_prof_expires = t->expires.cpu;
979 break;
980 }
981 t->firing = 1;
982 list_move_tail(&t->entry, firing);
983 }
984
985 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -0700986 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 tsk->it_virt_expires = cputime_zero;
988 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700989 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 struct cpu_timer_list,
991 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -0700992 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 tsk->it_virt_expires = t->expires.cpu;
994 break;
995 }
996 t->firing = 1;
997 list_move_tail(&t->entry, firing);
998 }
999
1000 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001001 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 tsk->it_sched_expires = 0;
1003 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001004 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 struct cpu_timer_list,
1006 entry);
Ingo Molnar41b86e92007-07-09 18:51:58 +02001007 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 tsk->it_sched_expires = t->expires.sched;
1009 break;
1010 }
1011 t->firing = 1;
1012 list_move_tail(&t->entry, firing);
1013 }
1014}
1015
1016/*
1017 * Check for any per-thread CPU timers that have fired and move them
1018 * off the tsk->*_timers list onto the firing list. Per-thread timers
1019 * have already been taken off.
1020 */
1021static void check_process_timers(struct task_struct *tsk,
1022 struct list_head *firing)
1023{
Linus Torvaldse80eda92005-10-23 10:02:50 -07001024 int maxfire;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 struct signal_struct *const sig = tsk->signal;
1026 cputime_t utime, stime, ptime, virt_expires, prof_expires;
Ingo Molnar41b86e92007-07-09 18:51:58 +02001027 unsigned long long sum_sched_runtime, sched_expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 struct task_struct *t;
1029 struct list_head *timers = sig->cpu_timers;
1030
1031 /*
1032 * Don't sample the current process CPU clocks if there are no timers.
1033 */
1034 if (list_empty(&timers[CPUCLOCK_PROF]) &&
1035 cputime_eq(sig->it_prof_expires, cputime_zero) &&
1036 sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1037 list_empty(&timers[CPUCLOCK_VIRT]) &&
1038 cputime_eq(sig->it_virt_expires, cputime_zero) &&
1039 list_empty(&timers[CPUCLOCK_SCHED]))
1040 return;
1041
1042 /*
1043 * Collect the current process totals.
1044 */
1045 utime = sig->utime;
1046 stime = sig->stime;
Ingo Molnar41b86e92007-07-09 18:51:58 +02001047 sum_sched_runtime = sig->sum_sched_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 t = tsk;
1049 do {
1050 utime = cputime_add(utime, t->utime);
1051 stime = cputime_add(stime, t->stime);
Ingo Molnar41b86e92007-07-09 18:51:58 +02001052 sum_sched_runtime += t->se.sum_exec_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 t = next_thread(t);
1054 } while (t != tsk);
1055 ptime = cputime_add(utime, stime);
1056
Linus Torvaldse80eda92005-10-23 10:02:50 -07001057 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 prof_expires = cputime_zero;
1059 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001060 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 struct cpu_timer_list,
1062 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -07001063 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 prof_expires = t->expires.cpu;
1065 break;
1066 }
1067 t->firing = 1;
1068 list_move_tail(&t->entry, firing);
1069 }
1070
1071 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001072 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 virt_expires = cputime_zero;
1074 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001075 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 struct cpu_timer_list,
1077 entry);
Linus Torvaldse80eda92005-10-23 10:02:50 -07001078 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 virt_expires = t->expires.cpu;
1080 break;
1081 }
1082 t->firing = 1;
1083 list_move_tail(&t->entry, firing);
1084 }
1085
1086 ++timers;
Linus Torvaldse80eda92005-10-23 10:02:50 -07001087 maxfire = 20;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 sched_expires = 0;
1089 while (!list_empty(timers)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001090 struct cpu_timer_list *t = list_first_entry(timers,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 struct cpu_timer_list,
1092 entry);
Ingo Molnar41b86e92007-07-09 18:51:58 +02001093 if (!--maxfire || sum_sched_runtime < t->expires.sched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 sched_expires = t->expires.sched;
1095 break;
1096 }
1097 t->firing = 1;
1098 list_move_tail(&t->entry, firing);
1099 }
1100
1101 /*
1102 * Check for the special case process timers.
1103 */
1104 if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1105 if (cputime_ge(ptime, sig->it_prof_expires)) {
1106 /* ITIMER_PROF fires and reloads. */
1107 sig->it_prof_expires = sig->it_prof_incr;
1108 if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1109 sig->it_prof_expires = cputime_add(
1110 sig->it_prof_expires, ptime);
1111 }
1112 __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1113 }
1114 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1115 (cputime_eq(prof_expires, cputime_zero) ||
1116 cputime_lt(sig->it_prof_expires, prof_expires))) {
1117 prof_expires = sig->it_prof_expires;
1118 }
1119 }
1120 if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1121 if (cputime_ge(utime, sig->it_virt_expires)) {
1122 /* ITIMER_VIRTUAL fires and reloads. */
1123 sig->it_virt_expires = sig->it_virt_incr;
1124 if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1125 sig->it_virt_expires = cputime_add(
1126 sig->it_virt_expires, utime);
1127 }
1128 __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1129 }
1130 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1131 (cputime_eq(virt_expires, cputime_zero) ||
1132 cputime_lt(sig->it_virt_expires, virt_expires))) {
1133 virt_expires = sig->it_virt_expires;
1134 }
1135 }
1136 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1137 unsigned long psecs = cputime_to_secs(ptime);
1138 cputime_t x;
1139 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1140 /*
1141 * At the hard limit, we just die.
1142 * No need to calculate anything else now.
1143 */
1144 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1145 return;
1146 }
1147 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1148 /*
1149 * At the soft limit, send a SIGXCPU every second.
1150 */
1151 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1152 if (sig->rlim[RLIMIT_CPU].rlim_cur
1153 < sig->rlim[RLIMIT_CPU].rlim_max) {
1154 sig->rlim[RLIMIT_CPU].rlim_cur++;
1155 }
1156 }
1157 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1158 if (cputime_eq(prof_expires, cputime_zero) ||
1159 cputime_lt(x, prof_expires)) {
1160 prof_expires = x;
1161 }
1162 }
1163
1164 if (!cputime_eq(prof_expires, cputime_zero) ||
1165 !cputime_eq(virt_expires, cputime_zero) ||
1166 sched_expires != 0) {
1167 /*
1168 * Rebalance the threads' expiry times for the remaining
1169 * process CPU timers.
1170 */
1171
1172 cputime_t prof_left, virt_left, ticks;
1173 unsigned long long sched_left, sched;
1174 const unsigned int nthreads = atomic_read(&sig->live);
1175
Oleg Nesterovca531a02005-10-24 14:36:28 +04001176 if (!nthreads)
1177 return;
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 prof_left = cputime_sub(prof_expires, utime);
1180 prof_left = cputime_sub(prof_left, stime);
Thomas Gleixnerac08c262006-10-17 00:09:39 -07001181 prof_left = cputime_div_non_zero(prof_left, nthreads);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 virt_left = cputime_sub(virt_expires, utime);
Thomas Gleixnerac08c262006-10-17 00:09:39 -07001183 virt_left = cputime_div_non_zero(virt_left, nthreads);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (sched_expires) {
Ingo Molnar41b86e92007-07-09 18:51:58 +02001185 sched_left = sched_expires - sum_sched_runtime;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 do_div(sched_left, nthreads);
Thomas Gleixnerac08c262006-10-17 00:09:39 -07001187 sched_left = max_t(unsigned long long, sched_left, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 } else {
1189 sched_left = 0;
1190 }
1191 t = tsk;
1192 do {
Oleg Nesterov8f17fc22006-06-15 20:11:15 +04001193 if (unlikely(t->flags & PF_EXITING))
1194 continue;
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 ticks = cputime_add(cputime_add(t->utime, t->stime),
1197 prof_left);
1198 if (!cputime_eq(prof_expires, cputime_zero) &&
1199 (cputime_eq(t->it_prof_expires, cputime_zero) ||
1200 cputime_gt(t->it_prof_expires, ticks))) {
1201 t->it_prof_expires = ticks;
1202 }
1203
1204 ticks = cputime_add(t->utime, virt_left);
1205 if (!cputime_eq(virt_expires, cputime_zero) &&
1206 (cputime_eq(t->it_virt_expires, cputime_zero) ||
1207 cputime_gt(t->it_virt_expires, ticks))) {
1208 t->it_virt_expires = ticks;
1209 }
1210
Ingo Molnar41b86e92007-07-09 18:51:58 +02001211 sched = t->se.sum_exec_runtime + sched_left;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (sched_expires && (t->it_sched_expires == 0 ||
1213 t->it_sched_expires > sched)) {
1214 t->it_sched_expires = sched;
1215 }
Oleg Nesterov8f17fc22006-06-15 20:11:15 +04001216 } while ((t = next_thread(t)) != tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
1218}
1219
1220/*
1221 * This is called from the signal code (via do_schedule_next_timer)
1222 * when the last timer signal was delivered and we have to reload the timer.
1223 */
1224void posix_cpu_timer_schedule(struct k_itimer *timer)
1225{
1226 struct task_struct *p = timer->it.cpu.task;
1227 union cpu_time_count now;
1228
1229 if (unlikely(p == NULL))
1230 /*
1231 * The task was cleaned up already, no future firings.
1232 */
Roland McGrath708f4302005-10-30 15:03:13 -08001233 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 /*
1236 * Fetch the current sample and update the timer's expiry time.
1237 */
1238 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1239 cpu_clock_sample(timer->it_clock, p, &now);
1240 bump_cpu_timer(timer, now);
1241 if (unlikely(p->exit_state)) {
1242 clear_dead_task(timer, now);
Roland McGrath708f4302005-10-30 15:03:13 -08001243 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245 read_lock(&tasklist_lock); /* arm_timer needs it. */
1246 } else {
1247 read_lock(&tasklist_lock);
1248 if (unlikely(p->signal == NULL)) {
1249 /*
1250 * The process has been reaped.
1251 * We can't even collect a sample any more.
1252 */
1253 put_task_struct(p);
1254 timer->it.cpu.task = p = NULL;
1255 timer->it.cpu.expires.sched = 0;
Roland McGrath708f4302005-10-30 15:03:13 -08001256 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1258 /*
1259 * We've noticed that the thread is dead, but
1260 * not yet reaped. Take this opportunity to
1261 * drop our task ref.
1262 */
1263 clear_dead_task(timer, now);
Roland McGrath708f4302005-10-30 15:03:13 -08001264 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 }
1266 cpu_clock_sample_group(timer->it_clock, p, &now);
1267 bump_cpu_timer(timer, now);
1268 /* Leave the tasklist_lock locked for the call below. */
1269 }
1270
1271 /*
1272 * Now re-arm for the new expiry time.
1273 */
1274 arm_timer(timer, now);
1275
Roland McGrath708f4302005-10-30 15:03:13 -08001276out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 read_unlock(&tasklist_lock);
Roland McGrath708f4302005-10-30 15:03:13 -08001278
1279out:
1280 timer->it_overrun_last = timer->it_overrun;
1281 timer->it_overrun = -1;
1282 ++timer->it_requeue_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
1285/*
1286 * This is called from the timer interrupt handler. The irq handler has
1287 * already updated our counts. We need to check if any timers fire now.
1288 * Interrupts are disabled.
1289 */
1290void run_posix_cpu_timers(struct task_struct *tsk)
1291{
1292 LIST_HEAD(firing);
1293 struct k_itimer *timer, *next;
1294
1295 BUG_ON(!irqs_disabled());
1296
1297#define UNEXPIRED(clock) \
1298 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1299 cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1300
1301 if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1302 (tsk->it_sched_expires == 0 ||
Ingo Molnar41b86e92007-07-09 18:51:58 +02001303 tsk->se.sum_exec_runtime < tsk->it_sched_expires))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 return;
1305
1306#undef UNEXPIRED
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 /*
1309 * Double-check with locks held.
1310 */
1311 read_lock(&tasklist_lock);
Oleg Nesterov30f1e3d2006-06-15 20:11:43 +04001312 if (likely(tsk->signal != NULL)) {
1313 spin_lock(&tsk->sighand->siglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Oleg Nesterov30f1e3d2006-06-15 20:11:43 +04001315 /*
1316 * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1317 * all the timers that are firing, and put them on the firing list.
1318 */
1319 check_thread_timers(tsk, &firing);
1320 check_process_timers(tsk, &firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Oleg Nesterov30f1e3d2006-06-15 20:11:43 +04001322 /*
1323 * We must release these locks before taking any timer's lock.
1324 * There is a potential race with timer deletion here, as the
1325 * siglock now protects our private firing list. We have set
1326 * the firing flag in each timer, so that a deletion attempt
1327 * that gets the timer lock before we do will give it up and
1328 * spin until we've taken care of that timer below.
1329 */
1330 spin_unlock(&tsk->sighand->siglock);
1331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 read_unlock(&tasklist_lock);
1333
1334 /*
1335 * Now that all the timers on our list have the firing flag,
1336 * noone will touch their list entries but us. We'll take
1337 * each timer's lock before clearing its firing flag, so no
1338 * timer call will interfere.
1339 */
1340 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1341 int firing;
1342 spin_lock(&timer->it_lock);
1343 list_del_init(&timer->it.cpu.entry);
1344 firing = timer->it.cpu.firing;
1345 timer->it.cpu.firing = 0;
1346 /*
1347 * The firing flag is -1 if we collided with a reset
1348 * of the timer, which already reported this
1349 * almost-firing as an overrun. So don't generate an event.
1350 */
1351 if (likely(firing >= 0)) {
1352 cpu_timer_fire(timer);
1353 }
1354 spin_unlock(&timer->it_lock);
1355 }
1356}
1357
1358/*
1359 * Set one of the process-wide special case CPU timers.
1360 * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1361 * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1362 * absolute; non-null for ITIMER_*, where *newval is relative and we update
1363 * it to be absolute, *oldval is absolute and we update it to be relative.
1364 */
1365void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1366 cputime_t *newval, cputime_t *oldval)
1367{
1368 union cpu_time_count now;
1369 struct list_head *head;
1370
1371 BUG_ON(clock_idx == CPUCLOCK_SCHED);
1372 cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1373
1374 if (oldval) {
1375 if (!cputime_eq(*oldval, cputime_zero)) {
1376 if (cputime_le(*oldval, now.cpu)) {
1377 /* Just about to fire. */
1378 *oldval = jiffies_to_cputime(1);
1379 } else {
1380 *oldval = cputime_sub(*oldval, now.cpu);
1381 }
1382 }
1383
1384 if (cputime_eq(*newval, cputime_zero))
1385 return;
1386 *newval = cputime_add(*newval, now.cpu);
1387
1388 /*
1389 * If the RLIMIT_CPU timer will expire before the
1390 * ITIMER_PROF timer, we have nothing else to do.
1391 */
1392 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1393 < cputime_to_secs(*newval))
1394 return;
1395 }
1396
1397 /*
1398 * Check whether there are any process timers already set to fire
1399 * before this one. If so, we don't have anything more to do.
1400 */
1401 head = &tsk->signal->cpu_timers[clock_idx];
1402 if (list_empty(head) ||
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001403 cputime_ge(list_first_entry(head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 struct cpu_timer_list, entry)->expires.cpu,
1405 *newval)) {
1406 /*
1407 * Rejigger each thread's expiry time so that one will
1408 * notice before we hit the process-cumulative expiry time.
1409 */
1410 union cpu_time_count expires = { .sched = 0 };
1411 expires.cpu = *newval;
1412 process_timer_rebalance(tsk, clock_idx, expires, now);
1413 }
1414}
1415
Toyo Abee4b76552006-09-29 02:00:29 -07001416static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1417 struct timespec *rqtp, struct itimerspec *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 struct k_itimer timer;
1420 int error;
1421
1422 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 * Set up a temporary timer and then wait for it to go off.
1424 */
1425 memset(&timer, 0, sizeof timer);
1426 spin_lock_init(&timer.it_lock);
1427 timer.it_clock = which_clock;
1428 timer.it_overrun = -1;
1429 error = posix_cpu_timer_create(&timer);
1430 timer.it_process = current;
1431 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 static struct itimerspec zero_it;
Toyo Abee4b76552006-09-29 02:00:29 -07001433
1434 memset(it, 0, sizeof *it);
1435 it->it_value = *rqtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 spin_lock_irq(&timer.it_lock);
Toyo Abee4b76552006-09-29 02:00:29 -07001438 error = posix_cpu_timer_set(&timer, flags, it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (error) {
1440 spin_unlock_irq(&timer.it_lock);
1441 return error;
1442 }
1443
1444 while (!signal_pending(current)) {
1445 if (timer.it.cpu.expires.sched == 0) {
1446 /*
1447 * Our timer fired and was reset.
1448 */
1449 spin_unlock_irq(&timer.it_lock);
1450 return 0;
1451 }
1452
1453 /*
1454 * Block until cpu_timer_fire (or a signal) wakes us.
1455 */
1456 __set_current_state(TASK_INTERRUPTIBLE);
1457 spin_unlock_irq(&timer.it_lock);
1458 schedule();
1459 spin_lock_irq(&timer.it_lock);
1460 }
1461
1462 /*
1463 * We were interrupted by a signal.
1464 */
1465 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
Toyo Abee4b76552006-09-29 02:00:29 -07001466 posix_cpu_timer_set(&timer, 0, &zero_it, it);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 spin_unlock_irq(&timer.it_lock);
1468
Toyo Abee4b76552006-09-29 02:00:29 -07001469 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 /*
1471 * It actually did fire already.
1472 */
1473 return 0;
1474 }
1475
Toyo Abee4b76552006-09-29 02:00:29 -07001476 error = -ERESTART_RESTARTBLOCK;
1477 }
1478
1479 return error;
1480}
1481
1482int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1483 struct timespec *rqtp, struct timespec __user *rmtp)
1484{
1485 struct restart_block *restart_block =
1486 &current_thread_info()->restart_block;
1487 struct itimerspec it;
1488 int error;
1489
1490 /*
1491 * Diagnose required errors first.
1492 */
1493 if (CPUCLOCK_PERTHREAD(which_clock) &&
1494 (CPUCLOCK_PID(which_clock) == 0 ||
1495 CPUCLOCK_PID(which_clock) == current->pid))
1496 return -EINVAL;
1497
1498 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1499
1500 if (error == -ERESTART_RESTARTBLOCK) {
1501
1502 if (flags & TIMER_ABSTIME)
1503 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 /*
Toyo Abee4b76552006-09-29 02:00:29 -07001505 * Report back to the user the time still remaining.
1506 */
1507 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return -EFAULT;
1509
Toyo Abe1711ef32006-09-29 02:00:28 -07001510 restart_block->fn = posix_cpu_nsleep_restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 restart_block->arg0 = which_clock;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001512 restart_block->arg1 = (unsigned long) rmtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 restart_block->arg2 = rqtp->tv_sec;
1514 restart_block->arg3 = rqtp->tv_nsec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return error;
1517}
1518
Toyo Abe1711ef32006-09-29 02:00:28 -07001519long posix_cpu_nsleep_restart(struct restart_block *restart_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
1521 clockid_t which_clock = restart_block->arg0;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001522 struct timespec __user *rmtp;
1523 struct timespec t;
Toyo Abee4b76552006-09-29 02:00:29 -07001524 struct itimerspec it;
1525 int error;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001526
1527 rmtp = (struct timespec __user *) restart_block->arg1;
1528 t.tv_sec = restart_block->arg2;
1529 t.tv_nsec = restart_block->arg3;
1530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 restart_block->fn = do_no_restart_syscall;
Toyo Abee4b76552006-09-29 02:00:29 -07001532 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1533
1534 if (error == -ERESTART_RESTARTBLOCK) {
1535 /*
1536 * Report back to the user the time still remaining.
1537 */
1538 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1539 return -EFAULT;
1540
1541 restart_block->fn = posix_cpu_nsleep_restart;
1542 restart_block->arg0 = which_clock;
1543 restart_block->arg1 = (unsigned long) rmtp;
1544 restart_block->arg2 = t.tv_sec;
1545 restart_block->arg3 = t.tv_nsec;
1546 }
1547 return error;
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549}
1550
1551
1552#define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1553#define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1554
Thomas Gleixnera924b042006-01-09 20:52:27 -08001555static int process_cpu_clock_getres(const clockid_t which_clock,
1556 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1559}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001560static int process_cpu_clock_get(const clockid_t which_clock,
1561 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562{
1563 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1564}
1565static int process_cpu_timer_create(struct k_itimer *timer)
1566{
1567 timer->it_clock = PROCESS_CLOCK;
1568 return posix_cpu_timer_create(timer);
1569}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001570static int process_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner97735f22006-01-09 20:52:37 -08001571 struct timespec *rqtp,
1572 struct timespec __user *rmtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Thomas Gleixner97735f22006-01-09 20:52:37 -08001574 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575}
Toyo Abe1711ef32006-09-29 02:00:28 -07001576static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1577{
1578 return -EINVAL;
1579}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001580static int thread_cpu_clock_getres(const clockid_t which_clock,
1581 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582{
1583 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1584}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001585static int thread_cpu_clock_get(const clockid_t which_clock,
1586 struct timespec *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
1588 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1589}
1590static int thread_cpu_timer_create(struct k_itimer *timer)
1591{
1592 timer->it_clock = THREAD_CLOCK;
1593 return posix_cpu_timer_create(timer);
1594}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001595static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner97735f22006-01-09 20:52:37 -08001596 struct timespec *rqtp, struct timespec __user *rmtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 return -EINVAL;
1599}
Toyo Abe1711ef32006-09-29 02:00:28 -07001600static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1601{
1602 return -EINVAL;
1603}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605static __init int init_posix_cpu_timers(void)
1606{
1607 struct k_clock process = {
1608 .clock_getres = process_cpu_clock_getres,
1609 .clock_get = process_cpu_clock_get,
1610 .clock_set = do_posix_clock_nosettime,
1611 .timer_create = process_cpu_timer_create,
1612 .nsleep = process_cpu_nsleep,
Toyo Abe1711ef32006-09-29 02:00:28 -07001613 .nsleep_restart = process_cpu_nsleep_restart,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 };
1615 struct k_clock thread = {
1616 .clock_getres = thread_cpu_clock_getres,
1617 .clock_get = thread_cpu_clock_get,
1618 .clock_set = do_posix_clock_nosettime,
1619 .timer_create = thread_cpu_timer_create,
1620 .nsleep = thread_cpu_nsleep,
Toyo Abe1711ef32006-09-29 02:00:28 -07001621 .nsleep_restart = thread_cpu_nsleep_restart,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 };
1623
1624 register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1625 register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1626
1627 return 0;
1628}
1629__initcall(init_posix_cpu_timers);