]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * Implement CPU time clocks for the POSIX clock interface. | |
4 | */ | |
5 | ||
3f07c014 | 6 | #include <linux/sched/signal.h> |
32ef5517 | 7 | #include <linux/sched/cputime.h> |
1da177e4 | 8 | #include <linux/posix-timers.h> |
1da177e4 | 9 | #include <linux/errno.h> |
f8bd2258 | 10 | #include <linux/math64.h> |
7c0f6ba6 | 11 | #include <linux/uaccess.h> |
bb34d92f | 12 | #include <linux/kernel_stat.h> |
3f0a525e | 13 | #include <trace/events/timer.h> |
a8572160 FW |
14 | #include <linux/tick.h> |
15 | #include <linux/workqueue.h> | |
edbeda46 | 16 | #include <linux/compat.h> |
1da177e4 | 17 | |
bab0aae9 TG |
18 | #include "posix-timers.h" |
19 | ||
f37fb0aa TG |
20 | static void posix_cpu_timer_rearm(struct k_itimer *timer); |
21 | ||
f06febc9 | 22 | /* |
f55db609 SG |
23 | * Called after updating RLIMIT_CPU to run cpu timer and update |
24 | * tsk->signal->cputime_expires expiration cache if necessary. Needs | |
25 | * siglock protection since other code may update expiration cache as | |
26 | * well. | |
f06febc9 | 27 | */ |
5ab46b34 | 28 | void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new) |
f06febc9 | 29 | { |
858cf3a8 | 30 | u64 nsecs = rlim_new * NSEC_PER_SEC; |
f06febc9 | 31 | |
5ab46b34 | 32 | spin_lock_irq(&task->sighand->siglock); |
858cf3a8 | 33 | set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL); |
5ab46b34 | 34 | spin_unlock_irq(&task->sighand->siglock); |
f06febc9 FM |
35 | } |
36 | ||
a924b04d | 37 | static int check_clock(const clockid_t which_clock) |
1da177e4 LT |
38 | { |
39 | int error = 0; | |
40 | struct task_struct *p; | |
41 | const pid_t pid = CPUCLOCK_PID(which_clock); | |
42 | ||
43 | if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX) | |
44 | return -EINVAL; | |
45 | ||
46 | if (pid == 0) | |
47 | return 0; | |
48 | ||
c0deae8c | 49 | rcu_read_lock(); |
8dc86af0 | 50 | p = find_task_by_vpid(pid); |
bac0abd6 | 51 | if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ? |
c0deae8c | 52 | same_thread_group(p, current) : has_group_leader_pid(p))) { |
1da177e4 LT |
53 | error = -EINVAL; |
54 | } | |
c0deae8c | 55 | rcu_read_unlock(); |
1da177e4 LT |
56 | |
57 | return error; | |
58 | } | |
59 | ||
1da177e4 LT |
60 | /* |
61 | * Update expiry time from increment, and increase overrun count, | |
62 | * given the current clock sample. | |
63 | */ | |
ebd7e7fc | 64 | static void bump_cpu_timer(struct k_itimer *timer, u64 now) |
1da177e4 LT |
65 | { |
66 | int i; | |
ebd7e7fc | 67 | u64 delta, incr; |
1da177e4 | 68 | |
55ccb616 | 69 | if (timer->it.cpu.incr == 0) |
1da177e4 LT |
70 | return; |
71 | ||
55ccb616 FW |
72 | if (now < timer->it.cpu.expires) |
73 | return; | |
1da177e4 | 74 | |
55ccb616 FW |
75 | incr = timer->it.cpu.incr; |
76 | delta = now + incr - timer->it.cpu.expires; | |
1da177e4 | 77 | |
55ccb616 FW |
78 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
79 | for (i = 0; incr < delta - incr; i++) | |
80 | incr = incr << 1; | |
81 | ||
82 | for (; i >= 0; incr >>= 1, i--) { | |
83 | if (delta < incr) | |
84 | continue; | |
85 | ||
86 | timer->it.cpu.expires += incr; | |
4729028b | 87 | timer->it_overrun += 1LL << i; |
55ccb616 | 88 | delta -= incr; |
1da177e4 LT |
89 | } |
90 | } | |
91 | ||
555347f6 FW |
92 | /** |
93 | * task_cputime_zero - Check a task_cputime struct for all zero fields. | |
94 | * | |
95 | * @cputime: The struct to compare. | |
96 | * | |
97 | * Checks @cputime to see if all fields are zero. Returns true if all fields | |
98 | * are zero, false if any field is nonzero. | |
99 | */ | |
ebd7e7fc | 100 | static inline int task_cputime_zero(const struct task_cputime *cputime) |
555347f6 FW |
101 | { |
102 | if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime) | |
103 | return 1; | |
104 | return 0; | |
105 | } | |
106 | ||
ebd7e7fc | 107 | static inline u64 prof_ticks(struct task_struct *p) |
1da177e4 | 108 | { |
ebd7e7fc | 109 | u64 utime, stime; |
6fac4829 | 110 | |
ebd7e7fc | 111 | task_cputime(p, &utime, &stime); |
6fac4829 | 112 | |
ebd7e7fc | 113 | return utime + stime; |
1da177e4 | 114 | } |
ebd7e7fc | 115 | static inline u64 virt_ticks(struct task_struct *p) |
1da177e4 | 116 | { |
ebd7e7fc | 117 | u64 utime, stime; |
6fac4829 | 118 | |
ebd7e7fc | 119 | task_cputime(p, &utime, &stime); |
6fac4829 | 120 | |
ebd7e7fc | 121 | return utime; |
1da177e4 | 122 | } |
1da177e4 | 123 | |
bc2c8ea4 | 124 | static int |
d2e3e0ca | 125 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp) |
1da177e4 LT |
126 | { |
127 | int error = check_clock(which_clock); | |
128 | if (!error) { | |
129 | tp->tv_sec = 0; | |
130 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); | |
131 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { | |
132 | /* | |
133 | * If sched_clock is using a cycle counter, we | |
134 | * don't have any idea of its true resolution | |
135 | * exported, but it is much more than 1s/HZ. | |
136 | */ | |
137 | tp->tv_nsec = 1; | |
138 | } | |
139 | } | |
140 | return error; | |
141 | } | |
142 | ||
bc2c8ea4 | 143 | static int |
0fe6afe3 | 144 | posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp) |
1da177e4 LT |
145 | { |
146 | /* | |
147 | * You can never reset a CPU clock, but we check for other errors | |
148 | * in the call before failing with EPERM. | |
149 | */ | |
150 | int error = check_clock(which_clock); | |
151 | if (error == 0) { | |
152 | error = -EPERM; | |
153 | } | |
154 | return error; | |
155 | } | |
156 | ||
157 | ||
158 | /* | |
159 | * Sample a per-thread clock for the given task. | |
160 | */ | |
ebd7e7fc FW |
161 | static int cpu_clock_sample(const clockid_t which_clock, |
162 | struct task_struct *p, u64 *sample) | |
1da177e4 LT |
163 | { |
164 | switch (CPUCLOCK_WHICH(which_clock)) { | |
165 | default: | |
166 | return -EINVAL; | |
167 | case CPUCLOCK_PROF: | |
55ccb616 | 168 | *sample = prof_ticks(p); |
1da177e4 LT |
169 | break; |
170 | case CPUCLOCK_VIRT: | |
55ccb616 | 171 | *sample = virt_ticks(p); |
1da177e4 LT |
172 | break; |
173 | case CPUCLOCK_SCHED: | |
55ccb616 | 174 | *sample = task_sched_runtime(p); |
1da177e4 LT |
175 | break; |
176 | } | |
177 | return 0; | |
178 | } | |
179 | ||
1018016c JL |
180 | /* |
181 | * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg | |
182 | * to avoid race conditions with concurrent updates to cputime. | |
183 | */ | |
184 | static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime) | |
4da94d49 | 185 | { |
1018016c JL |
186 | u64 curr_cputime; |
187 | retry: | |
188 | curr_cputime = atomic64_read(cputime); | |
189 | if (sum_cputime > curr_cputime) { | |
190 | if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime) | |
191 | goto retry; | |
192 | } | |
193 | } | |
4da94d49 | 194 | |
ebd7e7fc | 195 | static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum) |
1018016c | 196 | { |
71107445 JL |
197 | __update_gt_cputime(&cputime_atomic->utime, sum->utime); |
198 | __update_gt_cputime(&cputime_atomic->stime, sum->stime); | |
199 | __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime); | |
1018016c | 200 | } |
4da94d49 | 201 | |
71107445 | 202 | /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */ |
ebd7e7fc | 203 | static inline void sample_cputime_atomic(struct task_cputime *times, |
71107445 | 204 | struct task_cputime_atomic *atomic_times) |
1018016c | 205 | { |
71107445 JL |
206 | times->utime = atomic64_read(&atomic_times->utime); |
207 | times->stime = atomic64_read(&atomic_times->stime); | |
208 | times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime); | |
4da94d49 PZ |
209 | } |
210 | ||
ebd7e7fc | 211 | void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times) |
4da94d49 PZ |
212 | { |
213 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
ebd7e7fc | 214 | struct task_cputime sum; |
4da94d49 | 215 | |
1018016c JL |
216 | /* Check if cputimer isn't running. This is accessed without locking. */ |
217 | if (!READ_ONCE(cputimer->running)) { | |
4da94d49 PZ |
218 | /* |
219 | * The POSIX timer interface allows for absolute time expiry | |
220 | * values through the TIMER_ABSTIME flag, therefore we have | |
1018016c | 221 | * to synchronize the timer to the clock every time we start it. |
4da94d49 | 222 | */ |
ebd7e7fc | 223 | thread_group_cputime(tsk, &sum); |
71107445 | 224 | update_gt_cputime(&cputimer->cputime_atomic, &sum); |
1018016c JL |
225 | |
226 | /* | |
227 | * We're setting cputimer->running without a lock. Ensure | |
228 | * this only gets written to in one operation. We set | |
229 | * running after update_gt_cputime() as a small optimization, | |
230 | * but barriers are not required because update_gt_cputime() | |
231 | * can handle concurrent updates. | |
232 | */ | |
d5c373eb | 233 | WRITE_ONCE(cputimer->running, true); |
1018016c | 234 | } |
71107445 | 235 | sample_cputime_atomic(times, &cputimer->cputime_atomic); |
4da94d49 PZ |
236 | } |
237 | ||
1da177e4 LT |
238 | /* |
239 | * Sample a process (thread group) clock for the given group_leader task. | |
e73d84e3 FW |
240 | * Must be called with task sighand lock held for safe while_each_thread() |
241 | * traversal. | |
1da177e4 | 242 | */ |
bb34d92f FM |
243 | static int cpu_clock_sample_group(const clockid_t which_clock, |
244 | struct task_struct *p, | |
ebd7e7fc | 245 | u64 *sample) |
1da177e4 | 246 | { |
ebd7e7fc | 247 | struct task_cputime cputime; |
f06febc9 | 248 | |
eccdaeaf | 249 | switch (CPUCLOCK_WHICH(which_clock)) { |
1da177e4 LT |
250 | default: |
251 | return -EINVAL; | |
252 | case CPUCLOCK_PROF: | |
ebd7e7fc FW |
253 | thread_group_cputime(p, &cputime); |
254 | *sample = cputime.utime + cputime.stime; | |
1da177e4 LT |
255 | break; |
256 | case CPUCLOCK_VIRT: | |
ebd7e7fc FW |
257 | thread_group_cputime(p, &cputime); |
258 | *sample = cputime.utime; | |
1da177e4 LT |
259 | break; |
260 | case CPUCLOCK_SCHED: | |
ebd7e7fc | 261 | thread_group_cputime(p, &cputime); |
55ccb616 | 262 | *sample = cputime.sum_exec_runtime; |
1da177e4 LT |
263 | break; |
264 | } | |
265 | return 0; | |
266 | } | |
267 | ||
33ab0fec FW |
268 | static int posix_cpu_clock_get_task(struct task_struct *tsk, |
269 | const clockid_t which_clock, | |
3c9c12f4 | 270 | struct timespec64 *tp) |
33ab0fec FW |
271 | { |
272 | int err = -EINVAL; | |
ebd7e7fc | 273 | u64 rtn; |
33ab0fec FW |
274 | |
275 | if (CPUCLOCK_PERTHREAD(which_clock)) { | |
276 | if (same_thread_group(tsk, current)) | |
277 | err = cpu_clock_sample(which_clock, tsk, &rtn); | |
278 | } else { | |
50875788 | 279 | if (tsk == current || thread_group_leader(tsk)) |
33ab0fec | 280 | err = cpu_clock_sample_group(which_clock, tsk, &rtn); |
33ab0fec FW |
281 | } |
282 | ||
283 | if (!err) | |
3c9c12f4 | 284 | *tp = ns_to_timespec64(rtn); |
33ab0fec FW |
285 | |
286 | return err; | |
287 | } | |
288 | ||
1da177e4 | 289 | |
3c9c12f4 | 290 | static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp) |
1da177e4 LT |
291 | { |
292 | const pid_t pid = CPUCLOCK_PID(which_clock); | |
33ab0fec | 293 | int err = -EINVAL; |
1da177e4 LT |
294 | |
295 | if (pid == 0) { | |
296 | /* | |
297 | * Special case constant value for our own clocks. | |
298 | * We don't have to do any lookup to find ourselves. | |
299 | */ | |
33ab0fec | 300 | err = posix_cpu_clock_get_task(current, which_clock, tp); |
1da177e4 LT |
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; | |
1f2ea083 | 307 | rcu_read_lock(); |
8dc86af0 | 308 | p = find_task_by_vpid(pid); |
33ab0fec FW |
309 | if (p) |
310 | err = posix_cpu_clock_get_task(p, which_clock, tp); | |
1f2ea083 | 311 | rcu_read_unlock(); |
1da177e4 LT |
312 | } |
313 | ||
33ab0fec | 314 | return err; |
1da177e4 LT |
315 | } |
316 | ||
1da177e4 LT |
317 | /* |
318 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. | |
ba5ea951 SG |
319 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
320 | * new timer already all-zeros initialized. | |
1da177e4 | 321 | */ |
bc2c8ea4 | 322 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
1da177e4 LT |
323 | { |
324 | int ret = 0; | |
325 | const pid_t pid = CPUCLOCK_PID(new_timer->it_clock); | |
326 | struct task_struct *p; | |
327 | ||
328 | if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX) | |
329 | return -EINVAL; | |
330 | ||
d97bb75d TG |
331 | new_timer->kclock = &clock_posix_cpu; |
332 | ||
1da177e4 | 333 | INIT_LIST_HEAD(&new_timer->it.cpu.entry); |
1da177e4 | 334 | |
c0deae8c | 335 | rcu_read_lock(); |
1da177e4 LT |
336 | if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) { |
337 | if (pid == 0) { | |
338 | p = current; | |
339 | } else { | |
8dc86af0 | 340 | p = find_task_by_vpid(pid); |
bac0abd6 | 341 | if (p && !same_thread_group(p, current)) |
1da177e4 LT |
342 | p = NULL; |
343 | } | |
344 | } else { | |
345 | if (pid == 0) { | |
346 | p = current->group_leader; | |
347 | } else { | |
8dc86af0 | 348 | p = find_task_by_vpid(pid); |
c0deae8c | 349 | if (p && !has_group_leader_pid(p)) |
1da177e4 LT |
350 | p = NULL; |
351 | } | |
352 | } | |
353 | new_timer->it.cpu.task = p; | |
354 | if (p) { | |
355 | get_task_struct(p); | |
356 | } else { | |
357 | ret = -EINVAL; | |
358 | } | |
c0deae8c | 359 | rcu_read_unlock(); |
1da177e4 LT |
360 | |
361 | return ret; | |
362 | } | |
363 | ||
364 | /* | |
365 | * Clean up a CPU-clock timer that is about to be destroyed. | |
366 | * This is called from timer deletion with the timer already locked. | |
367 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
368 | * and try again. (This happens when the timer is in the middle of firing.) | |
369 | */ | |
bc2c8ea4 | 370 | static int posix_cpu_timer_del(struct k_itimer *timer) |
1da177e4 | 371 | { |
108150ea | 372 | int ret = 0; |
3d7a1427 FW |
373 | unsigned long flags; |
374 | struct sighand_struct *sighand; | |
375 | struct task_struct *p = timer->it.cpu.task; | |
1da177e4 | 376 | |
2a3a4479 TG |
377 | if (WARN_ON_ONCE(!p)) |
378 | return -EINVAL; | |
108150ea | 379 | |
3d7a1427 FW |
380 | /* |
381 | * Protect against sighand release/switch in exit/exec and process/ | |
382 | * thread timer list entry concurrent read/writes. | |
383 | */ | |
384 | sighand = lock_task_sighand(p, &flags); | |
385 | if (unlikely(sighand == NULL)) { | |
a3222f88 FW |
386 | /* |
387 | * We raced with the reaping of the task. | |
388 | * The deletion should have cleared us off the list. | |
389 | */ | |
531f64fd | 390 | WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry)); |
a3222f88 | 391 | } else { |
a3222f88 FW |
392 | if (timer->it.cpu.firing) |
393 | ret = TIMER_RETRY; | |
394 | else | |
395 | list_del(&timer->it.cpu.entry); | |
3d7a1427 FW |
396 | |
397 | unlock_task_sighand(p, &flags); | |
1da177e4 | 398 | } |
a3222f88 FW |
399 | |
400 | if (!ret) | |
401 | put_task_struct(p); | |
1da177e4 | 402 | |
108150ea | 403 | return ret; |
1da177e4 LT |
404 | } |
405 | ||
af82eb3c | 406 | static void cleanup_timers_list(struct list_head *head) |
1a7fa510 FW |
407 | { |
408 | struct cpu_timer_list *timer, *next; | |
409 | ||
a0b2062b | 410 | list_for_each_entry_safe(timer, next, head, entry) |
1a7fa510 | 411 | list_del_init(&timer->entry); |
1a7fa510 FW |
412 | } |
413 | ||
1da177e4 LT |
414 | /* |
415 | * Clean out CPU timers still ticking when a thread exited. The task | |
416 | * pointer is cleared, and the expiry time is replaced with the residual | |
417 | * time for later timer_gettime calls to return. | |
418 | * This must be called with the siglock held. | |
419 | */ | |
af82eb3c | 420 | static void cleanup_timers(struct list_head *head) |
1da177e4 | 421 | { |
af82eb3c FW |
422 | cleanup_timers_list(head); |
423 | cleanup_timers_list(++head); | |
424 | cleanup_timers_list(++head); | |
1da177e4 LT |
425 | } |
426 | ||
427 | /* | |
428 | * These are both called with the siglock held, when the current thread | |
429 | * is being reaped. When the final (leader) thread in the group is reaped, | |
430 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. | |
431 | */ | |
432 | void posix_cpu_timers_exit(struct task_struct *tsk) | |
433 | { | |
af82eb3c | 434 | cleanup_timers(tsk->cpu_timers); |
1da177e4 LT |
435 | } |
436 | void posix_cpu_timers_exit_group(struct task_struct *tsk) | |
437 | { | |
af82eb3c | 438 | cleanup_timers(tsk->signal->cpu_timers); |
1da177e4 LT |
439 | } |
440 | ||
ebd7e7fc | 441 | static inline int expires_gt(u64 expires, u64 new_exp) |
d1e3b6d1 | 442 | { |
64861634 | 443 | return expires == 0 || expires > new_exp; |
d1e3b6d1 SG |
444 | } |
445 | ||
1da177e4 LT |
446 | /* |
447 | * Insert the timer on the appropriate list before any timers that | |
e73d84e3 | 448 | * expire later. This must be called with the sighand lock held. |
1da177e4 | 449 | */ |
5eb9aa64 | 450 | static void arm_timer(struct k_itimer *timer) |
1da177e4 LT |
451 | { |
452 | struct task_struct *p = timer->it.cpu.task; | |
453 | struct list_head *head, *listpos; | |
ebd7e7fc | 454 | struct task_cputime *cputime_expires; |
1da177e4 LT |
455 | struct cpu_timer_list *const nt = &timer->it.cpu; |
456 | struct cpu_timer_list *next; | |
1da177e4 | 457 | |
5eb9aa64 SG |
458 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
459 | head = p->cpu_timers; | |
460 | cputime_expires = &p->cputime_expires; | |
461 | } else { | |
462 | head = p->signal->cpu_timers; | |
463 | cputime_expires = &p->signal->cputime_expires; | |
464 | } | |
1da177e4 LT |
465 | head += CPUCLOCK_WHICH(timer->it_clock); |
466 | ||
1da177e4 | 467 | listpos = head; |
5eb9aa64 | 468 | list_for_each_entry(next, head, entry) { |
55ccb616 | 469 | if (nt->expires < next->expires) |
5eb9aa64 SG |
470 | break; |
471 | listpos = &next->entry; | |
1da177e4 LT |
472 | } |
473 | list_add(&nt->entry, listpos); | |
474 | ||
475 | if (listpos == head) { | |
ebd7e7fc | 476 | u64 exp = nt->expires; |
5eb9aa64 | 477 | |
1da177e4 | 478 | /* |
5eb9aa64 SG |
479 | * We are the new earliest-expiring POSIX 1.b timer, hence |
480 | * need to update expiration cache. Take into account that | |
481 | * for process timers we share expiration cache with itimers | |
482 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. | |
1da177e4 LT |
483 | */ |
484 | ||
5eb9aa64 SG |
485 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
486 | case CPUCLOCK_PROF: | |
ebd7e7fc FW |
487 | if (expires_gt(cputime_expires->prof_exp, exp)) |
488 | cputime_expires->prof_exp = exp; | |
5eb9aa64 SG |
489 | break; |
490 | case CPUCLOCK_VIRT: | |
ebd7e7fc FW |
491 | if (expires_gt(cputime_expires->virt_exp, exp)) |
492 | cputime_expires->virt_exp = exp; | |
5eb9aa64 SG |
493 | break; |
494 | case CPUCLOCK_SCHED: | |
ebd7e7fc | 495 | if (expires_gt(cputime_expires->sched_exp, exp)) |
55ccb616 | 496 | cputime_expires->sched_exp = exp; |
5eb9aa64 | 497 | break; |
1da177e4 | 498 | } |
b7878300 FW |
499 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
500 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); | |
501 | else | |
502 | tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 | 503 | } |
1da177e4 LT |
504 | } |
505 | ||
506 | /* | |
507 | * The timer is locked, fire it and arrange for its reload. | |
508 | */ | |
509 | static void cpu_timer_fire(struct k_itimer *timer) | |
510 | { | |
1f169f84 SG |
511 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
512 | /* | |
513 | * User don't want any signal. | |
514 | */ | |
55ccb616 | 515 | timer->it.cpu.expires = 0; |
1f169f84 | 516 | } else if (unlikely(timer->sigq == NULL)) { |
1da177e4 LT |
517 | /* |
518 | * This a special case for clock_nanosleep, | |
519 | * not a normal timer from sys_timer_create. | |
520 | */ | |
521 | wake_up_process(timer->it_process); | |
55ccb616 FW |
522 | timer->it.cpu.expires = 0; |
523 | } else if (timer->it.cpu.incr == 0) { | |
1da177e4 LT |
524 | /* |
525 | * One-shot timer. Clear it as soon as it's fired. | |
526 | */ | |
527 | posix_timer_event(timer, 0); | |
55ccb616 | 528 | timer->it.cpu.expires = 0; |
1da177e4 LT |
529 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
530 | /* | |
531 | * The signal did not get queued because the signal | |
532 | * was ignored, so we won't get any callback to | |
533 | * reload the timer. But we need to keep it | |
534 | * ticking in case the signal is deliverable next time. | |
535 | */ | |
f37fb0aa | 536 | posix_cpu_timer_rearm(timer); |
af888d67 | 537 | ++timer->it_requeue_pending; |
1da177e4 LT |
538 | } |
539 | } | |
540 | ||
3997ad31 PZ |
541 | /* |
542 | * Sample a process (thread group) timer for the given group_leader task. | |
e73d84e3 FW |
543 | * Must be called with task sighand lock held for safe while_each_thread() |
544 | * traversal. | |
3997ad31 PZ |
545 | */ |
546 | static int cpu_timer_sample_group(const clockid_t which_clock, | |
ebd7e7fc | 547 | struct task_struct *p, u64 *sample) |
3997ad31 | 548 | { |
ebd7e7fc | 549 | struct task_cputime cputime; |
3997ad31 PZ |
550 | |
551 | thread_group_cputimer(p, &cputime); | |
552 | switch (CPUCLOCK_WHICH(which_clock)) { | |
553 | default: | |
554 | return -EINVAL; | |
555 | case CPUCLOCK_PROF: | |
ebd7e7fc | 556 | *sample = cputime.utime + cputime.stime; |
3997ad31 PZ |
557 | break; |
558 | case CPUCLOCK_VIRT: | |
ebd7e7fc | 559 | *sample = cputime.utime; |
3997ad31 PZ |
560 | break; |
561 | case CPUCLOCK_SCHED: | |
23cfa361 | 562 | *sample = cputime.sum_exec_runtime; |
3997ad31 PZ |
563 | break; |
564 | } | |
565 | return 0; | |
566 | } | |
567 | ||
1da177e4 LT |
568 | /* |
569 | * Guts of sys_timer_settime for CPU timers. | |
570 | * This is called with the timer locked and interrupts disabled. | |
571 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
572 | * and try again. (This happens when the timer is in the middle of firing.) | |
573 | */ | |
e73d84e3 | 574 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
5f252b32 | 575 | struct itimerspec64 *new, struct itimerspec64 *old) |
1da177e4 | 576 | { |
e73d84e3 FW |
577 | unsigned long flags; |
578 | struct sighand_struct *sighand; | |
1da177e4 | 579 | struct task_struct *p = timer->it.cpu.task; |
ebd7e7fc | 580 | u64 old_expires, new_expires, old_incr, val; |
1da177e4 LT |
581 | int ret; |
582 | ||
2a3a4479 TG |
583 | if (WARN_ON_ONCE(!p)) |
584 | return -EINVAL; | |
1da177e4 | 585 | |
098b0e01 TG |
586 | /* |
587 | * Use the to_ktime conversion because that clamps the maximum | |
588 | * value to KTIME_MAX and avoid multiplication overflows. | |
589 | */ | |
590 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); | |
1da177e4 | 591 | |
1da177e4 | 592 | /* |
e73d84e3 FW |
593 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
594 | * and p->signal->cpu_timers read/write in arm_timer() | |
595 | */ | |
596 | sighand = lock_task_sighand(p, &flags); | |
597 | /* | |
598 | * If p has just been reaped, we can no | |
1da177e4 LT |
599 | * longer get any information about it at all. |
600 | */ | |
e73d84e3 | 601 | if (unlikely(sighand == NULL)) { |
1da177e4 LT |
602 | return -ESRCH; |
603 | } | |
604 | ||
605 | /* | |
606 | * Disarm any old timer after extracting its expiry time. | |
607 | */ | |
a6968220 | 608 | lockdep_assert_irqs_disabled(); |
a69ac4a7 ON |
609 | |
610 | ret = 0; | |
ae1a78ee | 611 | old_incr = timer->it.cpu.incr; |
1da177e4 | 612 | old_expires = timer->it.cpu.expires; |
a69ac4a7 ON |
613 | if (unlikely(timer->it.cpu.firing)) { |
614 | timer->it.cpu.firing = -1; | |
615 | ret = TIMER_RETRY; | |
616 | } else | |
617 | list_del_init(&timer->it.cpu.entry); | |
1da177e4 LT |
618 | |
619 | /* | |
620 | * We need to sample the current value to convert the new | |
621 | * value from to relative and absolute, and to convert the | |
622 | * old value from absolute to relative. To set a process | |
623 | * timer, we need a sample to balance the thread expiry | |
624 | * times (in arm_timer). With an absolute time, we must | |
625 | * check if it's already passed. In short, we need a sample. | |
626 | */ | |
627 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
628 | cpu_clock_sample(timer->it_clock, p, &val); | |
629 | } else { | |
3997ad31 | 630 | cpu_timer_sample_group(timer->it_clock, p, &val); |
1da177e4 LT |
631 | } |
632 | ||
633 | if (old) { | |
55ccb616 | 634 | if (old_expires == 0) { |
1da177e4 LT |
635 | old->it_value.tv_sec = 0; |
636 | old->it_value.tv_nsec = 0; | |
637 | } else { | |
638 | /* | |
639 | * Update the timer in case it has | |
640 | * overrun already. If it has, | |
641 | * we'll report it as having overrun | |
642 | * and with the next reloaded timer | |
643 | * already ticking, though we are | |
644 | * swallowing that pending | |
645 | * notification here to install the | |
646 | * new setting. | |
647 | */ | |
648 | bump_cpu_timer(timer, val); | |
55ccb616 FW |
649 | if (val < timer->it.cpu.expires) { |
650 | old_expires = timer->it.cpu.expires - val; | |
5f252b32 | 651 | old->it_value = ns_to_timespec64(old_expires); |
1da177e4 LT |
652 | } else { |
653 | old->it_value.tv_nsec = 1; | |
654 | old->it_value.tv_sec = 0; | |
655 | } | |
656 | } | |
657 | } | |
658 | ||
a69ac4a7 | 659 | if (unlikely(ret)) { |
1da177e4 LT |
660 | /* |
661 | * We are colliding with the timer actually firing. | |
662 | * Punt after filling in the timer's old value, and | |
663 | * disable this firing since we are already reporting | |
664 | * it as an overrun (thanks to bump_cpu_timer above). | |
665 | */ | |
e73d84e3 | 666 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
667 | goto out; |
668 | } | |
669 | ||
e73d84e3 | 670 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
55ccb616 | 671 | new_expires += val; |
1da177e4 LT |
672 | } |
673 | ||
674 | /* | |
675 | * Install the new expiry time (or zero). | |
676 | * For a timer with no notification action, we don't actually | |
677 | * arm the timer (we'll just fake it for timer_gettime). | |
678 | */ | |
679 | timer->it.cpu.expires = new_expires; | |
55ccb616 | 680 | if (new_expires != 0 && val < new_expires) { |
5eb9aa64 | 681 | arm_timer(timer); |
1da177e4 LT |
682 | } |
683 | ||
e73d84e3 | 684 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
685 | /* |
686 | * Install the new reload setting, and | |
687 | * set up the signal and overrun bookkeeping. | |
688 | */ | |
5f252b32 | 689 | timer->it.cpu.incr = timespec64_to_ns(&new->it_interval); |
2ff23782 | 690 | timer->it_interval = ns_to_ktime(timer->it.cpu.incr); |
1da177e4 LT |
691 | |
692 | /* | |
693 | * This acts as a modification timestamp for the timer, | |
694 | * so any automatic reload attempt will punt on seeing | |
695 | * that we have reset the timer manually. | |
696 | */ | |
697 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & | |
698 | ~REQUEUE_PENDING; | |
699 | timer->it_overrun_last = 0; | |
700 | timer->it_overrun = -1; | |
701 | ||
55ccb616 | 702 | if (new_expires != 0 && !(val < new_expires)) { |
1da177e4 LT |
703 | /* |
704 | * The designated time already passed, so we notify | |
705 | * immediately, even if the thread never runs to | |
706 | * accumulate more time on this clock. | |
707 | */ | |
708 | cpu_timer_fire(timer); | |
709 | } | |
710 | ||
711 | ret = 0; | |
712 | out: | |
ebd7e7fc | 713 | if (old) |
5f252b32 | 714 | old->it_interval = ns_to_timespec64(old_incr); |
b7878300 | 715 | |
1da177e4 LT |
716 | return ret; |
717 | } | |
718 | ||
5f252b32 | 719 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
1da177e4 | 720 | { |
1da177e4 | 721 | struct task_struct *p = timer->it.cpu.task; |
2a3a4479 | 722 | u64 now; |
1da177e4 | 723 | |
2a3a4479 TG |
724 | if (WARN_ON_ONCE(!p)) |
725 | return; | |
a3222f88 | 726 | |
1da177e4 LT |
727 | /* |
728 | * Easy part: convert the reload time. | |
729 | */ | |
5f252b32 | 730 | itp->it_interval = ns_to_timespec64(timer->it.cpu.incr); |
1da177e4 | 731 | |
eabdec04 | 732 | if (!timer->it.cpu.expires) |
1da177e4 | 733 | return; |
1da177e4 | 734 | |
1da177e4 LT |
735 | /* |
736 | * Sample the clock to take the difference with the expiry time. | |
737 | */ | |
738 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
739 | cpu_clock_sample(timer->it_clock, p, &now); | |
1da177e4 | 740 | } else { |
e73d84e3 FW |
741 | struct sighand_struct *sighand; |
742 | unsigned long flags; | |
743 | ||
744 | /* | |
745 | * Protect against sighand release/switch in exit/exec and | |
746 | * also make timer sampling safe if it ends up calling | |
ebd7e7fc | 747 | * thread_group_cputime(). |
e73d84e3 FW |
748 | */ |
749 | sighand = lock_task_sighand(p, &flags); | |
750 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
751 | /* |
752 | * The process has been reaped. | |
753 | * We can't even collect a sample any more. | |
754 | * Call the timer disarmed, nothing else to do. | |
755 | */ | |
55ccb616 | 756 | timer->it.cpu.expires = 0; |
2c13ce8f | 757 | return; |
1da177e4 | 758 | } else { |
3997ad31 | 759 | cpu_timer_sample_group(timer->it_clock, p, &now); |
e73d84e3 | 760 | unlock_task_sighand(p, &flags); |
1da177e4 | 761 | } |
1da177e4 LT |
762 | } |
763 | ||
55ccb616 | 764 | if (now < timer->it.cpu.expires) { |
5f252b32 | 765 | itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now); |
1da177e4 LT |
766 | } else { |
767 | /* | |
768 | * The timer should have expired already, but the firing | |
769 | * hasn't taken place yet. Say it's just about to expire. | |
770 | */ | |
771 | itp->it_value.tv_nsec = 1; | |
772 | itp->it_value.tv_sec = 0; | |
773 | } | |
774 | } | |
775 | ||
2473f3e7 FW |
776 | static unsigned long long |
777 | check_timers_list(struct list_head *timers, | |
778 | struct list_head *firing, | |
779 | unsigned long long curr) | |
780 | { | |
781 | int maxfire = 20; | |
782 | ||
783 | while (!list_empty(timers)) { | |
784 | struct cpu_timer_list *t; | |
785 | ||
786 | t = list_first_entry(timers, struct cpu_timer_list, entry); | |
787 | ||
788 | if (!--maxfire || curr < t->expires) | |
789 | return t->expires; | |
790 | ||
791 | t->firing = 1; | |
792 | list_move_tail(&t->entry, firing); | |
793 | } | |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
1da177e4 LT |
798 | /* |
799 | * Check for any per-thread CPU timers that have fired and move them off | |
800 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the | |
801 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. | |
802 | */ | |
803 | static void check_thread_timers(struct task_struct *tsk, | |
804 | struct list_head *firing) | |
805 | { | |
806 | struct list_head *timers = tsk->cpu_timers; | |
ebd7e7fc FW |
807 | struct task_cputime *tsk_expires = &tsk->cputime_expires; |
808 | u64 expires; | |
d4bb5274 | 809 | unsigned long soft; |
1da177e4 | 810 | |
934715a1 JL |
811 | /* |
812 | * If cputime_expires is zero, then there are no active | |
813 | * per thread CPU timers. | |
814 | */ | |
815 | if (task_cputime_zero(&tsk->cputime_expires)) | |
816 | return; | |
817 | ||
2473f3e7 | 818 | expires = check_timers_list(timers, firing, prof_ticks(tsk)); |
ebd7e7fc | 819 | tsk_expires->prof_exp = expires; |
1da177e4 | 820 | |
2473f3e7 | 821 | expires = check_timers_list(++timers, firing, virt_ticks(tsk)); |
ebd7e7fc | 822 | tsk_expires->virt_exp = expires; |
1da177e4 | 823 | |
2473f3e7 FW |
824 | tsk_expires->sched_exp = check_timers_list(++timers, firing, |
825 | tsk->se.sum_exec_runtime); | |
78f2c7db PZ |
826 | |
827 | /* | |
828 | * Check for the special case thread timers. | |
829 | */ | |
3cf29496 | 830 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
d4bb5274 | 831 | if (soft != RLIM_INFINITY) { |
3cf29496 | 832 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
78f2c7db | 833 | |
5a52dd50 PZ |
834 | if (hard != RLIM_INFINITY && |
835 | tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) { | |
78f2c7db PZ |
836 | /* |
837 | * At the hard limit, we just die. | |
838 | * No need to calculate anything else now. | |
839 | */ | |
43fe8b8e TG |
840 | if (print_fatal_signals) { |
841 | pr_info("CPU Watchdog Timeout (hard): %s[%d]\n", | |
842 | tsk->comm, task_pid_nr(tsk)); | |
843 | } | |
78f2c7db PZ |
844 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
845 | return; | |
846 | } | |
d4bb5274 | 847 | if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) { |
78f2c7db PZ |
848 | /* |
849 | * At the soft limit, send a SIGXCPU every second. | |
850 | */ | |
d4bb5274 JS |
851 | if (soft < hard) { |
852 | soft += USEC_PER_SEC; | |
3cf29496 KO |
853 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = |
854 | soft; | |
78f2c7db | 855 | } |
43fe8b8e TG |
856 | if (print_fatal_signals) { |
857 | pr_info("RT Watchdog Timeout (soft): %s[%d]\n", | |
858 | tsk->comm, task_pid_nr(tsk)); | |
859 | } | |
78f2c7db PZ |
860 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
861 | } | |
862 | } | |
b7878300 FW |
863 | if (task_cputime_zero(tsk_expires)) |
864 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
865 | } |
866 | ||
1018016c | 867 | static inline void stop_process_timers(struct signal_struct *sig) |
3fccfd67 | 868 | { |
15365c10 | 869 | struct thread_group_cputimer *cputimer = &sig->cputimer; |
3fccfd67 | 870 | |
1018016c | 871 | /* Turn off cputimer->running. This is done without locking. */ |
d5c373eb | 872 | WRITE_ONCE(cputimer->running, false); |
b7878300 | 873 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
3fccfd67 PZ |
874 | } |
875 | ||
42c4ab41 | 876 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
ebd7e7fc | 877 | u64 *expires, u64 cur_time, int signo) |
42c4ab41 | 878 | { |
64861634 | 879 | if (!it->expires) |
42c4ab41 SG |
880 | return; |
881 | ||
858cf3a8 FW |
882 | if (cur_time >= it->expires) { |
883 | if (it->incr) | |
64861634 | 884 | it->expires += it->incr; |
858cf3a8 | 885 | else |
64861634 | 886 | it->expires = 0; |
42c4ab41 | 887 | |
3f0a525e XG |
888 | trace_itimer_expire(signo == SIGPROF ? |
889 | ITIMER_PROF : ITIMER_VIRTUAL, | |
890 | tsk->signal->leader_pid, cur_time); | |
42c4ab41 SG |
891 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
892 | } | |
893 | ||
858cf3a8 FW |
894 | if (it->expires && (!*expires || it->expires < *expires)) |
895 | *expires = it->expires; | |
42c4ab41 SG |
896 | } |
897 | ||
1da177e4 LT |
898 | /* |
899 | * Check for any per-thread CPU timers that have fired and move them | |
900 | * off the tsk->*_timers list onto the firing list. Per-thread timers | |
901 | * have already been taken off. | |
902 | */ | |
903 | static void check_process_timers(struct task_struct *tsk, | |
904 | struct list_head *firing) | |
905 | { | |
906 | struct signal_struct *const sig = tsk->signal; | |
ebd7e7fc FW |
907 | u64 utime, ptime, virt_expires, prof_expires; |
908 | u64 sum_sched_runtime, sched_expires; | |
1da177e4 | 909 | struct list_head *timers = sig->cpu_timers; |
ebd7e7fc | 910 | struct task_cputime cputime; |
d4bb5274 | 911 | unsigned long soft; |
1da177e4 | 912 | |
934715a1 JL |
913 | /* |
914 | * If cputimer is not running, then there are no active | |
915 | * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU). | |
916 | */ | |
917 | if (!READ_ONCE(tsk->signal->cputimer.running)) | |
918 | return; | |
919 | ||
c8d75aa4 JL |
920 | /* |
921 | * Signify that a thread is checking for process timers. | |
922 | * Write access to this field is protected by the sighand lock. | |
923 | */ | |
924 | sig->cputimer.checking_timer = true; | |
925 | ||
1da177e4 LT |
926 | /* |
927 | * Collect the current process totals. | |
928 | */ | |
4cd4c1b4 | 929 | thread_group_cputimer(tsk, &cputime); |
ebd7e7fc FW |
930 | utime = cputime.utime; |
931 | ptime = utime + cputime.stime; | |
f06febc9 | 932 | sum_sched_runtime = cputime.sum_exec_runtime; |
1da177e4 | 933 | |
2473f3e7 FW |
934 | prof_expires = check_timers_list(timers, firing, ptime); |
935 | virt_expires = check_timers_list(++timers, firing, utime); | |
936 | sched_expires = check_timers_list(++timers, firing, sum_sched_runtime); | |
1da177e4 LT |
937 | |
938 | /* | |
939 | * Check for the special case process timers. | |
940 | */ | |
42c4ab41 SG |
941 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime, |
942 | SIGPROF); | |
943 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime, | |
944 | SIGVTALRM); | |
3cf29496 | 945 | soft = task_rlimit(tsk, RLIMIT_CPU); |
d4bb5274 | 946 | if (soft != RLIM_INFINITY) { |
ebd7e7fc | 947 | unsigned long psecs = div_u64(ptime, NSEC_PER_SEC); |
3cf29496 | 948 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
ebd7e7fc | 949 | u64 x; |
d4bb5274 | 950 | if (psecs >= hard) { |
1da177e4 LT |
951 | /* |
952 | * At the hard limit, we just die. | |
953 | * No need to calculate anything else now. | |
954 | */ | |
43fe8b8e TG |
955 | if (print_fatal_signals) { |
956 | pr_info("RT Watchdog Timeout (hard): %s[%d]\n", | |
957 | tsk->comm, task_pid_nr(tsk)); | |
958 | } | |
1da177e4 LT |
959 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
960 | return; | |
961 | } | |
d4bb5274 | 962 | if (psecs >= soft) { |
1da177e4 LT |
963 | /* |
964 | * At the soft limit, send a SIGXCPU every second. | |
965 | */ | |
43fe8b8e TG |
966 | if (print_fatal_signals) { |
967 | pr_info("CPU Watchdog Timeout (soft): %s[%d]\n", | |
968 | tsk->comm, task_pid_nr(tsk)); | |
969 | } | |
1da177e4 | 970 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
d4bb5274 JS |
971 | if (soft < hard) { |
972 | soft++; | |
973 | sig->rlim[RLIMIT_CPU].rlim_cur = soft; | |
1da177e4 LT |
974 | } |
975 | } | |
ebd7e7fc FW |
976 | x = soft * NSEC_PER_SEC; |
977 | if (!prof_expires || x < prof_expires) | |
1da177e4 | 978 | prof_expires = x; |
1da177e4 LT |
979 | } |
980 | ||
ebd7e7fc FW |
981 | sig->cputime_expires.prof_exp = prof_expires; |
982 | sig->cputime_expires.virt_exp = virt_expires; | |
29f87b79 SG |
983 | sig->cputime_expires.sched_exp = sched_expires; |
984 | if (task_cputime_zero(&sig->cputime_expires)) | |
985 | stop_process_timers(sig); | |
c8d75aa4 JL |
986 | |
987 | sig->cputimer.checking_timer = false; | |
1da177e4 LT |
988 | } |
989 | ||
990 | /* | |
96fe3b07 | 991 | * This is called from the signal code (via posixtimer_rearm) |
1da177e4 LT |
992 | * when the last timer signal was delivered and we have to reload the timer. |
993 | */ | |
f37fb0aa | 994 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
1da177e4 | 995 | { |
2a3a4479 | 996 | struct task_struct *p = timer->it.cpu.task; |
e73d84e3 FW |
997 | struct sighand_struct *sighand; |
998 | unsigned long flags; | |
ebd7e7fc | 999 | u64 now; |
1da177e4 | 1000 | |
2a3a4479 TG |
1001 | if (WARN_ON_ONCE(!p)) |
1002 | return; | |
1da177e4 LT |
1003 | |
1004 | /* | |
1005 | * Fetch the current sample and update the timer's expiry time. | |
1006 | */ | |
1007 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
1008 | cpu_clock_sample(timer->it_clock, p, &now); | |
1009 | bump_cpu_timer(timer, now); | |
724a3713 | 1010 | if (unlikely(p->exit_state)) |
af888d67 | 1011 | return; |
724a3713 | 1012 | |
e73d84e3 FW |
1013 | /* Protect timer list r/w in arm_timer() */ |
1014 | sighand = lock_task_sighand(p, &flags); | |
1015 | if (!sighand) | |
af888d67 | 1016 | return; |
1da177e4 | 1017 | } else { |
e73d84e3 FW |
1018 | /* |
1019 | * Protect arm_timer() and timer sampling in case of call to | |
ebd7e7fc | 1020 | * thread_group_cputime(). |
e73d84e3 FW |
1021 | */ |
1022 | sighand = lock_task_sighand(p, &flags); | |
1023 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
1024 | /* |
1025 | * The process has been reaped. | |
1026 | * We can't even collect a sample any more. | |
1027 | */ | |
55ccb616 | 1028 | timer->it.cpu.expires = 0; |
af888d67 | 1029 | return; |
1da177e4 | 1030 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
af888d67 TG |
1031 | /* If the process is dying, no need to rearm */ |
1032 | goto unlock; | |
1da177e4 | 1033 | } |
3997ad31 | 1034 | cpu_timer_sample_group(timer->it_clock, p, &now); |
1da177e4 | 1035 | bump_cpu_timer(timer, now); |
e73d84e3 | 1036 | /* Leave the sighand locked for the call below. */ |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | /* | |
1040 | * Now re-arm for the new expiry time. | |
1041 | */ | |
a6968220 | 1042 | lockdep_assert_irqs_disabled(); |
5eb9aa64 | 1043 | arm_timer(timer); |
af888d67 | 1044 | unlock: |
e73d84e3 | 1045 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
1046 | } |
1047 | ||
f06febc9 FM |
1048 | /** |
1049 | * task_cputime_expired - Compare two task_cputime entities. | |
1050 | * | |
1051 | * @sample: The task_cputime structure to be checked for expiration. | |
1052 | * @expires: Expiration times, against which @sample will be checked. | |
1053 | * | |
1054 | * Checks @sample against @expires to see if any field of @sample has expired. | |
1055 | * Returns true if any field of the former is greater than the corresponding | |
1056 | * field of the latter if the latter field is set. Otherwise returns false. | |
1057 | */ | |
ebd7e7fc FW |
1058 | static inline int task_cputime_expired(const struct task_cputime *sample, |
1059 | const struct task_cputime *expires) | |
f06febc9 | 1060 | { |
64861634 | 1061 | if (expires->utime && sample->utime >= expires->utime) |
f06febc9 | 1062 | return 1; |
64861634 | 1063 | if (expires->stime && sample->utime + sample->stime >= expires->stime) |
f06febc9 FM |
1064 | return 1; |
1065 | if (expires->sum_exec_runtime != 0 && | |
1066 | sample->sum_exec_runtime >= expires->sum_exec_runtime) | |
1067 | return 1; | |
1068 | return 0; | |
1069 | } | |
1070 | ||
1071 | /** | |
1072 | * fastpath_timer_check - POSIX CPU timers fast path. | |
1073 | * | |
1074 | * @tsk: The task (thread) being checked. | |
f06febc9 | 1075 | * |
bb34d92f FM |
1076 | * Check the task and thread group timers. If both are zero (there are no |
1077 | * timers set) return false. Otherwise snapshot the task and thread group | |
1078 | * timers and compare them with the corresponding expiration times. Return | |
1079 | * true if a timer has expired, else return false. | |
f06febc9 | 1080 | */ |
bb34d92f | 1081 | static inline int fastpath_timer_check(struct task_struct *tsk) |
f06febc9 | 1082 | { |
ad133ba3 | 1083 | struct signal_struct *sig; |
bb34d92f | 1084 | |
bb34d92f | 1085 | if (!task_cputime_zero(&tsk->cputime_expires)) { |
ebd7e7fc | 1086 | struct task_cputime task_sample; |
bb34d92f | 1087 | |
ebd7e7fc | 1088 | task_cputime(tsk, &task_sample.utime, &task_sample.stime); |
7c177d99 | 1089 | task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime; |
bb34d92f FM |
1090 | if (task_cputime_expired(&task_sample, &tsk->cputime_expires)) |
1091 | return 1; | |
1092 | } | |
ad133ba3 ON |
1093 | |
1094 | sig = tsk->signal; | |
c8d75aa4 JL |
1095 | /* |
1096 | * Check if thread group timers expired when the cputimer is | |
1097 | * running and no other thread in the group is already checking | |
1098 | * for thread group cputimers. These fields are read without the | |
1099 | * sighand lock. However, this is fine because this is meant to | |
1100 | * be a fastpath heuristic to determine whether we should try to | |
1101 | * acquire the sighand lock to check/handle timers. | |
1102 | * | |
1103 | * In the worst case scenario, if 'running' or 'checking_timer' gets | |
1104 | * set but the current thread doesn't see the change yet, we'll wait | |
1105 | * until the next thread in the group gets a scheduler interrupt to | |
1106 | * handle the timer. This isn't an issue in practice because these | |
1107 | * types of delays with signals actually getting sent are expected. | |
1108 | */ | |
1109 | if (READ_ONCE(sig->cputimer.running) && | |
1110 | !READ_ONCE(sig->cputimer.checking_timer)) { | |
ebd7e7fc | 1111 | struct task_cputime group_sample; |
bb34d92f | 1112 | |
71107445 | 1113 | sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic); |
8d1f431c | 1114 | |
bb34d92f FM |
1115 | if (task_cputime_expired(&group_sample, &sig->cputime_expires)) |
1116 | return 1; | |
1117 | } | |
37bebc70 | 1118 | |
f55db609 | 1119 | return 0; |
f06febc9 FM |
1120 | } |
1121 | ||
1da177e4 LT |
1122 | /* |
1123 | * This is called from the timer interrupt handler. The irq handler has | |
1124 | * already updated our counts. We need to check if any timers fire now. | |
1125 | * Interrupts are disabled. | |
1126 | */ | |
1127 | void run_posix_cpu_timers(struct task_struct *tsk) | |
1128 | { | |
1129 | LIST_HEAD(firing); | |
1130 | struct k_itimer *timer, *next; | |
0bdd2ed4 | 1131 | unsigned long flags; |
1da177e4 | 1132 | |
a6968220 | 1133 | lockdep_assert_irqs_disabled(); |
1da177e4 | 1134 | |
1da177e4 | 1135 | /* |
f06febc9 | 1136 | * The fast path checks that there are no expired thread or thread |
bb34d92f | 1137 | * group timers. If that's so, just return. |
1da177e4 | 1138 | */ |
bb34d92f | 1139 | if (!fastpath_timer_check(tsk)) |
f06febc9 | 1140 | return; |
5ce73a4a | 1141 | |
0bdd2ed4 ON |
1142 | if (!lock_task_sighand(tsk, &flags)) |
1143 | return; | |
bb34d92f FM |
1144 | /* |
1145 | * Here we take off tsk->signal->cpu_timers[N] and | |
1146 | * tsk->cpu_timers[N] all the timers that are firing, and | |
1147 | * put them on the firing list. | |
1148 | */ | |
1149 | check_thread_timers(tsk, &firing); | |
934715a1 JL |
1150 | |
1151 | check_process_timers(tsk, &firing); | |
1da177e4 | 1152 | |
bb34d92f FM |
1153 | /* |
1154 | * We must release these locks before taking any timer's lock. | |
1155 | * There is a potential race with timer deletion here, as the | |
1156 | * siglock now protects our private firing list. We have set | |
1157 | * the firing flag in each timer, so that a deletion attempt | |
1158 | * that gets the timer lock before we do will give it up and | |
1159 | * spin until we've taken care of that timer below. | |
1160 | */ | |
0bdd2ed4 | 1161 | unlock_task_sighand(tsk, &flags); |
1da177e4 LT |
1162 | |
1163 | /* | |
1164 | * Now that all the timers on our list have the firing flag, | |
25985edc | 1165 | * no one will touch their list entries but us. We'll take |
1da177e4 LT |
1166 | * each timer's lock before clearing its firing flag, so no |
1167 | * timer call will interfere. | |
1168 | */ | |
1169 | list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) { | |
6e85c5ba HS |
1170 | int cpu_firing; |
1171 | ||
1da177e4 LT |
1172 | spin_lock(&timer->it_lock); |
1173 | list_del_init(&timer->it.cpu.entry); | |
6e85c5ba | 1174 | cpu_firing = timer->it.cpu.firing; |
1da177e4 LT |
1175 | timer->it.cpu.firing = 0; |
1176 | /* | |
1177 | * The firing flag is -1 if we collided with a reset | |
1178 | * of the timer, which already reported this | |
1179 | * almost-firing as an overrun. So don't generate an event. | |
1180 | */ | |
6e85c5ba | 1181 | if (likely(cpu_firing >= 0)) |
1da177e4 | 1182 | cpu_timer_fire(timer); |
1da177e4 LT |
1183 | spin_unlock(&timer->it_lock); |
1184 | } | |
1185 | } | |
1186 | ||
1187 | /* | |
f55db609 | 1188 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
f06febc9 | 1189 | * The tsk->sighand->siglock must be held by the caller. |
1da177e4 LT |
1190 | */ |
1191 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx, | |
858cf3a8 | 1192 | u64 *newval, u64 *oldval) |
1da177e4 | 1193 | { |
858cf3a8 | 1194 | u64 now; |
1da177e4 | 1195 | |
2a3a4479 TG |
1196 | if (WARN_ON_ONCE(clock_idx >= CPUCLOCK_SCHED)) |
1197 | return; | |
1198 | ||
4cd4c1b4 | 1199 | cpu_timer_sample_group(clock_idx, tsk, &now); |
1da177e4 LT |
1200 | |
1201 | if (oldval) { | |
f55db609 SG |
1202 | /* |
1203 | * We are setting itimer. The *oldval is absolute and we update | |
1204 | * it to be relative, *newval argument is relative and we update | |
1205 | * it to be absolute. | |
1206 | */ | |
64861634 | 1207 | if (*oldval) { |
858cf3a8 | 1208 | if (*oldval <= now) { |
1da177e4 | 1209 | /* Just about to fire. */ |
858cf3a8 | 1210 | *oldval = TICK_NSEC; |
1da177e4 | 1211 | } else { |
858cf3a8 | 1212 | *oldval -= now; |
1da177e4 LT |
1213 | } |
1214 | } | |
1215 | ||
64861634 | 1216 | if (!*newval) |
b7878300 | 1217 | return; |
858cf3a8 | 1218 | *newval += now; |
1da177e4 LT |
1219 | } |
1220 | ||
1221 | /* | |
f55db609 SG |
1222 | * Update expiration cache if we are the earliest timer, or eventually |
1223 | * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire. | |
1da177e4 | 1224 | */ |
f55db609 SG |
1225 | switch (clock_idx) { |
1226 | case CPUCLOCK_PROF: | |
858cf3a8 FW |
1227 | if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval)) |
1228 | tsk->signal->cputime_expires.prof_exp = *newval; | |
f55db609 SG |
1229 | break; |
1230 | case CPUCLOCK_VIRT: | |
858cf3a8 FW |
1231 | if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval)) |
1232 | tsk->signal->cputime_expires.virt_exp = *newval; | |
f55db609 | 1233 | break; |
1da177e4 | 1234 | } |
b7878300 FW |
1235 | |
1236 | tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
1237 | } |
1238 | ||
e4b76555 | 1239 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
343d8fc2 | 1240 | const struct timespec64 *rqtp) |
1da177e4 | 1241 | { |
86a9c446 | 1242 | struct itimerspec64 it; |
343d8fc2 TG |
1243 | struct k_itimer timer; |
1244 | u64 expires; | |
1da177e4 LT |
1245 | int error; |
1246 | ||
1da177e4 LT |
1247 | /* |
1248 | * Set up a temporary timer and then wait for it to go off. | |
1249 | */ | |
1250 | memset(&timer, 0, sizeof timer); | |
1251 | spin_lock_init(&timer.it_lock); | |
1252 | timer.it_clock = which_clock; | |
1253 | timer.it_overrun = -1; | |
1254 | error = posix_cpu_timer_create(&timer); | |
1255 | timer.it_process = current; | |
1256 | if (!error) { | |
5f252b32 | 1257 | static struct itimerspec64 zero_it; |
edbeda46 | 1258 | struct restart_block *restart; |
e4b76555 | 1259 | |
edbeda46 | 1260 | memset(&it, 0, sizeof(it)); |
86a9c446 | 1261 | it.it_value = *rqtp; |
1da177e4 LT |
1262 | |
1263 | spin_lock_irq(&timer.it_lock); | |
86a9c446 | 1264 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
1da177e4 LT |
1265 | if (error) { |
1266 | spin_unlock_irq(&timer.it_lock); | |
1267 | return error; | |
1268 | } | |
1269 | ||
1270 | while (!signal_pending(current)) { | |
55ccb616 | 1271 | if (timer.it.cpu.expires == 0) { |
1da177e4 | 1272 | /* |
e6c42c29 SG |
1273 | * Our timer fired and was reset, below |
1274 | * deletion can not fail. | |
1da177e4 | 1275 | */ |
e6c42c29 | 1276 | posix_cpu_timer_del(&timer); |
1da177e4 LT |
1277 | spin_unlock_irq(&timer.it_lock); |
1278 | return 0; | |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * Block until cpu_timer_fire (or a signal) wakes us. | |
1283 | */ | |
1284 | __set_current_state(TASK_INTERRUPTIBLE); | |
1285 | spin_unlock_irq(&timer.it_lock); | |
1286 | schedule(); | |
1287 | spin_lock_irq(&timer.it_lock); | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * We were interrupted by a signal. | |
1292 | */ | |
343d8fc2 | 1293 | expires = timer.it.cpu.expires; |
86a9c446 | 1294 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
e6c42c29 SG |
1295 | if (!error) { |
1296 | /* | |
1297 | * Timer is now unarmed, deletion can not fail. | |
1298 | */ | |
1299 | posix_cpu_timer_del(&timer); | |
1300 | } | |
1da177e4 LT |
1301 | spin_unlock_irq(&timer.it_lock); |
1302 | ||
e6c42c29 SG |
1303 | while (error == TIMER_RETRY) { |
1304 | /* | |
1305 | * We need to handle case when timer was or is in the | |
1306 | * middle of firing. In other cases we already freed | |
1307 | * resources. | |
1308 | */ | |
1309 | spin_lock_irq(&timer.it_lock); | |
1310 | error = posix_cpu_timer_del(&timer); | |
1311 | spin_unlock_irq(&timer.it_lock); | |
1312 | } | |
1313 | ||
86a9c446 | 1314 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
1da177e4 LT |
1315 | /* |
1316 | * It actually did fire already. | |
1317 | */ | |
1318 | return 0; | |
1319 | } | |
1320 | ||
e4b76555 | 1321 | error = -ERESTART_RESTARTBLOCK; |
86a9c446 AV |
1322 | /* |
1323 | * Report back to the user the time still remaining. | |
1324 | */ | |
edbeda46 | 1325 | restart = ¤t->restart_block; |
343d8fc2 | 1326 | restart->nanosleep.expires = expires; |
c0edd7c9 DD |
1327 | if (restart->nanosleep.type != TT_NONE) |
1328 | error = nanosleep_copyout(restart, &it.it_value); | |
e4b76555 TA |
1329 | } |
1330 | ||
1331 | return error; | |
1332 | } | |
1333 | ||
bc2c8ea4 TG |
1334 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
1335 | ||
1336 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, | |
938e7cf2 | 1337 | const struct timespec64 *rqtp) |
e4b76555 | 1338 | { |
f56141e3 | 1339 | struct restart_block *restart_block = ¤t->restart_block; |
e4b76555 TA |
1340 | int error; |
1341 | ||
1342 | /* | |
1343 | * Diagnose required errors first. | |
1344 | */ | |
1345 | if (CPUCLOCK_PERTHREAD(which_clock) && | |
1346 | (CPUCLOCK_PID(which_clock) == 0 || | |
01a21974 | 1347 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
e4b76555 TA |
1348 | return -EINVAL; |
1349 | ||
86a9c446 | 1350 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
e4b76555 TA |
1351 | |
1352 | if (error == -ERESTART_RESTARTBLOCK) { | |
1353 | ||
3751f9f2 | 1354 | if (flags & TIMER_ABSTIME) |
e4b76555 | 1355 | return -ERESTARTNOHAND; |
1da177e4 | 1356 | |
1711ef38 | 1357 | restart_block->fn = posix_cpu_nsleep_restart; |
ab8177bc | 1358 | restart_block->nanosleep.clockid = which_clock; |
1da177e4 | 1359 | } |
1da177e4 LT |
1360 | return error; |
1361 | } | |
1362 | ||
bc2c8ea4 | 1363 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1da177e4 | 1364 | { |
ab8177bc | 1365 | clockid_t which_clock = restart_block->nanosleep.clockid; |
ad196384 | 1366 | struct timespec64 t; |
97735f25 | 1367 | |
ad196384 | 1368 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
97735f25 | 1369 | |
86a9c446 | 1370 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
1da177e4 LT |
1371 | } |
1372 | ||
1da177e4 LT |
1373 | #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED) |
1374 | #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED) | |
1375 | ||
a924b04d | 1376 | static int process_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1377 | struct timespec64 *tp) |
1da177e4 LT |
1378 | { |
1379 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); | |
1380 | } | |
a924b04d | 1381 | static int process_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1382 | struct timespec64 *tp) |
1da177e4 LT |
1383 | { |
1384 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); | |
1385 | } | |
1386 | static int process_cpu_timer_create(struct k_itimer *timer) | |
1387 | { | |
1388 | timer->it_clock = PROCESS_CLOCK; | |
1389 | return posix_cpu_timer_create(timer); | |
1390 | } | |
a924b04d | 1391 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
938e7cf2 | 1392 | const struct timespec64 *rqtp) |
1da177e4 | 1393 | { |
99e6c0e6 | 1394 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
1da177e4 | 1395 | } |
a924b04d | 1396 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1397 | struct timespec64 *tp) |
1da177e4 LT |
1398 | { |
1399 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); | |
1400 | } | |
a924b04d | 1401 | static int thread_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1402 | struct timespec64 *tp) |
1da177e4 LT |
1403 | { |
1404 | return posix_cpu_clock_get(THREAD_CLOCK, tp); | |
1405 | } | |
1406 | static int thread_cpu_timer_create(struct k_itimer *timer) | |
1407 | { | |
1408 | timer->it_clock = THREAD_CLOCK; | |
1409 | return posix_cpu_timer_create(timer); | |
1410 | } | |
1da177e4 | 1411 | |
d3ba5a9a | 1412 | const struct k_clock clock_posix_cpu = { |
1976945e TG |
1413 | .clock_getres = posix_cpu_clock_getres, |
1414 | .clock_set = posix_cpu_clock_set, | |
1415 | .clock_get = posix_cpu_clock_get, | |
1416 | .timer_create = posix_cpu_timer_create, | |
1417 | .nsleep = posix_cpu_nsleep, | |
1976945e TG |
1418 | .timer_set = posix_cpu_timer_set, |
1419 | .timer_del = posix_cpu_timer_del, | |
1420 | .timer_get = posix_cpu_timer_get, | |
f37fb0aa | 1421 | .timer_rearm = posix_cpu_timer_rearm, |
1976945e TG |
1422 | }; |
1423 | ||
d3ba5a9a CH |
1424 | const struct k_clock clock_process = { |
1425 | .clock_getres = process_cpu_clock_getres, | |
1426 | .clock_get = process_cpu_clock_get, | |
1427 | .timer_create = process_cpu_timer_create, | |
1428 | .nsleep = process_cpu_nsleep, | |
d3ba5a9a | 1429 | }; |
1da177e4 | 1430 | |
d3ba5a9a CH |
1431 | const struct k_clock clock_thread = { |
1432 | .clock_getres = thread_cpu_clock_getres, | |
1433 | .clock_get = thread_cpu_clock_get, | |
1434 | .timer_create = thread_cpu_timer_create, | |
1435 | }; |