]>
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 | |
a3222f88 | 377 | WARN_ON_ONCE(p == NULL); |
108150ea | 378 | |
3d7a1427 FW |
379 | /* |
380 | * Protect against sighand release/switch in exit/exec and process/ | |
381 | * thread timer list entry concurrent read/writes. | |
382 | */ | |
383 | sighand = lock_task_sighand(p, &flags); | |
384 | if (unlikely(sighand == NULL)) { | |
a3222f88 FW |
385 | /* |
386 | * We raced with the reaping of the task. | |
387 | * The deletion should have cleared us off the list. | |
388 | */ | |
531f64fd | 389 | WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry)); |
a3222f88 | 390 | } else { |
a3222f88 FW |
391 | if (timer->it.cpu.firing) |
392 | ret = TIMER_RETRY; | |
393 | else | |
394 | list_del(&timer->it.cpu.entry); | |
3d7a1427 FW |
395 | |
396 | unlock_task_sighand(p, &flags); | |
1da177e4 | 397 | } |
a3222f88 FW |
398 | |
399 | if (!ret) | |
400 | put_task_struct(p); | |
1da177e4 | 401 | |
108150ea | 402 | return ret; |
1da177e4 LT |
403 | } |
404 | ||
af82eb3c | 405 | static void cleanup_timers_list(struct list_head *head) |
1a7fa510 FW |
406 | { |
407 | struct cpu_timer_list *timer, *next; | |
408 | ||
a0b2062b | 409 | list_for_each_entry_safe(timer, next, head, entry) |
1a7fa510 | 410 | list_del_init(&timer->entry); |
1a7fa510 FW |
411 | } |
412 | ||
1da177e4 LT |
413 | /* |
414 | * Clean out CPU timers still ticking when a thread exited. The task | |
415 | * pointer is cleared, and the expiry time is replaced with the residual | |
416 | * time for later timer_gettime calls to return. | |
417 | * This must be called with the siglock held. | |
418 | */ | |
af82eb3c | 419 | static void cleanup_timers(struct list_head *head) |
1da177e4 | 420 | { |
af82eb3c FW |
421 | cleanup_timers_list(head); |
422 | cleanup_timers_list(++head); | |
423 | cleanup_timers_list(++head); | |
1da177e4 LT |
424 | } |
425 | ||
426 | /* | |
427 | * These are both called with the siglock held, when the current thread | |
428 | * is being reaped. When the final (leader) thread in the group is reaped, | |
429 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. | |
430 | */ | |
431 | void posix_cpu_timers_exit(struct task_struct *tsk) | |
432 | { | |
af82eb3c | 433 | cleanup_timers(tsk->cpu_timers); |
1da177e4 LT |
434 | } |
435 | void posix_cpu_timers_exit_group(struct task_struct *tsk) | |
436 | { | |
af82eb3c | 437 | cleanup_timers(tsk->signal->cpu_timers); |
1da177e4 LT |
438 | } |
439 | ||
ebd7e7fc | 440 | static inline int expires_gt(u64 expires, u64 new_exp) |
d1e3b6d1 | 441 | { |
64861634 | 442 | return expires == 0 || expires > new_exp; |
d1e3b6d1 SG |
443 | } |
444 | ||
1da177e4 LT |
445 | /* |
446 | * Insert the timer on the appropriate list before any timers that | |
e73d84e3 | 447 | * expire later. This must be called with the sighand lock held. |
1da177e4 | 448 | */ |
5eb9aa64 | 449 | static void arm_timer(struct k_itimer *timer) |
1da177e4 LT |
450 | { |
451 | struct task_struct *p = timer->it.cpu.task; | |
452 | struct list_head *head, *listpos; | |
ebd7e7fc | 453 | struct task_cputime *cputime_expires; |
1da177e4 LT |
454 | struct cpu_timer_list *const nt = &timer->it.cpu; |
455 | struct cpu_timer_list *next; | |
1da177e4 | 456 | |
5eb9aa64 SG |
457 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
458 | head = p->cpu_timers; | |
459 | cputime_expires = &p->cputime_expires; | |
460 | } else { | |
461 | head = p->signal->cpu_timers; | |
462 | cputime_expires = &p->signal->cputime_expires; | |
463 | } | |
1da177e4 LT |
464 | head += CPUCLOCK_WHICH(timer->it_clock); |
465 | ||
1da177e4 | 466 | listpos = head; |
5eb9aa64 | 467 | list_for_each_entry(next, head, entry) { |
55ccb616 | 468 | if (nt->expires < next->expires) |
5eb9aa64 SG |
469 | break; |
470 | listpos = &next->entry; | |
1da177e4 LT |
471 | } |
472 | list_add(&nt->entry, listpos); | |
473 | ||
474 | if (listpos == head) { | |
ebd7e7fc | 475 | u64 exp = nt->expires; |
5eb9aa64 | 476 | |
1da177e4 | 477 | /* |
5eb9aa64 SG |
478 | * We are the new earliest-expiring POSIX 1.b timer, hence |
479 | * need to update expiration cache. Take into account that | |
480 | * for process timers we share expiration cache with itimers | |
481 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. | |
1da177e4 LT |
482 | */ |
483 | ||
5eb9aa64 SG |
484 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
485 | case CPUCLOCK_PROF: | |
ebd7e7fc FW |
486 | if (expires_gt(cputime_expires->prof_exp, exp)) |
487 | cputime_expires->prof_exp = exp; | |
5eb9aa64 SG |
488 | break; |
489 | case CPUCLOCK_VIRT: | |
ebd7e7fc FW |
490 | if (expires_gt(cputime_expires->virt_exp, exp)) |
491 | cputime_expires->virt_exp = exp; | |
5eb9aa64 SG |
492 | break; |
493 | case CPUCLOCK_SCHED: | |
ebd7e7fc | 494 | if (expires_gt(cputime_expires->sched_exp, exp)) |
55ccb616 | 495 | cputime_expires->sched_exp = exp; |
5eb9aa64 | 496 | break; |
1da177e4 | 497 | } |
b7878300 FW |
498 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
499 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); | |
500 | else | |
501 | tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 | 502 | } |
1da177e4 LT |
503 | } |
504 | ||
505 | /* | |
506 | * The timer is locked, fire it and arrange for its reload. | |
507 | */ | |
508 | static void cpu_timer_fire(struct k_itimer *timer) | |
509 | { | |
1f169f84 SG |
510 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
511 | /* | |
512 | * User don't want any signal. | |
513 | */ | |
55ccb616 | 514 | timer->it.cpu.expires = 0; |
1f169f84 | 515 | } else if (unlikely(timer->sigq == NULL)) { |
1da177e4 LT |
516 | /* |
517 | * This a special case for clock_nanosleep, | |
518 | * not a normal timer from sys_timer_create. | |
519 | */ | |
520 | wake_up_process(timer->it_process); | |
55ccb616 FW |
521 | timer->it.cpu.expires = 0; |
522 | } else if (timer->it.cpu.incr == 0) { | |
1da177e4 LT |
523 | /* |
524 | * One-shot timer. Clear it as soon as it's fired. | |
525 | */ | |
526 | posix_timer_event(timer, 0); | |
55ccb616 | 527 | timer->it.cpu.expires = 0; |
1da177e4 LT |
528 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
529 | /* | |
530 | * The signal did not get queued because the signal | |
531 | * was ignored, so we won't get any callback to | |
532 | * reload the timer. But we need to keep it | |
533 | * ticking in case the signal is deliverable next time. | |
534 | */ | |
f37fb0aa | 535 | posix_cpu_timer_rearm(timer); |
af888d67 | 536 | ++timer->it_requeue_pending; |
1da177e4 LT |
537 | } |
538 | } | |
539 | ||
3997ad31 PZ |
540 | /* |
541 | * Sample a process (thread group) timer for the given group_leader task. | |
e73d84e3 FW |
542 | * Must be called with task sighand lock held for safe while_each_thread() |
543 | * traversal. | |
3997ad31 PZ |
544 | */ |
545 | static int cpu_timer_sample_group(const clockid_t which_clock, | |
ebd7e7fc | 546 | struct task_struct *p, u64 *sample) |
3997ad31 | 547 | { |
ebd7e7fc | 548 | struct task_cputime cputime; |
3997ad31 PZ |
549 | |
550 | thread_group_cputimer(p, &cputime); | |
551 | switch (CPUCLOCK_WHICH(which_clock)) { | |
552 | default: | |
553 | return -EINVAL; | |
554 | case CPUCLOCK_PROF: | |
ebd7e7fc | 555 | *sample = cputime.utime + cputime.stime; |
3997ad31 PZ |
556 | break; |
557 | case CPUCLOCK_VIRT: | |
ebd7e7fc | 558 | *sample = cputime.utime; |
3997ad31 PZ |
559 | break; |
560 | case CPUCLOCK_SCHED: | |
23cfa361 | 561 | *sample = cputime.sum_exec_runtime; |
3997ad31 PZ |
562 | break; |
563 | } | |
564 | return 0; | |
565 | } | |
566 | ||
1da177e4 LT |
567 | /* |
568 | * Guts of sys_timer_settime for CPU timers. | |
569 | * This is called with the timer locked and interrupts disabled. | |
570 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
571 | * and try again. (This happens when the timer is in the middle of firing.) | |
572 | */ | |
e73d84e3 | 573 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
5f252b32 | 574 | struct itimerspec64 *new, struct itimerspec64 *old) |
1da177e4 | 575 | { |
e73d84e3 FW |
576 | unsigned long flags; |
577 | struct sighand_struct *sighand; | |
1da177e4 | 578 | struct task_struct *p = timer->it.cpu.task; |
ebd7e7fc | 579 | u64 old_expires, new_expires, old_incr, val; |
1da177e4 LT |
580 | int ret; |
581 | ||
a3222f88 | 582 | WARN_ON_ONCE(p == NULL); |
1da177e4 | 583 | |
098b0e01 TG |
584 | /* |
585 | * Use the to_ktime conversion because that clamps the maximum | |
586 | * value to KTIME_MAX and avoid multiplication overflows. | |
587 | */ | |
588 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); | |
1da177e4 | 589 | |
1da177e4 | 590 | /* |
e73d84e3 FW |
591 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
592 | * and p->signal->cpu_timers read/write in arm_timer() | |
593 | */ | |
594 | sighand = lock_task_sighand(p, &flags); | |
595 | /* | |
596 | * If p has just been reaped, we can no | |
1da177e4 LT |
597 | * longer get any information about it at all. |
598 | */ | |
e73d84e3 | 599 | if (unlikely(sighand == NULL)) { |
1da177e4 LT |
600 | return -ESRCH; |
601 | } | |
602 | ||
603 | /* | |
604 | * Disarm any old timer after extracting its expiry time. | |
605 | */ | |
a6968220 | 606 | lockdep_assert_irqs_disabled(); |
a69ac4a7 ON |
607 | |
608 | ret = 0; | |
ae1a78ee | 609 | old_incr = timer->it.cpu.incr; |
1da177e4 | 610 | old_expires = timer->it.cpu.expires; |
a69ac4a7 ON |
611 | if (unlikely(timer->it.cpu.firing)) { |
612 | timer->it.cpu.firing = -1; | |
613 | ret = TIMER_RETRY; | |
614 | } else | |
615 | list_del_init(&timer->it.cpu.entry); | |
1da177e4 LT |
616 | |
617 | /* | |
618 | * We need to sample the current value to convert the new | |
619 | * value from to relative and absolute, and to convert the | |
620 | * old value from absolute to relative. To set a process | |
621 | * timer, we need a sample to balance the thread expiry | |
622 | * times (in arm_timer). With an absolute time, we must | |
623 | * check if it's already passed. In short, we need a sample. | |
624 | */ | |
625 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
626 | cpu_clock_sample(timer->it_clock, p, &val); | |
627 | } else { | |
3997ad31 | 628 | cpu_timer_sample_group(timer->it_clock, p, &val); |
1da177e4 LT |
629 | } |
630 | ||
631 | if (old) { | |
55ccb616 | 632 | if (old_expires == 0) { |
1da177e4 LT |
633 | old->it_value.tv_sec = 0; |
634 | old->it_value.tv_nsec = 0; | |
635 | } else { | |
636 | /* | |
637 | * Update the timer in case it has | |
638 | * overrun already. If it has, | |
639 | * we'll report it as having overrun | |
640 | * and with the next reloaded timer | |
641 | * already ticking, though we are | |
642 | * swallowing that pending | |
643 | * notification here to install the | |
644 | * new setting. | |
645 | */ | |
646 | bump_cpu_timer(timer, val); | |
55ccb616 FW |
647 | if (val < timer->it.cpu.expires) { |
648 | old_expires = timer->it.cpu.expires - val; | |
5f252b32 | 649 | old->it_value = ns_to_timespec64(old_expires); |
1da177e4 LT |
650 | } else { |
651 | old->it_value.tv_nsec = 1; | |
652 | old->it_value.tv_sec = 0; | |
653 | } | |
654 | } | |
655 | } | |
656 | ||
a69ac4a7 | 657 | if (unlikely(ret)) { |
1da177e4 LT |
658 | /* |
659 | * We are colliding with the timer actually firing. | |
660 | * Punt after filling in the timer's old value, and | |
661 | * disable this firing since we are already reporting | |
662 | * it as an overrun (thanks to bump_cpu_timer above). | |
663 | */ | |
e73d84e3 | 664 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
665 | goto out; |
666 | } | |
667 | ||
e73d84e3 | 668 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
55ccb616 | 669 | new_expires += val; |
1da177e4 LT |
670 | } |
671 | ||
672 | /* | |
673 | * Install the new expiry time (or zero). | |
674 | * For a timer with no notification action, we don't actually | |
675 | * arm the timer (we'll just fake it for timer_gettime). | |
676 | */ | |
677 | timer->it.cpu.expires = new_expires; | |
55ccb616 | 678 | if (new_expires != 0 && val < new_expires) { |
5eb9aa64 | 679 | arm_timer(timer); |
1da177e4 LT |
680 | } |
681 | ||
e73d84e3 | 682 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
683 | /* |
684 | * Install the new reload setting, and | |
685 | * set up the signal and overrun bookkeeping. | |
686 | */ | |
5f252b32 | 687 | timer->it.cpu.incr = timespec64_to_ns(&new->it_interval); |
1da177e4 LT |
688 | |
689 | /* | |
690 | * This acts as a modification timestamp for the timer, | |
691 | * so any automatic reload attempt will punt on seeing | |
692 | * that we have reset the timer manually. | |
693 | */ | |
694 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & | |
695 | ~REQUEUE_PENDING; | |
696 | timer->it_overrun_last = 0; | |
697 | timer->it_overrun = -1; | |
698 | ||
55ccb616 | 699 | if (new_expires != 0 && !(val < new_expires)) { |
1da177e4 LT |
700 | /* |
701 | * The designated time already passed, so we notify | |
702 | * immediately, even if the thread never runs to | |
703 | * accumulate more time on this clock. | |
704 | */ | |
705 | cpu_timer_fire(timer); | |
706 | } | |
707 | ||
708 | ret = 0; | |
709 | out: | |
ebd7e7fc | 710 | if (old) |
5f252b32 | 711 | old->it_interval = ns_to_timespec64(old_incr); |
b7878300 | 712 | |
1da177e4 LT |
713 | return ret; |
714 | } | |
715 | ||
5f252b32 | 716 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
1da177e4 | 717 | { |
ebd7e7fc | 718 | u64 now; |
1da177e4 | 719 | struct task_struct *p = timer->it.cpu.task; |
1da177e4 | 720 | |
a3222f88 FW |
721 | WARN_ON_ONCE(p == NULL); |
722 | ||
1da177e4 LT |
723 | /* |
724 | * Easy part: convert the reload time. | |
725 | */ | |
5f252b32 | 726 | itp->it_interval = ns_to_timespec64(timer->it.cpu.incr); |
1da177e4 | 727 | |
eabdec04 | 728 | if (!timer->it.cpu.expires) |
1da177e4 | 729 | return; |
1da177e4 | 730 | |
1da177e4 LT |
731 | /* |
732 | * Sample the clock to take the difference with the expiry time. | |
733 | */ | |
734 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
735 | cpu_clock_sample(timer->it_clock, p, &now); | |
1da177e4 | 736 | } else { |
e73d84e3 FW |
737 | struct sighand_struct *sighand; |
738 | unsigned long flags; | |
739 | ||
740 | /* | |
741 | * Protect against sighand release/switch in exit/exec and | |
742 | * also make timer sampling safe if it ends up calling | |
ebd7e7fc | 743 | * thread_group_cputime(). |
e73d84e3 FW |
744 | */ |
745 | sighand = lock_task_sighand(p, &flags); | |
746 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
747 | /* |
748 | * The process has been reaped. | |
749 | * We can't even collect a sample any more. | |
750 | * Call the timer disarmed, nothing else to do. | |
751 | */ | |
55ccb616 | 752 | timer->it.cpu.expires = 0; |
2c13ce8f | 753 | return; |
1da177e4 | 754 | } else { |
3997ad31 | 755 | cpu_timer_sample_group(timer->it_clock, p, &now); |
e73d84e3 | 756 | unlock_task_sighand(p, &flags); |
1da177e4 | 757 | } |
1da177e4 LT |
758 | } |
759 | ||
55ccb616 | 760 | if (now < timer->it.cpu.expires) { |
5f252b32 | 761 | itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now); |
1da177e4 LT |
762 | } else { |
763 | /* | |
764 | * The timer should have expired already, but the firing | |
765 | * hasn't taken place yet. Say it's just about to expire. | |
766 | */ | |
767 | itp->it_value.tv_nsec = 1; | |
768 | itp->it_value.tv_sec = 0; | |
769 | } | |
770 | } | |
771 | ||
2473f3e7 FW |
772 | static unsigned long long |
773 | check_timers_list(struct list_head *timers, | |
774 | struct list_head *firing, | |
775 | unsigned long long curr) | |
776 | { | |
777 | int maxfire = 20; | |
778 | ||
779 | while (!list_empty(timers)) { | |
780 | struct cpu_timer_list *t; | |
781 | ||
782 | t = list_first_entry(timers, struct cpu_timer_list, entry); | |
783 | ||
784 | if (!--maxfire || curr < t->expires) | |
785 | return t->expires; | |
786 | ||
787 | t->firing = 1; | |
788 | list_move_tail(&t->entry, firing); | |
789 | } | |
790 | ||
791 | return 0; | |
792 | } | |
793 | ||
1da177e4 LT |
794 | /* |
795 | * Check for any per-thread CPU timers that have fired and move them off | |
796 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the | |
797 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. | |
798 | */ | |
799 | static void check_thread_timers(struct task_struct *tsk, | |
800 | struct list_head *firing) | |
801 | { | |
802 | struct list_head *timers = tsk->cpu_timers; | |
ebd7e7fc FW |
803 | struct task_cputime *tsk_expires = &tsk->cputime_expires; |
804 | u64 expires; | |
d4bb5274 | 805 | unsigned long soft; |
1da177e4 | 806 | |
934715a1 JL |
807 | /* |
808 | * If cputime_expires is zero, then there are no active | |
809 | * per thread CPU timers. | |
810 | */ | |
811 | if (task_cputime_zero(&tsk->cputime_expires)) | |
812 | return; | |
813 | ||
2473f3e7 | 814 | expires = check_timers_list(timers, firing, prof_ticks(tsk)); |
ebd7e7fc | 815 | tsk_expires->prof_exp = expires; |
1da177e4 | 816 | |
2473f3e7 | 817 | expires = check_timers_list(++timers, firing, virt_ticks(tsk)); |
ebd7e7fc | 818 | tsk_expires->virt_exp = expires; |
1da177e4 | 819 | |
2473f3e7 FW |
820 | tsk_expires->sched_exp = check_timers_list(++timers, firing, |
821 | tsk->se.sum_exec_runtime); | |
78f2c7db PZ |
822 | |
823 | /* | |
824 | * Check for the special case thread timers. | |
825 | */ | |
3cf29496 | 826 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
d4bb5274 | 827 | if (soft != RLIM_INFINITY) { |
3cf29496 | 828 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
78f2c7db | 829 | |
5a52dd50 PZ |
830 | if (hard != RLIM_INFINITY && |
831 | tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) { | |
78f2c7db PZ |
832 | /* |
833 | * At the hard limit, we just die. | |
834 | * No need to calculate anything else now. | |
835 | */ | |
43fe8b8e TG |
836 | if (print_fatal_signals) { |
837 | pr_info("CPU Watchdog Timeout (hard): %s[%d]\n", | |
838 | tsk->comm, task_pid_nr(tsk)); | |
839 | } | |
78f2c7db PZ |
840 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
841 | return; | |
842 | } | |
d4bb5274 | 843 | if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) { |
78f2c7db PZ |
844 | /* |
845 | * At the soft limit, send a SIGXCPU every second. | |
846 | */ | |
d4bb5274 JS |
847 | if (soft < hard) { |
848 | soft += USEC_PER_SEC; | |
3cf29496 KO |
849 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = |
850 | soft; | |
78f2c7db | 851 | } |
43fe8b8e TG |
852 | if (print_fatal_signals) { |
853 | pr_info("RT Watchdog Timeout (soft): %s[%d]\n", | |
854 | tsk->comm, task_pid_nr(tsk)); | |
855 | } | |
78f2c7db PZ |
856 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
857 | } | |
858 | } | |
b7878300 FW |
859 | if (task_cputime_zero(tsk_expires)) |
860 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
861 | } |
862 | ||
1018016c | 863 | static inline void stop_process_timers(struct signal_struct *sig) |
3fccfd67 | 864 | { |
15365c10 | 865 | struct thread_group_cputimer *cputimer = &sig->cputimer; |
3fccfd67 | 866 | |
1018016c | 867 | /* Turn off cputimer->running. This is done without locking. */ |
d5c373eb | 868 | WRITE_ONCE(cputimer->running, false); |
b7878300 | 869 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
3fccfd67 PZ |
870 | } |
871 | ||
42c4ab41 | 872 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
ebd7e7fc | 873 | u64 *expires, u64 cur_time, int signo) |
42c4ab41 | 874 | { |
64861634 | 875 | if (!it->expires) |
42c4ab41 SG |
876 | return; |
877 | ||
858cf3a8 FW |
878 | if (cur_time >= it->expires) { |
879 | if (it->incr) | |
64861634 | 880 | it->expires += it->incr; |
858cf3a8 | 881 | else |
64861634 | 882 | it->expires = 0; |
42c4ab41 | 883 | |
3f0a525e XG |
884 | trace_itimer_expire(signo == SIGPROF ? |
885 | ITIMER_PROF : ITIMER_VIRTUAL, | |
886 | tsk->signal->leader_pid, cur_time); | |
42c4ab41 SG |
887 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
888 | } | |
889 | ||
858cf3a8 FW |
890 | if (it->expires && (!*expires || it->expires < *expires)) |
891 | *expires = it->expires; | |
42c4ab41 SG |
892 | } |
893 | ||
1da177e4 LT |
894 | /* |
895 | * Check for any per-thread CPU timers that have fired and move them | |
896 | * off the tsk->*_timers list onto the firing list. Per-thread timers | |
897 | * have already been taken off. | |
898 | */ | |
899 | static void check_process_timers(struct task_struct *tsk, | |
900 | struct list_head *firing) | |
901 | { | |
902 | struct signal_struct *const sig = tsk->signal; | |
ebd7e7fc FW |
903 | u64 utime, ptime, virt_expires, prof_expires; |
904 | u64 sum_sched_runtime, sched_expires; | |
1da177e4 | 905 | struct list_head *timers = sig->cpu_timers; |
ebd7e7fc | 906 | struct task_cputime cputime; |
d4bb5274 | 907 | unsigned long soft; |
1da177e4 | 908 | |
934715a1 JL |
909 | /* |
910 | * If cputimer is not running, then there are no active | |
911 | * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU). | |
912 | */ | |
913 | if (!READ_ONCE(tsk->signal->cputimer.running)) | |
914 | return; | |
915 | ||
c8d75aa4 JL |
916 | /* |
917 | * Signify that a thread is checking for process timers. | |
918 | * Write access to this field is protected by the sighand lock. | |
919 | */ | |
920 | sig->cputimer.checking_timer = true; | |
921 | ||
1da177e4 LT |
922 | /* |
923 | * Collect the current process totals. | |
924 | */ | |
4cd4c1b4 | 925 | thread_group_cputimer(tsk, &cputime); |
ebd7e7fc FW |
926 | utime = cputime.utime; |
927 | ptime = utime + cputime.stime; | |
f06febc9 | 928 | sum_sched_runtime = cputime.sum_exec_runtime; |
1da177e4 | 929 | |
2473f3e7 FW |
930 | prof_expires = check_timers_list(timers, firing, ptime); |
931 | virt_expires = check_timers_list(++timers, firing, utime); | |
932 | sched_expires = check_timers_list(++timers, firing, sum_sched_runtime); | |
1da177e4 LT |
933 | |
934 | /* | |
935 | * Check for the special case process timers. | |
936 | */ | |
42c4ab41 SG |
937 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime, |
938 | SIGPROF); | |
939 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime, | |
940 | SIGVTALRM); | |
3cf29496 | 941 | soft = task_rlimit(tsk, RLIMIT_CPU); |
d4bb5274 | 942 | if (soft != RLIM_INFINITY) { |
ebd7e7fc | 943 | unsigned long psecs = div_u64(ptime, NSEC_PER_SEC); |
3cf29496 | 944 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
ebd7e7fc | 945 | u64 x; |
d4bb5274 | 946 | if (psecs >= hard) { |
1da177e4 LT |
947 | /* |
948 | * At the hard limit, we just die. | |
949 | * No need to calculate anything else now. | |
950 | */ | |
43fe8b8e TG |
951 | if (print_fatal_signals) { |
952 | pr_info("RT Watchdog Timeout (hard): %s[%d]\n", | |
953 | tsk->comm, task_pid_nr(tsk)); | |
954 | } | |
1da177e4 LT |
955 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
956 | return; | |
957 | } | |
d4bb5274 | 958 | if (psecs >= soft) { |
1da177e4 LT |
959 | /* |
960 | * At the soft limit, send a SIGXCPU every second. | |
961 | */ | |
43fe8b8e TG |
962 | if (print_fatal_signals) { |
963 | pr_info("CPU Watchdog Timeout (soft): %s[%d]\n", | |
964 | tsk->comm, task_pid_nr(tsk)); | |
965 | } | |
1da177e4 | 966 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
d4bb5274 JS |
967 | if (soft < hard) { |
968 | soft++; | |
969 | sig->rlim[RLIMIT_CPU].rlim_cur = soft; | |
1da177e4 LT |
970 | } |
971 | } | |
ebd7e7fc FW |
972 | x = soft * NSEC_PER_SEC; |
973 | if (!prof_expires || x < prof_expires) | |
1da177e4 | 974 | prof_expires = x; |
1da177e4 LT |
975 | } |
976 | ||
ebd7e7fc FW |
977 | sig->cputime_expires.prof_exp = prof_expires; |
978 | sig->cputime_expires.virt_exp = virt_expires; | |
29f87b79 SG |
979 | sig->cputime_expires.sched_exp = sched_expires; |
980 | if (task_cputime_zero(&sig->cputime_expires)) | |
981 | stop_process_timers(sig); | |
c8d75aa4 JL |
982 | |
983 | sig->cputimer.checking_timer = false; | |
1da177e4 LT |
984 | } |
985 | ||
986 | /* | |
96fe3b07 | 987 | * This is called from the signal code (via posixtimer_rearm) |
1da177e4 LT |
988 | * when the last timer signal was delivered and we have to reload the timer. |
989 | */ | |
f37fb0aa | 990 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
1da177e4 | 991 | { |
e73d84e3 FW |
992 | struct sighand_struct *sighand; |
993 | unsigned long flags; | |
1da177e4 | 994 | struct task_struct *p = timer->it.cpu.task; |
ebd7e7fc | 995 | u64 now; |
1da177e4 | 996 | |
a3222f88 | 997 | WARN_ON_ONCE(p == NULL); |
1da177e4 LT |
998 | |
999 | /* | |
1000 | * Fetch the current sample and update the timer's expiry time. | |
1001 | */ | |
1002 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
1003 | cpu_clock_sample(timer->it_clock, p, &now); | |
1004 | bump_cpu_timer(timer, now); | |
724a3713 | 1005 | if (unlikely(p->exit_state)) |
af888d67 | 1006 | return; |
724a3713 | 1007 | |
e73d84e3 FW |
1008 | /* Protect timer list r/w in arm_timer() */ |
1009 | sighand = lock_task_sighand(p, &flags); | |
1010 | if (!sighand) | |
af888d67 | 1011 | return; |
1da177e4 | 1012 | } else { |
e73d84e3 FW |
1013 | /* |
1014 | * Protect arm_timer() and timer sampling in case of call to | |
ebd7e7fc | 1015 | * thread_group_cputime(). |
e73d84e3 FW |
1016 | */ |
1017 | sighand = lock_task_sighand(p, &flags); | |
1018 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
1019 | /* |
1020 | * The process has been reaped. | |
1021 | * We can't even collect a sample any more. | |
1022 | */ | |
55ccb616 | 1023 | timer->it.cpu.expires = 0; |
af888d67 | 1024 | return; |
1da177e4 | 1025 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
af888d67 TG |
1026 | /* If the process is dying, no need to rearm */ |
1027 | goto unlock; | |
1da177e4 | 1028 | } |
3997ad31 | 1029 | cpu_timer_sample_group(timer->it_clock, p, &now); |
1da177e4 | 1030 | bump_cpu_timer(timer, now); |
e73d84e3 | 1031 | /* Leave the sighand locked for the call below. */ |
1da177e4 LT |
1032 | } |
1033 | ||
1034 | /* | |
1035 | * Now re-arm for the new expiry time. | |
1036 | */ | |
a6968220 | 1037 | lockdep_assert_irqs_disabled(); |
5eb9aa64 | 1038 | arm_timer(timer); |
af888d67 | 1039 | unlock: |
e73d84e3 | 1040 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
1041 | } |
1042 | ||
f06febc9 FM |
1043 | /** |
1044 | * task_cputime_expired - Compare two task_cputime entities. | |
1045 | * | |
1046 | * @sample: The task_cputime structure to be checked for expiration. | |
1047 | * @expires: Expiration times, against which @sample will be checked. | |
1048 | * | |
1049 | * Checks @sample against @expires to see if any field of @sample has expired. | |
1050 | * Returns true if any field of the former is greater than the corresponding | |
1051 | * field of the latter if the latter field is set. Otherwise returns false. | |
1052 | */ | |
ebd7e7fc FW |
1053 | static inline int task_cputime_expired(const struct task_cputime *sample, |
1054 | const struct task_cputime *expires) | |
f06febc9 | 1055 | { |
64861634 | 1056 | if (expires->utime && sample->utime >= expires->utime) |
f06febc9 | 1057 | return 1; |
64861634 | 1058 | if (expires->stime && sample->utime + sample->stime >= expires->stime) |
f06febc9 FM |
1059 | return 1; |
1060 | if (expires->sum_exec_runtime != 0 && | |
1061 | sample->sum_exec_runtime >= expires->sum_exec_runtime) | |
1062 | return 1; | |
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * fastpath_timer_check - POSIX CPU timers fast path. | |
1068 | * | |
1069 | * @tsk: The task (thread) being checked. | |
f06febc9 | 1070 | * |
bb34d92f FM |
1071 | * Check the task and thread group timers. If both are zero (there are no |
1072 | * timers set) return false. Otherwise snapshot the task and thread group | |
1073 | * timers and compare them with the corresponding expiration times. Return | |
1074 | * true if a timer has expired, else return false. | |
f06febc9 | 1075 | */ |
bb34d92f | 1076 | static inline int fastpath_timer_check(struct task_struct *tsk) |
f06febc9 | 1077 | { |
ad133ba3 | 1078 | struct signal_struct *sig; |
bb34d92f | 1079 | |
bb34d92f | 1080 | if (!task_cputime_zero(&tsk->cputime_expires)) { |
ebd7e7fc | 1081 | struct task_cputime task_sample; |
bb34d92f | 1082 | |
ebd7e7fc | 1083 | task_cputime(tsk, &task_sample.utime, &task_sample.stime); |
7c177d99 | 1084 | task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime; |
bb34d92f FM |
1085 | if (task_cputime_expired(&task_sample, &tsk->cputime_expires)) |
1086 | return 1; | |
1087 | } | |
ad133ba3 ON |
1088 | |
1089 | sig = tsk->signal; | |
c8d75aa4 JL |
1090 | /* |
1091 | * Check if thread group timers expired when the cputimer is | |
1092 | * running and no other thread in the group is already checking | |
1093 | * for thread group cputimers. These fields are read without the | |
1094 | * sighand lock. However, this is fine because this is meant to | |
1095 | * be a fastpath heuristic to determine whether we should try to | |
1096 | * acquire the sighand lock to check/handle timers. | |
1097 | * | |
1098 | * In the worst case scenario, if 'running' or 'checking_timer' gets | |
1099 | * set but the current thread doesn't see the change yet, we'll wait | |
1100 | * until the next thread in the group gets a scheduler interrupt to | |
1101 | * handle the timer. This isn't an issue in practice because these | |
1102 | * types of delays with signals actually getting sent are expected. | |
1103 | */ | |
1104 | if (READ_ONCE(sig->cputimer.running) && | |
1105 | !READ_ONCE(sig->cputimer.checking_timer)) { | |
ebd7e7fc | 1106 | struct task_cputime group_sample; |
bb34d92f | 1107 | |
71107445 | 1108 | sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic); |
8d1f431c | 1109 | |
bb34d92f FM |
1110 | if (task_cputime_expired(&group_sample, &sig->cputime_expires)) |
1111 | return 1; | |
1112 | } | |
37bebc70 | 1113 | |
f55db609 | 1114 | return 0; |
f06febc9 FM |
1115 | } |
1116 | ||
1da177e4 LT |
1117 | /* |
1118 | * This is called from the timer interrupt handler. The irq handler has | |
1119 | * already updated our counts. We need to check if any timers fire now. | |
1120 | * Interrupts are disabled. | |
1121 | */ | |
1122 | void run_posix_cpu_timers(struct task_struct *tsk) | |
1123 | { | |
1124 | LIST_HEAD(firing); | |
1125 | struct k_itimer *timer, *next; | |
0bdd2ed4 | 1126 | unsigned long flags; |
1da177e4 | 1127 | |
a6968220 | 1128 | lockdep_assert_irqs_disabled(); |
1da177e4 | 1129 | |
1da177e4 | 1130 | /* |
f06febc9 | 1131 | * The fast path checks that there are no expired thread or thread |
bb34d92f | 1132 | * group timers. If that's so, just return. |
1da177e4 | 1133 | */ |
bb34d92f | 1134 | if (!fastpath_timer_check(tsk)) |
f06febc9 | 1135 | return; |
5ce73a4a | 1136 | |
0bdd2ed4 ON |
1137 | if (!lock_task_sighand(tsk, &flags)) |
1138 | return; | |
bb34d92f FM |
1139 | /* |
1140 | * Here we take off tsk->signal->cpu_timers[N] and | |
1141 | * tsk->cpu_timers[N] all the timers that are firing, and | |
1142 | * put them on the firing list. | |
1143 | */ | |
1144 | check_thread_timers(tsk, &firing); | |
934715a1 JL |
1145 | |
1146 | check_process_timers(tsk, &firing); | |
1da177e4 | 1147 | |
bb34d92f FM |
1148 | /* |
1149 | * We must release these locks before taking any timer's lock. | |
1150 | * There is a potential race with timer deletion here, as the | |
1151 | * siglock now protects our private firing list. We have set | |
1152 | * the firing flag in each timer, so that a deletion attempt | |
1153 | * that gets the timer lock before we do will give it up and | |
1154 | * spin until we've taken care of that timer below. | |
1155 | */ | |
0bdd2ed4 | 1156 | unlock_task_sighand(tsk, &flags); |
1da177e4 LT |
1157 | |
1158 | /* | |
1159 | * Now that all the timers on our list have the firing flag, | |
25985edc | 1160 | * no one will touch their list entries but us. We'll take |
1da177e4 LT |
1161 | * each timer's lock before clearing its firing flag, so no |
1162 | * timer call will interfere. | |
1163 | */ | |
1164 | list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) { | |
6e85c5ba HS |
1165 | int cpu_firing; |
1166 | ||
1da177e4 LT |
1167 | spin_lock(&timer->it_lock); |
1168 | list_del_init(&timer->it.cpu.entry); | |
6e85c5ba | 1169 | cpu_firing = timer->it.cpu.firing; |
1da177e4 LT |
1170 | timer->it.cpu.firing = 0; |
1171 | /* | |
1172 | * The firing flag is -1 if we collided with a reset | |
1173 | * of the timer, which already reported this | |
1174 | * almost-firing as an overrun. So don't generate an event. | |
1175 | */ | |
6e85c5ba | 1176 | if (likely(cpu_firing >= 0)) |
1da177e4 | 1177 | cpu_timer_fire(timer); |
1da177e4 LT |
1178 | spin_unlock(&timer->it_lock); |
1179 | } | |
1180 | } | |
1181 | ||
1182 | /* | |
f55db609 | 1183 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
f06febc9 | 1184 | * The tsk->sighand->siglock must be held by the caller. |
1da177e4 LT |
1185 | */ |
1186 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx, | |
858cf3a8 | 1187 | u64 *newval, u64 *oldval) |
1da177e4 | 1188 | { |
858cf3a8 | 1189 | u64 now; |
1da177e4 | 1190 | |
531f64fd | 1191 | WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED); |
4cd4c1b4 | 1192 | cpu_timer_sample_group(clock_idx, tsk, &now); |
1da177e4 LT |
1193 | |
1194 | if (oldval) { | |
f55db609 SG |
1195 | /* |
1196 | * We are setting itimer. The *oldval is absolute and we update | |
1197 | * it to be relative, *newval argument is relative and we update | |
1198 | * it to be absolute. | |
1199 | */ | |
64861634 | 1200 | if (*oldval) { |
858cf3a8 | 1201 | if (*oldval <= now) { |
1da177e4 | 1202 | /* Just about to fire. */ |
858cf3a8 | 1203 | *oldval = TICK_NSEC; |
1da177e4 | 1204 | } else { |
858cf3a8 | 1205 | *oldval -= now; |
1da177e4 LT |
1206 | } |
1207 | } | |
1208 | ||
64861634 | 1209 | if (!*newval) |
b7878300 | 1210 | return; |
858cf3a8 | 1211 | *newval += now; |
1da177e4 LT |
1212 | } |
1213 | ||
1214 | /* | |
f55db609 SG |
1215 | * Update expiration cache if we are the earliest timer, or eventually |
1216 | * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire. | |
1da177e4 | 1217 | */ |
f55db609 SG |
1218 | switch (clock_idx) { |
1219 | case CPUCLOCK_PROF: | |
858cf3a8 FW |
1220 | if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval)) |
1221 | tsk->signal->cputime_expires.prof_exp = *newval; | |
f55db609 SG |
1222 | break; |
1223 | case CPUCLOCK_VIRT: | |
858cf3a8 FW |
1224 | if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval)) |
1225 | tsk->signal->cputime_expires.virt_exp = *newval; | |
f55db609 | 1226 | break; |
1da177e4 | 1227 | } |
b7878300 FW |
1228 | |
1229 | tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
1230 | } |
1231 | ||
e4b76555 | 1232 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
343d8fc2 | 1233 | const struct timespec64 *rqtp) |
1da177e4 | 1234 | { |
86a9c446 | 1235 | struct itimerspec64 it; |
343d8fc2 TG |
1236 | struct k_itimer timer; |
1237 | u64 expires; | |
1da177e4 LT |
1238 | int error; |
1239 | ||
1da177e4 LT |
1240 | /* |
1241 | * Set up a temporary timer and then wait for it to go off. | |
1242 | */ | |
1243 | memset(&timer, 0, sizeof timer); | |
1244 | spin_lock_init(&timer.it_lock); | |
1245 | timer.it_clock = which_clock; | |
1246 | timer.it_overrun = -1; | |
1247 | error = posix_cpu_timer_create(&timer); | |
1248 | timer.it_process = current; | |
1249 | if (!error) { | |
5f252b32 | 1250 | static struct itimerspec64 zero_it; |
edbeda46 | 1251 | struct restart_block *restart; |
e4b76555 | 1252 | |
edbeda46 | 1253 | memset(&it, 0, sizeof(it)); |
86a9c446 | 1254 | it.it_value = *rqtp; |
1da177e4 LT |
1255 | |
1256 | spin_lock_irq(&timer.it_lock); | |
86a9c446 | 1257 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
1da177e4 LT |
1258 | if (error) { |
1259 | spin_unlock_irq(&timer.it_lock); | |
1260 | return error; | |
1261 | } | |
1262 | ||
1263 | while (!signal_pending(current)) { | |
55ccb616 | 1264 | if (timer.it.cpu.expires == 0) { |
1da177e4 | 1265 | /* |
e6c42c29 SG |
1266 | * Our timer fired and was reset, below |
1267 | * deletion can not fail. | |
1da177e4 | 1268 | */ |
e6c42c29 | 1269 | posix_cpu_timer_del(&timer); |
1da177e4 LT |
1270 | spin_unlock_irq(&timer.it_lock); |
1271 | return 0; | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * Block until cpu_timer_fire (or a signal) wakes us. | |
1276 | */ | |
1277 | __set_current_state(TASK_INTERRUPTIBLE); | |
1278 | spin_unlock_irq(&timer.it_lock); | |
1279 | schedule(); | |
1280 | spin_lock_irq(&timer.it_lock); | |
1281 | } | |
1282 | ||
1283 | /* | |
1284 | * We were interrupted by a signal. | |
1285 | */ | |
343d8fc2 | 1286 | expires = timer.it.cpu.expires; |
86a9c446 | 1287 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
e6c42c29 SG |
1288 | if (!error) { |
1289 | /* | |
1290 | * Timer is now unarmed, deletion can not fail. | |
1291 | */ | |
1292 | posix_cpu_timer_del(&timer); | |
1293 | } | |
1da177e4 LT |
1294 | spin_unlock_irq(&timer.it_lock); |
1295 | ||
e6c42c29 SG |
1296 | while (error == TIMER_RETRY) { |
1297 | /* | |
1298 | * We need to handle case when timer was or is in the | |
1299 | * middle of firing. In other cases we already freed | |
1300 | * resources. | |
1301 | */ | |
1302 | spin_lock_irq(&timer.it_lock); | |
1303 | error = posix_cpu_timer_del(&timer); | |
1304 | spin_unlock_irq(&timer.it_lock); | |
1305 | } | |
1306 | ||
86a9c446 | 1307 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
1da177e4 LT |
1308 | /* |
1309 | * It actually did fire already. | |
1310 | */ | |
1311 | return 0; | |
1312 | } | |
1313 | ||
e4b76555 | 1314 | error = -ERESTART_RESTARTBLOCK; |
86a9c446 AV |
1315 | /* |
1316 | * Report back to the user the time still remaining. | |
1317 | */ | |
edbeda46 | 1318 | restart = ¤t->restart_block; |
343d8fc2 | 1319 | restart->nanosleep.expires = expires; |
c0edd7c9 DD |
1320 | if (restart->nanosleep.type != TT_NONE) |
1321 | error = nanosleep_copyout(restart, &it.it_value); | |
e4b76555 TA |
1322 | } |
1323 | ||
1324 | return error; | |
1325 | } | |
1326 | ||
bc2c8ea4 TG |
1327 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
1328 | ||
1329 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, | |
938e7cf2 | 1330 | const struct timespec64 *rqtp) |
e4b76555 | 1331 | { |
f56141e3 | 1332 | struct restart_block *restart_block = ¤t->restart_block; |
e4b76555 TA |
1333 | int error; |
1334 | ||
1335 | /* | |
1336 | * Diagnose required errors first. | |
1337 | */ | |
1338 | if (CPUCLOCK_PERTHREAD(which_clock) && | |
1339 | (CPUCLOCK_PID(which_clock) == 0 || | |
01a21974 | 1340 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
e4b76555 TA |
1341 | return -EINVAL; |
1342 | ||
86a9c446 | 1343 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
e4b76555 TA |
1344 | |
1345 | if (error == -ERESTART_RESTARTBLOCK) { | |
1346 | ||
3751f9f2 | 1347 | if (flags & TIMER_ABSTIME) |
e4b76555 | 1348 | return -ERESTARTNOHAND; |
1da177e4 | 1349 | |
1711ef38 | 1350 | restart_block->fn = posix_cpu_nsleep_restart; |
ab8177bc | 1351 | restart_block->nanosleep.clockid = which_clock; |
1da177e4 | 1352 | } |
1da177e4 LT |
1353 | return error; |
1354 | } | |
1355 | ||
bc2c8ea4 | 1356 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1da177e4 | 1357 | { |
ab8177bc | 1358 | clockid_t which_clock = restart_block->nanosleep.clockid; |
ad196384 | 1359 | struct timespec64 t; |
97735f25 | 1360 | |
ad196384 | 1361 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
97735f25 | 1362 | |
86a9c446 | 1363 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
1da177e4 LT |
1364 | } |
1365 | ||
1da177e4 LT |
1366 | #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED) |
1367 | #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED) | |
1368 | ||
a924b04d | 1369 | static int process_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1370 | struct timespec64 *tp) |
1da177e4 LT |
1371 | { |
1372 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); | |
1373 | } | |
a924b04d | 1374 | static int process_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1375 | struct timespec64 *tp) |
1da177e4 LT |
1376 | { |
1377 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); | |
1378 | } | |
1379 | static int process_cpu_timer_create(struct k_itimer *timer) | |
1380 | { | |
1381 | timer->it_clock = PROCESS_CLOCK; | |
1382 | return posix_cpu_timer_create(timer); | |
1383 | } | |
a924b04d | 1384 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
938e7cf2 | 1385 | const struct timespec64 *rqtp) |
1da177e4 | 1386 | { |
99e6c0e6 | 1387 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
1da177e4 | 1388 | } |
a924b04d | 1389 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1390 | struct timespec64 *tp) |
1da177e4 LT |
1391 | { |
1392 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); | |
1393 | } | |
a924b04d | 1394 | static int thread_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1395 | struct timespec64 *tp) |
1da177e4 LT |
1396 | { |
1397 | return posix_cpu_clock_get(THREAD_CLOCK, tp); | |
1398 | } | |
1399 | static int thread_cpu_timer_create(struct k_itimer *timer) | |
1400 | { | |
1401 | timer->it_clock = THREAD_CLOCK; | |
1402 | return posix_cpu_timer_create(timer); | |
1403 | } | |
1da177e4 | 1404 | |
d3ba5a9a | 1405 | const struct k_clock clock_posix_cpu = { |
1976945e TG |
1406 | .clock_getres = posix_cpu_clock_getres, |
1407 | .clock_set = posix_cpu_clock_set, | |
1408 | .clock_get = posix_cpu_clock_get, | |
1409 | .timer_create = posix_cpu_timer_create, | |
1410 | .nsleep = posix_cpu_nsleep, | |
1976945e TG |
1411 | .timer_set = posix_cpu_timer_set, |
1412 | .timer_del = posix_cpu_timer_del, | |
1413 | .timer_get = posix_cpu_timer_get, | |
f37fb0aa | 1414 | .timer_rearm = posix_cpu_timer_rearm, |
1976945e TG |
1415 | }; |
1416 | ||
d3ba5a9a CH |
1417 | const struct k_clock clock_process = { |
1418 | .clock_getres = process_cpu_clock_getres, | |
1419 | .clock_get = process_cpu_clock_get, | |
1420 | .timer_create = process_cpu_timer_create, | |
1421 | .nsleep = process_cpu_nsleep, | |
d3ba5a9a | 1422 | }; |
1da177e4 | 1423 | |
d3ba5a9a CH |
1424 | const struct k_clock clock_thread = { |
1425 | .clock_getres = thread_cpu_clock_getres, | |
1426 | .clock_get = thread_cpu_clock_get, | |
1427 | .timer_create = thread_cpu_timer_create, | |
1428 | }; |