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> |
34be3930 | 17 | #include <linux/sched/deadline.h> |
1da177e4 | 18 | |
bab0aae9 TG |
19 | #include "posix-timers.h" |
20 | ||
f37fb0aa TG |
21 | static void posix_cpu_timer_rearm(struct k_itimer *timer); |
22 | ||
f06febc9 | 23 | /* |
f55db609 SG |
24 | * Called after updating RLIMIT_CPU to run cpu timer and update |
25 | * tsk->signal->cputime_expires expiration cache if necessary. Needs | |
26 | * siglock protection since other code may update expiration cache as | |
27 | * well. | |
f06febc9 | 28 | */ |
5ab46b34 | 29 | void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new) |
f06febc9 | 30 | { |
858cf3a8 | 31 | u64 nsecs = rlim_new * NSEC_PER_SEC; |
f06febc9 | 32 | |
5ab46b34 | 33 | spin_lock_irq(&task->sighand->siglock); |
858cf3a8 | 34 | set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL); |
5ab46b34 | 35 | spin_unlock_irq(&task->sighand->siglock); |
f06febc9 FM |
36 | } |
37 | ||
6ae40e3f TG |
38 | /* |
39 | * Functions for validating access to tasks. | |
40 | */ | |
41 | static struct task_struct *lookup_task(const pid_t pid, bool thread) | |
1da177e4 | 42 | { |
1da177e4 | 43 | struct task_struct *p; |
1da177e4 | 44 | |
6ae40e3f TG |
45 | if (!pid) |
46 | return thread ? current : current->group_leader; | |
1da177e4 | 47 | |
6ae40e3f TG |
48 | p = find_task_by_vpid(pid); |
49 | if (!p || p == current) | |
50 | return p; | |
51 | if (thread) | |
52 | return same_thread_group(p, current) ? p : NULL; | |
53 | if (p == current) | |
54 | return p; | |
55 | return has_group_leader_pid(p) ? p : NULL; | |
56 | } | |
57 | ||
58 | static struct task_struct *__get_task_for_clock(const clockid_t clock, | |
59 | bool getref) | |
60 | { | |
61 | const bool thread = !!CPUCLOCK_PERTHREAD(clock); | |
62 | const pid_t pid = CPUCLOCK_PID(clock); | |
63 | struct task_struct *p; | |
64 | ||
65 | if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX) | |
66 | return NULL; | |
1da177e4 | 67 | |
c0deae8c | 68 | rcu_read_lock(); |
6ae40e3f TG |
69 | p = lookup_task(pid, thread); |
70 | if (p && getref) | |
71 | get_task_struct(p); | |
c0deae8c | 72 | rcu_read_unlock(); |
6ae40e3f TG |
73 | return p; |
74 | } | |
1da177e4 | 75 | |
6ae40e3f TG |
76 | static inline struct task_struct *get_task_for_clock(const clockid_t clock) |
77 | { | |
78 | return __get_task_for_clock(clock, true); | |
79 | } | |
80 | ||
81 | static inline int validate_clock_permissions(const clockid_t clock) | |
82 | { | |
83 | return __get_task_for_clock(clock, false) ? 0 : -EINVAL; | |
1da177e4 LT |
84 | } |
85 | ||
1da177e4 LT |
86 | /* |
87 | * Update expiry time from increment, and increase overrun count, | |
88 | * given the current clock sample. | |
89 | */ | |
ebd7e7fc | 90 | static void bump_cpu_timer(struct k_itimer *timer, u64 now) |
1da177e4 LT |
91 | { |
92 | int i; | |
ebd7e7fc | 93 | u64 delta, incr; |
1da177e4 | 94 | |
16118794 | 95 | if (!timer->it_interval) |
1da177e4 LT |
96 | return; |
97 | ||
55ccb616 FW |
98 | if (now < timer->it.cpu.expires) |
99 | return; | |
1da177e4 | 100 | |
16118794 | 101 | incr = timer->it_interval; |
55ccb616 | 102 | delta = now + incr - timer->it.cpu.expires; |
1da177e4 | 103 | |
55ccb616 FW |
104 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
105 | for (i = 0; incr < delta - incr; i++) | |
106 | incr = incr << 1; | |
107 | ||
108 | for (; i >= 0; incr >>= 1, i--) { | |
109 | if (delta < incr) | |
110 | continue; | |
111 | ||
112 | timer->it.cpu.expires += incr; | |
78c9c4df | 113 | timer->it_overrun += 1LL << i; |
55ccb616 | 114 | delta -= incr; |
1da177e4 LT |
115 | } |
116 | } | |
117 | ||
555347f6 FW |
118 | /** |
119 | * task_cputime_zero - Check a task_cputime struct for all zero fields. | |
120 | * | |
121 | * @cputime: The struct to compare. | |
122 | * | |
123 | * Checks @cputime to see if all fields are zero. Returns true if all fields | |
124 | * are zero, false if any field is nonzero. | |
125 | */ | |
ebd7e7fc | 126 | static inline int task_cputime_zero(const struct task_cputime *cputime) |
555347f6 FW |
127 | { |
128 | if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime) | |
129 | return 1; | |
130 | return 0; | |
131 | } | |
132 | ||
ebd7e7fc | 133 | static inline u64 prof_ticks(struct task_struct *p) |
1da177e4 | 134 | { |
ebd7e7fc | 135 | u64 utime, stime; |
6fac4829 | 136 | |
ebd7e7fc | 137 | task_cputime(p, &utime, &stime); |
6fac4829 | 138 | |
ebd7e7fc | 139 | return utime + stime; |
1da177e4 | 140 | } |
ebd7e7fc | 141 | static inline u64 virt_ticks(struct task_struct *p) |
1da177e4 | 142 | { |
ebd7e7fc | 143 | u64 utime, stime; |
6fac4829 | 144 | |
ebd7e7fc | 145 | task_cputime(p, &utime, &stime); |
6fac4829 | 146 | |
ebd7e7fc | 147 | return utime; |
1da177e4 | 148 | } |
1da177e4 | 149 | |
bc2c8ea4 | 150 | static int |
d2e3e0ca | 151 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp) |
1da177e4 | 152 | { |
6ae40e3f TG |
153 | int error = validate_clock_permissions(which_clock); |
154 | ||
1da177e4 LT |
155 | if (!error) { |
156 | tp->tv_sec = 0; | |
157 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); | |
158 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { | |
159 | /* | |
160 | * If sched_clock is using a cycle counter, we | |
161 | * don't have any idea of its true resolution | |
162 | * exported, but it is much more than 1s/HZ. | |
163 | */ | |
164 | tp->tv_nsec = 1; | |
165 | } | |
166 | } | |
167 | return error; | |
168 | } | |
169 | ||
bc2c8ea4 | 170 | static int |
6ae40e3f | 171 | posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp) |
1da177e4 | 172 | { |
6ae40e3f TG |
173 | int error = validate_clock_permissions(clock); |
174 | ||
1da177e4 LT |
175 | /* |
176 | * You can never reset a CPU clock, but we check for other errors | |
177 | * in the call before failing with EPERM. | |
178 | */ | |
6ae40e3f | 179 | return error ? : -EPERM; |
1da177e4 LT |
180 | } |
181 | ||
1da177e4 LT |
182 | /* |
183 | * Sample a per-thread clock for the given task. | |
184 | */ | |
ebd7e7fc FW |
185 | static int cpu_clock_sample(const clockid_t which_clock, |
186 | struct task_struct *p, u64 *sample) | |
1da177e4 LT |
187 | { |
188 | switch (CPUCLOCK_WHICH(which_clock)) { | |
189 | default: | |
190 | return -EINVAL; | |
191 | case CPUCLOCK_PROF: | |
55ccb616 | 192 | *sample = prof_ticks(p); |
1da177e4 LT |
193 | break; |
194 | case CPUCLOCK_VIRT: | |
55ccb616 | 195 | *sample = virt_ticks(p); |
1da177e4 LT |
196 | break; |
197 | case CPUCLOCK_SCHED: | |
55ccb616 | 198 | *sample = task_sched_runtime(p); |
1da177e4 LT |
199 | break; |
200 | } | |
201 | return 0; | |
202 | } | |
203 | ||
1018016c JL |
204 | /* |
205 | * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg | |
206 | * to avoid race conditions with concurrent updates to cputime. | |
207 | */ | |
208 | static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime) | |
4da94d49 | 209 | { |
1018016c JL |
210 | u64 curr_cputime; |
211 | retry: | |
212 | curr_cputime = atomic64_read(cputime); | |
213 | if (sum_cputime > curr_cputime) { | |
214 | if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime) | |
215 | goto retry; | |
216 | } | |
217 | } | |
4da94d49 | 218 | |
ebd7e7fc | 219 | static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum) |
1018016c | 220 | { |
71107445 JL |
221 | __update_gt_cputime(&cputime_atomic->utime, sum->utime); |
222 | __update_gt_cputime(&cputime_atomic->stime, sum->stime); | |
223 | __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime); | |
1018016c | 224 | } |
4da94d49 | 225 | |
71107445 | 226 | /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */ |
ebd7e7fc | 227 | static inline void sample_cputime_atomic(struct task_cputime *times, |
71107445 | 228 | struct task_cputime_atomic *atomic_times) |
1018016c | 229 | { |
71107445 JL |
230 | times->utime = atomic64_read(&atomic_times->utime); |
231 | times->stime = atomic64_read(&atomic_times->stime); | |
232 | times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime); | |
4da94d49 PZ |
233 | } |
234 | ||
19298fbf TG |
235 | /** |
236 | * thread_group_sample_cputime - Sample cputime for a given task | |
237 | * @tsk: Task for which cputime needs to be started | |
238 | * @iimes: Storage for time samples | |
239 | * | |
240 | * Called from sys_getitimer() to calculate the expiry time of an active | |
241 | * timer. That means group cputime accounting is already active. Called | |
242 | * with task sighand lock held. | |
243 | * | |
244 | * Updates @times with an uptodate sample of the thread group cputimes. | |
245 | */ | |
246 | void thread_group_sample_cputime(struct task_struct *tsk, | |
247 | struct task_cputime *times) | |
248 | { | |
249 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
250 | ||
251 | WARN_ON_ONCE(!cputimer->running); | |
252 | ||
253 | sample_cputime_atomic(times, &cputimer->cputime_atomic); | |
254 | } | |
255 | ||
ebd7e7fc | 256 | void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times) |
4da94d49 PZ |
257 | { |
258 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
ebd7e7fc | 259 | struct task_cputime sum; |
4da94d49 | 260 | |
1018016c JL |
261 | /* Check if cputimer isn't running. This is accessed without locking. */ |
262 | if (!READ_ONCE(cputimer->running)) { | |
4da94d49 PZ |
263 | /* |
264 | * The POSIX timer interface allows for absolute time expiry | |
265 | * values through the TIMER_ABSTIME flag, therefore we have | |
1018016c | 266 | * to synchronize the timer to the clock every time we start it. |
4da94d49 | 267 | */ |
ebd7e7fc | 268 | thread_group_cputime(tsk, &sum); |
71107445 | 269 | update_gt_cputime(&cputimer->cputime_atomic, &sum); |
1018016c JL |
270 | |
271 | /* | |
272 | * We're setting cputimer->running without a lock. Ensure | |
273 | * this only gets written to in one operation. We set | |
274 | * running after update_gt_cputime() as a small optimization, | |
275 | * but barriers are not required because update_gt_cputime() | |
276 | * can handle concurrent updates. | |
277 | */ | |
d5c373eb | 278 | WRITE_ONCE(cputimer->running, true); |
1018016c | 279 | } |
71107445 | 280 | sample_cputime_atomic(times, &cputimer->cputime_atomic); |
4da94d49 PZ |
281 | } |
282 | ||
1da177e4 LT |
283 | /* |
284 | * Sample a process (thread group) clock for the given group_leader task. | |
e73d84e3 FW |
285 | * Must be called with task sighand lock held for safe while_each_thread() |
286 | * traversal. | |
1da177e4 | 287 | */ |
bb34d92f FM |
288 | static int cpu_clock_sample_group(const clockid_t which_clock, |
289 | struct task_struct *p, | |
ebd7e7fc | 290 | u64 *sample) |
1da177e4 | 291 | { |
ebd7e7fc | 292 | struct task_cputime cputime; |
f06febc9 | 293 | |
eccdaeaf | 294 | switch (CPUCLOCK_WHICH(which_clock)) { |
1da177e4 LT |
295 | default: |
296 | return -EINVAL; | |
297 | case CPUCLOCK_PROF: | |
ebd7e7fc FW |
298 | thread_group_cputime(p, &cputime); |
299 | *sample = cputime.utime + cputime.stime; | |
1da177e4 LT |
300 | break; |
301 | case CPUCLOCK_VIRT: | |
ebd7e7fc FW |
302 | thread_group_cputime(p, &cputime); |
303 | *sample = cputime.utime; | |
1da177e4 LT |
304 | break; |
305 | case CPUCLOCK_SCHED: | |
ebd7e7fc | 306 | thread_group_cputime(p, &cputime); |
55ccb616 | 307 | *sample = cputime.sum_exec_runtime; |
1da177e4 LT |
308 | break; |
309 | } | |
310 | return 0; | |
311 | } | |
312 | ||
bfcf3e92 | 313 | static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp) |
33ab0fec | 314 | { |
bfcf3e92 TG |
315 | const clockid_t clkid = CPUCLOCK_WHICH(clock); |
316 | struct task_struct *tsk; | |
317 | u64 t; | |
33ab0fec | 318 | |
bfcf3e92 TG |
319 | tsk = get_task_for_clock(clock); |
320 | if (!tsk) | |
321 | return -EINVAL; | |
1da177e4 | 322 | |
bfcf3e92 TG |
323 | if (CPUCLOCK_PERTHREAD(clock)) |
324 | cpu_clock_sample(clkid, tsk, &t); | |
325 | else | |
326 | cpu_clock_sample_group(clkid, tsk, &t); | |
327 | put_task_struct(tsk); | |
1da177e4 | 328 | |
bfcf3e92 TG |
329 | *tp = ns_to_timespec64(t); |
330 | return 0; | |
1da177e4 LT |
331 | } |
332 | ||
1da177e4 LT |
333 | /* |
334 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. | |
ba5ea951 SG |
335 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
336 | * new timer already all-zeros initialized. | |
1da177e4 | 337 | */ |
bc2c8ea4 | 338 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
1da177e4 | 339 | { |
e5a8b65b | 340 | struct task_struct *p = get_task_for_clock(new_timer->it_clock); |
1da177e4 | 341 | |
e5a8b65b | 342 | if (!p) |
1da177e4 LT |
343 | return -EINVAL; |
344 | ||
d97bb75d | 345 | new_timer->kclock = &clock_posix_cpu; |
1da177e4 | 346 | INIT_LIST_HEAD(&new_timer->it.cpu.entry); |
1da177e4 | 347 | new_timer->it.cpu.task = p; |
e5a8b65b | 348 | return 0; |
1da177e4 LT |
349 | } |
350 | ||
351 | /* | |
352 | * Clean up a CPU-clock timer that is about to be destroyed. | |
353 | * This is called from timer deletion with the timer already locked. | |
354 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
355 | * and try again. (This happens when the timer is in the middle of firing.) | |
356 | */ | |
bc2c8ea4 | 357 | static int posix_cpu_timer_del(struct k_itimer *timer) |
1da177e4 | 358 | { |
108150ea | 359 | int ret = 0; |
3d7a1427 FW |
360 | unsigned long flags; |
361 | struct sighand_struct *sighand; | |
362 | struct task_struct *p = timer->it.cpu.task; | |
1da177e4 | 363 | |
692117c1 TG |
364 | if (WARN_ON_ONCE(!p)) |
365 | return -EINVAL; | |
108150ea | 366 | |
3d7a1427 FW |
367 | /* |
368 | * Protect against sighand release/switch in exit/exec and process/ | |
369 | * thread timer list entry concurrent read/writes. | |
370 | */ | |
371 | sighand = lock_task_sighand(p, &flags); | |
372 | if (unlikely(sighand == NULL)) { | |
a3222f88 FW |
373 | /* |
374 | * We raced with the reaping of the task. | |
375 | * The deletion should have cleared us off the list. | |
376 | */ | |
531f64fd | 377 | WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry)); |
a3222f88 | 378 | } else { |
a3222f88 FW |
379 | if (timer->it.cpu.firing) |
380 | ret = TIMER_RETRY; | |
381 | else | |
382 | list_del(&timer->it.cpu.entry); | |
3d7a1427 FW |
383 | |
384 | unlock_task_sighand(p, &flags); | |
1da177e4 | 385 | } |
a3222f88 FW |
386 | |
387 | if (!ret) | |
388 | put_task_struct(p); | |
1da177e4 | 389 | |
108150ea | 390 | return ret; |
1da177e4 LT |
391 | } |
392 | ||
af82eb3c | 393 | static void cleanup_timers_list(struct list_head *head) |
1a7fa510 FW |
394 | { |
395 | struct cpu_timer_list *timer, *next; | |
396 | ||
a0b2062b | 397 | list_for_each_entry_safe(timer, next, head, entry) |
1a7fa510 | 398 | list_del_init(&timer->entry); |
1a7fa510 FW |
399 | } |
400 | ||
1da177e4 | 401 | /* |
7cb9a94c TG |
402 | * Clean out CPU timers which are still armed when a thread exits. The |
403 | * timers are only removed from the list. No other updates are done. The | |
404 | * corresponding posix timers are still accessible, but cannot be rearmed. | |
405 | * | |
1da177e4 LT |
406 | * This must be called with the siglock held. |
407 | */ | |
af82eb3c | 408 | static void cleanup_timers(struct list_head *head) |
1da177e4 | 409 | { |
af82eb3c FW |
410 | cleanup_timers_list(head); |
411 | cleanup_timers_list(++head); | |
412 | cleanup_timers_list(++head); | |
1da177e4 LT |
413 | } |
414 | ||
415 | /* | |
416 | * These are both called with the siglock held, when the current thread | |
417 | * is being reaped. When the final (leader) thread in the group is reaped, | |
418 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. | |
419 | */ | |
420 | void posix_cpu_timers_exit(struct task_struct *tsk) | |
421 | { | |
af82eb3c | 422 | cleanup_timers(tsk->cpu_timers); |
1da177e4 LT |
423 | } |
424 | void posix_cpu_timers_exit_group(struct task_struct *tsk) | |
425 | { | |
af82eb3c | 426 | cleanup_timers(tsk->signal->cpu_timers); |
1da177e4 LT |
427 | } |
428 | ||
ebd7e7fc | 429 | static inline int expires_gt(u64 expires, u64 new_exp) |
d1e3b6d1 | 430 | { |
64861634 | 431 | return expires == 0 || expires > new_exp; |
d1e3b6d1 SG |
432 | } |
433 | ||
1da177e4 LT |
434 | /* |
435 | * Insert the timer on the appropriate list before any timers that | |
e73d84e3 | 436 | * expire later. This must be called with the sighand lock held. |
1da177e4 | 437 | */ |
5eb9aa64 | 438 | static void arm_timer(struct k_itimer *timer) |
1da177e4 LT |
439 | { |
440 | struct task_struct *p = timer->it.cpu.task; | |
441 | struct list_head *head, *listpos; | |
ebd7e7fc | 442 | struct task_cputime *cputime_expires; |
1da177e4 LT |
443 | struct cpu_timer_list *const nt = &timer->it.cpu; |
444 | struct cpu_timer_list *next; | |
1da177e4 | 445 | |
5eb9aa64 SG |
446 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
447 | head = p->cpu_timers; | |
448 | cputime_expires = &p->cputime_expires; | |
449 | } else { | |
450 | head = p->signal->cpu_timers; | |
451 | cputime_expires = &p->signal->cputime_expires; | |
452 | } | |
1da177e4 LT |
453 | head += CPUCLOCK_WHICH(timer->it_clock); |
454 | ||
1da177e4 | 455 | listpos = head; |
5eb9aa64 | 456 | list_for_each_entry(next, head, entry) { |
55ccb616 | 457 | if (nt->expires < next->expires) |
5eb9aa64 SG |
458 | break; |
459 | listpos = &next->entry; | |
1da177e4 LT |
460 | } |
461 | list_add(&nt->entry, listpos); | |
462 | ||
463 | if (listpos == head) { | |
ebd7e7fc | 464 | u64 exp = nt->expires; |
5eb9aa64 | 465 | |
1da177e4 | 466 | /* |
5eb9aa64 SG |
467 | * We are the new earliest-expiring POSIX 1.b timer, hence |
468 | * need to update expiration cache. Take into account that | |
469 | * for process timers we share expiration cache with itimers | |
470 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. | |
1da177e4 LT |
471 | */ |
472 | ||
5eb9aa64 SG |
473 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
474 | case CPUCLOCK_PROF: | |
ebd7e7fc FW |
475 | if (expires_gt(cputime_expires->prof_exp, exp)) |
476 | cputime_expires->prof_exp = exp; | |
5eb9aa64 SG |
477 | break; |
478 | case CPUCLOCK_VIRT: | |
ebd7e7fc FW |
479 | if (expires_gt(cputime_expires->virt_exp, exp)) |
480 | cputime_expires->virt_exp = exp; | |
5eb9aa64 SG |
481 | break; |
482 | case CPUCLOCK_SCHED: | |
ebd7e7fc | 483 | if (expires_gt(cputime_expires->sched_exp, exp)) |
55ccb616 | 484 | cputime_expires->sched_exp = exp; |
5eb9aa64 | 485 | break; |
1da177e4 | 486 | } |
b7878300 FW |
487 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
488 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); | |
489 | else | |
490 | tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 | 491 | } |
1da177e4 LT |
492 | } |
493 | ||
494 | /* | |
495 | * The timer is locked, fire it and arrange for its reload. | |
496 | */ | |
497 | static void cpu_timer_fire(struct k_itimer *timer) | |
498 | { | |
1f169f84 SG |
499 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
500 | /* | |
501 | * User don't want any signal. | |
502 | */ | |
55ccb616 | 503 | timer->it.cpu.expires = 0; |
1f169f84 | 504 | } else if (unlikely(timer->sigq == NULL)) { |
1da177e4 LT |
505 | /* |
506 | * This a special case for clock_nanosleep, | |
507 | * not a normal timer from sys_timer_create. | |
508 | */ | |
509 | wake_up_process(timer->it_process); | |
55ccb616 | 510 | timer->it.cpu.expires = 0; |
16118794 | 511 | } else if (!timer->it_interval) { |
1da177e4 LT |
512 | /* |
513 | * One-shot timer. Clear it as soon as it's fired. | |
514 | */ | |
515 | posix_timer_event(timer, 0); | |
55ccb616 | 516 | timer->it.cpu.expires = 0; |
1da177e4 LT |
517 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
518 | /* | |
519 | * The signal did not get queued because the signal | |
520 | * was ignored, so we won't get any callback to | |
521 | * reload the timer. But we need to keep it | |
522 | * ticking in case the signal is deliverable next time. | |
523 | */ | |
f37fb0aa | 524 | posix_cpu_timer_rearm(timer); |
af888d67 | 525 | ++timer->it_requeue_pending; |
1da177e4 LT |
526 | } |
527 | } | |
528 | ||
3997ad31 PZ |
529 | /* |
530 | * Sample a process (thread group) timer for the given group_leader task. | |
e73d84e3 FW |
531 | * Must be called with task sighand lock held for safe while_each_thread() |
532 | * traversal. | |
3997ad31 PZ |
533 | */ |
534 | static int cpu_timer_sample_group(const clockid_t which_clock, | |
ebd7e7fc | 535 | struct task_struct *p, u64 *sample) |
3997ad31 | 536 | { |
ebd7e7fc | 537 | struct task_cputime cputime; |
3997ad31 PZ |
538 | |
539 | thread_group_cputimer(p, &cputime); | |
540 | switch (CPUCLOCK_WHICH(which_clock)) { | |
541 | default: | |
542 | return -EINVAL; | |
543 | case CPUCLOCK_PROF: | |
ebd7e7fc | 544 | *sample = cputime.utime + cputime.stime; |
3997ad31 PZ |
545 | break; |
546 | case CPUCLOCK_VIRT: | |
ebd7e7fc | 547 | *sample = cputime.utime; |
3997ad31 PZ |
548 | break; |
549 | case CPUCLOCK_SCHED: | |
23cfa361 | 550 | *sample = cputime.sum_exec_runtime; |
3997ad31 PZ |
551 | break; |
552 | } | |
553 | return 0; | |
554 | } | |
555 | ||
1da177e4 LT |
556 | /* |
557 | * Guts of sys_timer_settime for CPU timers. | |
558 | * This is called with the timer locked and interrupts disabled. | |
559 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
560 | * and try again. (This happens when the timer is in the middle of firing.) | |
561 | */ | |
e73d84e3 | 562 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
5f252b32 | 563 | struct itimerspec64 *new, struct itimerspec64 *old) |
1da177e4 | 564 | { |
e73d84e3 FW |
565 | unsigned long flags; |
566 | struct sighand_struct *sighand; | |
1da177e4 | 567 | struct task_struct *p = timer->it.cpu.task; |
ebd7e7fc | 568 | u64 old_expires, new_expires, old_incr, val; |
1da177e4 LT |
569 | int ret; |
570 | ||
692117c1 TG |
571 | if (WARN_ON_ONCE(!p)) |
572 | return -EINVAL; | |
1da177e4 | 573 | |
098b0e01 TG |
574 | /* |
575 | * Use the to_ktime conversion because that clamps the maximum | |
576 | * value to KTIME_MAX and avoid multiplication overflows. | |
577 | */ | |
578 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); | |
1da177e4 | 579 | |
1da177e4 | 580 | /* |
e73d84e3 FW |
581 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
582 | * and p->signal->cpu_timers read/write in arm_timer() | |
583 | */ | |
584 | sighand = lock_task_sighand(p, &flags); | |
585 | /* | |
586 | * If p has just been reaped, we can no | |
1da177e4 LT |
587 | * longer get any information about it at all. |
588 | */ | |
e73d84e3 | 589 | if (unlikely(sighand == NULL)) { |
1da177e4 LT |
590 | return -ESRCH; |
591 | } | |
592 | ||
593 | /* | |
594 | * Disarm any old timer after extracting its expiry time. | |
595 | */ | |
a69ac4a7 ON |
596 | |
597 | ret = 0; | |
16118794 | 598 | old_incr = timer->it_interval; |
1da177e4 | 599 | old_expires = timer->it.cpu.expires; |
a69ac4a7 ON |
600 | if (unlikely(timer->it.cpu.firing)) { |
601 | timer->it.cpu.firing = -1; | |
602 | ret = TIMER_RETRY; | |
603 | } else | |
604 | list_del_init(&timer->it.cpu.entry); | |
1da177e4 LT |
605 | |
606 | /* | |
607 | * We need to sample the current value to convert the new | |
608 | * value from to relative and absolute, and to convert the | |
609 | * old value from absolute to relative. To set a process | |
610 | * timer, we need a sample to balance the thread expiry | |
611 | * times (in arm_timer). With an absolute time, we must | |
612 | * check if it's already passed. In short, we need a sample. | |
613 | */ | |
614 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
615 | cpu_clock_sample(timer->it_clock, p, &val); | |
616 | } else { | |
3997ad31 | 617 | cpu_timer_sample_group(timer->it_clock, p, &val); |
1da177e4 LT |
618 | } |
619 | ||
620 | if (old) { | |
55ccb616 | 621 | if (old_expires == 0) { |
1da177e4 LT |
622 | old->it_value.tv_sec = 0; |
623 | old->it_value.tv_nsec = 0; | |
624 | } else { | |
625 | /* | |
626 | * Update the timer in case it has | |
627 | * overrun already. If it has, | |
628 | * we'll report it as having overrun | |
629 | * and with the next reloaded timer | |
630 | * already ticking, though we are | |
631 | * swallowing that pending | |
632 | * notification here to install the | |
633 | * new setting. | |
634 | */ | |
635 | bump_cpu_timer(timer, val); | |
55ccb616 FW |
636 | if (val < timer->it.cpu.expires) { |
637 | old_expires = timer->it.cpu.expires - val; | |
5f252b32 | 638 | old->it_value = ns_to_timespec64(old_expires); |
1da177e4 LT |
639 | } else { |
640 | old->it_value.tv_nsec = 1; | |
641 | old->it_value.tv_sec = 0; | |
642 | } | |
643 | } | |
644 | } | |
645 | ||
a69ac4a7 | 646 | if (unlikely(ret)) { |
1da177e4 LT |
647 | /* |
648 | * We are colliding with the timer actually firing. | |
649 | * Punt after filling in the timer's old value, and | |
650 | * disable this firing since we are already reporting | |
651 | * it as an overrun (thanks to bump_cpu_timer above). | |
652 | */ | |
e73d84e3 | 653 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
654 | goto out; |
655 | } | |
656 | ||
e73d84e3 | 657 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
55ccb616 | 658 | new_expires += val; |
1da177e4 LT |
659 | } |
660 | ||
661 | /* | |
662 | * Install the new expiry time (or zero). | |
663 | * For a timer with no notification action, we don't actually | |
664 | * arm the timer (we'll just fake it for timer_gettime). | |
665 | */ | |
666 | timer->it.cpu.expires = new_expires; | |
55ccb616 | 667 | if (new_expires != 0 && val < new_expires) { |
5eb9aa64 | 668 | arm_timer(timer); |
1da177e4 LT |
669 | } |
670 | ||
e73d84e3 | 671 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
672 | /* |
673 | * Install the new reload setting, and | |
674 | * set up the signal and overrun bookkeeping. | |
675 | */ | |
16118794 | 676 | timer->it_interval = timespec64_to_ktime(new->it_interval); |
1da177e4 LT |
677 | |
678 | /* | |
679 | * This acts as a modification timestamp for the timer, | |
680 | * so any automatic reload attempt will punt on seeing | |
681 | * that we have reset the timer manually. | |
682 | */ | |
683 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & | |
684 | ~REQUEUE_PENDING; | |
685 | timer->it_overrun_last = 0; | |
686 | timer->it_overrun = -1; | |
687 | ||
55ccb616 | 688 | if (new_expires != 0 && !(val < new_expires)) { |
1da177e4 LT |
689 | /* |
690 | * The designated time already passed, so we notify | |
691 | * immediately, even if the thread never runs to | |
692 | * accumulate more time on this clock. | |
693 | */ | |
694 | cpu_timer_fire(timer); | |
695 | } | |
696 | ||
697 | ret = 0; | |
698 | out: | |
ebd7e7fc | 699 | if (old) |
5f252b32 | 700 | old->it_interval = ns_to_timespec64(old_incr); |
b7878300 | 701 | |
1da177e4 LT |
702 | return ret; |
703 | } | |
704 | ||
5f252b32 | 705 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
1da177e4 | 706 | { |
1da177e4 | 707 | struct task_struct *p = timer->it.cpu.task; |
692117c1 | 708 | u64 now; |
1da177e4 | 709 | |
692117c1 TG |
710 | if (WARN_ON_ONCE(!p)) |
711 | return; | |
a3222f88 | 712 | |
1da177e4 LT |
713 | /* |
714 | * Easy part: convert the reload time. | |
715 | */ | |
16118794 | 716 | itp->it_interval = ktime_to_timespec64(timer->it_interval); |
1da177e4 | 717 | |
eabdec04 | 718 | if (!timer->it.cpu.expires) |
1da177e4 | 719 | return; |
1da177e4 | 720 | |
1da177e4 LT |
721 | /* |
722 | * Sample the clock to take the difference with the expiry time. | |
723 | */ | |
724 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
725 | cpu_clock_sample(timer->it_clock, p, &now); | |
1da177e4 | 726 | } else { |
e73d84e3 FW |
727 | struct sighand_struct *sighand; |
728 | unsigned long flags; | |
729 | ||
730 | /* | |
731 | * Protect against sighand release/switch in exit/exec and | |
732 | * also make timer sampling safe if it ends up calling | |
ebd7e7fc | 733 | * thread_group_cputime(). |
e73d84e3 FW |
734 | */ |
735 | sighand = lock_task_sighand(p, &flags); | |
736 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
737 | /* |
738 | * The process has been reaped. | |
739 | * We can't even collect a sample any more. | |
740 | * Call the timer disarmed, nothing else to do. | |
741 | */ | |
55ccb616 | 742 | timer->it.cpu.expires = 0; |
2c13ce8f | 743 | return; |
1da177e4 | 744 | } else { |
3997ad31 | 745 | cpu_timer_sample_group(timer->it_clock, p, &now); |
e73d84e3 | 746 | unlock_task_sighand(p, &flags); |
1da177e4 | 747 | } |
1da177e4 LT |
748 | } |
749 | ||
55ccb616 | 750 | if (now < timer->it.cpu.expires) { |
5f252b32 | 751 | itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now); |
1da177e4 LT |
752 | } else { |
753 | /* | |
754 | * The timer should have expired already, but the firing | |
755 | * hasn't taken place yet. Say it's just about to expire. | |
756 | */ | |
757 | itp->it_value.tv_nsec = 1; | |
758 | itp->it_value.tv_sec = 0; | |
759 | } | |
760 | } | |
761 | ||
2473f3e7 FW |
762 | static unsigned long long |
763 | check_timers_list(struct list_head *timers, | |
764 | struct list_head *firing, | |
765 | unsigned long long curr) | |
766 | { | |
767 | int maxfire = 20; | |
768 | ||
769 | while (!list_empty(timers)) { | |
770 | struct cpu_timer_list *t; | |
771 | ||
772 | t = list_first_entry(timers, struct cpu_timer_list, entry); | |
773 | ||
774 | if (!--maxfire || curr < t->expires) | |
775 | return t->expires; | |
776 | ||
777 | t->firing = 1; | |
778 | list_move_tail(&t->entry, firing); | |
779 | } | |
780 | ||
781 | return 0; | |
782 | } | |
783 | ||
34be3930 JL |
784 | static inline void check_dl_overrun(struct task_struct *tsk) |
785 | { | |
786 | if (tsk->dl.dl_overrun) { | |
787 | tsk->dl.dl_overrun = 0; | |
788 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); | |
789 | } | |
790 | } | |
791 | ||
1da177e4 LT |
792 | /* |
793 | * Check for any per-thread CPU timers that have fired and move them off | |
794 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the | |
795 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. | |
796 | */ | |
797 | static void check_thread_timers(struct task_struct *tsk, | |
798 | struct list_head *firing) | |
799 | { | |
800 | struct list_head *timers = tsk->cpu_timers; | |
ebd7e7fc FW |
801 | struct task_cputime *tsk_expires = &tsk->cputime_expires; |
802 | u64 expires; | |
d4bb5274 | 803 | unsigned long soft; |
1da177e4 | 804 | |
34be3930 JL |
805 | if (dl_task(tsk)) |
806 | check_dl_overrun(tsk); | |
807 | ||
934715a1 JL |
808 | /* |
809 | * If cputime_expires is zero, then there are no active | |
810 | * per thread CPU timers. | |
811 | */ | |
812 | if (task_cputime_zero(&tsk->cputime_expires)) | |
813 | return; | |
814 | ||
2473f3e7 | 815 | expires = check_timers_list(timers, firing, prof_ticks(tsk)); |
ebd7e7fc | 816 | tsk_expires->prof_exp = expires; |
1da177e4 | 817 | |
2473f3e7 | 818 | expires = check_timers_list(++timers, firing, virt_ticks(tsk)); |
ebd7e7fc | 819 | tsk_expires->virt_exp = expires; |
1da177e4 | 820 | |
2473f3e7 FW |
821 | tsk_expires->sched_exp = check_timers_list(++timers, firing, |
822 | tsk->se.sum_exec_runtime); | |
78f2c7db PZ |
823 | |
824 | /* | |
825 | * Check for the special case thread timers. | |
826 | */ | |
3cf29496 | 827 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
d4bb5274 | 828 | if (soft != RLIM_INFINITY) { |
3cf29496 | 829 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
78f2c7db | 830 | |
5a52dd50 PZ |
831 | if (hard != RLIM_INFINITY && |
832 | tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) { | |
78f2c7db PZ |
833 | /* |
834 | * At the hard limit, we just die. | |
835 | * No need to calculate anything else now. | |
836 | */ | |
43fe8b8e TG |
837 | if (print_fatal_signals) { |
838 | pr_info("CPU Watchdog Timeout (hard): %s[%d]\n", | |
839 | tsk->comm, task_pid_nr(tsk)); | |
840 | } | |
78f2c7db PZ |
841 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
842 | return; | |
843 | } | |
d4bb5274 | 844 | if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) { |
78f2c7db PZ |
845 | /* |
846 | * At the soft limit, send a SIGXCPU every second. | |
847 | */ | |
d4bb5274 JS |
848 | if (soft < hard) { |
849 | soft += USEC_PER_SEC; | |
3cf29496 KO |
850 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = |
851 | soft; | |
78f2c7db | 852 | } |
43fe8b8e TG |
853 | if (print_fatal_signals) { |
854 | pr_info("RT Watchdog Timeout (soft): %s[%d]\n", | |
855 | tsk->comm, task_pid_nr(tsk)); | |
856 | } | |
78f2c7db PZ |
857 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
858 | } | |
859 | } | |
b7878300 FW |
860 | if (task_cputime_zero(tsk_expires)) |
861 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
862 | } |
863 | ||
1018016c | 864 | static inline void stop_process_timers(struct signal_struct *sig) |
3fccfd67 | 865 | { |
15365c10 | 866 | struct thread_group_cputimer *cputimer = &sig->cputimer; |
3fccfd67 | 867 | |
1018016c | 868 | /* Turn off cputimer->running. This is done without locking. */ |
d5c373eb | 869 | WRITE_ONCE(cputimer->running, false); |
b7878300 | 870 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
3fccfd67 PZ |
871 | } |
872 | ||
42c4ab41 | 873 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
ebd7e7fc | 874 | u64 *expires, u64 cur_time, int signo) |
42c4ab41 | 875 | { |
64861634 | 876 | if (!it->expires) |
42c4ab41 SG |
877 | return; |
878 | ||
858cf3a8 FW |
879 | if (cur_time >= it->expires) { |
880 | if (it->incr) | |
64861634 | 881 | it->expires += it->incr; |
858cf3a8 | 882 | else |
64861634 | 883 | it->expires = 0; |
42c4ab41 | 884 | |
3f0a525e XG |
885 | trace_itimer_expire(signo == SIGPROF ? |
886 | ITIMER_PROF : ITIMER_VIRTUAL, | |
6883f81a | 887 | task_tgid(tsk), cur_time); |
42c4ab41 SG |
888 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
889 | } | |
890 | ||
858cf3a8 FW |
891 | if (it->expires && (!*expires || it->expires < *expires)) |
892 | *expires = it->expires; | |
42c4ab41 SG |
893 | } |
894 | ||
1da177e4 LT |
895 | /* |
896 | * Check for any per-thread CPU timers that have fired and move them | |
897 | * off the tsk->*_timers list onto the firing list. Per-thread timers | |
898 | * have already been taken off. | |
899 | */ | |
900 | static void check_process_timers(struct task_struct *tsk, | |
901 | struct list_head *firing) | |
902 | { | |
903 | struct signal_struct *const sig = tsk->signal; | |
ebd7e7fc FW |
904 | u64 utime, ptime, virt_expires, prof_expires; |
905 | u64 sum_sched_runtime, sched_expires; | |
1da177e4 | 906 | struct list_head *timers = sig->cpu_timers; |
ebd7e7fc | 907 | struct task_cputime cputime; |
d4bb5274 | 908 | unsigned long soft; |
1da177e4 | 909 | |
934715a1 JL |
910 | /* |
911 | * If cputimer is not running, then there are no active | |
912 | * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU). | |
913 | */ | |
914 | if (!READ_ONCE(tsk->signal->cputimer.running)) | |
915 | return; | |
916 | ||
a324956f | 917 | /* |
c8d75aa4 JL |
918 | * Signify that a thread is checking for process timers. |
919 | * Write access to this field is protected by the sighand lock. | |
920 | */ | |
921 | sig->cputimer.checking_timer = true; | |
922 | ||
1da177e4 | 923 | /* |
a324956f TG |
924 | * Collect the current process totals. Group accounting is active |
925 | * so the sample can be taken directly. | |
1da177e4 | 926 | */ |
a324956f | 927 | sample_cputime_atomic(&cputime, &sig->cputimer.cputime_atomic); |
ebd7e7fc FW |
928 | utime = cputime.utime; |
929 | ptime = utime + cputime.stime; | |
f06febc9 | 930 | sum_sched_runtime = cputime.sum_exec_runtime; |
1da177e4 | 931 | |
2473f3e7 FW |
932 | prof_expires = check_timers_list(timers, firing, ptime); |
933 | virt_expires = check_timers_list(++timers, firing, utime); | |
934 | sched_expires = check_timers_list(++timers, firing, sum_sched_runtime); | |
1da177e4 LT |
935 | |
936 | /* | |
937 | * Check for the special case process timers. | |
938 | */ | |
42c4ab41 SG |
939 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime, |
940 | SIGPROF); | |
941 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime, | |
942 | SIGVTALRM); | |
3cf29496 | 943 | soft = task_rlimit(tsk, RLIMIT_CPU); |
d4bb5274 | 944 | if (soft != RLIM_INFINITY) { |
ebd7e7fc | 945 | unsigned long psecs = div_u64(ptime, NSEC_PER_SEC); |
3cf29496 | 946 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
ebd7e7fc | 947 | u64 x; |
d4bb5274 | 948 | if (psecs >= hard) { |
1da177e4 LT |
949 | /* |
950 | * At the hard limit, we just die. | |
951 | * No need to calculate anything else now. | |
952 | */ | |
43fe8b8e TG |
953 | if (print_fatal_signals) { |
954 | pr_info("RT Watchdog Timeout (hard): %s[%d]\n", | |
955 | tsk->comm, task_pid_nr(tsk)); | |
956 | } | |
1da177e4 LT |
957 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
958 | return; | |
959 | } | |
d4bb5274 | 960 | if (psecs >= soft) { |
1da177e4 LT |
961 | /* |
962 | * At the soft limit, send a SIGXCPU every second. | |
963 | */ | |
43fe8b8e TG |
964 | if (print_fatal_signals) { |
965 | pr_info("CPU Watchdog Timeout (soft): %s[%d]\n", | |
966 | tsk->comm, task_pid_nr(tsk)); | |
967 | } | |
1da177e4 | 968 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
d4bb5274 JS |
969 | if (soft < hard) { |
970 | soft++; | |
971 | sig->rlim[RLIMIT_CPU].rlim_cur = soft; | |
1da177e4 LT |
972 | } |
973 | } | |
ebd7e7fc FW |
974 | x = soft * NSEC_PER_SEC; |
975 | if (!prof_expires || x < prof_expires) | |
1da177e4 | 976 | prof_expires = x; |
1da177e4 LT |
977 | } |
978 | ||
ebd7e7fc FW |
979 | sig->cputime_expires.prof_exp = prof_expires; |
980 | sig->cputime_expires.virt_exp = virt_expires; | |
29f87b79 SG |
981 | sig->cputime_expires.sched_exp = sched_expires; |
982 | if (task_cputime_zero(&sig->cputime_expires)) | |
983 | stop_process_timers(sig); | |
c8d75aa4 JL |
984 | |
985 | sig->cputimer.checking_timer = false; | |
1da177e4 LT |
986 | } |
987 | ||
988 | /* | |
96fe3b07 | 989 | * This is called from the signal code (via posixtimer_rearm) |
1da177e4 LT |
990 | * when the last timer signal was delivered and we have to reload the timer. |
991 | */ | |
f37fb0aa | 992 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
1da177e4 | 993 | { |
692117c1 | 994 | struct task_struct *p = timer->it.cpu.task; |
e73d84e3 FW |
995 | struct sighand_struct *sighand; |
996 | unsigned long flags; | |
ebd7e7fc | 997 | u64 now; |
1da177e4 | 998 | |
692117c1 TG |
999 | if (WARN_ON_ONCE(!p)) |
1000 | return; | |
1da177e4 LT |
1001 | |
1002 | /* | |
1003 | * Fetch the current sample and update the timer's expiry time. | |
1004 | */ | |
1005 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
1006 | cpu_clock_sample(timer->it_clock, p, &now); | |
1007 | bump_cpu_timer(timer, now); | |
724a3713 | 1008 | if (unlikely(p->exit_state)) |
af888d67 | 1009 | return; |
724a3713 | 1010 | |
e73d84e3 FW |
1011 | /* Protect timer list r/w in arm_timer() */ |
1012 | sighand = lock_task_sighand(p, &flags); | |
1013 | if (!sighand) | |
af888d67 | 1014 | return; |
1da177e4 | 1015 | } else { |
e73d84e3 FW |
1016 | /* |
1017 | * Protect arm_timer() and timer sampling in case of call to | |
ebd7e7fc | 1018 | * thread_group_cputime(). |
e73d84e3 FW |
1019 | */ |
1020 | sighand = lock_task_sighand(p, &flags); | |
1021 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
1022 | /* |
1023 | * The process has been reaped. | |
1024 | * We can't even collect a sample any more. | |
1025 | */ | |
55ccb616 | 1026 | timer->it.cpu.expires = 0; |
af888d67 | 1027 | return; |
1da177e4 | 1028 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
af888d67 TG |
1029 | /* If the process is dying, no need to rearm */ |
1030 | goto unlock; | |
1da177e4 | 1031 | } |
3997ad31 | 1032 | cpu_timer_sample_group(timer->it_clock, p, &now); |
1da177e4 | 1033 | bump_cpu_timer(timer, now); |
e73d84e3 | 1034 | /* Leave the sighand locked for the call below. */ |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | /* | |
1038 | * Now re-arm for the new expiry time. | |
1039 | */ | |
5eb9aa64 | 1040 | arm_timer(timer); |
af888d67 | 1041 | unlock: |
e73d84e3 | 1042 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
1043 | } |
1044 | ||
f06febc9 FM |
1045 | /** |
1046 | * task_cputime_expired - Compare two task_cputime entities. | |
1047 | * | |
1048 | * @sample: The task_cputime structure to be checked for expiration. | |
1049 | * @expires: Expiration times, against which @sample will be checked. | |
1050 | * | |
1051 | * Checks @sample against @expires to see if any field of @sample has expired. | |
1052 | * Returns true if any field of the former is greater than the corresponding | |
1053 | * field of the latter if the latter field is set. Otherwise returns false. | |
1054 | */ | |
ebd7e7fc FW |
1055 | static inline int task_cputime_expired(const struct task_cputime *sample, |
1056 | const struct task_cputime *expires) | |
f06febc9 | 1057 | { |
64861634 | 1058 | if (expires->utime && sample->utime >= expires->utime) |
f06febc9 | 1059 | return 1; |
64861634 | 1060 | if (expires->stime && sample->utime + sample->stime >= expires->stime) |
f06febc9 FM |
1061 | return 1; |
1062 | if (expires->sum_exec_runtime != 0 && | |
1063 | sample->sum_exec_runtime >= expires->sum_exec_runtime) | |
1064 | return 1; | |
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | /** | |
1069 | * fastpath_timer_check - POSIX CPU timers fast path. | |
1070 | * | |
1071 | * @tsk: The task (thread) being checked. | |
f06febc9 | 1072 | * |
bb34d92f FM |
1073 | * Check the task and thread group timers. If both are zero (there are no |
1074 | * timers set) return false. Otherwise snapshot the task and thread group | |
1075 | * timers and compare them with the corresponding expiration times. Return | |
1076 | * true if a timer has expired, else return false. | |
f06febc9 | 1077 | */ |
bb34d92f | 1078 | static inline int fastpath_timer_check(struct task_struct *tsk) |
f06febc9 | 1079 | { |
ad133ba3 | 1080 | struct signal_struct *sig; |
bb34d92f | 1081 | |
bb34d92f | 1082 | if (!task_cputime_zero(&tsk->cputime_expires)) { |
ebd7e7fc | 1083 | struct task_cputime task_sample; |
bb34d92f | 1084 | |
ebd7e7fc | 1085 | task_cputime(tsk, &task_sample.utime, &task_sample.stime); |
7c177d99 | 1086 | task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime; |
bb34d92f FM |
1087 | if (task_cputime_expired(&task_sample, &tsk->cputime_expires)) |
1088 | return 1; | |
1089 | } | |
ad133ba3 ON |
1090 | |
1091 | sig = tsk->signal; | |
c8d75aa4 JL |
1092 | /* |
1093 | * Check if thread group timers expired when the cputimer is | |
1094 | * running and no other thread in the group is already checking | |
1095 | * for thread group cputimers. These fields are read without the | |
1096 | * sighand lock. However, this is fine because this is meant to | |
1097 | * be a fastpath heuristic to determine whether we should try to | |
1098 | * acquire the sighand lock to check/handle timers. | |
1099 | * | |
1100 | * In the worst case scenario, if 'running' or 'checking_timer' gets | |
1101 | * set but the current thread doesn't see the change yet, we'll wait | |
1102 | * until the next thread in the group gets a scheduler interrupt to | |
1103 | * handle the timer. This isn't an issue in practice because these | |
1104 | * types of delays with signals actually getting sent are expected. | |
1105 | */ | |
1106 | if (READ_ONCE(sig->cputimer.running) && | |
1107 | !READ_ONCE(sig->cputimer.checking_timer)) { | |
ebd7e7fc | 1108 | struct task_cputime group_sample; |
bb34d92f | 1109 | |
71107445 | 1110 | sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic); |
8d1f431c | 1111 | |
bb34d92f FM |
1112 | if (task_cputime_expired(&group_sample, &sig->cputime_expires)) |
1113 | return 1; | |
1114 | } | |
37bebc70 | 1115 | |
34be3930 JL |
1116 | if (dl_task(tsk) && tsk->dl.dl_overrun) |
1117 | return 1; | |
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 | */ | |
dce3e8fd | 1127 | void run_posix_cpu_timers(void) |
1da177e4 | 1128 | { |
dce3e8fd | 1129 | struct task_struct *tsk = current; |
1da177e4 | 1130 | struct k_itimer *timer, *next; |
0bdd2ed4 | 1131 | unsigned long flags; |
dce3e8fd | 1132 | LIST_HEAD(firing); |
1da177e4 | 1133 | |
a6968220 | 1134 | lockdep_assert_irqs_disabled(); |
1da177e4 | 1135 | |
1da177e4 | 1136 | /* |
f06febc9 | 1137 | * The fast path checks that there are no expired thread or thread |
bb34d92f | 1138 | * group timers. If that's so, just return. |
1da177e4 | 1139 | */ |
bb34d92f | 1140 | if (!fastpath_timer_check(tsk)) |
f06febc9 | 1141 | return; |
5ce73a4a | 1142 | |
0bdd2ed4 ON |
1143 | if (!lock_task_sighand(tsk, &flags)) |
1144 | return; | |
bb34d92f FM |
1145 | /* |
1146 | * Here we take off tsk->signal->cpu_timers[N] and | |
1147 | * tsk->cpu_timers[N] all the timers that are firing, and | |
1148 | * put them on the firing list. | |
1149 | */ | |
1150 | check_thread_timers(tsk, &firing); | |
934715a1 JL |
1151 | |
1152 | check_process_timers(tsk, &firing); | |
1da177e4 | 1153 | |
bb34d92f FM |
1154 | /* |
1155 | * We must release these locks before taking any timer's lock. | |
1156 | * There is a potential race with timer deletion here, as the | |
1157 | * siglock now protects our private firing list. We have set | |
1158 | * the firing flag in each timer, so that a deletion attempt | |
1159 | * that gets the timer lock before we do will give it up and | |
1160 | * spin until we've taken care of that timer below. | |
1161 | */ | |
0bdd2ed4 | 1162 | unlock_task_sighand(tsk, &flags); |
1da177e4 LT |
1163 | |
1164 | /* | |
1165 | * Now that all the timers on our list have the firing flag, | |
25985edc | 1166 | * no one will touch their list entries but us. We'll take |
1da177e4 LT |
1167 | * each timer's lock before clearing its firing flag, so no |
1168 | * timer call will interfere. | |
1169 | */ | |
1170 | list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) { | |
6e85c5ba HS |
1171 | int cpu_firing; |
1172 | ||
1da177e4 LT |
1173 | spin_lock(&timer->it_lock); |
1174 | list_del_init(&timer->it.cpu.entry); | |
6e85c5ba | 1175 | cpu_firing = timer->it.cpu.firing; |
1da177e4 LT |
1176 | timer->it.cpu.firing = 0; |
1177 | /* | |
1178 | * The firing flag is -1 if we collided with a reset | |
1179 | * of the timer, which already reported this | |
1180 | * almost-firing as an overrun. So don't generate an event. | |
1181 | */ | |
6e85c5ba | 1182 | if (likely(cpu_firing >= 0)) |
1da177e4 | 1183 | cpu_timer_fire(timer); |
1da177e4 LT |
1184 | spin_unlock(&timer->it_lock); |
1185 | } | |
1186 | } | |
1187 | ||
1188 | /* | |
f55db609 | 1189 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
f06febc9 | 1190 | * The tsk->sighand->siglock must be held by the caller. |
1da177e4 LT |
1191 | */ |
1192 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx, | |
858cf3a8 | 1193 | u64 *newval, u64 *oldval) |
1da177e4 | 1194 | { |
858cf3a8 | 1195 | u64 now; |
c3bca5d4 | 1196 | int ret; |
1da177e4 | 1197 | |
692117c1 TG |
1198 | if (WARN_ON_ONCE(clock_idx >= CPUCLOCK_SCHED)) |
1199 | return; | |
1200 | ||
c3bca5d4 | 1201 | ret = cpu_timer_sample_group(clock_idx, tsk, &now); |
1da177e4 | 1202 | |
c3bca5d4 | 1203 | if (oldval && ret != -EINVAL) { |
f55db609 SG |
1204 | /* |
1205 | * We are setting itimer. The *oldval is absolute and we update | |
1206 | * it to be relative, *newval argument is relative and we update | |
1207 | * it to be absolute. | |
1208 | */ | |
64861634 | 1209 | if (*oldval) { |
858cf3a8 | 1210 | if (*oldval <= now) { |
1da177e4 | 1211 | /* Just about to fire. */ |
858cf3a8 | 1212 | *oldval = TICK_NSEC; |
1da177e4 | 1213 | } else { |
858cf3a8 | 1214 | *oldval -= now; |
1da177e4 LT |
1215 | } |
1216 | } | |
1217 | ||
64861634 | 1218 | if (!*newval) |
b7878300 | 1219 | return; |
858cf3a8 | 1220 | *newval += now; |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
f55db609 SG |
1224 | * Update expiration cache if we are the earliest timer, or eventually |
1225 | * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire. | |
1da177e4 | 1226 | */ |
f55db609 SG |
1227 | switch (clock_idx) { |
1228 | case CPUCLOCK_PROF: | |
858cf3a8 FW |
1229 | if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval)) |
1230 | tsk->signal->cputime_expires.prof_exp = *newval; | |
f55db609 SG |
1231 | break; |
1232 | case CPUCLOCK_VIRT: | |
858cf3a8 FW |
1233 | if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval)) |
1234 | tsk->signal->cputime_expires.virt_exp = *newval; | |
f55db609 | 1235 | break; |
1da177e4 | 1236 | } |
b7878300 FW |
1237 | |
1238 | tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
1239 | } |
1240 | ||
e4b76555 | 1241 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
343d8fc2 | 1242 | const struct timespec64 *rqtp) |
1da177e4 | 1243 | { |
86a9c446 | 1244 | struct itimerspec64 it; |
343d8fc2 TG |
1245 | struct k_itimer timer; |
1246 | u64 expires; | |
1da177e4 LT |
1247 | int error; |
1248 | ||
1da177e4 LT |
1249 | /* |
1250 | * Set up a temporary timer and then wait for it to go off. | |
1251 | */ | |
1252 | memset(&timer, 0, sizeof timer); | |
1253 | spin_lock_init(&timer.it_lock); | |
1254 | timer.it_clock = which_clock; | |
1255 | timer.it_overrun = -1; | |
1256 | error = posix_cpu_timer_create(&timer); | |
1257 | timer.it_process = current; | |
1258 | if (!error) { | |
5f252b32 | 1259 | static struct itimerspec64 zero_it; |
edbeda46 | 1260 | struct restart_block *restart; |
e4b76555 | 1261 | |
edbeda46 | 1262 | memset(&it, 0, sizeof(it)); |
86a9c446 | 1263 | it.it_value = *rqtp; |
1da177e4 LT |
1264 | |
1265 | spin_lock_irq(&timer.it_lock); | |
86a9c446 | 1266 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
1da177e4 LT |
1267 | if (error) { |
1268 | spin_unlock_irq(&timer.it_lock); | |
1269 | return error; | |
1270 | } | |
1271 | ||
1272 | while (!signal_pending(current)) { | |
55ccb616 | 1273 | if (timer.it.cpu.expires == 0) { |
1da177e4 | 1274 | /* |
e6c42c29 SG |
1275 | * Our timer fired and was reset, below |
1276 | * deletion can not fail. | |
1da177e4 | 1277 | */ |
e6c42c29 | 1278 | posix_cpu_timer_del(&timer); |
1da177e4 LT |
1279 | spin_unlock_irq(&timer.it_lock); |
1280 | return 0; | |
1281 | } | |
1282 | ||
1283 | /* | |
1284 | * Block until cpu_timer_fire (or a signal) wakes us. | |
1285 | */ | |
1286 | __set_current_state(TASK_INTERRUPTIBLE); | |
1287 | spin_unlock_irq(&timer.it_lock); | |
1288 | schedule(); | |
1289 | spin_lock_irq(&timer.it_lock); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * We were interrupted by a signal. | |
1294 | */ | |
343d8fc2 | 1295 | expires = timer.it.cpu.expires; |
86a9c446 | 1296 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
e6c42c29 SG |
1297 | if (!error) { |
1298 | /* | |
1299 | * Timer is now unarmed, deletion can not fail. | |
1300 | */ | |
1301 | posix_cpu_timer_del(&timer); | |
1302 | } | |
1da177e4 LT |
1303 | spin_unlock_irq(&timer.it_lock); |
1304 | ||
e6c42c29 SG |
1305 | while (error == TIMER_RETRY) { |
1306 | /* | |
1307 | * We need to handle case when timer was or is in the | |
1308 | * middle of firing. In other cases we already freed | |
1309 | * resources. | |
1310 | */ | |
1311 | spin_lock_irq(&timer.it_lock); | |
1312 | error = posix_cpu_timer_del(&timer); | |
1313 | spin_unlock_irq(&timer.it_lock); | |
1314 | } | |
1315 | ||
86a9c446 | 1316 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
1da177e4 LT |
1317 | /* |
1318 | * It actually did fire already. | |
1319 | */ | |
1320 | return 0; | |
1321 | } | |
1322 | ||
e4b76555 | 1323 | error = -ERESTART_RESTARTBLOCK; |
86a9c446 AV |
1324 | /* |
1325 | * Report back to the user the time still remaining. | |
1326 | */ | |
edbeda46 | 1327 | restart = ¤t->restart_block; |
343d8fc2 | 1328 | restart->nanosleep.expires = expires; |
c0edd7c9 DD |
1329 | if (restart->nanosleep.type != TT_NONE) |
1330 | error = nanosleep_copyout(restart, &it.it_value); | |
e4b76555 TA |
1331 | } |
1332 | ||
1333 | return error; | |
1334 | } | |
1335 | ||
bc2c8ea4 TG |
1336 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
1337 | ||
1338 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, | |
938e7cf2 | 1339 | const struct timespec64 *rqtp) |
e4b76555 | 1340 | { |
f56141e3 | 1341 | struct restart_block *restart_block = ¤t->restart_block; |
e4b76555 TA |
1342 | int error; |
1343 | ||
1344 | /* | |
1345 | * Diagnose required errors first. | |
1346 | */ | |
1347 | if (CPUCLOCK_PERTHREAD(which_clock) && | |
1348 | (CPUCLOCK_PID(which_clock) == 0 || | |
01a21974 | 1349 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
e4b76555 TA |
1350 | return -EINVAL; |
1351 | ||
86a9c446 | 1352 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
e4b76555 TA |
1353 | |
1354 | if (error == -ERESTART_RESTARTBLOCK) { | |
1355 | ||
3751f9f2 | 1356 | if (flags & TIMER_ABSTIME) |
e4b76555 | 1357 | return -ERESTARTNOHAND; |
1da177e4 | 1358 | |
1711ef38 | 1359 | restart_block->fn = posix_cpu_nsleep_restart; |
ab8177bc | 1360 | restart_block->nanosleep.clockid = which_clock; |
1da177e4 | 1361 | } |
1da177e4 LT |
1362 | return error; |
1363 | } | |
1364 | ||
bc2c8ea4 | 1365 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1da177e4 | 1366 | { |
ab8177bc | 1367 | clockid_t which_clock = restart_block->nanosleep.clockid; |
ad196384 | 1368 | struct timespec64 t; |
97735f25 | 1369 | |
ad196384 | 1370 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
97735f25 | 1371 | |
86a9c446 | 1372 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
1da177e4 LT |
1373 | } |
1374 | ||
29f1b2b0 ND |
1375 | #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED) |
1376 | #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED) | |
1da177e4 | 1377 | |
a924b04d | 1378 | static int process_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1379 | struct timespec64 *tp) |
1da177e4 LT |
1380 | { |
1381 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); | |
1382 | } | |
a924b04d | 1383 | static int process_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1384 | struct timespec64 *tp) |
1da177e4 LT |
1385 | { |
1386 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); | |
1387 | } | |
1388 | static int process_cpu_timer_create(struct k_itimer *timer) | |
1389 | { | |
1390 | timer->it_clock = PROCESS_CLOCK; | |
1391 | return posix_cpu_timer_create(timer); | |
1392 | } | |
a924b04d | 1393 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
938e7cf2 | 1394 | const struct timespec64 *rqtp) |
1da177e4 | 1395 | { |
99e6c0e6 | 1396 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
1da177e4 | 1397 | } |
a924b04d | 1398 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1399 | struct timespec64 *tp) |
1da177e4 LT |
1400 | { |
1401 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); | |
1402 | } | |
a924b04d | 1403 | static int thread_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1404 | struct timespec64 *tp) |
1da177e4 LT |
1405 | { |
1406 | return posix_cpu_clock_get(THREAD_CLOCK, tp); | |
1407 | } | |
1408 | static int thread_cpu_timer_create(struct k_itimer *timer) | |
1409 | { | |
1410 | timer->it_clock = THREAD_CLOCK; | |
1411 | return posix_cpu_timer_create(timer); | |
1412 | } | |
1da177e4 | 1413 | |
d3ba5a9a | 1414 | const struct k_clock clock_posix_cpu = { |
1976945e TG |
1415 | .clock_getres = posix_cpu_clock_getres, |
1416 | .clock_set = posix_cpu_clock_set, | |
1417 | .clock_get = posix_cpu_clock_get, | |
1418 | .timer_create = posix_cpu_timer_create, | |
1419 | .nsleep = posix_cpu_nsleep, | |
1976945e TG |
1420 | .timer_set = posix_cpu_timer_set, |
1421 | .timer_del = posix_cpu_timer_del, | |
1422 | .timer_get = posix_cpu_timer_get, | |
f37fb0aa | 1423 | .timer_rearm = posix_cpu_timer_rearm, |
1976945e TG |
1424 | }; |
1425 | ||
d3ba5a9a CH |
1426 | const struct k_clock clock_process = { |
1427 | .clock_getres = process_cpu_clock_getres, | |
1428 | .clock_get = process_cpu_clock_get, | |
1429 | .timer_create = process_cpu_timer_create, | |
1430 | .nsleep = process_cpu_nsleep, | |
d3ba5a9a | 1431 | }; |
1da177e4 | 1432 | |
d3ba5a9a CH |
1433 | const struct k_clock clock_thread = { |
1434 | .clock_getres = thread_cpu_clock_getres, | |
1435 | .clock_get = thread_cpu_clock_get, | |
1436 | .timer_create = thread_cpu_timer_create, | |
1437 | }; |