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 | ||
3a245c0f TG |
23 | void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit) |
24 | { | |
25 | posix_cputimers_init(pct); | |
244d49e3 | 26 | if (cpu_limit != RLIM_INFINITY) { |
87dc6448 | 27 | pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC; |
244d49e3 TG |
28 | pct->timers_active = true; |
29 | } | |
3a245c0f TG |
30 | } |
31 | ||
f06febc9 | 32 | /* |
f55db609 | 33 | * Called after updating RLIMIT_CPU to run cpu timer and update |
87dc6448 TG |
34 | * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if |
35 | * necessary. Needs siglock protection since other code may update the | |
3a245c0f | 36 | * expiration cache as well. |
f06febc9 | 37 | */ |
5ab46b34 | 38 | void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new) |
f06febc9 | 39 | { |
858cf3a8 | 40 | u64 nsecs = rlim_new * NSEC_PER_SEC; |
f06febc9 | 41 | |
5ab46b34 | 42 | spin_lock_irq(&task->sighand->siglock); |
858cf3a8 | 43 | set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL); |
5ab46b34 | 44 | spin_unlock_irq(&task->sighand->siglock); |
f06febc9 FM |
45 | } |
46 | ||
6ae40e3f TG |
47 | /* |
48 | * Functions for validating access to tasks. | |
49 | */ | |
77b4b542 TG |
50 | static struct task_struct *lookup_task(const pid_t pid, bool thread, |
51 | bool gettime) | |
1da177e4 | 52 | { |
1da177e4 | 53 | struct task_struct *p; |
1da177e4 | 54 | |
77b4b542 TG |
55 | /* |
56 | * If the encoded PID is 0, then the timer is targeted at current | |
57 | * or the process to which current belongs. | |
58 | */ | |
6ae40e3f TG |
59 | if (!pid) |
60 | return thread ? current : current->group_leader; | |
1da177e4 | 61 | |
6ae40e3f | 62 | p = find_task_by_vpid(pid); |
77b4b542 | 63 | if (!p) |
6ae40e3f | 64 | return p; |
77b4b542 | 65 | |
6ae40e3f TG |
66 | if (thread) |
67 | return same_thread_group(p, current) ? p : NULL; | |
77b4b542 TG |
68 | |
69 | if (gettime) { | |
70 | /* | |
71 | * For clock_gettime(PROCESS) the task does not need to be | |
72 | * the actual group leader. tsk->sighand gives | |
73 | * access to the group's clock. | |
74 | * | |
75 | * Timers need the group leader because they take a | |
76 | * reference on it and store the task pointer until the | |
77 | * timer is destroyed. | |
78 | */ | |
79 | return (p == current || thread_group_leader(p)) ? p : NULL; | |
80 | } | |
81 | ||
82 | /* | |
83 | * For processes require that p is group leader. | |
84 | */ | |
6ae40e3f TG |
85 | return has_group_leader_pid(p) ? p : NULL; |
86 | } | |
87 | ||
88 | static struct task_struct *__get_task_for_clock(const clockid_t clock, | |
77b4b542 | 89 | bool getref, bool gettime) |
6ae40e3f TG |
90 | { |
91 | const bool thread = !!CPUCLOCK_PERTHREAD(clock); | |
92 | const pid_t pid = CPUCLOCK_PID(clock); | |
93 | struct task_struct *p; | |
94 | ||
95 | if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX) | |
96 | return NULL; | |
1da177e4 | 97 | |
c0deae8c | 98 | rcu_read_lock(); |
77b4b542 | 99 | p = lookup_task(pid, thread, gettime); |
6ae40e3f TG |
100 | if (p && getref) |
101 | get_task_struct(p); | |
c0deae8c | 102 | rcu_read_unlock(); |
6ae40e3f TG |
103 | return p; |
104 | } | |
1da177e4 | 105 | |
6ae40e3f TG |
106 | static inline struct task_struct *get_task_for_clock(const clockid_t clock) |
107 | { | |
77b4b542 TG |
108 | return __get_task_for_clock(clock, true, false); |
109 | } | |
110 | ||
111 | static inline struct task_struct *get_task_for_clock_get(const clockid_t clock) | |
112 | { | |
113 | return __get_task_for_clock(clock, true, true); | |
6ae40e3f TG |
114 | } |
115 | ||
116 | static inline int validate_clock_permissions(const clockid_t clock) | |
117 | { | |
77b4b542 | 118 | return __get_task_for_clock(clock, false, false) ? 0 : -EINVAL; |
1da177e4 LT |
119 | } |
120 | ||
1da177e4 LT |
121 | /* |
122 | * Update expiry time from increment, and increase overrun count, | |
123 | * given the current clock sample. | |
124 | */ | |
60bda037 | 125 | static u64 bump_cpu_timer(struct k_itimer *timer, u64 now) |
1da177e4 | 126 | { |
60bda037 | 127 | u64 delta, incr, expires = timer->it.cpu.node.expires; |
1da177e4 LT |
128 | int i; |
129 | ||
16118794 | 130 | if (!timer->it_interval) |
60bda037 | 131 | return expires; |
1da177e4 | 132 | |
60bda037 TG |
133 | if (now < expires) |
134 | return expires; | |
1da177e4 | 135 | |
16118794 | 136 | incr = timer->it_interval; |
60bda037 | 137 | delta = now + incr - expires; |
1da177e4 | 138 | |
55ccb616 FW |
139 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
140 | for (i = 0; incr < delta - incr; i++) | |
141 | incr = incr << 1; | |
142 | ||
143 | for (; i >= 0; incr >>= 1, i--) { | |
144 | if (delta < incr) | |
145 | continue; | |
146 | ||
60bda037 | 147 | timer->it.cpu.node.expires += incr; |
78c9c4df | 148 | timer->it_overrun += 1LL << i; |
55ccb616 | 149 | delta -= incr; |
1da177e4 | 150 | } |
60bda037 | 151 | return timer->it.cpu.node.expires; |
1da177e4 LT |
152 | } |
153 | ||
2bbdbdae TG |
154 | /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */ |
155 | static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct) | |
555347f6 | 156 | { |
2bbdbdae TG |
157 | return !(~pct->bases[CPUCLOCK_PROF].nextevt | |
158 | ~pct->bases[CPUCLOCK_VIRT].nextevt | | |
159 | ~pct->bases[CPUCLOCK_SCHED].nextevt); | |
555347f6 FW |
160 | } |
161 | ||
bc2c8ea4 | 162 | static int |
d2e3e0ca | 163 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp) |
1da177e4 | 164 | { |
6ae40e3f TG |
165 | int error = validate_clock_permissions(which_clock); |
166 | ||
1da177e4 LT |
167 | if (!error) { |
168 | tp->tv_sec = 0; | |
169 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); | |
170 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { | |
171 | /* | |
172 | * If sched_clock is using a cycle counter, we | |
173 | * don't have any idea of its true resolution | |
174 | * exported, but it is much more than 1s/HZ. | |
175 | */ | |
176 | tp->tv_nsec = 1; | |
177 | } | |
178 | } | |
179 | return error; | |
180 | } | |
181 | ||
bc2c8ea4 | 182 | static int |
6ae40e3f | 183 | posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp) |
1da177e4 | 184 | { |
6ae40e3f TG |
185 | int error = validate_clock_permissions(clock); |
186 | ||
1da177e4 LT |
187 | /* |
188 | * You can never reset a CPU clock, but we check for other errors | |
189 | * in the call before failing with EPERM. | |
190 | */ | |
6ae40e3f | 191 | return error ? : -EPERM; |
1da177e4 LT |
192 | } |
193 | ||
1da177e4 | 194 | /* |
2092c1d4 | 195 | * Sample a per-thread clock for the given task. clkid is validated. |
1da177e4 | 196 | */ |
8c2d74f0 | 197 | static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p) |
1da177e4 | 198 | { |
ab693c5a TG |
199 | u64 utime, stime; |
200 | ||
201 | if (clkid == CPUCLOCK_SCHED) | |
202 | return task_sched_runtime(p); | |
203 | ||
204 | task_cputime(p, &utime, &stime); | |
205 | ||
2092c1d4 | 206 | switch (clkid) { |
1da177e4 | 207 | case CPUCLOCK_PROF: |
ab693c5a | 208 | return utime + stime; |
1da177e4 | 209 | case CPUCLOCK_VIRT: |
ab693c5a | 210 | return utime; |
2092c1d4 TG |
211 | default: |
212 | WARN_ON_ONCE(1); | |
1da177e4 | 213 | } |
8c2d74f0 | 214 | return 0; |
1da177e4 LT |
215 | } |
216 | ||
b0d524f7 TG |
217 | static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime) |
218 | { | |
219 | samples[CPUCLOCK_PROF] = stime + utime; | |
220 | samples[CPUCLOCK_VIRT] = utime; | |
221 | samples[CPUCLOCK_SCHED] = rtime; | |
222 | } | |
223 | ||
224 | static void task_sample_cputime(struct task_struct *p, u64 *samples) | |
225 | { | |
226 | u64 stime, utime; | |
227 | ||
228 | task_cputime(p, &utime, &stime); | |
229 | store_samples(samples, stime, utime, p->se.sum_exec_runtime); | |
230 | } | |
231 | ||
232 | static void proc_sample_cputime_atomic(struct task_cputime_atomic *at, | |
233 | u64 *samples) | |
234 | { | |
235 | u64 stime, utime, rtime; | |
236 | ||
237 | utime = atomic64_read(&at->utime); | |
238 | stime = atomic64_read(&at->stime); | |
239 | rtime = atomic64_read(&at->sum_exec_runtime); | |
240 | store_samples(samples, stime, utime, rtime); | |
241 | } | |
242 | ||
1018016c JL |
243 | /* |
244 | * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg | |
245 | * to avoid race conditions with concurrent updates to cputime. | |
246 | */ | |
247 | static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime) | |
4da94d49 | 248 | { |
1018016c JL |
249 | u64 curr_cputime; |
250 | retry: | |
251 | curr_cputime = atomic64_read(cputime); | |
252 | if (sum_cputime > curr_cputime) { | |
253 | if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime) | |
254 | goto retry; | |
255 | } | |
256 | } | |
4da94d49 | 257 | |
b7be4ef1 TG |
258 | static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, |
259 | struct task_cputime *sum) | |
1018016c | 260 | { |
71107445 JL |
261 | __update_gt_cputime(&cputime_atomic->utime, sum->utime); |
262 | __update_gt_cputime(&cputime_atomic->stime, sum->stime); | |
263 | __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime); | |
1018016c | 264 | } |
4da94d49 | 265 | |
19298fbf TG |
266 | /** |
267 | * thread_group_sample_cputime - Sample cputime for a given task | |
268 | * @tsk: Task for which cputime needs to be started | |
269 | * @iimes: Storage for time samples | |
270 | * | |
271 | * Called from sys_getitimer() to calculate the expiry time of an active | |
272 | * timer. That means group cputime accounting is already active. Called | |
273 | * with task sighand lock held. | |
274 | * | |
275 | * Updates @times with an uptodate sample of the thread group cputimes. | |
276 | */ | |
b7be4ef1 | 277 | void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples) |
19298fbf TG |
278 | { |
279 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
244d49e3 | 280 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
19298fbf | 281 | |
244d49e3 | 282 | WARN_ON_ONCE(!pct->timers_active); |
19298fbf | 283 | |
b7be4ef1 | 284 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
19298fbf TG |
285 | } |
286 | ||
c506bef4 TG |
287 | /** |
288 | * thread_group_start_cputime - Start cputime and return a sample | |
289 | * @tsk: Task for which cputime needs to be started | |
b7be4ef1 | 290 | * @samples: Storage for time samples |
c506bef4 TG |
291 | * |
292 | * The thread group cputime accouting is avoided when there are no posix | |
293 | * CPU timers armed. Before starting a timer it's required to check whether | |
294 | * the time accounting is active. If not, a full update of the atomic | |
295 | * accounting store needs to be done and the accounting enabled. | |
296 | * | |
297 | * Updates @times with an uptodate sample of the thread group cputimes. | |
298 | */ | |
b7be4ef1 | 299 | static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples) |
4da94d49 PZ |
300 | { |
301 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
244d49e3 | 302 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
4da94d49 | 303 | |
1018016c | 304 | /* Check if cputimer isn't running. This is accessed without locking. */ |
244d49e3 | 305 | if (!READ_ONCE(pct->timers_active)) { |
b7be4ef1 TG |
306 | struct task_cputime sum; |
307 | ||
4da94d49 PZ |
308 | /* |
309 | * The POSIX timer interface allows for absolute time expiry | |
310 | * values through the TIMER_ABSTIME flag, therefore we have | |
1018016c | 311 | * to synchronize the timer to the clock every time we start it. |
4da94d49 | 312 | */ |
ebd7e7fc | 313 | thread_group_cputime(tsk, &sum); |
71107445 | 314 | update_gt_cputime(&cputimer->cputime_atomic, &sum); |
1018016c JL |
315 | |
316 | /* | |
244d49e3 TG |
317 | * We're setting timers_active without a lock. Ensure this |
318 | * only gets written to in one operation. We set it after | |
319 | * update_gt_cputime() as a small optimization, but | |
320 | * barriers are not required because update_gt_cputime() | |
1018016c JL |
321 | * can handle concurrent updates. |
322 | */ | |
244d49e3 | 323 | WRITE_ONCE(pct->timers_active, true); |
1018016c | 324 | } |
b7be4ef1 TG |
325 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
326 | } | |
327 | ||
328 | static void __thread_group_cputime(struct task_struct *tsk, u64 *samples) | |
329 | { | |
330 | struct task_cputime ct; | |
331 | ||
332 | thread_group_cputime(tsk, &ct); | |
333 | store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime); | |
4da94d49 PZ |
334 | } |
335 | ||
1da177e4 | 336 | /* |
24ab7f5a TG |
337 | * Sample a process (thread group) clock for the given task clkid. If the |
338 | * group's cputime accounting is already enabled, read the atomic | |
339 | * store. Otherwise a full update is required. Task's sighand lock must be | |
2092c1d4 TG |
340 | * held to protect the task traversal on a full update. clkid is already |
341 | * validated. | |
1da177e4 | 342 | */ |
8c2d74f0 TG |
343 | static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p, |
344 | bool start) | |
1da177e4 | 345 | { |
24ab7f5a | 346 | struct thread_group_cputimer *cputimer = &p->signal->cputimer; |
244d49e3 | 347 | struct posix_cputimers *pct = &p->signal->posix_cputimers; |
b7be4ef1 | 348 | u64 samples[CPUCLOCK_MAX]; |
f06febc9 | 349 | |
244d49e3 | 350 | if (!READ_ONCE(pct->timers_active)) { |
24ab7f5a | 351 | if (start) |
b7be4ef1 | 352 | thread_group_start_cputime(p, samples); |
24ab7f5a | 353 | else |
b7be4ef1 | 354 | __thread_group_cputime(p, samples); |
24ab7f5a | 355 | } else { |
b7be4ef1 | 356 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
24ab7f5a TG |
357 | } |
358 | ||
b7be4ef1 | 359 | return samples[clkid]; |
1da177e4 LT |
360 | } |
361 | ||
bfcf3e92 | 362 | static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp) |
33ab0fec | 363 | { |
bfcf3e92 TG |
364 | const clockid_t clkid = CPUCLOCK_WHICH(clock); |
365 | struct task_struct *tsk; | |
366 | u64 t; | |
33ab0fec | 367 | |
77b4b542 | 368 | tsk = get_task_for_clock_get(clock); |
bfcf3e92 TG |
369 | if (!tsk) |
370 | return -EINVAL; | |
1da177e4 | 371 | |
bfcf3e92 | 372 | if (CPUCLOCK_PERTHREAD(clock)) |
8c2d74f0 | 373 | t = cpu_clock_sample(clkid, tsk); |
bfcf3e92 | 374 | else |
8c2d74f0 | 375 | t = cpu_clock_sample_group(clkid, tsk, false); |
bfcf3e92 | 376 | put_task_struct(tsk); |
1da177e4 | 377 | |
bfcf3e92 TG |
378 | *tp = ns_to_timespec64(t); |
379 | return 0; | |
1da177e4 LT |
380 | } |
381 | ||
1da177e4 LT |
382 | /* |
383 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. | |
ba5ea951 SG |
384 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
385 | * new timer already all-zeros initialized. | |
1da177e4 | 386 | */ |
bc2c8ea4 | 387 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
1da177e4 | 388 | { |
e5a8b65b | 389 | struct task_struct *p = get_task_for_clock(new_timer->it_clock); |
1da177e4 | 390 | |
e5a8b65b | 391 | if (!p) |
1da177e4 LT |
392 | return -EINVAL; |
393 | ||
d97bb75d | 394 | new_timer->kclock = &clock_posix_cpu; |
60bda037 | 395 | timerqueue_init(&new_timer->it.cpu.node); |
1da177e4 | 396 | new_timer->it.cpu.task = p; |
e5a8b65b | 397 | return 0; |
1da177e4 LT |
398 | } |
399 | ||
400 | /* | |
401 | * Clean up a CPU-clock timer that is about to be destroyed. | |
402 | * This is called from timer deletion with the timer already locked. | |
403 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
404 | * and try again. (This happens when the timer is in the middle of firing.) | |
405 | */ | |
bc2c8ea4 | 406 | static int posix_cpu_timer_del(struct k_itimer *timer) |
1da177e4 | 407 | { |
60bda037 TG |
408 | struct cpu_timer *ctmr = &timer->it.cpu; |
409 | struct task_struct *p = ctmr->task; | |
3d7a1427 | 410 | struct sighand_struct *sighand; |
60bda037 TG |
411 | unsigned long flags; |
412 | int ret = 0; | |
1da177e4 | 413 | |
692117c1 TG |
414 | if (WARN_ON_ONCE(!p)) |
415 | return -EINVAL; | |
108150ea | 416 | |
3d7a1427 FW |
417 | /* |
418 | * Protect against sighand release/switch in exit/exec and process/ | |
419 | * thread timer list entry concurrent read/writes. | |
420 | */ | |
421 | sighand = lock_task_sighand(p, &flags); | |
422 | if (unlikely(sighand == NULL)) { | |
a3222f88 | 423 | /* |
60bda037 TG |
424 | * This raced with the reaping of the task. The exit cleanup |
425 | * should have removed this timer from the timer queue. | |
a3222f88 | 426 | */ |
60bda037 | 427 | WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node)); |
a3222f88 | 428 | } else { |
a3222f88 FW |
429 | if (timer->it.cpu.firing) |
430 | ret = TIMER_RETRY; | |
431 | else | |
60bda037 | 432 | cpu_timer_dequeue(ctmr); |
3d7a1427 FW |
433 | |
434 | unlock_task_sighand(p, &flags); | |
1da177e4 | 435 | } |
a3222f88 FW |
436 | |
437 | if (!ret) | |
438 | put_task_struct(p); | |
1da177e4 | 439 | |
108150ea | 440 | return ret; |
1da177e4 LT |
441 | } |
442 | ||
60bda037 | 443 | static void cleanup_timerqueue(struct timerqueue_head *head) |
1a7fa510 | 444 | { |
60bda037 TG |
445 | struct timerqueue_node *node; |
446 | struct cpu_timer *ctmr; | |
1a7fa510 | 447 | |
60bda037 TG |
448 | while ((node = timerqueue_getnext(head))) { |
449 | timerqueue_del(head, node); | |
450 | ctmr = container_of(node, struct cpu_timer, node); | |
451 | ctmr->head = NULL; | |
452 | } | |
1a7fa510 FW |
453 | } |
454 | ||
1da177e4 | 455 | /* |
7cb9a94c TG |
456 | * Clean out CPU timers which are still armed when a thread exits. The |
457 | * timers are only removed from the list. No other updates are done. The | |
458 | * corresponding posix timers are still accessible, but cannot be rearmed. | |
459 | * | |
1da177e4 LT |
460 | * This must be called with the siglock held. |
461 | */ | |
2b69942f | 462 | static void cleanup_timers(struct posix_cputimers *pct) |
1da177e4 | 463 | { |
60bda037 TG |
464 | cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead); |
465 | cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead); | |
466 | cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead); | |
1da177e4 LT |
467 | } |
468 | ||
469 | /* | |
470 | * These are both called with the siglock held, when the current thread | |
471 | * is being reaped. When the final (leader) thread in the group is reaped, | |
472 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. | |
473 | */ | |
474 | void posix_cpu_timers_exit(struct task_struct *tsk) | |
475 | { | |
2b69942f | 476 | cleanup_timers(&tsk->posix_cputimers); |
1da177e4 LT |
477 | } |
478 | void posix_cpu_timers_exit_group(struct task_struct *tsk) | |
479 | { | |
2b69942f | 480 | cleanup_timers(&tsk->signal->posix_cputimers); |
1da177e4 LT |
481 | } |
482 | ||
1da177e4 LT |
483 | /* |
484 | * Insert the timer on the appropriate list before any timers that | |
e73d84e3 | 485 | * expire later. This must be called with the sighand lock held. |
1da177e4 | 486 | */ |
5eb9aa64 | 487 | static void arm_timer(struct k_itimer *timer) |
1da177e4 | 488 | { |
3b495b22 | 489 | int clkidx = CPUCLOCK_WHICH(timer->it_clock); |
60bda037 TG |
490 | struct cpu_timer *ctmr = &timer->it.cpu; |
491 | u64 newexp = cpu_timer_getexpires(ctmr); | |
492 | struct task_struct *p = ctmr->task; | |
87dc6448 | 493 | struct posix_cputimer_base *base; |
1da177e4 | 494 | |
87dc6448 TG |
495 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
496 | base = p->posix_cputimers.bases + clkidx; | |
497 | else | |
498 | base = p->signal->posix_cputimers.bases + clkidx; | |
1da177e4 | 499 | |
60bda037 | 500 | if (!cpu_timer_enqueue(&base->tqhead, ctmr)) |
3b495b22 | 501 | return; |
5eb9aa64 | 502 | |
3b495b22 TG |
503 | /* |
504 | * We are the new earliest-expiring POSIX 1.b timer, hence | |
505 | * need to update expiration cache. Take into account that | |
506 | * for process timers we share expiration cache with itimers | |
507 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. | |
508 | */ | |
2bbdbdae | 509 | if (newexp < base->nextevt) |
87dc6448 | 510 | base->nextevt = newexp; |
1da177e4 | 511 | |
3b495b22 TG |
512 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
513 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); | |
514 | else | |
515 | tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
516 | } |
517 | ||
518 | /* | |
519 | * The timer is locked, fire it and arrange for its reload. | |
520 | */ | |
521 | static void cpu_timer_fire(struct k_itimer *timer) | |
522 | { | |
60bda037 TG |
523 | struct cpu_timer *ctmr = &timer->it.cpu; |
524 | ||
1f169f84 SG |
525 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
526 | /* | |
527 | * User don't want any signal. | |
528 | */ | |
60bda037 | 529 | cpu_timer_setexpires(ctmr, 0); |
1f169f84 | 530 | } else if (unlikely(timer->sigq == NULL)) { |
1da177e4 LT |
531 | /* |
532 | * This a special case for clock_nanosleep, | |
533 | * not a normal timer from sys_timer_create. | |
534 | */ | |
535 | wake_up_process(timer->it_process); | |
60bda037 | 536 | cpu_timer_setexpires(ctmr, 0); |
16118794 | 537 | } else if (!timer->it_interval) { |
1da177e4 LT |
538 | /* |
539 | * One-shot timer. Clear it as soon as it's fired. | |
540 | */ | |
541 | posix_timer_event(timer, 0); | |
60bda037 | 542 | cpu_timer_setexpires(ctmr, 0); |
1da177e4 LT |
543 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
544 | /* | |
545 | * The signal did not get queued because the signal | |
546 | * was ignored, so we won't get any callback to | |
547 | * reload the timer. But we need to keep it | |
548 | * ticking in case the signal is deliverable next time. | |
549 | */ | |
f37fb0aa | 550 | posix_cpu_timer_rearm(timer); |
af888d67 | 551 | ++timer->it_requeue_pending; |
1da177e4 LT |
552 | } |
553 | } | |
554 | ||
555 | /* | |
556 | * Guts of sys_timer_settime for CPU timers. | |
557 | * This is called with the timer locked and interrupts disabled. | |
558 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
559 | * and try again. (This happens when the timer is in the middle of firing.) | |
560 | */ | |
e73d84e3 | 561 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
5f252b32 | 562 | struct itimerspec64 *new, struct itimerspec64 *old) |
1da177e4 | 563 | { |
c7a37c6f | 564 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
ebd7e7fc | 565 | u64 old_expires, new_expires, old_incr, val; |
60bda037 TG |
566 | struct cpu_timer *ctmr = &timer->it.cpu; |
567 | struct task_struct *p = ctmr->task; | |
c7a37c6f TG |
568 | struct sighand_struct *sighand; |
569 | unsigned long flags; | |
60bda037 | 570 | int ret = 0; |
1da177e4 | 571 | |
692117c1 TG |
572 | if (WARN_ON_ONCE(!p)) |
573 | return -EINVAL; | |
1da177e4 | 574 | |
098b0e01 TG |
575 | /* |
576 | * Use the to_ktime conversion because that clamps the maximum | |
577 | * value to KTIME_MAX and avoid multiplication overflows. | |
578 | */ | |
579 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); | |
1da177e4 | 580 | |
1da177e4 | 581 | /* |
e73d84e3 FW |
582 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
583 | * and p->signal->cpu_timers read/write in arm_timer() | |
584 | */ | |
585 | sighand = lock_task_sighand(p, &flags); | |
586 | /* | |
587 | * If p has just been reaped, we can no | |
1da177e4 LT |
588 | * longer get any information about it at all. |
589 | */ | |
60bda037 | 590 | if (unlikely(sighand == NULL)) |
1da177e4 | 591 | return -ESRCH; |
1da177e4 LT |
592 | |
593 | /* | |
594 | * Disarm any old timer after extracting its expiry time. | |
595 | */ | |
16118794 | 596 | old_incr = timer->it_interval; |
60bda037 TG |
597 | old_expires = cpu_timer_getexpires(ctmr); |
598 | ||
a69ac4a7 ON |
599 | if (unlikely(timer->it.cpu.firing)) { |
600 | timer->it.cpu.firing = -1; | |
601 | ret = TIMER_RETRY; | |
60bda037 TG |
602 | } else { |
603 | cpu_timer_dequeue(ctmr); | |
604 | } | |
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 | */ | |
8c2d74f0 TG |
614 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
615 | val = cpu_clock_sample(clkid, p); | |
616 | else | |
617 | val = cpu_clock_sample_group(clkid, p, true); | |
1da177e4 LT |
618 | |
619 | if (old) { | |
55ccb616 | 620 | if (old_expires == 0) { |
1da177e4 LT |
621 | old->it_value.tv_sec = 0; |
622 | old->it_value.tv_nsec = 0; | |
623 | } else { | |
624 | /* | |
60bda037 TG |
625 | * Update the timer in case it has overrun already. |
626 | * If it has, we'll report it as having overrun and | |
627 | * with the next reloaded timer already ticking, | |
628 | * though we are swallowing that pending | |
629 | * notification here to install the new setting. | |
1da177e4 | 630 | */ |
60bda037 TG |
631 | u64 exp = bump_cpu_timer(timer, val); |
632 | ||
633 | if (val < exp) { | |
634 | old_expires = exp - val; | |
5f252b32 | 635 | old->it_value = ns_to_timespec64(old_expires); |
1da177e4 LT |
636 | } else { |
637 | old->it_value.tv_nsec = 1; | |
638 | old->it_value.tv_sec = 0; | |
639 | } | |
640 | } | |
641 | } | |
642 | ||
a69ac4a7 | 643 | if (unlikely(ret)) { |
1da177e4 LT |
644 | /* |
645 | * We are colliding with the timer actually firing. | |
646 | * Punt after filling in the timer's old value, and | |
647 | * disable this firing since we are already reporting | |
648 | * it as an overrun (thanks to bump_cpu_timer above). | |
649 | */ | |
e73d84e3 | 650 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
651 | goto out; |
652 | } | |
653 | ||
e73d84e3 | 654 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
55ccb616 | 655 | new_expires += val; |
1da177e4 LT |
656 | } |
657 | ||
658 | /* | |
659 | * Install the new expiry time (or zero). | |
660 | * For a timer with no notification action, we don't actually | |
661 | * arm the timer (we'll just fake it for timer_gettime). | |
662 | */ | |
60bda037 | 663 | cpu_timer_setexpires(ctmr, new_expires); |
55ccb616 | 664 | if (new_expires != 0 && val < new_expires) { |
5eb9aa64 | 665 | arm_timer(timer); |
1da177e4 LT |
666 | } |
667 | ||
e73d84e3 | 668 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
669 | /* |
670 | * Install the new reload setting, and | |
671 | * set up the signal and overrun bookkeeping. | |
672 | */ | |
16118794 | 673 | timer->it_interval = timespec64_to_ktime(new->it_interval); |
1da177e4 LT |
674 | |
675 | /* | |
676 | * This acts as a modification timestamp for the timer, | |
677 | * so any automatic reload attempt will punt on seeing | |
678 | * that we have reset the timer manually. | |
679 | */ | |
680 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & | |
681 | ~REQUEUE_PENDING; | |
682 | timer->it_overrun_last = 0; | |
683 | timer->it_overrun = -1; | |
684 | ||
55ccb616 | 685 | if (new_expires != 0 && !(val < new_expires)) { |
1da177e4 LT |
686 | /* |
687 | * The designated time already passed, so we notify | |
688 | * immediately, even if the thread never runs to | |
689 | * accumulate more time on this clock. | |
690 | */ | |
691 | cpu_timer_fire(timer); | |
692 | } | |
693 | ||
694 | ret = 0; | |
695 | out: | |
ebd7e7fc | 696 | if (old) |
5f252b32 | 697 | old->it_interval = ns_to_timespec64(old_incr); |
b7878300 | 698 | |
1da177e4 LT |
699 | return ret; |
700 | } | |
701 | ||
5f252b32 | 702 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
1da177e4 | 703 | { |
99093c5b | 704 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
60bda037 TG |
705 | struct cpu_timer *ctmr = &timer->it.cpu; |
706 | u64 now, expires = cpu_timer_getexpires(ctmr); | |
707 | struct task_struct *p = ctmr->task; | |
1da177e4 | 708 | |
692117c1 TG |
709 | if (WARN_ON_ONCE(!p)) |
710 | return; | |
a3222f88 | 711 | |
1da177e4 LT |
712 | /* |
713 | * Easy part: convert the reload time. | |
714 | */ | |
16118794 | 715 | itp->it_interval = ktime_to_timespec64(timer->it_interval); |
1da177e4 | 716 | |
60bda037 | 717 | if (!expires) |
1da177e4 | 718 | return; |
1da177e4 | 719 | |
1da177e4 LT |
720 | /* |
721 | * Sample the clock to take the difference with the expiry time. | |
722 | */ | |
723 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
8c2d74f0 | 724 | now = cpu_clock_sample(clkid, p); |
1da177e4 | 725 | } else { |
e73d84e3 FW |
726 | struct sighand_struct *sighand; |
727 | unsigned long flags; | |
728 | ||
729 | /* | |
730 | * Protect against sighand release/switch in exit/exec and | |
731 | * also make timer sampling safe if it ends up calling | |
ebd7e7fc | 732 | * thread_group_cputime(). |
e73d84e3 FW |
733 | */ |
734 | sighand = lock_task_sighand(p, &flags); | |
735 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
736 | /* |
737 | * The process has been reaped. | |
738 | * We can't even collect a sample any more. | |
60bda037 | 739 | * Disarm the timer, nothing else to do. |
1da177e4 | 740 | */ |
60bda037 | 741 | cpu_timer_setexpires(ctmr, 0); |
2c13ce8f | 742 | return; |
1da177e4 | 743 | } else { |
8c2d74f0 | 744 | now = cpu_clock_sample_group(clkid, p, false); |
e73d84e3 | 745 | unlock_task_sighand(p, &flags); |
1da177e4 | 746 | } |
1da177e4 LT |
747 | } |
748 | ||
60bda037 TG |
749 | if (now < expires) { |
750 | itp->it_value = ns_to_timespec64(expires - now); | |
1da177e4 LT |
751 | } else { |
752 | /* | |
753 | * The timer should have expired already, but the firing | |
754 | * hasn't taken place yet. Say it's just about to expire. | |
755 | */ | |
756 | itp->it_value.tv_nsec = 1; | |
757 | itp->it_value.tv_sec = 0; | |
758 | } | |
759 | } | |
760 | ||
60bda037 | 761 | #define MAX_COLLECTED 20 |
2473f3e7 | 762 | |
60bda037 TG |
763 | static u64 collect_timerqueue(struct timerqueue_head *head, |
764 | struct list_head *firing, u64 now) | |
765 | { | |
766 | struct timerqueue_node *next; | |
767 | int i = 0; | |
768 | ||
769 | while ((next = timerqueue_getnext(head))) { | |
770 | struct cpu_timer *ctmr; | |
771 | u64 expires; | |
772 | ||
773 | ctmr = container_of(next, struct cpu_timer, node); | |
774 | expires = cpu_timer_getexpires(ctmr); | |
775 | /* Limit the number of timers to expire at once */ | |
776 | if (++i == MAX_COLLECTED || now < expires) | |
777 | return expires; | |
778 | ||
779 | ctmr->firing = 1; | |
780 | cpu_timer_dequeue(ctmr); | |
781 | list_add_tail(&ctmr->elist, firing); | |
2473f3e7 FW |
782 | } |
783 | ||
2bbdbdae | 784 | return U64_MAX; |
2473f3e7 FW |
785 | } |
786 | ||
60bda037 TG |
787 | static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples, |
788 | struct list_head *firing) | |
1cd07c0b TG |
789 | { |
790 | struct posix_cputimer_base *base = pct->bases; | |
791 | int i; | |
792 | ||
793 | for (i = 0; i < CPUCLOCK_MAX; i++, base++) { | |
60bda037 TG |
794 | base->nextevt = collect_timerqueue(&base->tqhead, firing, |
795 | samples[i]); | |
1cd07c0b TG |
796 | } |
797 | } | |
798 | ||
34be3930 JL |
799 | static inline void check_dl_overrun(struct task_struct *tsk) |
800 | { | |
801 | if (tsk->dl.dl_overrun) { | |
802 | tsk->dl.dl_overrun = 0; | |
803 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); | |
804 | } | |
805 | } | |
806 | ||
8991afe2 TG |
807 | static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard) |
808 | { | |
809 | if (time < limit) | |
810 | return false; | |
811 | ||
812 | if (print_fatal_signals) { | |
813 | pr_info("%s Watchdog Timeout (%s): %s[%d]\n", | |
814 | rt ? "RT" : "CPU", hard ? "hard" : "soft", | |
815 | current->comm, task_pid_nr(current)); | |
816 | } | |
817 | __group_send_sig_info(signo, SEND_SIG_PRIV, current); | |
818 | return true; | |
819 | } | |
820 | ||
1da177e4 LT |
821 | /* |
822 | * Check for any per-thread CPU timers that have fired and move them off | |
823 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the | |
824 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. | |
825 | */ | |
826 | static void check_thread_timers(struct task_struct *tsk, | |
827 | struct list_head *firing) | |
828 | { | |
1cd07c0b TG |
829 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
830 | u64 samples[CPUCLOCK_MAX]; | |
d4bb5274 | 831 | unsigned long soft; |
1da177e4 | 832 | |
34be3930 JL |
833 | if (dl_task(tsk)) |
834 | check_dl_overrun(tsk); | |
835 | ||
1cd07c0b | 836 | if (expiry_cache_is_inactive(pct)) |
934715a1 JL |
837 | return; |
838 | ||
1cd07c0b TG |
839 | task_sample_cputime(tsk, samples); |
840 | collect_posix_cputimers(pct, samples, firing); | |
78f2c7db PZ |
841 | |
842 | /* | |
843 | * Check for the special case thread timers. | |
844 | */ | |
3cf29496 | 845 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
d4bb5274 | 846 | if (soft != RLIM_INFINITY) { |
8ea1de90 | 847 | /* Task RT timeout is accounted in jiffies. RTTIME is usec */ |
8991afe2 | 848 | unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ); |
3cf29496 | 849 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
78f2c7db | 850 | |
8991afe2 TG |
851 | /* At the hard limit, send SIGKILL. No further action. */ |
852 | if (hard != RLIM_INFINITY && | |
853 | check_rlimit(rttime, hard, SIGKILL, true, true)) | |
78f2c7db | 854 | return; |
dd670224 | 855 | |
8991afe2 TG |
856 | /* At the soft limit, send a SIGXCPU every second */ |
857 | if (check_rlimit(rttime, soft, SIGXCPU, true, false)) { | |
dd670224 TG |
858 | soft += USEC_PER_SEC; |
859 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft; | |
78f2c7db PZ |
860 | } |
861 | } | |
c02b078e | 862 | |
1cd07c0b | 863 | if (expiry_cache_is_inactive(pct)) |
b7878300 | 864 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); |
1da177e4 LT |
865 | } |
866 | ||
1018016c | 867 | static inline void stop_process_timers(struct signal_struct *sig) |
3fccfd67 | 868 | { |
244d49e3 | 869 | struct posix_cputimers *pct = &sig->posix_cputimers; |
3fccfd67 | 870 | |
244d49e3 TG |
871 | /* Turn off the active flag. This is done without locking. */ |
872 | WRITE_ONCE(pct->timers_active, false); | |
b7878300 | 873 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
3fccfd67 PZ |
874 | } |
875 | ||
42c4ab41 | 876 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
ebd7e7fc | 877 | u64 *expires, u64 cur_time, int signo) |
42c4ab41 | 878 | { |
64861634 | 879 | if (!it->expires) |
42c4ab41 SG |
880 | return; |
881 | ||
858cf3a8 FW |
882 | if (cur_time >= it->expires) { |
883 | if (it->incr) | |
64861634 | 884 | it->expires += it->incr; |
858cf3a8 | 885 | else |
64861634 | 886 | it->expires = 0; |
42c4ab41 | 887 | |
3f0a525e XG |
888 | trace_itimer_expire(signo == SIGPROF ? |
889 | ITIMER_PROF : ITIMER_VIRTUAL, | |
6883f81a | 890 | task_tgid(tsk), cur_time); |
42c4ab41 SG |
891 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
892 | } | |
893 | ||
2bbdbdae | 894 | if (it->expires && it->expires < *expires) |
858cf3a8 | 895 | *expires = it->expires; |
42c4ab41 SG |
896 | } |
897 | ||
1da177e4 LT |
898 | /* |
899 | * Check for any per-thread CPU timers that have fired and move them | |
900 | * off the tsk->*_timers list onto the firing list. Per-thread timers | |
901 | * have already been taken off. | |
902 | */ | |
903 | static void check_process_timers(struct task_struct *tsk, | |
904 | struct list_head *firing) | |
905 | { | |
906 | struct signal_struct *const sig = tsk->signal; | |
1cd07c0b TG |
907 | struct posix_cputimers *pct = &sig->posix_cputimers; |
908 | u64 samples[CPUCLOCK_MAX]; | |
d4bb5274 | 909 | unsigned long soft; |
1da177e4 | 910 | |
934715a1 | 911 | /* |
244d49e3 | 912 | * If there are no active process wide timers (POSIX 1.b, itimers, |
a2ed4fd6 TG |
913 | * RLIMIT_CPU) nothing to check. Also skip the process wide timer |
914 | * processing when there is already another task handling them. | |
934715a1 | 915 | */ |
a2ed4fd6 | 916 | if (!READ_ONCE(pct->timers_active) || pct->expiry_active) |
934715a1 JL |
917 | return; |
918 | ||
a2ed4fd6 | 919 | /* |
c8d75aa4 JL |
920 | * Signify that a thread is checking for process timers. |
921 | * Write access to this field is protected by the sighand lock. | |
922 | */ | |
a2ed4fd6 | 923 | pct->expiry_active = true; |
c8d75aa4 | 924 | |
1da177e4 | 925 | /* |
a324956f TG |
926 | * Collect the current process totals. Group accounting is active |
927 | * so the sample can be taken directly. | |
1da177e4 | 928 | */ |
b7be4ef1 | 929 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples); |
1cd07c0b | 930 | collect_posix_cputimers(pct, samples, firing); |
1da177e4 LT |
931 | |
932 | /* | |
933 | * Check for the special case process timers. | |
934 | */ | |
1cd07c0b TG |
935 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], |
936 | &pct->bases[CPUCLOCK_PROF].nextevt, | |
b7be4ef1 | 937 | samples[CPUCLOCK_PROF], SIGPROF); |
1cd07c0b TG |
938 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], |
939 | &pct->bases[CPUCLOCK_VIRT].nextevt, | |
940 | samples[CPUCLOCK_VIRT], SIGVTALRM); | |
b7be4ef1 | 941 | |
3cf29496 | 942 | soft = task_rlimit(tsk, RLIMIT_CPU); |
d4bb5274 | 943 | if (soft != RLIM_INFINITY) { |
8ea1de90 | 944 | /* RLIMIT_CPU is in seconds. Samples are nanoseconds */ |
3cf29496 | 945 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
8ea1de90 TG |
946 | u64 ptime = samples[CPUCLOCK_PROF]; |
947 | u64 softns = (u64)soft * NSEC_PER_SEC; | |
948 | u64 hardns = (u64)hard * NSEC_PER_SEC; | |
b7be4ef1 | 949 | |
8991afe2 TG |
950 | /* At the hard limit, send SIGKILL. No further action. */ |
951 | if (hard != RLIM_INFINITY && | |
952 | check_rlimit(ptime, hardns, SIGKILL, false, true)) | |
1da177e4 | 953 | return; |
dd670224 | 954 | |
8991afe2 TG |
955 | /* At the soft limit, send a SIGXCPU every second */ |
956 | if (check_rlimit(ptime, softns, SIGXCPU, false, false)) { | |
dd670224 TG |
957 | sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1; |
958 | softns += NSEC_PER_SEC; | |
1da177e4 | 959 | } |
8ea1de90 TG |
960 | |
961 | /* Update the expiry cache */ | |
1cd07c0b TG |
962 | if (softns < pct->bases[CPUCLOCK_PROF].nextevt) |
963 | pct->bases[CPUCLOCK_PROF].nextevt = softns; | |
1da177e4 LT |
964 | } |
965 | ||
1cd07c0b | 966 | if (expiry_cache_is_inactive(pct)) |
29f87b79 | 967 | stop_process_timers(sig); |
c8d75aa4 | 968 | |
244d49e3 | 969 | pct->expiry_active = false; |
1da177e4 LT |
970 | } |
971 | ||
972 | /* | |
96fe3b07 | 973 | * This is called from the signal code (via posixtimer_rearm) |
1da177e4 LT |
974 | * when the last timer signal was delivered and we have to reload the timer. |
975 | */ | |
f37fb0aa | 976 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
1da177e4 | 977 | { |
da020ce4 | 978 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
60bda037 TG |
979 | struct cpu_timer *ctmr = &timer->it.cpu; |
980 | struct task_struct *p = ctmr->task; | |
e73d84e3 FW |
981 | struct sighand_struct *sighand; |
982 | unsigned long flags; | |
ebd7e7fc | 983 | u64 now; |
1da177e4 | 984 | |
692117c1 TG |
985 | if (WARN_ON_ONCE(!p)) |
986 | return; | |
1da177e4 LT |
987 | |
988 | /* | |
989 | * Fetch the current sample and update the timer's expiry time. | |
990 | */ | |
991 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { | |
8c2d74f0 | 992 | now = cpu_clock_sample(clkid, p); |
1da177e4 | 993 | bump_cpu_timer(timer, now); |
724a3713 | 994 | if (unlikely(p->exit_state)) |
af888d67 | 995 | return; |
724a3713 | 996 | |
e73d84e3 FW |
997 | /* Protect timer list r/w in arm_timer() */ |
998 | sighand = lock_task_sighand(p, &flags); | |
999 | if (!sighand) | |
af888d67 | 1000 | return; |
1da177e4 | 1001 | } else { |
e73d84e3 FW |
1002 | /* |
1003 | * Protect arm_timer() and timer sampling in case of call to | |
ebd7e7fc | 1004 | * thread_group_cputime(). |
e73d84e3 FW |
1005 | */ |
1006 | sighand = lock_task_sighand(p, &flags); | |
1007 | if (unlikely(sighand == NULL)) { | |
1da177e4 LT |
1008 | /* |
1009 | * The process has been reaped. | |
1010 | * We can't even collect a sample any more. | |
1011 | */ | |
60bda037 | 1012 | cpu_timer_setexpires(ctmr, 0); |
af888d67 | 1013 | return; |
1da177e4 | 1014 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
af888d67 TG |
1015 | /* If the process is dying, no need to rearm */ |
1016 | goto unlock; | |
1da177e4 | 1017 | } |
8c2d74f0 | 1018 | now = cpu_clock_sample_group(clkid, p, true); |
1da177e4 | 1019 | bump_cpu_timer(timer, now); |
e73d84e3 | 1020 | /* Leave the sighand locked for the call below. */ |
1da177e4 LT |
1021 | } |
1022 | ||
1023 | /* | |
1024 | * Now re-arm for the new expiry time. | |
1025 | */ | |
5eb9aa64 | 1026 | arm_timer(timer); |
af888d67 | 1027 | unlock: |
e73d84e3 | 1028 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
1029 | } |
1030 | ||
f06febc9 | 1031 | /** |
87dc6448 | 1032 | * task_cputimers_expired - Check whether posix CPU timers are expired |
f06febc9 | 1033 | * |
001f7971 | 1034 | * @samples: Array of current samples for the CPUCLOCK clocks |
87dc6448 | 1035 | * @pct: Pointer to a posix_cputimers container |
f06febc9 | 1036 | * |
87dc6448 TG |
1037 | * Returns true if any member of @samples is greater than the corresponding |
1038 | * member of @pct->bases[CLK].nextevt. False otherwise | |
f06febc9 | 1039 | */ |
87dc6448 TG |
1040 | static inline bool |
1041 | task_cputimers_expired(const u64 *sample, struct posix_cputimers *pct) | |
f06febc9 | 1042 | { |
001f7971 TG |
1043 | int i; |
1044 | ||
1045 | for (i = 0; i < CPUCLOCK_MAX; i++) { | |
2bbdbdae | 1046 | if (sample[i] >= pct->bases[i].nextevt) |
001f7971 TG |
1047 | return true; |
1048 | } | |
1049 | return false; | |
f06febc9 FM |
1050 | } |
1051 | ||
1052 | /** | |
1053 | * fastpath_timer_check - POSIX CPU timers fast path. | |
1054 | * | |
1055 | * @tsk: The task (thread) being checked. | |
f06febc9 | 1056 | * |
bb34d92f FM |
1057 | * Check the task and thread group timers. If both are zero (there are no |
1058 | * timers set) return false. Otherwise snapshot the task and thread group | |
1059 | * timers and compare them with the corresponding expiration times. Return | |
1060 | * true if a timer has expired, else return false. | |
f06febc9 | 1061 | */ |
001f7971 | 1062 | static inline bool fastpath_timer_check(struct task_struct *tsk) |
f06febc9 | 1063 | { |
244d49e3 | 1064 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
ad133ba3 | 1065 | struct signal_struct *sig; |
bb34d92f | 1066 | |
244d49e3 | 1067 | if (!expiry_cache_is_inactive(pct)) { |
001f7971 | 1068 | u64 samples[CPUCLOCK_MAX]; |
bb34d92f | 1069 | |
001f7971 | 1070 | task_sample_cputime(tsk, samples); |
244d49e3 | 1071 | if (task_cputimers_expired(samples, pct)) |
001f7971 | 1072 | return true; |
bb34d92f | 1073 | } |
ad133ba3 ON |
1074 | |
1075 | sig = tsk->signal; | |
244d49e3 | 1076 | pct = &sig->posix_cputimers; |
c8d75aa4 | 1077 | /* |
244d49e3 TG |
1078 | * Check if thread group timers expired when timers are active and |
1079 | * no other thread in the group is already handling expiry for | |
1080 | * thread group cputimers. These fields are read without the | |
1081 | * sighand lock. However, this is fine because this is meant to be | |
1082 | * a fastpath heuristic to determine whether we should try to | |
1083 | * acquire the sighand lock to handle timer expiry. | |
c8d75aa4 | 1084 | * |
244d49e3 TG |
1085 | * In the worst case scenario, if concurrently timers_active is set |
1086 | * or expiry_active is cleared, but the current thread doesn't see | |
1087 | * the change yet, the timer checks are delayed until the next | |
1088 | * thread in the group gets a scheduler interrupt to handle the | |
1089 | * timer. This isn't an issue in practice because these types of | |
1090 | * delays with signals actually getting sent are expected. | |
c8d75aa4 | 1091 | */ |
244d49e3 | 1092 | if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) { |
001f7971 | 1093 | u64 samples[CPUCLOCK_MAX]; |
bb34d92f | 1094 | |
001f7971 TG |
1095 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, |
1096 | samples); | |
8d1f431c | 1097 | |
244d49e3 | 1098 | if (task_cputimers_expired(samples, pct)) |
001f7971 | 1099 | return true; |
bb34d92f | 1100 | } |
37bebc70 | 1101 | |
34be3930 | 1102 | if (dl_task(tsk) && tsk->dl.dl_overrun) |
001f7971 | 1103 | return true; |
34be3930 | 1104 | |
001f7971 | 1105 | return false; |
f06febc9 FM |
1106 | } |
1107 | ||
1da177e4 LT |
1108 | /* |
1109 | * This is called from the timer interrupt handler. The irq handler has | |
1110 | * already updated our counts. We need to check if any timers fire now. | |
1111 | * Interrupts are disabled. | |
1112 | */ | |
dce3e8fd | 1113 | void run_posix_cpu_timers(void) |
1da177e4 | 1114 | { |
dce3e8fd | 1115 | struct task_struct *tsk = current; |
1da177e4 | 1116 | struct k_itimer *timer, *next; |
0bdd2ed4 | 1117 | unsigned long flags; |
dce3e8fd | 1118 | LIST_HEAD(firing); |
1da177e4 | 1119 | |
a6968220 | 1120 | lockdep_assert_irqs_disabled(); |
1da177e4 | 1121 | |
1da177e4 | 1122 | /* |
f06febc9 | 1123 | * The fast path checks that there are no expired thread or thread |
bb34d92f | 1124 | * group timers. If that's so, just return. |
1da177e4 | 1125 | */ |
bb34d92f | 1126 | if (!fastpath_timer_check(tsk)) |
f06febc9 | 1127 | return; |
5ce73a4a | 1128 | |
0bdd2ed4 ON |
1129 | if (!lock_task_sighand(tsk, &flags)) |
1130 | return; | |
bb34d92f FM |
1131 | /* |
1132 | * Here we take off tsk->signal->cpu_timers[N] and | |
1133 | * tsk->cpu_timers[N] all the timers that are firing, and | |
1134 | * put them on the firing list. | |
1135 | */ | |
1136 | check_thread_timers(tsk, &firing); | |
934715a1 JL |
1137 | |
1138 | check_process_timers(tsk, &firing); | |
1da177e4 | 1139 | |
bb34d92f FM |
1140 | /* |
1141 | * We must release these locks before taking any timer's lock. | |
1142 | * There is a potential race with timer deletion here, as the | |
1143 | * siglock now protects our private firing list. We have set | |
1144 | * the firing flag in each timer, so that a deletion attempt | |
1145 | * that gets the timer lock before we do will give it up and | |
1146 | * spin until we've taken care of that timer below. | |
1147 | */ | |
0bdd2ed4 | 1148 | unlock_task_sighand(tsk, &flags); |
1da177e4 LT |
1149 | |
1150 | /* | |
1151 | * Now that all the timers on our list have the firing flag, | |
25985edc | 1152 | * no one will touch their list entries but us. We'll take |
1da177e4 LT |
1153 | * each timer's lock before clearing its firing flag, so no |
1154 | * timer call will interfere. | |
1155 | */ | |
60bda037 | 1156 | list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) { |
6e85c5ba HS |
1157 | int cpu_firing; |
1158 | ||
1da177e4 | 1159 | spin_lock(&timer->it_lock); |
60bda037 | 1160 | list_del_init(&timer->it.cpu.elist); |
6e85c5ba | 1161 | cpu_firing = timer->it.cpu.firing; |
1da177e4 LT |
1162 | timer->it.cpu.firing = 0; |
1163 | /* | |
1164 | * The firing flag is -1 if we collided with a reset | |
1165 | * of the timer, which already reported this | |
1166 | * almost-firing as an overrun. So don't generate an event. | |
1167 | */ | |
6e85c5ba | 1168 | if (likely(cpu_firing >= 0)) |
1da177e4 | 1169 | cpu_timer_fire(timer); |
1da177e4 LT |
1170 | spin_unlock(&timer->it_lock); |
1171 | } | |
1172 | } | |
1173 | ||
1174 | /* | |
f55db609 | 1175 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
f06febc9 | 1176 | * The tsk->sighand->siglock must be held by the caller. |
1da177e4 | 1177 | */ |
1b0dd96d | 1178 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid, |
858cf3a8 | 1179 | u64 *newval, u64 *oldval) |
1da177e4 | 1180 | { |
87dc6448 | 1181 | u64 now, *nextevt; |
1da177e4 | 1182 | |
1b0dd96d | 1183 | if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED)) |
692117c1 TG |
1184 | return; |
1185 | ||
87dc6448 | 1186 | nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt; |
1b0dd96d | 1187 | now = cpu_clock_sample_group(clkid, tsk, true); |
1da177e4 | 1188 | |
5405d005 | 1189 | if (oldval) { |
f55db609 SG |
1190 | /* |
1191 | * We are setting itimer. The *oldval is absolute and we update | |
1192 | * it to be relative, *newval argument is relative and we update | |
1193 | * it to be absolute. | |
1194 | */ | |
64861634 | 1195 | if (*oldval) { |
858cf3a8 | 1196 | if (*oldval <= now) { |
1da177e4 | 1197 | /* Just about to fire. */ |
858cf3a8 | 1198 | *oldval = TICK_NSEC; |
1da177e4 | 1199 | } else { |
858cf3a8 | 1200 | *oldval -= now; |
1da177e4 LT |
1201 | } |
1202 | } | |
1203 | ||
64861634 | 1204 | if (!*newval) |
b7878300 | 1205 | return; |
858cf3a8 | 1206 | *newval += now; |
1da177e4 LT |
1207 | } |
1208 | ||
1209 | /* | |
1b0dd96d TG |
1210 | * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF |
1211 | * expiry cache is also used by RLIMIT_CPU!. | |
1da177e4 | 1212 | */ |
2bbdbdae | 1213 | if (*newval < *nextevt) |
87dc6448 | 1214 | *nextevt = *newval; |
b7878300 FW |
1215 | |
1216 | tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER); | |
1da177e4 LT |
1217 | } |
1218 | ||
e4b76555 | 1219 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
343d8fc2 | 1220 | const struct timespec64 *rqtp) |
1da177e4 | 1221 | { |
86a9c446 | 1222 | struct itimerspec64 it; |
343d8fc2 TG |
1223 | struct k_itimer timer; |
1224 | u64 expires; | |
1da177e4 LT |
1225 | int error; |
1226 | ||
1da177e4 LT |
1227 | /* |
1228 | * Set up a temporary timer and then wait for it to go off. | |
1229 | */ | |
1230 | memset(&timer, 0, sizeof timer); | |
1231 | spin_lock_init(&timer.it_lock); | |
1232 | timer.it_clock = which_clock; | |
1233 | timer.it_overrun = -1; | |
1234 | error = posix_cpu_timer_create(&timer); | |
1235 | timer.it_process = current; | |
60bda037 | 1236 | |
1da177e4 | 1237 | if (!error) { |
5f252b32 | 1238 | static struct itimerspec64 zero_it; |
edbeda46 | 1239 | struct restart_block *restart; |
e4b76555 | 1240 | |
edbeda46 | 1241 | memset(&it, 0, sizeof(it)); |
86a9c446 | 1242 | it.it_value = *rqtp; |
1da177e4 LT |
1243 | |
1244 | spin_lock_irq(&timer.it_lock); | |
86a9c446 | 1245 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
1da177e4 LT |
1246 | if (error) { |
1247 | spin_unlock_irq(&timer.it_lock); | |
1248 | return error; | |
1249 | } | |
1250 | ||
1251 | while (!signal_pending(current)) { | |
60bda037 | 1252 | if (!cpu_timer_getexpires(&timer.it.cpu)) { |
1da177e4 | 1253 | /* |
e6c42c29 SG |
1254 | * Our timer fired and was reset, below |
1255 | * deletion can not fail. | |
1da177e4 | 1256 | */ |
e6c42c29 | 1257 | posix_cpu_timer_del(&timer); |
1da177e4 LT |
1258 | spin_unlock_irq(&timer.it_lock); |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Block until cpu_timer_fire (or a signal) wakes us. | |
1264 | */ | |
1265 | __set_current_state(TASK_INTERRUPTIBLE); | |
1266 | spin_unlock_irq(&timer.it_lock); | |
1267 | schedule(); | |
1268 | spin_lock_irq(&timer.it_lock); | |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * We were interrupted by a signal. | |
1273 | */ | |
60bda037 | 1274 | expires = cpu_timer_getexpires(&timer.it.cpu); |
86a9c446 | 1275 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
e6c42c29 SG |
1276 | if (!error) { |
1277 | /* | |
1278 | * Timer is now unarmed, deletion can not fail. | |
1279 | */ | |
1280 | posix_cpu_timer_del(&timer); | |
1281 | } | |
1da177e4 LT |
1282 | spin_unlock_irq(&timer.it_lock); |
1283 | ||
e6c42c29 SG |
1284 | while (error == TIMER_RETRY) { |
1285 | /* | |
1286 | * We need to handle case when timer was or is in the | |
1287 | * middle of firing. In other cases we already freed | |
1288 | * resources. | |
1289 | */ | |
1290 | spin_lock_irq(&timer.it_lock); | |
1291 | error = posix_cpu_timer_del(&timer); | |
1292 | spin_unlock_irq(&timer.it_lock); | |
1293 | } | |
1294 | ||
86a9c446 | 1295 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
1da177e4 LT |
1296 | /* |
1297 | * It actually did fire already. | |
1298 | */ | |
1299 | return 0; | |
1300 | } | |
1301 | ||
e4b76555 | 1302 | error = -ERESTART_RESTARTBLOCK; |
86a9c446 AV |
1303 | /* |
1304 | * Report back to the user the time still remaining. | |
1305 | */ | |
edbeda46 | 1306 | restart = ¤t->restart_block; |
343d8fc2 | 1307 | restart->nanosleep.expires = expires; |
c0edd7c9 DD |
1308 | if (restart->nanosleep.type != TT_NONE) |
1309 | error = nanosleep_copyout(restart, &it.it_value); | |
e4b76555 TA |
1310 | } |
1311 | ||
1312 | return error; | |
1313 | } | |
1314 | ||
bc2c8ea4 TG |
1315 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
1316 | ||
1317 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, | |
938e7cf2 | 1318 | const struct timespec64 *rqtp) |
e4b76555 | 1319 | { |
f56141e3 | 1320 | struct restart_block *restart_block = ¤t->restart_block; |
e4b76555 TA |
1321 | int error; |
1322 | ||
1323 | /* | |
1324 | * Diagnose required errors first. | |
1325 | */ | |
1326 | if (CPUCLOCK_PERTHREAD(which_clock) && | |
1327 | (CPUCLOCK_PID(which_clock) == 0 || | |
01a21974 | 1328 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
e4b76555 TA |
1329 | return -EINVAL; |
1330 | ||
86a9c446 | 1331 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
e4b76555 TA |
1332 | |
1333 | if (error == -ERESTART_RESTARTBLOCK) { | |
1334 | ||
3751f9f2 | 1335 | if (flags & TIMER_ABSTIME) |
e4b76555 | 1336 | return -ERESTARTNOHAND; |
1da177e4 | 1337 | |
1711ef38 | 1338 | restart_block->fn = posix_cpu_nsleep_restart; |
ab8177bc | 1339 | restart_block->nanosleep.clockid = which_clock; |
1da177e4 | 1340 | } |
1da177e4 LT |
1341 | return error; |
1342 | } | |
1343 | ||
bc2c8ea4 | 1344 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1da177e4 | 1345 | { |
ab8177bc | 1346 | clockid_t which_clock = restart_block->nanosleep.clockid; |
ad196384 | 1347 | struct timespec64 t; |
97735f25 | 1348 | |
ad196384 | 1349 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
97735f25 | 1350 | |
86a9c446 | 1351 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
1da177e4 LT |
1352 | } |
1353 | ||
29f1b2b0 ND |
1354 | #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED) |
1355 | #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED) | |
1da177e4 | 1356 | |
a924b04d | 1357 | static int process_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1358 | struct timespec64 *tp) |
1da177e4 LT |
1359 | { |
1360 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); | |
1361 | } | |
a924b04d | 1362 | static int process_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1363 | struct timespec64 *tp) |
1da177e4 LT |
1364 | { |
1365 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); | |
1366 | } | |
1367 | static int process_cpu_timer_create(struct k_itimer *timer) | |
1368 | { | |
1369 | timer->it_clock = PROCESS_CLOCK; | |
1370 | return posix_cpu_timer_create(timer); | |
1371 | } | |
a924b04d | 1372 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
938e7cf2 | 1373 | const struct timespec64 *rqtp) |
1da177e4 | 1374 | { |
99e6c0e6 | 1375 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
1da177e4 | 1376 | } |
a924b04d | 1377 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1378 | struct timespec64 *tp) |
1da177e4 LT |
1379 | { |
1380 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); | |
1381 | } | |
a924b04d | 1382 | static int thread_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1383 | struct timespec64 *tp) |
1da177e4 LT |
1384 | { |
1385 | return posix_cpu_clock_get(THREAD_CLOCK, tp); | |
1386 | } | |
1387 | static int thread_cpu_timer_create(struct k_itimer *timer) | |
1388 | { | |
1389 | timer->it_clock = THREAD_CLOCK; | |
1390 | return posix_cpu_timer_create(timer); | |
1391 | } | |
1da177e4 | 1392 | |
d3ba5a9a | 1393 | const struct k_clock clock_posix_cpu = { |
1976945e TG |
1394 | .clock_getres = posix_cpu_clock_getres, |
1395 | .clock_set = posix_cpu_clock_set, | |
1396 | .clock_get = posix_cpu_clock_get, | |
1397 | .timer_create = posix_cpu_timer_create, | |
1398 | .nsleep = posix_cpu_nsleep, | |
1976945e TG |
1399 | .timer_set = posix_cpu_timer_set, |
1400 | .timer_del = posix_cpu_timer_del, | |
1401 | .timer_get = posix_cpu_timer_get, | |
f37fb0aa | 1402 | .timer_rearm = posix_cpu_timer_rearm, |
1976945e TG |
1403 | }; |
1404 | ||
d3ba5a9a CH |
1405 | const struct k_clock clock_process = { |
1406 | .clock_getres = process_cpu_clock_getres, | |
1407 | .clock_get = process_cpu_clock_get, | |
1408 | .timer_create = process_cpu_timer_create, | |
1409 | .nsleep = process_cpu_nsleep, | |
d3ba5a9a | 1410 | }; |
1da177e4 | 1411 | |
d3ba5a9a CH |
1412 | const struct k_clock clock_thread = { |
1413 | .clock_getres = thread_cpu_clock_getres, | |
1414 | .clock_get = thread_cpu_clock_get, | |
1415 | .timer_create = thread_cpu_timer_create, | |
1416 | }; |