]>
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 | */ | |
96498773 | 50 | static struct pid *pid_for_clock(const clockid_t clock, bool gettime) |
1da177e4 | 51 | { |
96498773 EB |
52 | const bool thread = !!CPUCLOCK_PERTHREAD(clock); |
53 | const pid_t upid = CPUCLOCK_PID(clock); | |
54 | struct pid *pid; | |
55 | ||
56 | if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX) | |
57 | return NULL; | |
1da177e4 | 58 | |
77b4b542 TG |
59 | /* |
60 | * If the encoded PID is 0, then the timer is targeted at current | |
61 | * or the process to which current belongs. | |
62 | */ | |
96498773 EB |
63 | if (upid == 0) |
64 | return thread ? task_pid(current) : task_tgid(current); | |
1da177e4 | 65 | |
96498773 EB |
66 | pid = find_vpid(upid); |
67 | if (!pid) | |
68 | return NULL; | |
77b4b542 | 69 | |
96498773 EB |
70 | if (thread) { |
71 | struct task_struct *tsk = pid_task(pid, PIDTYPE_PID); | |
72 | return (tsk && same_thread_group(tsk, current)) ? pid : NULL; | |
73 | } | |
77b4b542 | 74 | |
c7f51940 | 75 | /* |
96498773 EB |
76 | * For clock_gettime(PROCESS) allow finding the process by |
77 | * with the pid of the current task. The code needs the tgid | |
78 | * of the process so that pid_task(pid, PIDTYPE_TGID) can be | |
79 | * used to find the process. | |
c7f51940 | 80 | */ |
96498773 EB |
81 | if (gettime && (pid == task_pid(current))) |
82 | return task_tgid(current); | |
77b4b542 TG |
83 | |
84 | /* | |
96498773 | 85 | * For processes require that pid identifies a process. |
77b4b542 | 86 | */ |
96498773 | 87 | return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL; |
6ae40e3f TG |
88 | } |
89 | ||
90 | static inline int validate_clock_permissions(const clockid_t clock) | |
91 | { | |
9bf7c324 EB |
92 | int ret; |
93 | ||
94 | rcu_read_lock(); | |
96498773 | 95 | ret = pid_for_clock(clock, false) ? 0 : -EINVAL; |
9bf7c324 EB |
96 | rcu_read_unlock(); |
97 | ||
98 | return ret; | |
1da177e4 LT |
99 | } |
100 | ||
fece9826 | 101 | static inline enum pid_type clock_pid_type(const clockid_t clock) |
55e8c8eb | 102 | { |
fece9826 | 103 | return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID; |
55e8c8eb EB |
104 | } |
105 | ||
106 | static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer) | |
107 | { | |
fece9826 | 108 | return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock)); |
55e8c8eb EB |
109 | } |
110 | ||
1da177e4 LT |
111 | /* |
112 | * Update expiry time from increment, and increase overrun count, | |
113 | * given the current clock sample. | |
114 | */ | |
60bda037 | 115 | static u64 bump_cpu_timer(struct k_itimer *timer, u64 now) |
1da177e4 | 116 | { |
60bda037 | 117 | u64 delta, incr, expires = timer->it.cpu.node.expires; |
1da177e4 LT |
118 | int i; |
119 | ||
16118794 | 120 | if (!timer->it_interval) |
60bda037 | 121 | return expires; |
1da177e4 | 122 | |
60bda037 TG |
123 | if (now < expires) |
124 | return expires; | |
1da177e4 | 125 | |
16118794 | 126 | incr = timer->it_interval; |
60bda037 | 127 | delta = now + incr - expires; |
1da177e4 | 128 | |
55ccb616 FW |
129 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
130 | for (i = 0; incr < delta - incr; i++) | |
131 | incr = incr << 1; | |
132 | ||
133 | for (; i >= 0; incr >>= 1, i--) { | |
134 | if (delta < incr) | |
135 | continue; | |
136 | ||
60bda037 | 137 | timer->it.cpu.node.expires += incr; |
78c9c4df | 138 | timer->it_overrun += 1LL << i; |
55ccb616 | 139 | delta -= incr; |
1da177e4 | 140 | } |
60bda037 | 141 | return timer->it.cpu.node.expires; |
1da177e4 LT |
142 | } |
143 | ||
2bbdbdae TG |
144 | /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */ |
145 | static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct) | |
555347f6 | 146 | { |
2bbdbdae TG |
147 | return !(~pct->bases[CPUCLOCK_PROF].nextevt | |
148 | ~pct->bases[CPUCLOCK_VIRT].nextevt | | |
149 | ~pct->bases[CPUCLOCK_SCHED].nextevt); | |
555347f6 FW |
150 | } |
151 | ||
bc2c8ea4 | 152 | static int |
d2e3e0ca | 153 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp) |
1da177e4 | 154 | { |
6ae40e3f TG |
155 | int error = validate_clock_permissions(which_clock); |
156 | ||
1da177e4 LT |
157 | if (!error) { |
158 | tp->tv_sec = 0; | |
159 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); | |
160 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { | |
161 | /* | |
162 | * If sched_clock is using a cycle counter, we | |
163 | * don't have any idea of its true resolution | |
164 | * exported, but it is much more than 1s/HZ. | |
165 | */ | |
166 | tp->tv_nsec = 1; | |
167 | } | |
168 | } | |
169 | return error; | |
170 | } | |
171 | ||
bc2c8ea4 | 172 | static int |
6ae40e3f | 173 | posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp) |
1da177e4 | 174 | { |
6ae40e3f TG |
175 | int error = validate_clock_permissions(clock); |
176 | ||
1da177e4 LT |
177 | /* |
178 | * You can never reset a CPU clock, but we check for other errors | |
179 | * in the call before failing with EPERM. | |
180 | */ | |
6ae40e3f | 181 | return error ? : -EPERM; |
1da177e4 LT |
182 | } |
183 | ||
1da177e4 | 184 | /* |
2092c1d4 | 185 | * Sample a per-thread clock for the given task. clkid is validated. |
1da177e4 | 186 | */ |
8c2d74f0 | 187 | static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p) |
1da177e4 | 188 | { |
ab693c5a TG |
189 | u64 utime, stime; |
190 | ||
191 | if (clkid == CPUCLOCK_SCHED) | |
192 | return task_sched_runtime(p); | |
193 | ||
194 | task_cputime(p, &utime, &stime); | |
195 | ||
2092c1d4 | 196 | switch (clkid) { |
1da177e4 | 197 | case CPUCLOCK_PROF: |
ab693c5a | 198 | return utime + stime; |
1da177e4 | 199 | case CPUCLOCK_VIRT: |
ab693c5a | 200 | return utime; |
2092c1d4 TG |
201 | default: |
202 | WARN_ON_ONCE(1); | |
1da177e4 | 203 | } |
8c2d74f0 | 204 | return 0; |
1da177e4 LT |
205 | } |
206 | ||
b0d524f7 TG |
207 | static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime) |
208 | { | |
209 | samples[CPUCLOCK_PROF] = stime + utime; | |
210 | samples[CPUCLOCK_VIRT] = utime; | |
211 | samples[CPUCLOCK_SCHED] = rtime; | |
212 | } | |
213 | ||
214 | static void task_sample_cputime(struct task_struct *p, u64 *samples) | |
215 | { | |
216 | u64 stime, utime; | |
217 | ||
218 | task_cputime(p, &utime, &stime); | |
219 | store_samples(samples, stime, utime, p->se.sum_exec_runtime); | |
220 | } | |
221 | ||
222 | static void proc_sample_cputime_atomic(struct task_cputime_atomic *at, | |
223 | u64 *samples) | |
224 | { | |
225 | u64 stime, utime, rtime; | |
226 | ||
227 | utime = atomic64_read(&at->utime); | |
228 | stime = atomic64_read(&at->stime); | |
229 | rtime = atomic64_read(&at->sum_exec_runtime); | |
230 | store_samples(samples, stime, utime, rtime); | |
231 | } | |
232 | ||
1018016c JL |
233 | /* |
234 | * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg | |
235 | * to avoid race conditions with concurrent updates to cputime. | |
236 | */ | |
237 | static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime) | |
4da94d49 | 238 | { |
1018016c JL |
239 | u64 curr_cputime; |
240 | retry: | |
241 | curr_cputime = atomic64_read(cputime); | |
242 | if (sum_cputime > curr_cputime) { | |
243 | if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime) | |
244 | goto retry; | |
245 | } | |
246 | } | |
4da94d49 | 247 | |
b7be4ef1 TG |
248 | static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, |
249 | struct task_cputime *sum) | |
1018016c | 250 | { |
71107445 JL |
251 | __update_gt_cputime(&cputime_atomic->utime, sum->utime); |
252 | __update_gt_cputime(&cputime_atomic->stime, sum->stime); | |
253 | __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime); | |
1018016c | 254 | } |
4da94d49 | 255 | |
19298fbf TG |
256 | /** |
257 | * thread_group_sample_cputime - Sample cputime for a given task | |
258 | * @tsk: Task for which cputime needs to be started | |
7f2cbcbc | 259 | * @samples: Storage for time samples |
19298fbf TG |
260 | * |
261 | * Called from sys_getitimer() to calculate the expiry time of an active | |
262 | * timer. That means group cputime accounting is already active. Called | |
263 | * with task sighand lock held. | |
264 | * | |
265 | * Updates @times with an uptodate sample of the thread group cputimes. | |
266 | */ | |
b7be4ef1 | 267 | void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples) |
19298fbf TG |
268 | { |
269 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
244d49e3 | 270 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
19298fbf | 271 | |
244d49e3 | 272 | WARN_ON_ONCE(!pct->timers_active); |
19298fbf | 273 | |
b7be4ef1 | 274 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
19298fbf TG |
275 | } |
276 | ||
c506bef4 TG |
277 | /** |
278 | * thread_group_start_cputime - Start cputime and return a sample | |
279 | * @tsk: Task for which cputime needs to be started | |
b7be4ef1 | 280 | * @samples: Storage for time samples |
c506bef4 | 281 | * |
4bf07f65 | 282 | * The thread group cputime accounting is avoided when there are no posix |
c506bef4 TG |
283 | * CPU timers armed. Before starting a timer it's required to check whether |
284 | * the time accounting is active. If not, a full update of the atomic | |
285 | * accounting store needs to be done and the accounting enabled. | |
286 | * | |
287 | * Updates @times with an uptodate sample of the thread group cputimes. | |
288 | */ | |
b7be4ef1 | 289 | static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples) |
4da94d49 PZ |
290 | { |
291 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; | |
244d49e3 | 292 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
4da94d49 | 293 | |
a5dec9f8 FW |
294 | lockdep_assert_task_sighand_held(tsk); |
295 | ||
1018016c | 296 | /* Check if cputimer isn't running. This is accessed without locking. */ |
244d49e3 | 297 | if (!READ_ONCE(pct->timers_active)) { |
b7be4ef1 TG |
298 | struct task_cputime sum; |
299 | ||
4da94d49 PZ |
300 | /* |
301 | * The POSIX timer interface allows for absolute time expiry | |
302 | * values through the TIMER_ABSTIME flag, therefore we have | |
1018016c | 303 | * to synchronize the timer to the clock every time we start it. |
4da94d49 | 304 | */ |
ebd7e7fc | 305 | thread_group_cputime(tsk, &sum); |
71107445 | 306 | update_gt_cputime(&cputimer->cputime_atomic, &sum); |
1018016c JL |
307 | |
308 | /* | |
244d49e3 TG |
309 | * We're setting timers_active without a lock. Ensure this |
310 | * only gets written to in one operation. We set it after | |
311 | * update_gt_cputime() as a small optimization, but | |
312 | * barriers are not required because update_gt_cputime() | |
1018016c JL |
313 | * can handle concurrent updates. |
314 | */ | |
244d49e3 | 315 | WRITE_ONCE(pct->timers_active, true); |
1018016c | 316 | } |
b7be4ef1 TG |
317 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
318 | } | |
319 | ||
320 | static void __thread_group_cputime(struct task_struct *tsk, u64 *samples) | |
321 | { | |
322 | struct task_cputime ct; | |
323 | ||
324 | thread_group_cputime(tsk, &ct); | |
325 | store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime); | |
4da94d49 PZ |
326 | } |
327 | ||
1da177e4 | 328 | /* |
24ab7f5a TG |
329 | * Sample a process (thread group) clock for the given task clkid. If the |
330 | * group's cputime accounting is already enabled, read the atomic | |
a2efdbf4 | 331 | * store. Otherwise a full update is required. clkid is already validated. |
1da177e4 | 332 | */ |
8c2d74f0 TG |
333 | static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p, |
334 | bool start) | |
1da177e4 | 335 | { |
24ab7f5a | 336 | struct thread_group_cputimer *cputimer = &p->signal->cputimer; |
244d49e3 | 337 | struct posix_cputimers *pct = &p->signal->posix_cputimers; |
b7be4ef1 | 338 | u64 samples[CPUCLOCK_MAX]; |
f06febc9 | 339 | |
244d49e3 | 340 | if (!READ_ONCE(pct->timers_active)) { |
24ab7f5a | 341 | if (start) |
b7be4ef1 | 342 | thread_group_start_cputime(p, samples); |
24ab7f5a | 343 | else |
b7be4ef1 | 344 | __thread_group_cputime(p, samples); |
24ab7f5a | 345 | } else { |
b7be4ef1 | 346 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
24ab7f5a TG |
347 | } |
348 | ||
b7be4ef1 | 349 | return samples[clkid]; |
1da177e4 LT |
350 | } |
351 | ||
bfcf3e92 | 352 | static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp) |
33ab0fec | 353 | { |
bfcf3e92 TG |
354 | const clockid_t clkid = CPUCLOCK_WHICH(clock); |
355 | struct task_struct *tsk; | |
356 | u64 t; | |
33ab0fec | 357 | |
9bf7c324 | 358 | rcu_read_lock(); |
96498773 | 359 | tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock)); |
9bf7c324 EB |
360 | if (!tsk) { |
361 | rcu_read_unlock(); | |
bfcf3e92 | 362 | return -EINVAL; |
9bf7c324 | 363 | } |
1da177e4 | 364 | |
bfcf3e92 | 365 | if (CPUCLOCK_PERTHREAD(clock)) |
8c2d74f0 | 366 | t = cpu_clock_sample(clkid, tsk); |
bfcf3e92 | 367 | else |
8c2d74f0 | 368 | t = cpu_clock_sample_group(clkid, tsk, false); |
9bf7c324 | 369 | rcu_read_unlock(); |
1da177e4 | 370 | |
bfcf3e92 TG |
371 | *tp = ns_to_timespec64(t); |
372 | return 0; | |
1da177e4 LT |
373 | } |
374 | ||
1da177e4 LT |
375 | /* |
376 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. | |
ba5ea951 SG |
377 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
378 | * new timer already all-zeros initialized. | |
1da177e4 | 379 | */ |
bc2c8ea4 | 380 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
1da177e4 | 381 | { |
1fb497dd | 382 | static struct lock_class_key posix_cpu_timers_key; |
96498773 | 383 | struct pid *pid; |
1da177e4 | 384 | |
9bf7c324 | 385 | rcu_read_lock(); |
96498773 EB |
386 | pid = pid_for_clock(new_timer->it_clock, false); |
387 | if (!pid) { | |
9bf7c324 | 388 | rcu_read_unlock(); |
1da177e4 | 389 | return -EINVAL; |
9bf7c324 | 390 | } |
1da177e4 | 391 | |
1fb497dd TG |
392 | /* |
393 | * If posix timer expiry is handled in task work context then | |
394 | * timer::it_lock can be taken without disabling interrupts as all | |
4bf07f65 | 395 | * other locking happens in task context. This requires a separate |
1fb497dd TG |
396 | * lock class key otherwise regular posix timer expiry would record |
397 | * the lock class being taken in interrupt context and generate a | |
398 | * false positive warning. | |
399 | */ | |
400 | if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK)) | |
401 | lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key); | |
402 | ||
d97bb75d | 403 | new_timer->kclock = &clock_posix_cpu; |
60bda037 | 404 | timerqueue_init(&new_timer->it.cpu.node); |
96498773 | 405 | new_timer->it.cpu.pid = get_pid(pid); |
9bf7c324 | 406 | rcu_read_unlock(); |
e5a8b65b | 407 | return 0; |
1da177e4 LT |
408 | } |
409 | ||
5c8f23e6 FW |
410 | static struct posix_cputimer_base *timer_base(struct k_itimer *timer, |
411 | struct task_struct *tsk) | |
412 | { | |
413 | int clkidx = CPUCLOCK_WHICH(timer->it_clock); | |
414 | ||
415 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) | |
416 | return tsk->posix_cputimers.bases + clkidx; | |
417 | else | |
418 | return tsk->signal->posix_cputimers.bases + clkidx; | |
419 | } | |
420 | ||
ee375328 FW |
421 | /* |
422 | * Force recalculating the base earliest expiration on the next tick. | |
423 | * This will also re-evaluate the need to keep around the process wide | |
424 | * cputime counter and tick dependency and eventually shut these down | |
425 | * if necessary. | |
426 | */ | |
427 | static void trigger_base_recalc_expires(struct k_itimer *timer, | |
428 | struct task_struct *tsk) | |
429 | { | |
430 | struct posix_cputimer_base *base = timer_base(timer, tsk); | |
431 | ||
432 | base->nextevt = 0; | |
433 | } | |
434 | ||
175cc3ab FW |
435 | /* |
436 | * Dequeue the timer and reset the base if it was its earliest expiration. | |
437 | * It makes sure the next tick recalculates the base next expiration so we | |
438 | * don't keep the costly process wide cputime counter around for a random | |
439 | * amount of time, along with the tick dependency. | |
440 | * | |
441 | * If another timer gets queued between this and the next tick, its | |
442 | * expiration will update the base next event if necessary on the next | |
443 | * tick. | |
444 | */ | |
445 | static void disarm_timer(struct k_itimer *timer, struct task_struct *p) | |
446 | { | |
447 | struct cpu_timer *ctmr = &timer->it.cpu; | |
448 | struct posix_cputimer_base *base; | |
175cc3ab FW |
449 | |
450 | if (!cpu_timer_dequeue(ctmr)) | |
451 | return; | |
452 | ||
5c8f23e6 | 453 | base = timer_base(timer, p); |
175cc3ab | 454 | if (cpu_timer_getexpires(ctmr) == base->nextevt) |
ee375328 | 455 | trigger_base_recalc_expires(timer, p); |
175cc3ab FW |
456 | } |
457 | ||
458 | ||
1da177e4 LT |
459 | /* |
460 | * Clean up a CPU-clock timer that is about to be destroyed. | |
461 | * This is called from timer deletion with the timer already locked. | |
462 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
463 | * and try again. (This happens when the timer is in the middle of firing.) | |
464 | */ | |
bc2c8ea4 | 465 | static int posix_cpu_timer_del(struct k_itimer *timer) |
1da177e4 | 466 | { |
60bda037 | 467 | struct cpu_timer *ctmr = &timer->it.cpu; |
3d7a1427 | 468 | struct sighand_struct *sighand; |
55e8c8eb | 469 | struct task_struct *p; |
60bda037 TG |
470 | unsigned long flags; |
471 | int ret = 0; | |
1da177e4 | 472 | |
55e8c8eb EB |
473 | rcu_read_lock(); |
474 | p = cpu_timer_task_rcu(timer); | |
475 | if (!p) | |
476 | goto out; | |
108150ea | 477 | |
3d7a1427 FW |
478 | /* |
479 | * Protect against sighand release/switch in exit/exec and process/ | |
480 | * thread timer list entry concurrent read/writes. | |
481 | */ | |
482 | sighand = lock_task_sighand(p, &flags); | |
483 | if (unlikely(sighand == NULL)) { | |
a3222f88 | 484 | /* |
60bda037 TG |
485 | * This raced with the reaping of the task. The exit cleanup |
486 | * should have removed this timer from the timer queue. | |
a3222f88 | 487 | */ |
60bda037 | 488 | WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node)); |
a3222f88 | 489 | } else { |
a3222f88 FW |
490 | if (timer->it.cpu.firing) |
491 | ret = TIMER_RETRY; | |
492 | else | |
175cc3ab | 493 | disarm_timer(timer, p); |
3d7a1427 FW |
494 | |
495 | unlock_task_sighand(p, &flags); | |
1da177e4 | 496 | } |
a3222f88 | 497 | |
55e8c8eb EB |
498 | out: |
499 | rcu_read_unlock(); | |
a3222f88 | 500 | if (!ret) |
55e8c8eb | 501 | put_pid(ctmr->pid); |
1da177e4 | 502 | |
108150ea | 503 | return ret; |
1da177e4 LT |
504 | } |
505 | ||
60bda037 | 506 | static void cleanup_timerqueue(struct timerqueue_head *head) |
1a7fa510 | 507 | { |
60bda037 TG |
508 | struct timerqueue_node *node; |
509 | struct cpu_timer *ctmr; | |
1a7fa510 | 510 | |
60bda037 TG |
511 | while ((node = timerqueue_getnext(head))) { |
512 | timerqueue_del(head, node); | |
513 | ctmr = container_of(node, struct cpu_timer, node); | |
514 | ctmr->head = NULL; | |
515 | } | |
1a7fa510 FW |
516 | } |
517 | ||
1da177e4 | 518 | /* |
7cb9a94c TG |
519 | * Clean out CPU timers which are still armed when a thread exits. The |
520 | * timers are only removed from the list. No other updates are done. The | |
521 | * corresponding posix timers are still accessible, but cannot be rearmed. | |
522 | * | |
1da177e4 LT |
523 | * This must be called with the siglock held. |
524 | */ | |
2b69942f | 525 | static void cleanup_timers(struct posix_cputimers *pct) |
1da177e4 | 526 | { |
60bda037 TG |
527 | cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead); |
528 | cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead); | |
529 | cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead); | |
1da177e4 LT |
530 | } |
531 | ||
532 | /* | |
533 | * These are both called with the siglock held, when the current thread | |
534 | * is being reaped. When the final (leader) thread in the group is reaped, | |
535 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. | |
536 | */ | |
537 | void posix_cpu_timers_exit(struct task_struct *tsk) | |
538 | { | |
2b69942f | 539 | cleanup_timers(&tsk->posix_cputimers); |
1da177e4 LT |
540 | } |
541 | void posix_cpu_timers_exit_group(struct task_struct *tsk) | |
542 | { | |
2b69942f | 543 | cleanup_timers(&tsk->signal->posix_cputimers); |
1da177e4 LT |
544 | } |
545 | ||
1da177e4 LT |
546 | /* |
547 | * Insert the timer on the appropriate list before any timers that | |
e73d84e3 | 548 | * expire later. This must be called with the sighand lock held. |
1da177e4 | 549 | */ |
beb41d9c | 550 | static void arm_timer(struct k_itimer *timer, struct task_struct *p) |
1da177e4 | 551 | { |
5c8f23e6 | 552 | struct posix_cputimer_base *base = timer_base(timer, p); |
60bda037 TG |
553 | struct cpu_timer *ctmr = &timer->it.cpu; |
554 | u64 newexp = cpu_timer_getexpires(ctmr); | |
1da177e4 | 555 | |
60bda037 | 556 | if (!cpu_timer_enqueue(&base->tqhead, ctmr)) |
3b495b22 | 557 | return; |
5eb9aa64 | 558 | |
3b495b22 TG |
559 | /* |
560 | * We are the new earliest-expiring POSIX 1.b timer, hence | |
561 | * need to update expiration cache. Take into account that | |
562 | * for process timers we share expiration cache with itimers | |
563 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. | |
564 | */ | |
2bbdbdae | 565 | if (newexp < base->nextevt) |
87dc6448 | 566 | base->nextevt = newexp; |
1da177e4 | 567 | |
3b495b22 TG |
568 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
569 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); | |
570 | else | |
1e4ca26d | 571 | tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER); |
1da177e4 LT |
572 | } |
573 | ||
574 | /* | |
575 | * The timer is locked, fire it and arrange for its reload. | |
576 | */ | |
577 | static void cpu_timer_fire(struct k_itimer *timer) | |
578 | { | |
60bda037 TG |
579 | struct cpu_timer *ctmr = &timer->it.cpu; |
580 | ||
1f169f84 SG |
581 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
582 | /* | |
583 | * User don't want any signal. | |
584 | */ | |
60bda037 | 585 | cpu_timer_setexpires(ctmr, 0); |
1f169f84 | 586 | } else if (unlikely(timer->sigq == NULL)) { |
1da177e4 LT |
587 | /* |
588 | * This a special case for clock_nanosleep, | |
589 | * not a normal timer from sys_timer_create. | |
590 | */ | |
591 | wake_up_process(timer->it_process); | |
60bda037 | 592 | cpu_timer_setexpires(ctmr, 0); |
16118794 | 593 | } else if (!timer->it_interval) { |
1da177e4 LT |
594 | /* |
595 | * One-shot timer. Clear it as soon as it's fired. | |
596 | */ | |
597 | posix_timer_event(timer, 0); | |
60bda037 | 598 | cpu_timer_setexpires(ctmr, 0); |
1da177e4 LT |
599 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
600 | /* | |
601 | * The signal did not get queued because the signal | |
602 | * was ignored, so we won't get any callback to | |
603 | * reload the timer. But we need to keep it | |
604 | * ticking in case the signal is deliverable next time. | |
605 | */ | |
f37fb0aa | 606 | posix_cpu_timer_rearm(timer); |
af888d67 | 607 | ++timer->it_requeue_pending; |
1da177e4 LT |
608 | } |
609 | } | |
610 | ||
611 | /* | |
612 | * Guts of sys_timer_settime for CPU timers. | |
613 | * This is called with the timer locked and interrupts disabled. | |
614 | * If we return TIMER_RETRY, it's necessary to release the timer's lock | |
615 | * and try again. (This happens when the timer is in the middle of firing.) | |
616 | */ | |
e73d84e3 | 617 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
5f252b32 | 618 | struct itimerspec64 *new, struct itimerspec64 *old) |
1da177e4 | 619 | { |
c7a37c6f | 620 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
ebd7e7fc | 621 | u64 old_expires, new_expires, old_incr, val; |
60bda037 | 622 | struct cpu_timer *ctmr = &timer->it.cpu; |
c7a37c6f | 623 | struct sighand_struct *sighand; |
55e8c8eb | 624 | struct task_struct *p; |
c7a37c6f | 625 | unsigned long flags; |
60bda037 | 626 | int ret = 0; |
1da177e4 | 627 | |
55e8c8eb EB |
628 | rcu_read_lock(); |
629 | p = cpu_timer_task_rcu(timer); | |
630 | if (!p) { | |
631 | /* | |
632 | * If p has just been reaped, we can no | |
633 | * longer get any information about it at all. | |
634 | */ | |
635 | rcu_read_unlock(); | |
636 | return -ESRCH; | |
637 | } | |
1da177e4 | 638 | |
098b0e01 TG |
639 | /* |
640 | * Use the to_ktime conversion because that clamps the maximum | |
641 | * value to KTIME_MAX and avoid multiplication overflows. | |
642 | */ | |
643 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); | |
1da177e4 | 644 | |
1da177e4 | 645 | /* |
e73d84e3 FW |
646 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
647 | * and p->signal->cpu_timers read/write in arm_timer() | |
648 | */ | |
649 | sighand = lock_task_sighand(p, &flags); | |
650 | /* | |
651 | * If p has just been reaped, we can no | |
1da177e4 LT |
652 | * longer get any information about it at all. |
653 | */ | |
55e8c8eb EB |
654 | if (unlikely(sighand == NULL)) { |
655 | rcu_read_unlock(); | |
1da177e4 | 656 | return -ESRCH; |
55e8c8eb | 657 | } |
1da177e4 LT |
658 | |
659 | /* | |
660 | * Disarm any old timer after extracting its expiry time. | |
661 | */ | |
16118794 | 662 | old_incr = timer->it_interval; |
60bda037 TG |
663 | old_expires = cpu_timer_getexpires(ctmr); |
664 | ||
a69ac4a7 ON |
665 | if (unlikely(timer->it.cpu.firing)) { |
666 | timer->it.cpu.firing = -1; | |
667 | ret = TIMER_RETRY; | |
60bda037 TG |
668 | } else { |
669 | cpu_timer_dequeue(ctmr); | |
670 | } | |
1da177e4 LT |
671 | |
672 | /* | |
673 | * We need to sample the current value to convert the new | |
674 | * value from to relative and absolute, and to convert the | |
675 | * old value from absolute to relative. To set a process | |
676 | * timer, we need a sample to balance the thread expiry | |
677 | * times (in arm_timer). With an absolute time, we must | |
678 | * check if it's already passed. In short, we need a sample. | |
679 | */ | |
8c2d74f0 TG |
680 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
681 | val = cpu_clock_sample(clkid, p); | |
682 | else | |
683 | val = cpu_clock_sample_group(clkid, p, true); | |
1da177e4 LT |
684 | |
685 | if (old) { | |
55ccb616 | 686 | if (old_expires == 0) { |
1da177e4 LT |
687 | old->it_value.tv_sec = 0; |
688 | old->it_value.tv_nsec = 0; | |
689 | } else { | |
690 | /* | |
60bda037 TG |
691 | * Update the timer in case it has overrun already. |
692 | * If it has, we'll report it as having overrun and | |
693 | * with the next reloaded timer already ticking, | |
694 | * though we are swallowing that pending | |
695 | * notification here to install the new setting. | |
1da177e4 | 696 | */ |
60bda037 TG |
697 | u64 exp = bump_cpu_timer(timer, val); |
698 | ||
699 | if (val < exp) { | |
700 | old_expires = exp - val; | |
5f252b32 | 701 | old->it_value = ns_to_timespec64(old_expires); |
1da177e4 LT |
702 | } else { |
703 | old->it_value.tv_nsec = 1; | |
704 | old->it_value.tv_sec = 0; | |
705 | } | |
706 | } | |
707 | } | |
708 | ||
a69ac4a7 | 709 | if (unlikely(ret)) { |
1da177e4 LT |
710 | /* |
711 | * We are colliding with the timer actually firing. | |
712 | * Punt after filling in the timer's old value, and | |
713 | * disable this firing since we are already reporting | |
714 | * it as an overrun (thanks to bump_cpu_timer above). | |
715 | */ | |
e73d84e3 | 716 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
717 | goto out; |
718 | } | |
719 | ||
e73d84e3 | 720 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
55ccb616 | 721 | new_expires += val; |
1da177e4 LT |
722 | } |
723 | ||
724 | /* | |
725 | * Install the new expiry time (or zero). | |
726 | * For a timer with no notification action, we don't actually | |
727 | * arm the timer (we'll just fake it for timer_gettime). | |
728 | */ | |
60bda037 | 729 | cpu_timer_setexpires(ctmr, new_expires); |
55ccb616 | 730 | if (new_expires != 0 && val < new_expires) { |
beb41d9c | 731 | arm_timer(timer, p); |
1da177e4 LT |
732 | } |
733 | ||
e73d84e3 | 734 | unlock_task_sighand(p, &flags); |
1da177e4 LT |
735 | /* |
736 | * Install the new reload setting, and | |
737 | * set up the signal and overrun bookkeeping. | |
738 | */ | |
16118794 | 739 | timer->it_interval = timespec64_to_ktime(new->it_interval); |
1da177e4 LT |
740 | |
741 | /* | |
742 | * This acts as a modification timestamp for the timer, | |
743 | * so any automatic reload attempt will punt on seeing | |
744 | * that we have reset the timer manually. | |
745 | */ | |
746 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & | |
747 | ~REQUEUE_PENDING; | |
748 | timer->it_overrun_last = 0; | |
749 | timer->it_overrun = -1; | |
750 | ||
ee375328 FW |
751 | if (val >= new_expires) { |
752 | if (new_expires != 0) { | |
753 | /* | |
754 | * The designated time already passed, so we notify | |
755 | * immediately, even if the thread never runs to | |
756 | * accumulate more time on this clock. | |
757 | */ | |
758 | cpu_timer_fire(timer); | |
759 | } | |
760 | ||
1da177e4 | 761 | /* |
ee375328 FW |
762 | * Make sure we don't keep around the process wide cputime |
763 | * counter or the tick dependency if they are not necessary. | |
1da177e4 | 764 | */ |
ee375328 FW |
765 | sighand = lock_task_sighand(p, &flags); |
766 | if (!sighand) | |
767 | goto out; | |
768 | ||
769 | if (!cpu_timer_queued(ctmr)) | |
770 | trigger_base_recalc_expires(timer, p); | |
771 | ||
772 | unlock_task_sighand(p, &flags); | |
1da177e4 | 773 | } |
1da177e4 | 774 | out: |
55e8c8eb | 775 | rcu_read_unlock(); |
ebd7e7fc | 776 | if (old) |
5f252b32 | 777 | old->it_interval = ns_to_timespec64(old_incr); |
b7878300 | 778 | |
1da177e4 LT |
779 | return ret; |
780 | } | |
781 | ||
5f252b32 | 782 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
1da177e4 | 783 | { |
99093c5b | 784 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
60bda037 TG |
785 | struct cpu_timer *ctmr = &timer->it.cpu; |
786 | u64 now, expires = cpu_timer_getexpires(ctmr); | |
55e8c8eb | 787 | struct task_struct *p; |
1da177e4 | 788 | |
55e8c8eb EB |
789 | rcu_read_lock(); |
790 | p = cpu_timer_task_rcu(timer); | |
791 | if (!p) | |
792 | goto out; | |
a3222f88 | 793 | |
1da177e4 LT |
794 | /* |
795 | * Easy part: convert the reload time. | |
796 | */ | |
16118794 | 797 | itp->it_interval = ktime_to_timespec64(timer->it_interval); |
1da177e4 | 798 | |
60bda037 | 799 | if (!expires) |
55e8c8eb | 800 | goto out; |
1da177e4 | 801 | |
1da177e4 LT |
802 | /* |
803 | * Sample the clock to take the difference with the expiry time. | |
804 | */ | |
60f2ceaa | 805 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
8c2d74f0 | 806 | now = cpu_clock_sample(clkid, p); |
60f2ceaa EB |
807 | else |
808 | now = cpu_clock_sample_group(clkid, p, false); | |
1da177e4 | 809 | |
60bda037 TG |
810 | if (now < expires) { |
811 | itp->it_value = ns_to_timespec64(expires - now); | |
1da177e4 LT |
812 | } else { |
813 | /* | |
814 | * The timer should have expired already, but the firing | |
815 | * hasn't taken place yet. Say it's just about to expire. | |
816 | */ | |
817 | itp->it_value.tv_nsec = 1; | |
818 | itp->it_value.tv_sec = 0; | |
819 | } | |
55e8c8eb EB |
820 | out: |
821 | rcu_read_unlock(); | |
1da177e4 LT |
822 | } |
823 | ||
60bda037 | 824 | #define MAX_COLLECTED 20 |
2473f3e7 | 825 | |
60bda037 TG |
826 | static u64 collect_timerqueue(struct timerqueue_head *head, |
827 | struct list_head *firing, u64 now) | |
828 | { | |
829 | struct timerqueue_node *next; | |
830 | int i = 0; | |
831 | ||
832 | while ((next = timerqueue_getnext(head))) { | |
833 | struct cpu_timer *ctmr; | |
834 | u64 expires; | |
835 | ||
836 | ctmr = container_of(next, struct cpu_timer, node); | |
837 | expires = cpu_timer_getexpires(ctmr); | |
838 | /* Limit the number of timers to expire at once */ | |
839 | if (++i == MAX_COLLECTED || now < expires) | |
840 | return expires; | |
841 | ||
842 | ctmr->firing = 1; | |
843 | cpu_timer_dequeue(ctmr); | |
844 | list_add_tail(&ctmr->elist, firing); | |
2473f3e7 FW |
845 | } |
846 | ||
2bbdbdae | 847 | return U64_MAX; |
2473f3e7 FW |
848 | } |
849 | ||
60bda037 TG |
850 | static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples, |
851 | struct list_head *firing) | |
1cd07c0b TG |
852 | { |
853 | struct posix_cputimer_base *base = pct->bases; | |
854 | int i; | |
855 | ||
856 | for (i = 0; i < CPUCLOCK_MAX; i++, base++) { | |
60bda037 TG |
857 | base->nextevt = collect_timerqueue(&base->tqhead, firing, |
858 | samples[i]); | |
1cd07c0b TG |
859 | } |
860 | } | |
861 | ||
34be3930 JL |
862 | static inline void check_dl_overrun(struct task_struct *tsk) |
863 | { | |
864 | if (tsk->dl.dl_overrun) { | |
865 | tsk->dl.dl_overrun = 0; | |
866 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); | |
867 | } | |
868 | } | |
869 | ||
8991afe2 TG |
870 | static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard) |
871 | { | |
872 | if (time < limit) | |
873 | return false; | |
874 | ||
875 | if (print_fatal_signals) { | |
876 | pr_info("%s Watchdog Timeout (%s): %s[%d]\n", | |
877 | rt ? "RT" : "CPU", hard ? "hard" : "soft", | |
878 | current->comm, task_pid_nr(current)); | |
879 | } | |
880 | __group_send_sig_info(signo, SEND_SIG_PRIV, current); | |
881 | return true; | |
882 | } | |
883 | ||
1da177e4 LT |
884 | /* |
885 | * Check for any per-thread CPU timers that have fired and move them off | |
886 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the | |
887 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. | |
888 | */ | |
889 | static void check_thread_timers(struct task_struct *tsk, | |
890 | struct list_head *firing) | |
891 | { | |
1cd07c0b TG |
892 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
893 | u64 samples[CPUCLOCK_MAX]; | |
d4bb5274 | 894 | unsigned long soft; |
1da177e4 | 895 | |
34be3930 JL |
896 | if (dl_task(tsk)) |
897 | check_dl_overrun(tsk); | |
898 | ||
1cd07c0b | 899 | if (expiry_cache_is_inactive(pct)) |
934715a1 JL |
900 | return; |
901 | ||
1cd07c0b TG |
902 | task_sample_cputime(tsk, samples); |
903 | collect_posix_cputimers(pct, samples, firing); | |
78f2c7db PZ |
904 | |
905 | /* | |
906 | * Check for the special case thread timers. | |
907 | */ | |
3cf29496 | 908 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
d4bb5274 | 909 | if (soft != RLIM_INFINITY) { |
8ea1de90 | 910 | /* Task RT timeout is accounted in jiffies. RTTIME is usec */ |
8991afe2 | 911 | unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ); |
3cf29496 | 912 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
78f2c7db | 913 | |
8991afe2 TG |
914 | /* At the hard limit, send SIGKILL. No further action. */ |
915 | if (hard != RLIM_INFINITY && | |
916 | check_rlimit(rttime, hard, SIGKILL, true, true)) | |
78f2c7db | 917 | return; |
dd670224 | 918 | |
8991afe2 TG |
919 | /* At the soft limit, send a SIGXCPU every second */ |
920 | if (check_rlimit(rttime, soft, SIGXCPU, true, false)) { | |
dd670224 TG |
921 | soft += USEC_PER_SEC; |
922 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft; | |
78f2c7db PZ |
923 | } |
924 | } | |
c02b078e | 925 | |
1cd07c0b | 926 | if (expiry_cache_is_inactive(pct)) |
b7878300 | 927 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); |
1da177e4 LT |
928 | } |
929 | ||
1018016c | 930 | static inline void stop_process_timers(struct signal_struct *sig) |
3fccfd67 | 931 | { |
244d49e3 | 932 | struct posix_cputimers *pct = &sig->posix_cputimers; |
3fccfd67 | 933 | |
244d49e3 TG |
934 | /* Turn off the active flag. This is done without locking. */ |
935 | WRITE_ONCE(pct->timers_active, false); | |
b7878300 | 936 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
3fccfd67 PZ |
937 | } |
938 | ||
42c4ab41 | 939 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
ebd7e7fc | 940 | u64 *expires, u64 cur_time, int signo) |
42c4ab41 | 941 | { |
64861634 | 942 | if (!it->expires) |
42c4ab41 SG |
943 | return; |
944 | ||
858cf3a8 FW |
945 | if (cur_time >= it->expires) { |
946 | if (it->incr) | |
64861634 | 947 | it->expires += it->incr; |
858cf3a8 | 948 | else |
64861634 | 949 | it->expires = 0; |
42c4ab41 | 950 | |
3f0a525e XG |
951 | trace_itimer_expire(signo == SIGPROF ? |
952 | ITIMER_PROF : ITIMER_VIRTUAL, | |
6883f81a | 953 | task_tgid(tsk), cur_time); |
42c4ab41 SG |
954 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
955 | } | |
956 | ||
2bbdbdae | 957 | if (it->expires && it->expires < *expires) |
858cf3a8 | 958 | *expires = it->expires; |
42c4ab41 SG |
959 | } |
960 | ||
1da177e4 LT |
961 | /* |
962 | * Check for any per-thread CPU timers that have fired and move them | |
963 | * off the tsk->*_timers list onto the firing list. Per-thread timers | |
964 | * have already been taken off. | |
965 | */ | |
966 | static void check_process_timers(struct task_struct *tsk, | |
967 | struct list_head *firing) | |
968 | { | |
969 | struct signal_struct *const sig = tsk->signal; | |
1cd07c0b TG |
970 | struct posix_cputimers *pct = &sig->posix_cputimers; |
971 | u64 samples[CPUCLOCK_MAX]; | |
d4bb5274 | 972 | unsigned long soft; |
1da177e4 | 973 | |
934715a1 | 974 | /* |
244d49e3 | 975 | * If there are no active process wide timers (POSIX 1.b, itimers, |
a2ed4fd6 TG |
976 | * RLIMIT_CPU) nothing to check. Also skip the process wide timer |
977 | * processing when there is already another task handling them. | |
934715a1 | 978 | */ |
a2ed4fd6 | 979 | if (!READ_ONCE(pct->timers_active) || pct->expiry_active) |
934715a1 JL |
980 | return; |
981 | ||
a2ed4fd6 | 982 | /* |
c8d75aa4 JL |
983 | * Signify that a thread is checking for process timers. |
984 | * Write access to this field is protected by the sighand lock. | |
985 | */ | |
a2ed4fd6 | 986 | pct->expiry_active = true; |
c8d75aa4 | 987 | |
1da177e4 | 988 | /* |
a324956f TG |
989 | * Collect the current process totals. Group accounting is active |
990 | * so the sample can be taken directly. | |
1da177e4 | 991 | */ |
b7be4ef1 | 992 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples); |
1cd07c0b | 993 | collect_posix_cputimers(pct, samples, firing); |
1da177e4 LT |
994 | |
995 | /* | |
996 | * Check for the special case process timers. | |
997 | */ | |
1cd07c0b TG |
998 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], |
999 | &pct->bases[CPUCLOCK_PROF].nextevt, | |
b7be4ef1 | 1000 | samples[CPUCLOCK_PROF], SIGPROF); |
1cd07c0b TG |
1001 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], |
1002 | &pct->bases[CPUCLOCK_VIRT].nextevt, | |
1003 | samples[CPUCLOCK_VIRT], SIGVTALRM); | |
b7be4ef1 | 1004 | |
3cf29496 | 1005 | soft = task_rlimit(tsk, RLIMIT_CPU); |
d4bb5274 | 1006 | if (soft != RLIM_INFINITY) { |
8ea1de90 | 1007 | /* RLIMIT_CPU is in seconds. Samples are nanoseconds */ |
3cf29496 | 1008 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
8ea1de90 TG |
1009 | u64 ptime = samples[CPUCLOCK_PROF]; |
1010 | u64 softns = (u64)soft * NSEC_PER_SEC; | |
1011 | u64 hardns = (u64)hard * NSEC_PER_SEC; | |
b7be4ef1 | 1012 | |
8991afe2 TG |
1013 | /* At the hard limit, send SIGKILL. No further action. */ |
1014 | if (hard != RLIM_INFINITY && | |
1015 | check_rlimit(ptime, hardns, SIGKILL, false, true)) | |
1da177e4 | 1016 | return; |
dd670224 | 1017 | |
8991afe2 TG |
1018 | /* At the soft limit, send a SIGXCPU every second */ |
1019 | if (check_rlimit(ptime, softns, SIGXCPU, false, false)) { | |
dd670224 TG |
1020 | sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1; |
1021 | softns += NSEC_PER_SEC; | |
1da177e4 | 1022 | } |
8ea1de90 TG |
1023 | |
1024 | /* Update the expiry cache */ | |
1cd07c0b TG |
1025 | if (softns < pct->bases[CPUCLOCK_PROF].nextevt) |
1026 | pct->bases[CPUCLOCK_PROF].nextevt = softns; | |
1da177e4 LT |
1027 | } |
1028 | ||
1cd07c0b | 1029 | if (expiry_cache_is_inactive(pct)) |
29f87b79 | 1030 | stop_process_timers(sig); |
c8d75aa4 | 1031 | |
244d49e3 | 1032 | pct->expiry_active = false; |
1da177e4 LT |
1033 | } |
1034 | ||
1035 | /* | |
96fe3b07 | 1036 | * This is called from the signal code (via posixtimer_rearm) |
1da177e4 LT |
1037 | * when the last timer signal was delivered and we have to reload the timer. |
1038 | */ | |
f37fb0aa | 1039 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
1da177e4 | 1040 | { |
da020ce4 | 1041 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
55e8c8eb | 1042 | struct task_struct *p; |
e73d84e3 FW |
1043 | struct sighand_struct *sighand; |
1044 | unsigned long flags; | |
ebd7e7fc | 1045 | u64 now; |
1da177e4 | 1046 | |
55e8c8eb EB |
1047 | rcu_read_lock(); |
1048 | p = cpu_timer_task_rcu(timer); | |
1049 | if (!p) | |
1050 | goto out; | |
1da177e4 | 1051 | |
1a3402d9 FW |
1052 | /* Protect timer list r/w in arm_timer() */ |
1053 | sighand = lock_task_sighand(p, &flags); | |
1054 | if (unlikely(sighand == NULL)) | |
1055 | goto out; | |
1056 | ||
1da177e4 LT |
1057 | /* |
1058 | * Fetch the current sample and update the timer's expiry time. | |
1059 | */ | |
60f2ceaa | 1060 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
8c2d74f0 | 1061 | now = cpu_clock_sample(clkid, p); |
60f2ceaa | 1062 | else |
8c2d74f0 | 1063 | now = cpu_clock_sample_group(clkid, p, true); |
60f2ceaa EB |
1064 | |
1065 | bump_cpu_timer(timer, now); | |
1066 | ||
1da177e4 LT |
1067 | /* |
1068 | * Now re-arm for the new expiry time. | |
1069 | */ | |
beb41d9c | 1070 | arm_timer(timer, p); |
e73d84e3 | 1071 | unlock_task_sighand(p, &flags); |
55e8c8eb EB |
1072 | out: |
1073 | rcu_read_unlock(); | |
1da177e4 LT |
1074 | } |
1075 | ||
f06febc9 | 1076 | /** |
87dc6448 | 1077 | * task_cputimers_expired - Check whether posix CPU timers are expired |
f06febc9 | 1078 | * |
001f7971 | 1079 | * @samples: Array of current samples for the CPUCLOCK clocks |
87dc6448 | 1080 | * @pct: Pointer to a posix_cputimers container |
f06febc9 | 1081 | * |
87dc6448 TG |
1082 | * Returns true if any member of @samples is greater than the corresponding |
1083 | * member of @pct->bases[CLK].nextevt. False otherwise | |
f06febc9 | 1084 | */ |
87dc6448 | 1085 | static inline bool |
7f2cbcbc | 1086 | task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct) |
f06febc9 | 1087 | { |
001f7971 TG |
1088 | int i; |
1089 | ||
1090 | for (i = 0; i < CPUCLOCK_MAX; i++) { | |
7f2cbcbc | 1091 | if (samples[i] >= pct->bases[i].nextevt) |
001f7971 TG |
1092 | return true; |
1093 | } | |
1094 | return false; | |
f06febc9 FM |
1095 | } |
1096 | ||
1097 | /** | |
1098 | * fastpath_timer_check - POSIX CPU timers fast path. | |
1099 | * | |
1100 | * @tsk: The task (thread) being checked. | |
f06febc9 | 1101 | * |
bb34d92f FM |
1102 | * Check the task and thread group timers. If both are zero (there are no |
1103 | * timers set) return false. Otherwise snapshot the task and thread group | |
1104 | * timers and compare them with the corresponding expiration times. Return | |
1105 | * true if a timer has expired, else return false. | |
f06febc9 | 1106 | */ |
001f7971 | 1107 | static inline bool fastpath_timer_check(struct task_struct *tsk) |
f06febc9 | 1108 | { |
244d49e3 | 1109 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
ad133ba3 | 1110 | struct signal_struct *sig; |
bb34d92f | 1111 | |
244d49e3 | 1112 | if (!expiry_cache_is_inactive(pct)) { |
001f7971 | 1113 | u64 samples[CPUCLOCK_MAX]; |
bb34d92f | 1114 | |
001f7971 | 1115 | task_sample_cputime(tsk, samples); |
244d49e3 | 1116 | if (task_cputimers_expired(samples, pct)) |
001f7971 | 1117 | return true; |
bb34d92f | 1118 | } |
ad133ba3 ON |
1119 | |
1120 | sig = tsk->signal; | |
244d49e3 | 1121 | pct = &sig->posix_cputimers; |
c8d75aa4 | 1122 | /* |
244d49e3 TG |
1123 | * Check if thread group timers expired when timers are active and |
1124 | * no other thread in the group is already handling expiry for | |
1125 | * thread group cputimers. These fields are read without the | |
1126 | * sighand lock. However, this is fine because this is meant to be | |
1127 | * a fastpath heuristic to determine whether we should try to | |
1128 | * acquire the sighand lock to handle timer expiry. | |
c8d75aa4 | 1129 | * |
244d49e3 TG |
1130 | * In the worst case scenario, if concurrently timers_active is set |
1131 | * or expiry_active is cleared, but the current thread doesn't see | |
1132 | * the change yet, the timer checks are delayed until the next | |
1133 | * thread in the group gets a scheduler interrupt to handle the | |
1134 | * timer. This isn't an issue in practice because these types of | |
1135 | * delays with signals actually getting sent are expected. | |
c8d75aa4 | 1136 | */ |
244d49e3 | 1137 | if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) { |
001f7971 | 1138 | u64 samples[CPUCLOCK_MAX]; |
bb34d92f | 1139 | |
001f7971 TG |
1140 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, |
1141 | samples); | |
8d1f431c | 1142 | |
244d49e3 | 1143 | if (task_cputimers_expired(samples, pct)) |
001f7971 | 1144 | return true; |
bb34d92f | 1145 | } |
37bebc70 | 1146 | |
34be3930 | 1147 | if (dl_task(tsk) && tsk->dl.dl_overrun) |
001f7971 | 1148 | return true; |
34be3930 | 1149 | |
001f7971 | 1150 | return false; |
f06febc9 FM |
1151 | } |
1152 | ||
1fb497dd TG |
1153 | static void handle_posix_cpu_timers(struct task_struct *tsk); |
1154 | ||
1155 | #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK | |
1156 | static void posix_cpu_timers_work(struct callback_head *work) | |
1157 | { | |
1158 | handle_posix_cpu_timers(current); | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * Initialize posix CPU timers task work in init task. Out of line to | |
1163 | * keep the callback static and to avoid header recursion hell. | |
1164 | */ | |
1165 | void __init posix_cputimers_init_work(void) | |
1166 | { | |
1167 | init_task_work(¤t->posix_cputimers_work.work, | |
1168 | posix_cpu_timers_work); | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * Note: All operations on tsk->posix_cputimer_work.scheduled happen either | |
1173 | * in hard interrupt context or in task context with interrupts | |
1174 | * disabled. Aside of that the writer/reader interaction is always in the | |
1175 | * context of the current task, which means they are strict per CPU. | |
1176 | */ | |
1177 | static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk) | |
1178 | { | |
1179 | return tsk->posix_cputimers_work.scheduled; | |
1180 | } | |
1181 | ||
1182 | static inline void __run_posix_cpu_timers(struct task_struct *tsk) | |
1183 | { | |
1184 | if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled)) | |
1185 | return; | |
1186 | ||
1187 | /* Schedule task work to actually expire the timers */ | |
1188 | tsk->posix_cputimers_work.scheduled = true; | |
1189 | task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME); | |
1190 | } | |
1191 | ||
1192 | static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk, | |
1193 | unsigned long start) | |
1194 | { | |
1195 | bool ret = true; | |
1196 | ||
1197 | /* | |
1198 | * On !RT kernels interrupts are disabled while collecting expired | |
1199 | * timers, so no tick can happen and the fast path check can be | |
1200 | * reenabled without further checks. | |
1201 | */ | |
1202 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
1203 | tsk->posix_cputimers_work.scheduled = false; | |
1204 | return true; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * On RT enabled kernels ticks can happen while the expired timers | |
1209 | * are collected under sighand lock. But any tick which observes | |
1210 | * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath | |
1211 | * checks. So reenabling the tick work has do be done carefully: | |
1212 | * | |
1213 | * Disable interrupts and run the fast path check if jiffies have | |
1214 | * advanced since the collecting of expired timers started. If | |
1215 | * jiffies have not advanced or the fast path check did not find | |
1216 | * newly expired timers, reenable the fast path check in the timer | |
1217 | * interrupt. If there are newly expired timers, return false and | |
1218 | * let the collection loop repeat. | |
1219 | */ | |
1220 | local_irq_disable(); | |
1221 | if (start != jiffies && fastpath_timer_check(tsk)) | |
1222 | ret = false; | |
1223 | else | |
1224 | tsk->posix_cputimers_work.scheduled = false; | |
1225 | local_irq_enable(); | |
1226 | ||
1227 | return ret; | |
1228 | } | |
1229 | #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */ | |
1230 | static inline void __run_posix_cpu_timers(struct task_struct *tsk) | |
1231 | { | |
1232 | lockdep_posixtimer_enter(); | |
1233 | handle_posix_cpu_timers(tsk); | |
1234 | lockdep_posixtimer_exit(); | |
1235 | } | |
1236 | ||
1237 | static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk) | |
1238 | { | |
1239 | return false; | |
1240 | } | |
1241 | ||
1242 | static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk, | |
1243 | unsigned long start) | |
1244 | { | |
1245 | return true; | |
1246 | } | |
1247 | #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */ | |
1248 | ||
1249 | static void handle_posix_cpu_timers(struct task_struct *tsk) | |
1da177e4 | 1250 | { |
1da177e4 | 1251 | struct k_itimer *timer, *next; |
1fb497dd | 1252 | unsigned long flags, start; |
dce3e8fd | 1253 | LIST_HEAD(firing); |
1da177e4 | 1254 | |
820903c7 | 1255 | if (!lock_task_sighand(tsk, &flags)) |
f06febc9 | 1256 | return; |
5ce73a4a | 1257 | |
1fb497dd TG |
1258 | do { |
1259 | /* | |
1260 | * On RT locking sighand lock does not disable interrupts, | |
1261 | * so this needs to be careful vs. ticks. Store the current | |
1262 | * jiffies value. | |
1263 | */ | |
1264 | start = READ_ONCE(jiffies); | |
1265 | barrier(); | |
934715a1 | 1266 | |
1fb497dd TG |
1267 | /* |
1268 | * Here we take off tsk->signal->cpu_timers[N] and | |
1269 | * tsk->cpu_timers[N] all the timers that are firing, and | |
1270 | * put them on the firing list. | |
1271 | */ | |
1272 | check_thread_timers(tsk, &firing); | |
1273 | ||
1274 | check_process_timers(tsk, &firing); | |
1275 | ||
1276 | /* | |
4bf07f65 | 1277 | * The above timer checks have updated the expiry cache and |
1fb497dd TG |
1278 | * because nothing can have queued or modified timers after |
1279 | * sighand lock was taken above it is guaranteed to be | |
1280 | * consistent. So the next timer interrupt fastpath check | |
1281 | * will find valid data. | |
1282 | * | |
1283 | * If timer expiry runs in the timer interrupt context then | |
1284 | * the loop is not relevant as timers will be directly | |
1285 | * expired in interrupt context. The stub function below | |
1286 | * returns always true which allows the compiler to | |
1287 | * optimize the loop out. | |
1288 | * | |
1289 | * If timer expiry is deferred to task work context then | |
1290 | * the following rules apply: | |
1291 | * | |
1292 | * - On !RT kernels no tick can have happened on this CPU | |
1293 | * after sighand lock was acquired because interrupts are | |
1294 | * disabled. So reenabling task work before dropping | |
1295 | * sighand lock and reenabling interrupts is race free. | |
1296 | * | |
1297 | * - On RT kernels ticks might have happened but the tick | |
1298 | * work ignored posix CPU timer handling because the | |
1299 | * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work | |
1300 | * must be done very carefully including a check whether | |
1301 | * ticks have happened since the start of the timer | |
1302 | * expiry checks. posix_cpu_timers_enable_work() takes | |
1303 | * care of that and eventually lets the expiry checks | |
1304 | * run again. | |
1305 | */ | |
1306 | } while (!posix_cpu_timers_enable_work(tsk, start)); | |
1da177e4 | 1307 | |
bb34d92f | 1308 | /* |
1fb497dd | 1309 | * We must release sighand lock before taking any timer's lock. |
bb34d92f FM |
1310 | * There is a potential race with timer deletion here, as the |
1311 | * siglock now protects our private firing list. We have set | |
1312 | * the firing flag in each timer, so that a deletion attempt | |
1313 | * that gets the timer lock before we do will give it up and | |
1314 | * spin until we've taken care of that timer below. | |
1315 | */ | |
0bdd2ed4 | 1316 | unlock_task_sighand(tsk, &flags); |
1da177e4 LT |
1317 | |
1318 | /* | |
1319 | * Now that all the timers on our list have the firing flag, | |
25985edc | 1320 | * no one will touch their list entries but us. We'll take |
1da177e4 LT |
1321 | * each timer's lock before clearing its firing flag, so no |
1322 | * timer call will interfere. | |
1323 | */ | |
60bda037 | 1324 | list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) { |
6e85c5ba HS |
1325 | int cpu_firing; |
1326 | ||
1fb497dd TG |
1327 | /* |
1328 | * spin_lock() is sufficient here even independent of the | |
1329 | * expiry context. If expiry happens in hard interrupt | |
1330 | * context it's obvious. For task work context it's safe | |
1331 | * because all other operations on timer::it_lock happen in | |
1332 | * task context (syscall or exit). | |
1333 | */ | |
1da177e4 | 1334 | spin_lock(&timer->it_lock); |
60bda037 | 1335 | list_del_init(&timer->it.cpu.elist); |
6e85c5ba | 1336 | cpu_firing = timer->it.cpu.firing; |
1da177e4 LT |
1337 | timer->it.cpu.firing = 0; |
1338 | /* | |
1339 | * The firing flag is -1 if we collided with a reset | |
1340 | * of the timer, which already reported this | |
1341 | * almost-firing as an overrun. So don't generate an event. | |
1342 | */ | |
6e85c5ba | 1343 | if (likely(cpu_firing >= 0)) |
1da177e4 | 1344 | cpu_timer_fire(timer); |
1da177e4 LT |
1345 | spin_unlock(&timer->it_lock); |
1346 | } | |
820903c7 TG |
1347 | } |
1348 | ||
1349 | /* | |
1350 | * This is called from the timer interrupt handler. The irq handler has | |
1351 | * already updated our counts. We need to check if any timers fire now. | |
1352 | * Interrupts are disabled. | |
1353 | */ | |
1354 | void run_posix_cpu_timers(void) | |
1355 | { | |
1356 | struct task_struct *tsk = current; | |
1357 | ||
1358 | lockdep_assert_irqs_disabled(); | |
1359 | ||
1fb497dd TG |
1360 | /* |
1361 | * If the actual expiry is deferred to task work context and the | |
1362 | * work is already scheduled there is no point to do anything here. | |
1363 | */ | |
1364 | if (posix_cpu_timers_work_scheduled(tsk)) | |
1365 | return; | |
1366 | ||
820903c7 TG |
1367 | /* |
1368 | * The fast path checks that there are no expired thread or thread | |
1369 | * group timers. If that's so, just return. | |
1370 | */ | |
1371 | if (!fastpath_timer_check(tsk)) | |
1372 | return; | |
1373 | ||
820903c7 | 1374 | __run_posix_cpu_timers(tsk); |
1da177e4 LT |
1375 | } |
1376 | ||
1377 | /* | |
f55db609 | 1378 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
f06febc9 | 1379 | * The tsk->sighand->siglock must be held by the caller. |
1da177e4 | 1380 | */ |
1b0dd96d | 1381 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid, |
858cf3a8 | 1382 | u64 *newval, u64 *oldval) |
1da177e4 | 1383 | { |
87dc6448 | 1384 | u64 now, *nextevt; |
1da177e4 | 1385 | |
1b0dd96d | 1386 | if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED)) |
692117c1 TG |
1387 | return; |
1388 | ||
87dc6448 | 1389 | nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt; |
1b0dd96d | 1390 | now = cpu_clock_sample_group(clkid, tsk, true); |
1da177e4 | 1391 | |
5405d005 | 1392 | if (oldval) { |
f55db609 SG |
1393 | /* |
1394 | * We are setting itimer. The *oldval is absolute and we update | |
1395 | * it to be relative, *newval argument is relative and we update | |
1396 | * it to be absolute. | |
1397 | */ | |
64861634 | 1398 | if (*oldval) { |
858cf3a8 | 1399 | if (*oldval <= now) { |
1da177e4 | 1400 | /* Just about to fire. */ |
858cf3a8 | 1401 | *oldval = TICK_NSEC; |
1da177e4 | 1402 | } else { |
858cf3a8 | 1403 | *oldval -= now; |
1da177e4 LT |
1404 | } |
1405 | } | |
1406 | ||
8cd9da85 FW |
1407 | if (*newval) |
1408 | *newval += now; | |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | /* | |
1b0dd96d TG |
1412 | * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF |
1413 | * expiry cache is also used by RLIMIT_CPU!. | |
1da177e4 | 1414 | */ |
2bbdbdae | 1415 | if (*newval < *nextevt) |
87dc6448 | 1416 | *nextevt = *newval; |
b7878300 | 1417 | |
1e4ca26d | 1418 | tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER); |
1da177e4 LT |
1419 | } |
1420 | ||
e4b76555 | 1421 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
343d8fc2 | 1422 | const struct timespec64 *rqtp) |
1da177e4 | 1423 | { |
86a9c446 | 1424 | struct itimerspec64 it; |
343d8fc2 TG |
1425 | struct k_itimer timer; |
1426 | u64 expires; | |
1da177e4 LT |
1427 | int error; |
1428 | ||
1da177e4 LT |
1429 | /* |
1430 | * Set up a temporary timer and then wait for it to go off. | |
1431 | */ | |
1432 | memset(&timer, 0, sizeof timer); | |
1433 | spin_lock_init(&timer.it_lock); | |
1434 | timer.it_clock = which_clock; | |
1435 | timer.it_overrun = -1; | |
1436 | error = posix_cpu_timer_create(&timer); | |
1437 | timer.it_process = current; | |
60bda037 | 1438 | |
1da177e4 | 1439 | if (!error) { |
5f252b32 | 1440 | static struct itimerspec64 zero_it; |
edbeda46 | 1441 | struct restart_block *restart; |
e4b76555 | 1442 | |
edbeda46 | 1443 | memset(&it, 0, sizeof(it)); |
86a9c446 | 1444 | it.it_value = *rqtp; |
1da177e4 LT |
1445 | |
1446 | spin_lock_irq(&timer.it_lock); | |
86a9c446 | 1447 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
1da177e4 LT |
1448 | if (error) { |
1449 | spin_unlock_irq(&timer.it_lock); | |
1450 | return error; | |
1451 | } | |
1452 | ||
1453 | while (!signal_pending(current)) { | |
60bda037 | 1454 | if (!cpu_timer_getexpires(&timer.it.cpu)) { |
1da177e4 | 1455 | /* |
e6c42c29 SG |
1456 | * Our timer fired and was reset, below |
1457 | * deletion can not fail. | |
1da177e4 | 1458 | */ |
e6c42c29 | 1459 | posix_cpu_timer_del(&timer); |
1da177e4 LT |
1460 | spin_unlock_irq(&timer.it_lock); |
1461 | return 0; | |
1462 | } | |
1463 | ||
1464 | /* | |
1465 | * Block until cpu_timer_fire (or a signal) wakes us. | |
1466 | */ | |
1467 | __set_current_state(TASK_INTERRUPTIBLE); | |
1468 | spin_unlock_irq(&timer.it_lock); | |
1469 | schedule(); | |
1470 | spin_lock_irq(&timer.it_lock); | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * We were interrupted by a signal. | |
1475 | */ | |
60bda037 | 1476 | expires = cpu_timer_getexpires(&timer.it.cpu); |
86a9c446 | 1477 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
e6c42c29 SG |
1478 | if (!error) { |
1479 | /* | |
1480 | * Timer is now unarmed, deletion can not fail. | |
1481 | */ | |
1482 | posix_cpu_timer_del(&timer); | |
1483 | } | |
1da177e4 LT |
1484 | spin_unlock_irq(&timer.it_lock); |
1485 | ||
e6c42c29 SG |
1486 | while (error == TIMER_RETRY) { |
1487 | /* | |
1488 | * We need to handle case when timer was or is in the | |
1489 | * middle of firing. In other cases we already freed | |
1490 | * resources. | |
1491 | */ | |
1492 | spin_lock_irq(&timer.it_lock); | |
1493 | error = posix_cpu_timer_del(&timer); | |
1494 | spin_unlock_irq(&timer.it_lock); | |
1495 | } | |
1496 | ||
86a9c446 | 1497 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
1da177e4 LT |
1498 | /* |
1499 | * It actually did fire already. | |
1500 | */ | |
1501 | return 0; | |
1502 | } | |
1503 | ||
e4b76555 | 1504 | error = -ERESTART_RESTARTBLOCK; |
86a9c446 AV |
1505 | /* |
1506 | * Report back to the user the time still remaining. | |
1507 | */ | |
edbeda46 | 1508 | restart = ¤t->restart_block; |
343d8fc2 | 1509 | restart->nanosleep.expires = expires; |
c0edd7c9 DD |
1510 | if (restart->nanosleep.type != TT_NONE) |
1511 | error = nanosleep_copyout(restart, &it.it_value); | |
e4b76555 TA |
1512 | } |
1513 | ||
1514 | return error; | |
1515 | } | |
1516 | ||
bc2c8ea4 TG |
1517 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
1518 | ||
1519 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, | |
938e7cf2 | 1520 | const struct timespec64 *rqtp) |
e4b76555 | 1521 | { |
f56141e3 | 1522 | struct restart_block *restart_block = ¤t->restart_block; |
e4b76555 TA |
1523 | int error; |
1524 | ||
1525 | /* | |
1526 | * Diagnose required errors first. | |
1527 | */ | |
1528 | if (CPUCLOCK_PERTHREAD(which_clock) && | |
1529 | (CPUCLOCK_PID(which_clock) == 0 || | |
01a21974 | 1530 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
e4b76555 TA |
1531 | return -EINVAL; |
1532 | ||
86a9c446 | 1533 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
e4b76555 TA |
1534 | |
1535 | if (error == -ERESTART_RESTARTBLOCK) { | |
1536 | ||
3751f9f2 | 1537 | if (flags & TIMER_ABSTIME) |
e4b76555 | 1538 | return -ERESTARTNOHAND; |
1da177e4 | 1539 | |
ab8177bc | 1540 | restart_block->nanosleep.clockid = which_clock; |
5abbe51a | 1541 | set_restart_fn(restart_block, posix_cpu_nsleep_restart); |
1da177e4 | 1542 | } |
1da177e4 LT |
1543 | return error; |
1544 | } | |
1545 | ||
bc2c8ea4 | 1546 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
1da177e4 | 1547 | { |
ab8177bc | 1548 | clockid_t which_clock = restart_block->nanosleep.clockid; |
ad196384 | 1549 | struct timespec64 t; |
97735f25 | 1550 | |
ad196384 | 1551 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
97735f25 | 1552 | |
86a9c446 | 1553 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
1da177e4 LT |
1554 | } |
1555 | ||
29f1b2b0 ND |
1556 | #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED) |
1557 | #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED) | |
1da177e4 | 1558 | |
a924b04d | 1559 | static int process_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1560 | struct timespec64 *tp) |
1da177e4 LT |
1561 | { |
1562 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); | |
1563 | } | |
a924b04d | 1564 | static int process_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1565 | struct timespec64 *tp) |
1da177e4 LT |
1566 | { |
1567 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); | |
1568 | } | |
1569 | static int process_cpu_timer_create(struct k_itimer *timer) | |
1570 | { | |
1571 | timer->it_clock = PROCESS_CLOCK; | |
1572 | return posix_cpu_timer_create(timer); | |
1573 | } | |
a924b04d | 1574 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
938e7cf2 | 1575 | const struct timespec64 *rqtp) |
1da177e4 | 1576 | { |
99e6c0e6 | 1577 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
1da177e4 | 1578 | } |
a924b04d | 1579 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
d2e3e0ca | 1580 | struct timespec64 *tp) |
1da177e4 LT |
1581 | { |
1582 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); | |
1583 | } | |
a924b04d | 1584 | static int thread_cpu_clock_get(const clockid_t which_clock, |
3c9c12f4 | 1585 | struct timespec64 *tp) |
1da177e4 LT |
1586 | { |
1587 | return posix_cpu_clock_get(THREAD_CLOCK, tp); | |
1588 | } | |
1589 | static int thread_cpu_timer_create(struct k_itimer *timer) | |
1590 | { | |
1591 | timer->it_clock = THREAD_CLOCK; | |
1592 | return posix_cpu_timer_create(timer); | |
1593 | } | |
1da177e4 | 1594 | |
d3ba5a9a | 1595 | const struct k_clock clock_posix_cpu = { |
819a95fe AV |
1596 | .clock_getres = posix_cpu_clock_getres, |
1597 | .clock_set = posix_cpu_clock_set, | |
1598 | .clock_get_timespec = posix_cpu_clock_get, | |
1599 | .timer_create = posix_cpu_timer_create, | |
1600 | .nsleep = posix_cpu_nsleep, | |
1601 | .timer_set = posix_cpu_timer_set, | |
1602 | .timer_del = posix_cpu_timer_del, | |
1603 | .timer_get = posix_cpu_timer_get, | |
1604 | .timer_rearm = posix_cpu_timer_rearm, | |
1976945e TG |
1605 | }; |
1606 | ||
d3ba5a9a | 1607 | const struct k_clock clock_process = { |
819a95fe AV |
1608 | .clock_getres = process_cpu_clock_getres, |
1609 | .clock_get_timespec = process_cpu_clock_get, | |
1610 | .timer_create = process_cpu_timer_create, | |
1611 | .nsleep = process_cpu_nsleep, | |
d3ba5a9a | 1612 | }; |
1da177e4 | 1613 | |
d3ba5a9a | 1614 | const struct k_clock clock_thread = { |
819a95fe AV |
1615 | .clock_getres = thread_cpu_clock_getres, |
1616 | .clock_get_timespec = thread_cpu_clock_get, | |
1617 | .timer_create = thread_cpu_timer_create, | |
d3ba5a9a | 1618 | }; |