1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Implement CPU time clocks for the POSIX clock interface.
4  */
5 
6 #include <linux/sched/signal.h>
7 #include <linux/sched/cputime.h>
8 #include <linux/posix-timers.h>
9 #include <linux/errno.h>
10 #include <linux/math64.h>
11 #include <linux/uaccess.h>
12 #include <linux/kernel_stat.h>
13 #include <trace/events/timer.h>
14 #include <linux/tick.h>
15 #include <linux/workqueue.h>
16 #include <linux/compat.h>
17 #include <linux/sched/deadline.h>
18 #include <linux/task_work.h>
19 
20 #include "posix-timers.h"
21 
22 static void posix_cpu_timer_rearm(struct k_itimer *timer);
23 
24 void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
25 {
26 	posix_cputimers_init(pct);
27 	if (cpu_limit != RLIM_INFINITY) {
28 		pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
29 		pct->timers_active = true;
30 	}
31 }
32 
33 /*
34  * Called after updating RLIMIT_CPU to run cpu timer and update
35  * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
36  * necessary. Needs siglock protection since other code may update the
37  * expiration cache as well.
38  *
39  * Returns 0 on success, -ESRCH on failure.  Can fail if the task is exiting and
40  * we cannot lock_task_sighand.  Cannot fail if task is current.
41  */
42 int update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
43 {
44 	u64 nsecs = rlim_new * NSEC_PER_SEC;
45 	unsigned long irq_fl;
46 
47 	if (!lock_task_sighand(task, &irq_fl))
48 		return -ESRCH;
49 	set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
50 	unlock_task_sighand(task, &irq_fl);
51 	return 0;
52 }
53 
54 /*
55  * Functions for validating access to tasks.
56  */
57 static struct pid *pid_for_clock(const clockid_t clock, bool gettime)
58 {
59 	const bool thread = !!CPUCLOCK_PERTHREAD(clock);
60 	const pid_t upid = CPUCLOCK_PID(clock);
61 	struct pid *pid;
62 
63 	if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
64 		return NULL;
65 
66 	/*
67 	 * If the encoded PID is 0, then the timer is targeted at current
68 	 * or the process to which current belongs.
69 	 */
70 	if (upid == 0)
71 		return thread ? task_pid(current) : task_tgid(current);
72 
73 	pid = find_vpid(upid);
74 	if (!pid)
75 		return NULL;
76 
77 	if (thread) {
78 		struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
79 		return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
80 	}
81 
82 	/*
83 	 * For clock_gettime(PROCESS) allow finding the process by
84 	 * with the pid of the current task.  The code needs the tgid
85 	 * of the process so that pid_task(pid, PIDTYPE_TGID) can be
86 	 * used to find the process.
87 	 */
88 	if (gettime && (pid == task_pid(current)))
89 		return task_tgid(current);
90 
91 	/*
92 	 * For processes require that pid identifies a process.
93 	 */
94 	return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
95 }
96 
97 static inline int validate_clock_permissions(const clockid_t clock)
98 {
99 	int ret;
100 
101 	rcu_read_lock();
102 	ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
103 	rcu_read_unlock();
104 
105 	return ret;
106 }
107 
108 static inline enum pid_type clock_pid_type(const clockid_t clock)
109 {
110 	return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
111 }
112 
113 static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
114 {
115 	return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
116 }
117 
118 /*
119  * Update expiry time from increment, and increase overrun count,
120  * given the current clock sample.
121  */
122 static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
123 {
124 	u64 delta, incr, expires = timer->it.cpu.node.expires;
125 	int i;
126 
127 	if (!timer->it_interval)
128 		return expires;
129 
130 	if (now < expires)
131 		return expires;
132 
133 	incr = timer->it_interval;
134 	delta = now + incr - expires;
135 
136 	/* Don't use (incr*2 < delta), incr*2 might overflow. */
137 	for (i = 0; incr < delta - incr; i++)
138 		incr = incr << 1;
139 
140 	for (; i >= 0; incr >>= 1, i--) {
141 		if (delta < incr)
142 			continue;
143 
144 		timer->it.cpu.node.expires += incr;
145 		timer->it_overrun += 1LL << i;
146 		delta -= incr;
147 	}
148 	return timer->it.cpu.node.expires;
149 }
150 
151 /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
152 static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
153 {
154 	return !(~pct->bases[CPUCLOCK_PROF].nextevt |
155 		 ~pct->bases[CPUCLOCK_VIRT].nextevt |
156 		 ~pct->bases[CPUCLOCK_SCHED].nextevt);
157 }
158 
159 static int
160 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
161 {
162 	int error = validate_clock_permissions(which_clock);
163 
164 	if (!error) {
165 		tp->tv_sec = 0;
166 		tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
167 		if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
168 			/*
169 			 * If sched_clock is using a cycle counter, we
170 			 * don't have any idea of its true resolution
171 			 * exported, but it is much more than 1s/HZ.
172 			 */
173 			tp->tv_nsec = 1;
174 		}
175 	}
176 	return error;
177 }
178 
179 static int
180 posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
181 {
182 	int error = validate_clock_permissions(clock);
183 
184 	/*
185 	 * You can never reset a CPU clock, but we check for other errors
186 	 * in the call before failing with EPERM.
187 	 */
188 	return error ? : -EPERM;
189 }
190 
191 /*
192  * Sample a per-thread clock for the given task. clkid is validated.
193  */
194 static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
195 {
196 	u64 utime, stime;
197 
198 	if (clkid == CPUCLOCK_SCHED)
199 		return task_sched_runtime(p);
200 
201 	task_cputime(p, &utime, &stime);
202 
203 	switch (clkid) {
204 	case CPUCLOCK_PROF:
205 		return utime + stime;
206 	case CPUCLOCK_VIRT:
207 		return utime;
208 	default:
209 		WARN_ON_ONCE(1);
210 	}
211 	return 0;
212 }
213 
214 static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime)
215 {
216 	samples[CPUCLOCK_PROF] = stime + utime;
217 	samples[CPUCLOCK_VIRT] = utime;
218 	samples[CPUCLOCK_SCHED] = rtime;
219 }
220 
221 static void task_sample_cputime(struct task_struct *p, u64 *samples)
222 {
223 	u64 stime, utime;
224 
225 	task_cputime(p, &utime, &stime);
226 	store_samples(samples, stime, utime, p->se.sum_exec_runtime);
227 }
228 
229 static void proc_sample_cputime_atomic(struct task_cputime_atomic *at,
230 				       u64 *samples)
231 {
232 	u64 stime, utime, rtime;
233 
234 	utime = atomic64_read(&at->utime);
235 	stime = atomic64_read(&at->stime);
236 	rtime = atomic64_read(&at->sum_exec_runtime);
237 	store_samples(samples, stime, utime, rtime);
238 }
239 
240 /*
241  * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
242  * to avoid race conditions with concurrent updates to cputime.
243  */
244 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
245 {
246 	u64 curr_cputime = atomic64_read(cputime);
247 
248 	do {
249 		if (sum_cputime <= curr_cputime)
250 			return;
251 	} while (!atomic64_try_cmpxchg(cputime, &curr_cputime, sum_cputime));
252 }
253 
254 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
255 			      struct task_cputime *sum)
256 {
257 	__update_gt_cputime(&cputime_atomic->utime, sum->utime);
258 	__update_gt_cputime(&cputime_atomic->stime, sum->stime);
259 	__update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
260 }
261 
262 /**
263  * thread_group_sample_cputime - Sample cputime for a given task
264  * @tsk:	Task for which cputime needs to be started
265  * @samples:	Storage for time samples
266  *
267  * Called from sys_getitimer() to calculate the expiry time of an active
268  * timer. That means group cputime accounting is already active. Called
269  * with task sighand lock held.
270  *
271  * Updates @times with an uptodate sample of the thread group cputimes.
272  */
273 void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
274 {
275 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
276 	struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
277 
278 	WARN_ON_ONCE(!pct->timers_active);
279 
280 	proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
281 }
282 
283 /**
284  * thread_group_start_cputime - Start cputime and return a sample
285  * @tsk:	Task for which cputime needs to be started
286  * @samples:	Storage for time samples
287  *
288  * The thread group cputime accounting is avoided when there are no posix
289  * CPU timers armed. Before starting a timer it's required to check whether
290  * the time accounting is active. If not, a full update of the atomic
291  * accounting store needs to be done and the accounting enabled.
292  *
293  * Updates @times with an uptodate sample of the thread group cputimes.
294  */
295 static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
296 {
297 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
298 	struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
299 
300 	lockdep_assert_task_sighand_held(tsk);
301 
302 	/* Check if cputimer isn't running. This is accessed without locking. */
303 	if (!READ_ONCE(pct->timers_active)) {
304 		struct task_cputime sum;
305 
306 		/*
307 		 * The POSIX timer interface allows for absolute time expiry
308 		 * values through the TIMER_ABSTIME flag, therefore we have
309 		 * to synchronize the timer to the clock every time we start it.
310 		 */
311 		thread_group_cputime(tsk, &sum);
312 		update_gt_cputime(&cputimer->cputime_atomic, &sum);
313 
314 		/*
315 		 * We're setting timers_active without a lock. Ensure this
316 		 * only gets written to in one operation. We set it after
317 		 * update_gt_cputime() as a small optimization, but
318 		 * barriers are not required because update_gt_cputime()
319 		 * can handle concurrent updates.
320 		 */
321 		WRITE_ONCE(pct->timers_active, true);
322 	}
323 	proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
324 }
325 
326 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
327 {
328 	struct task_cputime ct;
329 
330 	thread_group_cputime(tsk, &ct);
331 	store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
332 }
333 
334 /*
335  * Sample a process (thread group) clock for the given task clkid. If the
336  * group's cputime accounting is already enabled, read the atomic
337  * store. Otherwise a full update is required.  clkid is already validated.
338  */
339 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
340 				  bool start)
341 {
342 	struct thread_group_cputimer *cputimer = &p->signal->cputimer;
343 	struct posix_cputimers *pct = &p->signal->posix_cputimers;
344 	u64 samples[CPUCLOCK_MAX];
345 
346 	if (!READ_ONCE(pct->timers_active)) {
347 		if (start)
348 			thread_group_start_cputime(p, samples);
349 		else
350 			__thread_group_cputime(p, samples);
351 	} else {
352 		proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
353 	}
354 
355 	return samples[clkid];
356 }
357 
358 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
359 {
360 	const clockid_t clkid = CPUCLOCK_WHICH(clock);
361 	struct task_struct *tsk;
362 	u64 t;
363 
364 	rcu_read_lock();
365 	tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
366 	if (!tsk) {
367 		rcu_read_unlock();
368 		return -EINVAL;
369 	}
370 
371 	if (CPUCLOCK_PERTHREAD(clock))
372 		t = cpu_clock_sample(clkid, tsk);
373 	else
374 		t = cpu_clock_sample_group(clkid, tsk, false);
375 	rcu_read_unlock();
376 
377 	*tp = ns_to_timespec64(t);
378 	return 0;
379 }
380 
381 /*
382  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
383  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
384  * new timer already all-zeros initialized.
385  */
386 static int posix_cpu_timer_create(struct k_itimer *new_timer)
387 {
388 	static struct lock_class_key posix_cpu_timers_key;
389 	struct pid *pid;
390 
391 	rcu_read_lock();
392 	pid = pid_for_clock(new_timer->it_clock, false);
393 	if (!pid) {
394 		rcu_read_unlock();
395 		return -EINVAL;
396 	}
397 
398 	/*
399 	 * If posix timer expiry is handled in task work context then
400 	 * timer::it_lock can be taken without disabling interrupts as all
401 	 * other locking happens in task context. This requires a separate
402 	 * lock class key otherwise regular posix timer expiry would record
403 	 * the lock class being taken in interrupt context and generate a
404 	 * false positive warning.
405 	 */
406 	if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
407 		lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
408 
409 	new_timer->kclock = &clock_posix_cpu;
410 	timerqueue_init(&new_timer->it.cpu.node);
411 	new_timer->it.cpu.pid = get_pid(pid);
412 	rcu_read_unlock();
413 	return 0;
414 }
415 
416 static struct posix_cputimer_base *timer_base(struct k_itimer *timer,
417 					      struct task_struct *tsk)
418 {
419 	int clkidx = CPUCLOCK_WHICH(timer->it_clock);
420 
421 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
422 		return tsk->posix_cputimers.bases + clkidx;
423 	else
424 		return tsk->signal->posix_cputimers.bases + clkidx;
425 }
426 
427 /*
428  * Force recalculating the base earliest expiration on the next tick.
429  * This will also re-evaluate the need to keep around the process wide
430  * cputime counter and tick dependency and eventually shut these down
431  * if necessary.
432  */
433 static void trigger_base_recalc_expires(struct k_itimer *timer,
434 					struct task_struct *tsk)
435 {
436 	struct posix_cputimer_base *base = timer_base(timer, tsk);
437 
438 	base->nextevt = 0;
439 }
440 
441 /*
442  * Dequeue the timer and reset the base if it was its earliest expiration.
443  * It makes sure the next tick recalculates the base next expiration so we
444  * don't keep the costly process wide cputime counter around for a random
445  * amount of time, along with the tick dependency.
446  *
447  * If another timer gets queued between this and the next tick, its
448  * expiration will update the base next event if necessary on the next
449  * tick.
450  */
451 static void disarm_timer(struct k_itimer *timer, struct task_struct *p)
452 {
453 	struct cpu_timer *ctmr = &timer->it.cpu;
454 	struct posix_cputimer_base *base;
455 
456 	if (!cpu_timer_dequeue(ctmr))
457 		return;
458 
459 	base = timer_base(timer, p);
460 	if (cpu_timer_getexpires(ctmr) == base->nextevt)
461 		trigger_base_recalc_expires(timer, p);
462 }
463 
464 
465 /*
466  * Clean up a CPU-clock timer that is about to be destroyed.
467  * This is called from timer deletion with the timer already locked.
468  * If we return TIMER_RETRY, it's necessary to release the timer's lock
469  * and try again.  (This happens when the timer is in the middle of firing.)
470  */
471 static int posix_cpu_timer_del(struct k_itimer *timer)
472 {
473 	struct cpu_timer *ctmr = &timer->it.cpu;
474 	struct sighand_struct *sighand;
475 	struct task_struct *p;
476 	unsigned long flags;
477 	int ret = 0;
478 
479 	rcu_read_lock();
480 	p = cpu_timer_task_rcu(timer);
481 	if (!p)
482 		goto out;
483 
484 	/*
485 	 * Protect against sighand release/switch in exit/exec and process/
486 	 * thread timer list entry concurrent read/writes.
487 	 */
488 	sighand = lock_task_sighand(p, &flags);
489 	if (unlikely(sighand == NULL)) {
490 		/*
491 		 * This raced with the reaping of the task. The exit cleanup
492 		 * should have removed this timer from the timer queue.
493 		 */
494 		WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
495 	} else {
496 		if (timer->it.cpu.firing)
497 			ret = TIMER_RETRY;
498 		else
499 			disarm_timer(timer, p);
500 
501 		unlock_task_sighand(p, &flags);
502 	}
503 
504 out:
505 	rcu_read_unlock();
506 	if (!ret)
507 		put_pid(ctmr->pid);
508 
509 	return ret;
510 }
511 
512 static void cleanup_timerqueue(struct timerqueue_head *head)
513 {
514 	struct timerqueue_node *node;
515 	struct cpu_timer *ctmr;
516 
517 	while ((node = timerqueue_getnext(head))) {
518 		timerqueue_del(head, node);
519 		ctmr = container_of(node, struct cpu_timer, node);
520 		ctmr->head = NULL;
521 	}
522 }
523 
524 /*
525  * Clean out CPU timers which are still armed when a thread exits. The
526  * timers are only removed from the list. No other updates are done. The
527  * corresponding posix timers are still accessible, but cannot be rearmed.
528  *
529  * This must be called with the siglock held.
530  */
531 static void cleanup_timers(struct posix_cputimers *pct)
532 {
533 	cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
534 	cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
535 	cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
536 }
537 
538 /*
539  * These are both called with the siglock held, when the current thread
540  * is being reaped.  When the final (leader) thread in the group is reaped,
541  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
542  */
543 void posix_cpu_timers_exit(struct task_struct *tsk)
544 {
545 	cleanup_timers(&tsk->posix_cputimers);
546 }
547 void posix_cpu_timers_exit_group(struct task_struct *tsk)
548 {
549 	cleanup_timers(&tsk->signal->posix_cputimers);
550 }
551 
552 /*
553  * Insert the timer on the appropriate list before any timers that
554  * expire later.  This must be called with the sighand lock held.
555  */
556 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
557 {
558 	struct posix_cputimer_base *base = timer_base(timer, p);
559 	struct cpu_timer *ctmr = &timer->it.cpu;
560 	u64 newexp = cpu_timer_getexpires(ctmr);
561 
562 	if (!cpu_timer_enqueue(&base->tqhead, ctmr))
563 		return;
564 
565 	/*
566 	 * We are the new earliest-expiring POSIX 1.b timer, hence
567 	 * need to update expiration cache. Take into account that
568 	 * for process timers we share expiration cache with itimers
569 	 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
570 	 */
571 	if (newexp < base->nextevt)
572 		base->nextevt = newexp;
573 
574 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
575 		tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
576 	else
577 		tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER);
578 }
579 
580 /*
581  * The timer is locked, fire it and arrange for its reload.
582  */
583 static void cpu_timer_fire(struct k_itimer *timer)
584 {
585 	struct cpu_timer *ctmr = &timer->it.cpu;
586 
587 	if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
588 		/*
589 		 * User don't want any signal.
590 		 */
591 		cpu_timer_setexpires(ctmr, 0);
592 	} else if (unlikely(timer->sigq == NULL)) {
593 		/*
594 		 * This a special case for clock_nanosleep,
595 		 * not a normal timer from sys_timer_create.
596 		 */
597 		wake_up_process(timer->it_process);
598 		cpu_timer_setexpires(ctmr, 0);
599 	} else if (!timer->it_interval) {
600 		/*
601 		 * One-shot timer.  Clear it as soon as it's fired.
602 		 */
603 		posix_timer_event(timer, 0);
604 		cpu_timer_setexpires(ctmr, 0);
605 	} else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
606 		/*
607 		 * The signal did not get queued because the signal
608 		 * was ignored, so we won't get any callback to
609 		 * reload the timer.  But we need to keep it
610 		 * ticking in case the signal is deliverable next time.
611 		 */
612 		posix_cpu_timer_rearm(timer);
613 		++timer->it_requeue_pending;
614 	}
615 }
616 
617 /*
618  * Guts of sys_timer_settime for CPU timers.
619  * This is called with the timer locked and interrupts disabled.
620  * If we return TIMER_RETRY, it's necessary to release the timer's lock
621  * and try again.  (This happens when the timer is in the middle of firing.)
622  */
623 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
624 			       struct itimerspec64 *new, struct itimerspec64 *old)
625 {
626 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
627 	u64 old_expires, new_expires, old_incr, val;
628 	struct cpu_timer *ctmr = &timer->it.cpu;
629 	struct sighand_struct *sighand;
630 	struct task_struct *p;
631 	unsigned long flags;
632 	int ret = 0;
633 
634 	rcu_read_lock();
635 	p = cpu_timer_task_rcu(timer);
636 	if (!p) {
637 		/*
638 		 * If p has just been reaped, we can no
639 		 * longer get any information about it at all.
640 		 */
641 		rcu_read_unlock();
642 		return -ESRCH;
643 	}
644 
645 	/*
646 	 * Use the to_ktime conversion because that clamps the maximum
647 	 * value to KTIME_MAX and avoid multiplication overflows.
648 	 */
649 	new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
650 
651 	/*
652 	 * Protect against sighand release/switch in exit/exec and p->cpu_timers
653 	 * and p->signal->cpu_timers read/write in arm_timer()
654 	 */
655 	sighand = lock_task_sighand(p, &flags);
656 	/*
657 	 * If p has just been reaped, we can no
658 	 * longer get any information about it at all.
659 	 */
660 	if (unlikely(sighand == NULL)) {
661 		rcu_read_unlock();
662 		return -ESRCH;
663 	}
664 
665 	/*
666 	 * Disarm any old timer after extracting its expiry time.
667 	 */
668 	old_incr = timer->it_interval;
669 	old_expires = cpu_timer_getexpires(ctmr);
670 
671 	if (unlikely(timer->it.cpu.firing)) {
672 		timer->it.cpu.firing = -1;
673 		ret = TIMER_RETRY;
674 	} else {
675 		cpu_timer_dequeue(ctmr);
676 	}
677 
678 	/*
679 	 * We need to sample the current value to convert the new
680 	 * value from to relative and absolute, and to convert the
681 	 * old value from absolute to relative.  To set a process
682 	 * timer, we need a sample to balance the thread expiry
683 	 * times (in arm_timer).  With an absolute time, we must
684 	 * check if it's already passed.  In short, we need a sample.
685 	 */
686 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
687 		val = cpu_clock_sample(clkid, p);
688 	else
689 		val = cpu_clock_sample_group(clkid, p, true);
690 
691 	if (old) {
692 		if (old_expires == 0) {
693 			old->it_value.tv_sec = 0;
694 			old->it_value.tv_nsec = 0;
695 		} else {
696 			/*
697 			 * Update the timer in case it has overrun already.
698 			 * If it has, we'll report it as having overrun and
699 			 * with the next reloaded timer already ticking,
700 			 * though we are swallowing that pending
701 			 * notification here to install the new setting.
702 			 */
703 			u64 exp = bump_cpu_timer(timer, val);
704 
705 			if (val < exp) {
706 				old_expires = exp - val;
707 				old->it_value = ns_to_timespec64(old_expires);
708 			} else {
709 				old->it_value.tv_nsec = 1;
710 				old->it_value.tv_sec = 0;
711 			}
712 		}
713 	}
714 
715 	if (unlikely(ret)) {
716 		/*
717 		 * We are colliding with the timer actually firing.
718 		 * Punt after filling in the timer's old value, and
719 		 * disable this firing since we are already reporting
720 		 * it as an overrun (thanks to bump_cpu_timer above).
721 		 */
722 		unlock_task_sighand(p, &flags);
723 		goto out;
724 	}
725 
726 	if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
727 		new_expires += val;
728 	}
729 
730 	/*
731 	 * Install the new expiry time (or zero).
732 	 * For a timer with no notification action, we don't actually
733 	 * arm the timer (we'll just fake it for timer_gettime).
734 	 */
735 	cpu_timer_setexpires(ctmr, new_expires);
736 	if (new_expires != 0 && val < new_expires) {
737 		arm_timer(timer, p);
738 	}
739 
740 	unlock_task_sighand(p, &flags);
741 	/*
742 	 * Install the new reload setting, and
743 	 * set up the signal and overrun bookkeeping.
744 	 */
745 	timer->it_interval = timespec64_to_ktime(new->it_interval);
746 
747 	/*
748 	 * This acts as a modification timestamp for the timer,
749 	 * so any automatic reload attempt will punt on seeing
750 	 * that we have reset the timer manually.
751 	 */
752 	timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
753 		~REQUEUE_PENDING;
754 	timer->it_overrun_last = 0;
755 	timer->it_overrun = -1;
756 
757 	if (val >= new_expires) {
758 		if (new_expires != 0) {
759 			/*
760 			 * The designated time already passed, so we notify
761 			 * immediately, even if the thread never runs to
762 			 * accumulate more time on this clock.
763 			 */
764 			cpu_timer_fire(timer);
765 		}
766 
767 		/*
768 		 * Make sure we don't keep around the process wide cputime
769 		 * counter or the tick dependency if they are not necessary.
770 		 */
771 		sighand = lock_task_sighand(p, &flags);
772 		if (!sighand)
773 			goto out;
774 
775 		if (!cpu_timer_queued(ctmr))
776 			trigger_base_recalc_expires(timer, p);
777 
778 		unlock_task_sighand(p, &flags);
779 	}
780  out:
781 	rcu_read_unlock();
782 	if (old)
783 		old->it_interval = ns_to_timespec64(old_incr);
784 
785 	return ret;
786 }
787 
788 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
789 {
790 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
791 	struct cpu_timer *ctmr = &timer->it.cpu;
792 	u64 now, expires = cpu_timer_getexpires(ctmr);
793 	struct task_struct *p;
794 
795 	rcu_read_lock();
796 	p = cpu_timer_task_rcu(timer);
797 	if (!p)
798 		goto out;
799 
800 	/*
801 	 * Easy part: convert the reload time.
802 	 */
803 	itp->it_interval = ktime_to_timespec64(timer->it_interval);
804 
805 	if (!expires)
806 		goto out;
807 
808 	/*
809 	 * Sample the clock to take the difference with the expiry time.
810 	 */
811 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
812 		now = cpu_clock_sample(clkid, p);
813 	else
814 		now = cpu_clock_sample_group(clkid, p, false);
815 
816 	if (now < expires) {
817 		itp->it_value = ns_to_timespec64(expires - now);
818 	} else {
819 		/*
820 		 * The timer should have expired already, but the firing
821 		 * hasn't taken place yet.  Say it's just about to expire.
822 		 */
823 		itp->it_value.tv_nsec = 1;
824 		itp->it_value.tv_sec = 0;
825 	}
826 out:
827 	rcu_read_unlock();
828 }
829 
830 #define MAX_COLLECTED	20
831 
832 static u64 collect_timerqueue(struct timerqueue_head *head,
833 			      struct list_head *firing, u64 now)
834 {
835 	struct timerqueue_node *next;
836 	int i = 0;
837 
838 	while ((next = timerqueue_getnext(head))) {
839 		struct cpu_timer *ctmr;
840 		u64 expires;
841 
842 		ctmr = container_of(next, struct cpu_timer, node);
843 		expires = cpu_timer_getexpires(ctmr);
844 		/* Limit the number of timers to expire at once */
845 		if (++i == MAX_COLLECTED || now < expires)
846 			return expires;
847 
848 		ctmr->firing = 1;
849 		cpu_timer_dequeue(ctmr);
850 		list_add_tail(&ctmr->elist, firing);
851 	}
852 
853 	return U64_MAX;
854 }
855 
856 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
857 				    struct list_head *firing)
858 {
859 	struct posix_cputimer_base *base = pct->bases;
860 	int i;
861 
862 	for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
863 		base->nextevt = collect_timerqueue(&base->tqhead, firing,
864 						    samples[i]);
865 	}
866 }
867 
868 static inline void check_dl_overrun(struct task_struct *tsk)
869 {
870 	if (tsk->dl.dl_overrun) {
871 		tsk->dl.dl_overrun = 0;
872 		send_signal_locked(SIGXCPU, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
873 	}
874 }
875 
876 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
877 {
878 	if (time < limit)
879 		return false;
880 
881 	if (print_fatal_signals) {
882 		pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
883 			rt ? "RT" : "CPU", hard ? "hard" : "soft",
884 			current->comm, task_pid_nr(current));
885 	}
886 	send_signal_locked(signo, SEND_SIG_PRIV, current, PIDTYPE_TGID);
887 	return true;
888 }
889 
890 /*
891  * Check for any per-thread CPU timers that have fired and move them off
892  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
893  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
894  */
895 static void check_thread_timers(struct task_struct *tsk,
896 				struct list_head *firing)
897 {
898 	struct posix_cputimers *pct = &tsk->posix_cputimers;
899 	u64 samples[CPUCLOCK_MAX];
900 	unsigned long soft;
901 
902 	if (dl_task(tsk))
903 		check_dl_overrun(tsk);
904 
905 	if (expiry_cache_is_inactive(pct))
906 		return;
907 
908 	task_sample_cputime(tsk, samples);
909 	collect_posix_cputimers(pct, samples, firing);
910 
911 	/*
912 	 * Check for the special case thread timers.
913 	 */
914 	soft = task_rlimit(tsk, RLIMIT_RTTIME);
915 	if (soft != RLIM_INFINITY) {
916 		/* Task RT timeout is accounted in jiffies. RTTIME is usec */
917 		unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
918 		unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
919 
920 		/* At the hard limit, send SIGKILL. No further action. */
921 		if (hard != RLIM_INFINITY &&
922 		    check_rlimit(rttime, hard, SIGKILL, true, true))
923 			return;
924 
925 		/* At the soft limit, send a SIGXCPU every second */
926 		if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
927 			soft += USEC_PER_SEC;
928 			tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
929 		}
930 	}
931 
932 	if (expiry_cache_is_inactive(pct))
933 		tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
934 }
935 
936 static inline void stop_process_timers(struct signal_struct *sig)
937 {
938 	struct posix_cputimers *pct = &sig->posix_cputimers;
939 
940 	/* Turn off the active flag. This is done without locking. */
941 	WRITE_ONCE(pct->timers_active, false);
942 	tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
943 }
944 
945 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
946 			     u64 *expires, u64 cur_time, int signo)
947 {
948 	if (!it->expires)
949 		return;
950 
951 	if (cur_time >= it->expires) {
952 		if (it->incr)
953 			it->expires += it->incr;
954 		else
955 			it->expires = 0;
956 
957 		trace_itimer_expire(signo == SIGPROF ?
958 				    ITIMER_PROF : ITIMER_VIRTUAL,
959 				    task_tgid(tsk), cur_time);
960 		send_signal_locked(signo, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
961 	}
962 
963 	if (it->expires && it->expires < *expires)
964 		*expires = it->expires;
965 }
966 
967 /*
968  * Check for any per-thread CPU timers that have fired and move them
969  * off the tsk->*_timers list onto the firing list.  Per-thread timers
970  * have already been taken off.
971  */
972 static void check_process_timers(struct task_struct *tsk,
973 				 struct list_head *firing)
974 {
975 	struct signal_struct *const sig = tsk->signal;
976 	struct posix_cputimers *pct = &sig->posix_cputimers;
977 	u64 samples[CPUCLOCK_MAX];
978 	unsigned long soft;
979 
980 	/*
981 	 * If there are no active process wide timers (POSIX 1.b, itimers,
982 	 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
983 	 * processing when there is already another task handling them.
984 	 */
985 	if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
986 		return;
987 
988 	/*
989 	 * Signify that a thread is checking for process timers.
990 	 * Write access to this field is protected by the sighand lock.
991 	 */
992 	pct->expiry_active = true;
993 
994 	/*
995 	 * Collect the current process totals. Group accounting is active
996 	 * so the sample can be taken directly.
997 	 */
998 	proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
999 	collect_posix_cputimers(pct, samples, firing);
1000 
1001 	/*
1002 	 * Check for the special case process timers.
1003 	 */
1004 	check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
1005 			 &pct->bases[CPUCLOCK_PROF].nextevt,
1006 			 samples[CPUCLOCK_PROF], SIGPROF);
1007 	check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
1008 			 &pct->bases[CPUCLOCK_VIRT].nextevt,
1009 			 samples[CPUCLOCK_VIRT], SIGVTALRM);
1010 
1011 	soft = task_rlimit(tsk, RLIMIT_CPU);
1012 	if (soft != RLIM_INFINITY) {
1013 		/* RLIMIT_CPU is in seconds. Samples are nanoseconds */
1014 		unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
1015 		u64 ptime = samples[CPUCLOCK_PROF];
1016 		u64 softns = (u64)soft * NSEC_PER_SEC;
1017 		u64 hardns = (u64)hard * NSEC_PER_SEC;
1018 
1019 		/* At the hard limit, send SIGKILL. No further action. */
1020 		if (hard != RLIM_INFINITY &&
1021 		    check_rlimit(ptime, hardns, SIGKILL, false, true))
1022 			return;
1023 
1024 		/* At the soft limit, send a SIGXCPU every second */
1025 		if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
1026 			sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
1027 			softns += NSEC_PER_SEC;
1028 		}
1029 
1030 		/* Update the expiry cache */
1031 		if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
1032 			pct->bases[CPUCLOCK_PROF].nextevt = softns;
1033 	}
1034 
1035 	if (expiry_cache_is_inactive(pct))
1036 		stop_process_timers(sig);
1037 
1038 	pct->expiry_active = false;
1039 }
1040 
1041 /*
1042  * This is called from the signal code (via posixtimer_rearm)
1043  * when the last timer signal was delivered and we have to reload the timer.
1044  */
1045 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1046 {
1047 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
1048 	struct task_struct *p;
1049 	struct sighand_struct *sighand;
1050 	unsigned long flags;
1051 	u64 now;
1052 
1053 	rcu_read_lock();
1054 	p = cpu_timer_task_rcu(timer);
1055 	if (!p)
1056 		goto out;
1057 
1058 	/* Protect timer list r/w in arm_timer() */
1059 	sighand = lock_task_sighand(p, &flags);
1060 	if (unlikely(sighand == NULL))
1061 		goto out;
1062 
1063 	/*
1064 	 * Fetch the current sample and update the timer's expiry time.
1065 	 */
1066 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
1067 		now = cpu_clock_sample(clkid, p);
1068 	else
1069 		now = cpu_clock_sample_group(clkid, p, true);
1070 
1071 	bump_cpu_timer(timer, now);
1072 
1073 	/*
1074 	 * Now re-arm for the new expiry time.
1075 	 */
1076 	arm_timer(timer, p);
1077 	unlock_task_sighand(p, &flags);
1078 out:
1079 	rcu_read_unlock();
1080 }
1081 
1082 /**
1083  * task_cputimers_expired - Check whether posix CPU timers are expired
1084  *
1085  * @samples:	Array of current samples for the CPUCLOCK clocks
1086  * @pct:	Pointer to a posix_cputimers container
1087  *
1088  * Returns true if any member of @samples is greater than the corresponding
1089  * member of @pct->bases[CLK].nextevt. False otherwise
1090  */
1091 static inline bool
1092 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1093 {
1094 	int i;
1095 
1096 	for (i = 0; i < CPUCLOCK_MAX; i++) {
1097 		if (samples[i] >= pct->bases[i].nextevt)
1098 			return true;
1099 	}
1100 	return false;
1101 }
1102 
1103 /**
1104  * fastpath_timer_check - POSIX CPU timers fast path.
1105  *
1106  * @tsk:	The task (thread) being checked.
1107  *
1108  * Check the task and thread group timers.  If both are zero (there are no
1109  * timers set) return false.  Otherwise snapshot the task and thread group
1110  * timers and compare them with the corresponding expiration times.  Return
1111  * true if a timer has expired, else return false.
1112  */
1113 static inline bool fastpath_timer_check(struct task_struct *tsk)
1114 {
1115 	struct posix_cputimers *pct = &tsk->posix_cputimers;
1116 	struct signal_struct *sig;
1117 
1118 	if (!expiry_cache_is_inactive(pct)) {
1119 		u64 samples[CPUCLOCK_MAX];
1120 
1121 		task_sample_cputime(tsk, samples);
1122 		if (task_cputimers_expired(samples, pct))
1123 			return true;
1124 	}
1125 
1126 	sig = tsk->signal;
1127 	pct = &sig->posix_cputimers;
1128 	/*
1129 	 * Check if thread group timers expired when timers are active and
1130 	 * no other thread in the group is already handling expiry for
1131 	 * thread group cputimers. These fields are read without the
1132 	 * sighand lock. However, this is fine because this is meant to be
1133 	 * a fastpath heuristic to determine whether we should try to
1134 	 * acquire the sighand lock to handle timer expiry.
1135 	 *
1136 	 * In the worst case scenario, if concurrently timers_active is set
1137 	 * or expiry_active is cleared, but the current thread doesn't see
1138 	 * the change yet, the timer checks are delayed until the next
1139 	 * thread in the group gets a scheduler interrupt to handle the
1140 	 * timer. This isn't an issue in practice because these types of
1141 	 * delays with signals actually getting sent are expected.
1142 	 */
1143 	if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1144 		u64 samples[CPUCLOCK_MAX];
1145 
1146 		proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1147 					   samples);
1148 
1149 		if (task_cputimers_expired(samples, pct))
1150 			return true;
1151 	}
1152 
1153 	if (dl_task(tsk) && tsk->dl.dl_overrun)
1154 		return true;
1155 
1156 	return false;
1157 }
1158 
1159 static void handle_posix_cpu_timers(struct task_struct *tsk);
1160 
1161 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1162 static void posix_cpu_timers_work(struct callback_head *work)
1163 {
1164 	handle_posix_cpu_timers(current);
1165 }
1166 
1167 /*
1168  * Clear existing posix CPU timers task work.
1169  */
1170 void clear_posix_cputimers_work(struct task_struct *p)
1171 {
1172 	/*
1173 	 * A copied work entry from the old task is not meaningful, clear it.
1174 	 * N.B. init_task_work will not do this.
1175 	 */
1176 	memset(&p->posix_cputimers_work.work, 0,
1177 	       sizeof(p->posix_cputimers_work.work));
1178 	init_task_work(&p->posix_cputimers_work.work,
1179 		       posix_cpu_timers_work);
1180 	p->posix_cputimers_work.scheduled = false;
1181 }
1182 
1183 /*
1184  * Initialize posix CPU timers task work in init task. Out of line to
1185  * keep the callback static and to avoid header recursion hell.
1186  */
1187 void __init posix_cputimers_init_work(void)
1188 {
1189 	clear_posix_cputimers_work(current);
1190 }
1191 
1192 /*
1193  * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1194  * in hard interrupt context or in task context with interrupts
1195  * disabled. Aside of that the writer/reader interaction is always in the
1196  * context of the current task, which means they are strict per CPU.
1197  */
1198 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1199 {
1200 	return tsk->posix_cputimers_work.scheduled;
1201 }
1202 
1203 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1204 {
1205 	if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1206 		return;
1207 
1208 	/* Schedule task work to actually expire the timers */
1209 	tsk->posix_cputimers_work.scheduled = true;
1210 	task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1211 }
1212 
1213 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1214 						unsigned long start)
1215 {
1216 	bool ret = true;
1217 
1218 	/*
1219 	 * On !RT kernels interrupts are disabled while collecting expired
1220 	 * timers, so no tick can happen and the fast path check can be
1221 	 * reenabled without further checks.
1222 	 */
1223 	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1224 		tsk->posix_cputimers_work.scheduled = false;
1225 		return true;
1226 	}
1227 
1228 	/*
1229 	 * On RT enabled kernels ticks can happen while the expired timers
1230 	 * are collected under sighand lock. But any tick which observes
1231 	 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1232 	 * checks. So reenabling the tick work has do be done carefully:
1233 	 *
1234 	 * Disable interrupts and run the fast path check if jiffies have
1235 	 * advanced since the collecting of expired timers started. If
1236 	 * jiffies have not advanced or the fast path check did not find
1237 	 * newly expired timers, reenable the fast path check in the timer
1238 	 * interrupt. If there are newly expired timers, return false and
1239 	 * let the collection loop repeat.
1240 	 */
1241 	local_irq_disable();
1242 	if (start != jiffies && fastpath_timer_check(tsk))
1243 		ret = false;
1244 	else
1245 		tsk->posix_cputimers_work.scheduled = false;
1246 	local_irq_enable();
1247 
1248 	return ret;
1249 }
1250 #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1251 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1252 {
1253 	lockdep_posixtimer_enter();
1254 	handle_posix_cpu_timers(tsk);
1255 	lockdep_posixtimer_exit();
1256 }
1257 
1258 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1259 {
1260 	return false;
1261 }
1262 
1263 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1264 						unsigned long start)
1265 {
1266 	return true;
1267 }
1268 #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1269 
1270 static void handle_posix_cpu_timers(struct task_struct *tsk)
1271 {
1272 	struct k_itimer *timer, *next;
1273 	unsigned long flags, start;
1274 	LIST_HEAD(firing);
1275 
1276 	if (!lock_task_sighand(tsk, &flags))
1277 		return;
1278 
1279 	do {
1280 		/*
1281 		 * On RT locking sighand lock does not disable interrupts,
1282 		 * so this needs to be careful vs. ticks. Store the current
1283 		 * jiffies value.
1284 		 */
1285 		start = READ_ONCE(jiffies);
1286 		barrier();
1287 
1288 		/*
1289 		 * Here we take off tsk->signal->cpu_timers[N] and
1290 		 * tsk->cpu_timers[N] all the timers that are firing, and
1291 		 * put them on the firing list.
1292 		 */
1293 		check_thread_timers(tsk, &firing);
1294 
1295 		check_process_timers(tsk, &firing);
1296 
1297 		/*
1298 		 * The above timer checks have updated the expiry cache and
1299 		 * because nothing can have queued or modified timers after
1300 		 * sighand lock was taken above it is guaranteed to be
1301 		 * consistent. So the next timer interrupt fastpath check
1302 		 * will find valid data.
1303 		 *
1304 		 * If timer expiry runs in the timer interrupt context then
1305 		 * the loop is not relevant as timers will be directly
1306 		 * expired in interrupt context. The stub function below
1307 		 * returns always true which allows the compiler to
1308 		 * optimize the loop out.
1309 		 *
1310 		 * If timer expiry is deferred to task work context then
1311 		 * the following rules apply:
1312 		 *
1313 		 * - On !RT kernels no tick can have happened on this CPU
1314 		 *   after sighand lock was acquired because interrupts are
1315 		 *   disabled. So reenabling task work before dropping
1316 		 *   sighand lock and reenabling interrupts is race free.
1317 		 *
1318 		 * - On RT kernels ticks might have happened but the tick
1319 		 *   work ignored posix CPU timer handling because the
1320 		 *   CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1321 		 *   must be done very carefully including a check whether
1322 		 *   ticks have happened since the start of the timer
1323 		 *   expiry checks. posix_cpu_timers_enable_work() takes
1324 		 *   care of that and eventually lets the expiry checks
1325 		 *   run again.
1326 		 */
1327 	} while (!posix_cpu_timers_enable_work(tsk, start));
1328 
1329 	/*
1330 	 * We must release sighand lock before taking any timer's lock.
1331 	 * There is a potential race with timer deletion here, as the
1332 	 * siglock now protects our private firing list.  We have set
1333 	 * the firing flag in each timer, so that a deletion attempt
1334 	 * that gets the timer lock before we do will give it up and
1335 	 * spin until we've taken care of that timer below.
1336 	 */
1337 	unlock_task_sighand(tsk, &flags);
1338 
1339 	/*
1340 	 * Now that all the timers on our list have the firing flag,
1341 	 * no one will touch their list entries but us.  We'll take
1342 	 * each timer's lock before clearing its firing flag, so no
1343 	 * timer call will interfere.
1344 	 */
1345 	list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1346 		int cpu_firing;
1347 
1348 		/*
1349 		 * spin_lock() is sufficient here even independent of the
1350 		 * expiry context. If expiry happens in hard interrupt
1351 		 * context it's obvious. For task work context it's safe
1352 		 * because all other operations on timer::it_lock happen in
1353 		 * task context (syscall or exit).
1354 		 */
1355 		spin_lock(&timer->it_lock);
1356 		list_del_init(&timer->it.cpu.elist);
1357 		cpu_firing = timer->it.cpu.firing;
1358 		timer->it.cpu.firing = 0;
1359 		/*
1360 		 * The firing flag is -1 if we collided with a reset
1361 		 * of the timer, which already reported this
1362 		 * almost-firing as an overrun.  So don't generate an event.
1363 		 */
1364 		if (likely(cpu_firing >= 0))
1365 			cpu_timer_fire(timer);
1366 		spin_unlock(&timer->it_lock);
1367 	}
1368 }
1369 
1370 /*
1371  * This is called from the timer interrupt handler.  The irq handler has
1372  * already updated our counts.  We need to check if any timers fire now.
1373  * Interrupts are disabled.
1374  */
1375 void run_posix_cpu_timers(void)
1376 {
1377 	struct task_struct *tsk = current;
1378 
1379 	lockdep_assert_irqs_disabled();
1380 
1381 	/*
1382 	 * If the actual expiry is deferred to task work context and the
1383 	 * work is already scheduled there is no point to do anything here.
1384 	 */
1385 	if (posix_cpu_timers_work_scheduled(tsk))
1386 		return;
1387 
1388 	/*
1389 	 * The fast path checks that there are no expired thread or thread
1390 	 * group timers.  If that's so, just return.
1391 	 */
1392 	if (!fastpath_timer_check(tsk))
1393 		return;
1394 
1395 	__run_posix_cpu_timers(tsk);
1396 }
1397 
1398 /*
1399  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1400  * The tsk->sighand->siglock must be held by the caller.
1401  */
1402 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1403 			   u64 *newval, u64 *oldval)
1404 {
1405 	u64 now, *nextevt;
1406 
1407 	if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1408 		return;
1409 
1410 	nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1411 	now = cpu_clock_sample_group(clkid, tsk, true);
1412 
1413 	if (oldval) {
1414 		/*
1415 		 * We are setting itimer. The *oldval is absolute and we update
1416 		 * it to be relative, *newval argument is relative and we update
1417 		 * it to be absolute.
1418 		 */
1419 		if (*oldval) {
1420 			if (*oldval <= now) {
1421 				/* Just about to fire. */
1422 				*oldval = TICK_NSEC;
1423 			} else {
1424 				*oldval -= now;
1425 			}
1426 		}
1427 
1428 		if (*newval)
1429 			*newval += now;
1430 	}
1431 
1432 	/*
1433 	 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1434 	 * expiry cache is also used by RLIMIT_CPU!.
1435 	 */
1436 	if (*newval < *nextevt)
1437 		*nextevt = *newval;
1438 
1439 	tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER);
1440 }
1441 
1442 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1443 			    const struct timespec64 *rqtp)
1444 {
1445 	struct itimerspec64 it;
1446 	struct k_itimer timer;
1447 	u64 expires;
1448 	int error;
1449 
1450 	/*
1451 	 * Set up a temporary timer and then wait for it to go off.
1452 	 */
1453 	memset(&timer, 0, sizeof timer);
1454 	spin_lock_init(&timer.it_lock);
1455 	timer.it_clock = which_clock;
1456 	timer.it_overrun = -1;
1457 	error = posix_cpu_timer_create(&timer);
1458 	timer.it_process = current;
1459 
1460 	if (!error) {
1461 		static struct itimerspec64 zero_it;
1462 		struct restart_block *restart;
1463 
1464 		memset(&it, 0, sizeof(it));
1465 		it.it_value = *rqtp;
1466 
1467 		spin_lock_irq(&timer.it_lock);
1468 		error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1469 		if (error) {
1470 			spin_unlock_irq(&timer.it_lock);
1471 			return error;
1472 		}
1473 
1474 		while (!signal_pending(current)) {
1475 			if (!cpu_timer_getexpires(&timer.it.cpu)) {
1476 				/*
1477 				 * Our timer fired and was reset, below
1478 				 * deletion can not fail.
1479 				 */
1480 				posix_cpu_timer_del(&timer);
1481 				spin_unlock_irq(&timer.it_lock);
1482 				return 0;
1483 			}
1484 
1485 			/*
1486 			 * Block until cpu_timer_fire (or a signal) wakes us.
1487 			 */
1488 			__set_current_state(TASK_INTERRUPTIBLE);
1489 			spin_unlock_irq(&timer.it_lock);
1490 			schedule();
1491 			spin_lock_irq(&timer.it_lock);
1492 		}
1493 
1494 		/*
1495 		 * We were interrupted by a signal.
1496 		 */
1497 		expires = cpu_timer_getexpires(&timer.it.cpu);
1498 		error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1499 		if (!error) {
1500 			/*
1501 			 * Timer is now unarmed, deletion can not fail.
1502 			 */
1503 			posix_cpu_timer_del(&timer);
1504 		}
1505 		spin_unlock_irq(&timer.it_lock);
1506 
1507 		while (error == TIMER_RETRY) {
1508 			/*
1509 			 * We need to handle case when timer was or is in the
1510 			 * middle of firing. In other cases we already freed
1511 			 * resources.
1512 			 */
1513 			spin_lock_irq(&timer.it_lock);
1514 			error = posix_cpu_timer_del(&timer);
1515 			spin_unlock_irq(&timer.it_lock);
1516 		}
1517 
1518 		if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1519 			/*
1520 			 * It actually did fire already.
1521 			 */
1522 			return 0;
1523 		}
1524 
1525 		error = -ERESTART_RESTARTBLOCK;
1526 		/*
1527 		 * Report back to the user the time still remaining.
1528 		 */
1529 		restart = &current->restart_block;
1530 		restart->nanosleep.expires = expires;
1531 		if (restart->nanosleep.type != TT_NONE)
1532 			error = nanosleep_copyout(restart, &it.it_value);
1533 	}
1534 
1535 	return error;
1536 }
1537 
1538 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1539 
1540 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1541 			    const struct timespec64 *rqtp)
1542 {
1543 	struct restart_block *restart_block = &current->restart_block;
1544 	int error;
1545 
1546 	/*
1547 	 * Diagnose required errors first.
1548 	 */
1549 	if (CPUCLOCK_PERTHREAD(which_clock) &&
1550 	    (CPUCLOCK_PID(which_clock) == 0 ||
1551 	     CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1552 		return -EINVAL;
1553 
1554 	error = do_cpu_nanosleep(which_clock, flags, rqtp);
1555 
1556 	if (error == -ERESTART_RESTARTBLOCK) {
1557 
1558 		if (flags & TIMER_ABSTIME)
1559 			return -ERESTARTNOHAND;
1560 
1561 		restart_block->nanosleep.clockid = which_clock;
1562 		set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1563 	}
1564 	return error;
1565 }
1566 
1567 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1568 {
1569 	clockid_t which_clock = restart_block->nanosleep.clockid;
1570 	struct timespec64 t;
1571 
1572 	t = ns_to_timespec64(restart_block->nanosleep.expires);
1573 
1574 	return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1575 }
1576 
1577 #define PROCESS_CLOCK	make_process_cpuclock(0, CPUCLOCK_SCHED)
1578 #define THREAD_CLOCK	make_thread_cpuclock(0, CPUCLOCK_SCHED)
1579 
1580 static int process_cpu_clock_getres(const clockid_t which_clock,
1581 				    struct timespec64 *tp)
1582 {
1583 	return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1584 }
1585 static int process_cpu_clock_get(const clockid_t which_clock,
1586 				 struct timespec64 *tp)
1587 {
1588 	return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1589 }
1590 static int process_cpu_timer_create(struct k_itimer *timer)
1591 {
1592 	timer->it_clock = PROCESS_CLOCK;
1593 	return posix_cpu_timer_create(timer);
1594 }
1595 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1596 			      const struct timespec64 *rqtp)
1597 {
1598 	return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1599 }
1600 static int thread_cpu_clock_getres(const clockid_t which_clock,
1601 				   struct timespec64 *tp)
1602 {
1603 	return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1604 }
1605 static int thread_cpu_clock_get(const clockid_t which_clock,
1606 				struct timespec64 *tp)
1607 {
1608 	return posix_cpu_clock_get(THREAD_CLOCK, tp);
1609 }
1610 static int thread_cpu_timer_create(struct k_itimer *timer)
1611 {
1612 	timer->it_clock = THREAD_CLOCK;
1613 	return posix_cpu_timer_create(timer);
1614 }
1615 
1616 const struct k_clock clock_posix_cpu = {
1617 	.clock_getres		= posix_cpu_clock_getres,
1618 	.clock_set		= posix_cpu_clock_set,
1619 	.clock_get_timespec	= posix_cpu_clock_get,
1620 	.timer_create		= posix_cpu_timer_create,
1621 	.nsleep			= posix_cpu_nsleep,
1622 	.timer_set		= posix_cpu_timer_set,
1623 	.timer_del		= posix_cpu_timer_del,
1624 	.timer_get		= posix_cpu_timer_get,
1625 	.timer_rearm		= posix_cpu_timer_rearm,
1626 };
1627 
1628 const struct k_clock clock_process = {
1629 	.clock_getres		= process_cpu_clock_getres,
1630 	.clock_get_timespec	= process_cpu_clock_get,
1631 	.timer_create		= process_cpu_timer_create,
1632 	.nsleep			= process_cpu_nsleep,
1633 };
1634 
1635 const struct k_clock clock_thread = {
1636 	.clock_getres		= thread_cpu_clock_getres,
1637 	.clock_get_timespec	= thread_cpu_clock_get,
1638 	.timer_create		= thread_cpu_timer_create,
1639 };
1640