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;
247 retry:
248 	curr_cputime = atomic64_read(cputime);
249 	if (sum_cputime > curr_cputime) {
250 		if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
251 			goto retry;
252 	}
253 }
254 
255 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
256 			      struct task_cputime *sum)
257 {
258 	__update_gt_cputime(&cputime_atomic->utime, sum->utime);
259 	__update_gt_cputime(&cputime_atomic->stime, sum->stime);
260 	__update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
261 }
262 
263 /**
264  * thread_group_sample_cputime - Sample cputime for a given task
265  * @tsk:	Task for which cputime needs to be started
266  * @samples:	Storage for time samples
267  *
268  * Called from sys_getitimer() to calculate the expiry time of an active
269  * timer. That means group cputime accounting is already active. Called
270  * with task sighand lock held.
271  *
272  * Updates @times with an uptodate sample of the thread group cputimes.
273  */
274 void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
275 {
276 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
277 	struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
278 
279 	WARN_ON_ONCE(!pct->timers_active);
280 
281 	proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
282 }
283 
284 /**
285  * thread_group_start_cputime - Start cputime and return a sample
286  * @tsk:	Task for which cputime needs to be started
287  * @samples:	Storage for time samples
288  *
289  * The thread group cputime accounting is avoided when there are no posix
290  * CPU timers armed. Before starting a timer it's required to check whether
291  * the time accounting is active. If not, a full update of the atomic
292  * accounting store needs to be done and the accounting enabled.
293  *
294  * Updates @times with an uptodate sample of the thread group cputimes.
295  */
296 static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
297 {
298 	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
299 	struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
300 
301 	lockdep_assert_task_sighand_held(tsk);
302 
303 	/* Check if cputimer isn't running. This is accessed without locking. */
304 	if (!READ_ONCE(pct->timers_active)) {
305 		struct task_cputime sum;
306 
307 		/*
308 		 * The POSIX timer interface allows for absolute time expiry
309 		 * values through the TIMER_ABSTIME flag, therefore we have
310 		 * to synchronize the timer to the clock every time we start it.
311 		 */
312 		thread_group_cputime(tsk, &sum);
313 		update_gt_cputime(&cputimer->cputime_atomic, &sum);
314 
315 		/*
316 		 * We're setting timers_active without a lock. Ensure this
317 		 * only gets written to in one operation. We set it after
318 		 * update_gt_cputime() as a small optimization, but
319 		 * barriers are not required because update_gt_cputime()
320 		 * can handle concurrent updates.
321 		 */
322 		WRITE_ONCE(pct->timers_active, true);
323 	}
324 	proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
325 }
326 
327 static void __thread_group_cputime(struct task_struct *tsk, u64 *samples)
328 {
329 	struct task_cputime ct;
330 
331 	thread_group_cputime(tsk, &ct);
332 	store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime);
333 }
334 
335 /*
336  * Sample a process (thread group) clock for the given task clkid. If the
337  * group's cputime accounting is already enabled, read the atomic
338  * store. Otherwise a full update is required.  clkid is already validated.
339  */
340 static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
341 				  bool start)
342 {
343 	struct thread_group_cputimer *cputimer = &p->signal->cputimer;
344 	struct posix_cputimers *pct = &p->signal->posix_cputimers;
345 	u64 samples[CPUCLOCK_MAX];
346 
347 	if (!READ_ONCE(pct->timers_active)) {
348 		if (start)
349 			thread_group_start_cputime(p, samples);
350 		else
351 			__thread_group_cputime(p, samples);
352 	} else {
353 		proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
354 	}
355 
356 	return samples[clkid];
357 }
358 
359 static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
360 {
361 	const clockid_t clkid = CPUCLOCK_WHICH(clock);
362 	struct task_struct *tsk;
363 	u64 t;
364 
365 	rcu_read_lock();
366 	tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
367 	if (!tsk) {
368 		rcu_read_unlock();
369 		return -EINVAL;
370 	}
371 
372 	if (CPUCLOCK_PERTHREAD(clock))
373 		t = cpu_clock_sample(clkid, tsk);
374 	else
375 		t = cpu_clock_sample_group(clkid, tsk, false);
376 	rcu_read_unlock();
377 
378 	*tp = ns_to_timespec64(t);
379 	return 0;
380 }
381 
382 /*
383  * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
384  * This is called from sys_timer_create() and do_cpu_nanosleep() with the
385  * new timer already all-zeros initialized.
386  */
387 static int posix_cpu_timer_create(struct k_itimer *new_timer)
388 {
389 	static struct lock_class_key posix_cpu_timers_key;
390 	struct pid *pid;
391 
392 	rcu_read_lock();
393 	pid = pid_for_clock(new_timer->it_clock, false);
394 	if (!pid) {
395 		rcu_read_unlock();
396 		return -EINVAL;
397 	}
398 
399 	/*
400 	 * If posix timer expiry is handled in task work context then
401 	 * timer::it_lock can be taken without disabling interrupts as all
402 	 * other locking happens in task context. This requires a separate
403 	 * lock class key otherwise regular posix timer expiry would record
404 	 * the lock class being taken in interrupt context and generate a
405 	 * false positive warning.
406 	 */
407 	if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK))
408 		lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key);
409 
410 	new_timer->kclock = &clock_posix_cpu;
411 	timerqueue_init(&new_timer->it.cpu.node);
412 	new_timer->it.cpu.pid = get_pid(pid);
413 	rcu_read_unlock();
414 	return 0;
415 }
416 
417 static struct posix_cputimer_base *timer_base(struct k_itimer *timer,
418 					      struct task_struct *tsk)
419 {
420 	int clkidx = CPUCLOCK_WHICH(timer->it_clock);
421 
422 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
423 		return tsk->posix_cputimers.bases + clkidx;
424 	else
425 		return tsk->signal->posix_cputimers.bases + clkidx;
426 }
427 
428 /*
429  * Force recalculating the base earliest expiration on the next tick.
430  * This will also re-evaluate the need to keep around the process wide
431  * cputime counter and tick dependency and eventually shut these down
432  * if necessary.
433  */
434 static void trigger_base_recalc_expires(struct k_itimer *timer,
435 					struct task_struct *tsk)
436 {
437 	struct posix_cputimer_base *base = timer_base(timer, tsk);
438 
439 	base->nextevt = 0;
440 }
441 
442 /*
443  * Dequeue the timer and reset the base if it was its earliest expiration.
444  * It makes sure the next tick recalculates the base next expiration so we
445  * don't keep the costly process wide cputime counter around for a random
446  * amount of time, along with the tick dependency.
447  *
448  * If another timer gets queued between this and the next tick, its
449  * expiration will update the base next event if necessary on the next
450  * tick.
451  */
452 static void disarm_timer(struct k_itimer *timer, struct task_struct *p)
453 {
454 	struct cpu_timer *ctmr = &timer->it.cpu;
455 	struct posix_cputimer_base *base;
456 
457 	if (!cpu_timer_dequeue(ctmr))
458 		return;
459 
460 	base = timer_base(timer, p);
461 	if (cpu_timer_getexpires(ctmr) == base->nextevt)
462 		trigger_base_recalc_expires(timer, p);
463 }
464 
465 
466 /*
467  * Clean up a CPU-clock timer that is about to be destroyed.
468  * This is called from timer deletion with the timer already locked.
469  * If we return TIMER_RETRY, it's necessary to release the timer's lock
470  * and try again.  (This happens when the timer is in the middle of firing.)
471  */
472 static int posix_cpu_timer_del(struct k_itimer *timer)
473 {
474 	struct cpu_timer *ctmr = &timer->it.cpu;
475 	struct sighand_struct *sighand;
476 	struct task_struct *p;
477 	unsigned long flags;
478 	int ret = 0;
479 
480 	rcu_read_lock();
481 	p = cpu_timer_task_rcu(timer);
482 	if (!p)
483 		goto out;
484 
485 	/*
486 	 * Protect against sighand release/switch in exit/exec and process/
487 	 * thread timer list entry concurrent read/writes.
488 	 */
489 	sighand = lock_task_sighand(p, &flags);
490 	if (unlikely(sighand == NULL)) {
491 		/*
492 		 * This raced with the reaping of the task. The exit cleanup
493 		 * should have removed this timer from the timer queue.
494 		 */
495 		WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
496 	} else {
497 		if (timer->it.cpu.firing)
498 			ret = TIMER_RETRY;
499 		else
500 			disarm_timer(timer, p);
501 
502 		unlock_task_sighand(p, &flags);
503 	}
504 
505 out:
506 	rcu_read_unlock();
507 	if (!ret)
508 		put_pid(ctmr->pid);
509 
510 	return ret;
511 }
512 
513 static void cleanup_timerqueue(struct timerqueue_head *head)
514 {
515 	struct timerqueue_node *node;
516 	struct cpu_timer *ctmr;
517 
518 	while ((node = timerqueue_getnext(head))) {
519 		timerqueue_del(head, node);
520 		ctmr = container_of(node, struct cpu_timer, node);
521 		ctmr->head = NULL;
522 	}
523 }
524 
525 /*
526  * Clean out CPU timers which are still armed when a thread exits. The
527  * timers are only removed from the list. No other updates are done. The
528  * corresponding posix timers are still accessible, but cannot be rearmed.
529  *
530  * This must be called with the siglock held.
531  */
532 static void cleanup_timers(struct posix_cputimers *pct)
533 {
534 	cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
535 	cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
536 	cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
537 }
538 
539 /*
540  * These are both called with the siglock held, when the current thread
541  * is being reaped.  When the final (leader) thread in the group is reaped,
542  * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
543  */
544 void posix_cpu_timers_exit(struct task_struct *tsk)
545 {
546 	cleanup_timers(&tsk->posix_cputimers);
547 }
548 void posix_cpu_timers_exit_group(struct task_struct *tsk)
549 {
550 	cleanup_timers(&tsk->signal->posix_cputimers);
551 }
552 
553 /*
554  * Insert the timer on the appropriate list before any timers that
555  * expire later.  This must be called with the sighand lock held.
556  */
557 static void arm_timer(struct k_itimer *timer, struct task_struct *p)
558 {
559 	struct posix_cputimer_base *base = timer_base(timer, p);
560 	struct cpu_timer *ctmr = &timer->it.cpu;
561 	u64 newexp = cpu_timer_getexpires(ctmr);
562 
563 	if (!cpu_timer_enqueue(&base->tqhead, ctmr))
564 		return;
565 
566 	/*
567 	 * We are the new earliest-expiring POSIX 1.b timer, hence
568 	 * need to update expiration cache. Take into account that
569 	 * for process timers we share expiration cache with itimers
570 	 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
571 	 */
572 	if (newexp < base->nextevt)
573 		base->nextevt = newexp;
574 
575 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
576 		tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
577 	else
578 		tick_dep_set_signal(p, TICK_DEP_BIT_POSIX_TIMER);
579 }
580 
581 /*
582  * The timer is locked, fire it and arrange for its reload.
583  */
584 static void cpu_timer_fire(struct k_itimer *timer)
585 {
586 	struct cpu_timer *ctmr = &timer->it.cpu;
587 
588 	if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
589 		/*
590 		 * User don't want any signal.
591 		 */
592 		cpu_timer_setexpires(ctmr, 0);
593 	} else if (unlikely(timer->sigq == NULL)) {
594 		/*
595 		 * This a special case for clock_nanosleep,
596 		 * not a normal timer from sys_timer_create.
597 		 */
598 		wake_up_process(timer->it_process);
599 		cpu_timer_setexpires(ctmr, 0);
600 	} else if (!timer->it_interval) {
601 		/*
602 		 * One-shot timer.  Clear it as soon as it's fired.
603 		 */
604 		posix_timer_event(timer, 0);
605 		cpu_timer_setexpires(ctmr, 0);
606 	} else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
607 		/*
608 		 * The signal did not get queued because the signal
609 		 * was ignored, so we won't get any callback to
610 		 * reload the timer.  But we need to keep it
611 		 * ticking in case the signal is deliverable next time.
612 		 */
613 		posix_cpu_timer_rearm(timer);
614 		++timer->it_requeue_pending;
615 	}
616 }
617 
618 /*
619  * Guts of sys_timer_settime for CPU timers.
620  * This is called with the timer locked and interrupts disabled.
621  * If we return TIMER_RETRY, it's necessary to release the timer's lock
622  * and try again.  (This happens when the timer is in the middle of firing.)
623  */
624 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
625 			       struct itimerspec64 *new, struct itimerspec64 *old)
626 {
627 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
628 	u64 old_expires, new_expires, old_incr, val;
629 	struct cpu_timer *ctmr = &timer->it.cpu;
630 	struct sighand_struct *sighand;
631 	struct task_struct *p;
632 	unsigned long flags;
633 	int ret = 0;
634 
635 	rcu_read_lock();
636 	p = cpu_timer_task_rcu(timer);
637 	if (!p) {
638 		/*
639 		 * If p has just been reaped, we can no
640 		 * longer get any information about it at all.
641 		 */
642 		rcu_read_unlock();
643 		return -ESRCH;
644 	}
645 
646 	/*
647 	 * Use the to_ktime conversion because that clamps the maximum
648 	 * value to KTIME_MAX and avoid multiplication overflows.
649 	 */
650 	new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
651 
652 	/*
653 	 * Protect against sighand release/switch in exit/exec and p->cpu_timers
654 	 * and p->signal->cpu_timers read/write in arm_timer()
655 	 */
656 	sighand = lock_task_sighand(p, &flags);
657 	/*
658 	 * If p has just been reaped, we can no
659 	 * longer get any information about it at all.
660 	 */
661 	if (unlikely(sighand == NULL)) {
662 		rcu_read_unlock();
663 		return -ESRCH;
664 	}
665 
666 	/*
667 	 * Disarm any old timer after extracting its expiry time.
668 	 */
669 	old_incr = timer->it_interval;
670 	old_expires = cpu_timer_getexpires(ctmr);
671 
672 	if (unlikely(timer->it.cpu.firing)) {
673 		timer->it.cpu.firing = -1;
674 		ret = TIMER_RETRY;
675 	} else {
676 		cpu_timer_dequeue(ctmr);
677 	}
678 
679 	/*
680 	 * We need to sample the current value to convert the new
681 	 * value from to relative and absolute, and to convert the
682 	 * old value from absolute to relative.  To set a process
683 	 * timer, we need a sample to balance the thread expiry
684 	 * times (in arm_timer).  With an absolute time, we must
685 	 * check if it's already passed.  In short, we need a sample.
686 	 */
687 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
688 		val = cpu_clock_sample(clkid, p);
689 	else
690 		val = cpu_clock_sample_group(clkid, p, true);
691 
692 	if (old) {
693 		if (old_expires == 0) {
694 			old->it_value.tv_sec = 0;
695 			old->it_value.tv_nsec = 0;
696 		} else {
697 			/*
698 			 * Update the timer in case it has overrun already.
699 			 * If it has, we'll report it as having overrun and
700 			 * with the next reloaded timer already ticking,
701 			 * though we are swallowing that pending
702 			 * notification here to install the new setting.
703 			 */
704 			u64 exp = bump_cpu_timer(timer, val);
705 
706 			if (val < exp) {
707 				old_expires = exp - val;
708 				old->it_value = ns_to_timespec64(old_expires);
709 			} else {
710 				old->it_value.tv_nsec = 1;
711 				old->it_value.tv_sec = 0;
712 			}
713 		}
714 	}
715 
716 	if (unlikely(ret)) {
717 		/*
718 		 * We are colliding with the timer actually firing.
719 		 * Punt after filling in the timer's old value, and
720 		 * disable this firing since we are already reporting
721 		 * it as an overrun (thanks to bump_cpu_timer above).
722 		 */
723 		unlock_task_sighand(p, &flags);
724 		goto out;
725 	}
726 
727 	if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
728 		new_expires += val;
729 	}
730 
731 	/*
732 	 * Install the new expiry time (or zero).
733 	 * For a timer with no notification action, we don't actually
734 	 * arm the timer (we'll just fake it for timer_gettime).
735 	 */
736 	cpu_timer_setexpires(ctmr, new_expires);
737 	if (new_expires != 0 && val < new_expires) {
738 		arm_timer(timer, p);
739 	}
740 
741 	unlock_task_sighand(p, &flags);
742 	/*
743 	 * Install the new reload setting, and
744 	 * set up the signal and overrun bookkeeping.
745 	 */
746 	timer->it_interval = timespec64_to_ktime(new->it_interval);
747 
748 	/*
749 	 * This acts as a modification timestamp for the timer,
750 	 * so any automatic reload attempt will punt on seeing
751 	 * that we have reset the timer manually.
752 	 */
753 	timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
754 		~REQUEUE_PENDING;
755 	timer->it_overrun_last = 0;
756 	timer->it_overrun = -1;
757 
758 	if (val >= new_expires) {
759 		if (new_expires != 0) {
760 			/*
761 			 * The designated time already passed, so we notify
762 			 * immediately, even if the thread never runs to
763 			 * accumulate more time on this clock.
764 			 */
765 			cpu_timer_fire(timer);
766 		}
767 
768 		/*
769 		 * Make sure we don't keep around the process wide cputime
770 		 * counter or the tick dependency if they are not necessary.
771 		 */
772 		sighand = lock_task_sighand(p, &flags);
773 		if (!sighand)
774 			goto out;
775 
776 		if (!cpu_timer_queued(ctmr))
777 			trigger_base_recalc_expires(timer, p);
778 
779 		unlock_task_sighand(p, &flags);
780 	}
781  out:
782 	rcu_read_unlock();
783 	if (old)
784 		old->it_interval = ns_to_timespec64(old_incr);
785 
786 	return ret;
787 }
788 
789 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
790 {
791 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
792 	struct cpu_timer *ctmr = &timer->it.cpu;
793 	u64 now, expires = cpu_timer_getexpires(ctmr);
794 	struct task_struct *p;
795 
796 	rcu_read_lock();
797 	p = cpu_timer_task_rcu(timer);
798 	if (!p)
799 		goto out;
800 
801 	/*
802 	 * Easy part: convert the reload time.
803 	 */
804 	itp->it_interval = ktime_to_timespec64(timer->it_interval);
805 
806 	if (!expires)
807 		goto out;
808 
809 	/*
810 	 * Sample the clock to take the difference with the expiry time.
811 	 */
812 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
813 		now = cpu_clock_sample(clkid, p);
814 	else
815 		now = cpu_clock_sample_group(clkid, p, false);
816 
817 	if (now < expires) {
818 		itp->it_value = ns_to_timespec64(expires - now);
819 	} else {
820 		/*
821 		 * The timer should have expired already, but the firing
822 		 * hasn't taken place yet.  Say it's just about to expire.
823 		 */
824 		itp->it_value.tv_nsec = 1;
825 		itp->it_value.tv_sec = 0;
826 	}
827 out:
828 	rcu_read_unlock();
829 }
830 
831 #define MAX_COLLECTED	20
832 
833 static u64 collect_timerqueue(struct timerqueue_head *head,
834 			      struct list_head *firing, u64 now)
835 {
836 	struct timerqueue_node *next;
837 	int i = 0;
838 
839 	while ((next = timerqueue_getnext(head))) {
840 		struct cpu_timer *ctmr;
841 		u64 expires;
842 
843 		ctmr = container_of(next, struct cpu_timer, node);
844 		expires = cpu_timer_getexpires(ctmr);
845 		/* Limit the number of timers to expire at once */
846 		if (++i == MAX_COLLECTED || now < expires)
847 			return expires;
848 
849 		ctmr->firing = 1;
850 		cpu_timer_dequeue(ctmr);
851 		list_add_tail(&ctmr->elist, firing);
852 	}
853 
854 	return U64_MAX;
855 }
856 
857 static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
858 				    struct list_head *firing)
859 {
860 	struct posix_cputimer_base *base = pct->bases;
861 	int i;
862 
863 	for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
864 		base->nextevt = collect_timerqueue(&base->tqhead, firing,
865 						    samples[i]);
866 	}
867 }
868 
869 static inline void check_dl_overrun(struct task_struct *tsk)
870 {
871 	if (tsk->dl.dl_overrun) {
872 		tsk->dl.dl_overrun = 0;
873 		send_signal_locked(SIGXCPU, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
874 	}
875 }
876 
877 static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
878 {
879 	if (time < limit)
880 		return false;
881 
882 	if (print_fatal_signals) {
883 		pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
884 			rt ? "RT" : "CPU", hard ? "hard" : "soft",
885 			current->comm, task_pid_nr(current));
886 	}
887 	send_signal_locked(signo, SEND_SIG_PRIV, current, PIDTYPE_TGID);
888 	return true;
889 }
890 
891 /*
892  * Check for any per-thread CPU timers that have fired and move them off
893  * the tsk->cpu_timers[N] list onto the firing list.  Here we update the
894  * tsk->it_*_expires values to reflect the remaining thread CPU timers.
895  */
896 static void check_thread_timers(struct task_struct *tsk,
897 				struct list_head *firing)
898 {
899 	struct posix_cputimers *pct = &tsk->posix_cputimers;
900 	u64 samples[CPUCLOCK_MAX];
901 	unsigned long soft;
902 
903 	if (dl_task(tsk))
904 		check_dl_overrun(tsk);
905 
906 	if (expiry_cache_is_inactive(pct))
907 		return;
908 
909 	task_sample_cputime(tsk, samples);
910 	collect_posix_cputimers(pct, samples, firing);
911 
912 	/*
913 	 * Check for the special case thread timers.
914 	 */
915 	soft = task_rlimit(tsk, RLIMIT_RTTIME);
916 	if (soft != RLIM_INFINITY) {
917 		/* Task RT timeout is accounted in jiffies. RTTIME is usec */
918 		unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
919 		unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
920 
921 		/* At the hard limit, send SIGKILL. No further action. */
922 		if (hard != RLIM_INFINITY &&
923 		    check_rlimit(rttime, hard, SIGKILL, true, true))
924 			return;
925 
926 		/* At the soft limit, send a SIGXCPU every second */
927 		if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
928 			soft += USEC_PER_SEC;
929 			tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
930 		}
931 	}
932 
933 	if (expiry_cache_is_inactive(pct))
934 		tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
935 }
936 
937 static inline void stop_process_timers(struct signal_struct *sig)
938 {
939 	struct posix_cputimers *pct = &sig->posix_cputimers;
940 
941 	/* Turn off the active flag. This is done without locking. */
942 	WRITE_ONCE(pct->timers_active, false);
943 	tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
944 }
945 
946 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
947 			     u64 *expires, u64 cur_time, int signo)
948 {
949 	if (!it->expires)
950 		return;
951 
952 	if (cur_time >= it->expires) {
953 		if (it->incr)
954 			it->expires += it->incr;
955 		else
956 			it->expires = 0;
957 
958 		trace_itimer_expire(signo == SIGPROF ?
959 				    ITIMER_PROF : ITIMER_VIRTUAL,
960 				    task_tgid(tsk), cur_time);
961 		send_signal_locked(signo, SEND_SIG_PRIV, tsk, PIDTYPE_TGID);
962 	}
963 
964 	if (it->expires && it->expires < *expires)
965 		*expires = it->expires;
966 }
967 
968 /*
969  * Check for any per-thread CPU timers that have fired and move them
970  * off the tsk->*_timers list onto the firing list.  Per-thread timers
971  * have already been taken off.
972  */
973 static void check_process_timers(struct task_struct *tsk,
974 				 struct list_head *firing)
975 {
976 	struct signal_struct *const sig = tsk->signal;
977 	struct posix_cputimers *pct = &sig->posix_cputimers;
978 	u64 samples[CPUCLOCK_MAX];
979 	unsigned long soft;
980 
981 	/*
982 	 * If there are no active process wide timers (POSIX 1.b, itimers,
983 	 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
984 	 * processing when there is already another task handling them.
985 	 */
986 	if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
987 		return;
988 
989 	/*
990 	 * Signify that a thread is checking for process timers.
991 	 * Write access to this field is protected by the sighand lock.
992 	 */
993 	pct->expiry_active = true;
994 
995 	/*
996 	 * Collect the current process totals. Group accounting is active
997 	 * so the sample can be taken directly.
998 	 */
999 	proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
1000 	collect_posix_cputimers(pct, samples, firing);
1001 
1002 	/*
1003 	 * Check for the special case process timers.
1004 	 */
1005 	check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
1006 			 &pct->bases[CPUCLOCK_PROF].nextevt,
1007 			 samples[CPUCLOCK_PROF], SIGPROF);
1008 	check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
1009 			 &pct->bases[CPUCLOCK_VIRT].nextevt,
1010 			 samples[CPUCLOCK_VIRT], SIGVTALRM);
1011 
1012 	soft = task_rlimit(tsk, RLIMIT_CPU);
1013 	if (soft != RLIM_INFINITY) {
1014 		/* RLIMIT_CPU is in seconds. Samples are nanoseconds */
1015 		unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
1016 		u64 ptime = samples[CPUCLOCK_PROF];
1017 		u64 softns = (u64)soft * NSEC_PER_SEC;
1018 		u64 hardns = (u64)hard * NSEC_PER_SEC;
1019 
1020 		/* At the hard limit, send SIGKILL. No further action. */
1021 		if (hard != RLIM_INFINITY &&
1022 		    check_rlimit(ptime, hardns, SIGKILL, false, true))
1023 			return;
1024 
1025 		/* At the soft limit, send a SIGXCPU every second */
1026 		if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
1027 			sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
1028 			softns += NSEC_PER_SEC;
1029 		}
1030 
1031 		/* Update the expiry cache */
1032 		if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
1033 			pct->bases[CPUCLOCK_PROF].nextevt = softns;
1034 	}
1035 
1036 	if (expiry_cache_is_inactive(pct))
1037 		stop_process_timers(sig);
1038 
1039 	pct->expiry_active = false;
1040 }
1041 
1042 /*
1043  * This is called from the signal code (via posixtimer_rearm)
1044  * when the last timer signal was delivered and we have to reload the timer.
1045  */
1046 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1047 {
1048 	clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
1049 	struct task_struct *p;
1050 	struct sighand_struct *sighand;
1051 	unsigned long flags;
1052 	u64 now;
1053 
1054 	rcu_read_lock();
1055 	p = cpu_timer_task_rcu(timer);
1056 	if (!p)
1057 		goto out;
1058 
1059 	/* Protect timer list r/w in arm_timer() */
1060 	sighand = lock_task_sighand(p, &flags);
1061 	if (unlikely(sighand == NULL))
1062 		goto out;
1063 
1064 	/*
1065 	 * Fetch the current sample and update the timer's expiry time.
1066 	 */
1067 	if (CPUCLOCK_PERTHREAD(timer->it_clock))
1068 		now = cpu_clock_sample(clkid, p);
1069 	else
1070 		now = cpu_clock_sample_group(clkid, p, true);
1071 
1072 	bump_cpu_timer(timer, now);
1073 
1074 	/*
1075 	 * Now re-arm for the new expiry time.
1076 	 */
1077 	arm_timer(timer, p);
1078 	unlock_task_sighand(p, &flags);
1079 out:
1080 	rcu_read_unlock();
1081 }
1082 
1083 /**
1084  * task_cputimers_expired - Check whether posix CPU timers are expired
1085  *
1086  * @samples:	Array of current samples for the CPUCLOCK clocks
1087  * @pct:	Pointer to a posix_cputimers container
1088  *
1089  * Returns true if any member of @samples is greater than the corresponding
1090  * member of @pct->bases[CLK].nextevt. False otherwise
1091  */
1092 static inline bool
1093 task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
1094 {
1095 	int i;
1096 
1097 	for (i = 0; i < CPUCLOCK_MAX; i++) {
1098 		if (samples[i] >= pct->bases[i].nextevt)
1099 			return true;
1100 	}
1101 	return false;
1102 }
1103 
1104 /**
1105  * fastpath_timer_check - POSIX CPU timers fast path.
1106  *
1107  * @tsk:	The task (thread) being checked.
1108  *
1109  * Check the task and thread group timers.  If both are zero (there are no
1110  * timers set) return false.  Otherwise snapshot the task and thread group
1111  * timers and compare them with the corresponding expiration times.  Return
1112  * true if a timer has expired, else return false.
1113  */
1114 static inline bool fastpath_timer_check(struct task_struct *tsk)
1115 {
1116 	struct posix_cputimers *pct = &tsk->posix_cputimers;
1117 	struct signal_struct *sig;
1118 
1119 	if (!expiry_cache_is_inactive(pct)) {
1120 		u64 samples[CPUCLOCK_MAX];
1121 
1122 		task_sample_cputime(tsk, samples);
1123 		if (task_cputimers_expired(samples, pct))
1124 			return true;
1125 	}
1126 
1127 	sig = tsk->signal;
1128 	pct = &sig->posix_cputimers;
1129 	/*
1130 	 * Check if thread group timers expired when timers are active and
1131 	 * no other thread in the group is already handling expiry for
1132 	 * thread group cputimers. These fields are read without the
1133 	 * sighand lock. However, this is fine because this is meant to be
1134 	 * a fastpath heuristic to determine whether we should try to
1135 	 * acquire the sighand lock to handle timer expiry.
1136 	 *
1137 	 * In the worst case scenario, if concurrently timers_active is set
1138 	 * or expiry_active is cleared, but the current thread doesn't see
1139 	 * the change yet, the timer checks are delayed until the next
1140 	 * thread in the group gets a scheduler interrupt to handle the
1141 	 * timer. This isn't an issue in practice because these types of
1142 	 * delays with signals actually getting sent are expected.
1143 	 */
1144 	if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
1145 		u64 samples[CPUCLOCK_MAX];
1146 
1147 		proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1148 					   samples);
1149 
1150 		if (task_cputimers_expired(samples, pct))
1151 			return true;
1152 	}
1153 
1154 	if (dl_task(tsk) && tsk->dl.dl_overrun)
1155 		return true;
1156 
1157 	return false;
1158 }
1159 
1160 static void handle_posix_cpu_timers(struct task_struct *tsk);
1161 
1162 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1163 static void posix_cpu_timers_work(struct callback_head *work)
1164 {
1165 	handle_posix_cpu_timers(current);
1166 }
1167 
1168 /*
1169  * Clear existing posix CPU timers task work.
1170  */
1171 void clear_posix_cputimers_work(struct task_struct *p)
1172 {
1173 	/*
1174 	 * A copied work entry from the old task is not meaningful, clear it.
1175 	 * N.B. init_task_work will not do this.
1176 	 */
1177 	memset(&p->posix_cputimers_work.work, 0,
1178 	       sizeof(p->posix_cputimers_work.work));
1179 	init_task_work(&p->posix_cputimers_work.work,
1180 		       posix_cpu_timers_work);
1181 	p->posix_cputimers_work.scheduled = false;
1182 }
1183 
1184 /*
1185  * Initialize posix CPU timers task work in init task. Out of line to
1186  * keep the callback static and to avoid header recursion hell.
1187  */
1188 void __init posix_cputimers_init_work(void)
1189 {
1190 	clear_posix_cputimers_work(current);
1191 }
1192 
1193 /*
1194  * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
1195  * in hard interrupt context or in task context with interrupts
1196  * disabled. Aside of that the writer/reader interaction is always in the
1197  * context of the current task, which means they are strict per CPU.
1198  */
1199 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1200 {
1201 	return tsk->posix_cputimers_work.scheduled;
1202 }
1203 
1204 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1205 {
1206 	if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
1207 		return;
1208 
1209 	/* Schedule task work to actually expire the timers */
1210 	tsk->posix_cputimers_work.scheduled = true;
1211 	task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
1212 }
1213 
1214 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1215 						unsigned long start)
1216 {
1217 	bool ret = true;
1218 
1219 	/*
1220 	 * On !RT kernels interrupts are disabled while collecting expired
1221 	 * timers, so no tick can happen and the fast path check can be
1222 	 * reenabled without further checks.
1223 	 */
1224 	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
1225 		tsk->posix_cputimers_work.scheduled = false;
1226 		return true;
1227 	}
1228 
1229 	/*
1230 	 * On RT enabled kernels ticks can happen while the expired timers
1231 	 * are collected under sighand lock. But any tick which observes
1232 	 * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
1233 	 * checks. So reenabling the tick work has do be done carefully:
1234 	 *
1235 	 * Disable interrupts and run the fast path check if jiffies have
1236 	 * advanced since the collecting of expired timers started. If
1237 	 * jiffies have not advanced or the fast path check did not find
1238 	 * newly expired timers, reenable the fast path check in the timer
1239 	 * interrupt. If there are newly expired timers, return false and
1240 	 * let the collection loop repeat.
1241 	 */
1242 	local_irq_disable();
1243 	if (start != jiffies && fastpath_timer_check(tsk))
1244 		ret = false;
1245 	else
1246 		tsk->posix_cputimers_work.scheduled = false;
1247 	local_irq_enable();
1248 
1249 	return ret;
1250 }
1251 #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1252 static inline void __run_posix_cpu_timers(struct task_struct *tsk)
1253 {
1254 	lockdep_posixtimer_enter();
1255 	handle_posix_cpu_timers(tsk);
1256 	lockdep_posixtimer_exit();
1257 }
1258 
1259 static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1260 {
1261 	return false;
1262 }
1263 
1264 static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
1265 						unsigned long start)
1266 {
1267 	return true;
1268 }
1269 #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
1270 
1271 static void handle_posix_cpu_timers(struct task_struct *tsk)
1272 {
1273 	struct k_itimer *timer, *next;
1274 	unsigned long flags, start;
1275 	LIST_HEAD(firing);
1276 
1277 	if (!lock_task_sighand(tsk, &flags))
1278 		return;
1279 
1280 	do {
1281 		/*
1282 		 * On RT locking sighand lock does not disable interrupts,
1283 		 * so this needs to be careful vs. ticks. Store the current
1284 		 * jiffies value.
1285 		 */
1286 		start = READ_ONCE(jiffies);
1287 		barrier();
1288 
1289 		/*
1290 		 * Here we take off tsk->signal->cpu_timers[N] and
1291 		 * tsk->cpu_timers[N] all the timers that are firing, and
1292 		 * put them on the firing list.
1293 		 */
1294 		check_thread_timers(tsk, &firing);
1295 
1296 		check_process_timers(tsk, &firing);
1297 
1298 		/*
1299 		 * The above timer checks have updated the expiry cache and
1300 		 * because nothing can have queued or modified timers after
1301 		 * sighand lock was taken above it is guaranteed to be
1302 		 * consistent. So the next timer interrupt fastpath check
1303 		 * will find valid data.
1304 		 *
1305 		 * If timer expiry runs in the timer interrupt context then
1306 		 * the loop is not relevant as timers will be directly
1307 		 * expired in interrupt context. The stub function below
1308 		 * returns always true which allows the compiler to
1309 		 * optimize the loop out.
1310 		 *
1311 		 * If timer expiry is deferred to task work context then
1312 		 * the following rules apply:
1313 		 *
1314 		 * - On !RT kernels no tick can have happened on this CPU
1315 		 *   after sighand lock was acquired because interrupts are
1316 		 *   disabled. So reenabling task work before dropping
1317 		 *   sighand lock and reenabling interrupts is race free.
1318 		 *
1319 		 * - On RT kernels ticks might have happened but the tick
1320 		 *   work ignored posix CPU timer handling because the
1321 		 *   CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
1322 		 *   must be done very carefully including a check whether
1323 		 *   ticks have happened since the start of the timer
1324 		 *   expiry checks. posix_cpu_timers_enable_work() takes
1325 		 *   care of that and eventually lets the expiry checks
1326 		 *   run again.
1327 		 */
1328 	} while (!posix_cpu_timers_enable_work(tsk, start));
1329 
1330 	/*
1331 	 * We must release sighand lock before taking any timer's lock.
1332 	 * There is a potential race with timer deletion here, as the
1333 	 * siglock now protects our private firing list.  We have set
1334 	 * the firing flag in each timer, so that a deletion attempt
1335 	 * that gets the timer lock before we do will give it up and
1336 	 * spin until we've taken care of that timer below.
1337 	 */
1338 	unlock_task_sighand(tsk, &flags);
1339 
1340 	/*
1341 	 * Now that all the timers on our list have the firing flag,
1342 	 * no one will touch their list entries but us.  We'll take
1343 	 * each timer's lock before clearing its firing flag, so no
1344 	 * timer call will interfere.
1345 	 */
1346 	list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1347 		int cpu_firing;
1348 
1349 		/*
1350 		 * spin_lock() is sufficient here even independent of the
1351 		 * expiry context. If expiry happens in hard interrupt
1352 		 * context it's obvious. For task work context it's safe
1353 		 * because all other operations on timer::it_lock happen in
1354 		 * task context (syscall or exit).
1355 		 */
1356 		spin_lock(&timer->it_lock);
1357 		list_del_init(&timer->it.cpu.elist);
1358 		cpu_firing = timer->it.cpu.firing;
1359 		timer->it.cpu.firing = 0;
1360 		/*
1361 		 * The firing flag is -1 if we collided with a reset
1362 		 * of the timer, which already reported this
1363 		 * almost-firing as an overrun.  So don't generate an event.
1364 		 */
1365 		if (likely(cpu_firing >= 0))
1366 			cpu_timer_fire(timer);
1367 		spin_unlock(&timer->it_lock);
1368 	}
1369 }
1370 
1371 /*
1372  * This is called from the timer interrupt handler.  The irq handler has
1373  * already updated our counts.  We need to check if any timers fire now.
1374  * Interrupts are disabled.
1375  */
1376 void run_posix_cpu_timers(void)
1377 {
1378 	struct task_struct *tsk = current;
1379 
1380 	lockdep_assert_irqs_disabled();
1381 
1382 	/*
1383 	 * If the actual expiry is deferred to task work context and the
1384 	 * work is already scheduled there is no point to do anything here.
1385 	 */
1386 	if (posix_cpu_timers_work_scheduled(tsk))
1387 		return;
1388 
1389 	/*
1390 	 * The fast path checks that there are no expired thread or thread
1391 	 * group timers.  If that's so, just return.
1392 	 */
1393 	if (!fastpath_timer_check(tsk))
1394 		return;
1395 
1396 	__run_posix_cpu_timers(tsk);
1397 }
1398 
1399 /*
1400  * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1401  * The tsk->sighand->siglock must be held by the caller.
1402  */
1403 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1404 			   u64 *newval, u64 *oldval)
1405 {
1406 	u64 now, *nextevt;
1407 
1408 	if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1409 		return;
1410 
1411 	nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
1412 	now = cpu_clock_sample_group(clkid, tsk, true);
1413 
1414 	if (oldval) {
1415 		/*
1416 		 * We are setting itimer. The *oldval is absolute and we update
1417 		 * it to be relative, *newval argument is relative and we update
1418 		 * it to be absolute.
1419 		 */
1420 		if (*oldval) {
1421 			if (*oldval <= now) {
1422 				/* Just about to fire. */
1423 				*oldval = TICK_NSEC;
1424 			} else {
1425 				*oldval -= now;
1426 			}
1427 		}
1428 
1429 		if (*newval)
1430 			*newval += now;
1431 	}
1432 
1433 	/*
1434 	 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1435 	 * expiry cache is also used by RLIMIT_CPU!.
1436 	 */
1437 	if (*newval < *nextevt)
1438 		*nextevt = *newval;
1439 
1440 	tick_dep_set_signal(tsk, TICK_DEP_BIT_POSIX_TIMER);
1441 }
1442 
1443 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1444 			    const struct timespec64 *rqtp)
1445 {
1446 	struct itimerspec64 it;
1447 	struct k_itimer timer;
1448 	u64 expires;
1449 	int error;
1450 
1451 	/*
1452 	 * Set up a temporary timer and then wait for it to go off.
1453 	 */
1454 	memset(&timer, 0, sizeof timer);
1455 	spin_lock_init(&timer.it_lock);
1456 	timer.it_clock = which_clock;
1457 	timer.it_overrun = -1;
1458 	error = posix_cpu_timer_create(&timer);
1459 	timer.it_process = current;
1460 
1461 	if (!error) {
1462 		static struct itimerspec64 zero_it;
1463 		struct restart_block *restart;
1464 
1465 		memset(&it, 0, sizeof(it));
1466 		it.it_value = *rqtp;
1467 
1468 		spin_lock_irq(&timer.it_lock);
1469 		error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1470 		if (error) {
1471 			spin_unlock_irq(&timer.it_lock);
1472 			return error;
1473 		}
1474 
1475 		while (!signal_pending(current)) {
1476 			if (!cpu_timer_getexpires(&timer.it.cpu)) {
1477 				/*
1478 				 * Our timer fired and was reset, below
1479 				 * deletion can not fail.
1480 				 */
1481 				posix_cpu_timer_del(&timer);
1482 				spin_unlock_irq(&timer.it_lock);
1483 				return 0;
1484 			}
1485 
1486 			/*
1487 			 * Block until cpu_timer_fire (or a signal) wakes us.
1488 			 */
1489 			__set_current_state(TASK_INTERRUPTIBLE);
1490 			spin_unlock_irq(&timer.it_lock);
1491 			schedule();
1492 			spin_lock_irq(&timer.it_lock);
1493 		}
1494 
1495 		/*
1496 		 * We were interrupted by a signal.
1497 		 */
1498 		expires = cpu_timer_getexpires(&timer.it.cpu);
1499 		error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1500 		if (!error) {
1501 			/*
1502 			 * Timer is now unarmed, deletion can not fail.
1503 			 */
1504 			posix_cpu_timer_del(&timer);
1505 		}
1506 		spin_unlock_irq(&timer.it_lock);
1507 
1508 		while (error == TIMER_RETRY) {
1509 			/*
1510 			 * We need to handle case when timer was or is in the
1511 			 * middle of firing. In other cases we already freed
1512 			 * resources.
1513 			 */
1514 			spin_lock_irq(&timer.it_lock);
1515 			error = posix_cpu_timer_del(&timer);
1516 			spin_unlock_irq(&timer.it_lock);
1517 		}
1518 
1519 		if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1520 			/*
1521 			 * It actually did fire already.
1522 			 */
1523 			return 0;
1524 		}
1525 
1526 		error = -ERESTART_RESTARTBLOCK;
1527 		/*
1528 		 * Report back to the user the time still remaining.
1529 		 */
1530 		restart = &current->restart_block;
1531 		restart->nanosleep.expires = expires;
1532 		if (restart->nanosleep.type != TT_NONE)
1533 			error = nanosleep_copyout(restart, &it.it_value);
1534 	}
1535 
1536 	return error;
1537 }
1538 
1539 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1540 
1541 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1542 			    const struct timespec64 *rqtp)
1543 {
1544 	struct restart_block *restart_block = &current->restart_block;
1545 	int error;
1546 
1547 	/*
1548 	 * Diagnose required errors first.
1549 	 */
1550 	if (CPUCLOCK_PERTHREAD(which_clock) &&
1551 	    (CPUCLOCK_PID(which_clock) == 0 ||
1552 	     CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1553 		return -EINVAL;
1554 
1555 	error = do_cpu_nanosleep(which_clock, flags, rqtp);
1556 
1557 	if (error == -ERESTART_RESTARTBLOCK) {
1558 
1559 		if (flags & TIMER_ABSTIME)
1560 			return -ERESTARTNOHAND;
1561 
1562 		restart_block->nanosleep.clockid = which_clock;
1563 		set_restart_fn(restart_block, posix_cpu_nsleep_restart);
1564 	}
1565 	return error;
1566 }
1567 
1568 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1569 {
1570 	clockid_t which_clock = restart_block->nanosleep.clockid;
1571 	struct timespec64 t;
1572 
1573 	t = ns_to_timespec64(restart_block->nanosleep.expires);
1574 
1575 	return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1576 }
1577 
1578 #define PROCESS_CLOCK	make_process_cpuclock(0, CPUCLOCK_SCHED)
1579 #define THREAD_CLOCK	make_thread_cpuclock(0, CPUCLOCK_SCHED)
1580 
1581 static int process_cpu_clock_getres(const clockid_t which_clock,
1582 				    struct timespec64 *tp)
1583 {
1584 	return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1585 }
1586 static int process_cpu_clock_get(const clockid_t which_clock,
1587 				 struct timespec64 *tp)
1588 {
1589 	return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1590 }
1591 static int process_cpu_timer_create(struct k_itimer *timer)
1592 {
1593 	timer->it_clock = PROCESS_CLOCK;
1594 	return posix_cpu_timer_create(timer);
1595 }
1596 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1597 			      const struct timespec64 *rqtp)
1598 {
1599 	return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1600 }
1601 static int thread_cpu_clock_getres(const clockid_t which_clock,
1602 				   struct timespec64 *tp)
1603 {
1604 	return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1605 }
1606 static int thread_cpu_clock_get(const clockid_t which_clock,
1607 				struct timespec64 *tp)
1608 {
1609 	return posix_cpu_clock_get(THREAD_CLOCK, tp);
1610 }
1611 static int thread_cpu_timer_create(struct k_itimer *timer)
1612 {
1613 	timer->it_clock = THREAD_CLOCK;
1614 	return posix_cpu_timer_create(timer);
1615 }
1616 
1617 const struct k_clock clock_posix_cpu = {
1618 	.clock_getres		= posix_cpu_clock_getres,
1619 	.clock_set		= posix_cpu_clock_set,
1620 	.clock_get_timespec	= posix_cpu_clock_get,
1621 	.timer_create		= posix_cpu_timer_create,
1622 	.nsleep			= posix_cpu_nsleep,
1623 	.timer_set		= posix_cpu_timer_set,
1624 	.timer_del		= posix_cpu_timer_del,
1625 	.timer_get		= posix_cpu_timer_get,
1626 	.timer_rearm		= posix_cpu_timer_rearm,
1627 };
1628 
1629 const struct k_clock clock_process = {
1630 	.clock_getres		= process_cpu_clock_getres,
1631 	.clock_get_timespec	= process_cpu_clock_get,
1632 	.timer_create		= process_cpu_timer_create,
1633 	.nsleep			= process_cpu_nsleep,
1634 };
1635 
1636 const struct k_clock clock_thread = {
1637 	.clock_getres		= thread_cpu_clock_getres,
1638 	.clock_get_timespec	= thread_cpu_clock_get,
1639 	.timer_create		= thread_cpu_timer_create,
1640 };
1641