xref: /openbmc/linux/kernel/kthread.c (revision 8730046c)
1 /* Kernel thread helper functions.
2  *   Copyright (C) 2004 IBM Corporation, Rusty Russell.
3  *
4  * Creation is done via kthreadd, so that we get a clean environment
5  * even if we're invoked from userspace (think modprobe, hotplug cpu,
6  * etc.).
7  */
8 #include <linux/sched.h>
9 #include <linux/kthread.h>
10 #include <linux/completion.h>
11 #include <linux/err.h>
12 #include <linux/cpuset.h>
13 #include <linux/unistd.h>
14 #include <linux/file.h>
15 #include <linux/export.h>
16 #include <linux/mutex.h>
17 #include <linux/slab.h>
18 #include <linux/freezer.h>
19 #include <linux/ptrace.h>
20 #include <linux/uaccess.h>
21 #include <trace/events/sched.h>
22 
23 static DEFINE_SPINLOCK(kthread_create_lock);
24 static LIST_HEAD(kthread_create_list);
25 struct task_struct *kthreadd_task;
26 
27 struct kthread_create_info
28 {
29 	/* Information passed to kthread() from kthreadd. */
30 	int (*threadfn)(void *data);
31 	void *data;
32 	int node;
33 
34 	/* Result passed back to kthread_create() from kthreadd. */
35 	struct task_struct *result;
36 	struct completion *done;
37 
38 	struct list_head list;
39 };
40 
41 struct kthread {
42 	unsigned long flags;
43 	unsigned int cpu;
44 	void *data;
45 	struct completion parked;
46 	struct completion exited;
47 };
48 
49 enum KTHREAD_BITS {
50 	KTHREAD_IS_PER_CPU = 0,
51 	KTHREAD_SHOULD_STOP,
52 	KTHREAD_SHOULD_PARK,
53 	KTHREAD_IS_PARKED,
54 };
55 
56 static inline void set_kthread_struct(void *kthread)
57 {
58 	/*
59 	 * We abuse ->set_child_tid to avoid the new member and because it
60 	 * can't be wrongly copied by copy_process(). We also rely on fact
61 	 * that the caller can't exec, so PF_KTHREAD can't be cleared.
62 	 */
63 	current->set_child_tid = (__force void __user *)kthread;
64 }
65 
66 static inline struct kthread *to_kthread(struct task_struct *k)
67 {
68 	WARN_ON(!(k->flags & PF_KTHREAD));
69 	return (__force void *)k->set_child_tid;
70 }
71 
72 void free_kthread_struct(struct task_struct *k)
73 {
74 	/*
75 	 * Can be NULL if this kthread was created by kernel_thread()
76 	 * or if kmalloc() in kthread() failed.
77 	 */
78 	kfree(to_kthread(k));
79 }
80 
81 /**
82  * kthread_should_stop - should this kthread return now?
83  *
84  * When someone calls kthread_stop() on your kthread, it will be woken
85  * and this will return true.  You should then return, and your return
86  * value will be passed through to kthread_stop().
87  */
88 bool kthread_should_stop(void)
89 {
90 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
91 }
92 EXPORT_SYMBOL(kthread_should_stop);
93 
94 /**
95  * kthread_should_park - should this kthread park now?
96  *
97  * When someone calls kthread_park() on your kthread, it will be woken
98  * and this will return true.  You should then do the necessary
99  * cleanup and call kthread_parkme()
100  *
101  * Similar to kthread_should_stop(), but this keeps the thread alive
102  * and in a park position. kthread_unpark() "restarts" the thread and
103  * calls the thread function again.
104  */
105 bool kthread_should_park(void)
106 {
107 	return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
108 }
109 EXPORT_SYMBOL_GPL(kthread_should_park);
110 
111 /**
112  * kthread_freezable_should_stop - should this freezable kthread return now?
113  * @was_frozen: optional out parameter, indicates whether %current was frozen
114  *
115  * kthread_should_stop() for freezable kthreads, which will enter
116  * refrigerator if necessary.  This function is safe from kthread_stop() /
117  * freezer deadlock and freezable kthreads should use this function instead
118  * of calling try_to_freeze() directly.
119  */
120 bool kthread_freezable_should_stop(bool *was_frozen)
121 {
122 	bool frozen = false;
123 
124 	might_sleep();
125 
126 	if (unlikely(freezing(current)))
127 		frozen = __refrigerator(true);
128 
129 	if (was_frozen)
130 		*was_frozen = frozen;
131 
132 	return kthread_should_stop();
133 }
134 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
135 
136 /**
137  * kthread_data - return data value specified on kthread creation
138  * @task: kthread task in question
139  *
140  * Return the data value specified when kthread @task was created.
141  * The caller is responsible for ensuring the validity of @task when
142  * calling this function.
143  */
144 void *kthread_data(struct task_struct *task)
145 {
146 	return to_kthread(task)->data;
147 }
148 
149 /**
150  * kthread_probe_data - speculative version of kthread_data()
151  * @task: possible kthread task in question
152  *
153  * @task could be a kthread task.  Return the data value specified when it
154  * was created if accessible.  If @task isn't a kthread task or its data is
155  * inaccessible for any reason, %NULL is returned.  This function requires
156  * that @task itself is safe to dereference.
157  */
158 void *kthread_probe_data(struct task_struct *task)
159 {
160 	struct kthread *kthread = to_kthread(task);
161 	void *data = NULL;
162 
163 	probe_kernel_read(&data, &kthread->data, sizeof(data));
164 	return data;
165 }
166 
167 static void __kthread_parkme(struct kthread *self)
168 {
169 	__set_current_state(TASK_PARKED);
170 	while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
171 		if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
172 			complete(&self->parked);
173 		schedule();
174 		__set_current_state(TASK_PARKED);
175 	}
176 	clear_bit(KTHREAD_IS_PARKED, &self->flags);
177 	__set_current_state(TASK_RUNNING);
178 }
179 
180 void kthread_parkme(void)
181 {
182 	__kthread_parkme(to_kthread(current));
183 }
184 EXPORT_SYMBOL_GPL(kthread_parkme);
185 
186 static int kthread(void *_create)
187 {
188 	/* Copy data: it's on kthread's stack */
189 	struct kthread_create_info *create = _create;
190 	int (*threadfn)(void *data) = create->threadfn;
191 	void *data = create->data;
192 	struct completion *done;
193 	struct kthread *self;
194 	int ret;
195 
196 	self = kmalloc(sizeof(*self), GFP_KERNEL);
197 	set_kthread_struct(self);
198 
199 	/* If user was SIGKILLed, I release the structure. */
200 	done = xchg(&create->done, NULL);
201 	if (!done) {
202 		kfree(create);
203 		do_exit(-EINTR);
204 	}
205 
206 	if (!self) {
207 		create->result = ERR_PTR(-ENOMEM);
208 		complete(done);
209 		do_exit(-ENOMEM);
210 	}
211 
212 	self->flags = 0;
213 	self->data = data;
214 	init_completion(&self->exited);
215 	init_completion(&self->parked);
216 	current->vfork_done = &self->exited;
217 
218 	/* OK, tell user we're spawned, wait for stop or wakeup */
219 	__set_current_state(TASK_UNINTERRUPTIBLE);
220 	create->result = current;
221 	complete(done);
222 	schedule();
223 
224 	ret = -EINTR;
225 	if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
226 		__kthread_parkme(self);
227 		ret = threadfn(data);
228 	}
229 	do_exit(ret);
230 }
231 
232 /* called from do_fork() to get node information for about to be created task */
233 int tsk_fork_get_node(struct task_struct *tsk)
234 {
235 #ifdef CONFIG_NUMA
236 	if (tsk == kthreadd_task)
237 		return tsk->pref_node_fork;
238 #endif
239 	return NUMA_NO_NODE;
240 }
241 
242 static void create_kthread(struct kthread_create_info *create)
243 {
244 	int pid;
245 
246 #ifdef CONFIG_NUMA
247 	current->pref_node_fork = create->node;
248 #endif
249 	/* We want our own signal handler (we take no signals by default). */
250 	pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
251 	if (pid < 0) {
252 		/* If user was SIGKILLed, I release the structure. */
253 		struct completion *done = xchg(&create->done, NULL);
254 
255 		if (!done) {
256 			kfree(create);
257 			return;
258 		}
259 		create->result = ERR_PTR(pid);
260 		complete(done);
261 	}
262 }
263 
264 static __printf(4, 0)
265 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
266 						    void *data, int node,
267 						    const char namefmt[],
268 						    va_list args)
269 {
270 	DECLARE_COMPLETION_ONSTACK(done);
271 	struct task_struct *task;
272 	struct kthread_create_info *create = kmalloc(sizeof(*create),
273 						     GFP_KERNEL);
274 
275 	if (!create)
276 		return ERR_PTR(-ENOMEM);
277 	create->threadfn = threadfn;
278 	create->data = data;
279 	create->node = node;
280 	create->done = &done;
281 
282 	spin_lock(&kthread_create_lock);
283 	list_add_tail(&create->list, &kthread_create_list);
284 	spin_unlock(&kthread_create_lock);
285 
286 	wake_up_process(kthreadd_task);
287 	/*
288 	 * Wait for completion in killable state, for I might be chosen by
289 	 * the OOM killer while kthreadd is trying to allocate memory for
290 	 * new kernel thread.
291 	 */
292 	if (unlikely(wait_for_completion_killable(&done))) {
293 		/*
294 		 * If I was SIGKILLed before kthreadd (or new kernel thread)
295 		 * calls complete(), leave the cleanup of this structure to
296 		 * that thread.
297 		 */
298 		if (xchg(&create->done, NULL))
299 			return ERR_PTR(-EINTR);
300 		/*
301 		 * kthreadd (or new kernel thread) will call complete()
302 		 * shortly.
303 		 */
304 		wait_for_completion(&done);
305 	}
306 	task = create->result;
307 	if (!IS_ERR(task)) {
308 		static const struct sched_param param = { .sched_priority = 0 };
309 
310 		vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
311 		/*
312 		 * root may have changed our (kthreadd's) priority or CPU mask.
313 		 * The kernel thread should not inherit these properties.
314 		 */
315 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
316 		set_cpus_allowed_ptr(task, cpu_all_mask);
317 	}
318 	kfree(create);
319 	return task;
320 }
321 
322 /**
323  * kthread_create_on_node - create a kthread.
324  * @threadfn: the function to run until signal_pending(current).
325  * @data: data ptr for @threadfn.
326  * @node: task and thread structures for the thread are allocated on this node
327  * @namefmt: printf-style name for the thread.
328  *
329  * Description: This helper function creates and names a kernel
330  * thread.  The thread will be stopped: use wake_up_process() to start
331  * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
332  * is affine to all CPUs.
333  *
334  * If thread is going to be bound on a particular cpu, give its node
335  * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
336  * When woken, the thread will run @threadfn() with @data as its
337  * argument. @threadfn() can either call do_exit() directly if it is a
338  * standalone thread for which no one will call kthread_stop(), or
339  * return when 'kthread_should_stop()' is true (which means
340  * kthread_stop() has been called).  The return value should be zero
341  * or a negative error number; it will be passed to kthread_stop().
342  *
343  * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
344  */
345 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
346 					   void *data, int node,
347 					   const char namefmt[],
348 					   ...)
349 {
350 	struct task_struct *task;
351 	va_list args;
352 
353 	va_start(args, namefmt);
354 	task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
355 	va_end(args);
356 
357 	return task;
358 }
359 EXPORT_SYMBOL(kthread_create_on_node);
360 
361 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
362 {
363 	unsigned long flags;
364 
365 	if (!wait_task_inactive(p, state)) {
366 		WARN_ON(1);
367 		return;
368 	}
369 
370 	/* It's safe because the task is inactive. */
371 	raw_spin_lock_irqsave(&p->pi_lock, flags);
372 	do_set_cpus_allowed(p, mask);
373 	p->flags |= PF_NO_SETAFFINITY;
374 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
375 }
376 
377 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
378 {
379 	__kthread_bind_mask(p, cpumask_of(cpu), state);
380 }
381 
382 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
383 {
384 	__kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
385 }
386 
387 /**
388  * kthread_bind - bind a just-created kthread to a cpu.
389  * @p: thread created by kthread_create().
390  * @cpu: cpu (might not be online, must be possible) for @k to run on.
391  *
392  * Description: This function is equivalent to set_cpus_allowed(),
393  * except that @cpu doesn't need to be online, and the thread must be
394  * stopped (i.e., just returned from kthread_create()).
395  */
396 void kthread_bind(struct task_struct *p, unsigned int cpu)
397 {
398 	__kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
399 }
400 EXPORT_SYMBOL(kthread_bind);
401 
402 /**
403  * kthread_create_on_cpu - Create a cpu bound kthread
404  * @threadfn: the function to run until signal_pending(current).
405  * @data: data ptr for @threadfn.
406  * @cpu: The cpu on which the thread should be bound,
407  * @namefmt: printf-style name for the thread. Format is restricted
408  *	     to "name.*%u". Code fills in cpu number.
409  *
410  * Description: This helper function creates and names a kernel thread
411  * The thread will be woken and put into park mode.
412  */
413 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
414 					  void *data, unsigned int cpu,
415 					  const char *namefmt)
416 {
417 	struct task_struct *p;
418 
419 	p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
420 				   cpu);
421 	if (IS_ERR(p))
422 		return p;
423 	kthread_bind(p, cpu);
424 	/* CPU hotplug need to bind once again when unparking the thread. */
425 	set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
426 	to_kthread(p)->cpu = cpu;
427 	return p;
428 }
429 
430 /**
431  * kthread_unpark - unpark a thread created by kthread_create().
432  * @k:		thread created by kthread_create().
433  *
434  * Sets kthread_should_park() for @k to return false, wakes it, and
435  * waits for it to return. If the thread is marked percpu then its
436  * bound to the cpu again.
437  */
438 void kthread_unpark(struct task_struct *k)
439 {
440 	struct kthread *kthread = to_kthread(k);
441 
442 	clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
443 	/*
444 	 * We clear the IS_PARKED bit here as we don't wait
445 	 * until the task has left the park code. So if we'd
446 	 * park before that happens we'd see the IS_PARKED bit
447 	 * which might be about to be cleared.
448 	 */
449 	if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
450 		/*
451 		 * Newly created kthread was parked when the CPU was offline.
452 		 * The binding was lost and we need to set it again.
453 		 */
454 		if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
455 			__kthread_bind(k, kthread->cpu, TASK_PARKED);
456 		wake_up_state(k, TASK_PARKED);
457 	}
458 }
459 EXPORT_SYMBOL_GPL(kthread_unpark);
460 
461 /**
462  * kthread_park - park a thread created by kthread_create().
463  * @k: thread created by kthread_create().
464  *
465  * Sets kthread_should_park() for @k to return true, wakes it, and
466  * waits for it to return. This can also be called after kthread_create()
467  * instead of calling wake_up_process(): the thread will park without
468  * calling threadfn().
469  *
470  * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
471  * If called by the kthread itself just the park bit is set.
472  */
473 int kthread_park(struct task_struct *k)
474 {
475 	struct kthread *kthread = to_kthread(k);
476 
477 	if (WARN_ON(k->flags & PF_EXITING))
478 		return -ENOSYS;
479 
480 	if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
481 		set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
482 		if (k != current) {
483 			wake_up_process(k);
484 			wait_for_completion(&kthread->parked);
485 		}
486 	}
487 
488 	return 0;
489 }
490 EXPORT_SYMBOL_GPL(kthread_park);
491 
492 /**
493  * kthread_stop - stop a thread created by kthread_create().
494  * @k: thread created by kthread_create().
495  *
496  * Sets kthread_should_stop() for @k to return true, wakes it, and
497  * waits for it to exit. This can also be called after kthread_create()
498  * instead of calling wake_up_process(): the thread will exit without
499  * calling threadfn().
500  *
501  * If threadfn() may call do_exit() itself, the caller must ensure
502  * task_struct can't go away.
503  *
504  * Returns the result of threadfn(), or %-EINTR if wake_up_process()
505  * was never called.
506  */
507 int kthread_stop(struct task_struct *k)
508 {
509 	struct kthread *kthread;
510 	int ret;
511 
512 	trace_sched_kthread_stop(k);
513 
514 	get_task_struct(k);
515 	kthread = to_kthread(k);
516 	set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
517 	kthread_unpark(k);
518 	wake_up_process(k);
519 	wait_for_completion(&kthread->exited);
520 	ret = k->exit_code;
521 	put_task_struct(k);
522 
523 	trace_sched_kthread_stop_ret(ret);
524 	return ret;
525 }
526 EXPORT_SYMBOL(kthread_stop);
527 
528 int kthreadd(void *unused)
529 {
530 	struct task_struct *tsk = current;
531 
532 	/* Setup a clean context for our children to inherit. */
533 	set_task_comm(tsk, "kthreadd");
534 	ignore_signals(tsk);
535 	set_cpus_allowed_ptr(tsk, cpu_all_mask);
536 	set_mems_allowed(node_states[N_MEMORY]);
537 
538 	current->flags |= PF_NOFREEZE;
539 
540 	for (;;) {
541 		set_current_state(TASK_INTERRUPTIBLE);
542 		if (list_empty(&kthread_create_list))
543 			schedule();
544 		__set_current_state(TASK_RUNNING);
545 
546 		spin_lock(&kthread_create_lock);
547 		while (!list_empty(&kthread_create_list)) {
548 			struct kthread_create_info *create;
549 
550 			create = list_entry(kthread_create_list.next,
551 					    struct kthread_create_info, list);
552 			list_del_init(&create->list);
553 			spin_unlock(&kthread_create_lock);
554 
555 			create_kthread(create);
556 
557 			spin_lock(&kthread_create_lock);
558 		}
559 		spin_unlock(&kthread_create_lock);
560 	}
561 
562 	return 0;
563 }
564 
565 void __kthread_init_worker(struct kthread_worker *worker,
566 				const char *name,
567 				struct lock_class_key *key)
568 {
569 	memset(worker, 0, sizeof(struct kthread_worker));
570 	spin_lock_init(&worker->lock);
571 	lockdep_set_class_and_name(&worker->lock, key, name);
572 	INIT_LIST_HEAD(&worker->work_list);
573 	INIT_LIST_HEAD(&worker->delayed_work_list);
574 }
575 EXPORT_SYMBOL_GPL(__kthread_init_worker);
576 
577 /**
578  * kthread_worker_fn - kthread function to process kthread_worker
579  * @worker_ptr: pointer to initialized kthread_worker
580  *
581  * This function implements the main cycle of kthread worker. It processes
582  * work_list until it is stopped with kthread_stop(). It sleeps when the queue
583  * is empty.
584  *
585  * The works are not allowed to keep any locks, disable preemption or interrupts
586  * when they finish. There is defined a safe point for freezing when one work
587  * finishes and before a new one is started.
588  *
589  * Also the works must not be handled by more than one worker at the same time,
590  * see also kthread_queue_work().
591  */
592 int kthread_worker_fn(void *worker_ptr)
593 {
594 	struct kthread_worker *worker = worker_ptr;
595 	struct kthread_work *work;
596 
597 	/*
598 	 * FIXME: Update the check and remove the assignment when all kthread
599 	 * worker users are created using kthread_create_worker*() functions.
600 	 */
601 	WARN_ON(worker->task && worker->task != current);
602 	worker->task = current;
603 
604 	if (worker->flags & KTW_FREEZABLE)
605 		set_freezable();
606 
607 repeat:
608 	set_current_state(TASK_INTERRUPTIBLE);	/* mb paired w/ kthread_stop */
609 
610 	if (kthread_should_stop()) {
611 		__set_current_state(TASK_RUNNING);
612 		spin_lock_irq(&worker->lock);
613 		worker->task = NULL;
614 		spin_unlock_irq(&worker->lock);
615 		return 0;
616 	}
617 
618 	work = NULL;
619 	spin_lock_irq(&worker->lock);
620 	if (!list_empty(&worker->work_list)) {
621 		work = list_first_entry(&worker->work_list,
622 					struct kthread_work, node);
623 		list_del_init(&work->node);
624 	}
625 	worker->current_work = work;
626 	spin_unlock_irq(&worker->lock);
627 
628 	if (work) {
629 		__set_current_state(TASK_RUNNING);
630 		work->func(work);
631 	} else if (!freezing(current))
632 		schedule();
633 
634 	try_to_freeze();
635 	goto repeat;
636 }
637 EXPORT_SYMBOL_GPL(kthread_worker_fn);
638 
639 static __printf(3, 0) struct kthread_worker *
640 __kthread_create_worker(int cpu, unsigned int flags,
641 			const char namefmt[], va_list args)
642 {
643 	struct kthread_worker *worker;
644 	struct task_struct *task;
645 	int node = -1;
646 
647 	worker = kzalloc(sizeof(*worker), GFP_KERNEL);
648 	if (!worker)
649 		return ERR_PTR(-ENOMEM);
650 
651 	kthread_init_worker(worker);
652 
653 	if (cpu >= 0)
654 		node = cpu_to_node(cpu);
655 
656 	task = __kthread_create_on_node(kthread_worker_fn, worker,
657 						node, namefmt, args);
658 	if (IS_ERR(task))
659 		goto fail_task;
660 
661 	if (cpu >= 0)
662 		kthread_bind(task, cpu);
663 
664 	worker->flags = flags;
665 	worker->task = task;
666 	wake_up_process(task);
667 	return worker;
668 
669 fail_task:
670 	kfree(worker);
671 	return ERR_CAST(task);
672 }
673 
674 /**
675  * kthread_create_worker - create a kthread worker
676  * @flags: flags modifying the default behavior of the worker
677  * @namefmt: printf-style name for the kthread worker (task).
678  *
679  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
680  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
681  * when the worker was SIGKILLed.
682  */
683 struct kthread_worker *
684 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
685 {
686 	struct kthread_worker *worker;
687 	va_list args;
688 
689 	va_start(args, namefmt);
690 	worker = __kthread_create_worker(-1, flags, namefmt, args);
691 	va_end(args);
692 
693 	return worker;
694 }
695 EXPORT_SYMBOL(kthread_create_worker);
696 
697 /**
698  * kthread_create_worker_on_cpu - create a kthread worker and bind it
699  *	it to a given CPU and the associated NUMA node.
700  * @cpu: CPU number
701  * @flags: flags modifying the default behavior of the worker
702  * @namefmt: printf-style name for the kthread worker (task).
703  *
704  * Use a valid CPU number if you want to bind the kthread worker
705  * to the given CPU and the associated NUMA node.
706  *
707  * A good practice is to add the cpu number also into the worker name.
708  * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
709  *
710  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
711  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
712  * when the worker was SIGKILLed.
713  */
714 struct kthread_worker *
715 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
716 			     const char namefmt[], ...)
717 {
718 	struct kthread_worker *worker;
719 	va_list args;
720 
721 	va_start(args, namefmt);
722 	worker = __kthread_create_worker(cpu, flags, namefmt, args);
723 	va_end(args);
724 
725 	return worker;
726 }
727 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
728 
729 /*
730  * Returns true when the work could not be queued at the moment.
731  * It happens when it is already pending in a worker list
732  * or when it is being cancelled.
733  */
734 static inline bool queuing_blocked(struct kthread_worker *worker,
735 				   struct kthread_work *work)
736 {
737 	lockdep_assert_held(&worker->lock);
738 
739 	return !list_empty(&work->node) || work->canceling;
740 }
741 
742 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
743 					     struct kthread_work *work)
744 {
745 	lockdep_assert_held(&worker->lock);
746 	WARN_ON_ONCE(!list_empty(&work->node));
747 	/* Do not use a work with >1 worker, see kthread_queue_work() */
748 	WARN_ON_ONCE(work->worker && work->worker != worker);
749 }
750 
751 /* insert @work before @pos in @worker */
752 static void kthread_insert_work(struct kthread_worker *worker,
753 				struct kthread_work *work,
754 				struct list_head *pos)
755 {
756 	kthread_insert_work_sanity_check(worker, work);
757 
758 	list_add_tail(&work->node, pos);
759 	work->worker = worker;
760 	if (!worker->current_work && likely(worker->task))
761 		wake_up_process(worker->task);
762 }
763 
764 /**
765  * kthread_queue_work - queue a kthread_work
766  * @worker: target kthread_worker
767  * @work: kthread_work to queue
768  *
769  * Queue @work to work processor @task for async execution.  @task
770  * must have been created with kthread_worker_create().  Returns %true
771  * if @work was successfully queued, %false if it was already pending.
772  *
773  * Reinitialize the work if it needs to be used by another worker.
774  * For example, when the worker was stopped and started again.
775  */
776 bool kthread_queue_work(struct kthread_worker *worker,
777 			struct kthread_work *work)
778 {
779 	bool ret = false;
780 	unsigned long flags;
781 
782 	spin_lock_irqsave(&worker->lock, flags);
783 	if (!queuing_blocked(worker, work)) {
784 		kthread_insert_work(worker, work, &worker->work_list);
785 		ret = true;
786 	}
787 	spin_unlock_irqrestore(&worker->lock, flags);
788 	return ret;
789 }
790 EXPORT_SYMBOL_GPL(kthread_queue_work);
791 
792 /**
793  * kthread_delayed_work_timer_fn - callback that queues the associated kthread
794  *	delayed work when the timer expires.
795  * @__data: pointer to the data associated with the timer
796  *
797  * The format of the function is defined by struct timer_list.
798  * It should have been called from irqsafe timer with irq already off.
799  */
800 void kthread_delayed_work_timer_fn(unsigned long __data)
801 {
802 	struct kthread_delayed_work *dwork =
803 		(struct kthread_delayed_work *)__data;
804 	struct kthread_work *work = &dwork->work;
805 	struct kthread_worker *worker = work->worker;
806 
807 	/*
808 	 * This might happen when a pending work is reinitialized.
809 	 * It means that it is used a wrong way.
810 	 */
811 	if (WARN_ON_ONCE(!worker))
812 		return;
813 
814 	spin_lock(&worker->lock);
815 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
816 	WARN_ON_ONCE(work->worker != worker);
817 
818 	/* Move the work from worker->delayed_work_list. */
819 	WARN_ON_ONCE(list_empty(&work->node));
820 	list_del_init(&work->node);
821 	kthread_insert_work(worker, work, &worker->work_list);
822 
823 	spin_unlock(&worker->lock);
824 }
825 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
826 
827 void __kthread_queue_delayed_work(struct kthread_worker *worker,
828 				  struct kthread_delayed_work *dwork,
829 				  unsigned long delay)
830 {
831 	struct timer_list *timer = &dwork->timer;
832 	struct kthread_work *work = &dwork->work;
833 
834 	WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn ||
835 		     timer->data != (unsigned long)dwork);
836 
837 	/*
838 	 * If @delay is 0, queue @dwork->work immediately.  This is for
839 	 * both optimization and correctness.  The earliest @timer can
840 	 * expire is on the closest next tick and delayed_work users depend
841 	 * on that there's no such delay when @delay is 0.
842 	 */
843 	if (!delay) {
844 		kthread_insert_work(worker, work, &worker->work_list);
845 		return;
846 	}
847 
848 	/* Be paranoid and try to detect possible races already now. */
849 	kthread_insert_work_sanity_check(worker, work);
850 
851 	list_add(&work->node, &worker->delayed_work_list);
852 	work->worker = worker;
853 	timer_stats_timer_set_start_info(&dwork->timer);
854 	timer->expires = jiffies + delay;
855 	add_timer(timer);
856 }
857 
858 /**
859  * kthread_queue_delayed_work - queue the associated kthread work
860  *	after a delay.
861  * @worker: target kthread_worker
862  * @dwork: kthread_delayed_work to queue
863  * @delay: number of jiffies to wait before queuing
864  *
865  * If the work has not been pending it starts a timer that will queue
866  * the work after the given @delay. If @delay is zero, it queues the
867  * work immediately.
868  *
869  * Return: %false if the @work has already been pending. It means that
870  * either the timer was running or the work was queued. It returns %true
871  * otherwise.
872  */
873 bool kthread_queue_delayed_work(struct kthread_worker *worker,
874 				struct kthread_delayed_work *dwork,
875 				unsigned long delay)
876 {
877 	struct kthread_work *work = &dwork->work;
878 	unsigned long flags;
879 	bool ret = false;
880 
881 	spin_lock_irqsave(&worker->lock, flags);
882 
883 	if (!queuing_blocked(worker, work)) {
884 		__kthread_queue_delayed_work(worker, dwork, delay);
885 		ret = true;
886 	}
887 
888 	spin_unlock_irqrestore(&worker->lock, flags);
889 	return ret;
890 }
891 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
892 
893 struct kthread_flush_work {
894 	struct kthread_work	work;
895 	struct completion	done;
896 };
897 
898 static void kthread_flush_work_fn(struct kthread_work *work)
899 {
900 	struct kthread_flush_work *fwork =
901 		container_of(work, struct kthread_flush_work, work);
902 	complete(&fwork->done);
903 }
904 
905 /**
906  * kthread_flush_work - flush a kthread_work
907  * @work: work to flush
908  *
909  * If @work is queued or executing, wait for it to finish execution.
910  */
911 void kthread_flush_work(struct kthread_work *work)
912 {
913 	struct kthread_flush_work fwork = {
914 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
915 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
916 	};
917 	struct kthread_worker *worker;
918 	bool noop = false;
919 
920 	worker = work->worker;
921 	if (!worker)
922 		return;
923 
924 	spin_lock_irq(&worker->lock);
925 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
926 	WARN_ON_ONCE(work->worker != worker);
927 
928 	if (!list_empty(&work->node))
929 		kthread_insert_work(worker, &fwork.work, work->node.next);
930 	else if (worker->current_work == work)
931 		kthread_insert_work(worker, &fwork.work,
932 				    worker->work_list.next);
933 	else
934 		noop = true;
935 
936 	spin_unlock_irq(&worker->lock);
937 
938 	if (!noop)
939 		wait_for_completion(&fwork.done);
940 }
941 EXPORT_SYMBOL_GPL(kthread_flush_work);
942 
943 /*
944  * This function removes the work from the worker queue. Also it makes sure
945  * that it won't get queued later via the delayed work's timer.
946  *
947  * The work might still be in use when this function finishes. See the
948  * current_work proceed by the worker.
949  *
950  * Return: %true if @work was pending and successfully canceled,
951  *	%false if @work was not pending
952  */
953 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
954 				  unsigned long *flags)
955 {
956 	/* Try to cancel the timer if exists. */
957 	if (is_dwork) {
958 		struct kthread_delayed_work *dwork =
959 			container_of(work, struct kthread_delayed_work, work);
960 		struct kthread_worker *worker = work->worker;
961 
962 		/*
963 		 * del_timer_sync() must be called to make sure that the timer
964 		 * callback is not running. The lock must be temporary released
965 		 * to avoid a deadlock with the callback. In the meantime,
966 		 * any queuing is blocked by setting the canceling counter.
967 		 */
968 		work->canceling++;
969 		spin_unlock_irqrestore(&worker->lock, *flags);
970 		del_timer_sync(&dwork->timer);
971 		spin_lock_irqsave(&worker->lock, *flags);
972 		work->canceling--;
973 	}
974 
975 	/*
976 	 * Try to remove the work from a worker list. It might either
977 	 * be from worker->work_list or from worker->delayed_work_list.
978 	 */
979 	if (!list_empty(&work->node)) {
980 		list_del_init(&work->node);
981 		return true;
982 	}
983 
984 	return false;
985 }
986 
987 /**
988  * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
989  * @worker: kthread worker to use
990  * @dwork: kthread delayed work to queue
991  * @delay: number of jiffies to wait before queuing
992  *
993  * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
994  * modify @dwork's timer so that it expires after @delay. If @delay is zero,
995  * @work is guaranteed to be queued immediately.
996  *
997  * Return: %true if @dwork was pending and its timer was modified,
998  * %false otherwise.
999  *
1000  * A special case is when the work is being canceled in parallel.
1001  * It might be caused either by the real kthread_cancel_delayed_work_sync()
1002  * or yet another kthread_mod_delayed_work() call. We let the other command
1003  * win and return %false here. The caller is supposed to synchronize these
1004  * operations a reasonable way.
1005  *
1006  * This function is safe to call from any context including IRQ handler.
1007  * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1008  * for details.
1009  */
1010 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1011 			      struct kthread_delayed_work *dwork,
1012 			      unsigned long delay)
1013 {
1014 	struct kthread_work *work = &dwork->work;
1015 	unsigned long flags;
1016 	int ret = false;
1017 
1018 	spin_lock_irqsave(&worker->lock, flags);
1019 
1020 	/* Do not bother with canceling when never queued. */
1021 	if (!work->worker)
1022 		goto fast_queue;
1023 
1024 	/* Work must not be used with >1 worker, see kthread_queue_work() */
1025 	WARN_ON_ONCE(work->worker != worker);
1026 
1027 	/* Do not fight with another command that is canceling this work. */
1028 	if (work->canceling)
1029 		goto out;
1030 
1031 	ret = __kthread_cancel_work(work, true, &flags);
1032 fast_queue:
1033 	__kthread_queue_delayed_work(worker, dwork, delay);
1034 out:
1035 	spin_unlock_irqrestore(&worker->lock, flags);
1036 	return ret;
1037 }
1038 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1039 
1040 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1041 {
1042 	struct kthread_worker *worker = work->worker;
1043 	unsigned long flags;
1044 	int ret = false;
1045 
1046 	if (!worker)
1047 		goto out;
1048 
1049 	spin_lock_irqsave(&worker->lock, flags);
1050 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1051 	WARN_ON_ONCE(work->worker != worker);
1052 
1053 	ret = __kthread_cancel_work(work, is_dwork, &flags);
1054 
1055 	if (worker->current_work != work)
1056 		goto out_fast;
1057 
1058 	/*
1059 	 * The work is in progress and we need to wait with the lock released.
1060 	 * In the meantime, block any queuing by setting the canceling counter.
1061 	 */
1062 	work->canceling++;
1063 	spin_unlock_irqrestore(&worker->lock, flags);
1064 	kthread_flush_work(work);
1065 	spin_lock_irqsave(&worker->lock, flags);
1066 	work->canceling--;
1067 
1068 out_fast:
1069 	spin_unlock_irqrestore(&worker->lock, flags);
1070 out:
1071 	return ret;
1072 }
1073 
1074 /**
1075  * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1076  * @work: the kthread work to cancel
1077  *
1078  * Cancel @work and wait for its execution to finish.  This function
1079  * can be used even if the work re-queues itself. On return from this
1080  * function, @work is guaranteed to be not pending or executing on any CPU.
1081  *
1082  * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1083  * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1084  *
1085  * The caller must ensure that the worker on which @work was last
1086  * queued can't be destroyed before this function returns.
1087  *
1088  * Return: %true if @work was pending, %false otherwise.
1089  */
1090 bool kthread_cancel_work_sync(struct kthread_work *work)
1091 {
1092 	return __kthread_cancel_work_sync(work, false);
1093 }
1094 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1095 
1096 /**
1097  * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1098  *	wait for it to finish.
1099  * @dwork: the kthread delayed work to cancel
1100  *
1101  * This is kthread_cancel_work_sync() for delayed works.
1102  *
1103  * Return: %true if @dwork was pending, %false otherwise.
1104  */
1105 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1106 {
1107 	return __kthread_cancel_work_sync(&dwork->work, true);
1108 }
1109 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1110 
1111 /**
1112  * kthread_flush_worker - flush all current works on a kthread_worker
1113  * @worker: worker to flush
1114  *
1115  * Wait until all currently executing or pending works on @worker are
1116  * finished.
1117  */
1118 void kthread_flush_worker(struct kthread_worker *worker)
1119 {
1120 	struct kthread_flush_work fwork = {
1121 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1122 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1123 	};
1124 
1125 	kthread_queue_work(worker, &fwork.work);
1126 	wait_for_completion(&fwork.done);
1127 }
1128 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1129 
1130 /**
1131  * kthread_destroy_worker - destroy a kthread worker
1132  * @worker: worker to be destroyed
1133  *
1134  * Flush and destroy @worker.  The simple flush is enough because the kthread
1135  * worker API is used only in trivial scenarios.  There are no multi-step state
1136  * machines needed.
1137  */
1138 void kthread_destroy_worker(struct kthread_worker *worker)
1139 {
1140 	struct task_struct *task;
1141 
1142 	task = worker->task;
1143 	if (WARN_ON(!task))
1144 		return;
1145 
1146 	kthread_flush_worker(worker);
1147 	kthread_stop(task);
1148 	WARN_ON(!list_empty(&worker->work_list));
1149 	kfree(worker);
1150 }
1151 EXPORT_SYMBOL(kthread_destroy_worker);
1152