xref: /openbmc/linux/kernel/kthread.c (revision 84fbfc33)
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 <uapi/linux/sched/types.h>
9 #include <linux/sched.h>
10 #include <linux/sched/task.h>
11 #include <linux/kthread.h>
12 #include <linux/completion.h>
13 #include <linux/err.h>
14 #include <linux/cpuset.h>
15 #include <linux/unistd.h>
16 #include <linux/file.h>
17 #include <linux/export.h>
18 #include <linux/mutex.h>
19 #include <linux/slab.h>
20 #include <linux/freezer.h>
21 #include <linux/ptrace.h>
22 #include <linux/uaccess.h>
23 #include <linux/cgroup.h>
24 #include <trace/events/sched.h>
25 
26 static DEFINE_SPINLOCK(kthread_create_lock);
27 static LIST_HEAD(kthread_create_list);
28 struct task_struct *kthreadd_task;
29 
30 struct kthread_create_info
31 {
32 	/* Information passed to kthread() from kthreadd. */
33 	int (*threadfn)(void *data);
34 	void *data;
35 	int node;
36 
37 	/* Result passed back to kthread_create() from kthreadd. */
38 	struct task_struct *result;
39 	struct completion *done;
40 
41 	struct list_head list;
42 };
43 
44 struct kthread {
45 	unsigned long flags;
46 	unsigned int cpu;
47 	void *data;
48 	struct completion parked;
49 	struct completion exited;
50 };
51 
52 enum KTHREAD_BITS {
53 	KTHREAD_IS_PER_CPU = 0,
54 	KTHREAD_SHOULD_STOP,
55 	KTHREAD_SHOULD_PARK,
56 	KTHREAD_IS_PARKED,
57 };
58 
59 static inline void set_kthread_struct(void *kthread)
60 {
61 	/*
62 	 * We abuse ->set_child_tid to avoid the new member and because it
63 	 * can't be wrongly copied by copy_process(). We also rely on fact
64 	 * that the caller can't exec, so PF_KTHREAD can't be cleared.
65 	 */
66 	current->set_child_tid = (__force void __user *)kthread;
67 }
68 
69 static inline struct kthread *to_kthread(struct task_struct *k)
70 {
71 	WARN_ON(!(k->flags & PF_KTHREAD));
72 	return (__force void *)k->set_child_tid;
73 }
74 
75 void free_kthread_struct(struct task_struct *k)
76 {
77 	/*
78 	 * Can be NULL if this kthread was created by kernel_thread()
79 	 * or if kmalloc() in kthread() failed.
80 	 */
81 	kfree(to_kthread(k));
82 }
83 
84 /**
85  * kthread_should_stop - should this kthread return now?
86  *
87  * When someone calls kthread_stop() on your kthread, it will be woken
88  * and this will return true.  You should then return, and your return
89  * value will be passed through to kthread_stop().
90  */
91 bool kthread_should_stop(void)
92 {
93 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
94 }
95 EXPORT_SYMBOL(kthread_should_stop);
96 
97 /**
98  * kthread_should_park - should this kthread park now?
99  *
100  * When someone calls kthread_park() on your kthread, it will be woken
101  * and this will return true.  You should then do the necessary
102  * cleanup and call kthread_parkme()
103  *
104  * Similar to kthread_should_stop(), but this keeps the thread alive
105  * and in a park position. kthread_unpark() "restarts" the thread and
106  * calls the thread function again.
107  */
108 bool kthread_should_park(void)
109 {
110 	return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
111 }
112 EXPORT_SYMBOL_GPL(kthread_should_park);
113 
114 /**
115  * kthread_freezable_should_stop - should this freezable kthread return now?
116  * @was_frozen: optional out parameter, indicates whether %current was frozen
117  *
118  * kthread_should_stop() for freezable kthreads, which will enter
119  * refrigerator if necessary.  This function is safe from kthread_stop() /
120  * freezer deadlock and freezable kthreads should use this function instead
121  * of calling try_to_freeze() directly.
122  */
123 bool kthread_freezable_should_stop(bool *was_frozen)
124 {
125 	bool frozen = false;
126 
127 	might_sleep();
128 
129 	if (unlikely(freezing(current)))
130 		frozen = __refrigerator(true);
131 
132 	if (was_frozen)
133 		*was_frozen = frozen;
134 
135 	return kthread_should_stop();
136 }
137 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
138 
139 /**
140  * kthread_data - return data value specified on kthread creation
141  * @task: kthread task in question
142  *
143  * Return the data value specified when kthread @task was created.
144  * The caller is responsible for ensuring the validity of @task when
145  * calling this function.
146  */
147 void *kthread_data(struct task_struct *task)
148 {
149 	return to_kthread(task)->data;
150 }
151 
152 /**
153  * kthread_probe_data - speculative version of kthread_data()
154  * @task: possible kthread task in question
155  *
156  * @task could be a kthread task.  Return the data value specified when it
157  * was created if accessible.  If @task isn't a kthread task or its data is
158  * inaccessible for any reason, %NULL is returned.  This function requires
159  * that @task itself is safe to dereference.
160  */
161 void *kthread_probe_data(struct task_struct *task)
162 {
163 	struct kthread *kthread = to_kthread(task);
164 	void *data = NULL;
165 
166 	probe_kernel_read(&data, &kthread->data, sizeof(data));
167 	return data;
168 }
169 
170 static void __kthread_parkme(struct kthread *self)
171 {
172 	__set_current_state(TASK_PARKED);
173 	while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
174 		if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
175 			complete(&self->parked);
176 		schedule();
177 		__set_current_state(TASK_PARKED);
178 	}
179 	clear_bit(KTHREAD_IS_PARKED, &self->flags);
180 	__set_current_state(TASK_RUNNING);
181 }
182 
183 void kthread_parkme(void)
184 {
185 	__kthread_parkme(to_kthread(current));
186 }
187 EXPORT_SYMBOL_GPL(kthread_parkme);
188 
189 static int kthread(void *_create)
190 {
191 	/* Copy data: it's on kthread's stack */
192 	struct kthread_create_info *create = _create;
193 	int (*threadfn)(void *data) = create->threadfn;
194 	void *data = create->data;
195 	struct completion *done;
196 	struct kthread *self;
197 	int ret;
198 
199 	self = kmalloc(sizeof(*self), GFP_KERNEL);
200 	set_kthread_struct(self);
201 
202 	/* If user was SIGKILLed, I release the structure. */
203 	done = xchg(&create->done, NULL);
204 	if (!done) {
205 		kfree(create);
206 		do_exit(-EINTR);
207 	}
208 
209 	if (!self) {
210 		create->result = ERR_PTR(-ENOMEM);
211 		complete(done);
212 		do_exit(-ENOMEM);
213 	}
214 
215 	self->flags = 0;
216 	self->data = data;
217 	init_completion(&self->exited);
218 	init_completion(&self->parked);
219 	current->vfork_done = &self->exited;
220 
221 	/* OK, tell user we're spawned, wait for stop or wakeup */
222 	__set_current_state(TASK_UNINTERRUPTIBLE);
223 	create->result = current;
224 	complete(done);
225 	schedule();
226 
227 	ret = -EINTR;
228 	if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
229 		cgroup_kthread_ready();
230 		__kthread_parkme(self);
231 		ret = threadfn(data);
232 	}
233 	do_exit(ret);
234 }
235 
236 /* called from do_fork() to get node information for about to be created task */
237 int tsk_fork_get_node(struct task_struct *tsk)
238 {
239 #ifdef CONFIG_NUMA
240 	if (tsk == kthreadd_task)
241 		return tsk->pref_node_fork;
242 #endif
243 	return NUMA_NO_NODE;
244 }
245 
246 static void create_kthread(struct kthread_create_info *create)
247 {
248 	int pid;
249 
250 #ifdef CONFIG_NUMA
251 	current->pref_node_fork = create->node;
252 #endif
253 	/* We want our own signal handler (we take no signals by default). */
254 	pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
255 	if (pid < 0) {
256 		/* If user was SIGKILLed, I release the structure. */
257 		struct completion *done = xchg(&create->done, NULL);
258 
259 		if (!done) {
260 			kfree(create);
261 			return;
262 		}
263 		create->result = ERR_PTR(pid);
264 		complete(done);
265 	}
266 }
267 
268 static __printf(4, 0)
269 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
270 						    void *data, int node,
271 						    const char namefmt[],
272 						    va_list args)
273 {
274 	DECLARE_COMPLETION_ONSTACK(done);
275 	struct task_struct *task;
276 	struct kthread_create_info *create = kmalloc(sizeof(*create),
277 						     GFP_KERNEL);
278 
279 	if (!create)
280 		return ERR_PTR(-ENOMEM);
281 	create->threadfn = threadfn;
282 	create->data = data;
283 	create->node = node;
284 	create->done = &done;
285 
286 	spin_lock(&kthread_create_lock);
287 	list_add_tail(&create->list, &kthread_create_list);
288 	spin_unlock(&kthread_create_lock);
289 
290 	wake_up_process(kthreadd_task);
291 	/*
292 	 * Wait for completion in killable state, for I might be chosen by
293 	 * the OOM killer while kthreadd is trying to allocate memory for
294 	 * new kernel thread.
295 	 */
296 	if (unlikely(wait_for_completion_killable(&done))) {
297 		/*
298 		 * If I was SIGKILLed before kthreadd (or new kernel thread)
299 		 * calls complete(), leave the cleanup of this structure to
300 		 * that thread.
301 		 */
302 		if (xchg(&create->done, NULL))
303 			return ERR_PTR(-EINTR);
304 		/*
305 		 * kthreadd (or new kernel thread) will call complete()
306 		 * shortly.
307 		 */
308 		wait_for_completion(&done);
309 	}
310 	task = create->result;
311 	if (!IS_ERR(task)) {
312 		static const struct sched_param param = { .sched_priority = 0 };
313 
314 		vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
315 		/*
316 		 * root may have changed our (kthreadd's) priority or CPU mask.
317 		 * The kernel thread should not inherit these properties.
318 		 */
319 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
320 		set_cpus_allowed_ptr(task, cpu_all_mask);
321 	}
322 	kfree(create);
323 	return task;
324 }
325 
326 /**
327  * kthread_create_on_node - create a kthread.
328  * @threadfn: the function to run until signal_pending(current).
329  * @data: data ptr for @threadfn.
330  * @node: task and thread structures for the thread are allocated on this node
331  * @namefmt: printf-style name for the thread.
332  *
333  * Description: This helper function creates and names a kernel
334  * thread.  The thread will be stopped: use wake_up_process() to start
335  * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
336  * is affine to all CPUs.
337  *
338  * If thread is going to be bound on a particular cpu, give its node
339  * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
340  * When woken, the thread will run @threadfn() with @data as its
341  * argument. @threadfn() can either call do_exit() directly if it is a
342  * standalone thread for which no one will call kthread_stop(), or
343  * return when 'kthread_should_stop()' is true (which means
344  * kthread_stop() has been called).  The return value should be zero
345  * or a negative error number; it will be passed to kthread_stop().
346  *
347  * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
348  */
349 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
350 					   void *data, int node,
351 					   const char namefmt[],
352 					   ...)
353 {
354 	struct task_struct *task;
355 	va_list args;
356 
357 	va_start(args, namefmt);
358 	task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
359 	va_end(args);
360 
361 	return task;
362 }
363 EXPORT_SYMBOL(kthread_create_on_node);
364 
365 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
366 {
367 	unsigned long flags;
368 
369 	if (!wait_task_inactive(p, state)) {
370 		WARN_ON(1);
371 		return;
372 	}
373 
374 	/* It's safe because the task is inactive. */
375 	raw_spin_lock_irqsave(&p->pi_lock, flags);
376 	do_set_cpus_allowed(p, mask);
377 	p->flags |= PF_NO_SETAFFINITY;
378 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
379 }
380 
381 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
382 {
383 	__kthread_bind_mask(p, cpumask_of(cpu), state);
384 }
385 
386 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
387 {
388 	__kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
389 }
390 
391 /**
392  * kthread_bind - bind a just-created kthread to a cpu.
393  * @p: thread created by kthread_create().
394  * @cpu: cpu (might not be online, must be possible) for @k to run on.
395  *
396  * Description: This function is equivalent to set_cpus_allowed(),
397  * except that @cpu doesn't need to be online, and the thread must be
398  * stopped (i.e., just returned from kthread_create()).
399  */
400 void kthread_bind(struct task_struct *p, unsigned int cpu)
401 {
402 	__kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
403 }
404 EXPORT_SYMBOL(kthread_bind);
405 
406 /**
407  * kthread_create_on_cpu - Create a cpu bound kthread
408  * @threadfn: the function to run until signal_pending(current).
409  * @data: data ptr for @threadfn.
410  * @cpu: The cpu on which the thread should be bound,
411  * @namefmt: printf-style name for the thread. Format is restricted
412  *	     to "name.*%u". Code fills in cpu number.
413  *
414  * Description: This helper function creates and names a kernel thread
415  * The thread will be woken and put into park mode.
416  */
417 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
418 					  void *data, unsigned int cpu,
419 					  const char *namefmt)
420 {
421 	struct task_struct *p;
422 
423 	p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
424 				   cpu);
425 	if (IS_ERR(p))
426 		return p;
427 	kthread_bind(p, cpu);
428 	/* CPU hotplug need to bind once again when unparking the thread. */
429 	set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
430 	to_kthread(p)->cpu = cpu;
431 	return p;
432 }
433 
434 /**
435  * kthread_unpark - unpark a thread created by kthread_create().
436  * @k:		thread created by kthread_create().
437  *
438  * Sets kthread_should_park() for @k to return false, wakes it, and
439  * waits for it to return. If the thread is marked percpu then its
440  * bound to the cpu again.
441  */
442 void kthread_unpark(struct task_struct *k)
443 {
444 	struct kthread *kthread = to_kthread(k);
445 
446 	clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
447 	/*
448 	 * We clear the IS_PARKED bit here as we don't wait
449 	 * until the task has left the park code. So if we'd
450 	 * park before that happens we'd see the IS_PARKED bit
451 	 * which might be about to be cleared.
452 	 */
453 	if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
454 		/*
455 		 * Newly created kthread was parked when the CPU was offline.
456 		 * The binding was lost and we need to set it again.
457 		 */
458 		if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
459 			__kthread_bind(k, kthread->cpu, TASK_PARKED);
460 		wake_up_state(k, TASK_PARKED);
461 	}
462 }
463 EXPORT_SYMBOL_GPL(kthread_unpark);
464 
465 /**
466  * kthread_park - park a thread created by kthread_create().
467  * @k: thread created by kthread_create().
468  *
469  * Sets kthread_should_park() for @k to return true, wakes it, and
470  * waits for it to return. This can also be called after kthread_create()
471  * instead of calling wake_up_process(): the thread will park without
472  * calling threadfn().
473  *
474  * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
475  * If called by the kthread itself just the park bit is set.
476  */
477 int kthread_park(struct task_struct *k)
478 {
479 	struct kthread *kthread = to_kthread(k);
480 
481 	if (WARN_ON(k->flags & PF_EXITING))
482 		return -ENOSYS;
483 
484 	if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
485 		set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
486 		if (k != current) {
487 			wake_up_process(k);
488 			wait_for_completion(&kthread->parked);
489 		}
490 	}
491 
492 	return 0;
493 }
494 EXPORT_SYMBOL_GPL(kthread_park);
495 
496 /**
497  * kthread_stop - stop a thread created by kthread_create().
498  * @k: thread created by kthread_create().
499  *
500  * Sets kthread_should_stop() for @k to return true, wakes it, and
501  * waits for it to exit. This can also be called after kthread_create()
502  * instead of calling wake_up_process(): the thread will exit without
503  * calling threadfn().
504  *
505  * If threadfn() may call do_exit() itself, the caller must ensure
506  * task_struct can't go away.
507  *
508  * Returns the result of threadfn(), or %-EINTR if wake_up_process()
509  * was never called.
510  */
511 int kthread_stop(struct task_struct *k)
512 {
513 	struct kthread *kthread;
514 	int ret;
515 
516 	trace_sched_kthread_stop(k);
517 
518 	get_task_struct(k);
519 	kthread = to_kthread(k);
520 	set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
521 	kthread_unpark(k);
522 	wake_up_process(k);
523 	wait_for_completion(&kthread->exited);
524 	ret = k->exit_code;
525 	put_task_struct(k);
526 
527 	trace_sched_kthread_stop_ret(ret);
528 	return ret;
529 }
530 EXPORT_SYMBOL(kthread_stop);
531 
532 int kthreadd(void *unused)
533 {
534 	struct task_struct *tsk = current;
535 
536 	/* Setup a clean context for our children to inherit. */
537 	set_task_comm(tsk, "kthreadd");
538 	ignore_signals(tsk);
539 	set_cpus_allowed_ptr(tsk, cpu_all_mask);
540 	set_mems_allowed(node_states[N_MEMORY]);
541 
542 	current->flags |= PF_NOFREEZE;
543 	cgroup_init_kthreadd();
544 
545 	for (;;) {
546 		set_current_state(TASK_INTERRUPTIBLE);
547 		if (list_empty(&kthread_create_list))
548 			schedule();
549 		__set_current_state(TASK_RUNNING);
550 
551 		spin_lock(&kthread_create_lock);
552 		while (!list_empty(&kthread_create_list)) {
553 			struct kthread_create_info *create;
554 
555 			create = list_entry(kthread_create_list.next,
556 					    struct kthread_create_info, list);
557 			list_del_init(&create->list);
558 			spin_unlock(&kthread_create_lock);
559 
560 			create_kthread(create);
561 
562 			spin_lock(&kthread_create_lock);
563 		}
564 		spin_unlock(&kthread_create_lock);
565 	}
566 
567 	return 0;
568 }
569 
570 void __kthread_init_worker(struct kthread_worker *worker,
571 				const char *name,
572 				struct lock_class_key *key)
573 {
574 	memset(worker, 0, sizeof(struct kthread_worker));
575 	spin_lock_init(&worker->lock);
576 	lockdep_set_class_and_name(&worker->lock, key, name);
577 	INIT_LIST_HEAD(&worker->work_list);
578 	INIT_LIST_HEAD(&worker->delayed_work_list);
579 }
580 EXPORT_SYMBOL_GPL(__kthread_init_worker);
581 
582 /**
583  * kthread_worker_fn - kthread function to process kthread_worker
584  * @worker_ptr: pointer to initialized kthread_worker
585  *
586  * This function implements the main cycle of kthread worker. It processes
587  * work_list until it is stopped with kthread_stop(). It sleeps when the queue
588  * is empty.
589  *
590  * The works are not allowed to keep any locks, disable preemption or interrupts
591  * when they finish. There is defined a safe point for freezing when one work
592  * finishes and before a new one is started.
593  *
594  * Also the works must not be handled by more than one worker at the same time,
595  * see also kthread_queue_work().
596  */
597 int kthread_worker_fn(void *worker_ptr)
598 {
599 	struct kthread_worker *worker = worker_ptr;
600 	struct kthread_work *work;
601 
602 	/*
603 	 * FIXME: Update the check and remove the assignment when all kthread
604 	 * worker users are created using kthread_create_worker*() functions.
605 	 */
606 	WARN_ON(worker->task && worker->task != current);
607 	worker->task = current;
608 
609 	if (worker->flags & KTW_FREEZABLE)
610 		set_freezable();
611 
612 repeat:
613 	set_current_state(TASK_INTERRUPTIBLE);	/* mb paired w/ kthread_stop */
614 
615 	if (kthread_should_stop()) {
616 		__set_current_state(TASK_RUNNING);
617 		spin_lock_irq(&worker->lock);
618 		worker->task = NULL;
619 		spin_unlock_irq(&worker->lock);
620 		return 0;
621 	}
622 
623 	work = NULL;
624 	spin_lock_irq(&worker->lock);
625 	if (!list_empty(&worker->work_list)) {
626 		work = list_first_entry(&worker->work_list,
627 					struct kthread_work, node);
628 		list_del_init(&work->node);
629 	}
630 	worker->current_work = work;
631 	spin_unlock_irq(&worker->lock);
632 
633 	if (work) {
634 		__set_current_state(TASK_RUNNING);
635 		work->func(work);
636 	} else if (!freezing(current))
637 		schedule();
638 
639 	try_to_freeze();
640 	cond_resched();
641 	goto repeat;
642 }
643 EXPORT_SYMBOL_GPL(kthread_worker_fn);
644 
645 static __printf(3, 0) struct kthread_worker *
646 __kthread_create_worker(int cpu, unsigned int flags,
647 			const char namefmt[], va_list args)
648 {
649 	struct kthread_worker *worker;
650 	struct task_struct *task;
651 	int node = -1;
652 
653 	worker = kzalloc(sizeof(*worker), GFP_KERNEL);
654 	if (!worker)
655 		return ERR_PTR(-ENOMEM);
656 
657 	kthread_init_worker(worker);
658 
659 	if (cpu >= 0)
660 		node = cpu_to_node(cpu);
661 
662 	task = __kthread_create_on_node(kthread_worker_fn, worker,
663 						node, namefmt, args);
664 	if (IS_ERR(task))
665 		goto fail_task;
666 
667 	if (cpu >= 0)
668 		kthread_bind(task, cpu);
669 
670 	worker->flags = flags;
671 	worker->task = task;
672 	wake_up_process(task);
673 	return worker;
674 
675 fail_task:
676 	kfree(worker);
677 	return ERR_CAST(task);
678 }
679 
680 /**
681  * kthread_create_worker - create a kthread worker
682  * @flags: flags modifying the default behavior of the worker
683  * @namefmt: printf-style name for the kthread worker (task).
684  *
685  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
686  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
687  * when the worker was SIGKILLed.
688  */
689 struct kthread_worker *
690 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
691 {
692 	struct kthread_worker *worker;
693 	va_list args;
694 
695 	va_start(args, namefmt);
696 	worker = __kthread_create_worker(-1, flags, namefmt, args);
697 	va_end(args);
698 
699 	return worker;
700 }
701 EXPORT_SYMBOL(kthread_create_worker);
702 
703 /**
704  * kthread_create_worker_on_cpu - create a kthread worker and bind it
705  *	it to a given CPU and the associated NUMA node.
706  * @cpu: CPU number
707  * @flags: flags modifying the default behavior of the worker
708  * @namefmt: printf-style name for the kthread worker (task).
709  *
710  * Use a valid CPU number if you want to bind the kthread worker
711  * to the given CPU and the associated NUMA node.
712  *
713  * A good practice is to add the cpu number also into the worker name.
714  * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
715  *
716  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
717  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
718  * when the worker was SIGKILLed.
719  */
720 struct kthread_worker *
721 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
722 			     const char namefmt[], ...)
723 {
724 	struct kthread_worker *worker;
725 	va_list args;
726 
727 	va_start(args, namefmt);
728 	worker = __kthread_create_worker(cpu, flags, namefmt, args);
729 	va_end(args);
730 
731 	return worker;
732 }
733 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
734 
735 /*
736  * Returns true when the work could not be queued at the moment.
737  * It happens when it is already pending in a worker list
738  * or when it is being cancelled.
739  */
740 static inline bool queuing_blocked(struct kthread_worker *worker,
741 				   struct kthread_work *work)
742 {
743 	lockdep_assert_held(&worker->lock);
744 
745 	return !list_empty(&work->node) || work->canceling;
746 }
747 
748 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
749 					     struct kthread_work *work)
750 {
751 	lockdep_assert_held(&worker->lock);
752 	WARN_ON_ONCE(!list_empty(&work->node));
753 	/* Do not use a work with >1 worker, see kthread_queue_work() */
754 	WARN_ON_ONCE(work->worker && work->worker != worker);
755 }
756 
757 /* insert @work before @pos in @worker */
758 static void kthread_insert_work(struct kthread_worker *worker,
759 				struct kthread_work *work,
760 				struct list_head *pos)
761 {
762 	kthread_insert_work_sanity_check(worker, work);
763 
764 	list_add_tail(&work->node, pos);
765 	work->worker = worker;
766 	if (!worker->current_work && likely(worker->task))
767 		wake_up_process(worker->task);
768 }
769 
770 /**
771  * kthread_queue_work - queue a kthread_work
772  * @worker: target kthread_worker
773  * @work: kthread_work to queue
774  *
775  * Queue @work to work processor @task for async execution.  @task
776  * must have been created with kthread_worker_create().  Returns %true
777  * if @work was successfully queued, %false if it was already pending.
778  *
779  * Reinitialize the work if it needs to be used by another worker.
780  * For example, when the worker was stopped and started again.
781  */
782 bool kthread_queue_work(struct kthread_worker *worker,
783 			struct kthread_work *work)
784 {
785 	bool ret = false;
786 	unsigned long flags;
787 
788 	spin_lock_irqsave(&worker->lock, flags);
789 	if (!queuing_blocked(worker, work)) {
790 		kthread_insert_work(worker, work, &worker->work_list);
791 		ret = true;
792 	}
793 	spin_unlock_irqrestore(&worker->lock, flags);
794 	return ret;
795 }
796 EXPORT_SYMBOL_GPL(kthread_queue_work);
797 
798 /**
799  * kthread_delayed_work_timer_fn - callback that queues the associated kthread
800  *	delayed work when the timer expires.
801  * @__data: pointer to the data associated with the timer
802  *
803  * The format of the function is defined by struct timer_list.
804  * It should have been called from irqsafe timer with irq already off.
805  */
806 void kthread_delayed_work_timer_fn(unsigned long __data)
807 {
808 	struct kthread_delayed_work *dwork =
809 		(struct kthread_delayed_work *)__data;
810 	struct kthread_work *work = &dwork->work;
811 	struct kthread_worker *worker = work->worker;
812 
813 	/*
814 	 * This might happen when a pending work is reinitialized.
815 	 * It means that it is used a wrong way.
816 	 */
817 	if (WARN_ON_ONCE(!worker))
818 		return;
819 
820 	spin_lock(&worker->lock);
821 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
822 	WARN_ON_ONCE(work->worker != worker);
823 
824 	/* Move the work from worker->delayed_work_list. */
825 	WARN_ON_ONCE(list_empty(&work->node));
826 	list_del_init(&work->node);
827 	kthread_insert_work(worker, work, &worker->work_list);
828 
829 	spin_unlock(&worker->lock);
830 }
831 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
832 
833 void __kthread_queue_delayed_work(struct kthread_worker *worker,
834 				  struct kthread_delayed_work *dwork,
835 				  unsigned long delay)
836 {
837 	struct timer_list *timer = &dwork->timer;
838 	struct kthread_work *work = &dwork->work;
839 
840 	WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn ||
841 		     timer->data != (unsigned long)dwork);
842 
843 	/*
844 	 * If @delay is 0, queue @dwork->work immediately.  This is for
845 	 * both optimization and correctness.  The earliest @timer can
846 	 * expire is on the closest next tick and delayed_work users depend
847 	 * on that there's no such delay when @delay is 0.
848 	 */
849 	if (!delay) {
850 		kthread_insert_work(worker, work, &worker->work_list);
851 		return;
852 	}
853 
854 	/* Be paranoid and try to detect possible races already now. */
855 	kthread_insert_work_sanity_check(worker, work);
856 
857 	list_add(&work->node, &worker->delayed_work_list);
858 	work->worker = worker;
859 	timer->expires = jiffies + delay;
860 	add_timer(timer);
861 }
862 
863 /**
864  * kthread_queue_delayed_work - queue the associated kthread work
865  *	after a delay.
866  * @worker: target kthread_worker
867  * @dwork: kthread_delayed_work to queue
868  * @delay: number of jiffies to wait before queuing
869  *
870  * If the work has not been pending it starts a timer that will queue
871  * the work after the given @delay. If @delay is zero, it queues the
872  * work immediately.
873  *
874  * Return: %false if the @work has already been pending. It means that
875  * either the timer was running or the work was queued. It returns %true
876  * otherwise.
877  */
878 bool kthread_queue_delayed_work(struct kthread_worker *worker,
879 				struct kthread_delayed_work *dwork,
880 				unsigned long delay)
881 {
882 	struct kthread_work *work = &dwork->work;
883 	unsigned long flags;
884 	bool ret = false;
885 
886 	spin_lock_irqsave(&worker->lock, flags);
887 
888 	if (!queuing_blocked(worker, work)) {
889 		__kthread_queue_delayed_work(worker, dwork, delay);
890 		ret = true;
891 	}
892 
893 	spin_unlock_irqrestore(&worker->lock, flags);
894 	return ret;
895 }
896 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
897 
898 struct kthread_flush_work {
899 	struct kthread_work	work;
900 	struct completion	done;
901 };
902 
903 static void kthread_flush_work_fn(struct kthread_work *work)
904 {
905 	struct kthread_flush_work *fwork =
906 		container_of(work, struct kthread_flush_work, work);
907 	complete(&fwork->done);
908 }
909 
910 /**
911  * kthread_flush_work - flush a kthread_work
912  * @work: work to flush
913  *
914  * If @work is queued or executing, wait for it to finish execution.
915  */
916 void kthread_flush_work(struct kthread_work *work)
917 {
918 	struct kthread_flush_work fwork = {
919 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
920 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
921 	};
922 	struct kthread_worker *worker;
923 	bool noop = false;
924 
925 	worker = work->worker;
926 	if (!worker)
927 		return;
928 
929 	spin_lock_irq(&worker->lock);
930 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
931 	WARN_ON_ONCE(work->worker != worker);
932 
933 	if (!list_empty(&work->node))
934 		kthread_insert_work(worker, &fwork.work, work->node.next);
935 	else if (worker->current_work == work)
936 		kthread_insert_work(worker, &fwork.work,
937 				    worker->work_list.next);
938 	else
939 		noop = true;
940 
941 	spin_unlock_irq(&worker->lock);
942 
943 	if (!noop)
944 		wait_for_completion(&fwork.done);
945 }
946 EXPORT_SYMBOL_GPL(kthread_flush_work);
947 
948 /*
949  * This function removes the work from the worker queue. Also it makes sure
950  * that it won't get queued later via the delayed work's timer.
951  *
952  * The work might still be in use when this function finishes. See the
953  * current_work proceed by the worker.
954  *
955  * Return: %true if @work was pending and successfully canceled,
956  *	%false if @work was not pending
957  */
958 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
959 				  unsigned long *flags)
960 {
961 	/* Try to cancel the timer if exists. */
962 	if (is_dwork) {
963 		struct kthread_delayed_work *dwork =
964 			container_of(work, struct kthread_delayed_work, work);
965 		struct kthread_worker *worker = work->worker;
966 
967 		/*
968 		 * del_timer_sync() must be called to make sure that the timer
969 		 * callback is not running. The lock must be temporary released
970 		 * to avoid a deadlock with the callback. In the meantime,
971 		 * any queuing is blocked by setting the canceling counter.
972 		 */
973 		work->canceling++;
974 		spin_unlock_irqrestore(&worker->lock, *flags);
975 		del_timer_sync(&dwork->timer);
976 		spin_lock_irqsave(&worker->lock, *flags);
977 		work->canceling--;
978 	}
979 
980 	/*
981 	 * Try to remove the work from a worker list. It might either
982 	 * be from worker->work_list or from worker->delayed_work_list.
983 	 */
984 	if (!list_empty(&work->node)) {
985 		list_del_init(&work->node);
986 		return true;
987 	}
988 
989 	return false;
990 }
991 
992 /**
993  * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
994  * @worker: kthread worker to use
995  * @dwork: kthread delayed work to queue
996  * @delay: number of jiffies to wait before queuing
997  *
998  * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
999  * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1000  * @work is guaranteed to be queued immediately.
1001  *
1002  * Return: %true if @dwork was pending and its timer was modified,
1003  * %false otherwise.
1004  *
1005  * A special case is when the work is being canceled in parallel.
1006  * It might be caused either by the real kthread_cancel_delayed_work_sync()
1007  * or yet another kthread_mod_delayed_work() call. We let the other command
1008  * win and return %false here. The caller is supposed to synchronize these
1009  * operations a reasonable way.
1010  *
1011  * This function is safe to call from any context including IRQ handler.
1012  * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1013  * for details.
1014  */
1015 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1016 			      struct kthread_delayed_work *dwork,
1017 			      unsigned long delay)
1018 {
1019 	struct kthread_work *work = &dwork->work;
1020 	unsigned long flags;
1021 	int ret = false;
1022 
1023 	spin_lock_irqsave(&worker->lock, flags);
1024 
1025 	/* Do not bother with canceling when never queued. */
1026 	if (!work->worker)
1027 		goto fast_queue;
1028 
1029 	/* Work must not be used with >1 worker, see kthread_queue_work() */
1030 	WARN_ON_ONCE(work->worker != worker);
1031 
1032 	/* Do not fight with another command that is canceling this work. */
1033 	if (work->canceling)
1034 		goto out;
1035 
1036 	ret = __kthread_cancel_work(work, true, &flags);
1037 fast_queue:
1038 	__kthread_queue_delayed_work(worker, dwork, delay);
1039 out:
1040 	spin_unlock_irqrestore(&worker->lock, flags);
1041 	return ret;
1042 }
1043 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1044 
1045 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1046 {
1047 	struct kthread_worker *worker = work->worker;
1048 	unsigned long flags;
1049 	int ret = false;
1050 
1051 	if (!worker)
1052 		goto out;
1053 
1054 	spin_lock_irqsave(&worker->lock, flags);
1055 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1056 	WARN_ON_ONCE(work->worker != worker);
1057 
1058 	ret = __kthread_cancel_work(work, is_dwork, &flags);
1059 
1060 	if (worker->current_work != work)
1061 		goto out_fast;
1062 
1063 	/*
1064 	 * The work is in progress and we need to wait with the lock released.
1065 	 * In the meantime, block any queuing by setting the canceling counter.
1066 	 */
1067 	work->canceling++;
1068 	spin_unlock_irqrestore(&worker->lock, flags);
1069 	kthread_flush_work(work);
1070 	spin_lock_irqsave(&worker->lock, flags);
1071 	work->canceling--;
1072 
1073 out_fast:
1074 	spin_unlock_irqrestore(&worker->lock, flags);
1075 out:
1076 	return ret;
1077 }
1078 
1079 /**
1080  * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1081  * @work: the kthread work to cancel
1082  *
1083  * Cancel @work and wait for its execution to finish.  This function
1084  * can be used even if the work re-queues itself. On return from this
1085  * function, @work is guaranteed to be not pending or executing on any CPU.
1086  *
1087  * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1088  * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1089  *
1090  * The caller must ensure that the worker on which @work was last
1091  * queued can't be destroyed before this function returns.
1092  *
1093  * Return: %true if @work was pending, %false otherwise.
1094  */
1095 bool kthread_cancel_work_sync(struct kthread_work *work)
1096 {
1097 	return __kthread_cancel_work_sync(work, false);
1098 }
1099 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1100 
1101 /**
1102  * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1103  *	wait for it to finish.
1104  * @dwork: the kthread delayed work to cancel
1105  *
1106  * This is kthread_cancel_work_sync() for delayed works.
1107  *
1108  * Return: %true if @dwork was pending, %false otherwise.
1109  */
1110 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1111 {
1112 	return __kthread_cancel_work_sync(&dwork->work, true);
1113 }
1114 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1115 
1116 /**
1117  * kthread_flush_worker - flush all current works on a kthread_worker
1118  * @worker: worker to flush
1119  *
1120  * Wait until all currently executing or pending works on @worker are
1121  * finished.
1122  */
1123 void kthread_flush_worker(struct kthread_worker *worker)
1124 {
1125 	struct kthread_flush_work fwork = {
1126 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1127 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1128 	};
1129 
1130 	kthread_queue_work(worker, &fwork.work);
1131 	wait_for_completion(&fwork.done);
1132 }
1133 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1134 
1135 /**
1136  * kthread_destroy_worker - destroy a kthread worker
1137  * @worker: worker to be destroyed
1138  *
1139  * Flush and destroy @worker.  The simple flush is enough because the kthread
1140  * worker API is used only in trivial scenarios.  There are no multi-step state
1141  * machines needed.
1142  */
1143 void kthread_destroy_worker(struct kthread_worker *worker)
1144 {
1145 	struct task_struct *task;
1146 
1147 	task = worker->task;
1148 	if (WARN_ON(!task))
1149 		return;
1150 
1151 	kthread_flush_worker(worker);
1152 	kthread_stop(task);
1153 	WARN_ON(!list_empty(&worker->work_list));
1154 	kfree(worker);
1155 }
1156 EXPORT_SYMBOL(kthread_destroy_worker);
1157