xref: /linux/kernel/kthread.c (revision 0750b8fcf313845b21c71344b4bea8ad7d3cee84)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Kernel thread helper functions.
3  *   Copyright (C) 2004 IBM Corporation, Rusty Russell.
4  *   Copyright (C) 2009 Red Hat, Inc.
5  *
6  * Creation is done via kthreadd, so that we get a clean environment
7  * even if we're invoked from userspace (think modprobe, hotplug cpu,
8  * etc.).
9  */
10 #include <uapi/linux/sched/types.h>
11 #include <linux/mm.h>
12 #include <linux/mmu_context.h>
13 #include <linux/sched.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/task.h>
16 #include <linux/kthread.h>
17 #include <linux/completion.h>
18 #include <linux/err.h>
19 #include <linux/cgroup.h>
20 #include <linux/cpuset.h>
21 #include <linux/unistd.h>
22 #include <linux/file.h>
23 #include <linux/export.h>
24 #include <linux/mutex.h>
25 #include <linux/slab.h>
26 #include <linux/freezer.h>
27 #include <linux/ptrace.h>
28 #include <linux/uaccess.h>
29 #include <linux/numa.h>
30 #include <linux/sched/isolation.h>
31 #include <trace/events/sched.h>
32 
33 
34 static DEFINE_SPINLOCK(kthread_create_lock);
35 static LIST_HEAD(kthread_create_list);
36 struct task_struct *kthreadd_task;
37 
38 struct kthread_create_info
39 {
40 	/* Information passed to kthread() from kthreadd. */
41 	int (*threadfn)(void *data);
42 	void *data;
43 	int node;
44 
45 	/* Result passed back to kthread_create() from kthreadd. */
46 	struct task_struct *result;
47 	struct completion *done;
48 
49 	struct list_head list;
50 };
51 
52 struct kthread {
53 	unsigned long flags;
54 	unsigned int cpu;
55 	int result;
56 	int (*threadfn)(void *);
57 	void *data;
58 	mm_segment_t oldfs;
59 	struct completion parked;
60 	struct completion exited;
61 #ifdef CONFIG_BLK_CGROUP
62 	struct cgroup_subsys_state *blkcg_css;
63 #endif
64 	/* To store the full name if task comm is truncated. */
65 	char *full_name;
66 };
67 
68 enum KTHREAD_BITS {
69 	KTHREAD_IS_PER_CPU = 0,
70 	KTHREAD_SHOULD_STOP,
71 	KTHREAD_SHOULD_PARK,
72 };
73 
74 static inline struct kthread *to_kthread(struct task_struct *k)
75 {
76 	WARN_ON(!(k->flags & PF_KTHREAD));
77 	return k->worker_private;
78 }
79 
80 /*
81  * Variant of to_kthread() that doesn't assume @p is a kthread.
82  *
83  * Per construction; when:
84  *
85  *   (p->flags & PF_KTHREAD) && p->worker_private
86  *
87  * the task is both a kthread and struct kthread is persistent. However
88  * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
89  * begin_new_exec()).
90  */
91 static inline struct kthread *__to_kthread(struct task_struct *p)
92 {
93 	void *kthread = p->worker_private;
94 	if (kthread && !(p->flags & PF_KTHREAD))
95 		kthread = NULL;
96 	return kthread;
97 }
98 
99 void get_kthread_comm(char *buf, size_t buf_size, struct task_struct *tsk)
100 {
101 	struct kthread *kthread = to_kthread(tsk);
102 
103 	if (!kthread || !kthread->full_name) {
104 		__get_task_comm(buf, buf_size, tsk);
105 		return;
106 	}
107 
108 	strscpy_pad(buf, kthread->full_name, buf_size);
109 }
110 
111 bool set_kthread_struct(struct task_struct *p)
112 {
113 	struct kthread *kthread;
114 
115 	if (WARN_ON_ONCE(to_kthread(p)))
116 		return false;
117 
118 	kthread = kzalloc(sizeof(*kthread), GFP_KERNEL);
119 	if (!kthread)
120 		return false;
121 
122 	init_completion(&kthread->exited);
123 	init_completion(&kthread->parked);
124 	p->vfork_done = &kthread->exited;
125 
126 	p->worker_private = kthread;
127 	return true;
128 }
129 
130 void free_kthread_struct(struct task_struct *k)
131 {
132 	struct kthread *kthread;
133 
134 	/*
135 	 * Can be NULL if kmalloc() in set_kthread_struct() failed.
136 	 */
137 	kthread = to_kthread(k);
138 	if (!kthread)
139 		return;
140 
141 #ifdef CONFIG_BLK_CGROUP
142 	WARN_ON_ONCE(kthread->blkcg_css);
143 #endif
144 	k->worker_private = NULL;
145 	kfree(kthread->full_name);
146 	kfree(kthread);
147 }
148 
149 /**
150  * kthread_should_stop - should this kthread return now?
151  *
152  * When someone calls kthread_stop() on your kthread, it will be woken
153  * and this will return true.  You should then return, and your return
154  * value will be passed through to kthread_stop().
155  */
156 bool kthread_should_stop(void)
157 {
158 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
159 }
160 EXPORT_SYMBOL(kthread_should_stop);
161 
162 bool __kthread_should_park(struct task_struct *k)
163 {
164 	return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
165 }
166 EXPORT_SYMBOL_GPL(__kthread_should_park);
167 
168 /**
169  * kthread_should_park - should this kthread park now?
170  *
171  * When someone calls kthread_park() on your kthread, it will be woken
172  * and this will return true.  You should then do the necessary
173  * cleanup and call kthread_parkme()
174  *
175  * Similar to kthread_should_stop(), but this keeps the thread alive
176  * and in a park position. kthread_unpark() "restarts" the thread and
177  * calls the thread function again.
178  */
179 bool kthread_should_park(void)
180 {
181 	return __kthread_should_park(current);
182 }
183 EXPORT_SYMBOL_GPL(kthread_should_park);
184 
185 /**
186  * kthread_freezable_should_stop - should this freezable kthread return now?
187  * @was_frozen: optional out parameter, indicates whether %current was frozen
188  *
189  * kthread_should_stop() for freezable kthreads, which will enter
190  * refrigerator if necessary.  This function is safe from kthread_stop() /
191  * freezer deadlock and freezable kthreads should use this function instead
192  * of calling try_to_freeze() directly.
193  */
194 bool kthread_freezable_should_stop(bool *was_frozen)
195 {
196 	bool frozen = false;
197 
198 	might_sleep();
199 
200 	if (unlikely(freezing(current)))
201 		frozen = __refrigerator(true);
202 
203 	if (was_frozen)
204 		*was_frozen = frozen;
205 
206 	return kthread_should_stop();
207 }
208 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
209 
210 /**
211  * kthread_func - return the function specified on kthread creation
212  * @task: kthread task in question
213  *
214  * Returns NULL if the task is not a kthread.
215  */
216 void *kthread_func(struct task_struct *task)
217 {
218 	struct kthread *kthread = __to_kthread(task);
219 	if (kthread)
220 		return kthread->threadfn;
221 	return NULL;
222 }
223 EXPORT_SYMBOL_GPL(kthread_func);
224 
225 /**
226  * kthread_data - return data value specified on kthread creation
227  * @task: kthread task in question
228  *
229  * Return the data value specified when kthread @task was created.
230  * The caller is responsible for ensuring the validity of @task when
231  * calling this function.
232  */
233 void *kthread_data(struct task_struct *task)
234 {
235 	return to_kthread(task)->data;
236 }
237 EXPORT_SYMBOL_GPL(kthread_data);
238 
239 /**
240  * kthread_probe_data - speculative version of kthread_data()
241  * @task: possible kthread task in question
242  *
243  * @task could be a kthread task.  Return the data value specified when it
244  * was created if accessible.  If @task isn't a kthread task or its data is
245  * inaccessible for any reason, %NULL is returned.  This function requires
246  * that @task itself is safe to dereference.
247  */
248 void *kthread_probe_data(struct task_struct *task)
249 {
250 	struct kthread *kthread = __to_kthread(task);
251 	void *data = NULL;
252 
253 	if (kthread)
254 		copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
255 	return data;
256 }
257 
258 static void __kthread_parkme(struct kthread *self)
259 {
260 	for (;;) {
261 		/*
262 		 * TASK_PARKED is a special state; we must serialize against
263 		 * possible pending wakeups to avoid store-store collisions on
264 		 * task->state.
265 		 *
266 		 * Such a collision might possibly result in the task state
267 		 * changin from TASK_PARKED and us failing the
268 		 * wait_task_inactive() in kthread_park().
269 		 */
270 		set_special_state(TASK_PARKED);
271 		if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
272 			break;
273 
274 		/*
275 		 * Thread is going to call schedule(), do not preempt it,
276 		 * or the caller of kthread_park() may spend more time in
277 		 * wait_task_inactive().
278 		 */
279 		preempt_disable();
280 		complete(&self->parked);
281 		schedule_preempt_disabled();
282 		preempt_enable();
283 	}
284 	__set_current_state(TASK_RUNNING);
285 }
286 
287 void kthread_parkme(void)
288 {
289 	__kthread_parkme(to_kthread(current));
290 }
291 EXPORT_SYMBOL_GPL(kthread_parkme);
292 
293 /**
294  * kthread_exit - Cause the current kthread return @result to kthread_stop().
295  * @result: The integer value to return to kthread_stop().
296  *
297  * While kthread_exit can be called directly, it exists so that
298  * functions which do some additional work in non-modular code such as
299  * module_put_and_kthread_exit can be implemented.
300  *
301  * Does not return.
302  */
303 void __noreturn kthread_exit(long result)
304 {
305 	struct kthread *kthread = to_kthread(current);
306 	kthread->result = result;
307 	do_exit(0);
308 }
309 
310 /**
311  * kthread_complete_and_exit - Exit the current kthread.
312  * @comp: Completion to complete
313  * @code: The integer value to return to kthread_stop().
314  *
315  * If present complete @comp and the reuturn code to kthread_stop().
316  *
317  * A kernel thread whose module may be removed after the completion of
318  * @comp can use this function exit safely.
319  *
320  * Does not return.
321  */
322 void __noreturn kthread_complete_and_exit(struct completion *comp, long code)
323 {
324 	if (comp)
325 		complete(comp);
326 
327 	kthread_exit(code);
328 }
329 EXPORT_SYMBOL(kthread_complete_and_exit);
330 
331 static int kthread(void *_create)
332 {
333 	static const struct sched_param param = { .sched_priority = 0 };
334 	/* Copy data: it's on kthread's stack */
335 	struct kthread_create_info *create = _create;
336 	int (*threadfn)(void *data) = create->threadfn;
337 	void *data = create->data;
338 	struct completion *done;
339 	struct kthread *self;
340 	int ret;
341 
342 	self = to_kthread(current);
343 
344 	/* If user was SIGKILLed, I release the structure. */
345 	done = xchg(&create->done, NULL);
346 	if (!done) {
347 		kfree(create);
348 		kthread_exit(-EINTR);
349 	}
350 
351 	self->threadfn = threadfn;
352 	self->data = data;
353 
354 	/*
355 	 * The new thread inherited kthreadd's priority and CPU mask. Reset
356 	 * back to default in case they have been changed.
357 	 */
358 	sched_setscheduler_nocheck(current, SCHED_NORMAL, &param);
359 	set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_KTHREAD));
360 
361 	/* OK, tell user we're spawned, wait for stop or wakeup */
362 	__set_current_state(TASK_UNINTERRUPTIBLE);
363 	create->result = current;
364 	/*
365 	 * Thread is going to call schedule(), do not preempt it,
366 	 * or the creator may spend more time in wait_task_inactive().
367 	 */
368 	preempt_disable();
369 	complete(done);
370 	schedule_preempt_disabled();
371 	preempt_enable();
372 
373 	ret = -EINTR;
374 	if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
375 		cgroup_kthread_ready();
376 		__kthread_parkme(self);
377 		ret = threadfn(data);
378 	}
379 	kthread_exit(ret);
380 }
381 
382 /* called from kernel_clone() to get node information for about to be created task */
383 int tsk_fork_get_node(struct task_struct *tsk)
384 {
385 #ifdef CONFIG_NUMA
386 	if (tsk == kthreadd_task)
387 		return tsk->pref_node_fork;
388 #endif
389 	return NUMA_NO_NODE;
390 }
391 
392 static void create_kthread(struct kthread_create_info *create)
393 {
394 	int pid;
395 
396 #ifdef CONFIG_NUMA
397 	current->pref_node_fork = create->node;
398 #endif
399 	/* We want our own signal handler (we take no signals by default). */
400 	pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
401 	if (pid < 0) {
402 		/* If user was SIGKILLed, I release the structure. */
403 		struct completion *done = xchg(&create->done, NULL);
404 
405 		if (!done) {
406 			kfree(create);
407 			return;
408 		}
409 		create->result = ERR_PTR(pid);
410 		complete(done);
411 	}
412 }
413 
414 static __printf(4, 0)
415 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
416 						    void *data, int node,
417 						    const char namefmt[],
418 						    va_list args)
419 {
420 	DECLARE_COMPLETION_ONSTACK(done);
421 	struct task_struct *task;
422 	struct kthread_create_info *create = kmalloc(sizeof(*create),
423 						     GFP_KERNEL);
424 
425 	if (!create)
426 		return ERR_PTR(-ENOMEM);
427 	create->threadfn = threadfn;
428 	create->data = data;
429 	create->node = node;
430 	create->done = &done;
431 
432 	spin_lock(&kthread_create_lock);
433 	list_add_tail(&create->list, &kthread_create_list);
434 	spin_unlock(&kthread_create_lock);
435 
436 	wake_up_process(kthreadd_task);
437 	/*
438 	 * Wait for completion in killable state, for I might be chosen by
439 	 * the OOM killer while kthreadd is trying to allocate memory for
440 	 * new kernel thread.
441 	 */
442 	if (unlikely(wait_for_completion_killable(&done))) {
443 		/*
444 		 * If I was SIGKILLed before kthreadd (or new kernel thread)
445 		 * calls complete(), leave the cleanup of this structure to
446 		 * that thread.
447 		 */
448 		if (xchg(&create->done, NULL))
449 			return ERR_PTR(-EINTR);
450 		/*
451 		 * kthreadd (or new kernel thread) will call complete()
452 		 * shortly.
453 		 */
454 		wait_for_completion(&done);
455 	}
456 	task = create->result;
457 	if (!IS_ERR(task)) {
458 		char name[TASK_COMM_LEN];
459 		va_list aq;
460 		int len;
461 
462 		/*
463 		 * task is already visible to other tasks, so updating
464 		 * COMM must be protected.
465 		 */
466 		va_copy(aq, args);
467 		len = vsnprintf(name, sizeof(name), namefmt, aq);
468 		va_end(aq);
469 		if (len >= TASK_COMM_LEN) {
470 			struct kthread *kthread = to_kthread(task);
471 
472 			/* leave it truncated when out of memory. */
473 			kthread->full_name = kvasprintf(GFP_KERNEL, namefmt, args);
474 		}
475 		set_task_comm(task, name);
476 	}
477 	kfree(create);
478 	return task;
479 }
480 
481 /**
482  * kthread_create_on_node - create a kthread.
483  * @threadfn: the function to run until signal_pending(current).
484  * @data: data ptr for @threadfn.
485  * @node: task and thread structures for the thread are allocated on this node
486  * @namefmt: printf-style name for the thread.
487  *
488  * Description: This helper function creates and names a kernel
489  * thread.  The thread will be stopped: use wake_up_process() to start
490  * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
491  * is affine to all CPUs.
492  *
493  * If thread is going to be bound on a particular cpu, give its node
494  * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
495  * When woken, the thread will run @threadfn() with @data as its
496  * argument. @threadfn() can either return directly if it is a
497  * standalone thread for which no one will call kthread_stop(), or
498  * return when 'kthread_should_stop()' is true (which means
499  * kthread_stop() has been called).  The return value should be zero
500  * or a negative error number; it will be passed to kthread_stop().
501  *
502  * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
503  */
504 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
505 					   void *data, int node,
506 					   const char namefmt[],
507 					   ...)
508 {
509 	struct task_struct *task;
510 	va_list args;
511 
512 	va_start(args, namefmt);
513 	task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
514 	va_end(args);
515 
516 	return task;
517 }
518 EXPORT_SYMBOL(kthread_create_on_node);
519 
520 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
521 {
522 	unsigned long flags;
523 
524 	if (!wait_task_inactive(p, state)) {
525 		WARN_ON(1);
526 		return;
527 	}
528 
529 	/* It's safe because the task is inactive. */
530 	raw_spin_lock_irqsave(&p->pi_lock, flags);
531 	do_set_cpus_allowed(p, mask);
532 	p->flags |= PF_NO_SETAFFINITY;
533 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
534 }
535 
536 static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
537 {
538 	__kthread_bind_mask(p, cpumask_of(cpu), state);
539 }
540 
541 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
542 {
543 	__kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
544 }
545 
546 /**
547  * kthread_bind - bind a just-created kthread to a cpu.
548  * @p: thread created by kthread_create().
549  * @cpu: cpu (might not be online, must be possible) for @k to run on.
550  *
551  * Description: This function is equivalent to set_cpus_allowed(),
552  * except that @cpu doesn't need to be online, and the thread must be
553  * stopped (i.e., just returned from kthread_create()).
554  */
555 void kthread_bind(struct task_struct *p, unsigned int cpu)
556 {
557 	__kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
558 }
559 EXPORT_SYMBOL(kthread_bind);
560 
561 /**
562  * kthread_create_on_cpu - Create a cpu bound kthread
563  * @threadfn: the function to run until signal_pending(current).
564  * @data: data ptr for @threadfn.
565  * @cpu: The cpu on which the thread should be bound,
566  * @namefmt: printf-style name for the thread. Format is restricted
567  *	     to "name.*%u". Code fills in cpu number.
568  *
569  * Description: This helper function creates and names a kernel thread
570  */
571 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
572 					  void *data, unsigned int cpu,
573 					  const char *namefmt)
574 {
575 	struct task_struct *p;
576 
577 	p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
578 				   cpu);
579 	if (IS_ERR(p))
580 		return p;
581 	kthread_bind(p, cpu);
582 	/* CPU hotplug need to bind once again when unparking the thread. */
583 	to_kthread(p)->cpu = cpu;
584 	return p;
585 }
586 EXPORT_SYMBOL(kthread_create_on_cpu);
587 
588 void kthread_set_per_cpu(struct task_struct *k, int cpu)
589 {
590 	struct kthread *kthread = to_kthread(k);
591 	if (!kthread)
592 		return;
593 
594 	WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
595 
596 	if (cpu < 0) {
597 		clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
598 		return;
599 	}
600 
601 	kthread->cpu = cpu;
602 	set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
603 }
604 
605 bool kthread_is_per_cpu(struct task_struct *p)
606 {
607 	struct kthread *kthread = __to_kthread(p);
608 	if (!kthread)
609 		return false;
610 
611 	return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
612 }
613 
614 /**
615  * kthread_unpark - unpark a thread created by kthread_create().
616  * @k:		thread created by kthread_create().
617  *
618  * Sets kthread_should_park() for @k to return false, wakes it, and
619  * waits for it to return. If the thread is marked percpu then its
620  * bound to the cpu again.
621  */
622 void kthread_unpark(struct task_struct *k)
623 {
624 	struct kthread *kthread = to_kthread(k);
625 
626 	/*
627 	 * Newly created kthread was parked when the CPU was offline.
628 	 * The binding was lost and we need to set it again.
629 	 */
630 	if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
631 		__kthread_bind(k, kthread->cpu, TASK_PARKED);
632 
633 	clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
634 	/*
635 	 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
636 	 */
637 	wake_up_state(k, TASK_PARKED);
638 }
639 EXPORT_SYMBOL_GPL(kthread_unpark);
640 
641 /**
642  * kthread_park - park a thread created by kthread_create().
643  * @k: thread created by kthread_create().
644  *
645  * Sets kthread_should_park() for @k to return true, wakes it, and
646  * waits for it to return. This can also be called after kthread_create()
647  * instead of calling wake_up_process(): the thread will park without
648  * calling threadfn().
649  *
650  * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
651  * If called by the kthread itself just the park bit is set.
652  */
653 int kthread_park(struct task_struct *k)
654 {
655 	struct kthread *kthread = to_kthread(k);
656 
657 	if (WARN_ON(k->flags & PF_EXITING))
658 		return -ENOSYS;
659 
660 	if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
661 		return -EBUSY;
662 
663 	set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
664 	if (k != current) {
665 		wake_up_process(k);
666 		/*
667 		 * Wait for __kthread_parkme() to complete(), this means we
668 		 * _will_ have TASK_PARKED and are about to call schedule().
669 		 */
670 		wait_for_completion(&kthread->parked);
671 		/*
672 		 * Now wait for that schedule() to complete and the task to
673 		 * get scheduled out.
674 		 */
675 		WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
676 	}
677 
678 	return 0;
679 }
680 EXPORT_SYMBOL_GPL(kthread_park);
681 
682 /**
683  * kthread_stop - stop a thread created by kthread_create().
684  * @k: thread created by kthread_create().
685  *
686  * Sets kthread_should_stop() for @k to return true, wakes it, and
687  * waits for it to exit. This can also be called after kthread_create()
688  * instead of calling wake_up_process(): the thread will exit without
689  * calling threadfn().
690  *
691  * If threadfn() may call kthread_exit() itself, the caller must ensure
692  * task_struct can't go away.
693  *
694  * Returns the result of threadfn(), or %-EINTR if wake_up_process()
695  * was never called.
696  */
697 int kthread_stop(struct task_struct *k)
698 {
699 	struct kthread *kthread;
700 	int ret;
701 
702 	trace_sched_kthread_stop(k);
703 
704 	get_task_struct(k);
705 	kthread = to_kthread(k);
706 	set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
707 	kthread_unpark(k);
708 	wake_up_process(k);
709 	wait_for_completion(&kthread->exited);
710 	ret = kthread->result;
711 	put_task_struct(k);
712 
713 	trace_sched_kthread_stop_ret(ret);
714 	return ret;
715 }
716 EXPORT_SYMBOL(kthread_stop);
717 
718 int kthreadd(void *unused)
719 {
720 	struct task_struct *tsk = current;
721 
722 	/* Setup a clean context for our children to inherit. */
723 	set_task_comm(tsk, "kthreadd");
724 	ignore_signals(tsk);
725 	set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
726 	set_mems_allowed(node_states[N_MEMORY]);
727 
728 	current->flags |= PF_NOFREEZE;
729 	cgroup_init_kthreadd();
730 
731 	for (;;) {
732 		set_current_state(TASK_INTERRUPTIBLE);
733 		if (list_empty(&kthread_create_list))
734 			schedule();
735 		__set_current_state(TASK_RUNNING);
736 
737 		spin_lock(&kthread_create_lock);
738 		while (!list_empty(&kthread_create_list)) {
739 			struct kthread_create_info *create;
740 
741 			create = list_entry(kthread_create_list.next,
742 					    struct kthread_create_info, list);
743 			list_del_init(&create->list);
744 			spin_unlock(&kthread_create_lock);
745 
746 			create_kthread(create);
747 
748 			spin_lock(&kthread_create_lock);
749 		}
750 		spin_unlock(&kthread_create_lock);
751 	}
752 
753 	return 0;
754 }
755 
756 void __kthread_init_worker(struct kthread_worker *worker,
757 				const char *name,
758 				struct lock_class_key *key)
759 {
760 	memset(worker, 0, sizeof(struct kthread_worker));
761 	raw_spin_lock_init(&worker->lock);
762 	lockdep_set_class_and_name(&worker->lock, key, name);
763 	INIT_LIST_HEAD(&worker->work_list);
764 	INIT_LIST_HEAD(&worker->delayed_work_list);
765 }
766 EXPORT_SYMBOL_GPL(__kthread_init_worker);
767 
768 /**
769  * kthread_worker_fn - kthread function to process kthread_worker
770  * @worker_ptr: pointer to initialized kthread_worker
771  *
772  * This function implements the main cycle of kthread worker. It processes
773  * work_list until it is stopped with kthread_stop(). It sleeps when the queue
774  * is empty.
775  *
776  * The works are not allowed to keep any locks, disable preemption or interrupts
777  * when they finish. There is defined a safe point for freezing when one work
778  * finishes and before a new one is started.
779  *
780  * Also the works must not be handled by more than one worker at the same time,
781  * see also kthread_queue_work().
782  */
783 int kthread_worker_fn(void *worker_ptr)
784 {
785 	struct kthread_worker *worker = worker_ptr;
786 	struct kthread_work *work;
787 
788 	/*
789 	 * FIXME: Update the check and remove the assignment when all kthread
790 	 * worker users are created using kthread_create_worker*() functions.
791 	 */
792 	WARN_ON(worker->task && worker->task != current);
793 	worker->task = current;
794 
795 	if (worker->flags & KTW_FREEZABLE)
796 		set_freezable();
797 
798 repeat:
799 	set_current_state(TASK_INTERRUPTIBLE);	/* mb paired w/ kthread_stop */
800 
801 	if (kthread_should_stop()) {
802 		__set_current_state(TASK_RUNNING);
803 		raw_spin_lock_irq(&worker->lock);
804 		worker->task = NULL;
805 		raw_spin_unlock_irq(&worker->lock);
806 		return 0;
807 	}
808 
809 	work = NULL;
810 	raw_spin_lock_irq(&worker->lock);
811 	if (!list_empty(&worker->work_list)) {
812 		work = list_first_entry(&worker->work_list,
813 					struct kthread_work, node);
814 		list_del_init(&work->node);
815 	}
816 	worker->current_work = work;
817 	raw_spin_unlock_irq(&worker->lock);
818 
819 	if (work) {
820 		kthread_work_func_t func = work->func;
821 		__set_current_state(TASK_RUNNING);
822 		trace_sched_kthread_work_execute_start(work);
823 		work->func(work);
824 		/*
825 		 * Avoid dereferencing work after this point.  The trace
826 		 * event only cares about the address.
827 		 */
828 		trace_sched_kthread_work_execute_end(work, func);
829 	} else if (!freezing(current))
830 		schedule();
831 
832 	try_to_freeze();
833 	cond_resched();
834 	goto repeat;
835 }
836 EXPORT_SYMBOL_GPL(kthread_worker_fn);
837 
838 static __printf(3, 0) struct kthread_worker *
839 __kthread_create_worker(int cpu, unsigned int flags,
840 			const char namefmt[], va_list args)
841 {
842 	struct kthread_worker *worker;
843 	struct task_struct *task;
844 	int node = NUMA_NO_NODE;
845 
846 	worker = kzalloc(sizeof(*worker), GFP_KERNEL);
847 	if (!worker)
848 		return ERR_PTR(-ENOMEM);
849 
850 	kthread_init_worker(worker);
851 
852 	if (cpu >= 0)
853 		node = cpu_to_node(cpu);
854 
855 	task = __kthread_create_on_node(kthread_worker_fn, worker,
856 						node, namefmt, args);
857 	if (IS_ERR(task))
858 		goto fail_task;
859 
860 	if (cpu >= 0)
861 		kthread_bind(task, cpu);
862 
863 	worker->flags = flags;
864 	worker->task = task;
865 	wake_up_process(task);
866 	return worker;
867 
868 fail_task:
869 	kfree(worker);
870 	return ERR_CAST(task);
871 }
872 
873 /**
874  * kthread_create_worker - create a kthread worker
875  * @flags: flags modifying the default behavior of the worker
876  * @namefmt: printf-style name for the kthread worker (task).
877  *
878  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
879  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
880  * when the worker was SIGKILLed.
881  */
882 struct kthread_worker *
883 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
884 {
885 	struct kthread_worker *worker;
886 	va_list args;
887 
888 	va_start(args, namefmt);
889 	worker = __kthread_create_worker(-1, flags, namefmt, args);
890 	va_end(args);
891 
892 	return worker;
893 }
894 EXPORT_SYMBOL(kthread_create_worker);
895 
896 /**
897  * kthread_create_worker_on_cpu - create a kthread worker and bind it
898  *	to a given CPU and the associated NUMA node.
899  * @cpu: CPU number
900  * @flags: flags modifying the default behavior of the worker
901  * @namefmt: printf-style name for the kthread worker (task).
902  *
903  * Use a valid CPU number if you want to bind the kthread worker
904  * to the given CPU and the associated NUMA node.
905  *
906  * A good practice is to add the cpu number also into the worker name.
907  * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
908  *
909  * CPU hotplug:
910  * The kthread worker API is simple and generic. It just provides a way
911  * to create, use, and destroy workers.
912  *
913  * It is up to the API user how to handle CPU hotplug. They have to decide
914  * how to handle pending work items, prevent queuing new ones, and
915  * restore the functionality when the CPU goes off and on. There are a
916  * few catches:
917  *
918  *    - CPU affinity gets lost when it is scheduled on an offline CPU.
919  *
920  *    - The worker might not exist when the CPU was off when the user
921  *      created the workers.
922  *
923  * Good practice is to implement two CPU hotplug callbacks and to
924  * destroy/create the worker when the CPU goes down/up.
925  *
926  * Return:
927  * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
928  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
929  * when the worker was SIGKILLed.
930  */
931 struct kthread_worker *
932 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
933 			     const char namefmt[], ...)
934 {
935 	struct kthread_worker *worker;
936 	va_list args;
937 
938 	va_start(args, namefmt);
939 	worker = __kthread_create_worker(cpu, flags, namefmt, args);
940 	va_end(args);
941 
942 	return worker;
943 }
944 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
945 
946 /*
947  * Returns true when the work could not be queued at the moment.
948  * It happens when it is already pending in a worker list
949  * or when it is being cancelled.
950  */
951 static inline bool queuing_blocked(struct kthread_worker *worker,
952 				   struct kthread_work *work)
953 {
954 	lockdep_assert_held(&worker->lock);
955 
956 	return !list_empty(&work->node) || work->canceling;
957 }
958 
959 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
960 					     struct kthread_work *work)
961 {
962 	lockdep_assert_held(&worker->lock);
963 	WARN_ON_ONCE(!list_empty(&work->node));
964 	/* Do not use a work with >1 worker, see kthread_queue_work() */
965 	WARN_ON_ONCE(work->worker && work->worker != worker);
966 }
967 
968 /* insert @work before @pos in @worker */
969 static void kthread_insert_work(struct kthread_worker *worker,
970 				struct kthread_work *work,
971 				struct list_head *pos)
972 {
973 	kthread_insert_work_sanity_check(worker, work);
974 
975 	trace_sched_kthread_work_queue_work(worker, work);
976 
977 	list_add_tail(&work->node, pos);
978 	work->worker = worker;
979 	if (!worker->current_work && likely(worker->task))
980 		wake_up_process(worker->task);
981 }
982 
983 /**
984  * kthread_queue_work - queue a kthread_work
985  * @worker: target kthread_worker
986  * @work: kthread_work to queue
987  *
988  * Queue @work to work processor @task for async execution.  @task
989  * must have been created with kthread_worker_create().  Returns %true
990  * if @work was successfully queued, %false if it was already pending.
991  *
992  * Reinitialize the work if it needs to be used by another worker.
993  * For example, when the worker was stopped and started again.
994  */
995 bool kthread_queue_work(struct kthread_worker *worker,
996 			struct kthread_work *work)
997 {
998 	bool ret = false;
999 	unsigned long flags;
1000 
1001 	raw_spin_lock_irqsave(&worker->lock, flags);
1002 	if (!queuing_blocked(worker, work)) {
1003 		kthread_insert_work(worker, work, &worker->work_list);
1004 		ret = true;
1005 	}
1006 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1007 	return ret;
1008 }
1009 EXPORT_SYMBOL_GPL(kthread_queue_work);
1010 
1011 /**
1012  * kthread_delayed_work_timer_fn - callback that queues the associated kthread
1013  *	delayed work when the timer expires.
1014  * @t: pointer to the expired timer
1015  *
1016  * The format of the function is defined by struct timer_list.
1017  * It should have been called from irqsafe timer with irq already off.
1018  */
1019 void kthread_delayed_work_timer_fn(struct timer_list *t)
1020 {
1021 	struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
1022 	struct kthread_work *work = &dwork->work;
1023 	struct kthread_worker *worker = work->worker;
1024 	unsigned long flags;
1025 
1026 	/*
1027 	 * This might happen when a pending work is reinitialized.
1028 	 * It means that it is used a wrong way.
1029 	 */
1030 	if (WARN_ON_ONCE(!worker))
1031 		return;
1032 
1033 	raw_spin_lock_irqsave(&worker->lock, flags);
1034 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1035 	WARN_ON_ONCE(work->worker != worker);
1036 
1037 	/* Move the work from worker->delayed_work_list. */
1038 	WARN_ON_ONCE(list_empty(&work->node));
1039 	list_del_init(&work->node);
1040 	if (!work->canceling)
1041 		kthread_insert_work(worker, work, &worker->work_list);
1042 
1043 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1044 }
1045 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
1046 
1047 static void __kthread_queue_delayed_work(struct kthread_worker *worker,
1048 					 struct kthread_delayed_work *dwork,
1049 					 unsigned long delay)
1050 {
1051 	struct timer_list *timer = &dwork->timer;
1052 	struct kthread_work *work = &dwork->work;
1053 
1054 	WARN_ON_FUNCTION_MISMATCH(timer->function,
1055 				  kthread_delayed_work_timer_fn);
1056 
1057 	/*
1058 	 * If @delay is 0, queue @dwork->work immediately.  This is for
1059 	 * both optimization and correctness.  The earliest @timer can
1060 	 * expire is on the closest next tick and delayed_work users depend
1061 	 * on that there's no such delay when @delay is 0.
1062 	 */
1063 	if (!delay) {
1064 		kthread_insert_work(worker, work, &worker->work_list);
1065 		return;
1066 	}
1067 
1068 	/* Be paranoid and try to detect possible races already now. */
1069 	kthread_insert_work_sanity_check(worker, work);
1070 
1071 	list_add(&work->node, &worker->delayed_work_list);
1072 	work->worker = worker;
1073 	timer->expires = jiffies + delay;
1074 	add_timer(timer);
1075 }
1076 
1077 /**
1078  * kthread_queue_delayed_work - queue the associated kthread work
1079  *	after a delay.
1080  * @worker: target kthread_worker
1081  * @dwork: kthread_delayed_work to queue
1082  * @delay: number of jiffies to wait before queuing
1083  *
1084  * If the work has not been pending it starts a timer that will queue
1085  * the work after the given @delay. If @delay is zero, it queues the
1086  * work immediately.
1087  *
1088  * Return: %false if the @work has already been pending. It means that
1089  * either the timer was running or the work was queued. It returns %true
1090  * otherwise.
1091  */
1092 bool kthread_queue_delayed_work(struct kthread_worker *worker,
1093 				struct kthread_delayed_work *dwork,
1094 				unsigned long delay)
1095 {
1096 	struct kthread_work *work = &dwork->work;
1097 	unsigned long flags;
1098 	bool ret = false;
1099 
1100 	raw_spin_lock_irqsave(&worker->lock, flags);
1101 
1102 	if (!queuing_blocked(worker, work)) {
1103 		__kthread_queue_delayed_work(worker, dwork, delay);
1104 		ret = true;
1105 	}
1106 
1107 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1108 	return ret;
1109 }
1110 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1111 
1112 struct kthread_flush_work {
1113 	struct kthread_work	work;
1114 	struct completion	done;
1115 };
1116 
1117 static void kthread_flush_work_fn(struct kthread_work *work)
1118 {
1119 	struct kthread_flush_work *fwork =
1120 		container_of(work, struct kthread_flush_work, work);
1121 	complete(&fwork->done);
1122 }
1123 
1124 /**
1125  * kthread_flush_work - flush a kthread_work
1126  * @work: work to flush
1127  *
1128  * If @work is queued or executing, wait for it to finish execution.
1129  */
1130 void kthread_flush_work(struct kthread_work *work)
1131 {
1132 	struct kthread_flush_work fwork = {
1133 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1134 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1135 	};
1136 	struct kthread_worker *worker;
1137 	bool noop = false;
1138 
1139 	worker = work->worker;
1140 	if (!worker)
1141 		return;
1142 
1143 	raw_spin_lock_irq(&worker->lock);
1144 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1145 	WARN_ON_ONCE(work->worker != worker);
1146 
1147 	if (!list_empty(&work->node))
1148 		kthread_insert_work(worker, &fwork.work, work->node.next);
1149 	else if (worker->current_work == work)
1150 		kthread_insert_work(worker, &fwork.work,
1151 				    worker->work_list.next);
1152 	else
1153 		noop = true;
1154 
1155 	raw_spin_unlock_irq(&worker->lock);
1156 
1157 	if (!noop)
1158 		wait_for_completion(&fwork.done);
1159 }
1160 EXPORT_SYMBOL_GPL(kthread_flush_work);
1161 
1162 /*
1163  * Make sure that the timer is neither set nor running and could
1164  * not manipulate the work list_head any longer.
1165  *
1166  * The function is called under worker->lock. The lock is temporary
1167  * released but the timer can't be set again in the meantime.
1168  */
1169 static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1170 					      unsigned long *flags)
1171 {
1172 	struct kthread_delayed_work *dwork =
1173 		container_of(work, struct kthread_delayed_work, work);
1174 	struct kthread_worker *worker = work->worker;
1175 
1176 	/*
1177 	 * del_timer_sync() must be called to make sure that the timer
1178 	 * callback is not running. The lock must be temporary released
1179 	 * to avoid a deadlock with the callback. In the meantime,
1180 	 * any queuing is blocked by setting the canceling counter.
1181 	 */
1182 	work->canceling++;
1183 	raw_spin_unlock_irqrestore(&worker->lock, *flags);
1184 	del_timer_sync(&dwork->timer);
1185 	raw_spin_lock_irqsave(&worker->lock, *flags);
1186 	work->canceling--;
1187 }
1188 
1189 /*
1190  * This function removes the work from the worker queue.
1191  *
1192  * It is called under worker->lock. The caller must make sure that
1193  * the timer used by delayed work is not running, e.g. by calling
1194  * kthread_cancel_delayed_work_timer().
1195  *
1196  * The work might still be in use when this function finishes. See the
1197  * current_work proceed by the worker.
1198  *
1199  * Return: %true if @work was pending and successfully canceled,
1200  *	%false if @work was not pending
1201  */
1202 static bool __kthread_cancel_work(struct kthread_work *work)
1203 {
1204 	/*
1205 	 * Try to remove the work from a worker list. It might either
1206 	 * be from worker->work_list or from worker->delayed_work_list.
1207 	 */
1208 	if (!list_empty(&work->node)) {
1209 		list_del_init(&work->node);
1210 		return true;
1211 	}
1212 
1213 	return false;
1214 }
1215 
1216 /**
1217  * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1218  * @worker: kthread worker to use
1219  * @dwork: kthread delayed work to queue
1220  * @delay: number of jiffies to wait before queuing
1221  *
1222  * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1223  * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1224  * @work is guaranteed to be queued immediately.
1225  *
1226  * Return: %false if @dwork was idle and queued, %true otherwise.
1227  *
1228  * A special case is when the work is being canceled in parallel.
1229  * It might be caused either by the real kthread_cancel_delayed_work_sync()
1230  * or yet another kthread_mod_delayed_work() call. We let the other command
1231  * win and return %true here. The return value can be used for reference
1232  * counting and the number of queued works stays the same. Anyway, the caller
1233  * is supposed to synchronize these operations a reasonable way.
1234  *
1235  * This function is safe to call from any context including IRQ handler.
1236  * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1237  * for details.
1238  */
1239 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1240 			      struct kthread_delayed_work *dwork,
1241 			      unsigned long delay)
1242 {
1243 	struct kthread_work *work = &dwork->work;
1244 	unsigned long flags;
1245 	int ret;
1246 
1247 	raw_spin_lock_irqsave(&worker->lock, flags);
1248 
1249 	/* Do not bother with canceling when never queued. */
1250 	if (!work->worker) {
1251 		ret = false;
1252 		goto fast_queue;
1253 	}
1254 
1255 	/* Work must not be used with >1 worker, see kthread_queue_work() */
1256 	WARN_ON_ONCE(work->worker != worker);
1257 
1258 	/*
1259 	 * Temporary cancel the work but do not fight with another command
1260 	 * that is canceling the work as well.
1261 	 *
1262 	 * It is a bit tricky because of possible races with another
1263 	 * mod_delayed_work() and cancel_delayed_work() callers.
1264 	 *
1265 	 * The timer must be canceled first because worker->lock is released
1266 	 * when doing so. But the work can be removed from the queue (list)
1267 	 * only when it can be queued again so that the return value can
1268 	 * be used for reference counting.
1269 	 */
1270 	kthread_cancel_delayed_work_timer(work, &flags);
1271 	if (work->canceling) {
1272 		/* The number of works in the queue does not change. */
1273 		ret = true;
1274 		goto out;
1275 	}
1276 	ret = __kthread_cancel_work(work);
1277 
1278 fast_queue:
1279 	__kthread_queue_delayed_work(worker, dwork, delay);
1280 out:
1281 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1282 	return ret;
1283 }
1284 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1285 
1286 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1287 {
1288 	struct kthread_worker *worker = work->worker;
1289 	unsigned long flags;
1290 	int ret = false;
1291 
1292 	if (!worker)
1293 		goto out;
1294 
1295 	raw_spin_lock_irqsave(&worker->lock, flags);
1296 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1297 	WARN_ON_ONCE(work->worker != worker);
1298 
1299 	if (is_dwork)
1300 		kthread_cancel_delayed_work_timer(work, &flags);
1301 
1302 	ret = __kthread_cancel_work(work);
1303 
1304 	if (worker->current_work != work)
1305 		goto out_fast;
1306 
1307 	/*
1308 	 * The work is in progress and we need to wait with the lock released.
1309 	 * In the meantime, block any queuing by setting the canceling counter.
1310 	 */
1311 	work->canceling++;
1312 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1313 	kthread_flush_work(work);
1314 	raw_spin_lock_irqsave(&worker->lock, flags);
1315 	work->canceling--;
1316 
1317 out_fast:
1318 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1319 out:
1320 	return ret;
1321 }
1322 
1323 /**
1324  * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1325  * @work: the kthread work to cancel
1326  *
1327  * Cancel @work and wait for its execution to finish.  This function
1328  * can be used even if the work re-queues itself. On return from this
1329  * function, @work is guaranteed to be not pending or executing on any CPU.
1330  *
1331  * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1332  * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1333  *
1334  * The caller must ensure that the worker on which @work was last
1335  * queued can't be destroyed before this function returns.
1336  *
1337  * Return: %true if @work was pending, %false otherwise.
1338  */
1339 bool kthread_cancel_work_sync(struct kthread_work *work)
1340 {
1341 	return __kthread_cancel_work_sync(work, false);
1342 }
1343 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1344 
1345 /**
1346  * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1347  *	wait for it to finish.
1348  * @dwork: the kthread delayed work to cancel
1349  *
1350  * This is kthread_cancel_work_sync() for delayed works.
1351  *
1352  * Return: %true if @dwork was pending, %false otherwise.
1353  */
1354 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1355 {
1356 	return __kthread_cancel_work_sync(&dwork->work, true);
1357 }
1358 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1359 
1360 /**
1361  * kthread_flush_worker - flush all current works on a kthread_worker
1362  * @worker: worker to flush
1363  *
1364  * Wait until all currently executing or pending works on @worker are
1365  * finished.
1366  */
1367 void kthread_flush_worker(struct kthread_worker *worker)
1368 {
1369 	struct kthread_flush_work fwork = {
1370 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1371 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1372 	};
1373 
1374 	kthread_queue_work(worker, &fwork.work);
1375 	wait_for_completion(&fwork.done);
1376 }
1377 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1378 
1379 /**
1380  * kthread_destroy_worker - destroy a kthread worker
1381  * @worker: worker to be destroyed
1382  *
1383  * Flush and destroy @worker.  The simple flush is enough because the kthread
1384  * worker API is used only in trivial scenarios.  There are no multi-step state
1385  * machines needed.
1386  */
1387 void kthread_destroy_worker(struct kthread_worker *worker)
1388 {
1389 	struct task_struct *task;
1390 
1391 	task = worker->task;
1392 	if (WARN_ON(!task))
1393 		return;
1394 
1395 	kthread_flush_worker(worker);
1396 	kthread_stop(task);
1397 	WARN_ON(!list_empty(&worker->work_list));
1398 	kfree(worker);
1399 }
1400 EXPORT_SYMBOL(kthread_destroy_worker);
1401 
1402 /**
1403  * kthread_use_mm - make the calling kthread operate on an address space
1404  * @mm: address space to operate on
1405  */
1406 void kthread_use_mm(struct mm_struct *mm)
1407 {
1408 	struct mm_struct *active_mm;
1409 	struct task_struct *tsk = current;
1410 
1411 	WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1412 	WARN_ON_ONCE(tsk->mm);
1413 
1414 	task_lock(tsk);
1415 	/* Hold off tlb flush IPIs while switching mm's */
1416 	local_irq_disable();
1417 	active_mm = tsk->active_mm;
1418 	if (active_mm != mm) {
1419 		mmgrab(mm);
1420 		tsk->active_mm = mm;
1421 	}
1422 	tsk->mm = mm;
1423 	membarrier_update_current_mm(mm);
1424 	switch_mm_irqs_off(active_mm, mm, tsk);
1425 	local_irq_enable();
1426 	task_unlock(tsk);
1427 #ifdef finish_arch_post_lock_switch
1428 	finish_arch_post_lock_switch();
1429 #endif
1430 
1431 	/*
1432 	 * When a kthread starts operating on an address space, the loop
1433 	 * in membarrier_{private,global}_expedited() may not observe
1434 	 * that tsk->mm, and not issue an IPI. Membarrier requires a
1435 	 * memory barrier after storing to tsk->mm, before accessing
1436 	 * user-space memory. A full memory barrier for membarrier
1437 	 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1438 	 * mmdrop(), or explicitly with smp_mb().
1439 	 */
1440 	if (active_mm != mm)
1441 		mmdrop(active_mm);
1442 	else
1443 		smp_mb();
1444 
1445 	to_kthread(tsk)->oldfs = force_uaccess_begin();
1446 }
1447 EXPORT_SYMBOL_GPL(kthread_use_mm);
1448 
1449 /**
1450  * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1451  * @mm: address space to operate on
1452  */
1453 void kthread_unuse_mm(struct mm_struct *mm)
1454 {
1455 	struct task_struct *tsk = current;
1456 
1457 	WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1458 	WARN_ON_ONCE(!tsk->mm);
1459 
1460 	force_uaccess_end(to_kthread(tsk)->oldfs);
1461 
1462 	task_lock(tsk);
1463 	/*
1464 	 * When a kthread stops operating on an address space, the loop
1465 	 * in membarrier_{private,global}_expedited() may not observe
1466 	 * that tsk->mm, and not issue an IPI. Membarrier requires a
1467 	 * memory barrier after accessing user-space memory, before
1468 	 * clearing tsk->mm.
1469 	 */
1470 	smp_mb__after_spinlock();
1471 	sync_mm_rss(mm);
1472 	local_irq_disable();
1473 	tsk->mm = NULL;
1474 	membarrier_update_current_mm(NULL);
1475 	/* active_mm is still 'mm' */
1476 	enter_lazy_tlb(mm, tsk);
1477 	local_irq_enable();
1478 	task_unlock(tsk);
1479 }
1480 EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1481 
1482 #ifdef CONFIG_BLK_CGROUP
1483 /**
1484  * kthread_associate_blkcg - associate blkcg to current kthread
1485  * @css: the cgroup info
1486  *
1487  * Current thread must be a kthread. The thread is running jobs on behalf of
1488  * other threads. In some cases, we expect the jobs attach cgroup info of
1489  * original threads instead of that of current thread. This function stores
1490  * original thread's cgroup info in current kthread context for later
1491  * retrieval.
1492  */
1493 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1494 {
1495 	struct kthread *kthread;
1496 
1497 	if (!(current->flags & PF_KTHREAD))
1498 		return;
1499 	kthread = to_kthread(current);
1500 	if (!kthread)
1501 		return;
1502 
1503 	if (kthread->blkcg_css) {
1504 		css_put(kthread->blkcg_css);
1505 		kthread->blkcg_css = NULL;
1506 	}
1507 	if (css) {
1508 		css_get(css);
1509 		kthread->blkcg_css = css;
1510 	}
1511 }
1512 EXPORT_SYMBOL(kthread_associate_blkcg);
1513 
1514 /**
1515  * kthread_blkcg - get associated blkcg css of current kthread
1516  *
1517  * Current thread must be a kthread.
1518  */
1519 struct cgroup_subsys_state *kthread_blkcg(void)
1520 {
1521 	struct kthread *kthread;
1522 
1523 	if (current->flags & PF_KTHREAD) {
1524 		kthread = to_kthread(current);
1525 		if (kthread)
1526 			return kthread->blkcg_css;
1527 	}
1528 	return NULL;
1529 }
1530 EXPORT_SYMBOL(kthread_blkcg);
1531 #endif
1532