xref: /openbmc/linux/kernel/kthread.c (revision ba61bb17)
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 <trace/events/sched.h>
24 
25 static DEFINE_SPINLOCK(kthread_create_lock);
26 static LIST_HEAD(kthread_create_list);
27 struct task_struct *kthreadd_task;
28 
29 struct kthread_create_info
30 {
31 	/* Information passed to kthread() from kthreadd. */
32 	int (*threadfn)(void *data);
33 	void *data;
34 	int node;
35 
36 	/* Result passed back to kthread_create() from kthreadd. */
37 	struct task_struct *result;
38 	struct completion *done;
39 
40 	struct list_head list;
41 };
42 
43 struct kthread {
44 	unsigned long flags;
45 	unsigned int cpu;
46 	void *data;
47 	struct completion parked;
48 	struct completion exited;
49 #ifdef CONFIG_BLK_CGROUP
50 	struct cgroup_subsys_state *blkcg_css;
51 #endif
52 };
53 
54 enum KTHREAD_BITS {
55 	KTHREAD_IS_PER_CPU = 0,
56 	KTHREAD_SHOULD_STOP,
57 	KTHREAD_SHOULD_PARK,
58 };
59 
60 static inline void set_kthread_struct(void *kthread)
61 {
62 	/*
63 	 * We abuse ->set_child_tid to avoid the new member and because it
64 	 * can't be wrongly copied by copy_process(). We also rely on fact
65 	 * that the caller can't exec, so PF_KTHREAD can't be cleared.
66 	 */
67 	current->set_child_tid = (__force void __user *)kthread;
68 }
69 
70 static inline struct kthread *to_kthread(struct task_struct *k)
71 {
72 	WARN_ON(!(k->flags & PF_KTHREAD));
73 	return (__force void *)k->set_child_tid;
74 }
75 
76 void free_kthread_struct(struct task_struct *k)
77 {
78 	struct kthread *kthread;
79 
80 	/*
81 	 * Can be NULL if this kthread was created by kernel_thread()
82 	 * or if kmalloc() in kthread() failed.
83 	 */
84 	kthread = to_kthread(k);
85 #ifdef CONFIG_BLK_CGROUP
86 	WARN_ON_ONCE(kthread && kthread->blkcg_css);
87 #endif
88 	kfree(kthread);
89 }
90 
91 /**
92  * kthread_should_stop - should this kthread return now?
93  *
94  * When someone calls kthread_stop() on your kthread, it will be woken
95  * and this will return true.  You should then return, and your return
96  * value will be passed through to kthread_stop().
97  */
98 bool kthread_should_stop(void)
99 {
100 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
101 }
102 EXPORT_SYMBOL(kthread_should_stop);
103 
104 /**
105  * kthread_should_park - should this kthread park now?
106  *
107  * When someone calls kthread_park() on your kthread, it will be woken
108  * and this will return true.  You should then do the necessary
109  * cleanup and call kthread_parkme()
110  *
111  * Similar to kthread_should_stop(), but this keeps the thread alive
112  * and in a park position. kthread_unpark() "restarts" the thread and
113  * calls the thread function again.
114  */
115 bool kthread_should_park(void)
116 {
117 	return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
118 }
119 EXPORT_SYMBOL_GPL(kthread_should_park);
120 
121 /**
122  * kthread_freezable_should_stop - should this freezable kthread return now?
123  * @was_frozen: optional out parameter, indicates whether %current was frozen
124  *
125  * kthread_should_stop() for freezable kthreads, which will enter
126  * refrigerator if necessary.  This function is safe from kthread_stop() /
127  * freezer deadlock and freezable kthreads should use this function instead
128  * of calling try_to_freeze() directly.
129  */
130 bool kthread_freezable_should_stop(bool *was_frozen)
131 {
132 	bool frozen = false;
133 
134 	might_sleep();
135 
136 	if (unlikely(freezing(current)))
137 		frozen = __refrigerator(true);
138 
139 	if (was_frozen)
140 		*was_frozen = frozen;
141 
142 	return kthread_should_stop();
143 }
144 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
145 
146 /**
147  * kthread_data - return data value specified on kthread creation
148  * @task: kthread task in question
149  *
150  * Return the data value specified when kthread @task was created.
151  * The caller is responsible for ensuring the validity of @task when
152  * calling this function.
153  */
154 void *kthread_data(struct task_struct *task)
155 {
156 	return to_kthread(task)->data;
157 }
158 
159 /**
160  * kthread_probe_data - speculative version of kthread_data()
161  * @task: possible kthread task in question
162  *
163  * @task could be a kthread task.  Return the data value specified when it
164  * was created if accessible.  If @task isn't a kthread task or its data is
165  * inaccessible for any reason, %NULL is returned.  This function requires
166  * that @task itself is safe to dereference.
167  */
168 void *kthread_probe_data(struct task_struct *task)
169 {
170 	struct kthread *kthread = to_kthread(task);
171 	void *data = NULL;
172 
173 	probe_kernel_read(&data, &kthread->data, sizeof(data));
174 	return data;
175 }
176 
177 static void __kthread_parkme(struct kthread *self)
178 {
179 	for (;;) {
180 		set_current_state(TASK_PARKED);
181 		if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
182 			break;
183 		schedule();
184 	}
185 	__set_current_state(TASK_RUNNING);
186 }
187 
188 void kthread_parkme(void)
189 {
190 	__kthread_parkme(to_kthread(current));
191 }
192 EXPORT_SYMBOL_GPL(kthread_parkme);
193 
194 void kthread_park_complete(struct task_struct *k)
195 {
196 	complete_all(&to_kthread(k)->parked);
197 }
198 
199 static int kthread(void *_create)
200 {
201 	/* Copy data: it's on kthread's stack */
202 	struct kthread_create_info *create = _create;
203 	int (*threadfn)(void *data) = create->threadfn;
204 	void *data = create->data;
205 	struct completion *done;
206 	struct kthread *self;
207 	int ret;
208 
209 	self = kzalloc(sizeof(*self), GFP_KERNEL);
210 	set_kthread_struct(self);
211 
212 	/* If user was SIGKILLed, I release the structure. */
213 	done = xchg(&create->done, NULL);
214 	if (!done) {
215 		kfree(create);
216 		do_exit(-EINTR);
217 	}
218 
219 	if (!self) {
220 		create->result = ERR_PTR(-ENOMEM);
221 		complete(done);
222 		do_exit(-ENOMEM);
223 	}
224 
225 	self->data = data;
226 	init_completion(&self->exited);
227 	init_completion(&self->parked);
228 	current->vfork_done = &self->exited;
229 
230 	/* OK, tell user we're spawned, wait for stop or wakeup */
231 	__set_current_state(TASK_UNINTERRUPTIBLE);
232 	create->result = current;
233 	complete(done);
234 	schedule();
235 
236 	ret = -EINTR;
237 	if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
238 		cgroup_kthread_ready();
239 		__kthread_parkme(self);
240 		ret = threadfn(data);
241 	}
242 	do_exit(ret);
243 }
244 
245 /* called from do_fork() to get node information for about to be created task */
246 int tsk_fork_get_node(struct task_struct *tsk)
247 {
248 #ifdef CONFIG_NUMA
249 	if (tsk == kthreadd_task)
250 		return tsk->pref_node_fork;
251 #endif
252 	return NUMA_NO_NODE;
253 }
254 
255 static void create_kthread(struct kthread_create_info *create)
256 {
257 	int pid;
258 
259 #ifdef CONFIG_NUMA
260 	current->pref_node_fork = create->node;
261 #endif
262 	/* We want our own signal handler (we take no signals by default). */
263 	pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
264 	if (pid < 0) {
265 		/* If user was SIGKILLed, I release the structure. */
266 		struct completion *done = xchg(&create->done, NULL);
267 
268 		if (!done) {
269 			kfree(create);
270 			return;
271 		}
272 		create->result = ERR_PTR(pid);
273 		complete(done);
274 	}
275 }
276 
277 static __printf(4, 0)
278 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
279 						    void *data, int node,
280 						    const char namefmt[],
281 						    va_list args)
282 {
283 	DECLARE_COMPLETION_ONSTACK(done);
284 	struct task_struct *task;
285 	struct kthread_create_info *create = kmalloc(sizeof(*create),
286 						     GFP_KERNEL);
287 
288 	if (!create)
289 		return ERR_PTR(-ENOMEM);
290 	create->threadfn = threadfn;
291 	create->data = data;
292 	create->node = node;
293 	create->done = &done;
294 
295 	spin_lock(&kthread_create_lock);
296 	list_add_tail(&create->list, &kthread_create_list);
297 	spin_unlock(&kthread_create_lock);
298 
299 	wake_up_process(kthreadd_task);
300 	/*
301 	 * Wait for completion in killable state, for I might be chosen by
302 	 * the OOM killer while kthreadd is trying to allocate memory for
303 	 * new kernel thread.
304 	 */
305 	if (unlikely(wait_for_completion_killable(&done))) {
306 		/*
307 		 * If I was SIGKILLed before kthreadd (or new kernel thread)
308 		 * calls complete(), leave the cleanup of this structure to
309 		 * that thread.
310 		 */
311 		if (xchg(&create->done, NULL))
312 			return ERR_PTR(-EINTR);
313 		/*
314 		 * kthreadd (or new kernel thread) will call complete()
315 		 * shortly.
316 		 */
317 		wait_for_completion(&done);
318 	}
319 	task = create->result;
320 	if (!IS_ERR(task)) {
321 		static const struct sched_param param = { .sched_priority = 0 };
322 
323 		vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
324 		/*
325 		 * root may have changed our (kthreadd's) priority or CPU mask.
326 		 * The kernel thread should not inherit these properties.
327 		 */
328 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
329 		set_cpus_allowed_ptr(task, cpu_all_mask);
330 	}
331 	kfree(create);
332 	return task;
333 }
334 
335 /**
336  * kthread_create_on_node - create a kthread.
337  * @threadfn: the function to run until signal_pending(current).
338  * @data: data ptr for @threadfn.
339  * @node: task and thread structures for the thread are allocated on this node
340  * @namefmt: printf-style name for the thread.
341  *
342  * Description: This helper function creates and names a kernel
343  * thread.  The thread will be stopped: use wake_up_process() to start
344  * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
345  * is affine to all CPUs.
346  *
347  * If thread is going to be bound on a particular cpu, give its node
348  * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
349  * When woken, the thread will run @threadfn() with @data as its
350  * argument. @threadfn() can either call do_exit() directly if it is a
351  * standalone thread for which no one will call kthread_stop(), or
352  * return when 'kthread_should_stop()' is true (which means
353  * kthread_stop() has been called).  The return value should be zero
354  * or a negative error number; it will be passed to kthread_stop().
355  *
356  * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
357  */
358 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
359 					   void *data, int node,
360 					   const char namefmt[],
361 					   ...)
362 {
363 	struct task_struct *task;
364 	va_list args;
365 
366 	va_start(args, namefmt);
367 	task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
368 	va_end(args);
369 
370 	return task;
371 }
372 EXPORT_SYMBOL(kthread_create_on_node);
373 
374 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
375 {
376 	unsigned long flags;
377 
378 	if (!wait_task_inactive(p, state)) {
379 		WARN_ON(1);
380 		return;
381 	}
382 
383 	/* It's safe because the task is inactive. */
384 	raw_spin_lock_irqsave(&p->pi_lock, flags);
385 	do_set_cpus_allowed(p, mask);
386 	p->flags |= PF_NO_SETAFFINITY;
387 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
388 }
389 
390 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
391 {
392 	__kthread_bind_mask(p, cpumask_of(cpu), state);
393 }
394 
395 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
396 {
397 	__kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
398 }
399 
400 /**
401  * kthread_bind - bind a just-created kthread to a cpu.
402  * @p: thread created by kthread_create().
403  * @cpu: cpu (might not be online, must be possible) for @k to run on.
404  *
405  * Description: This function is equivalent to set_cpus_allowed(),
406  * except that @cpu doesn't need to be online, and the thread must be
407  * stopped (i.e., just returned from kthread_create()).
408  */
409 void kthread_bind(struct task_struct *p, unsigned int cpu)
410 {
411 	__kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
412 }
413 EXPORT_SYMBOL(kthread_bind);
414 
415 /**
416  * kthread_create_on_cpu - Create a cpu bound kthread
417  * @threadfn: the function to run until signal_pending(current).
418  * @data: data ptr for @threadfn.
419  * @cpu: The cpu on which the thread should be bound,
420  * @namefmt: printf-style name for the thread. Format is restricted
421  *	     to "name.*%u". Code fills in cpu number.
422  *
423  * Description: This helper function creates and names a kernel thread
424  * The thread will be woken and put into park mode.
425  */
426 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
427 					  void *data, unsigned int cpu,
428 					  const char *namefmt)
429 {
430 	struct task_struct *p;
431 
432 	p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
433 				   cpu);
434 	if (IS_ERR(p))
435 		return p;
436 	kthread_bind(p, cpu);
437 	/* CPU hotplug need to bind once again when unparking the thread. */
438 	set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
439 	to_kthread(p)->cpu = cpu;
440 	return p;
441 }
442 
443 /**
444  * kthread_unpark - unpark a thread created by kthread_create().
445  * @k:		thread created by kthread_create().
446  *
447  * Sets kthread_should_park() for @k to return false, wakes it, and
448  * waits for it to return. If the thread is marked percpu then its
449  * bound to the cpu again.
450  */
451 void kthread_unpark(struct task_struct *k)
452 {
453 	struct kthread *kthread = to_kthread(k);
454 
455 	/*
456 	 * Newly created kthread was parked when the CPU was offline.
457 	 * The binding was lost and we need to set it again.
458 	 */
459 	if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
460 		__kthread_bind(k, kthread->cpu, TASK_PARKED);
461 
462 	reinit_completion(&kthread->parked);
463 	clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
464 	wake_up_state(k, TASK_PARKED);
465 }
466 EXPORT_SYMBOL_GPL(kthread_unpark);
467 
468 /**
469  * kthread_park - park a thread created by kthread_create().
470  * @k: thread created by kthread_create().
471  *
472  * Sets kthread_should_park() for @k to return true, wakes it, and
473  * waits for it to return. This can also be called after kthread_create()
474  * instead of calling wake_up_process(): the thread will park without
475  * calling threadfn().
476  *
477  * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
478  * If called by the kthread itself just the park bit is set.
479  */
480 int kthread_park(struct task_struct *k)
481 {
482 	struct kthread *kthread = to_kthread(k);
483 
484 	if (WARN_ON(k->flags & PF_EXITING))
485 		return -ENOSYS;
486 
487 	set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
488 	if (k != current) {
489 		wake_up_process(k);
490 		wait_for_completion(&kthread->parked);
491 	}
492 
493 	return 0;
494 }
495 EXPORT_SYMBOL_GPL(kthread_park);
496 
497 /**
498  * kthread_stop - stop a thread created by kthread_create().
499  * @k: thread created by kthread_create().
500  *
501  * Sets kthread_should_stop() for @k to return true, wakes it, and
502  * waits for it to exit. This can also be called after kthread_create()
503  * instead of calling wake_up_process(): the thread will exit without
504  * calling threadfn().
505  *
506  * If threadfn() may call do_exit() itself, the caller must ensure
507  * task_struct can't go away.
508  *
509  * Returns the result of threadfn(), or %-EINTR if wake_up_process()
510  * was never called.
511  */
512 int kthread_stop(struct task_struct *k)
513 {
514 	struct kthread *kthread;
515 	int ret;
516 
517 	trace_sched_kthread_stop(k);
518 
519 	get_task_struct(k);
520 	kthread = to_kthread(k);
521 	set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
522 	kthread_unpark(k);
523 	wake_up_process(k);
524 	wait_for_completion(&kthread->exited);
525 	ret = k->exit_code;
526 	put_task_struct(k);
527 
528 	trace_sched_kthread_stop_ret(ret);
529 	return ret;
530 }
531 EXPORT_SYMBOL(kthread_stop);
532 
533 int kthreadd(void *unused)
534 {
535 	struct task_struct *tsk = current;
536 
537 	/* Setup a clean context for our children to inherit. */
538 	set_task_comm(tsk, "kthreadd");
539 	ignore_signals(tsk);
540 	set_cpus_allowed_ptr(tsk, cpu_all_mask);
541 	set_mems_allowed(node_states[N_MEMORY]);
542 
543 	current->flags |= PF_NOFREEZE;
544 	cgroup_init_kthreadd();
545 
546 	for (;;) {
547 		set_current_state(TASK_INTERRUPTIBLE);
548 		if (list_empty(&kthread_create_list))
549 			schedule();
550 		__set_current_state(TASK_RUNNING);
551 
552 		spin_lock(&kthread_create_lock);
553 		while (!list_empty(&kthread_create_list)) {
554 			struct kthread_create_info *create;
555 
556 			create = list_entry(kthread_create_list.next,
557 					    struct kthread_create_info, list);
558 			list_del_init(&create->list);
559 			spin_unlock(&kthread_create_lock);
560 
561 			create_kthread(create);
562 
563 			spin_lock(&kthread_create_lock);
564 		}
565 		spin_unlock(&kthread_create_lock);
566 	}
567 
568 	return 0;
569 }
570 
571 void __kthread_init_worker(struct kthread_worker *worker,
572 				const char *name,
573 				struct lock_class_key *key)
574 {
575 	memset(worker, 0, sizeof(struct kthread_worker));
576 	spin_lock_init(&worker->lock);
577 	lockdep_set_class_and_name(&worker->lock, key, name);
578 	INIT_LIST_HEAD(&worker->work_list);
579 	INIT_LIST_HEAD(&worker->delayed_work_list);
580 }
581 EXPORT_SYMBOL_GPL(__kthread_init_worker);
582 
583 /**
584  * kthread_worker_fn - kthread function to process kthread_worker
585  * @worker_ptr: pointer to initialized kthread_worker
586  *
587  * This function implements the main cycle of kthread worker. It processes
588  * work_list until it is stopped with kthread_stop(). It sleeps when the queue
589  * is empty.
590  *
591  * The works are not allowed to keep any locks, disable preemption or interrupts
592  * when they finish. There is defined a safe point for freezing when one work
593  * finishes and before a new one is started.
594  *
595  * Also the works must not be handled by more than one worker at the same time,
596  * see also kthread_queue_work().
597  */
598 int kthread_worker_fn(void *worker_ptr)
599 {
600 	struct kthread_worker *worker = worker_ptr;
601 	struct kthread_work *work;
602 
603 	/*
604 	 * FIXME: Update the check and remove the assignment when all kthread
605 	 * worker users are created using kthread_create_worker*() functions.
606 	 */
607 	WARN_ON(worker->task && worker->task != current);
608 	worker->task = current;
609 
610 	if (worker->flags & KTW_FREEZABLE)
611 		set_freezable();
612 
613 repeat:
614 	set_current_state(TASK_INTERRUPTIBLE);	/* mb paired w/ kthread_stop */
615 
616 	if (kthread_should_stop()) {
617 		__set_current_state(TASK_RUNNING);
618 		spin_lock_irq(&worker->lock);
619 		worker->task = NULL;
620 		spin_unlock_irq(&worker->lock);
621 		return 0;
622 	}
623 
624 	work = NULL;
625 	spin_lock_irq(&worker->lock);
626 	if (!list_empty(&worker->work_list)) {
627 		work = list_first_entry(&worker->work_list,
628 					struct kthread_work, node);
629 		list_del_init(&work->node);
630 	}
631 	worker->current_work = work;
632 	spin_unlock_irq(&worker->lock);
633 
634 	if (work) {
635 		__set_current_state(TASK_RUNNING);
636 		work->func(work);
637 	} else if (!freezing(current))
638 		schedule();
639 
640 	try_to_freeze();
641 	cond_resched();
642 	goto repeat;
643 }
644 EXPORT_SYMBOL_GPL(kthread_worker_fn);
645 
646 static __printf(3, 0) struct kthread_worker *
647 __kthread_create_worker(int cpu, unsigned int flags,
648 			const char namefmt[], va_list args)
649 {
650 	struct kthread_worker *worker;
651 	struct task_struct *task;
652 	int node = -1;
653 
654 	worker = kzalloc(sizeof(*worker), GFP_KERNEL);
655 	if (!worker)
656 		return ERR_PTR(-ENOMEM);
657 
658 	kthread_init_worker(worker);
659 
660 	if (cpu >= 0)
661 		node = cpu_to_node(cpu);
662 
663 	task = __kthread_create_on_node(kthread_worker_fn, worker,
664 						node, namefmt, args);
665 	if (IS_ERR(task))
666 		goto fail_task;
667 
668 	if (cpu >= 0)
669 		kthread_bind(task, cpu);
670 
671 	worker->flags = flags;
672 	worker->task = task;
673 	wake_up_process(task);
674 	return worker;
675 
676 fail_task:
677 	kfree(worker);
678 	return ERR_CAST(task);
679 }
680 
681 /**
682  * kthread_create_worker - create a kthread worker
683  * @flags: flags modifying the default behavior of the worker
684  * @namefmt: printf-style name for the kthread worker (task).
685  *
686  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
687  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
688  * when the worker was SIGKILLed.
689  */
690 struct kthread_worker *
691 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
692 {
693 	struct kthread_worker *worker;
694 	va_list args;
695 
696 	va_start(args, namefmt);
697 	worker = __kthread_create_worker(-1, flags, namefmt, args);
698 	va_end(args);
699 
700 	return worker;
701 }
702 EXPORT_SYMBOL(kthread_create_worker);
703 
704 /**
705  * kthread_create_worker_on_cpu - create a kthread worker and bind it
706  *	it to a given CPU and the associated NUMA node.
707  * @cpu: CPU number
708  * @flags: flags modifying the default behavior of the worker
709  * @namefmt: printf-style name for the kthread worker (task).
710  *
711  * Use a valid CPU number if you want to bind the kthread worker
712  * to the given CPU and the associated NUMA node.
713  *
714  * A good practice is to add the cpu number also into the worker name.
715  * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
716  *
717  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
718  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
719  * when the worker was SIGKILLed.
720  */
721 struct kthread_worker *
722 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
723 			     const char namefmt[], ...)
724 {
725 	struct kthread_worker *worker;
726 	va_list args;
727 
728 	va_start(args, namefmt);
729 	worker = __kthread_create_worker(cpu, flags, namefmt, args);
730 	va_end(args);
731 
732 	return worker;
733 }
734 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
735 
736 /*
737  * Returns true when the work could not be queued at the moment.
738  * It happens when it is already pending in a worker list
739  * or when it is being cancelled.
740  */
741 static inline bool queuing_blocked(struct kthread_worker *worker,
742 				   struct kthread_work *work)
743 {
744 	lockdep_assert_held(&worker->lock);
745 
746 	return !list_empty(&work->node) || work->canceling;
747 }
748 
749 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
750 					     struct kthread_work *work)
751 {
752 	lockdep_assert_held(&worker->lock);
753 	WARN_ON_ONCE(!list_empty(&work->node));
754 	/* Do not use a work with >1 worker, see kthread_queue_work() */
755 	WARN_ON_ONCE(work->worker && work->worker != worker);
756 }
757 
758 /* insert @work before @pos in @worker */
759 static void kthread_insert_work(struct kthread_worker *worker,
760 				struct kthread_work *work,
761 				struct list_head *pos)
762 {
763 	kthread_insert_work_sanity_check(worker, work);
764 
765 	list_add_tail(&work->node, pos);
766 	work->worker = worker;
767 	if (!worker->current_work && likely(worker->task))
768 		wake_up_process(worker->task);
769 }
770 
771 /**
772  * kthread_queue_work - queue a kthread_work
773  * @worker: target kthread_worker
774  * @work: kthread_work to queue
775  *
776  * Queue @work to work processor @task for async execution.  @task
777  * must have been created with kthread_worker_create().  Returns %true
778  * if @work was successfully queued, %false if it was already pending.
779  *
780  * Reinitialize the work if it needs to be used by another worker.
781  * For example, when the worker was stopped and started again.
782  */
783 bool kthread_queue_work(struct kthread_worker *worker,
784 			struct kthread_work *work)
785 {
786 	bool ret = false;
787 	unsigned long flags;
788 
789 	spin_lock_irqsave(&worker->lock, flags);
790 	if (!queuing_blocked(worker, work)) {
791 		kthread_insert_work(worker, work, &worker->work_list);
792 		ret = true;
793 	}
794 	spin_unlock_irqrestore(&worker->lock, flags);
795 	return ret;
796 }
797 EXPORT_SYMBOL_GPL(kthread_queue_work);
798 
799 /**
800  * kthread_delayed_work_timer_fn - callback that queues the associated kthread
801  *	delayed work when the timer expires.
802  * @t: pointer to the expired timer
803  *
804  * The format of the function is defined by struct timer_list.
805  * It should have been called from irqsafe timer with irq already off.
806  */
807 void kthread_delayed_work_timer_fn(struct timer_list *t)
808 {
809 	struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
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 
842 	/*
843 	 * If @delay is 0, queue @dwork->work immediately.  This is for
844 	 * both optimization and correctness.  The earliest @timer can
845 	 * expire is on the closest next tick and delayed_work users depend
846 	 * on that there's no such delay when @delay is 0.
847 	 */
848 	if (!delay) {
849 		kthread_insert_work(worker, work, &worker->work_list);
850 		return;
851 	}
852 
853 	/* Be paranoid and try to detect possible races already now. */
854 	kthread_insert_work_sanity_check(worker, work);
855 
856 	list_add(&work->node, &worker->delayed_work_list);
857 	work->worker = worker;
858 	timer->expires = jiffies + delay;
859 	add_timer(timer);
860 }
861 
862 /**
863  * kthread_queue_delayed_work - queue the associated kthread work
864  *	after a delay.
865  * @worker: target kthread_worker
866  * @dwork: kthread_delayed_work to queue
867  * @delay: number of jiffies to wait before queuing
868  *
869  * If the work has not been pending it starts a timer that will queue
870  * the work after the given @delay. If @delay is zero, it queues the
871  * work immediately.
872  *
873  * Return: %false if the @work has already been pending. It means that
874  * either the timer was running or the work was queued. It returns %true
875  * otherwise.
876  */
877 bool kthread_queue_delayed_work(struct kthread_worker *worker,
878 				struct kthread_delayed_work *dwork,
879 				unsigned long delay)
880 {
881 	struct kthread_work *work = &dwork->work;
882 	unsigned long flags;
883 	bool ret = false;
884 
885 	spin_lock_irqsave(&worker->lock, flags);
886 
887 	if (!queuing_blocked(worker, work)) {
888 		__kthread_queue_delayed_work(worker, dwork, delay);
889 		ret = true;
890 	}
891 
892 	spin_unlock_irqrestore(&worker->lock, flags);
893 	return ret;
894 }
895 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
896 
897 struct kthread_flush_work {
898 	struct kthread_work	work;
899 	struct completion	done;
900 };
901 
902 static void kthread_flush_work_fn(struct kthread_work *work)
903 {
904 	struct kthread_flush_work *fwork =
905 		container_of(work, struct kthread_flush_work, work);
906 	complete(&fwork->done);
907 }
908 
909 /**
910  * kthread_flush_work - flush a kthread_work
911  * @work: work to flush
912  *
913  * If @work is queued or executing, wait for it to finish execution.
914  */
915 void kthread_flush_work(struct kthread_work *work)
916 {
917 	struct kthread_flush_work fwork = {
918 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
919 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
920 	};
921 	struct kthread_worker *worker;
922 	bool noop = false;
923 
924 	worker = work->worker;
925 	if (!worker)
926 		return;
927 
928 	spin_lock_irq(&worker->lock);
929 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
930 	WARN_ON_ONCE(work->worker != worker);
931 
932 	if (!list_empty(&work->node))
933 		kthread_insert_work(worker, &fwork.work, work->node.next);
934 	else if (worker->current_work == work)
935 		kthread_insert_work(worker, &fwork.work,
936 				    worker->work_list.next);
937 	else
938 		noop = true;
939 
940 	spin_unlock_irq(&worker->lock);
941 
942 	if (!noop)
943 		wait_for_completion(&fwork.done);
944 }
945 EXPORT_SYMBOL_GPL(kthread_flush_work);
946 
947 /*
948  * This function removes the work from the worker queue. Also it makes sure
949  * that it won't get queued later via the delayed work's timer.
950  *
951  * The work might still be in use when this function finishes. See the
952  * current_work proceed by the worker.
953  *
954  * Return: %true if @work was pending and successfully canceled,
955  *	%false if @work was not pending
956  */
957 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
958 				  unsigned long *flags)
959 {
960 	/* Try to cancel the timer if exists. */
961 	if (is_dwork) {
962 		struct kthread_delayed_work *dwork =
963 			container_of(work, struct kthread_delayed_work, work);
964 		struct kthread_worker *worker = work->worker;
965 
966 		/*
967 		 * del_timer_sync() must be called to make sure that the timer
968 		 * callback is not running. The lock must be temporary released
969 		 * to avoid a deadlock with the callback. In the meantime,
970 		 * any queuing is blocked by setting the canceling counter.
971 		 */
972 		work->canceling++;
973 		spin_unlock_irqrestore(&worker->lock, *flags);
974 		del_timer_sync(&dwork->timer);
975 		spin_lock_irqsave(&worker->lock, *flags);
976 		work->canceling--;
977 	}
978 
979 	/*
980 	 * Try to remove the work from a worker list. It might either
981 	 * be from worker->work_list or from worker->delayed_work_list.
982 	 */
983 	if (!list_empty(&work->node)) {
984 		list_del_init(&work->node);
985 		return true;
986 	}
987 
988 	return false;
989 }
990 
991 /**
992  * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
993  * @worker: kthread worker to use
994  * @dwork: kthread delayed work to queue
995  * @delay: number of jiffies to wait before queuing
996  *
997  * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
998  * modify @dwork's timer so that it expires after @delay. If @delay is zero,
999  * @work is guaranteed to be queued immediately.
1000  *
1001  * Return: %true if @dwork was pending and its timer was modified,
1002  * %false otherwise.
1003  *
1004  * A special case is when the work is being canceled in parallel.
1005  * It might be caused either by the real kthread_cancel_delayed_work_sync()
1006  * or yet another kthread_mod_delayed_work() call. We let the other command
1007  * win and return %false here. The caller is supposed to synchronize these
1008  * operations a reasonable way.
1009  *
1010  * This function is safe to call from any context including IRQ handler.
1011  * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1012  * for details.
1013  */
1014 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1015 			      struct kthread_delayed_work *dwork,
1016 			      unsigned long delay)
1017 {
1018 	struct kthread_work *work = &dwork->work;
1019 	unsigned long flags;
1020 	int ret = false;
1021 
1022 	spin_lock_irqsave(&worker->lock, flags);
1023 
1024 	/* Do not bother with canceling when never queued. */
1025 	if (!work->worker)
1026 		goto fast_queue;
1027 
1028 	/* Work must not be used with >1 worker, see kthread_queue_work() */
1029 	WARN_ON_ONCE(work->worker != worker);
1030 
1031 	/* Do not fight with another command that is canceling this work. */
1032 	if (work->canceling)
1033 		goto out;
1034 
1035 	ret = __kthread_cancel_work(work, true, &flags);
1036 fast_queue:
1037 	__kthread_queue_delayed_work(worker, dwork, delay);
1038 out:
1039 	spin_unlock_irqrestore(&worker->lock, flags);
1040 	return ret;
1041 }
1042 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1043 
1044 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1045 {
1046 	struct kthread_worker *worker = work->worker;
1047 	unsigned long flags;
1048 	int ret = false;
1049 
1050 	if (!worker)
1051 		goto out;
1052 
1053 	spin_lock_irqsave(&worker->lock, flags);
1054 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1055 	WARN_ON_ONCE(work->worker != worker);
1056 
1057 	ret = __kthread_cancel_work(work, is_dwork, &flags);
1058 
1059 	if (worker->current_work != work)
1060 		goto out_fast;
1061 
1062 	/*
1063 	 * The work is in progress and we need to wait with the lock released.
1064 	 * In the meantime, block any queuing by setting the canceling counter.
1065 	 */
1066 	work->canceling++;
1067 	spin_unlock_irqrestore(&worker->lock, flags);
1068 	kthread_flush_work(work);
1069 	spin_lock_irqsave(&worker->lock, flags);
1070 	work->canceling--;
1071 
1072 out_fast:
1073 	spin_unlock_irqrestore(&worker->lock, flags);
1074 out:
1075 	return ret;
1076 }
1077 
1078 /**
1079  * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1080  * @work: the kthread work to cancel
1081  *
1082  * Cancel @work and wait for its execution to finish.  This function
1083  * can be used even if the work re-queues itself. On return from this
1084  * function, @work is guaranteed to be not pending or executing on any CPU.
1085  *
1086  * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1087  * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1088  *
1089  * The caller must ensure that the worker on which @work was last
1090  * queued can't be destroyed before this function returns.
1091  *
1092  * Return: %true if @work was pending, %false otherwise.
1093  */
1094 bool kthread_cancel_work_sync(struct kthread_work *work)
1095 {
1096 	return __kthread_cancel_work_sync(work, false);
1097 }
1098 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1099 
1100 /**
1101  * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1102  *	wait for it to finish.
1103  * @dwork: the kthread delayed work to cancel
1104  *
1105  * This is kthread_cancel_work_sync() for delayed works.
1106  *
1107  * Return: %true if @dwork was pending, %false otherwise.
1108  */
1109 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1110 {
1111 	return __kthread_cancel_work_sync(&dwork->work, true);
1112 }
1113 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1114 
1115 /**
1116  * kthread_flush_worker - flush all current works on a kthread_worker
1117  * @worker: worker to flush
1118  *
1119  * Wait until all currently executing or pending works on @worker are
1120  * finished.
1121  */
1122 void kthread_flush_worker(struct kthread_worker *worker)
1123 {
1124 	struct kthread_flush_work fwork = {
1125 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1126 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1127 	};
1128 
1129 	kthread_queue_work(worker, &fwork.work);
1130 	wait_for_completion(&fwork.done);
1131 }
1132 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1133 
1134 /**
1135  * kthread_destroy_worker - destroy a kthread worker
1136  * @worker: worker to be destroyed
1137  *
1138  * Flush and destroy @worker.  The simple flush is enough because the kthread
1139  * worker API is used only in trivial scenarios.  There are no multi-step state
1140  * machines needed.
1141  */
1142 void kthread_destroy_worker(struct kthread_worker *worker)
1143 {
1144 	struct task_struct *task;
1145 
1146 	task = worker->task;
1147 	if (WARN_ON(!task))
1148 		return;
1149 
1150 	kthread_flush_worker(worker);
1151 	kthread_stop(task);
1152 	WARN_ON(!list_empty(&worker->work_list));
1153 	kfree(worker);
1154 }
1155 EXPORT_SYMBOL(kthread_destroy_worker);
1156 
1157 #ifdef CONFIG_BLK_CGROUP
1158 /**
1159  * kthread_associate_blkcg - associate blkcg to current kthread
1160  * @css: the cgroup info
1161  *
1162  * Current thread must be a kthread. The thread is running jobs on behalf of
1163  * other threads. In some cases, we expect the jobs attach cgroup info of
1164  * original threads instead of that of current thread. This function stores
1165  * original thread's cgroup info in current kthread context for later
1166  * retrieval.
1167  */
1168 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1169 {
1170 	struct kthread *kthread;
1171 
1172 	if (!(current->flags & PF_KTHREAD))
1173 		return;
1174 	kthread = to_kthread(current);
1175 	if (!kthread)
1176 		return;
1177 
1178 	if (kthread->blkcg_css) {
1179 		css_put(kthread->blkcg_css);
1180 		kthread->blkcg_css = NULL;
1181 	}
1182 	if (css) {
1183 		css_get(css);
1184 		kthread->blkcg_css = css;
1185 	}
1186 }
1187 EXPORT_SYMBOL(kthread_associate_blkcg);
1188 
1189 /**
1190  * kthread_blkcg - get associated blkcg css of current kthread
1191  *
1192  * Current thread must be a kthread.
1193  */
1194 struct cgroup_subsys_state *kthread_blkcg(void)
1195 {
1196 	struct kthread *kthread;
1197 
1198 	if (current->flags & PF_KTHREAD) {
1199 		kthread = to_kthread(current);
1200 		if (kthread)
1201 			return kthread->blkcg_css;
1202 	}
1203 	return NULL;
1204 }
1205 EXPORT_SYMBOL(kthread_blkcg);
1206 #endif
1207