]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - kernel/kthread.c
mm: replace all open encodings for NUMA_NO_NODE
[mirror_ubuntu-hirsute-kernel.git] / kernel / kthread.c
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/numa.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 #ifdef CONFIG_BLK_CGROUP
51 struct cgroup_subsys_state *blkcg_css;
52 #endif
53 };
54
55 enum KTHREAD_BITS {
56 KTHREAD_IS_PER_CPU = 0,
57 KTHREAD_SHOULD_STOP,
58 KTHREAD_SHOULD_PARK,
59 };
60
61 static inline void set_kthread_struct(void *kthread)
62 {
63 /*
64 * We abuse ->set_child_tid to avoid the new member and because it
65 * can't be wrongly copied by copy_process(). We also rely on fact
66 * that the caller can't exec, so PF_KTHREAD can't be cleared.
67 */
68 current->set_child_tid = (__force void __user *)kthread;
69 }
70
71 static inline struct kthread *to_kthread(struct task_struct *k)
72 {
73 WARN_ON(!(k->flags & PF_KTHREAD));
74 return (__force void *)k->set_child_tid;
75 }
76
77 void free_kthread_struct(struct task_struct *k)
78 {
79 struct kthread *kthread;
80
81 /*
82 * Can be NULL if this kthread was created by kernel_thread()
83 * or if kmalloc() in kthread() failed.
84 */
85 kthread = to_kthread(k);
86 #ifdef CONFIG_BLK_CGROUP
87 WARN_ON_ONCE(kthread && kthread->blkcg_css);
88 #endif
89 kfree(kthread);
90 }
91
92 /**
93 * kthread_should_stop - should this kthread return now?
94 *
95 * When someone calls kthread_stop() on your kthread, it will be woken
96 * and this will return true. You should then return, and your return
97 * value will be passed through to kthread_stop().
98 */
99 bool kthread_should_stop(void)
100 {
101 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
102 }
103 EXPORT_SYMBOL(kthread_should_stop);
104
105 /**
106 * kthread_should_park - should this kthread park now?
107 *
108 * When someone calls kthread_park() on your kthread, it will be woken
109 * and this will return true. You should then do the necessary
110 * cleanup and call kthread_parkme()
111 *
112 * Similar to kthread_should_stop(), but this keeps the thread alive
113 * and in a park position. kthread_unpark() "restarts" the thread and
114 * calls the thread function again.
115 */
116 bool kthread_should_park(void)
117 {
118 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
119 }
120 EXPORT_SYMBOL_GPL(kthread_should_park);
121
122 /**
123 * kthread_freezable_should_stop - should this freezable kthread return now?
124 * @was_frozen: optional out parameter, indicates whether %current was frozen
125 *
126 * kthread_should_stop() for freezable kthreads, which will enter
127 * refrigerator if necessary. This function is safe from kthread_stop() /
128 * freezer deadlock and freezable kthreads should use this function instead
129 * of calling try_to_freeze() directly.
130 */
131 bool kthread_freezable_should_stop(bool *was_frozen)
132 {
133 bool frozen = false;
134
135 might_sleep();
136
137 if (unlikely(freezing(current)))
138 frozen = __refrigerator(true);
139
140 if (was_frozen)
141 *was_frozen = frozen;
142
143 return kthread_should_stop();
144 }
145 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
146
147 /**
148 * kthread_data - return data value specified on kthread creation
149 * @task: kthread task in question
150 *
151 * Return the data value specified when kthread @task was created.
152 * The caller is responsible for ensuring the validity of @task when
153 * calling this function.
154 */
155 void *kthread_data(struct task_struct *task)
156 {
157 return to_kthread(task)->data;
158 }
159
160 /**
161 * kthread_probe_data - speculative version of kthread_data()
162 * @task: possible kthread task in question
163 *
164 * @task could be a kthread task. Return the data value specified when it
165 * was created if accessible. If @task isn't a kthread task or its data is
166 * inaccessible for any reason, %NULL is returned. This function requires
167 * that @task itself is safe to dereference.
168 */
169 void *kthread_probe_data(struct task_struct *task)
170 {
171 struct kthread *kthread = to_kthread(task);
172 void *data = NULL;
173
174 probe_kernel_read(&data, &kthread->data, sizeof(data));
175 return data;
176 }
177
178 static void __kthread_parkme(struct kthread *self)
179 {
180 for (;;) {
181 /*
182 * TASK_PARKED is a special state; we must serialize against
183 * possible pending wakeups to avoid store-store collisions on
184 * task->state.
185 *
186 * Such a collision might possibly result in the task state
187 * changin from TASK_PARKED and us failing the
188 * wait_task_inactive() in kthread_park().
189 */
190 set_special_state(TASK_PARKED);
191 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
192 break;
193
194 complete(&self->parked);
195 schedule();
196 }
197 __set_current_state(TASK_RUNNING);
198 }
199
200 void kthread_parkme(void)
201 {
202 __kthread_parkme(to_kthread(current));
203 }
204 EXPORT_SYMBOL_GPL(kthread_parkme);
205
206 static int kthread(void *_create)
207 {
208 /* Copy data: it's on kthread's stack */
209 struct kthread_create_info *create = _create;
210 int (*threadfn)(void *data) = create->threadfn;
211 void *data = create->data;
212 struct completion *done;
213 struct kthread *self;
214 int ret;
215
216 self = kzalloc(sizeof(*self), GFP_KERNEL);
217 set_kthread_struct(self);
218
219 /* If user was SIGKILLed, I release the structure. */
220 done = xchg(&create->done, NULL);
221 if (!done) {
222 kfree(create);
223 do_exit(-EINTR);
224 }
225
226 if (!self) {
227 create->result = ERR_PTR(-ENOMEM);
228 complete(done);
229 do_exit(-ENOMEM);
230 }
231
232 self->data = data;
233 init_completion(&self->exited);
234 init_completion(&self->parked);
235 current->vfork_done = &self->exited;
236
237 /* OK, tell user we're spawned, wait for stop or wakeup */
238 __set_current_state(TASK_UNINTERRUPTIBLE);
239 create->result = current;
240 complete(done);
241 schedule();
242
243 ret = -EINTR;
244 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
245 cgroup_kthread_ready();
246 __kthread_parkme(self);
247 ret = threadfn(data);
248 }
249 do_exit(ret);
250 }
251
252 /* called from do_fork() to get node information for about to be created task */
253 int tsk_fork_get_node(struct task_struct *tsk)
254 {
255 #ifdef CONFIG_NUMA
256 if (tsk == kthreadd_task)
257 return tsk->pref_node_fork;
258 #endif
259 return NUMA_NO_NODE;
260 }
261
262 static void create_kthread(struct kthread_create_info *create)
263 {
264 int pid;
265
266 #ifdef CONFIG_NUMA
267 current->pref_node_fork = create->node;
268 #endif
269 /* We want our own signal handler (we take no signals by default). */
270 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
271 if (pid < 0) {
272 /* If user was SIGKILLed, I release the structure. */
273 struct completion *done = xchg(&create->done, NULL);
274
275 if (!done) {
276 kfree(create);
277 return;
278 }
279 create->result = ERR_PTR(pid);
280 complete(done);
281 }
282 }
283
284 static __printf(4, 0)
285 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
286 void *data, int node,
287 const char namefmt[],
288 va_list args)
289 {
290 DECLARE_COMPLETION_ONSTACK(done);
291 struct task_struct *task;
292 struct kthread_create_info *create = kmalloc(sizeof(*create),
293 GFP_KERNEL);
294
295 if (!create)
296 return ERR_PTR(-ENOMEM);
297 create->threadfn = threadfn;
298 create->data = data;
299 create->node = node;
300 create->done = &done;
301
302 spin_lock(&kthread_create_lock);
303 list_add_tail(&create->list, &kthread_create_list);
304 spin_unlock(&kthread_create_lock);
305
306 wake_up_process(kthreadd_task);
307 /*
308 * Wait for completion in killable state, for I might be chosen by
309 * the OOM killer while kthreadd is trying to allocate memory for
310 * new kernel thread.
311 */
312 if (unlikely(wait_for_completion_killable(&done))) {
313 /*
314 * If I was SIGKILLed before kthreadd (or new kernel thread)
315 * calls complete(), leave the cleanup of this structure to
316 * that thread.
317 */
318 if (xchg(&create->done, NULL))
319 return ERR_PTR(-EINTR);
320 /*
321 * kthreadd (or new kernel thread) will call complete()
322 * shortly.
323 */
324 wait_for_completion(&done);
325 }
326 task = create->result;
327 if (!IS_ERR(task)) {
328 static const struct sched_param param = { .sched_priority = 0 };
329 char name[TASK_COMM_LEN];
330
331 /*
332 * task is already visible to other tasks, so updating
333 * COMM must be protected.
334 */
335 vsnprintf(name, sizeof(name), namefmt, args);
336 set_task_comm(task, name);
337 /*
338 * root may have changed our (kthreadd's) priority or CPU mask.
339 * The kernel thread should not inherit these properties.
340 */
341 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
342 set_cpus_allowed_ptr(task, cpu_all_mask);
343 }
344 kfree(create);
345 return task;
346 }
347
348 /**
349 * kthread_create_on_node - create a kthread.
350 * @threadfn: the function to run until signal_pending(current).
351 * @data: data ptr for @threadfn.
352 * @node: task and thread structures for the thread are allocated on this node
353 * @namefmt: printf-style name for the thread.
354 *
355 * Description: This helper function creates and names a kernel
356 * thread. The thread will be stopped: use wake_up_process() to start
357 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
358 * is affine to all CPUs.
359 *
360 * If thread is going to be bound on a particular cpu, give its node
361 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
362 * When woken, the thread will run @threadfn() with @data as its
363 * argument. @threadfn() can either call do_exit() directly if it is a
364 * standalone thread for which no one will call kthread_stop(), or
365 * return when 'kthread_should_stop()' is true (which means
366 * kthread_stop() has been called). The return value should be zero
367 * or a negative error number; it will be passed to kthread_stop().
368 *
369 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
370 */
371 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
372 void *data, int node,
373 const char namefmt[],
374 ...)
375 {
376 struct task_struct *task;
377 va_list args;
378
379 va_start(args, namefmt);
380 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
381 va_end(args);
382
383 return task;
384 }
385 EXPORT_SYMBOL(kthread_create_on_node);
386
387 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
388 {
389 unsigned long flags;
390
391 if (!wait_task_inactive(p, state)) {
392 WARN_ON(1);
393 return;
394 }
395
396 /* It's safe because the task is inactive. */
397 raw_spin_lock_irqsave(&p->pi_lock, flags);
398 do_set_cpus_allowed(p, mask);
399 p->flags |= PF_NO_SETAFFINITY;
400 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
401 }
402
403 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
404 {
405 __kthread_bind_mask(p, cpumask_of(cpu), state);
406 }
407
408 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
409 {
410 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
411 }
412
413 /**
414 * kthread_bind - bind a just-created kthread to a cpu.
415 * @p: thread created by kthread_create().
416 * @cpu: cpu (might not be online, must be possible) for @k to run on.
417 *
418 * Description: This function is equivalent to set_cpus_allowed(),
419 * except that @cpu doesn't need to be online, and the thread must be
420 * stopped (i.e., just returned from kthread_create()).
421 */
422 void kthread_bind(struct task_struct *p, unsigned int cpu)
423 {
424 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
425 }
426 EXPORT_SYMBOL(kthread_bind);
427
428 /**
429 * kthread_create_on_cpu - Create a cpu bound kthread
430 * @threadfn: the function to run until signal_pending(current).
431 * @data: data ptr for @threadfn.
432 * @cpu: The cpu on which the thread should be bound,
433 * @namefmt: printf-style name for the thread. Format is restricted
434 * to "name.*%u". Code fills in cpu number.
435 *
436 * Description: This helper function creates and names a kernel thread
437 * The thread will be woken and put into park mode.
438 */
439 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
440 void *data, unsigned int cpu,
441 const char *namefmt)
442 {
443 struct task_struct *p;
444
445 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
446 cpu);
447 if (IS_ERR(p))
448 return p;
449 kthread_bind(p, cpu);
450 /* CPU hotplug need to bind once again when unparking the thread. */
451 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
452 to_kthread(p)->cpu = cpu;
453 return p;
454 }
455
456 /**
457 * kthread_unpark - unpark a thread created by kthread_create().
458 * @k: thread created by kthread_create().
459 *
460 * Sets kthread_should_park() for @k to return false, wakes it, and
461 * waits for it to return. If the thread is marked percpu then its
462 * bound to the cpu again.
463 */
464 void kthread_unpark(struct task_struct *k)
465 {
466 struct kthread *kthread = to_kthread(k);
467
468 /*
469 * Newly created kthread was parked when the CPU was offline.
470 * The binding was lost and we need to set it again.
471 */
472 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
473 __kthread_bind(k, kthread->cpu, TASK_PARKED);
474
475 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
476 /*
477 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
478 */
479 wake_up_state(k, TASK_PARKED);
480 }
481 EXPORT_SYMBOL_GPL(kthread_unpark);
482
483 /**
484 * kthread_park - park a thread created by kthread_create().
485 * @k: thread created by kthread_create().
486 *
487 * Sets kthread_should_park() for @k to return true, wakes it, and
488 * waits for it to return. This can also be called after kthread_create()
489 * instead of calling wake_up_process(): the thread will park without
490 * calling threadfn().
491 *
492 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
493 * If called by the kthread itself just the park bit is set.
494 */
495 int kthread_park(struct task_struct *k)
496 {
497 struct kthread *kthread = to_kthread(k);
498
499 if (WARN_ON(k->flags & PF_EXITING))
500 return -ENOSYS;
501
502 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
503 return -EBUSY;
504
505 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
506 if (k != current) {
507 wake_up_process(k);
508 /*
509 * Wait for __kthread_parkme() to complete(), this means we
510 * _will_ have TASK_PARKED and are about to call schedule().
511 */
512 wait_for_completion(&kthread->parked);
513 /*
514 * Now wait for that schedule() to complete and the task to
515 * get scheduled out.
516 */
517 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
518 }
519
520 return 0;
521 }
522 EXPORT_SYMBOL_GPL(kthread_park);
523
524 /**
525 * kthread_stop - stop a thread created by kthread_create().
526 * @k: thread created by kthread_create().
527 *
528 * Sets kthread_should_stop() for @k to return true, wakes it, and
529 * waits for it to exit. This can also be called after kthread_create()
530 * instead of calling wake_up_process(): the thread will exit without
531 * calling threadfn().
532 *
533 * If threadfn() may call do_exit() itself, the caller must ensure
534 * task_struct can't go away.
535 *
536 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
537 * was never called.
538 */
539 int kthread_stop(struct task_struct *k)
540 {
541 struct kthread *kthread;
542 int ret;
543
544 trace_sched_kthread_stop(k);
545
546 get_task_struct(k);
547 kthread = to_kthread(k);
548 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
549 kthread_unpark(k);
550 wake_up_process(k);
551 wait_for_completion(&kthread->exited);
552 ret = k->exit_code;
553 put_task_struct(k);
554
555 trace_sched_kthread_stop_ret(ret);
556 return ret;
557 }
558 EXPORT_SYMBOL(kthread_stop);
559
560 int kthreadd(void *unused)
561 {
562 struct task_struct *tsk = current;
563
564 /* Setup a clean context for our children to inherit. */
565 set_task_comm(tsk, "kthreadd");
566 ignore_signals(tsk);
567 set_cpus_allowed_ptr(tsk, cpu_all_mask);
568 set_mems_allowed(node_states[N_MEMORY]);
569
570 current->flags |= PF_NOFREEZE;
571 cgroup_init_kthreadd();
572
573 for (;;) {
574 set_current_state(TASK_INTERRUPTIBLE);
575 if (list_empty(&kthread_create_list))
576 schedule();
577 __set_current_state(TASK_RUNNING);
578
579 spin_lock(&kthread_create_lock);
580 while (!list_empty(&kthread_create_list)) {
581 struct kthread_create_info *create;
582
583 create = list_entry(kthread_create_list.next,
584 struct kthread_create_info, list);
585 list_del_init(&create->list);
586 spin_unlock(&kthread_create_lock);
587
588 create_kthread(create);
589
590 spin_lock(&kthread_create_lock);
591 }
592 spin_unlock(&kthread_create_lock);
593 }
594
595 return 0;
596 }
597
598 void __kthread_init_worker(struct kthread_worker *worker,
599 const char *name,
600 struct lock_class_key *key)
601 {
602 memset(worker, 0, sizeof(struct kthread_worker));
603 spin_lock_init(&worker->lock);
604 lockdep_set_class_and_name(&worker->lock, key, name);
605 INIT_LIST_HEAD(&worker->work_list);
606 INIT_LIST_HEAD(&worker->delayed_work_list);
607 }
608 EXPORT_SYMBOL_GPL(__kthread_init_worker);
609
610 /**
611 * kthread_worker_fn - kthread function to process kthread_worker
612 * @worker_ptr: pointer to initialized kthread_worker
613 *
614 * This function implements the main cycle of kthread worker. It processes
615 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
616 * is empty.
617 *
618 * The works are not allowed to keep any locks, disable preemption or interrupts
619 * when they finish. There is defined a safe point for freezing when one work
620 * finishes and before a new one is started.
621 *
622 * Also the works must not be handled by more than one worker at the same time,
623 * see also kthread_queue_work().
624 */
625 int kthread_worker_fn(void *worker_ptr)
626 {
627 struct kthread_worker *worker = worker_ptr;
628 struct kthread_work *work;
629
630 /*
631 * FIXME: Update the check and remove the assignment when all kthread
632 * worker users are created using kthread_create_worker*() functions.
633 */
634 WARN_ON(worker->task && worker->task != current);
635 worker->task = current;
636
637 if (worker->flags & KTW_FREEZABLE)
638 set_freezable();
639
640 repeat:
641 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
642
643 if (kthread_should_stop()) {
644 __set_current_state(TASK_RUNNING);
645 spin_lock_irq(&worker->lock);
646 worker->task = NULL;
647 spin_unlock_irq(&worker->lock);
648 return 0;
649 }
650
651 work = NULL;
652 spin_lock_irq(&worker->lock);
653 if (!list_empty(&worker->work_list)) {
654 work = list_first_entry(&worker->work_list,
655 struct kthread_work, node);
656 list_del_init(&work->node);
657 }
658 worker->current_work = work;
659 spin_unlock_irq(&worker->lock);
660
661 if (work) {
662 __set_current_state(TASK_RUNNING);
663 work->func(work);
664 } else if (!freezing(current))
665 schedule();
666
667 try_to_freeze();
668 cond_resched();
669 goto repeat;
670 }
671 EXPORT_SYMBOL_GPL(kthread_worker_fn);
672
673 static __printf(3, 0) struct kthread_worker *
674 __kthread_create_worker(int cpu, unsigned int flags,
675 const char namefmt[], va_list args)
676 {
677 struct kthread_worker *worker;
678 struct task_struct *task;
679 int node = NUMA_NO_NODE;
680
681 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
682 if (!worker)
683 return ERR_PTR(-ENOMEM);
684
685 kthread_init_worker(worker);
686
687 if (cpu >= 0)
688 node = cpu_to_node(cpu);
689
690 task = __kthread_create_on_node(kthread_worker_fn, worker,
691 node, namefmt, args);
692 if (IS_ERR(task))
693 goto fail_task;
694
695 if (cpu >= 0)
696 kthread_bind(task, cpu);
697
698 worker->flags = flags;
699 worker->task = task;
700 wake_up_process(task);
701 return worker;
702
703 fail_task:
704 kfree(worker);
705 return ERR_CAST(task);
706 }
707
708 /**
709 * kthread_create_worker - create a kthread worker
710 * @flags: flags modifying the default behavior of the worker
711 * @namefmt: printf-style name for the kthread worker (task).
712 *
713 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
714 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
715 * when the worker was SIGKILLed.
716 */
717 struct kthread_worker *
718 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
719 {
720 struct kthread_worker *worker;
721 va_list args;
722
723 va_start(args, namefmt);
724 worker = __kthread_create_worker(-1, flags, namefmt, args);
725 va_end(args);
726
727 return worker;
728 }
729 EXPORT_SYMBOL(kthread_create_worker);
730
731 /**
732 * kthread_create_worker_on_cpu - create a kthread worker and bind it
733 * it to a given CPU and the associated NUMA node.
734 * @cpu: CPU number
735 * @flags: flags modifying the default behavior of the worker
736 * @namefmt: printf-style name for the kthread worker (task).
737 *
738 * Use a valid CPU number if you want to bind the kthread worker
739 * to the given CPU and the associated NUMA node.
740 *
741 * A good practice is to add the cpu number also into the worker name.
742 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
743 *
744 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
745 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
746 * when the worker was SIGKILLed.
747 */
748 struct kthread_worker *
749 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
750 const char namefmt[], ...)
751 {
752 struct kthread_worker *worker;
753 va_list args;
754
755 va_start(args, namefmt);
756 worker = __kthread_create_worker(cpu, flags, namefmt, args);
757 va_end(args);
758
759 return worker;
760 }
761 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
762
763 /*
764 * Returns true when the work could not be queued at the moment.
765 * It happens when it is already pending in a worker list
766 * or when it is being cancelled.
767 */
768 static inline bool queuing_blocked(struct kthread_worker *worker,
769 struct kthread_work *work)
770 {
771 lockdep_assert_held(&worker->lock);
772
773 return !list_empty(&work->node) || work->canceling;
774 }
775
776 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
777 struct kthread_work *work)
778 {
779 lockdep_assert_held(&worker->lock);
780 WARN_ON_ONCE(!list_empty(&work->node));
781 /* Do not use a work with >1 worker, see kthread_queue_work() */
782 WARN_ON_ONCE(work->worker && work->worker != worker);
783 }
784
785 /* insert @work before @pos in @worker */
786 static void kthread_insert_work(struct kthread_worker *worker,
787 struct kthread_work *work,
788 struct list_head *pos)
789 {
790 kthread_insert_work_sanity_check(worker, work);
791
792 list_add_tail(&work->node, pos);
793 work->worker = worker;
794 if (!worker->current_work && likely(worker->task))
795 wake_up_process(worker->task);
796 }
797
798 /**
799 * kthread_queue_work - queue a kthread_work
800 * @worker: target kthread_worker
801 * @work: kthread_work to queue
802 *
803 * Queue @work to work processor @task for async execution. @task
804 * must have been created with kthread_worker_create(). Returns %true
805 * if @work was successfully queued, %false if it was already pending.
806 *
807 * Reinitialize the work if it needs to be used by another worker.
808 * For example, when the worker was stopped and started again.
809 */
810 bool kthread_queue_work(struct kthread_worker *worker,
811 struct kthread_work *work)
812 {
813 bool ret = false;
814 unsigned long flags;
815
816 spin_lock_irqsave(&worker->lock, flags);
817 if (!queuing_blocked(worker, work)) {
818 kthread_insert_work(worker, work, &worker->work_list);
819 ret = true;
820 }
821 spin_unlock_irqrestore(&worker->lock, flags);
822 return ret;
823 }
824 EXPORT_SYMBOL_GPL(kthread_queue_work);
825
826 /**
827 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
828 * delayed work when the timer expires.
829 * @t: pointer to the expired timer
830 *
831 * The format of the function is defined by struct timer_list.
832 * It should have been called from irqsafe timer with irq already off.
833 */
834 void kthread_delayed_work_timer_fn(struct timer_list *t)
835 {
836 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
837 struct kthread_work *work = &dwork->work;
838 struct kthread_worker *worker = work->worker;
839
840 /*
841 * This might happen when a pending work is reinitialized.
842 * It means that it is used a wrong way.
843 */
844 if (WARN_ON_ONCE(!worker))
845 return;
846
847 spin_lock(&worker->lock);
848 /* Work must not be used with >1 worker, see kthread_queue_work(). */
849 WARN_ON_ONCE(work->worker != worker);
850
851 /* Move the work from worker->delayed_work_list. */
852 WARN_ON_ONCE(list_empty(&work->node));
853 list_del_init(&work->node);
854 kthread_insert_work(worker, work, &worker->work_list);
855
856 spin_unlock(&worker->lock);
857 }
858 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
859
860 void __kthread_queue_delayed_work(struct kthread_worker *worker,
861 struct kthread_delayed_work *dwork,
862 unsigned long delay)
863 {
864 struct timer_list *timer = &dwork->timer;
865 struct kthread_work *work = &dwork->work;
866
867 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
868
869 /*
870 * If @delay is 0, queue @dwork->work immediately. This is for
871 * both optimization and correctness. The earliest @timer can
872 * expire is on the closest next tick and delayed_work users depend
873 * on that there's no such delay when @delay is 0.
874 */
875 if (!delay) {
876 kthread_insert_work(worker, work, &worker->work_list);
877 return;
878 }
879
880 /* Be paranoid and try to detect possible races already now. */
881 kthread_insert_work_sanity_check(worker, work);
882
883 list_add(&work->node, &worker->delayed_work_list);
884 work->worker = worker;
885 timer->expires = jiffies + delay;
886 add_timer(timer);
887 }
888
889 /**
890 * kthread_queue_delayed_work - queue the associated kthread work
891 * after a delay.
892 * @worker: target kthread_worker
893 * @dwork: kthread_delayed_work to queue
894 * @delay: number of jiffies to wait before queuing
895 *
896 * If the work has not been pending it starts a timer that will queue
897 * the work after the given @delay. If @delay is zero, it queues the
898 * work immediately.
899 *
900 * Return: %false if the @work has already been pending. It means that
901 * either the timer was running or the work was queued. It returns %true
902 * otherwise.
903 */
904 bool kthread_queue_delayed_work(struct kthread_worker *worker,
905 struct kthread_delayed_work *dwork,
906 unsigned long delay)
907 {
908 struct kthread_work *work = &dwork->work;
909 unsigned long flags;
910 bool ret = false;
911
912 spin_lock_irqsave(&worker->lock, flags);
913
914 if (!queuing_blocked(worker, work)) {
915 __kthread_queue_delayed_work(worker, dwork, delay);
916 ret = true;
917 }
918
919 spin_unlock_irqrestore(&worker->lock, flags);
920 return ret;
921 }
922 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
923
924 struct kthread_flush_work {
925 struct kthread_work work;
926 struct completion done;
927 };
928
929 static void kthread_flush_work_fn(struct kthread_work *work)
930 {
931 struct kthread_flush_work *fwork =
932 container_of(work, struct kthread_flush_work, work);
933 complete(&fwork->done);
934 }
935
936 /**
937 * kthread_flush_work - flush a kthread_work
938 * @work: work to flush
939 *
940 * If @work is queued or executing, wait for it to finish execution.
941 */
942 void kthread_flush_work(struct kthread_work *work)
943 {
944 struct kthread_flush_work fwork = {
945 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
946 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
947 };
948 struct kthread_worker *worker;
949 bool noop = false;
950
951 worker = work->worker;
952 if (!worker)
953 return;
954
955 spin_lock_irq(&worker->lock);
956 /* Work must not be used with >1 worker, see kthread_queue_work(). */
957 WARN_ON_ONCE(work->worker != worker);
958
959 if (!list_empty(&work->node))
960 kthread_insert_work(worker, &fwork.work, work->node.next);
961 else if (worker->current_work == work)
962 kthread_insert_work(worker, &fwork.work,
963 worker->work_list.next);
964 else
965 noop = true;
966
967 spin_unlock_irq(&worker->lock);
968
969 if (!noop)
970 wait_for_completion(&fwork.done);
971 }
972 EXPORT_SYMBOL_GPL(kthread_flush_work);
973
974 /*
975 * This function removes the work from the worker queue. Also it makes sure
976 * that it won't get queued later via the delayed work's timer.
977 *
978 * The work might still be in use when this function finishes. See the
979 * current_work proceed by the worker.
980 *
981 * Return: %true if @work was pending and successfully canceled,
982 * %false if @work was not pending
983 */
984 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
985 unsigned long *flags)
986 {
987 /* Try to cancel the timer if exists. */
988 if (is_dwork) {
989 struct kthread_delayed_work *dwork =
990 container_of(work, struct kthread_delayed_work, work);
991 struct kthread_worker *worker = work->worker;
992
993 /*
994 * del_timer_sync() must be called to make sure that the timer
995 * callback is not running. The lock must be temporary released
996 * to avoid a deadlock with the callback. In the meantime,
997 * any queuing is blocked by setting the canceling counter.
998 */
999 work->canceling++;
1000 spin_unlock_irqrestore(&worker->lock, *flags);
1001 del_timer_sync(&dwork->timer);
1002 spin_lock_irqsave(&worker->lock, *flags);
1003 work->canceling--;
1004 }
1005
1006 /*
1007 * Try to remove the work from a worker list. It might either
1008 * be from worker->work_list or from worker->delayed_work_list.
1009 */
1010 if (!list_empty(&work->node)) {
1011 list_del_init(&work->node);
1012 return true;
1013 }
1014
1015 return false;
1016 }
1017
1018 /**
1019 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1020 * @worker: kthread worker to use
1021 * @dwork: kthread delayed work to queue
1022 * @delay: number of jiffies to wait before queuing
1023 *
1024 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1025 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1026 * @work is guaranteed to be queued immediately.
1027 *
1028 * Return: %true if @dwork was pending and its timer was modified,
1029 * %false otherwise.
1030 *
1031 * A special case is when the work is being canceled in parallel.
1032 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1033 * or yet another kthread_mod_delayed_work() call. We let the other command
1034 * win and return %false here. The caller is supposed to synchronize these
1035 * operations a reasonable way.
1036 *
1037 * This function is safe to call from any context including IRQ handler.
1038 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1039 * for details.
1040 */
1041 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1042 struct kthread_delayed_work *dwork,
1043 unsigned long delay)
1044 {
1045 struct kthread_work *work = &dwork->work;
1046 unsigned long flags;
1047 int ret = false;
1048
1049 spin_lock_irqsave(&worker->lock, flags);
1050
1051 /* Do not bother with canceling when never queued. */
1052 if (!work->worker)
1053 goto fast_queue;
1054
1055 /* Work must not be used with >1 worker, see kthread_queue_work() */
1056 WARN_ON_ONCE(work->worker != worker);
1057
1058 /* Do not fight with another command that is canceling this work. */
1059 if (work->canceling)
1060 goto out;
1061
1062 ret = __kthread_cancel_work(work, true, &flags);
1063 fast_queue:
1064 __kthread_queue_delayed_work(worker, dwork, delay);
1065 out:
1066 spin_unlock_irqrestore(&worker->lock, flags);
1067 return ret;
1068 }
1069 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1070
1071 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1072 {
1073 struct kthread_worker *worker = work->worker;
1074 unsigned long flags;
1075 int ret = false;
1076
1077 if (!worker)
1078 goto out;
1079
1080 spin_lock_irqsave(&worker->lock, flags);
1081 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1082 WARN_ON_ONCE(work->worker != worker);
1083
1084 ret = __kthread_cancel_work(work, is_dwork, &flags);
1085
1086 if (worker->current_work != work)
1087 goto out_fast;
1088
1089 /*
1090 * The work is in progress and we need to wait with the lock released.
1091 * In the meantime, block any queuing by setting the canceling counter.
1092 */
1093 work->canceling++;
1094 spin_unlock_irqrestore(&worker->lock, flags);
1095 kthread_flush_work(work);
1096 spin_lock_irqsave(&worker->lock, flags);
1097 work->canceling--;
1098
1099 out_fast:
1100 spin_unlock_irqrestore(&worker->lock, flags);
1101 out:
1102 return ret;
1103 }
1104
1105 /**
1106 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1107 * @work: the kthread work to cancel
1108 *
1109 * Cancel @work and wait for its execution to finish. This function
1110 * can be used even if the work re-queues itself. On return from this
1111 * function, @work is guaranteed to be not pending or executing on any CPU.
1112 *
1113 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1114 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1115 *
1116 * The caller must ensure that the worker on which @work was last
1117 * queued can't be destroyed before this function returns.
1118 *
1119 * Return: %true if @work was pending, %false otherwise.
1120 */
1121 bool kthread_cancel_work_sync(struct kthread_work *work)
1122 {
1123 return __kthread_cancel_work_sync(work, false);
1124 }
1125 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1126
1127 /**
1128 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1129 * wait for it to finish.
1130 * @dwork: the kthread delayed work to cancel
1131 *
1132 * This is kthread_cancel_work_sync() for delayed works.
1133 *
1134 * Return: %true if @dwork was pending, %false otherwise.
1135 */
1136 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1137 {
1138 return __kthread_cancel_work_sync(&dwork->work, true);
1139 }
1140 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1141
1142 /**
1143 * kthread_flush_worker - flush all current works on a kthread_worker
1144 * @worker: worker to flush
1145 *
1146 * Wait until all currently executing or pending works on @worker are
1147 * finished.
1148 */
1149 void kthread_flush_worker(struct kthread_worker *worker)
1150 {
1151 struct kthread_flush_work fwork = {
1152 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1153 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1154 };
1155
1156 kthread_queue_work(worker, &fwork.work);
1157 wait_for_completion(&fwork.done);
1158 }
1159 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1160
1161 /**
1162 * kthread_destroy_worker - destroy a kthread worker
1163 * @worker: worker to be destroyed
1164 *
1165 * Flush and destroy @worker. The simple flush is enough because the kthread
1166 * worker API is used only in trivial scenarios. There are no multi-step state
1167 * machines needed.
1168 */
1169 void kthread_destroy_worker(struct kthread_worker *worker)
1170 {
1171 struct task_struct *task;
1172
1173 task = worker->task;
1174 if (WARN_ON(!task))
1175 return;
1176
1177 kthread_flush_worker(worker);
1178 kthread_stop(task);
1179 WARN_ON(!list_empty(&worker->work_list));
1180 kfree(worker);
1181 }
1182 EXPORT_SYMBOL(kthread_destroy_worker);
1183
1184 #ifdef CONFIG_BLK_CGROUP
1185 /**
1186 * kthread_associate_blkcg - associate blkcg to current kthread
1187 * @css: the cgroup info
1188 *
1189 * Current thread must be a kthread. The thread is running jobs on behalf of
1190 * other threads. In some cases, we expect the jobs attach cgroup info of
1191 * original threads instead of that of current thread. This function stores
1192 * original thread's cgroup info in current kthread context for later
1193 * retrieval.
1194 */
1195 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1196 {
1197 struct kthread *kthread;
1198
1199 if (!(current->flags & PF_KTHREAD))
1200 return;
1201 kthread = to_kthread(current);
1202 if (!kthread)
1203 return;
1204
1205 if (kthread->blkcg_css) {
1206 css_put(kthread->blkcg_css);
1207 kthread->blkcg_css = NULL;
1208 }
1209 if (css) {
1210 css_get(css);
1211 kthread->blkcg_css = css;
1212 }
1213 }
1214 EXPORT_SYMBOL(kthread_associate_blkcg);
1215
1216 /**
1217 * kthread_blkcg - get associated blkcg css of current kthread
1218 *
1219 * Current thread must be a kthread.
1220 */
1221 struct cgroup_subsys_state *kthread_blkcg(void)
1222 {
1223 struct kthread *kthread;
1224
1225 if (current->flags & PF_KTHREAD) {
1226 kthread = to_kthread(current);
1227 if (kthread)
1228 return kthread->blkcg_css;
1229 }
1230 return NULL;
1231 }
1232 EXPORT_SYMBOL(kthread_blkcg);
1233 #endif