]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/kthread.c
x86/kvm: Update spectre-v1 mitigation
[mirror_ubuntu-bionic-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 <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 KTHREAD_IS_PARKED,
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 __set_current_state(TASK_PARKED);
181 while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
182 if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
183 complete(&self->parked);
184 schedule();
185 __set_current_state(TASK_PARKED);
186 }
187 clear_bit(KTHREAD_IS_PARKED, &self->flags);
188 __set_current_state(TASK_RUNNING);
189 }
190
191 void kthread_parkme(void)
192 {
193 __kthread_parkme(to_kthread(current));
194 }
195 EXPORT_SYMBOL_GPL(kthread_parkme);
196
197 static int kthread(void *_create)
198 {
199 /* Copy data: it's on kthread's stack */
200 struct kthread_create_info *create = _create;
201 int (*threadfn)(void *data) = create->threadfn;
202 void *data = create->data;
203 struct completion *done;
204 struct kthread *self;
205 int ret;
206
207 self = kzalloc(sizeof(*self), GFP_KERNEL);
208 set_kthread_struct(self);
209
210 /* If user was SIGKILLed, I release the structure. */
211 done = xchg(&create->done, NULL);
212 if (!done) {
213 kfree(create);
214 do_exit(-EINTR);
215 }
216
217 if (!self) {
218 create->result = ERR_PTR(-ENOMEM);
219 complete(done);
220 do_exit(-ENOMEM);
221 }
222
223 self->data = data;
224 init_completion(&self->exited);
225 init_completion(&self->parked);
226 current->vfork_done = &self->exited;
227
228 /* OK, tell user we're spawned, wait for stop or wakeup */
229 __set_current_state(TASK_UNINTERRUPTIBLE);
230 create->result = current;
231 complete(done);
232 schedule();
233
234 ret = -EINTR;
235 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
236 cgroup_kthread_ready();
237 __kthread_parkme(self);
238 ret = threadfn(data);
239 }
240 do_exit(ret);
241 }
242
243 /* called from do_fork() to get node information for about to be created task */
244 int tsk_fork_get_node(struct task_struct *tsk)
245 {
246 #ifdef CONFIG_NUMA
247 if (tsk == kthreadd_task)
248 return tsk->pref_node_fork;
249 #endif
250 return NUMA_NO_NODE;
251 }
252
253 static void create_kthread(struct kthread_create_info *create)
254 {
255 int pid;
256
257 #ifdef CONFIG_NUMA
258 current->pref_node_fork = create->node;
259 #endif
260 /* We want our own signal handler (we take no signals by default). */
261 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
262 if (pid < 0) {
263 /* If user was SIGKILLed, I release the structure. */
264 struct completion *done = xchg(&create->done, NULL);
265
266 if (!done) {
267 kfree(create);
268 return;
269 }
270 create->result = ERR_PTR(pid);
271 complete(done);
272 }
273 }
274
275 static __printf(4, 0)
276 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
277 void *data, int node,
278 const char namefmt[],
279 va_list args)
280 {
281 DECLARE_COMPLETION_ONSTACK(done);
282 struct task_struct *task;
283 struct kthread_create_info *create = kmalloc(sizeof(*create),
284 GFP_KERNEL);
285
286 if (!create)
287 return ERR_PTR(-ENOMEM);
288 create->threadfn = threadfn;
289 create->data = data;
290 create->node = node;
291 create->done = &done;
292
293 spin_lock(&kthread_create_lock);
294 list_add_tail(&create->list, &kthread_create_list);
295 spin_unlock(&kthread_create_lock);
296
297 wake_up_process(kthreadd_task);
298 /*
299 * Wait for completion in killable state, for I might be chosen by
300 * the OOM killer while kthreadd is trying to allocate memory for
301 * new kernel thread.
302 */
303 if (unlikely(wait_for_completion_killable(&done))) {
304 int i = 0;
305
306 /*
307 * I got SIGKILL, but wait for 10 more seconds for completion
308 * unless chosen by the OOM killer. This delay is there as a
309 * workaround for boot failure caused by SIGKILL upon device
310 * driver initialization timeout.
311 */
312 while (i++ < 10 && !test_tsk_thread_flag(current, TIF_MEMDIE))
313 if (wait_for_completion_timeout(&done, HZ))
314 goto ready;
315 /*
316 * If I was SIGKILLed before kthreadd (or new kernel thread)
317 * calls complete(), leave the cleanup of this structure to
318 * that thread.
319 */
320 if (xchg(&create->done, NULL))
321 return ERR_PTR(-EINTR);
322 /*
323 * kthreadd (or new kernel thread) will call complete()
324 * shortly.
325 */
326 wait_for_completion(&done);
327 }
328 ready:
329 task = create->result;
330 if (!IS_ERR(task)) {
331 static const struct sched_param param = { .sched_priority = 0 };
332
333 vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
334 /*
335 * root may have changed our (kthreadd's) priority or CPU mask.
336 * The kernel thread should not inherit these properties.
337 */
338 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
339 set_cpus_allowed_ptr(task, cpu_all_mask);
340 }
341 kfree(create);
342 return task;
343 }
344
345 /**
346 * kthread_create_on_node - create a kthread.
347 * @threadfn: the function to run until signal_pending(current).
348 * @data: data ptr for @threadfn.
349 * @node: task and thread structures for the thread are allocated on this node
350 * @namefmt: printf-style name for the thread.
351 *
352 * Description: This helper function creates and names a kernel
353 * thread. The thread will be stopped: use wake_up_process() to start
354 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
355 * is affine to all CPUs.
356 *
357 * If thread is going to be bound on a particular cpu, give its node
358 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
359 * When woken, the thread will run @threadfn() with @data as its
360 * argument. @threadfn() can either call do_exit() directly if it is a
361 * standalone thread for which no one will call kthread_stop(), or
362 * return when 'kthread_should_stop()' is true (which means
363 * kthread_stop() has been called). The return value should be zero
364 * or a negative error number; it will be passed to kthread_stop().
365 *
366 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
367 */
368 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
369 void *data, int node,
370 const char namefmt[],
371 ...)
372 {
373 struct task_struct *task;
374 va_list args;
375
376 va_start(args, namefmt);
377 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
378 va_end(args);
379
380 return task;
381 }
382 EXPORT_SYMBOL(kthread_create_on_node);
383
384 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
385 {
386 unsigned long flags;
387
388 if (!wait_task_inactive(p, state)) {
389 WARN_ON(1);
390 return;
391 }
392
393 /* It's safe because the task is inactive. */
394 raw_spin_lock_irqsave(&p->pi_lock, flags);
395 do_set_cpus_allowed(p, mask);
396 p->flags |= PF_NO_SETAFFINITY;
397 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
398 }
399
400 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
401 {
402 __kthread_bind_mask(p, cpumask_of(cpu), state);
403 }
404
405 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
406 {
407 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
408 }
409
410 /**
411 * kthread_bind - bind a just-created kthread to a cpu.
412 * @p: thread created by kthread_create().
413 * @cpu: cpu (might not be online, must be possible) for @k to run on.
414 *
415 * Description: This function is equivalent to set_cpus_allowed(),
416 * except that @cpu doesn't need to be online, and the thread must be
417 * stopped (i.e., just returned from kthread_create()).
418 */
419 void kthread_bind(struct task_struct *p, unsigned int cpu)
420 {
421 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
422 }
423 EXPORT_SYMBOL(kthread_bind);
424
425 /**
426 * kthread_create_on_cpu - Create a cpu bound kthread
427 * @threadfn: the function to run until signal_pending(current).
428 * @data: data ptr for @threadfn.
429 * @cpu: The cpu on which the thread should be bound,
430 * @namefmt: printf-style name for the thread. Format is restricted
431 * to "name.*%u". Code fills in cpu number.
432 *
433 * Description: This helper function creates and names a kernel thread
434 * The thread will be woken and put into park mode.
435 */
436 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
437 void *data, unsigned int cpu,
438 const char *namefmt)
439 {
440 struct task_struct *p;
441
442 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
443 cpu);
444 if (IS_ERR(p))
445 return p;
446 kthread_bind(p, cpu);
447 /* CPU hotplug need to bind once again when unparking the thread. */
448 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
449 to_kthread(p)->cpu = cpu;
450 return p;
451 }
452
453 /**
454 * kthread_unpark - unpark a thread created by kthread_create().
455 * @k: thread created by kthread_create().
456 *
457 * Sets kthread_should_park() for @k to return false, wakes it, and
458 * waits for it to return. If the thread is marked percpu then its
459 * bound to the cpu again.
460 */
461 void kthread_unpark(struct task_struct *k)
462 {
463 struct kthread *kthread = to_kthread(k);
464
465 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
466 /*
467 * We clear the IS_PARKED bit here as we don't wait
468 * until the task has left the park code. So if we'd
469 * park before that happens we'd see the IS_PARKED bit
470 * which might be about to be cleared.
471 */
472 if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
473 /*
474 * Newly created kthread was parked when the CPU was offline.
475 * The binding was lost and we need to set it again.
476 */
477 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
478 __kthread_bind(k, kthread->cpu, TASK_PARKED);
479 wake_up_state(k, TASK_PARKED);
480 }
481 }
482 EXPORT_SYMBOL_GPL(kthread_unpark);
483
484 /**
485 * kthread_park - park a thread created by kthread_create().
486 * @k: thread created by kthread_create().
487 *
488 * Sets kthread_should_park() for @k to return true, wakes it, and
489 * waits for it to return. This can also be called after kthread_create()
490 * instead of calling wake_up_process(): the thread will park without
491 * calling threadfn().
492 *
493 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
494 * If called by the kthread itself just the park bit is set.
495 */
496 int kthread_park(struct task_struct *k)
497 {
498 struct kthread *kthread = to_kthread(k);
499
500 if (WARN_ON(k->flags & PF_EXITING))
501 return -ENOSYS;
502
503 if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
504 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
505 if (k != current) {
506 wake_up_process(k);
507 wait_for_completion(&kthread->parked);
508 }
509 }
510
511 return 0;
512 }
513 EXPORT_SYMBOL_GPL(kthread_park);
514
515 /**
516 * kthread_stop - stop a thread created by kthread_create().
517 * @k: thread created by kthread_create().
518 *
519 * Sets kthread_should_stop() for @k to return true, wakes it, and
520 * waits for it to exit. This can also be called after kthread_create()
521 * instead of calling wake_up_process(): the thread will exit without
522 * calling threadfn().
523 *
524 * If threadfn() may call do_exit() itself, the caller must ensure
525 * task_struct can't go away.
526 *
527 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
528 * was never called.
529 */
530 int kthread_stop(struct task_struct *k)
531 {
532 struct kthread *kthread;
533 int ret;
534
535 trace_sched_kthread_stop(k);
536
537 get_task_struct(k);
538 kthread = to_kthread(k);
539 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
540 kthread_unpark(k);
541 wake_up_process(k);
542 wait_for_completion(&kthread->exited);
543 ret = k->exit_code;
544 put_task_struct(k);
545
546 trace_sched_kthread_stop_ret(ret);
547 return ret;
548 }
549 EXPORT_SYMBOL(kthread_stop);
550
551 int kthreadd(void *unused)
552 {
553 struct task_struct *tsk = current;
554
555 /* Setup a clean context for our children to inherit. */
556 set_task_comm(tsk, "kthreadd");
557 ignore_signals(tsk);
558 set_cpus_allowed_ptr(tsk, cpu_all_mask);
559 set_mems_allowed(node_states[N_MEMORY]);
560
561 current->flags |= PF_NOFREEZE;
562 cgroup_init_kthreadd();
563
564 for (;;) {
565 set_current_state(TASK_INTERRUPTIBLE);
566 if (list_empty(&kthread_create_list))
567 schedule();
568 __set_current_state(TASK_RUNNING);
569
570 spin_lock(&kthread_create_lock);
571 while (!list_empty(&kthread_create_list)) {
572 struct kthread_create_info *create;
573
574 create = list_entry(kthread_create_list.next,
575 struct kthread_create_info, list);
576 list_del_init(&create->list);
577 spin_unlock(&kthread_create_lock);
578
579 create_kthread(create);
580
581 spin_lock(&kthread_create_lock);
582 }
583 spin_unlock(&kthread_create_lock);
584 }
585
586 return 0;
587 }
588
589 void __kthread_init_worker(struct kthread_worker *worker,
590 const char *name,
591 struct lock_class_key *key)
592 {
593 memset(worker, 0, sizeof(struct kthread_worker));
594 spin_lock_init(&worker->lock);
595 lockdep_set_class_and_name(&worker->lock, key, name);
596 INIT_LIST_HEAD(&worker->work_list);
597 INIT_LIST_HEAD(&worker->delayed_work_list);
598 }
599 EXPORT_SYMBOL_GPL(__kthread_init_worker);
600
601 /**
602 * kthread_worker_fn - kthread function to process kthread_worker
603 * @worker_ptr: pointer to initialized kthread_worker
604 *
605 * This function implements the main cycle of kthread worker. It processes
606 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
607 * is empty.
608 *
609 * The works are not allowed to keep any locks, disable preemption or interrupts
610 * when they finish. There is defined a safe point for freezing when one work
611 * finishes and before a new one is started.
612 *
613 * Also the works must not be handled by more than one worker at the same time,
614 * see also kthread_queue_work().
615 */
616 int kthread_worker_fn(void *worker_ptr)
617 {
618 struct kthread_worker *worker = worker_ptr;
619 struct kthread_work *work;
620
621 /*
622 * FIXME: Update the check and remove the assignment when all kthread
623 * worker users are created using kthread_create_worker*() functions.
624 */
625 WARN_ON(worker->task && worker->task != current);
626 worker->task = current;
627
628 if (worker->flags & KTW_FREEZABLE)
629 set_freezable();
630
631 repeat:
632 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
633
634 if (kthread_should_stop()) {
635 __set_current_state(TASK_RUNNING);
636 spin_lock_irq(&worker->lock);
637 worker->task = NULL;
638 spin_unlock_irq(&worker->lock);
639 return 0;
640 }
641
642 work = NULL;
643 spin_lock_irq(&worker->lock);
644 if (!list_empty(&worker->work_list)) {
645 work = list_first_entry(&worker->work_list,
646 struct kthread_work, node);
647 list_del_init(&work->node);
648 }
649 worker->current_work = work;
650 spin_unlock_irq(&worker->lock);
651
652 if (work) {
653 __set_current_state(TASK_RUNNING);
654 work->func(work);
655 } else if (!freezing(current))
656 schedule();
657
658 try_to_freeze();
659 cond_resched();
660 goto repeat;
661 }
662 EXPORT_SYMBOL_GPL(kthread_worker_fn);
663
664 static __printf(3, 0) struct kthread_worker *
665 __kthread_create_worker(int cpu, unsigned int flags,
666 const char namefmt[], va_list args)
667 {
668 struct kthread_worker *worker;
669 struct task_struct *task;
670 int node = -1;
671
672 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
673 if (!worker)
674 return ERR_PTR(-ENOMEM);
675
676 kthread_init_worker(worker);
677
678 if (cpu >= 0)
679 node = cpu_to_node(cpu);
680
681 task = __kthread_create_on_node(kthread_worker_fn, worker,
682 node, namefmt, args);
683 if (IS_ERR(task))
684 goto fail_task;
685
686 if (cpu >= 0)
687 kthread_bind(task, cpu);
688
689 worker->flags = flags;
690 worker->task = task;
691 wake_up_process(task);
692 return worker;
693
694 fail_task:
695 kfree(worker);
696 return ERR_CAST(task);
697 }
698
699 /**
700 * kthread_create_worker - create a kthread worker
701 * @flags: flags modifying the default behavior of the worker
702 * @namefmt: printf-style name for the kthread worker (task).
703 *
704 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
705 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
706 * when the worker was SIGKILLed.
707 */
708 struct kthread_worker *
709 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
710 {
711 struct kthread_worker *worker;
712 va_list args;
713
714 va_start(args, namefmt);
715 worker = __kthread_create_worker(-1, flags, namefmt, args);
716 va_end(args);
717
718 return worker;
719 }
720 EXPORT_SYMBOL(kthread_create_worker);
721
722 /**
723 * kthread_create_worker_on_cpu - create a kthread worker and bind it
724 * it to a given CPU and the associated NUMA node.
725 * @cpu: CPU number
726 * @flags: flags modifying the default behavior of the worker
727 * @namefmt: printf-style name for the kthread worker (task).
728 *
729 * Use a valid CPU number if you want to bind the kthread worker
730 * to the given CPU and the associated NUMA node.
731 *
732 * A good practice is to add the cpu number also into the worker name.
733 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
734 *
735 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
736 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
737 * when the worker was SIGKILLed.
738 */
739 struct kthread_worker *
740 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
741 const char namefmt[], ...)
742 {
743 struct kthread_worker *worker;
744 va_list args;
745
746 va_start(args, namefmt);
747 worker = __kthread_create_worker(cpu, flags, namefmt, args);
748 va_end(args);
749
750 return worker;
751 }
752 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
753
754 /*
755 * Returns true when the work could not be queued at the moment.
756 * It happens when it is already pending in a worker list
757 * or when it is being cancelled.
758 */
759 static inline bool queuing_blocked(struct kthread_worker *worker,
760 struct kthread_work *work)
761 {
762 lockdep_assert_held(&worker->lock);
763
764 return !list_empty(&work->node) || work->canceling;
765 }
766
767 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
768 struct kthread_work *work)
769 {
770 lockdep_assert_held(&worker->lock);
771 WARN_ON_ONCE(!list_empty(&work->node));
772 /* Do not use a work with >1 worker, see kthread_queue_work() */
773 WARN_ON_ONCE(work->worker && work->worker != worker);
774 }
775
776 /* insert @work before @pos in @worker */
777 static void kthread_insert_work(struct kthread_worker *worker,
778 struct kthread_work *work,
779 struct list_head *pos)
780 {
781 kthread_insert_work_sanity_check(worker, work);
782
783 list_add_tail(&work->node, pos);
784 work->worker = worker;
785 if (!worker->current_work && likely(worker->task))
786 wake_up_process(worker->task);
787 }
788
789 /**
790 * kthread_queue_work - queue a kthread_work
791 * @worker: target kthread_worker
792 * @work: kthread_work to queue
793 *
794 * Queue @work to work processor @task for async execution. @task
795 * must have been created with kthread_worker_create(). Returns %true
796 * if @work was successfully queued, %false if it was already pending.
797 *
798 * Reinitialize the work if it needs to be used by another worker.
799 * For example, when the worker was stopped and started again.
800 */
801 bool kthread_queue_work(struct kthread_worker *worker,
802 struct kthread_work *work)
803 {
804 bool ret = false;
805 unsigned long flags;
806
807 spin_lock_irqsave(&worker->lock, flags);
808 if (!queuing_blocked(worker, work)) {
809 kthread_insert_work(worker, work, &worker->work_list);
810 ret = true;
811 }
812 spin_unlock_irqrestore(&worker->lock, flags);
813 return ret;
814 }
815 EXPORT_SYMBOL_GPL(kthread_queue_work);
816
817 /**
818 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
819 * delayed work when the timer expires.
820 * @t: pointer to the expired timer
821 *
822 * The format of the function is defined by struct timer_list.
823 * It should have been called from irqsafe timer with irq already off.
824 */
825 void kthread_delayed_work_timer_fn(struct timer_list *t)
826 {
827 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
828 struct kthread_work *work = &dwork->work;
829 struct kthread_worker *worker = work->worker;
830
831 /*
832 * This might happen when a pending work is reinitialized.
833 * It means that it is used a wrong way.
834 */
835 if (WARN_ON_ONCE(!worker))
836 return;
837
838 spin_lock(&worker->lock);
839 /* Work must not be used with >1 worker, see kthread_queue_work(). */
840 WARN_ON_ONCE(work->worker != worker);
841
842 /* Move the work from worker->delayed_work_list. */
843 WARN_ON_ONCE(list_empty(&work->node));
844 list_del_init(&work->node);
845 kthread_insert_work(worker, work, &worker->work_list);
846
847 spin_unlock(&worker->lock);
848 }
849 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
850
851 void __kthread_queue_delayed_work(struct kthread_worker *worker,
852 struct kthread_delayed_work *dwork,
853 unsigned long delay)
854 {
855 struct timer_list *timer = &dwork->timer;
856 struct kthread_work *work = &dwork->work;
857
858 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
859
860 /*
861 * If @delay is 0, queue @dwork->work immediately. This is for
862 * both optimization and correctness. The earliest @timer can
863 * expire is on the closest next tick and delayed_work users depend
864 * on that there's no such delay when @delay is 0.
865 */
866 if (!delay) {
867 kthread_insert_work(worker, work, &worker->work_list);
868 return;
869 }
870
871 /* Be paranoid and try to detect possible races already now. */
872 kthread_insert_work_sanity_check(worker, work);
873
874 list_add(&work->node, &worker->delayed_work_list);
875 work->worker = worker;
876 timer->expires = jiffies + delay;
877 add_timer(timer);
878 }
879
880 /**
881 * kthread_queue_delayed_work - queue the associated kthread work
882 * after a delay.
883 * @worker: target kthread_worker
884 * @dwork: kthread_delayed_work to queue
885 * @delay: number of jiffies to wait before queuing
886 *
887 * If the work has not been pending it starts a timer that will queue
888 * the work after the given @delay. If @delay is zero, it queues the
889 * work immediately.
890 *
891 * Return: %false if the @work has already been pending. It means that
892 * either the timer was running or the work was queued. It returns %true
893 * otherwise.
894 */
895 bool kthread_queue_delayed_work(struct kthread_worker *worker,
896 struct kthread_delayed_work *dwork,
897 unsigned long delay)
898 {
899 struct kthread_work *work = &dwork->work;
900 unsigned long flags;
901 bool ret = false;
902
903 spin_lock_irqsave(&worker->lock, flags);
904
905 if (!queuing_blocked(worker, work)) {
906 __kthread_queue_delayed_work(worker, dwork, delay);
907 ret = true;
908 }
909
910 spin_unlock_irqrestore(&worker->lock, flags);
911 return ret;
912 }
913 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
914
915 struct kthread_flush_work {
916 struct kthread_work work;
917 struct completion done;
918 };
919
920 static void kthread_flush_work_fn(struct kthread_work *work)
921 {
922 struct kthread_flush_work *fwork =
923 container_of(work, struct kthread_flush_work, work);
924 complete(&fwork->done);
925 }
926
927 /**
928 * kthread_flush_work - flush a kthread_work
929 * @work: work to flush
930 *
931 * If @work is queued or executing, wait for it to finish execution.
932 */
933 void kthread_flush_work(struct kthread_work *work)
934 {
935 struct kthread_flush_work fwork = {
936 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
937 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
938 };
939 struct kthread_worker *worker;
940 bool noop = false;
941
942 worker = work->worker;
943 if (!worker)
944 return;
945
946 spin_lock_irq(&worker->lock);
947 /* Work must not be used with >1 worker, see kthread_queue_work(). */
948 WARN_ON_ONCE(work->worker != worker);
949
950 if (!list_empty(&work->node))
951 kthread_insert_work(worker, &fwork.work, work->node.next);
952 else if (worker->current_work == work)
953 kthread_insert_work(worker, &fwork.work,
954 worker->work_list.next);
955 else
956 noop = true;
957
958 spin_unlock_irq(&worker->lock);
959
960 if (!noop)
961 wait_for_completion(&fwork.done);
962 }
963 EXPORT_SYMBOL_GPL(kthread_flush_work);
964
965 /*
966 * This function removes the work from the worker queue. Also it makes sure
967 * that it won't get queued later via the delayed work's timer.
968 *
969 * The work might still be in use when this function finishes. See the
970 * current_work proceed by the worker.
971 *
972 * Return: %true if @work was pending and successfully canceled,
973 * %false if @work was not pending
974 */
975 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
976 unsigned long *flags)
977 {
978 /* Try to cancel the timer if exists. */
979 if (is_dwork) {
980 struct kthread_delayed_work *dwork =
981 container_of(work, struct kthread_delayed_work, work);
982 struct kthread_worker *worker = work->worker;
983
984 /*
985 * del_timer_sync() must be called to make sure that the timer
986 * callback is not running. The lock must be temporary released
987 * to avoid a deadlock with the callback. In the meantime,
988 * any queuing is blocked by setting the canceling counter.
989 */
990 work->canceling++;
991 spin_unlock_irqrestore(&worker->lock, *flags);
992 del_timer_sync(&dwork->timer);
993 spin_lock_irqsave(&worker->lock, *flags);
994 work->canceling--;
995 }
996
997 /*
998 * Try to remove the work from a worker list. It might either
999 * be from worker->work_list or from worker->delayed_work_list.
1000 */
1001 if (!list_empty(&work->node)) {
1002 list_del_init(&work->node);
1003 return true;
1004 }
1005
1006 return false;
1007 }
1008
1009 /**
1010 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1011 * @worker: kthread worker to use
1012 * @dwork: kthread delayed work to queue
1013 * @delay: number of jiffies to wait before queuing
1014 *
1015 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1016 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1017 * @work is guaranteed to be queued immediately.
1018 *
1019 * Return: %true if @dwork was pending and its timer was modified,
1020 * %false otherwise.
1021 *
1022 * A special case is when the work is being canceled in parallel.
1023 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1024 * or yet another kthread_mod_delayed_work() call. We let the other command
1025 * win and return %false here. The caller is supposed to synchronize these
1026 * operations a reasonable way.
1027 *
1028 * This function is safe to call from any context including IRQ handler.
1029 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1030 * for details.
1031 */
1032 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1033 struct kthread_delayed_work *dwork,
1034 unsigned long delay)
1035 {
1036 struct kthread_work *work = &dwork->work;
1037 unsigned long flags;
1038 int ret = false;
1039
1040 spin_lock_irqsave(&worker->lock, flags);
1041
1042 /* Do not bother with canceling when never queued. */
1043 if (!work->worker)
1044 goto fast_queue;
1045
1046 /* Work must not be used with >1 worker, see kthread_queue_work() */
1047 WARN_ON_ONCE(work->worker != worker);
1048
1049 /* Do not fight with another command that is canceling this work. */
1050 if (work->canceling)
1051 goto out;
1052
1053 ret = __kthread_cancel_work(work, true, &flags);
1054 fast_queue:
1055 __kthread_queue_delayed_work(worker, dwork, delay);
1056 out:
1057 spin_unlock_irqrestore(&worker->lock, flags);
1058 return ret;
1059 }
1060 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1061
1062 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1063 {
1064 struct kthread_worker *worker = work->worker;
1065 unsigned long flags;
1066 int ret = false;
1067
1068 if (!worker)
1069 goto out;
1070
1071 spin_lock_irqsave(&worker->lock, flags);
1072 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1073 WARN_ON_ONCE(work->worker != worker);
1074
1075 ret = __kthread_cancel_work(work, is_dwork, &flags);
1076
1077 if (worker->current_work != work)
1078 goto out_fast;
1079
1080 /*
1081 * The work is in progress and we need to wait with the lock released.
1082 * In the meantime, block any queuing by setting the canceling counter.
1083 */
1084 work->canceling++;
1085 spin_unlock_irqrestore(&worker->lock, flags);
1086 kthread_flush_work(work);
1087 spin_lock_irqsave(&worker->lock, flags);
1088 work->canceling--;
1089
1090 out_fast:
1091 spin_unlock_irqrestore(&worker->lock, flags);
1092 out:
1093 return ret;
1094 }
1095
1096 /**
1097 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1098 * @work: the kthread work to cancel
1099 *
1100 * Cancel @work and wait for its execution to finish. This function
1101 * can be used even if the work re-queues itself. On return from this
1102 * function, @work is guaranteed to be not pending or executing on any CPU.
1103 *
1104 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1105 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1106 *
1107 * The caller must ensure that the worker on which @work was last
1108 * queued can't be destroyed before this function returns.
1109 *
1110 * Return: %true if @work was pending, %false otherwise.
1111 */
1112 bool kthread_cancel_work_sync(struct kthread_work *work)
1113 {
1114 return __kthread_cancel_work_sync(work, false);
1115 }
1116 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1117
1118 /**
1119 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1120 * wait for it to finish.
1121 * @dwork: the kthread delayed work to cancel
1122 *
1123 * This is kthread_cancel_work_sync() for delayed works.
1124 *
1125 * Return: %true if @dwork was pending, %false otherwise.
1126 */
1127 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1128 {
1129 return __kthread_cancel_work_sync(&dwork->work, true);
1130 }
1131 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1132
1133 /**
1134 * kthread_flush_worker - flush all current works on a kthread_worker
1135 * @worker: worker to flush
1136 *
1137 * Wait until all currently executing or pending works on @worker are
1138 * finished.
1139 */
1140 void kthread_flush_worker(struct kthread_worker *worker)
1141 {
1142 struct kthread_flush_work fwork = {
1143 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1144 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1145 };
1146
1147 kthread_queue_work(worker, &fwork.work);
1148 wait_for_completion(&fwork.done);
1149 }
1150 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1151
1152 /**
1153 * kthread_destroy_worker - destroy a kthread worker
1154 * @worker: worker to be destroyed
1155 *
1156 * Flush and destroy @worker. The simple flush is enough because the kthread
1157 * worker API is used only in trivial scenarios. There are no multi-step state
1158 * machines needed.
1159 */
1160 void kthread_destroy_worker(struct kthread_worker *worker)
1161 {
1162 struct task_struct *task;
1163
1164 task = worker->task;
1165 if (WARN_ON(!task))
1166 return;
1167
1168 kthread_flush_worker(worker);
1169 kthread_stop(task);
1170 WARN_ON(!list_empty(&worker->work_list));
1171 kfree(worker);
1172 }
1173 EXPORT_SYMBOL(kthread_destroy_worker);
1174
1175 #ifdef CONFIG_BLK_CGROUP
1176 /**
1177 * kthread_associate_blkcg - associate blkcg to current kthread
1178 * @css: the cgroup info
1179 *
1180 * Current thread must be a kthread. The thread is running jobs on behalf of
1181 * other threads. In some cases, we expect the jobs attach cgroup info of
1182 * original threads instead of that of current thread. This function stores
1183 * original thread's cgroup info in current kthread context for later
1184 * retrieval.
1185 */
1186 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1187 {
1188 struct kthread *kthread;
1189
1190 if (!(current->flags & PF_KTHREAD))
1191 return;
1192 kthread = to_kthread(current);
1193 if (!kthread)
1194 return;
1195
1196 if (kthread->blkcg_css) {
1197 css_put(kthread->blkcg_css);
1198 kthread->blkcg_css = NULL;
1199 }
1200 if (css) {
1201 css_get(css);
1202 kthread->blkcg_css = css;
1203 }
1204 }
1205 EXPORT_SYMBOL(kthread_associate_blkcg);
1206
1207 /**
1208 * kthread_blkcg - get associated blkcg css of current kthread
1209 *
1210 * Current thread must be a kthread.
1211 */
1212 struct cgroup_subsys_state *kthread_blkcg(void)
1213 {
1214 struct kthread *kthread;
1215
1216 if (current->flags & PF_KTHREAD) {
1217 kthread = to_kthread(current);
1218 if (kthread)
1219 return kthread->blkcg_css;
1220 }
1221 return NULL;
1222 }
1223 EXPORT_SYMBOL(kthread_blkcg);
1224 #endif