]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - kernel/kthread.c
KVM: PPC: Use preregistered memory API to access TCE list
[mirror_ubuntu-zesty-kernel.git] / kernel / kthread.c
CommitLineData
1da177e4
LT
1/* Kernel thread helper functions.
2 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
3 *
73c27992 4 * Creation is done via kthreadd, so that we get a clean environment
1da177e4
LT
5 * even if we're invoked from userspace (think modprobe, hotplug cpu,
6 * etc.).
7 */
8#include <linux/sched.h>
9#include <linux/kthread.h>
10#include <linux/completion.h>
11#include <linux/err.h>
58568d2a 12#include <linux/cpuset.h>
1da177e4
LT
13#include <linux/unistd.h>
14#include <linux/file.h>
9984de1a 15#include <linux/export.h>
97d1f15b 16#include <linux/mutex.h>
b56c0d89
TH
17#include <linux/slab.h>
18#include <linux/freezer.h>
a74fb73c 19#include <linux/ptrace.h>
cd42d559 20#include <linux/uaccess.h>
793adb1b 21#include <linux/cgroup.h>
ad8d75ff 22#include <trace/events/sched.h>
1da177e4 23
73c27992
EB
24static DEFINE_SPINLOCK(kthread_create_lock);
25static LIST_HEAD(kthread_create_list);
26struct task_struct *kthreadd_task;
1da177e4
LT
27
28struct kthread_create_info
29{
73c27992 30 /* Information passed to kthread() from kthreadd. */
1da177e4
LT
31 int (*threadfn)(void *data);
32 void *data;
207205a2 33 int node;
1da177e4 34
73c27992 35 /* Result passed back to kthread_create() from kthreadd. */
1da177e4 36 struct task_struct *result;
786235ee 37 struct completion *done;
65f27f38 38
73c27992 39 struct list_head list;
1da177e4
LT
40};
41
63706172 42struct kthread {
2a1d4460
TG
43 unsigned long flags;
44 unsigned int cpu;
82805ab7 45 void *data;
2a1d4460 46 struct completion parked;
63706172 47 struct completion exited;
1da177e4
LT
48};
49
2a1d4460
TG
50enum KTHREAD_BITS {
51 KTHREAD_IS_PER_CPU = 0,
52 KTHREAD_SHOULD_STOP,
53 KTHREAD_SHOULD_PARK,
54 KTHREAD_IS_PARKED,
55};
56
1da5c46f
ON
57static inline void set_kthread_struct(void *kthread)
58{
59 /*
60 * We abuse ->set_child_tid to avoid the new member and because it
61 * can't be wrongly copied by copy_process(). We also rely on fact
62 * that the caller can't exec, so PF_KTHREAD can't be cleared.
63 */
64 current->set_child_tid = (__force void __user *)kthread;
65}
4ecdafc8
ON
66
67static inline struct kthread *to_kthread(struct task_struct *k)
68{
1da5c46f
ON
69 WARN_ON(!(k->flags & PF_KTHREAD));
70 return (__force void *)k->set_child_tid;
4ecdafc8
ON
71}
72
1da5c46f
ON
73void free_kthread_struct(struct task_struct *k)
74{
75 /*
76 * Can be NULL if this kthread was created by kernel_thread()
77 * or if kmalloc() in kthread() failed.
78 */
79 kfree(to_kthread(k));
80}
81
9e37bd30
RD
82/**
83 * kthread_should_stop - should this kthread return now?
84 *
72fd4a35 85 * When someone calls kthread_stop() on your kthread, it will be woken
9e37bd30
RD
86 * and this will return true. You should then return, and your return
87 * value will be passed through to kthread_stop().
88 */
2a1d4460 89bool kthread_should_stop(void)
1da177e4 90{
2a1d4460 91 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
1da177e4
LT
92}
93EXPORT_SYMBOL(kthread_should_stop);
94
2a1d4460
TG
95/**
96 * kthread_should_park - should this kthread park now?
97 *
98 * When someone calls kthread_park() on your kthread, it will be woken
99 * and this will return true. You should then do the necessary
100 * cleanup and call kthread_parkme()
101 *
102 * Similar to kthread_should_stop(), but this keeps the thread alive
103 * and in a park position. kthread_unpark() "restarts" the thread and
104 * calls the thread function again.
105 */
106bool kthread_should_park(void)
107{
108 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
109}
18896451 110EXPORT_SYMBOL_GPL(kthread_should_park);
2a1d4460 111
8a32c441
TH
112/**
113 * kthread_freezable_should_stop - should this freezable kthread return now?
114 * @was_frozen: optional out parameter, indicates whether %current was frozen
115 *
116 * kthread_should_stop() for freezable kthreads, which will enter
117 * refrigerator if necessary. This function is safe from kthread_stop() /
118 * freezer deadlock and freezable kthreads should use this function instead
119 * of calling try_to_freeze() directly.
120 */
121bool kthread_freezable_should_stop(bool *was_frozen)
122{
123 bool frozen = false;
124
125 might_sleep();
126
127 if (unlikely(freezing(current)))
128 frozen = __refrigerator(true);
129
130 if (was_frozen)
131 *was_frozen = frozen;
132
133 return kthread_should_stop();
134}
135EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
136
82805ab7
TH
137/**
138 * kthread_data - return data value specified on kthread creation
139 * @task: kthread task in question
140 *
141 * Return the data value specified when kthread @task was created.
142 * The caller is responsible for ensuring the validity of @task when
143 * calling this function.
144 */
145void *kthread_data(struct task_struct *task)
146{
147 return to_kthread(task)->data;
148}
149
cd42d559 150/**
e700591a 151 * kthread_probe_data - speculative version of kthread_data()
cd42d559
TH
152 * @task: possible kthread task in question
153 *
154 * @task could be a kthread task. Return the data value specified when it
155 * was created if accessible. If @task isn't a kthread task or its data is
156 * inaccessible for any reason, %NULL is returned. This function requires
157 * that @task itself is safe to dereference.
158 */
e700591a 159void *kthread_probe_data(struct task_struct *task)
cd42d559
TH
160{
161 struct kthread *kthread = to_kthread(task);
162 void *data = NULL;
163
164 probe_kernel_read(&data, &kthread->data, sizeof(data));
165 return data;
166}
167
2a1d4460
TG
168static void __kthread_parkme(struct kthread *self)
169{
f2530dc7 170 __set_current_state(TASK_PARKED);
2a1d4460
TG
171 while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
172 if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
173 complete(&self->parked);
174 schedule();
f2530dc7 175 __set_current_state(TASK_PARKED);
2a1d4460
TG
176 }
177 clear_bit(KTHREAD_IS_PARKED, &self->flags);
178 __set_current_state(TASK_RUNNING);
179}
180
181void kthread_parkme(void)
182{
183 __kthread_parkme(to_kthread(current));
184}
18896451 185EXPORT_SYMBOL_GPL(kthread_parkme);
2a1d4460 186
1da177e4
LT
187static int kthread(void *_create)
188{
63706172 189 /* Copy data: it's on kthread's stack */
1da177e4 190 struct kthread_create_info *create = _create;
63706172
ON
191 int (*threadfn)(void *data) = create->threadfn;
192 void *data = create->data;
786235ee 193 struct completion *done;
1da5c46f 194 struct kthread *self;
63706172 195 int ret;
1da177e4 196
1da5c46f
ON
197 self = kmalloc(sizeof(*self), GFP_KERNEL);
198 set_kthread_struct(self);
1da177e4 199
786235ee
TH
200 /* If user was SIGKILLed, I release the structure. */
201 done = xchg(&create->done, NULL);
202 if (!done) {
203 kfree(create);
204 do_exit(-EINTR);
205 }
1da5c46f
ON
206
207 if (!self) {
208 create->result = ERR_PTR(-ENOMEM);
209 complete(done);
210 do_exit(-ENOMEM);
211 }
212
213 self->flags = 0;
214 self->data = data;
215 init_completion(&self->exited);
216 init_completion(&self->parked);
217 current->vfork_done = &self->exited;
218
1da177e4 219 /* OK, tell user we're spawned, wait for stop or wakeup */
a076e4bc 220 __set_current_state(TASK_UNINTERRUPTIBLE);
3217ab97 221 create->result = current;
786235ee 222 complete(done);
1da177e4
LT
223 schedule();
224
63706172 225 ret = -EINTR;
1da5c46f 226 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
793adb1b 227 cgroup_kthread_ready();
1da5c46f 228 __kthread_parkme(self);
2a1d4460
TG
229 ret = threadfn(data);
230 }
63706172 231 do_exit(ret);
1da177e4
LT
232}
233
207205a2
ED
234/* called from do_fork() to get node information for about to be created task */
235int tsk_fork_get_node(struct task_struct *tsk)
236{
237#ifdef CONFIG_NUMA
238 if (tsk == kthreadd_task)
239 return tsk->pref_node_fork;
240#endif
81c98869 241 return NUMA_NO_NODE;
207205a2
ED
242}
243
73c27992 244static void create_kthread(struct kthread_create_info *create)
1da177e4 245{
1da177e4
LT
246 int pid;
247
207205a2
ED
248#ifdef CONFIG_NUMA
249 current->pref_node_fork = create->node;
250#endif
1da177e4
LT
251 /* We want our own signal handler (we take no signals by default). */
252 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
cdd140bd 253 if (pid < 0) {
786235ee
TH
254 /* If user was SIGKILLed, I release the structure. */
255 struct completion *done = xchg(&create->done, NULL);
256
257 if (!done) {
258 kfree(create);
259 return;
260 }
1da177e4 261 create->result = ERR_PTR(pid);
786235ee 262 complete(done);
cdd140bd 263 }
1da177e4
LT
264}
265
c0b942a7
NI
266static __printf(4, 0)
267struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
255451e4
PM
268 void *data, int node,
269 const char namefmt[],
270 va_list args)
1da177e4 271{
786235ee
TH
272 DECLARE_COMPLETION_ONSTACK(done);
273 struct task_struct *task;
274 struct kthread_create_info *create = kmalloc(sizeof(*create),
275 GFP_KERNEL);
276
277 if (!create)
278 return ERR_PTR(-ENOMEM);
279 create->threadfn = threadfn;
280 create->data = data;
281 create->node = node;
282 create->done = &done;
73c27992
EB
283
284 spin_lock(&kthread_create_lock);
786235ee 285 list_add_tail(&create->list, &kthread_create_list);
73c27992
EB
286 spin_unlock(&kthread_create_lock);
287
cbd9b67b 288 wake_up_process(kthreadd_task);
786235ee
TH
289 /*
290 * Wait for completion in killable state, for I might be chosen by
291 * the OOM killer while kthreadd is trying to allocate memory for
292 * new kernel thread.
293 */
294 if (unlikely(wait_for_completion_killable(&done))) {
cdbbd331
TH
295 int i = 0;
296
297 /*
298 * I got SIGKILL, but wait for 10 more seconds for completion
299 * unless chosen by the OOM killer. This delay is there as a
300 * workaround for boot failure caused by SIGKILL upon device
301 * driver initialization timeout.
302 */
303 while (i++ < 10 && !test_tsk_thread_flag(current, TIF_MEMDIE))
304 if (wait_for_completion_timeout(&done, HZ))
305 goto ready;
786235ee
TH
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))
8fe6929c 312 return ERR_PTR(-EINTR);
786235ee
TH
313 /*
314 * kthreadd (or new kernel thread) will call complete()
315 * shortly.
316 */
317 wait_for_completion(&done);
318 }
cdbbd331 319ready:
786235ee
TH
320 task = create->result;
321 if (!IS_ERR(task)) {
c9b5f501 322 static const struct sched_param param = { .sched_priority = 0 };
1c99315b 323
786235ee 324 vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
1c99315b
ON
325 /*
326 * root may have changed our (kthreadd's) priority or CPU mask.
327 * The kernel thread should not inherit these properties.
328 */
786235ee
TH
329 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
330 set_cpus_allowed_ptr(task, cpu_all_mask);
1da177e4 331 }
786235ee
TH
332 kfree(create);
333 return task;
1da177e4 334}
255451e4
PM
335
336/**
337 * kthread_create_on_node - create a kthread.
338 * @threadfn: the function to run until signal_pending(current).
339 * @data: data ptr for @threadfn.
340 * @node: task and thread structures for the thread are allocated on this node
341 * @namefmt: printf-style name for the thread.
342 *
343 * Description: This helper function creates and names a kernel
344 * thread. The thread will be stopped: use wake_up_process() to start
345 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
346 * is affine to all CPUs.
347 *
348 * If thread is going to be bound on a particular cpu, give its node
349 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
350 * When woken, the thread will run @threadfn() with @data as its
351 * argument. @threadfn() can either call do_exit() directly if it is a
352 * standalone thread for which no one will call kthread_stop(), or
353 * return when 'kthread_should_stop()' is true (which means
354 * kthread_stop() has been called). The return value should be zero
355 * or a negative error number; it will be passed to kthread_stop().
356 *
357 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
358 */
359struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
360 void *data, int node,
361 const char namefmt[],
362 ...)
363{
364 struct task_struct *task;
365 va_list args;
366
367 va_start(args, namefmt);
368 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
369 va_end(args);
370
371 return task;
372}
207205a2 373EXPORT_SYMBOL(kthread_create_on_node);
1da177e4 374
25834c73 375static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
2a1d4460 376{
25834c73
PZ
377 unsigned long flags;
378
f2530dc7
TG
379 if (!wait_task_inactive(p, state)) {
380 WARN_ON(1);
381 return;
382 }
25834c73 383
2a1d4460 384 /* It's safe because the task is inactive. */
25834c73
PZ
385 raw_spin_lock_irqsave(&p->pi_lock, flags);
386 do_set_cpus_allowed(p, mask);
14a40ffc 387 p->flags |= PF_NO_SETAFFINITY;
25834c73
PZ
388 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
389}
390
391static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
392{
393 __kthread_bind_mask(p, cpumask_of(cpu), state);
394}
395
396void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
397{
398 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
2a1d4460
TG
399}
400
881232b7
PZ
401/**
402 * kthread_bind - bind a just-created kthread to a cpu.
403 * @p: thread created by kthread_create().
404 * @cpu: cpu (might not be online, must be possible) for @k to run on.
405 *
406 * Description: This function is equivalent to set_cpus_allowed(),
407 * except that @cpu doesn't need to be online, and the thread must be
408 * stopped (i.e., just returned from kthread_create()).
409 */
410void kthread_bind(struct task_struct *p, unsigned int cpu)
411{
f2530dc7 412 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
881232b7
PZ
413}
414EXPORT_SYMBOL(kthread_bind);
415
2a1d4460
TG
416/**
417 * kthread_create_on_cpu - Create a cpu bound kthread
418 * @threadfn: the function to run until signal_pending(current).
419 * @data: data ptr for @threadfn.
420 * @cpu: The cpu on which the thread should be bound,
421 * @namefmt: printf-style name for the thread. Format is restricted
422 * to "name.*%u". Code fills in cpu number.
423 *
424 * Description: This helper function creates and names a kernel thread
425 * The thread will be woken and put into park mode.
426 */
427struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
428 void *data, unsigned int cpu,
429 const char *namefmt)
430{
431 struct task_struct *p;
432
10922838 433 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
2a1d4460
TG
434 cpu);
435 if (IS_ERR(p))
436 return p;
a65d4096
PM
437 kthread_bind(p, cpu);
438 /* CPU hotplug need to bind once again when unparking the thread. */
2a1d4460
TG
439 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
440 to_kthread(p)->cpu = cpu;
2a1d4460
TG
441 return p;
442}
443
cf380a4a
ON
444/**
445 * kthread_unpark - unpark a thread created by kthread_create().
446 * @k: thread created by kthread_create().
447 *
448 * Sets kthread_should_park() for @k to return false, wakes it, and
449 * waits for it to return. If the thread is marked percpu then its
450 * bound to the cpu again.
451 */
452void kthread_unpark(struct task_struct *k)
f2530dc7 453{
cf380a4a
ON
454 struct kthread *kthread = to_kthread(k);
455
f2530dc7
TG
456 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
457 /*
458 * We clear the IS_PARKED bit here as we don't wait
459 * until the task has left the park code. So if we'd
460 * park before that happens we'd see the IS_PARKED bit
461 * which might be about to be cleared.
462 */
463 if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
a65d4096
PM
464 /*
465 * Newly created kthread was parked when the CPU was offline.
466 * The binding was lost and we need to set it again.
467 */
f2530dc7
TG
468 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
469 __kthread_bind(k, kthread->cpu, TASK_PARKED);
470 wake_up_state(k, TASK_PARKED);
471 }
472}
18896451 473EXPORT_SYMBOL_GPL(kthread_unpark);
2a1d4460
TG
474
475/**
476 * kthread_park - park a thread created by kthread_create().
477 * @k: thread created by kthread_create().
478 *
479 * Sets kthread_should_park() for @k to return true, wakes it, and
480 * waits for it to return. This can also be called after kthread_create()
481 * instead of calling wake_up_process(): the thread will park without
482 * calling threadfn().
483 *
484 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
485 * If called by the kthread itself just the park bit is set.
486 */
487int kthread_park(struct task_struct *k)
488{
cf380a4a
ON
489 struct kthread *kthread = to_kthread(k);
490
491 if (WARN_ON(k->flags & PF_EXITING))
492 return -ENOSYS;
493
494 if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
495 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
496 if (k != current) {
497 wake_up_process(k);
498 wait_for_completion(&kthread->parked);
2a1d4460 499 }
2a1d4460 500 }
cf380a4a
ON
501
502 return 0;
2a1d4460 503}
18896451 504EXPORT_SYMBOL_GPL(kthread_park);
2a1d4460 505
9e37bd30
RD
506/**
507 * kthread_stop - stop a thread created by kthread_create().
508 * @k: thread created by kthread_create().
509 *
510 * Sets kthread_should_stop() for @k to return true, wakes it, and
9ae26027
ON
511 * waits for it to exit. This can also be called after kthread_create()
512 * instead of calling wake_up_process(): the thread will exit without
513 * calling threadfn().
514 *
515 * If threadfn() may call do_exit() itself, the caller must ensure
516 * task_struct can't go away.
9e37bd30
RD
517 *
518 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
519 * was never called.
520 */
1da177e4
LT
521int kthread_stop(struct task_struct *k)
522{
b5c5442b 523 struct kthread *kthread;
1da177e4
LT
524 int ret;
525
0a16b607 526 trace_sched_kthread_stop(k);
b5c5442b
ON
527
528 get_task_struct(k);
efb29fbf
ON
529 kthread = to_kthread(k);
530 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
cf380a4a 531 kthread_unpark(k);
efb29fbf
ON
532 wake_up_process(k);
533 wait_for_completion(&kthread->exited);
63706172 534 ret = k->exit_code;
1da177e4 535 put_task_struct(k);
0a16b607 536
b5c5442b 537 trace_sched_kthread_stop_ret(ret);
1da177e4
LT
538 return ret;
539}
52e92e57 540EXPORT_SYMBOL(kthread_stop);
1da177e4 541
e804a4a4 542int kthreadd(void *unused)
1da177e4 543{
73c27992 544 struct task_struct *tsk = current;
1da177e4 545
e804a4a4 546 /* Setup a clean context for our children to inherit. */
73c27992 547 set_task_comm(tsk, "kthreadd");
10ab825b 548 ignore_signals(tsk);
1a2142af 549 set_cpus_allowed_ptr(tsk, cpu_all_mask);
aee4faa4 550 set_mems_allowed(node_states[N_MEMORY]);
73c27992 551
34b087e4 552 current->flags |= PF_NOFREEZE;
793adb1b 553 cgroup_init_kthreadd();
73c27992
EB
554
555 for (;;) {
556 set_current_state(TASK_INTERRUPTIBLE);
557 if (list_empty(&kthread_create_list))
558 schedule();
559 __set_current_state(TASK_RUNNING);
560
561 spin_lock(&kthread_create_lock);
562 while (!list_empty(&kthread_create_list)) {
563 struct kthread_create_info *create;
564
565 create = list_entry(kthread_create_list.next,
566 struct kthread_create_info, list);
567 list_del_init(&create->list);
568 spin_unlock(&kthread_create_lock);
569
570 create_kthread(create);
571
572 spin_lock(&kthread_create_lock);
573 }
574 spin_unlock(&kthread_create_lock);
575 }
576
577 return 0;
578}
b56c0d89 579
3989144f 580void __kthread_init_worker(struct kthread_worker *worker,
4f32e9b1
YZ
581 const char *name,
582 struct lock_class_key *key)
583{
dbf52682 584 memset(worker, 0, sizeof(struct kthread_worker));
4f32e9b1
YZ
585 spin_lock_init(&worker->lock);
586 lockdep_set_class_and_name(&worker->lock, key, name);
587 INIT_LIST_HEAD(&worker->work_list);
22597dc3 588 INIT_LIST_HEAD(&worker->delayed_work_list);
4f32e9b1 589}
3989144f 590EXPORT_SYMBOL_GPL(__kthread_init_worker);
4f32e9b1 591
b56c0d89
TH
592/**
593 * kthread_worker_fn - kthread function to process kthread_worker
594 * @worker_ptr: pointer to initialized kthread_worker
595 *
fbae2d44
PM
596 * This function implements the main cycle of kthread worker. It processes
597 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
598 * is empty.
b56c0d89 599 *
fbae2d44
PM
600 * The works are not allowed to keep any locks, disable preemption or interrupts
601 * when they finish. There is defined a safe point for freezing when one work
602 * finishes and before a new one is started.
8197b3d4
PM
603 *
604 * Also the works must not be handled by more than one worker at the same time,
605 * see also kthread_queue_work().
b56c0d89
TH
606 */
607int kthread_worker_fn(void *worker_ptr)
608{
609 struct kthread_worker *worker = worker_ptr;
610 struct kthread_work *work;
611
fbae2d44
PM
612 /*
613 * FIXME: Update the check and remove the assignment when all kthread
614 * worker users are created using kthread_create_worker*() functions.
615 */
616 WARN_ON(worker->task && worker->task != current);
b56c0d89 617 worker->task = current;
dbf52682
PM
618
619 if (worker->flags & KTW_FREEZABLE)
620 set_freezable();
621
b56c0d89
TH
622repeat:
623 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
624
625 if (kthread_should_stop()) {
626 __set_current_state(TASK_RUNNING);
627 spin_lock_irq(&worker->lock);
628 worker->task = NULL;
629 spin_unlock_irq(&worker->lock);
630 return 0;
631 }
632
633 work = NULL;
634 spin_lock_irq(&worker->lock);
635 if (!list_empty(&worker->work_list)) {
636 work = list_first_entry(&worker->work_list,
637 struct kthread_work, node);
638 list_del_init(&work->node);
639 }
46f3d976 640 worker->current_work = work;
b56c0d89
TH
641 spin_unlock_irq(&worker->lock);
642
643 if (work) {
644 __set_current_state(TASK_RUNNING);
645 work->func(work);
b56c0d89
TH
646 } else if (!freezing(current))
647 schedule();
648
649 try_to_freeze();
650 goto repeat;
651}
652EXPORT_SYMBOL_GPL(kthread_worker_fn);
653
c0b942a7 654static __printf(3, 0) struct kthread_worker *
dbf52682
PM
655__kthread_create_worker(int cpu, unsigned int flags,
656 const char namefmt[], va_list args)
fbae2d44
PM
657{
658 struct kthread_worker *worker;
659 struct task_struct *task;
8fb9dcbd 660 int node = -1;
fbae2d44
PM
661
662 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
663 if (!worker)
664 return ERR_PTR(-ENOMEM);
665
666 kthread_init_worker(worker);
667
8fb9dcbd
ON
668 if (cpu >= 0)
669 node = cpu_to_node(cpu);
fbae2d44 670
8fb9dcbd
ON
671 task = __kthread_create_on_node(kthread_worker_fn, worker,
672 node, namefmt, args);
fbae2d44
PM
673 if (IS_ERR(task))
674 goto fail_task;
675
8fb9dcbd
ON
676 if (cpu >= 0)
677 kthread_bind(task, cpu);
678
dbf52682 679 worker->flags = flags;
fbae2d44
PM
680 worker->task = task;
681 wake_up_process(task);
682 return worker;
683
684fail_task:
685 kfree(worker);
686 return ERR_CAST(task);
687}
688
689/**
690 * kthread_create_worker - create a kthread worker
dbf52682 691 * @flags: flags modifying the default behavior of the worker
fbae2d44
PM
692 * @namefmt: printf-style name for the kthread worker (task).
693 *
694 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
695 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
696 * when the worker was SIGKILLed.
697 */
698struct kthread_worker *
dbf52682 699kthread_create_worker(unsigned int flags, const char namefmt[], ...)
fbae2d44
PM
700{
701 struct kthread_worker *worker;
702 va_list args;
703
704 va_start(args, namefmt);
dbf52682 705 worker = __kthread_create_worker(-1, flags, namefmt, args);
fbae2d44
PM
706 va_end(args);
707
708 return worker;
709}
710EXPORT_SYMBOL(kthread_create_worker);
711
712/**
713 * kthread_create_worker_on_cpu - create a kthread worker and bind it
714 * it to a given CPU and the associated NUMA node.
715 * @cpu: CPU number
dbf52682 716 * @flags: flags modifying the default behavior of the worker
fbae2d44
PM
717 * @namefmt: printf-style name for the kthread worker (task).
718 *
719 * Use a valid CPU number if you want to bind the kthread worker
720 * to the given CPU and the associated NUMA node.
721 *
722 * A good practice is to add the cpu number also into the worker name.
723 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
724 *
725 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
726 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
727 * when the worker was SIGKILLed.
728 */
729struct kthread_worker *
dbf52682
PM
730kthread_create_worker_on_cpu(int cpu, unsigned int flags,
731 const char namefmt[], ...)
fbae2d44
PM
732{
733 struct kthread_worker *worker;
734 va_list args;
735
736 va_start(args, namefmt);
dbf52682 737 worker = __kthread_create_worker(cpu, flags, namefmt, args);
fbae2d44
PM
738 va_end(args);
739
740 return worker;
741}
742EXPORT_SYMBOL(kthread_create_worker_on_cpu);
743
37be45d4
PM
744/*
745 * Returns true when the work could not be queued at the moment.
746 * It happens when it is already pending in a worker list
747 * or when it is being cancelled.
748 */
749static inline bool queuing_blocked(struct kthread_worker *worker,
750 struct kthread_work *work)
751{
752 lockdep_assert_held(&worker->lock);
753
754 return !list_empty(&work->node) || work->canceling;
755}
756
8197b3d4
PM
757static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
758 struct kthread_work *work)
759{
760 lockdep_assert_held(&worker->lock);
761 WARN_ON_ONCE(!list_empty(&work->node));
762 /* Do not use a work with >1 worker, see kthread_queue_work() */
763 WARN_ON_ONCE(work->worker && work->worker != worker);
764}
765
9a2e03d8 766/* insert @work before @pos in @worker */
3989144f 767static void kthread_insert_work(struct kthread_worker *worker,
8197b3d4
PM
768 struct kthread_work *work,
769 struct list_head *pos)
9a2e03d8 770{
8197b3d4 771 kthread_insert_work_sanity_check(worker, work);
9a2e03d8
TH
772
773 list_add_tail(&work->node, pos);
46f3d976 774 work->worker = worker;
ed1403ec 775 if (!worker->current_work && likely(worker->task))
9a2e03d8
TH
776 wake_up_process(worker->task);
777}
778
b56c0d89 779/**
3989144f 780 * kthread_queue_work - queue a kthread_work
b56c0d89
TH
781 * @worker: target kthread_worker
782 * @work: kthread_work to queue
783 *
784 * Queue @work to work processor @task for async execution. @task
785 * must have been created with kthread_worker_create(). Returns %true
786 * if @work was successfully queued, %false if it was already pending.
8197b3d4
PM
787 *
788 * Reinitialize the work if it needs to be used by another worker.
789 * For example, when the worker was stopped and started again.
b56c0d89 790 */
3989144f 791bool kthread_queue_work(struct kthread_worker *worker,
b56c0d89
TH
792 struct kthread_work *work)
793{
794 bool ret = false;
795 unsigned long flags;
796
797 spin_lock_irqsave(&worker->lock, flags);
37be45d4 798 if (!queuing_blocked(worker, work)) {
3989144f 799 kthread_insert_work(worker, work, &worker->work_list);
b56c0d89
TH
800 ret = true;
801 }
802 spin_unlock_irqrestore(&worker->lock, flags);
803 return ret;
804}
3989144f 805EXPORT_SYMBOL_GPL(kthread_queue_work);
b56c0d89 806
22597dc3
PM
807/**
808 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
809 * delayed work when the timer expires.
810 * @__data: pointer to the data associated with the timer
811 *
812 * The format of the function is defined by struct timer_list.
813 * It should have been called from irqsafe timer with irq already off.
814 */
815void kthread_delayed_work_timer_fn(unsigned long __data)
816{
817 struct kthread_delayed_work *dwork =
818 (struct kthread_delayed_work *)__data;
819 struct kthread_work *work = &dwork->work;
820 struct kthread_worker *worker = work->worker;
821
822 /*
823 * This might happen when a pending work is reinitialized.
824 * It means that it is used a wrong way.
825 */
826 if (WARN_ON_ONCE(!worker))
827 return;
828
829 spin_lock(&worker->lock);
830 /* Work must not be used with >1 worker, see kthread_queue_work(). */
831 WARN_ON_ONCE(work->worker != worker);
832
833 /* Move the work from worker->delayed_work_list. */
834 WARN_ON_ONCE(list_empty(&work->node));
835 list_del_init(&work->node);
836 kthread_insert_work(worker, work, &worker->work_list);
837
838 spin_unlock(&worker->lock);
839}
840EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
841
842void __kthread_queue_delayed_work(struct kthread_worker *worker,
843 struct kthread_delayed_work *dwork,
844 unsigned long delay)
845{
846 struct timer_list *timer = &dwork->timer;
847 struct kthread_work *work = &dwork->work;
848
849 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn ||
850 timer->data != (unsigned long)dwork);
851
852 /*
853 * If @delay is 0, queue @dwork->work immediately. This is for
854 * both optimization and correctness. The earliest @timer can
855 * expire is on the closest next tick and delayed_work users depend
856 * on that there's no such delay when @delay is 0.
857 */
858 if (!delay) {
859 kthread_insert_work(worker, work, &worker->work_list);
860 return;
861 }
862
863 /* Be paranoid and try to detect possible races already now. */
864 kthread_insert_work_sanity_check(worker, work);
865
866 list_add(&work->node, &worker->delayed_work_list);
867 work->worker = worker;
868 timer_stats_timer_set_start_info(&dwork->timer);
869 timer->expires = jiffies + delay;
870 add_timer(timer);
871}
872
873/**
874 * kthread_queue_delayed_work - queue the associated kthread work
875 * after a delay.
876 * @worker: target kthread_worker
877 * @dwork: kthread_delayed_work to queue
878 * @delay: number of jiffies to wait before queuing
879 *
880 * If the work has not been pending it starts a timer that will queue
881 * the work after the given @delay. If @delay is zero, it queues the
882 * work immediately.
883 *
884 * Return: %false if the @work has already been pending. It means that
885 * either the timer was running or the work was queued. It returns %true
886 * otherwise.
887 */
888bool kthread_queue_delayed_work(struct kthread_worker *worker,
889 struct kthread_delayed_work *dwork,
890 unsigned long delay)
891{
892 struct kthread_work *work = &dwork->work;
893 unsigned long flags;
894 bool ret = false;
895
896 spin_lock_irqsave(&worker->lock, flags);
897
37be45d4 898 if (!queuing_blocked(worker, work)) {
22597dc3
PM
899 __kthread_queue_delayed_work(worker, dwork, delay);
900 ret = true;
901 }
902
903 spin_unlock_irqrestore(&worker->lock, flags);
904 return ret;
905}
906EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
907
9a2e03d8
TH
908struct kthread_flush_work {
909 struct kthread_work work;
910 struct completion done;
911};
912
913static void kthread_flush_work_fn(struct kthread_work *work)
914{
915 struct kthread_flush_work *fwork =
916 container_of(work, struct kthread_flush_work, work);
917 complete(&fwork->done);
918}
919
b56c0d89 920/**
3989144f 921 * kthread_flush_work - flush a kthread_work
b56c0d89
TH
922 * @work: work to flush
923 *
924 * If @work is queued or executing, wait for it to finish execution.
925 */
3989144f 926void kthread_flush_work(struct kthread_work *work)
b56c0d89 927{
46f3d976
TH
928 struct kthread_flush_work fwork = {
929 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
930 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
931 };
932 struct kthread_worker *worker;
933 bool noop = false;
934
46f3d976
TH
935 worker = work->worker;
936 if (!worker)
937 return;
b56c0d89 938
46f3d976 939 spin_lock_irq(&worker->lock);
8197b3d4
PM
940 /* Work must not be used with >1 worker, see kthread_queue_work(). */
941 WARN_ON_ONCE(work->worker != worker);
b56c0d89 942
46f3d976 943 if (!list_empty(&work->node))
3989144f 944 kthread_insert_work(worker, &fwork.work, work->node.next);
46f3d976 945 else if (worker->current_work == work)
3989144f
PM
946 kthread_insert_work(worker, &fwork.work,
947 worker->work_list.next);
46f3d976
TH
948 else
949 noop = true;
b56c0d89 950
46f3d976 951 spin_unlock_irq(&worker->lock);
b56c0d89 952
46f3d976
TH
953 if (!noop)
954 wait_for_completion(&fwork.done);
b56c0d89 955}
3989144f 956EXPORT_SYMBOL_GPL(kthread_flush_work);
b56c0d89 957
37be45d4
PM
958/*
959 * This function removes the work from the worker queue. Also it makes sure
960 * that it won't get queued later via the delayed work's timer.
961 *
962 * The work might still be in use when this function finishes. See the
963 * current_work proceed by the worker.
964 *
965 * Return: %true if @work was pending and successfully canceled,
966 * %false if @work was not pending
967 */
968static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
969 unsigned long *flags)
970{
971 /* Try to cancel the timer if exists. */
972 if (is_dwork) {
973 struct kthread_delayed_work *dwork =
974 container_of(work, struct kthread_delayed_work, work);
975 struct kthread_worker *worker = work->worker;
976
977 /*
978 * del_timer_sync() must be called to make sure that the timer
979 * callback is not running. The lock must be temporary released
980 * to avoid a deadlock with the callback. In the meantime,
981 * any queuing is blocked by setting the canceling counter.
982 */
983 work->canceling++;
984 spin_unlock_irqrestore(&worker->lock, *flags);
985 del_timer_sync(&dwork->timer);
986 spin_lock_irqsave(&worker->lock, *flags);
987 work->canceling--;
988 }
989
990 /*
991 * Try to remove the work from a worker list. It might either
992 * be from worker->work_list or from worker->delayed_work_list.
993 */
994 if (!list_empty(&work->node)) {
995 list_del_init(&work->node);
996 return true;
997 }
998
999 return false;
1000}
1001
9a6b06c8
PM
1002/**
1003 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1004 * @worker: kthread worker to use
1005 * @dwork: kthread delayed work to queue
1006 * @delay: number of jiffies to wait before queuing
1007 *
1008 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1009 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1010 * @work is guaranteed to be queued immediately.
1011 *
1012 * Return: %true if @dwork was pending and its timer was modified,
1013 * %false otherwise.
1014 *
1015 * A special case is when the work is being canceled in parallel.
1016 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1017 * or yet another kthread_mod_delayed_work() call. We let the other command
1018 * win and return %false here. The caller is supposed to synchronize these
1019 * operations a reasonable way.
1020 *
1021 * This function is safe to call from any context including IRQ handler.
1022 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1023 * for details.
1024 */
1025bool kthread_mod_delayed_work(struct kthread_worker *worker,
1026 struct kthread_delayed_work *dwork,
1027 unsigned long delay)
1028{
1029 struct kthread_work *work = &dwork->work;
1030 unsigned long flags;
1031 int ret = false;
1032
1033 spin_lock_irqsave(&worker->lock, flags);
1034
1035 /* Do not bother with canceling when never queued. */
1036 if (!work->worker)
1037 goto fast_queue;
1038
1039 /* Work must not be used with >1 worker, see kthread_queue_work() */
1040 WARN_ON_ONCE(work->worker != worker);
1041
1042 /* Do not fight with another command that is canceling this work. */
1043 if (work->canceling)
1044 goto out;
1045
1046 ret = __kthread_cancel_work(work, true, &flags);
1047fast_queue:
1048 __kthread_queue_delayed_work(worker, dwork, delay);
1049out:
1050 spin_unlock_irqrestore(&worker->lock, flags);
1051 return ret;
1052}
1053EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1054
37be45d4
PM
1055static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1056{
1057 struct kthread_worker *worker = work->worker;
1058 unsigned long flags;
1059 int ret = false;
1060
1061 if (!worker)
1062 goto out;
1063
1064 spin_lock_irqsave(&worker->lock, flags);
1065 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1066 WARN_ON_ONCE(work->worker != worker);
1067
1068 ret = __kthread_cancel_work(work, is_dwork, &flags);
1069
1070 if (worker->current_work != work)
1071 goto out_fast;
1072
1073 /*
1074 * The work is in progress and we need to wait with the lock released.
1075 * In the meantime, block any queuing by setting the canceling counter.
1076 */
1077 work->canceling++;
1078 spin_unlock_irqrestore(&worker->lock, flags);
1079 kthread_flush_work(work);
1080 spin_lock_irqsave(&worker->lock, flags);
1081 work->canceling--;
1082
1083out_fast:
1084 spin_unlock_irqrestore(&worker->lock, flags);
1085out:
1086 return ret;
1087}
1088
1089/**
1090 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1091 * @work: the kthread work to cancel
1092 *
1093 * Cancel @work and wait for its execution to finish. This function
1094 * can be used even if the work re-queues itself. On return from this
1095 * function, @work is guaranteed to be not pending or executing on any CPU.
1096 *
1097 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1098 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1099 *
1100 * The caller must ensure that the worker on which @work was last
1101 * queued can't be destroyed before this function returns.
1102 *
1103 * Return: %true if @work was pending, %false otherwise.
1104 */
1105bool kthread_cancel_work_sync(struct kthread_work *work)
1106{
1107 return __kthread_cancel_work_sync(work, false);
1108}
1109EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1110
1111/**
1112 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1113 * wait for it to finish.
1114 * @dwork: the kthread delayed work to cancel
1115 *
1116 * This is kthread_cancel_work_sync() for delayed works.
1117 *
1118 * Return: %true if @dwork was pending, %false otherwise.
1119 */
1120bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1121{
1122 return __kthread_cancel_work_sync(&dwork->work, true);
1123}
1124EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1125
b56c0d89 1126/**
3989144f 1127 * kthread_flush_worker - flush all current works on a kthread_worker
b56c0d89
TH
1128 * @worker: worker to flush
1129 *
1130 * Wait until all currently executing or pending works on @worker are
1131 * finished.
1132 */
3989144f 1133void kthread_flush_worker(struct kthread_worker *worker)
b56c0d89
TH
1134{
1135 struct kthread_flush_work fwork = {
1136 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1137 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1138 };
1139
3989144f 1140 kthread_queue_work(worker, &fwork.work);
b56c0d89
TH
1141 wait_for_completion(&fwork.done);
1142}
3989144f 1143EXPORT_SYMBOL_GPL(kthread_flush_worker);
35033fe9
PM
1144
1145/**
1146 * kthread_destroy_worker - destroy a kthread worker
1147 * @worker: worker to be destroyed
1148 *
1149 * Flush and destroy @worker. The simple flush is enough because the kthread
1150 * worker API is used only in trivial scenarios. There are no multi-step state
1151 * machines needed.
1152 */
1153void kthread_destroy_worker(struct kthread_worker *worker)
1154{
1155 struct task_struct *task;
1156
1157 task = worker->task;
1158 if (WARN_ON(!task))
1159 return;
1160
1161 kthread_flush_worker(worker);
1162 kthread_stop(task);
1163 WARN_ON(!list_empty(&worker->work_list));
1164 kfree(worker);
1165}
1166EXPORT_SYMBOL(kthread_destroy_worker);