Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* Kernel thread helper functions. |
3 | * Copyright (C) 2004 IBM Corporation, Rusty Russell. | |
9bf5b9eb | 4 | * Copyright (C) 2009 Red Hat, Inc. |
1da177e4 | 5 | * |
73c27992 | 6 | * Creation is done via kthreadd, so that we get a clean environment |
1da177e4 LT |
7 | * even if we're invoked from userspace (think modprobe, hotplug cpu, |
8 | * etc.). | |
9 | */ | |
ae7e81c0 | 10 | #include <uapi/linux/sched/types.h> |
9bf5b9eb CH |
11 | #include <linux/mm.h> |
12 | #include <linux/mmu_context.h> | |
1da177e4 | 13 | #include <linux/sched.h> |
9bf5b9eb | 14 | #include <linux/sched/mm.h> |
29930025 | 15 | #include <linux/sched/task.h> |
1da177e4 LT |
16 | #include <linux/kthread.h> |
17 | #include <linux/completion.h> | |
18 | #include <linux/err.h> | |
8af0c18a | 19 | #include <linux/cgroup.h> |
58568d2a | 20 | #include <linux/cpuset.h> |
1da177e4 LT |
21 | #include <linux/unistd.h> |
22 | #include <linux/file.h> | |
9984de1a | 23 | #include <linux/export.h> |
97d1f15b | 24 | #include <linux/mutex.h> |
b56c0d89 TH |
25 | #include <linux/slab.h> |
26 | #include <linux/freezer.h> | |
a74fb73c | 27 | #include <linux/ptrace.h> |
cd42d559 | 28 | #include <linux/uaccess.h> |
98fa15f3 | 29 | #include <linux/numa.h> |
9cc5b865 | 30 | #include <linux/sched/isolation.h> |
ad8d75ff | 31 | #include <trace/events/sched.h> |
1da177e4 | 32 | |
9bf5b9eb | 33 | |
73c27992 EB |
34 | static DEFINE_SPINLOCK(kthread_create_lock); |
35 | static LIST_HEAD(kthread_create_list); | |
36 | struct task_struct *kthreadd_task; | |
1da177e4 LT |
37 | |
38 | struct kthread_create_info | |
39 | { | |
73c27992 | 40 | /* Information passed to kthread() from kthreadd. */ |
1da177e4 LT |
41 | int (*threadfn)(void *data); |
42 | void *data; | |
207205a2 | 43 | int node; |
1da177e4 | 44 | |
73c27992 | 45 | /* Result passed back to kthread_create() from kthreadd. */ |
1da177e4 | 46 | struct task_struct *result; |
786235ee | 47 | struct completion *done; |
65f27f38 | 48 | |
73c27992 | 49 | struct list_head list; |
1da177e4 LT |
50 | }; |
51 | ||
63706172 | 52 | struct kthread { |
2a1d4460 TG |
53 | unsigned long flags; |
54 | unsigned int cpu; | |
52782c92 | 55 | int (*threadfn)(void *); |
82805ab7 | 56 | void *data; |
37c54f9b | 57 | mm_segment_t oldfs; |
2a1d4460 | 58 | struct completion parked; |
63706172 | 59 | struct completion exited; |
0b508bc9 | 60 | #ifdef CONFIG_BLK_CGROUP |
05e3db95 SL |
61 | struct cgroup_subsys_state *blkcg_css; |
62 | #endif | |
1da177e4 LT |
63 | }; |
64 | ||
2a1d4460 TG |
65 | enum KTHREAD_BITS { |
66 | KTHREAD_IS_PER_CPU = 0, | |
67 | KTHREAD_SHOULD_STOP, | |
68 | KTHREAD_SHOULD_PARK, | |
2a1d4460 TG |
69 | }; |
70 | ||
4ecdafc8 ON |
71 | static inline struct kthread *to_kthread(struct task_struct *k) |
72 | { | |
1da5c46f ON |
73 | WARN_ON(!(k->flags & PF_KTHREAD)); |
74 | return (__force void *)k->set_child_tid; | |
4ecdafc8 ON |
75 | } |
76 | ||
3a7956e2 PZ |
77 | /* |
78 | * Variant of to_kthread() that doesn't assume @p is a kthread. | |
79 | * | |
80 | * Per construction; when: | |
81 | * | |
82 | * (p->flags & PF_KTHREAD) && p->set_child_tid | |
83 | * | |
84 | * the task is both a kthread and struct kthread is persistent. However | |
85 | * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and | |
86 | * begin_new_exec()). | |
87 | */ | |
88 | static inline struct kthread *__to_kthread(struct task_struct *p) | |
89 | { | |
90 | void *kthread = (__force void *)p->set_child_tid; | |
91 | if (kthread && !(p->flags & PF_KTHREAD)) | |
92 | kthread = NULL; | |
93 | return kthread; | |
94 | } | |
95 | ||
00b89fe0 VS |
96 | void set_kthread_struct(struct task_struct *p) |
97 | { | |
98 | struct kthread *kthread; | |
99 | ||
100 | if (__to_kthread(p)) | |
101 | return; | |
102 | ||
103 | kthread = kzalloc(sizeof(*kthread), GFP_KERNEL); | |
104 | /* | |
105 | * We abuse ->set_child_tid to avoid the new member and because it | |
106 | * can't be wrongly copied by copy_process(). We also rely on fact | |
107 | * that the caller can't exec, so PF_KTHREAD can't be cleared. | |
108 | */ | |
109 | p->set_child_tid = (__force void __user *)kthread; | |
110 | } | |
111 | ||
1da5c46f ON |
112 | void free_kthread_struct(struct task_struct *k) |
113 | { | |
05e3db95 SL |
114 | struct kthread *kthread; |
115 | ||
1da5c46f ON |
116 | /* |
117 | * Can be NULL if this kthread was created by kernel_thread() | |
118 | * or if kmalloc() in kthread() failed. | |
119 | */ | |
05e3db95 | 120 | kthread = to_kthread(k); |
0b508bc9 | 121 | #ifdef CONFIG_BLK_CGROUP |
05e3db95 SL |
122 | WARN_ON_ONCE(kthread && kthread->blkcg_css); |
123 | #endif | |
124 | kfree(kthread); | |
1da5c46f ON |
125 | } |
126 | ||
9e37bd30 RD |
127 | /** |
128 | * kthread_should_stop - should this kthread return now? | |
129 | * | |
72fd4a35 | 130 | * When someone calls kthread_stop() on your kthread, it will be woken |
9e37bd30 RD |
131 | * and this will return true. You should then return, and your return |
132 | * value will be passed through to kthread_stop(). | |
133 | */ | |
2a1d4460 | 134 | bool kthread_should_stop(void) |
1da177e4 | 135 | { |
2a1d4460 | 136 | return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags); |
1da177e4 LT |
137 | } |
138 | EXPORT_SYMBOL(kthread_should_stop); | |
139 | ||
0121805d MK |
140 | bool __kthread_should_park(struct task_struct *k) |
141 | { | |
142 | return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags); | |
143 | } | |
144 | EXPORT_SYMBOL_GPL(__kthread_should_park); | |
145 | ||
2a1d4460 TG |
146 | /** |
147 | * kthread_should_park - should this kthread park now? | |
148 | * | |
149 | * When someone calls kthread_park() on your kthread, it will be woken | |
150 | * and this will return true. You should then do the necessary | |
151 | * cleanup and call kthread_parkme() | |
152 | * | |
153 | * Similar to kthread_should_stop(), but this keeps the thread alive | |
154 | * and in a park position. kthread_unpark() "restarts" the thread and | |
155 | * calls the thread function again. | |
156 | */ | |
157 | bool kthread_should_park(void) | |
158 | { | |
0121805d | 159 | return __kthread_should_park(current); |
2a1d4460 | 160 | } |
18896451 | 161 | EXPORT_SYMBOL_GPL(kthread_should_park); |
2a1d4460 | 162 | |
8a32c441 TH |
163 | /** |
164 | * kthread_freezable_should_stop - should this freezable kthread return now? | |
165 | * @was_frozen: optional out parameter, indicates whether %current was frozen | |
166 | * | |
167 | * kthread_should_stop() for freezable kthreads, which will enter | |
168 | * refrigerator if necessary. This function is safe from kthread_stop() / | |
169 | * freezer deadlock and freezable kthreads should use this function instead | |
170 | * of calling try_to_freeze() directly. | |
171 | */ | |
172 | bool kthread_freezable_should_stop(bool *was_frozen) | |
173 | { | |
174 | bool frozen = false; | |
175 | ||
176 | might_sleep(); | |
177 | ||
178 | if (unlikely(freezing(current))) | |
179 | frozen = __refrigerator(true); | |
180 | ||
181 | if (was_frozen) | |
182 | *was_frozen = frozen; | |
183 | ||
184 | return kthread_should_stop(); | |
185 | } | |
186 | EXPORT_SYMBOL_GPL(kthread_freezable_should_stop); | |
187 | ||
52782c92 BF |
188 | /** |
189 | * kthread_func - return the function specified on kthread creation | |
190 | * @task: kthread task in question | |
191 | * | |
192 | * Returns NULL if the task is not a kthread. | |
193 | */ | |
194 | void *kthread_func(struct task_struct *task) | |
195 | { | |
3a7956e2 PZ |
196 | struct kthread *kthread = __to_kthread(task); |
197 | if (kthread) | |
198 | return kthread->threadfn; | |
52782c92 BF |
199 | return NULL; |
200 | } | |
201 | EXPORT_SYMBOL_GPL(kthread_func); | |
202 | ||
82805ab7 TH |
203 | /** |
204 | * kthread_data - return data value specified on kthread creation | |
205 | * @task: kthread task in question | |
206 | * | |
207 | * Return the data value specified when kthread @task was created. | |
208 | * The caller is responsible for ensuring the validity of @task when | |
209 | * calling this function. | |
210 | */ | |
211 | void *kthread_data(struct task_struct *task) | |
212 | { | |
213 | return to_kthread(task)->data; | |
214 | } | |
52782c92 | 215 | EXPORT_SYMBOL_GPL(kthread_data); |
82805ab7 | 216 | |
cd42d559 | 217 | /** |
e700591a | 218 | * kthread_probe_data - speculative version of kthread_data() |
cd42d559 TH |
219 | * @task: possible kthread task in question |
220 | * | |
221 | * @task could be a kthread task. Return the data value specified when it | |
222 | * was created if accessible. If @task isn't a kthread task or its data is | |
223 | * inaccessible for any reason, %NULL is returned. This function requires | |
224 | * that @task itself is safe to dereference. | |
225 | */ | |
e700591a | 226 | void *kthread_probe_data(struct task_struct *task) |
cd42d559 | 227 | { |
3a7956e2 | 228 | struct kthread *kthread = __to_kthread(task); |
cd42d559 TH |
229 | void *data = NULL; |
230 | ||
3a7956e2 PZ |
231 | if (kthread) |
232 | copy_from_kernel_nofault(&data, &kthread->data, sizeof(data)); | |
cd42d559 TH |
233 | return data; |
234 | } | |
235 | ||
2a1d4460 TG |
236 | static void __kthread_parkme(struct kthread *self) |
237 | { | |
741a76b3 | 238 | for (;;) { |
1cef1150 PZ |
239 | /* |
240 | * TASK_PARKED is a special state; we must serialize against | |
241 | * possible pending wakeups to avoid store-store collisions on | |
242 | * task->state. | |
243 | * | |
244 | * Such a collision might possibly result in the task state | |
245 | * changin from TASK_PARKED and us failing the | |
246 | * wait_task_inactive() in kthread_park(). | |
247 | */ | |
248 | set_special_state(TASK_PARKED); | |
741a76b3 PZ |
249 | if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags)) |
250 | break; | |
1cef1150 | 251 | |
26c7295b LC |
252 | /* |
253 | * Thread is going to call schedule(), do not preempt it, | |
254 | * or the caller of kthread_park() may spend more time in | |
255 | * wait_task_inactive(). | |
256 | */ | |
257 | preempt_disable(); | |
f83ee19b | 258 | complete(&self->parked); |
26c7295b LC |
259 | schedule_preempt_disabled(); |
260 | preempt_enable(); | |
2a1d4460 | 261 | } |
2a1d4460 TG |
262 | __set_current_state(TASK_RUNNING); |
263 | } | |
264 | ||
265 | void kthread_parkme(void) | |
266 | { | |
267 | __kthread_parkme(to_kthread(current)); | |
268 | } | |
18896451 | 269 | EXPORT_SYMBOL_GPL(kthread_parkme); |
2a1d4460 | 270 | |
1da177e4 LT |
271 | static int kthread(void *_create) |
272 | { | |
63706172 | 273 | /* Copy data: it's on kthread's stack */ |
1da177e4 | 274 | struct kthread_create_info *create = _create; |
63706172 ON |
275 | int (*threadfn)(void *data) = create->threadfn; |
276 | void *data = create->data; | |
786235ee | 277 | struct completion *done; |
1da5c46f | 278 | struct kthread *self; |
63706172 | 279 | int ret; |
1da177e4 | 280 | |
00b89fe0 VS |
281 | set_kthread_struct(current); |
282 | self = to_kthread(current); | |
1da177e4 | 283 | |
786235ee TH |
284 | /* If user was SIGKILLed, I release the structure. */ |
285 | done = xchg(&create->done, NULL); | |
286 | if (!done) { | |
287 | kfree(create); | |
288 | do_exit(-EINTR); | |
289 | } | |
1da5c46f ON |
290 | |
291 | if (!self) { | |
292 | create->result = ERR_PTR(-ENOMEM); | |
293 | complete(done); | |
294 | do_exit(-ENOMEM); | |
295 | } | |
296 | ||
52782c92 | 297 | self->threadfn = threadfn; |
1da5c46f ON |
298 | self->data = data; |
299 | init_completion(&self->exited); | |
300 | init_completion(&self->parked); | |
301 | current->vfork_done = &self->exited; | |
302 | ||
1da177e4 | 303 | /* OK, tell user we're spawned, wait for stop or wakeup */ |
a076e4bc | 304 | __set_current_state(TASK_UNINTERRUPTIBLE); |
3217ab97 | 305 | create->result = current; |
26c7295b LC |
306 | /* |
307 | * Thread is going to call schedule(), do not preempt it, | |
308 | * or the creator may spend more time in wait_task_inactive(). | |
309 | */ | |
310 | preempt_disable(); | |
786235ee | 311 | complete(done); |
26c7295b LC |
312 | schedule_preempt_disabled(); |
313 | preempt_enable(); | |
1da177e4 | 314 | |
63706172 | 315 | ret = -EINTR; |
1da5c46f | 316 | if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) { |
77f88796 | 317 | cgroup_kthread_ready(); |
1da5c46f | 318 | __kthread_parkme(self); |
2a1d4460 TG |
319 | ret = threadfn(data); |
320 | } | |
63706172 | 321 | do_exit(ret); |
1da177e4 LT |
322 | } |
323 | ||
cb5021ca | 324 | /* called from kernel_clone() to get node information for about to be created task */ |
207205a2 ED |
325 | int tsk_fork_get_node(struct task_struct *tsk) |
326 | { | |
327 | #ifdef CONFIG_NUMA | |
328 | if (tsk == kthreadd_task) | |
329 | return tsk->pref_node_fork; | |
330 | #endif | |
81c98869 | 331 | return NUMA_NO_NODE; |
207205a2 ED |
332 | } |
333 | ||
73c27992 | 334 | static void create_kthread(struct kthread_create_info *create) |
1da177e4 | 335 | { |
1da177e4 LT |
336 | int pid; |
337 | ||
207205a2 ED |
338 | #ifdef CONFIG_NUMA |
339 | current->pref_node_fork = create->node; | |
340 | #endif | |
1da177e4 LT |
341 | /* We want our own signal handler (we take no signals by default). */ |
342 | pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD); | |
cdd140bd | 343 | if (pid < 0) { |
786235ee TH |
344 | /* If user was SIGKILLed, I release the structure. */ |
345 | struct completion *done = xchg(&create->done, NULL); | |
346 | ||
347 | if (!done) { | |
348 | kfree(create); | |
349 | return; | |
350 | } | |
1da177e4 | 351 | create->result = ERR_PTR(pid); |
786235ee | 352 | complete(done); |
cdd140bd | 353 | } |
1da177e4 LT |
354 | } |
355 | ||
c0b942a7 NI |
356 | static __printf(4, 0) |
357 | struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data), | |
255451e4 PM |
358 | void *data, int node, |
359 | const char namefmt[], | |
360 | va_list args) | |
1da177e4 | 361 | { |
786235ee TH |
362 | DECLARE_COMPLETION_ONSTACK(done); |
363 | struct task_struct *task; | |
364 | struct kthread_create_info *create = kmalloc(sizeof(*create), | |
365 | GFP_KERNEL); | |
366 | ||
367 | if (!create) | |
368 | return ERR_PTR(-ENOMEM); | |
369 | create->threadfn = threadfn; | |
370 | create->data = data; | |
371 | create->node = node; | |
372 | create->done = &done; | |
73c27992 EB |
373 | |
374 | spin_lock(&kthread_create_lock); | |
786235ee | 375 | list_add_tail(&create->list, &kthread_create_list); |
73c27992 EB |
376 | spin_unlock(&kthread_create_lock); |
377 | ||
cbd9b67b | 378 | wake_up_process(kthreadd_task); |
786235ee TH |
379 | /* |
380 | * Wait for completion in killable state, for I might be chosen by | |
381 | * the OOM killer while kthreadd is trying to allocate memory for | |
382 | * new kernel thread. | |
383 | */ | |
384 | if (unlikely(wait_for_completion_killable(&done))) { | |
385 | /* | |
386 | * If I was SIGKILLed before kthreadd (or new kernel thread) | |
387 | * calls complete(), leave the cleanup of this structure to | |
388 | * that thread. | |
389 | */ | |
390 | if (xchg(&create->done, NULL)) | |
8fe6929c | 391 | return ERR_PTR(-EINTR); |
786235ee TH |
392 | /* |
393 | * kthreadd (or new kernel thread) will call complete() | |
394 | * shortly. | |
395 | */ | |
396 | wait_for_completion(&done); | |
397 | } | |
398 | task = create->result; | |
399 | if (!IS_ERR(task)) { | |
c9b5f501 | 400 | static const struct sched_param param = { .sched_priority = 0 }; |
3e536e22 | 401 | char name[TASK_COMM_LEN]; |
1c99315b | 402 | |
3e536e22 SD |
403 | /* |
404 | * task is already visible to other tasks, so updating | |
405 | * COMM must be protected. | |
406 | */ | |
407 | vsnprintf(name, sizeof(name), namefmt, args); | |
408 | set_task_comm(task, name); | |
1c99315b ON |
409 | /* |
410 | * root may have changed our (kthreadd's) priority or CPU mask. | |
411 | * The kernel thread should not inherit these properties. | |
412 | */ | |
786235ee | 413 | sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m); |
9cc5b865 MT |
414 | set_cpus_allowed_ptr(task, |
415 | housekeeping_cpumask(HK_FLAG_KTHREAD)); | |
1da177e4 | 416 | } |
786235ee TH |
417 | kfree(create); |
418 | return task; | |
1da177e4 | 419 | } |
255451e4 PM |
420 | |
421 | /** | |
422 | * kthread_create_on_node - create a kthread. | |
423 | * @threadfn: the function to run until signal_pending(current). | |
424 | * @data: data ptr for @threadfn. | |
425 | * @node: task and thread structures for the thread are allocated on this node | |
426 | * @namefmt: printf-style name for the thread. | |
427 | * | |
428 | * Description: This helper function creates and names a kernel | |
429 | * thread. The thread will be stopped: use wake_up_process() to start | |
430 | * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and | |
431 | * is affine to all CPUs. | |
432 | * | |
433 | * If thread is going to be bound on a particular cpu, give its node | |
434 | * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE. | |
435 | * When woken, the thread will run @threadfn() with @data as its | |
436 | * argument. @threadfn() can either call do_exit() directly if it is a | |
437 | * standalone thread for which no one will call kthread_stop(), or | |
438 | * return when 'kthread_should_stop()' is true (which means | |
439 | * kthread_stop() has been called). The return value should be zero | |
440 | * or a negative error number; it will be passed to kthread_stop(). | |
441 | * | |
442 | * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR). | |
443 | */ | |
444 | struct task_struct *kthread_create_on_node(int (*threadfn)(void *data), | |
445 | void *data, int node, | |
446 | const char namefmt[], | |
447 | ...) | |
448 | { | |
449 | struct task_struct *task; | |
450 | va_list args; | |
451 | ||
452 | va_start(args, namefmt); | |
453 | task = __kthread_create_on_node(threadfn, data, node, namefmt, args); | |
454 | va_end(args); | |
455 | ||
456 | return task; | |
457 | } | |
207205a2 | 458 | EXPORT_SYMBOL(kthread_create_on_node); |
1da177e4 | 459 | |
2f064a59 | 460 | static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state) |
2a1d4460 | 461 | { |
25834c73 PZ |
462 | unsigned long flags; |
463 | ||
f2530dc7 TG |
464 | if (!wait_task_inactive(p, state)) { |
465 | WARN_ON(1); | |
466 | return; | |
467 | } | |
25834c73 | 468 | |
2a1d4460 | 469 | /* It's safe because the task is inactive. */ |
25834c73 PZ |
470 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
471 | do_set_cpus_allowed(p, mask); | |
14a40ffc | 472 | p->flags |= PF_NO_SETAFFINITY; |
25834c73 PZ |
473 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
474 | } | |
475 | ||
2f064a59 | 476 | static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state) |
25834c73 PZ |
477 | { |
478 | __kthread_bind_mask(p, cpumask_of(cpu), state); | |
479 | } | |
480 | ||
481 | void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask) | |
482 | { | |
483 | __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE); | |
2a1d4460 TG |
484 | } |
485 | ||
881232b7 PZ |
486 | /** |
487 | * kthread_bind - bind a just-created kthread to a cpu. | |
488 | * @p: thread created by kthread_create(). | |
489 | * @cpu: cpu (might not be online, must be possible) for @k to run on. | |
490 | * | |
491 | * Description: This function is equivalent to set_cpus_allowed(), | |
492 | * except that @cpu doesn't need to be online, and the thread must be | |
493 | * stopped (i.e., just returned from kthread_create()). | |
494 | */ | |
495 | void kthread_bind(struct task_struct *p, unsigned int cpu) | |
496 | { | |
f2530dc7 | 497 | __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE); |
881232b7 PZ |
498 | } |
499 | EXPORT_SYMBOL(kthread_bind); | |
500 | ||
2a1d4460 TG |
501 | /** |
502 | * kthread_create_on_cpu - Create a cpu bound kthread | |
503 | * @threadfn: the function to run until signal_pending(current). | |
504 | * @data: data ptr for @threadfn. | |
505 | * @cpu: The cpu on which the thread should be bound, | |
506 | * @namefmt: printf-style name for the thread. Format is restricted | |
507 | * to "name.*%u". Code fills in cpu number. | |
508 | * | |
509 | * Description: This helper function creates and names a kernel thread | |
2a1d4460 TG |
510 | */ |
511 | struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data), | |
512 | void *data, unsigned int cpu, | |
513 | const char *namefmt) | |
514 | { | |
515 | struct task_struct *p; | |
516 | ||
10922838 | 517 | p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt, |
2a1d4460 TG |
518 | cpu); |
519 | if (IS_ERR(p)) | |
520 | return p; | |
a65d4096 PM |
521 | kthread_bind(p, cpu); |
522 | /* CPU hotplug need to bind once again when unparking the thread. */ | |
2a1d4460 | 523 | to_kthread(p)->cpu = cpu; |
2a1d4460 TG |
524 | return p; |
525 | } | |
526 | ||
ac687e6e PZ |
527 | void kthread_set_per_cpu(struct task_struct *k, int cpu) |
528 | { | |
529 | struct kthread *kthread = to_kthread(k); | |
530 | if (!kthread) | |
531 | return; | |
532 | ||
533 | WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY)); | |
534 | ||
535 | if (cpu < 0) { | |
536 | clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
537 | return; | |
538 | } | |
539 | ||
540 | kthread->cpu = cpu; | |
541 | set_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
542 | } | |
543 | ||
3a7956e2 | 544 | bool kthread_is_per_cpu(struct task_struct *p) |
ac687e6e | 545 | { |
3a7956e2 | 546 | struct kthread *kthread = __to_kthread(p); |
ac687e6e PZ |
547 | if (!kthread) |
548 | return false; | |
549 | ||
550 | return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
551 | } | |
552 | ||
cf380a4a ON |
553 | /** |
554 | * kthread_unpark - unpark a thread created by kthread_create(). | |
555 | * @k: thread created by kthread_create(). | |
556 | * | |
557 | * Sets kthread_should_park() for @k to return false, wakes it, and | |
558 | * waits for it to return. If the thread is marked percpu then its | |
559 | * bound to the cpu again. | |
560 | */ | |
561 | void kthread_unpark(struct task_struct *k) | |
f2530dc7 | 562 | { |
cf380a4a ON |
563 | struct kthread *kthread = to_kthread(k); |
564 | ||
f2530dc7 | 565 | /* |
85f1abe0 PZ |
566 | * Newly created kthread was parked when the CPU was offline. |
567 | * The binding was lost and we need to set it again. | |
f2530dc7 | 568 | */ |
85f1abe0 PZ |
569 | if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags)) |
570 | __kthread_bind(k, kthread->cpu, TASK_PARKED); | |
571 | ||
572 | clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags); | |
1cef1150 PZ |
573 | /* |
574 | * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup. | |
575 | */ | |
85f1abe0 | 576 | wake_up_state(k, TASK_PARKED); |
f2530dc7 | 577 | } |
18896451 | 578 | EXPORT_SYMBOL_GPL(kthread_unpark); |
2a1d4460 TG |
579 | |
580 | /** | |
581 | * kthread_park - park a thread created by kthread_create(). | |
582 | * @k: thread created by kthread_create(). | |
583 | * | |
584 | * Sets kthread_should_park() for @k to return true, wakes it, and | |
585 | * waits for it to return. This can also be called after kthread_create() | |
586 | * instead of calling wake_up_process(): the thread will park without | |
587 | * calling threadfn(). | |
588 | * | |
589 | * Returns 0 if the thread is parked, -ENOSYS if the thread exited. | |
590 | * If called by the kthread itself just the park bit is set. | |
591 | */ | |
592 | int kthread_park(struct task_struct *k) | |
593 | { | |
cf380a4a ON |
594 | struct kthread *kthread = to_kthread(k); |
595 | ||
596 | if (WARN_ON(k->flags & PF_EXITING)) | |
597 | return -ENOSYS; | |
598 | ||
f83ee19b PZ |
599 | if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags))) |
600 | return -EBUSY; | |
601 | ||
85f1abe0 PZ |
602 | set_bit(KTHREAD_SHOULD_PARK, &kthread->flags); |
603 | if (k != current) { | |
604 | wake_up_process(k); | |
1cef1150 PZ |
605 | /* |
606 | * Wait for __kthread_parkme() to complete(), this means we | |
607 | * _will_ have TASK_PARKED and are about to call schedule(). | |
608 | */ | |
85f1abe0 | 609 | wait_for_completion(&kthread->parked); |
1cef1150 PZ |
610 | /* |
611 | * Now wait for that schedule() to complete and the task to | |
612 | * get scheduled out. | |
613 | */ | |
614 | WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED)); | |
2a1d4460 | 615 | } |
cf380a4a ON |
616 | |
617 | return 0; | |
2a1d4460 | 618 | } |
18896451 | 619 | EXPORT_SYMBOL_GPL(kthread_park); |
2a1d4460 | 620 | |
9e37bd30 RD |
621 | /** |
622 | * kthread_stop - stop a thread created by kthread_create(). | |
623 | * @k: thread created by kthread_create(). | |
624 | * | |
625 | * Sets kthread_should_stop() for @k to return true, wakes it, and | |
9ae26027 ON |
626 | * waits for it to exit. This can also be called after kthread_create() |
627 | * instead of calling wake_up_process(): the thread will exit without | |
628 | * calling threadfn(). | |
629 | * | |
630 | * If threadfn() may call do_exit() itself, the caller must ensure | |
631 | * task_struct can't go away. | |
9e37bd30 RD |
632 | * |
633 | * Returns the result of threadfn(), or %-EINTR if wake_up_process() | |
634 | * was never called. | |
635 | */ | |
1da177e4 LT |
636 | int kthread_stop(struct task_struct *k) |
637 | { | |
b5c5442b | 638 | struct kthread *kthread; |
1da177e4 LT |
639 | int ret; |
640 | ||
0a16b607 | 641 | trace_sched_kthread_stop(k); |
b5c5442b ON |
642 | |
643 | get_task_struct(k); | |
efb29fbf ON |
644 | kthread = to_kthread(k); |
645 | set_bit(KTHREAD_SHOULD_STOP, &kthread->flags); | |
cf380a4a | 646 | kthread_unpark(k); |
efb29fbf ON |
647 | wake_up_process(k); |
648 | wait_for_completion(&kthread->exited); | |
63706172 | 649 | ret = k->exit_code; |
1da177e4 | 650 | put_task_struct(k); |
0a16b607 | 651 | |
b5c5442b | 652 | trace_sched_kthread_stop_ret(ret); |
1da177e4 LT |
653 | return ret; |
654 | } | |
52e92e57 | 655 | EXPORT_SYMBOL(kthread_stop); |
1da177e4 | 656 | |
e804a4a4 | 657 | int kthreadd(void *unused) |
1da177e4 | 658 | { |
73c27992 | 659 | struct task_struct *tsk = current; |
1da177e4 | 660 | |
e804a4a4 | 661 | /* Setup a clean context for our children to inherit. */ |
73c27992 | 662 | set_task_comm(tsk, "kthreadd"); |
10ab825b | 663 | ignore_signals(tsk); |
9cc5b865 | 664 | set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD)); |
aee4faa4 | 665 | set_mems_allowed(node_states[N_MEMORY]); |
73c27992 | 666 | |
34b087e4 | 667 | current->flags |= PF_NOFREEZE; |
77f88796 | 668 | cgroup_init_kthreadd(); |
73c27992 EB |
669 | |
670 | for (;;) { | |
671 | set_current_state(TASK_INTERRUPTIBLE); | |
672 | if (list_empty(&kthread_create_list)) | |
673 | schedule(); | |
674 | __set_current_state(TASK_RUNNING); | |
675 | ||
676 | spin_lock(&kthread_create_lock); | |
677 | while (!list_empty(&kthread_create_list)) { | |
678 | struct kthread_create_info *create; | |
679 | ||
680 | create = list_entry(kthread_create_list.next, | |
681 | struct kthread_create_info, list); | |
682 | list_del_init(&create->list); | |
683 | spin_unlock(&kthread_create_lock); | |
684 | ||
685 | create_kthread(create); | |
686 | ||
687 | spin_lock(&kthread_create_lock); | |
688 | } | |
689 | spin_unlock(&kthread_create_lock); | |
690 | } | |
691 | ||
692 | return 0; | |
693 | } | |
b56c0d89 | 694 | |
3989144f | 695 | void __kthread_init_worker(struct kthread_worker *worker, |
4f32e9b1 YZ |
696 | const char *name, |
697 | struct lock_class_key *key) | |
698 | { | |
dbf52682 | 699 | memset(worker, 0, sizeof(struct kthread_worker)); |
fe99a4f4 | 700 | raw_spin_lock_init(&worker->lock); |
4f32e9b1 YZ |
701 | lockdep_set_class_and_name(&worker->lock, key, name); |
702 | INIT_LIST_HEAD(&worker->work_list); | |
22597dc3 | 703 | INIT_LIST_HEAD(&worker->delayed_work_list); |
4f32e9b1 | 704 | } |
3989144f | 705 | EXPORT_SYMBOL_GPL(__kthread_init_worker); |
4f32e9b1 | 706 | |
b56c0d89 TH |
707 | /** |
708 | * kthread_worker_fn - kthread function to process kthread_worker | |
709 | * @worker_ptr: pointer to initialized kthread_worker | |
710 | * | |
fbae2d44 PM |
711 | * This function implements the main cycle of kthread worker. It processes |
712 | * work_list until it is stopped with kthread_stop(). It sleeps when the queue | |
713 | * is empty. | |
b56c0d89 | 714 | * |
fbae2d44 PM |
715 | * The works are not allowed to keep any locks, disable preemption or interrupts |
716 | * when they finish. There is defined a safe point for freezing when one work | |
717 | * finishes and before a new one is started. | |
8197b3d4 PM |
718 | * |
719 | * Also the works must not be handled by more than one worker at the same time, | |
720 | * see also kthread_queue_work(). | |
b56c0d89 TH |
721 | */ |
722 | int kthread_worker_fn(void *worker_ptr) | |
723 | { | |
724 | struct kthread_worker *worker = worker_ptr; | |
725 | struct kthread_work *work; | |
726 | ||
fbae2d44 PM |
727 | /* |
728 | * FIXME: Update the check and remove the assignment when all kthread | |
729 | * worker users are created using kthread_create_worker*() functions. | |
730 | */ | |
731 | WARN_ON(worker->task && worker->task != current); | |
b56c0d89 | 732 | worker->task = current; |
dbf52682 PM |
733 | |
734 | if (worker->flags & KTW_FREEZABLE) | |
735 | set_freezable(); | |
736 | ||
b56c0d89 TH |
737 | repeat: |
738 | set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */ | |
739 | ||
740 | if (kthread_should_stop()) { | |
741 | __set_current_state(TASK_RUNNING); | |
fe99a4f4 | 742 | raw_spin_lock_irq(&worker->lock); |
b56c0d89 | 743 | worker->task = NULL; |
fe99a4f4 | 744 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 TH |
745 | return 0; |
746 | } | |
747 | ||
748 | work = NULL; | |
fe99a4f4 | 749 | raw_spin_lock_irq(&worker->lock); |
b56c0d89 TH |
750 | if (!list_empty(&worker->work_list)) { |
751 | work = list_first_entry(&worker->work_list, | |
752 | struct kthread_work, node); | |
753 | list_del_init(&work->node); | |
754 | } | |
46f3d976 | 755 | worker->current_work = work; |
fe99a4f4 | 756 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 TH |
757 | |
758 | if (work) { | |
f630c7c6 | 759 | kthread_work_func_t func = work->func; |
b56c0d89 | 760 | __set_current_state(TASK_RUNNING); |
f630c7c6 | 761 | trace_sched_kthread_work_execute_start(work); |
b56c0d89 | 762 | work->func(work); |
f630c7c6 RC |
763 | /* |
764 | * Avoid dereferencing work after this point. The trace | |
765 | * event only cares about the address. | |
766 | */ | |
767 | trace_sched_kthread_work_execute_end(work, func); | |
b56c0d89 TH |
768 | } else if (!freezing(current)) |
769 | schedule(); | |
770 | ||
771 | try_to_freeze(); | |
22cf8bc6 | 772 | cond_resched(); |
b56c0d89 TH |
773 | goto repeat; |
774 | } | |
775 | EXPORT_SYMBOL_GPL(kthread_worker_fn); | |
776 | ||
c0b942a7 | 777 | static __printf(3, 0) struct kthread_worker * |
dbf52682 PM |
778 | __kthread_create_worker(int cpu, unsigned int flags, |
779 | const char namefmt[], va_list args) | |
fbae2d44 PM |
780 | { |
781 | struct kthread_worker *worker; | |
782 | struct task_struct *task; | |
98fa15f3 | 783 | int node = NUMA_NO_NODE; |
fbae2d44 PM |
784 | |
785 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
786 | if (!worker) | |
787 | return ERR_PTR(-ENOMEM); | |
788 | ||
789 | kthread_init_worker(worker); | |
790 | ||
8fb9dcbd ON |
791 | if (cpu >= 0) |
792 | node = cpu_to_node(cpu); | |
fbae2d44 | 793 | |
8fb9dcbd ON |
794 | task = __kthread_create_on_node(kthread_worker_fn, worker, |
795 | node, namefmt, args); | |
fbae2d44 PM |
796 | if (IS_ERR(task)) |
797 | goto fail_task; | |
798 | ||
8fb9dcbd ON |
799 | if (cpu >= 0) |
800 | kthread_bind(task, cpu); | |
801 | ||
dbf52682 | 802 | worker->flags = flags; |
fbae2d44 PM |
803 | worker->task = task; |
804 | wake_up_process(task); | |
805 | return worker; | |
806 | ||
807 | fail_task: | |
808 | kfree(worker); | |
809 | return ERR_CAST(task); | |
810 | } | |
811 | ||
812 | /** | |
813 | * kthread_create_worker - create a kthread worker | |
dbf52682 | 814 | * @flags: flags modifying the default behavior of the worker |
fbae2d44 PM |
815 | * @namefmt: printf-style name for the kthread worker (task). |
816 | * | |
817 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
818 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) | |
819 | * when the worker was SIGKILLed. | |
820 | */ | |
821 | struct kthread_worker * | |
dbf52682 | 822 | kthread_create_worker(unsigned int flags, const char namefmt[], ...) |
fbae2d44 PM |
823 | { |
824 | struct kthread_worker *worker; | |
825 | va_list args; | |
826 | ||
827 | va_start(args, namefmt); | |
dbf52682 | 828 | worker = __kthread_create_worker(-1, flags, namefmt, args); |
fbae2d44 PM |
829 | va_end(args); |
830 | ||
831 | return worker; | |
832 | } | |
833 | EXPORT_SYMBOL(kthread_create_worker); | |
834 | ||
835 | /** | |
836 | * kthread_create_worker_on_cpu - create a kthread worker and bind it | |
7b7b8a2c | 837 | * to a given CPU and the associated NUMA node. |
fbae2d44 | 838 | * @cpu: CPU number |
dbf52682 | 839 | * @flags: flags modifying the default behavior of the worker |
fbae2d44 PM |
840 | * @namefmt: printf-style name for the kthread worker (task). |
841 | * | |
842 | * Use a valid CPU number if you want to bind the kthread worker | |
843 | * to the given CPU and the associated NUMA node. | |
844 | * | |
845 | * A good practice is to add the cpu number also into the worker name. | |
846 | * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu). | |
847 | * | |
ebb2bdce PM |
848 | * CPU hotplug: |
849 | * The kthread worker API is simple and generic. It just provides a way | |
850 | * to create, use, and destroy workers. | |
851 | * | |
852 | * It is up to the API user how to handle CPU hotplug. They have to decide | |
853 | * how to handle pending work items, prevent queuing new ones, and | |
854 | * restore the functionality when the CPU goes off and on. There are a | |
855 | * few catches: | |
856 | * | |
857 | * - CPU affinity gets lost when it is scheduled on an offline CPU. | |
858 | * | |
859 | * - The worker might not exist when the CPU was off when the user | |
860 | * created the workers. | |
861 | * | |
862 | * Good practice is to implement two CPU hotplug callbacks and to | |
863 | * destroy/create the worker when the CPU goes down/up. | |
864 | * | |
865 | * Return: | |
866 | * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
fbae2d44 PM |
867 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) |
868 | * when the worker was SIGKILLed. | |
869 | */ | |
870 | struct kthread_worker * | |
dbf52682 PM |
871 | kthread_create_worker_on_cpu(int cpu, unsigned int flags, |
872 | const char namefmt[], ...) | |
fbae2d44 PM |
873 | { |
874 | struct kthread_worker *worker; | |
875 | va_list args; | |
876 | ||
877 | va_start(args, namefmt); | |
dbf52682 | 878 | worker = __kthread_create_worker(cpu, flags, namefmt, args); |
fbae2d44 PM |
879 | va_end(args); |
880 | ||
881 | return worker; | |
882 | } | |
883 | EXPORT_SYMBOL(kthread_create_worker_on_cpu); | |
884 | ||
37be45d4 PM |
885 | /* |
886 | * Returns true when the work could not be queued at the moment. | |
887 | * It happens when it is already pending in a worker list | |
888 | * or when it is being cancelled. | |
889 | */ | |
890 | static inline bool queuing_blocked(struct kthread_worker *worker, | |
891 | struct kthread_work *work) | |
892 | { | |
893 | lockdep_assert_held(&worker->lock); | |
894 | ||
895 | return !list_empty(&work->node) || work->canceling; | |
896 | } | |
897 | ||
8197b3d4 PM |
898 | static void kthread_insert_work_sanity_check(struct kthread_worker *worker, |
899 | struct kthread_work *work) | |
900 | { | |
901 | lockdep_assert_held(&worker->lock); | |
902 | WARN_ON_ONCE(!list_empty(&work->node)); | |
903 | /* Do not use a work with >1 worker, see kthread_queue_work() */ | |
904 | WARN_ON_ONCE(work->worker && work->worker != worker); | |
905 | } | |
906 | ||
9a2e03d8 | 907 | /* insert @work before @pos in @worker */ |
3989144f | 908 | static void kthread_insert_work(struct kthread_worker *worker, |
8197b3d4 PM |
909 | struct kthread_work *work, |
910 | struct list_head *pos) | |
9a2e03d8 | 911 | { |
8197b3d4 | 912 | kthread_insert_work_sanity_check(worker, work); |
9a2e03d8 | 913 | |
f630c7c6 RC |
914 | trace_sched_kthread_work_queue_work(worker, work); |
915 | ||
9a2e03d8 | 916 | list_add_tail(&work->node, pos); |
46f3d976 | 917 | work->worker = worker; |
ed1403ec | 918 | if (!worker->current_work && likely(worker->task)) |
9a2e03d8 TH |
919 | wake_up_process(worker->task); |
920 | } | |
921 | ||
b56c0d89 | 922 | /** |
3989144f | 923 | * kthread_queue_work - queue a kthread_work |
b56c0d89 TH |
924 | * @worker: target kthread_worker |
925 | * @work: kthread_work to queue | |
926 | * | |
927 | * Queue @work to work processor @task for async execution. @task | |
928 | * must have been created with kthread_worker_create(). Returns %true | |
929 | * if @work was successfully queued, %false if it was already pending. | |
8197b3d4 PM |
930 | * |
931 | * Reinitialize the work if it needs to be used by another worker. | |
932 | * For example, when the worker was stopped and started again. | |
b56c0d89 | 933 | */ |
3989144f | 934 | bool kthread_queue_work(struct kthread_worker *worker, |
b56c0d89 TH |
935 | struct kthread_work *work) |
936 | { | |
937 | bool ret = false; | |
938 | unsigned long flags; | |
939 | ||
fe99a4f4 | 940 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 | 941 | if (!queuing_blocked(worker, work)) { |
3989144f | 942 | kthread_insert_work(worker, work, &worker->work_list); |
b56c0d89 TH |
943 | ret = true; |
944 | } | |
fe99a4f4 | 945 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
b56c0d89 TH |
946 | return ret; |
947 | } | |
3989144f | 948 | EXPORT_SYMBOL_GPL(kthread_queue_work); |
b56c0d89 | 949 | |
22597dc3 PM |
950 | /** |
951 | * kthread_delayed_work_timer_fn - callback that queues the associated kthread | |
952 | * delayed work when the timer expires. | |
fe5c3b69 | 953 | * @t: pointer to the expired timer |
22597dc3 PM |
954 | * |
955 | * The format of the function is defined by struct timer_list. | |
956 | * It should have been called from irqsafe timer with irq already off. | |
957 | */ | |
fe5c3b69 | 958 | void kthread_delayed_work_timer_fn(struct timer_list *t) |
22597dc3 | 959 | { |
fe5c3b69 | 960 | struct kthread_delayed_work *dwork = from_timer(dwork, t, timer); |
22597dc3 PM |
961 | struct kthread_work *work = &dwork->work; |
962 | struct kthread_worker *worker = work->worker; | |
ad01423a | 963 | unsigned long flags; |
22597dc3 PM |
964 | |
965 | /* | |
966 | * This might happen when a pending work is reinitialized. | |
967 | * It means that it is used a wrong way. | |
968 | */ | |
969 | if (WARN_ON_ONCE(!worker)) | |
970 | return; | |
971 | ||
ad01423a | 972 | raw_spin_lock_irqsave(&worker->lock, flags); |
22597dc3 PM |
973 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
974 | WARN_ON_ONCE(work->worker != worker); | |
975 | ||
976 | /* Move the work from worker->delayed_work_list. */ | |
977 | WARN_ON_ONCE(list_empty(&work->node)); | |
978 | list_del_init(&work->node); | |
6993d0fd Z |
979 | if (!work->canceling) |
980 | kthread_insert_work(worker, work, &worker->work_list); | |
22597dc3 | 981 | |
ad01423a | 982 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
22597dc3 PM |
983 | } |
984 | EXPORT_SYMBOL(kthread_delayed_work_timer_fn); | |
985 | ||
bc88f85c BD |
986 | static void __kthread_queue_delayed_work(struct kthread_worker *worker, |
987 | struct kthread_delayed_work *dwork, | |
988 | unsigned long delay) | |
22597dc3 PM |
989 | { |
990 | struct timer_list *timer = &dwork->timer; | |
991 | struct kthread_work *work = &dwork->work; | |
992 | ||
0a5b4128 ST |
993 | WARN_ON_FUNCTION_MISMATCH(timer->function, |
994 | kthread_delayed_work_timer_fn); | |
22597dc3 PM |
995 | |
996 | /* | |
997 | * If @delay is 0, queue @dwork->work immediately. This is for | |
998 | * both optimization and correctness. The earliest @timer can | |
999 | * expire is on the closest next tick and delayed_work users depend | |
1000 | * on that there's no such delay when @delay is 0. | |
1001 | */ | |
1002 | if (!delay) { | |
1003 | kthread_insert_work(worker, work, &worker->work_list); | |
1004 | return; | |
1005 | } | |
1006 | ||
1007 | /* Be paranoid and try to detect possible races already now. */ | |
1008 | kthread_insert_work_sanity_check(worker, work); | |
1009 | ||
1010 | list_add(&work->node, &worker->delayed_work_list); | |
1011 | work->worker = worker; | |
22597dc3 PM |
1012 | timer->expires = jiffies + delay; |
1013 | add_timer(timer); | |
1014 | } | |
1015 | ||
1016 | /** | |
1017 | * kthread_queue_delayed_work - queue the associated kthread work | |
1018 | * after a delay. | |
1019 | * @worker: target kthread_worker | |
1020 | * @dwork: kthread_delayed_work to queue | |
1021 | * @delay: number of jiffies to wait before queuing | |
1022 | * | |
1023 | * If the work has not been pending it starts a timer that will queue | |
1024 | * the work after the given @delay. If @delay is zero, it queues the | |
1025 | * work immediately. | |
1026 | * | |
1027 | * Return: %false if the @work has already been pending. It means that | |
1028 | * either the timer was running or the work was queued. It returns %true | |
1029 | * otherwise. | |
1030 | */ | |
1031 | bool kthread_queue_delayed_work(struct kthread_worker *worker, | |
1032 | struct kthread_delayed_work *dwork, | |
1033 | unsigned long delay) | |
1034 | { | |
1035 | struct kthread_work *work = &dwork->work; | |
1036 | unsigned long flags; | |
1037 | bool ret = false; | |
1038 | ||
fe99a4f4 | 1039 | raw_spin_lock_irqsave(&worker->lock, flags); |
22597dc3 | 1040 | |
37be45d4 | 1041 | if (!queuing_blocked(worker, work)) { |
22597dc3 PM |
1042 | __kthread_queue_delayed_work(worker, dwork, delay); |
1043 | ret = true; | |
1044 | } | |
1045 | ||
fe99a4f4 | 1046 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
22597dc3 PM |
1047 | return ret; |
1048 | } | |
1049 | EXPORT_SYMBOL_GPL(kthread_queue_delayed_work); | |
1050 | ||
9a2e03d8 TH |
1051 | struct kthread_flush_work { |
1052 | struct kthread_work work; | |
1053 | struct completion done; | |
1054 | }; | |
1055 | ||
1056 | static void kthread_flush_work_fn(struct kthread_work *work) | |
1057 | { | |
1058 | struct kthread_flush_work *fwork = | |
1059 | container_of(work, struct kthread_flush_work, work); | |
1060 | complete(&fwork->done); | |
1061 | } | |
1062 | ||
b56c0d89 | 1063 | /** |
3989144f | 1064 | * kthread_flush_work - flush a kthread_work |
b56c0d89 TH |
1065 | * @work: work to flush |
1066 | * | |
1067 | * If @work is queued or executing, wait for it to finish execution. | |
1068 | */ | |
3989144f | 1069 | void kthread_flush_work(struct kthread_work *work) |
b56c0d89 | 1070 | { |
46f3d976 TH |
1071 | struct kthread_flush_work fwork = { |
1072 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
1073 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
1074 | }; | |
1075 | struct kthread_worker *worker; | |
1076 | bool noop = false; | |
1077 | ||
46f3d976 TH |
1078 | worker = work->worker; |
1079 | if (!worker) | |
1080 | return; | |
b56c0d89 | 1081 | |
fe99a4f4 | 1082 | raw_spin_lock_irq(&worker->lock); |
8197b3d4 PM |
1083 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
1084 | WARN_ON_ONCE(work->worker != worker); | |
b56c0d89 | 1085 | |
46f3d976 | 1086 | if (!list_empty(&work->node)) |
3989144f | 1087 | kthread_insert_work(worker, &fwork.work, work->node.next); |
46f3d976 | 1088 | else if (worker->current_work == work) |
3989144f PM |
1089 | kthread_insert_work(worker, &fwork.work, |
1090 | worker->work_list.next); | |
46f3d976 TH |
1091 | else |
1092 | noop = true; | |
b56c0d89 | 1093 | |
fe99a4f4 | 1094 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 | 1095 | |
46f3d976 TH |
1096 | if (!noop) |
1097 | wait_for_completion(&fwork.done); | |
b56c0d89 | 1098 | } |
3989144f | 1099 | EXPORT_SYMBOL_GPL(kthread_flush_work); |
b56c0d89 | 1100 | |
34b3d534 PM |
1101 | /* |
1102 | * Make sure that the timer is neither set nor running and could | |
1103 | * not manipulate the work list_head any longer. | |
1104 | * | |
1105 | * The function is called under worker->lock. The lock is temporary | |
1106 | * released but the timer can't be set again in the meantime. | |
1107 | */ | |
1108 | static void kthread_cancel_delayed_work_timer(struct kthread_work *work, | |
1109 | unsigned long *flags) | |
1110 | { | |
1111 | struct kthread_delayed_work *dwork = | |
1112 | container_of(work, struct kthread_delayed_work, work); | |
1113 | struct kthread_worker *worker = work->worker; | |
1114 | ||
1115 | /* | |
1116 | * del_timer_sync() must be called to make sure that the timer | |
1117 | * callback is not running. The lock must be temporary released | |
1118 | * to avoid a deadlock with the callback. In the meantime, | |
1119 | * any queuing is blocked by setting the canceling counter. | |
1120 | */ | |
1121 | work->canceling++; | |
1122 | raw_spin_unlock_irqrestore(&worker->lock, *flags); | |
1123 | del_timer_sync(&dwork->timer); | |
1124 | raw_spin_lock_irqsave(&worker->lock, *flags); | |
1125 | work->canceling--; | |
1126 | } | |
1127 | ||
37be45d4 | 1128 | /* |
5fa54346 PM |
1129 | * This function removes the work from the worker queue. |
1130 | * | |
1131 | * It is called under worker->lock. The caller must make sure that | |
1132 | * the timer used by delayed work is not running, e.g. by calling | |
1133 | * kthread_cancel_delayed_work_timer(). | |
37be45d4 PM |
1134 | * |
1135 | * The work might still be in use when this function finishes. See the | |
1136 | * current_work proceed by the worker. | |
1137 | * | |
1138 | * Return: %true if @work was pending and successfully canceled, | |
1139 | * %false if @work was not pending | |
1140 | */ | |
5fa54346 | 1141 | static bool __kthread_cancel_work(struct kthread_work *work) |
37be45d4 | 1142 | { |
37be45d4 PM |
1143 | /* |
1144 | * Try to remove the work from a worker list. It might either | |
1145 | * be from worker->work_list or from worker->delayed_work_list. | |
1146 | */ | |
1147 | if (!list_empty(&work->node)) { | |
1148 | list_del_init(&work->node); | |
1149 | return true; | |
1150 | } | |
1151 | ||
1152 | return false; | |
1153 | } | |
1154 | ||
9a6b06c8 PM |
1155 | /** |
1156 | * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work | |
1157 | * @worker: kthread worker to use | |
1158 | * @dwork: kthread delayed work to queue | |
1159 | * @delay: number of jiffies to wait before queuing | |
1160 | * | |
1161 | * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise, | |
1162 | * modify @dwork's timer so that it expires after @delay. If @delay is zero, | |
1163 | * @work is guaranteed to be queued immediately. | |
1164 | * | |
d71ba164 | 1165 | * Return: %false if @dwork was idle and queued, %true otherwise. |
9a6b06c8 PM |
1166 | * |
1167 | * A special case is when the work is being canceled in parallel. | |
1168 | * It might be caused either by the real kthread_cancel_delayed_work_sync() | |
1169 | * or yet another kthread_mod_delayed_work() call. We let the other command | |
d71ba164 PM |
1170 | * win and return %true here. The return value can be used for reference |
1171 | * counting and the number of queued works stays the same. Anyway, the caller | |
1172 | * is supposed to synchronize these operations a reasonable way. | |
9a6b06c8 PM |
1173 | * |
1174 | * This function is safe to call from any context including IRQ handler. | |
1175 | * See __kthread_cancel_work() and kthread_delayed_work_timer_fn() | |
1176 | * for details. | |
1177 | */ | |
1178 | bool kthread_mod_delayed_work(struct kthread_worker *worker, | |
1179 | struct kthread_delayed_work *dwork, | |
1180 | unsigned long delay) | |
1181 | { | |
1182 | struct kthread_work *work = &dwork->work; | |
1183 | unsigned long flags; | |
d71ba164 | 1184 | int ret; |
9a6b06c8 | 1185 | |
fe99a4f4 | 1186 | raw_spin_lock_irqsave(&worker->lock, flags); |
9a6b06c8 PM |
1187 | |
1188 | /* Do not bother with canceling when never queued. */ | |
d71ba164 PM |
1189 | if (!work->worker) { |
1190 | ret = false; | |
9a6b06c8 | 1191 | goto fast_queue; |
d71ba164 | 1192 | } |
9a6b06c8 PM |
1193 | |
1194 | /* Work must not be used with >1 worker, see kthread_queue_work() */ | |
1195 | WARN_ON_ONCE(work->worker != worker); | |
1196 | ||
5fa54346 PM |
1197 | /* |
1198 | * Temporary cancel the work but do not fight with another command | |
1199 | * that is canceling the work as well. | |
1200 | * | |
1201 | * It is a bit tricky because of possible races with another | |
1202 | * mod_delayed_work() and cancel_delayed_work() callers. | |
1203 | * | |
1204 | * The timer must be canceled first because worker->lock is released | |
1205 | * when doing so. But the work can be removed from the queue (list) | |
1206 | * only when it can be queued again so that the return value can | |
1207 | * be used for reference counting. | |
1208 | */ | |
1209 | kthread_cancel_delayed_work_timer(work, &flags); | |
d71ba164 PM |
1210 | if (work->canceling) { |
1211 | /* The number of works in the queue does not change. */ | |
1212 | ret = true; | |
9a6b06c8 | 1213 | goto out; |
d71ba164 | 1214 | } |
5fa54346 | 1215 | ret = __kthread_cancel_work(work); |
9a6b06c8 | 1216 | |
9a6b06c8 PM |
1217 | fast_queue: |
1218 | __kthread_queue_delayed_work(worker, dwork, delay); | |
1219 | out: | |
fe99a4f4 | 1220 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
9a6b06c8 PM |
1221 | return ret; |
1222 | } | |
1223 | EXPORT_SYMBOL_GPL(kthread_mod_delayed_work); | |
1224 | ||
37be45d4 PM |
1225 | static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork) |
1226 | { | |
1227 | struct kthread_worker *worker = work->worker; | |
1228 | unsigned long flags; | |
1229 | int ret = false; | |
1230 | ||
1231 | if (!worker) | |
1232 | goto out; | |
1233 | ||
fe99a4f4 | 1234 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 PM |
1235 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
1236 | WARN_ON_ONCE(work->worker != worker); | |
1237 | ||
5fa54346 PM |
1238 | if (is_dwork) |
1239 | kthread_cancel_delayed_work_timer(work, &flags); | |
1240 | ||
1241 | ret = __kthread_cancel_work(work); | |
37be45d4 PM |
1242 | |
1243 | if (worker->current_work != work) | |
1244 | goto out_fast; | |
1245 | ||
1246 | /* | |
1247 | * The work is in progress and we need to wait with the lock released. | |
1248 | * In the meantime, block any queuing by setting the canceling counter. | |
1249 | */ | |
1250 | work->canceling++; | |
fe99a4f4 | 1251 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
37be45d4 | 1252 | kthread_flush_work(work); |
fe99a4f4 | 1253 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 PM |
1254 | work->canceling--; |
1255 | ||
1256 | out_fast: | |
fe99a4f4 | 1257 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
37be45d4 PM |
1258 | out: |
1259 | return ret; | |
1260 | } | |
1261 | ||
1262 | /** | |
1263 | * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish | |
1264 | * @work: the kthread work to cancel | |
1265 | * | |
1266 | * Cancel @work and wait for its execution to finish. This function | |
1267 | * can be used even if the work re-queues itself. On return from this | |
1268 | * function, @work is guaranteed to be not pending or executing on any CPU. | |
1269 | * | |
1270 | * kthread_cancel_work_sync(&delayed_work->work) must not be used for | |
1271 | * delayed_work's. Use kthread_cancel_delayed_work_sync() instead. | |
1272 | * | |
1273 | * The caller must ensure that the worker on which @work was last | |
1274 | * queued can't be destroyed before this function returns. | |
1275 | * | |
1276 | * Return: %true if @work was pending, %false otherwise. | |
1277 | */ | |
1278 | bool kthread_cancel_work_sync(struct kthread_work *work) | |
1279 | { | |
1280 | return __kthread_cancel_work_sync(work, false); | |
1281 | } | |
1282 | EXPORT_SYMBOL_GPL(kthread_cancel_work_sync); | |
1283 | ||
1284 | /** | |
1285 | * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and | |
1286 | * wait for it to finish. | |
1287 | * @dwork: the kthread delayed work to cancel | |
1288 | * | |
1289 | * This is kthread_cancel_work_sync() for delayed works. | |
1290 | * | |
1291 | * Return: %true if @dwork was pending, %false otherwise. | |
1292 | */ | |
1293 | bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork) | |
1294 | { | |
1295 | return __kthread_cancel_work_sync(&dwork->work, true); | |
1296 | } | |
1297 | EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync); | |
1298 | ||
b56c0d89 | 1299 | /** |
3989144f | 1300 | * kthread_flush_worker - flush all current works on a kthread_worker |
b56c0d89 TH |
1301 | * @worker: worker to flush |
1302 | * | |
1303 | * Wait until all currently executing or pending works on @worker are | |
1304 | * finished. | |
1305 | */ | |
3989144f | 1306 | void kthread_flush_worker(struct kthread_worker *worker) |
b56c0d89 TH |
1307 | { |
1308 | struct kthread_flush_work fwork = { | |
1309 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
1310 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
1311 | }; | |
1312 | ||
3989144f | 1313 | kthread_queue_work(worker, &fwork.work); |
b56c0d89 TH |
1314 | wait_for_completion(&fwork.done); |
1315 | } | |
3989144f | 1316 | EXPORT_SYMBOL_GPL(kthread_flush_worker); |
35033fe9 PM |
1317 | |
1318 | /** | |
1319 | * kthread_destroy_worker - destroy a kthread worker | |
1320 | * @worker: worker to be destroyed | |
1321 | * | |
1322 | * Flush and destroy @worker. The simple flush is enough because the kthread | |
1323 | * worker API is used only in trivial scenarios. There are no multi-step state | |
1324 | * machines needed. | |
1325 | */ | |
1326 | void kthread_destroy_worker(struct kthread_worker *worker) | |
1327 | { | |
1328 | struct task_struct *task; | |
1329 | ||
1330 | task = worker->task; | |
1331 | if (WARN_ON(!task)) | |
1332 | return; | |
1333 | ||
1334 | kthread_flush_worker(worker); | |
1335 | kthread_stop(task); | |
1336 | WARN_ON(!list_empty(&worker->work_list)); | |
1337 | kfree(worker); | |
1338 | } | |
1339 | EXPORT_SYMBOL(kthread_destroy_worker); | |
05e3db95 | 1340 | |
f5678e7f CH |
1341 | /** |
1342 | * kthread_use_mm - make the calling kthread operate on an address space | |
1343 | * @mm: address space to operate on | |
9bf5b9eb | 1344 | */ |
f5678e7f | 1345 | void kthread_use_mm(struct mm_struct *mm) |
9bf5b9eb CH |
1346 | { |
1347 | struct mm_struct *active_mm; | |
1348 | struct task_struct *tsk = current; | |
1349 | ||
f5678e7f CH |
1350 | WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD)); |
1351 | WARN_ON_ONCE(tsk->mm); | |
1352 | ||
9bf5b9eb | 1353 | task_lock(tsk); |
38cf307c PZ |
1354 | /* Hold off tlb flush IPIs while switching mm's */ |
1355 | local_irq_disable(); | |
9bf5b9eb CH |
1356 | active_mm = tsk->active_mm; |
1357 | if (active_mm != mm) { | |
1358 | mmgrab(mm); | |
1359 | tsk->active_mm = mm; | |
1360 | } | |
1361 | tsk->mm = mm; | |
618758ed | 1362 | membarrier_update_current_mm(mm); |
38cf307c PZ |
1363 | switch_mm_irqs_off(active_mm, mm, tsk); |
1364 | local_irq_enable(); | |
9bf5b9eb CH |
1365 | task_unlock(tsk); |
1366 | #ifdef finish_arch_post_lock_switch | |
1367 | finish_arch_post_lock_switch(); | |
1368 | #endif | |
1369 | ||
618758ed MD |
1370 | /* |
1371 | * When a kthread starts operating on an address space, the loop | |
1372 | * in membarrier_{private,global}_expedited() may not observe | |
1373 | * that tsk->mm, and not issue an IPI. Membarrier requires a | |
1374 | * memory barrier after storing to tsk->mm, before accessing | |
1375 | * user-space memory. A full memory barrier for membarrier | |
1376 | * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by | |
1377 | * mmdrop(), or explicitly with smp_mb(). | |
1378 | */ | |
9bf5b9eb CH |
1379 | if (active_mm != mm) |
1380 | mmdrop(active_mm); | |
618758ed MD |
1381 | else |
1382 | smp_mb(); | |
37c54f9b | 1383 | |
3d13f313 | 1384 | to_kthread(tsk)->oldfs = force_uaccess_begin(); |
9bf5b9eb | 1385 | } |
f5678e7f | 1386 | EXPORT_SYMBOL_GPL(kthread_use_mm); |
9bf5b9eb | 1387 | |
f5678e7f CH |
1388 | /** |
1389 | * kthread_unuse_mm - reverse the effect of kthread_use_mm() | |
1390 | * @mm: address space to operate on | |
9bf5b9eb | 1391 | */ |
f5678e7f | 1392 | void kthread_unuse_mm(struct mm_struct *mm) |
9bf5b9eb CH |
1393 | { |
1394 | struct task_struct *tsk = current; | |
1395 | ||
f5678e7f CH |
1396 | WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD)); |
1397 | WARN_ON_ONCE(!tsk->mm); | |
1398 | ||
3d13f313 | 1399 | force_uaccess_end(to_kthread(tsk)->oldfs); |
37c54f9b | 1400 | |
9bf5b9eb | 1401 | task_lock(tsk); |
618758ed MD |
1402 | /* |
1403 | * When a kthread stops operating on an address space, the loop | |
1404 | * in membarrier_{private,global}_expedited() may not observe | |
1405 | * that tsk->mm, and not issue an IPI. Membarrier requires a | |
1406 | * memory barrier after accessing user-space memory, before | |
1407 | * clearing tsk->mm. | |
1408 | */ | |
1409 | smp_mb__after_spinlock(); | |
9bf5b9eb | 1410 | sync_mm_rss(mm); |
38cf307c | 1411 | local_irq_disable(); |
9bf5b9eb | 1412 | tsk->mm = NULL; |
618758ed | 1413 | membarrier_update_current_mm(NULL); |
9bf5b9eb CH |
1414 | /* active_mm is still 'mm' */ |
1415 | enter_lazy_tlb(mm, tsk); | |
38cf307c | 1416 | local_irq_enable(); |
9bf5b9eb CH |
1417 | task_unlock(tsk); |
1418 | } | |
f5678e7f | 1419 | EXPORT_SYMBOL_GPL(kthread_unuse_mm); |
9bf5b9eb | 1420 | |
0b508bc9 | 1421 | #ifdef CONFIG_BLK_CGROUP |
05e3db95 SL |
1422 | /** |
1423 | * kthread_associate_blkcg - associate blkcg to current kthread | |
1424 | * @css: the cgroup info | |
1425 | * | |
1426 | * Current thread must be a kthread. The thread is running jobs on behalf of | |
1427 | * other threads. In some cases, we expect the jobs attach cgroup info of | |
1428 | * original threads instead of that of current thread. This function stores | |
1429 | * original thread's cgroup info in current kthread context for later | |
1430 | * retrieval. | |
1431 | */ | |
1432 | void kthread_associate_blkcg(struct cgroup_subsys_state *css) | |
1433 | { | |
1434 | struct kthread *kthread; | |
1435 | ||
1436 | if (!(current->flags & PF_KTHREAD)) | |
1437 | return; | |
1438 | kthread = to_kthread(current); | |
1439 | if (!kthread) | |
1440 | return; | |
1441 | ||
1442 | if (kthread->blkcg_css) { | |
1443 | css_put(kthread->blkcg_css); | |
1444 | kthread->blkcg_css = NULL; | |
1445 | } | |
1446 | if (css) { | |
1447 | css_get(css); | |
1448 | kthread->blkcg_css = css; | |
1449 | } | |
1450 | } | |
1451 | EXPORT_SYMBOL(kthread_associate_blkcg); | |
1452 | ||
1453 | /** | |
1454 | * kthread_blkcg - get associated blkcg css of current kthread | |
1455 | * | |
1456 | * Current thread must be a kthread. | |
1457 | */ | |
1458 | struct cgroup_subsys_state *kthread_blkcg(void) | |
1459 | { | |
1460 | struct kthread *kthread; | |
1461 | ||
1462 | if (current->flags & PF_KTHREAD) { | |
1463 | kthread = to_kthread(current); | |
1464 | if (kthread) | |
1465 | return kthread->blkcg_css; | |
1466 | } | |
1467 | return NULL; | |
1468 | } | |
1469 | EXPORT_SYMBOL(kthread_blkcg); | |
1470 | #endif |