]>
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))) { | |
2bad71c3 TH |
385 | int i = 0; |
386 | ||
387 | /* | |
388 | * I got SIGKILL, but wait for 10 more seconds for completion | |
389 | * unless chosen by the OOM killer. This delay is there as a | |
390 | * workaround for boot failure caused by SIGKILL upon device | |
391 | * driver initialization timeout. | |
392 | */ | |
393 | while (i++ < 10 && !test_tsk_thread_flag(current, TIF_MEMDIE)) | |
394 | if (wait_for_completion_timeout(&done, HZ)) | |
395 | goto ready; | |
786235ee TH |
396 | /* |
397 | * If I was SIGKILLed before kthreadd (or new kernel thread) | |
398 | * calls complete(), leave the cleanup of this structure to | |
399 | * that thread. | |
400 | */ | |
401 | if (xchg(&create->done, NULL)) | |
8fe6929c | 402 | return ERR_PTR(-EINTR); |
786235ee TH |
403 | /* |
404 | * kthreadd (or new kernel thread) will call complete() | |
405 | * shortly. | |
406 | */ | |
407 | wait_for_completion(&done); | |
408 | } | |
2bad71c3 | 409 | ready: |
786235ee TH |
410 | task = create->result; |
411 | if (!IS_ERR(task)) { | |
c9b5f501 | 412 | static const struct sched_param param = { .sched_priority = 0 }; |
3e536e22 | 413 | char name[TASK_COMM_LEN]; |
1c99315b | 414 | |
3e536e22 SD |
415 | /* |
416 | * task is already visible to other tasks, so updating | |
417 | * COMM must be protected. | |
418 | */ | |
419 | vsnprintf(name, sizeof(name), namefmt, args); | |
420 | set_task_comm(task, name); | |
1c99315b ON |
421 | /* |
422 | * root may have changed our (kthreadd's) priority or CPU mask. | |
423 | * The kernel thread should not inherit these properties. | |
424 | */ | |
786235ee | 425 | sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m); |
9cc5b865 MT |
426 | set_cpus_allowed_ptr(task, |
427 | housekeeping_cpumask(HK_FLAG_KTHREAD)); | |
1da177e4 | 428 | } |
786235ee TH |
429 | kfree(create); |
430 | return task; | |
1da177e4 | 431 | } |
255451e4 PM |
432 | |
433 | /** | |
434 | * kthread_create_on_node - create a kthread. | |
435 | * @threadfn: the function to run until signal_pending(current). | |
436 | * @data: data ptr for @threadfn. | |
437 | * @node: task and thread structures for the thread are allocated on this node | |
438 | * @namefmt: printf-style name for the thread. | |
439 | * | |
440 | * Description: This helper function creates and names a kernel | |
441 | * thread. The thread will be stopped: use wake_up_process() to start | |
442 | * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and | |
443 | * is affine to all CPUs. | |
444 | * | |
445 | * If thread is going to be bound on a particular cpu, give its node | |
446 | * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE. | |
447 | * When woken, the thread will run @threadfn() with @data as its | |
448 | * argument. @threadfn() can either call do_exit() directly if it is a | |
449 | * standalone thread for which no one will call kthread_stop(), or | |
450 | * return when 'kthread_should_stop()' is true (which means | |
451 | * kthread_stop() has been called). The return value should be zero | |
452 | * or a negative error number; it will be passed to kthread_stop(). | |
453 | * | |
454 | * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR). | |
455 | */ | |
456 | struct task_struct *kthread_create_on_node(int (*threadfn)(void *data), | |
457 | void *data, int node, | |
458 | const char namefmt[], | |
459 | ...) | |
460 | { | |
461 | struct task_struct *task; | |
462 | va_list args; | |
463 | ||
464 | va_start(args, namefmt); | |
465 | task = __kthread_create_on_node(threadfn, data, node, namefmt, args); | |
466 | va_end(args); | |
467 | ||
468 | return task; | |
469 | } | |
207205a2 | 470 | EXPORT_SYMBOL(kthread_create_on_node); |
1da177e4 | 471 | |
2f064a59 | 472 | static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state) |
2a1d4460 | 473 | { |
25834c73 PZ |
474 | unsigned long flags; |
475 | ||
f2530dc7 TG |
476 | if (!wait_task_inactive(p, state)) { |
477 | WARN_ON(1); | |
478 | return; | |
479 | } | |
25834c73 | 480 | |
2a1d4460 | 481 | /* It's safe because the task is inactive. */ |
25834c73 PZ |
482 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
483 | do_set_cpus_allowed(p, mask); | |
14a40ffc | 484 | p->flags |= PF_NO_SETAFFINITY; |
25834c73 PZ |
485 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
486 | } | |
487 | ||
2f064a59 | 488 | static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state) |
25834c73 PZ |
489 | { |
490 | __kthread_bind_mask(p, cpumask_of(cpu), state); | |
491 | } | |
492 | ||
493 | void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask) | |
494 | { | |
495 | __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE); | |
2a1d4460 TG |
496 | } |
497 | ||
881232b7 PZ |
498 | /** |
499 | * kthread_bind - bind a just-created kthread to a cpu. | |
500 | * @p: thread created by kthread_create(). | |
501 | * @cpu: cpu (might not be online, must be possible) for @k to run on. | |
502 | * | |
503 | * Description: This function is equivalent to set_cpus_allowed(), | |
504 | * except that @cpu doesn't need to be online, and the thread must be | |
505 | * stopped (i.e., just returned from kthread_create()). | |
506 | */ | |
507 | void kthread_bind(struct task_struct *p, unsigned int cpu) | |
508 | { | |
f2530dc7 | 509 | __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE); |
881232b7 PZ |
510 | } |
511 | EXPORT_SYMBOL(kthread_bind); | |
512 | ||
2a1d4460 TG |
513 | /** |
514 | * kthread_create_on_cpu - Create a cpu bound kthread | |
515 | * @threadfn: the function to run until signal_pending(current). | |
516 | * @data: data ptr for @threadfn. | |
517 | * @cpu: The cpu on which the thread should be bound, | |
518 | * @namefmt: printf-style name for the thread. Format is restricted | |
519 | * to "name.*%u". Code fills in cpu number. | |
520 | * | |
521 | * Description: This helper function creates and names a kernel thread | |
2a1d4460 TG |
522 | */ |
523 | struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data), | |
524 | void *data, unsigned int cpu, | |
525 | const char *namefmt) | |
526 | { | |
527 | struct task_struct *p; | |
528 | ||
10922838 | 529 | p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt, |
2a1d4460 TG |
530 | cpu); |
531 | if (IS_ERR(p)) | |
532 | return p; | |
a65d4096 PM |
533 | kthread_bind(p, cpu); |
534 | /* CPU hotplug need to bind once again when unparking the thread. */ | |
2a1d4460 | 535 | to_kthread(p)->cpu = cpu; |
2a1d4460 TG |
536 | return p; |
537 | } | |
538 | ||
ac687e6e PZ |
539 | void kthread_set_per_cpu(struct task_struct *k, int cpu) |
540 | { | |
541 | struct kthread *kthread = to_kthread(k); | |
542 | if (!kthread) | |
543 | return; | |
544 | ||
545 | WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY)); | |
546 | ||
547 | if (cpu < 0) { | |
548 | clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
549 | return; | |
550 | } | |
551 | ||
552 | kthread->cpu = cpu; | |
553 | set_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
554 | } | |
555 | ||
3a7956e2 | 556 | bool kthread_is_per_cpu(struct task_struct *p) |
ac687e6e | 557 | { |
3a7956e2 | 558 | struct kthread *kthread = __to_kthread(p); |
ac687e6e PZ |
559 | if (!kthread) |
560 | return false; | |
561 | ||
562 | return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags); | |
563 | } | |
564 | ||
cf380a4a ON |
565 | /** |
566 | * kthread_unpark - unpark a thread created by kthread_create(). | |
567 | * @k: thread created by kthread_create(). | |
568 | * | |
569 | * Sets kthread_should_park() for @k to return false, wakes it, and | |
570 | * waits for it to return. If the thread is marked percpu then its | |
571 | * bound to the cpu again. | |
572 | */ | |
573 | void kthread_unpark(struct task_struct *k) | |
f2530dc7 | 574 | { |
cf380a4a ON |
575 | struct kthread *kthread = to_kthread(k); |
576 | ||
f2530dc7 | 577 | /* |
85f1abe0 PZ |
578 | * Newly created kthread was parked when the CPU was offline. |
579 | * The binding was lost and we need to set it again. | |
f2530dc7 | 580 | */ |
85f1abe0 PZ |
581 | if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags)) |
582 | __kthread_bind(k, kthread->cpu, TASK_PARKED); | |
583 | ||
584 | clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags); | |
1cef1150 PZ |
585 | /* |
586 | * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup. | |
587 | */ | |
85f1abe0 | 588 | wake_up_state(k, TASK_PARKED); |
f2530dc7 | 589 | } |
18896451 | 590 | EXPORT_SYMBOL_GPL(kthread_unpark); |
2a1d4460 TG |
591 | |
592 | /** | |
593 | * kthread_park - park a thread created by kthread_create(). | |
594 | * @k: thread created by kthread_create(). | |
595 | * | |
596 | * Sets kthread_should_park() for @k to return true, wakes it, and | |
597 | * waits for it to return. This can also be called after kthread_create() | |
598 | * instead of calling wake_up_process(): the thread will park without | |
599 | * calling threadfn(). | |
600 | * | |
601 | * Returns 0 if the thread is parked, -ENOSYS if the thread exited. | |
602 | * If called by the kthread itself just the park bit is set. | |
603 | */ | |
604 | int kthread_park(struct task_struct *k) | |
605 | { | |
cf380a4a ON |
606 | struct kthread *kthread = to_kthread(k); |
607 | ||
608 | if (WARN_ON(k->flags & PF_EXITING)) | |
609 | return -ENOSYS; | |
610 | ||
f83ee19b PZ |
611 | if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags))) |
612 | return -EBUSY; | |
613 | ||
85f1abe0 PZ |
614 | set_bit(KTHREAD_SHOULD_PARK, &kthread->flags); |
615 | if (k != current) { | |
616 | wake_up_process(k); | |
1cef1150 PZ |
617 | /* |
618 | * Wait for __kthread_parkme() to complete(), this means we | |
619 | * _will_ have TASK_PARKED and are about to call schedule(). | |
620 | */ | |
85f1abe0 | 621 | wait_for_completion(&kthread->parked); |
1cef1150 PZ |
622 | /* |
623 | * Now wait for that schedule() to complete and the task to | |
624 | * get scheduled out. | |
625 | */ | |
626 | WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED)); | |
2a1d4460 | 627 | } |
cf380a4a ON |
628 | |
629 | return 0; | |
2a1d4460 | 630 | } |
18896451 | 631 | EXPORT_SYMBOL_GPL(kthread_park); |
2a1d4460 | 632 | |
9e37bd30 RD |
633 | /** |
634 | * kthread_stop - stop a thread created by kthread_create(). | |
635 | * @k: thread created by kthread_create(). | |
636 | * | |
637 | * Sets kthread_should_stop() for @k to return true, wakes it, and | |
9ae26027 ON |
638 | * waits for it to exit. This can also be called after kthread_create() |
639 | * instead of calling wake_up_process(): the thread will exit without | |
640 | * calling threadfn(). | |
641 | * | |
642 | * If threadfn() may call do_exit() itself, the caller must ensure | |
643 | * task_struct can't go away. | |
9e37bd30 RD |
644 | * |
645 | * Returns the result of threadfn(), or %-EINTR if wake_up_process() | |
646 | * was never called. | |
647 | */ | |
1da177e4 LT |
648 | int kthread_stop(struct task_struct *k) |
649 | { | |
b5c5442b | 650 | struct kthread *kthread; |
1da177e4 LT |
651 | int ret; |
652 | ||
0a16b607 | 653 | trace_sched_kthread_stop(k); |
b5c5442b ON |
654 | |
655 | get_task_struct(k); | |
efb29fbf ON |
656 | kthread = to_kthread(k); |
657 | set_bit(KTHREAD_SHOULD_STOP, &kthread->flags); | |
cf380a4a | 658 | kthread_unpark(k); |
efb29fbf ON |
659 | wake_up_process(k); |
660 | wait_for_completion(&kthread->exited); | |
63706172 | 661 | ret = k->exit_code; |
1da177e4 | 662 | put_task_struct(k); |
0a16b607 | 663 | |
b5c5442b | 664 | trace_sched_kthread_stop_ret(ret); |
1da177e4 LT |
665 | return ret; |
666 | } | |
52e92e57 | 667 | EXPORT_SYMBOL(kthread_stop); |
1da177e4 | 668 | |
e804a4a4 | 669 | int kthreadd(void *unused) |
1da177e4 | 670 | { |
73c27992 | 671 | struct task_struct *tsk = current; |
1da177e4 | 672 | |
e804a4a4 | 673 | /* Setup a clean context for our children to inherit. */ |
73c27992 | 674 | set_task_comm(tsk, "kthreadd"); |
10ab825b | 675 | ignore_signals(tsk); |
9cc5b865 | 676 | set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD)); |
aee4faa4 | 677 | set_mems_allowed(node_states[N_MEMORY]); |
73c27992 | 678 | |
34b087e4 | 679 | current->flags |= PF_NOFREEZE; |
77f88796 | 680 | cgroup_init_kthreadd(); |
73c27992 EB |
681 | |
682 | for (;;) { | |
683 | set_current_state(TASK_INTERRUPTIBLE); | |
684 | if (list_empty(&kthread_create_list)) | |
685 | schedule(); | |
686 | __set_current_state(TASK_RUNNING); | |
687 | ||
688 | spin_lock(&kthread_create_lock); | |
689 | while (!list_empty(&kthread_create_list)) { | |
690 | struct kthread_create_info *create; | |
691 | ||
692 | create = list_entry(kthread_create_list.next, | |
693 | struct kthread_create_info, list); | |
694 | list_del_init(&create->list); | |
695 | spin_unlock(&kthread_create_lock); | |
696 | ||
697 | create_kthread(create); | |
698 | ||
699 | spin_lock(&kthread_create_lock); | |
700 | } | |
701 | spin_unlock(&kthread_create_lock); | |
702 | } | |
703 | ||
704 | return 0; | |
705 | } | |
b56c0d89 | 706 | |
3989144f | 707 | void __kthread_init_worker(struct kthread_worker *worker, |
4f32e9b1 YZ |
708 | const char *name, |
709 | struct lock_class_key *key) | |
710 | { | |
dbf52682 | 711 | memset(worker, 0, sizeof(struct kthread_worker)); |
fe99a4f4 | 712 | raw_spin_lock_init(&worker->lock); |
4f32e9b1 YZ |
713 | lockdep_set_class_and_name(&worker->lock, key, name); |
714 | INIT_LIST_HEAD(&worker->work_list); | |
22597dc3 | 715 | INIT_LIST_HEAD(&worker->delayed_work_list); |
4f32e9b1 | 716 | } |
3989144f | 717 | EXPORT_SYMBOL_GPL(__kthread_init_worker); |
4f32e9b1 | 718 | |
b56c0d89 TH |
719 | /** |
720 | * kthread_worker_fn - kthread function to process kthread_worker | |
721 | * @worker_ptr: pointer to initialized kthread_worker | |
722 | * | |
fbae2d44 PM |
723 | * This function implements the main cycle of kthread worker. It processes |
724 | * work_list until it is stopped with kthread_stop(). It sleeps when the queue | |
725 | * is empty. | |
b56c0d89 | 726 | * |
fbae2d44 PM |
727 | * The works are not allowed to keep any locks, disable preemption or interrupts |
728 | * when they finish. There is defined a safe point for freezing when one work | |
729 | * finishes and before a new one is started. | |
8197b3d4 PM |
730 | * |
731 | * Also the works must not be handled by more than one worker at the same time, | |
732 | * see also kthread_queue_work(). | |
b56c0d89 TH |
733 | */ |
734 | int kthread_worker_fn(void *worker_ptr) | |
735 | { | |
736 | struct kthread_worker *worker = worker_ptr; | |
737 | struct kthread_work *work; | |
738 | ||
fbae2d44 PM |
739 | /* |
740 | * FIXME: Update the check and remove the assignment when all kthread | |
741 | * worker users are created using kthread_create_worker*() functions. | |
742 | */ | |
743 | WARN_ON(worker->task && worker->task != current); | |
b56c0d89 | 744 | worker->task = current; |
dbf52682 PM |
745 | |
746 | if (worker->flags & KTW_FREEZABLE) | |
747 | set_freezable(); | |
748 | ||
b56c0d89 TH |
749 | repeat: |
750 | set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */ | |
751 | ||
752 | if (kthread_should_stop()) { | |
753 | __set_current_state(TASK_RUNNING); | |
fe99a4f4 | 754 | raw_spin_lock_irq(&worker->lock); |
b56c0d89 | 755 | worker->task = NULL; |
fe99a4f4 | 756 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 TH |
757 | return 0; |
758 | } | |
759 | ||
760 | work = NULL; | |
fe99a4f4 | 761 | raw_spin_lock_irq(&worker->lock); |
b56c0d89 TH |
762 | if (!list_empty(&worker->work_list)) { |
763 | work = list_first_entry(&worker->work_list, | |
764 | struct kthread_work, node); | |
765 | list_del_init(&work->node); | |
766 | } | |
46f3d976 | 767 | worker->current_work = work; |
fe99a4f4 | 768 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 TH |
769 | |
770 | if (work) { | |
f630c7c6 | 771 | kthread_work_func_t func = work->func; |
b56c0d89 | 772 | __set_current_state(TASK_RUNNING); |
f630c7c6 | 773 | trace_sched_kthread_work_execute_start(work); |
b56c0d89 | 774 | work->func(work); |
f630c7c6 RC |
775 | /* |
776 | * Avoid dereferencing work after this point. The trace | |
777 | * event only cares about the address. | |
778 | */ | |
779 | trace_sched_kthread_work_execute_end(work, func); | |
b56c0d89 TH |
780 | } else if (!freezing(current)) |
781 | schedule(); | |
782 | ||
783 | try_to_freeze(); | |
22cf8bc6 | 784 | cond_resched(); |
b56c0d89 TH |
785 | goto repeat; |
786 | } | |
787 | EXPORT_SYMBOL_GPL(kthread_worker_fn); | |
788 | ||
c0b942a7 | 789 | static __printf(3, 0) struct kthread_worker * |
dbf52682 PM |
790 | __kthread_create_worker(int cpu, unsigned int flags, |
791 | const char namefmt[], va_list args) | |
fbae2d44 PM |
792 | { |
793 | struct kthread_worker *worker; | |
794 | struct task_struct *task; | |
98fa15f3 | 795 | int node = NUMA_NO_NODE; |
fbae2d44 PM |
796 | |
797 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
798 | if (!worker) | |
799 | return ERR_PTR(-ENOMEM); | |
800 | ||
801 | kthread_init_worker(worker); | |
802 | ||
8fb9dcbd ON |
803 | if (cpu >= 0) |
804 | node = cpu_to_node(cpu); | |
fbae2d44 | 805 | |
8fb9dcbd ON |
806 | task = __kthread_create_on_node(kthread_worker_fn, worker, |
807 | node, namefmt, args); | |
fbae2d44 PM |
808 | if (IS_ERR(task)) |
809 | goto fail_task; | |
810 | ||
8fb9dcbd ON |
811 | if (cpu >= 0) |
812 | kthread_bind(task, cpu); | |
813 | ||
dbf52682 | 814 | worker->flags = flags; |
fbae2d44 PM |
815 | worker->task = task; |
816 | wake_up_process(task); | |
817 | return worker; | |
818 | ||
819 | fail_task: | |
820 | kfree(worker); | |
821 | return ERR_CAST(task); | |
822 | } | |
823 | ||
824 | /** | |
825 | * kthread_create_worker - create a kthread worker | |
dbf52682 | 826 | * @flags: flags modifying the default behavior of the worker |
fbae2d44 PM |
827 | * @namefmt: printf-style name for the kthread worker (task). |
828 | * | |
829 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
830 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) | |
831 | * when the worker was SIGKILLed. | |
832 | */ | |
833 | struct kthread_worker * | |
dbf52682 | 834 | kthread_create_worker(unsigned int flags, const char namefmt[], ...) |
fbae2d44 PM |
835 | { |
836 | struct kthread_worker *worker; | |
837 | va_list args; | |
838 | ||
839 | va_start(args, namefmt); | |
dbf52682 | 840 | worker = __kthread_create_worker(-1, flags, namefmt, args); |
fbae2d44 PM |
841 | va_end(args); |
842 | ||
843 | return worker; | |
844 | } | |
845 | EXPORT_SYMBOL(kthread_create_worker); | |
846 | ||
847 | /** | |
848 | * kthread_create_worker_on_cpu - create a kthread worker and bind it | |
7b7b8a2c | 849 | * to a given CPU and the associated NUMA node. |
fbae2d44 | 850 | * @cpu: CPU number |
dbf52682 | 851 | * @flags: flags modifying the default behavior of the worker |
fbae2d44 PM |
852 | * @namefmt: printf-style name for the kthread worker (task). |
853 | * | |
854 | * Use a valid CPU number if you want to bind the kthread worker | |
855 | * to the given CPU and the associated NUMA node. | |
856 | * | |
857 | * A good practice is to add the cpu number also into the worker name. | |
858 | * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu). | |
859 | * | |
ebb2bdce PM |
860 | * CPU hotplug: |
861 | * The kthread worker API is simple and generic. It just provides a way | |
862 | * to create, use, and destroy workers. | |
863 | * | |
864 | * It is up to the API user how to handle CPU hotplug. They have to decide | |
865 | * how to handle pending work items, prevent queuing new ones, and | |
866 | * restore the functionality when the CPU goes off and on. There are a | |
867 | * few catches: | |
868 | * | |
869 | * - CPU affinity gets lost when it is scheduled on an offline CPU. | |
870 | * | |
871 | * - The worker might not exist when the CPU was off when the user | |
872 | * created the workers. | |
873 | * | |
874 | * Good practice is to implement two CPU hotplug callbacks and to | |
875 | * destroy/create the worker when the CPU goes down/up. | |
876 | * | |
877 | * Return: | |
878 | * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
fbae2d44 PM |
879 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) |
880 | * when the worker was SIGKILLed. | |
881 | */ | |
882 | struct kthread_worker * | |
dbf52682 PM |
883 | kthread_create_worker_on_cpu(int cpu, unsigned int flags, |
884 | const char namefmt[], ...) | |
fbae2d44 PM |
885 | { |
886 | struct kthread_worker *worker; | |
887 | va_list args; | |
888 | ||
889 | va_start(args, namefmt); | |
dbf52682 | 890 | worker = __kthread_create_worker(cpu, flags, namefmt, args); |
fbae2d44 PM |
891 | va_end(args); |
892 | ||
893 | return worker; | |
894 | } | |
895 | EXPORT_SYMBOL(kthread_create_worker_on_cpu); | |
896 | ||
37be45d4 PM |
897 | /* |
898 | * Returns true when the work could not be queued at the moment. | |
899 | * It happens when it is already pending in a worker list | |
900 | * or when it is being cancelled. | |
901 | */ | |
902 | static inline bool queuing_blocked(struct kthread_worker *worker, | |
903 | struct kthread_work *work) | |
904 | { | |
905 | lockdep_assert_held(&worker->lock); | |
906 | ||
907 | return !list_empty(&work->node) || work->canceling; | |
908 | } | |
909 | ||
8197b3d4 PM |
910 | static void kthread_insert_work_sanity_check(struct kthread_worker *worker, |
911 | struct kthread_work *work) | |
912 | { | |
913 | lockdep_assert_held(&worker->lock); | |
914 | WARN_ON_ONCE(!list_empty(&work->node)); | |
915 | /* Do not use a work with >1 worker, see kthread_queue_work() */ | |
916 | WARN_ON_ONCE(work->worker && work->worker != worker); | |
917 | } | |
918 | ||
9a2e03d8 | 919 | /* insert @work before @pos in @worker */ |
3989144f | 920 | static void kthread_insert_work(struct kthread_worker *worker, |
8197b3d4 PM |
921 | struct kthread_work *work, |
922 | struct list_head *pos) | |
9a2e03d8 | 923 | { |
8197b3d4 | 924 | kthread_insert_work_sanity_check(worker, work); |
9a2e03d8 | 925 | |
f630c7c6 RC |
926 | trace_sched_kthread_work_queue_work(worker, work); |
927 | ||
9a2e03d8 | 928 | list_add_tail(&work->node, pos); |
46f3d976 | 929 | work->worker = worker; |
ed1403ec | 930 | if (!worker->current_work && likely(worker->task)) |
9a2e03d8 TH |
931 | wake_up_process(worker->task); |
932 | } | |
933 | ||
b56c0d89 | 934 | /** |
3989144f | 935 | * kthread_queue_work - queue a kthread_work |
b56c0d89 TH |
936 | * @worker: target kthread_worker |
937 | * @work: kthread_work to queue | |
938 | * | |
939 | * Queue @work to work processor @task for async execution. @task | |
940 | * must have been created with kthread_worker_create(). Returns %true | |
941 | * if @work was successfully queued, %false if it was already pending. | |
8197b3d4 PM |
942 | * |
943 | * Reinitialize the work if it needs to be used by another worker. | |
944 | * For example, when the worker was stopped and started again. | |
b56c0d89 | 945 | */ |
3989144f | 946 | bool kthread_queue_work(struct kthread_worker *worker, |
b56c0d89 TH |
947 | struct kthread_work *work) |
948 | { | |
949 | bool ret = false; | |
950 | unsigned long flags; | |
951 | ||
fe99a4f4 | 952 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 | 953 | if (!queuing_blocked(worker, work)) { |
3989144f | 954 | kthread_insert_work(worker, work, &worker->work_list); |
b56c0d89 TH |
955 | ret = true; |
956 | } | |
fe99a4f4 | 957 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
b56c0d89 TH |
958 | return ret; |
959 | } | |
3989144f | 960 | EXPORT_SYMBOL_GPL(kthread_queue_work); |
b56c0d89 | 961 | |
22597dc3 PM |
962 | /** |
963 | * kthread_delayed_work_timer_fn - callback that queues the associated kthread | |
964 | * delayed work when the timer expires. | |
fe5c3b69 | 965 | * @t: pointer to the expired timer |
22597dc3 PM |
966 | * |
967 | * The format of the function is defined by struct timer_list. | |
968 | * It should have been called from irqsafe timer with irq already off. | |
969 | */ | |
fe5c3b69 | 970 | void kthread_delayed_work_timer_fn(struct timer_list *t) |
22597dc3 | 971 | { |
fe5c3b69 | 972 | struct kthread_delayed_work *dwork = from_timer(dwork, t, timer); |
22597dc3 PM |
973 | struct kthread_work *work = &dwork->work; |
974 | struct kthread_worker *worker = work->worker; | |
ad01423a | 975 | unsigned long flags; |
22597dc3 PM |
976 | |
977 | /* | |
978 | * This might happen when a pending work is reinitialized. | |
979 | * It means that it is used a wrong way. | |
980 | */ | |
981 | if (WARN_ON_ONCE(!worker)) | |
982 | return; | |
983 | ||
ad01423a | 984 | raw_spin_lock_irqsave(&worker->lock, flags); |
22597dc3 PM |
985 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
986 | WARN_ON_ONCE(work->worker != worker); | |
987 | ||
988 | /* Move the work from worker->delayed_work_list. */ | |
989 | WARN_ON_ONCE(list_empty(&work->node)); | |
990 | list_del_init(&work->node); | |
6993d0fd Z |
991 | if (!work->canceling) |
992 | kthread_insert_work(worker, work, &worker->work_list); | |
22597dc3 | 993 | |
ad01423a | 994 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
22597dc3 PM |
995 | } |
996 | EXPORT_SYMBOL(kthread_delayed_work_timer_fn); | |
997 | ||
bc88f85c BD |
998 | static void __kthread_queue_delayed_work(struct kthread_worker *worker, |
999 | struct kthread_delayed_work *dwork, | |
1000 | unsigned long delay) | |
22597dc3 PM |
1001 | { |
1002 | struct timer_list *timer = &dwork->timer; | |
1003 | struct kthread_work *work = &dwork->work; | |
1004 | ||
0a5b4128 ST |
1005 | WARN_ON_FUNCTION_MISMATCH(timer->function, |
1006 | kthread_delayed_work_timer_fn); | |
22597dc3 PM |
1007 | |
1008 | /* | |
1009 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1010 | * both optimization and correctness. The earliest @timer can | |
1011 | * expire is on the closest next tick and delayed_work users depend | |
1012 | * on that there's no such delay when @delay is 0. | |
1013 | */ | |
1014 | if (!delay) { | |
1015 | kthread_insert_work(worker, work, &worker->work_list); | |
1016 | return; | |
1017 | } | |
1018 | ||
1019 | /* Be paranoid and try to detect possible races already now. */ | |
1020 | kthread_insert_work_sanity_check(worker, work); | |
1021 | ||
1022 | list_add(&work->node, &worker->delayed_work_list); | |
1023 | work->worker = worker; | |
22597dc3 PM |
1024 | timer->expires = jiffies + delay; |
1025 | add_timer(timer); | |
1026 | } | |
1027 | ||
1028 | /** | |
1029 | * kthread_queue_delayed_work - queue the associated kthread work | |
1030 | * after a delay. | |
1031 | * @worker: target kthread_worker | |
1032 | * @dwork: kthread_delayed_work to queue | |
1033 | * @delay: number of jiffies to wait before queuing | |
1034 | * | |
1035 | * If the work has not been pending it starts a timer that will queue | |
1036 | * the work after the given @delay. If @delay is zero, it queues the | |
1037 | * work immediately. | |
1038 | * | |
1039 | * Return: %false if the @work has already been pending. It means that | |
1040 | * either the timer was running or the work was queued. It returns %true | |
1041 | * otherwise. | |
1042 | */ | |
1043 | bool kthread_queue_delayed_work(struct kthread_worker *worker, | |
1044 | struct kthread_delayed_work *dwork, | |
1045 | unsigned long delay) | |
1046 | { | |
1047 | struct kthread_work *work = &dwork->work; | |
1048 | unsigned long flags; | |
1049 | bool ret = false; | |
1050 | ||
fe99a4f4 | 1051 | raw_spin_lock_irqsave(&worker->lock, flags); |
22597dc3 | 1052 | |
37be45d4 | 1053 | if (!queuing_blocked(worker, work)) { |
22597dc3 PM |
1054 | __kthread_queue_delayed_work(worker, dwork, delay); |
1055 | ret = true; | |
1056 | } | |
1057 | ||
fe99a4f4 | 1058 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
22597dc3 PM |
1059 | return ret; |
1060 | } | |
1061 | EXPORT_SYMBOL_GPL(kthread_queue_delayed_work); | |
1062 | ||
9a2e03d8 TH |
1063 | struct kthread_flush_work { |
1064 | struct kthread_work work; | |
1065 | struct completion done; | |
1066 | }; | |
1067 | ||
1068 | static void kthread_flush_work_fn(struct kthread_work *work) | |
1069 | { | |
1070 | struct kthread_flush_work *fwork = | |
1071 | container_of(work, struct kthread_flush_work, work); | |
1072 | complete(&fwork->done); | |
1073 | } | |
1074 | ||
b56c0d89 | 1075 | /** |
3989144f | 1076 | * kthread_flush_work - flush a kthread_work |
b56c0d89 TH |
1077 | * @work: work to flush |
1078 | * | |
1079 | * If @work is queued or executing, wait for it to finish execution. | |
1080 | */ | |
3989144f | 1081 | void kthread_flush_work(struct kthread_work *work) |
b56c0d89 | 1082 | { |
46f3d976 TH |
1083 | struct kthread_flush_work fwork = { |
1084 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
1085 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
1086 | }; | |
1087 | struct kthread_worker *worker; | |
1088 | bool noop = false; | |
1089 | ||
46f3d976 TH |
1090 | worker = work->worker; |
1091 | if (!worker) | |
1092 | return; | |
b56c0d89 | 1093 | |
fe99a4f4 | 1094 | raw_spin_lock_irq(&worker->lock); |
8197b3d4 PM |
1095 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
1096 | WARN_ON_ONCE(work->worker != worker); | |
b56c0d89 | 1097 | |
46f3d976 | 1098 | if (!list_empty(&work->node)) |
3989144f | 1099 | kthread_insert_work(worker, &fwork.work, work->node.next); |
46f3d976 | 1100 | else if (worker->current_work == work) |
3989144f PM |
1101 | kthread_insert_work(worker, &fwork.work, |
1102 | worker->work_list.next); | |
46f3d976 TH |
1103 | else |
1104 | noop = true; | |
b56c0d89 | 1105 | |
fe99a4f4 | 1106 | raw_spin_unlock_irq(&worker->lock); |
b56c0d89 | 1107 | |
46f3d976 TH |
1108 | if (!noop) |
1109 | wait_for_completion(&fwork.done); | |
b56c0d89 | 1110 | } |
3989144f | 1111 | EXPORT_SYMBOL_GPL(kthread_flush_work); |
b56c0d89 | 1112 | |
34b3d534 PM |
1113 | /* |
1114 | * Make sure that the timer is neither set nor running and could | |
1115 | * not manipulate the work list_head any longer. | |
1116 | * | |
1117 | * The function is called under worker->lock. The lock is temporary | |
1118 | * released but the timer can't be set again in the meantime. | |
1119 | */ | |
1120 | static void kthread_cancel_delayed_work_timer(struct kthread_work *work, | |
1121 | unsigned long *flags) | |
1122 | { | |
1123 | struct kthread_delayed_work *dwork = | |
1124 | container_of(work, struct kthread_delayed_work, work); | |
1125 | struct kthread_worker *worker = work->worker; | |
1126 | ||
1127 | /* | |
1128 | * del_timer_sync() must be called to make sure that the timer | |
1129 | * callback is not running. The lock must be temporary released | |
1130 | * to avoid a deadlock with the callback. In the meantime, | |
1131 | * any queuing is blocked by setting the canceling counter. | |
1132 | */ | |
1133 | work->canceling++; | |
1134 | raw_spin_unlock_irqrestore(&worker->lock, *flags); | |
1135 | del_timer_sync(&dwork->timer); | |
1136 | raw_spin_lock_irqsave(&worker->lock, *flags); | |
1137 | work->canceling--; | |
1138 | } | |
1139 | ||
37be45d4 | 1140 | /* |
5fa54346 PM |
1141 | * This function removes the work from the worker queue. |
1142 | * | |
1143 | * It is called under worker->lock. The caller must make sure that | |
1144 | * the timer used by delayed work is not running, e.g. by calling | |
1145 | * kthread_cancel_delayed_work_timer(). | |
37be45d4 PM |
1146 | * |
1147 | * The work might still be in use when this function finishes. See the | |
1148 | * current_work proceed by the worker. | |
1149 | * | |
1150 | * Return: %true if @work was pending and successfully canceled, | |
1151 | * %false if @work was not pending | |
1152 | */ | |
5fa54346 | 1153 | static bool __kthread_cancel_work(struct kthread_work *work) |
37be45d4 | 1154 | { |
37be45d4 PM |
1155 | /* |
1156 | * Try to remove the work from a worker list. It might either | |
1157 | * be from worker->work_list or from worker->delayed_work_list. | |
1158 | */ | |
1159 | if (!list_empty(&work->node)) { | |
1160 | list_del_init(&work->node); | |
1161 | return true; | |
1162 | } | |
1163 | ||
1164 | return false; | |
1165 | } | |
1166 | ||
9a6b06c8 PM |
1167 | /** |
1168 | * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work | |
1169 | * @worker: kthread worker to use | |
1170 | * @dwork: kthread delayed work to queue | |
1171 | * @delay: number of jiffies to wait before queuing | |
1172 | * | |
1173 | * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise, | |
1174 | * modify @dwork's timer so that it expires after @delay. If @delay is zero, | |
1175 | * @work is guaranteed to be queued immediately. | |
1176 | * | |
d71ba164 | 1177 | * Return: %false if @dwork was idle and queued, %true otherwise. |
9a6b06c8 PM |
1178 | * |
1179 | * A special case is when the work is being canceled in parallel. | |
1180 | * It might be caused either by the real kthread_cancel_delayed_work_sync() | |
1181 | * or yet another kthread_mod_delayed_work() call. We let the other command | |
d71ba164 PM |
1182 | * win and return %true here. The return value can be used for reference |
1183 | * counting and the number of queued works stays the same. Anyway, the caller | |
1184 | * is supposed to synchronize these operations a reasonable way. | |
9a6b06c8 PM |
1185 | * |
1186 | * This function is safe to call from any context including IRQ handler. | |
1187 | * See __kthread_cancel_work() and kthread_delayed_work_timer_fn() | |
1188 | * for details. | |
1189 | */ | |
1190 | bool kthread_mod_delayed_work(struct kthread_worker *worker, | |
1191 | struct kthread_delayed_work *dwork, | |
1192 | unsigned long delay) | |
1193 | { | |
1194 | struct kthread_work *work = &dwork->work; | |
1195 | unsigned long flags; | |
d71ba164 | 1196 | int ret; |
9a6b06c8 | 1197 | |
fe99a4f4 | 1198 | raw_spin_lock_irqsave(&worker->lock, flags); |
9a6b06c8 PM |
1199 | |
1200 | /* Do not bother with canceling when never queued. */ | |
d71ba164 PM |
1201 | if (!work->worker) { |
1202 | ret = false; | |
9a6b06c8 | 1203 | goto fast_queue; |
d71ba164 | 1204 | } |
9a6b06c8 PM |
1205 | |
1206 | /* Work must not be used with >1 worker, see kthread_queue_work() */ | |
1207 | WARN_ON_ONCE(work->worker != worker); | |
1208 | ||
5fa54346 PM |
1209 | /* |
1210 | * Temporary cancel the work but do not fight with another command | |
1211 | * that is canceling the work as well. | |
1212 | * | |
1213 | * It is a bit tricky because of possible races with another | |
1214 | * mod_delayed_work() and cancel_delayed_work() callers. | |
1215 | * | |
1216 | * The timer must be canceled first because worker->lock is released | |
1217 | * when doing so. But the work can be removed from the queue (list) | |
1218 | * only when it can be queued again so that the return value can | |
1219 | * be used for reference counting. | |
1220 | */ | |
1221 | kthread_cancel_delayed_work_timer(work, &flags); | |
d71ba164 PM |
1222 | if (work->canceling) { |
1223 | /* The number of works in the queue does not change. */ | |
1224 | ret = true; | |
9a6b06c8 | 1225 | goto out; |
d71ba164 | 1226 | } |
5fa54346 | 1227 | ret = __kthread_cancel_work(work); |
9a6b06c8 | 1228 | |
9a6b06c8 PM |
1229 | fast_queue: |
1230 | __kthread_queue_delayed_work(worker, dwork, delay); | |
1231 | out: | |
fe99a4f4 | 1232 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
9a6b06c8 PM |
1233 | return ret; |
1234 | } | |
1235 | EXPORT_SYMBOL_GPL(kthread_mod_delayed_work); | |
1236 | ||
37be45d4 PM |
1237 | static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork) |
1238 | { | |
1239 | struct kthread_worker *worker = work->worker; | |
1240 | unsigned long flags; | |
1241 | int ret = false; | |
1242 | ||
1243 | if (!worker) | |
1244 | goto out; | |
1245 | ||
fe99a4f4 | 1246 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 PM |
1247 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
1248 | WARN_ON_ONCE(work->worker != worker); | |
1249 | ||
5fa54346 PM |
1250 | if (is_dwork) |
1251 | kthread_cancel_delayed_work_timer(work, &flags); | |
1252 | ||
1253 | ret = __kthread_cancel_work(work); | |
37be45d4 PM |
1254 | |
1255 | if (worker->current_work != work) | |
1256 | goto out_fast; | |
1257 | ||
1258 | /* | |
1259 | * The work is in progress and we need to wait with the lock released. | |
1260 | * In the meantime, block any queuing by setting the canceling counter. | |
1261 | */ | |
1262 | work->canceling++; | |
fe99a4f4 | 1263 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
37be45d4 | 1264 | kthread_flush_work(work); |
fe99a4f4 | 1265 | raw_spin_lock_irqsave(&worker->lock, flags); |
37be45d4 PM |
1266 | work->canceling--; |
1267 | ||
1268 | out_fast: | |
fe99a4f4 | 1269 | raw_spin_unlock_irqrestore(&worker->lock, flags); |
37be45d4 PM |
1270 | out: |
1271 | return ret; | |
1272 | } | |
1273 | ||
1274 | /** | |
1275 | * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish | |
1276 | * @work: the kthread work to cancel | |
1277 | * | |
1278 | * Cancel @work and wait for its execution to finish. This function | |
1279 | * can be used even if the work re-queues itself. On return from this | |
1280 | * function, @work is guaranteed to be not pending or executing on any CPU. | |
1281 | * | |
1282 | * kthread_cancel_work_sync(&delayed_work->work) must not be used for | |
1283 | * delayed_work's. Use kthread_cancel_delayed_work_sync() instead. | |
1284 | * | |
1285 | * The caller must ensure that the worker on which @work was last | |
1286 | * queued can't be destroyed before this function returns. | |
1287 | * | |
1288 | * Return: %true if @work was pending, %false otherwise. | |
1289 | */ | |
1290 | bool kthread_cancel_work_sync(struct kthread_work *work) | |
1291 | { | |
1292 | return __kthread_cancel_work_sync(work, false); | |
1293 | } | |
1294 | EXPORT_SYMBOL_GPL(kthread_cancel_work_sync); | |
1295 | ||
1296 | /** | |
1297 | * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and | |
1298 | * wait for it to finish. | |
1299 | * @dwork: the kthread delayed work to cancel | |
1300 | * | |
1301 | * This is kthread_cancel_work_sync() for delayed works. | |
1302 | * | |
1303 | * Return: %true if @dwork was pending, %false otherwise. | |
1304 | */ | |
1305 | bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork) | |
1306 | { | |
1307 | return __kthread_cancel_work_sync(&dwork->work, true); | |
1308 | } | |
1309 | EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync); | |
1310 | ||
b56c0d89 | 1311 | /** |
3989144f | 1312 | * kthread_flush_worker - flush all current works on a kthread_worker |
b56c0d89 TH |
1313 | * @worker: worker to flush |
1314 | * | |
1315 | * Wait until all currently executing or pending works on @worker are | |
1316 | * finished. | |
1317 | */ | |
3989144f | 1318 | void kthread_flush_worker(struct kthread_worker *worker) |
b56c0d89 TH |
1319 | { |
1320 | struct kthread_flush_work fwork = { | |
1321 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
1322 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
1323 | }; | |
1324 | ||
3989144f | 1325 | kthread_queue_work(worker, &fwork.work); |
b56c0d89 TH |
1326 | wait_for_completion(&fwork.done); |
1327 | } | |
3989144f | 1328 | EXPORT_SYMBOL_GPL(kthread_flush_worker); |
35033fe9 PM |
1329 | |
1330 | /** | |
1331 | * kthread_destroy_worker - destroy a kthread worker | |
1332 | * @worker: worker to be destroyed | |
1333 | * | |
1334 | * Flush and destroy @worker. The simple flush is enough because the kthread | |
1335 | * worker API is used only in trivial scenarios. There are no multi-step state | |
1336 | * machines needed. | |
1337 | */ | |
1338 | void kthread_destroy_worker(struct kthread_worker *worker) | |
1339 | { | |
1340 | struct task_struct *task; | |
1341 | ||
1342 | task = worker->task; | |
1343 | if (WARN_ON(!task)) | |
1344 | return; | |
1345 | ||
1346 | kthread_flush_worker(worker); | |
1347 | kthread_stop(task); | |
1348 | WARN_ON(!list_empty(&worker->work_list)); | |
1349 | kfree(worker); | |
1350 | } | |
1351 | EXPORT_SYMBOL(kthread_destroy_worker); | |
05e3db95 | 1352 | |
f5678e7f CH |
1353 | /** |
1354 | * kthread_use_mm - make the calling kthread operate on an address space | |
1355 | * @mm: address space to operate on | |
9bf5b9eb | 1356 | */ |
f5678e7f | 1357 | void kthread_use_mm(struct mm_struct *mm) |
9bf5b9eb CH |
1358 | { |
1359 | struct mm_struct *active_mm; | |
1360 | struct task_struct *tsk = current; | |
1361 | ||
f5678e7f CH |
1362 | WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD)); |
1363 | WARN_ON_ONCE(tsk->mm); | |
1364 | ||
9bf5b9eb | 1365 | task_lock(tsk); |
38cf307c PZ |
1366 | /* Hold off tlb flush IPIs while switching mm's */ |
1367 | local_irq_disable(); | |
9bf5b9eb CH |
1368 | active_mm = tsk->active_mm; |
1369 | if (active_mm != mm) { | |
1370 | mmgrab(mm); | |
1371 | tsk->active_mm = mm; | |
1372 | } | |
1373 | tsk->mm = mm; | |
618758ed | 1374 | membarrier_update_current_mm(mm); |
38cf307c PZ |
1375 | switch_mm_irqs_off(active_mm, mm, tsk); |
1376 | local_irq_enable(); | |
9bf5b9eb CH |
1377 | task_unlock(tsk); |
1378 | #ifdef finish_arch_post_lock_switch | |
1379 | finish_arch_post_lock_switch(); | |
1380 | #endif | |
1381 | ||
618758ed MD |
1382 | /* |
1383 | * When a kthread starts operating on an address space, the loop | |
1384 | * in membarrier_{private,global}_expedited() may not observe | |
1385 | * that tsk->mm, and not issue an IPI. Membarrier requires a | |
1386 | * memory barrier after storing to tsk->mm, before accessing | |
1387 | * user-space memory. A full memory barrier for membarrier | |
1388 | * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by | |
1389 | * mmdrop(), or explicitly with smp_mb(). | |
1390 | */ | |
9bf5b9eb CH |
1391 | if (active_mm != mm) |
1392 | mmdrop(active_mm); | |
618758ed MD |
1393 | else |
1394 | smp_mb(); | |
37c54f9b | 1395 | |
3d13f313 | 1396 | to_kthread(tsk)->oldfs = force_uaccess_begin(); |
9bf5b9eb | 1397 | } |
f5678e7f | 1398 | EXPORT_SYMBOL_GPL(kthread_use_mm); |
9bf5b9eb | 1399 | |
f5678e7f CH |
1400 | /** |
1401 | * kthread_unuse_mm - reverse the effect of kthread_use_mm() | |
1402 | * @mm: address space to operate on | |
9bf5b9eb | 1403 | */ |
f5678e7f | 1404 | void kthread_unuse_mm(struct mm_struct *mm) |
9bf5b9eb CH |
1405 | { |
1406 | struct task_struct *tsk = current; | |
1407 | ||
f5678e7f CH |
1408 | WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD)); |
1409 | WARN_ON_ONCE(!tsk->mm); | |
1410 | ||
3d13f313 | 1411 | force_uaccess_end(to_kthread(tsk)->oldfs); |
37c54f9b | 1412 | |
9bf5b9eb | 1413 | task_lock(tsk); |
618758ed MD |
1414 | /* |
1415 | * When a kthread stops operating on an address space, the loop | |
1416 | * in membarrier_{private,global}_expedited() may not observe | |
1417 | * that tsk->mm, and not issue an IPI. Membarrier requires a | |
1418 | * memory barrier after accessing user-space memory, before | |
1419 | * clearing tsk->mm. | |
1420 | */ | |
1421 | smp_mb__after_spinlock(); | |
9bf5b9eb | 1422 | sync_mm_rss(mm); |
38cf307c | 1423 | local_irq_disable(); |
9bf5b9eb | 1424 | tsk->mm = NULL; |
618758ed | 1425 | membarrier_update_current_mm(NULL); |
9bf5b9eb CH |
1426 | /* active_mm is still 'mm' */ |
1427 | enter_lazy_tlb(mm, tsk); | |
38cf307c | 1428 | local_irq_enable(); |
9bf5b9eb CH |
1429 | task_unlock(tsk); |
1430 | } | |
f5678e7f | 1431 | EXPORT_SYMBOL_GPL(kthread_unuse_mm); |
9bf5b9eb | 1432 | |
0b508bc9 | 1433 | #ifdef CONFIG_BLK_CGROUP |
05e3db95 SL |
1434 | /** |
1435 | * kthread_associate_blkcg - associate blkcg to current kthread | |
1436 | * @css: the cgroup info | |
1437 | * | |
1438 | * Current thread must be a kthread. The thread is running jobs on behalf of | |
1439 | * other threads. In some cases, we expect the jobs attach cgroup info of | |
1440 | * original threads instead of that of current thread. This function stores | |
1441 | * original thread's cgroup info in current kthread context for later | |
1442 | * retrieval. | |
1443 | */ | |
1444 | void kthread_associate_blkcg(struct cgroup_subsys_state *css) | |
1445 | { | |
1446 | struct kthread *kthread; | |
1447 | ||
1448 | if (!(current->flags & PF_KTHREAD)) | |
1449 | return; | |
1450 | kthread = to_kthread(current); | |
1451 | if (!kthread) | |
1452 | return; | |
1453 | ||
1454 | if (kthread->blkcg_css) { | |
1455 | css_put(kthread->blkcg_css); | |
1456 | kthread->blkcg_css = NULL; | |
1457 | } | |
1458 | if (css) { | |
1459 | css_get(css); | |
1460 | kthread->blkcg_css = css; | |
1461 | } | |
1462 | } | |
1463 | EXPORT_SYMBOL(kthread_associate_blkcg); | |
1464 | ||
1465 | /** | |
1466 | * kthread_blkcg - get associated blkcg css of current kthread | |
1467 | * | |
1468 | * Current thread must be a kthread. | |
1469 | */ | |
1470 | struct cgroup_subsys_state *kthread_blkcg(void) | |
1471 | { | |
1472 | struct kthread *kthread; | |
1473 | ||
1474 | if (current->flags & PF_KTHREAD) { | |
1475 | kthread = to_kthread(current); | |
1476 | if (kthread) | |
1477 | return kthread->blkcg_css; | |
1478 | } | |
1479 | return NULL; | |
1480 | } | |
1481 | EXPORT_SYMBOL(kthread_blkcg); | |
1482 | #endif |