]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* CPU control. |
2 | * (C) 2001, 2002, 2003, 2004 Rusty Russell | |
3 | * | |
4 | * This code is licenced under the GPL. | |
5 | */ | |
6 | #include <linux/proc_fs.h> | |
7 | #include <linux/smp.h> | |
8 | #include <linux/init.h> | |
9 | #include <linux/notifier.h> | |
10 | #include <linux/sched.h> | |
11 | #include <linux/unistd.h> | |
12 | #include <linux/cpu.h> | |
cb79295e AV |
13 | #include <linux/oom.h> |
14 | #include <linux/rcupdate.h> | |
9984de1a | 15 | #include <linux/export.h> |
e4cc2f87 | 16 | #include <linux/bug.h> |
1da177e4 LT |
17 | #include <linux/kthread.h> |
18 | #include <linux/stop_machine.h> | |
81615b62 | 19 | #include <linux/mutex.h> |
5a0e3ad6 | 20 | #include <linux/gfp.h> |
79cfbdfa | 21 | #include <linux/suspend.h> |
a19423b9 | 22 | #include <linux/lockdep.h> |
bb3632c6 | 23 | #include <trace/events/power.h> |
1da177e4 | 24 | |
38498a67 TG |
25 | #include "smpboot.h" |
26 | ||
98a79d6a | 27 | #ifdef CONFIG_SMP |
b3199c02 | 28 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 29 | static DEFINE_MUTEX(cpu_add_remove_lock); |
1da177e4 | 30 | |
79a6cdeb | 31 | /* |
93ae4f97 SB |
32 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
33 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
34 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
35 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
36 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
37 | */ |
38 | void cpu_maps_update_begin(void) | |
39 | { | |
40 | mutex_lock(&cpu_add_remove_lock); | |
41 | } | |
93ae4f97 | 42 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
43 | |
44 | void cpu_maps_update_done(void) | |
45 | { | |
46 | mutex_unlock(&cpu_add_remove_lock); | |
47 | } | |
93ae4f97 | 48 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 49 | |
5c113fbe | 50 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 51 | |
e3920fb4 RW |
52 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
53 | * Should always be manipulated under cpu_add_remove_lock | |
54 | */ | |
55 | static int cpu_hotplug_disabled; | |
56 | ||
79a6cdeb LJ |
57 | #ifdef CONFIG_HOTPLUG_CPU |
58 | ||
d221938c GS |
59 | static struct { |
60 | struct task_struct *active_writer; | |
87af9e7f DH |
61 | /* wait queue to wake up the active_writer */ |
62 | wait_queue_head_t wq; | |
63 | /* verifies that no writer will get active while readers are active */ | |
64 | struct mutex lock; | |
d221938c GS |
65 | /* |
66 | * Also blocks the new readers during | |
67 | * an ongoing cpu hotplug operation. | |
68 | */ | |
87af9e7f | 69 | atomic_t refcount; |
a19423b9 GS |
70 | |
71 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
72 | struct lockdep_map dep_map; | |
73 | #endif | |
31950eb6 LT |
74 | } cpu_hotplug = { |
75 | .active_writer = NULL, | |
87af9e7f | 76 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 77 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
78 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
79 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
80 | #endif | |
31950eb6 | 81 | }; |
d221938c | 82 | |
a19423b9 GS |
83 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
84 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
85 | #define cpuhp_lock_acquire_tryread() \ |
86 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
87 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
88 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
89 | ||
62db99f4 | 90 | |
86ef5c9a | 91 | void get_online_cpus(void) |
a9d9baa1 | 92 | { |
d221938c GS |
93 | might_sleep(); |
94 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 95 | return; |
a19423b9 | 96 | cpuhp_lock_acquire_read(); |
d221938c | 97 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 98 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 99 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 100 | } |
86ef5c9a | 101 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 102 | |
dd56af42 PM |
103 | bool try_get_online_cpus(void) |
104 | { | |
105 | if (cpu_hotplug.active_writer == current) | |
106 | return true; | |
107 | if (!mutex_trylock(&cpu_hotplug.lock)) | |
108 | return false; | |
109 | cpuhp_lock_acquire_tryread(); | |
87af9e7f | 110 | atomic_inc(&cpu_hotplug.refcount); |
dd56af42 PM |
111 | mutex_unlock(&cpu_hotplug.lock); |
112 | return true; | |
113 | } | |
114 | EXPORT_SYMBOL_GPL(try_get_online_cpus); | |
115 | ||
86ef5c9a | 116 | void put_online_cpus(void) |
a9d9baa1 | 117 | { |
87af9e7f DH |
118 | int refcount; |
119 | ||
d221938c | 120 | if (cpu_hotplug.active_writer == current) |
aa953877 | 121 | return; |
075663d1 | 122 | |
87af9e7f DH |
123 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
124 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
125 | atomic_inc(&cpu_hotplug.refcount); | |
126 | ||
127 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
128 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 129 | |
a19423b9 | 130 | cpuhp_lock_release(); |
d221938c | 131 | |
a9d9baa1 | 132 | } |
86ef5c9a | 133 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 134 | |
d221938c GS |
135 | /* |
136 | * This ensures that the hotplug operation can begin only when the | |
137 | * refcount goes to zero. | |
138 | * | |
139 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
140 | * will be blocked by the cpu_hotplug.lock | |
141 | * | |
d2ba7e2a ON |
142 | * Since cpu_hotplug_begin() is always called after invoking |
143 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
144 | * |
145 | * Note that theoretically, there is a possibility of a livelock: | |
146 | * - Refcount goes to zero, last reader wakes up the sleeping | |
147 | * writer. | |
148 | * - Last reader unlocks the cpu_hotplug.lock. | |
149 | * - A new reader arrives at this moment, bumps up the refcount. | |
150 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
151 | * non zero and goes to sleep again. | |
152 | * | |
153 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 154 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
155 | * |
156 | */ | |
b9d10be7 | 157 | void cpu_hotplug_begin(void) |
d221938c | 158 | { |
87af9e7f | 159 | DEFINE_WAIT(wait); |
d2ba7e2a | 160 | |
87af9e7f | 161 | cpu_hotplug.active_writer = current; |
a19423b9 | 162 | cpuhp_lock_acquire(); |
87af9e7f | 163 | |
d2ba7e2a ON |
164 | for (;;) { |
165 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
166 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
167 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
168 | break; | |
d221938c GS |
169 | mutex_unlock(&cpu_hotplug.lock); |
170 | schedule(); | |
d221938c | 171 | } |
87af9e7f | 172 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
173 | } |
174 | ||
b9d10be7 | 175 | void cpu_hotplug_done(void) |
d221938c GS |
176 | { |
177 | cpu_hotplug.active_writer = NULL; | |
178 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 179 | cpuhp_lock_release(); |
d221938c | 180 | } |
79a6cdeb | 181 | |
16e53dbf SB |
182 | /* |
183 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
184 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
185 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
186 | * hotplug path before performing hotplug operations. So acquiring that lock | |
187 | * guarantees mutual exclusion from any currently running hotplug operations. | |
188 | */ | |
189 | void cpu_hotplug_disable(void) | |
190 | { | |
191 | cpu_maps_update_begin(); | |
192 | cpu_hotplug_disabled = 1; | |
193 | cpu_maps_update_done(); | |
194 | } | |
195 | ||
196 | void cpu_hotplug_enable(void) | |
197 | { | |
198 | cpu_maps_update_begin(); | |
199 | cpu_hotplug_disabled = 0; | |
200 | cpu_maps_update_done(); | |
201 | } | |
202 | ||
b9d10be7 | 203 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 204 | |
1da177e4 | 205 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 206 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 207 | { |
bd5349cf | 208 | int ret; |
d221938c | 209 | cpu_maps_update_begin(); |
bd5349cf | 210 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 211 | cpu_maps_update_done(); |
bd5349cf | 212 | return ret; |
1da177e4 | 213 | } |
65edc68c | 214 | |
93ae4f97 SB |
215 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
216 | { | |
217 | return raw_notifier_chain_register(&cpu_chain, nb); | |
218 | } | |
219 | ||
e9fb7631 AM |
220 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
221 | int *nr_calls) | |
222 | { | |
e6bde73b AM |
223 | int ret; |
224 | ||
225 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 226 | nr_calls); |
e6bde73b AM |
227 | |
228 | return notifier_to_errno(ret); | |
e9fb7631 AM |
229 | } |
230 | ||
231 | static int cpu_notify(unsigned long val, void *v) | |
232 | { | |
233 | return __cpu_notify(val, v, -1, NULL); | |
234 | } | |
235 | ||
00b9b0af LT |
236 | #ifdef CONFIG_HOTPLUG_CPU |
237 | ||
e9fb7631 AM |
238 | static void cpu_notify_nofail(unsigned long val, void *v) |
239 | { | |
00b9b0af | 240 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 241 | } |
1da177e4 | 242 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 243 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 244 | |
9647155f | 245 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 246 | { |
d221938c | 247 | cpu_maps_update_begin(); |
bd5349cf | 248 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 249 | cpu_maps_update_done(); |
1da177e4 LT |
250 | } |
251 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
252 | ||
93ae4f97 SB |
253 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
254 | { | |
255 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
256 | } | |
257 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
258 | ||
e4cc2f87 AV |
259 | /** |
260 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
261 | * @cpu: a CPU id | |
262 | * | |
263 | * This function walks all processes, finds a valid mm struct for each one and | |
264 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
265 | * trivial, there are various non-obvious corner cases, which this function | |
266 | * tries to solve in a safe manner. | |
267 | * | |
268 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
269 | * be called only for an already offlined CPU. | |
270 | */ | |
cb79295e AV |
271 | void clear_tasks_mm_cpumask(int cpu) |
272 | { | |
273 | struct task_struct *p; | |
274 | ||
275 | /* | |
276 | * This function is called after the cpu is taken down and marked | |
277 | * offline, so its not like new tasks will ever get this cpu set in | |
278 | * their mm mask. -- Peter Zijlstra | |
279 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
280 | * full-fledged tasklist_lock. | |
281 | */ | |
e4cc2f87 | 282 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
283 | rcu_read_lock(); |
284 | for_each_process(p) { | |
285 | struct task_struct *t; | |
286 | ||
e4cc2f87 AV |
287 | /* |
288 | * Main thread might exit, but other threads may still have | |
289 | * a valid mm. Find one. | |
290 | */ | |
cb79295e AV |
291 | t = find_lock_task_mm(p); |
292 | if (!t) | |
293 | continue; | |
294 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
295 | task_unlock(t); | |
296 | } | |
297 | rcu_read_unlock(); | |
298 | } | |
299 | ||
b728ca06 | 300 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 301 | { |
b728ca06 | 302 | struct task_struct *g, *p; |
1da177e4 | 303 | |
b728ca06 KT |
304 | read_lock_irq(&tasklist_lock); |
305 | do_each_thread(g, p) { | |
306 | if (!p->on_rq) | |
307 | continue; | |
308 | /* | |
309 | * We do the check with unlocked task_rq(p)->lock. | |
310 | * Order the reading to do not warn about a task, | |
311 | * which was running on this cpu in the past, and | |
312 | * it's just been woken on another cpu. | |
313 | */ | |
314 | rmb(); | |
315 | if (task_cpu(p) != dead_cpu) | |
316 | continue; | |
317 | ||
318 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
319 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
320 | } while_each_thread(g, p); | |
321 | read_unlock_irq(&tasklist_lock); | |
1da177e4 LT |
322 | } |
323 | ||
db912f96 AK |
324 | struct take_cpu_down_param { |
325 | unsigned long mod; | |
326 | void *hcpu; | |
327 | }; | |
328 | ||
1da177e4 | 329 | /* Take this CPU down. */ |
514a20a5 | 330 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 331 | { |
db912f96 | 332 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
333 | int err; |
334 | ||
1da177e4 LT |
335 | /* Ensure this CPU doesn't handle any more interrupts. */ |
336 | err = __cpu_disable(); | |
337 | if (err < 0) | |
f3705136 | 338 | return err; |
1da177e4 | 339 | |
e9fb7631 | 340 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
14e568e7 TG |
341 | /* Park the stopper thread */ |
342 | kthread_park(current); | |
f3705136 | 343 | return 0; |
1da177e4 LT |
344 | } |
345 | ||
e3920fb4 | 346 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 347 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 348 | { |
e7407dcc | 349 | int err, nr_calls = 0; |
e7407dcc | 350 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 351 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
352 | struct take_cpu_down_param tcd_param = { |
353 | .mod = mod, | |
354 | .hcpu = hcpu, | |
355 | }; | |
1da177e4 | 356 | |
e3920fb4 RW |
357 | if (num_online_cpus() == 1) |
358 | return -EBUSY; | |
1da177e4 | 359 | |
e3920fb4 RW |
360 | if (!cpu_online(cpu)) |
361 | return -EINVAL; | |
1da177e4 | 362 | |
d221938c | 363 | cpu_hotplug_begin(); |
4d51985e | 364 | |
e9fb7631 | 365 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 366 | if (err) { |
a0d8cdb6 | 367 | nr_calls--; |
e9fb7631 | 368 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
369 | pr_warn("%s: attempt to take down CPU %u failed\n", |
370 | __func__, cpu); | |
baaca49f | 371 | goto out_release; |
1da177e4 LT |
372 | } |
373 | ||
6acce3ef PZ |
374 | /* |
375 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
376 | * and RCU users of this state to go away such that all new such users | |
377 | * will observe it. | |
378 | * | |
379 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
380 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
381 | * |
382 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
383 | */ |
384 | #ifdef CONFIG_PREEMPT | |
385 | synchronize_sched(); | |
386 | #endif | |
387 | synchronize_rcu(); | |
388 | ||
106dd5af M |
389 | smpboot_park_threads(cpu); |
390 | ||
6acce3ef PZ |
391 | /* |
392 | * So now all preempt/rcu users must observe !cpu_active(). | |
393 | */ | |
394 | ||
e0b582ec | 395 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 396 | if (err) { |
1da177e4 | 397 | /* CPU didn't die: tell everyone. Can't complain. */ |
f97f8f06 | 398 | smpboot_unpark_threads(cpu); |
e9fb7631 | 399 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 400 | goto out_release; |
8fa1d7d3 | 401 | } |
04321587 | 402 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 403 | |
48c5ccae PZ |
404 | /* |
405 | * The migration_call() CPU_DYING callback will have removed all | |
406 | * runnable tasks from the cpu, there's only the idle task left now | |
407 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
408 | * |
409 | * Wait for the stop thread to go away. | |
48c5ccae | 410 | */ |
51a96c77 PZ |
411 | while (!idle_cpu(cpu)) |
412 | cpu_relax(); | |
1da177e4 LT |
413 | |
414 | /* This actually kills the CPU. */ | |
415 | __cpu_die(cpu); | |
416 | ||
1da177e4 | 417 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 418 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
419 | |
420 | check_for_tasks(cpu); | |
421 | ||
baaca49f | 422 | out_release: |
d221938c | 423 | cpu_hotplug_done(); |
e9fb7631 AM |
424 | if (!err) |
425 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
426 | return err; |
427 | } | |
428 | ||
514a20a5 | 429 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 430 | { |
9ea09af3 | 431 | int err; |
e3920fb4 | 432 | |
d221938c | 433 | cpu_maps_update_begin(); |
e761b772 MK |
434 | |
435 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 436 | err = -EBUSY; |
e761b772 MK |
437 | goto out; |
438 | } | |
439 | ||
e761b772 | 440 | err = _cpu_down(cpu, 0); |
e3920fb4 | 441 | |
e761b772 | 442 | out: |
d221938c | 443 | cpu_maps_update_done(); |
1da177e4 LT |
444 | return err; |
445 | } | |
b62b8ef9 | 446 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
447 | #endif /*CONFIG_HOTPLUG_CPU*/ |
448 | ||
e3920fb4 | 449 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 450 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 451 | { |
baaca49f | 452 | int ret, nr_calls = 0; |
1da177e4 | 453 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 454 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 455 | struct task_struct *idle; |
1da177e4 | 456 | |
d221938c | 457 | cpu_hotplug_begin(); |
38498a67 | 458 | |
5e5041f3 YI |
459 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
460 | ret = -EINVAL; | |
461 | goto out; | |
462 | } | |
463 | ||
3bb5d2ee SS |
464 | idle = idle_thread_get(cpu); |
465 | if (IS_ERR(idle)) { | |
466 | ret = PTR_ERR(idle); | |
38498a67 | 467 | goto out; |
3bb5d2ee | 468 | } |
38498a67 | 469 | |
f97f8f06 TG |
470 | ret = smpboot_create_threads(cpu); |
471 | if (ret) | |
472 | goto out; | |
473 | ||
e9fb7631 | 474 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 475 | if (ret) { |
a0d8cdb6 | 476 | nr_calls--; |
84117da5 FF |
477 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
478 | __func__, cpu); | |
1da177e4 LT |
479 | goto out_notify; |
480 | } | |
481 | ||
482 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 483 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
484 | if (ret != 0) |
485 | goto out_notify; | |
6978c705 | 486 | BUG_ON(!cpu_online(cpu)); |
1da177e4 | 487 | |
f97f8f06 TG |
488 | /* Wake the per cpu threads */ |
489 | smpboot_unpark_threads(cpu); | |
490 | ||
1da177e4 | 491 | /* Now call notifier in preparation. */ |
e9fb7631 | 492 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
493 | |
494 | out_notify: | |
495 | if (ret != 0) | |
e9fb7631 | 496 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 497 | out: |
d221938c | 498 | cpu_hotplug_done(); |
e3920fb4 RW |
499 | |
500 | return ret; | |
501 | } | |
502 | ||
0db0628d | 503 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
504 | { |
505 | int err = 0; | |
cf23422b | 506 | |
e0b582ec | 507 | if (!cpu_possible(cpu)) { |
84117da5 FF |
508 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
509 | cpu); | |
87d5e023 | 510 | #if defined(CONFIG_IA64) |
84117da5 | 511 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
512 | #endif |
513 | return -EINVAL; | |
514 | } | |
e3920fb4 | 515 | |
01b0f197 TK |
516 | err = try_online_node(cpu_to_node(cpu)); |
517 | if (err) | |
518 | return err; | |
cf23422b | 519 | |
d221938c | 520 | cpu_maps_update_begin(); |
e761b772 MK |
521 | |
522 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 523 | err = -EBUSY; |
e761b772 MK |
524 | goto out; |
525 | } | |
526 | ||
527 | err = _cpu_up(cpu, 0); | |
528 | ||
e761b772 | 529 | out: |
d221938c | 530 | cpu_maps_update_done(); |
e3920fb4 RW |
531 | return err; |
532 | } | |
a513f6ba | 533 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 534 | |
f3de4be9 | 535 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 536 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
537 | |
538 | int disable_nonboot_cpus(void) | |
539 | { | |
e9a5f426 | 540 | int cpu, first_cpu, error = 0; |
e3920fb4 | 541 | |
d221938c | 542 | cpu_maps_update_begin(); |
e0b582ec | 543 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
544 | /* |
545 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
546 | * with the userspace trying to use the CPU hotplug at the same time |
547 | */ | |
e0b582ec | 548 | cpumask_clear(frozen_cpus); |
6ad4c188 | 549 | |
84117da5 | 550 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
551 | for_each_online_cpu(cpu) { |
552 | if (cpu == first_cpu) | |
553 | continue; | |
bb3632c6 | 554 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 555 | error = _cpu_down(cpu, 1); |
bb3632c6 | 556 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 557 | if (!error) |
e0b582ec | 558 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 559 | else { |
84117da5 | 560 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
561 | break; |
562 | } | |
563 | } | |
86886e55 | 564 | |
e3920fb4 RW |
565 | if (!error) { |
566 | BUG_ON(num_online_cpus() > 1); | |
567 | /* Make sure the CPUs won't be enabled by someone else */ | |
568 | cpu_hotplug_disabled = 1; | |
569 | } else { | |
84117da5 | 570 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 571 | } |
d221938c | 572 | cpu_maps_update_done(); |
e3920fb4 RW |
573 | return error; |
574 | } | |
575 | ||
d0af9eed SS |
576 | void __weak arch_enable_nonboot_cpus_begin(void) |
577 | { | |
578 | } | |
579 | ||
580 | void __weak arch_enable_nonboot_cpus_end(void) | |
581 | { | |
582 | } | |
583 | ||
fa7303e2 | 584 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
585 | { |
586 | int cpu, error; | |
587 | ||
588 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 589 | cpu_maps_update_begin(); |
e3920fb4 | 590 | cpu_hotplug_disabled = 0; |
e0b582ec | 591 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 592 | goto out; |
e3920fb4 | 593 | |
84117da5 | 594 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
595 | |
596 | arch_enable_nonboot_cpus_begin(); | |
597 | ||
e0b582ec | 598 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 599 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 600 | error = _cpu_up(cpu, 1); |
bb3632c6 | 601 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 602 | if (!error) { |
84117da5 | 603 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
604 | continue; |
605 | } | |
84117da5 | 606 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 607 | } |
d0af9eed SS |
608 | |
609 | arch_enable_nonboot_cpus_end(); | |
610 | ||
e0b582ec | 611 | cpumask_clear(frozen_cpus); |
1d64b9cb | 612 | out: |
d221938c | 613 | cpu_maps_update_done(); |
1da177e4 | 614 | } |
e0b582ec | 615 | |
d7268a31 | 616 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
617 | { |
618 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
619 | return -ENOMEM; | |
620 | return 0; | |
621 | } | |
622 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 623 | |
79cfbdfa SB |
624 | /* |
625 | * When callbacks for CPU hotplug notifications are being executed, we must | |
626 | * ensure that the state of the system with respect to the tasks being frozen | |
627 | * or not, as reported by the notification, remains unchanged *throughout the | |
628 | * duration* of the execution of the callbacks. | |
629 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
630 | * | |
631 | * This synchronization is implemented by mutually excluding regular CPU | |
632 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
633 | * Hibernate notifications. | |
634 | */ | |
635 | static int | |
636 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
637 | unsigned long action, void *ptr) | |
638 | { | |
639 | switch (action) { | |
640 | ||
641 | case PM_SUSPEND_PREPARE: | |
642 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 643 | cpu_hotplug_disable(); |
79cfbdfa SB |
644 | break; |
645 | ||
646 | case PM_POST_SUSPEND: | |
647 | case PM_POST_HIBERNATION: | |
16e53dbf | 648 | cpu_hotplug_enable(); |
79cfbdfa SB |
649 | break; |
650 | ||
651 | default: | |
652 | return NOTIFY_DONE; | |
653 | } | |
654 | ||
655 | return NOTIFY_OK; | |
656 | } | |
657 | ||
658 | ||
d7268a31 | 659 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 660 | { |
6e32d479 FY |
661 | /* |
662 | * cpu_hotplug_pm_callback has higher priority than x86 | |
663 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
664 | * to disable cpu hotplug to avoid cpu hotplug race. | |
665 | */ | |
79cfbdfa SB |
666 | pm_notifier(cpu_hotplug_pm_callback, 0); |
667 | return 0; | |
668 | } | |
669 | core_initcall(cpu_hotplug_pm_sync_init); | |
670 | ||
f3de4be9 | 671 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 672 | |
e545a614 MS |
673 | /** |
674 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
675 | * @cpu: cpu that just started | |
676 | * | |
677 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
678 | * It must be called by the arch code on the new cpu, before the new cpu | |
679 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
680 | */ | |
0db0628d | 681 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
682 | { |
683 | unsigned long val = CPU_STARTING; | |
684 | ||
685 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 686 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
687 | val = CPU_STARTING_FROZEN; |
688 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 689 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
690 | } |
691 | ||
68f4f1ec | 692 | #endif /* CONFIG_SMP */ |
b8d317d1 | 693 | |
e56b3bc7 LT |
694 | /* |
695 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
696 | * represents all NR_CPUS bits binary values of 1<<nr. | |
697 | * | |
e0b582ec | 698 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
699 | * mask value that has a single bit set only. |
700 | */ | |
b8d317d1 | 701 | |
e56b3bc7 | 702 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 703 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
704 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
705 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
706 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 707 | |
e56b3bc7 LT |
708 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
709 | ||
710 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
711 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
712 | #if BITS_PER_LONG > 32 | |
713 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
714 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
715 | #endif |
716 | }; | |
e56b3bc7 | 717 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
718 | |
719 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
720 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
721 | |
722 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
723 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
724 | = CPU_BITS_ALL; | |
725 | #else | |
726 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
727 | #endif | |
728 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
729 | EXPORT_SYMBOL(cpu_possible_mask); | |
730 | ||
731 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
732 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
733 | EXPORT_SYMBOL(cpu_online_mask); | |
734 | ||
735 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
736 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
737 | EXPORT_SYMBOL(cpu_present_mask); | |
738 | ||
739 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
740 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
741 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
742 | |
743 | void set_cpu_possible(unsigned int cpu, bool possible) | |
744 | { | |
745 | if (possible) | |
746 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
747 | else | |
748 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
749 | } | |
750 | ||
751 | void set_cpu_present(unsigned int cpu, bool present) | |
752 | { | |
753 | if (present) | |
754 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
755 | else | |
756 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
757 | } | |
758 | ||
759 | void set_cpu_online(unsigned int cpu, bool online) | |
760 | { | |
6acbfb96 | 761 | if (online) { |
3fa41520 | 762 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
763 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
764 | } else { | |
3fa41520 | 765 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 766 | } |
3fa41520 RR |
767 | } |
768 | ||
769 | void set_cpu_active(unsigned int cpu, bool active) | |
770 | { | |
771 | if (active) | |
772 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
773 | else | |
774 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
775 | } | |
776 | ||
777 | void init_cpu_present(const struct cpumask *src) | |
778 | { | |
779 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
780 | } | |
781 | ||
782 | void init_cpu_possible(const struct cpumask *src) | |
783 | { | |
784 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
785 | } | |
786 | ||
787 | void init_cpu_online(const struct cpumask *src) | |
788 | { | |
789 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
790 | } |