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; | |
61 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
62 | /* | |
63 | * Also blocks the new readers during | |
64 | * an ongoing cpu hotplug operation. | |
65 | */ | |
66 | int refcount; | |
b2c4623d PM |
67 | /* And allows lockless put_online_cpus(). */ |
68 | atomic_t puts_pending; | |
a19423b9 GS |
69 | |
70 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
71 | struct lockdep_map dep_map; | |
72 | #endif | |
31950eb6 LT |
73 | } cpu_hotplug = { |
74 | .active_writer = NULL, | |
75 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), | |
76 | .refcount = 0, | |
a19423b9 GS |
77 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
78 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
79 | #endif | |
31950eb6 | 80 | }; |
d221938c | 81 | |
a19423b9 GS |
82 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
83 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
84 | #define cpuhp_lock_acquire_tryread() \ |
85 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
86 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
87 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
88 | ||
86ef5c9a | 89 | void get_online_cpus(void) |
a9d9baa1 | 90 | { |
d221938c GS |
91 | might_sleep(); |
92 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 93 | return; |
a19423b9 | 94 | cpuhp_lock_acquire_read(); |
d221938c GS |
95 | mutex_lock(&cpu_hotplug.lock); |
96 | cpu_hotplug.refcount++; | |
97 | mutex_unlock(&cpu_hotplug.lock); | |
a9d9baa1 | 98 | } |
86ef5c9a | 99 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 100 | |
dd56af42 PM |
101 | bool try_get_online_cpus(void) |
102 | { | |
103 | if (cpu_hotplug.active_writer == current) | |
104 | return true; | |
105 | if (!mutex_trylock(&cpu_hotplug.lock)) | |
106 | return false; | |
107 | cpuhp_lock_acquire_tryread(); | |
108 | cpu_hotplug.refcount++; | |
109 | mutex_unlock(&cpu_hotplug.lock); | |
110 | return true; | |
111 | } | |
112 | EXPORT_SYMBOL_GPL(try_get_online_cpus); | |
113 | ||
86ef5c9a | 114 | void put_online_cpus(void) |
a9d9baa1 | 115 | { |
d221938c | 116 | if (cpu_hotplug.active_writer == current) |
aa953877 | 117 | return; |
b2c4623d PM |
118 | if (!mutex_trylock(&cpu_hotplug.lock)) { |
119 | atomic_inc(&cpu_hotplug.puts_pending); | |
120 | cpuhp_lock_release(); | |
121 | return; | |
122 | } | |
075663d1 SB |
123 | |
124 | if (WARN_ON(!cpu_hotplug.refcount)) | |
125 | cpu_hotplug.refcount++; /* try to fix things up */ | |
126 | ||
d2ba7e2a ON |
127 | if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) |
128 | wake_up_process(cpu_hotplug.active_writer); | |
d221938c | 129 | mutex_unlock(&cpu_hotplug.lock); |
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 | { |
d221938c | 159 | cpu_hotplug.active_writer = current; |
d2ba7e2a | 160 | |
a19423b9 | 161 | cpuhp_lock_acquire(); |
d2ba7e2a ON |
162 | for (;;) { |
163 | mutex_lock(&cpu_hotplug.lock); | |
b2c4623d PM |
164 | if (atomic_read(&cpu_hotplug.puts_pending)) { |
165 | int delta; | |
166 | ||
167 | delta = atomic_xchg(&cpu_hotplug.puts_pending, 0); | |
168 | cpu_hotplug.refcount -= delta; | |
169 | } | |
d2ba7e2a ON |
170 | if (likely(!cpu_hotplug.refcount)) |
171 | break; | |
172 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
d221938c GS |
173 | mutex_unlock(&cpu_hotplug.lock); |
174 | schedule(); | |
d221938c | 175 | } |
d221938c GS |
176 | } |
177 | ||
b9d10be7 | 178 | void cpu_hotplug_done(void) |
d221938c GS |
179 | { |
180 | cpu_hotplug.active_writer = NULL; | |
181 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 182 | cpuhp_lock_release(); |
d221938c | 183 | } |
79a6cdeb | 184 | |
16e53dbf SB |
185 | /* |
186 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
187 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
188 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
189 | * hotplug path before performing hotplug operations. So acquiring that lock | |
190 | * guarantees mutual exclusion from any currently running hotplug operations. | |
191 | */ | |
192 | void cpu_hotplug_disable(void) | |
193 | { | |
194 | cpu_maps_update_begin(); | |
195 | cpu_hotplug_disabled = 1; | |
196 | cpu_maps_update_done(); | |
197 | } | |
198 | ||
199 | void cpu_hotplug_enable(void) | |
200 | { | |
201 | cpu_maps_update_begin(); | |
202 | cpu_hotplug_disabled = 0; | |
203 | cpu_maps_update_done(); | |
204 | } | |
205 | ||
b9d10be7 | 206 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 207 | |
1da177e4 | 208 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 209 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 210 | { |
bd5349cf | 211 | int ret; |
d221938c | 212 | cpu_maps_update_begin(); |
bd5349cf | 213 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 214 | cpu_maps_update_done(); |
bd5349cf | 215 | return ret; |
1da177e4 | 216 | } |
65edc68c | 217 | |
93ae4f97 SB |
218 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
219 | { | |
220 | return raw_notifier_chain_register(&cpu_chain, nb); | |
221 | } | |
222 | ||
e9fb7631 AM |
223 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
224 | int *nr_calls) | |
225 | { | |
e6bde73b AM |
226 | int ret; |
227 | ||
228 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 229 | nr_calls); |
e6bde73b AM |
230 | |
231 | return notifier_to_errno(ret); | |
e9fb7631 AM |
232 | } |
233 | ||
234 | static int cpu_notify(unsigned long val, void *v) | |
235 | { | |
236 | return __cpu_notify(val, v, -1, NULL); | |
237 | } | |
238 | ||
00b9b0af LT |
239 | #ifdef CONFIG_HOTPLUG_CPU |
240 | ||
e9fb7631 AM |
241 | static void cpu_notify_nofail(unsigned long val, void *v) |
242 | { | |
00b9b0af | 243 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 244 | } |
1da177e4 | 245 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 246 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 247 | |
9647155f | 248 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 249 | { |
d221938c | 250 | cpu_maps_update_begin(); |
bd5349cf | 251 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 252 | cpu_maps_update_done(); |
1da177e4 LT |
253 | } |
254 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
255 | ||
93ae4f97 SB |
256 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
257 | { | |
258 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
259 | } | |
260 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
261 | ||
e4cc2f87 AV |
262 | /** |
263 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
264 | * @cpu: a CPU id | |
265 | * | |
266 | * This function walks all processes, finds a valid mm struct for each one and | |
267 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
268 | * trivial, there are various non-obvious corner cases, which this function | |
269 | * tries to solve in a safe manner. | |
270 | * | |
271 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
272 | * be called only for an already offlined CPU. | |
273 | */ | |
cb79295e AV |
274 | void clear_tasks_mm_cpumask(int cpu) |
275 | { | |
276 | struct task_struct *p; | |
277 | ||
278 | /* | |
279 | * This function is called after the cpu is taken down and marked | |
280 | * offline, so its not like new tasks will ever get this cpu set in | |
281 | * their mm mask. -- Peter Zijlstra | |
282 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
283 | * full-fledged tasklist_lock. | |
284 | */ | |
e4cc2f87 | 285 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
286 | rcu_read_lock(); |
287 | for_each_process(p) { | |
288 | struct task_struct *t; | |
289 | ||
e4cc2f87 AV |
290 | /* |
291 | * Main thread might exit, but other threads may still have | |
292 | * a valid mm. Find one. | |
293 | */ | |
cb79295e AV |
294 | t = find_lock_task_mm(p); |
295 | if (!t) | |
296 | continue; | |
297 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
298 | task_unlock(t); | |
299 | } | |
300 | rcu_read_unlock(); | |
301 | } | |
302 | ||
b728ca06 | 303 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 304 | { |
b728ca06 | 305 | struct task_struct *g, *p; |
1da177e4 | 306 | |
b728ca06 KT |
307 | read_lock_irq(&tasklist_lock); |
308 | do_each_thread(g, p) { | |
309 | if (!p->on_rq) | |
310 | continue; | |
311 | /* | |
312 | * We do the check with unlocked task_rq(p)->lock. | |
313 | * Order the reading to do not warn about a task, | |
314 | * which was running on this cpu in the past, and | |
315 | * it's just been woken on another cpu. | |
316 | */ | |
317 | rmb(); | |
318 | if (task_cpu(p) != dead_cpu) | |
319 | continue; | |
320 | ||
321 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
322 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
323 | } while_each_thread(g, p); | |
324 | read_unlock_irq(&tasklist_lock); | |
1da177e4 LT |
325 | } |
326 | ||
db912f96 AK |
327 | struct take_cpu_down_param { |
328 | unsigned long mod; | |
329 | void *hcpu; | |
330 | }; | |
331 | ||
1da177e4 | 332 | /* Take this CPU down. */ |
514a20a5 | 333 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 334 | { |
db912f96 | 335 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
336 | int err; |
337 | ||
1da177e4 LT |
338 | /* Ensure this CPU doesn't handle any more interrupts. */ |
339 | err = __cpu_disable(); | |
340 | if (err < 0) | |
f3705136 | 341 | return err; |
1da177e4 | 342 | |
e9fb7631 | 343 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
14e568e7 TG |
344 | /* Park the stopper thread */ |
345 | kthread_park(current); | |
f3705136 | 346 | return 0; |
1da177e4 LT |
347 | } |
348 | ||
e3920fb4 | 349 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 350 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 351 | { |
e7407dcc | 352 | int err, nr_calls = 0; |
e7407dcc | 353 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 354 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
355 | struct take_cpu_down_param tcd_param = { |
356 | .mod = mod, | |
357 | .hcpu = hcpu, | |
358 | }; | |
1da177e4 | 359 | |
e3920fb4 RW |
360 | if (num_online_cpus() == 1) |
361 | return -EBUSY; | |
1da177e4 | 362 | |
e3920fb4 RW |
363 | if (!cpu_online(cpu)) |
364 | return -EINVAL; | |
1da177e4 | 365 | |
d221938c | 366 | cpu_hotplug_begin(); |
4d51985e | 367 | |
e9fb7631 | 368 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 369 | if (err) { |
a0d8cdb6 | 370 | nr_calls--; |
e9fb7631 | 371 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
372 | pr_warn("%s: attempt to take down CPU %u failed\n", |
373 | __func__, cpu); | |
baaca49f | 374 | goto out_release; |
1da177e4 LT |
375 | } |
376 | ||
6acce3ef PZ |
377 | /* |
378 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
379 | * and RCU users of this state to go away such that all new such users | |
380 | * will observe it. | |
381 | * | |
382 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
383 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
384 | * |
385 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
386 | */ |
387 | #ifdef CONFIG_PREEMPT | |
388 | synchronize_sched(); | |
389 | #endif | |
390 | synchronize_rcu(); | |
391 | ||
106dd5af M |
392 | smpboot_park_threads(cpu); |
393 | ||
6acce3ef PZ |
394 | /* |
395 | * So now all preempt/rcu users must observe !cpu_active(). | |
396 | */ | |
397 | ||
e0b582ec | 398 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 399 | if (err) { |
1da177e4 | 400 | /* CPU didn't die: tell everyone. Can't complain. */ |
f97f8f06 | 401 | smpboot_unpark_threads(cpu); |
e9fb7631 | 402 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 403 | goto out_release; |
8fa1d7d3 | 404 | } |
04321587 | 405 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 406 | |
48c5ccae PZ |
407 | /* |
408 | * The migration_call() CPU_DYING callback will have removed all | |
409 | * runnable tasks from the cpu, there's only the idle task left now | |
410 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
411 | * |
412 | * Wait for the stop thread to go away. | |
48c5ccae | 413 | */ |
51a96c77 PZ |
414 | while (!idle_cpu(cpu)) |
415 | cpu_relax(); | |
1da177e4 LT |
416 | |
417 | /* This actually kills the CPU. */ | |
418 | __cpu_die(cpu); | |
419 | ||
1da177e4 | 420 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 421 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
422 | |
423 | check_for_tasks(cpu); | |
424 | ||
baaca49f | 425 | out_release: |
d221938c | 426 | cpu_hotplug_done(); |
e9fb7631 AM |
427 | if (!err) |
428 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
429 | return err; |
430 | } | |
431 | ||
514a20a5 | 432 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 433 | { |
9ea09af3 | 434 | int err; |
e3920fb4 | 435 | |
d221938c | 436 | cpu_maps_update_begin(); |
e761b772 MK |
437 | |
438 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 439 | err = -EBUSY; |
e761b772 MK |
440 | goto out; |
441 | } | |
442 | ||
e761b772 | 443 | err = _cpu_down(cpu, 0); |
e3920fb4 | 444 | |
e761b772 | 445 | out: |
d221938c | 446 | cpu_maps_update_done(); |
1da177e4 LT |
447 | return err; |
448 | } | |
b62b8ef9 | 449 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
450 | #endif /*CONFIG_HOTPLUG_CPU*/ |
451 | ||
e3920fb4 | 452 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 453 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 454 | { |
baaca49f | 455 | int ret, nr_calls = 0; |
1da177e4 | 456 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 457 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 458 | struct task_struct *idle; |
1da177e4 | 459 | |
d221938c | 460 | cpu_hotplug_begin(); |
38498a67 | 461 | |
5e5041f3 YI |
462 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
463 | ret = -EINVAL; | |
464 | goto out; | |
465 | } | |
466 | ||
3bb5d2ee SS |
467 | idle = idle_thread_get(cpu); |
468 | if (IS_ERR(idle)) { | |
469 | ret = PTR_ERR(idle); | |
38498a67 | 470 | goto out; |
3bb5d2ee | 471 | } |
38498a67 | 472 | |
f97f8f06 TG |
473 | ret = smpboot_create_threads(cpu); |
474 | if (ret) | |
475 | goto out; | |
476 | ||
e9fb7631 | 477 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 478 | if (ret) { |
a0d8cdb6 | 479 | nr_calls--; |
84117da5 FF |
480 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
481 | __func__, cpu); | |
1da177e4 LT |
482 | goto out_notify; |
483 | } | |
484 | ||
485 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 486 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
487 | if (ret != 0) |
488 | goto out_notify; | |
6978c705 | 489 | BUG_ON(!cpu_online(cpu)); |
1da177e4 | 490 | |
f97f8f06 TG |
491 | /* Wake the per cpu threads */ |
492 | smpboot_unpark_threads(cpu); | |
493 | ||
1da177e4 | 494 | /* Now call notifier in preparation. */ |
e9fb7631 | 495 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
496 | |
497 | out_notify: | |
498 | if (ret != 0) | |
e9fb7631 | 499 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 500 | out: |
d221938c | 501 | cpu_hotplug_done(); |
e3920fb4 RW |
502 | |
503 | return ret; | |
504 | } | |
505 | ||
0db0628d | 506 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
507 | { |
508 | int err = 0; | |
cf23422b | 509 | |
e0b582ec | 510 | if (!cpu_possible(cpu)) { |
84117da5 FF |
511 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
512 | cpu); | |
87d5e023 | 513 | #if defined(CONFIG_IA64) |
84117da5 | 514 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
515 | #endif |
516 | return -EINVAL; | |
517 | } | |
e3920fb4 | 518 | |
01b0f197 TK |
519 | err = try_online_node(cpu_to_node(cpu)); |
520 | if (err) | |
521 | return err; | |
cf23422b | 522 | |
d221938c | 523 | cpu_maps_update_begin(); |
e761b772 MK |
524 | |
525 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 526 | err = -EBUSY; |
e761b772 MK |
527 | goto out; |
528 | } | |
529 | ||
530 | err = _cpu_up(cpu, 0); | |
531 | ||
e761b772 | 532 | out: |
d221938c | 533 | cpu_maps_update_done(); |
e3920fb4 RW |
534 | return err; |
535 | } | |
a513f6ba | 536 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 537 | |
f3de4be9 | 538 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 539 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
540 | |
541 | int disable_nonboot_cpus(void) | |
542 | { | |
e9a5f426 | 543 | int cpu, first_cpu, error = 0; |
e3920fb4 | 544 | |
d221938c | 545 | cpu_maps_update_begin(); |
e0b582ec | 546 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
547 | /* |
548 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
549 | * with the userspace trying to use the CPU hotplug at the same time |
550 | */ | |
e0b582ec | 551 | cpumask_clear(frozen_cpus); |
6ad4c188 | 552 | |
84117da5 | 553 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
554 | for_each_online_cpu(cpu) { |
555 | if (cpu == first_cpu) | |
556 | continue; | |
bb3632c6 | 557 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 558 | error = _cpu_down(cpu, 1); |
bb3632c6 | 559 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 560 | if (!error) |
e0b582ec | 561 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 562 | else { |
84117da5 | 563 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
564 | break; |
565 | } | |
566 | } | |
86886e55 | 567 | |
e3920fb4 RW |
568 | if (!error) { |
569 | BUG_ON(num_online_cpus() > 1); | |
570 | /* Make sure the CPUs won't be enabled by someone else */ | |
571 | cpu_hotplug_disabled = 1; | |
572 | } else { | |
84117da5 | 573 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 574 | } |
d221938c | 575 | cpu_maps_update_done(); |
e3920fb4 RW |
576 | return error; |
577 | } | |
578 | ||
d0af9eed SS |
579 | void __weak arch_enable_nonboot_cpus_begin(void) |
580 | { | |
581 | } | |
582 | ||
583 | void __weak arch_enable_nonboot_cpus_end(void) | |
584 | { | |
585 | } | |
586 | ||
fa7303e2 | 587 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
588 | { |
589 | int cpu, error; | |
590 | ||
591 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 592 | cpu_maps_update_begin(); |
e3920fb4 | 593 | cpu_hotplug_disabled = 0; |
e0b582ec | 594 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 595 | goto out; |
e3920fb4 | 596 | |
84117da5 | 597 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
598 | |
599 | arch_enable_nonboot_cpus_begin(); | |
600 | ||
e0b582ec | 601 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 602 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 603 | error = _cpu_up(cpu, 1); |
bb3632c6 | 604 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 605 | if (!error) { |
84117da5 | 606 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
607 | continue; |
608 | } | |
84117da5 | 609 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 610 | } |
d0af9eed SS |
611 | |
612 | arch_enable_nonboot_cpus_end(); | |
613 | ||
e0b582ec | 614 | cpumask_clear(frozen_cpus); |
1d64b9cb | 615 | out: |
d221938c | 616 | cpu_maps_update_done(); |
1da177e4 | 617 | } |
e0b582ec | 618 | |
d7268a31 | 619 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
620 | { |
621 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
622 | return -ENOMEM; | |
623 | return 0; | |
624 | } | |
625 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 626 | |
79cfbdfa SB |
627 | /* |
628 | * When callbacks for CPU hotplug notifications are being executed, we must | |
629 | * ensure that the state of the system with respect to the tasks being frozen | |
630 | * or not, as reported by the notification, remains unchanged *throughout the | |
631 | * duration* of the execution of the callbacks. | |
632 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
633 | * | |
634 | * This synchronization is implemented by mutually excluding regular CPU | |
635 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
636 | * Hibernate notifications. | |
637 | */ | |
638 | static int | |
639 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
640 | unsigned long action, void *ptr) | |
641 | { | |
642 | switch (action) { | |
643 | ||
644 | case PM_SUSPEND_PREPARE: | |
645 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 646 | cpu_hotplug_disable(); |
79cfbdfa SB |
647 | break; |
648 | ||
649 | case PM_POST_SUSPEND: | |
650 | case PM_POST_HIBERNATION: | |
16e53dbf | 651 | cpu_hotplug_enable(); |
79cfbdfa SB |
652 | break; |
653 | ||
654 | default: | |
655 | return NOTIFY_DONE; | |
656 | } | |
657 | ||
658 | return NOTIFY_OK; | |
659 | } | |
660 | ||
661 | ||
d7268a31 | 662 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 663 | { |
6e32d479 FY |
664 | /* |
665 | * cpu_hotplug_pm_callback has higher priority than x86 | |
666 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
667 | * to disable cpu hotplug to avoid cpu hotplug race. | |
668 | */ | |
79cfbdfa SB |
669 | pm_notifier(cpu_hotplug_pm_callback, 0); |
670 | return 0; | |
671 | } | |
672 | core_initcall(cpu_hotplug_pm_sync_init); | |
673 | ||
f3de4be9 | 674 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 675 | |
e545a614 MS |
676 | /** |
677 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
678 | * @cpu: cpu that just started | |
679 | * | |
680 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
681 | * It must be called by the arch code on the new cpu, before the new cpu | |
682 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
683 | */ | |
0db0628d | 684 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
685 | { |
686 | unsigned long val = CPU_STARTING; | |
687 | ||
688 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 689 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
690 | val = CPU_STARTING_FROZEN; |
691 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 692 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
693 | } |
694 | ||
68f4f1ec | 695 | #endif /* CONFIG_SMP */ |
b8d317d1 | 696 | |
e56b3bc7 LT |
697 | /* |
698 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
699 | * represents all NR_CPUS bits binary values of 1<<nr. | |
700 | * | |
e0b582ec | 701 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
702 | * mask value that has a single bit set only. |
703 | */ | |
b8d317d1 | 704 | |
e56b3bc7 | 705 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 706 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
707 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
708 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
709 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 710 | |
e56b3bc7 LT |
711 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
712 | ||
713 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
714 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
715 | #if BITS_PER_LONG > 32 | |
716 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
717 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
718 | #endif |
719 | }; | |
e56b3bc7 | 720 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
721 | |
722 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
723 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
724 | |
725 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
726 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
727 | = CPU_BITS_ALL; | |
728 | #else | |
729 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
730 | #endif | |
731 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
732 | EXPORT_SYMBOL(cpu_possible_mask); | |
733 | ||
734 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
735 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
736 | EXPORT_SYMBOL(cpu_online_mask); | |
737 | ||
738 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
739 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
740 | EXPORT_SYMBOL(cpu_present_mask); | |
741 | ||
742 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
743 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
744 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
745 | |
746 | void set_cpu_possible(unsigned int cpu, bool possible) | |
747 | { | |
748 | if (possible) | |
749 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
750 | else | |
751 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
752 | } | |
753 | ||
754 | void set_cpu_present(unsigned int cpu, bool present) | |
755 | { | |
756 | if (present) | |
757 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
758 | else | |
759 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
760 | } | |
761 | ||
762 | void set_cpu_online(unsigned int cpu, bool online) | |
763 | { | |
6acbfb96 | 764 | if (online) { |
3fa41520 | 765 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
766 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
767 | } else { | |
3fa41520 | 768 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 769 | } |
3fa41520 RR |
770 | } |
771 | ||
772 | void set_cpu_active(unsigned int cpu, bool active) | |
773 | { | |
774 | if (active) | |
775 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
776 | else | |
777 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
778 | } | |
779 | ||
780 | void init_cpu_present(const struct cpumask *src) | |
781 | { | |
782 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
783 | } | |
784 | ||
785 | void init_cpu_possible(const struct cpumask *src) | |
786 | { | |
787 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
788 | } | |
789 | ||
790 | void init_cpu_online(const struct cpumask *src) | |
791 | { | |
792 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
793 | } |