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 | ||
62db99f4 PM |
89 | static void apply_puts_pending(int max) |
90 | { | |
91 | int delta; | |
92 | ||
93 | if (atomic_read(&cpu_hotplug.puts_pending) >= max) { | |
94 | delta = atomic_xchg(&cpu_hotplug.puts_pending, 0); | |
95 | cpu_hotplug.refcount -= delta; | |
96 | } | |
97 | } | |
98 | ||
86ef5c9a | 99 | void get_online_cpus(void) |
a9d9baa1 | 100 | { |
d221938c GS |
101 | might_sleep(); |
102 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 103 | return; |
a19423b9 | 104 | cpuhp_lock_acquire_read(); |
d221938c | 105 | mutex_lock(&cpu_hotplug.lock); |
62db99f4 | 106 | apply_puts_pending(65536); |
d221938c GS |
107 | cpu_hotplug.refcount++; |
108 | mutex_unlock(&cpu_hotplug.lock); | |
a9d9baa1 | 109 | } |
86ef5c9a | 110 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 111 | |
dd56af42 PM |
112 | bool try_get_online_cpus(void) |
113 | { | |
114 | if (cpu_hotplug.active_writer == current) | |
115 | return true; | |
116 | if (!mutex_trylock(&cpu_hotplug.lock)) | |
117 | return false; | |
118 | cpuhp_lock_acquire_tryread(); | |
62db99f4 | 119 | apply_puts_pending(65536); |
dd56af42 PM |
120 | cpu_hotplug.refcount++; |
121 | mutex_unlock(&cpu_hotplug.lock); | |
122 | return true; | |
123 | } | |
124 | EXPORT_SYMBOL_GPL(try_get_online_cpus); | |
125 | ||
86ef5c9a | 126 | void put_online_cpus(void) |
a9d9baa1 | 127 | { |
d221938c | 128 | if (cpu_hotplug.active_writer == current) |
aa953877 | 129 | return; |
b2c4623d PM |
130 | if (!mutex_trylock(&cpu_hotplug.lock)) { |
131 | atomic_inc(&cpu_hotplug.puts_pending); | |
132 | cpuhp_lock_release(); | |
133 | return; | |
134 | } | |
075663d1 SB |
135 | |
136 | if (WARN_ON(!cpu_hotplug.refcount)) | |
137 | cpu_hotplug.refcount++; /* try to fix things up */ | |
138 | ||
d2ba7e2a ON |
139 | if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) |
140 | wake_up_process(cpu_hotplug.active_writer); | |
d221938c | 141 | mutex_unlock(&cpu_hotplug.lock); |
a19423b9 | 142 | cpuhp_lock_release(); |
d221938c | 143 | |
a9d9baa1 | 144 | } |
86ef5c9a | 145 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 146 | |
d221938c GS |
147 | /* |
148 | * This ensures that the hotplug operation can begin only when the | |
149 | * refcount goes to zero. | |
150 | * | |
151 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
152 | * will be blocked by the cpu_hotplug.lock | |
153 | * | |
d2ba7e2a ON |
154 | * Since cpu_hotplug_begin() is always called after invoking |
155 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
156 | * |
157 | * Note that theoretically, there is a possibility of a livelock: | |
158 | * - Refcount goes to zero, last reader wakes up the sleeping | |
159 | * writer. | |
160 | * - Last reader unlocks the cpu_hotplug.lock. | |
161 | * - A new reader arrives at this moment, bumps up the refcount. | |
162 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
163 | * non zero and goes to sleep again. | |
164 | * | |
165 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 166 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
167 | * |
168 | */ | |
b9d10be7 | 169 | void cpu_hotplug_begin(void) |
d221938c | 170 | { |
d221938c | 171 | cpu_hotplug.active_writer = current; |
d2ba7e2a | 172 | |
a19423b9 | 173 | cpuhp_lock_acquire(); |
d2ba7e2a ON |
174 | for (;;) { |
175 | mutex_lock(&cpu_hotplug.lock); | |
62db99f4 | 176 | apply_puts_pending(1); |
d2ba7e2a ON |
177 | if (likely(!cpu_hotplug.refcount)) |
178 | break; | |
179 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
d221938c GS |
180 | mutex_unlock(&cpu_hotplug.lock); |
181 | schedule(); | |
d221938c | 182 | } |
d221938c GS |
183 | } |
184 | ||
b9d10be7 | 185 | void cpu_hotplug_done(void) |
d221938c GS |
186 | { |
187 | cpu_hotplug.active_writer = NULL; | |
188 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 189 | cpuhp_lock_release(); |
d221938c | 190 | } |
79a6cdeb | 191 | |
16e53dbf SB |
192 | /* |
193 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
194 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
195 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
196 | * hotplug path before performing hotplug operations. So acquiring that lock | |
197 | * guarantees mutual exclusion from any currently running hotplug operations. | |
198 | */ | |
199 | void cpu_hotplug_disable(void) | |
200 | { | |
201 | cpu_maps_update_begin(); | |
202 | cpu_hotplug_disabled = 1; | |
203 | cpu_maps_update_done(); | |
204 | } | |
205 | ||
206 | void cpu_hotplug_enable(void) | |
207 | { | |
208 | cpu_maps_update_begin(); | |
209 | cpu_hotplug_disabled = 0; | |
210 | cpu_maps_update_done(); | |
211 | } | |
212 | ||
b9d10be7 | 213 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 214 | |
1da177e4 | 215 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 216 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 217 | { |
bd5349cf | 218 | int ret; |
d221938c | 219 | cpu_maps_update_begin(); |
bd5349cf | 220 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 221 | cpu_maps_update_done(); |
bd5349cf | 222 | return ret; |
1da177e4 | 223 | } |
65edc68c | 224 | |
93ae4f97 SB |
225 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
226 | { | |
227 | return raw_notifier_chain_register(&cpu_chain, nb); | |
228 | } | |
229 | ||
e9fb7631 AM |
230 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
231 | int *nr_calls) | |
232 | { | |
e6bde73b AM |
233 | int ret; |
234 | ||
235 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 236 | nr_calls); |
e6bde73b AM |
237 | |
238 | return notifier_to_errno(ret); | |
e9fb7631 AM |
239 | } |
240 | ||
241 | static int cpu_notify(unsigned long val, void *v) | |
242 | { | |
243 | return __cpu_notify(val, v, -1, NULL); | |
244 | } | |
245 | ||
00b9b0af LT |
246 | #ifdef CONFIG_HOTPLUG_CPU |
247 | ||
e9fb7631 AM |
248 | static void cpu_notify_nofail(unsigned long val, void *v) |
249 | { | |
00b9b0af | 250 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 251 | } |
1da177e4 | 252 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 253 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 254 | |
9647155f | 255 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 256 | { |
d221938c | 257 | cpu_maps_update_begin(); |
bd5349cf | 258 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 259 | cpu_maps_update_done(); |
1da177e4 LT |
260 | } |
261 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
262 | ||
93ae4f97 SB |
263 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
264 | { | |
265 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
266 | } | |
267 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
268 | ||
e4cc2f87 AV |
269 | /** |
270 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
271 | * @cpu: a CPU id | |
272 | * | |
273 | * This function walks all processes, finds a valid mm struct for each one and | |
274 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
275 | * trivial, there are various non-obvious corner cases, which this function | |
276 | * tries to solve in a safe manner. | |
277 | * | |
278 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
279 | * be called only for an already offlined CPU. | |
280 | */ | |
cb79295e AV |
281 | void clear_tasks_mm_cpumask(int cpu) |
282 | { | |
283 | struct task_struct *p; | |
284 | ||
285 | /* | |
286 | * This function is called after the cpu is taken down and marked | |
287 | * offline, so its not like new tasks will ever get this cpu set in | |
288 | * their mm mask. -- Peter Zijlstra | |
289 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
290 | * full-fledged tasklist_lock. | |
291 | */ | |
e4cc2f87 | 292 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
293 | rcu_read_lock(); |
294 | for_each_process(p) { | |
295 | struct task_struct *t; | |
296 | ||
e4cc2f87 AV |
297 | /* |
298 | * Main thread might exit, but other threads may still have | |
299 | * a valid mm. Find one. | |
300 | */ | |
cb79295e AV |
301 | t = find_lock_task_mm(p); |
302 | if (!t) | |
303 | continue; | |
304 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
305 | task_unlock(t); | |
306 | } | |
307 | rcu_read_unlock(); | |
308 | } | |
309 | ||
b728ca06 | 310 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 311 | { |
b728ca06 | 312 | struct task_struct *g, *p; |
1da177e4 | 313 | |
b728ca06 KT |
314 | read_lock_irq(&tasklist_lock); |
315 | do_each_thread(g, p) { | |
316 | if (!p->on_rq) | |
317 | continue; | |
318 | /* | |
319 | * We do the check with unlocked task_rq(p)->lock. | |
320 | * Order the reading to do not warn about a task, | |
321 | * which was running on this cpu in the past, and | |
322 | * it's just been woken on another cpu. | |
323 | */ | |
324 | rmb(); | |
325 | if (task_cpu(p) != dead_cpu) | |
326 | continue; | |
327 | ||
328 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
329 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
330 | } while_each_thread(g, p); | |
331 | read_unlock_irq(&tasklist_lock); | |
1da177e4 LT |
332 | } |
333 | ||
db912f96 AK |
334 | struct take_cpu_down_param { |
335 | unsigned long mod; | |
336 | void *hcpu; | |
337 | }; | |
338 | ||
1da177e4 | 339 | /* Take this CPU down. */ |
514a20a5 | 340 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 341 | { |
db912f96 | 342 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
343 | int err; |
344 | ||
1da177e4 LT |
345 | /* Ensure this CPU doesn't handle any more interrupts. */ |
346 | err = __cpu_disable(); | |
347 | if (err < 0) | |
f3705136 | 348 | return err; |
1da177e4 | 349 | |
e9fb7631 | 350 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
14e568e7 TG |
351 | /* Park the stopper thread */ |
352 | kthread_park(current); | |
f3705136 | 353 | return 0; |
1da177e4 LT |
354 | } |
355 | ||
e3920fb4 | 356 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 357 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 358 | { |
e7407dcc | 359 | int err, nr_calls = 0; |
e7407dcc | 360 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 361 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
362 | struct take_cpu_down_param tcd_param = { |
363 | .mod = mod, | |
364 | .hcpu = hcpu, | |
365 | }; | |
1da177e4 | 366 | |
e3920fb4 RW |
367 | if (num_online_cpus() == 1) |
368 | return -EBUSY; | |
1da177e4 | 369 | |
e3920fb4 RW |
370 | if (!cpu_online(cpu)) |
371 | return -EINVAL; | |
1da177e4 | 372 | |
d221938c | 373 | cpu_hotplug_begin(); |
4d51985e | 374 | |
e9fb7631 | 375 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 376 | if (err) { |
a0d8cdb6 | 377 | nr_calls--; |
e9fb7631 | 378 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
379 | pr_warn("%s: attempt to take down CPU %u failed\n", |
380 | __func__, cpu); | |
baaca49f | 381 | goto out_release; |
1da177e4 LT |
382 | } |
383 | ||
6acce3ef PZ |
384 | /* |
385 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
386 | * and RCU users of this state to go away such that all new such users | |
387 | * will observe it. | |
388 | * | |
389 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
390 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
391 | * |
392 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
393 | */ |
394 | #ifdef CONFIG_PREEMPT | |
395 | synchronize_sched(); | |
396 | #endif | |
397 | synchronize_rcu(); | |
398 | ||
106dd5af M |
399 | smpboot_park_threads(cpu); |
400 | ||
6acce3ef PZ |
401 | /* |
402 | * So now all preempt/rcu users must observe !cpu_active(). | |
403 | */ | |
404 | ||
e0b582ec | 405 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 406 | if (err) { |
1da177e4 | 407 | /* CPU didn't die: tell everyone. Can't complain. */ |
f97f8f06 | 408 | smpboot_unpark_threads(cpu); |
e9fb7631 | 409 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 410 | goto out_release; |
8fa1d7d3 | 411 | } |
04321587 | 412 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 413 | |
48c5ccae PZ |
414 | /* |
415 | * The migration_call() CPU_DYING callback will have removed all | |
416 | * runnable tasks from the cpu, there's only the idle task left now | |
417 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
418 | * |
419 | * Wait for the stop thread to go away. | |
48c5ccae | 420 | */ |
51a96c77 PZ |
421 | while (!idle_cpu(cpu)) |
422 | cpu_relax(); | |
1da177e4 LT |
423 | |
424 | /* This actually kills the CPU. */ | |
425 | __cpu_die(cpu); | |
426 | ||
1da177e4 | 427 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 428 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
429 | |
430 | check_for_tasks(cpu); | |
431 | ||
baaca49f | 432 | out_release: |
d221938c | 433 | cpu_hotplug_done(); |
e9fb7631 AM |
434 | if (!err) |
435 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
436 | return err; |
437 | } | |
438 | ||
514a20a5 | 439 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 440 | { |
9ea09af3 | 441 | int err; |
e3920fb4 | 442 | |
d221938c | 443 | cpu_maps_update_begin(); |
e761b772 MK |
444 | |
445 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 446 | err = -EBUSY; |
e761b772 MK |
447 | goto out; |
448 | } | |
449 | ||
e761b772 | 450 | err = _cpu_down(cpu, 0); |
e3920fb4 | 451 | |
e761b772 | 452 | out: |
d221938c | 453 | cpu_maps_update_done(); |
1da177e4 LT |
454 | return err; |
455 | } | |
b62b8ef9 | 456 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
457 | #endif /*CONFIG_HOTPLUG_CPU*/ |
458 | ||
e3920fb4 | 459 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 460 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 461 | { |
baaca49f | 462 | int ret, nr_calls = 0; |
1da177e4 | 463 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 464 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 465 | struct task_struct *idle; |
1da177e4 | 466 | |
d221938c | 467 | cpu_hotplug_begin(); |
38498a67 | 468 | |
5e5041f3 YI |
469 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
470 | ret = -EINVAL; | |
471 | goto out; | |
472 | } | |
473 | ||
3bb5d2ee SS |
474 | idle = idle_thread_get(cpu); |
475 | if (IS_ERR(idle)) { | |
476 | ret = PTR_ERR(idle); | |
38498a67 | 477 | goto out; |
3bb5d2ee | 478 | } |
38498a67 | 479 | |
f97f8f06 TG |
480 | ret = smpboot_create_threads(cpu); |
481 | if (ret) | |
482 | goto out; | |
483 | ||
e9fb7631 | 484 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 485 | if (ret) { |
a0d8cdb6 | 486 | nr_calls--; |
84117da5 FF |
487 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
488 | __func__, cpu); | |
1da177e4 LT |
489 | goto out_notify; |
490 | } | |
491 | ||
492 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 493 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
494 | if (ret != 0) |
495 | goto out_notify; | |
6978c705 | 496 | BUG_ON(!cpu_online(cpu)); |
1da177e4 | 497 | |
f97f8f06 TG |
498 | /* Wake the per cpu threads */ |
499 | smpboot_unpark_threads(cpu); | |
500 | ||
1da177e4 | 501 | /* Now call notifier in preparation. */ |
e9fb7631 | 502 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
503 | |
504 | out_notify: | |
505 | if (ret != 0) | |
e9fb7631 | 506 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 507 | out: |
d221938c | 508 | cpu_hotplug_done(); |
e3920fb4 RW |
509 | |
510 | return ret; | |
511 | } | |
512 | ||
0db0628d | 513 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
514 | { |
515 | int err = 0; | |
cf23422b | 516 | |
e0b582ec | 517 | if (!cpu_possible(cpu)) { |
84117da5 FF |
518 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
519 | cpu); | |
87d5e023 | 520 | #if defined(CONFIG_IA64) |
84117da5 | 521 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
522 | #endif |
523 | return -EINVAL; | |
524 | } | |
e3920fb4 | 525 | |
01b0f197 TK |
526 | err = try_online_node(cpu_to_node(cpu)); |
527 | if (err) | |
528 | return err; | |
cf23422b | 529 | |
d221938c | 530 | cpu_maps_update_begin(); |
e761b772 MK |
531 | |
532 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 533 | err = -EBUSY; |
e761b772 MK |
534 | goto out; |
535 | } | |
536 | ||
537 | err = _cpu_up(cpu, 0); | |
538 | ||
e761b772 | 539 | out: |
d221938c | 540 | cpu_maps_update_done(); |
e3920fb4 RW |
541 | return err; |
542 | } | |
a513f6ba | 543 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 544 | |
f3de4be9 | 545 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 546 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
547 | |
548 | int disable_nonboot_cpus(void) | |
549 | { | |
e9a5f426 | 550 | int cpu, first_cpu, error = 0; |
e3920fb4 | 551 | |
d221938c | 552 | cpu_maps_update_begin(); |
e0b582ec | 553 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
554 | /* |
555 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
556 | * with the userspace trying to use the CPU hotplug at the same time |
557 | */ | |
e0b582ec | 558 | cpumask_clear(frozen_cpus); |
6ad4c188 | 559 | |
84117da5 | 560 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
561 | for_each_online_cpu(cpu) { |
562 | if (cpu == first_cpu) | |
563 | continue; | |
bb3632c6 | 564 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 565 | error = _cpu_down(cpu, 1); |
bb3632c6 | 566 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 567 | if (!error) |
e0b582ec | 568 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 569 | else { |
84117da5 | 570 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
571 | break; |
572 | } | |
573 | } | |
86886e55 | 574 | |
e3920fb4 RW |
575 | if (!error) { |
576 | BUG_ON(num_online_cpus() > 1); | |
577 | /* Make sure the CPUs won't be enabled by someone else */ | |
578 | cpu_hotplug_disabled = 1; | |
579 | } else { | |
84117da5 | 580 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 581 | } |
d221938c | 582 | cpu_maps_update_done(); |
e3920fb4 RW |
583 | return error; |
584 | } | |
585 | ||
d0af9eed SS |
586 | void __weak arch_enable_nonboot_cpus_begin(void) |
587 | { | |
588 | } | |
589 | ||
590 | void __weak arch_enable_nonboot_cpus_end(void) | |
591 | { | |
592 | } | |
593 | ||
fa7303e2 | 594 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
595 | { |
596 | int cpu, error; | |
597 | ||
598 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 599 | cpu_maps_update_begin(); |
e3920fb4 | 600 | cpu_hotplug_disabled = 0; |
e0b582ec | 601 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 602 | goto out; |
e3920fb4 | 603 | |
84117da5 | 604 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
605 | |
606 | arch_enable_nonboot_cpus_begin(); | |
607 | ||
e0b582ec | 608 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 609 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 610 | error = _cpu_up(cpu, 1); |
bb3632c6 | 611 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 612 | if (!error) { |
84117da5 | 613 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
614 | continue; |
615 | } | |
84117da5 | 616 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 617 | } |
d0af9eed SS |
618 | |
619 | arch_enable_nonboot_cpus_end(); | |
620 | ||
e0b582ec | 621 | cpumask_clear(frozen_cpus); |
1d64b9cb | 622 | out: |
d221938c | 623 | cpu_maps_update_done(); |
1da177e4 | 624 | } |
e0b582ec | 625 | |
d7268a31 | 626 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
627 | { |
628 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
629 | return -ENOMEM; | |
630 | return 0; | |
631 | } | |
632 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 633 | |
79cfbdfa SB |
634 | /* |
635 | * When callbacks for CPU hotplug notifications are being executed, we must | |
636 | * ensure that the state of the system with respect to the tasks being frozen | |
637 | * or not, as reported by the notification, remains unchanged *throughout the | |
638 | * duration* of the execution of the callbacks. | |
639 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
640 | * | |
641 | * This synchronization is implemented by mutually excluding regular CPU | |
642 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
643 | * Hibernate notifications. | |
644 | */ | |
645 | static int | |
646 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
647 | unsigned long action, void *ptr) | |
648 | { | |
649 | switch (action) { | |
650 | ||
651 | case PM_SUSPEND_PREPARE: | |
652 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 653 | cpu_hotplug_disable(); |
79cfbdfa SB |
654 | break; |
655 | ||
656 | case PM_POST_SUSPEND: | |
657 | case PM_POST_HIBERNATION: | |
16e53dbf | 658 | cpu_hotplug_enable(); |
79cfbdfa SB |
659 | break; |
660 | ||
661 | default: | |
662 | return NOTIFY_DONE; | |
663 | } | |
664 | ||
665 | return NOTIFY_OK; | |
666 | } | |
667 | ||
668 | ||
d7268a31 | 669 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 670 | { |
6e32d479 FY |
671 | /* |
672 | * cpu_hotplug_pm_callback has higher priority than x86 | |
673 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
674 | * to disable cpu hotplug to avoid cpu hotplug race. | |
675 | */ | |
79cfbdfa SB |
676 | pm_notifier(cpu_hotplug_pm_callback, 0); |
677 | return 0; | |
678 | } | |
679 | core_initcall(cpu_hotplug_pm_sync_init); | |
680 | ||
f3de4be9 | 681 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 682 | |
e545a614 MS |
683 | /** |
684 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
685 | * @cpu: cpu that just started | |
686 | * | |
687 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
688 | * It must be called by the arch code on the new cpu, before the new cpu | |
689 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
690 | */ | |
0db0628d | 691 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
692 | { |
693 | unsigned long val = CPU_STARTING; | |
694 | ||
695 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 696 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
697 | val = CPU_STARTING_FROZEN; |
698 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 699 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
700 | } |
701 | ||
68f4f1ec | 702 | #endif /* CONFIG_SMP */ |
b8d317d1 | 703 | |
e56b3bc7 LT |
704 | /* |
705 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
706 | * represents all NR_CPUS bits binary values of 1<<nr. | |
707 | * | |
e0b582ec | 708 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
709 | * mask value that has a single bit set only. |
710 | */ | |
b8d317d1 | 711 | |
e56b3bc7 | 712 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 713 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
714 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
715 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
716 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 717 | |
e56b3bc7 LT |
718 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
719 | ||
720 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
721 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
722 | #if BITS_PER_LONG > 32 | |
723 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
724 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
725 | #endif |
726 | }; | |
e56b3bc7 | 727 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
728 | |
729 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
730 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
731 | |
732 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
733 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
734 | = CPU_BITS_ALL; | |
735 | #else | |
736 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
737 | #endif | |
738 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
739 | EXPORT_SYMBOL(cpu_possible_mask); | |
740 | ||
741 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
742 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
743 | EXPORT_SYMBOL(cpu_online_mask); | |
744 | ||
745 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
746 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
747 | EXPORT_SYMBOL(cpu_present_mask); | |
748 | ||
749 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
750 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
751 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
752 | |
753 | void set_cpu_possible(unsigned int cpu, bool possible) | |
754 | { | |
755 | if (possible) | |
756 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
757 | else | |
758 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
759 | } | |
760 | ||
761 | void set_cpu_present(unsigned int cpu, bool present) | |
762 | { | |
763 | if (present) | |
764 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
765 | else | |
766 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
767 | } | |
768 | ||
769 | void set_cpu_online(unsigned int cpu, bool online) | |
770 | { | |
6acbfb96 | 771 | if (online) { |
3fa41520 | 772 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
773 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
774 | } else { | |
3fa41520 | 775 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 776 | } |
3fa41520 RR |
777 | } |
778 | ||
779 | void set_cpu_active(unsigned int cpu, bool active) | |
780 | { | |
781 | if (active) | |
782 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
783 | else | |
784 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
785 | } | |
786 | ||
787 | void init_cpu_present(const struct cpumask *src) | |
788 | { | |
789 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
790 | } | |
791 | ||
792 | void init_cpu_possible(const struct cpumask *src) | |
793 | { | |
794 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
795 | } | |
796 | ||
797 | void init_cpu_online(const struct cpumask *src) | |
798 | { | |
799 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
800 | } |