]>
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> | |
19a284b2 | 11 | #include <linux/sched/smt.h> |
1da177e4 LT |
12 | #include <linux/unistd.h> |
13 | #include <linux/cpu.h> | |
cb79295e AV |
14 | #include <linux/oom.h> |
15 | #include <linux/rcupdate.h> | |
9984de1a | 16 | #include <linux/export.h> |
e4cc2f87 | 17 | #include <linux/bug.h> |
1da177e4 LT |
18 | #include <linux/kthread.h> |
19 | #include <linux/stop_machine.h> | |
81615b62 | 20 | #include <linux/mutex.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
79cfbdfa | 22 | #include <linux/suspend.h> |
a19423b9 | 23 | #include <linux/lockdep.h> |
bb3632c6 | 24 | #include <trace/events/power.h> |
1da177e4 | 25 | |
38498a67 TG |
26 | #include "smpboot.h" |
27 | ||
98a79d6a | 28 | #ifdef CONFIG_SMP |
b3199c02 | 29 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 30 | static DEFINE_MUTEX(cpu_add_remove_lock); |
1da177e4 | 31 | |
79a6cdeb | 32 | /* |
93ae4f97 SB |
33 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
34 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
35 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
36 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
37 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
38 | */ |
39 | void cpu_maps_update_begin(void) | |
40 | { | |
41 | mutex_lock(&cpu_add_remove_lock); | |
42 | } | |
93ae4f97 | 43 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
44 | |
45 | void cpu_maps_update_done(void) | |
46 | { | |
47 | mutex_unlock(&cpu_add_remove_lock); | |
48 | } | |
93ae4f97 | 49 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 50 | |
5c113fbe | 51 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 52 | |
e3920fb4 RW |
53 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
54 | * Should always be manipulated under cpu_add_remove_lock | |
55 | */ | |
56 | static int cpu_hotplug_disabled; | |
57 | ||
79a6cdeb LJ |
58 | #ifdef CONFIG_HOTPLUG_CPU |
59 | ||
d221938c GS |
60 | static struct { |
61 | struct task_struct *active_writer; | |
62 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
63 | /* | |
64 | * Also blocks the new readers during | |
65 | * an ongoing cpu hotplug operation. | |
66 | */ | |
67 | int refcount; | |
a19423b9 GS |
68 | |
69 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
70 | struct lockdep_map dep_map; | |
71 | #endif | |
31950eb6 LT |
72 | } cpu_hotplug = { |
73 | .active_writer = NULL, | |
74 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), | |
75 | .refcount = 0, | |
a19423b9 GS |
76 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
77 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
78 | #endif | |
31950eb6 | 79 | }; |
d221938c | 80 | |
a19423b9 GS |
81 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
82 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
83 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) | |
84 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
85 | ||
86ef5c9a | 86 | void get_online_cpus(void) |
a9d9baa1 | 87 | { |
d221938c GS |
88 | might_sleep(); |
89 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 90 | return; |
a19423b9 | 91 | cpuhp_lock_acquire_read(); |
d221938c GS |
92 | mutex_lock(&cpu_hotplug.lock); |
93 | cpu_hotplug.refcount++; | |
94 | mutex_unlock(&cpu_hotplug.lock); | |
95 | ||
a9d9baa1 | 96 | } |
86ef5c9a | 97 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 98 | |
86ef5c9a | 99 | void put_online_cpus(void) |
a9d9baa1 | 100 | { |
d221938c | 101 | if (cpu_hotplug.active_writer == current) |
aa953877 | 102 | return; |
d221938c | 103 | mutex_lock(&cpu_hotplug.lock); |
075663d1 SB |
104 | |
105 | if (WARN_ON(!cpu_hotplug.refcount)) | |
106 | cpu_hotplug.refcount++; /* try to fix things up */ | |
107 | ||
d2ba7e2a ON |
108 | if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) |
109 | wake_up_process(cpu_hotplug.active_writer); | |
d221938c | 110 | mutex_unlock(&cpu_hotplug.lock); |
a19423b9 | 111 | cpuhp_lock_release(); |
d221938c | 112 | |
a9d9baa1 | 113 | } |
86ef5c9a | 114 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 115 | |
d221938c GS |
116 | /* |
117 | * This ensures that the hotplug operation can begin only when the | |
118 | * refcount goes to zero. | |
119 | * | |
120 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
121 | * will be blocked by the cpu_hotplug.lock | |
122 | * | |
d2ba7e2a ON |
123 | * Since cpu_hotplug_begin() is always called after invoking |
124 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
125 | * |
126 | * Note that theoretically, there is a possibility of a livelock: | |
127 | * - Refcount goes to zero, last reader wakes up the sleeping | |
128 | * writer. | |
129 | * - Last reader unlocks the cpu_hotplug.lock. | |
130 | * - A new reader arrives at this moment, bumps up the refcount. | |
131 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
132 | * non zero and goes to sleep again. | |
133 | * | |
134 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 135 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
136 | * |
137 | */ | |
b9d10be7 | 138 | void cpu_hotplug_begin(void) |
d221938c | 139 | { |
d221938c | 140 | cpu_hotplug.active_writer = current; |
d2ba7e2a | 141 | |
a19423b9 | 142 | cpuhp_lock_acquire(); |
d2ba7e2a ON |
143 | for (;;) { |
144 | mutex_lock(&cpu_hotplug.lock); | |
145 | if (likely(!cpu_hotplug.refcount)) | |
146 | break; | |
147 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
d221938c GS |
148 | mutex_unlock(&cpu_hotplug.lock); |
149 | schedule(); | |
d221938c | 150 | } |
d221938c GS |
151 | } |
152 | ||
b9d10be7 | 153 | void cpu_hotplug_done(void) |
d221938c GS |
154 | { |
155 | cpu_hotplug.active_writer = NULL; | |
156 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 157 | cpuhp_lock_release(); |
d221938c | 158 | } |
79a6cdeb | 159 | |
16e53dbf SB |
160 | /* |
161 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
162 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
163 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
164 | * hotplug path before performing hotplug operations. So acquiring that lock | |
165 | * guarantees mutual exclusion from any currently running hotplug operations. | |
166 | */ | |
167 | void cpu_hotplug_disable(void) | |
168 | { | |
169 | cpu_maps_update_begin(); | |
170 | cpu_hotplug_disabled = 1; | |
171 | cpu_maps_update_done(); | |
172 | } | |
173 | ||
174 | void cpu_hotplug_enable(void) | |
175 | { | |
176 | cpu_maps_update_begin(); | |
177 | cpu_hotplug_disabled = 0; | |
178 | cpu_maps_update_done(); | |
179 | } | |
180 | ||
b9d10be7 | 181 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 182 | |
19a284b2 TG |
183 | /* |
184 | * Architectures that need SMT-specific errata handling during SMT hotplug | |
185 | * should override this. | |
186 | */ | |
187 | void __weak arch_smt_update(void) { } | |
188 | ||
1da177e4 | 189 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 190 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 191 | { |
bd5349cf | 192 | int ret; |
d221938c | 193 | cpu_maps_update_begin(); |
bd5349cf | 194 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 195 | cpu_maps_update_done(); |
bd5349cf | 196 | return ret; |
1da177e4 | 197 | } |
65edc68c | 198 | |
93ae4f97 SB |
199 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
200 | { | |
201 | return raw_notifier_chain_register(&cpu_chain, nb); | |
202 | } | |
203 | ||
e9fb7631 AM |
204 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
205 | int *nr_calls) | |
206 | { | |
e6bde73b AM |
207 | int ret; |
208 | ||
209 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 210 | nr_calls); |
e6bde73b AM |
211 | |
212 | return notifier_to_errno(ret); | |
e9fb7631 AM |
213 | } |
214 | ||
215 | static int cpu_notify(unsigned long val, void *v) | |
216 | { | |
217 | return __cpu_notify(val, v, -1, NULL); | |
218 | } | |
219 | ||
1da177e4 | 220 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 221 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 222 | |
9647155f | 223 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 224 | { |
d221938c | 225 | cpu_maps_update_begin(); |
bd5349cf | 226 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 227 | cpu_maps_update_done(); |
1da177e4 LT |
228 | } |
229 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
230 | ||
93ae4f97 SB |
231 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
232 | { | |
233 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
234 | } | |
235 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
236 | ||
9ad4ebf4 MH |
237 | #ifdef CONFIG_HOTPLUG_CPU |
238 | ||
239 | static void cpu_notify_nofail(unsigned long val, void *v) | |
240 | { | |
241 | BUG_ON(cpu_notify(val, v)); | |
242 | } | |
243 | ||
e4cc2f87 AV |
244 | /** |
245 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
246 | * @cpu: a CPU id | |
247 | * | |
248 | * This function walks all processes, finds a valid mm struct for each one and | |
249 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
250 | * trivial, there are various non-obvious corner cases, which this function | |
251 | * tries to solve in a safe manner. | |
252 | * | |
253 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
254 | * be called only for an already offlined CPU. | |
255 | */ | |
cb79295e AV |
256 | void clear_tasks_mm_cpumask(int cpu) |
257 | { | |
258 | struct task_struct *p; | |
259 | ||
260 | /* | |
261 | * This function is called after the cpu is taken down and marked | |
262 | * offline, so its not like new tasks will ever get this cpu set in | |
263 | * their mm mask. -- Peter Zijlstra | |
264 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
265 | * full-fledged tasklist_lock. | |
266 | */ | |
e4cc2f87 | 267 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
268 | rcu_read_lock(); |
269 | for_each_process(p) { | |
270 | struct task_struct *t; | |
271 | ||
e4cc2f87 AV |
272 | /* |
273 | * Main thread might exit, but other threads may still have | |
274 | * a valid mm. Find one. | |
275 | */ | |
cb79295e AV |
276 | t = find_lock_task_mm(p); |
277 | if (!t) | |
278 | continue; | |
279 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
280 | task_unlock(t); | |
281 | } | |
282 | rcu_read_unlock(); | |
283 | } | |
284 | ||
1da177e4 LT |
285 | static inline void check_for_tasks(int cpu) |
286 | { | |
287 | struct task_struct *p; | |
6fac4829 | 288 | cputime_t utime, stime; |
1da177e4 LT |
289 | |
290 | write_lock_irq(&tasklist_lock); | |
291 | for_each_process(p) { | |
6fac4829 | 292 | task_cputime(p, &utime, &stime); |
11854247 | 293 | if (task_cpu(p) == cpu && p->state == TASK_RUNNING && |
6fac4829 | 294 | (utime || stime)) |
84117da5 | 295 | pr_warn("Task %s (pid = %d) is on cpu %d (state = %ld, flags = %x)\n", |
9d3cfc4c FP |
296 | p->comm, task_pid_nr(p), cpu, |
297 | p->state, p->flags); | |
1da177e4 LT |
298 | } |
299 | write_unlock_irq(&tasklist_lock); | |
300 | } | |
301 | ||
db912f96 AK |
302 | struct take_cpu_down_param { |
303 | unsigned long mod; | |
304 | void *hcpu; | |
305 | }; | |
306 | ||
1da177e4 | 307 | /* Take this CPU down. */ |
514a20a5 | 308 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 309 | { |
db912f96 | 310 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
311 | int err; |
312 | ||
1da177e4 LT |
313 | /* Ensure this CPU doesn't handle any more interrupts. */ |
314 | err = __cpu_disable(); | |
315 | if (err < 0) | |
f3705136 | 316 | return err; |
1da177e4 | 317 | |
e9fb7631 | 318 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
14e568e7 TG |
319 | /* Park the stopper thread */ |
320 | kthread_park(current); | |
f3705136 | 321 | return 0; |
1da177e4 LT |
322 | } |
323 | ||
e3920fb4 | 324 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 325 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 326 | { |
e7407dcc | 327 | int err, nr_calls = 0; |
e7407dcc | 328 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 329 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
330 | struct take_cpu_down_param tcd_param = { |
331 | .mod = mod, | |
332 | .hcpu = hcpu, | |
333 | }; | |
1da177e4 | 334 | |
e3920fb4 RW |
335 | if (num_online_cpus() == 1) |
336 | return -EBUSY; | |
1da177e4 | 337 | |
e3920fb4 RW |
338 | if (!cpu_online(cpu)) |
339 | return -EINVAL; | |
1da177e4 | 340 | |
d221938c | 341 | cpu_hotplug_begin(); |
4d51985e | 342 | |
e9fb7631 | 343 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 344 | if (err) { |
a0d8cdb6 | 345 | nr_calls--; |
e9fb7631 | 346 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
347 | pr_warn("%s: attempt to take down CPU %u failed\n", |
348 | __func__, cpu); | |
baaca49f | 349 | goto out_release; |
1da177e4 LT |
350 | } |
351 | ||
6acce3ef PZ |
352 | /* |
353 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
354 | * and RCU users of this state to go away such that all new such users | |
355 | * will observe it. | |
356 | * | |
357 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
358 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
359 | * |
360 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
361 | */ |
362 | #ifdef CONFIG_PREEMPT | |
363 | synchronize_sched(); | |
364 | #endif | |
365 | synchronize_rcu(); | |
366 | ||
106dd5af M |
367 | smpboot_park_threads(cpu); |
368 | ||
6acce3ef PZ |
369 | /* |
370 | * So now all preempt/rcu users must observe !cpu_active(). | |
371 | */ | |
372 | ||
e0b582ec | 373 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 374 | if (err) { |
1da177e4 | 375 | /* CPU didn't die: tell everyone. Can't complain. */ |
f97f8f06 | 376 | smpboot_unpark_threads(cpu); |
e9fb7631 | 377 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 378 | goto out_release; |
8fa1d7d3 | 379 | } |
04321587 | 380 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 381 | |
48c5ccae PZ |
382 | /* |
383 | * The migration_call() CPU_DYING callback will have removed all | |
384 | * runnable tasks from the cpu, there's only the idle task left now | |
385 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
386 | * |
387 | * Wait for the stop thread to go away. | |
48c5ccae | 388 | */ |
51a96c77 PZ |
389 | while (!idle_cpu(cpu)) |
390 | cpu_relax(); | |
1da177e4 LT |
391 | |
392 | /* This actually kills the CPU. */ | |
393 | __cpu_die(cpu); | |
394 | ||
1da177e4 | 395 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 396 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
397 | |
398 | check_for_tasks(cpu); | |
399 | ||
baaca49f | 400 | out_release: |
d221938c | 401 | cpu_hotplug_done(); |
e9fb7631 AM |
402 | if (!err) |
403 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
19a284b2 | 404 | arch_smt_update(); |
e3920fb4 RW |
405 | return err; |
406 | } | |
407 | ||
514a20a5 | 408 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 409 | { |
9ea09af3 | 410 | int err; |
e3920fb4 | 411 | |
d221938c | 412 | cpu_maps_update_begin(); |
e761b772 MK |
413 | |
414 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 415 | err = -EBUSY; |
e761b772 MK |
416 | goto out; |
417 | } | |
418 | ||
e761b772 | 419 | err = _cpu_down(cpu, 0); |
e3920fb4 | 420 | |
e761b772 | 421 | out: |
d221938c | 422 | cpu_maps_update_done(); |
1da177e4 LT |
423 | return err; |
424 | } | |
b62b8ef9 | 425 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
426 | #endif /*CONFIG_HOTPLUG_CPU*/ |
427 | ||
477a64e6 PM |
428 | /* |
429 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
430 | */ | |
431 | static int smpboot_thread_call(struct notifier_block *nfb, | |
432 | unsigned long action, void *hcpu) | |
433 | { | |
434 | int cpu = (long)hcpu; | |
435 | ||
436 | switch (action & ~CPU_TASKS_FROZEN) { | |
437 | ||
438 | case CPU_ONLINE: | |
439 | smpboot_unpark_threads(cpu); | |
440 | break; | |
441 | ||
442 | default: | |
443 | break; | |
444 | } | |
445 | ||
446 | return NOTIFY_OK; | |
447 | } | |
448 | ||
449 | static struct notifier_block smpboot_thread_notifier = { | |
450 | .notifier_call = smpboot_thread_call, | |
451 | .priority = CPU_PRI_SMPBOOT, | |
452 | }; | |
453 | ||
454 | void __cpuinit smpboot_thread_init(void) | |
455 | { | |
456 | register_cpu_notifier(&smpboot_thread_notifier); | |
457 | } | |
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 LT |
497 | |
498 | /* Now call notifier in preparation. */ | |
e9fb7631 | 499 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
500 | |
501 | out_notify: | |
502 | if (ret != 0) | |
e9fb7631 | 503 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 504 | out: |
d221938c | 505 | cpu_hotplug_done(); |
19a284b2 | 506 | arch_smt_update(); |
e3920fb4 RW |
507 | return ret; |
508 | } | |
509 | ||
0db0628d | 510 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
511 | { |
512 | int err = 0; | |
cf23422b | 513 | |
e0b582ec | 514 | if (!cpu_possible(cpu)) { |
84117da5 FF |
515 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
516 | cpu); | |
87d5e023 | 517 | #if defined(CONFIG_IA64) |
84117da5 | 518 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
519 | #endif |
520 | return -EINVAL; | |
521 | } | |
e3920fb4 | 522 | |
01b0f197 TK |
523 | err = try_online_node(cpu_to_node(cpu)); |
524 | if (err) | |
525 | return err; | |
cf23422b | 526 | |
d221938c | 527 | cpu_maps_update_begin(); |
e761b772 MK |
528 | |
529 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 530 | err = -EBUSY; |
e761b772 MK |
531 | goto out; |
532 | } | |
533 | ||
534 | err = _cpu_up(cpu, 0); | |
535 | ||
e761b772 | 536 | out: |
d221938c | 537 | cpu_maps_update_done(); |
e3920fb4 RW |
538 | return err; |
539 | } | |
a513f6ba | 540 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 541 | |
f3de4be9 | 542 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 543 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
544 | |
545 | int disable_nonboot_cpus(void) | |
546 | { | |
e9a5f426 | 547 | int cpu, first_cpu, error = 0; |
e3920fb4 | 548 | |
d221938c | 549 | cpu_maps_update_begin(); |
e0b582ec | 550 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
551 | /* |
552 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
553 | * with the userspace trying to use the CPU hotplug at the same time |
554 | */ | |
e0b582ec | 555 | cpumask_clear(frozen_cpus); |
6ad4c188 | 556 | |
84117da5 | 557 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
558 | for_each_online_cpu(cpu) { |
559 | if (cpu == first_cpu) | |
560 | continue; | |
bb3632c6 | 561 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 562 | error = _cpu_down(cpu, 1); |
bb3632c6 | 563 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 564 | if (!error) |
e0b582ec | 565 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 566 | else { |
84117da5 | 567 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
568 | break; |
569 | } | |
570 | } | |
86886e55 | 571 | |
e3920fb4 RW |
572 | if (!error) { |
573 | BUG_ON(num_online_cpus() > 1); | |
574 | /* Make sure the CPUs won't be enabled by someone else */ | |
575 | cpu_hotplug_disabled = 1; | |
576 | } else { | |
84117da5 | 577 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 578 | } |
d221938c | 579 | cpu_maps_update_done(); |
e3920fb4 RW |
580 | return error; |
581 | } | |
582 | ||
d0af9eed SS |
583 | void __weak arch_enable_nonboot_cpus_begin(void) |
584 | { | |
585 | } | |
586 | ||
587 | void __weak arch_enable_nonboot_cpus_end(void) | |
588 | { | |
589 | } | |
590 | ||
fa7303e2 | 591 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
592 | { |
593 | int cpu, error; | |
594 | ||
595 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 596 | cpu_maps_update_begin(); |
e3920fb4 | 597 | cpu_hotplug_disabled = 0; |
e0b582ec | 598 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 599 | goto out; |
e3920fb4 | 600 | |
84117da5 | 601 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
602 | |
603 | arch_enable_nonboot_cpus_begin(); | |
604 | ||
e0b582ec | 605 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 606 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 607 | error = _cpu_up(cpu, 1); |
bb3632c6 | 608 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 609 | if (!error) { |
84117da5 | 610 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
611 | continue; |
612 | } | |
84117da5 | 613 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 614 | } |
d0af9eed SS |
615 | |
616 | arch_enable_nonboot_cpus_end(); | |
617 | ||
e0b582ec | 618 | cpumask_clear(frozen_cpus); |
1d64b9cb | 619 | out: |
d221938c | 620 | cpu_maps_update_done(); |
1da177e4 | 621 | } |
e0b582ec | 622 | |
d7268a31 | 623 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
624 | { |
625 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
626 | return -ENOMEM; | |
627 | return 0; | |
628 | } | |
629 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 630 | |
79cfbdfa SB |
631 | /* |
632 | * When callbacks for CPU hotplug notifications are being executed, we must | |
633 | * ensure that the state of the system with respect to the tasks being frozen | |
634 | * or not, as reported by the notification, remains unchanged *throughout the | |
635 | * duration* of the execution of the callbacks. | |
636 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
637 | * | |
638 | * This synchronization is implemented by mutually excluding regular CPU | |
639 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
640 | * Hibernate notifications. | |
641 | */ | |
642 | static int | |
643 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
644 | unsigned long action, void *ptr) | |
645 | { | |
646 | switch (action) { | |
647 | ||
648 | case PM_SUSPEND_PREPARE: | |
649 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 650 | cpu_hotplug_disable(); |
79cfbdfa SB |
651 | break; |
652 | ||
653 | case PM_POST_SUSPEND: | |
654 | case PM_POST_HIBERNATION: | |
16e53dbf | 655 | cpu_hotplug_enable(); |
79cfbdfa SB |
656 | break; |
657 | ||
658 | default: | |
659 | return NOTIFY_DONE; | |
660 | } | |
661 | ||
662 | return NOTIFY_OK; | |
663 | } | |
664 | ||
665 | ||
d7268a31 | 666 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 667 | { |
6e32d479 FY |
668 | /* |
669 | * cpu_hotplug_pm_callback has higher priority than x86 | |
670 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
671 | * to disable cpu hotplug to avoid cpu hotplug race. | |
672 | */ | |
79cfbdfa SB |
673 | pm_notifier(cpu_hotplug_pm_callback, 0); |
674 | return 0; | |
675 | } | |
676 | core_initcall(cpu_hotplug_pm_sync_init); | |
677 | ||
f3de4be9 | 678 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 679 | |
e545a614 MS |
680 | /** |
681 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
682 | * @cpu: cpu that just started | |
683 | * | |
684 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
685 | * It must be called by the arch code on the new cpu, before the new cpu | |
686 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
687 | */ | |
0db0628d | 688 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
689 | { |
690 | unsigned long val = CPU_STARTING; | |
691 | ||
692 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 693 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
694 | val = CPU_STARTING_FROZEN; |
695 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 696 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
697 | } |
698 | ||
68f4f1ec | 699 | #endif /* CONFIG_SMP */ |
b8d317d1 | 700 | |
e56b3bc7 LT |
701 | /* |
702 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
703 | * represents all NR_CPUS bits binary values of 1<<nr. | |
704 | * | |
e0b582ec | 705 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
706 | * mask value that has a single bit set only. |
707 | */ | |
b8d317d1 | 708 | |
e56b3bc7 | 709 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 710 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
711 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
712 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
713 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 714 | |
e56b3bc7 LT |
715 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
716 | ||
717 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
718 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
719 | #if BITS_PER_LONG > 32 | |
720 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
721 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
722 | #endif |
723 | }; | |
e56b3bc7 | 724 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
725 | |
726 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
727 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
728 | |
729 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
730 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
731 | = CPU_BITS_ALL; | |
732 | #else | |
733 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
734 | #endif | |
735 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
736 | EXPORT_SYMBOL(cpu_possible_mask); | |
737 | ||
738 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
739 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
740 | EXPORT_SYMBOL(cpu_online_mask); | |
741 | ||
742 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
743 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
744 | EXPORT_SYMBOL(cpu_present_mask); | |
745 | ||
746 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
747 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
748 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
749 | |
750 | void set_cpu_possible(unsigned int cpu, bool possible) | |
751 | { | |
752 | if (possible) | |
753 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
754 | else | |
755 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
756 | } | |
757 | ||
758 | void set_cpu_present(unsigned int cpu, bool present) | |
759 | { | |
760 | if (present) | |
761 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
762 | else | |
763 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
764 | } | |
765 | ||
766 | void set_cpu_online(unsigned int cpu, bool online) | |
767 | { | |
6acbfb96 | 768 | if (online) { |
3fa41520 | 769 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
770 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
771 | } else { | |
3fa41520 | 772 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 773 | } |
3fa41520 RR |
774 | } |
775 | ||
776 | void set_cpu_active(unsigned int cpu, bool active) | |
777 | { | |
778 | if (active) | |
779 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
780 | else | |
781 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
782 | } | |
783 | ||
784 | void init_cpu_present(const struct cpumask *src) | |
785 | { | |
786 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
787 | } | |
788 | ||
789 | void init_cpu_possible(const struct cpumask *src) | |
790 | { | |
791 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
792 | } | |
793 | ||
794 | void init_cpu_online(const struct cpumask *src) | |
795 | { | |
796 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
797 | } |