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> |
1da177e4 LT |
16 | #include <linux/kthread.h> |
17 | #include <linux/stop_machine.h> | |
81615b62 | 18 | #include <linux/mutex.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
79cfbdfa | 20 | #include <linux/suspend.h> |
1da177e4 | 21 | |
38498a67 TG |
22 | #include "smpboot.h" |
23 | ||
98a79d6a | 24 | #ifdef CONFIG_SMP |
b3199c02 | 25 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 26 | static DEFINE_MUTEX(cpu_add_remove_lock); |
1da177e4 | 27 | |
79a6cdeb LJ |
28 | /* |
29 | * The following two API's must be used when attempting | |
30 | * to serialize the updates to cpu_online_mask, cpu_present_mask. | |
31 | */ | |
32 | void cpu_maps_update_begin(void) | |
33 | { | |
34 | mutex_lock(&cpu_add_remove_lock); | |
35 | } | |
36 | ||
37 | void cpu_maps_update_done(void) | |
38 | { | |
39 | mutex_unlock(&cpu_add_remove_lock); | |
40 | } | |
41 | ||
5c113fbe | 42 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 43 | |
e3920fb4 RW |
44 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
45 | * Should always be manipulated under cpu_add_remove_lock | |
46 | */ | |
47 | static int cpu_hotplug_disabled; | |
48 | ||
79a6cdeb LJ |
49 | #ifdef CONFIG_HOTPLUG_CPU |
50 | ||
d221938c GS |
51 | static struct { |
52 | struct task_struct *active_writer; | |
53 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
54 | /* | |
55 | * Also blocks the new readers during | |
56 | * an ongoing cpu hotplug operation. | |
57 | */ | |
58 | int refcount; | |
31950eb6 LT |
59 | } cpu_hotplug = { |
60 | .active_writer = NULL, | |
61 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), | |
62 | .refcount = 0, | |
63 | }; | |
d221938c | 64 | |
86ef5c9a | 65 | void get_online_cpus(void) |
a9d9baa1 | 66 | { |
d221938c GS |
67 | might_sleep(); |
68 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 69 | return; |
d221938c GS |
70 | mutex_lock(&cpu_hotplug.lock); |
71 | cpu_hotplug.refcount++; | |
72 | mutex_unlock(&cpu_hotplug.lock); | |
73 | ||
a9d9baa1 | 74 | } |
86ef5c9a | 75 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 76 | |
86ef5c9a | 77 | void put_online_cpus(void) |
a9d9baa1 | 78 | { |
d221938c | 79 | if (cpu_hotplug.active_writer == current) |
aa953877 | 80 | return; |
d221938c | 81 | mutex_lock(&cpu_hotplug.lock); |
d2ba7e2a ON |
82 | if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) |
83 | wake_up_process(cpu_hotplug.active_writer); | |
d221938c GS |
84 | mutex_unlock(&cpu_hotplug.lock); |
85 | ||
a9d9baa1 | 86 | } |
86ef5c9a | 87 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 88 | |
d221938c GS |
89 | /* |
90 | * This ensures that the hotplug operation can begin only when the | |
91 | * refcount goes to zero. | |
92 | * | |
93 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
94 | * will be blocked by the cpu_hotplug.lock | |
95 | * | |
d2ba7e2a ON |
96 | * Since cpu_hotplug_begin() is always called after invoking |
97 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
98 | * |
99 | * Note that theoretically, there is a possibility of a livelock: | |
100 | * - Refcount goes to zero, last reader wakes up the sleeping | |
101 | * writer. | |
102 | * - Last reader unlocks the cpu_hotplug.lock. | |
103 | * - A new reader arrives at this moment, bumps up the refcount. | |
104 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
105 | * non zero and goes to sleep again. | |
106 | * | |
107 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 108 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
109 | * |
110 | */ | |
111 | static void cpu_hotplug_begin(void) | |
112 | { | |
d221938c | 113 | cpu_hotplug.active_writer = current; |
d2ba7e2a ON |
114 | |
115 | for (;;) { | |
116 | mutex_lock(&cpu_hotplug.lock); | |
117 | if (likely(!cpu_hotplug.refcount)) | |
118 | break; | |
119 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
d221938c GS |
120 | mutex_unlock(&cpu_hotplug.lock); |
121 | schedule(); | |
d221938c | 122 | } |
d221938c GS |
123 | } |
124 | ||
125 | static void cpu_hotplug_done(void) | |
126 | { | |
127 | cpu_hotplug.active_writer = NULL; | |
128 | mutex_unlock(&cpu_hotplug.lock); | |
129 | } | |
79a6cdeb LJ |
130 | |
131 | #else /* #if CONFIG_HOTPLUG_CPU */ | |
132 | static void cpu_hotplug_begin(void) {} | |
133 | static void cpu_hotplug_done(void) {} | |
25985edc | 134 | #endif /* #else #if CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 135 | |
1da177e4 | 136 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 137 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 138 | { |
bd5349cf | 139 | int ret; |
d221938c | 140 | cpu_maps_update_begin(); |
bd5349cf | 141 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 142 | cpu_maps_update_done(); |
bd5349cf | 143 | return ret; |
1da177e4 | 144 | } |
65edc68c | 145 | |
e9fb7631 AM |
146 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
147 | int *nr_calls) | |
148 | { | |
e6bde73b AM |
149 | int ret; |
150 | ||
151 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 152 | nr_calls); |
e6bde73b AM |
153 | |
154 | return notifier_to_errno(ret); | |
e9fb7631 AM |
155 | } |
156 | ||
157 | static int cpu_notify(unsigned long val, void *v) | |
158 | { | |
159 | return __cpu_notify(val, v, -1, NULL); | |
160 | } | |
161 | ||
00b9b0af LT |
162 | #ifdef CONFIG_HOTPLUG_CPU |
163 | ||
e9fb7631 AM |
164 | static void cpu_notify_nofail(unsigned long val, void *v) |
165 | { | |
00b9b0af | 166 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 167 | } |
1da177e4 LT |
168 | EXPORT_SYMBOL(register_cpu_notifier); |
169 | ||
9647155f | 170 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 171 | { |
d221938c | 172 | cpu_maps_update_begin(); |
bd5349cf | 173 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 174 | cpu_maps_update_done(); |
1da177e4 LT |
175 | } |
176 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
177 | ||
cb79295e AV |
178 | void clear_tasks_mm_cpumask(int cpu) |
179 | { | |
180 | struct task_struct *p; | |
181 | ||
182 | /* | |
183 | * This function is called after the cpu is taken down and marked | |
184 | * offline, so its not like new tasks will ever get this cpu set in | |
185 | * their mm mask. -- Peter Zijlstra | |
186 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
187 | * full-fledged tasklist_lock. | |
188 | */ | |
189 | rcu_read_lock(); | |
190 | for_each_process(p) { | |
191 | struct task_struct *t; | |
192 | ||
193 | t = find_lock_task_mm(p); | |
194 | if (!t) | |
195 | continue; | |
196 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
197 | task_unlock(t); | |
198 | } | |
199 | rcu_read_unlock(); | |
200 | } | |
201 | ||
1da177e4 LT |
202 | static inline void check_for_tasks(int cpu) |
203 | { | |
204 | struct task_struct *p; | |
205 | ||
206 | write_lock_irq(&tasklist_lock); | |
207 | for_each_process(p) { | |
11854247 | 208 | if (task_cpu(p) == cpu && p->state == TASK_RUNNING && |
64861634 | 209 | (p->utime || p->stime)) |
9d3cfc4c FP |
210 | printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d " |
211 | "(state = %ld, flags = %x)\n", | |
212 | p->comm, task_pid_nr(p), cpu, | |
213 | p->state, p->flags); | |
1da177e4 LT |
214 | } |
215 | write_unlock_irq(&tasklist_lock); | |
216 | } | |
217 | ||
db912f96 AK |
218 | struct take_cpu_down_param { |
219 | unsigned long mod; | |
220 | void *hcpu; | |
221 | }; | |
222 | ||
1da177e4 | 223 | /* Take this CPU down. */ |
514a20a5 | 224 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 225 | { |
db912f96 | 226 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
227 | int err; |
228 | ||
1da177e4 LT |
229 | /* Ensure this CPU doesn't handle any more interrupts. */ |
230 | err = __cpu_disable(); | |
231 | if (err < 0) | |
f3705136 | 232 | return err; |
1da177e4 | 233 | |
e9fb7631 | 234 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
f3705136 | 235 | return 0; |
1da177e4 LT |
236 | } |
237 | ||
e3920fb4 | 238 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 239 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 240 | { |
e7407dcc | 241 | int err, nr_calls = 0; |
e7407dcc | 242 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 243 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
244 | struct take_cpu_down_param tcd_param = { |
245 | .mod = mod, | |
246 | .hcpu = hcpu, | |
247 | }; | |
1da177e4 | 248 | |
e3920fb4 RW |
249 | if (num_online_cpus() == 1) |
250 | return -EBUSY; | |
1da177e4 | 251 | |
e3920fb4 RW |
252 | if (!cpu_online(cpu)) |
253 | return -EINVAL; | |
1da177e4 | 254 | |
d221938c | 255 | cpu_hotplug_begin(); |
4d51985e | 256 | |
e9fb7631 | 257 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 258 | if (err) { |
a0d8cdb6 | 259 | nr_calls--; |
e9fb7631 | 260 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
1da177e4 | 261 | printk("%s: attempt to take down CPU %u failed\n", |
af1f16d0 | 262 | __func__, cpu); |
baaca49f | 263 | goto out_release; |
1da177e4 LT |
264 | } |
265 | ||
e0b582ec | 266 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 267 | if (err) { |
1da177e4 | 268 | /* CPU didn't die: tell everyone. Can't complain. */ |
e9fb7631 | 269 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
1da177e4 | 270 | |
6a1bdc1b | 271 | goto out_release; |
8fa1d7d3 | 272 | } |
04321587 | 273 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 274 | |
48c5ccae PZ |
275 | /* |
276 | * The migration_call() CPU_DYING callback will have removed all | |
277 | * runnable tasks from the cpu, there's only the idle task left now | |
278 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
279 | * |
280 | * Wait for the stop thread to go away. | |
48c5ccae | 281 | */ |
51a96c77 PZ |
282 | while (!idle_cpu(cpu)) |
283 | cpu_relax(); | |
1da177e4 LT |
284 | |
285 | /* This actually kills the CPU. */ | |
286 | __cpu_die(cpu); | |
287 | ||
1da177e4 | 288 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 289 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
290 | |
291 | check_for_tasks(cpu); | |
292 | ||
baaca49f | 293 | out_release: |
d221938c | 294 | cpu_hotplug_done(); |
e9fb7631 AM |
295 | if (!err) |
296 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
297 | return err; |
298 | } | |
299 | ||
514a20a5 | 300 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 301 | { |
9ea09af3 | 302 | int err; |
e3920fb4 | 303 | |
d221938c | 304 | cpu_maps_update_begin(); |
e761b772 MK |
305 | |
306 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 307 | err = -EBUSY; |
e761b772 MK |
308 | goto out; |
309 | } | |
310 | ||
e761b772 | 311 | err = _cpu_down(cpu, 0); |
e3920fb4 | 312 | |
e761b772 | 313 | out: |
d221938c | 314 | cpu_maps_update_done(); |
1da177e4 LT |
315 | return err; |
316 | } | |
b62b8ef9 | 317 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
318 | #endif /*CONFIG_HOTPLUG_CPU*/ |
319 | ||
e3920fb4 | 320 | /* Requires cpu_add_remove_lock to be held */ |
8bb78442 | 321 | static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 322 | { |
baaca49f | 323 | int ret, nr_calls = 0; |
1da177e4 | 324 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 325 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 326 | struct task_struct *idle; |
1da177e4 | 327 | |
e3920fb4 RW |
328 | if (cpu_online(cpu) || !cpu_present(cpu)) |
329 | return -EINVAL; | |
90d45d17 | 330 | |
d221938c | 331 | cpu_hotplug_begin(); |
38498a67 | 332 | |
3bb5d2ee SS |
333 | idle = idle_thread_get(cpu); |
334 | if (IS_ERR(idle)) { | |
335 | ret = PTR_ERR(idle); | |
38498a67 | 336 | goto out; |
3bb5d2ee | 337 | } |
38498a67 | 338 | |
e9fb7631 | 339 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 340 | if (ret) { |
a0d8cdb6 | 341 | nr_calls--; |
4d51985e | 342 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", |
af1f16d0 | 343 | __func__, cpu); |
1da177e4 LT |
344 | goto out_notify; |
345 | } | |
346 | ||
347 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 348 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
349 | if (ret != 0) |
350 | goto out_notify; | |
6978c705 | 351 | BUG_ON(!cpu_online(cpu)); |
1da177e4 LT |
352 | |
353 | /* Now call notifier in preparation. */ | |
e9fb7631 | 354 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
355 | |
356 | out_notify: | |
357 | if (ret != 0) | |
e9fb7631 | 358 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 359 | out: |
d221938c | 360 | cpu_hotplug_done(); |
e3920fb4 RW |
361 | |
362 | return ret; | |
363 | } | |
364 | ||
b282b6f8 | 365 | int __cpuinit cpu_up(unsigned int cpu) |
e3920fb4 RW |
366 | { |
367 | int err = 0; | |
cf23422b | 368 | |
369 | #ifdef CONFIG_MEMORY_HOTPLUG | |
370 | int nid; | |
371 | pg_data_t *pgdat; | |
372 | #endif | |
373 | ||
e0b582ec | 374 | if (!cpu_possible(cpu)) { |
73e753a5 KH |
375 | printk(KERN_ERR "can't online cpu %d because it is not " |
376 | "configured as may-hotadd at boot time\n", cpu); | |
87d5e023 | 377 | #if defined(CONFIG_IA64) |
73e753a5 KH |
378 | printk(KERN_ERR "please check additional_cpus= boot " |
379 | "parameter\n"); | |
380 | #endif | |
381 | return -EINVAL; | |
382 | } | |
e3920fb4 | 383 | |
cf23422b | 384 | #ifdef CONFIG_MEMORY_HOTPLUG |
385 | nid = cpu_to_node(cpu); | |
386 | if (!node_online(nid)) { | |
387 | err = mem_online_node(nid); | |
388 | if (err) | |
389 | return err; | |
390 | } | |
391 | ||
392 | pgdat = NODE_DATA(nid); | |
393 | if (!pgdat) { | |
394 | printk(KERN_ERR | |
395 | "Can't online cpu %d due to NULL pgdat\n", cpu); | |
396 | return -ENOMEM; | |
397 | } | |
398 | ||
4eaf3f64 HL |
399 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
400 | mutex_lock(&zonelists_mutex); | |
1f522509 | 401 | build_all_zonelists(NULL); |
4eaf3f64 HL |
402 | mutex_unlock(&zonelists_mutex); |
403 | } | |
cf23422b | 404 | #endif |
405 | ||
d221938c | 406 | cpu_maps_update_begin(); |
e761b772 MK |
407 | |
408 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 409 | err = -EBUSY; |
e761b772 MK |
410 | goto out; |
411 | } | |
412 | ||
413 | err = _cpu_up(cpu, 0); | |
414 | ||
e761b772 | 415 | out: |
d221938c | 416 | cpu_maps_update_done(); |
e3920fb4 RW |
417 | return err; |
418 | } | |
a513f6ba | 419 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 420 | |
f3de4be9 | 421 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 422 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 423 | |
3fb82d56 SS |
424 | void __weak arch_disable_nonboot_cpus_begin(void) |
425 | { | |
426 | } | |
427 | ||
428 | void __weak arch_disable_nonboot_cpus_end(void) | |
429 | { | |
430 | } | |
431 | ||
e3920fb4 RW |
432 | int disable_nonboot_cpus(void) |
433 | { | |
e9a5f426 | 434 | int cpu, first_cpu, error = 0; |
e3920fb4 | 435 | |
d221938c | 436 | cpu_maps_update_begin(); |
e0b582ec | 437 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
438 | /* |
439 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
440 | * with the userspace trying to use the CPU hotplug at the same time |
441 | */ | |
e0b582ec | 442 | cpumask_clear(frozen_cpus); |
3fb82d56 | 443 | arch_disable_nonboot_cpus_begin(); |
6ad4c188 | 444 | |
e3920fb4 RW |
445 | printk("Disabling non-boot CPUs ...\n"); |
446 | for_each_online_cpu(cpu) { | |
447 | if (cpu == first_cpu) | |
448 | continue; | |
8bb78442 | 449 | error = _cpu_down(cpu, 1); |
feae3203 | 450 | if (!error) |
e0b582ec | 451 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 452 | else { |
e3920fb4 RW |
453 | printk(KERN_ERR "Error taking CPU%d down: %d\n", |
454 | cpu, error); | |
455 | break; | |
456 | } | |
457 | } | |
86886e55 | 458 | |
3fb82d56 SS |
459 | arch_disable_nonboot_cpus_end(); |
460 | ||
e3920fb4 RW |
461 | if (!error) { |
462 | BUG_ON(num_online_cpus() > 1); | |
463 | /* Make sure the CPUs won't be enabled by someone else */ | |
464 | cpu_hotplug_disabled = 1; | |
465 | } else { | |
e1d9fd2e | 466 | printk(KERN_ERR "Non-boot CPUs are not disabled\n"); |
e3920fb4 | 467 | } |
d221938c | 468 | cpu_maps_update_done(); |
e3920fb4 RW |
469 | return error; |
470 | } | |
471 | ||
d0af9eed SS |
472 | void __weak arch_enable_nonboot_cpus_begin(void) |
473 | { | |
474 | } | |
475 | ||
476 | void __weak arch_enable_nonboot_cpus_end(void) | |
477 | { | |
478 | } | |
479 | ||
fa7303e2 | 480 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
481 | { |
482 | int cpu, error; | |
483 | ||
484 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 485 | cpu_maps_update_begin(); |
e3920fb4 | 486 | cpu_hotplug_disabled = 0; |
e0b582ec | 487 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 488 | goto out; |
e3920fb4 | 489 | |
4d51985e | 490 | printk(KERN_INFO "Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
491 | |
492 | arch_enable_nonboot_cpus_begin(); | |
493 | ||
e0b582ec | 494 | for_each_cpu(cpu, frozen_cpus) { |
8bb78442 | 495 | error = _cpu_up(cpu, 1); |
e3920fb4 | 496 | if (!error) { |
4d51985e | 497 | printk(KERN_INFO "CPU%d is up\n", cpu); |
e3920fb4 RW |
498 | continue; |
499 | } | |
1d64b9cb | 500 | printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 501 | } |
d0af9eed SS |
502 | |
503 | arch_enable_nonboot_cpus_end(); | |
504 | ||
e0b582ec | 505 | cpumask_clear(frozen_cpus); |
1d64b9cb | 506 | out: |
d221938c | 507 | cpu_maps_update_done(); |
1da177e4 | 508 | } |
e0b582ec | 509 | |
d7268a31 | 510 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
511 | { |
512 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
513 | return -ENOMEM; | |
514 | return 0; | |
515 | } | |
516 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa SB |
517 | |
518 | /* | |
519 | * Prevent regular CPU hotplug from racing with the freezer, by disabling CPU | |
520 | * hotplug when tasks are about to be frozen. Also, don't allow the freezer | |
521 | * to continue until any currently running CPU hotplug operation gets | |
522 | * completed. | |
523 | * To modify the 'cpu_hotplug_disabled' flag, we need to acquire the | |
524 | * 'cpu_add_remove_lock'. And this same lock is also taken by the regular | |
525 | * CPU hotplug path and released only after it is complete. Thus, we | |
526 | * (and hence the freezer) will block here until any currently running CPU | |
527 | * hotplug operation gets completed. | |
528 | */ | |
529 | void cpu_hotplug_disable_before_freeze(void) | |
530 | { | |
531 | cpu_maps_update_begin(); | |
532 | cpu_hotplug_disabled = 1; | |
533 | cpu_maps_update_done(); | |
534 | } | |
535 | ||
536 | ||
537 | /* | |
538 | * When tasks have been thawed, re-enable regular CPU hotplug (which had been | |
539 | * disabled while beginning to freeze tasks). | |
540 | */ | |
541 | void cpu_hotplug_enable_after_thaw(void) | |
542 | { | |
543 | cpu_maps_update_begin(); | |
544 | cpu_hotplug_disabled = 0; | |
545 | cpu_maps_update_done(); | |
546 | } | |
547 | ||
548 | /* | |
549 | * When callbacks for CPU hotplug notifications are being executed, we must | |
550 | * ensure that the state of the system with respect to the tasks being frozen | |
551 | * or not, as reported by the notification, remains unchanged *throughout the | |
552 | * duration* of the execution of the callbacks. | |
553 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
554 | * | |
555 | * This synchronization is implemented by mutually excluding regular CPU | |
556 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
557 | * Hibernate notifications. | |
558 | */ | |
559 | static int | |
560 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
561 | unsigned long action, void *ptr) | |
562 | { | |
563 | switch (action) { | |
564 | ||
565 | case PM_SUSPEND_PREPARE: | |
566 | case PM_HIBERNATION_PREPARE: | |
567 | cpu_hotplug_disable_before_freeze(); | |
568 | break; | |
569 | ||
570 | case PM_POST_SUSPEND: | |
571 | case PM_POST_HIBERNATION: | |
572 | cpu_hotplug_enable_after_thaw(); | |
573 | break; | |
574 | ||
575 | default: | |
576 | return NOTIFY_DONE; | |
577 | } | |
578 | ||
579 | return NOTIFY_OK; | |
580 | } | |
581 | ||
582 | ||
d7268a31 | 583 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa SB |
584 | { |
585 | pm_notifier(cpu_hotplug_pm_callback, 0); | |
586 | return 0; | |
587 | } | |
588 | core_initcall(cpu_hotplug_pm_sync_init); | |
589 | ||
f3de4be9 | 590 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 591 | |
e545a614 MS |
592 | /** |
593 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
594 | * @cpu: cpu that just started | |
595 | * | |
596 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
597 | * It must be called by the arch code on the new cpu, before the new cpu | |
598 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
599 | */ | |
84196414 | 600 | void __cpuinit notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
601 | { |
602 | unsigned long val = CPU_STARTING; | |
603 | ||
604 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 605 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
606 | val = CPU_STARTING_FROZEN; |
607 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 608 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
609 | } |
610 | ||
68f4f1ec | 611 | #endif /* CONFIG_SMP */ |
b8d317d1 | 612 | |
e56b3bc7 LT |
613 | /* |
614 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
615 | * represents all NR_CPUS bits binary values of 1<<nr. | |
616 | * | |
e0b582ec | 617 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
618 | * mask value that has a single bit set only. |
619 | */ | |
b8d317d1 | 620 | |
e56b3bc7 | 621 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 622 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
623 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
624 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
625 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 626 | |
e56b3bc7 LT |
627 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
628 | ||
629 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
630 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
631 | #if BITS_PER_LONG > 32 | |
632 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
633 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
634 | #endif |
635 | }; | |
e56b3bc7 | 636 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
637 | |
638 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
639 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
640 | |
641 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
642 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
643 | = CPU_BITS_ALL; | |
644 | #else | |
645 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
646 | #endif | |
647 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
648 | EXPORT_SYMBOL(cpu_possible_mask); | |
649 | ||
650 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
651 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
652 | EXPORT_SYMBOL(cpu_online_mask); | |
653 | ||
654 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
655 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
656 | EXPORT_SYMBOL(cpu_present_mask); | |
657 | ||
658 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
659 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
660 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
661 | |
662 | void set_cpu_possible(unsigned int cpu, bool possible) | |
663 | { | |
664 | if (possible) | |
665 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
666 | else | |
667 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
668 | } | |
669 | ||
670 | void set_cpu_present(unsigned int cpu, bool present) | |
671 | { | |
672 | if (present) | |
673 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
674 | else | |
675 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
676 | } | |
677 | ||
678 | void set_cpu_online(unsigned int cpu, bool online) | |
679 | { | |
680 | if (online) | |
681 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); | |
682 | else | |
683 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); | |
684 | } | |
685 | ||
686 | void set_cpu_active(unsigned int cpu, bool active) | |
687 | { | |
688 | if (active) | |
689 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
690 | else | |
691 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
692 | } | |
693 | ||
694 | void init_cpu_present(const struct cpumask *src) | |
695 | { | |
696 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
697 | } | |
698 | ||
699 | void init_cpu_possible(const struct cpumask *src) | |
700 | { | |
701 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
702 | } | |
703 | ||
704 | void init_cpu_online(const struct cpumask *src) | |
705 | { | |
706 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
707 | } |