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> |
1da177e4 | 22 | |
38498a67 TG |
23 | #include "smpboot.h" |
24 | ||
98a79d6a | 25 | #ifdef CONFIG_SMP |
b3199c02 | 26 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 27 | static DEFINE_MUTEX(cpu_add_remove_lock); |
1da177e4 | 28 | |
79a6cdeb LJ |
29 | /* |
30 | * The following two API's must be used when attempting | |
31 | * to serialize the updates to cpu_online_mask, cpu_present_mask. | |
32 | */ | |
33 | void cpu_maps_update_begin(void) | |
34 | { | |
35 | mutex_lock(&cpu_add_remove_lock); | |
36 | } | |
37 | ||
38 | void cpu_maps_update_done(void) | |
39 | { | |
40 | mutex_unlock(&cpu_add_remove_lock); | |
41 | } | |
42 | ||
5c113fbe | 43 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 44 | |
e3920fb4 RW |
45 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
46 | * Should always be manipulated under cpu_add_remove_lock | |
47 | */ | |
48 | static int cpu_hotplug_disabled; | |
49 | ||
79a6cdeb LJ |
50 | #ifdef CONFIG_HOTPLUG_CPU |
51 | ||
d221938c GS |
52 | static struct { |
53 | struct task_struct *active_writer; | |
54 | struct mutex lock; /* Synchronizes accesses to refcount, */ | |
55 | /* | |
56 | * Also blocks the new readers during | |
57 | * an ongoing cpu hotplug operation. | |
58 | */ | |
59 | int refcount; | |
31950eb6 LT |
60 | } cpu_hotplug = { |
61 | .active_writer = NULL, | |
62 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), | |
63 | .refcount = 0, | |
64 | }; | |
d221938c | 65 | |
86ef5c9a | 66 | void get_online_cpus(void) |
a9d9baa1 | 67 | { |
d221938c GS |
68 | might_sleep(); |
69 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 70 | return; |
d221938c GS |
71 | mutex_lock(&cpu_hotplug.lock); |
72 | cpu_hotplug.refcount++; | |
73 | mutex_unlock(&cpu_hotplug.lock); | |
74 | ||
a9d9baa1 | 75 | } |
86ef5c9a | 76 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 77 | |
86ef5c9a | 78 | void put_online_cpus(void) |
a9d9baa1 | 79 | { |
d221938c | 80 | if (cpu_hotplug.active_writer == current) |
aa953877 | 81 | return; |
d221938c | 82 | mutex_lock(&cpu_hotplug.lock); |
075663d1 SB |
83 | |
84 | if (WARN_ON(!cpu_hotplug.refcount)) | |
85 | cpu_hotplug.refcount++; /* try to fix things up */ | |
86 | ||
d2ba7e2a ON |
87 | if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) |
88 | wake_up_process(cpu_hotplug.active_writer); | |
d221938c GS |
89 | mutex_unlock(&cpu_hotplug.lock); |
90 | ||
a9d9baa1 | 91 | } |
86ef5c9a | 92 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 93 | |
d221938c GS |
94 | /* |
95 | * This ensures that the hotplug operation can begin only when the | |
96 | * refcount goes to zero. | |
97 | * | |
98 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
99 | * will be blocked by the cpu_hotplug.lock | |
100 | * | |
d2ba7e2a ON |
101 | * Since cpu_hotplug_begin() is always called after invoking |
102 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
103 | * |
104 | * Note that theoretically, there is a possibility of a livelock: | |
105 | * - Refcount goes to zero, last reader wakes up the sleeping | |
106 | * writer. | |
107 | * - Last reader unlocks the cpu_hotplug.lock. | |
108 | * - A new reader arrives at this moment, bumps up the refcount. | |
109 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
110 | * non zero and goes to sleep again. | |
111 | * | |
112 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 113 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
114 | * |
115 | */ | |
116 | static void cpu_hotplug_begin(void) | |
117 | { | |
d221938c | 118 | cpu_hotplug.active_writer = current; |
d2ba7e2a ON |
119 | |
120 | for (;;) { | |
121 | mutex_lock(&cpu_hotplug.lock); | |
122 | if (likely(!cpu_hotplug.refcount)) | |
123 | break; | |
124 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
d221938c GS |
125 | mutex_unlock(&cpu_hotplug.lock); |
126 | schedule(); | |
d221938c | 127 | } |
d221938c GS |
128 | } |
129 | ||
130 | static void cpu_hotplug_done(void) | |
131 | { | |
132 | cpu_hotplug.active_writer = NULL; | |
133 | mutex_unlock(&cpu_hotplug.lock); | |
134 | } | |
79a6cdeb LJ |
135 | |
136 | #else /* #if CONFIG_HOTPLUG_CPU */ | |
137 | static void cpu_hotplug_begin(void) {} | |
138 | static void cpu_hotplug_done(void) {} | |
25985edc | 139 | #endif /* #else #if CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 140 | |
1da177e4 | 141 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 142 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 143 | { |
bd5349cf | 144 | int ret; |
d221938c | 145 | cpu_maps_update_begin(); |
bd5349cf | 146 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 147 | cpu_maps_update_done(); |
bd5349cf | 148 | return ret; |
1da177e4 | 149 | } |
65edc68c | 150 | |
e9fb7631 AM |
151 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
152 | int *nr_calls) | |
153 | { | |
e6bde73b AM |
154 | int ret; |
155 | ||
156 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 157 | nr_calls); |
e6bde73b AM |
158 | |
159 | return notifier_to_errno(ret); | |
e9fb7631 AM |
160 | } |
161 | ||
162 | static int cpu_notify(unsigned long val, void *v) | |
163 | { | |
164 | return __cpu_notify(val, v, -1, NULL); | |
165 | } | |
166 | ||
00b9b0af LT |
167 | #ifdef CONFIG_HOTPLUG_CPU |
168 | ||
e9fb7631 AM |
169 | static void cpu_notify_nofail(unsigned long val, void *v) |
170 | { | |
00b9b0af | 171 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 172 | } |
1da177e4 LT |
173 | EXPORT_SYMBOL(register_cpu_notifier); |
174 | ||
9647155f | 175 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 176 | { |
d221938c | 177 | cpu_maps_update_begin(); |
bd5349cf | 178 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 179 | cpu_maps_update_done(); |
1da177e4 LT |
180 | } |
181 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
182 | ||
e4cc2f87 AV |
183 | /** |
184 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
185 | * @cpu: a CPU id | |
186 | * | |
187 | * This function walks all processes, finds a valid mm struct for each one and | |
188 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
189 | * trivial, there are various non-obvious corner cases, which this function | |
190 | * tries to solve in a safe manner. | |
191 | * | |
192 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
193 | * be called only for an already offlined CPU. | |
194 | */ | |
cb79295e AV |
195 | void clear_tasks_mm_cpumask(int cpu) |
196 | { | |
197 | struct task_struct *p; | |
198 | ||
199 | /* | |
200 | * This function is called after the cpu is taken down and marked | |
201 | * offline, so its not like new tasks will ever get this cpu set in | |
202 | * their mm mask. -- Peter Zijlstra | |
203 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
204 | * full-fledged tasklist_lock. | |
205 | */ | |
e4cc2f87 | 206 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
207 | rcu_read_lock(); |
208 | for_each_process(p) { | |
209 | struct task_struct *t; | |
210 | ||
e4cc2f87 AV |
211 | /* |
212 | * Main thread might exit, but other threads may still have | |
213 | * a valid mm. Find one. | |
214 | */ | |
cb79295e AV |
215 | t = find_lock_task_mm(p); |
216 | if (!t) | |
217 | continue; | |
218 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
219 | task_unlock(t); | |
220 | } | |
221 | rcu_read_unlock(); | |
222 | } | |
223 | ||
1da177e4 LT |
224 | static inline void check_for_tasks(int cpu) |
225 | { | |
226 | struct task_struct *p; | |
227 | ||
228 | write_lock_irq(&tasklist_lock); | |
229 | for_each_process(p) { | |
11854247 | 230 | if (task_cpu(p) == cpu && p->state == TASK_RUNNING && |
64861634 | 231 | (p->utime || p->stime)) |
9d3cfc4c FP |
232 | printk(KERN_WARNING "Task %s (pid = %d) is on cpu %d " |
233 | "(state = %ld, flags = %x)\n", | |
234 | p->comm, task_pid_nr(p), cpu, | |
235 | p->state, p->flags); | |
1da177e4 LT |
236 | } |
237 | write_unlock_irq(&tasklist_lock); | |
238 | } | |
239 | ||
db912f96 AK |
240 | struct take_cpu_down_param { |
241 | unsigned long mod; | |
242 | void *hcpu; | |
243 | }; | |
244 | ||
1da177e4 | 245 | /* Take this CPU down. */ |
514a20a5 | 246 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 247 | { |
db912f96 | 248 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
249 | int err; |
250 | ||
1da177e4 LT |
251 | /* Ensure this CPU doesn't handle any more interrupts. */ |
252 | err = __cpu_disable(); | |
253 | if (err < 0) | |
f3705136 | 254 | return err; |
1da177e4 | 255 | |
e9fb7631 | 256 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
f3705136 | 257 | return 0; |
1da177e4 LT |
258 | } |
259 | ||
e3920fb4 | 260 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 261 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 262 | { |
e7407dcc | 263 | int err, nr_calls = 0; |
e7407dcc | 264 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 265 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
266 | struct take_cpu_down_param tcd_param = { |
267 | .mod = mod, | |
268 | .hcpu = hcpu, | |
269 | }; | |
1da177e4 | 270 | |
e3920fb4 RW |
271 | if (num_online_cpus() == 1) |
272 | return -EBUSY; | |
1da177e4 | 273 | |
e3920fb4 RW |
274 | if (!cpu_online(cpu)) |
275 | return -EINVAL; | |
1da177e4 | 276 | |
d221938c | 277 | cpu_hotplug_begin(); |
4d51985e | 278 | |
e9fb7631 | 279 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 280 | if (err) { |
a0d8cdb6 | 281 | nr_calls--; |
e9fb7631 | 282 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
1da177e4 | 283 | printk("%s: attempt to take down CPU %u failed\n", |
af1f16d0 | 284 | __func__, cpu); |
baaca49f | 285 | goto out_release; |
1da177e4 | 286 | } |
f97f8f06 | 287 | smpboot_park_threads(cpu); |
1da177e4 | 288 | |
e0b582ec | 289 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 290 | if (err) { |
1da177e4 | 291 | /* CPU didn't die: tell everyone. Can't complain. */ |
f97f8f06 | 292 | smpboot_unpark_threads(cpu); |
e9fb7631 | 293 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 294 | goto out_release; |
8fa1d7d3 | 295 | } |
04321587 | 296 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 297 | |
48c5ccae PZ |
298 | /* |
299 | * The migration_call() CPU_DYING callback will have removed all | |
300 | * runnable tasks from the cpu, there's only the idle task left now | |
301 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
302 | * |
303 | * Wait for the stop thread to go away. | |
48c5ccae | 304 | */ |
51a96c77 PZ |
305 | while (!idle_cpu(cpu)) |
306 | cpu_relax(); | |
1da177e4 LT |
307 | |
308 | /* This actually kills the CPU. */ | |
309 | __cpu_die(cpu); | |
310 | ||
1da177e4 | 311 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
e9fb7631 | 312 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
313 | |
314 | check_for_tasks(cpu); | |
315 | ||
baaca49f | 316 | out_release: |
d221938c | 317 | cpu_hotplug_done(); |
e9fb7631 AM |
318 | if (!err) |
319 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
320 | return err; |
321 | } | |
322 | ||
514a20a5 | 323 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 324 | { |
9ea09af3 | 325 | int err; |
e3920fb4 | 326 | |
d221938c | 327 | cpu_maps_update_begin(); |
e761b772 MK |
328 | |
329 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 330 | err = -EBUSY; |
e761b772 MK |
331 | goto out; |
332 | } | |
333 | ||
e761b772 | 334 | err = _cpu_down(cpu, 0); |
e3920fb4 | 335 | |
e761b772 | 336 | out: |
d221938c | 337 | cpu_maps_update_done(); |
1da177e4 LT |
338 | return err; |
339 | } | |
b62b8ef9 | 340 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
341 | #endif /*CONFIG_HOTPLUG_CPU*/ |
342 | ||
e3920fb4 | 343 | /* Requires cpu_add_remove_lock to be held */ |
8bb78442 | 344 | static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 345 | { |
baaca49f | 346 | int ret, nr_calls = 0; |
1da177e4 | 347 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 348 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 349 | struct task_struct *idle; |
1da177e4 | 350 | |
e3920fb4 RW |
351 | if (cpu_online(cpu) || !cpu_present(cpu)) |
352 | return -EINVAL; | |
90d45d17 | 353 | |
d221938c | 354 | cpu_hotplug_begin(); |
38498a67 | 355 | |
3bb5d2ee SS |
356 | idle = idle_thread_get(cpu); |
357 | if (IS_ERR(idle)) { | |
358 | ret = PTR_ERR(idle); | |
38498a67 | 359 | goto out; |
3bb5d2ee | 360 | } |
38498a67 | 361 | |
f97f8f06 TG |
362 | ret = smpboot_create_threads(cpu); |
363 | if (ret) | |
364 | goto out; | |
365 | ||
e9fb7631 | 366 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 367 | if (ret) { |
a0d8cdb6 | 368 | nr_calls--; |
4d51985e | 369 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", |
af1f16d0 | 370 | __func__, cpu); |
1da177e4 LT |
371 | goto out_notify; |
372 | } | |
373 | ||
374 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 375 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
376 | if (ret != 0) |
377 | goto out_notify; | |
6978c705 | 378 | BUG_ON(!cpu_online(cpu)); |
1da177e4 | 379 | |
f97f8f06 TG |
380 | /* Wake the per cpu threads */ |
381 | smpboot_unpark_threads(cpu); | |
382 | ||
1da177e4 | 383 | /* Now call notifier in preparation. */ |
e9fb7631 | 384 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
385 | |
386 | out_notify: | |
387 | if (ret != 0) | |
e9fb7631 | 388 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 389 | out: |
d221938c | 390 | cpu_hotplug_done(); |
e3920fb4 RW |
391 | |
392 | return ret; | |
393 | } | |
394 | ||
b282b6f8 | 395 | int __cpuinit cpu_up(unsigned int cpu) |
e3920fb4 RW |
396 | { |
397 | int err = 0; | |
cf23422b | 398 | |
399 | #ifdef CONFIG_MEMORY_HOTPLUG | |
400 | int nid; | |
401 | pg_data_t *pgdat; | |
402 | #endif | |
403 | ||
e0b582ec | 404 | if (!cpu_possible(cpu)) { |
73e753a5 KH |
405 | printk(KERN_ERR "can't online cpu %d because it is not " |
406 | "configured as may-hotadd at boot time\n", cpu); | |
87d5e023 | 407 | #if defined(CONFIG_IA64) |
73e753a5 KH |
408 | printk(KERN_ERR "please check additional_cpus= boot " |
409 | "parameter\n"); | |
410 | #endif | |
411 | return -EINVAL; | |
412 | } | |
e3920fb4 | 413 | |
cf23422b | 414 | #ifdef CONFIG_MEMORY_HOTPLUG |
415 | nid = cpu_to_node(cpu); | |
416 | if (!node_online(nid)) { | |
417 | err = mem_online_node(nid); | |
418 | if (err) | |
419 | return err; | |
420 | } | |
421 | ||
422 | pgdat = NODE_DATA(nid); | |
423 | if (!pgdat) { | |
424 | printk(KERN_ERR | |
425 | "Can't online cpu %d due to NULL pgdat\n", cpu); | |
426 | return -ENOMEM; | |
427 | } | |
428 | ||
4eaf3f64 HL |
429 | if (pgdat->node_zonelists->_zonerefs->zone == NULL) { |
430 | mutex_lock(&zonelists_mutex); | |
9adb62a5 | 431 | build_all_zonelists(NULL, NULL); |
4eaf3f64 HL |
432 | mutex_unlock(&zonelists_mutex); |
433 | } | |
cf23422b | 434 | #endif |
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 | ||
443 | err = _cpu_up(cpu, 0); | |
444 | ||
e761b772 | 445 | out: |
d221938c | 446 | cpu_maps_update_done(); |
e3920fb4 RW |
447 | return err; |
448 | } | |
a513f6ba | 449 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 450 | |
f3de4be9 | 451 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 452 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
453 | |
454 | int disable_nonboot_cpus(void) | |
455 | { | |
e9a5f426 | 456 | int cpu, first_cpu, error = 0; |
e3920fb4 | 457 | |
d221938c | 458 | cpu_maps_update_begin(); |
e0b582ec | 459 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
460 | /* |
461 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
462 | * with the userspace trying to use the CPU hotplug at the same time |
463 | */ | |
e0b582ec | 464 | cpumask_clear(frozen_cpus); |
6ad4c188 | 465 | |
e3920fb4 RW |
466 | printk("Disabling non-boot CPUs ...\n"); |
467 | for_each_online_cpu(cpu) { | |
468 | if (cpu == first_cpu) | |
469 | continue; | |
8bb78442 | 470 | error = _cpu_down(cpu, 1); |
feae3203 | 471 | if (!error) |
e0b582ec | 472 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 473 | else { |
e3920fb4 RW |
474 | printk(KERN_ERR "Error taking CPU%d down: %d\n", |
475 | cpu, error); | |
476 | break; | |
477 | } | |
478 | } | |
86886e55 | 479 | |
e3920fb4 RW |
480 | if (!error) { |
481 | BUG_ON(num_online_cpus() > 1); | |
482 | /* Make sure the CPUs won't be enabled by someone else */ | |
483 | cpu_hotplug_disabled = 1; | |
484 | } else { | |
e1d9fd2e | 485 | printk(KERN_ERR "Non-boot CPUs are not disabled\n"); |
e3920fb4 | 486 | } |
d221938c | 487 | cpu_maps_update_done(); |
e3920fb4 RW |
488 | return error; |
489 | } | |
490 | ||
d0af9eed SS |
491 | void __weak arch_enable_nonboot_cpus_begin(void) |
492 | { | |
493 | } | |
494 | ||
495 | void __weak arch_enable_nonboot_cpus_end(void) | |
496 | { | |
497 | } | |
498 | ||
fa7303e2 | 499 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
500 | { |
501 | int cpu, error; | |
502 | ||
503 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 504 | cpu_maps_update_begin(); |
e3920fb4 | 505 | cpu_hotplug_disabled = 0; |
e0b582ec | 506 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 507 | goto out; |
e3920fb4 | 508 | |
4d51985e | 509 | printk(KERN_INFO "Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
510 | |
511 | arch_enable_nonboot_cpus_begin(); | |
512 | ||
e0b582ec | 513 | for_each_cpu(cpu, frozen_cpus) { |
8bb78442 | 514 | error = _cpu_up(cpu, 1); |
e3920fb4 | 515 | if (!error) { |
4d51985e | 516 | printk(KERN_INFO "CPU%d is up\n", cpu); |
e3920fb4 RW |
517 | continue; |
518 | } | |
1d64b9cb | 519 | printk(KERN_WARNING "Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 520 | } |
d0af9eed SS |
521 | |
522 | arch_enable_nonboot_cpus_end(); | |
523 | ||
e0b582ec | 524 | cpumask_clear(frozen_cpus); |
1d64b9cb | 525 | out: |
d221938c | 526 | cpu_maps_update_done(); |
1da177e4 | 527 | } |
e0b582ec | 528 | |
d7268a31 | 529 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
530 | { |
531 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
532 | return -ENOMEM; | |
533 | return 0; | |
534 | } | |
535 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa SB |
536 | |
537 | /* | |
538 | * Prevent regular CPU hotplug from racing with the freezer, by disabling CPU | |
539 | * hotplug when tasks are about to be frozen. Also, don't allow the freezer | |
540 | * to continue until any currently running CPU hotplug operation gets | |
541 | * completed. | |
542 | * To modify the 'cpu_hotplug_disabled' flag, we need to acquire the | |
543 | * 'cpu_add_remove_lock'. And this same lock is also taken by the regular | |
544 | * CPU hotplug path and released only after it is complete. Thus, we | |
545 | * (and hence the freezer) will block here until any currently running CPU | |
546 | * hotplug operation gets completed. | |
547 | */ | |
548 | void cpu_hotplug_disable_before_freeze(void) | |
549 | { | |
550 | cpu_maps_update_begin(); | |
551 | cpu_hotplug_disabled = 1; | |
552 | cpu_maps_update_done(); | |
553 | } | |
554 | ||
555 | ||
556 | /* | |
557 | * When tasks have been thawed, re-enable regular CPU hotplug (which had been | |
558 | * disabled while beginning to freeze tasks). | |
559 | */ | |
560 | void cpu_hotplug_enable_after_thaw(void) | |
561 | { | |
562 | cpu_maps_update_begin(); | |
563 | cpu_hotplug_disabled = 0; | |
564 | cpu_maps_update_done(); | |
565 | } | |
566 | ||
567 | /* | |
568 | * When callbacks for CPU hotplug notifications are being executed, we must | |
569 | * ensure that the state of the system with respect to the tasks being frozen | |
570 | * or not, as reported by the notification, remains unchanged *throughout the | |
571 | * duration* of the execution of the callbacks. | |
572 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
573 | * | |
574 | * This synchronization is implemented by mutually excluding regular CPU | |
575 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
576 | * Hibernate notifications. | |
577 | */ | |
578 | static int | |
579 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
580 | unsigned long action, void *ptr) | |
581 | { | |
582 | switch (action) { | |
583 | ||
584 | case PM_SUSPEND_PREPARE: | |
585 | case PM_HIBERNATION_PREPARE: | |
586 | cpu_hotplug_disable_before_freeze(); | |
587 | break; | |
588 | ||
589 | case PM_POST_SUSPEND: | |
590 | case PM_POST_HIBERNATION: | |
591 | cpu_hotplug_enable_after_thaw(); | |
592 | break; | |
593 | ||
594 | default: | |
595 | return NOTIFY_DONE; | |
596 | } | |
597 | ||
598 | return NOTIFY_OK; | |
599 | } | |
600 | ||
601 | ||
d7268a31 | 602 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa SB |
603 | { |
604 | pm_notifier(cpu_hotplug_pm_callback, 0); | |
605 | return 0; | |
606 | } | |
607 | core_initcall(cpu_hotplug_pm_sync_init); | |
608 | ||
f3de4be9 | 609 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 610 | |
e545a614 MS |
611 | /** |
612 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
613 | * @cpu: cpu that just started | |
614 | * | |
615 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
616 | * It must be called by the arch code on the new cpu, before the new cpu | |
617 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
618 | */ | |
84196414 | 619 | void __cpuinit notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
620 | { |
621 | unsigned long val = CPU_STARTING; | |
622 | ||
623 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 624 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
625 | val = CPU_STARTING_FROZEN; |
626 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 627 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
628 | } |
629 | ||
68f4f1ec | 630 | #endif /* CONFIG_SMP */ |
b8d317d1 | 631 | |
e56b3bc7 LT |
632 | /* |
633 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
634 | * represents all NR_CPUS bits binary values of 1<<nr. | |
635 | * | |
e0b582ec | 636 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
637 | * mask value that has a single bit set only. |
638 | */ | |
b8d317d1 | 639 | |
e56b3bc7 | 640 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 641 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
642 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
643 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
644 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 645 | |
e56b3bc7 LT |
646 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
647 | ||
648 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
649 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
650 | #if BITS_PER_LONG > 32 | |
651 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
652 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
653 | #endif |
654 | }; | |
e56b3bc7 | 655 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
656 | |
657 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
658 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
659 | |
660 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
661 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
662 | = CPU_BITS_ALL; | |
663 | #else | |
664 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
665 | #endif | |
666 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
667 | EXPORT_SYMBOL(cpu_possible_mask); | |
668 | ||
669 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
670 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
671 | EXPORT_SYMBOL(cpu_online_mask); | |
672 | ||
673 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
674 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
675 | EXPORT_SYMBOL(cpu_present_mask); | |
676 | ||
677 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
678 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
679 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
680 | |
681 | void set_cpu_possible(unsigned int cpu, bool possible) | |
682 | { | |
683 | if (possible) | |
684 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
685 | else | |
686 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
687 | } | |
688 | ||
689 | void set_cpu_present(unsigned int cpu, bool present) | |
690 | { | |
691 | if (present) | |
692 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
693 | else | |
694 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
695 | } | |
696 | ||
697 | void set_cpu_online(unsigned int cpu, bool online) | |
698 | { | |
699 | if (online) | |
700 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); | |
701 | else | |
702 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); | |
703 | } | |
704 | ||
705 | void set_cpu_active(unsigned int cpu, bool active) | |
706 | { | |
707 | if (active) | |
708 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
709 | else | |
710 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
711 | } | |
712 | ||
713 | void init_cpu_present(const struct cpumask *src) | |
714 | { | |
715 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
716 | } | |
717 | ||
718 | void init_cpu_possible(const struct cpumask *src) | |
719 | { | |
720 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
721 | } | |
722 | ||
723 | void init_cpu_online(const struct cpumask *src) | |
724 | { | |
725 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
726 | } |