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> |
345527b1 | 23 | #include <linux/tick.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; | |
87af9e7f DH |
62 | /* wait queue to wake up the active_writer */ |
63 | wait_queue_head_t wq; | |
64 | /* verifies that no writer will get active while readers are active */ | |
65 | struct mutex lock; | |
d221938c GS |
66 | /* |
67 | * Also blocks the new readers during | |
68 | * an ongoing cpu hotplug operation. | |
69 | */ | |
87af9e7f | 70 | atomic_t refcount; |
a19423b9 GS |
71 | |
72 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
73 | struct lockdep_map dep_map; | |
74 | #endif | |
31950eb6 LT |
75 | } cpu_hotplug = { |
76 | .active_writer = NULL, | |
87af9e7f | 77 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 78 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
79 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
80 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
81 | #endif | |
31950eb6 | 82 | }; |
d221938c | 83 | |
a19423b9 GS |
84 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
85 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
86 | #define cpuhp_lock_acquire_tryread() \ |
87 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
88 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
89 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
90 | ||
62db99f4 | 91 | |
86ef5c9a | 92 | void get_online_cpus(void) |
a9d9baa1 | 93 | { |
d221938c GS |
94 | might_sleep(); |
95 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 96 | return; |
a19423b9 | 97 | cpuhp_lock_acquire_read(); |
d221938c | 98 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 99 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 100 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 101 | } |
86ef5c9a | 102 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 103 | |
dd56af42 PM |
104 | bool try_get_online_cpus(void) |
105 | { | |
106 | if (cpu_hotplug.active_writer == current) | |
107 | return true; | |
108 | if (!mutex_trylock(&cpu_hotplug.lock)) | |
109 | return false; | |
110 | cpuhp_lock_acquire_tryread(); | |
87af9e7f | 111 | atomic_inc(&cpu_hotplug.refcount); |
dd56af42 PM |
112 | mutex_unlock(&cpu_hotplug.lock); |
113 | return true; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(try_get_online_cpus); | |
116 | ||
86ef5c9a | 117 | void put_online_cpus(void) |
a9d9baa1 | 118 | { |
87af9e7f DH |
119 | int refcount; |
120 | ||
d221938c | 121 | if (cpu_hotplug.active_writer == current) |
aa953877 | 122 | return; |
075663d1 | 123 | |
87af9e7f DH |
124 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
125 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
126 | atomic_inc(&cpu_hotplug.refcount); | |
127 | ||
128 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
129 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 130 | |
a19423b9 | 131 | cpuhp_lock_release(); |
d221938c | 132 | |
a9d9baa1 | 133 | } |
86ef5c9a | 134 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 135 | |
d221938c GS |
136 | /* |
137 | * This ensures that the hotplug operation can begin only when the | |
138 | * refcount goes to zero. | |
139 | * | |
140 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
141 | * will be blocked by the cpu_hotplug.lock | |
142 | * | |
d2ba7e2a ON |
143 | * Since cpu_hotplug_begin() is always called after invoking |
144 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
145 | * |
146 | * Note that theoretically, there is a possibility of a livelock: | |
147 | * - Refcount goes to zero, last reader wakes up the sleeping | |
148 | * writer. | |
149 | * - Last reader unlocks the cpu_hotplug.lock. | |
150 | * - A new reader arrives at this moment, bumps up the refcount. | |
151 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
152 | * non zero and goes to sleep again. | |
153 | * | |
154 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 155 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
156 | * |
157 | */ | |
b9d10be7 | 158 | void cpu_hotplug_begin(void) |
d221938c | 159 | { |
87af9e7f | 160 | DEFINE_WAIT(wait); |
d2ba7e2a | 161 | |
87af9e7f | 162 | cpu_hotplug.active_writer = current; |
a19423b9 | 163 | cpuhp_lock_acquire(); |
87af9e7f | 164 | |
d2ba7e2a ON |
165 | for (;;) { |
166 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
167 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
168 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
169 | break; | |
d221938c GS |
170 | mutex_unlock(&cpu_hotplug.lock); |
171 | schedule(); | |
d221938c | 172 | } |
87af9e7f | 173 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
174 | } |
175 | ||
b9d10be7 | 176 | void cpu_hotplug_done(void) |
d221938c GS |
177 | { |
178 | cpu_hotplug.active_writer = NULL; | |
179 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 180 | cpuhp_lock_release(); |
d221938c | 181 | } |
79a6cdeb | 182 | |
16e53dbf SB |
183 | /* |
184 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
185 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
186 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
187 | * hotplug path before performing hotplug operations. So acquiring that lock | |
188 | * guarantees mutual exclusion from any currently running hotplug operations. | |
189 | */ | |
190 | void cpu_hotplug_disable(void) | |
191 | { | |
192 | cpu_maps_update_begin(); | |
193 | cpu_hotplug_disabled = 1; | |
194 | cpu_maps_update_done(); | |
195 | } | |
196 | ||
197 | void cpu_hotplug_enable(void) | |
198 | { | |
199 | cpu_maps_update_begin(); | |
200 | cpu_hotplug_disabled = 0; | |
201 | cpu_maps_update_done(); | |
202 | } | |
203 | ||
b9d10be7 | 204 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 205 | |
1da177e4 | 206 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 207 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 208 | { |
bd5349cf | 209 | int ret; |
d221938c | 210 | cpu_maps_update_begin(); |
bd5349cf | 211 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 212 | cpu_maps_update_done(); |
bd5349cf | 213 | return ret; |
1da177e4 | 214 | } |
65edc68c | 215 | |
93ae4f97 SB |
216 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
217 | { | |
218 | return raw_notifier_chain_register(&cpu_chain, nb); | |
219 | } | |
220 | ||
e9fb7631 AM |
221 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
222 | int *nr_calls) | |
223 | { | |
e6bde73b AM |
224 | int ret; |
225 | ||
226 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 227 | nr_calls); |
e6bde73b AM |
228 | |
229 | return notifier_to_errno(ret); | |
e9fb7631 AM |
230 | } |
231 | ||
232 | static int cpu_notify(unsigned long val, void *v) | |
233 | { | |
234 | return __cpu_notify(val, v, -1, NULL); | |
235 | } | |
236 | ||
00b9b0af LT |
237 | #ifdef CONFIG_HOTPLUG_CPU |
238 | ||
e9fb7631 AM |
239 | static void cpu_notify_nofail(unsigned long val, void *v) |
240 | { | |
00b9b0af | 241 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 242 | } |
1da177e4 | 243 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 244 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 245 | |
9647155f | 246 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 247 | { |
d221938c | 248 | cpu_maps_update_begin(); |
bd5349cf | 249 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 250 | cpu_maps_update_done(); |
1da177e4 LT |
251 | } |
252 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
253 | ||
93ae4f97 SB |
254 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
255 | { | |
256 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
257 | } | |
258 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
259 | ||
e4cc2f87 AV |
260 | /** |
261 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
262 | * @cpu: a CPU id | |
263 | * | |
264 | * This function walks all processes, finds a valid mm struct for each one and | |
265 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
266 | * trivial, there are various non-obvious corner cases, which this function | |
267 | * tries to solve in a safe manner. | |
268 | * | |
269 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
270 | * be called only for an already offlined CPU. | |
271 | */ | |
cb79295e AV |
272 | void clear_tasks_mm_cpumask(int cpu) |
273 | { | |
274 | struct task_struct *p; | |
275 | ||
276 | /* | |
277 | * This function is called after the cpu is taken down and marked | |
278 | * offline, so its not like new tasks will ever get this cpu set in | |
279 | * their mm mask. -- Peter Zijlstra | |
280 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
281 | * full-fledged tasklist_lock. | |
282 | */ | |
e4cc2f87 | 283 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
284 | rcu_read_lock(); |
285 | for_each_process(p) { | |
286 | struct task_struct *t; | |
287 | ||
e4cc2f87 AV |
288 | /* |
289 | * Main thread might exit, but other threads may still have | |
290 | * a valid mm. Find one. | |
291 | */ | |
cb79295e AV |
292 | t = find_lock_task_mm(p); |
293 | if (!t) | |
294 | continue; | |
295 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
296 | task_unlock(t); | |
297 | } | |
298 | rcu_read_unlock(); | |
299 | } | |
300 | ||
b728ca06 | 301 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 302 | { |
b728ca06 | 303 | struct task_struct *g, *p; |
1da177e4 | 304 | |
b728ca06 KT |
305 | read_lock_irq(&tasklist_lock); |
306 | do_each_thread(g, p) { | |
307 | if (!p->on_rq) | |
308 | continue; | |
309 | /* | |
310 | * We do the check with unlocked task_rq(p)->lock. | |
311 | * Order the reading to do not warn about a task, | |
312 | * which was running on this cpu in the past, and | |
313 | * it's just been woken on another cpu. | |
314 | */ | |
315 | rmb(); | |
316 | if (task_cpu(p) != dead_cpu) | |
317 | continue; | |
318 | ||
319 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
320 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
321 | } while_each_thread(g, p); | |
322 | read_unlock_irq(&tasklist_lock); | |
1da177e4 LT |
323 | } |
324 | ||
db912f96 AK |
325 | struct take_cpu_down_param { |
326 | unsigned long mod; | |
327 | void *hcpu; | |
328 | }; | |
329 | ||
1da177e4 | 330 | /* Take this CPU down. */ |
514a20a5 | 331 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 332 | { |
db912f96 | 333 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
334 | int err; |
335 | ||
1da177e4 LT |
336 | /* Ensure this CPU doesn't handle any more interrupts. */ |
337 | err = __cpu_disable(); | |
338 | if (err < 0) | |
f3705136 | 339 | return err; |
1da177e4 | 340 | |
e9fb7631 | 341 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
52c063d1 TG |
342 | /* Give up timekeeping duties */ |
343 | tick_handover_do_timer(); | |
14e568e7 TG |
344 | /* Park the stopper thread */ |
345 | kthread_park(current); | |
f3705136 | 346 | return 0; |
1da177e4 LT |
347 | } |
348 | ||
e3920fb4 | 349 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 350 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 351 | { |
e7407dcc | 352 | int err, nr_calls = 0; |
e7407dcc | 353 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 354 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
355 | struct take_cpu_down_param tcd_param = { |
356 | .mod = mod, | |
357 | .hcpu = hcpu, | |
358 | }; | |
1da177e4 | 359 | |
e3920fb4 RW |
360 | if (num_online_cpus() == 1) |
361 | return -EBUSY; | |
1da177e4 | 362 | |
e3920fb4 RW |
363 | if (!cpu_online(cpu)) |
364 | return -EINVAL; | |
1da177e4 | 365 | |
d221938c | 366 | cpu_hotplug_begin(); |
4d51985e | 367 | |
e9fb7631 | 368 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 369 | if (err) { |
a0d8cdb6 | 370 | nr_calls--; |
e9fb7631 | 371 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
372 | pr_warn("%s: attempt to take down CPU %u failed\n", |
373 | __func__, cpu); | |
baaca49f | 374 | goto out_release; |
1da177e4 LT |
375 | } |
376 | ||
6acce3ef PZ |
377 | /* |
378 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
379 | * and RCU users of this state to go away such that all new such users | |
380 | * will observe it. | |
381 | * | |
382 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
383 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
384 | * |
385 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
386 | */ |
387 | #ifdef CONFIG_PREEMPT | |
388 | synchronize_sched(); | |
389 | #endif | |
390 | synchronize_rcu(); | |
391 | ||
106dd5af M |
392 | smpboot_park_threads(cpu); |
393 | ||
6acce3ef PZ |
394 | /* |
395 | * So now all preempt/rcu users must observe !cpu_active(). | |
396 | */ | |
397 | ||
e0b582ec | 398 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 399 | if (err) { |
1da177e4 | 400 | /* CPU didn't die: tell everyone. Can't complain. */ |
e9fb7631 | 401 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
6a1bdc1b | 402 | goto out_release; |
8fa1d7d3 | 403 | } |
04321587 | 404 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 405 | |
48c5ccae PZ |
406 | /* |
407 | * The migration_call() CPU_DYING callback will have removed all | |
408 | * runnable tasks from the cpu, there's only the idle task left now | |
409 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
410 | * |
411 | * Wait for the stop thread to go away. | |
48c5ccae | 412 | */ |
528a25b0 | 413 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 414 | cpu_relax(); |
528a25b0 PM |
415 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
416 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 417 | |
345527b1 | 418 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
419 | /* This actually kills the CPU. */ |
420 | __cpu_die(cpu); | |
421 | ||
1da177e4 | 422 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
a49b116d | 423 | tick_cleanup_dead_cpu(cpu); |
e9fb7631 | 424 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
425 | |
426 | check_for_tasks(cpu); | |
427 | ||
baaca49f | 428 | out_release: |
d221938c | 429 | cpu_hotplug_done(); |
e9fb7631 AM |
430 | if (!err) |
431 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
432 | return err; |
433 | } | |
434 | ||
514a20a5 | 435 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 436 | { |
9ea09af3 | 437 | int err; |
e3920fb4 | 438 | |
d221938c | 439 | cpu_maps_update_begin(); |
e761b772 MK |
440 | |
441 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 442 | err = -EBUSY; |
e761b772 MK |
443 | goto out; |
444 | } | |
445 | ||
e761b772 | 446 | err = _cpu_down(cpu, 0); |
e3920fb4 | 447 | |
e761b772 | 448 | out: |
d221938c | 449 | cpu_maps_update_done(); |
1da177e4 LT |
450 | return err; |
451 | } | |
b62b8ef9 | 452 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
453 | #endif /*CONFIG_HOTPLUG_CPU*/ |
454 | ||
00df35f9 PM |
455 | /* |
456 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
457 | */ | |
458 | static int smpboot_thread_call(struct notifier_block *nfb, | |
459 | unsigned long action, void *hcpu) | |
460 | { | |
461 | int cpu = (long)hcpu; | |
462 | ||
463 | switch (action & ~CPU_TASKS_FROZEN) { | |
464 | ||
64eaf974 | 465 | case CPU_DOWN_FAILED: |
00df35f9 PM |
466 | case CPU_ONLINE: |
467 | smpboot_unpark_threads(cpu); | |
468 | break; | |
469 | ||
470 | default: | |
471 | break; | |
472 | } | |
473 | ||
474 | return NOTIFY_OK; | |
475 | } | |
476 | ||
477 | static struct notifier_block smpboot_thread_notifier = { | |
478 | .notifier_call = smpboot_thread_call, | |
479 | .priority = CPU_PRI_SMPBOOT, | |
480 | }; | |
481 | ||
927da9df | 482 | void smpboot_thread_init(void) |
00df35f9 PM |
483 | { |
484 | register_cpu_notifier(&smpboot_thread_notifier); | |
485 | } | |
486 | ||
e3920fb4 | 487 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 488 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 489 | { |
baaca49f | 490 | int ret, nr_calls = 0; |
1da177e4 | 491 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 492 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 493 | struct task_struct *idle; |
1da177e4 | 494 | |
d221938c | 495 | cpu_hotplug_begin(); |
38498a67 | 496 | |
5e5041f3 YI |
497 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
498 | ret = -EINVAL; | |
499 | goto out; | |
500 | } | |
501 | ||
3bb5d2ee SS |
502 | idle = idle_thread_get(cpu); |
503 | if (IS_ERR(idle)) { | |
504 | ret = PTR_ERR(idle); | |
38498a67 | 505 | goto out; |
3bb5d2ee | 506 | } |
38498a67 | 507 | |
f97f8f06 TG |
508 | ret = smpboot_create_threads(cpu); |
509 | if (ret) | |
510 | goto out; | |
511 | ||
e9fb7631 | 512 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 513 | if (ret) { |
a0d8cdb6 | 514 | nr_calls--; |
84117da5 FF |
515 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
516 | __func__, cpu); | |
1da177e4 LT |
517 | goto out_notify; |
518 | } | |
519 | ||
520 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 521 | ret = __cpu_up(cpu, idle); |
1da177e4 LT |
522 | if (ret != 0) |
523 | goto out_notify; | |
6978c705 | 524 | BUG_ON(!cpu_online(cpu)); |
1da177e4 LT |
525 | |
526 | /* Now call notifier in preparation. */ | |
e9fb7631 | 527 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
528 | |
529 | out_notify: | |
530 | if (ret != 0) | |
e9fb7631 | 531 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 532 | out: |
d221938c | 533 | cpu_hotplug_done(); |
e3920fb4 RW |
534 | |
535 | return ret; | |
536 | } | |
537 | ||
0db0628d | 538 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
539 | { |
540 | int err = 0; | |
cf23422b | 541 | |
e0b582ec | 542 | if (!cpu_possible(cpu)) { |
84117da5 FF |
543 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
544 | cpu); | |
87d5e023 | 545 | #if defined(CONFIG_IA64) |
84117da5 | 546 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
547 | #endif |
548 | return -EINVAL; | |
549 | } | |
e3920fb4 | 550 | |
01b0f197 TK |
551 | err = try_online_node(cpu_to_node(cpu)); |
552 | if (err) | |
553 | return err; | |
cf23422b | 554 | |
d221938c | 555 | cpu_maps_update_begin(); |
e761b772 MK |
556 | |
557 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 558 | err = -EBUSY; |
e761b772 MK |
559 | goto out; |
560 | } | |
561 | ||
562 | err = _cpu_up(cpu, 0); | |
563 | ||
e761b772 | 564 | out: |
d221938c | 565 | cpu_maps_update_done(); |
e3920fb4 RW |
566 | return err; |
567 | } | |
a513f6ba | 568 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 569 | |
f3de4be9 | 570 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 571 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
572 | |
573 | int disable_nonboot_cpus(void) | |
574 | { | |
e9a5f426 | 575 | int cpu, first_cpu, error = 0; |
e3920fb4 | 576 | |
d221938c | 577 | cpu_maps_update_begin(); |
e0b582ec | 578 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
579 | /* |
580 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
581 | * with the userspace trying to use the CPU hotplug at the same time |
582 | */ | |
e0b582ec | 583 | cpumask_clear(frozen_cpus); |
6ad4c188 | 584 | |
84117da5 | 585 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
586 | for_each_online_cpu(cpu) { |
587 | if (cpu == first_cpu) | |
588 | continue; | |
bb3632c6 | 589 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 590 | error = _cpu_down(cpu, 1); |
bb3632c6 | 591 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 592 | if (!error) |
e0b582ec | 593 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 594 | else { |
84117da5 | 595 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
596 | break; |
597 | } | |
598 | } | |
86886e55 | 599 | |
e3920fb4 RW |
600 | if (!error) { |
601 | BUG_ON(num_online_cpus() > 1); | |
602 | /* Make sure the CPUs won't be enabled by someone else */ | |
603 | cpu_hotplug_disabled = 1; | |
604 | } else { | |
84117da5 | 605 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 606 | } |
d221938c | 607 | cpu_maps_update_done(); |
e3920fb4 RW |
608 | return error; |
609 | } | |
610 | ||
d0af9eed SS |
611 | void __weak arch_enable_nonboot_cpus_begin(void) |
612 | { | |
613 | } | |
614 | ||
615 | void __weak arch_enable_nonboot_cpus_end(void) | |
616 | { | |
617 | } | |
618 | ||
fa7303e2 | 619 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
620 | { |
621 | int cpu, error; | |
622 | ||
623 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 624 | cpu_maps_update_begin(); |
e3920fb4 | 625 | cpu_hotplug_disabled = 0; |
e0b582ec | 626 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 627 | goto out; |
e3920fb4 | 628 | |
84117da5 | 629 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
630 | |
631 | arch_enable_nonboot_cpus_begin(); | |
632 | ||
e0b582ec | 633 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 634 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 635 | error = _cpu_up(cpu, 1); |
bb3632c6 | 636 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 637 | if (!error) { |
84117da5 | 638 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
639 | continue; |
640 | } | |
84117da5 | 641 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 642 | } |
d0af9eed SS |
643 | |
644 | arch_enable_nonboot_cpus_end(); | |
645 | ||
e0b582ec | 646 | cpumask_clear(frozen_cpus); |
1d64b9cb | 647 | out: |
d221938c | 648 | cpu_maps_update_done(); |
1da177e4 | 649 | } |
e0b582ec | 650 | |
d7268a31 | 651 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
652 | { |
653 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
654 | return -ENOMEM; | |
655 | return 0; | |
656 | } | |
657 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 658 | |
79cfbdfa SB |
659 | /* |
660 | * When callbacks for CPU hotplug notifications are being executed, we must | |
661 | * ensure that the state of the system with respect to the tasks being frozen | |
662 | * or not, as reported by the notification, remains unchanged *throughout the | |
663 | * duration* of the execution of the callbacks. | |
664 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
665 | * | |
666 | * This synchronization is implemented by mutually excluding regular CPU | |
667 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
668 | * Hibernate notifications. | |
669 | */ | |
670 | static int | |
671 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
672 | unsigned long action, void *ptr) | |
673 | { | |
674 | switch (action) { | |
675 | ||
676 | case PM_SUSPEND_PREPARE: | |
677 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 678 | cpu_hotplug_disable(); |
79cfbdfa SB |
679 | break; |
680 | ||
681 | case PM_POST_SUSPEND: | |
682 | case PM_POST_HIBERNATION: | |
16e53dbf | 683 | cpu_hotplug_enable(); |
79cfbdfa SB |
684 | break; |
685 | ||
686 | default: | |
687 | return NOTIFY_DONE; | |
688 | } | |
689 | ||
690 | return NOTIFY_OK; | |
691 | } | |
692 | ||
693 | ||
d7268a31 | 694 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 695 | { |
6e32d479 FY |
696 | /* |
697 | * cpu_hotplug_pm_callback has higher priority than x86 | |
698 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
699 | * to disable cpu hotplug to avoid cpu hotplug race. | |
700 | */ | |
79cfbdfa SB |
701 | pm_notifier(cpu_hotplug_pm_callback, 0); |
702 | return 0; | |
703 | } | |
704 | core_initcall(cpu_hotplug_pm_sync_init); | |
705 | ||
f3de4be9 | 706 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 707 | |
e545a614 MS |
708 | /** |
709 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
710 | * @cpu: cpu that just started | |
711 | * | |
712 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
713 | * It must be called by the arch code on the new cpu, before the new cpu | |
714 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
715 | */ | |
0db0628d | 716 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
717 | { |
718 | unsigned long val = CPU_STARTING; | |
719 | ||
720 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 721 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
722 | val = CPU_STARTING_FROZEN; |
723 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 724 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
725 | } |
726 | ||
68f4f1ec | 727 | #endif /* CONFIG_SMP */ |
b8d317d1 | 728 | |
e56b3bc7 LT |
729 | /* |
730 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
731 | * represents all NR_CPUS bits binary values of 1<<nr. | |
732 | * | |
e0b582ec | 733 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
734 | * mask value that has a single bit set only. |
735 | */ | |
b8d317d1 | 736 | |
e56b3bc7 | 737 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 738 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
739 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
740 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
741 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 742 | |
e56b3bc7 LT |
743 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
744 | ||
745 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
746 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
747 | #if BITS_PER_LONG > 32 | |
748 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
749 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
750 | #endif |
751 | }; | |
e56b3bc7 | 752 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
753 | |
754 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
755 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
756 | |
757 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
758 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
759 | = CPU_BITS_ALL; | |
760 | #else | |
761 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
762 | #endif | |
763 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
764 | EXPORT_SYMBOL(cpu_possible_mask); | |
765 | ||
766 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
767 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
768 | EXPORT_SYMBOL(cpu_online_mask); | |
769 | ||
770 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
771 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
772 | EXPORT_SYMBOL(cpu_present_mask); | |
773 | ||
774 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
775 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
776 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
777 | |
778 | void set_cpu_possible(unsigned int cpu, bool possible) | |
779 | { | |
780 | if (possible) | |
781 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
782 | else | |
783 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
784 | } | |
785 | ||
786 | void set_cpu_present(unsigned int cpu, bool present) | |
787 | { | |
788 | if (present) | |
789 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
790 | else | |
791 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
792 | } | |
793 | ||
794 | void set_cpu_online(unsigned int cpu, bool online) | |
795 | { | |
6acbfb96 | 796 | if (online) { |
3fa41520 | 797 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
798 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
799 | } else { | |
3fa41520 | 800 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 801 | } |
3fa41520 RR |
802 | } |
803 | ||
804 | void set_cpu_active(unsigned int cpu, bool active) | |
805 | { | |
806 | if (active) | |
807 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
808 | else | |
809 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
810 | } | |
811 | ||
812 | void init_cpu_present(const struct cpumask *src) | |
813 | { | |
814 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
815 | } | |
816 | ||
817 | void init_cpu_possible(const struct cpumask *src) | |
818 | { | |
819 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
820 | } | |
821 | ||
822 | void init_cpu_online(const struct cpumask *src) | |
823 | { | |
824 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
825 | } |