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> |
a8994181 | 24 | #include <linux/irq.h> |
bb3632c6 | 25 | #include <trace/events/power.h> |
1da177e4 | 26 | |
38498a67 TG |
27 | #include "smpboot.h" |
28 | ||
98a79d6a | 29 | #ifdef CONFIG_SMP |
b3199c02 | 30 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 31 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
32 | bool cpuhp_tasks_frozen; |
33 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 34 | |
79a6cdeb | 35 | /* |
93ae4f97 SB |
36 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
37 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
38 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
39 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
40 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
41 | */ |
42 | void cpu_maps_update_begin(void) | |
43 | { | |
44 | mutex_lock(&cpu_add_remove_lock); | |
45 | } | |
93ae4f97 | 46 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
47 | |
48 | void cpu_maps_update_done(void) | |
49 | { | |
50 | mutex_unlock(&cpu_add_remove_lock); | |
51 | } | |
93ae4f97 | 52 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 53 | |
5c113fbe | 54 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 55 | |
e3920fb4 RW |
56 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
57 | * Should always be manipulated under cpu_add_remove_lock | |
58 | */ | |
59 | static int cpu_hotplug_disabled; | |
60 | ||
79a6cdeb LJ |
61 | #ifdef CONFIG_HOTPLUG_CPU |
62 | ||
d221938c GS |
63 | static struct { |
64 | struct task_struct *active_writer; | |
87af9e7f DH |
65 | /* wait queue to wake up the active_writer */ |
66 | wait_queue_head_t wq; | |
67 | /* verifies that no writer will get active while readers are active */ | |
68 | struct mutex lock; | |
d221938c GS |
69 | /* |
70 | * Also blocks the new readers during | |
71 | * an ongoing cpu hotplug operation. | |
72 | */ | |
87af9e7f | 73 | atomic_t refcount; |
a19423b9 GS |
74 | |
75 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
76 | struct lockdep_map dep_map; | |
77 | #endif | |
31950eb6 LT |
78 | } cpu_hotplug = { |
79 | .active_writer = NULL, | |
87af9e7f | 80 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 81 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
82 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
83 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
84 | #endif | |
31950eb6 | 85 | }; |
d221938c | 86 | |
a19423b9 GS |
87 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
88 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
89 | #define cpuhp_lock_acquire_tryread() \ |
90 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
91 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
92 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
93 | ||
62db99f4 | 94 | |
86ef5c9a | 95 | void get_online_cpus(void) |
a9d9baa1 | 96 | { |
d221938c GS |
97 | might_sleep(); |
98 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 99 | return; |
a19423b9 | 100 | cpuhp_lock_acquire_read(); |
d221938c | 101 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 102 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 103 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 104 | } |
86ef5c9a | 105 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 106 | |
86ef5c9a | 107 | void put_online_cpus(void) |
a9d9baa1 | 108 | { |
87af9e7f DH |
109 | int refcount; |
110 | ||
d221938c | 111 | if (cpu_hotplug.active_writer == current) |
aa953877 | 112 | return; |
075663d1 | 113 | |
87af9e7f DH |
114 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
115 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
116 | atomic_inc(&cpu_hotplug.refcount); | |
117 | ||
118 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
119 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 120 | |
a19423b9 | 121 | cpuhp_lock_release(); |
d221938c | 122 | |
a9d9baa1 | 123 | } |
86ef5c9a | 124 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 125 | |
d221938c GS |
126 | /* |
127 | * This ensures that the hotplug operation can begin only when the | |
128 | * refcount goes to zero. | |
129 | * | |
130 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
131 | * will be blocked by the cpu_hotplug.lock | |
132 | * | |
d2ba7e2a ON |
133 | * Since cpu_hotplug_begin() is always called after invoking |
134 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
135 | * |
136 | * Note that theoretically, there is a possibility of a livelock: | |
137 | * - Refcount goes to zero, last reader wakes up the sleeping | |
138 | * writer. | |
139 | * - Last reader unlocks the cpu_hotplug.lock. | |
140 | * - A new reader arrives at this moment, bumps up the refcount. | |
141 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
142 | * non zero and goes to sleep again. | |
143 | * | |
144 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 145 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
146 | * |
147 | */ | |
b9d10be7 | 148 | void cpu_hotplug_begin(void) |
d221938c | 149 | { |
87af9e7f | 150 | DEFINE_WAIT(wait); |
d2ba7e2a | 151 | |
87af9e7f | 152 | cpu_hotplug.active_writer = current; |
a19423b9 | 153 | cpuhp_lock_acquire(); |
87af9e7f | 154 | |
d2ba7e2a ON |
155 | for (;;) { |
156 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
157 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
158 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
159 | break; | |
d221938c GS |
160 | mutex_unlock(&cpu_hotplug.lock); |
161 | schedule(); | |
d221938c | 162 | } |
87af9e7f | 163 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
164 | } |
165 | ||
b9d10be7 | 166 | void cpu_hotplug_done(void) |
d221938c GS |
167 | { |
168 | cpu_hotplug.active_writer = NULL; | |
169 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 170 | cpuhp_lock_release(); |
d221938c | 171 | } |
79a6cdeb | 172 | |
16e53dbf SB |
173 | /* |
174 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
175 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
176 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
177 | * hotplug path before performing hotplug operations. So acquiring that lock | |
178 | * guarantees mutual exclusion from any currently running hotplug operations. | |
179 | */ | |
180 | void cpu_hotplug_disable(void) | |
181 | { | |
182 | cpu_maps_update_begin(); | |
89af7ba5 | 183 | cpu_hotplug_disabled++; |
16e53dbf SB |
184 | cpu_maps_update_done(); |
185 | } | |
32145c46 | 186 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf SB |
187 | |
188 | void cpu_hotplug_enable(void) | |
189 | { | |
190 | cpu_maps_update_begin(); | |
89af7ba5 | 191 | WARN_ON(--cpu_hotplug_disabled < 0); |
16e53dbf SB |
192 | cpu_maps_update_done(); |
193 | } | |
32145c46 | 194 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 195 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 196 | |
1da177e4 | 197 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 198 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 199 | { |
bd5349cf | 200 | int ret; |
d221938c | 201 | cpu_maps_update_begin(); |
bd5349cf | 202 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 203 | cpu_maps_update_done(); |
bd5349cf | 204 | return ret; |
1da177e4 | 205 | } |
65edc68c | 206 | |
71cf5aee | 207 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
208 | { |
209 | return raw_notifier_chain_register(&cpu_chain, nb); | |
210 | } | |
211 | ||
090e77c3 | 212 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
213 | int *nr_calls) |
214 | { | |
090e77c3 TG |
215 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
216 | void *hcpu = (void *)(long)cpu; | |
217 | ||
e6bde73b AM |
218 | int ret; |
219 | ||
090e77c3 | 220 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 221 | nr_calls); |
e6bde73b AM |
222 | |
223 | return notifier_to_errno(ret); | |
e9fb7631 AM |
224 | } |
225 | ||
090e77c3 | 226 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 227 | { |
090e77c3 | 228 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
229 | } |
230 | ||
00b9b0af LT |
231 | #ifdef CONFIG_HOTPLUG_CPU |
232 | ||
090e77c3 | 233 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
e9fb7631 | 234 | { |
090e77c3 | 235 | BUG_ON(cpu_notify(val, cpu)); |
e9fb7631 | 236 | } |
1da177e4 | 237 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 238 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 239 | |
71cf5aee | 240 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 241 | { |
d221938c | 242 | cpu_maps_update_begin(); |
bd5349cf | 243 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 244 | cpu_maps_update_done(); |
1da177e4 LT |
245 | } |
246 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
247 | ||
71cf5aee | 248 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
249 | { |
250 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
251 | } | |
252 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
253 | ||
e4cc2f87 AV |
254 | /** |
255 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
256 | * @cpu: a CPU id | |
257 | * | |
258 | * This function walks all processes, finds a valid mm struct for each one and | |
259 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
260 | * trivial, there are various non-obvious corner cases, which this function | |
261 | * tries to solve in a safe manner. | |
262 | * | |
263 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
264 | * be called only for an already offlined CPU. | |
265 | */ | |
cb79295e AV |
266 | void clear_tasks_mm_cpumask(int cpu) |
267 | { | |
268 | struct task_struct *p; | |
269 | ||
270 | /* | |
271 | * This function is called after the cpu is taken down and marked | |
272 | * offline, so its not like new tasks will ever get this cpu set in | |
273 | * their mm mask. -- Peter Zijlstra | |
274 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
275 | * full-fledged tasklist_lock. | |
276 | */ | |
e4cc2f87 | 277 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
278 | rcu_read_lock(); |
279 | for_each_process(p) { | |
280 | struct task_struct *t; | |
281 | ||
e4cc2f87 AV |
282 | /* |
283 | * Main thread might exit, but other threads may still have | |
284 | * a valid mm. Find one. | |
285 | */ | |
cb79295e AV |
286 | t = find_lock_task_mm(p); |
287 | if (!t) | |
288 | continue; | |
289 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
290 | task_unlock(t); | |
291 | } | |
292 | rcu_read_unlock(); | |
293 | } | |
294 | ||
b728ca06 | 295 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 296 | { |
b728ca06 | 297 | struct task_struct *g, *p; |
1da177e4 | 298 | |
a75a6068 ON |
299 | read_lock(&tasklist_lock); |
300 | for_each_process_thread(g, p) { | |
b728ca06 KT |
301 | if (!p->on_rq) |
302 | continue; | |
303 | /* | |
304 | * We do the check with unlocked task_rq(p)->lock. | |
305 | * Order the reading to do not warn about a task, | |
306 | * which was running on this cpu in the past, and | |
307 | * it's just been woken on another cpu. | |
308 | */ | |
309 | rmb(); | |
310 | if (task_cpu(p) != dead_cpu) | |
311 | continue; | |
312 | ||
313 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
314 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
315 | } |
316 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
317 | } |
318 | ||
319 | /* Take this CPU down. */ | |
71cf5aee | 320 | static int take_cpu_down(void *_param) |
1da177e4 | 321 | { |
090e77c3 | 322 | int err, cpu = smp_processor_id(); |
1da177e4 | 323 | |
1da177e4 LT |
324 | /* Ensure this CPU doesn't handle any more interrupts. */ |
325 | err = __cpu_disable(); | |
326 | if (err < 0) | |
f3705136 | 327 | return err; |
1da177e4 | 328 | |
090e77c3 | 329 | cpu_notify(CPU_DYING, cpu); |
52c063d1 TG |
330 | /* Give up timekeeping duties */ |
331 | tick_handover_do_timer(); | |
14e568e7 | 332 | /* Park the stopper thread */ |
090e77c3 | 333 | stop_machine_park(cpu); |
f3705136 | 334 | return 0; |
1da177e4 LT |
335 | } |
336 | ||
e3920fb4 | 337 | /* Requires cpu_add_remove_lock to be held */ |
71cf5aee | 338 | static int _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 339 | { |
e7407dcc | 340 | int err, nr_calls = 0; |
1da177e4 | 341 | |
e3920fb4 RW |
342 | if (num_online_cpus() == 1) |
343 | return -EBUSY; | |
1da177e4 | 344 | |
e3920fb4 RW |
345 | if (!cpu_online(cpu)) |
346 | return -EINVAL; | |
1da177e4 | 347 | |
d221938c | 348 | cpu_hotplug_begin(); |
4d51985e | 349 | |
090e77c3 TG |
350 | cpuhp_tasks_frozen = tasks_frozen; |
351 | ||
352 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
e6bde73b | 353 | if (err) { |
a0d8cdb6 | 354 | nr_calls--; |
090e77c3 | 355 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); |
84117da5 FF |
356 | pr_warn("%s: attempt to take down CPU %u failed\n", |
357 | __func__, cpu); | |
baaca49f | 358 | goto out_release; |
1da177e4 LT |
359 | } |
360 | ||
6acce3ef PZ |
361 | /* |
362 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
363 | * and RCU users of this state to go away such that all new such users | |
364 | * will observe it. | |
365 | * | |
366 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
779de6ce | 367 | * not imply sync_sched(), so wait for both. |
106dd5af M |
368 | * |
369 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef | 370 | */ |
779de6ce PM |
371 | if (IS_ENABLED(CONFIG_PREEMPT)) |
372 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
373 | else | |
374 | synchronize_rcu(); | |
6acce3ef | 375 | |
106dd5af M |
376 | smpboot_park_threads(cpu); |
377 | ||
6acce3ef | 378 | /* |
a8994181 TG |
379 | * Prevent irq alloc/free while the dying cpu reorganizes the |
380 | * interrupt affinities. | |
6acce3ef | 381 | */ |
a8994181 | 382 | irq_lock_sparse(); |
6acce3ef | 383 | |
a8994181 TG |
384 | /* |
385 | * So now all preempt/rcu users must observe !cpu_active(). | |
386 | */ | |
090e77c3 | 387 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 388 | if (err) { |
1da177e4 | 389 | /* CPU didn't die: tell everyone. Can't complain. */ |
090e77c3 | 390 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); |
a8994181 | 391 | irq_unlock_sparse(); |
6a1bdc1b | 392 | goto out_release; |
8fa1d7d3 | 393 | } |
04321587 | 394 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 395 | |
48c5ccae PZ |
396 | /* |
397 | * The migration_call() CPU_DYING callback will have removed all | |
398 | * runnable tasks from the cpu, there's only the idle task left now | |
399 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
400 | * |
401 | * Wait for the stop thread to go away. | |
48c5ccae | 402 | */ |
528a25b0 | 403 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 404 | cpu_relax(); |
528a25b0 PM |
405 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
406 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 407 | |
a8994181 TG |
408 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
409 | irq_unlock_sparse(); | |
410 | ||
345527b1 | 411 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
412 | /* This actually kills the CPU. */ |
413 | __cpu_die(cpu); | |
414 | ||
1da177e4 | 415 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
a49b116d | 416 | tick_cleanup_dead_cpu(cpu); |
090e77c3 | 417 | cpu_notify_nofail(CPU_DEAD, cpu); |
1da177e4 LT |
418 | |
419 | check_for_tasks(cpu); | |
420 | ||
baaca49f | 421 | out_release: |
d221938c | 422 | cpu_hotplug_done(); |
e9fb7631 | 423 | if (!err) |
090e77c3 | 424 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
e3920fb4 RW |
425 | return err; |
426 | } | |
427 | ||
71cf5aee | 428 | int cpu_down(unsigned int cpu) |
e3920fb4 | 429 | { |
9ea09af3 | 430 | int err; |
e3920fb4 | 431 | |
d221938c | 432 | cpu_maps_update_begin(); |
e761b772 MK |
433 | |
434 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 435 | err = -EBUSY; |
e761b772 MK |
436 | goto out; |
437 | } | |
438 | ||
e761b772 | 439 | err = _cpu_down(cpu, 0); |
e3920fb4 | 440 | |
e761b772 | 441 | out: |
d221938c | 442 | cpu_maps_update_done(); |
1da177e4 LT |
443 | return err; |
444 | } | |
b62b8ef9 | 445 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
446 | #endif /*CONFIG_HOTPLUG_CPU*/ |
447 | ||
00df35f9 PM |
448 | /* |
449 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
450 | */ | |
451 | static int smpboot_thread_call(struct notifier_block *nfb, | |
452 | unsigned long action, void *hcpu) | |
453 | { | |
454 | int cpu = (long)hcpu; | |
455 | ||
456 | switch (action & ~CPU_TASKS_FROZEN) { | |
457 | ||
64eaf974 | 458 | case CPU_DOWN_FAILED: |
00df35f9 PM |
459 | case CPU_ONLINE: |
460 | smpboot_unpark_threads(cpu); | |
461 | break; | |
462 | ||
463 | default: | |
464 | break; | |
465 | } | |
466 | ||
467 | return NOTIFY_OK; | |
468 | } | |
469 | ||
470 | static struct notifier_block smpboot_thread_notifier = { | |
471 | .notifier_call = smpboot_thread_call, | |
472 | .priority = CPU_PRI_SMPBOOT, | |
473 | }; | |
474 | ||
927da9df | 475 | void smpboot_thread_init(void) |
00df35f9 PM |
476 | { |
477 | register_cpu_notifier(&smpboot_thread_notifier); | |
478 | } | |
479 | ||
e3920fb4 | 480 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 481 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 482 | { |
3bb5d2ee | 483 | struct task_struct *idle; |
090e77c3 | 484 | int ret, nr_calls = 0; |
1da177e4 | 485 | |
d221938c | 486 | cpu_hotplug_begin(); |
38498a67 | 487 | |
5e5041f3 YI |
488 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
489 | ret = -EINVAL; | |
490 | goto out; | |
491 | } | |
492 | ||
3bb5d2ee SS |
493 | idle = idle_thread_get(cpu); |
494 | if (IS_ERR(idle)) { | |
495 | ret = PTR_ERR(idle); | |
38498a67 | 496 | goto out; |
3bb5d2ee | 497 | } |
38498a67 | 498 | |
f97f8f06 TG |
499 | ret = smpboot_create_threads(cpu); |
500 | if (ret) | |
501 | goto out; | |
502 | ||
090e77c3 TG |
503 | cpuhp_tasks_frozen = tasks_frozen; |
504 | ||
505 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
e6bde73b | 506 | if (ret) { |
a0d8cdb6 | 507 | nr_calls--; |
84117da5 FF |
508 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
509 | __func__, cpu); | |
1da177e4 LT |
510 | goto out_notify; |
511 | } | |
512 | ||
513 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 514 | ret = __cpu_up(cpu, idle); |
a8994181 | 515 | |
1da177e4 LT |
516 | if (ret != 0) |
517 | goto out_notify; | |
6978c705 | 518 | BUG_ON(!cpu_online(cpu)); |
1da177e4 LT |
519 | |
520 | /* Now call notifier in preparation. */ | |
090e77c3 | 521 | cpu_notify(CPU_ONLINE, cpu); |
1da177e4 LT |
522 | |
523 | out_notify: | |
524 | if (ret != 0) | |
090e77c3 | 525 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); |
38498a67 | 526 | out: |
d221938c | 527 | cpu_hotplug_done(); |
e3920fb4 RW |
528 | |
529 | return ret; | |
530 | } | |
531 | ||
0db0628d | 532 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
533 | { |
534 | int err = 0; | |
cf23422b | 535 | |
e0b582ec | 536 | if (!cpu_possible(cpu)) { |
84117da5 FF |
537 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
538 | cpu); | |
87d5e023 | 539 | #if defined(CONFIG_IA64) |
84117da5 | 540 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
541 | #endif |
542 | return -EINVAL; | |
543 | } | |
e3920fb4 | 544 | |
01b0f197 TK |
545 | err = try_online_node(cpu_to_node(cpu)); |
546 | if (err) | |
547 | return err; | |
cf23422b | 548 | |
d221938c | 549 | cpu_maps_update_begin(); |
e761b772 MK |
550 | |
551 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 552 | err = -EBUSY; |
e761b772 MK |
553 | goto out; |
554 | } | |
555 | ||
556 | err = _cpu_up(cpu, 0); | |
557 | ||
e761b772 | 558 | out: |
d221938c | 559 | cpu_maps_update_done(); |
e3920fb4 RW |
560 | return err; |
561 | } | |
a513f6ba | 562 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 563 | |
f3de4be9 | 564 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 565 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
566 | |
567 | int disable_nonboot_cpus(void) | |
568 | { | |
e9a5f426 | 569 | int cpu, first_cpu, error = 0; |
e3920fb4 | 570 | |
d221938c | 571 | cpu_maps_update_begin(); |
e0b582ec | 572 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
573 | /* |
574 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
575 | * with the userspace trying to use the CPU hotplug at the same time |
576 | */ | |
e0b582ec | 577 | cpumask_clear(frozen_cpus); |
6ad4c188 | 578 | |
84117da5 | 579 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
580 | for_each_online_cpu(cpu) { |
581 | if (cpu == first_cpu) | |
582 | continue; | |
bb3632c6 | 583 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 584 | error = _cpu_down(cpu, 1); |
bb3632c6 | 585 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 586 | if (!error) |
e0b582ec | 587 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 588 | else { |
84117da5 | 589 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
590 | break; |
591 | } | |
592 | } | |
86886e55 | 593 | |
89af7ba5 | 594 | if (!error) |
e3920fb4 | 595 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 596 | else |
84117da5 | 597 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
598 | |
599 | /* | |
600 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
601 | * this even in case of failure as all disable_nonboot_cpus() users are | |
602 | * supposed to do enable_nonboot_cpus() on the failure path. | |
603 | */ | |
604 | cpu_hotplug_disabled++; | |
605 | ||
d221938c | 606 | cpu_maps_update_done(); |
e3920fb4 RW |
607 | return error; |
608 | } | |
609 | ||
d0af9eed SS |
610 | void __weak arch_enable_nonboot_cpus_begin(void) |
611 | { | |
612 | } | |
613 | ||
614 | void __weak arch_enable_nonboot_cpus_end(void) | |
615 | { | |
616 | } | |
617 | ||
71cf5aee | 618 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
619 | { |
620 | int cpu, error; | |
621 | ||
622 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 623 | cpu_maps_update_begin(); |
89af7ba5 | 624 | WARN_ON(--cpu_hotplug_disabled < 0); |
e0b582ec | 625 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 626 | goto out; |
e3920fb4 | 627 | |
84117da5 | 628 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
629 | |
630 | arch_enable_nonboot_cpus_begin(); | |
631 | ||
e0b582ec | 632 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 633 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 634 | error = _cpu_up(cpu, 1); |
bb3632c6 | 635 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 636 | if (!error) { |
84117da5 | 637 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
638 | continue; |
639 | } | |
84117da5 | 640 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 641 | } |
d0af9eed SS |
642 | |
643 | arch_enable_nonboot_cpus_end(); | |
644 | ||
e0b582ec | 645 | cpumask_clear(frozen_cpus); |
1d64b9cb | 646 | out: |
d221938c | 647 | cpu_maps_update_done(); |
1da177e4 | 648 | } |
e0b582ec | 649 | |
d7268a31 | 650 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
651 | { |
652 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
653 | return -ENOMEM; | |
654 | return 0; | |
655 | } | |
656 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 657 | |
79cfbdfa SB |
658 | /* |
659 | * When callbacks for CPU hotplug notifications are being executed, we must | |
660 | * ensure that the state of the system with respect to the tasks being frozen | |
661 | * or not, as reported by the notification, remains unchanged *throughout the | |
662 | * duration* of the execution of the callbacks. | |
663 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
664 | * | |
665 | * This synchronization is implemented by mutually excluding regular CPU | |
666 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
667 | * Hibernate notifications. | |
668 | */ | |
669 | static int | |
670 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
671 | unsigned long action, void *ptr) | |
672 | { | |
673 | switch (action) { | |
674 | ||
675 | case PM_SUSPEND_PREPARE: | |
676 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 677 | cpu_hotplug_disable(); |
79cfbdfa SB |
678 | break; |
679 | ||
680 | case PM_POST_SUSPEND: | |
681 | case PM_POST_HIBERNATION: | |
16e53dbf | 682 | cpu_hotplug_enable(); |
79cfbdfa SB |
683 | break; |
684 | ||
685 | default: | |
686 | return NOTIFY_DONE; | |
687 | } | |
688 | ||
689 | return NOTIFY_OK; | |
690 | } | |
691 | ||
692 | ||
d7268a31 | 693 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 694 | { |
6e32d479 FY |
695 | /* |
696 | * cpu_hotplug_pm_callback has higher priority than x86 | |
697 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
698 | * to disable cpu hotplug to avoid cpu hotplug race. | |
699 | */ | |
79cfbdfa SB |
700 | pm_notifier(cpu_hotplug_pm_callback, 0); |
701 | return 0; | |
702 | } | |
703 | core_initcall(cpu_hotplug_pm_sync_init); | |
704 | ||
f3de4be9 | 705 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 706 | |
e545a614 MS |
707 | /** |
708 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
709 | * @cpu: cpu that just started | |
710 | * | |
711 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
712 | * It must be called by the arch code on the new cpu, before the new cpu | |
713 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
714 | */ | |
0db0628d | 715 | void notify_cpu_starting(unsigned int cpu) |
e545a614 | 716 | { |
090e77c3 | 717 | cpu_notify(CPU_STARTING, cpu); |
e545a614 MS |
718 | } |
719 | ||
68f4f1ec | 720 | #endif /* CONFIG_SMP */ |
b8d317d1 | 721 | |
e56b3bc7 LT |
722 | /* |
723 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
724 | * represents all NR_CPUS bits binary values of 1<<nr. | |
725 | * | |
e0b582ec | 726 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
727 | * mask value that has a single bit set only. |
728 | */ | |
b8d317d1 | 729 | |
e56b3bc7 | 730 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 731 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
732 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
733 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
734 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 735 | |
e56b3bc7 LT |
736 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
737 | ||
738 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
739 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
740 | #if BITS_PER_LONG > 32 | |
741 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
742 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
743 | #endif |
744 | }; | |
e56b3bc7 | 745 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
746 | |
747 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
748 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
749 | |
750 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 751 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 752 | = {CPU_BITS_ALL}; |
b3199c02 | 753 | #else |
4b804c85 | 754 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 755 | #endif |
4b804c85 | 756 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 757 | |
4b804c85 RV |
758 | struct cpumask __cpu_online_mask __read_mostly; |
759 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 760 | |
4b804c85 RV |
761 | struct cpumask __cpu_present_mask __read_mostly; |
762 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 763 | |
4b804c85 RV |
764 | struct cpumask __cpu_active_mask __read_mostly; |
765 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 766 | |
3fa41520 RR |
767 | void init_cpu_present(const struct cpumask *src) |
768 | { | |
c4c54dd1 | 769 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
770 | } |
771 | ||
772 | void init_cpu_possible(const struct cpumask *src) | |
773 | { | |
c4c54dd1 | 774 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
775 | } |
776 | ||
777 | void init_cpu_online(const struct cpumask *src) | |
778 | { | |
c4c54dd1 | 779 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 780 | } |