]>
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> |
cff7d378 | 25 | |
bb3632c6 | 26 | #include <trace/events/power.h> |
cff7d378 TG |
27 | #define CREATE_TRACE_POINTS |
28 | #include <trace/events/cpuhp.h> | |
1da177e4 | 29 | |
38498a67 TG |
30 | #include "smpboot.h" |
31 | ||
cff7d378 TG |
32 | /** |
33 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
34 | * @state: The current cpu state | |
35 | * @target: The target state | |
36 | */ | |
37 | struct cpuhp_cpu_state { | |
38 | enum cpuhp_state state; | |
39 | enum cpuhp_state target; | |
40 | }; | |
41 | ||
42 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
43 | ||
44 | /** | |
45 | * cpuhp_step - Hotplug state machine step | |
46 | * @name: Name of the step | |
47 | * @startup: Startup function of the step | |
48 | * @teardown: Teardown function of the step | |
49 | * @skip_onerr: Do not invoke the functions on error rollback | |
50 | * Will go away once the notifiers are gone | |
51 | */ | |
52 | struct cpuhp_step { | |
53 | const char *name; | |
54 | int (*startup)(unsigned int cpu); | |
55 | int (*teardown)(unsigned int cpu); | |
56 | bool skip_onerr; | |
57 | }; | |
58 | ||
59 | static struct cpuhp_step cpuhp_bp_states[]; | |
60 | ||
61 | /** | |
62 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
63 | * @cpu: The cpu for which the callback should be invoked | |
64 | * @step: The step in the state machine | |
65 | * @cb: The callback function to invoke | |
66 | * | |
67 | * Called from cpu hotplug and from the state register machinery | |
68 | */ | |
69 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step, | |
70 | int (*cb)(unsigned int)) | |
71 | { | |
72 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
73 | int ret = 0; | |
74 | ||
75 | if (cb) { | |
76 | trace_cpuhp_enter(cpu, st->target, step, cb); | |
77 | ret = cb(cpu); | |
78 | trace_cpuhp_exit(cpu, st->state, step, ret); | |
79 | } | |
80 | return ret; | |
81 | } | |
82 | ||
98a79d6a | 83 | #ifdef CONFIG_SMP |
b3199c02 | 84 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 85 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
86 | bool cpuhp_tasks_frozen; |
87 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 88 | |
79a6cdeb | 89 | /* |
93ae4f97 SB |
90 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
91 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
92 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
93 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
94 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
95 | */ |
96 | void cpu_maps_update_begin(void) | |
97 | { | |
98 | mutex_lock(&cpu_add_remove_lock); | |
99 | } | |
93ae4f97 | 100 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
101 | |
102 | void cpu_maps_update_done(void) | |
103 | { | |
104 | mutex_unlock(&cpu_add_remove_lock); | |
105 | } | |
93ae4f97 | 106 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 107 | |
5c113fbe | 108 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 109 | |
e3920fb4 RW |
110 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
111 | * Should always be manipulated under cpu_add_remove_lock | |
112 | */ | |
113 | static int cpu_hotplug_disabled; | |
114 | ||
79a6cdeb LJ |
115 | #ifdef CONFIG_HOTPLUG_CPU |
116 | ||
d221938c GS |
117 | static struct { |
118 | struct task_struct *active_writer; | |
87af9e7f DH |
119 | /* wait queue to wake up the active_writer */ |
120 | wait_queue_head_t wq; | |
121 | /* verifies that no writer will get active while readers are active */ | |
122 | struct mutex lock; | |
d221938c GS |
123 | /* |
124 | * Also blocks the new readers during | |
125 | * an ongoing cpu hotplug operation. | |
126 | */ | |
87af9e7f | 127 | atomic_t refcount; |
a19423b9 GS |
128 | |
129 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
130 | struct lockdep_map dep_map; | |
131 | #endif | |
31950eb6 LT |
132 | } cpu_hotplug = { |
133 | .active_writer = NULL, | |
87af9e7f | 134 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 135 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
136 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
137 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
138 | #endif | |
31950eb6 | 139 | }; |
d221938c | 140 | |
a19423b9 GS |
141 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
142 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
143 | #define cpuhp_lock_acquire_tryread() \ |
144 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
145 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
146 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
147 | ||
62db99f4 | 148 | |
86ef5c9a | 149 | void get_online_cpus(void) |
a9d9baa1 | 150 | { |
d221938c GS |
151 | might_sleep(); |
152 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 153 | return; |
a19423b9 | 154 | cpuhp_lock_acquire_read(); |
d221938c | 155 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 156 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 157 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 158 | } |
86ef5c9a | 159 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 160 | |
86ef5c9a | 161 | void put_online_cpus(void) |
a9d9baa1 | 162 | { |
87af9e7f DH |
163 | int refcount; |
164 | ||
d221938c | 165 | if (cpu_hotplug.active_writer == current) |
aa953877 | 166 | return; |
075663d1 | 167 | |
87af9e7f DH |
168 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
169 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
170 | atomic_inc(&cpu_hotplug.refcount); | |
171 | ||
172 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
173 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 174 | |
a19423b9 | 175 | cpuhp_lock_release(); |
d221938c | 176 | |
a9d9baa1 | 177 | } |
86ef5c9a | 178 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 179 | |
d221938c GS |
180 | /* |
181 | * This ensures that the hotplug operation can begin only when the | |
182 | * refcount goes to zero. | |
183 | * | |
184 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
185 | * will be blocked by the cpu_hotplug.lock | |
186 | * | |
d2ba7e2a ON |
187 | * Since cpu_hotplug_begin() is always called after invoking |
188 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
189 | * |
190 | * Note that theoretically, there is a possibility of a livelock: | |
191 | * - Refcount goes to zero, last reader wakes up the sleeping | |
192 | * writer. | |
193 | * - Last reader unlocks the cpu_hotplug.lock. | |
194 | * - A new reader arrives at this moment, bumps up the refcount. | |
195 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
196 | * non zero and goes to sleep again. | |
197 | * | |
198 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 199 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
200 | * |
201 | */ | |
b9d10be7 | 202 | void cpu_hotplug_begin(void) |
d221938c | 203 | { |
87af9e7f | 204 | DEFINE_WAIT(wait); |
d2ba7e2a | 205 | |
87af9e7f | 206 | cpu_hotplug.active_writer = current; |
a19423b9 | 207 | cpuhp_lock_acquire(); |
87af9e7f | 208 | |
d2ba7e2a ON |
209 | for (;;) { |
210 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
211 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
212 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
213 | break; | |
d221938c GS |
214 | mutex_unlock(&cpu_hotplug.lock); |
215 | schedule(); | |
d221938c | 216 | } |
87af9e7f | 217 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
218 | } |
219 | ||
b9d10be7 | 220 | void cpu_hotplug_done(void) |
d221938c GS |
221 | { |
222 | cpu_hotplug.active_writer = NULL; | |
223 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 224 | cpuhp_lock_release(); |
d221938c | 225 | } |
79a6cdeb | 226 | |
16e53dbf SB |
227 | /* |
228 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
229 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
230 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
231 | * hotplug path before performing hotplug operations. So acquiring that lock | |
232 | * guarantees mutual exclusion from any currently running hotplug operations. | |
233 | */ | |
234 | void cpu_hotplug_disable(void) | |
235 | { | |
236 | cpu_maps_update_begin(); | |
89af7ba5 | 237 | cpu_hotplug_disabled++; |
16e53dbf SB |
238 | cpu_maps_update_done(); |
239 | } | |
32145c46 | 240 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf SB |
241 | |
242 | void cpu_hotplug_enable(void) | |
243 | { | |
244 | cpu_maps_update_begin(); | |
89af7ba5 | 245 | WARN_ON(--cpu_hotplug_disabled < 0); |
16e53dbf SB |
246 | cpu_maps_update_done(); |
247 | } | |
32145c46 | 248 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 249 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 250 | |
1da177e4 | 251 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 252 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 253 | { |
bd5349cf | 254 | int ret; |
d221938c | 255 | cpu_maps_update_begin(); |
bd5349cf | 256 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 257 | cpu_maps_update_done(); |
bd5349cf | 258 | return ret; |
1da177e4 | 259 | } |
65edc68c | 260 | |
71cf5aee | 261 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
262 | { |
263 | return raw_notifier_chain_register(&cpu_chain, nb); | |
264 | } | |
265 | ||
090e77c3 | 266 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
267 | int *nr_calls) |
268 | { | |
090e77c3 TG |
269 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
270 | void *hcpu = (void *)(long)cpu; | |
271 | ||
e6bde73b AM |
272 | int ret; |
273 | ||
090e77c3 | 274 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 275 | nr_calls); |
e6bde73b AM |
276 | |
277 | return notifier_to_errno(ret); | |
e9fb7631 AM |
278 | } |
279 | ||
090e77c3 | 280 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 281 | { |
090e77c3 | 282 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
283 | } |
284 | ||
ba997462 TG |
285 | /* Notifier wrappers for transitioning to state machine */ |
286 | static int notify_prepare(unsigned int cpu) | |
287 | { | |
288 | int nr_calls = 0; | |
289 | int ret; | |
290 | ||
291 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
292 | if (ret) { | |
293 | nr_calls--; | |
294 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
295 | __func__, cpu); | |
296 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
297 | } | |
298 | return ret; | |
299 | } | |
300 | ||
301 | static int notify_online(unsigned int cpu) | |
302 | { | |
303 | cpu_notify(CPU_ONLINE, cpu); | |
304 | return 0; | |
305 | } | |
306 | ||
307 | static int bringup_cpu(unsigned int cpu) | |
308 | { | |
309 | struct task_struct *idle = idle_thread_get(cpu); | |
310 | int ret; | |
311 | ||
312 | /* Arch-specific enabling code. */ | |
313 | ret = __cpu_up(cpu, idle); | |
314 | if (ret) { | |
315 | cpu_notify(CPU_UP_CANCELED, cpu); | |
316 | return ret; | |
317 | } | |
318 | BUG_ON(!cpu_online(cpu)); | |
319 | return 0; | |
320 | } | |
321 | ||
00b9b0af | 322 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 323 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 324 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 325 | |
71cf5aee | 326 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 327 | { |
d221938c | 328 | cpu_maps_update_begin(); |
bd5349cf | 329 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 330 | cpu_maps_update_done(); |
1da177e4 LT |
331 | } |
332 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
333 | ||
71cf5aee | 334 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
335 | { |
336 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
337 | } | |
338 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
339 | ||
e4cc2f87 AV |
340 | /** |
341 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
342 | * @cpu: a CPU id | |
343 | * | |
344 | * This function walks all processes, finds a valid mm struct for each one and | |
345 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
346 | * trivial, there are various non-obvious corner cases, which this function | |
347 | * tries to solve in a safe manner. | |
348 | * | |
349 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
350 | * be called only for an already offlined CPU. | |
351 | */ | |
cb79295e AV |
352 | void clear_tasks_mm_cpumask(int cpu) |
353 | { | |
354 | struct task_struct *p; | |
355 | ||
356 | /* | |
357 | * This function is called after the cpu is taken down and marked | |
358 | * offline, so its not like new tasks will ever get this cpu set in | |
359 | * their mm mask. -- Peter Zijlstra | |
360 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
361 | * full-fledged tasklist_lock. | |
362 | */ | |
e4cc2f87 | 363 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
364 | rcu_read_lock(); |
365 | for_each_process(p) { | |
366 | struct task_struct *t; | |
367 | ||
e4cc2f87 AV |
368 | /* |
369 | * Main thread might exit, but other threads may still have | |
370 | * a valid mm. Find one. | |
371 | */ | |
cb79295e AV |
372 | t = find_lock_task_mm(p); |
373 | if (!t) | |
374 | continue; | |
375 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
376 | task_unlock(t); | |
377 | } | |
378 | rcu_read_unlock(); | |
379 | } | |
380 | ||
b728ca06 | 381 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 382 | { |
b728ca06 | 383 | struct task_struct *g, *p; |
1da177e4 | 384 | |
a75a6068 ON |
385 | read_lock(&tasklist_lock); |
386 | for_each_process_thread(g, p) { | |
b728ca06 KT |
387 | if (!p->on_rq) |
388 | continue; | |
389 | /* | |
390 | * We do the check with unlocked task_rq(p)->lock. | |
391 | * Order the reading to do not warn about a task, | |
392 | * which was running on this cpu in the past, and | |
393 | * it's just been woken on another cpu. | |
394 | */ | |
395 | rmb(); | |
396 | if (task_cpu(p) != dead_cpu) | |
397 | continue; | |
398 | ||
399 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
400 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
401 | } |
402 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
403 | } |
404 | ||
98458172 TG |
405 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
406 | { | |
407 | BUG_ON(cpu_notify(val, cpu)); | |
408 | } | |
409 | ||
410 | static int notify_down_prepare(unsigned int cpu) | |
411 | { | |
412 | int err, nr_calls = 0; | |
413 | ||
414 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
415 | if (err) { | |
416 | nr_calls--; | |
417 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
418 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
419 | __func__, cpu); | |
420 | } | |
421 | return err; | |
422 | } | |
423 | ||
1da177e4 | 424 | /* Take this CPU down. */ |
71cf5aee | 425 | static int take_cpu_down(void *_param) |
1da177e4 | 426 | { |
090e77c3 | 427 | int err, cpu = smp_processor_id(); |
1da177e4 | 428 | |
1da177e4 LT |
429 | /* Ensure this CPU doesn't handle any more interrupts. */ |
430 | err = __cpu_disable(); | |
431 | if (err < 0) | |
f3705136 | 432 | return err; |
1da177e4 | 433 | |
090e77c3 | 434 | cpu_notify(CPU_DYING, cpu); |
52c063d1 TG |
435 | /* Give up timekeeping duties */ |
436 | tick_handover_do_timer(); | |
14e568e7 | 437 | /* Park the stopper thread */ |
090e77c3 | 438 | stop_machine_park(cpu); |
f3705136 | 439 | return 0; |
1da177e4 LT |
440 | } |
441 | ||
98458172 | 442 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 443 | { |
98458172 | 444 | int err; |
1da177e4 | 445 | |
6acce3ef PZ |
446 | /* |
447 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
448 | * and RCU users of this state to go away such that all new such users | |
449 | * will observe it. | |
450 | * | |
451 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
779de6ce | 452 | * not imply sync_sched(), so wait for both. |
106dd5af M |
453 | * |
454 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef | 455 | */ |
779de6ce PM |
456 | if (IS_ENABLED(CONFIG_PREEMPT)) |
457 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
458 | else | |
459 | synchronize_rcu(); | |
6acce3ef | 460 | |
106dd5af M |
461 | smpboot_park_threads(cpu); |
462 | ||
6acce3ef | 463 | /* |
a8994181 TG |
464 | * Prevent irq alloc/free while the dying cpu reorganizes the |
465 | * interrupt affinities. | |
6acce3ef | 466 | */ |
a8994181 | 467 | irq_lock_sparse(); |
6acce3ef | 468 | |
a8994181 TG |
469 | /* |
470 | * So now all preempt/rcu users must observe !cpu_active(). | |
471 | */ | |
090e77c3 | 472 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 473 | if (err) { |
1da177e4 | 474 | /* CPU didn't die: tell everyone. Can't complain. */ |
090e77c3 | 475 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); |
a8994181 | 476 | irq_unlock_sparse(); |
98458172 | 477 | return err; |
8fa1d7d3 | 478 | } |
04321587 | 479 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 480 | |
48c5ccae PZ |
481 | /* |
482 | * The migration_call() CPU_DYING callback will have removed all | |
483 | * runnable tasks from the cpu, there's only the idle task left now | |
484 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
485 | * |
486 | * Wait for the stop thread to go away. | |
48c5ccae | 487 | */ |
528a25b0 | 488 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 489 | cpu_relax(); |
528a25b0 PM |
490 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
491 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 492 | |
a8994181 TG |
493 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
494 | irq_unlock_sparse(); | |
495 | ||
345527b1 | 496 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
497 | /* This actually kills the CPU. */ |
498 | __cpu_die(cpu); | |
499 | ||
a49b116d | 500 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
501 | return 0; |
502 | } | |
1da177e4 | 503 | |
98458172 TG |
504 | static int notify_dead(unsigned int cpu) |
505 | { | |
506 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 507 | check_for_tasks(cpu); |
98458172 TG |
508 | return 0; |
509 | } | |
510 | ||
cff7d378 TG |
511 | #else |
512 | #define notify_down_prepare NULL | |
513 | #define takedown_cpu NULL | |
514 | #define notify_dead NULL | |
515 | #endif | |
516 | ||
517 | #ifdef CONFIG_HOTPLUG_CPU | |
518 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) | |
519 | { | |
520 | for (st->state++; st->state < st->target; st->state++) { | |
521 | struct cpuhp_step *step = cpuhp_bp_states + st->state; | |
522 | ||
523 | if (!step->skip_onerr) | |
524 | cpuhp_invoke_callback(cpu, st->state, step->startup); | |
525 | } | |
526 | } | |
527 | ||
98458172 TG |
528 | /* Requires cpu_add_remove_lock to be held */ |
529 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) | |
530 | { | |
cff7d378 TG |
531 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
532 | int prev_state, ret = 0; | |
533 | bool hasdied = false; | |
98458172 TG |
534 | |
535 | if (num_online_cpus() == 1) | |
536 | return -EBUSY; | |
537 | ||
538 | if (!cpu_online(cpu)) | |
539 | return -EINVAL; | |
540 | ||
541 | cpu_hotplug_begin(); | |
542 | ||
543 | cpuhp_tasks_frozen = tasks_frozen; | |
544 | ||
cff7d378 TG |
545 | prev_state = st->state; |
546 | st->target = CPUHP_OFFLINE; | |
547 | for (; st->state > st->target; st->state--) { | |
548 | struct cpuhp_step *step = cpuhp_bp_states + st->state; | |
98458172 | 549 | |
cff7d378 TG |
550 | ret = cpuhp_invoke_callback(cpu, st->state, step->teardown); |
551 | if (ret) { | |
552 | st->target = prev_state; | |
553 | undo_cpu_down(cpu, st); | |
554 | break; | |
555 | } | |
556 | } | |
557 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; | |
1da177e4 | 558 | |
d221938c | 559 | cpu_hotplug_done(); |
cff7d378 TG |
560 | /* This post dead nonsense must die */ |
561 | if (!ret && hasdied) | |
090e77c3 | 562 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 563 | return ret; |
e3920fb4 RW |
564 | } |
565 | ||
71cf5aee | 566 | int cpu_down(unsigned int cpu) |
e3920fb4 | 567 | { |
9ea09af3 | 568 | int err; |
e3920fb4 | 569 | |
d221938c | 570 | cpu_maps_update_begin(); |
e761b772 MK |
571 | |
572 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 573 | err = -EBUSY; |
e761b772 MK |
574 | goto out; |
575 | } | |
576 | ||
e761b772 | 577 | err = _cpu_down(cpu, 0); |
e3920fb4 | 578 | |
e761b772 | 579 | out: |
d221938c | 580 | cpu_maps_update_done(); |
1da177e4 LT |
581 | return err; |
582 | } | |
b62b8ef9 | 583 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
584 | #endif /*CONFIG_HOTPLUG_CPU*/ |
585 | ||
00df35f9 PM |
586 | /* |
587 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
588 | */ | |
589 | static int smpboot_thread_call(struct notifier_block *nfb, | |
590 | unsigned long action, void *hcpu) | |
591 | { | |
592 | int cpu = (long)hcpu; | |
593 | ||
594 | switch (action & ~CPU_TASKS_FROZEN) { | |
595 | ||
64eaf974 | 596 | case CPU_DOWN_FAILED: |
00df35f9 PM |
597 | case CPU_ONLINE: |
598 | smpboot_unpark_threads(cpu); | |
599 | break; | |
600 | ||
601 | default: | |
602 | break; | |
603 | } | |
604 | ||
605 | return NOTIFY_OK; | |
606 | } | |
607 | ||
608 | static struct notifier_block smpboot_thread_notifier = { | |
609 | .notifier_call = smpboot_thread_call, | |
610 | .priority = CPU_PRI_SMPBOOT, | |
611 | }; | |
612 | ||
927da9df | 613 | void smpboot_thread_init(void) |
00df35f9 PM |
614 | { |
615 | register_cpu_notifier(&smpboot_thread_notifier); | |
616 | } | |
617 | ||
cff7d378 TG |
618 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
619 | { | |
620 | for (st->state--; st->state > st->target; st->state--) { | |
621 | struct cpuhp_step *step = cpuhp_bp_states + st->state; | |
622 | ||
623 | if (!step->skip_onerr) | |
624 | cpuhp_invoke_callback(cpu, st->state, step->teardown); | |
625 | } | |
626 | } | |
627 | ||
e3920fb4 | 628 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 629 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 630 | { |
cff7d378 | 631 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 632 | struct task_struct *idle; |
cff7d378 | 633 | int prev_state, ret = 0; |
1da177e4 | 634 | |
d221938c | 635 | cpu_hotplug_begin(); |
38498a67 | 636 | |
5e5041f3 YI |
637 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
638 | ret = -EINVAL; | |
639 | goto out; | |
640 | } | |
641 | ||
cff7d378 | 642 | /* Let it fail before we try to bring the cpu up */ |
3bb5d2ee SS |
643 | idle = idle_thread_get(cpu); |
644 | if (IS_ERR(idle)) { | |
645 | ret = PTR_ERR(idle); | |
38498a67 | 646 | goto out; |
3bb5d2ee | 647 | } |
38498a67 | 648 | |
ba997462 TG |
649 | cpuhp_tasks_frozen = tasks_frozen; |
650 | ||
cff7d378 TG |
651 | prev_state = st->state; |
652 | st->target = CPUHP_ONLINE; | |
653 | while (st->state < st->target) { | |
654 | struct cpuhp_step *step; | |
655 | ||
656 | st->state++; | |
657 | step = cpuhp_bp_states + st->state; | |
658 | ret = cpuhp_invoke_callback(cpu, st->state, step->startup); | |
659 | if (ret) { | |
660 | st->target = prev_state; | |
661 | undo_cpu_up(cpu, st); | |
662 | break; | |
663 | } | |
664 | } | |
38498a67 | 665 | out: |
d221938c | 666 | cpu_hotplug_done(); |
e3920fb4 RW |
667 | return ret; |
668 | } | |
669 | ||
0db0628d | 670 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
671 | { |
672 | int err = 0; | |
cf23422b | 673 | |
e0b582ec | 674 | if (!cpu_possible(cpu)) { |
84117da5 FF |
675 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
676 | cpu); | |
87d5e023 | 677 | #if defined(CONFIG_IA64) |
84117da5 | 678 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
679 | #endif |
680 | return -EINVAL; | |
681 | } | |
e3920fb4 | 682 | |
01b0f197 TK |
683 | err = try_online_node(cpu_to_node(cpu)); |
684 | if (err) | |
685 | return err; | |
cf23422b | 686 | |
d221938c | 687 | cpu_maps_update_begin(); |
e761b772 MK |
688 | |
689 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 690 | err = -EBUSY; |
e761b772 MK |
691 | goto out; |
692 | } | |
693 | ||
694 | err = _cpu_up(cpu, 0); | |
695 | ||
e761b772 | 696 | out: |
d221938c | 697 | cpu_maps_update_done(); |
e3920fb4 RW |
698 | return err; |
699 | } | |
a513f6ba | 700 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 701 | |
f3de4be9 | 702 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 703 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
704 | |
705 | int disable_nonboot_cpus(void) | |
706 | { | |
e9a5f426 | 707 | int cpu, first_cpu, error = 0; |
e3920fb4 | 708 | |
d221938c | 709 | cpu_maps_update_begin(); |
e0b582ec | 710 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
711 | /* |
712 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
713 | * with the userspace trying to use the CPU hotplug at the same time |
714 | */ | |
e0b582ec | 715 | cpumask_clear(frozen_cpus); |
6ad4c188 | 716 | |
84117da5 | 717 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
718 | for_each_online_cpu(cpu) { |
719 | if (cpu == first_cpu) | |
720 | continue; | |
bb3632c6 | 721 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 722 | error = _cpu_down(cpu, 1); |
bb3632c6 | 723 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 724 | if (!error) |
e0b582ec | 725 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 726 | else { |
84117da5 | 727 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
728 | break; |
729 | } | |
730 | } | |
86886e55 | 731 | |
89af7ba5 | 732 | if (!error) |
e3920fb4 | 733 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 734 | else |
84117da5 | 735 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
736 | |
737 | /* | |
738 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
739 | * this even in case of failure as all disable_nonboot_cpus() users are | |
740 | * supposed to do enable_nonboot_cpus() on the failure path. | |
741 | */ | |
742 | cpu_hotplug_disabled++; | |
743 | ||
d221938c | 744 | cpu_maps_update_done(); |
e3920fb4 RW |
745 | return error; |
746 | } | |
747 | ||
d0af9eed SS |
748 | void __weak arch_enable_nonboot_cpus_begin(void) |
749 | { | |
750 | } | |
751 | ||
752 | void __weak arch_enable_nonboot_cpus_end(void) | |
753 | { | |
754 | } | |
755 | ||
71cf5aee | 756 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
757 | { |
758 | int cpu, error; | |
759 | ||
760 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 761 | cpu_maps_update_begin(); |
89af7ba5 | 762 | WARN_ON(--cpu_hotplug_disabled < 0); |
e0b582ec | 763 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 764 | goto out; |
e3920fb4 | 765 | |
84117da5 | 766 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
767 | |
768 | arch_enable_nonboot_cpus_begin(); | |
769 | ||
e0b582ec | 770 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 771 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 772 | error = _cpu_up(cpu, 1); |
bb3632c6 | 773 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 774 | if (!error) { |
84117da5 | 775 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
776 | continue; |
777 | } | |
84117da5 | 778 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 779 | } |
d0af9eed SS |
780 | |
781 | arch_enable_nonboot_cpus_end(); | |
782 | ||
e0b582ec | 783 | cpumask_clear(frozen_cpus); |
1d64b9cb | 784 | out: |
d221938c | 785 | cpu_maps_update_done(); |
1da177e4 | 786 | } |
e0b582ec | 787 | |
d7268a31 | 788 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
789 | { |
790 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
791 | return -ENOMEM; | |
792 | return 0; | |
793 | } | |
794 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 795 | |
79cfbdfa SB |
796 | /* |
797 | * When callbacks for CPU hotplug notifications are being executed, we must | |
798 | * ensure that the state of the system with respect to the tasks being frozen | |
799 | * or not, as reported by the notification, remains unchanged *throughout the | |
800 | * duration* of the execution of the callbacks. | |
801 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
802 | * | |
803 | * This synchronization is implemented by mutually excluding regular CPU | |
804 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
805 | * Hibernate notifications. | |
806 | */ | |
807 | static int | |
808 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
809 | unsigned long action, void *ptr) | |
810 | { | |
811 | switch (action) { | |
812 | ||
813 | case PM_SUSPEND_PREPARE: | |
814 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 815 | cpu_hotplug_disable(); |
79cfbdfa SB |
816 | break; |
817 | ||
818 | case PM_POST_SUSPEND: | |
819 | case PM_POST_HIBERNATION: | |
16e53dbf | 820 | cpu_hotplug_enable(); |
79cfbdfa SB |
821 | break; |
822 | ||
823 | default: | |
824 | return NOTIFY_DONE; | |
825 | } | |
826 | ||
827 | return NOTIFY_OK; | |
828 | } | |
829 | ||
830 | ||
d7268a31 | 831 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 832 | { |
6e32d479 FY |
833 | /* |
834 | * cpu_hotplug_pm_callback has higher priority than x86 | |
835 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
836 | * to disable cpu hotplug to avoid cpu hotplug race. | |
837 | */ | |
79cfbdfa SB |
838 | pm_notifier(cpu_hotplug_pm_callback, 0); |
839 | return 0; | |
840 | } | |
841 | core_initcall(cpu_hotplug_pm_sync_init); | |
842 | ||
f3de4be9 | 843 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 844 | |
e545a614 MS |
845 | /** |
846 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
847 | * @cpu: cpu that just started | |
848 | * | |
849 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
850 | * It must be called by the arch code on the new cpu, before the new cpu | |
851 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
852 | */ | |
0db0628d | 853 | void notify_cpu_starting(unsigned int cpu) |
e545a614 | 854 | { |
090e77c3 | 855 | cpu_notify(CPU_STARTING, cpu); |
e545a614 MS |
856 | } |
857 | ||
68f4f1ec | 858 | #endif /* CONFIG_SMP */ |
b8d317d1 | 859 | |
cff7d378 TG |
860 | /* Boot processor state steps */ |
861 | static struct cpuhp_step cpuhp_bp_states[] = { | |
862 | [CPUHP_OFFLINE] = { | |
863 | .name = "offline", | |
864 | .startup = NULL, | |
865 | .teardown = NULL, | |
866 | }, | |
867 | #ifdef CONFIG_SMP | |
868 | [CPUHP_CREATE_THREADS]= { | |
869 | .name = "threads:create", | |
870 | .startup = smpboot_create_threads, | |
871 | .teardown = NULL, | |
872 | }, | |
873 | [CPUHP_NOTIFY_PREPARE] = { | |
874 | .name = "notify:prepare", | |
875 | .startup = notify_prepare, | |
876 | .teardown = notify_dead, | |
877 | .skip_onerr = true, | |
878 | }, | |
879 | [CPUHP_BRINGUP_CPU] = { | |
880 | .name = "cpu:bringup", | |
881 | .startup = bringup_cpu, | |
882 | .teardown = takedown_cpu, | |
883 | .skip_onerr = true, | |
884 | }, | |
885 | [CPUHP_NOTIFY_ONLINE] = { | |
886 | .name = "notify:online", | |
887 | .startup = notify_online, | |
888 | .teardown = notify_down_prepare, | |
889 | }, | |
890 | #endif | |
891 | [CPUHP_ONLINE] = { | |
892 | .name = "online", | |
893 | .startup = NULL, | |
894 | .teardown = NULL, | |
895 | }, | |
896 | }; | |
897 | ||
e56b3bc7 LT |
898 | /* |
899 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
900 | * represents all NR_CPUS bits binary values of 1<<nr. | |
901 | * | |
e0b582ec | 902 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
903 | * mask value that has a single bit set only. |
904 | */ | |
b8d317d1 | 905 | |
e56b3bc7 | 906 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 907 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
908 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
909 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
910 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 911 | |
e56b3bc7 LT |
912 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
913 | ||
914 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
915 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
916 | #if BITS_PER_LONG > 32 | |
917 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
918 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
919 | #endif |
920 | }; | |
e56b3bc7 | 921 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
922 | |
923 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
924 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
925 | |
926 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 927 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 928 | = {CPU_BITS_ALL}; |
b3199c02 | 929 | #else |
4b804c85 | 930 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 931 | #endif |
4b804c85 | 932 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 933 | |
4b804c85 RV |
934 | struct cpumask __cpu_online_mask __read_mostly; |
935 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 936 | |
4b804c85 RV |
937 | struct cpumask __cpu_present_mask __read_mostly; |
938 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 939 | |
4b804c85 RV |
940 | struct cpumask __cpu_active_mask __read_mostly; |
941 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 942 | |
3fa41520 RR |
943 | void init_cpu_present(const struct cpumask *src) |
944 | { | |
c4c54dd1 | 945 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
946 | } |
947 | ||
948 | void init_cpu_possible(const struct cpumask *src) | |
949 | { | |
c4c54dd1 | 950 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
951 | } |
952 | ||
953 | void init_cpu_online(const struct cpumask *src) | |
954 | { | |
c4c54dd1 | 955 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 956 | } |
cff7d378 TG |
957 | |
958 | /* | |
959 | * Activate the first processor. | |
960 | */ | |
961 | void __init boot_cpu_init(void) | |
962 | { | |
963 | int cpu = smp_processor_id(); | |
964 | ||
965 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
966 | set_cpu_online(cpu, true); | |
967 | set_cpu_active(cpu, true); | |
968 | set_cpu_present(cpu, true); | |
969 | set_cpu_possible(cpu, true); | |
970 | } | |
971 | ||
972 | /* | |
973 | * Must be called _AFTER_ setting up the per_cpu areas | |
974 | */ | |
975 | void __init boot_cpu_state_init(void) | |
976 | { | |
977 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
978 | } |