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); |
1da177e4 | 32 | |
79a6cdeb | 33 | /* |
93ae4f97 SB |
34 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
35 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
36 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
37 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
38 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
39 | */ |
40 | void cpu_maps_update_begin(void) | |
41 | { | |
42 | mutex_lock(&cpu_add_remove_lock); | |
43 | } | |
93ae4f97 | 44 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
45 | |
46 | void cpu_maps_update_done(void) | |
47 | { | |
48 | mutex_unlock(&cpu_add_remove_lock); | |
49 | } | |
93ae4f97 | 50 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 51 | |
5c113fbe | 52 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 53 | |
e3920fb4 RW |
54 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
55 | * Should always be manipulated under cpu_add_remove_lock | |
56 | */ | |
57 | static int cpu_hotplug_disabled; | |
58 | ||
79a6cdeb LJ |
59 | #ifdef CONFIG_HOTPLUG_CPU |
60 | ||
d221938c GS |
61 | static struct { |
62 | struct task_struct *active_writer; | |
87af9e7f DH |
63 | /* wait queue to wake up the active_writer */ |
64 | wait_queue_head_t wq; | |
65 | /* verifies that no writer will get active while readers are active */ | |
66 | struct mutex lock; | |
d221938c GS |
67 | /* |
68 | * Also blocks the new readers during | |
69 | * an ongoing cpu hotplug operation. | |
70 | */ | |
87af9e7f | 71 | atomic_t refcount; |
a19423b9 GS |
72 | |
73 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
74 | struct lockdep_map dep_map; | |
75 | #endif | |
31950eb6 LT |
76 | } cpu_hotplug = { |
77 | .active_writer = NULL, | |
87af9e7f | 78 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 79 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
80 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
81 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
82 | #endif | |
31950eb6 | 83 | }; |
d221938c | 84 | |
a19423b9 GS |
85 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
86 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
87 | #define cpuhp_lock_acquire_tryread() \ |
88 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
89 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
90 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
91 | ||
62db99f4 | 92 | |
86ef5c9a | 93 | void get_online_cpus(void) |
a9d9baa1 | 94 | { |
d221938c GS |
95 | might_sleep(); |
96 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 97 | return; |
a19423b9 | 98 | cpuhp_lock_acquire_read(); |
d221938c | 99 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 100 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 101 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 102 | } |
86ef5c9a | 103 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 104 | |
86ef5c9a | 105 | void put_online_cpus(void) |
a9d9baa1 | 106 | { |
87af9e7f DH |
107 | int refcount; |
108 | ||
d221938c | 109 | if (cpu_hotplug.active_writer == current) |
aa953877 | 110 | return; |
075663d1 | 111 | |
87af9e7f DH |
112 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
113 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
114 | atomic_inc(&cpu_hotplug.refcount); | |
115 | ||
116 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
117 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 118 | |
a19423b9 | 119 | cpuhp_lock_release(); |
d221938c | 120 | |
a9d9baa1 | 121 | } |
86ef5c9a | 122 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 123 | |
d221938c GS |
124 | /* |
125 | * This ensures that the hotplug operation can begin only when the | |
126 | * refcount goes to zero. | |
127 | * | |
128 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
129 | * will be blocked by the cpu_hotplug.lock | |
130 | * | |
d2ba7e2a ON |
131 | * Since cpu_hotplug_begin() is always called after invoking |
132 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
133 | * |
134 | * Note that theoretically, there is a possibility of a livelock: | |
135 | * - Refcount goes to zero, last reader wakes up the sleeping | |
136 | * writer. | |
137 | * - Last reader unlocks the cpu_hotplug.lock. | |
138 | * - A new reader arrives at this moment, bumps up the refcount. | |
139 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
140 | * non zero and goes to sleep again. | |
141 | * | |
142 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 143 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
144 | * |
145 | */ | |
b9d10be7 | 146 | void cpu_hotplug_begin(void) |
d221938c | 147 | { |
87af9e7f | 148 | DEFINE_WAIT(wait); |
d2ba7e2a | 149 | |
87af9e7f | 150 | cpu_hotplug.active_writer = current; |
a19423b9 | 151 | cpuhp_lock_acquire(); |
87af9e7f | 152 | |
d2ba7e2a ON |
153 | for (;;) { |
154 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
155 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
156 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
157 | break; | |
d221938c GS |
158 | mutex_unlock(&cpu_hotplug.lock); |
159 | schedule(); | |
d221938c | 160 | } |
87af9e7f | 161 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
162 | } |
163 | ||
b9d10be7 | 164 | void cpu_hotplug_done(void) |
d221938c GS |
165 | { |
166 | cpu_hotplug.active_writer = NULL; | |
167 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 168 | cpuhp_lock_release(); |
d221938c | 169 | } |
79a6cdeb | 170 | |
16e53dbf SB |
171 | /* |
172 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
173 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
174 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
175 | * hotplug path before performing hotplug operations. So acquiring that lock | |
176 | * guarantees mutual exclusion from any currently running hotplug operations. | |
177 | */ | |
178 | void cpu_hotplug_disable(void) | |
179 | { | |
180 | cpu_maps_update_begin(); | |
89af7ba5 | 181 | cpu_hotplug_disabled++; |
16e53dbf SB |
182 | cpu_maps_update_done(); |
183 | } | |
32145c46 | 184 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf SB |
185 | |
186 | void cpu_hotplug_enable(void) | |
187 | { | |
188 | cpu_maps_update_begin(); | |
89af7ba5 | 189 | WARN_ON(--cpu_hotplug_disabled < 0); |
16e53dbf SB |
190 | cpu_maps_update_done(); |
191 | } | |
32145c46 | 192 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 193 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 194 | |
1da177e4 | 195 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 196 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 197 | { |
bd5349cf | 198 | int ret; |
d221938c | 199 | cpu_maps_update_begin(); |
bd5349cf | 200 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 201 | cpu_maps_update_done(); |
bd5349cf | 202 | return ret; |
1da177e4 | 203 | } |
65edc68c | 204 | |
71cf5aee | 205 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
206 | { |
207 | return raw_notifier_chain_register(&cpu_chain, nb); | |
208 | } | |
209 | ||
e9fb7631 AM |
210 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
211 | int *nr_calls) | |
212 | { | |
e6bde73b AM |
213 | int ret; |
214 | ||
215 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 216 | nr_calls); |
e6bde73b AM |
217 | |
218 | return notifier_to_errno(ret); | |
e9fb7631 AM |
219 | } |
220 | ||
221 | static int cpu_notify(unsigned long val, void *v) | |
222 | { | |
223 | return __cpu_notify(val, v, -1, NULL); | |
224 | } | |
225 | ||
00b9b0af LT |
226 | #ifdef CONFIG_HOTPLUG_CPU |
227 | ||
e9fb7631 AM |
228 | static void cpu_notify_nofail(unsigned long val, void *v) |
229 | { | |
00b9b0af | 230 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 231 | } |
1da177e4 | 232 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 233 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 234 | |
71cf5aee | 235 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 236 | { |
d221938c | 237 | cpu_maps_update_begin(); |
bd5349cf | 238 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 239 | cpu_maps_update_done(); |
1da177e4 LT |
240 | } |
241 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
242 | ||
71cf5aee | 243 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
244 | { |
245 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
246 | } | |
247 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
248 | ||
e4cc2f87 AV |
249 | /** |
250 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
251 | * @cpu: a CPU id | |
252 | * | |
253 | * This function walks all processes, finds a valid mm struct for each one and | |
254 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
255 | * trivial, there are various non-obvious corner cases, which this function | |
256 | * tries to solve in a safe manner. | |
257 | * | |
258 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
259 | * be called only for an already offlined CPU. | |
260 | */ | |
cb79295e AV |
261 | void clear_tasks_mm_cpumask(int cpu) |
262 | { | |
263 | struct task_struct *p; | |
264 | ||
265 | /* | |
266 | * This function is called after the cpu is taken down and marked | |
267 | * offline, so its not like new tasks will ever get this cpu set in | |
268 | * their mm mask. -- Peter Zijlstra | |
269 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
270 | * full-fledged tasklist_lock. | |
271 | */ | |
e4cc2f87 | 272 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
273 | rcu_read_lock(); |
274 | for_each_process(p) { | |
275 | struct task_struct *t; | |
276 | ||
e4cc2f87 AV |
277 | /* |
278 | * Main thread might exit, but other threads may still have | |
279 | * a valid mm. Find one. | |
280 | */ | |
cb79295e AV |
281 | t = find_lock_task_mm(p); |
282 | if (!t) | |
283 | continue; | |
284 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
285 | task_unlock(t); | |
286 | } | |
287 | rcu_read_unlock(); | |
288 | } | |
289 | ||
b728ca06 | 290 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 291 | { |
b728ca06 | 292 | struct task_struct *g, *p; |
1da177e4 | 293 | |
a75a6068 ON |
294 | read_lock(&tasklist_lock); |
295 | for_each_process_thread(g, p) { | |
b728ca06 KT |
296 | if (!p->on_rq) |
297 | continue; | |
298 | /* | |
299 | * We do the check with unlocked task_rq(p)->lock. | |
300 | * Order the reading to do not warn about a task, | |
301 | * which was running on this cpu in the past, and | |
302 | * it's just been woken on another cpu. | |
303 | */ | |
304 | rmb(); | |
305 | if (task_cpu(p) != dead_cpu) | |
306 | continue; | |
307 | ||
308 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
309 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
310 | } |
311 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
312 | } |
313 | ||
db912f96 AK |
314 | struct take_cpu_down_param { |
315 | unsigned long mod; | |
316 | void *hcpu; | |
317 | }; | |
318 | ||
1da177e4 | 319 | /* Take this CPU down. */ |
71cf5aee | 320 | static int take_cpu_down(void *_param) |
1da177e4 | 321 | { |
db912f96 | 322 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
323 | int err; |
324 | ||
1da177e4 LT |
325 | /* Ensure this CPU doesn't handle any more interrupts. */ |
326 | err = __cpu_disable(); | |
327 | if (err < 0) | |
f3705136 | 328 | return err; |
1da177e4 | 329 | |
e9fb7631 | 330 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
52c063d1 TG |
331 | /* Give up timekeeping duties */ |
332 | tick_handover_do_timer(); | |
14e568e7 | 333 | /* Park the stopper thread */ |
233e7f26 | 334 | stop_machine_park((long)param->hcpu); |
f3705136 | 335 | return 0; |
1da177e4 LT |
336 | } |
337 | ||
e3920fb4 | 338 | /* Requires cpu_add_remove_lock to be held */ |
71cf5aee | 339 | static int _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 340 | { |
e7407dcc | 341 | int err, nr_calls = 0; |
e7407dcc | 342 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 343 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
344 | struct take_cpu_down_param tcd_param = { |
345 | .mod = mod, | |
346 | .hcpu = hcpu, | |
347 | }; | |
1da177e4 | 348 | |
e3920fb4 RW |
349 | if (num_online_cpus() == 1) |
350 | return -EBUSY; | |
1da177e4 | 351 | |
e3920fb4 RW |
352 | if (!cpu_online(cpu)) |
353 | return -EINVAL; | |
1da177e4 | 354 | |
d221938c | 355 | cpu_hotplug_begin(); |
4d51985e | 356 | |
e9fb7631 | 357 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 358 | if (err) { |
a0d8cdb6 | 359 | nr_calls--; |
e9fb7631 | 360 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
361 | pr_warn("%s: attempt to take down CPU %u failed\n", |
362 | __func__, cpu); | |
baaca49f | 363 | goto out_release; |
1da177e4 LT |
364 | } |
365 | ||
6acce3ef PZ |
366 | /* |
367 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
368 | * and RCU users of this state to go away such that all new such users | |
369 | * will observe it. | |
370 | * | |
371 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
779de6ce | 372 | * not imply sync_sched(), so wait for both. |
106dd5af M |
373 | * |
374 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef | 375 | */ |
779de6ce PM |
376 | if (IS_ENABLED(CONFIG_PREEMPT)) |
377 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
378 | else | |
379 | synchronize_rcu(); | |
6acce3ef | 380 | |
106dd5af M |
381 | smpboot_park_threads(cpu); |
382 | ||
6acce3ef | 383 | /* |
a8994181 TG |
384 | * Prevent irq alloc/free while the dying cpu reorganizes the |
385 | * interrupt affinities. | |
6acce3ef | 386 | */ |
a8994181 | 387 | irq_lock_sparse(); |
6acce3ef | 388 | |
a8994181 TG |
389 | /* |
390 | * So now all preempt/rcu users must observe !cpu_active(). | |
391 | */ | |
7eeb088e | 392 | err = stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 393 | if (err) { |
1da177e4 | 394 | /* CPU didn't die: tell everyone. Can't complain. */ |
e9fb7631 | 395 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
a8994181 | 396 | irq_unlock_sparse(); |
6a1bdc1b | 397 | goto out_release; |
8fa1d7d3 | 398 | } |
04321587 | 399 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 400 | |
48c5ccae PZ |
401 | /* |
402 | * The migration_call() CPU_DYING callback will have removed all | |
403 | * runnable tasks from the cpu, there's only the idle task left now | |
404 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
405 | * |
406 | * Wait for the stop thread to go away. | |
48c5ccae | 407 | */ |
528a25b0 | 408 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 409 | cpu_relax(); |
528a25b0 PM |
410 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
411 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 412 | |
a8994181 TG |
413 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
414 | irq_unlock_sparse(); | |
415 | ||
345527b1 | 416 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
417 | /* This actually kills the CPU. */ |
418 | __cpu_die(cpu); | |
419 | ||
1da177e4 | 420 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
a49b116d | 421 | tick_cleanup_dead_cpu(cpu); |
e9fb7631 | 422 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
423 | |
424 | check_for_tasks(cpu); | |
425 | ||
baaca49f | 426 | out_release: |
d221938c | 427 | cpu_hotplug_done(); |
e9fb7631 AM |
428 | if (!err) |
429 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
430 | return err; |
431 | } | |
432 | ||
71cf5aee | 433 | int cpu_down(unsigned int cpu) |
e3920fb4 | 434 | { |
9ea09af3 | 435 | int err; |
e3920fb4 | 436 | |
d221938c | 437 | cpu_maps_update_begin(); |
e761b772 MK |
438 | |
439 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 440 | err = -EBUSY; |
e761b772 MK |
441 | goto out; |
442 | } | |
443 | ||
e761b772 | 444 | err = _cpu_down(cpu, 0); |
e3920fb4 | 445 | |
e761b772 | 446 | out: |
d221938c | 447 | cpu_maps_update_done(); |
1da177e4 LT |
448 | return err; |
449 | } | |
b62b8ef9 | 450 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
451 | #endif /*CONFIG_HOTPLUG_CPU*/ |
452 | ||
00df35f9 PM |
453 | /* |
454 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
455 | */ | |
456 | static int smpboot_thread_call(struct notifier_block *nfb, | |
457 | unsigned long action, void *hcpu) | |
458 | { | |
459 | int cpu = (long)hcpu; | |
460 | ||
461 | switch (action & ~CPU_TASKS_FROZEN) { | |
462 | ||
64eaf974 | 463 | case CPU_DOWN_FAILED: |
00df35f9 PM |
464 | case CPU_ONLINE: |
465 | smpboot_unpark_threads(cpu); | |
466 | break; | |
467 | ||
468 | default: | |
469 | break; | |
470 | } | |
471 | ||
472 | return NOTIFY_OK; | |
473 | } | |
474 | ||
475 | static struct notifier_block smpboot_thread_notifier = { | |
476 | .notifier_call = smpboot_thread_call, | |
477 | .priority = CPU_PRI_SMPBOOT, | |
478 | }; | |
479 | ||
927da9df | 480 | void smpboot_thread_init(void) |
00df35f9 PM |
481 | { |
482 | register_cpu_notifier(&smpboot_thread_notifier); | |
483 | } | |
484 | ||
e3920fb4 | 485 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 486 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 487 | { |
baaca49f | 488 | int ret, nr_calls = 0; |
1da177e4 | 489 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 490 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 491 | struct task_struct *idle; |
1da177e4 | 492 | |
d221938c | 493 | cpu_hotplug_begin(); |
38498a67 | 494 | |
5e5041f3 YI |
495 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
496 | ret = -EINVAL; | |
497 | goto out; | |
498 | } | |
499 | ||
3bb5d2ee SS |
500 | idle = idle_thread_get(cpu); |
501 | if (IS_ERR(idle)) { | |
502 | ret = PTR_ERR(idle); | |
38498a67 | 503 | goto out; |
3bb5d2ee | 504 | } |
38498a67 | 505 | |
f97f8f06 TG |
506 | ret = smpboot_create_threads(cpu); |
507 | if (ret) | |
508 | goto out; | |
509 | ||
e9fb7631 | 510 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 511 | if (ret) { |
a0d8cdb6 | 512 | nr_calls--; |
84117da5 FF |
513 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
514 | __func__, cpu); | |
1da177e4 LT |
515 | goto out_notify; |
516 | } | |
517 | ||
518 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 519 | ret = __cpu_up(cpu, idle); |
a8994181 | 520 | |
1da177e4 LT |
521 | if (ret != 0) |
522 | goto out_notify; | |
6978c705 | 523 | BUG_ON(!cpu_online(cpu)); |
1da177e4 LT |
524 | |
525 | /* Now call notifier in preparation. */ | |
e9fb7631 | 526 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
527 | |
528 | out_notify: | |
529 | if (ret != 0) | |
e9fb7631 | 530 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 531 | out: |
d221938c | 532 | cpu_hotplug_done(); |
e3920fb4 RW |
533 | |
534 | return ret; | |
535 | } | |
536 | ||
0db0628d | 537 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
538 | { |
539 | int err = 0; | |
cf23422b | 540 | |
e0b582ec | 541 | if (!cpu_possible(cpu)) { |
84117da5 FF |
542 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
543 | cpu); | |
87d5e023 | 544 | #if defined(CONFIG_IA64) |
84117da5 | 545 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
546 | #endif |
547 | return -EINVAL; | |
548 | } | |
e3920fb4 | 549 | |
01b0f197 TK |
550 | err = try_online_node(cpu_to_node(cpu)); |
551 | if (err) | |
552 | return err; | |
cf23422b | 553 | |
d221938c | 554 | cpu_maps_update_begin(); |
e761b772 MK |
555 | |
556 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 557 | err = -EBUSY; |
e761b772 MK |
558 | goto out; |
559 | } | |
560 | ||
561 | err = _cpu_up(cpu, 0); | |
562 | ||
e761b772 | 563 | out: |
d221938c | 564 | cpu_maps_update_done(); |
e3920fb4 RW |
565 | return err; |
566 | } | |
a513f6ba | 567 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 568 | |
f3de4be9 | 569 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 570 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
571 | |
572 | int disable_nonboot_cpus(void) | |
573 | { | |
e9a5f426 | 574 | int cpu, first_cpu, error = 0; |
e3920fb4 | 575 | |
d221938c | 576 | cpu_maps_update_begin(); |
e0b582ec | 577 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
578 | /* |
579 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
580 | * with the userspace trying to use the CPU hotplug at the same time |
581 | */ | |
e0b582ec | 582 | cpumask_clear(frozen_cpus); |
6ad4c188 | 583 | |
84117da5 | 584 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
585 | for_each_online_cpu(cpu) { |
586 | if (cpu == first_cpu) | |
587 | continue; | |
bb3632c6 | 588 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 589 | error = _cpu_down(cpu, 1); |
bb3632c6 | 590 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 591 | if (!error) |
e0b582ec | 592 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 593 | else { |
84117da5 | 594 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
595 | break; |
596 | } | |
597 | } | |
86886e55 | 598 | |
89af7ba5 | 599 | if (!error) |
e3920fb4 | 600 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 601 | else |
84117da5 | 602 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
603 | |
604 | /* | |
605 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
606 | * this even in case of failure as all disable_nonboot_cpus() users are | |
607 | * supposed to do enable_nonboot_cpus() on the failure path. | |
608 | */ | |
609 | cpu_hotplug_disabled++; | |
610 | ||
d221938c | 611 | cpu_maps_update_done(); |
e3920fb4 RW |
612 | return error; |
613 | } | |
614 | ||
d0af9eed SS |
615 | void __weak arch_enable_nonboot_cpus_begin(void) |
616 | { | |
617 | } | |
618 | ||
619 | void __weak arch_enable_nonboot_cpus_end(void) | |
620 | { | |
621 | } | |
622 | ||
71cf5aee | 623 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
624 | { |
625 | int cpu, error; | |
626 | ||
627 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 628 | cpu_maps_update_begin(); |
89af7ba5 | 629 | WARN_ON(--cpu_hotplug_disabled < 0); |
e0b582ec | 630 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 631 | goto out; |
e3920fb4 | 632 | |
84117da5 | 633 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
634 | |
635 | arch_enable_nonboot_cpus_begin(); | |
636 | ||
e0b582ec | 637 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 638 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 639 | error = _cpu_up(cpu, 1); |
bb3632c6 | 640 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 641 | if (!error) { |
84117da5 | 642 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
643 | continue; |
644 | } | |
84117da5 | 645 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 646 | } |
d0af9eed SS |
647 | |
648 | arch_enable_nonboot_cpus_end(); | |
649 | ||
e0b582ec | 650 | cpumask_clear(frozen_cpus); |
1d64b9cb | 651 | out: |
d221938c | 652 | cpu_maps_update_done(); |
1da177e4 | 653 | } |
e0b582ec | 654 | |
d7268a31 | 655 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
656 | { |
657 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
658 | return -ENOMEM; | |
659 | return 0; | |
660 | } | |
661 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 662 | |
79cfbdfa SB |
663 | /* |
664 | * When callbacks for CPU hotplug notifications are being executed, we must | |
665 | * ensure that the state of the system with respect to the tasks being frozen | |
666 | * or not, as reported by the notification, remains unchanged *throughout the | |
667 | * duration* of the execution of the callbacks. | |
668 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
669 | * | |
670 | * This synchronization is implemented by mutually excluding regular CPU | |
671 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
672 | * Hibernate notifications. | |
673 | */ | |
674 | static int | |
675 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
676 | unsigned long action, void *ptr) | |
677 | { | |
678 | switch (action) { | |
679 | ||
680 | case PM_SUSPEND_PREPARE: | |
681 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 682 | cpu_hotplug_disable(); |
79cfbdfa SB |
683 | break; |
684 | ||
685 | case PM_POST_SUSPEND: | |
686 | case PM_POST_HIBERNATION: | |
16e53dbf | 687 | cpu_hotplug_enable(); |
79cfbdfa SB |
688 | break; |
689 | ||
690 | default: | |
691 | return NOTIFY_DONE; | |
692 | } | |
693 | ||
694 | return NOTIFY_OK; | |
695 | } | |
696 | ||
697 | ||
d7268a31 | 698 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 699 | { |
6e32d479 FY |
700 | /* |
701 | * cpu_hotplug_pm_callback has higher priority than x86 | |
702 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
703 | * to disable cpu hotplug to avoid cpu hotplug race. | |
704 | */ | |
79cfbdfa SB |
705 | pm_notifier(cpu_hotplug_pm_callback, 0); |
706 | return 0; | |
707 | } | |
708 | core_initcall(cpu_hotplug_pm_sync_init); | |
709 | ||
f3de4be9 | 710 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 711 | |
e545a614 MS |
712 | /** |
713 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
714 | * @cpu: cpu that just started | |
715 | * | |
716 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
717 | * It must be called by the arch code on the new cpu, before the new cpu | |
718 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
719 | */ | |
0db0628d | 720 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
721 | { |
722 | unsigned long val = CPU_STARTING; | |
723 | ||
724 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 725 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
726 | val = CPU_STARTING_FROZEN; |
727 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 728 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
729 | } |
730 | ||
68f4f1ec | 731 | #endif /* CONFIG_SMP */ |
b8d317d1 | 732 | |
e56b3bc7 LT |
733 | /* |
734 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
735 | * represents all NR_CPUS bits binary values of 1<<nr. | |
736 | * | |
e0b582ec | 737 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
738 | * mask value that has a single bit set only. |
739 | */ | |
b8d317d1 | 740 | |
e56b3bc7 | 741 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 742 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
743 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
744 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
745 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 746 | |
e56b3bc7 LT |
747 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
748 | ||
749 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
750 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
751 | #if BITS_PER_LONG > 32 | |
752 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
753 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
754 | #endif |
755 | }; | |
e56b3bc7 | 756 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
757 | |
758 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
759 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
760 | |
761 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 762 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 763 | = {CPU_BITS_ALL}; |
b3199c02 | 764 | #else |
4b804c85 | 765 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 766 | #endif |
4b804c85 | 767 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 768 | |
4b804c85 RV |
769 | struct cpumask __cpu_online_mask __read_mostly; |
770 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 771 | |
4b804c85 RV |
772 | struct cpumask __cpu_present_mask __read_mostly; |
773 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 774 | |
4b804c85 RV |
775 | struct cpumask __cpu_active_mask __read_mostly; |
776 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 777 | |
3fa41520 RR |
778 | void init_cpu_present(const struct cpumask *src) |
779 | { | |
c4c54dd1 | 780 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
781 | } |
782 | ||
783 | void init_cpu_possible(const struct cpumask *src) | |
784 | { | |
c4c54dd1 | 785 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
786 | } |
787 | ||
788 | void init_cpu_online(const struct cpumask *src) | |
789 | { | |
c4c54dd1 | 790 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 791 | } |