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 | |
dd56af42 PM |
105 | bool try_get_online_cpus(void) |
106 | { | |
107 | if (cpu_hotplug.active_writer == current) | |
108 | return true; | |
109 | if (!mutex_trylock(&cpu_hotplug.lock)) | |
110 | return false; | |
111 | cpuhp_lock_acquire_tryread(); | |
87af9e7f | 112 | atomic_inc(&cpu_hotplug.refcount); |
dd56af42 PM |
113 | mutex_unlock(&cpu_hotplug.lock); |
114 | return true; | |
115 | } | |
116 | EXPORT_SYMBOL_GPL(try_get_online_cpus); | |
117 | ||
86ef5c9a | 118 | void put_online_cpus(void) |
a9d9baa1 | 119 | { |
87af9e7f DH |
120 | int refcount; |
121 | ||
d221938c | 122 | if (cpu_hotplug.active_writer == current) |
aa953877 | 123 | return; |
075663d1 | 124 | |
87af9e7f DH |
125 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
126 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
127 | atomic_inc(&cpu_hotplug.refcount); | |
128 | ||
129 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
130 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 131 | |
a19423b9 | 132 | cpuhp_lock_release(); |
d221938c | 133 | |
a9d9baa1 | 134 | } |
86ef5c9a | 135 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 136 | |
d221938c GS |
137 | /* |
138 | * This ensures that the hotplug operation can begin only when the | |
139 | * refcount goes to zero. | |
140 | * | |
141 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
142 | * will be blocked by the cpu_hotplug.lock | |
143 | * | |
d2ba7e2a ON |
144 | * Since cpu_hotplug_begin() is always called after invoking |
145 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
146 | * |
147 | * Note that theoretically, there is a possibility of a livelock: | |
148 | * - Refcount goes to zero, last reader wakes up the sleeping | |
149 | * writer. | |
150 | * - Last reader unlocks the cpu_hotplug.lock. | |
151 | * - A new reader arrives at this moment, bumps up the refcount. | |
152 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
153 | * non zero and goes to sleep again. | |
154 | * | |
155 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 156 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
157 | * |
158 | */ | |
b9d10be7 | 159 | void cpu_hotplug_begin(void) |
d221938c | 160 | { |
87af9e7f | 161 | DEFINE_WAIT(wait); |
d2ba7e2a | 162 | |
87af9e7f | 163 | cpu_hotplug.active_writer = current; |
a19423b9 | 164 | cpuhp_lock_acquire(); |
87af9e7f | 165 | |
d2ba7e2a ON |
166 | for (;;) { |
167 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
168 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
169 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
170 | break; | |
d221938c GS |
171 | mutex_unlock(&cpu_hotplug.lock); |
172 | schedule(); | |
d221938c | 173 | } |
87af9e7f | 174 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
175 | } |
176 | ||
b9d10be7 | 177 | void cpu_hotplug_done(void) |
d221938c GS |
178 | { |
179 | cpu_hotplug.active_writer = NULL; | |
180 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 181 | cpuhp_lock_release(); |
d221938c | 182 | } |
79a6cdeb | 183 | |
16e53dbf SB |
184 | /* |
185 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
186 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
187 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
188 | * hotplug path before performing hotplug operations. So acquiring that lock | |
189 | * guarantees mutual exclusion from any currently running hotplug operations. | |
190 | */ | |
191 | void cpu_hotplug_disable(void) | |
192 | { | |
193 | cpu_maps_update_begin(); | |
194 | cpu_hotplug_disabled = 1; | |
195 | cpu_maps_update_done(); | |
196 | } | |
197 | ||
198 | void cpu_hotplug_enable(void) | |
199 | { | |
200 | cpu_maps_update_begin(); | |
201 | cpu_hotplug_disabled = 0; | |
202 | cpu_maps_update_done(); | |
203 | } | |
204 | ||
b9d10be7 | 205 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 206 | |
1da177e4 | 207 | /* Need to know about CPUs going up/down? */ |
f7b16c10 | 208 | int __ref register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 209 | { |
bd5349cf | 210 | int ret; |
d221938c | 211 | cpu_maps_update_begin(); |
bd5349cf | 212 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 213 | cpu_maps_update_done(); |
bd5349cf | 214 | return ret; |
1da177e4 | 215 | } |
65edc68c | 216 | |
93ae4f97 SB |
217 | int __ref __register_cpu_notifier(struct notifier_block *nb) |
218 | { | |
219 | return raw_notifier_chain_register(&cpu_chain, nb); | |
220 | } | |
221 | ||
e9fb7631 AM |
222 | static int __cpu_notify(unsigned long val, void *v, int nr_to_call, |
223 | int *nr_calls) | |
224 | { | |
e6bde73b AM |
225 | int ret; |
226 | ||
227 | ret = __raw_notifier_call_chain(&cpu_chain, val, v, nr_to_call, | |
e9fb7631 | 228 | nr_calls); |
e6bde73b AM |
229 | |
230 | return notifier_to_errno(ret); | |
e9fb7631 AM |
231 | } |
232 | ||
233 | static int cpu_notify(unsigned long val, void *v) | |
234 | { | |
235 | return __cpu_notify(val, v, -1, NULL); | |
236 | } | |
237 | ||
00b9b0af LT |
238 | #ifdef CONFIG_HOTPLUG_CPU |
239 | ||
e9fb7631 AM |
240 | static void cpu_notify_nofail(unsigned long val, void *v) |
241 | { | |
00b9b0af | 242 | BUG_ON(cpu_notify(val, v)); |
e9fb7631 | 243 | } |
1da177e4 | 244 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 245 | EXPORT_SYMBOL(__register_cpu_notifier); |
1da177e4 | 246 | |
9647155f | 247 | void __ref unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 248 | { |
d221938c | 249 | cpu_maps_update_begin(); |
bd5349cf | 250 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 251 | cpu_maps_update_done(); |
1da177e4 LT |
252 | } |
253 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
254 | ||
93ae4f97 SB |
255 | void __ref __unregister_cpu_notifier(struct notifier_block *nb) |
256 | { | |
257 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
258 | } | |
259 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
260 | ||
e4cc2f87 AV |
261 | /** |
262 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
263 | * @cpu: a CPU id | |
264 | * | |
265 | * This function walks all processes, finds a valid mm struct for each one and | |
266 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
267 | * trivial, there are various non-obvious corner cases, which this function | |
268 | * tries to solve in a safe manner. | |
269 | * | |
270 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
271 | * be called only for an already offlined CPU. | |
272 | */ | |
cb79295e AV |
273 | void clear_tasks_mm_cpumask(int cpu) |
274 | { | |
275 | struct task_struct *p; | |
276 | ||
277 | /* | |
278 | * This function is called after the cpu is taken down and marked | |
279 | * offline, so its not like new tasks will ever get this cpu set in | |
280 | * their mm mask. -- Peter Zijlstra | |
281 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
282 | * full-fledged tasklist_lock. | |
283 | */ | |
e4cc2f87 | 284 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
285 | rcu_read_lock(); |
286 | for_each_process(p) { | |
287 | struct task_struct *t; | |
288 | ||
e4cc2f87 AV |
289 | /* |
290 | * Main thread might exit, but other threads may still have | |
291 | * a valid mm. Find one. | |
292 | */ | |
cb79295e AV |
293 | t = find_lock_task_mm(p); |
294 | if (!t) | |
295 | continue; | |
296 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
297 | task_unlock(t); | |
298 | } | |
299 | rcu_read_unlock(); | |
300 | } | |
301 | ||
b728ca06 | 302 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 303 | { |
b728ca06 | 304 | struct task_struct *g, *p; |
1da177e4 | 305 | |
b728ca06 KT |
306 | read_lock_irq(&tasklist_lock); |
307 | do_each_thread(g, p) { | |
308 | if (!p->on_rq) | |
309 | continue; | |
310 | /* | |
311 | * We do the check with unlocked task_rq(p)->lock. | |
312 | * Order the reading to do not warn about a task, | |
313 | * which was running on this cpu in the past, and | |
314 | * it's just been woken on another cpu. | |
315 | */ | |
316 | rmb(); | |
317 | if (task_cpu(p) != dead_cpu) | |
318 | continue; | |
319 | ||
320 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
321 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
322 | } while_each_thread(g, p); | |
323 | read_unlock_irq(&tasklist_lock); | |
1da177e4 LT |
324 | } |
325 | ||
db912f96 AK |
326 | struct take_cpu_down_param { |
327 | unsigned long mod; | |
328 | void *hcpu; | |
329 | }; | |
330 | ||
1da177e4 | 331 | /* Take this CPU down. */ |
514a20a5 | 332 | static int __ref take_cpu_down(void *_param) |
1da177e4 | 333 | { |
db912f96 | 334 | struct take_cpu_down_param *param = _param; |
1da177e4 LT |
335 | int err; |
336 | ||
1da177e4 LT |
337 | /* Ensure this CPU doesn't handle any more interrupts. */ |
338 | err = __cpu_disable(); | |
339 | if (err < 0) | |
f3705136 | 340 | return err; |
1da177e4 | 341 | |
e9fb7631 | 342 | cpu_notify(CPU_DYING | param->mod, param->hcpu); |
52c063d1 TG |
343 | /* Give up timekeeping duties */ |
344 | tick_handover_do_timer(); | |
14e568e7 TG |
345 | /* Park the stopper thread */ |
346 | kthread_park(current); | |
f3705136 | 347 | return 0; |
1da177e4 LT |
348 | } |
349 | ||
e3920fb4 | 350 | /* Requires cpu_add_remove_lock to be held */ |
514a20a5 | 351 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen) |
1da177e4 | 352 | { |
e7407dcc | 353 | int err, nr_calls = 0; |
e7407dcc | 354 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 355 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
db912f96 AK |
356 | struct take_cpu_down_param tcd_param = { |
357 | .mod = mod, | |
358 | .hcpu = hcpu, | |
359 | }; | |
1da177e4 | 360 | |
e3920fb4 RW |
361 | if (num_online_cpus() == 1) |
362 | return -EBUSY; | |
1da177e4 | 363 | |
e3920fb4 RW |
364 | if (!cpu_online(cpu)) |
365 | return -EINVAL; | |
1da177e4 | 366 | |
d221938c | 367 | cpu_hotplug_begin(); |
4d51985e | 368 | |
e9fb7631 | 369 | err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 370 | if (err) { |
a0d8cdb6 | 371 | nr_calls--; |
e9fb7631 | 372 | __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL); |
84117da5 FF |
373 | pr_warn("%s: attempt to take down CPU %u failed\n", |
374 | __func__, cpu); | |
baaca49f | 375 | goto out_release; |
1da177e4 LT |
376 | } |
377 | ||
6acce3ef PZ |
378 | /* |
379 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
380 | * and RCU users of this state to go away such that all new such users | |
381 | * will observe it. | |
382 | * | |
383 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
384 | * not imply sync_sched(), so explicitly call both. | |
106dd5af M |
385 | * |
386 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef PZ |
387 | */ |
388 | #ifdef CONFIG_PREEMPT | |
389 | synchronize_sched(); | |
390 | #endif | |
391 | synchronize_rcu(); | |
392 | ||
106dd5af M |
393 | smpboot_park_threads(cpu); |
394 | ||
6acce3ef | 395 | /* |
a8994181 TG |
396 | * Prevent irq alloc/free while the dying cpu reorganizes the |
397 | * interrupt affinities. | |
6acce3ef | 398 | */ |
a8994181 | 399 | irq_lock_sparse(); |
6acce3ef | 400 | |
a8994181 TG |
401 | /* |
402 | * So now all preempt/rcu users must observe !cpu_active(). | |
403 | */ | |
e0b582ec | 404 | err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); |
04321587 | 405 | if (err) { |
1da177e4 | 406 | /* CPU didn't die: tell everyone. Can't complain. */ |
e9fb7631 | 407 | cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); |
a8994181 | 408 | irq_unlock_sparse(); |
6a1bdc1b | 409 | goto out_release; |
8fa1d7d3 | 410 | } |
04321587 | 411 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 412 | |
48c5ccae PZ |
413 | /* |
414 | * The migration_call() CPU_DYING callback will have removed all | |
415 | * runnable tasks from the cpu, there's only the idle task left now | |
416 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
417 | * |
418 | * Wait for the stop thread to go away. | |
48c5ccae | 419 | */ |
528a25b0 | 420 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 421 | cpu_relax(); |
528a25b0 PM |
422 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
423 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 424 | |
a8994181 TG |
425 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
426 | irq_unlock_sparse(); | |
427 | ||
345527b1 | 428 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
429 | /* This actually kills the CPU. */ |
430 | __cpu_die(cpu); | |
431 | ||
1da177e4 | 432 | /* CPU is completely dead: tell everyone. Too late to complain. */ |
a49b116d | 433 | tick_cleanup_dead_cpu(cpu); |
e9fb7631 | 434 | cpu_notify_nofail(CPU_DEAD | mod, hcpu); |
1da177e4 LT |
435 | |
436 | check_for_tasks(cpu); | |
437 | ||
baaca49f | 438 | out_release: |
d221938c | 439 | cpu_hotplug_done(); |
e9fb7631 AM |
440 | if (!err) |
441 | cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu); | |
e3920fb4 RW |
442 | return err; |
443 | } | |
444 | ||
514a20a5 | 445 | int __ref cpu_down(unsigned int cpu) |
e3920fb4 | 446 | { |
9ea09af3 | 447 | int err; |
e3920fb4 | 448 | |
d221938c | 449 | cpu_maps_update_begin(); |
e761b772 MK |
450 | |
451 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 452 | err = -EBUSY; |
e761b772 MK |
453 | goto out; |
454 | } | |
455 | ||
e761b772 | 456 | err = _cpu_down(cpu, 0); |
e3920fb4 | 457 | |
e761b772 | 458 | out: |
d221938c | 459 | cpu_maps_update_done(); |
1da177e4 LT |
460 | return err; |
461 | } | |
b62b8ef9 | 462 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
463 | #endif /*CONFIG_HOTPLUG_CPU*/ |
464 | ||
00df35f9 PM |
465 | /* |
466 | * Unpark per-CPU smpboot kthreads at CPU-online time. | |
467 | */ | |
468 | static int smpboot_thread_call(struct notifier_block *nfb, | |
469 | unsigned long action, void *hcpu) | |
470 | { | |
471 | int cpu = (long)hcpu; | |
472 | ||
473 | switch (action & ~CPU_TASKS_FROZEN) { | |
474 | ||
64eaf974 | 475 | case CPU_DOWN_FAILED: |
00df35f9 PM |
476 | case CPU_ONLINE: |
477 | smpboot_unpark_threads(cpu); | |
478 | break; | |
479 | ||
480 | default: | |
481 | break; | |
482 | } | |
483 | ||
484 | return NOTIFY_OK; | |
485 | } | |
486 | ||
487 | static struct notifier_block smpboot_thread_notifier = { | |
488 | .notifier_call = smpboot_thread_call, | |
489 | .priority = CPU_PRI_SMPBOOT, | |
490 | }; | |
491 | ||
927da9df | 492 | void smpboot_thread_init(void) |
00df35f9 PM |
493 | { |
494 | register_cpu_notifier(&smpboot_thread_notifier); | |
495 | } | |
496 | ||
e3920fb4 | 497 | /* Requires cpu_add_remove_lock to be held */ |
0db0628d | 498 | static int _cpu_up(unsigned int cpu, int tasks_frozen) |
1da177e4 | 499 | { |
baaca49f | 500 | int ret, nr_calls = 0; |
1da177e4 | 501 | void *hcpu = (void *)(long)cpu; |
8bb78442 | 502 | unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0; |
3bb5d2ee | 503 | struct task_struct *idle; |
1da177e4 | 504 | |
d221938c | 505 | cpu_hotplug_begin(); |
38498a67 | 506 | |
5e5041f3 YI |
507 | if (cpu_online(cpu) || !cpu_present(cpu)) { |
508 | ret = -EINVAL; | |
509 | goto out; | |
510 | } | |
511 | ||
3bb5d2ee SS |
512 | idle = idle_thread_get(cpu); |
513 | if (IS_ERR(idle)) { | |
514 | ret = PTR_ERR(idle); | |
38498a67 | 515 | goto out; |
3bb5d2ee | 516 | } |
38498a67 | 517 | |
f97f8f06 TG |
518 | ret = smpboot_create_threads(cpu); |
519 | if (ret) | |
520 | goto out; | |
521 | ||
e9fb7631 | 522 | ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); |
e6bde73b | 523 | if (ret) { |
a0d8cdb6 | 524 | nr_calls--; |
84117da5 FF |
525 | pr_warn("%s: attempt to bring up CPU %u failed\n", |
526 | __func__, cpu); | |
1da177e4 LT |
527 | goto out_notify; |
528 | } | |
529 | ||
530 | /* Arch-specific enabling code. */ | |
3bb5d2ee | 531 | ret = __cpu_up(cpu, idle); |
a8994181 | 532 | |
1da177e4 LT |
533 | if (ret != 0) |
534 | goto out_notify; | |
6978c705 | 535 | BUG_ON(!cpu_online(cpu)); |
1da177e4 LT |
536 | |
537 | /* Now call notifier in preparation. */ | |
e9fb7631 | 538 | cpu_notify(CPU_ONLINE | mod, hcpu); |
1da177e4 LT |
539 | |
540 | out_notify: | |
541 | if (ret != 0) | |
e9fb7631 | 542 | __cpu_notify(CPU_UP_CANCELED | mod, hcpu, nr_calls, NULL); |
38498a67 | 543 | out: |
d221938c | 544 | cpu_hotplug_done(); |
e3920fb4 RW |
545 | |
546 | return ret; | |
547 | } | |
548 | ||
0db0628d | 549 | int cpu_up(unsigned int cpu) |
e3920fb4 RW |
550 | { |
551 | int err = 0; | |
cf23422b | 552 | |
e0b582ec | 553 | if (!cpu_possible(cpu)) { |
84117da5 FF |
554 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
555 | cpu); | |
87d5e023 | 556 | #if defined(CONFIG_IA64) |
84117da5 | 557 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
558 | #endif |
559 | return -EINVAL; | |
560 | } | |
e3920fb4 | 561 | |
01b0f197 TK |
562 | err = try_online_node(cpu_to_node(cpu)); |
563 | if (err) | |
564 | return err; | |
cf23422b | 565 | |
d221938c | 566 | cpu_maps_update_begin(); |
e761b772 MK |
567 | |
568 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 569 | err = -EBUSY; |
e761b772 MK |
570 | goto out; |
571 | } | |
572 | ||
573 | err = _cpu_up(cpu, 0); | |
574 | ||
e761b772 | 575 | out: |
d221938c | 576 | cpu_maps_update_done(); |
e3920fb4 RW |
577 | return err; |
578 | } | |
a513f6ba | 579 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 580 | |
f3de4be9 | 581 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 582 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
583 | |
584 | int disable_nonboot_cpus(void) | |
585 | { | |
e9a5f426 | 586 | int cpu, first_cpu, error = 0; |
e3920fb4 | 587 | |
d221938c | 588 | cpu_maps_update_begin(); |
e0b582ec | 589 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
590 | /* |
591 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
592 | * with the userspace trying to use the CPU hotplug at the same time |
593 | */ | |
e0b582ec | 594 | cpumask_clear(frozen_cpus); |
6ad4c188 | 595 | |
84117da5 | 596 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
597 | for_each_online_cpu(cpu) { |
598 | if (cpu == first_cpu) | |
599 | continue; | |
bb3632c6 | 600 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
8bb78442 | 601 | error = _cpu_down(cpu, 1); |
bb3632c6 | 602 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 603 | if (!error) |
e0b582ec | 604 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 605 | else { |
84117da5 | 606 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
607 | break; |
608 | } | |
609 | } | |
86886e55 | 610 | |
e3920fb4 RW |
611 | if (!error) { |
612 | BUG_ON(num_online_cpus() > 1); | |
613 | /* Make sure the CPUs won't be enabled by someone else */ | |
614 | cpu_hotplug_disabled = 1; | |
615 | } else { | |
84117da5 | 616 | pr_err("Non-boot CPUs are not disabled\n"); |
e3920fb4 | 617 | } |
d221938c | 618 | cpu_maps_update_done(); |
e3920fb4 RW |
619 | return error; |
620 | } | |
621 | ||
d0af9eed SS |
622 | void __weak arch_enable_nonboot_cpus_begin(void) |
623 | { | |
624 | } | |
625 | ||
626 | void __weak arch_enable_nonboot_cpus_end(void) | |
627 | { | |
628 | } | |
629 | ||
fa7303e2 | 630 | void __ref enable_nonboot_cpus(void) |
e3920fb4 RW |
631 | { |
632 | int cpu, error; | |
633 | ||
634 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 635 | cpu_maps_update_begin(); |
e3920fb4 | 636 | cpu_hotplug_disabled = 0; |
e0b582ec | 637 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 638 | goto out; |
e3920fb4 | 639 | |
84117da5 | 640 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
641 | |
642 | arch_enable_nonboot_cpus_begin(); | |
643 | ||
e0b582ec | 644 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 645 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
8bb78442 | 646 | error = _cpu_up(cpu, 1); |
bb3632c6 | 647 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 648 | if (!error) { |
84117da5 | 649 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
650 | continue; |
651 | } | |
84117da5 | 652 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 653 | } |
d0af9eed SS |
654 | |
655 | arch_enable_nonboot_cpus_end(); | |
656 | ||
e0b582ec | 657 | cpumask_clear(frozen_cpus); |
1d64b9cb | 658 | out: |
d221938c | 659 | cpu_maps_update_done(); |
1da177e4 | 660 | } |
e0b582ec | 661 | |
d7268a31 | 662 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
663 | { |
664 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
665 | return -ENOMEM; | |
666 | return 0; | |
667 | } | |
668 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 669 | |
79cfbdfa SB |
670 | /* |
671 | * When callbacks for CPU hotplug notifications are being executed, we must | |
672 | * ensure that the state of the system with respect to the tasks being frozen | |
673 | * or not, as reported by the notification, remains unchanged *throughout the | |
674 | * duration* of the execution of the callbacks. | |
675 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
676 | * | |
677 | * This synchronization is implemented by mutually excluding regular CPU | |
678 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
679 | * Hibernate notifications. | |
680 | */ | |
681 | static int | |
682 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
683 | unsigned long action, void *ptr) | |
684 | { | |
685 | switch (action) { | |
686 | ||
687 | case PM_SUSPEND_PREPARE: | |
688 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 689 | cpu_hotplug_disable(); |
79cfbdfa SB |
690 | break; |
691 | ||
692 | case PM_POST_SUSPEND: | |
693 | case PM_POST_HIBERNATION: | |
16e53dbf | 694 | cpu_hotplug_enable(); |
79cfbdfa SB |
695 | break; |
696 | ||
697 | default: | |
698 | return NOTIFY_DONE; | |
699 | } | |
700 | ||
701 | return NOTIFY_OK; | |
702 | } | |
703 | ||
704 | ||
d7268a31 | 705 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 706 | { |
6e32d479 FY |
707 | /* |
708 | * cpu_hotplug_pm_callback has higher priority than x86 | |
709 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
710 | * to disable cpu hotplug to avoid cpu hotplug race. | |
711 | */ | |
79cfbdfa SB |
712 | pm_notifier(cpu_hotplug_pm_callback, 0); |
713 | return 0; | |
714 | } | |
715 | core_initcall(cpu_hotplug_pm_sync_init); | |
716 | ||
f3de4be9 | 717 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 718 | |
e545a614 MS |
719 | /** |
720 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
721 | * @cpu: cpu that just started | |
722 | * | |
723 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
724 | * It must be called by the arch code on the new cpu, before the new cpu | |
725 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
726 | */ | |
0db0628d | 727 | void notify_cpu_starting(unsigned int cpu) |
e545a614 MS |
728 | { |
729 | unsigned long val = CPU_STARTING; | |
730 | ||
731 | #ifdef CONFIG_PM_SLEEP_SMP | |
e0b582ec | 732 | if (frozen_cpus != NULL && cpumask_test_cpu(cpu, frozen_cpus)) |
e545a614 MS |
733 | val = CPU_STARTING_FROZEN; |
734 | #endif /* CONFIG_PM_SLEEP_SMP */ | |
e9fb7631 | 735 | cpu_notify(val, (void *)(long)cpu); |
e545a614 MS |
736 | } |
737 | ||
68f4f1ec | 738 | #endif /* CONFIG_SMP */ |
b8d317d1 | 739 | |
e56b3bc7 LT |
740 | /* |
741 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
742 | * represents all NR_CPUS bits binary values of 1<<nr. | |
743 | * | |
e0b582ec | 744 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
745 | * mask value that has a single bit set only. |
746 | */ | |
b8d317d1 | 747 | |
e56b3bc7 | 748 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 749 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
750 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
751 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
752 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 753 | |
e56b3bc7 LT |
754 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
755 | ||
756 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
757 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
758 | #if BITS_PER_LONG > 32 | |
759 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
760 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
761 | #endif |
762 | }; | |
e56b3bc7 | 763 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
764 | |
765 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
766 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
767 | |
768 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
769 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly | |
770 | = CPU_BITS_ALL; | |
771 | #else | |
772 | static DECLARE_BITMAP(cpu_possible_bits, CONFIG_NR_CPUS) __read_mostly; | |
773 | #endif | |
774 | const struct cpumask *const cpu_possible_mask = to_cpumask(cpu_possible_bits); | |
775 | EXPORT_SYMBOL(cpu_possible_mask); | |
776 | ||
777 | static DECLARE_BITMAP(cpu_online_bits, CONFIG_NR_CPUS) __read_mostly; | |
778 | const struct cpumask *const cpu_online_mask = to_cpumask(cpu_online_bits); | |
779 | EXPORT_SYMBOL(cpu_online_mask); | |
780 | ||
781 | static DECLARE_BITMAP(cpu_present_bits, CONFIG_NR_CPUS) __read_mostly; | |
782 | const struct cpumask *const cpu_present_mask = to_cpumask(cpu_present_bits); | |
783 | EXPORT_SYMBOL(cpu_present_mask); | |
784 | ||
785 | static DECLARE_BITMAP(cpu_active_bits, CONFIG_NR_CPUS) __read_mostly; | |
786 | const struct cpumask *const cpu_active_mask = to_cpumask(cpu_active_bits); | |
787 | EXPORT_SYMBOL(cpu_active_mask); | |
3fa41520 RR |
788 | |
789 | void set_cpu_possible(unsigned int cpu, bool possible) | |
790 | { | |
791 | if (possible) | |
792 | cpumask_set_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
793 | else | |
794 | cpumask_clear_cpu(cpu, to_cpumask(cpu_possible_bits)); | |
795 | } | |
796 | ||
797 | void set_cpu_present(unsigned int cpu, bool present) | |
798 | { | |
799 | if (present) | |
800 | cpumask_set_cpu(cpu, to_cpumask(cpu_present_bits)); | |
801 | else | |
802 | cpumask_clear_cpu(cpu, to_cpumask(cpu_present_bits)); | |
803 | } | |
804 | ||
805 | void set_cpu_online(unsigned int cpu, bool online) | |
806 | { | |
6acbfb96 | 807 | if (online) { |
3fa41520 | 808 | cpumask_set_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 LJ |
809 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); |
810 | } else { | |
3fa41520 | 811 | cpumask_clear_cpu(cpu, to_cpumask(cpu_online_bits)); |
6acbfb96 | 812 | } |
3fa41520 RR |
813 | } |
814 | ||
815 | void set_cpu_active(unsigned int cpu, bool active) | |
816 | { | |
817 | if (active) | |
818 | cpumask_set_cpu(cpu, to_cpumask(cpu_active_bits)); | |
819 | else | |
820 | cpumask_clear_cpu(cpu, to_cpumask(cpu_active_bits)); | |
821 | } | |
822 | ||
823 | void init_cpu_present(const struct cpumask *src) | |
824 | { | |
825 | cpumask_copy(to_cpumask(cpu_present_bits), src); | |
826 | } | |
827 | ||
828 | void init_cpu_possible(const struct cpumask *src) | |
829 | { | |
830 | cpumask_copy(to_cpumask(cpu_possible_bits), src); | |
831 | } | |
832 | ||
833 | void init_cpu_online(const struct cpumask *src) | |
834 | { | |
835 | cpumask_copy(to_cpumask(cpu_online_bits), src); | |
836 | } |