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> |
4cb28ced | 25 | #include <linux/smpboot.h> |
e6d4989a | 26 | #include <linux/relay.h> |
6731d4f1 | 27 | #include <linux/slab.h> |
cff7d378 | 28 | |
bb3632c6 | 29 | #include <trace/events/power.h> |
cff7d378 TG |
30 | #define CREATE_TRACE_POINTS |
31 | #include <trace/events/cpuhp.h> | |
1da177e4 | 32 | |
38498a67 TG |
33 | #include "smpboot.h" |
34 | ||
cff7d378 TG |
35 | /** |
36 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
37 | * @state: The current cpu state | |
38 | * @target: The target state | |
4cb28ced TG |
39 | * @thread: Pointer to the hotplug thread |
40 | * @should_run: Thread should execute | |
3b9d6da6 | 41 | * @rollback: Perform a rollback |
a724632c TG |
42 | * @single: Single callback invocation |
43 | * @bringup: Single callback bringup or teardown selector | |
44 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced TG |
45 | * @result: Result of the operation |
46 | * @done: Signal completion to the issuer of the task | |
cff7d378 TG |
47 | */ |
48 | struct cpuhp_cpu_state { | |
49 | enum cpuhp_state state; | |
50 | enum cpuhp_state target; | |
4cb28ced TG |
51 | #ifdef CONFIG_SMP |
52 | struct task_struct *thread; | |
53 | bool should_run; | |
3b9d6da6 | 54 | bool rollback; |
a724632c TG |
55 | bool single; |
56 | bool bringup; | |
cf392d10 | 57 | struct hlist_node *node; |
4cb28ced | 58 | enum cpuhp_state cb_state; |
4cb28ced TG |
59 | int result; |
60 | struct completion done; | |
61 | #endif | |
cff7d378 TG |
62 | }; |
63 | ||
64 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
65 | ||
66 | /** | |
67 | * cpuhp_step - Hotplug state machine step | |
68 | * @name: Name of the step | |
69 | * @startup: Startup function of the step | |
70 | * @teardown: Teardown function of the step | |
71 | * @skip_onerr: Do not invoke the functions on error rollback | |
72 | * Will go away once the notifiers are gone | |
757c989b | 73 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
74 | */ |
75 | struct cpuhp_step { | |
cf392d10 TG |
76 | const char *name; |
77 | union { | |
3c1627e9 TG |
78 | int (*single)(unsigned int cpu); |
79 | int (*multi)(unsigned int cpu, | |
80 | struct hlist_node *node); | |
81 | } startup; | |
cf392d10 | 82 | union { |
3c1627e9 TG |
83 | int (*single)(unsigned int cpu); |
84 | int (*multi)(unsigned int cpu, | |
85 | struct hlist_node *node); | |
86 | } teardown; | |
cf392d10 TG |
87 | struct hlist_head list; |
88 | bool skip_onerr; | |
89 | bool cant_stop; | |
90 | bool multi_instance; | |
cff7d378 TG |
91 | }; |
92 | ||
98f8cdce | 93 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 94 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 95 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 96 | |
a724632c TG |
97 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
98 | { | |
99 | /* | |
100 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
101 | * purposes as that state is handled explicitly in cpu_down. | |
102 | */ | |
103 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
104 | } | |
105 | ||
106 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
107 | { | |
108 | struct cpuhp_step *sp; | |
109 | ||
110 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
111 | return sp + state; | |
112 | } | |
113 | ||
cff7d378 TG |
114 | /** |
115 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
116 | * @cpu: The cpu for which the callback should be invoked | |
117 | * @step: The step in the state machine | |
a724632c | 118 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 119 | * |
cf392d10 | 120 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 121 | */ |
a724632c | 122 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 123 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
124 | { |
125 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 126 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
127 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
128 | int (*cb)(unsigned int cpu); | |
129 | int ret, cnt; | |
130 | ||
131 | if (!step->multi_instance) { | |
3c1627e9 | 132 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
133 | if (!cb) |
134 | return 0; | |
a724632c | 135 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 136 | ret = cb(cpu); |
a724632c | 137 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
138 | return ret; |
139 | } | |
3c1627e9 | 140 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
141 | if (!cbm) |
142 | return 0; | |
143 | ||
144 | /* Single invocation for instance add/remove */ | |
145 | if (node) { | |
146 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
147 | ret = cbm(cpu, node); | |
148 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
149 | return ret; | |
150 | } | |
151 | ||
152 | /* State transition. Invoke on all instances */ | |
153 | cnt = 0; | |
154 | hlist_for_each(node, &step->list) { | |
155 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
156 | ret = cbm(cpu, node); | |
157 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
158 | if (ret) | |
159 | goto err; | |
160 | cnt++; | |
161 | } | |
162 | return 0; | |
163 | err: | |
164 | /* Rollback the instances if one failed */ | |
3c1627e9 | 165 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
166 | if (!cbm) |
167 | return ret; | |
168 | ||
169 | hlist_for_each(node, &step->list) { | |
170 | if (!cnt--) | |
171 | break; | |
172 | cbm(cpu, node); | |
cff7d378 TG |
173 | } |
174 | return ret; | |
175 | } | |
176 | ||
98a79d6a | 177 | #ifdef CONFIG_SMP |
b3199c02 | 178 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 179 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
180 | bool cpuhp_tasks_frozen; |
181 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 182 | |
79a6cdeb | 183 | /* |
93ae4f97 SB |
184 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
185 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
186 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
187 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
188 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
189 | */ |
190 | void cpu_maps_update_begin(void) | |
191 | { | |
192 | mutex_lock(&cpu_add_remove_lock); | |
193 | } | |
93ae4f97 | 194 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
195 | |
196 | void cpu_maps_update_done(void) | |
197 | { | |
198 | mutex_unlock(&cpu_add_remove_lock); | |
199 | } | |
93ae4f97 | 200 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 201 | |
5c113fbe | 202 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 203 | |
e3920fb4 RW |
204 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
205 | * Should always be manipulated under cpu_add_remove_lock | |
206 | */ | |
207 | static int cpu_hotplug_disabled; | |
208 | ||
79a6cdeb LJ |
209 | #ifdef CONFIG_HOTPLUG_CPU |
210 | ||
d221938c GS |
211 | static struct { |
212 | struct task_struct *active_writer; | |
87af9e7f DH |
213 | /* wait queue to wake up the active_writer */ |
214 | wait_queue_head_t wq; | |
215 | /* verifies that no writer will get active while readers are active */ | |
216 | struct mutex lock; | |
d221938c GS |
217 | /* |
218 | * Also blocks the new readers during | |
219 | * an ongoing cpu hotplug operation. | |
220 | */ | |
87af9e7f | 221 | atomic_t refcount; |
a19423b9 GS |
222 | |
223 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
224 | struct lockdep_map dep_map; | |
225 | #endif | |
31950eb6 LT |
226 | } cpu_hotplug = { |
227 | .active_writer = NULL, | |
87af9e7f | 228 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 229 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 | 230 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a705e07b | 231 | .dep_map = STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug.dep_map), |
a19423b9 | 232 | #endif |
31950eb6 | 233 | }; |
d221938c | 234 | |
a19423b9 GS |
235 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
236 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
237 | #define cpuhp_lock_acquire_tryread() \ |
238 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
239 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
240 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
241 | ||
62db99f4 | 242 | |
86ef5c9a | 243 | void get_online_cpus(void) |
a9d9baa1 | 244 | { |
d221938c GS |
245 | might_sleep(); |
246 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 247 | return; |
a19423b9 | 248 | cpuhp_lock_acquire_read(); |
d221938c | 249 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 250 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 251 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 252 | } |
86ef5c9a | 253 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 254 | |
86ef5c9a | 255 | void put_online_cpus(void) |
a9d9baa1 | 256 | { |
87af9e7f DH |
257 | int refcount; |
258 | ||
d221938c | 259 | if (cpu_hotplug.active_writer == current) |
aa953877 | 260 | return; |
075663d1 | 261 | |
87af9e7f DH |
262 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
263 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
264 | atomic_inc(&cpu_hotplug.refcount); | |
265 | ||
266 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
267 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 268 | |
a19423b9 | 269 | cpuhp_lock_release(); |
d221938c | 270 | |
a9d9baa1 | 271 | } |
86ef5c9a | 272 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 273 | |
d221938c GS |
274 | /* |
275 | * This ensures that the hotplug operation can begin only when the | |
276 | * refcount goes to zero. | |
277 | * | |
278 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
279 | * will be blocked by the cpu_hotplug.lock | |
280 | * | |
d2ba7e2a ON |
281 | * Since cpu_hotplug_begin() is always called after invoking |
282 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
283 | * |
284 | * Note that theoretically, there is a possibility of a livelock: | |
285 | * - Refcount goes to zero, last reader wakes up the sleeping | |
286 | * writer. | |
287 | * - Last reader unlocks the cpu_hotplug.lock. | |
288 | * - A new reader arrives at this moment, bumps up the refcount. | |
289 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
290 | * non zero and goes to sleep again. | |
291 | * | |
292 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 293 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
294 | * |
295 | */ | |
b9d10be7 | 296 | void cpu_hotplug_begin(void) |
d221938c | 297 | { |
87af9e7f | 298 | DEFINE_WAIT(wait); |
d2ba7e2a | 299 | |
87af9e7f | 300 | cpu_hotplug.active_writer = current; |
a19423b9 | 301 | cpuhp_lock_acquire(); |
87af9e7f | 302 | |
d2ba7e2a ON |
303 | for (;;) { |
304 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
305 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
306 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
307 | break; | |
d221938c GS |
308 | mutex_unlock(&cpu_hotplug.lock); |
309 | schedule(); | |
d221938c | 310 | } |
87af9e7f | 311 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
312 | } |
313 | ||
b9d10be7 | 314 | void cpu_hotplug_done(void) |
d221938c GS |
315 | { |
316 | cpu_hotplug.active_writer = NULL; | |
317 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 318 | cpuhp_lock_release(); |
d221938c | 319 | } |
79a6cdeb | 320 | |
16e53dbf SB |
321 | /* |
322 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
323 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
324 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
325 | * hotplug path before performing hotplug operations. So acquiring that lock | |
326 | * guarantees mutual exclusion from any currently running hotplug operations. | |
327 | */ | |
328 | void cpu_hotplug_disable(void) | |
329 | { | |
330 | cpu_maps_update_begin(); | |
89af7ba5 | 331 | cpu_hotplug_disabled++; |
16e53dbf SB |
332 | cpu_maps_update_done(); |
333 | } | |
32145c46 | 334 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 335 | |
01b41159 LW |
336 | static void __cpu_hotplug_enable(void) |
337 | { | |
338 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
339 | return; | |
340 | cpu_hotplug_disabled--; | |
341 | } | |
342 | ||
16e53dbf SB |
343 | void cpu_hotplug_enable(void) |
344 | { | |
345 | cpu_maps_update_begin(); | |
01b41159 | 346 | __cpu_hotplug_enable(); |
16e53dbf SB |
347 | cpu_maps_update_done(); |
348 | } | |
32145c46 | 349 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 350 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 351 | |
1da177e4 | 352 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 353 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 354 | { |
bd5349cf | 355 | int ret; |
d221938c | 356 | cpu_maps_update_begin(); |
bd5349cf | 357 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 358 | cpu_maps_update_done(); |
bd5349cf | 359 | return ret; |
1da177e4 | 360 | } |
65edc68c | 361 | |
71cf5aee | 362 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
363 | { |
364 | return raw_notifier_chain_register(&cpu_chain, nb); | |
365 | } | |
366 | ||
090e77c3 | 367 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
368 | int *nr_calls) |
369 | { | |
090e77c3 TG |
370 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
371 | void *hcpu = (void *)(long)cpu; | |
372 | ||
e6bde73b AM |
373 | int ret; |
374 | ||
090e77c3 | 375 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 376 | nr_calls); |
e6bde73b AM |
377 | |
378 | return notifier_to_errno(ret); | |
e9fb7631 AM |
379 | } |
380 | ||
090e77c3 | 381 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 382 | { |
090e77c3 | 383 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
384 | } |
385 | ||
3b9d6da6 SAS |
386 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
387 | { | |
388 | BUG_ON(cpu_notify(val, cpu)); | |
389 | } | |
390 | ||
ba997462 TG |
391 | /* Notifier wrappers for transitioning to state machine */ |
392 | static int notify_prepare(unsigned int cpu) | |
393 | { | |
394 | int nr_calls = 0; | |
395 | int ret; | |
396 | ||
397 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
398 | if (ret) { | |
399 | nr_calls--; | |
400 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
401 | __func__, cpu); | |
402 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
403 | } | |
404 | return ret; | |
405 | } | |
406 | ||
407 | static int notify_online(unsigned int cpu) | |
408 | { | |
409 | cpu_notify(CPU_ONLINE, cpu); | |
410 | return 0; | |
411 | } | |
412 | ||
8df3e07e TG |
413 | static int bringup_wait_for_ap(unsigned int cpu) |
414 | { | |
415 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
416 | ||
417 | wait_for_completion(&st->done); | |
418 | return st->result; | |
419 | } | |
420 | ||
ba997462 TG |
421 | static int bringup_cpu(unsigned int cpu) |
422 | { | |
423 | struct task_struct *idle = idle_thread_get(cpu); | |
424 | int ret; | |
425 | ||
aa877175 BO |
426 | /* |
427 | * Some architectures have to walk the irq descriptors to | |
428 | * setup the vector space for the cpu which comes online. | |
429 | * Prevent irq alloc/free across the bringup. | |
430 | */ | |
431 | irq_lock_sparse(); | |
432 | ||
ba997462 TG |
433 | /* Arch-specific enabling code. */ |
434 | ret = __cpu_up(cpu, idle); | |
aa877175 | 435 | irq_unlock_sparse(); |
ba997462 TG |
436 | if (ret) { |
437 | cpu_notify(CPU_UP_CANCELED, cpu); | |
438 | return ret; | |
439 | } | |
8df3e07e | 440 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 441 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 442 | return ret; |
ba997462 TG |
443 | } |
444 | ||
2e1a3483 TG |
445 | /* |
446 | * Hotplug state machine related functions | |
447 | */ | |
a724632c | 448 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
449 | { |
450 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 451 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
452 | |
453 | if (!step->skip_onerr) | |
cf392d10 | 454 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
455 | } |
456 | } | |
457 | ||
458 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 459 | enum cpuhp_state target) |
2e1a3483 TG |
460 | { |
461 | enum cpuhp_state prev_state = st->state; | |
462 | int ret = 0; | |
463 | ||
464 | for (; st->state > target; st->state--) { | |
cf392d10 | 465 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
466 | if (ret) { |
467 | st->target = prev_state; | |
a724632c | 468 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
469 | break; |
470 | } | |
471 | } | |
472 | return ret; | |
473 | } | |
474 | ||
a724632c | 475 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
476 | { |
477 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 478 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
479 | |
480 | if (!step->skip_onerr) | |
cf392d10 | 481 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
482 | } |
483 | } | |
484 | ||
485 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 486 | enum cpuhp_state target) |
2e1a3483 TG |
487 | { |
488 | enum cpuhp_state prev_state = st->state; | |
489 | int ret = 0; | |
490 | ||
491 | while (st->state < target) { | |
2e1a3483 | 492 | st->state++; |
cf392d10 | 493 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
494 | if (ret) { |
495 | st->target = prev_state; | |
a724632c | 496 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
497 | break; |
498 | } | |
499 | } | |
500 | return ret; | |
501 | } | |
502 | ||
4cb28ced TG |
503 | /* |
504 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
505 | */ | |
506 | static void cpuhp_create(unsigned int cpu) | |
507 | { | |
508 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
509 | ||
510 | init_completion(&st->done); | |
511 | } | |
512 | ||
513 | static int cpuhp_should_run(unsigned int cpu) | |
514 | { | |
515 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
516 | ||
517 | return st->should_run; | |
518 | } | |
519 | ||
520 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
521 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
522 | { | |
1cf4f629 | 523 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 524 | |
a724632c | 525 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
526 | } |
527 | ||
528 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
529 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
530 | { | |
a724632c | 531 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
532 | } |
533 | ||
534 | /* | |
535 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
536 | * callbacks when a state gets [un]installed at runtime. | |
537 | */ | |
538 | static void cpuhp_thread_fun(unsigned int cpu) | |
539 | { | |
540 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
541 | int ret = 0; | |
542 | ||
543 | /* | |
544 | * Paired with the mb() in cpuhp_kick_ap_work and | |
545 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
546 | */ | |
547 | smp_mb(); | |
548 | if (!st->should_run) | |
549 | return; | |
550 | ||
551 | st->should_run = false; | |
552 | ||
553 | /* Single callback invocation for [un]install ? */ | |
a724632c | 554 | if (st->single) { |
4cb28ced TG |
555 | if (st->cb_state < CPUHP_AP_ONLINE) { |
556 | local_irq_disable(); | |
a724632c | 557 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 558 | st->bringup, st->node); |
4cb28ced TG |
559 | local_irq_enable(); |
560 | } else { | |
a724632c | 561 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 562 | st->bringup, st->node); |
4cb28ced | 563 | } |
3b9d6da6 SAS |
564 | } else if (st->rollback) { |
565 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
566 | ||
a724632c | 567 | undo_cpu_down(cpu, st); |
3b9d6da6 SAS |
568 | /* |
569 | * This is a momentary workaround to keep the notifier users | |
570 | * happy. Will go away once we got rid of the notifiers. | |
571 | */ | |
572 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); | |
573 | st->rollback = false; | |
4cb28ced | 574 | } else { |
1cf4f629 | 575 | /* Cannot happen .... */ |
8df3e07e | 576 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 577 | |
4cb28ced TG |
578 | /* Regular hotplug work */ |
579 | if (st->state < st->target) | |
580 | ret = cpuhp_ap_online(cpu, st); | |
581 | else if (st->state > st->target) | |
582 | ret = cpuhp_ap_offline(cpu, st); | |
583 | } | |
584 | st->result = ret; | |
585 | complete(&st->done); | |
586 | } | |
587 | ||
588 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 589 | static int |
cf392d10 TG |
590 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
591 | struct hlist_node *node) | |
4cb28ced TG |
592 | { |
593 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
594 | ||
595 | if (!cpu_online(cpu)) | |
596 | return 0; | |
597 | ||
6a4e2451 TG |
598 | /* |
599 | * If we are up and running, use the hotplug thread. For early calls | |
600 | * we invoke the thread function directly. | |
601 | */ | |
602 | if (!st->thread) | |
cf392d10 | 603 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 604 | |
4cb28ced | 605 | st->cb_state = state; |
a724632c TG |
606 | st->single = true; |
607 | st->bringup = bringup; | |
cf392d10 | 608 | st->node = node; |
a724632c | 609 | |
4cb28ced TG |
610 | /* |
611 | * Make sure the above stores are visible before should_run becomes | |
612 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
613 | */ | |
614 | smp_mb(); | |
615 | st->should_run = true; | |
616 | wake_up_process(st->thread); | |
617 | wait_for_completion(&st->done); | |
618 | return st->result; | |
619 | } | |
620 | ||
621 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 622 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 623 | { |
4cb28ced | 624 | st->result = 0; |
a724632c | 625 | st->single = false; |
4cb28ced TG |
626 | /* |
627 | * Make sure the above stores are visible before should_run becomes | |
628 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
629 | */ | |
630 | smp_mb(); | |
631 | st->should_run = true; | |
632 | wake_up_process(st->thread); | |
1cf4f629 TG |
633 | } |
634 | ||
635 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
636 | { | |
637 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
638 | enum cpuhp_state state = st->state; | |
639 | ||
640 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
641 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
642 | wait_for_completion(&st->done); |
643 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
644 | return st->result; | |
645 | } | |
646 | ||
647 | static struct smp_hotplug_thread cpuhp_threads = { | |
648 | .store = &cpuhp_state.thread, | |
649 | .create = &cpuhp_create, | |
650 | .thread_should_run = cpuhp_should_run, | |
651 | .thread_fn = cpuhp_thread_fun, | |
652 | .thread_comm = "cpuhp/%u", | |
653 | .selfparking = true, | |
654 | }; | |
655 | ||
656 | void __init cpuhp_threads_init(void) | |
657 | { | |
658 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
659 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
660 | } | |
661 | ||
00b9b0af | 662 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 663 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 664 | EXPORT_SYMBOL(__register_cpu_notifier); |
71cf5aee | 665 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 666 | { |
d221938c | 667 | cpu_maps_update_begin(); |
bd5349cf | 668 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 669 | cpu_maps_update_done(); |
1da177e4 LT |
670 | } |
671 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
672 | ||
71cf5aee | 673 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
674 | { |
675 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
676 | } | |
677 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
678 | ||
e4cc2f87 AV |
679 | /** |
680 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
681 | * @cpu: a CPU id | |
682 | * | |
683 | * This function walks all processes, finds a valid mm struct for each one and | |
684 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
685 | * trivial, there are various non-obvious corner cases, which this function | |
686 | * tries to solve in a safe manner. | |
687 | * | |
688 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
689 | * be called only for an already offlined CPU. | |
690 | */ | |
cb79295e AV |
691 | void clear_tasks_mm_cpumask(int cpu) |
692 | { | |
693 | struct task_struct *p; | |
694 | ||
695 | /* | |
696 | * This function is called after the cpu is taken down and marked | |
697 | * offline, so its not like new tasks will ever get this cpu set in | |
698 | * their mm mask. -- Peter Zijlstra | |
699 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
700 | * full-fledged tasklist_lock. | |
701 | */ | |
e4cc2f87 | 702 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
703 | rcu_read_lock(); |
704 | for_each_process(p) { | |
705 | struct task_struct *t; | |
706 | ||
e4cc2f87 AV |
707 | /* |
708 | * Main thread might exit, but other threads may still have | |
709 | * a valid mm. Find one. | |
710 | */ | |
cb79295e AV |
711 | t = find_lock_task_mm(p); |
712 | if (!t) | |
713 | continue; | |
714 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
715 | task_unlock(t); | |
716 | } | |
717 | rcu_read_unlock(); | |
718 | } | |
719 | ||
b728ca06 | 720 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 721 | { |
b728ca06 | 722 | struct task_struct *g, *p; |
1da177e4 | 723 | |
a75a6068 ON |
724 | read_lock(&tasklist_lock); |
725 | for_each_process_thread(g, p) { | |
b728ca06 KT |
726 | if (!p->on_rq) |
727 | continue; | |
728 | /* | |
729 | * We do the check with unlocked task_rq(p)->lock. | |
730 | * Order the reading to do not warn about a task, | |
731 | * which was running on this cpu in the past, and | |
732 | * it's just been woken on another cpu. | |
733 | */ | |
734 | rmb(); | |
735 | if (task_cpu(p) != dead_cpu) | |
736 | continue; | |
737 | ||
738 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
739 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
740 | } |
741 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
742 | } |
743 | ||
98458172 TG |
744 | static int notify_down_prepare(unsigned int cpu) |
745 | { | |
746 | int err, nr_calls = 0; | |
747 | ||
748 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
749 | if (err) { | |
750 | nr_calls--; | |
751 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
752 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
753 | __func__, cpu); | |
754 | } | |
755 | return err; | |
756 | } | |
757 | ||
1da177e4 | 758 | /* Take this CPU down. */ |
71cf5aee | 759 | static int take_cpu_down(void *_param) |
1da177e4 | 760 | { |
4baa0afc TG |
761 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
762 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 763 | int err, cpu = smp_processor_id(); |
1da177e4 | 764 | |
1da177e4 LT |
765 | /* Ensure this CPU doesn't handle any more interrupts. */ |
766 | err = __cpu_disable(); | |
767 | if (err < 0) | |
f3705136 | 768 | return err; |
1da177e4 | 769 | |
a724632c TG |
770 | /* |
771 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
772 | * do this step again. | |
773 | */ | |
774 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
775 | st->state--; | |
4baa0afc | 776 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 777 | for (; st->state > target; st->state--) |
cf392d10 | 778 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 779 | |
52c063d1 TG |
780 | /* Give up timekeeping duties */ |
781 | tick_handover_do_timer(); | |
14e568e7 | 782 | /* Park the stopper thread */ |
090e77c3 | 783 | stop_machine_park(cpu); |
f3705136 | 784 | return 0; |
1da177e4 LT |
785 | } |
786 | ||
98458172 | 787 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 788 | { |
e69aab13 | 789 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 790 | int err; |
1da177e4 | 791 | |
2a58c527 | 792 | /* Park the smpboot threads */ |
1cf4f629 | 793 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 794 | smpboot_park_threads(cpu); |
1cf4f629 | 795 | |
6acce3ef | 796 | /* |
a8994181 TG |
797 | * Prevent irq alloc/free while the dying cpu reorganizes the |
798 | * interrupt affinities. | |
6acce3ef | 799 | */ |
a8994181 | 800 | irq_lock_sparse(); |
6acce3ef | 801 | |
a8994181 TG |
802 | /* |
803 | * So now all preempt/rcu users must observe !cpu_active(). | |
804 | */ | |
090e77c3 | 805 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 806 | if (err) { |
3b9d6da6 | 807 | /* CPU refused to die */ |
a8994181 | 808 | irq_unlock_sparse(); |
3b9d6da6 SAS |
809 | /* Unpark the hotplug thread so we can rollback there */ |
810 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 811 | return err; |
8fa1d7d3 | 812 | } |
04321587 | 813 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 814 | |
48c5ccae | 815 | /* |
ee1e714b | 816 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
817 | * runnable tasks from the cpu, there's only the idle task left now |
818 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
819 | * |
820 | * Wait for the stop thread to go away. | |
48c5ccae | 821 | */ |
e69aab13 TG |
822 | wait_for_completion(&st->done); |
823 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 824 | |
a8994181 TG |
825 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
826 | irq_unlock_sparse(); | |
827 | ||
345527b1 | 828 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
829 | /* This actually kills the CPU. */ |
830 | __cpu_die(cpu); | |
831 | ||
a49b116d | 832 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
833 | return 0; |
834 | } | |
1da177e4 | 835 | |
98458172 TG |
836 | static int notify_dead(unsigned int cpu) |
837 | { | |
838 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 839 | check_for_tasks(cpu); |
98458172 TG |
840 | return 0; |
841 | } | |
842 | ||
71f87b2f TG |
843 | static void cpuhp_complete_idle_dead(void *arg) |
844 | { | |
845 | struct cpuhp_cpu_state *st = arg; | |
846 | ||
847 | complete(&st->done); | |
848 | } | |
849 | ||
e69aab13 TG |
850 | void cpuhp_report_idle_dead(void) |
851 | { | |
852 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
853 | ||
854 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 855 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
856 | st->state = CPUHP_AP_IDLE_DEAD; |
857 | /* | |
858 | * We cannot call complete after rcu_report_dead() so we delegate it | |
859 | * to an online cpu. | |
860 | */ | |
861 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
862 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
863 | } |
864 | ||
cff7d378 TG |
865 | #else |
866 | #define notify_down_prepare NULL | |
867 | #define takedown_cpu NULL | |
868 | #define notify_dead NULL | |
869 | #endif | |
870 | ||
871 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 872 | |
98458172 | 873 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
874 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
875 | enum cpuhp_state target) | |
98458172 | 876 | { |
cff7d378 TG |
877 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
878 | int prev_state, ret = 0; | |
879 | bool hasdied = false; | |
98458172 TG |
880 | |
881 | if (num_online_cpus() == 1) | |
882 | return -EBUSY; | |
883 | ||
757c989b | 884 | if (!cpu_present(cpu)) |
98458172 TG |
885 | return -EINVAL; |
886 | ||
887 | cpu_hotplug_begin(); | |
888 | ||
889 | cpuhp_tasks_frozen = tasks_frozen; | |
890 | ||
cff7d378 | 891 | prev_state = st->state; |
af1f4045 | 892 | st->target = target; |
1cf4f629 TG |
893 | /* |
894 | * If the current CPU state is in the range of the AP hotplug thread, | |
895 | * then we need to kick the thread. | |
896 | */ | |
8df3e07e | 897 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
898 | ret = cpuhp_kick_ap_work(cpu); |
899 | /* | |
900 | * The AP side has done the error rollback already. Just | |
901 | * return the error code.. | |
902 | */ | |
903 | if (ret) | |
904 | goto out; | |
905 | ||
906 | /* | |
907 | * We might have stopped still in the range of the AP hotplug | |
908 | * thread. Nothing to do anymore. | |
909 | */ | |
8df3e07e | 910 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
911 | goto out; |
912 | } | |
913 | /* | |
8df3e07e | 914 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
915 | * to do the further cleanups. |
916 | */ | |
a724632c | 917 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
918 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
919 | st->target = prev_state; | |
920 | st->rollback = true; | |
921 | cpuhp_kick_ap_work(cpu); | |
922 | } | |
98458172 | 923 | |
cff7d378 | 924 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; |
1cf4f629 | 925 | out: |
d221938c | 926 | cpu_hotplug_done(); |
cff7d378 TG |
927 | /* This post dead nonsense must die */ |
928 | if (!ret && hasdied) | |
090e77c3 | 929 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 930 | return ret; |
e3920fb4 RW |
931 | } |
932 | ||
af1f4045 | 933 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 934 | { |
9ea09af3 | 935 | int err; |
e3920fb4 | 936 | |
d221938c | 937 | cpu_maps_update_begin(); |
e761b772 MK |
938 | |
939 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 940 | err = -EBUSY; |
e761b772 MK |
941 | goto out; |
942 | } | |
943 | ||
af1f4045 | 944 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 945 | |
e761b772 | 946 | out: |
d221938c | 947 | cpu_maps_update_done(); |
1da177e4 LT |
948 | return err; |
949 | } | |
af1f4045 TG |
950 | int cpu_down(unsigned int cpu) |
951 | { | |
952 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
953 | } | |
b62b8ef9 | 954 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
955 | #endif /*CONFIG_HOTPLUG_CPU*/ |
956 | ||
4baa0afc | 957 | /** |
ee1e714b | 958 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
959 | * @cpu: cpu that just started |
960 | * | |
4baa0afc TG |
961 | * It must be called by the arch code on the new cpu, before the new cpu |
962 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
963 | */ | |
964 | void notify_cpu_starting(unsigned int cpu) | |
965 | { | |
966 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
967 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
968 | ||
0c6d4576 | 969 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 970 | while (st->state < target) { |
4baa0afc | 971 | st->state++; |
cf392d10 | 972 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
973 | } |
974 | } | |
975 | ||
949338e3 TG |
976 | /* |
977 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
978 | * the stopper thread and unpark the smpboot threads. If the target state is |
979 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
980 | * cpu further. | |
949338e3 | 981 | */ |
8df3e07e | 982 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 983 | { |
8df3e07e TG |
984 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
985 | unsigned int cpu = smp_processor_id(); | |
986 | ||
987 | /* Happens for the boot cpu */ | |
988 | if (state != CPUHP_AP_ONLINE_IDLE) | |
989 | return; | |
990 | ||
991 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 992 | |
8df3e07e | 993 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 994 | stop_machine_unpark(cpu); |
1cf4f629 | 995 | kthread_unpark(st->thread); |
8df3e07e TG |
996 | |
997 | /* Should we go further up ? */ | |
998 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
999 | __cpuhp_kick_ap_work(st); | |
1000 | else | |
1001 | complete(&st->done); | |
949338e3 TG |
1002 | } |
1003 | ||
e3920fb4 | 1004 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1005 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1006 | { |
cff7d378 | 1007 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1008 | struct task_struct *idle; |
2e1a3483 | 1009 | int ret = 0; |
1da177e4 | 1010 | |
d221938c | 1011 | cpu_hotplug_begin(); |
38498a67 | 1012 | |
757c989b | 1013 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1014 | ret = -EINVAL; |
1015 | goto out; | |
1016 | } | |
1017 | ||
757c989b TG |
1018 | /* |
1019 | * The caller of do_cpu_up might have raced with another | |
1020 | * caller. Ignore it for now. | |
1021 | */ | |
1022 | if (st->state >= target) | |
38498a67 | 1023 | goto out; |
757c989b TG |
1024 | |
1025 | if (st->state == CPUHP_OFFLINE) { | |
1026 | /* Let it fail before we try to bring the cpu up */ | |
1027 | idle = idle_thread_get(cpu); | |
1028 | if (IS_ERR(idle)) { | |
1029 | ret = PTR_ERR(idle); | |
1030 | goto out; | |
1031 | } | |
3bb5d2ee | 1032 | } |
38498a67 | 1033 | |
ba997462 TG |
1034 | cpuhp_tasks_frozen = tasks_frozen; |
1035 | ||
af1f4045 | 1036 | st->target = target; |
1cf4f629 TG |
1037 | /* |
1038 | * If the current CPU state is in the range of the AP hotplug thread, | |
1039 | * then we need to kick the thread once more. | |
1040 | */ | |
8df3e07e | 1041 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1042 | ret = cpuhp_kick_ap_work(cpu); |
1043 | /* | |
1044 | * The AP side has done the error rollback already. Just | |
1045 | * return the error code.. | |
1046 | */ | |
1047 | if (ret) | |
1048 | goto out; | |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1053 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1054 | * responsible for bringing it up to the target state. |
1055 | */ | |
8df3e07e | 1056 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1057 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1058 | out: |
d221938c | 1059 | cpu_hotplug_done(); |
e3920fb4 RW |
1060 | return ret; |
1061 | } | |
1062 | ||
af1f4045 | 1063 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1064 | { |
1065 | int err = 0; | |
cf23422b | 1066 | |
e0b582ec | 1067 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1068 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1069 | cpu); | |
87d5e023 | 1070 | #if defined(CONFIG_IA64) |
84117da5 | 1071 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1072 | #endif |
1073 | return -EINVAL; | |
1074 | } | |
e3920fb4 | 1075 | |
01b0f197 TK |
1076 | err = try_online_node(cpu_to_node(cpu)); |
1077 | if (err) | |
1078 | return err; | |
cf23422b | 1079 | |
d221938c | 1080 | cpu_maps_update_begin(); |
e761b772 MK |
1081 | |
1082 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1083 | err = -EBUSY; |
e761b772 MK |
1084 | goto out; |
1085 | } | |
1086 | ||
af1f4045 | 1087 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1088 | out: |
d221938c | 1089 | cpu_maps_update_done(); |
e3920fb4 RW |
1090 | return err; |
1091 | } | |
af1f4045 TG |
1092 | |
1093 | int cpu_up(unsigned int cpu) | |
1094 | { | |
1095 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1096 | } | |
a513f6ba | 1097 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1098 | |
f3de4be9 | 1099 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1100 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1101 | |
d391e552 | 1102 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1103 | { |
d391e552 | 1104 | int cpu, error = 0; |
e3920fb4 | 1105 | |
d221938c | 1106 | cpu_maps_update_begin(); |
d391e552 JM |
1107 | if (!cpu_online(primary)) |
1108 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
1109 | /* |
1110 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1111 | * with the userspace trying to use the CPU hotplug at the same time |
1112 | */ | |
e0b582ec | 1113 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1114 | |
84117da5 | 1115 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1116 | for_each_online_cpu(cpu) { |
d391e552 | 1117 | if (cpu == primary) |
e3920fb4 | 1118 | continue; |
bb3632c6 | 1119 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1120 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1121 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1122 | if (!error) |
e0b582ec | 1123 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1124 | else { |
84117da5 | 1125 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1126 | break; |
1127 | } | |
1128 | } | |
86886e55 | 1129 | |
89af7ba5 | 1130 | if (!error) |
e3920fb4 | 1131 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1132 | else |
84117da5 | 1133 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1134 | |
1135 | /* | |
1136 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1137 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1138 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1139 | */ | |
1140 | cpu_hotplug_disabled++; | |
1141 | ||
d221938c | 1142 | cpu_maps_update_done(); |
e3920fb4 RW |
1143 | return error; |
1144 | } | |
1145 | ||
d0af9eed SS |
1146 | void __weak arch_enable_nonboot_cpus_begin(void) |
1147 | { | |
1148 | } | |
1149 | ||
1150 | void __weak arch_enable_nonboot_cpus_end(void) | |
1151 | { | |
1152 | } | |
1153 | ||
71cf5aee | 1154 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1155 | { |
1156 | int cpu, error; | |
1157 | ||
1158 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1159 | cpu_maps_update_begin(); |
01b41159 | 1160 | __cpu_hotplug_enable(); |
e0b582ec | 1161 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1162 | goto out; |
e3920fb4 | 1163 | |
84117da5 | 1164 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1165 | |
1166 | arch_enable_nonboot_cpus_begin(); | |
1167 | ||
e0b582ec | 1168 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1169 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1170 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1171 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1172 | if (!error) { |
84117da5 | 1173 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1174 | continue; |
1175 | } | |
84117da5 | 1176 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1177 | } |
d0af9eed SS |
1178 | |
1179 | arch_enable_nonboot_cpus_end(); | |
1180 | ||
e0b582ec | 1181 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1182 | out: |
d221938c | 1183 | cpu_maps_update_done(); |
1da177e4 | 1184 | } |
e0b582ec | 1185 | |
d7268a31 | 1186 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1187 | { |
1188 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1189 | return -ENOMEM; | |
1190 | return 0; | |
1191 | } | |
1192 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1193 | |
79cfbdfa SB |
1194 | /* |
1195 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1196 | * ensure that the state of the system with respect to the tasks being frozen | |
1197 | * or not, as reported by the notification, remains unchanged *throughout the | |
1198 | * duration* of the execution of the callbacks. | |
1199 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1200 | * | |
1201 | * This synchronization is implemented by mutually excluding regular CPU | |
1202 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1203 | * Hibernate notifications. | |
1204 | */ | |
1205 | static int | |
1206 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1207 | unsigned long action, void *ptr) | |
1208 | { | |
1209 | switch (action) { | |
1210 | ||
1211 | case PM_SUSPEND_PREPARE: | |
1212 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1213 | cpu_hotplug_disable(); |
79cfbdfa SB |
1214 | break; |
1215 | ||
1216 | case PM_POST_SUSPEND: | |
1217 | case PM_POST_HIBERNATION: | |
16e53dbf | 1218 | cpu_hotplug_enable(); |
79cfbdfa SB |
1219 | break; |
1220 | ||
1221 | default: | |
1222 | return NOTIFY_DONE; | |
1223 | } | |
1224 | ||
1225 | return NOTIFY_OK; | |
1226 | } | |
1227 | ||
1228 | ||
d7268a31 | 1229 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1230 | { |
6e32d479 FY |
1231 | /* |
1232 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1233 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1234 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1235 | */ | |
79cfbdfa SB |
1236 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1237 | return 0; | |
1238 | } | |
1239 | core_initcall(cpu_hotplug_pm_sync_init); | |
1240 | ||
f3de4be9 | 1241 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
1242 | |
1243 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 1244 | |
cff7d378 TG |
1245 | /* Boot processor state steps */ |
1246 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1247 | [CPUHP_OFFLINE] = { | |
1248 | .name = "offline", | |
3c1627e9 TG |
1249 | .startup.single = NULL, |
1250 | .teardown.single = NULL, | |
cff7d378 TG |
1251 | }, |
1252 | #ifdef CONFIG_SMP | |
1253 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1254 | .name = "threads:prepare", |
3c1627e9 TG |
1255 | .startup.single = smpboot_create_threads, |
1256 | .teardown.single = NULL, | |
757c989b | 1257 | .cant_stop = true, |
cff7d378 | 1258 | }, |
00e16c3d | 1259 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1260 | .name = "perf:prepare", |
1261 | .startup.single = perf_event_init_cpu, | |
1262 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1263 | }, |
7ee681b2 | 1264 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1265 | .name = "workqueue:prepare", |
1266 | .startup.single = workqueue_prepare_cpu, | |
1267 | .teardown.single = NULL, | |
7ee681b2 | 1268 | }, |
27590dc1 | 1269 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1270 | .name = "hrtimers:prepare", |
1271 | .startup.single = hrtimers_prepare_cpu, | |
1272 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1273 | }, |
31487f83 | 1274 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1275 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1276 | .startup.single = smpcfd_prepare_cpu, |
1277 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1278 | }, |
e6d4989a RW |
1279 | [CPUHP_RELAY_PREPARE] = { |
1280 | .name = "relay:prepare", | |
1281 | .startup.single = relay_prepare_cpu, | |
1282 | .teardown.single = NULL, | |
1283 | }, | |
6731d4f1 SAS |
1284 | [CPUHP_SLAB_PREPARE] = { |
1285 | .name = "slab:prepare", | |
1286 | .startup.single = slab_prepare_cpu, | |
1287 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1288 | }, |
4df83742 | 1289 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1290 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1291 | .startup.single = rcutree_prepare_cpu, |
1292 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1293 | }, |
d10ef6f9 TG |
1294 | /* |
1295 | * Preparatory and dead notifiers. Will be replaced once the notifiers | |
1296 | * are converted to states. | |
1297 | */ | |
cff7d378 TG |
1298 | [CPUHP_NOTIFY_PREPARE] = { |
1299 | .name = "notify:prepare", | |
3c1627e9 TG |
1300 | .startup.single = notify_prepare, |
1301 | .teardown.single = notify_dead, | |
cff7d378 | 1302 | .skip_onerr = true, |
757c989b | 1303 | .cant_stop = true, |
cff7d378 | 1304 | }, |
4fae16df RC |
1305 | /* |
1306 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1307 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1308 | * otherwise a RCU stall occurs. | |
1309 | */ | |
1310 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1311 | .name = "timers:dead", |
1312 | .startup.single = NULL, | |
1313 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1314 | }, |
d10ef6f9 | 1315 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1316 | [CPUHP_BRINGUP_CPU] = { |
1317 | .name = "cpu:bringup", | |
3c1627e9 TG |
1318 | .startup.single = bringup_cpu, |
1319 | .teardown.single = NULL, | |
757c989b | 1320 | .cant_stop = true, |
4baa0afc | 1321 | }, |
31487f83 | 1322 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1323 | .name = "smpcfd:dying", |
3c1627e9 TG |
1324 | .startup.single = NULL, |
1325 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1326 | }, |
d10ef6f9 TG |
1327 | /* |
1328 | * Handled on controll processor until the plugged processor manages | |
1329 | * this itself. | |
1330 | */ | |
4baa0afc TG |
1331 | [CPUHP_TEARDOWN_CPU] = { |
1332 | .name = "cpu:teardown", | |
3c1627e9 TG |
1333 | .startup.single = NULL, |
1334 | .teardown.single = takedown_cpu, | |
757c989b | 1335 | .cant_stop = true, |
cff7d378 | 1336 | }, |
a7c73414 TG |
1337 | #else |
1338 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1339 | #endif |
cff7d378 TG |
1340 | }; |
1341 | ||
4baa0afc TG |
1342 | /* Application processor state steps */ |
1343 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1344 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1345 | /* Final state before CPU kills itself */ |
1346 | [CPUHP_AP_IDLE_DEAD] = { | |
1347 | .name = "idle:dead", | |
1348 | }, | |
1349 | /* | |
1350 | * Last state before CPU enters the idle loop to die. Transient state | |
1351 | * for synchronization. | |
1352 | */ | |
1353 | [CPUHP_AP_OFFLINE] = { | |
1354 | .name = "ap:offline", | |
1355 | .cant_stop = true, | |
1356 | }, | |
9cf7243d TG |
1357 | /* First state is scheduler control. Interrupts are disabled */ |
1358 | [CPUHP_AP_SCHED_STARTING] = { | |
1359 | .name = "sched:starting", | |
3c1627e9 TG |
1360 | .startup.single = sched_cpu_starting, |
1361 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1362 | }, |
4df83742 | 1363 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1364 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1365 | .startup.single = NULL, |
1366 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1367 | }, |
d10ef6f9 TG |
1368 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1369 | * state for synchronsization */ | |
1370 | [CPUHP_AP_ONLINE] = { | |
1371 | .name = "ap:online", | |
1372 | }, | |
1373 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1374 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1375 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1376 | .startup.single = smpboot_unpark_threads, |
1377 | .teardown.single = NULL, | |
1cf4f629 | 1378 | }, |
00e16c3d | 1379 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1380 | .name = "perf:online", |
1381 | .startup.single = perf_event_init_cpu, | |
1382 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1383 | }, |
7ee681b2 | 1384 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1385 | .name = "workqueue:online", |
1386 | .startup.single = workqueue_online_cpu, | |
1387 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1388 | }, |
4df83742 | 1389 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1390 | .name = "RCU/tree:online", |
3c1627e9 TG |
1391 | .startup.single = rcutree_online_cpu, |
1392 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1393 | }, |
00e16c3d | 1394 | |
d10ef6f9 TG |
1395 | /* |
1396 | * Online/down_prepare notifiers. Will be removed once the notifiers | |
1397 | * are converted to states. | |
1398 | */ | |
1cf4f629 TG |
1399 | [CPUHP_AP_NOTIFY_ONLINE] = { |
1400 | .name = "notify:online", | |
3c1627e9 TG |
1401 | .startup.single = notify_online, |
1402 | .teardown.single = notify_down_prepare, | |
3b9d6da6 | 1403 | .skip_onerr = true, |
1cf4f629 | 1404 | }, |
4baa0afc | 1405 | #endif |
d10ef6f9 TG |
1406 | /* |
1407 | * The dynamically registered state space is here | |
1408 | */ | |
1409 | ||
aaddd7d1 TG |
1410 | #ifdef CONFIG_SMP |
1411 | /* Last state is scheduler control setting the cpu active */ | |
1412 | [CPUHP_AP_ACTIVE] = { | |
1413 | .name = "sched:active", | |
3c1627e9 TG |
1414 | .startup.single = sched_cpu_activate, |
1415 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1416 | }, |
1417 | #endif | |
1418 | ||
d10ef6f9 | 1419 | /* CPU is fully up and running. */ |
4baa0afc TG |
1420 | [CPUHP_ONLINE] = { |
1421 | .name = "online", | |
3c1627e9 TG |
1422 | .startup.single = NULL, |
1423 | .teardown.single = NULL, | |
4baa0afc TG |
1424 | }, |
1425 | }; | |
1426 | ||
5b7aa87e TG |
1427 | /* Sanity check for callbacks */ |
1428 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1429 | { | |
1430 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1431 | return -EINVAL; | |
1432 | return 0; | |
1433 | } | |
1434 | ||
5b7aa87e TG |
1435 | static void cpuhp_store_callbacks(enum cpuhp_state state, |
1436 | const char *name, | |
1437 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1438 | int (*teardown)(unsigned int cpu), |
1439 | bool multi_instance) | |
5b7aa87e TG |
1440 | { |
1441 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1442 | struct cpuhp_step *sp; | |
1443 | ||
1444 | mutex_lock(&cpuhp_state_mutex); | |
1445 | sp = cpuhp_get_step(state); | |
3c1627e9 TG |
1446 | sp->startup.single = startup; |
1447 | sp->teardown.single = teardown; | |
5b7aa87e | 1448 | sp->name = name; |
cf392d10 TG |
1449 | sp->multi_instance = multi_instance; |
1450 | INIT_HLIST_HEAD(&sp->list); | |
5b7aa87e TG |
1451 | mutex_unlock(&cpuhp_state_mutex); |
1452 | } | |
1453 | ||
1454 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1455 | { | |
3c1627e9 | 1456 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1457 | } |
1458 | ||
5b7aa87e TG |
1459 | /* |
1460 | * Call the startup/teardown function for a step either on the AP or | |
1461 | * on the current CPU. | |
1462 | */ | |
cf392d10 TG |
1463 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1464 | struct hlist_node *node) | |
5b7aa87e | 1465 | { |
a724632c | 1466 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1467 | int ret; |
1468 | ||
3c1627e9 TG |
1469 | if ((bringup && !sp->startup.single) || |
1470 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1471 | return 0; |
5b7aa87e TG |
1472 | /* |
1473 | * The non AP bound callbacks can fail on bringup. On teardown | |
1474 | * e.g. module removal we crash for now. | |
1475 | */ | |
1cf4f629 TG |
1476 | #ifdef CONFIG_SMP |
1477 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1478 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1479 | else |
cf392d10 | 1480 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1481 | #else |
cf392d10 | 1482 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1483 | #endif |
5b7aa87e TG |
1484 | BUG_ON(ret && !bringup); |
1485 | return ret; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1490 | * | |
1491 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1492 | */ | |
1493 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1494 | struct hlist_node *node) |
5b7aa87e TG |
1495 | { |
1496 | int cpu; | |
1497 | ||
5b7aa87e TG |
1498 | /* Roll back the already executed steps on the other cpus */ |
1499 | for_each_present_cpu(cpu) { | |
1500 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1501 | int cpustate = st->state; | |
1502 | ||
1503 | if (cpu >= failedcpu) | |
1504 | break; | |
1505 | ||
1506 | /* Did we invoke the startup call on that cpu ? */ | |
1507 | if (cpustate >= state) | |
cf392d10 | 1508 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1509 | } |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1514 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1515 | * by having no name assigned. | |
1516 | */ | |
1517 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1518 | { | |
1519 | enum cpuhp_state i; | |
1520 | ||
1521 | mutex_lock(&cpuhp_state_mutex); | |
1cf4f629 TG |
1522 | for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) { |
1523 | if (cpuhp_ap_states[i].name) | |
5b7aa87e TG |
1524 | continue; |
1525 | ||
1cf4f629 | 1526 | cpuhp_ap_states[i].name = "Reserved"; |
5b7aa87e TG |
1527 | mutex_unlock(&cpuhp_state_mutex); |
1528 | return i; | |
1529 | } | |
1530 | mutex_unlock(&cpuhp_state_mutex); | |
1531 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1532 | return -ENOSPC; | |
1533 | } | |
1534 | ||
cf392d10 TG |
1535 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, |
1536 | bool invoke) | |
1537 | { | |
1538 | struct cpuhp_step *sp; | |
1539 | int cpu; | |
1540 | int ret; | |
1541 | ||
1542 | sp = cpuhp_get_step(state); | |
1543 | if (sp->multi_instance == false) | |
1544 | return -EINVAL; | |
1545 | ||
1546 | get_online_cpus(); | |
1547 | ||
3c1627e9 | 1548 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1549 | goto add_node; |
1550 | ||
1551 | /* | |
1552 | * Try to call the startup callback for each present cpu | |
1553 | * depending on the hotplug state of the cpu. | |
1554 | */ | |
1555 | for_each_present_cpu(cpu) { | |
1556 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1557 | int cpustate = st->state; | |
1558 | ||
1559 | if (cpustate < state) | |
1560 | continue; | |
1561 | ||
1562 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1563 | if (ret) { | |
3c1627e9 | 1564 | if (sp->teardown.multi) |
cf392d10 TG |
1565 | cpuhp_rollback_install(cpu, state, node); |
1566 | goto err; | |
1567 | } | |
1568 | } | |
1569 | add_node: | |
1570 | ret = 0; | |
1571 | mutex_lock(&cpuhp_state_mutex); | |
1572 | hlist_add_head(node, &sp->list); | |
1573 | mutex_unlock(&cpuhp_state_mutex); | |
1574 | ||
1575 | err: | |
1576 | put_online_cpus(); | |
1577 | return ret; | |
1578 | } | |
1579 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1580 | ||
5b7aa87e TG |
1581 | /** |
1582 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
1583 | * @state: The state to setup | |
1584 | * @invoke: If true, the startup function is invoked for cpus where | |
1585 | * cpu state >= @state | |
1586 | * @startup: startup callback function | |
1587 | * @teardown: teardown callback function | |
1588 | * | |
1358e038 BO |
1589 | * Returns: |
1590 | * On success: | |
1591 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1592 | * 0 for all other states | |
1593 | * On failure: proper (negative) error code | |
5b7aa87e TG |
1594 | */ |
1595 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1596 | const char *name, bool invoke, | |
1597 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1598 | int (*teardown)(unsigned int cpu), |
1599 | bool multi_instance) | |
5b7aa87e TG |
1600 | { |
1601 | int cpu, ret = 0; | |
1602 | int dyn_state = 0; | |
1603 | ||
1604 | if (cpuhp_cb_check(state) || !name) | |
1605 | return -EINVAL; | |
1606 | ||
1607 | get_online_cpus(); | |
1608 | ||
1609 | /* currently assignments for the ONLINE state are possible */ | |
1cf4f629 | 1610 | if (state == CPUHP_AP_ONLINE_DYN) { |
5b7aa87e TG |
1611 | dyn_state = 1; |
1612 | ret = cpuhp_reserve_state(state); | |
1613 | if (ret < 0) | |
1614 | goto out; | |
1615 | state = ret; | |
1616 | } | |
1617 | ||
cf392d10 | 1618 | cpuhp_store_callbacks(state, name, startup, teardown, multi_instance); |
5b7aa87e TG |
1619 | |
1620 | if (!invoke || !startup) | |
1621 | goto out; | |
1622 | ||
1623 | /* | |
1624 | * Try to call the startup callback for each present cpu | |
1625 | * depending on the hotplug state of the cpu. | |
1626 | */ | |
1627 | for_each_present_cpu(cpu) { | |
1628 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1629 | int cpustate = st->state; | |
1630 | ||
1631 | if (cpustate < state) | |
1632 | continue; | |
1633 | ||
cf392d10 | 1634 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1635 | if (ret) { |
a724632c | 1636 | if (teardown) |
cf392d10 TG |
1637 | cpuhp_rollback_install(cpu, state, NULL); |
1638 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1639 | goto out; |
1640 | } | |
1641 | } | |
1642 | out: | |
1643 | put_online_cpus(); | |
1644 | if (!ret && dyn_state) | |
1645 | return state; | |
1646 | return ret; | |
1647 | } | |
1648 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1649 | ||
cf392d10 TG |
1650 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1651 | struct hlist_node *node, bool invoke) | |
1652 | { | |
1653 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1654 | int cpu; | |
1655 | ||
1656 | BUG_ON(cpuhp_cb_check(state)); | |
1657 | ||
1658 | if (!sp->multi_instance) | |
1659 | return -EINVAL; | |
1660 | ||
1661 | get_online_cpus(); | |
1662 | if (!invoke || !cpuhp_get_teardown_cb(state)) | |
1663 | goto remove; | |
1664 | /* | |
1665 | * Call the teardown callback for each present cpu depending | |
1666 | * on the hotplug state of the cpu. This function is not | |
1667 | * allowed to fail currently! | |
1668 | */ | |
1669 | for_each_present_cpu(cpu) { | |
1670 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1671 | int cpustate = st->state; | |
1672 | ||
1673 | if (cpustate >= state) | |
1674 | cpuhp_issue_call(cpu, state, false, node); | |
1675 | } | |
1676 | ||
1677 | remove: | |
1678 | mutex_lock(&cpuhp_state_mutex); | |
1679 | hlist_del(node); | |
1680 | mutex_unlock(&cpuhp_state_mutex); | |
1681 | put_online_cpus(); | |
1682 | ||
1683 | return 0; | |
1684 | } | |
1685 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
5b7aa87e TG |
1686 | /** |
1687 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1688 | * @state: The state to remove | |
1689 | * @invoke: If true, the teardown function is invoked for cpus where | |
1690 | * cpu state >= @state | |
1691 | * | |
1692 | * The teardown callback is currently not allowed to fail. Think | |
1693 | * about module removal! | |
1694 | */ | |
1695 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1696 | { | |
cf392d10 | 1697 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1698 | int cpu; |
1699 | ||
1700 | BUG_ON(cpuhp_cb_check(state)); | |
1701 | ||
1702 | get_online_cpus(); | |
1703 | ||
cf392d10 TG |
1704 | if (sp->multi_instance) { |
1705 | WARN(!hlist_empty(&sp->list), | |
1706 | "Error: Removing state %d which has instances left.\n", | |
1707 | state); | |
1708 | goto remove; | |
1709 | } | |
1710 | ||
a724632c | 1711 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1712 | goto remove; |
1713 | ||
1714 | /* | |
1715 | * Call the teardown callback for each present cpu depending | |
1716 | * on the hotplug state of the cpu. This function is not | |
1717 | * allowed to fail currently! | |
1718 | */ | |
1719 | for_each_present_cpu(cpu) { | |
1720 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1721 | int cpustate = st->state; | |
1722 | ||
1723 | if (cpustate >= state) | |
cf392d10 | 1724 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1725 | } |
1726 | remove: | |
cf392d10 | 1727 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
5b7aa87e TG |
1728 | put_online_cpus(); |
1729 | } | |
1730 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1731 | ||
98f8cdce TG |
1732 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1733 | static ssize_t show_cpuhp_state(struct device *dev, | |
1734 | struct device_attribute *attr, char *buf) | |
1735 | { | |
1736 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1737 | ||
1738 | return sprintf(buf, "%d\n", st->state); | |
1739 | } | |
1740 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1741 | ||
757c989b TG |
1742 | static ssize_t write_cpuhp_target(struct device *dev, |
1743 | struct device_attribute *attr, | |
1744 | const char *buf, size_t count) | |
1745 | { | |
1746 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1747 | struct cpuhp_step *sp; | |
1748 | int target, ret; | |
1749 | ||
1750 | ret = kstrtoint(buf, 10, &target); | |
1751 | if (ret) | |
1752 | return ret; | |
1753 | ||
1754 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1755 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1756 | return -EINVAL; | |
1757 | #else | |
1758 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1759 | return -EINVAL; | |
1760 | #endif | |
1761 | ||
1762 | ret = lock_device_hotplug_sysfs(); | |
1763 | if (ret) | |
1764 | return ret; | |
1765 | ||
1766 | mutex_lock(&cpuhp_state_mutex); | |
1767 | sp = cpuhp_get_step(target); | |
1768 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1769 | mutex_unlock(&cpuhp_state_mutex); | |
1770 | if (ret) | |
1771 | return ret; | |
1772 | ||
1773 | if (st->state < target) | |
1774 | ret = do_cpu_up(dev->id, target); | |
1775 | else | |
1776 | ret = do_cpu_down(dev->id, target); | |
1777 | ||
1778 | unlock_device_hotplug(); | |
1779 | return ret ? ret : count; | |
1780 | } | |
1781 | ||
98f8cdce TG |
1782 | static ssize_t show_cpuhp_target(struct device *dev, |
1783 | struct device_attribute *attr, char *buf) | |
1784 | { | |
1785 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1786 | ||
1787 | return sprintf(buf, "%d\n", st->target); | |
1788 | } | |
757c989b | 1789 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1790 | |
1791 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1792 | &dev_attr_state.attr, | |
1793 | &dev_attr_target.attr, | |
1794 | NULL | |
1795 | }; | |
1796 | ||
1797 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1798 | .attrs = cpuhp_cpu_attrs, | |
1799 | .name = "hotplug", | |
1800 | NULL | |
1801 | }; | |
1802 | ||
1803 | static ssize_t show_cpuhp_states(struct device *dev, | |
1804 | struct device_attribute *attr, char *buf) | |
1805 | { | |
1806 | ssize_t cur, res = 0; | |
1807 | int i; | |
1808 | ||
1809 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1810 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1811 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1812 | ||
1813 | if (sp->name) { | |
1814 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1815 | buf += cur; | |
1816 | res += cur; | |
1817 | } | |
1818 | } | |
1819 | mutex_unlock(&cpuhp_state_mutex); | |
1820 | return res; | |
1821 | } | |
1822 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1823 | ||
1824 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1825 | &dev_attr_states.attr, | |
1826 | NULL | |
1827 | }; | |
1828 | ||
1829 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1830 | .attrs = cpuhp_cpu_root_attrs, | |
1831 | .name = "hotplug", | |
1832 | NULL | |
1833 | }; | |
1834 | ||
1835 | static int __init cpuhp_sysfs_init(void) | |
1836 | { | |
1837 | int cpu, ret; | |
1838 | ||
1839 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1840 | &cpuhp_cpu_root_attr_group); | |
1841 | if (ret) | |
1842 | return ret; | |
1843 | ||
1844 | for_each_possible_cpu(cpu) { | |
1845 | struct device *dev = get_cpu_device(cpu); | |
1846 | ||
1847 | if (!dev) | |
1848 | continue; | |
1849 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1850 | if (ret) | |
1851 | return ret; | |
1852 | } | |
1853 | return 0; | |
1854 | } | |
1855 | device_initcall(cpuhp_sysfs_init); | |
1856 | #endif | |
1857 | ||
e56b3bc7 LT |
1858 | /* |
1859 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1860 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1861 | * | |
e0b582ec | 1862 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1863 | * mask value that has a single bit set only. |
1864 | */ | |
b8d317d1 | 1865 | |
e56b3bc7 | 1866 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1867 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1868 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1869 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1870 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1871 | |
e56b3bc7 LT |
1872 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1873 | ||
1874 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1875 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1876 | #if BITS_PER_LONG > 32 | |
1877 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1878 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1879 | #endif |
1880 | }; | |
e56b3bc7 | 1881 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1882 | |
1883 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1884 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1885 | |
1886 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1887 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1888 | = {CPU_BITS_ALL}; |
b3199c02 | 1889 | #else |
4b804c85 | 1890 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1891 | #endif |
4b804c85 | 1892 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1893 | |
4b804c85 RV |
1894 | struct cpumask __cpu_online_mask __read_mostly; |
1895 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1896 | |
4b804c85 RV |
1897 | struct cpumask __cpu_present_mask __read_mostly; |
1898 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1899 | |
4b804c85 RV |
1900 | struct cpumask __cpu_active_mask __read_mostly; |
1901 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1902 | |
3fa41520 RR |
1903 | void init_cpu_present(const struct cpumask *src) |
1904 | { | |
c4c54dd1 | 1905 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1906 | } |
1907 | ||
1908 | void init_cpu_possible(const struct cpumask *src) | |
1909 | { | |
c4c54dd1 | 1910 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1911 | } |
1912 | ||
1913 | void init_cpu_online(const struct cpumask *src) | |
1914 | { | |
c4c54dd1 | 1915 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1916 | } |
cff7d378 TG |
1917 | |
1918 | /* | |
1919 | * Activate the first processor. | |
1920 | */ | |
1921 | void __init boot_cpu_init(void) | |
1922 | { | |
1923 | int cpu = smp_processor_id(); | |
1924 | ||
1925 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1926 | set_cpu_online(cpu, true); | |
1927 | set_cpu_active(cpu, true); | |
1928 | set_cpu_present(cpu, true); | |
1929 | set_cpu_possible(cpu, true); | |
1930 | } | |
1931 | ||
1932 | /* | |
1933 | * Must be called _AFTER_ setting up the per_cpu areas | |
1934 | */ | |
1935 | void __init boot_cpu_state_init(void) | |
1936 | { | |
1937 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1938 | } |