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> |
cff7d378 | 27 | |
bb3632c6 | 28 | #include <trace/events/power.h> |
cff7d378 TG |
29 | #define CREATE_TRACE_POINTS |
30 | #include <trace/events/cpuhp.h> | |
1da177e4 | 31 | |
38498a67 TG |
32 | #include "smpboot.h" |
33 | ||
cff7d378 TG |
34 | /** |
35 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
36 | * @state: The current cpu state | |
37 | * @target: The target state | |
4cb28ced TG |
38 | * @thread: Pointer to the hotplug thread |
39 | * @should_run: Thread should execute | |
3b9d6da6 | 40 | * @rollback: Perform a rollback |
a724632c TG |
41 | * @single: Single callback invocation |
42 | * @bringup: Single callback bringup or teardown selector | |
43 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced TG |
44 | * @result: Result of the operation |
45 | * @done: Signal completion to the issuer of the task | |
cff7d378 TG |
46 | */ |
47 | struct cpuhp_cpu_state { | |
48 | enum cpuhp_state state; | |
49 | enum cpuhp_state target; | |
4cb28ced TG |
50 | #ifdef CONFIG_SMP |
51 | struct task_struct *thread; | |
52 | bool should_run; | |
3b9d6da6 | 53 | bool rollback; |
a724632c TG |
54 | bool single; |
55 | bool bringup; | |
cf392d10 | 56 | struct hlist_node *node; |
4cb28ced | 57 | enum cpuhp_state cb_state; |
4cb28ced TG |
58 | int result; |
59 | struct completion done; | |
60 | #endif | |
cff7d378 TG |
61 | }; |
62 | ||
63 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
64 | ||
65 | /** | |
66 | * cpuhp_step - Hotplug state machine step | |
67 | * @name: Name of the step | |
68 | * @startup: Startup function of the step | |
69 | * @teardown: Teardown function of the step | |
70 | * @skip_onerr: Do not invoke the functions on error rollback | |
71 | * Will go away once the notifiers are gone | |
757c989b | 72 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
73 | */ |
74 | struct cpuhp_step { | |
cf392d10 TG |
75 | const char *name; |
76 | union { | |
3c1627e9 TG |
77 | int (*single)(unsigned int cpu); |
78 | int (*multi)(unsigned int cpu, | |
79 | struct hlist_node *node); | |
80 | } startup; | |
cf392d10 | 81 | union { |
3c1627e9 TG |
82 | int (*single)(unsigned int cpu); |
83 | int (*multi)(unsigned int cpu, | |
84 | struct hlist_node *node); | |
85 | } teardown; | |
cf392d10 TG |
86 | struct hlist_head list; |
87 | bool skip_onerr; | |
88 | bool cant_stop; | |
89 | bool multi_instance; | |
cff7d378 TG |
90 | }; |
91 | ||
98f8cdce | 92 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 93 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 94 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 95 | |
a724632c TG |
96 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
97 | { | |
98 | /* | |
99 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
100 | * purposes as that state is handled explicitly in cpu_down. | |
101 | */ | |
102 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
103 | } | |
104 | ||
105 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
106 | { | |
107 | struct cpuhp_step *sp; | |
108 | ||
109 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
110 | return sp + state; | |
111 | } | |
112 | ||
cff7d378 TG |
113 | /** |
114 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
115 | * @cpu: The cpu for which the callback should be invoked | |
116 | * @step: The step in the state machine | |
a724632c | 117 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 118 | * |
cf392d10 | 119 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 120 | */ |
a724632c | 121 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 122 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
123 | { |
124 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 125 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
126 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
127 | int (*cb)(unsigned int cpu); | |
128 | int ret, cnt; | |
129 | ||
130 | if (!step->multi_instance) { | |
3c1627e9 | 131 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
132 | if (!cb) |
133 | return 0; | |
a724632c | 134 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 135 | ret = cb(cpu); |
a724632c | 136 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
137 | return ret; |
138 | } | |
3c1627e9 | 139 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
140 | if (!cbm) |
141 | return 0; | |
142 | ||
143 | /* Single invocation for instance add/remove */ | |
144 | if (node) { | |
145 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
146 | ret = cbm(cpu, node); | |
147 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
148 | return ret; | |
149 | } | |
150 | ||
151 | /* State transition. Invoke on all instances */ | |
152 | cnt = 0; | |
153 | hlist_for_each(node, &step->list) { | |
154 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
155 | ret = cbm(cpu, node); | |
156 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
157 | if (ret) | |
158 | goto err; | |
159 | cnt++; | |
160 | } | |
161 | return 0; | |
162 | err: | |
163 | /* Rollback the instances if one failed */ | |
3c1627e9 | 164 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
165 | if (!cbm) |
166 | return ret; | |
167 | ||
168 | hlist_for_each(node, &step->list) { | |
169 | if (!cnt--) | |
170 | break; | |
171 | cbm(cpu, node); | |
cff7d378 TG |
172 | } |
173 | return ret; | |
174 | } | |
175 | ||
98a79d6a | 176 | #ifdef CONFIG_SMP |
b3199c02 | 177 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 178 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
179 | bool cpuhp_tasks_frozen; |
180 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 181 | |
79a6cdeb | 182 | /* |
93ae4f97 SB |
183 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
184 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
185 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
186 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
187 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
188 | */ |
189 | void cpu_maps_update_begin(void) | |
190 | { | |
191 | mutex_lock(&cpu_add_remove_lock); | |
192 | } | |
93ae4f97 | 193 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
194 | |
195 | void cpu_maps_update_done(void) | |
196 | { | |
197 | mutex_unlock(&cpu_add_remove_lock); | |
198 | } | |
93ae4f97 | 199 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 200 | |
5c113fbe | 201 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 202 | |
e3920fb4 RW |
203 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
204 | * Should always be manipulated under cpu_add_remove_lock | |
205 | */ | |
206 | static int cpu_hotplug_disabled; | |
207 | ||
79a6cdeb LJ |
208 | #ifdef CONFIG_HOTPLUG_CPU |
209 | ||
d221938c GS |
210 | static struct { |
211 | struct task_struct *active_writer; | |
87af9e7f DH |
212 | /* wait queue to wake up the active_writer */ |
213 | wait_queue_head_t wq; | |
214 | /* verifies that no writer will get active while readers are active */ | |
215 | struct mutex lock; | |
d221938c GS |
216 | /* |
217 | * Also blocks the new readers during | |
218 | * an ongoing cpu hotplug operation. | |
219 | */ | |
87af9e7f | 220 | atomic_t refcount; |
a19423b9 GS |
221 | |
222 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
223 | struct lockdep_map dep_map; | |
224 | #endif | |
31950eb6 LT |
225 | } cpu_hotplug = { |
226 | .active_writer = NULL, | |
87af9e7f | 227 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 228 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
229 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
230 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
231 | #endif | |
31950eb6 | 232 | }; |
d221938c | 233 | |
a19423b9 GS |
234 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
235 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
236 | #define cpuhp_lock_acquire_tryread() \ |
237 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
238 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
239 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
240 | ||
62db99f4 | 241 | |
86ef5c9a | 242 | void get_online_cpus(void) |
a9d9baa1 | 243 | { |
d221938c GS |
244 | might_sleep(); |
245 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 246 | return; |
a19423b9 | 247 | cpuhp_lock_acquire_read(); |
d221938c | 248 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 249 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 250 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 251 | } |
86ef5c9a | 252 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 253 | |
86ef5c9a | 254 | void put_online_cpus(void) |
a9d9baa1 | 255 | { |
87af9e7f DH |
256 | int refcount; |
257 | ||
d221938c | 258 | if (cpu_hotplug.active_writer == current) |
aa953877 | 259 | return; |
075663d1 | 260 | |
87af9e7f DH |
261 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
262 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
263 | atomic_inc(&cpu_hotplug.refcount); | |
264 | ||
265 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
266 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 267 | |
a19423b9 | 268 | cpuhp_lock_release(); |
d221938c | 269 | |
a9d9baa1 | 270 | } |
86ef5c9a | 271 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 272 | |
d221938c GS |
273 | /* |
274 | * This ensures that the hotplug operation can begin only when the | |
275 | * refcount goes to zero. | |
276 | * | |
277 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
278 | * will be blocked by the cpu_hotplug.lock | |
279 | * | |
d2ba7e2a ON |
280 | * Since cpu_hotplug_begin() is always called after invoking |
281 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
282 | * |
283 | * Note that theoretically, there is a possibility of a livelock: | |
284 | * - Refcount goes to zero, last reader wakes up the sleeping | |
285 | * writer. | |
286 | * - Last reader unlocks the cpu_hotplug.lock. | |
287 | * - A new reader arrives at this moment, bumps up the refcount. | |
288 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
289 | * non zero and goes to sleep again. | |
290 | * | |
291 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 292 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
293 | * |
294 | */ | |
b9d10be7 | 295 | void cpu_hotplug_begin(void) |
d221938c | 296 | { |
87af9e7f | 297 | DEFINE_WAIT(wait); |
d2ba7e2a | 298 | |
87af9e7f | 299 | cpu_hotplug.active_writer = current; |
a19423b9 | 300 | cpuhp_lock_acquire(); |
87af9e7f | 301 | |
d2ba7e2a ON |
302 | for (;;) { |
303 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
304 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
305 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
306 | break; | |
d221938c GS |
307 | mutex_unlock(&cpu_hotplug.lock); |
308 | schedule(); | |
d221938c | 309 | } |
87af9e7f | 310 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
311 | } |
312 | ||
b9d10be7 | 313 | void cpu_hotplug_done(void) |
d221938c GS |
314 | { |
315 | cpu_hotplug.active_writer = NULL; | |
316 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 317 | cpuhp_lock_release(); |
d221938c | 318 | } |
79a6cdeb | 319 | |
16e53dbf SB |
320 | /* |
321 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
322 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
323 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
324 | * hotplug path before performing hotplug operations. So acquiring that lock | |
325 | * guarantees mutual exclusion from any currently running hotplug operations. | |
326 | */ | |
327 | void cpu_hotplug_disable(void) | |
328 | { | |
329 | cpu_maps_update_begin(); | |
89af7ba5 | 330 | cpu_hotplug_disabled++; |
16e53dbf SB |
331 | cpu_maps_update_done(); |
332 | } | |
32145c46 | 333 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 334 | |
01b41159 LW |
335 | static void __cpu_hotplug_enable(void) |
336 | { | |
337 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
338 | return; | |
339 | cpu_hotplug_disabled--; | |
340 | } | |
341 | ||
16e53dbf SB |
342 | void cpu_hotplug_enable(void) |
343 | { | |
344 | cpu_maps_update_begin(); | |
01b41159 | 345 | __cpu_hotplug_enable(); |
16e53dbf SB |
346 | cpu_maps_update_done(); |
347 | } | |
32145c46 | 348 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 349 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 350 | |
1da177e4 | 351 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 352 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 353 | { |
bd5349cf | 354 | int ret; |
d221938c | 355 | cpu_maps_update_begin(); |
bd5349cf | 356 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 357 | cpu_maps_update_done(); |
bd5349cf | 358 | return ret; |
1da177e4 | 359 | } |
65edc68c | 360 | |
71cf5aee | 361 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
362 | { |
363 | return raw_notifier_chain_register(&cpu_chain, nb); | |
364 | } | |
365 | ||
090e77c3 | 366 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
367 | int *nr_calls) |
368 | { | |
090e77c3 TG |
369 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
370 | void *hcpu = (void *)(long)cpu; | |
371 | ||
e6bde73b AM |
372 | int ret; |
373 | ||
090e77c3 | 374 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 375 | nr_calls); |
e6bde73b AM |
376 | |
377 | return notifier_to_errno(ret); | |
e9fb7631 AM |
378 | } |
379 | ||
090e77c3 | 380 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 381 | { |
090e77c3 | 382 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
383 | } |
384 | ||
3b9d6da6 SAS |
385 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
386 | { | |
387 | BUG_ON(cpu_notify(val, cpu)); | |
388 | } | |
389 | ||
ba997462 TG |
390 | /* Notifier wrappers for transitioning to state machine */ |
391 | static int notify_prepare(unsigned int cpu) | |
392 | { | |
393 | int nr_calls = 0; | |
394 | int ret; | |
395 | ||
396 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
397 | if (ret) { | |
398 | nr_calls--; | |
399 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
400 | __func__, cpu); | |
401 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
402 | } | |
403 | return ret; | |
404 | } | |
405 | ||
406 | static int notify_online(unsigned int cpu) | |
407 | { | |
408 | cpu_notify(CPU_ONLINE, cpu); | |
409 | return 0; | |
410 | } | |
411 | ||
8df3e07e TG |
412 | static int bringup_wait_for_ap(unsigned int cpu) |
413 | { | |
414 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
415 | ||
416 | wait_for_completion(&st->done); | |
417 | return st->result; | |
418 | } | |
419 | ||
ba997462 TG |
420 | static int bringup_cpu(unsigned int cpu) |
421 | { | |
422 | struct task_struct *idle = idle_thread_get(cpu); | |
423 | int ret; | |
424 | ||
aa877175 BO |
425 | /* |
426 | * Some architectures have to walk the irq descriptors to | |
427 | * setup the vector space for the cpu which comes online. | |
428 | * Prevent irq alloc/free across the bringup. | |
429 | */ | |
430 | irq_lock_sparse(); | |
431 | ||
ba997462 TG |
432 | /* Arch-specific enabling code. */ |
433 | ret = __cpu_up(cpu, idle); | |
aa877175 | 434 | irq_unlock_sparse(); |
ba997462 TG |
435 | if (ret) { |
436 | cpu_notify(CPU_UP_CANCELED, cpu); | |
437 | return ret; | |
438 | } | |
8df3e07e | 439 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 440 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 441 | return ret; |
ba997462 TG |
442 | } |
443 | ||
2e1a3483 TG |
444 | /* |
445 | * Hotplug state machine related functions | |
446 | */ | |
a724632c | 447 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
448 | { |
449 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 450 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
451 | |
452 | if (!step->skip_onerr) | |
cf392d10 | 453 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
454 | } |
455 | } | |
456 | ||
457 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 458 | enum cpuhp_state target) |
2e1a3483 TG |
459 | { |
460 | enum cpuhp_state prev_state = st->state; | |
461 | int ret = 0; | |
462 | ||
463 | for (; st->state > target; st->state--) { | |
cf392d10 | 464 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
465 | if (ret) { |
466 | st->target = prev_state; | |
a724632c | 467 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
468 | break; |
469 | } | |
470 | } | |
471 | return ret; | |
472 | } | |
473 | ||
a724632c | 474 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
475 | { |
476 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 477 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
478 | |
479 | if (!step->skip_onerr) | |
cf392d10 | 480 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
481 | } |
482 | } | |
483 | ||
484 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 485 | enum cpuhp_state target) |
2e1a3483 TG |
486 | { |
487 | enum cpuhp_state prev_state = st->state; | |
488 | int ret = 0; | |
489 | ||
490 | while (st->state < target) { | |
2e1a3483 | 491 | st->state++; |
cf392d10 | 492 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
493 | if (ret) { |
494 | st->target = prev_state; | |
a724632c | 495 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
496 | break; |
497 | } | |
498 | } | |
499 | return ret; | |
500 | } | |
501 | ||
4cb28ced TG |
502 | /* |
503 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
504 | */ | |
505 | static void cpuhp_create(unsigned int cpu) | |
506 | { | |
507 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
508 | ||
509 | init_completion(&st->done); | |
510 | } | |
511 | ||
512 | static int cpuhp_should_run(unsigned int cpu) | |
513 | { | |
514 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
515 | ||
516 | return st->should_run; | |
517 | } | |
518 | ||
519 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
520 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
521 | { | |
1cf4f629 | 522 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 523 | |
a724632c | 524 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
525 | } |
526 | ||
527 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
528 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
529 | { | |
a724632c | 530 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
531 | } |
532 | ||
533 | /* | |
534 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
535 | * callbacks when a state gets [un]installed at runtime. | |
536 | */ | |
537 | static void cpuhp_thread_fun(unsigned int cpu) | |
538 | { | |
539 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
540 | int ret = 0; | |
541 | ||
542 | /* | |
543 | * Paired with the mb() in cpuhp_kick_ap_work and | |
544 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
545 | */ | |
546 | smp_mb(); | |
547 | if (!st->should_run) | |
548 | return; | |
549 | ||
550 | st->should_run = false; | |
551 | ||
552 | /* Single callback invocation for [un]install ? */ | |
a724632c | 553 | if (st->single) { |
4cb28ced TG |
554 | if (st->cb_state < CPUHP_AP_ONLINE) { |
555 | local_irq_disable(); | |
a724632c | 556 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 557 | st->bringup, st->node); |
4cb28ced TG |
558 | local_irq_enable(); |
559 | } else { | |
a724632c | 560 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 561 | st->bringup, st->node); |
4cb28ced | 562 | } |
3b9d6da6 SAS |
563 | } else if (st->rollback) { |
564 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
565 | ||
a724632c | 566 | undo_cpu_down(cpu, st); |
3b9d6da6 SAS |
567 | /* |
568 | * This is a momentary workaround to keep the notifier users | |
569 | * happy. Will go away once we got rid of the notifiers. | |
570 | */ | |
571 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); | |
572 | st->rollback = false; | |
4cb28ced | 573 | } else { |
1cf4f629 | 574 | /* Cannot happen .... */ |
8df3e07e | 575 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 576 | |
4cb28ced TG |
577 | /* Regular hotplug work */ |
578 | if (st->state < st->target) | |
579 | ret = cpuhp_ap_online(cpu, st); | |
580 | else if (st->state > st->target) | |
581 | ret = cpuhp_ap_offline(cpu, st); | |
582 | } | |
583 | st->result = ret; | |
584 | complete(&st->done); | |
585 | } | |
586 | ||
587 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 588 | static int |
cf392d10 TG |
589 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
590 | struct hlist_node *node) | |
4cb28ced TG |
591 | { |
592 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
593 | ||
594 | if (!cpu_online(cpu)) | |
595 | return 0; | |
596 | ||
6a4e2451 TG |
597 | /* |
598 | * If we are up and running, use the hotplug thread. For early calls | |
599 | * we invoke the thread function directly. | |
600 | */ | |
601 | if (!st->thread) | |
cf392d10 | 602 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 603 | |
4cb28ced | 604 | st->cb_state = state; |
a724632c TG |
605 | st->single = true; |
606 | st->bringup = bringup; | |
cf392d10 | 607 | st->node = node; |
a724632c | 608 | |
4cb28ced TG |
609 | /* |
610 | * Make sure the above stores are visible before should_run becomes | |
611 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
612 | */ | |
613 | smp_mb(); | |
614 | st->should_run = true; | |
615 | wake_up_process(st->thread); | |
616 | wait_for_completion(&st->done); | |
617 | return st->result; | |
618 | } | |
619 | ||
620 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 621 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 622 | { |
4cb28ced | 623 | st->result = 0; |
a724632c | 624 | st->single = false; |
4cb28ced TG |
625 | /* |
626 | * Make sure the above stores are visible before should_run becomes | |
627 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
628 | */ | |
629 | smp_mb(); | |
630 | st->should_run = true; | |
631 | wake_up_process(st->thread); | |
1cf4f629 TG |
632 | } |
633 | ||
634 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
635 | { | |
636 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
637 | enum cpuhp_state state = st->state; | |
638 | ||
639 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
640 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
641 | wait_for_completion(&st->done); |
642 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
643 | return st->result; | |
644 | } | |
645 | ||
646 | static struct smp_hotplug_thread cpuhp_threads = { | |
647 | .store = &cpuhp_state.thread, | |
648 | .create = &cpuhp_create, | |
649 | .thread_should_run = cpuhp_should_run, | |
650 | .thread_fn = cpuhp_thread_fun, | |
651 | .thread_comm = "cpuhp/%u", | |
652 | .selfparking = true, | |
653 | }; | |
654 | ||
655 | void __init cpuhp_threads_init(void) | |
656 | { | |
657 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
658 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
659 | } | |
660 | ||
00b9b0af | 661 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 662 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 663 | EXPORT_SYMBOL(__register_cpu_notifier); |
71cf5aee | 664 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 665 | { |
d221938c | 666 | cpu_maps_update_begin(); |
bd5349cf | 667 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 668 | cpu_maps_update_done(); |
1da177e4 LT |
669 | } |
670 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
671 | ||
71cf5aee | 672 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
673 | { |
674 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
675 | } | |
676 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
677 | ||
e4cc2f87 AV |
678 | /** |
679 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
680 | * @cpu: a CPU id | |
681 | * | |
682 | * This function walks all processes, finds a valid mm struct for each one and | |
683 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
684 | * trivial, there are various non-obvious corner cases, which this function | |
685 | * tries to solve in a safe manner. | |
686 | * | |
687 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
688 | * be called only for an already offlined CPU. | |
689 | */ | |
cb79295e AV |
690 | void clear_tasks_mm_cpumask(int cpu) |
691 | { | |
692 | struct task_struct *p; | |
693 | ||
694 | /* | |
695 | * This function is called after the cpu is taken down and marked | |
696 | * offline, so its not like new tasks will ever get this cpu set in | |
697 | * their mm mask. -- Peter Zijlstra | |
698 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
699 | * full-fledged tasklist_lock. | |
700 | */ | |
e4cc2f87 | 701 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
702 | rcu_read_lock(); |
703 | for_each_process(p) { | |
704 | struct task_struct *t; | |
705 | ||
e4cc2f87 AV |
706 | /* |
707 | * Main thread might exit, but other threads may still have | |
708 | * a valid mm. Find one. | |
709 | */ | |
cb79295e AV |
710 | t = find_lock_task_mm(p); |
711 | if (!t) | |
712 | continue; | |
713 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
714 | task_unlock(t); | |
715 | } | |
716 | rcu_read_unlock(); | |
717 | } | |
718 | ||
b728ca06 | 719 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 720 | { |
b728ca06 | 721 | struct task_struct *g, *p; |
1da177e4 | 722 | |
a75a6068 ON |
723 | read_lock(&tasklist_lock); |
724 | for_each_process_thread(g, p) { | |
b728ca06 KT |
725 | if (!p->on_rq) |
726 | continue; | |
727 | /* | |
728 | * We do the check with unlocked task_rq(p)->lock. | |
729 | * Order the reading to do not warn about a task, | |
730 | * which was running on this cpu in the past, and | |
731 | * it's just been woken on another cpu. | |
732 | */ | |
733 | rmb(); | |
734 | if (task_cpu(p) != dead_cpu) | |
735 | continue; | |
736 | ||
737 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
738 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
739 | } |
740 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
741 | } |
742 | ||
98458172 TG |
743 | static int notify_down_prepare(unsigned int cpu) |
744 | { | |
745 | int err, nr_calls = 0; | |
746 | ||
747 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
748 | if (err) { | |
749 | nr_calls--; | |
750 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
751 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
752 | __func__, cpu); | |
753 | } | |
754 | return err; | |
755 | } | |
756 | ||
1da177e4 | 757 | /* Take this CPU down. */ |
71cf5aee | 758 | static int take_cpu_down(void *_param) |
1da177e4 | 759 | { |
4baa0afc TG |
760 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
761 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 762 | int err, cpu = smp_processor_id(); |
1da177e4 | 763 | |
1da177e4 LT |
764 | /* Ensure this CPU doesn't handle any more interrupts. */ |
765 | err = __cpu_disable(); | |
766 | if (err < 0) | |
f3705136 | 767 | return err; |
1da177e4 | 768 | |
a724632c TG |
769 | /* |
770 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
771 | * do this step again. | |
772 | */ | |
773 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
774 | st->state--; | |
4baa0afc | 775 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 776 | for (; st->state > target; st->state--) |
cf392d10 | 777 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 778 | |
52c063d1 TG |
779 | /* Give up timekeeping duties */ |
780 | tick_handover_do_timer(); | |
14e568e7 | 781 | /* Park the stopper thread */ |
090e77c3 | 782 | stop_machine_park(cpu); |
f3705136 | 783 | return 0; |
1da177e4 LT |
784 | } |
785 | ||
98458172 | 786 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 787 | { |
e69aab13 | 788 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 789 | int err; |
1da177e4 | 790 | |
2a58c527 | 791 | /* Park the smpboot threads */ |
1cf4f629 | 792 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 793 | smpboot_park_threads(cpu); |
1cf4f629 | 794 | |
6acce3ef | 795 | /* |
a8994181 TG |
796 | * Prevent irq alloc/free while the dying cpu reorganizes the |
797 | * interrupt affinities. | |
6acce3ef | 798 | */ |
a8994181 | 799 | irq_lock_sparse(); |
6acce3ef | 800 | |
a8994181 TG |
801 | /* |
802 | * So now all preempt/rcu users must observe !cpu_active(). | |
803 | */ | |
090e77c3 | 804 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 805 | if (err) { |
3b9d6da6 | 806 | /* CPU refused to die */ |
a8994181 | 807 | irq_unlock_sparse(); |
3b9d6da6 SAS |
808 | /* Unpark the hotplug thread so we can rollback there */ |
809 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 810 | return err; |
8fa1d7d3 | 811 | } |
04321587 | 812 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 813 | |
48c5ccae | 814 | /* |
ee1e714b | 815 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
816 | * runnable tasks from the cpu, there's only the idle task left now |
817 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
818 | * |
819 | * Wait for the stop thread to go away. | |
48c5ccae | 820 | */ |
e69aab13 TG |
821 | wait_for_completion(&st->done); |
822 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 823 | |
a8994181 TG |
824 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
825 | irq_unlock_sparse(); | |
826 | ||
345527b1 | 827 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
828 | /* This actually kills the CPU. */ |
829 | __cpu_die(cpu); | |
830 | ||
a49b116d | 831 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
832 | return 0; |
833 | } | |
1da177e4 | 834 | |
98458172 TG |
835 | static int notify_dead(unsigned int cpu) |
836 | { | |
837 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 838 | check_for_tasks(cpu); |
98458172 TG |
839 | return 0; |
840 | } | |
841 | ||
71f87b2f TG |
842 | static void cpuhp_complete_idle_dead(void *arg) |
843 | { | |
844 | struct cpuhp_cpu_state *st = arg; | |
845 | ||
846 | complete(&st->done); | |
847 | } | |
848 | ||
e69aab13 TG |
849 | void cpuhp_report_idle_dead(void) |
850 | { | |
851 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
852 | ||
853 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 854 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
855 | st->state = CPUHP_AP_IDLE_DEAD; |
856 | /* | |
857 | * We cannot call complete after rcu_report_dead() so we delegate it | |
858 | * to an online cpu. | |
859 | */ | |
860 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
861 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
862 | } |
863 | ||
cff7d378 TG |
864 | #else |
865 | #define notify_down_prepare NULL | |
866 | #define takedown_cpu NULL | |
867 | #define notify_dead NULL | |
868 | #endif | |
869 | ||
870 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 871 | |
98458172 | 872 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
873 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
874 | enum cpuhp_state target) | |
98458172 | 875 | { |
cff7d378 TG |
876 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
877 | int prev_state, ret = 0; | |
878 | bool hasdied = false; | |
98458172 TG |
879 | |
880 | if (num_online_cpus() == 1) | |
881 | return -EBUSY; | |
882 | ||
757c989b | 883 | if (!cpu_present(cpu)) |
98458172 TG |
884 | return -EINVAL; |
885 | ||
886 | cpu_hotplug_begin(); | |
887 | ||
888 | cpuhp_tasks_frozen = tasks_frozen; | |
889 | ||
cff7d378 | 890 | prev_state = st->state; |
af1f4045 | 891 | st->target = target; |
1cf4f629 TG |
892 | /* |
893 | * If the current CPU state is in the range of the AP hotplug thread, | |
894 | * then we need to kick the thread. | |
895 | */ | |
8df3e07e | 896 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
897 | ret = cpuhp_kick_ap_work(cpu); |
898 | /* | |
899 | * The AP side has done the error rollback already. Just | |
900 | * return the error code.. | |
901 | */ | |
902 | if (ret) | |
903 | goto out; | |
904 | ||
905 | /* | |
906 | * We might have stopped still in the range of the AP hotplug | |
907 | * thread. Nothing to do anymore. | |
908 | */ | |
8df3e07e | 909 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
910 | goto out; |
911 | } | |
912 | /* | |
8df3e07e | 913 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
914 | * to do the further cleanups. |
915 | */ | |
a724632c | 916 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
917 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
918 | st->target = prev_state; | |
919 | st->rollback = true; | |
920 | cpuhp_kick_ap_work(cpu); | |
921 | } | |
98458172 | 922 | |
cff7d378 | 923 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; |
1cf4f629 | 924 | out: |
d221938c | 925 | cpu_hotplug_done(); |
cff7d378 TG |
926 | /* This post dead nonsense must die */ |
927 | if (!ret && hasdied) | |
090e77c3 | 928 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 929 | return ret; |
e3920fb4 RW |
930 | } |
931 | ||
af1f4045 | 932 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 933 | { |
9ea09af3 | 934 | int err; |
e3920fb4 | 935 | |
d221938c | 936 | cpu_maps_update_begin(); |
e761b772 MK |
937 | |
938 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 939 | err = -EBUSY; |
e761b772 MK |
940 | goto out; |
941 | } | |
942 | ||
af1f4045 | 943 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 944 | |
e761b772 | 945 | out: |
d221938c | 946 | cpu_maps_update_done(); |
1da177e4 LT |
947 | return err; |
948 | } | |
af1f4045 TG |
949 | int cpu_down(unsigned int cpu) |
950 | { | |
951 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
952 | } | |
b62b8ef9 | 953 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
954 | #endif /*CONFIG_HOTPLUG_CPU*/ |
955 | ||
4baa0afc | 956 | /** |
ee1e714b | 957 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
958 | * @cpu: cpu that just started |
959 | * | |
4baa0afc TG |
960 | * It must be called by the arch code on the new cpu, before the new cpu |
961 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
962 | */ | |
963 | void notify_cpu_starting(unsigned int cpu) | |
964 | { | |
965 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
966 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
967 | ||
968 | while (st->state < target) { | |
4baa0afc | 969 | st->state++; |
cf392d10 | 970 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
971 | } |
972 | } | |
973 | ||
949338e3 TG |
974 | /* |
975 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
976 | * the stopper thread and unpark the smpboot threads. If the target state is |
977 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
978 | * cpu further. | |
949338e3 | 979 | */ |
8df3e07e | 980 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 981 | { |
8df3e07e TG |
982 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
983 | unsigned int cpu = smp_processor_id(); | |
984 | ||
985 | /* Happens for the boot cpu */ | |
986 | if (state != CPUHP_AP_ONLINE_IDLE) | |
987 | return; | |
988 | ||
989 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 990 | |
8df3e07e | 991 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 992 | stop_machine_unpark(cpu); |
1cf4f629 | 993 | kthread_unpark(st->thread); |
8df3e07e TG |
994 | |
995 | /* Should we go further up ? */ | |
996 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
997 | __cpuhp_kick_ap_work(st); | |
998 | else | |
999 | complete(&st->done); | |
949338e3 TG |
1000 | } |
1001 | ||
e3920fb4 | 1002 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1003 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1004 | { |
cff7d378 | 1005 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1006 | struct task_struct *idle; |
2e1a3483 | 1007 | int ret = 0; |
1da177e4 | 1008 | |
d221938c | 1009 | cpu_hotplug_begin(); |
38498a67 | 1010 | |
757c989b | 1011 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1012 | ret = -EINVAL; |
1013 | goto out; | |
1014 | } | |
1015 | ||
757c989b TG |
1016 | /* |
1017 | * The caller of do_cpu_up might have raced with another | |
1018 | * caller. Ignore it for now. | |
1019 | */ | |
1020 | if (st->state >= target) | |
38498a67 | 1021 | goto out; |
757c989b TG |
1022 | |
1023 | if (st->state == CPUHP_OFFLINE) { | |
1024 | /* Let it fail before we try to bring the cpu up */ | |
1025 | idle = idle_thread_get(cpu); | |
1026 | if (IS_ERR(idle)) { | |
1027 | ret = PTR_ERR(idle); | |
1028 | goto out; | |
1029 | } | |
3bb5d2ee | 1030 | } |
38498a67 | 1031 | |
ba997462 TG |
1032 | cpuhp_tasks_frozen = tasks_frozen; |
1033 | ||
af1f4045 | 1034 | st->target = target; |
1cf4f629 TG |
1035 | /* |
1036 | * If the current CPU state is in the range of the AP hotplug thread, | |
1037 | * then we need to kick the thread once more. | |
1038 | */ | |
8df3e07e | 1039 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1040 | ret = cpuhp_kick_ap_work(cpu); |
1041 | /* | |
1042 | * The AP side has done the error rollback already. Just | |
1043 | * return the error code.. | |
1044 | */ | |
1045 | if (ret) | |
1046 | goto out; | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1051 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1052 | * responsible for bringing it up to the target state. |
1053 | */ | |
8df3e07e | 1054 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1055 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1056 | out: |
d221938c | 1057 | cpu_hotplug_done(); |
e3920fb4 RW |
1058 | return ret; |
1059 | } | |
1060 | ||
af1f4045 | 1061 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1062 | { |
1063 | int err = 0; | |
cf23422b | 1064 | |
e0b582ec | 1065 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1066 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1067 | cpu); | |
87d5e023 | 1068 | #if defined(CONFIG_IA64) |
84117da5 | 1069 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1070 | #endif |
1071 | return -EINVAL; | |
1072 | } | |
e3920fb4 | 1073 | |
01b0f197 TK |
1074 | err = try_online_node(cpu_to_node(cpu)); |
1075 | if (err) | |
1076 | return err; | |
cf23422b | 1077 | |
d221938c | 1078 | cpu_maps_update_begin(); |
e761b772 MK |
1079 | |
1080 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1081 | err = -EBUSY; |
e761b772 MK |
1082 | goto out; |
1083 | } | |
1084 | ||
af1f4045 | 1085 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1086 | out: |
d221938c | 1087 | cpu_maps_update_done(); |
e3920fb4 RW |
1088 | return err; |
1089 | } | |
af1f4045 TG |
1090 | |
1091 | int cpu_up(unsigned int cpu) | |
1092 | { | |
1093 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1094 | } | |
a513f6ba | 1095 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1096 | |
f3de4be9 | 1097 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1098 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
1099 | |
1100 | int disable_nonboot_cpus(void) | |
1101 | { | |
e9a5f426 | 1102 | int cpu, first_cpu, error = 0; |
e3920fb4 | 1103 | |
d221938c | 1104 | cpu_maps_update_begin(); |
e0b582ec | 1105 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
1106 | /* |
1107 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1108 | * with the userspace trying to use the CPU hotplug at the same time |
1109 | */ | |
e0b582ec | 1110 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1111 | |
84117da5 | 1112 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
1113 | for_each_online_cpu(cpu) { |
1114 | if (cpu == first_cpu) | |
1115 | continue; | |
bb3632c6 | 1116 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1117 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1118 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1119 | if (!error) |
e0b582ec | 1120 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1121 | else { |
84117da5 | 1122 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1123 | break; |
1124 | } | |
1125 | } | |
86886e55 | 1126 | |
89af7ba5 | 1127 | if (!error) |
e3920fb4 | 1128 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1129 | else |
84117da5 | 1130 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1131 | |
1132 | /* | |
1133 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1134 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1135 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1136 | */ | |
1137 | cpu_hotplug_disabled++; | |
1138 | ||
d221938c | 1139 | cpu_maps_update_done(); |
e3920fb4 RW |
1140 | return error; |
1141 | } | |
1142 | ||
d0af9eed SS |
1143 | void __weak arch_enable_nonboot_cpus_begin(void) |
1144 | { | |
1145 | } | |
1146 | ||
1147 | void __weak arch_enable_nonboot_cpus_end(void) | |
1148 | { | |
1149 | } | |
1150 | ||
71cf5aee | 1151 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1152 | { |
1153 | int cpu, error; | |
1154 | ||
1155 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1156 | cpu_maps_update_begin(); |
01b41159 | 1157 | __cpu_hotplug_enable(); |
e0b582ec | 1158 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1159 | goto out; |
e3920fb4 | 1160 | |
84117da5 | 1161 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1162 | |
1163 | arch_enable_nonboot_cpus_begin(); | |
1164 | ||
e0b582ec | 1165 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1166 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1167 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1168 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1169 | if (!error) { |
84117da5 | 1170 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1171 | continue; |
1172 | } | |
84117da5 | 1173 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1174 | } |
d0af9eed SS |
1175 | |
1176 | arch_enable_nonboot_cpus_end(); | |
1177 | ||
e0b582ec | 1178 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1179 | out: |
d221938c | 1180 | cpu_maps_update_done(); |
1da177e4 | 1181 | } |
e0b582ec | 1182 | |
d7268a31 | 1183 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1184 | { |
1185 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1186 | return -ENOMEM; | |
1187 | return 0; | |
1188 | } | |
1189 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1190 | |
79cfbdfa SB |
1191 | /* |
1192 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1193 | * ensure that the state of the system with respect to the tasks being frozen | |
1194 | * or not, as reported by the notification, remains unchanged *throughout the | |
1195 | * duration* of the execution of the callbacks. | |
1196 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1197 | * | |
1198 | * This synchronization is implemented by mutually excluding regular CPU | |
1199 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1200 | * Hibernate notifications. | |
1201 | */ | |
1202 | static int | |
1203 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1204 | unsigned long action, void *ptr) | |
1205 | { | |
1206 | switch (action) { | |
1207 | ||
1208 | case PM_SUSPEND_PREPARE: | |
1209 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1210 | cpu_hotplug_disable(); |
79cfbdfa SB |
1211 | break; |
1212 | ||
1213 | case PM_POST_SUSPEND: | |
1214 | case PM_POST_HIBERNATION: | |
16e53dbf | 1215 | cpu_hotplug_enable(); |
79cfbdfa SB |
1216 | break; |
1217 | ||
1218 | default: | |
1219 | return NOTIFY_DONE; | |
1220 | } | |
1221 | ||
1222 | return NOTIFY_OK; | |
1223 | } | |
1224 | ||
1225 | ||
d7268a31 | 1226 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1227 | { |
6e32d479 FY |
1228 | /* |
1229 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1230 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1231 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1232 | */ | |
79cfbdfa SB |
1233 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1234 | return 0; | |
1235 | } | |
1236 | core_initcall(cpu_hotplug_pm_sync_init); | |
1237 | ||
f3de4be9 | 1238 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
1239 | |
1240 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 1241 | |
cff7d378 TG |
1242 | /* Boot processor state steps */ |
1243 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1244 | [CPUHP_OFFLINE] = { | |
1245 | .name = "offline", | |
3c1627e9 TG |
1246 | .startup.single = NULL, |
1247 | .teardown.single = NULL, | |
cff7d378 TG |
1248 | }, |
1249 | #ifdef CONFIG_SMP | |
1250 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1251 | .name = "threads:prepare", |
3c1627e9 TG |
1252 | .startup.single = smpboot_create_threads, |
1253 | .teardown.single = NULL, | |
757c989b | 1254 | .cant_stop = true, |
cff7d378 | 1255 | }, |
00e16c3d | 1256 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1257 | .name = "perf:prepare", |
1258 | .startup.single = perf_event_init_cpu, | |
1259 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1260 | }, |
7ee681b2 | 1261 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1262 | .name = "workqueue:prepare", |
1263 | .startup.single = workqueue_prepare_cpu, | |
1264 | .teardown.single = NULL, | |
7ee681b2 | 1265 | }, |
27590dc1 | 1266 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1267 | .name = "hrtimers:prepare", |
1268 | .startup.single = hrtimers_prepare_cpu, | |
1269 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1270 | }, |
31487f83 | 1271 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1272 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1273 | .startup.single = smpcfd_prepare_cpu, |
1274 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1275 | }, |
e6d4989a RW |
1276 | [CPUHP_RELAY_PREPARE] = { |
1277 | .name = "relay:prepare", | |
1278 | .startup.single = relay_prepare_cpu, | |
1279 | .teardown.single = NULL, | |
1280 | }, | |
4df83742 | 1281 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1282 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1283 | .startup.single = rcutree_prepare_cpu, |
1284 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1285 | }, |
d10ef6f9 TG |
1286 | /* |
1287 | * Preparatory and dead notifiers. Will be replaced once the notifiers | |
1288 | * are converted to states. | |
1289 | */ | |
cff7d378 TG |
1290 | [CPUHP_NOTIFY_PREPARE] = { |
1291 | .name = "notify:prepare", | |
3c1627e9 TG |
1292 | .startup.single = notify_prepare, |
1293 | .teardown.single = notify_dead, | |
cff7d378 | 1294 | .skip_onerr = true, |
757c989b | 1295 | .cant_stop = true, |
cff7d378 | 1296 | }, |
4fae16df RC |
1297 | /* |
1298 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1299 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1300 | * otherwise a RCU stall occurs. | |
1301 | */ | |
1302 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1303 | .name = "timers:dead", |
1304 | .startup.single = NULL, | |
1305 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1306 | }, |
d10ef6f9 | 1307 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1308 | [CPUHP_BRINGUP_CPU] = { |
1309 | .name = "cpu:bringup", | |
3c1627e9 TG |
1310 | .startup.single = bringup_cpu, |
1311 | .teardown.single = NULL, | |
757c989b | 1312 | .cant_stop = true, |
4baa0afc | 1313 | }, |
31487f83 | 1314 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1315 | .name = "smpcfd:dying", |
3c1627e9 TG |
1316 | .startup.single = NULL, |
1317 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1318 | }, |
d10ef6f9 TG |
1319 | /* |
1320 | * Handled on controll processor until the plugged processor manages | |
1321 | * this itself. | |
1322 | */ | |
4baa0afc TG |
1323 | [CPUHP_TEARDOWN_CPU] = { |
1324 | .name = "cpu:teardown", | |
3c1627e9 TG |
1325 | .startup.single = NULL, |
1326 | .teardown.single = takedown_cpu, | |
757c989b | 1327 | .cant_stop = true, |
cff7d378 | 1328 | }, |
a7c73414 TG |
1329 | #else |
1330 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1331 | #endif |
cff7d378 TG |
1332 | }; |
1333 | ||
4baa0afc TG |
1334 | /* Application processor state steps */ |
1335 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1336 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1337 | /* Final state before CPU kills itself */ |
1338 | [CPUHP_AP_IDLE_DEAD] = { | |
1339 | .name = "idle:dead", | |
1340 | }, | |
1341 | /* | |
1342 | * Last state before CPU enters the idle loop to die. Transient state | |
1343 | * for synchronization. | |
1344 | */ | |
1345 | [CPUHP_AP_OFFLINE] = { | |
1346 | .name = "ap:offline", | |
1347 | .cant_stop = true, | |
1348 | }, | |
9cf7243d TG |
1349 | /* First state is scheduler control. Interrupts are disabled */ |
1350 | [CPUHP_AP_SCHED_STARTING] = { | |
1351 | .name = "sched:starting", | |
3c1627e9 TG |
1352 | .startup.single = sched_cpu_starting, |
1353 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1354 | }, |
4df83742 | 1355 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1356 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1357 | .startup.single = NULL, |
1358 | .teardown.single = rcutree_dying_cpu, | |
4df83742 | 1359 | }, |
d10ef6f9 TG |
1360 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1361 | * state for synchronsization */ | |
1362 | [CPUHP_AP_ONLINE] = { | |
1363 | .name = "ap:online", | |
1364 | }, | |
1365 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1366 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1367 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1368 | .startup.single = smpboot_unpark_threads, |
1369 | .teardown.single = NULL, | |
1cf4f629 | 1370 | }, |
00e16c3d | 1371 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1372 | .name = "perf:online", |
1373 | .startup.single = perf_event_init_cpu, | |
1374 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1375 | }, |
7ee681b2 | 1376 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1377 | .name = "workqueue:online", |
1378 | .startup.single = workqueue_online_cpu, | |
1379 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1380 | }, |
4df83742 | 1381 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1382 | .name = "RCU/tree:online", |
3c1627e9 TG |
1383 | .startup.single = rcutree_online_cpu, |
1384 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1385 | }, |
00e16c3d | 1386 | |
d10ef6f9 TG |
1387 | /* |
1388 | * Online/down_prepare notifiers. Will be removed once the notifiers | |
1389 | * are converted to states. | |
1390 | */ | |
1cf4f629 TG |
1391 | [CPUHP_AP_NOTIFY_ONLINE] = { |
1392 | .name = "notify:online", | |
3c1627e9 TG |
1393 | .startup.single = notify_online, |
1394 | .teardown.single = notify_down_prepare, | |
3b9d6da6 | 1395 | .skip_onerr = true, |
1cf4f629 | 1396 | }, |
4baa0afc | 1397 | #endif |
d10ef6f9 TG |
1398 | /* |
1399 | * The dynamically registered state space is here | |
1400 | */ | |
1401 | ||
aaddd7d1 TG |
1402 | #ifdef CONFIG_SMP |
1403 | /* Last state is scheduler control setting the cpu active */ | |
1404 | [CPUHP_AP_ACTIVE] = { | |
1405 | .name = "sched:active", | |
3c1627e9 TG |
1406 | .startup.single = sched_cpu_activate, |
1407 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1408 | }, |
1409 | #endif | |
1410 | ||
d10ef6f9 | 1411 | /* CPU is fully up and running. */ |
4baa0afc TG |
1412 | [CPUHP_ONLINE] = { |
1413 | .name = "online", | |
3c1627e9 TG |
1414 | .startup.single = NULL, |
1415 | .teardown.single = NULL, | |
4baa0afc TG |
1416 | }, |
1417 | }; | |
1418 | ||
5b7aa87e TG |
1419 | /* Sanity check for callbacks */ |
1420 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1421 | { | |
1422 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1423 | return -EINVAL; | |
1424 | return 0; | |
1425 | } | |
1426 | ||
5b7aa87e TG |
1427 | static void cpuhp_store_callbacks(enum cpuhp_state state, |
1428 | const char *name, | |
1429 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1430 | int (*teardown)(unsigned int cpu), |
1431 | bool multi_instance) | |
5b7aa87e TG |
1432 | { |
1433 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1434 | struct cpuhp_step *sp; | |
1435 | ||
1436 | mutex_lock(&cpuhp_state_mutex); | |
1437 | sp = cpuhp_get_step(state); | |
3c1627e9 TG |
1438 | sp->startup.single = startup; |
1439 | sp->teardown.single = teardown; | |
5b7aa87e | 1440 | sp->name = name; |
cf392d10 TG |
1441 | sp->multi_instance = multi_instance; |
1442 | INIT_HLIST_HEAD(&sp->list); | |
5b7aa87e TG |
1443 | mutex_unlock(&cpuhp_state_mutex); |
1444 | } | |
1445 | ||
1446 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1447 | { | |
3c1627e9 | 1448 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1449 | } |
1450 | ||
5b7aa87e TG |
1451 | /* |
1452 | * Call the startup/teardown function for a step either on the AP or | |
1453 | * on the current CPU. | |
1454 | */ | |
cf392d10 TG |
1455 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1456 | struct hlist_node *node) | |
5b7aa87e | 1457 | { |
a724632c | 1458 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1459 | int ret; |
1460 | ||
3c1627e9 TG |
1461 | if ((bringup && !sp->startup.single) || |
1462 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1463 | return 0; |
5b7aa87e TG |
1464 | /* |
1465 | * The non AP bound callbacks can fail on bringup. On teardown | |
1466 | * e.g. module removal we crash for now. | |
1467 | */ | |
1cf4f629 TG |
1468 | #ifdef CONFIG_SMP |
1469 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1470 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1471 | else |
cf392d10 | 1472 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1473 | #else |
cf392d10 | 1474 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1475 | #endif |
5b7aa87e TG |
1476 | BUG_ON(ret && !bringup); |
1477 | return ret; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1482 | * | |
1483 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1484 | */ | |
cf392d10 TG |
1485 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, |
1486 | struct hlist_node *node) | |
5b7aa87e TG |
1487 | { |
1488 | int cpu; | |
1489 | ||
5b7aa87e TG |
1490 | /* Roll back the already executed steps on the other cpus */ |
1491 | for_each_present_cpu(cpu) { | |
1492 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1493 | int cpustate = st->state; | |
1494 | ||
1495 | if (cpu >= failedcpu) | |
1496 | break; | |
1497 | ||
1498 | /* Did we invoke the startup call on that cpu ? */ | |
1499 | if (cpustate >= state) | |
cf392d10 | 1500 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1501 | } |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1506 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1507 | * by having no name assigned. | |
1508 | */ | |
1509 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1510 | { | |
1511 | enum cpuhp_state i; | |
1512 | ||
1513 | mutex_lock(&cpuhp_state_mutex); | |
1cf4f629 TG |
1514 | for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) { |
1515 | if (cpuhp_ap_states[i].name) | |
5b7aa87e TG |
1516 | continue; |
1517 | ||
1cf4f629 | 1518 | cpuhp_ap_states[i].name = "Reserved"; |
5b7aa87e TG |
1519 | mutex_unlock(&cpuhp_state_mutex); |
1520 | return i; | |
1521 | } | |
1522 | mutex_unlock(&cpuhp_state_mutex); | |
1523 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1524 | return -ENOSPC; | |
1525 | } | |
1526 | ||
cf392d10 TG |
1527 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, |
1528 | bool invoke) | |
1529 | { | |
1530 | struct cpuhp_step *sp; | |
1531 | int cpu; | |
1532 | int ret; | |
1533 | ||
1534 | sp = cpuhp_get_step(state); | |
1535 | if (sp->multi_instance == false) | |
1536 | return -EINVAL; | |
1537 | ||
1538 | get_online_cpus(); | |
1539 | ||
3c1627e9 | 1540 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1541 | goto add_node; |
1542 | ||
1543 | /* | |
1544 | * Try to call the startup callback for each present cpu | |
1545 | * depending on the hotplug state of the cpu. | |
1546 | */ | |
1547 | for_each_present_cpu(cpu) { | |
1548 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1549 | int cpustate = st->state; | |
1550 | ||
1551 | if (cpustate < state) | |
1552 | continue; | |
1553 | ||
1554 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1555 | if (ret) { | |
3c1627e9 | 1556 | if (sp->teardown.multi) |
cf392d10 TG |
1557 | cpuhp_rollback_install(cpu, state, node); |
1558 | goto err; | |
1559 | } | |
1560 | } | |
1561 | add_node: | |
1562 | ret = 0; | |
1563 | mutex_lock(&cpuhp_state_mutex); | |
1564 | hlist_add_head(node, &sp->list); | |
1565 | mutex_unlock(&cpuhp_state_mutex); | |
1566 | ||
1567 | err: | |
1568 | put_online_cpus(); | |
1569 | return ret; | |
1570 | } | |
1571 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1572 | ||
5b7aa87e TG |
1573 | /** |
1574 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
1575 | * @state: The state to setup | |
1576 | * @invoke: If true, the startup function is invoked for cpus where | |
1577 | * cpu state >= @state | |
1578 | * @startup: startup callback function | |
1579 | * @teardown: teardown callback function | |
1580 | * | |
1581 | * Returns 0 if successful, otherwise a proper error code | |
1582 | */ | |
1583 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1584 | const char *name, bool invoke, | |
1585 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1586 | int (*teardown)(unsigned int cpu), |
1587 | bool multi_instance) | |
5b7aa87e TG |
1588 | { |
1589 | int cpu, ret = 0; | |
1590 | int dyn_state = 0; | |
1591 | ||
1592 | if (cpuhp_cb_check(state) || !name) | |
1593 | return -EINVAL; | |
1594 | ||
1595 | get_online_cpus(); | |
1596 | ||
1597 | /* currently assignments for the ONLINE state are possible */ | |
1cf4f629 | 1598 | if (state == CPUHP_AP_ONLINE_DYN) { |
5b7aa87e TG |
1599 | dyn_state = 1; |
1600 | ret = cpuhp_reserve_state(state); | |
1601 | if (ret < 0) | |
1602 | goto out; | |
1603 | state = ret; | |
1604 | } | |
1605 | ||
cf392d10 | 1606 | cpuhp_store_callbacks(state, name, startup, teardown, multi_instance); |
5b7aa87e TG |
1607 | |
1608 | if (!invoke || !startup) | |
1609 | goto out; | |
1610 | ||
1611 | /* | |
1612 | * Try to call the startup callback for each present cpu | |
1613 | * depending on the hotplug state of the cpu. | |
1614 | */ | |
1615 | for_each_present_cpu(cpu) { | |
1616 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1617 | int cpustate = st->state; | |
1618 | ||
1619 | if (cpustate < state) | |
1620 | continue; | |
1621 | ||
cf392d10 | 1622 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1623 | if (ret) { |
a724632c | 1624 | if (teardown) |
cf392d10 TG |
1625 | cpuhp_rollback_install(cpu, state, NULL); |
1626 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1627 | goto out; |
1628 | } | |
1629 | } | |
1630 | out: | |
1631 | put_online_cpus(); | |
1632 | if (!ret && dyn_state) | |
1633 | return state; | |
1634 | return ret; | |
1635 | } | |
1636 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1637 | ||
cf392d10 TG |
1638 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1639 | struct hlist_node *node, bool invoke) | |
1640 | { | |
1641 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1642 | int cpu; | |
1643 | ||
1644 | BUG_ON(cpuhp_cb_check(state)); | |
1645 | ||
1646 | if (!sp->multi_instance) | |
1647 | return -EINVAL; | |
1648 | ||
1649 | get_online_cpus(); | |
1650 | if (!invoke || !cpuhp_get_teardown_cb(state)) | |
1651 | goto remove; | |
1652 | /* | |
1653 | * Call the teardown callback for each present cpu depending | |
1654 | * on the hotplug state of the cpu. This function is not | |
1655 | * allowed to fail currently! | |
1656 | */ | |
1657 | for_each_present_cpu(cpu) { | |
1658 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1659 | int cpustate = st->state; | |
1660 | ||
1661 | if (cpustate >= state) | |
1662 | cpuhp_issue_call(cpu, state, false, node); | |
1663 | } | |
1664 | ||
1665 | remove: | |
1666 | mutex_lock(&cpuhp_state_mutex); | |
1667 | hlist_del(node); | |
1668 | mutex_unlock(&cpuhp_state_mutex); | |
1669 | put_online_cpus(); | |
1670 | ||
1671 | return 0; | |
1672 | } | |
1673 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
5b7aa87e TG |
1674 | /** |
1675 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1676 | * @state: The state to remove | |
1677 | * @invoke: If true, the teardown function is invoked for cpus where | |
1678 | * cpu state >= @state | |
1679 | * | |
1680 | * The teardown callback is currently not allowed to fail. Think | |
1681 | * about module removal! | |
1682 | */ | |
1683 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1684 | { | |
cf392d10 | 1685 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1686 | int cpu; |
1687 | ||
1688 | BUG_ON(cpuhp_cb_check(state)); | |
1689 | ||
1690 | get_online_cpus(); | |
1691 | ||
cf392d10 TG |
1692 | if (sp->multi_instance) { |
1693 | WARN(!hlist_empty(&sp->list), | |
1694 | "Error: Removing state %d which has instances left.\n", | |
1695 | state); | |
1696 | goto remove; | |
1697 | } | |
1698 | ||
a724632c | 1699 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1700 | goto remove; |
1701 | ||
1702 | /* | |
1703 | * Call the teardown callback for each present cpu depending | |
1704 | * on the hotplug state of the cpu. This function is not | |
1705 | * allowed to fail currently! | |
1706 | */ | |
1707 | for_each_present_cpu(cpu) { | |
1708 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1709 | int cpustate = st->state; | |
1710 | ||
1711 | if (cpustate >= state) | |
cf392d10 | 1712 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1713 | } |
1714 | remove: | |
cf392d10 | 1715 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
5b7aa87e TG |
1716 | put_online_cpus(); |
1717 | } | |
1718 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1719 | ||
98f8cdce TG |
1720 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1721 | static ssize_t show_cpuhp_state(struct device *dev, | |
1722 | struct device_attribute *attr, char *buf) | |
1723 | { | |
1724 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1725 | ||
1726 | return sprintf(buf, "%d\n", st->state); | |
1727 | } | |
1728 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1729 | ||
757c989b TG |
1730 | static ssize_t write_cpuhp_target(struct device *dev, |
1731 | struct device_attribute *attr, | |
1732 | const char *buf, size_t count) | |
1733 | { | |
1734 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1735 | struct cpuhp_step *sp; | |
1736 | int target, ret; | |
1737 | ||
1738 | ret = kstrtoint(buf, 10, &target); | |
1739 | if (ret) | |
1740 | return ret; | |
1741 | ||
1742 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1743 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1744 | return -EINVAL; | |
1745 | #else | |
1746 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1747 | return -EINVAL; | |
1748 | #endif | |
1749 | ||
1750 | ret = lock_device_hotplug_sysfs(); | |
1751 | if (ret) | |
1752 | return ret; | |
1753 | ||
1754 | mutex_lock(&cpuhp_state_mutex); | |
1755 | sp = cpuhp_get_step(target); | |
1756 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1757 | mutex_unlock(&cpuhp_state_mutex); | |
1758 | if (ret) | |
1759 | return ret; | |
1760 | ||
1761 | if (st->state < target) | |
1762 | ret = do_cpu_up(dev->id, target); | |
1763 | else | |
1764 | ret = do_cpu_down(dev->id, target); | |
1765 | ||
1766 | unlock_device_hotplug(); | |
1767 | return ret ? ret : count; | |
1768 | } | |
1769 | ||
98f8cdce TG |
1770 | static ssize_t show_cpuhp_target(struct device *dev, |
1771 | struct device_attribute *attr, char *buf) | |
1772 | { | |
1773 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1774 | ||
1775 | return sprintf(buf, "%d\n", st->target); | |
1776 | } | |
757c989b | 1777 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1778 | |
1779 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1780 | &dev_attr_state.attr, | |
1781 | &dev_attr_target.attr, | |
1782 | NULL | |
1783 | }; | |
1784 | ||
1785 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1786 | .attrs = cpuhp_cpu_attrs, | |
1787 | .name = "hotplug", | |
1788 | NULL | |
1789 | }; | |
1790 | ||
1791 | static ssize_t show_cpuhp_states(struct device *dev, | |
1792 | struct device_attribute *attr, char *buf) | |
1793 | { | |
1794 | ssize_t cur, res = 0; | |
1795 | int i; | |
1796 | ||
1797 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1798 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1799 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1800 | ||
1801 | if (sp->name) { | |
1802 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1803 | buf += cur; | |
1804 | res += cur; | |
1805 | } | |
1806 | } | |
1807 | mutex_unlock(&cpuhp_state_mutex); | |
1808 | return res; | |
1809 | } | |
1810 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1811 | ||
1812 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1813 | &dev_attr_states.attr, | |
1814 | NULL | |
1815 | }; | |
1816 | ||
1817 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1818 | .attrs = cpuhp_cpu_root_attrs, | |
1819 | .name = "hotplug", | |
1820 | NULL | |
1821 | }; | |
1822 | ||
1823 | static int __init cpuhp_sysfs_init(void) | |
1824 | { | |
1825 | int cpu, ret; | |
1826 | ||
1827 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1828 | &cpuhp_cpu_root_attr_group); | |
1829 | if (ret) | |
1830 | return ret; | |
1831 | ||
1832 | for_each_possible_cpu(cpu) { | |
1833 | struct device *dev = get_cpu_device(cpu); | |
1834 | ||
1835 | if (!dev) | |
1836 | continue; | |
1837 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1838 | if (ret) | |
1839 | return ret; | |
1840 | } | |
1841 | return 0; | |
1842 | } | |
1843 | device_initcall(cpuhp_sysfs_init); | |
1844 | #endif | |
1845 | ||
e56b3bc7 LT |
1846 | /* |
1847 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1848 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1849 | * | |
e0b582ec | 1850 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1851 | * mask value that has a single bit set only. |
1852 | */ | |
b8d317d1 | 1853 | |
e56b3bc7 | 1854 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1855 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1856 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1857 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1858 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1859 | |
e56b3bc7 LT |
1860 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1861 | ||
1862 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1863 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1864 | #if BITS_PER_LONG > 32 | |
1865 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1866 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1867 | #endif |
1868 | }; | |
e56b3bc7 | 1869 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1870 | |
1871 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1872 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1873 | |
1874 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1875 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1876 | = {CPU_BITS_ALL}; |
b3199c02 | 1877 | #else |
4b804c85 | 1878 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1879 | #endif |
4b804c85 | 1880 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1881 | |
4b804c85 RV |
1882 | struct cpumask __cpu_online_mask __read_mostly; |
1883 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1884 | |
4b804c85 RV |
1885 | struct cpumask __cpu_present_mask __read_mostly; |
1886 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1887 | |
4b804c85 RV |
1888 | struct cpumask __cpu_active_mask __read_mostly; |
1889 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1890 | |
3fa41520 RR |
1891 | void init_cpu_present(const struct cpumask *src) |
1892 | { | |
c4c54dd1 | 1893 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1894 | } |
1895 | ||
1896 | void init_cpu_possible(const struct cpumask *src) | |
1897 | { | |
c4c54dd1 | 1898 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1899 | } |
1900 | ||
1901 | void init_cpu_online(const struct cpumask *src) | |
1902 | { | |
c4c54dd1 | 1903 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1904 | } |
cff7d378 TG |
1905 | |
1906 | /* | |
1907 | * Activate the first processor. | |
1908 | */ | |
1909 | void __init boot_cpu_init(void) | |
1910 | { | |
1911 | int cpu = smp_processor_id(); | |
1912 | ||
1913 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1914 | set_cpu_online(cpu, true); | |
1915 | set_cpu_active(cpu, true); | |
1916 | set_cpu_present(cpu, true); | |
1917 | set_cpu_possible(cpu, true); | |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * Must be called _AFTER_ setting up the per_cpu areas | |
1922 | */ | |
1923 | void __init boot_cpu_state_init(void) | |
1924 | { | |
1925 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1926 | } |