]>
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> |
cff7d378 | 26 | |
bb3632c6 | 27 | #include <trace/events/power.h> |
cff7d378 TG |
28 | #define CREATE_TRACE_POINTS |
29 | #include <trace/events/cpuhp.h> | |
1da177e4 | 30 | |
38498a67 TG |
31 | #include "smpboot.h" |
32 | ||
cff7d378 TG |
33 | /** |
34 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
35 | * @state: The current cpu state | |
36 | * @target: The target state | |
4cb28ced TG |
37 | * @thread: Pointer to the hotplug thread |
38 | * @should_run: Thread should execute | |
3b9d6da6 | 39 | * @rollback: Perform a rollback |
a724632c TG |
40 | * @single: Single callback invocation |
41 | * @bringup: Single callback bringup or teardown selector | |
42 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced TG |
43 | * @result: Result of the operation |
44 | * @done: Signal completion to the issuer of the task | |
cff7d378 TG |
45 | */ |
46 | struct cpuhp_cpu_state { | |
47 | enum cpuhp_state state; | |
48 | enum cpuhp_state target; | |
4cb28ced TG |
49 | #ifdef CONFIG_SMP |
50 | struct task_struct *thread; | |
51 | bool should_run; | |
3b9d6da6 | 52 | bool rollback; |
a724632c TG |
53 | bool single; |
54 | bool bringup; | |
cf392d10 | 55 | struct hlist_node *node; |
4cb28ced | 56 | enum cpuhp_state cb_state; |
4cb28ced TG |
57 | int result; |
58 | struct completion done; | |
59 | #endif | |
cff7d378 TG |
60 | }; |
61 | ||
62 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
63 | ||
64 | /** | |
65 | * cpuhp_step - Hotplug state machine step | |
66 | * @name: Name of the step | |
67 | * @startup: Startup function of the step | |
68 | * @teardown: Teardown function of the step | |
69 | * @skip_onerr: Do not invoke the functions on error rollback | |
70 | * Will go away once the notifiers are gone | |
757c989b | 71 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
72 | */ |
73 | struct cpuhp_step { | |
cf392d10 TG |
74 | const char *name; |
75 | union { | |
3c1627e9 TG |
76 | int (*single)(unsigned int cpu); |
77 | int (*multi)(unsigned int cpu, | |
78 | struct hlist_node *node); | |
79 | } startup; | |
cf392d10 | 80 | union { |
3c1627e9 TG |
81 | int (*single)(unsigned int cpu); |
82 | int (*multi)(unsigned int cpu, | |
83 | struct hlist_node *node); | |
84 | } teardown; | |
cf392d10 TG |
85 | struct hlist_head list; |
86 | bool skip_onerr; | |
87 | bool cant_stop; | |
88 | bool multi_instance; | |
cff7d378 TG |
89 | }; |
90 | ||
98f8cdce | 91 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 92 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 93 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 94 | |
a724632c TG |
95 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
96 | { | |
97 | /* | |
98 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
99 | * purposes as that state is handled explicitly in cpu_down. | |
100 | */ | |
101 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
102 | } | |
103 | ||
104 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
105 | { | |
106 | struct cpuhp_step *sp; | |
107 | ||
108 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
109 | return sp + state; | |
110 | } | |
111 | ||
cff7d378 TG |
112 | /** |
113 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
114 | * @cpu: The cpu for which the callback should be invoked | |
115 | * @step: The step in the state machine | |
a724632c | 116 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 117 | * |
cf392d10 | 118 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 119 | */ |
a724632c | 120 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 121 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
122 | { |
123 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 124 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
125 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
126 | int (*cb)(unsigned int cpu); | |
127 | int ret, cnt; | |
128 | ||
129 | if (!step->multi_instance) { | |
3c1627e9 | 130 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
131 | if (!cb) |
132 | return 0; | |
a724632c | 133 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 134 | ret = cb(cpu); |
a724632c | 135 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
136 | return ret; |
137 | } | |
3c1627e9 | 138 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
139 | if (!cbm) |
140 | return 0; | |
141 | ||
142 | /* Single invocation for instance add/remove */ | |
143 | if (node) { | |
144 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
145 | ret = cbm(cpu, node); | |
146 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
147 | return ret; | |
148 | } | |
149 | ||
150 | /* State transition. Invoke on all instances */ | |
151 | cnt = 0; | |
152 | hlist_for_each(node, &step->list) { | |
153 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
154 | ret = cbm(cpu, node); | |
155 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
156 | if (ret) | |
157 | goto err; | |
158 | cnt++; | |
159 | } | |
160 | return 0; | |
161 | err: | |
162 | /* Rollback the instances if one failed */ | |
3c1627e9 | 163 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
164 | if (!cbm) |
165 | return ret; | |
166 | ||
167 | hlist_for_each(node, &step->list) { | |
168 | if (!cnt--) | |
169 | break; | |
170 | cbm(cpu, node); | |
cff7d378 TG |
171 | } |
172 | return ret; | |
173 | } | |
174 | ||
98a79d6a | 175 | #ifdef CONFIG_SMP |
b3199c02 | 176 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 177 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
178 | bool cpuhp_tasks_frozen; |
179 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 180 | |
79a6cdeb | 181 | /* |
93ae4f97 SB |
182 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
183 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
184 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
185 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
186 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
187 | */ |
188 | void cpu_maps_update_begin(void) | |
189 | { | |
190 | mutex_lock(&cpu_add_remove_lock); | |
191 | } | |
93ae4f97 | 192 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
193 | |
194 | void cpu_maps_update_done(void) | |
195 | { | |
196 | mutex_unlock(&cpu_add_remove_lock); | |
197 | } | |
93ae4f97 | 198 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 199 | |
5c113fbe | 200 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 201 | |
e3920fb4 RW |
202 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
203 | * Should always be manipulated under cpu_add_remove_lock | |
204 | */ | |
205 | static int cpu_hotplug_disabled; | |
206 | ||
79a6cdeb LJ |
207 | #ifdef CONFIG_HOTPLUG_CPU |
208 | ||
d221938c GS |
209 | static struct { |
210 | struct task_struct *active_writer; | |
87af9e7f DH |
211 | /* wait queue to wake up the active_writer */ |
212 | wait_queue_head_t wq; | |
213 | /* verifies that no writer will get active while readers are active */ | |
214 | struct mutex lock; | |
d221938c GS |
215 | /* |
216 | * Also blocks the new readers during | |
217 | * an ongoing cpu hotplug operation. | |
218 | */ | |
87af9e7f | 219 | atomic_t refcount; |
a19423b9 GS |
220 | |
221 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
222 | struct lockdep_map dep_map; | |
223 | #endif | |
31950eb6 LT |
224 | } cpu_hotplug = { |
225 | .active_writer = NULL, | |
87af9e7f | 226 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 227 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
228 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
229 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
230 | #endif | |
31950eb6 | 231 | }; |
d221938c | 232 | |
a19423b9 GS |
233 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
234 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
235 | #define cpuhp_lock_acquire_tryread() \ |
236 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
237 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
238 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
239 | ||
62db99f4 | 240 | |
86ef5c9a | 241 | void get_online_cpus(void) |
a9d9baa1 | 242 | { |
d221938c GS |
243 | might_sleep(); |
244 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 245 | return; |
a19423b9 | 246 | cpuhp_lock_acquire_read(); |
d221938c | 247 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 248 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 249 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 250 | } |
86ef5c9a | 251 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 252 | |
86ef5c9a | 253 | void put_online_cpus(void) |
a9d9baa1 | 254 | { |
87af9e7f DH |
255 | int refcount; |
256 | ||
d221938c | 257 | if (cpu_hotplug.active_writer == current) |
aa953877 | 258 | return; |
075663d1 | 259 | |
87af9e7f DH |
260 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
261 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
262 | atomic_inc(&cpu_hotplug.refcount); | |
263 | ||
264 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
265 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 266 | |
a19423b9 | 267 | cpuhp_lock_release(); |
d221938c | 268 | |
a9d9baa1 | 269 | } |
86ef5c9a | 270 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 271 | |
d221938c GS |
272 | /* |
273 | * This ensures that the hotplug operation can begin only when the | |
274 | * refcount goes to zero. | |
275 | * | |
276 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
277 | * will be blocked by the cpu_hotplug.lock | |
278 | * | |
d2ba7e2a ON |
279 | * Since cpu_hotplug_begin() is always called after invoking |
280 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
281 | * |
282 | * Note that theoretically, there is a possibility of a livelock: | |
283 | * - Refcount goes to zero, last reader wakes up the sleeping | |
284 | * writer. | |
285 | * - Last reader unlocks the cpu_hotplug.lock. | |
286 | * - A new reader arrives at this moment, bumps up the refcount. | |
287 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
288 | * non zero and goes to sleep again. | |
289 | * | |
290 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 291 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
292 | * |
293 | */ | |
b9d10be7 | 294 | void cpu_hotplug_begin(void) |
d221938c | 295 | { |
87af9e7f | 296 | DEFINE_WAIT(wait); |
d2ba7e2a | 297 | |
87af9e7f | 298 | cpu_hotplug.active_writer = current; |
a19423b9 | 299 | cpuhp_lock_acquire(); |
87af9e7f | 300 | |
d2ba7e2a ON |
301 | for (;;) { |
302 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
303 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
304 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
305 | break; | |
d221938c GS |
306 | mutex_unlock(&cpu_hotplug.lock); |
307 | schedule(); | |
d221938c | 308 | } |
87af9e7f | 309 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
310 | } |
311 | ||
b9d10be7 | 312 | void cpu_hotplug_done(void) |
d221938c GS |
313 | { |
314 | cpu_hotplug.active_writer = NULL; | |
315 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 316 | cpuhp_lock_release(); |
d221938c | 317 | } |
79a6cdeb | 318 | |
16e53dbf SB |
319 | /* |
320 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
321 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
322 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
323 | * hotplug path before performing hotplug operations. So acquiring that lock | |
324 | * guarantees mutual exclusion from any currently running hotplug operations. | |
325 | */ | |
326 | void cpu_hotplug_disable(void) | |
327 | { | |
328 | cpu_maps_update_begin(); | |
89af7ba5 | 329 | cpu_hotplug_disabled++; |
16e53dbf SB |
330 | cpu_maps_update_done(); |
331 | } | |
32145c46 | 332 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 333 | |
01b41159 LW |
334 | static void __cpu_hotplug_enable(void) |
335 | { | |
336 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
337 | return; | |
338 | cpu_hotplug_disabled--; | |
339 | } | |
340 | ||
16e53dbf SB |
341 | void cpu_hotplug_enable(void) |
342 | { | |
343 | cpu_maps_update_begin(); | |
01b41159 | 344 | __cpu_hotplug_enable(); |
16e53dbf SB |
345 | cpu_maps_update_done(); |
346 | } | |
32145c46 | 347 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 348 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 349 | |
1da177e4 | 350 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 351 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 352 | { |
bd5349cf | 353 | int ret; |
d221938c | 354 | cpu_maps_update_begin(); |
bd5349cf | 355 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 356 | cpu_maps_update_done(); |
bd5349cf | 357 | return ret; |
1da177e4 | 358 | } |
65edc68c | 359 | |
71cf5aee | 360 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
361 | { |
362 | return raw_notifier_chain_register(&cpu_chain, nb); | |
363 | } | |
364 | ||
090e77c3 | 365 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
366 | int *nr_calls) |
367 | { | |
090e77c3 TG |
368 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
369 | void *hcpu = (void *)(long)cpu; | |
370 | ||
e6bde73b AM |
371 | int ret; |
372 | ||
090e77c3 | 373 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 374 | nr_calls); |
e6bde73b AM |
375 | |
376 | return notifier_to_errno(ret); | |
e9fb7631 AM |
377 | } |
378 | ||
090e77c3 | 379 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 380 | { |
090e77c3 | 381 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
382 | } |
383 | ||
3b9d6da6 SAS |
384 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
385 | { | |
386 | BUG_ON(cpu_notify(val, cpu)); | |
387 | } | |
388 | ||
ba997462 TG |
389 | /* Notifier wrappers for transitioning to state machine */ |
390 | static int notify_prepare(unsigned int cpu) | |
391 | { | |
392 | int nr_calls = 0; | |
393 | int ret; | |
394 | ||
395 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
396 | if (ret) { | |
397 | nr_calls--; | |
398 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
399 | __func__, cpu); | |
400 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
401 | } | |
402 | return ret; | |
403 | } | |
404 | ||
405 | static int notify_online(unsigned int cpu) | |
406 | { | |
407 | cpu_notify(CPU_ONLINE, cpu); | |
408 | return 0; | |
409 | } | |
410 | ||
4baa0afc TG |
411 | static int notify_starting(unsigned int cpu) |
412 | { | |
413 | cpu_notify(CPU_STARTING, cpu); | |
414 | return 0; | |
415 | } | |
416 | ||
8df3e07e TG |
417 | static int bringup_wait_for_ap(unsigned int cpu) |
418 | { | |
419 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
420 | ||
421 | wait_for_completion(&st->done); | |
422 | return st->result; | |
423 | } | |
424 | ||
ba997462 TG |
425 | static int bringup_cpu(unsigned int cpu) |
426 | { | |
427 | struct task_struct *idle = idle_thread_get(cpu); | |
428 | int ret; | |
429 | ||
aa877175 BO |
430 | /* |
431 | * Some architectures have to walk the irq descriptors to | |
432 | * setup the vector space for the cpu which comes online. | |
433 | * Prevent irq alloc/free across the bringup. | |
434 | */ | |
435 | irq_lock_sparse(); | |
436 | ||
ba997462 TG |
437 | /* Arch-specific enabling code. */ |
438 | ret = __cpu_up(cpu, idle); | |
aa877175 | 439 | irq_unlock_sparse(); |
ba997462 TG |
440 | if (ret) { |
441 | cpu_notify(CPU_UP_CANCELED, cpu); | |
442 | return ret; | |
443 | } | |
8df3e07e | 444 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 445 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 446 | return ret; |
ba997462 TG |
447 | } |
448 | ||
2e1a3483 TG |
449 | /* |
450 | * Hotplug state machine related functions | |
451 | */ | |
a724632c | 452 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
453 | { |
454 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 455 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
456 | |
457 | if (!step->skip_onerr) | |
cf392d10 | 458 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
459 | } |
460 | } | |
461 | ||
462 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 463 | enum cpuhp_state target) |
2e1a3483 TG |
464 | { |
465 | enum cpuhp_state prev_state = st->state; | |
466 | int ret = 0; | |
467 | ||
468 | for (; st->state > target; st->state--) { | |
cf392d10 | 469 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
470 | if (ret) { |
471 | st->target = prev_state; | |
a724632c | 472 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
473 | break; |
474 | } | |
475 | } | |
476 | return ret; | |
477 | } | |
478 | ||
a724632c | 479 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
480 | { |
481 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 482 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
483 | |
484 | if (!step->skip_onerr) | |
cf392d10 | 485 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
486 | } |
487 | } | |
488 | ||
489 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 490 | enum cpuhp_state target) |
2e1a3483 TG |
491 | { |
492 | enum cpuhp_state prev_state = st->state; | |
493 | int ret = 0; | |
494 | ||
495 | while (st->state < target) { | |
2e1a3483 | 496 | st->state++; |
cf392d10 | 497 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
498 | if (ret) { |
499 | st->target = prev_state; | |
a724632c | 500 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
501 | break; |
502 | } | |
503 | } | |
504 | return ret; | |
505 | } | |
506 | ||
4cb28ced TG |
507 | /* |
508 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
509 | */ | |
510 | static void cpuhp_create(unsigned int cpu) | |
511 | { | |
512 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
513 | ||
514 | init_completion(&st->done); | |
515 | } | |
516 | ||
517 | static int cpuhp_should_run(unsigned int cpu) | |
518 | { | |
519 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
520 | ||
521 | return st->should_run; | |
522 | } | |
523 | ||
524 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
525 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
526 | { | |
1cf4f629 | 527 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 528 | |
a724632c | 529 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
530 | } |
531 | ||
532 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
533 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
534 | { | |
a724632c | 535 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
536 | } |
537 | ||
538 | /* | |
539 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
540 | * callbacks when a state gets [un]installed at runtime. | |
541 | */ | |
542 | static void cpuhp_thread_fun(unsigned int cpu) | |
543 | { | |
544 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
545 | int ret = 0; | |
546 | ||
547 | /* | |
548 | * Paired with the mb() in cpuhp_kick_ap_work and | |
549 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
550 | */ | |
551 | smp_mb(); | |
552 | if (!st->should_run) | |
553 | return; | |
554 | ||
555 | st->should_run = false; | |
556 | ||
557 | /* Single callback invocation for [un]install ? */ | |
a724632c | 558 | if (st->single) { |
4cb28ced TG |
559 | if (st->cb_state < CPUHP_AP_ONLINE) { |
560 | local_irq_disable(); | |
a724632c | 561 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 562 | st->bringup, st->node); |
4cb28ced TG |
563 | local_irq_enable(); |
564 | } else { | |
a724632c | 565 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 566 | st->bringup, st->node); |
4cb28ced | 567 | } |
3b9d6da6 SAS |
568 | } else if (st->rollback) { |
569 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
570 | ||
a724632c | 571 | undo_cpu_down(cpu, st); |
3b9d6da6 SAS |
572 | /* |
573 | * This is a momentary workaround to keep the notifier users | |
574 | * happy. Will go away once we got rid of the notifiers. | |
575 | */ | |
576 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); | |
577 | st->rollback = false; | |
4cb28ced | 578 | } else { |
1cf4f629 | 579 | /* Cannot happen .... */ |
8df3e07e | 580 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 581 | |
4cb28ced TG |
582 | /* Regular hotplug work */ |
583 | if (st->state < st->target) | |
584 | ret = cpuhp_ap_online(cpu, st); | |
585 | else if (st->state > st->target) | |
586 | ret = cpuhp_ap_offline(cpu, st); | |
587 | } | |
588 | st->result = ret; | |
589 | complete(&st->done); | |
590 | } | |
591 | ||
592 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 593 | static int |
cf392d10 TG |
594 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
595 | struct hlist_node *node) | |
4cb28ced TG |
596 | { |
597 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
598 | ||
599 | if (!cpu_online(cpu)) | |
600 | return 0; | |
601 | ||
6a4e2451 TG |
602 | /* |
603 | * If we are up and running, use the hotplug thread. For early calls | |
604 | * we invoke the thread function directly. | |
605 | */ | |
606 | if (!st->thread) | |
cf392d10 | 607 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 608 | |
4cb28ced | 609 | st->cb_state = state; |
a724632c TG |
610 | st->single = true; |
611 | st->bringup = bringup; | |
cf392d10 | 612 | st->node = node; |
a724632c | 613 | |
4cb28ced TG |
614 | /* |
615 | * Make sure the above stores are visible before should_run becomes | |
616 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
617 | */ | |
618 | smp_mb(); | |
619 | st->should_run = true; | |
620 | wake_up_process(st->thread); | |
621 | wait_for_completion(&st->done); | |
622 | return st->result; | |
623 | } | |
624 | ||
625 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 626 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 627 | { |
4cb28ced | 628 | st->result = 0; |
a724632c | 629 | st->single = false; |
4cb28ced TG |
630 | /* |
631 | * Make sure the above stores are visible before should_run becomes | |
632 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
633 | */ | |
634 | smp_mb(); | |
635 | st->should_run = true; | |
636 | wake_up_process(st->thread); | |
1cf4f629 TG |
637 | } |
638 | ||
639 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
640 | { | |
641 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
642 | enum cpuhp_state state = st->state; | |
643 | ||
644 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
645 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
646 | wait_for_completion(&st->done); |
647 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
648 | return st->result; | |
649 | } | |
650 | ||
651 | static struct smp_hotplug_thread cpuhp_threads = { | |
652 | .store = &cpuhp_state.thread, | |
653 | .create = &cpuhp_create, | |
654 | .thread_should_run = cpuhp_should_run, | |
655 | .thread_fn = cpuhp_thread_fun, | |
656 | .thread_comm = "cpuhp/%u", | |
657 | .selfparking = true, | |
658 | }; | |
659 | ||
660 | void __init cpuhp_threads_init(void) | |
661 | { | |
662 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
663 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
664 | } | |
665 | ||
00b9b0af | 666 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 667 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 668 | EXPORT_SYMBOL(__register_cpu_notifier); |
71cf5aee | 669 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 670 | { |
d221938c | 671 | cpu_maps_update_begin(); |
bd5349cf | 672 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 673 | cpu_maps_update_done(); |
1da177e4 LT |
674 | } |
675 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
676 | ||
71cf5aee | 677 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
678 | { |
679 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
680 | } | |
681 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
682 | ||
e4cc2f87 AV |
683 | /** |
684 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
685 | * @cpu: a CPU id | |
686 | * | |
687 | * This function walks all processes, finds a valid mm struct for each one and | |
688 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
689 | * trivial, there are various non-obvious corner cases, which this function | |
690 | * tries to solve in a safe manner. | |
691 | * | |
692 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
693 | * be called only for an already offlined CPU. | |
694 | */ | |
cb79295e AV |
695 | void clear_tasks_mm_cpumask(int cpu) |
696 | { | |
697 | struct task_struct *p; | |
698 | ||
699 | /* | |
700 | * This function is called after the cpu is taken down and marked | |
701 | * offline, so its not like new tasks will ever get this cpu set in | |
702 | * their mm mask. -- Peter Zijlstra | |
703 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
704 | * full-fledged tasklist_lock. | |
705 | */ | |
e4cc2f87 | 706 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
707 | rcu_read_lock(); |
708 | for_each_process(p) { | |
709 | struct task_struct *t; | |
710 | ||
e4cc2f87 AV |
711 | /* |
712 | * Main thread might exit, but other threads may still have | |
713 | * a valid mm. Find one. | |
714 | */ | |
cb79295e AV |
715 | t = find_lock_task_mm(p); |
716 | if (!t) | |
717 | continue; | |
718 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
719 | task_unlock(t); | |
720 | } | |
721 | rcu_read_unlock(); | |
722 | } | |
723 | ||
b728ca06 | 724 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 725 | { |
b728ca06 | 726 | struct task_struct *g, *p; |
1da177e4 | 727 | |
a75a6068 ON |
728 | read_lock(&tasklist_lock); |
729 | for_each_process_thread(g, p) { | |
b728ca06 KT |
730 | if (!p->on_rq) |
731 | continue; | |
732 | /* | |
733 | * We do the check with unlocked task_rq(p)->lock. | |
734 | * Order the reading to do not warn about a task, | |
735 | * which was running on this cpu in the past, and | |
736 | * it's just been woken on another cpu. | |
737 | */ | |
738 | rmb(); | |
739 | if (task_cpu(p) != dead_cpu) | |
740 | continue; | |
741 | ||
742 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
743 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
744 | } |
745 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
746 | } |
747 | ||
98458172 TG |
748 | static int notify_down_prepare(unsigned int cpu) |
749 | { | |
750 | int err, nr_calls = 0; | |
751 | ||
752 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
753 | if (err) { | |
754 | nr_calls--; | |
755 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
756 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
757 | __func__, cpu); | |
758 | } | |
759 | return err; | |
760 | } | |
761 | ||
4baa0afc TG |
762 | static int notify_dying(unsigned int cpu) |
763 | { | |
764 | cpu_notify(CPU_DYING, cpu); | |
765 | return 0; | |
766 | } | |
767 | ||
1da177e4 | 768 | /* Take this CPU down. */ |
71cf5aee | 769 | static int take_cpu_down(void *_param) |
1da177e4 | 770 | { |
4baa0afc TG |
771 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
772 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 773 | int err, cpu = smp_processor_id(); |
1da177e4 | 774 | |
1da177e4 LT |
775 | /* Ensure this CPU doesn't handle any more interrupts. */ |
776 | err = __cpu_disable(); | |
777 | if (err < 0) | |
f3705136 | 778 | return err; |
1da177e4 | 779 | |
a724632c TG |
780 | /* |
781 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
782 | * do this step again. | |
783 | */ | |
784 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
785 | st->state--; | |
4baa0afc | 786 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 787 | for (; st->state > target; st->state--) |
cf392d10 | 788 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 789 | |
52c063d1 TG |
790 | /* Give up timekeeping duties */ |
791 | tick_handover_do_timer(); | |
14e568e7 | 792 | /* Park the stopper thread */ |
090e77c3 | 793 | stop_machine_park(cpu); |
f3705136 | 794 | return 0; |
1da177e4 LT |
795 | } |
796 | ||
98458172 | 797 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 798 | { |
e69aab13 | 799 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 800 | int err; |
1da177e4 | 801 | |
2a58c527 | 802 | /* Park the smpboot threads */ |
1cf4f629 | 803 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 804 | smpboot_park_threads(cpu); |
1cf4f629 | 805 | |
6acce3ef | 806 | /* |
a8994181 TG |
807 | * Prevent irq alloc/free while the dying cpu reorganizes the |
808 | * interrupt affinities. | |
6acce3ef | 809 | */ |
a8994181 | 810 | irq_lock_sparse(); |
6acce3ef | 811 | |
a8994181 TG |
812 | /* |
813 | * So now all preempt/rcu users must observe !cpu_active(). | |
814 | */ | |
090e77c3 | 815 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 816 | if (err) { |
3b9d6da6 | 817 | /* CPU refused to die */ |
a8994181 | 818 | irq_unlock_sparse(); |
3b9d6da6 SAS |
819 | /* Unpark the hotplug thread so we can rollback there */ |
820 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 821 | return err; |
8fa1d7d3 | 822 | } |
04321587 | 823 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 824 | |
48c5ccae PZ |
825 | /* |
826 | * The migration_call() CPU_DYING callback will have removed all | |
827 | * runnable tasks from the cpu, there's only the idle task left now | |
828 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
829 | * |
830 | * Wait for the stop thread to go away. | |
48c5ccae | 831 | */ |
e69aab13 TG |
832 | wait_for_completion(&st->done); |
833 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 834 | |
a8994181 TG |
835 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
836 | irq_unlock_sparse(); | |
837 | ||
345527b1 | 838 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
839 | /* This actually kills the CPU. */ |
840 | __cpu_die(cpu); | |
841 | ||
a49b116d | 842 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
843 | return 0; |
844 | } | |
1da177e4 | 845 | |
98458172 TG |
846 | static int notify_dead(unsigned int cpu) |
847 | { | |
848 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 849 | check_for_tasks(cpu); |
98458172 TG |
850 | return 0; |
851 | } | |
852 | ||
71f87b2f TG |
853 | static void cpuhp_complete_idle_dead(void *arg) |
854 | { | |
855 | struct cpuhp_cpu_state *st = arg; | |
856 | ||
857 | complete(&st->done); | |
858 | } | |
859 | ||
e69aab13 TG |
860 | void cpuhp_report_idle_dead(void) |
861 | { | |
862 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
863 | ||
864 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 865 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
866 | st->state = CPUHP_AP_IDLE_DEAD; |
867 | /* | |
868 | * We cannot call complete after rcu_report_dead() so we delegate it | |
869 | * to an online cpu. | |
870 | */ | |
871 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
872 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
873 | } |
874 | ||
cff7d378 TG |
875 | #else |
876 | #define notify_down_prepare NULL | |
877 | #define takedown_cpu NULL | |
878 | #define notify_dead NULL | |
4baa0afc | 879 | #define notify_dying NULL |
cff7d378 TG |
880 | #endif |
881 | ||
882 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 883 | |
98458172 | 884 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
885 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
886 | enum cpuhp_state target) | |
98458172 | 887 | { |
cff7d378 TG |
888 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
889 | int prev_state, ret = 0; | |
890 | bool hasdied = false; | |
98458172 TG |
891 | |
892 | if (num_online_cpus() == 1) | |
893 | return -EBUSY; | |
894 | ||
757c989b | 895 | if (!cpu_present(cpu)) |
98458172 TG |
896 | return -EINVAL; |
897 | ||
898 | cpu_hotplug_begin(); | |
899 | ||
900 | cpuhp_tasks_frozen = tasks_frozen; | |
901 | ||
cff7d378 | 902 | prev_state = st->state; |
af1f4045 | 903 | st->target = target; |
1cf4f629 TG |
904 | /* |
905 | * If the current CPU state is in the range of the AP hotplug thread, | |
906 | * then we need to kick the thread. | |
907 | */ | |
8df3e07e | 908 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
909 | ret = cpuhp_kick_ap_work(cpu); |
910 | /* | |
911 | * The AP side has done the error rollback already. Just | |
912 | * return the error code.. | |
913 | */ | |
914 | if (ret) | |
915 | goto out; | |
916 | ||
917 | /* | |
918 | * We might have stopped still in the range of the AP hotplug | |
919 | * thread. Nothing to do anymore. | |
920 | */ | |
8df3e07e | 921 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
922 | goto out; |
923 | } | |
924 | /* | |
8df3e07e | 925 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
926 | * to do the further cleanups. |
927 | */ | |
a724632c | 928 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
929 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
930 | st->target = prev_state; | |
931 | st->rollback = true; | |
932 | cpuhp_kick_ap_work(cpu); | |
933 | } | |
98458172 | 934 | |
cff7d378 | 935 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; |
1cf4f629 | 936 | out: |
d221938c | 937 | cpu_hotplug_done(); |
cff7d378 TG |
938 | /* This post dead nonsense must die */ |
939 | if (!ret && hasdied) | |
090e77c3 | 940 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 941 | return ret; |
e3920fb4 RW |
942 | } |
943 | ||
af1f4045 | 944 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 945 | { |
9ea09af3 | 946 | int err; |
e3920fb4 | 947 | |
d221938c | 948 | cpu_maps_update_begin(); |
e761b772 MK |
949 | |
950 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 951 | err = -EBUSY; |
e761b772 MK |
952 | goto out; |
953 | } | |
954 | ||
af1f4045 | 955 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 956 | |
e761b772 | 957 | out: |
d221938c | 958 | cpu_maps_update_done(); |
1da177e4 LT |
959 | return err; |
960 | } | |
af1f4045 TG |
961 | int cpu_down(unsigned int cpu) |
962 | { | |
963 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
964 | } | |
b62b8ef9 | 965 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
966 | #endif /*CONFIG_HOTPLUG_CPU*/ |
967 | ||
4baa0afc TG |
968 | /** |
969 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
970 | * @cpu: cpu that just started | |
971 | * | |
972 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
973 | * It must be called by the arch code on the new cpu, before the new cpu | |
974 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
975 | */ | |
976 | void notify_cpu_starting(unsigned int cpu) | |
977 | { | |
978 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
979 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
980 | ||
981 | while (st->state < target) { | |
4baa0afc | 982 | st->state++; |
cf392d10 | 983 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
984 | } |
985 | } | |
986 | ||
949338e3 TG |
987 | /* |
988 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
989 | * the stopper thread and unpark the smpboot threads. If the target state is |
990 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
991 | * cpu further. | |
949338e3 | 992 | */ |
8df3e07e | 993 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 994 | { |
8df3e07e TG |
995 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
996 | unsigned int cpu = smp_processor_id(); | |
997 | ||
998 | /* Happens for the boot cpu */ | |
999 | if (state != CPUHP_AP_ONLINE_IDLE) | |
1000 | return; | |
1001 | ||
1002 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 1003 | |
8df3e07e | 1004 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 1005 | stop_machine_unpark(cpu); |
1cf4f629 | 1006 | kthread_unpark(st->thread); |
8df3e07e TG |
1007 | |
1008 | /* Should we go further up ? */ | |
1009 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
1010 | __cpuhp_kick_ap_work(st); | |
1011 | else | |
1012 | complete(&st->done); | |
949338e3 TG |
1013 | } |
1014 | ||
e3920fb4 | 1015 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1016 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1017 | { |
cff7d378 | 1018 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1019 | struct task_struct *idle; |
2e1a3483 | 1020 | int ret = 0; |
1da177e4 | 1021 | |
d221938c | 1022 | cpu_hotplug_begin(); |
38498a67 | 1023 | |
757c989b | 1024 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1025 | ret = -EINVAL; |
1026 | goto out; | |
1027 | } | |
1028 | ||
757c989b TG |
1029 | /* |
1030 | * The caller of do_cpu_up might have raced with another | |
1031 | * caller. Ignore it for now. | |
1032 | */ | |
1033 | if (st->state >= target) | |
38498a67 | 1034 | goto out; |
757c989b TG |
1035 | |
1036 | if (st->state == CPUHP_OFFLINE) { | |
1037 | /* Let it fail before we try to bring the cpu up */ | |
1038 | idle = idle_thread_get(cpu); | |
1039 | if (IS_ERR(idle)) { | |
1040 | ret = PTR_ERR(idle); | |
1041 | goto out; | |
1042 | } | |
3bb5d2ee | 1043 | } |
38498a67 | 1044 | |
ba997462 TG |
1045 | cpuhp_tasks_frozen = tasks_frozen; |
1046 | ||
af1f4045 | 1047 | st->target = target; |
1cf4f629 TG |
1048 | /* |
1049 | * If the current CPU state is in the range of the AP hotplug thread, | |
1050 | * then we need to kick the thread once more. | |
1051 | */ | |
8df3e07e | 1052 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1053 | ret = cpuhp_kick_ap_work(cpu); |
1054 | /* | |
1055 | * The AP side has done the error rollback already. Just | |
1056 | * return the error code.. | |
1057 | */ | |
1058 | if (ret) | |
1059 | goto out; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1064 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1065 | * responsible for bringing it up to the target state. |
1066 | */ | |
8df3e07e | 1067 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1068 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1069 | out: |
d221938c | 1070 | cpu_hotplug_done(); |
e3920fb4 RW |
1071 | return ret; |
1072 | } | |
1073 | ||
af1f4045 | 1074 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1075 | { |
1076 | int err = 0; | |
cf23422b | 1077 | |
e0b582ec | 1078 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1079 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1080 | cpu); | |
87d5e023 | 1081 | #if defined(CONFIG_IA64) |
84117da5 | 1082 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1083 | #endif |
1084 | return -EINVAL; | |
1085 | } | |
e3920fb4 | 1086 | |
01b0f197 TK |
1087 | err = try_online_node(cpu_to_node(cpu)); |
1088 | if (err) | |
1089 | return err; | |
cf23422b | 1090 | |
d221938c | 1091 | cpu_maps_update_begin(); |
e761b772 MK |
1092 | |
1093 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1094 | err = -EBUSY; |
e761b772 MK |
1095 | goto out; |
1096 | } | |
1097 | ||
af1f4045 | 1098 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1099 | out: |
d221938c | 1100 | cpu_maps_update_done(); |
e3920fb4 RW |
1101 | return err; |
1102 | } | |
af1f4045 TG |
1103 | |
1104 | int cpu_up(unsigned int cpu) | |
1105 | { | |
1106 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1107 | } | |
a513f6ba | 1108 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1109 | |
f3de4be9 | 1110 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1111 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
1112 | |
1113 | int disable_nonboot_cpus(void) | |
1114 | { | |
e9a5f426 | 1115 | int cpu, first_cpu, error = 0; |
e3920fb4 | 1116 | |
d221938c | 1117 | cpu_maps_update_begin(); |
e0b582ec | 1118 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
1119 | /* |
1120 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1121 | * with the userspace trying to use the CPU hotplug at the same time |
1122 | */ | |
e0b582ec | 1123 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1124 | |
84117da5 | 1125 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
1126 | for_each_online_cpu(cpu) { |
1127 | if (cpu == first_cpu) | |
1128 | continue; | |
bb3632c6 | 1129 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1130 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1131 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1132 | if (!error) |
e0b582ec | 1133 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1134 | else { |
84117da5 | 1135 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1136 | break; |
1137 | } | |
1138 | } | |
86886e55 | 1139 | |
89af7ba5 | 1140 | if (!error) |
e3920fb4 | 1141 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1142 | else |
84117da5 | 1143 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1144 | |
1145 | /* | |
1146 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1147 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1148 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1149 | */ | |
1150 | cpu_hotplug_disabled++; | |
1151 | ||
d221938c | 1152 | cpu_maps_update_done(); |
e3920fb4 RW |
1153 | return error; |
1154 | } | |
1155 | ||
d0af9eed SS |
1156 | void __weak arch_enable_nonboot_cpus_begin(void) |
1157 | { | |
1158 | } | |
1159 | ||
1160 | void __weak arch_enable_nonboot_cpus_end(void) | |
1161 | { | |
1162 | } | |
1163 | ||
71cf5aee | 1164 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1165 | { |
1166 | int cpu, error; | |
1167 | ||
1168 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1169 | cpu_maps_update_begin(); |
01b41159 | 1170 | __cpu_hotplug_enable(); |
e0b582ec | 1171 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1172 | goto out; |
e3920fb4 | 1173 | |
84117da5 | 1174 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1175 | |
1176 | arch_enable_nonboot_cpus_begin(); | |
1177 | ||
e0b582ec | 1178 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1179 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1180 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1181 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1182 | if (!error) { |
84117da5 | 1183 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1184 | continue; |
1185 | } | |
84117da5 | 1186 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1187 | } |
d0af9eed SS |
1188 | |
1189 | arch_enable_nonboot_cpus_end(); | |
1190 | ||
e0b582ec | 1191 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1192 | out: |
d221938c | 1193 | cpu_maps_update_done(); |
1da177e4 | 1194 | } |
e0b582ec | 1195 | |
d7268a31 | 1196 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1197 | { |
1198 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1199 | return -ENOMEM; | |
1200 | return 0; | |
1201 | } | |
1202 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1203 | |
79cfbdfa SB |
1204 | /* |
1205 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1206 | * ensure that the state of the system with respect to the tasks being frozen | |
1207 | * or not, as reported by the notification, remains unchanged *throughout the | |
1208 | * duration* of the execution of the callbacks. | |
1209 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1210 | * | |
1211 | * This synchronization is implemented by mutually excluding regular CPU | |
1212 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1213 | * Hibernate notifications. | |
1214 | */ | |
1215 | static int | |
1216 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1217 | unsigned long action, void *ptr) | |
1218 | { | |
1219 | switch (action) { | |
1220 | ||
1221 | case PM_SUSPEND_PREPARE: | |
1222 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1223 | cpu_hotplug_disable(); |
79cfbdfa SB |
1224 | break; |
1225 | ||
1226 | case PM_POST_SUSPEND: | |
1227 | case PM_POST_HIBERNATION: | |
16e53dbf | 1228 | cpu_hotplug_enable(); |
79cfbdfa SB |
1229 | break; |
1230 | ||
1231 | default: | |
1232 | return NOTIFY_DONE; | |
1233 | } | |
1234 | ||
1235 | return NOTIFY_OK; | |
1236 | } | |
1237 | ||
1238 | ||
d7268a31 | 1239 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1240 | { |
6e32d479 FY |
1241 | /* |
1242 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1243 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1244 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1245 | */ | |
79cfbdfa SB |
1246 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1247 | return 0; | |
1248 | } | |
1249 | core_initcall(cpu_hotplug_pm_sync_init); | |
1250 | ||
f3de4be9 | 1251 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
1252 | |
1253 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 1254 | |
cff7d378 TG |
1255 | /* Boot processor state steps */ |
1256 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1257 | [CPUHP_OFFLINE] = { | |
1258 | .name = "offline", | |
3c1627e9 TG |
1259 | .startup.single = NULL, |
1260 | .teardown.single = NULL, | |
cff7d378 TG |
1261 | }, |
1262 | #ifdef CONFIG_SMP | |
1263 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1264 | .name = "threads:prepare", |
3c1627e9 TG |
1265 | .startup.single = smpboot_create_threads, |
1266 | .teardown.single = NULL, | |
757c989b | 1267 | .cant_stop = true, |
cff7d378 | 1268 | }, |
00e16c3d | 1269 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1270 | .name = "perf:prepare", |
1271 | .startup.single = perf_event_init_cpu, | |
1272 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1273 | }, |
7ee681b2 | 1274 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1275 | .name = "workqueue:prepare", |
1276 | .startup.single = workqueue_prepare_cpu, | |
1277 | .teardown.single = NULL, | |
7ee681b2 | 1278 | }, |
27590dc1 | 1279 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1280 | .name = "hrtimers:prepare", |
1281 | .startup.single = hrtimers_prepare_cpu, | |
1282 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1283 | }, |
31487f83 | 1284 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1285 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1286 | .startup.single = smpcfd_prepare_cpu, |
1287 | .teardown.single = smpcfd_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, | |
4df83742 | 1367 | }, |
d10ef6f9 | 1368 | /* |
3c1627e9 | 1369 | * Low level startup.single/teardown notifiers. Run with interrupts |
d10ef6f9 TG |
1370 | * disabled. Will be removed once the notifiers are converted to |
1371 | * states. | |
1372 | */ | |
4baa0afc TG |
1373 | [CPUHP_AP_NOTIFY_STARTING] = { |
1374 | .name = "notify:starting", | |
3c1627e9 TG |
1375 | .startup.single = notify_starting, |
1376 | .teardown.single = notify_dying, | |
4baa0afc | 1377 | .skip_onerr = true, |
757c989b | 1378 | .cant_stop = true, |
4baa0afc | 1379 | }, |
d10ef6f9 TG |
1380 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1381 | * state for synchronsization */ | |
1382 | [CPUHP_AP_ONLINE] = { | |
1383 | .name = "ap:online", | |
1384 | }, | |
1385 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1386 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1387 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1388 | .startup.single = smpboot_unpark_threads, |
1389 | .teardown.single = NULL, | |
1cf4f629 | 1390 | }, |
00e16c3d | 1391 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1392 | .name = "perf:online", |
1393 | .startup.single = perf_event_init_cpu, | |
1394 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1395 | }, |
7ee681b2 | 1396 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1397 | .name = "workqueue:online", |
1398 | .startup.single = workqueue_online_cpu, | |
1399 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1400 | }, |
4df83742 | 1401 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1402 | .name = "RCU/tree:online", |
3c1627e9 TG |
1403 | .startup.single = rcutree_online_cpu, |
1404 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1405 | }, |
00e16c3d | 1406 | |
d10ef6f9 TG |
1407 | /* |
1408 | * Online/down_prepare notifiers. Will be removed once the notifiers | |
1409 | * are converted to states. | |
1410 | */ | |
1cf4f629 TG |
1411 | [CPUHP_AP_NOTIFY_ONLINE] = { |
1412 | .name = "notify:online", | |
3c1627e9 TG |
1413 | .startup.single = notify_online, |
1414 | .teardown.single = notify_down_prepare, | |
3b9d6da6 | 1415 | .skip_onerr = true, |
1cf4f629 | 1416 | }, |
4baa0afc | 1417 | #endif |
d10ef6f9 TG |
1418 | /* |
1419 | * The dynamically registered state space is here | |
1420 | */ | |
1421 | ||
aaddd7d1 TG |
1422 | #ifdef CONFIG_SMP |
1423 | /* Last state is scheduler control setting the cpu active */ | |
1424 | [CPUHP_AP_ACTIVE] = { | |
1425 | .name = "sched:active", | |
3c1627e9 TG |
1426 | .startup.single = sched_cpu_activate, |
1427 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1428 | }, |
1429 | #endif | |
1430 | ||
d10ef6f9 | 1431 | /* CPU is fully up and running. */ |
4baa0afc TG |
1432 | [CPUHP_ONLINE] = { |
1433 | .name = "online", | |
3c1627e9 TG |
1434 | .startup.single = NULL, |
1435 | .teardown.single = NULL, | |
4baa0afc TG |
1436 | }, |
1437 | }; | |
1438 | ||
5b7aa87e TG |
1439 | /* Sanity check for callbacks */ |
1440 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1441 | { | |
1442 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1443 | return -EINVAL; | |
1444 | return 0; | |
1445 | } | |
1446 | ||
5b7aa87e TG |
1447 | static void cpuhp_store_callbacks(enum cpuhp_state state, |
1448 | const char *name, | |
1449 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1450 | int (*teardown)(unsigned int cpu), |
1451 | bool multi_instance) | |
5b7aa87e TG |
1452 | { |
1453 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1454 | struct cpuhp_step *sp; | |
1455 | ||
1456 | mutex_lock(&cpuhp_state_mutex); | |
1457 | sp = cpuhp_get_step(state); | |
3c1627e9 TG |
1458 | sp->startup.single = startup; |
1459 | sp->teardown.single = teardown; | |
5b7aa87e | 1460 | sp->name = name; |
cf392d10 TG |
1461 | sp->multi_instance = multi_instance; |
1462 | INIT_HLIST_HEAD(&sp->list); | |
5b7aa87e TG |
1463 | mutex_unlock(&cpuhp_state_mutex); |
1464 | } | |
1465 | ||
1466 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1467 | { | |
3c1627e9 | 1468 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1469 | } |
1470 | ||
5b7aa87e TG |
1471 | /* |
1472 | * Call the startup/teardown function for a step either on the AP or | |
1473 | * on the current CPU. | |
1474 | */ | |
cf392d10 TG |
1475 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1476 | struct hlist_node *node) | |
5b7aa87e | 1477 | { |
a724632c | 1478 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1479 | int ret; |
1480 | ||
3c1627e9 TG |
1481 | if ((bringup && !sp->startup.single) || |
1482 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1483 | return 0; |
5b7aa87e TG |
1484 | /* |
1485 | * The non AP bound callbacks can fail on bringup. On teardown | |
1486 | * e.g. module removal we crash for now. | |
1487 | */ | |
1cf4f629 TG |
1488 | #ifdef CONFIG_SMP |
1489 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1490 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1491 | else |
cf392d10 | 1492 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1493 | #else |
cf392d10 | 1494 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1495 | #endif |
5b7aa87e TG |
1496 | BUG_ON(ret && !bringup); |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1502 | * | |
1503 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1504 | */ | |
cf392d10 TG |
1505 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, |
1506 | struct hlist_node *node) | |
5b7aa87e TG |
1507 | { |
1508 | int cpu; | |
1509 | ||
5b7aa87e TG |
1510 | /* Roll back the already executed steps on the other cpus */ |
1511 | for_each_present_cpu(cpu) { | |
1512 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1513 | int cpustate = st->state; | |
1514 | ||
1515 | if (cpu >= failedcpu) | |
1516 | break; | |
1517 | ||
1518 | /* Did we invoke the startup call on that cpu ? */ | |
1519 | if (cpustate >= state) | |
cf392d10 | 1520 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1521 | } |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1526 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1527 | * by having no name assigned. | |
1528 | */ | |
1529 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1530 | { | |
1531 | enum cpuhp_state i; | |
1532 | ||
1533 | mutex_lock(&cpuhp_state_mutex); | |
1cf4f629 TG |
1534 | for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) { |
1535 | if (cpuhp_ap_states[i].name) | |
5b7aa87e TG |
1536 | continue; |
1537 | ||
1cf4f629 | 1538 | cpuhp_ap_states[i].name = "Reserved"; |
5b7aa87e TG |
1539 | mutex_unlock(&cpuhp_state_mutex); |
1540 | return i; | |
1541 | } | |
1542 | mutex_unlock(&cpuhp_state_mutex); | |
1543 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1544 | return -ENOSPC; | |
1545 | } | |
1546 | ||
cf392d10 TG |
1547 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, |
1548 | bool invoke) | |
1549 | { | |
1550 | struct cpuhp_step *sp; | |
1551 | int cpu; | |
1552 | int ret; | |
1553 | ||
1554 | sp = cpuhp_get_step(state); | |
1555 | if (sp->multi_instance == false) | |
1556 | return -EINVAL; | |
1557 | ||
1558 | get_online_cpus(); | |
1559 | ||
3c1627e9 | 1560 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1561 | goto add_node; |
1562 | ||
1563 | /* | |
1564 | * Try to call the startup callback for each present cpu | |
1565 | * depending on the hotplug state of the cpu. | |
1566 | */ | |
1567 | for_each_present_cpu(cpu) { | |
1568 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1569 | int cpustate = st->state; | |
1570 | ||
1571 | if (cpustate < state) | |
1572 | continue; | |
1573 | ||
1574 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1575 | if (ret) { | |
3c1627e9 | 1576 | if (sp->teardown.multi) |
cf392d10 TG |
1577 | cpuhp_rollback_install(cpu, state, node); |
1578 | goto err; | |
1579 | } | |
1580 | } | |
1581 | add_node: | |
1582 | ret = 0; | |
1583 | mutex_lock(&cpuhp_state_mutex); | |
1584 | hlist_add_head(node, &sp->list); | |
1585 | mutex_unlock(&cpuhp_state_mutex); | |
1586 | ||
1587 | err: | |
1588 | put_online_cpus(); | |
1589 | return ret; | |
1590 | } | |
1591 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1592 | ||
5b7aa87e TG |
1593 | /** |
1594 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
1595 | * @state: The state to setup | |
1596 | * @invoke: If true, the startup function is invoked for cpus where | |
1597 | * cpu state >= @state | |
1598 | * @startup: startup callback function | |
1599 | * @teardown: teardown callback function | |
1600 | * | |
1601 | * Returns 0 if successful, otherwise a proper error code | |
1602 | */ | |
1603 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1604 | const char *name, bool invoke, | |
1605 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1606 | int (*teardown)(unsigned int cpu), |
1607 | bool multi_instance) | |
5b7aa87e TG |
1608 | { |
1609 | int cpu, ret = 0; | |
1610 | int dyn_state = 0; | |
1611 | ||
1612 | if (cpuhp_cb_check(state) || !name) | |
1613 | return -EINVAL; | |
1614 | ||
1615 | get_online_cpus(); | |
1616 | ||
1617 | /* currently assignments for the ONLINE state are possible */ | |
1cf4f629 | 1618 | if (state == CPUHP_AP_ONLINE_DYN) { |
5b7aa87e TG |
1619 | dyn_state = 1; |
1620 | ret = cpuhp_reserve_state(state); | |
1621 | if (ret < 0) | |
1622 | goto out; | |
1623 | state = ret; | |
1624 | } | |
1625 | ||
cf392d10 | 1626 | cpuhp_store_callbacks(state, name, startup, teardown, multi_instance); |
5b7aa87e TG |
1627 | |
1628 | if (!invoke || !startup) | |
1629 | goto out; | |
1630 | ||
1631 | /* | |
1632 | * Try to call the startup callback for each present cpu | |
1633 | * depending on the hotplug state of the cpu. | |
1634 | */ | |
1635 | for_each_present_cpu(cpu) { | |
1636 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1637 | int cpustate = st->state; | |
1638 | ||
1639 | if (cpustate < state) | |
1640 | continue; | |
1641 | ||
cf392d10 | 1642 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1643 | if (ret) { |
a724632c | 1644 | if (teardown) |
cf392d10 TG |
1645 | cpuhp_rollback_install(cpu, state, NULL); |
1646 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1647 | goto out; |
1648 | } | |
1649 | } | |
1650 | out: | |
1651 | put_online_cpus(); | |
1652 | if (!ret && dyn_state) | |
1653 | return state; | |
1654 | return ret; | |
1655 | } | |
1656 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1657 | ||
cf392d10 TG |
1658 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1659 | struct hlist_node *node, bool invoke) | |
1660 | { | |
1661 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1662 | int cpu; | |
1663 | ||
1664 | BUG_ON(cpuhp_cb_check(state)); | |
1665 | ||
1666 | if (!sp->multi_instance) | |
1667 | return -EINVAL; | |
1668 | ||
1669 | get_online_cpus(); | |
1670 | if (!invoke || !cpuhp_get_teardown_cb(state)) | |
1671 | goto remove; | |
1672 | /* | |
1673 | * Call the teardown callback for each present cpu depending | |
1674 | * on the hotplug state of the cpu. This function is not | |
1675 | * allowed to fail currently! | |
1676 | */ | |
1677 | for_each_present_cpu(cpu) { | |
1678 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1679 | int cpustate = st->state; | |
1680 | ||
1681 | if (cpustate >= state) | |
1682 | cpuhp_issue_call(cpu, state, false, node); | |
1683 | } | |
1684 | ||
1685 | remove: | |
1686 | mutex_lock(&cpuhp_state_mutex); | |
1687 | hlist_del(node); | |
1688 | mutex_unlock(&cpuhp_state_mutex); | |
1689 | put_online_cpus(); | |
1690 | ||
1691 | return 0; | |
1692 | } | |
1693 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
5b7aa87e TG |
1694 | /** |
1695 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1696 | * @state: The state to remove | |
1697 | * @invoke: If true, the teardown function is invoked for cpus where | |
1698 | * cpu state >= @state | |
1699 | * | |
1700 | * The teardown callback is currently not allowed to fail. Think | |
1701 | * about module removal! | |
1702 | */ | |
1703 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1704 | { | |
cf392d10 | 1705 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1706 | int cpu; |
1707 | ||
1708 | BUG_ON(cpuhp_cb_check(state)); | |
1709 | ||
1710 | get_online_cpus(); | |
1711 | ||
cf392d10 TG |
1712 | if (sp->multi_instance) { |
1713 | WARN(!hlist_empty(&sp->list), | |
1714 | "Error: Removing state %d which has instances left.\n", | |
1715 | state); | |
1716 | goto remove; | |
1717 | } | |
1718 | ||
a724632c | 1719 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1720 | goto remove; |
1721 | ||
1722 | /* | |
1723 | * Call the teardown callback for each present cpu depending | |
1724 | * on the hotplug state of the cpu. This function is not | |
1725 | * allowed to fail currently! | |
1726 | */ | |
1727 | for_each_present_cpu(cpu) { | |
1728 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1729 | int cpustate = st->state; | |
1730 | ||
1731 | if (cpustate >= state) | |
cf392d10 | 1732 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1733 | } |
1734 | remove: | |
cf392d10 | 1735 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
5b7aa87e TG |
1736 | put_online_cpus(); |
1737 | } | |
1738 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1739 | ||
98f8cdce TG |
1740 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1741 | static ssize_t show_cpuhp_state(struct device *dev, | |
1742 | struct device_attribute *attr, char *buf) | |
1743 | { | |
1744 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1745 | ||
1746 | return sprintf(buf, "%d\n", st->state); | |
1747 | } | |
1748 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1749 | ||
757c989b TG |
1750 | static ssize_t write_cpuhp_target(struct device *dev, |
1751 | struct device_attribute *attr, | |
1752 | const char *buf, size_t count) | |
1753 | { | |
1754 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1755 | struct cpuhp_step *sp; | |
1756 | int target, ret; | |
1757 | ||
1758 | ret = kstrtoint(buf, 10, &target); | |
1759 | if (ret) | |
1760 | return ret; | |
1761 | ||
1762 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1763 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1764 | return -EINVAL; | |
1765 | #else | |
1766 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1767 | return -EINVAL; | |
1768 | #endif | |
1769 | ||
1770 | ret = lock_device_hotplug_sysfs(); | |
1771 | if (ret) | |
1772 | return ret; | |
1773 | ||
1774 | mutex_lock(&cpuhp_state_mutex); | |
1775 | sp = cpuhp_get_step(target); | |
1776 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1777 | mutex_unlock(&cpuhp_state_mutex); | |
1778 | if (ret) | |
1779 | return ret; | |
1780 | ||
1781 | if (st->state < target) | |
1782 | ret = do_cpu_up(dev->id, target); | |
1783 | else | |
1784 | ret = do_cpu_down(dev->id, target); | |
1785 | ||
1786 | unlock_device_hotplug(); | |
1787 | return ret ? ret : count; | |
1788 | } | |
1789 | ||
98f8cdce TG |
1790 | static ssize_t show_cpuhp_target(struct device *dev, |
1791 | struct device_attribute *attr, char *buf) | |
1792 | { | |
1793 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1794 | ||
1795 | return sprintf(buf, "%d\n", st->target); | |
1796 | } | |
757c989b | 1797 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1798 | |
1799 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1800 | &dev_attr_state.attr, | |
1801 | &dev_attr_target.attr, | |
1802 | NULL | |
1803 | }; | |
1804 | ||
1805 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1806 | .attrs = cpuhp_cpu_attrs, | |
1807 | .name = "hotplug", | |
1808 | NULL | |
1809 | }; | |
1810 | ||
1811 | static ssize_t show_cpuhp_states(struct device *dev, | |
1812 | struct device_attribute *attr, char *buf) | |
1813 | { | |
1814 | ssize_t cur, res = 0; | |
1815 | int i; | |
1816 | ||
1817 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1818 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1819 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1820 | ||
1821 | if (sp->name) { | |
1822 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1823 | buf += cur; | |
1824 | res += cur; | |
1825 | } | |
1826 | } | |
1827 | mutex_unlock(&cpuhp_state_mutex); | |
1828 | return res; | |
1829 | } | |
1830 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1831 | ||
1832 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1833 | &dev_attr_states.attr, | |
1834 | NULL | |
1835 | }; | |
1836 | ||
1837 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1838 | .attrs = cpuhp_cpu_root_attrs, | |
1839 | .name = "hotplug", | |
1840 | NULL | |
1841 | }; | |
1842 | ||
1843 | static int __init cpuhp_sysfs_init(void) | |
1844 | { | |
1845 | int cpu, ret; | |
1846 | ||
1847 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1848 | &cpuhp_cpu_root_attr_group); | |
1849 | if (ret) | |
1850 | return ret; | |
1851 | ||
1852 | for_each_possible_cpu(cpu) { | |
1853 | struct device *dev = get_cpu_device(cpu); | |
1854 | ||
1855 | if (!dev) | |
1856 | continue; | |
1857 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1858 | if (ret) | |
1859 | return ret; | |
1860 | } | |
1861 | return 0; | |
1862 | } | |
1863 | device_initcall(cpuhp_sysfs_init); | |
1864 | #endif | |
1865 | ||
e56b3bc7 LT |
1866 | /* |
1867 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1868 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1869 | * | |
e0b582ec | 1870 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1871 | * mask value that has a single bit set only. |
1872 | */ | |
b8d317d1 | 1873 | |
e56b3bc7 | 1874 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1875 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1876 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1877 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1878 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1879 | |
e56b3bc7 LT |
1880 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1881 | ||
1882 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1883 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1884 | #if BITS_PER_LONG > 32 | |
1885 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1886 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1887 | #endif |
1888 | }; | |
e56b3bc7 | 1889 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1890 | |
1891 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1892 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1893 | |
1894 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1895 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1896 | = {CPU_BITS_ALL}; |
b3199c02 | 1897 | #else |
4b804c85 | 1898 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1899 | #endif |
4b804c85 | 1900 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1901 | |
4b804c85 RV |
1902 | struct cpumask __cpu_online_mask __read_mostly; |
1903 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1904 | |
4b804c85 RV |
1905 | struct cpumask __cpu_present_mask __read_mostly; |
1906 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1907 | |
4b804c85 RV |
1908 | struct cpumask __cpu_active_mask __read_mostly; |
1909 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1910 | |
3fa41520 RR |
1911 | void init_cpu_present(const struct cpumask *src) |
1912 | { | |
c4c54dd1 | 1913 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1914 | } |
1915 | ||
1916 | void init_cpu_possible(const struct cpumask *src) | |
1917 | { | |
c4c54dd1 | 1918 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1919 | } |
1920 | ||
1921 | void init_cpu_online(const struct cpumask *src) | |
1922 | { | |
c4c54dd1 | 1923 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1924 | } |
cff7d378 TG |
1925 | |
1926 | /* | |
1927 | * Activate the first processor. | |
1928 | */ | |
1929 | void __init boot_cpu_init(void) | |
1930 | { | |
1931 | int cpu = smp_processor_id(); | |
1932 | ||
1933 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1934 | set_cpu_online(cpu, true); | |
1935 | set_cpu_active(cpu, true); | |
1936 | set_cpu_present(cpu, true); | |
1937 | set_cpu_possible(cpu, true); | |
1938 | } | |
1939 | ||
1940 | /* | |
1941 | * Must be called _AFTER_ setting up the per_cpu areas | |
1942 | */ | |
1943 | void __init boot_cpu_state_init(void) | |
1944 | { | |
1945 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1946 | } |