]>
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; | |
4cb28ced | 55 | enum cpuhp_state cb_state; |
4cb28ced TG |
56 | int result; |
57 | struct completion done; | |
58 | #endif | |
cff7d378 TG |
59 | }; |
60 | ||
61 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
62 | ||
63 | /** | |
64 | * cpuhp_step - Hotplug state machine step | |
65 | * @name: Name of the step | |
66 | * @startup: Startup function of the step | |
67 | * @teardown: Teardown function of the step | |
68 | * @skip_onerr: Do not invoke the functions on error rollback | |
69 | * Will go away once the notifiers are gone | |
757c989b | 70 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
71 | */ |
72 | struct cpuhp_step { | |
73 | const char *name; | |
74 | int (*startup)(unsigned int cpu); | |
75 | int (*teardown)(unsigned int cpu); | |
76 | bool skip_onerr; | |
757c989b | 77 | bool cant_stop; |
cff7d378 TG |
78 | }; |
79 | ||
98f8cdce | 80 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 81 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 82 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 83 | |
a724632c TG |
84 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
85 | { | |
86 | /* | |
87 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
88 | * purposes as that state is handled explicitly in cpu_down. | |
89 | */ | |
90 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
91 | } | |
92 | ||
93 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
94 | { | |
95 | struct cpuhp_step *sp; | |
96 | ||
97 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
98 | return sp + state; | |
99 | } | |
100 | ||
cff7d378 TG |
101 | /** |
102 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
103 | * @cpu: The cpu for which the callback should be invoked | |
104 | * @step: The step in the state machine | |
a724632c | 105 | * @bringup: True if the bringup callback should be invoked |
cff7d378 TG |
106 | * |
107 | * Called from cpu hotplug and from the state register machinery | |
108 | */ | |
a724632c TG |
109 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
110 | bool bringup) | |
cff7d378 TG |
111 | { |
112 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c TG |
113 | struct cpuhp_step *step = cpuhp_get_step(state); |
114 | int (*cb)(unsigned int cpu) = bringup ? step->startup : step->teardown; | |
cff7d378 TG |
115 | int ret = 0; |
116 | ||
117 | if (cb) { | |
a724632c | 118 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 119 | ret = cb(cpu); |
a724632c | 120 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cff7d378 TG |
121 | } |
122 | return ret; | |
123 | } | |
124 | ||
98a79d6a | 125 | #ifdef CONFIG_SMP |
b3199c02 | 126 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 127 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
128 | bool cpuhp_tasks_frozen; |
129 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 130 | |
79a6cdeb | 131 | /* |
93ae4f97 SB |
132 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
133 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
134 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
135 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
136 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
137 | */ |
138 | void cpu_maps_update_begin(void) | |
139 | { | |
140 | mutex_lock(&cpu_add_remove_lock); | |
141 | } | |
93ae4f97 | 142 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
143 | |
144 | void cpu_maps_update_done(void) | |
145 | { | |
146 | mutex_unlock(&cpu_add_remove_lock); | |
147 | } | |
93ae4f97 | 148 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 149 | |
5c113fbe | 150 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 151 | |
e3920fb4 RW |
152 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
153 | * Should always be manipulated under cpu_add_remove_lock | |
154 | */ | |
155 | static int cpu_hotplug_disabled; | |
156 | ||
79a6cdeb LJ |
157 | #ifdef CONFIG_HOTPLUG_CPU |
158 | ||
d221938c GS |
159 | static struct { |
160 | struct task_struct *active_writer; | |
87af9e7f DH |
161 | /* wait queue to wake up the active_writer */ |
162 | wait_queue_head_t wq; | |
163 | /* verifies that no writer will get active while readers are active */ | |
164 | struct mutex lock; | |
d221938c GS |
165 | /* |
166 | * Also blocks the new readers during | |
167 | * an ongoing cpu hotplug operation. | |
168 | */ | |
87af9e7f | 169 | atomic_t refcount; |
a19423b9 GS |
170 | |
171 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
172 | struct lockdep_map dep_map; | |
173 | #endif | |
31950eb6 LT |
174 | } cpu_hotplug = { |
175 | .active_writer = NULL, | |
87af9e7f | 176 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 177 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
178 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
179 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
180 | #endif | |
31950eb6 | 181 | }; |
d221938c | 182 | |
a19423b9 GS |
183 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
184 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
185 | #define cpuhp_lock_acquire_tryread() \ |
186 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
187 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
188 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
189 | ||
62db99f4 | 190 | |
86ef5c9a | 191 | void get_online_cpus(void) |
a9d9baa1 | 192 | { |
d221938c GS |
193 | might_sleep(); |
194 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 195 | return; |
a19423b9 | 196 | cpuhp_lock_acquire_read(); |
d221938c | 197 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 198 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 199 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 200 | } |
86ef5c9a | 201 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 202 | |
86ef5c9a | 203 | void put_online_cpus(void) |
a9d9baa1 | 204 | { |
87af9e7f DH |
205 | int refcount; |
206 | ||
d221938c | 207 | if (cpu_hotplug.active_writer == current) |
aa953877 | 208 | return; |
075663d1 | 209 | |
87af9e7f DH |
210 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
211 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
212 | atomic_inc(&cpu_hotplug.refcount); | |
213 | ||
214 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
215 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 216 | |
a19423b9 | 217 | cpuhp_lock_release(); |
d221938c | 218 | |
a9d9baa1 | 219 | } |
86ef5c9a | 220 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 221 | |
d221938c GS |
222 | /* |
223 | * This ensures that the hotplug operation can begin only when the | |
224 | * refcount goes to zero. | |
225 | * | |
226 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
227 | * will be blocked by the cpu_hotplug.lock | |
228 | * | |
d2ba7e2a ON |
229 | * Since cpu_hotplug_begin() is always called after invoking |
230 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
231 | * |
232 | * Note that theoretically, there is a possibility of a livelock: | |
233 | * - Refcount goes to zero, last reader wakes up the sleeping | |
234 | * writer. | |
235 | * - Last reader unlocks the cpu_hotplug.lock. | |
236 | * - A new reader arrives at this moment, bumps up the refcount. | |
237 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
238 | * non zero and goes to sleep again. | |
239 | * | |
240 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 241 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
242 | * |
243 | */ | |
b9d10be7 | 244 | void cpu_hotplug_begin(void) |
d221938c | 245 | { |
87af9e7f | 246 | DEFINE_WAIT(wait); |
d2ba7e2a | 247 | |
87af9e7f | 248 | cpu_hotplug.active_writer = current; |
a19423b9 | 249 | cpuhp_lock_acquire(); |
87af9e7f | 250 | |
d2ba7e2a ON |
251 | for (;;) { |
252 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
253 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
254 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
255 | break; | |
d221938c GS |
256 | mutex_unlock(&cpu_hotplug.lock); |
257 | schedule(); | |
d221938c | 258 | } |
87af9e7f | 259 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
260 | } |
261 | ||
b9d10be7 | 262 | void cpu_hotplug_done(void) |
d221938c GS |
263 | { |
264 | cpu_hotplug.active_writer = NULL; | |
265 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 266 | cpuhp_lock_release(); |
d221938c | 267 | } |
79a6cdeb | 268 | |
16e53dbf SB |
269 | /* |
270 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
271 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
272 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
273 | * hotplug path before performing hotplug operations. So acquiring that lock | |
274 | * guarantees mutual exclusion from any currently running hotplug operations. | |
275 | */ | |
276 | void cpu_hotplug_disable(void) | |
277 | { | |
278 | cpu_maps_update_begin(); | |
89af7ba5 | 279 | cpu_hotplug_disabled++; |
16e53dbf SB |
280 | cpu_maps_update_done(); |
281 | } | |
32145c46 | 282 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf SB |
283 | |
284 | void cpu_hotplug_enable(void) | |
285 | { | |
286 | cpu_maps_update_begin(); | |
89af7ba5 | 287 | WARN_ON(--cpu_hotplug_disabled < 0); |
16e53dbf SB |
288 | cpu_maps_update_done(); |
289 | } | |
32145c46 | 290 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 291 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 292 | |
1da177e4 | 293 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 294 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 295 | { |
bd5349cf | 296 | int ret; |
d221938c | 297 | cpu_maps_update_begin(); |
bd5349cf | 298 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 299 | cpu_maps_update_done(); |
bd5349cf | 300 | return ret; |
1da177e4 | 301 | } |
65edc68c | 302 | |
71cf5aee | 303 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
304 | { |
305 | return raw_notifier_chain_register(&cpu_chain, nb); | |
306 | } | |
307 | ||
090e77c3 | 308 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
309 | int *nr_calls) |
310 | { | |
090e77c3 TG |
311 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
312 | void *hcpu = (void *)(long)cpu; | |
313 | ||
e6bde73b AM |
314 | int ret; |
315 | ||
090e77c3 | 316 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 317 | nr_calls); |
e6bde73b AM |
318 | |
319 | return notifier_to_errno(ret); | |
e9fb7631 AM |
320 | } |
321 | ||
090e77c3 | 322 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 323 | { |
090e77c3 | 324 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
325 | } |
326 | ||
3b9d6da6 SAS |
327 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
328 | { | |
329 | BUG_ON(cpu_notify(val, cpu)); | |
330 | } | |
331 | ||
ba997462 TG |
332 | /* Notifier wrappers for transitioning to state machine */ |
333 | static int notify_prepare(unsigned int cpu) | |
334 | { | |
335 | int nr_calls = 0; | |
336 | int ret; | |
337 | ||
338 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
339 | if (ret) { | |
340 | nr_calls--; | |
341 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
342 | __func__, cpu); | |
343 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
344 | } | |
345 | return ret; | |
346 | } | |
347 | ||
348 | static int notify_online(unsigned int cpu) | |
349 | { | |
350 | cpu_notify(CPU_ONLINE, cpu); | |
351 | return 0; | |
352 | } | |
353 | ||
4baa0afc TG |
354 | static int notify_starting(unsigned int cpu) |
355 | { | |
356 | cpu_notify(CPU_STARTING, cpu); | |
357 | return 0; | |
358 | } | |
359 | ||
8df3e07e TG |
360 | static int bringup_wait_for_ap(unsigned int cpu) |
361 | { | |
362 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
363 | ||
364 | wait_for_completion(&st->done); | |
365 | return st->result; | |
366 | } | |
367 | ||
ba997462 TG |
368 | static int bringup_cpu(unsigned int cpu) |
369 | { | |
370 | struct task_struct *idle = idle_thread_get(cpu); | |
371 | int ret; | |
372 | ||
aa877175 BO |
373 | /* |
374 | * Some architectures have to walk the irq descriptors to | |
375 | * setup the vector space for the cpu which comes online. | |
376 | * Prevent irq alloc/free across the bringup. | |
377 | */ | |
378 | irq_lock_sparse(); | |
379 | ||
ba997462 TG |
380 | /* Arch-specific enabling code. */ |
381 | ret = __cpu_up(cpu, idle); | |
aa877175 | 382 | irq_unlock_sparse(); |
ba997462 TG |
383 | if (ret) { |
384 | cpu_notify(CPU_UP_CANCELED, cpu); | |
385 | return ret; | |
386 | } | |
8df3e07e | 387 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 388 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 389 | return ret; |
ba997462 TG |
390 | } |
391 | ||
2e1a3483 TG |
392 | /* |
393 | * Hotplug state machine related functions | |
394 | */ | |
a724632c | 395 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
396 | { |
397 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 398 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
399 | |
400 | if (!step->skip_onerr) | |
a724632c | 401 | cpuhp_invoke_callback(cpu, st->state, true); |
2e1a3483 TG |
402 | } |
403 | } | |
404 | ||
405 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 406 | enum cpuhp_state target) |
2e1a3483 TG |
407 | { |
408 | enum cpuhp_state prev_state = st->state; | |
409 | int ret = 0; | |
410 | ||
411 | for (; st->state > target; st->state--) { | |
a724632c | 412 | ret = cpuhp_invoke_callback(cpu, st->state, false); |
2e1a3483 TG |
413 | if (ret) { |
414 | st->target = prev_state; | |
a724632c | 415 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
416 | break; |
417 | } | |
418 | } | |
419 | return ret; | |
420 | } | |
421 | ||
a724632c | 422 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
423 | { |
424 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 425 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
426 | |
427 | if (!step->skip_onerr) | |
a724632c | 428 | cpuhp_invoke_callback(cpu, st->state, false); |
2e1a3483 TG |
429 | } |
430 | } | |
431 | ||
432 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 433 | enum cpuhp_state target) |
2e1a3483 TG |
434 | { |
435 | enum cpuhp_state prev_state = st->state; | |
436 | int ret = 0; | |
437 | ||
438 | while (st->state < target) { | |
2e1a3483 | 439 | st->state++; |
a724632c | 440 | ret = cpuhp_invoke_callback(cpu, st->state, true); |
2e1a3483 TG |
441 | if (ret) { |
442 | st->target = prev_state; | |
a724632c | 443 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
444 | break; |
445 | } | |
446 | } | |
447 | return ret; | |
448 | } | |
449 | ||
4cb28ced TG |
450 | /* |
451 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
452 | */ | |
453 | static void cpuhp_create(unsigned int cpu) | |
454 | { | |
455 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
456 | ||
457 | init_completion(&st->done); | |
458 | } | |
459 | ||
460 | static int cpuhp_should_run(unsigned int cpu) | |
461 | { | |
462 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
463 | ||
464 | return st->should_run; | |
465 | } | |
466 | ||
467 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
468 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
469 | { | |
1cf4f629 | 470 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 471 | |
a724632c | 472 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
473 | } |
474 | ||
475 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
476 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
477 | { | |
a724632c | 478 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
479 | } |
480 | ||
481 | /* | |
482 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
483 | * callbacks when a state gets [un]installed at runtime. | |
484 | */ | |
485 | static void cpuhp_thread_fun(unsigned int cpu) | |
486 | { | |
487 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
488 | int ret = 0; | |
489 | ||
490 | /* | |
491 | * Paired with the mb() in cpuhp_kick_ap_work and | |
492 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
493 | */ | |
494 | smp_mb(); | |
495 | if (!st->should_run) | |
496 | return; | |
497 | ||
498 | st->should_run = false; | |
499 | ||
500 | /* Single callback invocation for [un]install ? */ | |
a724632c | 501 | if (st->single) { |
4cb28ced TG |
502 | if (st->cb_state < CPUHP_AP_ONLINE) { |
503 | local_irq_disable(); | |
a724632c TG |
504 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
505 | st->bringup); | |
4cb28ced TG |
506 | local_irq_enable(); |
507 | } else { | |
a724632c TG |
508 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
509 | st->bringup); | |
4cb28ced | 510 | } |
3b9d6da6 SAS |
511 | } else if (st->rollback) { |
512 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
513 | ||
a724632c | 514 | undo_cpu_down(cpu, st); |
3b9d6da6 SAS |
515 | /* |
516 | * This is a momentary workaround to keep the notifier users | |
517 | * happy. Will go away once we got rid of the notifiers. | |
518 | */ | |
519 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); | |
520 | st->rollback = false; | |
4cb28ced | 521 | } else { |
1cf4f629 | 522 | /* Cannot happen .... */ |
8df3e07e | 523 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 524 | |
4cb28ced TG |
525 | /* Regular hotplug work */ |
526 | if (st->state < st->target) | |
527 | ret = cpuhp_ap_online(cpu, st); | |
528 | else if (st->state > st->target) | |
529 | ret = cpuhp_ap_offline(cpu, st); | |
530 | } | |
531 | st->result = ret; | |
532 | complete(&st->done); | |
533 | } | |
534 | ||
535 | /* Invoke a single callback on a remote cpu */ | |
a724632c TG |
536 | static int |
537 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup) | |
4cb28ced TG |
538 | { |
539 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
540 | ||
541 | if (!cpu_online(cpu)) | |
542 | return 0; | |
543 | ||
6a4e2451 TG |
544 | /* |
545 | * If we are up and running, use the hotplug thread. For early calls | |
546 | * we invoke the thread function directly. | |
547 | */ | |
548 | if (!st->thread) | |
a724632c | 549 | return cpuhp_invoke_callback(cpu, state, bringup); |
6a4e2451 | 550 | |
4cb28ced | 551 | st->cb_state = state; |
a724632c TG |
552 | st->single = true; |
553 | st->bringup = bringup; | |
554 | ||
4cb28ced TG |
555 | /* |
556 | * Make sure the above stores are visible before should_run becomes | |
557 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
558 | */ | |
559 | smp_mb(); | |
560 | st->should_run = true; | |
561 | wake_up_process(st->thread); | |
562 | wait_for_completion(&st->done); | |
563 | return st->result; | |
564 | } | |
565 | ||
566 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 567 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 568 | { |
4cb28ced | 569 | st->result = 0; |
a724632c | 570 | st->single = false; |
4cb28ced TG |
571 | /* |
572 | * Make sure the above stores are visible before should_run becomes | |
573 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
574 | */ | |
575 | smp_mb(); | |
576 | st->should_run = true; | |
577 | wake_up_process(st->thread); | |
1cf4f629 TG |
578 | } |
579 | ||
580 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
581 | { | |
582 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
583 | enum cpuhp_state state = st->state; | |
584 | ||
585 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
586 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
587 | wait_for_completion(&st->done); |
588 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
589 | return st->result; | |
590 | } | |
591 | ||
592 | static struct smp_hotplug_thread cpuhp_threads = { | |
593 | .store = &cpuhp_state.thread, | |
594 | .create = &cpuhp_create, | |
595 | .thread_should_run = cpuhp_should_run, | |
596 | .thread_fn = cpuhp_thread_fun, | |
597 | .thread_comm = "cpuhp/%u", | |
598 | .selfparking = true, | |
599 | }; | |
600 | ||
601 | void __init cpuhp_threads_init(void) | |
602 | { | |
603 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
604 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
605 | } | |
606 | ||
00b9b0af | 607 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 608 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 609 | EXPORT_SYMBOL(__register_cpu_notifier); |
71cf5aee | 610 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 611 | { |
d221938c | 612 | cpu_maps_update_begin(); |
bd5349cf | 613 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 614 | cpu_maps_update_done(); |
1da177e4 LT |
615 | } |
616 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
617 | ||
71cf5aee | 618 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
619 | { |
620 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
621 | } | |
622 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
623 | ||
e4cc2f87 AV |
624 | /** |
625 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
626 | * @cpu: a CPU id | |
627 | * | |
628 | * This function walks all processes, finds a valid mm struct for each one and | |
629 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
630 | * trivial, there are various non-obvious corner cases, which this function | |
631 | * tries to solve in a safe manner. | |
632 | * | |
633 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
634 | * be called only for an already offlined CPU. | |
635 | */ | |
cb79295e AV |
636 | void clear_tasks_mm_cpumask(int cpu) |
637 | { | |
638 | struct task_struct *p; | |
639 | ||
640 | /* | |
641 | * This function is called after the cpu is taken down and marked | |
642 | * offline, so its not like new tasks will ever get this cpu set in | |
643 | * their mm mask. -- Peter Zijlstra | |
644 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
645 | * full-fledged tasklist_lock. | |
646 | */ | |
e4cc2f87 | 647 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
648 | rcu_read_lock(); |
649 | for_each_process(p) { | |
650 | struct task_struct *t; | |
651 | ||
e4cc2f87 AV |
652 | /* |
653 | * Main thread might exit, but other threads may still have | |
654 | * a valid mm. Find one. | |
655 | */ | |
cb79295e AV |
656 | t = find_lock_task_mm(p); |
657 | if (!t) | |
658 | continue; | |
659 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
660 | task_unlock(t); | |
661 | } | |
662 | rcu_read_unlock(); | |
663 | } | |
664 | ||
b728ca06 | 665 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 666 | { |
b728ca06 | 667 | struct task_struct *g, *p; |
1da177e4 | 668 | |
a75a6068 ON |
669 | read_lock(&tasklist_lock); |
670 | for_each_process_thread(g, p) { | |
b728ca06 KT |
671 | if (!p->on_rq) |
672 | continue; | |
673 | /* | |
674 | * We do the check with unlocked task_rq(p)->lock. | |
675 | * Order the reading to do not warn about a task, | |
676 | * which was running on this cpu in the past, and | |
677 | * it's just been woken on another cpu. | |
678 | */ | |
679 | rmb(); | |
680 | if (task_cpu(p) != dead_cpu) | |
681 | continue; | |
682 | ||
683 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
684 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
685 | } |
686 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
687 | } |
688 | ||
98458172 TG |
689 | static int notify_down_prepare(unsigned int cpu) |
690 | { | |
691 | int err, nr_calls = 0; | |
692 | ||
693 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
694 | if (err) { | |
695 | nr_calls--; | |
696 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
697 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
698 | __func__, cpu); | |
699 | } | |
700 | return err; | |
701 | } | |
702 | ||
4baa0afc TG |
703 | static int notify_dying(unsigned int cpu) |
704 | { | |
705 | cpu_notify(CPU_DYING, cpu); | |
706 | return 0; | |
707 | } | |
708 | ||
1da177e4 | 709 | /* Take this CPU down. */ |
71cf5aee | 710 | static int take_cpu_down(void *_param) |
1da177e4 | 711 | { |
4baa0afc TG |
712 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
713 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 714 | int err, cpu = smp_processor_id(); |
1da177e4 | 715 | |
1da177e4 LT |
716 | /* Ensure this CPU doesn't handle any more interrupts. */ |
717 | err = __cpu_disable(); | |
718 | if (err < 0) | |
f3705136 | 719 | return err; |
1da177e4 | 720 | |
a724632c TG |
721 | /* |
722 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
723 | * do this step again. | |
724 | */ | |
725 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
726 | st->state--; | |
4baa0afc | 727 | /* Invoke the former CPU_DYING callbacks */ |
a724632c TG |
728 | for (; st->state > target; st->state--) |
729 | cpuhp_invoke_callback(cpu, st->state, false); | |
4baa0afc | 730 | |
52c063d1 TG |
731 | /* Give up timekeeping duties */ |
732 | tick_handover_do_timer(); | |
14e568e7 | 733 | /* Park the stopper thread */ |
090e77c3 | 734 | stop_machine_park(cpu); |
f3705136 | 735 | return 0; |
1da177e4 LT |
736 | } |
737 | ||
98458172 | 738 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 739 | { |
e69aab13 | 740 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 741 | int err; |
1da177e4 | 742 | |
2a58c527 | 743 | /* Park the smpboot threads */ |
1cf4f629 | 744 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 745 | smpboot_park_threads(cpu); |
1cf4f629 | 746 | |
6acce3ef | 747 | /* |
a8994181 TG |
748 | * Prevent irq alloc/free while the dying cpu reorganizes the |
749 | * interrupt affinities. | |
6acce3ef | 750 | */ |
a8994181 | 751 | irq_lock_sparse(); |
6acce3ef | 752 | |
a8994181 TG |
753 | /* |
754 | * So now all preempt/rcu users must observe !cpu_active(). | |
755 | */ | |
090e77c3 | 756 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 757 | if (err) { |
3b9d6da6 | 758 | /* CPU refused to die */ |
a8994181 | 759 | irq_unlock_sparse(); |
3b9d6da6 SAS |
760 | /* Unpark the hotplug thread so we can rollback there */ |
761 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 762 | return err; |
8fa1d7d3 | 763 | } |
04321587 | 764 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 765 | |
48c5ccae PZ |
766 | /* |
767 | * The migration_call() CPU_DYING callback will have removed all | |
768 | * runnable tasks from the cpu, there's only the idle task left now | |
769 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
770 | * |
771 | * Wait for the stop thread to go away. | |
48c5ccae | 772 | */ |
e69aab13 TG |
773 | wait_for_completion(&st->done); |
774 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 775 | |
a8994181 TG |
776 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
777 | irq_unlock_sparse(); | |
778 | ||
345527b1 | 779 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
780 | /* This actually kills the CPU. */ |
781 | __cpu_die(cpu); | |
782 | ||
a49b116d | 783 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
784 | return 0; |
785 | } | |
1da177e4 | 786 | |
98458172 TG |
787 | static int notify_dead(unsigned int cpu) |
788 | { | |
789 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 790 | check_for_tasks(cpu); |
98458172 TG |
791 | return 0; |
792 | } | |
793 | ||
71f87b2f TG |
794 | static void cpuhp_complete_idle_dead(void *arg) |
795 | { | |
796 | struct cpuhp_cpu_state *st = arg; | |
797 | ||
798 | complete(&st->done); | |
799 | } | |
800 | ||
e69aab13 TG |
801 | void cpuhp_report_idle_dead(void) |
802 | { | |
803 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
804 | ||
805 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 806 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
807 | st->state = CPUHP_AP_IDLE_DEAD; |
808 | /* | |
809 | * We cannot call complete after rcu_report_dead() so we delegate it | |
810 | * to an online cpu. | |
811 | */ | |
812 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
813 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
814 | } |
815 | ||
cff7d378 TG |
816 | #else |
817 | #define notify_down_prepare NULL | |
818 | #define takedown_cpu NULL | |
819 | #define notify_dead NULL | |
4baa0afc | 820 | #define notify_dying NULL |
cff7d378 TG |
821 | #endif |
822 | ||
823 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 824 | |
98458172 | 825 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
826 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
827 | enum cpuhp_state target) | |
98458172 | 828 | { |
cff7d378 TG |
829 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
830 | int prev_state, ret = 0; | |
831 | bool hasdied = false; | |
98458172 TG |
832 | |
833 | if (num_online_cpus() == 1) | |
834 | return -EBUSY; | |
835 | ||
757c989b | 836 | if (!cpu_present(cpu)) |
98458172 TG |
837 | return -EINVAL; |
838 | ||
839 | cpu_hotplug_begin(); | |
840 | ||
841 | cpuhp_tasks_frozen = tasks_frozen; | |
842 | ||
cff7d378 | 843 | prev_state = st->state; |
af1f4045 | 844 | st->target = target; |
1cf4f629 TG |
845 | /* |
846 | * If the current CPU state is in the range of the AP hotplug thread, | |
847 | * then we need to kick the thread. | |
848 | */ | |
8df3e07e | 849 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
850 | ret = cpuhp_kick_ap_work(cpu); |
851 | /* | |
852 | * The AP side has done the error rollback already. Just | |
853 | * return the error code.. | |
854 | */ | |
855 | if (ret) | |
856 | goto out; | |
857 | ||
858 | /* | |
859 | * We might have stopped still in the range of the AP hotplug | |
860 | * thread. Nothing to do anymore. | |
861 | */ | |
8df3e07e | 862 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
863 | goto out; |
864 | } | |
865 | /* | |
8df3e07e | 866 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
867 | * to do the further cleanups. |
868 | */ | |
a724632c | 869 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
870 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
871 | st->target = prev_state; | |
872 | st->rollback = true; | |
873 | cpuhp_kick_ap_work(cpu); | |
874 | } | |
98458172 | 875 | |
cff7d378 | 876 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; |
1cf4f629 | 877 | out: |
d221938c | 878 | cpu_hotplug_done(); |
cff7d378 TG |
879 | /* This post dead nonsense must die */ |
880 | if (!ret && hasdied) | |
090e77c3 | 881 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 882 | return ret; |
e3920fb4 RW |
883 | } |
884 | ||
af1f4045 | 885 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 886 | { |
9ea09af3 | 887 | int err; |
e3920fb4 | 888 | |
d221938c | 889 | cpu_maps_update_begin(); |
e761b772 MK |
890 | |
891 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 892 | err = -EBUSY; |
e761b772 MK |
893 | goto out; |
894 | } | |
895 | ||
af1f4045 | 896 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 897 | |
e761b772 | 898 | out: |
d221938c | 899 | cpu_maps_update_done(); |
1da177e4 LT |
900 | return err; |
901 | } | |
af1f4045 TG |
902 | int cpu_down(unsigned int cpu) |
903 | { | |
904 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
905 | } | |
b62b8ef9 | 906 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
907 | #endif /*CONFIG_HOTPLUG_CPU*/ |
908 | ||
4baa0afc TG |
909 | /** |
910 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
911 | * @cpu: cpu that just started | |
912 | * | |
913 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
914 | * It must be called by the arch code on the new cpu, before the new cpu | |
915 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
916 | */ | |
917 | void notify_cpu_starting(unsigned int cpu) | |
918 | { | |
919 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
920 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
921 | ||
922 | while (st->state < target) { | |
4baa0afc | 923 | st->state++; |
a724632c | 924 | cpuhp_invoke_callback(cpu, st->state, true); |
4baa0afc TG |
925 | } |
926 | } | |
927 | ||
949338e3 TG |
928 | /* |
929 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
930 | * the stopper thread and unpark the smpboot threads. If the target state is |
931 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
932 | * cpu further. | |
949338e3 | 933 | */ |
8df3e07e | 934 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 935 | { |
8df3e07e TG |
936 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
937 | unsigned int cpu = smp_processor_id(); | |
938 | ||
939 | /* Happens for the boot cpu */ | |
940 | if (state != CPUHP_AP_ONLINE_IDLE) | |
941 | return; | |
942 | ||
943 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 944 | |
8df3e07e | 945 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 946 | stop_machine_unpark(cpu); |
1cf4f629 | 947 | kthread_unpark(st->thread); |
8df3e07e TG |
948 | |
949 | /* Should we go further up ? */ | |
950 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
951 | __cpuhp_kick_ap_work(st); | |
952 | else | |
953 | complete(&st->done); | |
949338e3 TG |
954 | } |
955 | ||
e3920fb4 | 956 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 957 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 958 | { |
cff7d378 | 959 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 960 | struct task_struct *idle; |
2e1a3483 | 961 | int ret = 0; |
1da177e4 | 962 | |
d221938c | 963 | cpu_hotplug_begin(); |
38498a67 | 964 | |
757c989b | 965 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
966 | ret = -EINVAL; |
967 | goto out; | |
968 | } | |
969 | ||
757c989b TG |
970 | /* |
971 | * The caller of do_cpu_up might have raced with another | |
972 | * caller. Ignore it for now. | |
973 | */ | |
974 | if (st->state >= target) | |
38498a67 | 975 | goto out; |
757c989b TG |
976 | |
977 | if (st->state == CPUHP_OFFLINE) { | |
978 | /* Let it fail before we try to bring the cpu up */ | |
979 | idle = idle_thread_get(cpu); | |
980 | if (IS_ERR(idle)) { | |
981 | ret = PTR_ERR(idle); | |
982 | goto out; | |
983 | } | |
3bb5d2ee | 984 | } |
38498a67 | 985 | |
ba997462 TG |
986 | cpuhp_tasks_frozen = tasks_frozen; |
987 | ||
af1f4045 | 988 | st->target = target; |
1cf4f629 TG |
989 | /* |
990 | * If the current CPU state is in the range of the AP hotplug thread, | |
991 | * then we need to kick the thread once more. | |
992 | */ | |
8df3e07e | 993 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
994 | ret = cpuhp_kick_ap_work(cpu); |
995 | /* | |
996 | * The AP side has done the error rollback already. Just | |
997 | * return the error code.. | |
998 | */ | |
999 | if (ret) | |
1000 | goto out; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1005 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1006 | * responsible for bringing it up to the target state. |
1007 | */ | |
8df3e07e | 1008 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1009 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1010 | out: |
d221938c | 1011 | cpu_hotplug_done(); |
e3920fb4 RW |
1012 | return ret; |
1013 | } | |
1014 | ||
af1f4045 | 1015 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1016 | { |
1017 | int err = 0; | |
cf23422b | 1018 | |
e0b582ec | 1019 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1020 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1021 | cpu); | |
87d5e023 | 1022 | #if defined(CONFIG_IA64) |
84117da5 | 1023 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1024 | #endif |
1025 | return -EINVAL; | |
1026 | } | |
e3920fb4 | 1027 | |
01b0f197 TK |
1028 | err = try_online_node(cpu_to_node(cpu)); |
1029 | if (err) | |
1030 | return err; | |
cf23422b | 1031 | |
d221938c | 1032 | cpu_maps_update_begin(); |
e761b772 MK |
1033 | |
1034 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1035 | err = -EBUSY; |
e761b772 MK |
1036 | goto out; |
1037 | } | |
1038 | ||
af1f4045 | 1039 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1040 | out: |
d221938c | 1041 | cpu_maps_update_done(); |
e3920fb4 RW |
1042 | return err; |
1043 | } | |
af1f4045 TG |
1044 | |
1045 | int cpu_up(unsigned int cpu) | |
1046 | { | |
1047 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1048 | } | |
a513f6ba | 1049 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1050 | |
f3de4be9 | 1051 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1052 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
1053 | |
1054 | int disable_nonboot_cpus(void) | |
1055 | { | |
e9a5f426 | 1056 | int cpu, first_cpu, error = 0; |
e3920fb4 | 1057 | |
d221938c | 1058 | cpu_maps_update_begin(); |
e0b582ec | 1059 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
1060 | /* |
1061 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1062 | * with the userspace trying to use the CPU hotplug at the same time |
1063 | */ | |
e0b582ec | 1064 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1065 | |
84117da5 | 1066 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
1067 | for_each_online_cpu(cpu) { |
1068 | if (cpu == first_cpu) | |
1069 | continue; | |
bb3632c6 | 1070 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1071 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1072 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1073 | if (!error) |
e0b582ec | 1074 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1075 | else { |
84117da5 | 1076 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1077 | break; |
1078 | } | |
1079 | } | |
86886e55 | 1080 | |
89af7ba5 | 1081 | if (!error) |
e3920fb4 | 1082 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1083 | else |
84117da5 | 1084 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1085 | |
1086 | /* | |
1087 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1088 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1089 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1090 | */ | |
1091 | cpu_hotplug_disabled++; | |
1092 | ||
d221938c | 1093 | cpu_maps_update_done(); |
e3920fb4 RW |
1094 | return error; |
1095 | } | |
1096 | ||
d0af9eed SS |
1097 | void __weak arch_enable_nonboot_cpus_begin(void) |
1098 | { | |
1099 | } | |
1100 | ||
1101 | void __weak arch_enable_nonboot_cpus_end(void) | |
1102 | { | |
1103 | } | |
1104 | ||
71cf5aee | 1105 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1106 | { |
1107 | int cpu, error; | |
1108 | ||
1109 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1110 | cpu_maps_update_begin(); |
89af7ba5 | 1111 | WARN_ON(--cpu_hotplug_disabled < 0); |
e0b582ec | 1112 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1113 | goto out; |
e3920fb4 | 1114 | |
84117da5 | 1115 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1116 | |
1117 | arch_enable_nonboot_cpus_begin(); | |
1118 | ||
e0b582ec | 1119 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1120 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1121 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1122 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1123 | if (!error) { |
84117da5 | 1124 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1125 | continue; |
1126 | } | |
84117da5 | 1127 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1128 | } |
d0af9eed SS |
1129 | |
1130 | arch_enable_nonboot_cpus_end(); | |
1131 | ||
e0b582ec | 1132 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1133 | out: |
d221938c | 1134 | cpu_maps_update_done(); |
1da177e4 | 1135 | } |
e0b582ec | 1136 | |
d7268a31 | 1137 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1138 | { |
1139 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1140 | return -ENOMEM; | |
1141 | return 0; | |
1142 | } | |
1143 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1144 | |
79cfbdfa SB |
1145 | /* |
1146 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1147 | * ensure that the state of the system with respect to the tasks being frozen | |
1148 | * or not, as reported by the notification, remains unchanged *throughout the | |
1149 | * duration* of the execution of the callbacks. | |
1150 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1151 | * | |
1152 | * This synchronization is implemented by mutually excluding regular CPU | |
1153 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1154 | * Hibernate notifications. | |
1155 | */ | |
1156 | static int | |
1157 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1158 | unsigned long action, void *ptr) | |
1159 | { | |
1160 | switch (action) { | |
1161 | ||
1162 | case PM_SUSPEND_PREPARE: | |
1163 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1164 | cpu_hotplug_disable(); |
79cfbdfa SB |
1165 | break; |
1166 | ||
1167 | case PM_POST_SUSPEND: | |
1168 | case PM_POST_HIBERNATION: | |
16e53dbf | 1169 | cpu_hotplug_enable(); |
79cfbdfa SB |
1170 | break; |
1171 | ||
1172 | default: | |
1173 | return NOTIFY_DONE; | |
1174 | } | |
1175 | ||
1176 | return NOTIFY_OK; | |
1177 | } | |
1178 | ||
1179 | ||
d7268a31 | 1180 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1181 | { |
6e32d479 FY |
1182 | /* |
1183 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1184 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1185 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1186 | */ | |
79cfbdfa SB |
1187 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1188 | return 0; | |
1189 | } | |
1190 | core_initcall(cpu_hotplug_pm_sync_init); | |
1191 | ||
f3de4be9 | 1192 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
1193 | |
1194 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 1195 | |
cff7d378 TG |
1196 | /* Boot processor state steps */ |
1197 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1198 | [CPUHP_OFFLINE] = { | |
1199 | .name = "offline", | |
1200 | .startup = NULL, | |
1201 | .teardown = NULL, | |
1202 | }, | |
1203 | #ifdef CONFIG_SMP | |
1204 | [CPUHP_CREATE_THREADS]= { | |
1205 | .name = "threads:create", | |
1206 | .startup = smpboot_create_threads, | |
1207 | .teardown = NULL, | |
757c989b | 1208 | .cant_stop = true, |
cff7d378 | 1209 | }, |
00e16c3d TG |
1210 | [CPUHP_PERF_PREPARE] = { |
1211 | .name = "perf prepare", | |
1212 | .startup = perf_event_init_cpu, | |
1213 | .teardown = perf_event_exit_cpu, | |
1214 | }, | |
7ee681b2 TG |
1215 | [CPUHP_WORKQUEUE_PREP] = { |
1216 | .name = "workqueue prepare", | |
1217 | .startup = workqueue_prepare_cpu, | |
1218 | .teardown = NULL, | |
1219 | }, | |
27590dc1 TG |
1220 | [CPUHP_HRTIMERS_PREPARE] = { |
1221 | .name = "hrtimers prepare", | |
1222 | .startup = hrtimers_prepare_cpu, | |
1223 | .teardown = hrtimers_dead_cpu, | |
1224 | }, | |
31487f83 RW |
1225 | [CPUHP_SMPCFD_PREPARE] = { |
1226 | .name = "SMPCFD prepare", | |
1227 | .startup = smpcfd_prepare_cpu, | |
1228 | .teardown = smpcfd_dead_cpu, | |
1229 | }, | |
4df83742 TG |
1230 | [CPUHP_RCUTREE_PREP] = { |
1231 | .name = "RCU-tree prepare", | |
1232 | .startup = rcutree_prepare_cpu, | |
1233 | .teardown = rcutree_dead_cpu, | |
1234 | }, | |
d10ef6f9 TG |
1235 | /* |
1236 | * Preparatory and dead notifiers. Will be replaced once the notifiers | |
1237 | * are converted to states. | |
1238 | */ | |
cff7d378 TG |
1239 | [CPUHP_NOTIFY_PREPARE] = { |
1240 | .name = "notify:prepare", | |
1241 | .startup = notify_prepare, | |
1242 | .teardown = notify_dead, | |
1243 | .skip_onerr = true, | |
757c989b | 1244 | .cant_stop = true, |
cff7d378 | 1245 | }, |
4fae16df RC |
1246 | /* |
1247 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1248 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1249 | * otherwise a RCU stall occurs. | |
1250 | */ | |
1251 | [CPUHP_TIMERS_DEAD] = { | |
1252 | .name = "timers dead", | |
1253 | .startup = NULL, | |
1254 | .teardown = timers_dead_cpu, | |
1255 | }, | |
d10ef6f9 | 1256 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1257 | [CPUHP_BRINGUP_CPU] = { |
1258 | .name = "cpu:bringup", | |
1259 | .startup = bringup_cpu, | |
4baa0afc | 1260 | .teardown = NULL, |
757c989b | 1261 | .cant_stop = true, |
4baa0afc | 1262 | }, |
31487f83 RW |
1263 | [CPUHP_AP_SMPCFD_DYING] = { |
1264 | .startup = NULL, | |
1265 | .teardown = smpcfd_dying_cpu, | |
1266 | }, | |
d10ef6f9 TG |
1267 | /* |
1268 | * Handled on controll processor until the plugged processor manages | |
1269 | * this itself. | |
1270 | */ | |
4baa0afc TG |
1271 | [CPUHP_TEARDOWN_CPU] = { |
1272 | .name = "cpu:teardown", | |
1273 | .startup = NULL, | |
cff7d378 | 1274 | .teardown = takedown_cpu, |
757c989b | 1275 | .cant_stop = true, |
cff7d378 | 1276 | }, |
a7c73414 TG |
1277 | #else |
1278 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1279 | #endif |
cff7d378 TG |
1280 | }; |
1281 | ||
4baa0afc TG |
1282 | /* Application processor state steps */ |
1283 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1284 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1285 | /* Final state before CPU kills itself */ |
1286 | [CPUHP_AP_IDLE_DEAD] = { | |
1287 | .name = "idle:dead", | |
1288 | }, | |
1289 | /* | |
1290 | * Last state before CPU enters the idle loop to die. Transient state | |
1291 | * for synchronization. | |
1292 | */ | |
1293 | [CPUHP_AP_OFFLINE] = { | |
1294 | .name = "ap:offline", | |
1295 | .cant_stop = true, | |
1296 | }, | |
9cf7243d TG |
1297 | /* First state is scheduler control. Interrupts are disabled */ |
1298 | [CPUHP_AP_SCHED_STARTING] = { | |
1299 | .name = "sched:starting", | |
1300 | .startup = sched_cpu_starting, | |
f2785ddb | 1301 | .teardown = sched_cpu_dying, |
9cf7243d | 1302 | }, |
4df83742 TG |
1303 | [CPUHP_AP_RCUTREE_DYING] = { |
1304 | .startup = NULL, | |
1305 | .teardown = rcutree_dying_cpu, | |
1306 | }, | |
d10ef6f9 TG |
1307 | /* |
1308 | * Low level startup/teardown notifiers. Run with interrupts | |
1309 | * disabled. Will be removed once the notifiers are converted to | |
1310 | * states. | |
1311 | */ | |
4baa0afc TG |
1312 | [CPUHP_AP_NOTIFY_STARTING] = { |
1313 | .name = "notify:starting", | |
1314 | .startup = notify_starting, | |
1315 | .teardown = notify_dying, | |
1316 | .skip_onerr = true, | |
757c989b | 1317 | .cant_stop = true, |
4baa0afc | 1318 | }, |
d10ef6f9 TG |
1319 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1320 | * state for synchronsization */ | |
1321 | [CPUHP_AP_ONLINE] = { | |
1322 | .name = "ap:online", | |
1323 | }, | |
1324 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 TG |
1325 | [CPUHP_AP_SMPBOOT_THREADS] = { |
1326 | .name = "smpboot:threads", | |
1327 | .startup = smpboot_unpark_threads, | |
2a58c527 | 1328 | .teardown = NULL, |
1cf4f629 | 1329 | }, |
00e16c3d TG |
1330 | [CPUHP_AP_PERF_ONLINE] = { |
1331 | .name = "perf online", | |
1332 | .startup = perf_event_init_cpu, | |
1333 | .teardown = perf_event_exit_cpu, | |
1334 | }, | |
7ee681b2 TG |
1335 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
1336 | .name = "workqueue online", | |
1337 | .startup = workqueue_online_cpu, | |
1338 | .teardown = workqueue_offline_cpu, | |
1339 | }, | |
4df83742 TG |
1340 | [CPUHP_AP_RCUTREE_ONLINE] = { |
1341 | .name = "RCU-tree online", | |
1342 | .startup = rcutree_online_cpu, | |
1343 | .teardown = rcutree_offline_cpu, | |
1344 | }, | |
00e16c3d | 1345 | |
d10ef6f9 TG |
1346 | /* |
1347 | * Online/down_prepare notifiers. Will be removed once the notifiers | |
1348 | * are converted to states. | |
1349 | */ | |
1cf4f629 TG |
1350 | [CPUHP_AP_NOTIFY_ONLINE] = { |
1351 | .name = "notify:online", | |
1352 | .startup = notify_online, | |
1353 | .teardown = notify_down_prepare, | |
3b9d6da6 | 1354 | .skip_onerr = true, |
1cf4f629 | 1355 | }, |
4baa0afc | 1356 | #endif |
d10ef6f9 TG |
1357 | /* |
1358 | * The dynamically registered state space is here | |
1359 | */ | |
1360 | ||
aaddd7d1 TG |
1361 | #ifdef CONFIG_SMP |
1362 | /* Last state is scheduler control setting the cpu active */ | |
1363 | [CPUHP_AP_ACTIVE] = { | |
1364 | .name = "sched:active", | |
1365 | .startup = sched_cpu_activate, | |
1366 | .teardown = sched_cpu_deactivate, | |
1367 | }, | |
1368 | #endif | |
1369 | ||
d10ef6f9 | 1370 | /* CPU is fully up and running. */ |
4baa0afc TG |
1371 | [CPUHP_ONLINE] = { |
1372 | .name = "online", | |
1373 | .startup = NULL, | |
1374 | .teardown = NULL, | |
1375 | }, | |
1376 | }; | |
1377 | ||
5b7aa87e TG |
1378 | /* Sanity check for callbacks */ |
1379 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1380 | { | |
1381 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1382 | return -EINVAL; | |
1383 | return 0; | |
1384 | } | |
1385 | ||
5b7aa87e TG |
1386 | static void cpuhp_store_callbacks(enum cpuhp_state state, |
1387 | const char *name, | |
1388 | int (*startup)(unsigned int cpu), | |
1389 | int (*teardown)(unsigned int cpu)) | |
1390 | { | |
1391 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1392 | struct cpuhp_step *sp; | |
1393 | ||
1394 | mutex_lock(&cpuhp_state_mutex); | |
1395 | sp = cpuhp_get_step(state); | |
1396 | sp->startup = startup; | |
1397 | sp->teardown = teardown; | |
1398 | sp->name = name; | |
1399 | mutex_unlock(&cpuhp_state_mutex); | |
1400 | } | |
1401 | ||
1402 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1403 | { | |
1404 | return cpuhp_get_step(state)->teardown; | |
1405 | } | |
1406 | ||
5b7aa87e TG |
1407 | /* |
1408 | * Call the startup/teardown function for a step either on the AP or | |
1409 | * on the current CPU. | |
1410 | */ | |
a724632c | 1411 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup) |
5b7aa87e | 1412 | { |
a724632c | 1413 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1414 | int ret; |
1415 | ||
a724632c | 1416 | if ((bringup && !sp->startup) || (!bringup && !sp->teardown)) |
5b7aa87e | 1417 | return 0; |
5b7aa87e TG |
1418 | /* |
1419 | * The non AP bound callbacks can fail on bringup. On teardown | |
1420 | * e.g. module removal we crash for now. | |
1421 | */ | |
1cf4f629 TG |
1422 | #ifdef CONFIG_SMP |
1423 | if (cpuhp_is_ap_state(state)) | |
a724632c | 1424 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup); |
1cf4f629 | 1425 | else |
a724632c | 1426 | ret = cpuhp_invoke_callback(cpu, state, bringup); |
1cf4f629 | 1427 | #else |
a724632c | 1428 | ret = cpuhp_invoke_callback(cpu, state, bringup); |
1cf4f629 | 1429 | #endif |
5b7aa87e TG |
1430 | BUG_ON(ret && !bringup); |
1431 | return ret; | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1436 | * | |
1437 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1438 | */ | |
a724632c | 1439 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state) |
5b7aa87e TG |
1440 | { |
1441 | int cpu; | |
1442 | ||
5b7aa87e TG |
1443 | /* Roll back the already executed steps on the other cpus */ |
1444 | for_each_present_cpu(cpu) { | |
1445 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1446 | int cpustate = st->state; | |
1447 | ||
1448 | if (cpu >= failedcpu) | |
1449 | break; | |
1450 | ||
1451 | /* Did we invoke the startup call on that cpu ? */ | |
1452 | if (cpustate >= state) | |
a724632c | 1453 | cpuhp_issue_call(cpu, state, false); |
5b7aa87e TG |
1454 | } |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1459 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1460 | * by having no name assigned. | |
1461 | */ | |
1462 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1463 | { | |
1464 | enum cpuhp_state i; | |
1465 | ||
1466 | mutex_lock(&cpuhp_state_mutex); | |
1cf4f629 TG |
1467 | for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) { |
1468 | if (cpuhp_ap_states[i].name) | |
5b7aa87e TG |
1469 | continue; |
1470 | ||
1cf4f629 | 1471 | cpuhp_ap_states[i].name = "Reserved"; |
5b7aa87e TG |
1472 | mutex_unlock(&cpuhp_state_mutex); |
1473 | return i; | |
1474 | } | |
1475 | mutex_unlock(&cpuhp_state_mutex); | |
1476 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1477 | return -ENOSPC; | |
1478 | } | |
1479 | ||
1480 | /** | |
1481 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
1482 | * @state: The state to setup | |
1483 | * @invoke: If true, the startup function is invoked for cpus where | |
1484 | * cpu state >= @state | |
1485 | * @startup: startup callback function | |
1486 | * @teardown: teardown callback function | |
1487 | * | |
1488 | * Returns 0 if successful, otherwise a proper error code | |
1489 | */ | |
1490 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1491 | const char *name, bool invoke, | |
1492 | int (*startup)(unsigned int cpu), | |
1493 | int (*teardown)(unsigned int cpu)) | |
1494 | { | |
1495 | int cpu, ret = 0; | |
1496 | int dyn_state = 0; | |
1497 | ||
1498 | if (cpuhp_cb_check(state) || !name) | |
1499 | return -EINVAL; | |
1500 | ||
1501 | get_online_cpus(); | |
1502 | ||
1503 | /* currently assignments for the ONLINE state are possible */ | |
1cf4f629 | 1504 | if (state == CPUHP_AP_ONLINE_DYN) { |
5b7aa87e TG |
1505 | dyn_state = 1; |
1506 | ret = cpuhp_reserve_state(state); | |
1507 | if (ret < 0) | |
1508 | goto out; | |
1509 | state = ret; | |
1510 | } | |
1511 | ||
1512 | cpuhp_store_callbacks(state, name, startup, teardown); | |
1513 | ||
1514 | if (!invoke || !startup) | |
1515 | goto out; | |
1516 | ||
1517 | /* | |
1518 | * Try to call the startup callback for each present cpu | |
1519 | * depending on the hotplug state of the cpu. | |
1520 | */ | |
1521 | for_each_present_cpu(cpu) { | |
1522 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1523 | int cpustate = st->state; | |
1524 | ||
1525 | if (cpustate < state) | |
1526 | continue; | |
1527 | ||
a724632c | 1528 | ret = cpuhp_issue_call(cpu, state, true); |
5b7aa87e | 1529 | if (ret) { |
a724632c TG |
1530 | if (teardown) |
1531 | cpuhp_rollback_install(cpu, state); | |
5b7aa87e TG |
1532 | cpuhp_store_callbacks(state, NULL, NULL, NULL); |
1533 | goto out; | |
1534 | } | |
1535 | } | |
1536 | out: | |
1537 | put_online_cpus(); | |
1538 | if (!ret && dyn_state) | |
1539 | return state; | |
1540 | return ret; | |
1541 | } | |
1542 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1543 | ||
1544 | /** | |
1545 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1546 | * @state: The state to remove | |
1547 | * @invoke: If true, the teardown function is invoked for cpus where | |
1548 | * cpu state >= @state | |
1549 | * | |
1550 | * The teardown callback is currently not allowed to fail. Think | |
1551 | * about module removal! | |
1552 | */ | |
1553 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1554 | { | |
5b7aa87e TG |
1555 | int cpu; |
1556 | ||
1557 | BUG_ON(cpuhp_cb_check(state)); | |
1558 | ||
1559 | get_online_cpus(); | |
1560 | ||
a724632c | 1561 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1562 | goto remove; |
1563 | ||
1564 | /* | |
1565 | * Call the teardown callback for each present cpu depending | |
1566 | * on the hotplug state of the cpu. This function is not | |
1567 | * allowed to fail currently! | |
1568 | */ | |
1569 | for_each_present_cpu(cpu) { | |
1570 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1571 | int cpustate = st->state; | |
1572 | ||
1573 | if (cpustate >= state) | |
a724632c | 1574 | cpuhp_issue_call(cpu, state, false); |
5b7aa87e TG |
1575 | } |
1576 | remove: | |
1577 | cpuhp_store_callbacks(state, NULL, NULL, NULL); | |
1578 | put_online_cpus(); | |
1579 | } | |
1580 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1581 | ||
98f8cdce TG |
1582 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1583 | static ssize_t show_cpuhp_state(struct device *dev, | |
1584 | struct device_attribute *attr, char *buf) | |
1585 | { | |
1586 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1587 | ||
1588 | return sprintf(buf, "%d\n", st->state); | |
1589 | } | |
1590 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1591 | ||
757c989b TG |
1592 | static ssize_t write_cpuhp_target(struct device *dev, |
1593 | struct device_attribute *attr, | |
1594 | const char *buf, size_t count) | |
1595 | { | |
1596 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1597 | struct cpuhp_step *sp; | |
1598 | int target, ret; | |
1599 | ||
1600 | ret = kstrtoint(buf, 10, &target); | |
1601 | if (ret) | |
1602 | return ret; | |
1603 | ||
1604 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1605 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1606 | return -EINVAL; | |
1607 | #else | |
1608 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1609 | return -EINVAL; | |
1610 | #endif | |
1611 | ||
1612 | ret = lock_device_hotplug_sysfs(); | |
1613 | if (ret) | |
1614 | return ret; | |
1615 | ||
1616 | mutex_lock(&cpuhp_state_mutex); | |
1617 | sp = cpuhp_get_step(target); | |
1618 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1619 | mutex_unlock(&cpuhp_state_mutex); | |
1620 | if (ret) | |
1621 | return ret; | |
1622 | ||
1623 | if (st->state < target) | |
1624 | ret = do_cpu_up(dev->id, target); | |
1625 | else | |
1626 | ret = do_cpu_down(dev->id, target); | |
1627 | ||
1628 | unlock_device_hotplug(); | |
1629 | return ret ? ret : count; | |
1630 | } | |
1631 | ||
98f8cdce TG |
1632 | static ssize_t show_cpuhp_target(struct device *dev, |
1633 | struct device_attribute *attr, char *buf) | |
1634 | { | |
1635 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1636 | ||
1637 | return sprintf(buf, "%d\n", st->target); | |
1638 | } | |
757c989b | 1639 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1640 | |
1641 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1642 | &dev_attr_state.attr, | |
1643 | &dev_attr_target.attr, | |
1644 | NULL | |
1645 | }; | |
1646 | ||
1647 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1648 | .attrs = cpuhp_cpu_attrs, | |
1649 | .name = "hotplug", | |
1650 | NULL | |
1651 | }; | |
1652 | ||
1653 | static ssize_t show_cpuhp_states(struct device *dev, | |
1654 | struct device_attribute *attr, char *buf) | |
1655 | { | |
1656 | ssize_t cur, res = 0; | |
1657 | int i; | |
1658 | ||
1659 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1660 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1661 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1662 | ||
1663 | if (sp->name) { | |
1664 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1665 | buf += cur; | |
1666 | res += cur; | |
1667 | } | |
1668 | } | |
1669 | mutex_unlock(&cpuhp_state_mutex); | |
1670 | return res; | |
1671 | } | |
1672 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1673 | ||
1674 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1675 | &dev_attr_states.attr, | |
1676 | NULL | |
1677 | }; | |
1678 | ||
1679 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1680 | .attrs = cpuhp_cpu_root_attrs, | |
1681 | .name = "hotplug", | |
1682 | NULL | |
1683 | }; | |
1684 | ||
1685 | static int __init cpuhp_sysfs_init(void) | |
1686 | { | |
1687 | int cpu, ret; | |
1688 | ||
1689 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1690 | &cpuhp_cpu_root_attr_group); | |
1691 | if (ret) | |
1692 | return ret; | |
1693 | ||
1694 | for_each_possible_cpu(cpu) { | |
1695 | struct device *dev = get_cpu_device(cpu); | |
1696 | ||
1697 | if (!dev) | |
1698 | continue; | |
1699 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1700 | if (ret) | |
1701 | return ret; | |
1702 | } | |
1703 | return 0; | |
1704 | } | |
1705 | device_initcall(cpuhp_sysfs_init); | |
1706 | #endif | |
1707 | ||
e56b3bc7 LT |
1708 | /* |
1709 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1710 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1711 | * | |
e0b582ec | 1712 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1713 | * mask value that has a single bit set only. |
1714 | */ | |
b8d317d1 | 1715 | |
e56b3bc7 | 1716 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1717 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1718 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1719 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1720 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1721 | |
e56b3bc7 LT |
1722 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1723 | ||
1724 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1725 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1726 | #if BITS_PER_LONG > 32 | |
1727 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1728 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1729 | #endif |
1730 | }; | |
e56b3bc7 | 1731 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1732 | |
1733 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1734 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1735 | |
1736 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1737 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1738 | = {CPU_BITS_ALL}; |
b3199c02 | 1739 | #else |
4b804c85 | 1740 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1741 | #endif |
4b804c85 | 1742 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1743 | |
4b804c85 RV |
1744 | struct cpumask __cpu_online_mask __read_mostly; |
1745 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1746 | |
4b804c85 RV |
1747 | struct cpumask __cpu_present_mask __read_mostly; |
1748 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1749 | |
4b804c85 RV |
1750 | struct cpumask __cpu_active_mask __read_mostly; |
1751 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1752 | |
3fa41520 RR |
1753 | void init_cpu_present(const struct cpumask *src) |
1754 | { | |
c4c54dd1 | 1755 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1756 | } |
1757 | ||
1758 | void init_cpu_possible(const struct cpumask *src) | |
1759 | { | |
c4c54dd1 | 1760 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1761 | } |
1762 | ||
1763 | void init_cpu_online(const struct cpumask *src) | |
1764 | { | |
c4c54dd1 | 1765 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1766 | } |
cff7d378 TG |
1767 | |
1768 | /* | |
1769 | * Activate the first processor. | |
1770 | */ | |
1771 | void __init boot_cpu_init(void) | |
1772 | { | |
1773 | int cpu = smp_processor_id(); | |
1774 | ||
1775 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1776 | set_cpu_online(cpu, true); | |
1777 | set_cpu_active(cpu, true); | |
1778 | set_cpu_present(cpu, true); | |
1779 | set_cpu_possible(cpu, true); | |
1780 | } | |
1781 | ||
1782 | /* | |
1783 | * Must be called _AFTER_ setting up the per_cpu areas | |
1784 | */ | |
1785 | void __init boot_cpu_state_init(void) | |
1786 | { | |
1787 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1788 | } |