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