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> | |
3f07c014 | 10 | #include <linux/sched/signal.h> |
ef8bd77f | 11 | #include <linux/sched/hotplug.h> |
1da177e4 LT |
12 | #include <linux/unistd.h> |
13 | #include <linux/cpu.h> | |
cb79295e AV |
14 | #include <linux/oom.h> |
15 | #include <linux/rcupdate.h> | |
9984de1a | 16 | #include <linux/export.h> |
e4cc2f87 | 17 | #include <linux/bug.h> |
1da177e4 LT |
18 | #include <linux/kthread.h> |
19 | #include <linux/stop_machine.h> | |
81615b62 | 20 | #include <linux/mutex.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
79cfbdfa | 22 | #include <linux/suspend.h> |
a19423b9 | 23 | #include <linux/lockdep.h> |
345527b1 | 24 | #include <linux/tick.h> |
a8994181 | 25 | #include <linux/irq.h> |
4cb28ced | 26 | #include <linux/smpboot.h> |
e6d4989a | 27 | #include <linux/relay.h> |
6731d4f1 | 28 | #include <linux/slab.h> |
cff7d378 | 29 | |
bb3632c6 | 30 | #include <trace/events/power.h> |
cff7d378 TG |
31 | #define CREATE_TRACE_POINTS |
32 | #include <trace/events/cpuhp.h> | |
1da177e4 | 33 | |
38498a67 TG |
34 | #include "smpboot.h" |
35 | ||
cff7d378 TG |
36 | /** |
37 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
38 | * @state: The current cpu state | |
39 | * @target: The target state | |
4cb28ced TG |
40 | * @thread: Pointer to the hotplug thread |
41 | * @should_run: Thread should execute | |
3b9d6da6 | 42 | * @rollback: Perform a rollback |
a724632c TG |
43 | * @single: Single callback invocation |
44 | * @bringup: Single callback bringup or teardown selector | |
45 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced TG |
46 | * @result: Result of the operation |
47 | * @done: Signal completion to the issuer of the task | |
cff7d378 TG |
48 | */ |
49 | struct cpuhp_cpu_state { | |
50 | enum cpuhp_state state; | |
51 | enum cpuhp_state target; | |
4cb28ced TG |
52 | #ifdef CONFIG_SMP |
53 | struct task_struct *thread; | |
54 | bool should_run; | |
3b9d6da6 | 55 | bool rollback; |
a724632c TG |
56 | bool single; |
57 | bool bringup; | |
cf392d10 | 58 | struct hlist_node *node; |
4cb28ced | 59 | enum cpuhp_state cb_state; |
4cb28ced TG |
60 | int result; |
61 | struct completion done; | |
62 | #endif | |
cff7d378 TG |
63 | }; |
64 | ||
65 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
66 | ||
67 | /** | |
68 | * cpuhp_step - Hotplug state machine step | |
69 | * @name: Name of the step | |
70 | * @startup: Startup function of the step | |
71 | * @teardown: Teardown function of the step | |
72 | * @skip_onerr: Do not invoke the functions on error rollback | |
73 | * Will go away once the notifiers are gone | |
757c989b | 74 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
75 | */ |
76 | struct cpuhp_step { | |
cf392d10 TG |
77 | const char *name; |
78 | union { | |
3c1627e9 TG |
79 | int (*single)(unsigned int cpu); |
80 | int (*multi)(unsigned int cpu, | |
81 | struct hlist_node *node); | |
82 | } startup; | |
cf392d10 | 83 | union { |
3c1627e9 TG |
84 | int (*single)(unsigned int cpu); |
85 | int (*multi)(unsigned int cpu, | |
86 | struct hlist_node *node); | |
87 | } teardown; | |
cf392d10 TG |
88 | struct hlist_head list; |
89 | bool skip_onerr; | |
90 | bool cant_stop; | |
91 | bool multi_instance; | |
cff7d378 TG |
92 | }; |
93 | ||
98f8cdce | 94 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 95 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 96 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 97 | |
a724632c TG |
98 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
99 | { | |
100 | /* | |
101 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
102 | * purposes as that state is handled explicitly in cpu_down. | |
103 | */ | |
104 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
105 | } | |
106 | ||
107 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
108 | { | |
109 | struct cpuhp_step *sp; | |
110 | ||
111 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
112 | return sp + state; | |
113 | } | |
114 | ||
cff7d378 TG |
115 | /** |
116 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
117 | * @cpu: The cpu for which the callback should be invoked | |
118 | * @step: The step in the state machine | |
a724632c | 119 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 120 | * |
cf392d10 | 121 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 122 | */ |
a724632c | 123 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 124 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
125 | { |
126 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 127 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
128 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
129 | int (*cb)(unsigned int cpu); | |
130 | int ret, cnt; | |
131 | ||
132 | if (!step->multi_instance) { | |
3c1627e9 | 133 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
134 | if (!cb) |
135 | return 0; | |
a724632c | 136 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 137 | ret = cb(cpu); |
a724632c | 138 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
139 | return ret; |
140 | } | |
3c1627e9 | 141 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
142 | if (!cbm) |
143 | return 0; | |
144 | ||
145 | /* Single invocation for instance add/remove */ | |
146 | if (node) { | |
147 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
148 | ret = cbm(cpu, node); | |
149 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
150 | return ret; | |
151 | } | |
152 | ||
153 | /* State transition. Invoke on all instances */ | |
154 | cnt = 0; | |
155 | hlist_for_each(node, &step->list) { | |
156 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
157 | ret = cbm(cpu, node); | |
158 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
159 | if (ret) | |
160 | goto err; | |
161 | cnt++; | |
162 | } | |
163 | return 0; | |
164 | err: | |
165 | /* Rollback the instances if one failed */ | |
3c1627e9 | 166 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
167 | if (!cbm) |
168 | return ret; | |
169 | ||
170 | hlist_for_each(node, &step->list) { | |
171 | if (!cnt--) | |
172 | break; | |
173 | cbm(cpu, node); | |
cff7d378 TG |
174 | } |
175 | return ret; | |
176 | } | |
177 | ||
98a79d6a | 178 | #ifdef CONFIG_SMP |
b3199c02 | 179 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 180 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
181 | bool cpuhp_tasks_frozen; |
182 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 183 | |
79a6cdeb | 184 | /* |
93ae4f97 SB |
185 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
186 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
187 | */ |
188 | void cpu_maps_update_begin(void) | |
189 | { | |
190 | mutex_lock(&cpu_add_remove_lock); | |
191 | } | |
192 | ||
193 | void cpu_maps_update_done(void) | |
194 | { | |
195 | mutex_unlock(&cpu_add_remove_lock); | |
196 | } | |
1da177e4 | 197 | |
e3920fb4 RW |
198 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
199 | * Should always be manipulated under cpu_add_remove_lock | |
200 | */ | |
201 | static int cpu_hotplug_disabled; | |
202 | ||
79a6cdeb LJ |
203 | #ifdef CONFIG_HOTPLUG_CPU |
204 | ||
d221938c GS |
205 | static struct { |
206 | struct task_struct *active_writer; | |
87af9e7f DH |
207 | /* wait queue to wake up the active_writer */ |
208 | wait_queue_head_t wq; | |
209 | /* verifies that no writer will get active while readers are active */ | |
210 | struct mutex lock; | |
d221938c GS |
211 | /* |
212 | * Also blocks the new readers during | |
213 | * an ongoing cpu hotplug operation. | |
214 | */ | |
87af9e7f | 215 | atomic_t refcount; |
a19423b9 GS |
216 | |
217 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
218 | struct lockdep_map dep_map; | |
219 | #endif | |
31950eb6 LT |
220 | } cpu_hotplug = { |
221 | .active_writer = NULL, | |
87af9e7f | 222 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 223 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 | 224 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
a705e07b | 225 | .dep_map = STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug.dep_map), |
a19423b9 | 226 | #endif |
31950eb6 | 227 | }; |
d221938c | 228 | |
a19423b9 GS |
229 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
230 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
231 | #define cpuhp_lock_acquire_tryread() \ |
232 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
233 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
234 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
235 | ||
62db99f4 | 236 | |
86ef5c9a | 237 | void get_online_cpus(void) |
a9d9baa1 | 238 | { |
d221938c GS |
239 | might_sleep(); |
240 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 241 | return; |
a19423b9 | 242 | cpuhp_lock_acquire_read(); |
d221938c | 243 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 244 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 245 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 246 | } |
86ef5c9a | 247 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 248 | |
86ef5c9a | 249 | void put_online_cpus(void) |
a9d9baa1 | 250 | { |
87af9e7f DH |
251 | int refcount; |
252 | ||
d221938c | 253 | if (cpu_hotplug.active_writer == current) |
aa953877 | 254 | return; |
075663d1 | 255 | |
87af9e7f DH |
256 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
257 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
258 | atomic_inc(&cpu_hotplug.refcount); | |
259 | ||
260 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
261 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 262 | |
a19423b9 | 263 | cpuhp_lock_release(); |
d221938c | 264 | |
a9d9baa1 | 265 | } |
86ef5c9a | 266 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 267 | |
d221938c GS |
268 | /* |
269 | * This ensures that the hotplug operation can begin only when the | |
270 | * refcount goes to zero. | |
271 | * | |
272 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
273 | * will be blocked by the cpu_hotplug.lock | |
274 | * | |
d2ba7e2a ON |
275 | * Since cpu_hotplug_begin() is always called after invoking |
276 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
277 | * |
278 | * Note that theoretically, there is a possibility of a livelock: | |
279 | * - Refcount goes to zero, last reader wakes up the sleeping | |
280 | * writer. | |
281 | * - Last reader unlocks the cpu_hotplug.lock. | |
282 | * - A new reader arrives at this moment, bumps up the refcount. | |
283 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
284 | * non zero and goes to sleep again. | |
285 | * | |
286 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 287 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
288 | * |
289 | */ | |
b9d10be7 | 290 | void cpu_hotplug_begin(void) |
d221938c | 291 | { |
87af9e7f | 292 | DEFINE_WAIT(wait); |
d2ba7e2a | 293 | |
87af9e7f | 294 | cpu_hotplug.active_writer = current; |
a19423b9 | 295 | cpuhp_lock_acquire(); |
87af9e7f | 296 | |
d2ba7e2a ON |
297 | for (;;) { |
298 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
299 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
300 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
301 | break; | |
d221938c GS |
302 | mutex_unlock(&cpu_hotplug.lock); |
303 | schedule(); | |
d221938c | 304 | } |
87af9e7f | 305 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
306 | } |
307 | ||
b9d10be7 | 308 | void cpu_hotplug_done(void) |
d221938c GS |
309 | { |
310 | cpu_hotplug.active_writer = NULL; | |
311 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 312 | cpuhp_lock_release(); |
d221938c | 313 | } |
79a6cdeb | 314 | |
16e53dbf SB |
315 | /* |
316 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
317 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
318 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
319 | * hotplug path before performing hotplug operations. So acquiring that lock | |
320 | * guarantees mutual exclusion from any currently running hotplug operations. | |
321 | */ | |
322 | void cpu_hotplug_disable(void) | |
323 | { | |
324 | cpu_maps_update_begin(); | |
89af7ba5 | 325 | cpu_hotplug_disabled++; |
16e53dbf SB |
326 | cpu_maps_update_done(); |
327 | } | |
32145c46 | 328 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 329 | |
01b41159 LW |
330 | static void __cpu_hotplug_enable(void) |
331 | { | |
332 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
333 | return; | |
334 | cpu_hotplug_disabled--; | |
335 | } | |
336 | ||
16e53dbf SB |
337 | void cpu_hotplug_enable(void) |
338 | { | |
339 | cpu_maps_update_begin(); | |
01b41159 | 340 | __cpu_hotplug_enable(); |
16e53dbf SB |
341 | cpu_maps_update_done(); |
342 | } | |
32145c46 | 343 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 344 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 345 | |
ba997462 | 346 | /* Notifier wrappers for transitioning to state machine */ |
ba997462 | 347 | |
8df3e07e TG |
348 | static int bringup_wait_for_ap(unsigned int cpu) |
349 | { | |
350 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
351 | ||
352 | wait_for_completion(&st->done); | |
353 | return st->result; | |
354 | } | |
355 | ||
ba997462 TG |
356 | static int bringup_cpu(unsigned int cpu) |
357 | { | |
358 | struct task_struct *idle = idle_thread_get(cpu); | |
359 | int ret; | |
360 | ||
aa877175 BO |
361 | /* |
362 | * Some architectures have to walk the irq descriptors to | |
363 | * setup the vector space for the cpu which comes online. | |
364 | * Prevent irq alloc/free across the bringup. | |
365 | */ | |
366 | irq_lock_sparse(); | |
367 | ||
ba997462 TG |
368 | /* Arch-specific enabling code. */ |
369 | ret = __cpu_up(cpu, idle); | |
aa877175 | 370 | irq_unlock_sparse(); |
530e9b76 | 371 | if (ret) |
ba997462 | 372 | return ret; |
8df3e07e | 373 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 374 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 375 | return ret; |
ba997462 TG |
376 | } |
377 | ||
2e1a3483 TG |
378 | /* |
379 | * Hotplug state machine related functions | |
380 | */ | |
a724632c | 381 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
382 | { |
383 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 384 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
385 | |
386 | if (!step->skip_onerr) | |
cf392d10 | 387 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
388 | } |
389 | } | |
390 | ||
391 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 392 | enum cpuhp_state target) |
2e1a3483 TG |
393 | { |
394 | enum cpuhp_state prev_state = st->state; | |
395 | int ret = 0; | |
396 | ||
397 | for (; st->state > target; st->state--) { | |
cf392d10 | 398 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
399 | if (ret) { |
400 | st->target = prev_state; | |
a724632c | 401 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
402 | break; |
403 | } | |
404 | } | |
405 | return ret; | |
406 | } | |
407 | ||
a724632c | 408 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
409 | { |
410 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 411 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
412 | |
413 | if (!step->skip_onerr) | |
cf392d10 | 414 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
415 | } |
416 | } | |
417 | ||
418 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 419 | enum cpuhp_state target) |
2e1a3483 TG |
420 | { |
421 | enum cpuhp_state prev_state = st->state; | |
422 | int ret = 0; | |
423 | ||
424 | while (st->state < target) { | |
2e1a3483 | 425 | st->state++; |
cf392d10 | 426 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
427 | if (ret) { |
428 | st->target = prev_state; | |
a724632c | 429 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
430 | break; |
431 | } | |
432 | } | |
433 | return ret; | |
434 | } | |
435 | ||
4cb28ced TG |
436 | /* |
437 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
438 | */ | |
439 | static void cpuhp_create(unsigned int cpu) | |
440 | { | |
441 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
442 | ||
443 | init_completion(&st->done); | |
444 | } | |
445 | ||
446 | static int cpuhp_should_run(unsigned int cpu) | |
447 | { | |
448 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
449 | ||
450 | return st->should_run; | |
451 | } | |
452 | ||
453 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
454 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
455 | { | |
1cf4f629 | 456 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 457 | |
a724632c | 458 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
459 | } |
460 | ||
461 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
462 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
463 | { | |
a724632c | 464 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
465 | } |
466 | ||
467 | /* | |
468 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
469 | * callbacks when a state gets [un]installed at runtime. | |
470 | */ | |
471 | static void cpuhp_thread_fun(unsigned int cpu) | |
472 | { | |
473 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
474 | int ret = 0; | |
475 | ||
476 | /* | |
477 | * Paired with the mb() in cpuhp_kick_ap_work and | |
478 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
479 | */ | |
480 | smp_mb(); | |
481 | if (!st->should_run) | |
482 | return; | |
483 | ||
484 | st->should_run = false; | |
485 | ||
486 | /* Single callback invocation for [un]install ? */ | |
a724632c | 487 | if (st->single) { |
4cb28ced TG |
488 | if (st->cb_state < CPUHP_AP_ONLINE) { |
489 | local_irq_disable(); | |
a724632c | 490 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 491 | st->bringup, st->node); |
4cb28ced TG |
492 | local_irq_enable(); |
493 | } else { | |
a724632c | 494 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 495 | st->bringup, st->node); |
4cb28ced | 496 | } |
3b9d6da6 SAS |
497 | } else if (st->rollback) { |
498 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
499 | ||
a724632c | 500 | undo_cpu_down(cpu, st); |
3b9d6da6 | 501 | st->rollback = false; |
4cb28ced | 502 | } else { |
1cf4f629 | 503 | /* Cannot happen .... */ |
8df3e07e | 504 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 505 | |
4cb28ced TG |
506 | /* Regular hotplug work */ |
507 | if (st->state < st->target) | |
508 | ret = cpuhp_ap_online(cpu, st); | |
509 | else if (st->state > st->target) | |
510 | ret = cpuhp_ap_offline(cpu, st); | |
511 | } | |
512 | st->result = ret; | |
513 | complete(&st->done); | |
514 | } | |
515 | ||
516 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 517 | static int |
cf392d10 TG |
518 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
519 | struct hlist_node *node) | |
4cb28ced TG |
520 | { |
521 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
522 | ||
523 | if (!cpu_online(cpu)) | |
524 | return 0; | |
525 | ||
6a4e2451 TG |
526 | /* |
527 | * If we are up and running, use the hotplug thread. For early calls | |
528 | * we invoke the thread function directly. | |
529 | */ | |
530 | if (!st->thread) | |
cf392d10 | 531 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 532 | |
4cb28ced | 533 | st->cb_state = state; |
a724632c TG |
534 | st->single = true; |
535 | st->bringup = bringup; | |
cf392d10 | 536 | st->node = node; |
a724632c | 537 | |
4cb28ced TG |
538 | /* |
539 | * Make sure the above stores are visible before should_run becomes | |
540 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
541 | */ | |
542 | smp_mb(); | |
543 | st->should_run = true; | |
544 | wake_up_process(st->thread); | |
545 | wait_for_completion(&st->done); | |
546 | return st->result; | |
547 | } | |
548 | ||
549 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 550 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 551 | { |
4cb28ced | 552 | st->result = 0; |
a724632c | 553 | st->single = false; |
4cb28ced TG |
554 | /* |
555 | * Make sure the above stores are visible before should_run becomes | |
556 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
557 | */ | |
558 | smp_mb(); | |
559 | st->should_run = true; | |
560 | wake_up_process(st->thread); | |
1cf4f629 TG |
561 | } |
562 | ||
563 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
564 | { | |
565 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
566 | enum cpuhp_state state = st->state; | |
567 | ||
568 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
569 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
570 | wait_for_completion(&st->done); |
571 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
572 | return st->result; | |
573 | } | |
574 | ||
575 | static struct smp_hotplug_thread cpuhp_threads = { | |
576 | .store = &cpuhp_state.thread, | |
577 | .create = &cpuhp_create, | |
578 | .thread_should_run = cpuhp_should_run, | |
579 | .thread_fn = cpuhp_thread_fun, | |
580 | .thread_comm = "cpuhp/%u", | |
581 | .selfparking = true, | |
582 | }; | |
583 | ||
584 | void __init cpuhp_threads_init(void) | |
585 | { | |
586 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
587 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
588 | } | |
589 | ||
777c6e0d | 590 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
591 | /** |
592 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
593 | * @cpu: a CPU id | |
594 | * | |
595 | * This function walks all processes, finds a valid mm struct for each one and | |
596 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
597 | * trivial, there are various non-obvious corner cases, which this function | |
598 | * tries to solve in a safe manner. | |
599 | * | |
600 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
601 | * be called only for an already offlined CPU. | |
602 | */ | |
cb79295e AV |
603 | void clear_tasks_mm_cpumask(int cpu) |
604 | { | |
605 | struct task_struct *p; | |
606 | ||
607 | /* | |
608 | * This function is called after the cpu is taken down and marked | |
609 | * offline, so its not like new tasks will ever get this cpu set in | |
610 | * their mm mask. -- Peter Zijlstra | |
611 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
612 | * full-fledged tasklist_lock. | |
613 | */ | |
e4cc2f87 | 614 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
615 | rcu_read_lock(); |
616 | for_each_process(p) { | |
617 | struct task_struct *t; | |
618 | ||
e4cc2f87 AV |
619 | /* |
620 | * Main thread might exit, but other threads may still have | |
621 | * a valid mm. Find one. | |
622 | */ | |
cb79295e AV |
623 | t = find_lock_task_mm(p); |
624 | if (!t) | |
625 | continue; | |
626 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
627 | task_unlock(t); | |
628 | } | |
629 | rcu_read_unlock(); | |
630 | } | |
631 | ||
b728ca06 | 632 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 633 | { |
b728ca06 | 634 | struct task_struct *g, *p; |
1da177e4 | 635 | |
a75a6068 ON |
636 | read_lock(&tasklist_lock); |
637 | for_each_process_thread(g, p) { | |
b728ca06 KT |
638 | if (!p->on_rq) |
639 | continue; | |
640 | /* | |
641 | * We do the check with unlocked task_rq(p)->lock. | |
642 | * Order the reading to do not warn about a task, | |
643 | * which was running on this cpu in the past, and | |
644 | * it's just been woken on another cpu. | |
645 | */ | |
646 | rmb(); | |
647 | if (task_cpu(p) != dead_cpu) | |
648 | continue; | |
649 | ||
650 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
651 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
652 | } |
653 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
654 | } |
655 | ||
656 | /* Take this CPU down. */ | |
71cf5aee | 657 | static int take_cpu_down(void *_param) |
1da177e4 | 658 | { |
4baa0afc TG |
659 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
660 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 661 | int err, cpu = smp_processor_id(); |
1da177e4 | 662 | |
1da177e4 LT |
663 | /* Ensure this CPU doesn't handle any more interrupts. */ |
664 | err = __cpu_disable(); | |
665 | if (err < 0) | |
f3705136 | 666 | return err; |
1da177e4 | 667 | |
a724632c TG |
668 | /* |
669 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
670 | * do this step again. | |
671 | */ | |
672 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
673 | st->state--; | |
4baa0afc | 674 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 675 | for (; st->state > target; st->state--) |
cf392d10 | 676 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 677 | |
52c063d1 TG |
678 | /* Give up timekeeping duties */ |
679 | tick_handover_do_timer(); | |
14e568e7 | 680 | /* Park the stopper thread */ |
090e77c3 | 681 | stop_machine_park(cpu); |
f3705136 | 682 | return 0; |
1da177e4 LT |
683 | } |
684 | ||
98458172 | 685 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 686 | { |
e69aab13 | 687 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 688 | int err; |
1da177e4 | 689 | |
2a58c527 | 690 | /* Park the smpboot threads */ |
1cf4f629 | 691 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 692 | smpboot_park_threads(cpu); |
1cf4f629 | 693 | |
6acce3ef | 694 | /* |
a8994181 TG |
695 | * Prevent irq alloc/free while the dying cpu reorganizes the |
696 | * interrupt affinities. | |
6acce3ef | 697 | */ |
a8994181 | 698 | irq_lock_sparse(); |
6acce3ef | 699 | |
a8994181 TG |
700 | /* |
701 | * So now all preempt/rcu users must observe !cpu_active(). | |
702 | */ | |
090e77c3 | 703 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 704 | if (err) { |
3b9d6da6 | 705 | /* CPU refused to die */ |
a8994181 | 706 | irq_unlock_sparse(); |
3b9d6da6 SAS |
707 | /* Unpark the hotplug thread so we can rollback there */ |
708 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 709 | return err; |
8fa1d7d3 | 710 | } |
04321587 | 711 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 712 | |
48c5ccae | 713 | /* |
ee1e714b | 714 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
715 | * runnable tasks from the cpu, there's only the idle task left now |
716 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
717 | * |
718 | * Wait for the stop thread to go away. | |
48c5ccae | 719 | */ |
e69aab13 TG |
720 | wait_for_completion(&st->done); |
721 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 722 | |
a8994181 TG |
723 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
724 | irq_unlock_sparse(); | |
725 | ||
345527b1 | 726 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
727 | /* This actually kills the CPU. */ |
728 | __cpu_die(cpu); | |
729 | ||
a49b116d | 730 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
731 | return 0; |
732 | } | |
1da177e4 | 733 | |
71f87b2f TG |
734 | static void cpuhp_complete_idle_dead(void *arg) |
735 | { | |
736 | struct cpuhp_cpu_state *st = arg; | |
737 | ||
738 | complete(&st->done); | |
739 | } | |
740 | ||
e69aab13 TG |
741 | void cpuhp_report_idle_dead(void) |
742 | { | |
743 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
744 | ||
745 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 746 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
747 | st->state = CPUHP_AP_IDLE_DEAD; |
748 | /* | |
749 | * We cannot call complete after rcu_report_dead() so we delegate it | |
750 | * to an online cpu. | |
751 | */ | |
752 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
753 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
754 | } |
755 | ||
cff7d378 | 756 | #else |
cff7d378 | 757 | #define takedown_cpu NULL |
cff7d378 TG |
758 | #endif |
759 | ||
760 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 761 | |
98458172 | 762 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
763 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
764 | enum cpuhp_state target) | |
98458172 | 765 | { |
cff7d378 TG |
766 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
767 | int prev_state, ret = 0; | |
98458172 TG |
768 | |
769 | if (num_online_cpus() == 1) | |
770 | return -EBUSY; | |
771 | ||
757c989b | 772 | if (!cpu_present(cpu)) |
98458172 TG |
773 | return -EINVAL; |
774 | ||
775 | cpu_hotplug_begin(); | |
776 | ||
777 | cpuhp_tasks_frozen = tasks_frozen; | |
778 | ||
cff7d378 | 779 | prev_state = st->state; |
af1f4045 | 780 | st->target = target; |
1cf4f629 TG |
781 | /* |
782 | * If the current CPU state is in the range of the AP hotplug thread, | |
783 | * then we need to kick the thread. | |
784 | */ | |
8df3e07e | 785 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
786 | ret = cpuhp_kick_ap_work(cpu); |
787 | /* | |
788 | * The AP side has done the error rollback already. Just | |
789 | * return the error code.. | |
790 | */ | |
791 | if (ret) | |
792 | goto out; | |
793 | ||
794 | /* | |
795 | * We might have stopped still in the range of the AP hotplug | |
796 | * thread. Nothing to do anymore. | |
797 | */ | |
8df3e07e | 798 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
799 | goto out; |
800 | } | |
801 | /* | |
8df3e07e | 802 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
803 | * to do the further cleanups. |
804 | */ | |
a724632c | 805 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
806 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
807 | st->target = prev_state; | |
808 | st->rollback = true; | |
809 | cpuhp_kick_ap_work(cpu); | |
810 | } | |
98458172 | 811 | |
1cf4f629 | 812 | out: |
d221938c | 813 | cpu_hotplug_done(); |
cff7d378 | 814 | return ret; |
e3920fb4 RW |
815 | } |
816 | ||
af1f4045 | 817 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 818 | { |
9ea09af3 | 819 | int err; |
e3920fb4 | 820 | |
d221938c | 821 | cpu_maps_update_begin(); |
e761b772 MK |
822 | |
823 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 824 | err = -EBUSY; |
e761b772 MK |
825 | goto out; |
826 | } | |
827 | ||
af1f4045 | 828 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 829 | |
e761b772 | 830 | out: |
d221938c | 831 | cpu_maps_update_done(); |
1da177e4 LT |
832 | return err; |
833 | } | |
af1f4045 TG |
834 | int cpu_down(unsigned int cpu) |
835 | { | |
836 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
837 | } | |
b62b8ef9 | 838 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
839 | #endif /*CONFIG_HOTPLUG_CPU*/ |
840 | ||
4baa0afc | 841 | /** |
ee1e714b | 842 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
843 | * @cpu: cpu that just started |
844 | * | |
4baa0afc TG |
845 | * It must be called by the arch code on the new cpu, before the new cpu |
846 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
847 | */ | |
848 | void notify_cpu_starting(unsigned int cpu) | |
849 | { | |
850 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
851 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
852 | ||
0c6d4576 | 853 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 854 | while (st->state < target) { |
4baa0afc | 855 | st->state++; |
cf392d10 | 856 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
857 | } |
858 | } | |
859 | ||
949338e3 TG |
860 | /* |
861 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
862 | * the stopper thread and unpark the smpboot threads. If the target state is |
863 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
864 | * cpu further. | |
949338e3 | 865 | */ |
8df3e07e | 866 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 867 | { |
8df3e07e TG |
868 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
869 | unsigned int cpu = smp_processor_id(); | |
870 | ||
871 | /* Happens for the boot cpu */ | |
872 | if (state != CPUHP_AP_ONLINE_IDLE) | |
873 | return; | |
874 | ||
875 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 876 | |
8df3e07e | 877 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 878 | stop_machine_unpark(cpu); |
1cf4f629 | 879 | kthread_unpark(st->thread); |
8df3e07e TG |
880 | |
881 | /* Should we go further up ? */ | |
882 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
883 | __cpuhp_kick_ap_work(st); | |
884 | else | |
885 | complete(&st->done); | |
949338e3 TG |
886 | } |
887 | ||
e3920fb4 | 888 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 889 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 890 | { |
cff7d378 | 891 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 892 | struct task_struct *idle; |
2e1a3483 | 893 | int ret = 0; |
1da177e4 | 894 | |
d221938c | 895 | cpu_hotplug_begin(); |
38498a67 | 896 | |
757c989b | 897 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
898 | ret = -EINVAL; |
899 | goto out; | |
900 | } | |
901 | ||
757c989b TG |
902 | /* |
903 | * The caller of do_cpu_up might have raced with another | |
904 | * caller. Ignore it for now. | |
905 | */ | |
906 | if (st->state >= target) | |
38498a67 | 907 | goto out; |
757c989b TG |
908 | |
909 | if (st->state == CPUHP_OFFLINE) { | |
910 | /* Let it fail before we try to bring the cpu up */ | |
911 | idle = idle_thread_get(cpu); | |
912 | if (IS_ERR(idle)) { | |
913 | ret = PTR_ERR(idle); | |
914 | goto out; | |
915 | } | |
3bb5d2ee | 916 | } |
38498a67 | 917 | |
ba997462 TG |
918 | cpuhp_tasks_frozen = tasks_frozen; |
919 | ||
af1f4045 | 920 | st->target = target; |
1cf4f629 TG |
921 | /* |
922 | * If the current CPU state is in the range of the AP hotplug thread, | |
923 | * then we need to kick the thread once more. | |
924 | */ | |
8df3e07e | 925 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
926 | ret = cpuhp_kick_ap_work(cpu); |
927 | /* | |
928 | * The AP side has done the error rollback already. Just | |
929 | * return the error code.. | |
930 | */ | |
931 | if (ret) | |
932 | goto out; | |
933 | } | |
934 | ||
935 | /* | |
936 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 937 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
938 | * responsible for bringing it up to the target state. |
939 | */ | |
8df3e07e | 940 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 941 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 942 | out: |
d221938c | 943 | cpu_hotplug_done(); |
e3920fb4 RW |
944 | return ret; |
945 | } | |
946 | ||
af1f4045 | 947 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
948 | { |
949 | int err = 0; | |
cf23422b | 950 | |
e0b582ec | 951 | if (!cpu_possible(cpu)) { |
84117da5 FF |
952 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
953 | cpu); | |
87d5e023 | 954 | #if defined(CONFIG_IA64) |
84117da5 | 955 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
956 | #endif |
957 | return -EINVAL; | |
958 | } | |
e3920fb4 | 959 | |
01b0f197 TK |
960 | err = try_online_node(cpu_to_node(cpu)); |
961 | if (err) | |
962 | return err; | |
cf23422b | 963 | |
d221938c | 964 | cpu_maps_update_begin(); |
e761b772 MK |
965 | |
966 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 967 | err = -EBUSY; |
e761b772 MK |
968 | goto out; |
969 | } | |
970 | ||
af1f4045 | 971 | err = _cpu_up(cpu, 0, target); |
e761b772 | 972 | out: |
d221938c | 973 | cpu_maps_update_done(); |
e3920fb4 RW |
974 | return err; |
975 | } | |
af1f4045 TG |
976 | |
977 | int cpu_up(unsigned int cpu) | |
978 | { | |
979 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
980 | } | |
a513f6ba | 981 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 982 | |
f3de4be9 | 983 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 984 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 985 | |
d391e552 | 986 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 987 | { |
d391e552 | 988 | int cpu, error = 0; |
e3920fb4 | 989 | |
d221938c | 990 | cpu_maps_update_begin(); |
d391e552 JM |
991 | if (!cpu_online(primary)) |
992 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
993 | /* |
994 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
995 | * with the userspace trying to use the CPU hotplug at the same time |
996 | */ | |
e0b582ec | 997 | cpumask_clear(frozen_cpus); |
6ad4c188 | 998 | |
84117da5 | 999 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1000 | for_each_online_cpu(cpu) { |
d391e552 | 1001 | if (cpu == primary) |
e3920fb4 | 1002 | continue; |
bb3632c6 | 1003 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1004 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1005 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1006 | if (!error) |
e0b582ec | 1007 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1008 | else { |
84117da5 | 1009 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1010 | break; |
1011 | } | |
1012 | } | |
86886e55 | 1013 | |
89af7ba5 | 1014 | if (!error) |
e3920fb4 | 1015 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1016 | else |
84117da5 | 1017 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1018 | |
1019 | /* | |
1020 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1021 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1022 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1023 | */ | |
1024 | cpu_hotplug_disabled++; | |
1025 | ||
d221938c | 1026 | cpu_maps_update_done(); |
e3920fb4 RW |
1027 | return error; |
1028 | } | |
1029 | ||
d0af9eed SS |
1030 | void __weak arch_enable_nonboot_cpus_begin(void) |
1031 | { | |
1032 | } | |
1033 | ||
1034 | void __weak arch_enable_nonboot_cpus_end(void) | |
1035 | { | |
1036 | } | |
1037 | ||
71cf5aee | 1038 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1039 | { |
1040 | int cpu, error; | |
1041 | ||
1042 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1043 | cpu_maps_update_begin(); |
01b41159 | 1044 | __cpu_hotplug_enable(); |
e0b582ec | 1045 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1046 | goto out; |
e3920fb4 | 1047 | |
84117da5 | 1048 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1049 | |
1050 | arch_enable_nonboot_cpus_begin(); | |
1051 | ||
e0b582ec | 1052 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1053 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1054 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1055 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1056 | if (!error) { |
84117da5 | 1057 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1058 | continue; |
1059 | } | |
84117da5 | 1060 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1061 | } |
d0af9eed SS |
1062 | |
1063 | arch_enable_nonboot_cpus_end(); | |
1064 | ||
e0b582ec | 1065 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1066 | out: |
d221938c | 1067 | cpu_maps_update_done(); |
1da177e4 | 1068 | } |
e0b582ec | 1069 | |
d7268a31 | 1070 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1071 | { |
1072 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1073 | return -ENOMEM; | |
1074 | return 0; | |
1075 | } | |
1076 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1077 | |
79cfbdfa SB |
1078 | /* |
1079 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1080 | * ensure that the state of the system with respect to the tasks being frozen | |
1081 | * or not, as reported by the notification, remains unchanged *throughout the | |
1082 | * duration* of the execution of the callbacks. | |
1083 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1084 | * | |
1085 | * This synchronization is implemented by mutually excluding regular CPU | |
1086 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1087 | * Hibernate notifications. | |
1088 | */ | |
1089 | static int | |
1090 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1091 | unsigned long action, void *ptr) | |
1092 | { | |
1093 | switch (action) { | |
1094 | ||
1095 | case PM_SUSPEND_PREPARE: | |
1096 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1097 | cpu_hotplug_disable(); |
79cfbdfa SB |
1098 | break; |
1099 | ||
1100 | case PM_POST_SUSPEND: | |
1101 | case PM_POST_HIBERNATION: | |
16e53dbf | 1102 | cpu_hotplug_enable(); |
79cfbdfa SB |
1103 | break; |
1104 | ||
1105 | default: | |
1106 | return NOTIFY_DONE; | |
1107 | } | |
1108 | ||
1109 | return NOTIFY_OK; | |
1110 | } | |
1111 | ||
1112 | ||
d7268a31 | 1113 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1114 | { |
6e32d479 FY |
1115 | /* |
1116 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1117 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1118 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1119 | */ | |
79cfbdfa SB |
1120 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1121 | return 0; | |
1122 | } | |
1123 | core_initcall(cpu_hotplug_pm_sync_init); | |
1124 | ||
f3de4be9 | 1125 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
1126 | |
1127 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 1128 | |
cff7d378 TG |
1129 | /* Boot processor state steps */ |
1130 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1131 | [CPUHP_OFFLINE] = { | |
1132 | .name = "offline", | |
3c1627e9 TG |
1133 | .startup.single = NULL, |
1134 | .teardown.single = NULL, | |
cff7d378 TG |
1135 | }, |
1136 | #ifdef CONFIG_SMP | |
1137 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1138 | .name = "threads:prepare", |
3c1627e9 TG |
1139 | .startup.single = smpboot_create_threads, |
1140 | .teardown.single = NULL, | |
757c989b | 1141 | .cant_stop = true, |
cff7d378 | 1142 | }, |
00e16c3d | 1143 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1144 | .name = "perf:prepare", |
1145 | .startup.single = perf_event_init_cpu, | |
1146 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1147 | }, |
7ee681b2 | 1148 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1149 | .name = "workqueue:prepare", |
1150 | .startup.single = workqueue_prepare_cpu, | |
1151 | .teardown.single = NULL, | |
7ee681b2 | 1152 | }, |
27590dc1 | 1153 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1154 | .name = "hrtimers:prepare", |
1155 | .startup.single = hrtimers_prepare_cpu, | |
1156 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1157 | }, |
31487f83 | 1158 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1159 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1160 | .startup.single = smpcfd_prepare_cpu, |
1161 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1162 | }, |
e6d4989a RW |
1163 | [CPUHP_RELAY_PREPARE] = { |
1164 | .name = "relay:prepare", | |
1165 | .startup.single = relay_prepare_cpu, | |
1166 | .teardown.single = NULL, | |
1167 | }, | |
6731d4f1 SAS |
1168 | [CPUHP_SLAB_PREPARE] = { |
1169 | .name = "slab:prepare", | |
1170 | .startup.single = slab_prepare_cpu, | |
1171 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1172 | }, |
4df83742 | 1173 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1174 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1175 | .startup.single = rcutree_prepare_cpu, |
1176 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1177 | }, |
4fae16df RC |
1178 | /* |
1179 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1180 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1181 | * otherwise a RCU stall occurs. | |
1182 | */ | |
1183 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1184 | .name = "timers:dead", |
1185 | .startup.single = NULL, | |
1186 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1187 | }, |
d10ef6f9 | 1188 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1189 | [CPUHP_BRINGUP_CPU] = { |
1190 | .name = "cpu:bringup", | |
3c1627e9 TG |
1191 | .startup.single = bringup_cpu, |
1192 | .teardown.single = NULL, | |
757c989b | 1193 | .cant_stop = true, |
4baa0afc | 1194 | }, |
31487f83 | 1195 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1196 | .name = "smpcfd:dying", |
3c1627e9 TG |
1197 | .startup.single = NULL, |
1198 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1199 | }, |
d10ef6f9 TG |
1200 | /* |
1201 | * Handled on controll processor until the plugged processor manages | |
1202 | * this itself. | |
1203 | */ | |
4baa0afc TG |
1204 | [CPUHP_TEARDOWN_CPU] = { |
1205 | .name = "cpu:teardown", | |
3c1627e9 TG |
1206 | .startup.single = NULL, |
1207 | .teardown.single = takedown_cpu, | |
757c989b | 1208 | .cant_stop = true, |
cff7d378 | 1209 | }, |
a7c73414 TG |
1210 | #else |
1211 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1212 | #endif |
cff7d378 TG |
1213 | }; |
1214 | ||
4baa0afc TG |
1215 | /* Application processor state steps */ |
1216 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1217 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1218 | /* Final state before CPU kills itself */ |
1219 | [CPUHP_AP_IDLE_DEAD] = { | |
1220 | .name = "idle:dead", | |
1221 | }, | |
1222 | /* | |
1223 | * Last state before CPU enters the idle loop to die. Transient state | |
1224 | * for synchronization. | |
1225 | */ | |
1226 | [CPUHP_AP_OFFLINE] = { | |
1227 | .name = "ap:offline", | |
1228 | .cant_stop = true, | |
1229 | }, | |
9cf7243d TG |
1230 | /* First state is scheduler control. Interrupts are disabled */ |
1231 | [CPUHP_AP_SCHED_STARTING] = { | |
1232 | .name = "sched:starting", | |
3c1627e9 TG |
1233 | .startup.single = sched_cpu_starting, |
1234 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1235 | }, |
4df83742 | 1236 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1237 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1238 | .startup.single = NULL, |
1239 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1240 | }, |
d10ef6f9 TG |
1241 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1242 | * state for synchronsization */ | |
1243 | [CPUHP_AP_ONLINE] = { | |
1244 | .name = "ap:online", | |
1245 | }, | |
1246 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1247 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1248 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1249 | .startup.single = smpboot_unpark_threads, |
1250 | .teardown.single = NULL, | |
1cf4f629 | 1251 | }, |
00e16c3d | 1252 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1253 | .name = "perf:online", |
1254 | .startup.single = perf_event_init_cpu, | |
1255 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1256 | }, |
7ee681b2 | 1257 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1258 | .name = "workqueue:online", |
1259 | .startup.single = workqueue_online_cpu, | |
1260 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1261 | }, |
4df83742 | 1262 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1263 | .name = "RCU/tree:online", |
3c1627e9 TG |
1264 | .startup.single = rcutree_online_cpu, |
1265 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1266 | }, |
4baa0afc | 1267 | #endif |
d10ef6f9 TG |
1268 | /* |
1269 | * The dynamically registered state space is here | |
1270 | */ | |
1271 | ||
aaddd7d1 TG |
1272 | #ifdef CONFIG_SMP |
1273 | /* Last state is scheduler control setting the cpu active */ | |
1274 | [CPUHP_AP_ACTIVE] = { | |
1275 | .name = "sched:active", | |
3c1627e9 TG |
1276 | .startup.single = sched_cpu_activate, |
1277 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1278 | }, |
1279 | #endif | |
1280 | ||
d10ef6f9 | 1281 | /* CPU is fully up and running. */ |
4baa0afc TG |
1282 | [CPUHP_ONLINE] = { |
1283 | .name = "online", | |
3c1627e9 TG |
1284 | .startup.single = NULL, |
1285 | .teardown.single = NULL, | |
4baa0afc TG |
1286 | }, |
1287 | }; | |
1288 | ||
5b7aa87e TG |
1289 | /* Sanity check for callbacks */ |
1290 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1291 | { | |
1292 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1293 | return -EINVAL; | |
1294 | return 0; | |
1295 | } | |
1296 | ||
dc280d93 TG |
1297 | /* |
1298 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1299 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1300 | * by having no name assigned. | |
1301 | */ | |
1302 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1303 | { | |
4205e478 TG |
1304 | enum cpuhp_state i, end; |
1305 | struct cpuhp_step *step; | |
dc280d93 | 1306 | |
4205e478 TG |
1307 | switch (state) { |
1308 | case CPUHP_AP_ONLINE_DYN: | |
1309 | step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; | |
1310 | end = CPUHP_AP_ONLINE_DYN_END; | |
1311 | break; | |
1312 | case CPUHP_BP_PREPARE_DYN: | |
1313 | step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; | |
1314 | end = CPUHP_BP_PREPARE_DYN_END; | |
1315 | break; | |
1316 | default: | |
1317 | return -EINVAL; | |
1318 | } | |
1319 | ||
1320 | for (i = state; i <= end; i++, step++) { | |
1321 | if (!step->name) | |
dc280d93 TG |
1322 | return i; |
1323 | } | |
1324 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1325 | return -ENOSPC; | |
1326 | } | |
1327 | ||
1328 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1329 | int (*startup)(unsigned int cpu), | |
1330 | int (*teardown)(unsigned int cpu), | |
1331 | bool multi_instance) | |
5b7aa87e TG |
1332 | { |
1333 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1334 | struct cpuhp_step *sp; | |
dc280d93 | 1335 | int ret = 0; |
5b7aa87e TG |
1336 | |
1337 | mutex_lock(&cpuhp_state_mutex); | |
dc280d93 | 1338 | |
4205e478 | 1339 | if (state == CPUHP_AP_ONLINE_DYN || state == CPUHP_BP_PREPARE_DYN) { |
dc280d93 TG |
1340 | ret = cpuhp_reserve_state(state); |
1341 | if (ret < 0) | |
1342 | goto out; | |
1343 | state = ret; | |
1344 | } | |
5b7aa87e | 1345 | sp = cpuhp_get_step(state); |
dc280d93 TG |
1346 | if (name && sp->name) { |
1347 | ret = -EBUSY; | |
1348 | goto out; | |
1349 | } | |
3c1627e9 TG |
1350 | sp->startup.single = startup; |
1351 | sp->teardown.single = teardown; | |
5b7aa87e | 1352 | sp->name = name; |
cf392d10 TG |
1353 | sp->multi_instance = multi_instance; |
1354 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1355 | out: |
5b7aa87e | 1356 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 | 1357 | return ret; |
5b7aa87e TG |
1358 | } |
1359 | ||
1360 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1361 | { | |
3c1627e9 | 1362 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1363 | } |
1364 | ||
5b7aa87e TG |
1365 | /* |
1366 | * Call the startup/teardown function for a step either on the AP or | |
1367 | * on the current CPU. | |
1368 | */ | |
cf392d10 TG |
1369 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1370 | struct hlist_node *node) | |
5b7aa87e | 1371 | { |
a724632c | 1372 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1373 | int ret; |
1374 | ||
3c1627e9 TG |
1375 | if ((bringup && !sp->startup.single) || |
1376 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1377 | return 0; |
5b7aa87e TG |
1378 | /* |
1379 | * The non AP bound callbacks can fail on bringup. On teardown | |
1380 | * e.g. module removal we crash for now. | |
1381 | */ | |
1cf4f629 TG |
1382 | #ifdef CONFIG_SMP |
1383 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1384 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1385 | else |
cf392d10 | 1386 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1387 | #else |
cf392d10 | 1388 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1389 | #endif |
5b7aa87e TG |
1390 | BUG_ON(ret && !bringup); |
1391 | return ret; | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1396 | * | |
1397 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1398 | */ | |
1399 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1400 | struct hlist_node *node) |
5b7aa87e TG |
1401 | { |
1402 | int cpu; | |
1403 | ||
5b7aa87e TG |
1404 | /* Roll back the already executed steps on the other cpus */ |
1405 | for_each_present_cpu(cpu) { | |
1406 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1407 | int cpustate = st->state; | |
1408 | ||
1409 | if (cpu >= failedcpu) | |
1410 | break; | |
1411 | ||
1412 | /* Did we invoke the startup call on that cpu ? */ | |
1413 | if (cpustate >= state) | |
cf392d10 | 1414 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1415 | } |
1416 | } | |
1417 | ||
cf392d10 TG |
1418 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, |
1419 | bool invoke) | |
1420 | { | |
1421 | struct cpuhp_step *sp; | |
1422 | int cpu; | |
1423 | int ret; | |
1424 | ||
1425 | sp = cpuhp_get_step(state); | |
1426 | if (sp->multi_instance == false) | |
1427 | return -EINVAL; | |
1428 | ||
1429 | get_online_cpus(); | |
1430 | ||
3c1627e9 | 1431 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1432 | goto add_node; |
1433 | ||
1434 | /* | |
1435 | * Try to call the startup callback for each present cpu | |
1436 | * depending on the hotplug state of the cpu. | |
1437 | */ | |
1438 | for_each_present_cpu(cpu) { | |
1439 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1440 | int cpustate = st->state; | |
1441 | ||
1442 | if (cpustate < state) | |
1443 | continue; | |
1444 | ||
1445 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1446 | if (ret) { | |
3c1627e9 | 1447 | if (sp->teardown.multi) |
cf392d10 TG |
1448 | cpuhp_rollback_install(cpu, state, node); |
1449 | goto err; | |
1450 | } | |
1451 | } | |
1452 | add_node: | |
1453 | ret = 0; | |
1454 | mutex_lock(&cpuhp_state_mutex); | |
1455 | hlist_add_head(node, &sp->list); | |
1456 | mutex_unlock(&cpuhp_state_mutex); | |
1457 | ||
1458 | err: | |
1459 | put_online_cpus(); | |
1460 | return ret; | |
1461 | } | |
1462 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1463 | ||
5b7aa87e TG |
1464 | /** |
1465 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
dc280d93 TG |
1466 | * @state: The state to setup |
1467 | * @invoke: If true, the startup function is invoked for cpus where | |
1468 | * cpu state >= @state | |
1469 | * @startup: startup callback function | |
1470 | * @teardown: teardown callback function | |
1471 | * @multi_instance: State is set up for multiple instances which get | |
1472 | * added afterwards. | |
5b7aa87e | 1473 | * |
512f0980 BO |
1474 | * Returns: |
1475 | * On success: | |
1476 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1477 | * 0 for all other states | |
1478 | * On failure: proper (negative) error code | |
5b7aa87e TG |
1479 | */ |
1480 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1481 | const char *name, bool invoke, | |
1482 | int (*startup)(unsigned int cpu), | |
cf392d10 TG |
1483 | int (*teardown)(unsigned int cpu), |
1484 | bool multi_instance) | |
5b7aa87e TG |
1485 | { |
1486 | int cpu, ret = 0; | |
b9d9d691 | 1487 | bool dynstate; |
5b7aa87e TG |
1488 | |
1489 | if (cpuhp_cb_check(state) || !name) | |
1490 | return -EINVAL; | |
1491 | ||
1492 | get_online_cpus(); | |
1493 | ||
dc280d93 TG |
1494 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1495 | multi_instance); | |
5b7aa87e | 1496 | |
b9d9d691 TG |
1497 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1498 | if (ret > 0 && dynstate) { | |
1499 | state = ret; | |
1500 | ret = 0; | |
1501 | } | |
1502 | ||
dc280d93 | 1503 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1504 | goto out; |
1505 | ||
1506 | /* | |
1507 | * Try to call the startup callback for each present cpu | |
1508 | * depending on the hotplug state of the cpu. | |
1509 | */ | |
1510 | for_each_present_cpu(cpu) { | |
1511 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1512 | int cpustate = st->state; | |
1513 | ||
1514 | if (cpustate < state) | |
1515 | continue; | |
1516 | ||
cf392d10 | 1517 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1518 | if (ret) { |
a724632c | 1519 | if (teardown) |
cf392d10 TG |
1520 | cpuhp_rollback_install(cpu, state, NULL); |
1521 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1522 | goto out; |
1523 | } | |
1524 | } | |
1525 | out: | |
1526 | put_online_cpus(); | |
dc280d93 TG |
1527 | /* |
1528 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1529 | * dynamically allocated state in case of success. | |
1530 | */ | |
b9d9d691 | 1531 | if (!ret && dynstate) |
5b7aa87e TG |
1532 | return state; |
1533 | return ret; | |
1534 | } | |
1535 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1536 | ||
cf392d10 TG |
1537 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1538 | struct hlist_node *node, bool invoke) | |
1539 | { | |
1540 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1541 | int cpu; | |
1542 | ||
1543 | BUG_ON(cpuhp_cb_check(state)); | |
1544 | ||
1545 | if (!sp->multi_instance) | |
1546 | return -EINVAL; | |
1547 | ||
1548 | get_online_cpus(); | |
1549 | if (!invoke || !cpuhp_get_teardown_cb(state)) | |
1550 | goto remove; | |
1551 | /* | |
1552 | * Call the teardown callback for each present cpu depending | |
1553 | * on the hotplug state of the cpu. This function is not | |
1554 | * allowed to fail currently! | |
1555 | */ | |
1556 | for_each_present_cpu(cpu) { | |
1557 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1558 | int cpustate = st->state; | |
1559 | ||
1560 | if (cpustate >= state) | |
1561 | cpuhp_issue_call(cpu, state, false, node); | |
1562 | } | |
1563 | ||
1564 | remove: | |
1565 | mutex_lock(&cpuhp_state_mutex); | |
1566 | hlist_del(node); | |
1567 | mutex_unlock(&cpuhp_state_mutex); | |
1568 | put_online_cpus(); | |
1569 | ||
1570 | return 0; | |
1571 | } | |
1572 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
5b7aa87e TG |
1573 | /** |
1574 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1575 | * @state: The state to remove | |
1576 | * @invoke: If true, the teardown function is invoked for cpus where | |
1577 | * cpu state >= @state | |
1578 | * | |
1579 | * The teardown callback is currently not allowed to fail. Think | |
1580 | * about module removal! | |
1581 | */ | |
1582 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1583 | { | |
cf392d10 | 1584 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1585 | int cpu; |
1586 | ||
1587 | BUG_ON(cpuhp_cb_check(state)); | |
1588 | ||
1589 | get_online_cpus(); | |
1590 | ||
cf392d10 TG |
1591 | if (sp->multi_instance) { |
1592 | WARN(!hlist_empty(&sp->list), | |
1593 | "Error: Removing state %d which has instances left.\n", | |
1594 | state); | |
1595 | goto remove; | |
1596 | } | |
1597 | ||
a724632c | 1598 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1599 | goto remove; |
1600 | ||
1601 | /* | |
1602 | * Call the teardown callback for each present cpu depending | |
1603 | * on the hotplug state of the cpu. This function is not | |
1604 | * allowed to fail currently! | |
1605 | */ | |
1606 | for_each_present_cpu(cpu) { | |
1607 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1608 | int cpustate = st->state; | |
1609 | ||
1610 | if (cpustate >= state) | |
cf392d10 | 1611 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1612 | } |
1613 | remove: | |
cf392d10 | 1614 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
5b7aa87e TG |
1615 | put_online_cpus(); |
1616 | } | |
1617 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1618 | ||
98f8cdce TG |
1619 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1620 | static ssize_t show_cpuhp_state(struct device *dev, | |
1621 | struct device_attribute *attr, char *buf) | |
1622 | { | |
1623 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1624 | ||
1625 | return sprintf(buf, "%d\n", st->state); | |
1626 | } | |
1627 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1628 | ||
757c989b TG |
1629 | static ssize_t write_cpuhp_target(struct device *dev, |
1630 | struct device_attribute *attr, | |
1631 | const char *buf, size_t count) | |
1632 | { | |
1633 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1634 | struct cpuhp_step *sp; | |
1635 | int target, ret; | |
1636 | ||
1637 | ret = kstrtoint(buf, 10, &target); | |
1638 | if (ret) | |
1639 | return ret; | |
1640 | ||
1641 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1642 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1643 | return -EINVAL; | |
1644 | #else | |
1645 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1646 | return -EINVAL; | |
1647 | #endif | |
1648 | ||
1649 | ret = lock_device_hotplug_sysfs(); | |
1650 | if (ret) | |
1651 | return ret; | |
1652 | ||
1653 | mutex_lock(&cpuhp_state_mutex); | |
1654 | sp = cpuhp_get_step(target); | |
1655 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1656 | mutex_unlock(&cpuhp_state_mutex); | |
1657 | if (ret) | |
1658 | return ret; | |
1659 | ||
1660 | if (st->state < target) | |
1661 | ret = do_cpu_up(dev->id, target); | |
1662 | else | |
1663 | ret = do_cpu_down(dev->id, target); | |
1664 | ||
1665 | unlock_device_hotplug(); | |
1666 | return ret ? ret : count; | |
1667 | } | |
1668 | ||
98f8cdce TG |
1669 | static ssize_t show_cpuhp_target(struct device *dev, |
1670 | struct device_attribute *attr, char *buf) | |
1671 | { | |
1672 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1673 | ||
1674 | return sprintf(buf, "%d\n", st->target); | |
1675 | } | |
757c989b | 1676 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1677 | |
1678 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1679 | &dev_attr_state.attr, | |
1680 | &dev_attr_target.attr, | |
1681 | NULL | |
1682 | }; | |
1683 | ||
1684 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1685 | .attrs = cpuhp_cpu_attrs, | |
1686 | .name = "hotplug", | |
1687 | NULL | |
1688 | }; | |
1689 | ||
1690 | static ssize_t show_cpuhp_states(struct device *dev, | |
1691 | struct device_attribute *attr, char *buf) | |
1692 | { | |
1693 | ssize_t cur, res = 0; | |
1694 | int i; | |
1695 | ||
1696 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1697 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1698 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1699 | ||
1700 | if (sp->name) { | |
1701 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1702 | buf += cur; | |
1703 | res += cur; | |
1704 | } | |
1705 | } | |
1706 | mutex_unlock(&cpuhp_state_mutex); | |
1707 | return res; | |
1708 | } | |
1709 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1710 | ||
1711 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1712 | &dev_attr_states.attr, | |
1713 | NULL | |
1714 | }; | |
1715 | ||
1716 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1717 | .attrs = cpuhp_cpu_root_attrs, | |
1718 | .name = "hotplug", | |
1719 | NULL | |
1720 | }; | |
1721 | ||
1722 | static int __init cpuhp_sysfs_init(void) | |
1723 | { | |
1724 | int cpu, ret; | |
1725 | ||
1726 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1727 | &cpuhp_cpu_root_attr_group); | |
1728 | if (ret) | |
1729 | return ret; | |
1730 | ||
1731 | for_each_possible_cpu(cpu) { | |
1732 | struct device *dev = get_cpu_device(cpu); | |
1733 | ||
1734 | if (!dev) | |
1735 | continue; | |
1736 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1737 | if (ret) | |
1738 | return ret; | |
1739 | } | |
1740 | return 0; | |
1741 | } | |
1742 | device_initcall(cpuhp_sysfs_init); | |
1743 | #endif | |
1744 | ||
e56b3bc7 LT |
1745 | /* |
1746 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1747 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1748 | * | |
e0b582ec | 1749 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1750 | * mask value that has a single bit set only. |
1751 | */ | |
b8d317d1 | 1752 | |
e56b3bc7 | 1753 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1754 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1755 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1756 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1757 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1758 | |
e56b3bc7 LT |
1759 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1760 | ||
1761 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1762 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1763 | #if BITS_PER_LONG > 32 | |
1764 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1765 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1766 | #endif |
1767 | }; | |
e56b3bc7 | 1768 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1769 | |
1770 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1771 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1772 | |
1773 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1774 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1775 | = {CPU_BITS_ALL}; |
b3199c02 | 1776 | #else |
4b804c85 | 1777 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1778 | #endif |
4b804c85 | 1779 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1780 | |
4b804c85 RV |
1781 | struct cpumask __cpu_online_mask __read_mostly; |
1782 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1783 | |
4b804c85 RV |
1784 | struct cpumask __cpu_present_mask __read_mostly; |
1785 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1786 | |
4b804c85 RV |
1787 | struct cpumask __cpu_active_mask __read_mostly; |
1788 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1789 | |
3fa41520 RR |
1790 | void init_cpu_present(const struct cpumask *src) |
1791 | { | |
c4c54dd1 | 1792 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1793 | } |
1794 | ||
1795 | void init_cpu_possible(const struct cpumask *src) | |
1796 | { | |
c4c54dd1 | 1797 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1798 | } |
1799 | ||
1800 | void init_cpu_online(const struct cpumask *src) | |
1801 | { | |
c4c54dd1 | 1802 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1803 | } |
cff7d378 TG |
1804 | |
1805 | /* | |
1806 | * Activate the first processor. | |
1807 | */ | |
1808 | void __init boot_cpu_init(void) | |
1809 | { | |
1810 | int cpu = smp_processor_id(); | |
1811 | ||
1812 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1813 | set_cpu_online(cpu, true); | |
1814 | set_cpu_active(cpu, true); | |
1815 | set_cpu_present(cpu, true); | |
1816 | set_cpu_possible(cpu, true); | |
1817 | } | |
1818 | ||
1819 | /* | |
1820 | * Must be called _AFTER_ setting up the per_cpu areas | |
1821 | */ | |
1822 | void __init boot_cpu_state_init(void) | |
1823 | { | |
1824 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1825 | } |