]>
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> |
29930025 | 12 | #include <linux/sched/task.h> |
1da177e4 LT |
13 | #include <linux/unistd.h> |
14 | #include <linux/cpu.h> | |
cb79295e AV |
15 | #include <linux/oom.h> |
16 | #include <linux/rcupdate.h> | |
9984de1a | 17 | #include <linux/export.h> |
e4cc2f87 | 18 | #include <linux/bug.h> |
1da177e4 LT |
19 | #include <linux/kthread.h> |
20 | #include <linux/stop_machine.h> | |
81615b62 | 21 | #include <linux/mutex.h> |
5a0e3ad6 | 22 | #include <linux/gfp.h> |
79cfbdfa | 23 | #include <linux/suspend.h> |
a19423b9 | 24 | #include <linux/lockdep.h> |
345527b1 | 25 | #include <linux/tick.h> |
a8994181 | 26 | #include <linux/irq.h> |
4cb28ced | 27 | #include <linux/smpboot.h> |
e6d4989a | 28 | #include <linux/relay.h> |
6731d4f1 | 29 | #include <linux/slab.h> |
fc8dffd3 | 30 | #include <linux/percpu-rwsem.h> |
cff7d378 | 31 | |
bb3632c6 | 32 | #include <trace/events/power.h> |
cff7d378 TG |
33 | #define CREATE_TRACE_POINTS |
34 | #include <trace/events/cpuhp.h> | |
1da177e4 | 35 | |
38498a67 TG |
36 | #include "smpboot.h" |
37 | ||
cff7d378 TG |
38 | /** |
39 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
40 | * @state: The current cpu state | |
41 | * @target: The target state | |
4cb28ced TG |
42 | * @thread: Pointer to the hotplug thread |
43 | * @should_run: Thread should execute | |
3b9d6da6 | 44 | * @rollback: Perform a rollback |
a724632c TG |
45 | * @single: Single callback invocation |
46 | * @bringup: Single callback bringup or teardown selector | |
47 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced TG |
48 | * @result: Result of the operation |
49 | * @done: Signal completion to the issuer of the task | |
cff7d378 TG |
50 | */ |
51 | struct cpuhp_cpu_state { | |
52 | enum cpuhp_state state; | |
53 | enum cpuhp_state target; | |
4cb28ced TG |
54 | #ifdef CONFIG_SMP |
55 | struct task_struct *thread; | |
56 | bool should_run; | |
3b9d6da6 | 57 | bool rollback; |
a724632c TG |
58 | bool single; |
59 | bool bringup; | |
cf392d10 | 60 | struct hlist_node *node; |
4cb28ced | 61 | enum cpuhp_state cb_state; |
4cb28ced TG |
62 | int result; |
63 | struct completion done; | |
64 | #endif | |
cff7d378 TG |
65 | }; |
66 | ||
67 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
68 | ||
49dfe2a6 TG |
69 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
70 | static struct lock_class_key cpuhp_state_key; | |
71 | static struct lockdep_map cpuhp_state_lock_map = | |
72 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state", &cpuhp_state_key); | |
73 | #endif | |
74 | ||
cff7d378 TG |
75 | /** |
76 | * cpuhp_step - Hotplug state machine step | |
77 | * @name: Name of the step | |
78 | * @startup: Startup function of the step | |
79 | * @teardown: Teardown function of the step | |
80 | * @skip_onerr: Do not invoke the functions on error rollback | |
81 | * Will go away once the notifiers are gone | |
757c989b | 82 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
83 | */ |
84 | struct cpuhp_step { | |
cf392d10 TG |
85 | const char *name; |
86 | union { | |
3c1627e9 TG |
87 | int (*single)(unsigned int cpu); |
88 | int (*multi)(unsigned int cpu, | |
89 | struct hlist_node *node); | |
90 | } startup; | |
cf392d10 | 91 | union { |
3c1627e9 TG |
92 | int (*single)(unsigned int cpu); |
93 | int (*multi)(unsigned int cpu, | |
94 | struct hlist_node *node); | |
95 | } teardown; | |
cf392d10 TG |
96 | struct hlist_head list; |
97 | bool skip_onerr; | |
98 | bool cant_stop; | |
99 | bool multi_instance; | |
cff7d378 TG |
100 | }; |
101 | ||
98f8cdce | 102 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 103 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 104 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 105 | |
a724632c TG |
106 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
107 | { | |
108 | /* | |
109 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
110 | * purposes as that state is handled explicitly in cpu_down. | |
111 | */ | |
112 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
113 | } | |
114 | ||
115 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
116 | { | |
117 | struct cpuhp_step *sp; | |
118 | ||
119 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
120 | return sp + state; | |
121 | } | |
122 | ||
cff7d378 TG |
123 | /** |
124 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
125 | * @cpu: The cpu for which the callback should be invoked | |
126 | * @step: The step in the state machine | |
a724632c | 127 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 128 | * |
cf392d10 | 129 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 130 | */ |
a724632c | 131 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 132 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
133 | { |
134 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 135 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
136 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
137 | int (*cb)(unsigned int cpu); | |
138 | int ret, cnt; | |
139 | ||
140 | if (!step->multi_instance) { | |
3c1627e9 | 141 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
142 | if (!cb) |
143 | return 0; | |
a724632c | 144 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 145 | ret = cb(cpu); |
a724632c | 146 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
147 | return ret; |
148 | } | |
3c1627e9 | 149 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
150 | if (!cbm) |
151 | return 0; | |
152 | ||
153 | /* Single invocation for instance add/remove */ | |
154 | if (node) { | |
155 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
156 | ret = cbm(cpu, node); | |
157 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
158 | return ret; | |
159 | } | |
160 | ||
161 | /* State transition. Invoke on all instances */ | |
162 | cnt = 0; | |
163 | hlist_for_each(node, &step->list) { | |
164 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
165 | ret = cbm(cpu, node); | |
166 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
167 | if (ret) | |
168 | goto err; | |
169 | cnt++; | |
170 | } | |
171 | return 0; | |
172 | err: | |
173 | /* Rollback the instances if one failed */ | |
3c1627e9 | 174 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
175 | if (!cbm) |
176 | return ret; | |
177 | ||
178 | hlist_for_each(node, &step->list) { | |
179 | if (!cnt--) | |
180 | break; | |
181 | cbm(cpu, node); | |
cff7d378 TG |
182 | } |
183 | return ret; | |
184 | } | |
185 | ||
98a79d6a | 186 | #ifdef CONFIG_SMP |
b3199c02 | 187 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 188 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
189 | bool cpuhp_tasks_frozen; |
190 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 191 | |
79a6cdeb | 192 | /* |
93ae4f97 SB |
193 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
194 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
195 | */ |
196 | void cpu_maps_update_begin(void) | |
197 | { | |
198 | mutex_lock(&cpu_add_remove_lock); | |
199 | } | |
200 | ||
201 | void cpu_maps_update_done(void) | |
202 | { | |
203 | mutex_unlock(&cpu_add_remove_lock); | |
204 | } | |
1da177e4 | 205 | |
fc8dffd3 TG |
206 | /* |
207 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
208 | * Should always be manipulated under cpu_add_remove_lock |
209 | */ | |
210 | static int cpu_hotplug_disabled; | |
211 | ||
79a6cdeb LJ |
212 | #ifdef CONFIG_HOTPLUG_CPU |
213 | ||
fc8dffd3 | 214 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 215 | |
8f553c49 | 216 | void cpus_read_lock(void) |
a9d9baa1 | 217 | { |
fc8dffd3 | 218 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 219 | } |
8f553c49 | 220 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 221 | |
8f553c49 | 222 | void cpus_read_unlock(void) |
a9d9baa1 | 223 | { |
fc8dffd3 | 224 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 225 | } |
8f553c49 | 226 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 227 | |
8f553c49 | 228 | void cpus_write_lock(void) |
d221938c | 229 | { |
fc8dffd3 | 230 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 231 | } |
87af9e7f | 232 | |
8f553c49 | 233 | void cpus_write_unlock(void) |
d221938c | 234 | { |
fc8dffd3 | 235 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
236 | } |
237 | ||
fc8dffd3 | 238 | void lockdep_assert_cpus_held(void) |
d221938c | 239 | { |
fc8dffd3 | 240 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 241 | } |
79a6cdeb | 242 | |
16e53dbf SB |
243 | /* |
244 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
245 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
246 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
247 | * hotplug path before performing hotplug operations. So acquiring that lock | |
248 | * guarantees mutual exclusion from any currently running hotplug operations. | |
249 | */ | |
250 | void cpu_hotplug_disable(void) | |
251 | { | |
252 | cpu_maps_update_begin(); | |
89af7ba5 | 253 | cpu_hotplug_disabled++; |
16e53dbf SB |
254 | cpu_maps_update_done(); |
255 | } | |
32145c46 | 256 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 257 | |
01b41159 LW |
258 | static void __cpu_hotplug_enable(void) |
259 | { | |
260 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
261 | return; | |
262 | cpu_hotplug_disabled--; | |
263 | } | |
264 | ||
16e53dbf SB |
265 | void cpu_hotplug_enable(void) |
266 | { | |
267 | cpu_maps_update_begin(); | |
01b41159 | 268 | __cpu_hotplug_enable(); |
16e53dbf SB |
269 | cpu_maps_update_done(); |
270 | } | |
32145c46 | 271 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 272 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 273 | |
9cd4f1a4 TG |
274 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st); |
275 | ||
8df3e07e TG |
276 | static int bringup_wait_for_ap(unsigned int cpu) |
277 | { | |
278 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
279 | ||
9cd4f1a4 | 280 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
8df3e07e | 281 | wait_for_completion(&st->done); |
9cd4f1a4 TG |
282 | BUG_ON(!cpu_online(cpu)); |
283 | ||
284 | /* Unpark the stopper thread and the hotplug thread of the target cpu */ | |
285 | stop_machine_unpark(cpu); | |
286 | kthread_unpark(st->thread); | |
287 | ||
288 | /* Should we go further up ? */ | |
289 | if (st->target > CPUHP_AP_ONLINE_IDLE) { | |
290 | __cpuhp_kick_ap_work(st); | |
291 | wait_for_completion(&st->done); | |
292 | } | |
8df3e07e TG |
293 | return st->result; |
294 | } | |
295 | ||
ba997462 TG |
296 | static int bringup_cpu(unsigned int cpu) |
297 | { | |
298 | struct task_struct *idle = idle_thread_get(cpu); | |
299 | int ret; | |
300 | ||
aa877175 BO |
301 | /* |
302 | * Some architectures have to walk the irq descriptors to | |
303 | * setup the vector space for the cpu which comes online. | |
304 | * Prevent irq alloc/free across the bringup. | |
305 | */ | |
306 | irq_lock_sparse(); | |
307 | ||
ba997462 TG |
308 | /* Arch-specific enabling code. */ |
309 | ret = __cpu_up(cpu, idle); | |
aa877175 | 310 | irq_unlock_sparse(); |
530e9b76 | 311 | if (ret) |
ba997462 | 312 | return ret; |
9cd4f1a4 | 313 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
314 | } |
315 | ||
2e1a3483 TG |
316 | /* |
317 | * Hotplug state machine related functions | |
318 | */ | |
a724632c | 319 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
320 | { |
321 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 322 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
323 | |
324 | if (!step->skip_onerr) | |
cf392d10 | 325 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
326 | } |
327 | } | |
328 | ||
329 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 330 | enum cpuhp_state target) |
2e1a3483 TG |
331 | { |
332 | enum cpuhp_state prev_state = st->state; | |
333 | int ret = 0; | |
334 | ||
335 | for (; st->state > target; st->state--) { | |
cf392d10 | 336 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
337 | if (ret) { |
338 | st->target = prev_state; | |
a724632c | 339 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
340 | break; |
341 | } | |
342 | } | |
343 | return ret; | |
344 | } | |
345 | ||
a724632c | 346 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
347 | { |
348 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 349 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
350 | |
351 | if (!step->skip_onerr) | |
cf392d10 | 352 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
353 | } |
354 | } | |
355 | ||
356 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 357 | enum cpuhp_state target) |
2e1a3483 TG |
358 | { |
359 | enum cpuhp_state prev_state = st->state; | |
360 | int ret = 0; | |
361 | ||
362 | while (st->state < target) { | |
2e1a3483 | 363 | st->state++; |
cf392d10 | 364 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
365 | if (ret) { |
366 | st->target = prev_state; | |
a724632c | 367 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
368 | break; |
369 | } | |
370 | } | |
371 | return ret; | |
372 | } | |
373 | ||
4cb28ced TG |
374 | /* |
375 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
376 | */ | |
377 | static void cpuhp_create(unsigned int cpu) | |
378 | { | |
379 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
380 | ||
381 | init_completion(&st->done); | |
382 | } | |
383 | ||
384 | static int cpuhp_should_run(unsigned int cpu) | |
385 | { | |
386 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
387 | ||
388 | return st->should_run; | |
389 | } | |
390 | ||
391 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
392 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
393 | { | |
1cf4f629 | 394 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 395 | |
a724632c | 396 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
397 | } |
398 | ||
399 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
400 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
401 | { | |
a724632c | 402 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
403 | } |
404 | ||
405 | /* | |
406 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
407 | * callbacks when a state gets [un]installed at runtime. | |
408 | */ | |
409 | static void cpuhp_thread_fun(unsigned int cpu) | |
410 | { | |
411 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
412 | int ret = 0; | |
413 | ||
414 | /* | |
415 | * Paired with the mb() in cpuhp_kick_ap_work and | |
416 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
417 | */ | |
418 | smp_mb(); | |
419 | if (!st->should_run) | |
420 | return; | |
421 | ||
422 | st->should_run = false; | |
423 | ||
49dfe2a6 | 424 | lock_map_acquire(&cpuhp_state_lock_map); |
4cb28ced | 425 | /* Single callback invocation for [un]install ? */ |
a724632c | 426 | if (st->single) { |
4cb28ced TG |
427 | if (st->cb_state < CPUHP_AP_ONLINE) { |
428 | local_irq_disable(); | |
a724632c | 429 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 430 | st->bringup, st->node); |
4cb28ced TG |
431 | local_irq_enable(); |
432 | } else { | |
a724632c | 433 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 434 | st->bringup, st->node); |
4cb28ced | 435 | } |
3b9d6da6 SAS |
436 | } else if (st->rollback) { |
437 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
438 | ||
a724632c | 439 | undo_cpu_down(cpu, st); |
3b9d6da6 | 440 | st->rollback = false; |
4cb28ced | 441 | } else { |
1cf4f629 | 442 | /* Cannot happen .... */ |
8df3e07e | 443 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 444 | |
4cb28ced TG |
445 | /* Regular hotplug work */ |
446 | if (st->state < st->target) | |
447 | ret = cpuhp_ap_online(cpu, st); | |
448 | else if (st->state > st->target) | |
449 | ret = cpuhp_ap_offline(cpu, st); | |
450 | } | |
49dfe2a6 | 451 | lock_map_release(&cpuhp_state_lock_map); |
4cb28ced TG |
452 | st->result = ret; |
453 | complete(&st->done); | |
454 | } | |
455 | ||
456 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 457 | static int |
cf392d10 TG |
458 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
459 | struct hlist_node *node) | |
4cb28ced TG |
460 | { |
461 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
462 | ||
463 | if (!cpu_online(cpu)) | |
464 | return 0; | |
465 | ||
49dfe2a6 TG |
466 | lock_map_acquire(&cpuhp_state_lock_map); |
467 | lock_map_release(&cpuhp_state_lock_map); | |
468 | ||
6a4e2451 TG |
469 | /* |
470 | * If we are up and running, use the hotplug thread. For early calls | |
471 | * we invoke the thread function directly. | |
472 | */ | |
473 | if (!st->thread) | |
cf392d10 | 474 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 475 | |
4cb28ced | 476 | st->cb_state = state; |
a724632c TG |
477 | st->single = true; |
478 | st->bringup = bringup; | |
cf392d10 | 479 | st->node = node; |
a724632c | 480 | |
4cb28ced TG |
481 | /* |
482 | * Make sure the above stores are visible before should_run becomes | |
483 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
484 | */ | |
485 | smp_mb(); | |
486 | st->should_run = true; | |
487 | wake_up_process(st->thread); | |
488 | wait_for_completion(&st->done); | |
489 | return st->result; | |
490 | } | |
491 | ||
492 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 493 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 494 | { |
4cb28ced | 495 | st->result = 0; |
a724632c | 496 | st->single = false; |
4cb28ced TG |
497 | /* |
498 | * Make sure the above stores are visible before should_run becomes | |
499 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
500 | */ | |
501 | smp_mb(); | |
502 | st->should_run = true; | |
503 | wake_up_process(st->thread); | |
1cf4f629 TG |
504 | } |
505 | ||
506 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
507 | { | |
508 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
509 | enum cpuhp_state state = st->state; | |
510 | ||
511 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
49dfe2a6 TG |
512 | lock_map_acquire(&cpuhp_state_lock_map); |
513 | lock_map_release(&cpuhp_state_lock_map); | |
1cf4f629 | 514 | __cpuhp_kick_ap_work(st); |
4cb28ced TG |
515 | wait_for_completion(&st->done); |
516 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
517 | return st->result; | |
518 | } | |
519 | ||
520 | static struct smp_hotplug_thread cpuhp_threads = { | |
521 | .store = &cpuhp_state.thread, | |
522 | .create = &cpuhp_create, | |
523 | .thread_should_run = cpuhp_should_run, | |
524 | .thread_fn = cpuhp_thread_fun, | |
525 | .thread_comm = "cpuhp/%u", | |
526 | .selfparking = true, | |
527 | }; | |
528 | ||
529 | void __init cpuhp_threads_init(void) | |
530 | { | |
531 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
532 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
533 | } | |
534 | ||
777c6e0d | 535 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
536 | /** |
537 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
538 | * @cpu: a CPU id | |
539 | * | |
540 | * This function walks all processes, finds a valid mm struct for each one and | |
541 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
542 | * trivial, there are various non-obvious corner cases, which this function | |
543 | * tries to solve in a safe manner. | |
544 | * | |
545 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
546 | * be called only for an already offlined CPU. | |
547 | */ | |
cb79295e AV |
548 | void clear_tasks_mm_cpumask(int cpu) |
549 | { | |
550 | struct task_struct *p; | |
551 | ||
552 | /* | |
553 | * This function is called after the cpu is taken down and marked | |
554 | * offline, so its not like new tasks will ever get this cpu set in | |
555 | * their mm mask. -- Peter Zijlstra | |
556 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
557 | * full-fledged tasklist_lock. | |
558 | */ | |
e4cc2f87 | 559 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
560 | rcu_read_lock(); |
561 | for_each_process(p) { | |
562 | struct task_struct *t; | |
563 | ||
e4cc2f87 AV |
564 | /* |
565 | * Main thread might exit, but other threads may still have | |
566 | * a valid mm. Find one. | |
567 | */ | |
cb79295e AV |
568 | t = find_lock_task_mm(p); |
569 | if (!t) | |
570 | continue; | |
571 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
572 | task_unlock(t); | |
573 | } | |
574 | rcu_read_unlock(); | |
575 | } | |
576 | ||
1da177e4 | 577 | /* Take this CPU down. */ |
71cf5aee | 578 | static int take_cpu_down(void *_param) |
1da177e4 | 579 | { |
4baa0afc TG |
580 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
581 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 582 | int err, cpu = smp_processor_id(); |
1da177e4 | 583 | |
1da177e4 LT |
584 | /* Ensure this CPU doesn't handle any more interrupts. */ |
585 | err = __cpu_disable(); | |
586 | if (err < 0) | |
f3705136 | 587 | return err; |
1da177e4 | 588 | |
a724632c TG |
589 | /* |
590 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
591 | * do this step again. | |
592 | */ | |
593 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
594 | st->state--; | |
4baa0afc | 595 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 596 | for (; st->state > target; st->state--) |
cf392d10 | 597 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 598 | |
52c063d1 TG |
599 | /* Give up timekeeping duties */ |
600 | tick_handover_do_timer(); | |
14e568e7 | 601 | /* Park the stopper thread */ |
090e77c3 | 602 | stop_machine_park(cpu); |
f3705136 | 603 | return 0; |
1da177e4 LT |
604 | } |
605 | ||
98458172 | 606 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 607 | { |
e69aab13 | 608 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 609 | int err; |
1da177e4 | 610 | |
2a58c527 | 611 | /* Park the smpboot threads */ |
1cf4f629 | 612 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 613 | smpboot_park_threads(cpu); |
1cf4f629 | 614 | |
6acce3ef | 615 | /* |
a8994181 TG |
616 | * Prevent irq alloc/free while the dying cpu reorganizes the |
617 | * interrupt affinities. | |
6acce3ef | 618 | */ |
a8994181 | 619 | irq_lock_sparse(); |
6acce3ef | 620 | |
a8994181 TG |
621 | /* |
622 | * So now all preempt/rcu users must observe !cpu_active(). | |
623 | */ | |
210e2133 | 624 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 625 | if (err) { |
3b9d6da6 | 626 | /* CPU refused to die */ |
a8994181 | 627 | irq_unlock_sparse(); |
3b9d6da6 SAS |
628 | /* Unpark the hotplug thread so we can rollback there */ |
629 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 630 | return err; |
8fa1d7d3 | 631 | } |
04321587 | 632 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 633 | |
48c5ccae | 634 | /* |
ee1e714b | 635 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
636 | * runnable tasks from the cpu, there's only the idle task left now |
637 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
638 | * |
639 | * Wait for the stop thread to go away. | |
48c5ccae | 640 | */ |
e69aab13 TG |
641 | wait_for_completion(&st->done); |
642 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 643 | |
a8994181 TG |
644 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
645 | irq_unlock_sparse(); | |
646 | ||
345527b1 | 647 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
648 | /* This actually kills the CPU. */ |
649 | __cpu_die(cpu); | |
650 | ||
a49b116d | 651 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
652 | return 0; |
653 | } | |
1da177e4 | 654 | |
71f87b2f TG |
655 | static void cpuhp_complete_idle_dead(void *arg) |
656 | { | |
657 | struct cpuhp_cpu_state *st = arg; | |
658 | ||
659 | complete(&st->done); | |
660 | } | |
661 | ||
e69aab13 TG |
662 | void cpuhp_report_idle_dead(void) |
663 | { | |
664 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
665 | ||
666 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 667 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
668 | st->state = CPUHP_AP_IDLE_DEAD; |
669 | /* | |
670 | * We cannot call complete after rcu_report_dead() so we delegate it | |
671 | * to an online cpu. | |
672 | */ | |
673 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
674 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
675 | } |
676 | ||
cff7d378 | 677 | #else |
cff7d378 | 678 | #define takedown_cpu NULL |
cff7d378 TG |
679 | #endif |
680 | ||
681 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 682 | |
98458172 | 683 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
684 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
685 | enum cpuhp_state target) | |
98458172 | 686 | { |
cff7d378 TG |
687 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
688 | int prev_state, ret = 0; | |
98458172 TG |
689 | |
690 | if (num_online_cpus() == 1) | |
691 | return -EBUSY; | |
692 | ||
757c989b | 693 | if (!cpu_present(cpu)) |
98458172 TG |
694 | return -EINVAL; |
695 | ||
8f553c49 | 696 | cpus_write_lock(); |
98458172 TG |
697 | |
698 | cpuhp_tasks_frozen = tasks_frozen; | |
699 | ||
cff7d378 | 700 | prev_state = st->state; |
af1f4045 | 701 | st->target = target; |
1cf4f629 TG |
702 | /* |
703 | * If the current CPU state is in the range of the AP hotplug thread, | |
704 | * then we need to kick the thread. | |
705 | */ | |
8df3e07e | 706 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
707 | ret = cpuhp_kick_ap_work(cpu); |
708 | /* | |
709 | * The AP side has done the error rollback already. Just | |
710 | * return the error code.. | |
711 | */ | |
712 | if (ret) | |
713 | goto out; | |
714 | ||
715 | /* | |
716 | * We might have stopped still in the range of the AP hotplug | |
717 | * thread. Nothing to do anymore. | |
718 | */ | |
8df3e07e | 719 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
720 | goto out; |
721 | } | |
722 | /* | |
8df3e07e | 723 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
724 | * to do the further cleanups. |
725 | */ | |
a724632c | 726 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
727 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
728 | st->target = prev_state; | |
729 | st->rollback = true; | |
730 | cpuhp_kick_ap_work(cpu); | |
731 | } | |
98458172 | 732 | |
1cf4f629 | 733 | out: |
8f553c49 | 734 | cpus_write_unlock(); |
cff7d378 | 735 | return ret; |
e3920fb4 RW |
736 | } |
737 | ||
af1f4045 | 738 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 739 | { |
9ea09af3 | 740 | int err; |
e3920fb4 | 741 | |
d221938c | 742 | cpu_maps_update_begin(); |
e761b772 MK |
743 | |
744 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 745 | err = -EBUSY; |
e761b772 MK |
746 | goto out; |
747 | } | |
748 | ||
af1f4045 | 749 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 750 | |
e761b772 | 751 | out: |
d221938c | 752 | cpu_maps_update_done(); |
1da177e4 LT |
753 | return err; |
754 | } | |
af1f4045 TG |
755 | int cpu_down(unsigned int cpu) |
756 | { | |
757 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
758 | } | |
b62b8ef9 | 759 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
760 | #endif /*CONFIG_HOTPLUG_CPU*/ |
761 | ||
4baa0afc | 762 | /** |
ee1e714b | 763 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
764 | * @cpu: cpu that just started |
765 | * | |
4baa0afc TG |
766 | * It must be called by the arch code on the new cpu, before the new cpu |
767 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
768 | */ | |
769 | void notify_cpu_starting(unsigned int cpu) | |
770 | { | |
771 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
772 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
773 | ||
0c6d4576 | 774 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 775 | while (st->state < target) { |
4baa0afc | 776 | st->state++; |
cf392d10 | 777 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
778 | } |
779 | } | |
780 | ||
949338e3 | 781 | /* |
9cd4f1a4 TG |
782 | * Called from the idle task. Wake up the controlling task which brings the |
783 | * stopper and the hotplug thread of the upcoming CPU up and then delegates | |
784 | * the rest of the online bringup to the hotplug thread. | |
949338e3 | 785 | */ |
8df3e07e | 786 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 787 | { |
8df3e07e | 788 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
789 | |
790 | /* Happens for the boot cpu */ | |
791 | if (state != CPUHP_AP_ONLINE_IDLE) | |
792 | return; | |
793 | ||
794 | st->state = CPUHP_AP_ONLINE_IDLE; | |
9cd4f1a4 | 795 | complete(&st->done); |
949338e3 TG |
796 | } |
797 | ||
e3920fb4 | 798 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 799 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 800 | { |
cff7d378 | 801 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 802 | struct task_struct *idle; |
2e1a3483 | 803 | int ret = 0; |
1da177e4 | 804 | |
8f553c49 | 805 | cpus_write_lock(); |
38498a67 | 806 | |
757c989b | 807 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
808 | ret = -EINVAL; |
809 | goto out; | |
810 | } | |
811 | ||
757c989b TG |
812 | /* |
813 | * The caller of do_cpu_up might have raced with another | |
814 | * caller. Ignore it for now. | |
815 | */ | |
816 | if (st->state >= target) | |
38498a67 | 817 | goto out; |
757c989b TG |
818 | |
819 | if (st->state == CPUHP_OFFLINE) { | |
820 | /* Let it fail before we try to bring the cpu up */ | |
821 | idle = idle_thread_get(cpu); | |
822 | if (IS_ERR(idle)) { | |
823 | ret = PTR_ERR(idle); | |
824 | goto out; | |
825 | } | |
3bb5d2ee | 826 | } |
38498a67 | 827 | |
ba997462 TG |
828 | cpuhp_tasks_frozen = tasks_frozen; |
829 | ||
af1f4045 | 830 | st->target = target; |
1cf4f629 TG |
831 | /* |
832 | * If the current CPU state is in the range of the AP hotplug thread, | |
833 | * then we need to kick the thread once more. | |
834 | */ | |
8df3e07e | 835 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
836 | ret = cpuhp_kick_ap_work(cpu); |
837 | /* | |
838 | * The AP side has done the error rollback already. Just | |
839 | * return the error code.. | |
840 | */ | |
841 | if (ret) | |
842 | goto out; | |
843 | } | |
844 | ||
845 | /* | |
846 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 847 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
848 | * responsible for bringing it up to the target state. |
849 | */ | |
8df3e07e | 850 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 851 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 852 | out: |
8f553c49 | 853 | cpus_write_unlock(); |
e3920fb4 RW |
854 | return ret; |
855 | } | |
856 | ||
af1f4045 | 857 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
858 | { |
859 | int err = 0; | |
cf23422b | 860 | |
e0b582ec | 861 | if (!cpu_possible(cpu)) { |
84117da5 FF |
862 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
863 | cpu); | |
87d5e023 | 864 | #if defined(CONFIG_IA64) |
84117da5 | 865 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
866 | #endif |
867 | return -EINVAL; | |
868 | } | |
e3920fb4 | 869 | |
01b0f197 TK |
870 | err = try_online_node(cpu_to_node(cpu)); |
871 | if (err) | |
872 | return err; | |
cf23422b | 873 | |
d221938c | 874 | cpu_maps_update_begin(); |
e761b772 MK |
875 | |
876 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 877 | err = -EBUSY; |
e761b772 MK |
878 | goto out; |
879 | } | |
880 | ||
af1f4045 | 881 | err = _cpu_up(cpu, 0, target); |
e761b772 | 882 | out: |
d221938c | 883 | cpu_maps_update_done(); |
e3920fb4 RW |
884 | return err; |
885 | } | |
af1f4045 TG |
886 | |
887 | int cpu_up(unsigned int cpu) | |
888 | { | |
889 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
890 | } | |
a513f6ba | 891 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 892 | |
f3de4be9 | 893 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 894 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 895 | |
d391e552 | 896 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 897 | { |
d391e552 | 898 | int cpu, error = 0; |
e3920fb4 | 899 | |
d221938c | 900 | cpu_maps_update_begin(); |
d391e552 JM |
901 | if (!cpu_online(primary)) |
902 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
903 | /* |
904 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
905 | * with the userspace trying to use the CPU hotplug at the same time |
906 | */ | |
e0b582ec | 907 | cpumask_clear(frozen_cpus); |
6ad4c188 | 908 | |
84117da5 | 909 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 910 | for_each_online_cpu(cpu) { |
d391e552 | 911 | if (cpu == primary) |
e3920fb4 | 912 | continue; |
bb3632c6 | 913 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 914 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 915 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 916 | if (!error) |
e0b582ec | 917 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 918 | else { |
84117da5 | 919 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
920 | break; |
921 | } | |
922 | } | |
86886e55 | 923 | |
89af7ba5 | 924 | if (!error) |
e3920fb4 | 925 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 926 | else |
84117da5 | 927 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
928 | |
929 | /* | |
930 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
931 | * this even in case of failure as all disable_nonboot_cpus() users are | |
932 | * supposed to do enable_nonboot_cpus() on the failure path. | |
933 | */ | |
934 | cpu_hotplug_disabled++; | |
935 | ||
d221938c | 936 | cpu_maps_update_done(); |
e3920fb4 RW |
937 | return error; |
938 | } | |
939 | ||
d0af9eed SS |
940 | void __weak arch_enable_nonboot_cpus_begin(void) |
941 | { | |
942 | } | |
943 | ||
944 | void __weak arch_enable_nonboot_cpus_end(void) | |
945 | { | |
946 | } | |
947 | ||
71cf5aee | 948 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
949 | { |
950 | int cpu, error; | |
951 | ||
952 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 953 | cpu_maps_update_begin(); |
01b41159 | 954 | __cpu_hotplug_enable(); |
e0b582ec | 955 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 956 | goto out; |
e3920fb4 | 957 | |
84117da5 | 958 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
959 | |
960 | arch_enable_nonboot_cpus_begin(); | |
961 | ||
e0b582ec | 962 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 963 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 964 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 965 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 966 | if (!error) { |
84117da5 | 967 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
968 | continue; |
969 | } | |
84117da5 | 970 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 971 | } |
d0af9eed SS |
972 | |
973 | arch_enable_nonboot_cpus_end(); | |
974 | ||
e0b582ec | 975 | cpumask_clear(frozen_cpus); |
1d64b9cb | 976 | out: |
d221938c | 977 | cpu_maps_update_done(); |
1da177e4 | 978 | } |
e0b582ec | 979 | |
d7268a31 | 980 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
981 | { |
982 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
983 | return -ENOMEM; | |
984 | return 0; | |
985 | } | |
986 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 987 | |
79cfbdfa SB |
988 | /* |
989 | * When callbacks for CPU hotplug notifications are being executed, we must | |
990 | * ensure that the state of the system with respect to the tasks being frozen | |
991 | * or not, as reported by the notification, remains unchanged *throughout the | |
992 | * duration* of the execution of the callbacks. | |
993 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
994 | * | |
995 | * This synchronization is implemented by mutually excluding regular CPU | |
996 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
997 | * Hibernate notifications. | |
998 | */ | |
999 | static int | |
1000 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1001 | unsigned long action, void *ptr) | |
1002 | { | |
1003 | switch (action) { | |
1004 | ||
1005 | case PM_SUSPEND_PREPARE: | |
1006 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1007 | cpu_hotplug_disable(); |
79cfbdfa SB |
1008 | break; |
1009 | ||
1010 | case PM_POST_SUSPEND: | |
1011 | case PM_POST_HIBERNATION: | |
16e53dbf | 1012 | cpu_hotplug_enable(); |
79cfbdfa SB |
1013 | break; |
1014 | ||
1015 | default: | |
1016 | return NOTIFY_DONE; | |
1017 | } | |
1018 | ||
1019 | return NOTIFY_OK; | |
1020 | } | |
1021 | ||
1022 | ||
d7268a31 | 1023 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1024 | { |
6e32d479 FY |
1025 | /* |
1026 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1027 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1028 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1029 | */ | |
79cfbdfa SB |
1030 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1031 | return 0; | |
1032 | } | |
1033 | core_initcall(cpu_hotplug_pm_sync_init); | |
1034 | ||
f3de4be9 | 1035 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1036 | |
8ce371f9 PZ |
1037 | int __boot_cpu_id; |
1038 | ||
68f4f1ec | 1039 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1040 | |
cff7d378 TG |
1041 | /* Boot processor state steps */ |
1042 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1043 | [CPUHP_OFFLINE] = { | |
1044 | .name = "offline", | |
3c1627e9 TG |
1045 | .startup.single = NULL, |
1046 | .teardown.single = NULL, | |
cff7d378 TG |
1047 | }, |
1048 | #ifdef CONFIG_SMP | |
1049 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1050 | .name = "threads:prepare", |
3c1627e9 TG |
1051 | .startup.single = smpboot_create_threads, |
1052 | .teardown.single = NULL, | |
757c989b | 1053 | .cant_stop = true, |
cff7d378 | 1054 | }, |
00e16c3d | 1055 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1056 | .name = "perf:prepare", |
1057 | .startup.single = perf_event_init_cpu, | |
1058 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1059 | }, |
7ee681b2 | 1060 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1061 | .name = "workqueue:prepare", |
1062 | .startup.single = workqueue_prepare_cpu, | |
1063 | .teardown.single = NULL, | |
7ee681b2 | 1064 | }, |
27590dc1 | 1065 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1066 | .name = "hrtimers:prepare", |
1067 | .startup.single = hrtimers_prepare_cpu, | |
1068 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1069 | }, |
31487f83 | 1070 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1071 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1072 | .startup.single = smpcfd_prepare_cpu, |
1073 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1074 | }, |
e6d4989a RW |
1075 | [CPUHP_RELAY_PREPARE] = { |
1076 | .name = "relay:prepare", | |
1077 | .startup.single = relay_prepare_cpu, | |
1078 | .teardown.single = NULL, | |
1079 | }, | |
6731d4f1 SAS |
1080 | [CPUHP_SLAB_PREPARE] = { |
1081 | .name = "slab:prepare", | |
1082 | .startup.single = slab_prepare_cpu, | |
1083 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1084 | }, |
4df83742 | 1085 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1086 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1087 | .startup.single = rcutree_prepare_cpu, |
1088 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1089 | }, |
4fae16df RC |
1090 | /* |
1091 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1092 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1093 | * otherwise a RCU stall occurs. | |
1094 | */ | |
1095 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1096 | .name = "timers:dead", |
1097 | .startup.single = NULL, | |
1098 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1099 | }, |
d10ef6f9 | 1100 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1101 | [CPUHP_BRINGUP_CPU] = { |
1102 | .name = "cpu:bringup", | |
3c1627e9 TG |
1103 | .startup.single = bringup_cpu, |
1104 | .teardown.single = NULL, | |
757c989b | 1105 | .cant_stop = true, |
4baa0afc | 1106 | }, |
31487f83 | 1107 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1108 | .name = "smpcfd:dying", |
3c1627e9 TG |
1109 | .startup.single = NULL, |
1110 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1111 | }, |
d10ef6f9 TG |
1112 | /* |
1113 | * Handled on controll processor until the plugged processor manages | |
1114 | * this itself. | |
1115 | */ | |
4baa0afc TG |
1116 | [CPUHP_TEARDOWN_CPU] = { |
1117 | .name = "cpu:teardown", | |
3c1627e9 TG |
1118 | .startup.single = NULL, |
1119 | .teardown.single = takedown_cpu, | |
757c989b | 1120 | .cant_stop = true, |
cff7d378 | 1121 | }, |
a7c73414 TG |
1122 | #else |
1123 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1124 | #endif |
cff7d378 TG |
1125 | }; |
1126 | ||
4baa0afc TG |
1127 | /* Application processor state steps */ |
1128 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1129 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1130 | /* Final state before CPU kills itself */ |
1131 | [CPUHP_AP_IDLE_DEAD] = { | |
1132 | .name = "idle:dead", | |
1133 | }, | |
1134 | /* | |
1135 | * Last state before CPU enters the idle loop to die. Transient state | |
1136 | * for synchronization. | |
1137 | */ | |
1138 | [CPUHP_AP_OFFLINE] = { | |
1139 | .name = "ap:offline", | |
1140 | .cant_stop = true, | |
1141 | }, | |
9cf7243d TG |
1142 | /* First state is scheduler control. Interrupts are disabled */ |
1143 | [CPUHP_AP_SCHED_STARTING] = { | |
1144 | .name = "sched:starting", | |
3c1627e9 TG |
1145 | .startup.single = sched_cpu_starting, |
1146 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1147 | }, |
4df83742 | 1148 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1149 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1150 | .startup.single = NULL, |
1151 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1152 | }, |
d10ef6f9 TG |
1153 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1154 | * state for synchronsization */ | |
1155 | [CPUHP_AP_ONLINE] = { | |
1156 | .name = "ap:online", | |
1157 | }, | |
1158 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1159 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1160 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1161 | .startup.single = smpboot_unpark_threads, |
1162 | .teardown.single = NULL, | |
1cf4f629 | 1163 | }, |
c5cb83bb TG |
1164 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1165 | .name = "irq/affinity:online", | |
1166 | .startup.single = irq_affinity_online_cpu, | |
1167 | .teardown.single = NULL, | |
1168 | }, | |
00e16c3d | 1169 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1170 | .name = "perf:online", |
1171 | .startup.single = perf_event_init_cpu, | |
1172 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1173 | }, |
7ee681b2 | 1174 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1175 | .name = "workqueue:online", |
1176 | .startup.single = workqueue_online_cpu, | |
1177 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1178 | }, |
4df83742 | 1179 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1180 | .name = "RCU/tree:online", |
3c1627e9 TG |
1181 | .startup.single = rcutree_online_cpu, |
1182 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1183 | }, |
4baa0afc | 1184 | #endif |
d10ef6f9 TG |
1185 | /* |
1186 | * The dynamically registered state space is here | |
1187 | */ | |
1188 | ||
aaddd7d1 TG |
1189 | #ifdef CONFIG_SMP |
1190 | /* Last state is scheduler control setting the cpu active */ | |
1191 | [CPUHP_AP_ACTIVE] = { | |
1192 | .name = "sched:active", | |
3c1627e9 TG |
1193 | .startup.single = sched_cpu_activate, |
1194 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1195 | }, |
1196 | #endif | |
1197 | ||
d10ef6f9 | 1198 | /* CPU is fully up and running. */ |
4baa0afc TG |
1199 | [CPUHP_ONLINE] = { |
1200 | .name = "online", | |
3c1627e9 TG |
1201 | .startup.single = NULL, |
1202 | .teardown.single = NULL, | |
4baa0afc TG |
1203 | }, |
1204 | }; | |
1205 | ||
5b7aa87e TG |
1206 | /* Sanity check for callbacks */ |
1207 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1208 | { | |
1209 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1210 | return -EINVAL; | |
1211 | return 0; | |
1212 | } | |
1213 | ||
dc280d93 TG |
1214 | /* |
1215 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1216 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1217 | * by having no name assigned. | |
1218 | */ | |
1219 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1220 | { | |
4205e478 TG |
1221 | enum cpuhp_state i, end; |
1222 | struct cpuhp_step *step; | |
dc280d93 | 1223 | |
4205e478 TG |
1224 | switch (state) { |
1225 | case CPUHP_AP_ONLINE_DYN: | |
1226 | step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; | |
1227 | end = CPUHP_AP_ONLINE_DYN_END; | |
1228 | break; | |
1229 | case CPUHP_BP_PREPARE_DYN: | |
1230 | step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; | |
1231 | end = CPUHP_BP_PREPARE_DYN_END; | |
1232 | break; | |
1233 | default: | |
1234 | return -EINVAL; | |
1235 | } | |
1236 | ||
1237 | for (i = state; i <= end; i++, step++) { | |
1238 | if (!step->name) | |
dc280d93 TG |
1239 | return i; |
1240 | } | |
1241 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1242 | return -ENOSPC; | |
1243 | } | |
1244 | ||
1245 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1246 | int (*startup)(unsigned int cpu), | |
1247 | int (*teardown)(unsigned int cpu), | |
1248 | bool multi_instance) | |
5b7aa87e TG |
1249 | { |
1250 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1251 | struct cpuhp_step *sp; | |
dc280d93 | 1252 | int ret = 0; |
5b7aa87e | 1253 | |
4205e478 | 1254 | if (state == CPUHP_AP_ONLINE_DYN || state == CPUHP_BP_PREPARE_DYN) { |
dc280d93 TG |
1255 | ret = cpuhp_reserve_state(state); |
1256 | if (ret < 0) | |
dc434e05 | 1257 | return ret; |
dc280d93 TG |
1258 | state = ret; |
1259 | } | |
5b7aa87e | 1260 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1261 | if (name && sp->name) |
1262 | return -EBUSY; | |
1263 | ||
3c1627e9 TG |
1264 | sp->startup.single = startup; |
1265 | sp->teardown.single = teardown; | |
5b7aa87e | 1266 | sp->name = name; |
cf392d10 TG |
1267 | sp->multi_instance = multi_instance; |
1268 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1269 | return ret; |
5b7aa87e TG |
1270 | } |
1271 | ||
1272 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1273 | { | |
3c1627e9 | 1274 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1275 | } |
1276 | ||
5b7aa87e TG |
1277 | /* |
1278 | * Call the startup/teardown function for a step either on the AP or | |
1279 | * on the current CPU. | |
1280 | */ | |
cf392d10 TG |
1281 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1282 | struct hlist_node *node) | |
5b7aa87e | 1283 | { |
a724632c | 1284 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1285 | int ret; |
1286 | ||
3c1627e9 TG |
1287 | if ((bringup && !sp->startup.single) || |
1288 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1289 | return 0; |
5b7aa87e TG |
1290 | /* |
1291 | * The non AP bound callbacks can fail on bringup. On teardown | |
1292 | * e.g. module removal we crash for now. | |
1293 | */ | |
1cf4f629 TG |
1294 | #ifdef CONFIG_SMP |
1295 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1296 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1297 | else |
cf392d10 | 1298 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1299 | #else |
cf392d10 | 1300 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1301 | #endif |
5b7aa87e TG |
1302 | BUG_ON(ret && !bringup); |
1303 | return ret; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1308 | * | |
1309 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1310 | */ | |
1311 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1312 | struct hlist_node *node) |
5b7aa87e TG |
1313 | { |
1314 | int cpu; | |
1315 | ||
5b7aa87e TG |
1316 | /* Roll back the already executed steps on the other cpus */ |
1317 | for_each_present_cpu(cpu) { | |
1318 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1319 | int cpustate = st->state; | |
1320 | ||
1321 | if (cpu >= failedcpu) | |
1322 | break; | |
1323 | ||
1324 | /* Did we invoke the startup call on that cpu ? */ | |
1325 | if (cpustate >= state) | |
cf392d10 | 1326 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1327 | } |
1328 | } | |
1329 | ||
9805c673 TG |
1330 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1331 | struct hlist_node *node, | |
1332 | bool invoke) | |
cf392d10 TG |
1333 | { |
1334 | struct cpuhp_step *sp; | |
1335 | int cpu; | |
1336 | int ret; | |
1337 | ||
9805c673 TG |
1338 | lockdep_assert_cpus_held(); |
1339 | ||
cf392d10 TG |
1340 | sp = cpuhp_get_step(state); |
1341 | if (sp->multi_instance == false) | |
1342 | return -EINVAL; | |
1343 | ||
dc434e05 | 1344 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1345 | |
3c1627e9 | 1346 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1347 | goto add_node; |
1348 | ||
1349 | /* | |
1350 | * Try to call the startup callback for each present cpu | |
1351 | * depending on the hotplug state of the cpu. | |
1352 | */ | |
1353 | for_each_present_cpu(cpu) { | |
1354 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1355 | int cpustate = st->state; | |
1356 | ||
1357 | if (cpustate < state) | |
1358 | continue; | |
1359 | ||
1360 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1361 | if (ret) { | |
3c1627e9 | 1362 | if (sp->teardown.multi) |
cf392d10 | 1363 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1364 | goto unlock; |
cf392d10 TG |
1365 | } |
1366 | } | |
1367 | add_node: | |
1368 | ret = 0; | |
cf392d10 | 1369 | hlist_add_head(node, &sp->list); |
dc434e05 | 1370 | unlock: |
cf392d10 | 1371 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1372 | return ret; |
1373 | } | |
1374 | ||
1375 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1376 | bool invoke) | |
1377 | { | |
1378 | int ret; | |
1379 | ||
1380 | cpus_read_lock(); | |
1381 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1382 | cpus_read_unlock(); |
cf392d10 TG |
1383 | return ret; |
1384 | } | |
1385 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1386 | ||
5b7aa87e | 1387 | /** |
71def423 | 1388 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1389 | * @state: The state to setup |
1390 | * @invoke: If true, the startup function is invoked for cpus where | |
1391 | * cpu state >= @state | |
1392 | * @startup: startup callback function | |
1393 | * @teardown: teardown callback function | |
1394 | * @multi_instance: State is set up for multiple instances which get | |
1395 | * added afterwards. | |
5b7aa87e | 1396 | * |
71def423 | 1397 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1398 | * Returns: |
1399 | * On success: | |
1400 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1401 | * 0 for all other states | |
1402 | * On failure: proper (negative) error code | |
5b7aa87e | 1403 | */ |
71def423 SAS |
1404 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1405 | const char *name, bool invoke, | |
1406 | int (*startup)(unsigned int cpu), | |
1407 | int (*teardown)(unsigned int cpu), | |
1408 | bool multi_instance) | |
5b7aa87e TG |
1409 | { |
1410 | int cpu, ret = 0; | |
b9d9d691 | 1411 | bool dynstate; |
5b7aa87e | 1412 | |
71def423 SAS |
1413 | lockdep_assert_cpus_held(); |
1414 | ||
5b7aa87e TG |
1415 | if (cpuhp_cb_check(state) || !name) |
1416 | return -EINVAL; | |
1417 | ||
dc434e05 | 1418 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1419 | |
dc280d93 TG |
1420 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1421 | multi_instance); | |
5b7aa87e | 1422 | |
b9d9d691 TG |
1423 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1424 | if (ret > 0 && dynstate) { | |
1425 | state = ret; | |
1426 | ret = 0; | |
1427 | } | |
1428 | ||
dc280d93 | 1429 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1430 | goto out; |
1431 | ||
1432 | /* | |
1433 | * Try to call the startup callback for each present cpu | |
1434 | * depending on the hotplug state of the cpu. | |
1435 | */ | |
1436 | for_each_present_cpu(cpu) { | |
1437 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1438 | int cpustate = st->state; | |
1439 | ||
1440 | if (cpustate < state) | |
1441 | continue; | |
1442 | ||
cf392d10 | 1443 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1444 | if (ret) { |
a724632c | 1445 | if (teardown) |
cf392d10 TG |
1446 | cpuhp_rollback_install(cpu, state, NULL); |
1447 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1448 | goto out; |
1449 | } | |
1450 | } | |
1451 | out: | |
dc434e05 | 1452 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1453 | /* |
1454 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1455 | * dynamically allocated state in case of success. | |
1456 | */ | |
b9d9d691 | 1457 | if (!ret && dynstate) |
5b7aa87e TG |
1458 | return state; |
1459 | return ret; | |
1460 | } | |
71def423 SAS |
1461 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1462 | ||
1463 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1464 | const char *name, bool invoke, | |
1465 | int (*startup)(unsigned int cpu), | |
1466 | int (*teardown)(unsigned int cpu), | |
1467 | bool multi_instance) | |
1468 | { | |
1469 | int ret; | |
1470 | ||
1471 | cpus_read_lock(); | |
1472 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1473 | teardown, multi_instance); | |
1474 | cpus_read_unlock(); | |
1475 | return ret; | |
1476 | } | |
5b7aa87e TG |
1477 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1478 | ||
cf392d10 TG |
1479 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1480 | struct hlist_node *node, bool invoke) | |
1481 | { | |
1482 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1483 | int cpu; | |
1484 | ||
1485 | BUG_ON(cpuhp_cb_check(state)); | |
1486 | ||
1487 | if (!sp->multi_instance) | |
1488 | return -EINVAL; | |
1489 | ||
8f553c49 | 1490 | cpus_read_lock(); |
dc434e05 SAS |
1491 | mutex_lock(&cpuhp_state_mutex); |
1492 | ||
cf392d10 TG |
1493 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1494 | goto remove; | |
1495 | /* | |
1496 | * Call the teardown callback for each present cpu depending | |
1497 | * on the hotplug state of the cpu. This function is not | |
1498 | * allowed to fail currently! | |
1499 | */ | |
1500 | for_each_present_cpu(cpu) { | |
1501 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1502 | int cpustate = st->state; | |
1503 | ||
1504 | if (cpustate >= state) | |
1505 | cpuhp_issue_call(cpu, state, false, node); | |
1506 | } | |
1507 | ||
1508 | remove: | |
cf392d10 TG |
1509 | hlist_del(node); |
1510 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1511 | cpus_read_unlock(); |
cf392d10 TG |
1512 | |
1513 | return 0; | |
1514 | } | |
1515 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1516 | |
5b7aa87e | 1517 | /** |
71def423 | 1518 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1519 | * @state: The state to remove |
1520 | * @invoke: If true, the teardown function is invoked for cpus where | |
1521 | * cpu state >= @state | |
1522 | * | |
71def423 | 1523 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1524 | * The teardown callback is currently not allowed to fail. Think |
1525 | * about module removal! | |
1526 | */ | |
71def423 | 1527 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1528 | { |
cf392d10 | 1529 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1530 | int cpu; |
1531 | ||
1532 | BUG_ON(cpuhp_cb_check(state)); | |
1533 | ||
71def423 | 1534 | lockdep_assert_cpus_held(); |
5b7aa87e | 1535 | |
dc434e05 | 1536 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1537 | if (sp->multi_instance) { |
1538 | WARN(!hlist_empty(&sp->list), | |
1539 | "Error: Removing state %d which has instances left.\n", | |
1540 | state); | |
1541 | goto remove; | |
1542 | } | |
1543 | ||
a724632c | 1544 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1545 | goto remove; |
1546 | ||
1547 | /* | |
1548 | * Call the teardown callback for each present cpu depending | |
1549 | * on the hotplug state of the cpu. This function is not | |
1550 | * allowed to fail currently! | |
1551 | */ | |
1552 | for_each_present_cpu(cpu) { | |
1553 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1554 | int cpustate = st->state; | |
1555 | ||
1556 | if (cpustate >= state) | |
cf392d10 | 1557 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1558 | } |
1559 | remove: | |
cf392d10 | 1560 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1561 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1562 | } |
1563 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1564 | ||
1565 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1566 | { | |
1567 | cpus_read_lock(); | |
1568 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1569 | cpus_read_unlock(); |
5b7aa87e TG |
1570 | } |
1571 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1572 | ||
98f8cdce TG |
1573 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1574 | static ssize_t show_cpuhp_state(struct device *dev, | |
1575 | struct device_attribute *attr, char *buf) | |
1576 | { | |
1577 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1578 | ||
1579 | return sprintf(buf, "%d\n", st->state); | |
1580 | } | |
1581 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1582 | ||
757c989b TG |
1583 | static ssize_t write_cpuhp_target(struct device *dev, |
1584 | struct device_attribute *attr, | |
1585 | const char *buf, size_t count) | |
1586 | { | |
1587 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1588 | struct cpuhp_step *sp; | |
1589 | int target, ret; | |
1590 | ||
1591 | ret = kstrtoint(buf, 10, &target); | |
1592 | if (ret) | |
1593 | return ret; | |
1594 | ||
1595 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1596 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1597 | return -EINVAL; | |
1598 | #else | |
1599 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1600 | return -EINVAL; | |
1601 | #endif | |
1602 | ||
1603 | ret = lock_device_hotplug_sysfs(); | |
1604 | if (ret) | |
1605 | return ret; | |
1606 | ||
1607 | mutex_lock(&cpuhp_state_mutex); | |
1608 | sp = cpuhp_get_step(target); | |
1609 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1610 | mutex_unlock(&cpuhp_state_mutex); | |
1611 | if (ret) | |
40da1b11 | 1612 | goto out; |
757c989b TG |
1613 | |
1614 | if (st->state < target) | |
1615 | ret = do_cpu_up(dev->id, target); | |
1616 | else | |
1617 | ret = do_cpu_down(dev->id, target); | |
40da1b11 | 1618 | out: |
757c989b TG |
1619 | unlock_device_hotplug(); |
1620 | return ret ? ret : count; | |
1621 | } | |
1622 | ||
98f8cdce TG |
1623 | static ssize_t show_cpuhp_target(struct device *dev, |
1624 | struct device_attribute *attr, char *buf) | |
1625 | { | |
1626 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1627 | ||
1628 | return sprintf(buf, "%d\n", st->target); | |
1629 | } | |
757c989b | 1630 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1631 | |
1632 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1633 | &dev_attr_state.attr, | |
1634 | &dev_attr_target.attr, | |
1635 | NULL | |
1636 | }; | |
1637 | ||
993647a2 | 1638 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
1639 | .attrs = cpuhp_cpu_attrs, |
1640 | .name = "hotplug", | |
1641 | NULL | |
1642 | }; | |
1643 | ||
1644 | static ssize_t show_cpuhp_states(struct device *dev, | |
1645 | struct device_attribute *attr, char *buf) | |
1646 | { | |
1647 | ssize_t cur, res = 0; | |
1648 | int i; | |
1649 | ||
1650 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1651 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1652 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1653 | ||
1654 | if (sp->name) { | |
1655 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1656 | buf += cur; | |
1657 | res += cur; | |
1658 | } | |
1659 | } | |
1660 | mutex_unlock(&cpuhp_state_mutex); | |
1661 | return res; | |
1662 | } | |
1663 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1664 | ||
1665 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1666 | &dev_attr_states.attr, | |
1667 | NULL | |
1668 | }; | |
1669 | ||
993647a2 | 1670 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
1671 | .attrs = cpuhp_cpu_root_attrs, |
1672 | .name = "hotplug", | |
1673 | NULL | |
1674 | }; | |
1675 | ||
1676 | static int __init cpuhp_sysfs_init(void) | |
1677 | { | |
1678 | int cpu, ret; | |
1679 | ||
1680 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1681 | &cpuhp_cpu_root_attr_group); | |
1682 | if (ret) | |
1683 | return ret; | |
1684 | ||
1685 | for_each_possible_cpu(cpu) { | |
1686 | struct device *dev = get_cpu_device(cpu); | |
1687 | ||
1688 | if (!dev) | |
1689 | continue; | |
1690 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1691 | if (ret) | |
1692 | return ret; | |
1693 | } | |
1694 | return 0; | |
1695 | } | |
1696 | device_initcall(cpuhp_sysfs_init); | |
1697 | #endif | |
1698 | ||
e56b3bc7 LT |
1699 | /* |
1700 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1701 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1702 | * | |
e0b582ec | 1703 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1704 | * mask value that has a single bit set only. |
1705 | */ | |
b8d317d1 | 1706 | |
e56b3bc7 | 1707 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1708 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1709 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1710 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1711 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1712 | |
e56b3bc7 LT |
1713 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1714 | ||
1715 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1716 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1717 | #if BITS_PER_LONG > 32 | |
1718 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1719 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1720 | #endif |
1721 | }; | |
e56b3bc7 | 1722 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1723 | |
1724 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1725 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1726 | |
1727 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1728 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1729 | = {CPU_BITS_ALL}; |
b3199c02 | 1730 | #else |
4b804c85 | 1731 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1732 | #endif |
4b804c85 | 1733 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1734 | |
4b804c85 RV |
1735 | struct cpumask __cpu_online_mask __read_mostly; |
1736 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1737 | |
4b804c85 RV |
1738 | struct cpumask __cpu_present_mask __read_mostly; |
1739 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1740 | |
4b804c85 RV |
1741 | struct cpumask __cpu_active_mask __read_mostly; |
1742 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1743 | |
3fa41520 RR |
1744 | void init_cpu_present(const struct cpumask *src) |
1745 | { | |
c4c54dd1 | 1746 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1747 | } |
1748 | ||
1749 | void init_cpu_possible(const struct cpumask *src) | |
1750 | { | |
c4c54dd1 | 1751 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1752 | } |
1753 | ||
1754 | void init_cpu_online(const struct cpumask *src) | |
1755 | { | |
c4c54dd1 | 1756 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1757 | } |
cff7d378 TG |
1758 | |
1759 | /* | |
1760 | * Activate the first processor. | |
1761 | */ | |
1762 | void __init boot_cpu_init(void) | |
1763 | { | |
1764 | int cpu = smp_processor_id(); | |
1765 | ||
1766 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1767 | set_cpu_online(cpu, true); | |
1768 | set_cpu_active(cpu, true); | |
1769 | set_cpu_present(cpu, true); | |
1770 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
1771 | |
1772 | #ifdef CONFIG_SMP | |
1773 | __boot_cpu_id = cpu; | |
1774 | #endif | |
cff7d378 TG |
1775 | } |
1776 | ||
1777 | /* | |
1778 | * Must be called _AFTER_ setting up the per_cpu areas | |
1779 | */ | |
1780 | void __init boot_cpu_state_init(void) | |
1781 | { | |
1782 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1783 | } |