]>
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 | ||
69 | /** | |
70 | * cpuhp_step - Hotplug state machine step | |
71 | * @name: Name of the step | |
72 | * @startup: Startup function of the step | |
73 | * @teardown: Teardown function of the step | |
74 | * @skip_onerr: Do not invoke the functions on error rollback | |
75 | * Will go away once the notifiers are gone | |
757c989b | 76 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
77 | */ |
78 | struct cpuhp_step { | |
cf392d10 TG |
79 | const char *name; |
80 | union { | |
3c1627e9 TG |
81 | int (*single)(unsigned int cpu); |
82 | int (*multi)(unsigned int cpu, | |
83 | struct hlist_node *node); | |
84 | } startup; | |
cf392d10 | 85 | union { |
3c1627e9 TG |
86 | int (*single)(unsigned int cpu); |
87 | int (*multi)(unsigned int cpu, | |
88 | struct hlist_node *node); | |
89 | } teardown; | |
cf392d10 TG |
90 | struct hlist_head list; |
91 | bool skip_onerr; | |
92 | bool cant_stop; | |
93 | bool multi_instance; | |
cff7d378 TG |
94 | }; |
95 | ||
98f8cdce | 96 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 97 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 98 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 99 | |
a724632c TG |
100 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
101 | { | |
102 | /* | |
103 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
104 | * purposes as that state is handled explicitly in cpu_down. | |
105 | */ | |
106 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
107 | } | |
108 | ||
109 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
110 | { | |
111 | struct cpuhp_step *sp; | |
112 | ||
113 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
114 | return sp + state; | |
115 | } | |
116 | ||
cff7d378 TG |
117 | /** |
118 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
119 | * @cpu: The cpu for which the callback should be invoked | |
120 | * @step: The step in the state machine | |
a724632c | 121 | * @bringup: True if the bringup callback should be invoked |
cff7d378 | 122 | * |
cf392d10 | 123 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 124 | */ |
a724632c | 125 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
cf392d10 | 126 | bool bringup, struct hlist_node *node) |
cff7d378 TG |
127 | { |
128 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 129 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
130 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
131 | int (*cb)(unsigned int cpu); | |
132 | int ret, cnt; | |
133 | ||
134 | if (!step->multi_instance) { | |
3c1627e9 | 135 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
136 | if (!cb) |
137 | return 0; | |
a724632c | 138 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 139 | ret = cb(cpu); |
a724632c | 140 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
141 | return ret; |
142 | } | |
3c1627e9 | 143 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
144 | if (!cbm) |
145 | return 0; | |
146 | ||
147 | /* Single invocation for instance add/remove */ | |
148 | if (node) { | |
149 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
150 | ret = cbm(cpu, node); | |
151 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
152 | return ret; | |
153 | } | |
154 | ||
155 | /* State transition. Invoke on all instances */ | |
156 | cnt = 0; | |
157 | hlist_for_each(node, &step->list) { | |
158 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
159 | ret = cbm(cpu, node); | |
160 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
161 | if (ret) | |
162 | goto err; | |
163 | cnt++; | |
164 | } | |
165 | return 0; | |
166 | err: | |
167 | /* Rollback the instances if one failed */ | |
3c1627e9 | 168 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
169 | if (!cbm) |
170 | return ret; | |
171 | ||
172 | hlist_for_each(node, &step->list) { | |
173 | if (!cnt--) | |
174 | break; | |
175 | cbm(cpu, node); | |
cff7d378 TG |
176 | } |
177 | return ret; | |
178 | } | |
179 | ||
98a79d6a | 180 | #ifdef CONFIG_SMP |
b3199c02 | 181 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 182 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
183 | bool cpuhp_tasks_frozen; |
184 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 185 | |
79a6cdeb | 186 | /* |
93ae4f97 SB |
187 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
188 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
189 | */ |
190 | void cpu_maps_update_begin(void) | |
191 | { | |
192 | mutex_lock(&cpu_add_remove_lock); | |
193 | } | |
194 | ||
195 | void cpu_maps_update_done(void) | |
196 | { | |
197 | mutex_unlock(&cpu_add_remove_lock); | |
198 | } | |
1da177e4 | 199 | |
fc8dffd3 TG |
200 | /* |
201 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
202 | * Should always be manipulated under cpu_add_remove_lock |
203 | */ | |
204 | static int cpu_hotplug_disabled; | |
205 | ||
79a6cdeb LJ |
206 | #ifdef CONFIG_HOTPLUG_CPU |
207 | ||
fc8dffd3 | 208 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
62db99f4 | 209 | |
8f553c49 | 210 | void cpus_read_lock(void) |
a9d9baa1 | 211 | { |
fc8dffd3 | 212 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 213 | } |
8f553c49 | 214 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 215 | |
8f553c49 | 216 | void cpus_read_unlock(void) |
a9d9baa1 | 217 | { |
fc8dffd3 | 218 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 219 | } |
8f553c49 | 220 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 221 | |
8f553c49 | 222 | void cpus_write_lock(void) |
d221938c | 223 | { |
fc8dffd3 | 224 | percpu_down_write(&cpu_hotplug_lock); |
d221938c GS |
225 | } |
226 | ||
8f553c49 | 227 | void cpus_write_unlock(void) |
d221938c | 228 | { |
fc8dffd3 TG |
229 | percpu_up_write(&cpu_hotplug_lock); |
230 | } | |
231 | ||
232 | void lockdep_assert_cpus_held(void) | |
233 | { | |
234 | percpu_rwsem_assert_held(&cpu_hotplug_lock); | |
d221938c | 235 | } |
79a6cdeb | 236 | |
16e53dbf SB |
237 | /* |
238 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
239 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
240 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
241 | * hotplug path before performing hotplug operations. So acquiring that lock | |
242 | * guarantees mutual exclusion from any currently running hotplug operations. | |
243 | */ | |
244 | void cpu_hotplug_disable(void) | |
245 | { | |
246 | cpu_maps_update_begin(); | |
89af7ba5 | 247 | cpu_hotplug_disabled++; |
16e53dbf SB |
248 | cpu_maps_update_done(); |
249 | } | |
32145c46 | 250 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 251 | |
01b41159 LW |
252 | static void __cpu_hotplug_enable(void) |
253 | { | |
254 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
255 | return; | |
256 | cpu_hotplug_disabled--; | |
257 | } | |
258 | ||
16e53dbf SB |
259 | void cpu_hotplug_enable(void) |
260 | { | |
261 | cpu_maps_update_begin(); | |
01b41159 | 262 | __cpu_hotplug_enable(); |
16e53dbf SB |
263 | cpu_maps_update_done(); |
264 | } | |
32145c46 | 265 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 266 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 267 | |
8df3e07e TG |
268 | static int bringup_wait_for_ap(unsigned int cpu) |
269 | { | |
270 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
271 | ||
272 | wait_for_completion(&st->done); | |
273 | return st->result; | |
274 | } | |
275 | ||
ba997462 TG |
276 | static int bringup_cpu(unsigned int cpu) |
277 | { | |
278 | struct task_struct *idle = idle_thread_get(cpu); | |
279 | int ret; | |
280 | ||
aa877175 BO |
281 | /* |
282 | * Some architectures have to walk the irq descriptors to | |
283 | * setup the vector space for the cpu which comes online. | |
284 | * Prevent irq alloc/free across the bringup. | |
285 | */ | |
286 | irq_lock_sparse(); | |
287 | ||
ba997462 TG |
288 | /* Arch-specific enabling code. */ |
289 | ret = __cpu_up(cpu, idle); | |
aa877175 | 290 | irq_unlock_sparse(); |
530e9b76 | 291 | if (ret) |
ba997462 | 292 | return ret; |
8df3e07e | 293 | ret = bringup_wait_for_ap(cpu); |
ba997462 | 294 | BUG_ON(!cpu_online(cpu)); |
8df3e07e | 295 | return ret; |
ba997462 TG |
296 | } |
297 | ||
2e1a3483 TG |
298 | /* |
299 | * Hotplug state machine related functions | |
300 | */ | |
a724632c | 301 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
302 | { |
303 | for (st->state++; st->state < st->target; st->state++) { | |
a724632c | 304 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
305 | |
306 | if (!step->skip_onerr) | |
cf392d10 | 307 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
308 | } |
309 | } | |
310 | ||
311 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 312 | enum cpuhp_state target) |
2e1a3483 TG |
313 | { |
314 | enum cpuhp_state prev_state = st->state; | |
315 | int ret = 0; | |
316 | ||
317 | for (; st->state > target; st->state--) { | |
cf392d10 | 318 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
319 | if (ret) { |
320 | st->target = prev_state; | |
a724632c | 321 | undo_cpu_down(cpu, st); |
2e1a3483 TG |
322 | break; |
323 | } | |
324 | } | |
325 | return ret; | |
326 | } | |
327 | ||
a724632c | 328 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
329 | { |
330 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 331 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
332 | |
333 | if (!step->skip_onerr) | |
cf392d10 | 334 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
2e1a3483 TG |
335 | } |
336 | } | |
337 | ||
338 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 339 | enum cpuhp_state target) |
2e1a3483 TG |
340 | { |
341 | enum cpuhp_state prev_state = st->state; | |
342 | int ret = 0; | |
343 | ||
344 | while (st->state < target) { | |
2e1a3483 | 345 | st->state++; |
cf392d10 | 346 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL); |
2e1a3483 TG |
347 | if (ret) { |
348 | st->target = prev_state; | |
a724632c | 349 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
350 | break; |
351 | } | |
352 | } | |
353 | return ret; | |
354 | } | |
355 | ||
4cb28ced TG |
356 | /* |
357 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
358 | */ | |
359 | static void cpuhp_create(unsigned int cpu) | |
360 | { | |
361 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
362 | ||
363 | init_completion(&st->done); | |
364 | } | |
365 | ||
366 | static int cpuhp_should_run(unsigned int cpu) | |
367 | { | |
368 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
369 | ||
370 | return st->should_run; | |
371 | } | |
372 | ||
373 | /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */ | |
374 | static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st) | |
375 | { | |
1cf4f629 | 376 | enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU); |
4cb28ced | 377 | |
a724632c | 378 | return cpuhp_down_callbacks(cpu, st, target); |
4cb28ced TG |
379 | } |
380 | ||
381 | /* Execute the online startup callbacks. Used to be CPU_ONLINE */ | |
382 | static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st) | |
383 | { | |
a724632c | 384 | return cpuhp_up_callbacks(cpu, st, st->target); |
4cb28ced TG |
385 | } |
386 | ||
387 | /* | |
388 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
389 | * callbacks when a state gets [un]installed at runtime. | |
390 | */ | |
391 | static void cpuhp_thread_fun(unsigned int cpu) | |
392 | { | |
393 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
394 | int ret = 0; | |
395 | ||
396 | /* | |
397 | * Paired with the mb() in cpuhp_kick_ap_work and | |
398 | * cpuhp_invoke_ap_callback, so the work set is consistent visible. | |
399 | */ | |
400 | smp_mb(); | |
401 | if (!st->should_run) | |
402 | return; | |
403 | ||
404 | st->should_run = false; | |
405 | ||
406 | /* Single callback invocation for [un]install ? */ | |
a724632c | 407 | if (st->single) { |
4cb28ced TG |
408 | if (st->cb_state < CPUHP_AP_ONLINE) { |
409 | local_irq_disable(); | |
a724632c | 410 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 411 | st->bringup, st->node); |
4cb28ced TG |
412 | local_irq_enable(); |
413 | } else { | |
a724632c | 414 | ret = cpuhp_invoke_callback(cpu, st->cb_state, |
cf392d10 | 415 | st->bringup, st->node); |
4cb28ced | 416 | } |
3b9d6da6 SAS |
417 | } else if (st->rollback) { |
418 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); | |
419 | ||
a724632c | 420 | undo_cpu_down(cpu, st); |
3b9d6da6 | 421 | st->rollback = false; |
4cb28ced | 422 | } else { |
1cf4f629 | 423 | /* Cannot happen .... */ |
8df3e07e | 424 | BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE); |
1cf4f629 | 425 | |
4cb28ced TG |
426 | /* Regular hotplug work */ |
427 | if (st->state < st->target) | |
428 | ret = cpuhp_ap_online(cpu, st); | |
429 | else if (st->state > st->target) | |
430 | ret = cpuhp_ap_offline(cpu, st); | |
431 | } | |
432 | st->result = ret; | |
433 | complete(&st->done); | |
434 | } | |
435 | ||
436 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 437 | static int |
cf392d10 TG |
438 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
439 | struct hlist_node *node) | |
4cb28ced TG |
440 | { |
441 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
442 | ||
443 | if (!cpu_online(cpu)) | |
444 | return 0; | |
445 | ||
6a4e2451 TG |
446 | /* |
447 | * If we are up and running, use the hotplug thread. For early calls | |
448 | * we invoke the thread function directly. | |
449 | */ | |
450 | if (!st->thread) | |
cf392d10 | 451 | return cpuhp_invoke_callback(cpu, state, bringup, node); |
6a4e2451 | 452 | |
4cb28ced | 453 | st->cb_state = state; |
a724632c TG |
454 | st->single = true; |
455 | st->bringup = bringup; | |
cf392d10 | 456 | st->node = node; |
a724632c | 457 | |
4cb28ced TG |
458 | /* |
459 | * Make sure the above stores are visible before should_run becomes | |
460 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
461 | */ | |
462 | smp_mb(); | |
463 | st->should_run = true; | |
464 | wake_up_process(st->thread); | |
465 | wait_for_completion(&st->done); | |
466 | return st->result; | |
467 | } | |
468 | ||
469 | /* Regular hotplug invocation of the AP hotplug thread */ | |
1cf4f629 | 470 | static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st) |
4cb28ced | 471 | { |
4cb28ced | 472 | st->result = 0; |
a724632c | 473 | st->single = false; |
4cb28ced TG |
474 | /* |
475 | * Make sure the above stores are visible before should_run becomes | |
476 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
477 | */ | |
478 | smp_mb(); | |
479 | st->should_run = true; | |
480 | wake_up_process(st->thread); | |
1cf4f629 TG |
481 | } |
482 | ||
483 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
484 | { | |
485 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
486 | enum cpuhp_state state = st->state; | |
487 | ||
488 | trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work); | |
489 | __cpuhp_kick_ap_work(st); | |
4cb28ced TG |
490 | wait_for_completion(&st->done); |
491 | trace_cpuhp_exit(cpu, st->state, state, st->result); | |
492 | return st->result; | |
493 | } | |
494 | ||
495 | static struct smp_hotplug_thread cpuhp_threads = { | |
496 | .store = &cpuhp_state.thread, | |
497 | .create = &cpuhp_create, | |
498 | .thread_should_run = cpuhp_should_run, | |
499 | .thread_fn = cpuhp_thread_fun, | |
500 | .thread_comm = "cpuhp/%u", | |
501 | .selfparking = true, | |
502 | }; | |
503 | ||
504 | void __init cpuhp_threads_init(void) | |
505 | { | |
506 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
507 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
508 | } | |
509 | ||
777c6e0d | 510 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
511 | /** |
512 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
513 | * @cpu: a CPU id | |
514 | * | |
515 | * This function walks all processes, finds a valid mm struct for each one and | |
516 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
517 | * trivial, there are various non-obvious corner cases, which this function | |
518 | * tries to solve in a safe manner. | |
519 | * | |
520 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
521 | * be called only for an already offlined CPU. | |
522 | */ | |
cb79295e AV |
523 | void clear_tasks_mm_cpumask(int cpu) |
524 | { | |
525 | struct task_struct *p; | |
526 | ||
527 | /* | |
528 | * This function is called after the cpu is taken down and marked | |
529 | * offline, so its not like new tasks will ever get this cpu set in | |
530 | * their mm mask. -- Peter Zijlstra | |
531 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
532 | * full-fledged tasklist_lock. | |
533 | */ | |
e4cc2f87 | 534 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
535 | rcu_read_lock(); |
536 | for_each_process(p) { | |
537 | struct task_struct *t; | |
538 | ||
e4cc2f87 AV |
539 | /* |
540 | * Main thread might exit, but other threads may still have | |
541 | * a valid mm. Find one. | |
542 | */ | |
cb79295e AV |
543 | t = find_lock_task_mm(p); |
544 | if (!t) | |
545 | continue; | |
546 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
547 | task_unlock(t); | |
548 | } | |
549 | rcu_read_unlock(); | |
550 | } | |
551 | ||
b728ca06 | 552 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 553 | { |
b728ca06 | 554 | struct task_struct *g, *p; |
1da177e4 | 555 | |
a75a6068 ON |
556 | read_lock(&tasklist_lock); |
557 | for_each_process_thread(g, p) { | |
b728ca06 KT |
558 | if (!p->on_rq) |
559 | continue; | |
560 | /* | |
561 | * We do the check with unlocked task_rq(p)->lock. | |
562 | * Order the reading to do not warn about a task, | |
563 | * which was running on this cpu in the past, and | |
564 | * it's just been woken on another cpu. | |
565 | */ | |
566 | rmb(); | |
567 | if (task_cpu(p) != dead_cpu) | |
568 | continue; | |
569 | ||
570 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
571 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
572 | } |
573 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
574 | } |
575 | ||
576 | /* Take this CPU down. */ | |
71cf5aee | 577 | static int take_cpu_down(void *_param) |
1da177e4 | 578 | { |
4baa0afc TG |
579 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
580 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 581 | int err, cpu = smp_processor_id(); |
1da177e4 | 582 | |
1da177e4 LT |
583 | /* Ensure this CPU doesn't handle any more interrupts. */ |
584 | err = __cpu_disable(); | |
585 | if (err < 0) | |
f3705136 | 586 | return err; |
1da177e4 | 587 | |
a724632c TG |
588 | /* |
589 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
590 | * do this step again. | |
591 | */ | |
592 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
593 | st->state--; | |
4baa0afc | 594 | /* Invoke the former CPU_DYING callbacks */ |
a724632c | 595 | for (; st->state > target; st->state--) |
cf392d10 | 596 | cpuhp_invoke_callback(cpu, st->state, false, NULL); |
4baa0afc | 597 | |
52c063d1 TG |
598 | /* Give up timekeeping duties */ |
599 | tick_handover_do_timer(); | |
14e568e7 | 600 | /* Park the stopper thread */ |
090e77c3 | 601 | stop_machine_park(cpu); |
f3705136 | 602 | return 0; |
1da177e4 LT |
603 | } |
604 | ||
98458172 | 605 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 606 | { |
e69aab13 | 607 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 608 | int err; |
1da177e4 | 609 | |
2a58c527 | 610 | /* Park the smpboot threads */ |
1cf4f629 | 611 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 612 | smpboot_park_threads(cpu); |
1cf4f629 | 613 | |
6acce3ef | 614 | /* |
a8994181 TG |
615 | * Prevent irq alloc/free while the dying cpu reorganizes the |
616 | * interrupt affinities. | |
6acce3ef | 617 | */ |
a8994181 | 618 | irq_lock_sparse(); |
6acce3ef | 619 | |
a8994181 TG |
620 | /* |
621 | * So now all preempt/rcu users must observe !cpu_active(). | |
622 | */ | |
210e2133 | 623 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 624 | if (err) { |
3b9d6da6 | 625 | /* CPU refused to die */ |
a8994181 | 626 | irq_unlock_sparse(); |
3b9d6da6 SAS |
627 | /* Unpark the hotplug thread so we can rollback there */ |
628 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 629 | return err; |
8fa1d7d3 | 630 | } |
04321587 | 631 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 632 | |
48c5ccae | 633 | /* |
ee1e714b | 634 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
635 | * runnable tasks from the cpu, there's only the idle task left now |
636 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
637 | * |
638 | * Wait for the stop thread to go away. | |
48c5ccae | 639 | */ |
e69aab13 TG |
640 | wait_for_completion(&st->done); |
641 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 642 | |
a8994181 TG |
643 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
644 | irq_unlock_sparse(); | |
645 | ||
345527b1 | 646 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
647 | /* This actually kills the CPU. */ |
648 | __cpu_die(cpu); | |
649 | ||
a49b116d | 650 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
651 | return 0; |
652 | } | |
1da177e4 | 653 | |
71f87b2f TG |
654 | static void cpuhp_complete_idle_dead(void *arg) |
655 | { | |
656 | struct cpuhp_cpu_state *st = arg; | |
657 | ||
658 | complete(&st->done); | |
659 | } | |
660 | ||
e69aab13 TG |
661 | void cpuhp_report_idle_dead(void) |
662 | { | |
663 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
664 | ||
665 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 666 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
667 | st->state = CPUHP_AP_IDLE_DEAD; |
668 | /* | |
669 | * We cannot call complete after rcu_report_dead() so we delegate it | |
670 | * to an online cpu. | |
671 | */ | |
672 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
673 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
674 | } |
675 | ||
cff7d378 | 676 | #else |
cff7d378 | 677 | #define takedown_cpu NULL |
cff7d378 TG |
678 | #endif |
679 | ||
680 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 681 | |
98458172 | 682 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
683 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
684 | enum cpuhp_state target) | |
98458172 | 685 | { |
cff7d378 TG |
686 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
687 | int prev_state, ret = 0; | |
98458172 TG |
688 | |
689 | if (num_online_cpus() == 1) | |
690 | return -EBUSY; | |
691 | ||
757c989b | 692 | if (!cpu_present(cpu)) |
98458172 TG |
693 | return -EINVAL; |
694 | ||
8f553c49 | 695 | cpus_write_lock(); |
98458172 TG |
696 | |
697 | cpuhp_tasks_frozen = tasks_frozen; | |
698 | ||
cff7d378 | 699 | prev_state = st->state; |
af1f4045 | 700 | st->target = target; |
1cf4f629 TG |
701 | /* |
702 | * If the current CPU state is in the range of the AP hotplug thread, | |
703 | * then we need to kick the thread. | |
704 | */ | |
8df3e07e | 705 | if (st->state > CPUHP_TEARDOWN_CPU) { |
1cf4f629 TG |
706 | ret = cpuhp_kick_ap_work(cpu); |
707 | /* | |
708 | * The AP side has done the error rollback already. Just | |
709 | * return the error code.. | |
710 | */ | |
711 | if (ret) | |
712 | goto out; | |
713 | ||
714 | /* | |
715 | * We might have stopped still in the range of the AP hotplug | |
716 | * thread. Nothing to do anymore. | |
717 | */ | |
8df3e07e | 718 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 TG |
719 | goto out; |
720 | } | |
721 | /* | |
8df3e07e | 722 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
723 | * to do the further cleanups. |
724 | */ | |
a724632c | 725 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 SAS |
726 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
727 | st->target = prev_state; | |
728 | st->rollback = true; | |
729 | cpuhp_kick_ap_work(cpu); | |
730 | } | |
98458172 | 731 | |
1cf4f629 | 732 | out: |
8f553c49 | 733 | cpus_write_unlock(); |
cff7d378 | 734 | return ret; |
e3920fb4 RW |
735 | } |
736 | ||
af1f4045 | 737 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 738 | { |
9ea09af3 | 739 | int err; |
e3920fb4 | 740 | |
d221938c | 741 | cpu_maps_update_begin(); |
e761b772 MK |
742 | |
743 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 744 | err = -EBUSY; |
e761b772 MK |
745 | goto out; |
746 | } | |
747 | ||
af1f4045 | 748 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 749 | |
e761b772 | 750 | out: |
d221938c | 751 | cpu_maps_update_done(); |
1da177e4 LT |
752 | return err; |
753 | } | |
af1f4045 TG |
754 | int cpu_down(unsigned int cpu) |
755 | { | |
756 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
757 | } | |
b62b8ef9 | 758 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
759 | #endif /*CONFIG_HOTPLUG_CPU*/ |
760 | ||
4baa0afc | 761 | /** |
ee1e714b | 762 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
763 | * @cpu: cpu that just started |
764 | * | |
4baa0afc TG |
765 | * It must be called by the arch code on the new cpu, before the new cpu |
766 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
767 | */ | |
768 | void notify_cpu_starting(unsigned int cpu) | |
769 | { | |
770 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
771 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
772 | ||
0c6d4576 | 773 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 774 | while (st->state < target) { |
4baa0afc | 775 | st->state++; |
cf392d10 | 776 | cpuhp_invoke_callback(cpu, st->state, true, NULL); |
4baa0afc TG |
777 | } |
778 | } | |
779 | ||
949338e3 TG |
780 | /* |
781 | * Called from the idle task. We need to set active here, so we can kick off | |
8df3e07e TG |
782 | * the stopper thread and unpark the smpboot threads. If the target state is |
783 | * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the | |
784 | * cpu further. | |
949338e3 | 785 | */ |
8df3e07e | 786 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 787 | { |
8df3e07e TG |
788 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
789 | unsigned int cpu = smp_processor_id(); | |
790 | ||
791 | /* Happens for the boot cpu */ | |
792 | if (state != CPUHP_AP_ONLINE_IDLE) | |
793 | return; | |
794 | ||
795 | st->state = CPUHP_AP_ONLINE_IDLE; | |
1cf4f629 | 796 | |
8df3e07e | 797 | /* Unpark the stopper thread and the hotplug thread of this cpu */ |
949338e3 | 798 | stop_machine_unpark(cpu); |
1cf4f629 | 799 | kthread_unpark(st->thread); |
8df3e07e TG |
800 | |
801 | /* Should we go further up ? */ | |
802 | if (st->target > CPUHP_AP_ONLINE_IDLE) | |
803 | __cpuhp_kick_ap_work(st); | |
804 | else | |
805 | complete(&st->done); | |
949338e3 TG |
806 | } |
807 | ||
e3920fb4 | 808 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 809 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 810 | { |
cff7d378 | 811 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 812 | struct task_struct *idle; |
2e1a3483 | 813 | int ret = 0; |
1da177e4 | 814 | |
8f553c49 | 815 | cpus_write_lock(); |
38498a67 | 816 | |
757c989b | 817 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
818 | ret = -EINVAL; |
819 | goto out; | |
820 | } | |
821 | ||
757c989b TG |
822 | /* |
823 | * The caller of do_cpu_up might have raced with another | |
824 | * caller. Ignore it for now. | |
825 | */ | |
826 | if (st->state >= target) | |
38498a67 | 827 | goto out; |
757c989b TG |
828 | |
829 | if (st->state == CPUHP_OFFLINE) { | |
830 | /* Let it fail before we try to bring the cpu up */ | |
831 | idle = idle_thread_get(cpu); | |
832 | if (IS_ERR(idle)) { | |
833 | ret = PTR_ERR(idle); | |
834 | goto out; | |
835 | } | |
3bb5d2ee | 836 | } |
38498a67 | 837 | |
ba997462 TG |
838 | cpuhp_tasks_frozen = tasks_frozen; |
839 | ||
af1f4045 | 840 | st->target = target; |
1cf4f629 TG |
841 | /* |
842 | * If the current CPU state is in the range of the AP hotplug thread, | |
843 | * then we need to kick the thread once more. | |
844 | */ | |
8df3e07e | 845 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
846 | ret = cpuhp_kick_ap_work(cpu); |
847 | /* | |
848 | * The AP side has done the error rollback already. Just | |
849 | * return the error code.. | |
850 | */ | |
851 | if (ret) | |
852 | goto out; | |
853 | } | |
854 | ||
855 | /* | |
856 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 857 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
858 | * responsible for bringing it up to the target state. |
859 | */ | |
8df3e07e | 860 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 861 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 862 | out: |
8f553c49 | 863 | cpus_write_unlock(); |
e3920fb4 RW |
864 | return ret; |
865 | } | |
866 | ||
af1f4045 | 867 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
868 | { |
869 | int err = 0; | |
cf23422b | 870 | |
e0b582ec | 871 | if (!cpu_possible(cpu)) { |
84117da5 FF |
872 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
873 | cpu); | |
87d5e023 | 874 | #if defined(CONFIG_IA64) |
84117da5 | 875 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
876 | #endif |
877 | return -EINVAL; | |
878 | } | |
e3920fb4 | 879 | |
01b0f197 TK |
880 | err = try_online_node(cpu_to_node(cpu)); |
881 | if (err) | |
882 | return err; | |
cf23422b | 883 | |
d221938c | 884 | cpu_maps_update_begin(); |
e761b772 MK |
885 | |
886 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 887 | err = -EBUSY; |
e761b772 MK |
888 | goto out; |
889 | } | |
890 | ||
af1f4045 | 891 | err = _cpu_up(cpu, 0, target); |
e761b772 | 892 | out: |
d221938c | 893 | cpu_maps_update_done(); |
e3920fb4 RW |
894 | return err; |
895 | } | |
af1f4045 TG |
896 | |
897 | int cpu_up(unsigned int cpu) | |
898 | { | |
899 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
900 | } | |
a513f6ba | 901 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 902 | |
f3de4be9 | 903 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 904 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 905 | |
d391e552 | 906 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 907 | { |
d391e552 | 908 | int cpu, error = 0; |
e3920fb4 | 909 | |
d221938c | 910 | cpu_maps_update_begin(); |
d391e552 JM |
911 | if (!cpu_online(primary)) |
912 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
913 | /* |
914 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
915 | * with the userspace trying to use the CPU hotplug at the same time |
916 | */ | |
e0b582ec | 917 | cpumask_clear(frozen_cpus); |
6ad4c188 | 918 | |
84117da5 | 919 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 920 | for_each_online_cpu(cpu) { |
d391e552 | 921 | if (cpu == primary) |
e3920fb4 | 922 | continue; |
bb3632c6 | 923 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 924 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 925 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 926 | if (!error) |
e0b582ec | 927 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 928 | else { |
84117da5 | 929 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
930 | break; |
931 | } | |
932 | } | |
86886e55 | 933 | |
89af7ba5 | 934 | if (!error) |
e3920fb4 | 935 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 936 | else |
84117da5 | 937 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
938 | |
939 | /* | |
940 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
941 | * this even in case of failure as all disable_nonboot_cpus() users are | |
942 | * supposed to do enable_nonboot_cpus() on the failure path. | |
943 | */ | |
944 | cpu_hotplug_disabled++; | |
945 | ||
d221938c | 946 | cpu_maps_update_done(); |
e3920fb4 RW |
947 | return error; |
948 | } | |
949 | ||
d0af9eed SS |
950 | void __weak arch_enable_nonboot_cpus_begin(void) |
951 | { | |
952 | } | |
953 | ||
954 | void __weak arch_enable_nonboot_cpus_end(void) | |
955 | { | |
956 | } | |
957 | ||
71cf5aee | 958 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
959 | { |
960 | int cpu, error; | |
961 | ||
962 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 963 | cpu_maps_update_begin(); |
01b41159 | 964 | __cpu_hotplug_enable(); |
e0b582ec | 965 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 966 | goto out; |
e3920fb4 | 967 | |
84117da5 | 968 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
969 | |
970 | arch_enable_nonboot_cpus_begin(); | |
971 | ||
e0b582ec | 972 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 973 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 974 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 975 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 976 | if (!error) { |
84117da5 | 977 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
978 | continue; |
979 | } | |
84117da5 | 980 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 981 | } |
d0af9eed SS |
982 | |
983 | arch_enable_nonboot_cpus_end(); | |
984 | ||
e0b582ec | 985 | cpumask_clear(frozen_cpus); |
1d64b9cb | 986 | out: |
d221938c | 987 | cpu_maps_update_done(); |
1da177e4 | 988 | } |
e0b582ec | 989 | |
d7268a31 | 990 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
991 | { |
992 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
993 | return -ENOMEM; | |
994 | return 0; | |
995 | } | |
996 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 997 | |
79cfbdfa SB |
998 | /* |
999 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1000 | * ensure that the state of the system with respect to the tasks being frozen | |
1001 | * or not, as reported by the notification, remains unchanged *throughout the | |
1002 | * duration* of the execution of the callbacks. | |
1003 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1004 | * | |
1005 | * This synchronization is implemented by mutually excluding regular CPU | |
1006 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1007 | * Hibernate notifications. | |
1008 | */ | |
1009 | static int | |
1010 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1011 | unsigned long action, void *ptr) | |
1012 | { | |
1013 | switch (action) { | |
1014 | ||
1015 | case PM_SUSPEND_PREPARE: | |
1016 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1017 | cpu_hotplug_disable(); |
79cfbdfa SB |
1018 | break; |
1019 | ||
1020 | case PM_POST_SUSPEND: | |
1021 | case PM_POST_HIBERNATION: | |
16e53dbf | 1022 | cpu_hotplug_enable(); |
79cfbdfa SB |
1023 | break; |
1024 | ||
1025 | default: | |
1026 | return NOTIFY_DONE; | |
1027 | } | |
1028 | ||
1029 | return NOTIFY_OK; | |
1030 | } | |
1031 | ||
1032 | ||
d7268a31 | 1033 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1034 | { |
6e32d479 FY |
1035 | /* |
1036 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1037 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1038 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1039 | */ | |
79cfbdfa SB |
1040 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1041 | return 0; | |
1042 | } | |
1043 | core_initcall(cpu_hotplug_pm_sync_init); | |
1044 | ||
f3de4be9 | 1045 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1046 | |
8ce371f9 PZ |
1047 | int __boot_cpu_id; |
1048 | ||
68f4f1ec | 1049 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1050 | |
cff7d378 TG |
1051 | /* Boot processor state steps */ |
1052 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1053 | [CPUHP_OFFLINE] = { | |
1054 | .name = "offline", | |
3c1627e9 TG |
1055 | .startup.single = NULL, |
1056 | .teardown.single = NULL, | |
cff7d378 TG |
1057 | }, |
1058 | #ifdef CONFIG_SMP | |
1059 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1060 | .name = "threads:prepare", |
3c1627e9 TG |
1061 | .startup.single = smpboot_create_threads, |
1062 | .teardown.single = NULL, | |
757c989b | 1063 | .cant_stop = true, |
cff7d378 | 1064 | }, |
00e16c3d | 1065 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1066 | .name = "perf:prepare", |
1067 | .startup.single = perf_event_init_cpu, | |
1068 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1069 | }, |
7ee681b2 | 1070 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1071 | .name = "workqueue:prepare", |
1072 | .startup.single = workqueue_prepare_cpu, | |
1073 | .teardown.single = NULL, | |
7ee681b2 | 1074 | }, |
27590dc1 | 1075 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1076 | .name = "hrtimers:prepare", |
1077 | .startup.single = hrtimers_prepare_cpu, | |
1078 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1079 | }, |
31487f83 | 1080 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1081 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1082 | .startup.single = smpcfd_prepare_cpu, |
1083 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1084 | }, |
e6d4989a RW |
1085 | [CPUHP_RELAY_PREPARE] = { |
1086 | .name = "relay:prepare", | |
1087 | .startup.single = relay_prepare_cpu, | |
1088 | .teardown.single = NULL, | |
1089 | }, | |
6731d4f1 SAS |
1090 | [CPUHP_SLAB_PREPARE] = { |
1091 | .name = "slab:prepare", | |
1092 | .startup.single = slab_prepare_cpu, | |
1093 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1094 | }, |
4df83742 | 1095 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1096 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1097 | .startup.single = rcutree_prepare_cpu, |
1098 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1099 | }, |
4fae16df RC |
1100 | /* |
1101 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1102 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1103 | * otherwise a RCU stall occurs. | |
1104 | */ | |
1105 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1106 | .name = "timers:dead", |
1107 | .startup.single = NULL, | |
1108 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1109 | }, |
d10ef6f9 | 1110 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1111 | [CPUHP_BRINGUP_CPU] = { |
1112 | .name = "cpu:bringup", | |
3c1627e9 TG |
1113 | .startup.single = bringup_cpu, |
1114 | .teardown.single = NULL, | |
757c989b | 1115 | .cant_stop = true, |
4baa0afc | 1116 | }, |
31487f83 | 1117 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1118 | .name = "smpcfd:dying", |
3c1627e9 TG |
1119 | .startup.single = NULL, |
1120 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1121 | }, |
d10ef6f9 TG |
1122 | /* |
1123 | * Handled on controll processor until the plugged processor manages | |
1124 | * this itself. | |
1125 | */ | |
4baa0afc TG |
1126 | [CPUHP_TEARDOWN_CPU] = { |
1127 | .name = "cpu:teardown", | |
3c1627e9 TG |
1128 | .startup.single = NULL, |
1129 | .teardown.single = takedown_cpu, | |
757c989b | 1130 | .cant_stop = true, |
cff7d378 | 1131 | }, |
a7c73414 TG |
1132 | #else |
1133 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1134 | #endif |
cff7d378 TG |
1135 | }; |
1136 | ||
4baa0afc TG |
1137 | /* Application processor state steps */ |
1138 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1139 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1140 | /* Final state before CPU kills itself */ |
1141 | [CPUHP_AP_IDLE_DEAD] = { | |
1142 | .name = "idle:dead", | |
1143 | }, | |
1144 | /* | |
1145 | * Last state before CPU enters the idle loop to die. Transient state | |
1146 | * for synchronization. | |
1147 | */ | |
1148 | [CPUHP_AP_OFFLINE] = { | |
1149 | .name = "ap:offline", | |
1150 | .cant_stop = true, | |
1151 | }, | |
9cf7243d TG |
1152 | /* First state is scheduler control. Interrupts are disabled */ |
1153 | [CPUHP_AP_SCHED_STARTING] = { | |
1154 | .name = "sched:starting", | |
3c1627e9 TG |
1155 | .startup.single = sched_cpu_starting, |
1156 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1157 | }, |
4df83742 | 1158 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1159 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1160 | .startup.single = NULL, |
1161 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1162 | }, |
d10ef6f9 TG |
1163 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1164 | * state for synchronsization */ | |
1165 | [CPUHP_AP_ONLINE] = { | |
1166 | .name = "ap:online", | |
1167 | }, | |
1168 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1169 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1170 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1171 | .startup.single = smpboot_unpark_threads, |
1172 | .teardown.single = NULL, | |
1cf4f629 | 1173 | }, |
00e16c3d | 1174 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1175 | .name = "perf:online", |
1176 | .startup.single = perf_event_init_cpu, | |
1177 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1178 | }, |
7ee681b2 | 1179 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1180 | .name = "workqueue:online", |
1181 | .startup.single = workqueue_online_cpu, | |
1182 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1183 | }, |
4df83742 | 1184 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1185 | .name = "RCU/tree:online", |
3c1627e9 TG |
1186 | .startup.single = rcutree_online_cpu, |
1187 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1188 | }, |
4baa0afc | 1189 | #endif |
d10ef6f9 TG |
1190 | /* |
1191 | * The dynamically registered state space is here | |
1192 | */ | |
1193 | ||
aaddd7d1 TG |
1194 | #ifdef CONFIG_SMP |
1195 | /* Last state is scheduler control setting the cpu active */ | |
1196 | [CPUHP_AP_ACTIVE] = { | |
1197 | .name = "sched:active", | |
3c1627e9 TG |
1198 | .startup.single = sched_cpu_activate, |
1199 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1200 | }, |
1201 | #endif | |
1202 | ||
d10ef6f9 | 1203 | /* CPU is fully up and running. */ |
4baa0afc TG |
1204 | [CPUHP_ONLINE] = { |
1205 | .name = "online", | |
3c1627e9 TG |
1206 | .startup.single = NULL, |
1207 | .teardown.single = NULL, | |
4baa0afc TG |
1208 | }, |
1209 | }; | |
1210 | ||
5b7aa87e TG |
1211 | /* Sanity check for callbacks */ |
1212 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1213 | { | |
1214 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1215 | return -EINVAL; | |
1216 | return 0; | |
1217 | } | |
1218 | ||
dc280d93 TG |
1219 | /* |
1220 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1221 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1222 | * by having no name assigned. | |
1223 | */ | |
1224 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1225 | { | |
4205e478 TG |
1226 | enum cpuhp_state i, end; |
1227 | struct cpuhp_step *step; | |
dc280d93 | 1228 | |
4205e478 TG |
1229 | switch (state) { |
1230 | case CPUHP_AP_ONLINE_DYN: | |
1231 | step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; | |
1232 | end = CPUHP_AP_ONLINE_DYN_END; | |
1233 | break; | |
1234 | case CPUHP_BP_PREPARE_DYN: | |
1235 | step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; | |
1236 | end = CPUHP_BP_PREPARE_DYN_END; | |
1237 | break; | |
1238 | default: | |
1239 | return -EINVAL; | |
1240 | } | |
1241 | ||
1242 | for (i = state; i <= end; i++, step++) { | |
1243 | if (!step->name) | |
dc280d93 TG |
1244 | return i; |
1245 | } | |
1246 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1247 | return -ENOSPC; | |
1248 | } | |
1249 | ||
1250 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1251 | int (*startup)(unsigned int cpu), | |
1252 | int (*teardown)(unsigned int cpu), | |
1253 | bool multi_instance) | |
5b7aa87e TG |
1254 | { |
1255 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1256 | struct cpuhp_step *sp; | |
dc280d93 | 1257 | int ret = 0; |
5b7aa87e | 1258 | |
4205e478 | 1259 | if (state == CPUHP_AP_ONLINE_DYN || state == CPUHP_BP_PREPARE_DYN) { |
dc280d93 TG |
1260 | ret = cpuhp_reserve_state(state); |
1261 | if (ret < 0) | |
dc434e05 | 1262 | return ret; |
dc280d93 TG |
1263 | state = ret; |
1264 | } | |
5b7aa87e | 1265 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1266 | if (name && sp->name) |
1267 | return -EBUSY; | |
1268 | ||
3c1627e9 TG |
1269 | sp->startup.single = startup; |
1270 | sp->teardown.single = teardown; | |
5b7aa87e | 1271 | sp->name = name; |
cf392d10 TG |
1272 | sp->multi_instance = multi_instance; |
1273 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1274 | return ret; |
5b7aa87e TG |
1275 | } |
1276 | ||
1277 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1278 | { | |
3c1627e9 | 1279 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1280 | } |
1281 | ||
5b7aa87e TG |
1282 | /* |
1283 | * Call the startup/teardown function for a step either on the AP or | |
1284 | * on the current CPU. | |
1285 | */ | |
cf392d10 TG |
1286 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1287 | struct hlist_node *node) | |
5b7aa87e | 1288 | { |
a724632c | 1289 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1290 | int ret; |
1291 | ||
3c1627e9 TG |
1292 | if ((bringup && !sp->startup.single) || |
1293 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1294 | return 0; |
5b7aa87e TG |
1295 | /* |
1296 | * The non AP bound callbacks can fail on bringup. On teardown | |
1297 | * e.g. module removal we crash for now. | |
1298 | */ | |
1cf4f629 TG |
1299 | #ifdef CONFIG_SMP |
1300 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1301 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1302 | else |
cf392d10 | 1303 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1304 | #else |
cf392d10 | 1305 | ret = cpuhp_invoke_callback(cpu, state, bringup, node); |
1cf4f629 | 1306 | #endif |
5b7aa87e TG |
1307 | BUG_ON(ret && !bringup); |
1308 | return ret; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1313 | * | |
1314 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1315 | */ | |
1316 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1317 | struct hlist_node *node) |
5b7aa87e TG |
1318 | { |
1319 | int cpu; | |
1320 | ||
5b7aa87e TG |
1321 | /* Roll back the already executed steps on the other cpus */ |
1322 | for_each_present_cpu(cpu) { | |
1323 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1324 | int cpustate = st->state; | |
1325 | ||
1326 | if (cpu >= failedcpu) | |
1327 | break; | |
1328 | ||
1329 | /* Did we invoke the startup call on that cpu ? */ | |
1330 | if (cpustate >= state) | |
cf392d10 | 1331 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1332 | } |
1333 | } | |
1334 | ||
9805c673 TG |
1335 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1336 | struct hlist_node *node, | |
1337 | bool invoke) | |
cf392d10 TG |
1338 | { |
1339 | struct cpuhp_step *sp; | |
1340 | int cpu; | |
1341 | int ret; | |
1342 | ||
9805c673 TG |
1343 | lockdep_assert_cpus_held(); |
1344 | ||
cf392d10 TG |
1345 | sp = cpuhp_get_step(state); |
1346 | if (sp->multi_instance == false) | |
1347 | return -EINVAL; | |
1348 | ||
dc434e05 | 1349 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1350 | |
3c1627e9 | 1351 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1352 | goto add_node; |
1353 | ||
1354 | /* | |
1355 | * Try to call the startup callback for each present cpu | |
1356 | * depending on the hotplug state of the cpu. | |
1357 | */ | |
1358 | for_each_present_cpu(cpu) { | |
1359 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1360 | int cpustate = st->state; | |
1361 | ||
1362 | if (cpustate < state) | |
1363 | continue; | |
1364 | ||
1365 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1366 | if (ret) { | |
3c1627e9 | 1367 | if (sp->teardown.multi) |
cf392d10 | 1368 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1369 | goto unlock; |
cf392d10 TG |
1370 | } |
1371 | } | |
1372 | add_node: | |
1373 | ret = 0; | |
cf392d10 | 1374 | hlist_add_head(node, &sp->list); |
dc434e05 | 1375 | unlock: |
cf392d10 | 1376 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1377 | return ret; |
1378 | } | |
1379 | ||
1380 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1381 | bool invoke) | |
1382 | { | |
1383 | int ret; | |
1384 | ||
1385 | cpus_read_lock(); | |
1386 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1387 | cpus_read_unlock(); |
cf392d10 TG |
1388 | return ret; |
1389 | } | |
1390 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1391 | ||
5b7aa87e | 1392 | /** |
71def423 | 1393 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1394 | * @state: The state to setup |
1395 | * @invoke: If true, the startup function is invoked for cpus where | |
1396 | * cpu state >= @state | |
1397 | * @startup: startup callback function | |
1398 | * @teardown: teardown callback function | |
1399 | * @multi_instance: State is set up for multiple instances which get | |
1400 | * added afterwards. | |
5b7aa87e | 1401 | * |
71def423 | 1402 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1403 | * Returns: |
1404 | * On success: | |
1405 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1406 | * 0 for all other states | |
1407 | * On failure: proper (negative) error code | |
5b7aa87e | 1408 | */ |
71def423 SAS |
1409 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1410 | const char *name, bool invoke, | |
1411 | int (*startup)(unsigned int cpu), | |
1412 | int (*teardown)(unsigned int cpu), | |
1413 | bool multi_instance) | |
5b7aa87e TG |
1414 | { |
1415 | int cpu, ret = 0; | |
b9d9d691 | 1416 | bool dynstate; |
5b7aa87e | 1417 | |
71def423 SAS |
1418 | lockdep_assert_cpus_held(); |
1419 | ||
5b7aa87e TG |
1420 | if (cpuhp_cb_check(state) || !name) |
1421 | return -EINVAL; | |
1422 | ||
dc434e05 | 1423 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1424 | |
dc280d93 TG |
1425 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1426 | multi_instance); | |
5b7aa87e | 1427 | |
b9d9d691 TG |
1428 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1429 | if (ret > 0 && dynstate) { | |
1430 | state = ret; | |
1431 | ret = 0; | |
1432 | } | |
1433 | ||
dc280d93 | 1434 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1435 | goto out; |
1436 | ||
1437 | /* | |
1438 | * Try to call the startup callback for each present cpu | |
1439 | * depending on the hotplug state of the cpu. | |
1440 | */ | |
1441 | for_each_present_cpu(cpu) { | |
1442 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1443 | int cpustate = st->state; | |
1444 | ||
1445 | if (cpustate < state) | |
1446 | continue; | |
1447 | ||
cf392d10 | 1448 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1449 | if (ret) { |
a724632c | 1450 | if (teardown) |
cf392d10 TG |
1451 | cpuhp_rollback_install(cpu, state, NULL); |
1452 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1453 | goto out; |
1454 | } | |
1455 | } | |
1456 | out: | |
dc434e05 | 1457 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1458 | /* |
1459 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1460 | * dynamically allocated state in case of success. | |
1461 | */ | |
b9d9d691 | 1462 | if (!ret && dynstate) |
5b7aa87e TG |
1463 | return state; |
1464 | return ret; | |
1465 | } | |
71def423 SAS |
1466 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1467 | ||
1468 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1469 | const char *name, bool invoke, | |
1470 | int (*startup)(unsigned int cpu), | |
1471 | int (*teardown)(unsigned int cpu), | |
1472 | bool multi_instance) | |
1473 | { | |
1474 | int ret; | |
1475 | ||
1476 | cpus_read_lock(); | |
1477 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1478 | teardown, multi_instance); | |
1479 | cpus_read_unlock(); | |
1480 | return ret; | |
1481 | } | |
5b7aa87e TG |
1482 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1483 | ||
cf392d10 TG |
1484 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1485 | struct hlist_node *node, bool invoke) | |
1486 | { | |
1487 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1488 | int cpu; | |
1489 | ||
1490 | BUG_ON(cpuhp_cb_check(state)); | |
1491 | ||
1492 | if (!sp->multi_instance) | |
1493 | return -EINVAL; | |
1494 | ||
8f553c49 | 1495 | cpus_read_lock(); |
dc434e05 SAS |
1496 | mutex_lock(&cpuhp_state_mutex); |
1497 | ||
cf392d10 TG |
1498 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1499 | goto remove; | |
1500 | /* | |
1501 | * Call the teardown callback for each present cpu depending | |
1502 | * on the hotplug state of the cpu. This function is not | |
1503 | * allowed to fail currently! | |
1504 | */ | |
1505 | for_each_present_cpu(cpu) { | |
1506 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1507 | int cpustate = st->state; | |
1508 | ||
1509 | if (cpustate >= state) | |
1510 | cpuhp_issue_call(cpu, state, false, node); | |
1511 | } | |
1512 | ||
1513 | remove: | |
cf392d10 TG |
1514 | hlist_del(node); |
1515 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1516 | cpus_read_unlock(); |
cf392d10 TG |
1517 | |
1518 | return 0; | |
1519 | } | |
1520 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1521 | |
5b7aa87e | 1522 | /** |
71def423 | 1523 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1524 | * @state: The state to remove |
1525 | * @invoke: If true, the teardown function is invoked for cpus where | |
1526 | * cpu state >= @state | |
1527 | * | |
71def423 | 1528 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1529 | * The teardown callback is currently not allowed to fail. Think |
1530 | * about module removal! | |
1531 | */ | |
71def423 | 1532 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1533 | { |
cf392d10 | 1534 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1535 | int cpu; |
1536 | ||
1537 | BUG_ON(cpuhp_cb_check(state)); | |
1538 | ||
71def423 | 1539 | lockdep_assert_cpus_held(); |
5b7aa87e | 1540 | |
dc434e05 | 1541 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1542 | if (sp->multi_instance) { |
1543 | WARN(!hlist_empty(&sp->list), | |
1544 | "Error: Removing state %d which has instances left.\n", | |
1545 | state); | |
1546 | goto remove; | |
1547 | } | |
1548 | ||
a724632c | 1549 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1550 | goto remove; |
1551 | ||
1552 | /* | |
1553 | * Call the teardown callback for each present cpu depending | |
1554 | * on the hotplug state of the cpu. This function is not | |
1555 | * allowed to fail currently! | |
1556 | */ | |
1557 | for_each_present_cpu(cpu) { | |
1558 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1559 | int cpustate = st->state; | |
1560 | ||
1561 | if (cpustate >= state) | |
cf392d10 | 1562 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1563 | } |
1564 | remove: | |
cf392d10 | 1565 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1566 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1567 | } |
1568 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1569 | ||
1570 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1571 | { | |
1572 | cpus_read_lock(); | |
1573 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1574 | cpus_read_unlock(); |
5b7aa87e TG |
1575 | } |
1576 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1577 | ||
98f8cdce TG |
1578 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1579 | static ssize_t show_cpuhp_state(struct device *dev, | |
1580 | struct device_attribute *attr, char *buf) | |
1581 | { | |
1582 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1583 | ||
1584 | return sprintf(buf, "%d\n", st->state); | |
1585 | } | |
1586 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1587 | ||
757c989b TG |
1588 | static ssize_t write_cpuhp_target(struct device *dev, |
1589 | struct device_attribute *attr, | |
1590 | const char *buf, size_t count) | |
1591 | { | |
1592 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1593 | struct cpuhp_step *sp; | |
1594 | int target, ret; | |
1595 | ||
1596 | ret = kstrtoint(buf, 10, &target); | |
1597 | if (ret) | |
1598 | return ret; | |
1599 | ||
1600 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1601 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1602 | return -EINVAL; | |
1603 | #else | |
1604 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1605 | return -EINVAL; | |
1606 | #endif | |
1607 | ||
1608 | ret = lock_device_hotplug_sysfs(); | |
1609 | if (ret) | |
1610 | return ret; | |
1611 | ||
1612 | mutex_lock(&cpuhp_state_mutex); | |
1613 | sp = cpuhp_get_step(target); | |
1614 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1615 | mutex_unlock(&cpuhp_state_mutex); | |
1616 | if (ret) | |
1617 | return ret; | |
1618 | ||
1619 | if (st->state < target) | |
1620 | ret = do_cpu_up(dev->id, target); | |
1621 | else | |
1622 | ret = do_cpu_down(dev->id, target); | |
1623 | ||
1624 | unlock_device_hotplug(); | |
1625 | return ret ? ret : count; | |
1626 | } | |
1627 | ||
98f8cdce TG |
1628 | static ssize_t show_cpuhp_target(struct device *dev, |
1629 | struct device_attribute *attr, char *buf) | |
1630 | { | |
1631 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1632 | ||
1633 | return sprintf(buf, "%d\n", st->target); | |
1634 | } | |
757c989b | 1635 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1636 | |
1637 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1638 | &dev_attr_state.attr, | |
1639 | &dev_attr_target.attr, | |
1640 | NULL | |
1641 | }; | |
1642 | ||
1643 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1644 | .attrs = cpuhp_cpu_attrs, | |
1645 | .name = "hotplug", | |
1646 | NULL | |
1647 | }; | |
1648 | ||
1649 | static ssize_t show_cpuhp_states(struct device *dev, | |
1650 | struct device_attribute *attr, char *buf) | |
1651 | { | |
1652 | ssize_t cur, res = 0; | |
1653 | int i; | |
1654 | ||
1655 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1656 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1657 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1658 | ||
1659 | if (sp->name) { | |
1660 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1661 | buf += cur; | |
1662 | res += cur; | |
1663 | } | |
1664 | } | |
1665 | mutex_unlock(&cpuhp_state_mutex); | |
1666 | return res; | |
1667 | } | |
1668 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1669 | ||
1670 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1671 | &dev_attr_states.attr, | |
1672 | NULL | |
1673 | }; | |
1674 | ||
1675 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1676 | .attrs = cpuhp_cpu_root_attrs, | |
1677 | .name = "hotplug", | |
1678 | NULL | |
1679 | }; | |
1680 | ||
1681 | static int __init cpuhp_sysfs_init(void) | |
1682 | { | |
1683 | int cpu, ret; | |
1684 | ||
1685 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1686 | &cpuhp_cpu_root_attr_group); | |
1687 | if (ret) | |
1688 | return ret; | |
1689 | ||
1690 | for_each_possible_cpu(cpu) { | |
1691 | struct device *dev = get_cpu_device(cpu); | |
1692 | ||
1693 | if (!dev) | |
1694 | continue; | |
1695 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1696 | if (ret) | |
1697 | return ret; | |
1698 | } | |
1699 | return 0; | |
1700 | } | |
1701 | device_initcall(cpuhp_sysfs_init); | |
1702 | #endif | |
1703 | ||
e56b3bc7 LT |
1704 | /* |
1705 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1706 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1707 | * | |
e0b582ec | 1708 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1709 | * mask value that has a single bit set only. |
1710 | */ | |
b8d317d1 | 1711 | |
e56b3bc7 | 1712 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1713 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1714 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1715 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1716 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1717 | |
e56b3bc7 LT |
1718 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1719 | ||
1720 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1721 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1722 | #if BITS_PER_LONG > 32 | |
1723 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1724 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1725 | #endif |
1726 | }; | |
e56b3bc7 | 1727 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1728 | |
1729 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1730 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1731 | |
1732 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1733 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1734 | = {CPU_BITS_ALL}; |
b3199c02 | 1735 | #else |
4b804c85 | 1736 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1737 | #endif |
4b804c85 | 1738 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1739 | |
4b804c85 RV |
1740 | struct cpumask __cpu_online_mask __read_mostly; |
1741 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1742 | |
4b804c85 RV |
1743 | struct cpumask __cpu_present_mask __read_mostly; |
1744 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1745 | |
4b804c85 RV |
1746 | struct cpumask __cpu_active_mask __read_mostly; |
1747 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1748 | |
3fa41520 RR |
1749 | void init_cpu_present(const struct cpumask *src) |
1750 | { | |
c4c54dd1 | 1751 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1752 | } |
1753 | ||
1754 | void init_cpu_possible(const struct cpumask *src) | |
1755 | { | |
c4c54dd1 | 1756 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1757 | } |
1758 | ||
1759 | void init_cpu_online(const struct cpumask *src) | |
1760 | { | |
c4c54dd1 | 1761 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1762 | } |
cff7d378 TG |
1763 | |
1764 | /* | |
1765 | * Activate the first processor. | |
1766 | */ | |
1767 | void __init boot_cpu_init(void) | |
1768 | { | |
1769 | int cpu = smp_processor_id(); | |
1770 | ||
1771 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1772 | set_cpu_online(cpu, true); | |
1773 | set_cpu_active(cpu, true); | |
1774 | set_cpu_present(cpu, true); | |
1775 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
1776 | |
1777 | #ifdef CONFIG_SMP | |
1778 | __boot_cpu_id = cpu; | |
1779 | #endif | |
cff7d378 TG |
1780 | } |
1781 | ||
1782 | /* | |
1783 | * Must be called _AFTER_ setting up the per_cpu areas | |
1784 | */ | |
1785 | void __init boot_cpu_state_init(void) | |
1786 | { | |
1787 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1788 | } |