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; |
4dddfb5f | 61 | struct hlist_node *last; |
4cb28ced | 62 | enum cpuhp_state cb_state; |
4cb28ced TG |
63 | int result; |
64 | struct completion done; | |
65 | #endif | |
cff7d378 TG |
66 | }; |
67 | ||
68 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
69 | ||
49dfe2a6 | 70 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
71 | static struct lockdep_map cpuhp_state_up_map = |
72 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
73 | static struct lockdep_map cpuhp_state_down_map = | |
74 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
75 | ||
76 | ||
77 | static void inline cpuhp_lock_acquire(bool bringup) | |
78 | { | |
79 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
80 | } | |
81 | ||
82 | static void inline cpuhp_lock_release(bool bringup) | |
83 | { | |
84 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
85 | } | |
86 | #else | |
87 | ||
88 | static void inline cpuhp_lock_acquire(bool bringup) { } | |
89 | static void inline cpuhp_lock_release(bool bringup) { } | |
90 | ||
49dfe2a6 TG |
91 | #endif |
92 | ||
cff7d378 TG |
93 | /** |
94 | * cpuhp_step - Hotplug state machine step | |
95 | * @name: Name of the step | |
96 | * @startup: Startup function of the step | |
97 | * @teardown: Teardown function of the step | |
98 | * @skip_onerr: Do not invoke the functions on error rollback | |
99 | * Will go away once the notifiers are gone | |
757c989b | 100 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
101 | */ |
102 | struct cpuhp_step { | |
cf392d10 TG |
103 | const char *name; |
104 | union { | |
3c1627e9 TG |
105 | int (*single)(unsigned int cpu); |
106 | int (*multi)(unsigned int cpu, | |
107 | struct hlist_node *node); | |
108 | } startup; | |
cf392d10 | 109 | union { |
3c1627e9 TG |
110 | int (*single)(unsigned int cpu); |
111 | int (*multi)(unsigned int cpu, | |
112 | struct hlist_node *node); | |
113 | } teardown; | |
cf392d10 TG |
114 | struct hlist_head list; |
115 | bool skip_onerr; | |
116 | bool cant_stop; | |
117 | bool multi_instance; | |
cff7d378 TG |
118 | }; |
119 | ||
98f8cdce | 120 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 121 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 122 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 123 | |
a724632c TG |
124 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
125 | { | |
126 | /* | |
127 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
128 | * purposes as that state is handled explicitly in cpu_down. | |
129 | */ | |
130 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
131 | } | |
132 | ||
4dddfb5f PZ |
133 | /* |
134 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
135 | */ | |
136 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
137 | { | |
138 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
139 | } | |
140 | ||
a724632c TG |
141 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) |
142 | { | |
143 | struct cpuhp_step *sp; | |
144 | ||
145 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
146 | return sp + state; | |
147 | } | |
148 | ||
cff7d378 TG |
149 | /** |
150 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
151 | * @cpu: The cpu for which the callback should be invoked | |
96abb968 | 152 | * @state: The state to do callbacks for |
a724632c | 153 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
154 | * @node: For multi-instance, do a single entry callback for install/remove |
155 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 156 | * |
cf392d10 | 157 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 158 | */ |
a724632c | 159 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
160 | bool bringup, struct hlist_node *node, |
161 | struct hlist_node **lastp) | |
cff7d378 TG |
162 | { |
163 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 164 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
165 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
166 | int (*cb)(unsigned int cpu); | |
167 | int ret, cnt; | |
168 | ||
169 | if (!step->multi_instance) { | |
96abb968 | 170 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 171 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
172 | if (!cb) |
173 | return 0; | |
a724632c | 174 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 175 | ret = cb(cpu); |
a724632c | 176 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
177 | return ret; |
178 | } | |
3c1627e9 | 179 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
180 | if (!cbm) |
181 | return 0; | |
182 | ||
183 | /* Single invocation for instance add/remove */ | |
184 | if (node) { | |
96abb968 | 185 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
186 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
187 | ret = cbm(cpu, node); | |
188 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
189 | return ret; | |
190 | } | |
191 | ||
192 | /* State transition. Invoke on all instances */ | |
193 | cnt = 0; | |
194 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
195 | if (lastp && node == *lastp) |
196 | break; | |
197 | ||
cf392d10 TG |
198 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
199 | ret = cbm(cpu, node); | |
200 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
201 | if (ret) { |
202 | if (!lastp) | |
203 | goto err; | |
204 | ||
205 | *lastp = node; | |
206 | return ret; | |
207 | } | |
cf392d10 TG |
208 | cnt++; |
209 | } | |
96abb968 PZ |
210 | if (lastp) |
211 | *lastp = NULL; | |
cf392d10 TG |
212 | return 0; |
213 | err: | |
214 | /* Rollback the instances if one failed */ | |
3c1627e9 | 215 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
216 | if (!cbm) |
217 | return ret; | |
218 | ||
219 | hlist_for_each(node, &step->list) { | |
220 | if (!cnt--) | |
221 | break; | |
724a8688 PZ |
222 | |
223 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
224 | ret = cbm(cpu, node); | |
225 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
226 | /* | |
227 | * Rollback must not fail, | |
228 | */ | |
229 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
230 | } |
231 | return ret; | |
232 | } | |
233 | ||
98a79d6a | 234 | #ifdef CONFIG_SMP |
b3199c02 | 235 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 236 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
237 | bool cpuhp_tasks_frozen; |
238 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 239 | |
79a6cdeb | 240 | /* |
93ae4f97 SB |
241 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
242 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
243 | */ |
244 | void cpu_maps_update_begin(void) | |
245 | { | |
246 | mutex_lock(&cpu_add_remove_lock); | |
247 | } | |
248 | ||
249 | void cpu_maps_update_done(void) | |
250 | { | |
251 | mutex_unlock(&cpu_add_remove_lock); | |
252 | } | |
1da177e4 | 253 | |
fc8dffd3 TG |
254 | /* |
255 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
256 | * Should always be manipulated under cpu_add_remove_lock |
257 | */ | |
258 | static int cpu_hotplug_disabled; | |
259 | ||
79a6cdeb LJ |
260 | #ifdef CONFIG_HOTPLUG_CPU |
261 | ||
fc8dffd3 | 262 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 263 | |
8f553c49 | 264 | void cpus_read_lock(void) |
a9d9baa1 | 265 | { |
fc8dffd3 | 266 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 267 | } |
8f553c49 | 268 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 269 | |
8f553c49 | 270 | void cpus_read_unlock(void) |
a9d9baa1 | 271 | { |
fc8dffd3 | 272 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 273 | } |
8f553c49 | 274 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 275 | |
8f553c49 | 276 | void cpus_write_lock(void) |
d221938c | 277 | { |
fc8dffd3 | 278 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 279 | } |
87af9e7f | 280 | |
8f553c49 | 281 | void cpus_write_unlock(void) |
d221938c | 282 | { |
fc8dffd3 | 283 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
284 | } |
285 | ||
fc8dffd3 | 286 | void lockdep_assert_cpus_held(void) |
d221938c | 287 | { |
fc8dffd3 | 288 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 289 | } |
79a6cdeb | 290 | |
16e53dbf SB |
291 | /* |
292 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
293 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
294 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
295 | * hotplug path before performing hotplug operations. So acquiring that lock | |
296 | * guarantees mutual exclusion from any currently running hotplug operations. | |
297 | */ | |
298 | void cpu_hotplug_disable(void) | |
299 | { | |
300 | cpu_maps_update_begin(); | |
89af7ba5 | 301 | cpu_hotplug_disabled++; |
16e53dbf SB |
302 | cpu_maps_update_done(); |
303 | } | |
32145c46 | 304 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 305 | |
01b41159 LW |
306 | static void __cpu_hotplug_enable(void) |
307 | { | |
308 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
309 | return; | |
310 | cpu_hotplug_disabled--; | |
311 | } | |
312 | ||
16e53dbf SB |
313 | void cpu_hotplug_enable(void) |
314 | { | |
315 | cpu_maps_update_begin(); | |
01b41159 | 316 | __cpu_hotplug_enable(); |
16e53dbf SB |
317 | cpu_maps_update_done(); |
318 | } | |
32145c46 | 319 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 320 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 321 | |
4dddfb5f PZ |
322 | static inline enum cpuhp_state |
323 | cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
324 | { | |
325 | enum cpuhp_state prev_state = st->state; | |
326 | ||
327 | st->rollback = false; | |
328 | st->last = NULL; | |
329 | ||
330 | st->target = target; | |
331 | st->single = false; | |
332 | st->bringup = st->state < target; | |
333 | ||
334 | return prev_state; | |
335 | } | |
336 | ||
337 | static inline void | |
338 | cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state) | |
339 | { | |
340 | st->rollback = true; | |
341 | ||
342 | /* | |
343 | * If we have st->last we need to undo partial multi_instance of this | |
344 | * state first. Otherwise start undo at the previous state. | |
345 | */ | |
346 | if (!st->last) { | |
347 | if (st->bringup) | |
348 | st->state--; | |
349 | else | |
350 | st->state++; | |
351 | } | |
352 | ||
353 | st->target = prev_state; | |
354 | st->bringup = !st->bringup; | |
355 | } | |
356 | ||
357 | /* Regular hotplug invocation of the AP hotplug thread */ | |
358 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
359 | { | |
360 | if (!st->single && st->state == st->target) | |
361 | return; | |
362 | ||
363 | st->result = 0; | |
364 | /* | |
365 | * Make sure the above stores are visible before should_run becomes | |
366 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
367 | */ | |
368 | smp_mb(); | |
369 | st->should_run = true; | |
370 | wake_up_process(st->thread); | |
371 | wait_for_completion(&st->done); | |
372 | } | |
373 | ||
374 | static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
375 | { | |
376 | enum cpuhp_state prev_state; | |
377 | int ret; | |
378 | ||
379 | prev_state = cpuhp_set_state(st, target); | |
380 | __cpuhp_kick_ap(st); | |
381 | if ((ret = st->result)) { | |
382 | cpuhp_reset_state(st, prev_state); | |
383 | __cpuhp_kick_ap(st); | |
384 | } | |
385 | ||
386 | return ret; | |
387 | } | |
9cd4f1a4 | 388 | |
8df3e07e TG |
389 | static int bringup_wait_for_ap(unsigned int cpu) |
390 | { | |
391 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
392 | ||
9cd4f1a4 | 393 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
8df3e07e | 394 | wait_for_completion(&st->done); |
dea1d0f5 TG |
395 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
396 | return -ECANCELED; | |
9cd4f1a4 TG |
397 | |
398 | /* Unpark the stopper thread and the hotplug thread of the target cpu */ | |
399 | stop_machine_unpark(cpu); | |
400 | kthread_unpark(st->thread); | |
401 | ||
4dddfb5f PZ |
402 | if (st->target <= CPUHP_AP_ONLINE_IDLE) |
403 | return 0; | |
404 | ||
405 | return cpuhp_kick_ap(st, st->target); | |
8df3e07e TG |
406 | } |
407 | ||
ba997462 TG |
408 | static int bringup_cpu(unsigned int cpu) |
409 | { | |
410 | struct task_struct *idle = idle_thread_get(cpu); | |
411 | int ret; | |
412 | ||
aa877175 BO |
413 | /* |
414 | * Some architectures have to walk the irq descriptors to | |
415 | * setup the vector space for the cpu which comes online. | |
416 | * Prevent irq alloc/free across the bringup. | |
417 | */ | |
418 | irq_lock_sparse(); | |
419 | ||
ba997462 TG |
420 | /* Arch-specific enabling code. */ |
421 | ret = __cpu_up(cpu, idle); | |
aa877175 | 422 | irq_unlock_sparse(); |
530e9b76 | 423 | if (ret) |
ba997462 | 424 | return ret; |
9cd4f1a4 | 425 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
426 | } |
427 | ||
2e1a3483 TG |
428 | /* |
429 | * Hotplug state machine related functions | |
430 | */ | |
2e1a3483 | 431 | |
a724632c | 432 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
433 | { |
434 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 435 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
436 | |
437 | if (!step->skip_onerr) | |
96abb968 | 438 | cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); |
2e1a3483 TG |
439 | } |
440 | } | |
441 | ||
442 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 443 | enum cpuhp_state target) |
2e1a3483 TG |
444 | { |
445 | enum cpuhp_state prev_state = st->state; | |
446 | int ret = 0; | |
447 | ||
448 | while (st->state < target) { | |
2e1a3483 | 449 | st->state++; |
96abb968 | 450 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
2e1a3483 TG |
451 | if (ret) { |
452 | st->target = prev_state; | |
a724632c | 453 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
454 | break; |
455 | } | |
456 | } | |
457 | return ret; | |
458 | } | |
459 | ||
4cb28ced TG |
460 | /* |
461 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
462 | */ | |
463 | static void cpuhp_create(unsigned int cpu) | |
464 | { | |
465 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
466 | ||
467 | init_completion(&st->done); | |
468 | } | |
469 | ||
470 | static int cpuhp_should_run(unsigned int cpu) | |
471 | { | |
472 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
473 | ||
474 | return st->should_run; | |
475 | } | |
476 | ||
4cb28ced TG |
477 | /* |
478 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
479 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
480 | * |
481 | * Each invocation of this function by the smpboot thread does a single AP | |
482 | * state callback. | |
483 | * | |
484 | * It has 3 modes of operation: | |
485 | * - single: runs st->cb_state | |
486 | * - up: runs ++st->state, while st->state < st->target | |
487 | * - down: runs st->state--, while st->state > st->target | |
488 | * | |
489 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
490 | */ |
491 | static void cpuhp_thread_fun(unsigned int cpu) | |
492 | { | |
493 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
494 | bool bringup = st->bringup; |
495 | enum cpuhp_state state; | |
4cb28ced TG |
496 | |
497 | /* | |
4dddfb5f PZ |
498 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
499 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
500 | */ |
501 | smp_mb(); | |
4cb28ced | 502 | |
4dddfb5f PZ |
503 | if (WARN_ON_ONCE(!st->should_run)) |
504 | return; | |
4cb28ced | 505 | |
5f4b55e1 | 506 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 507 | |
a724632c | 508 | if (st->single) { |
4dddfb5f PZ |
509 | state = st->cb_state; |
510 | st->should_run = false; | |
511 | } else { | |
512 | if (bringup) { | |
513 | st->state++; | |
514 | state = st->state; | |
515 | st->should_run = (st->state < st->target); | |
516 | WARN_ON_ONCE(st->state > st->target); | |
4cb28ced | 517 | } else { |
4dddfb5f PZ |
518 | state = st->state; |
519 | st->state--; | |
520 | st->should_run = (st->state > st->target); | |
521 | WARN_ON_ONCE(st->state < st->target); | |
4cb28ced | 522 | } |
4dddfb5f PZ |
523 | } |
524 | ||
525 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
526 | ||
527 | if (st->rollback) { | |
528 | struct cpuhp_step *step = cpuhp_get_step(state); | |
529 | if (step->skip_onerr) | |
530 | goto next; | |
531 | } | |
532 | ||
533 | if (cpuhp_is_atomic_state(state)) { | |
534 | local_irq_disable(); | |
535 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
536 | local_irq_enable(); | |
3b9d6da6 | 537 | |
4dddfb5f PZ |
538 | /* |
539 | * STARTING/DYING must not fail! | |
540 | */ | |
541 | WARN_ON_ONCE(st->result); | |
4cb28ced | 542 | } else { |
4dddfb5f PZ |
543 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
544 | } | |
545 | ||
546 | if (st->result) { | |
547 | /* | |
548 | * If we fail on a rollback, we're up a creek without no | |
549 | * paddle, no way forward, no way back. We loose, thanks for | |
550 | * playing. | |
551 | */ | |
552 | WARN_ON_ONCE(st->rollback); | |
553 | st->should_run = false; | |
4cb28ced | 554 | } |
4dddfb5f PZ |
555 | |
556 | next: | |
5f4b55e1 | 557 | cpuhp_lock_release(bringup); |
4dddfb5f PZ |
558 | |
559 | if (!st->should_run) | |
560 | complete(&st->done); | |
4cb28ced TG |
561 | } |
562 | ||
563 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 564 | static int |
cf392d10 TG |
565 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
566 | struct hlist_node *node) | |
4cb28ced TG |
567 | { |
568 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 569 | int ret; |
4cb28ced TG |
570 | |
571 | if (!cpu_online(cpu)) | |
572 | return 0; | |
573 | ||
5f4b55e1 PZ |
574 | cpuhp_lock_acquire(false); |
575 | cpuhp_lock_release(false); | |
576 | ||
577 | cpuhp_lock_acquire(true); | |
578 | cpuhp_lock_release(true); | |
49dfe2a6 | 579 | |
6a4e2451 TG |
580 | /* |
581 | * If we are up and running, use the hotplug thread. For early calls | |
582 | * we invoke the thread function directly. | |
583 | */ | |
584 | if (!st->thread) | |
96abb968 | 585 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 586 | |
4dddfb5f PZ |
587 | st->rollback = false; |
588 | st->last = NULL; | |
589 | ||
590 | st->node = node; | |
591 | st->bringup = bringup; | |
4cb28ced | 592 | st->cb_state = state; |
a724632c | 593 | st->single = true; |
a724632c | 594 | |
4dddfb5f | 595 | __cpuhp_kick_ap(st); |
4cb28ced | 596 | |
4cb28ced | 597 | /* |
4dddfb5f | 598 | * If we failed and did a partial, do a rollback. |
4cb28ced | 599 | */ |
4dddfb5f PZ |
600 | if ((ret = st->result) && st->last) { |
601 | st->rollback = true; | |
602 | st->bringup = !bringup; | |
603 | ||
604 | __cpuhp_kick_ap(st); | |
605 | } | |
606 | ||
607 | return ret; | |
1cf4f629 TG |
608 | } |
609 | ||
610 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
611 | { | |
612 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
613 | enum cpuhp_state prev_state = st->state; |
614 | int ret; | |
1cf4f629 | 615 | |
5f4b55e1 PZ |
616 | cpuhp_lock_acquire(false); |
617 | cpuhp_lock_release(false); | |
618 | ||
619 | cpuhp_lock_acquire(true); | |
620 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
621 | |
622 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
623 | ret = cpuhp_kick_ap(st, st->target); | |
624 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); | |
625 | ||
626 | return ret; | |
4cb28ced TG |
627 | } |
628 | ||
629 | static struct smp_hotplug_thread cpuhp_threads = { | |
630 | .store = &cpuhp_state.thread, | |
631 | .create = &cpuhp_create, | |
632 | .thread_should_run = cpuhp_should_run, | |
633 | .thread_fn = cpuhp_thread_fun, | |
634 | .thread_comm = "cpuhp/%u", | |
635 | .selfparking = true, | |
636 | }; | |
637 | ||
638 | void __init cpuhp_threads_init(void) | |
639 | { | |
640 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
641 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
642 | } | |
643 | ||
777c6e0d | 644 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
645 | /** |
646 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
647 | * @cpu: a CPU id | |
648 | * | |
649 | * This function walks all processes, finds a valid mm struct for each one and | |
650 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
651 | * trivial, there are various non-obvious corner cases, which this function | |
652 | * tries to solve in a safe manner. | |
653 | * | |
654 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
655 | * be called only for an already offlined CPU. | |
656 | */ | |
cb79295e AV |
657 | void clear_tasks_mm_cpumask(int cpu) |
658 | { | |
659 | struct task_struct *p; | |
660 | ||
661 | /* | |
662 | * This function is called after the cpu is taken down and marked | |
663 | * offline, so its not like new tasks will ever get this cpu set in | |
664 | * their mm mask. -- Peter Zijlstra | |
665 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
666 | * full-fledged tasklist_lock. | |
667 | */ | |
e4cc2f87 | 668 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
669 | rcu_read_lock(); |
670 | for_each_process(p) { | |
671 | struct task_struct *t; | |
672 | ||
e4cc2f87 AV |
673 | /* |
674 | * Main thread might exit, but other threads may still have | |
675 | * a valid mm. Find one. | |
676 | */ | |
cb79295e AV |
677 | t = find_lock_task_mm(p); |
678 | if (!t) | |
679 | continue; | |
680 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
681 | task_unlock(t); | |
682 | } | |
683 | rcu_read_unlock(); | |
684 | } | |
685 | ||
1da177e4 | 686 | /* Take this CPU down. */ |
71cf5aee | 687 | static int take_cpu_down(void *_param) |
1da177e4 | 688 | { |
4baa0afc TG |
689 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
690 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 691 | int err, cpu = smp_processor_id(); |
724a8688 | 692 | int ret; |
1da177e4 | 693 | |
1da177e4 LT |
694 | /* Ensure this CPU doesn't handle any more interrupts. */ |
695 | err = __cpu_disable(); | |
696 | if (err < 0) | |
f3705136 | 697 | return err; |
1da177e4 | 698 | |
a724632c TG |
699 | /* |
700 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
701 | * do this step again. | |
702 | */ | |
703 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
704 | st->state--; | |
4baa0afc | 705 | /* Invoke the former CPU_DYING callbacks */ |
724a8688 PZ |
706 | for (; st->state > target; st->state--) { |
707 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
708 | /* | |
709 | * DYING must not fail! | |
710 | */ | |
711 | WARN_ON_ONCE(ret); | |
712 | } | |
4baa0afc | 713 | |
52c063d1 TG |
714 | /* Give up timekeeping duties */ |
715 | tick_handover_do_timer(); | |
14e568e7 | 716 | /* Park the stopper thread */ |
090e77c3 | 717 | stop_machine_park(cpu); |
f3705136 | 718 | return 0; |
1da177e4 LT |
719 | } |
720 | ||
98458172 | 721 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 722 | { |
e69aab13 | 723 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 724 | int err; |
1da177e4 | 725 | |
2a58c527 | 726 | /* Park the smpboot threads */ |
1cf4f629 | 727 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 728 | smpboot_park_threads(cpu); |
1cf4f629 | 729 | |
6acce3ef | 730 | /* |
a8994181 TG |
731 | * Prevent irq alloc/free while the dying cpu reorganizes the |
732 | * interrupt affinities. | |
6acce3ef | 733 | */ |
a8994181 | 734 | irq_lock_sparse(); |
6acce3ef | 735 | |
a8994181 TG |
736 | /* |
737 | * So now all preempt/rcu users must observe !cpu_active(). | |
738 | */ | |
210e2133 | 739 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 740 | if (err) { |
3b9d6da6 | 741 | /* CPU refused to die */ |
a8994181 | 742 | irq_unlock_sparse(); |
3b9d6da6 SAS |
743 | /* Unpark the hotplug thread so we can rollback there */ |
744 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 745 | return err; |
8fa1d7d3 | 746 | } |
04321587 | 747 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 748 | |
48c5ccae | 749 | /* |
ee1e714b | 750 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
751 | * runnable tasks from the cpu, there's only the idle task left now |
752 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
753 | * |
754 | * Wait for the stop thread to go away. | |
48c5ccae | 755 | */ |
e69aab13 TG |
756 | wait_for_completion(&st->done); |
757 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); | |
1da177e4 | 758 | |
a8994181 TG |
759 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
760 | irq_unlock_sparse(); | |
761 | ||
345527b1 | 762 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
763 | /* This actually kills the CPU. */ |
764 | __cpu_die(cpu); | |
765 | ||
a49b116d | 766 | tick_cleanup_dead_cpu(cpu); |
a58163d8 | 767 | rcutree_migrate_callbacks(cpu); |
98458172 TG |
768 | return 0; |
769 | } | |
1da177e4 | 770 | |
71f87b2f TG |
771 | static void cpuhp_complete_idle_dead(void *arg) |
772 | { | |
773 | struct cpuhp_cpu_state *st = arg; | |
774 | ||
775 | complete(&st->done); | |
776 | } | |
777 | ||
e69aab13 TG |
778 | void cpuhp_report_idle_dead(void) |
779 | { | |
780 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
781 | ||
782 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 783 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
784 | st->state = CPUHP_AP_IDLE_DEAD; |
785 | /* | |
786 | * We cannot call complete after rcu_report_dead() so we delegate it | |
787 | * to an online cpu. | |
788 | */ | |
789 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
790 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
791 | } |
792 | ||
4dddfb5f PZ |
793 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
794 | { | |
795 | for (st->state++; st->state < st->target; st->state++) { | |
796 | struct cpuhp_step *step = cpuhp_get_step(st->state); | |
cff7d378 | 797 | |
4dddfb5f PZ |
798 | if (!step->skip_onerr) |
799 | cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); | |
800 | } | |
801 | } | |
802 | ||
803 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
804 | enum cpuhp_state target) | |
805 | { | |
806 | enum cpuhp_state prev_state = st->state; | |
807 | int ret = 0; | |
808 | ||
809 | for (; st->state > target; st->state--) { | |
810 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
811 | if (ret) { | |
812 | st->target = prev_state; | |
813 | undo_cpu_down(cpu, st); | |
814 | break; | |
815 | } | |
816 | } | |
817 | return ret; | |
818 | } | |
cff7d378 | 819 | |
98458172 | 820 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
821 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
822 | enum cpuhp_state target) | |
98458172 | 823 | { |
cff7d378 TG |
824 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
825 | int prev_state, ret = 0; | |
98458172 TG |
826 | |
827 | if (num_online_cpus() == 1) | |
828 | return -EBUSY; | |
829 | ||
757c989b | 830 | if (!cpu_present(cpu)) |
98458172 TG |
831 | return -EINVAL; |
832 | ||
8f553c49 | 833 | cpus_write_lock(); |
98458172 TG |
834 | |
835 | cpuhp_tasks_frozen = tasks_frozen; | |
836 | ||
4dddfb5f | 837 | prev_state = cpuhp_set_state(st, target); |
1cf4f629 TG |
838 | /* |
839 | * If the current CPU state is in the range of the AP hotplug thread, | |
840 | * then we need to kick the thread. | |
841 | */ | |
8df3e07e | 842 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 843 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
844 | ret = cpuhp_kick_ap_work(cpu); |
845 | /* | |
846 | * The AP side has done the error rollback already. Just | |
847 | * return the error code.. | |
848 | */ | |
849 | if (ret) | |
850 | goto out; | |
851 | ||
852 | /* | |
853 | * We might have stopped still in the range of the AP hotplug | |
854 | * thread. Nothing to do anymore. | |
855 | */ | |
8df3e07e | 856 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 857 | goto out; |
4dddfb5f PZ |
858 | |
859 | st->target = target; | |
1cf4f629 TG |
860 | } |
861 | /* | |
8df3e07e | 862 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
863 | * to do the further cleanups. |
864 | */ | |
a724632c | 865 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 | 866 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
4dddfb5f PZ |
867 | cpuhp_reset_state(st, prev_state); |
868 | __cpuhp_kick_ap(st); | |
3b9d6da6 | 869 | } |
98458172 | 870 | |
1cf4f629 | 871 | out: |
8f553c49 | 872 | cpus_write_unlock(); |
cff7d378 | 873 | return ret; |
e3920fb4 RW |
874 | } |
875 | ||
af1f4045 | 876 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 877 | { |
9ea09af3 | 878 | int err; |
e3920fb4 | 879 | |
d221938c | 880 | cpu_maps_update_begin(); |
e761b772 MK |
881 | |
882 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 883 | err = -EBUSY; |
e761b772 MK |
884 | goto out; |
885 | } | |
886 | ||
af1f4045 | 887 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 888 | |
e761b772 | 889 | out: |
d221938c | 890 | cpu_maps_update_done(); |
1da177e4 LT |
891 | return err; |
892 | } | |
4dddfb5f | 893 | |
af1f4045 TG |
894 | int cpu_down(unsigned int cpu) |
895 | { | |
896 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
897 | } | |
b62b8ef9 | 898 | EXPORT_SYMBOL(cpu_down); |
4dddfb5f PZ |
899 | |
900 | #else | |
901 | #define takedown_cpu NULL | |
1da177e4 LT |
902 | #endif /*CONFIG_HOTPLUG_CPU*/ |
903 | ||
4baa0afc | 904 | /** |
ee1e714b | 905 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
906 | * @cpu: cpu that just started |
907 | * | |
4baa0afc TG |
908 | * It must be called by the arch code on the new cpu, before the new cpu |
909 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
910 | */ | |
911 | void notify_cpu_starting(unsigned int cpu) | |
912 | { | |
913 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
914 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
724a8688 | 915 | int ret; |
4baa0afc | 916 | |
0c6d4576 | 917 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 918 | while (st->state < target) { |
4baa0afc | 919 | st->state++; |
724a8688 PZ |
920 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
921 | /* | |
922 | * STARTING must not fail! | |
923 | */ | |
924 | WARN_ON_ONCE(ret); | |
4baa0afc TG |
925 | } |
926 | } | |
927 | ||
949338e3 | 928 | /* |
9cd4f1a4 TG |
929 | * Called from the idle task. Wake up the controlling task which brings the |
930 | * stopper and the hotplug thread of the upcoming CPU up and then delegates | |
931 | * the rest of the online bringup to the hotplug thread. | |
949338e3 | 932 | */ |
8df3e07e | 933 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 934 | { |
8df3e07e | 935 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
936 | |
937 | /* Happens for the boot cpu */ | |
938 | if (state != CPUHP_AP_ONLINE_IDLE) | |
939 | return; | |
940 | ||
941 | st->state = CPUHP_AP_ONLINE_IDLE; | |
9cd4f1a4 | 942 | complete(&st->done); |
949338e3 TG |
943 | } |
944 | ||
e3920fb4 | 945 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 946 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 947 | { |
cff7d378 | 948 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 949 | struct task_struct *idle; |
2e1a3483 | 950 | int ret = 0; |
1da177e4 | 951 | |
8f553c49 | 952 | cpus_write_lock(); |
38498a67 | 953 | |
757c989b | 954 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
955 | ret = -EINVAL; |
956 | goto out; | |
957 | } | |
958 | ||
757c989b TG |
959 | /* |
960 | * The caller of do_cpu_up might have raced with another | |
961 | * caller. Ignore it for now. | |
962 | */ | |
963 | if (st->state >= target) | |
38498a67 | 964 | goto out; |
757c989b TG |
965 | |
966 | if (st->state == CPUHP_OFFLINE) { | |
967 | /* Let it fail before we try to bring the cpu up */ | |
968 | idle = idle_thread_get(cpu); | |
969 | if (IS_ERR(idle)) { | |
970 | ret = PTR_ERR(idle); | |
971 | goto out; | |
972 | } | |
3bb5d2ee | 973 | } |
38498a67 | 974 | |
ba997462 TG |
975 | cpuhp_tasks_frozen = tasks_frozen; |
976 | ||
4dddfb5f | 977 | cpuhp_set_state(st, target); |
1cf4f629 TG |
978 | /* |
979 | * If the current CPU state is in the range of the AP hotplug thread, | |
980 | * then we need to kick the thread once more. | |
981 | */ | |
8df3e07e | 982 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
983 | ret = cpuhp_kick_ap_work(cpu); |
984 | /* | |
985 | * The AP side has done the error rollback already. Just | |
986 | * return the error code.. | |
987 | */ | |
988 | if (ret) | |
989 | goto out; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 994 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
995 | * responsible for bringing it up to the target state. |
996 | */ | |
8df3e07e | 997 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 998 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 999 | out: |
8f553c49 | 1000 | cpus_write_unlock(); |
e3920fb4 RW |
1001 | return ret; |
1002 | } | |
1003 | ||
af1f4045 | 1004 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1005 | { |
1006 | int err = 0; | |
cf23422b | 1007 | |
e0b582ec | 1008 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1009 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1010 | cpu); | |
87d5e023 | 1011 | #if defined(CONFIG_IA64) |
84117da5 | 1012 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1013 | #endif |
1014 | return -EINVAL; | |
1015 | } | |
e3920fb4 | 1016 | |
01b0f197 TK |
1017 | err = try_online_node(cpu_to_node(cpu)); |
1018 | if (err) | |
1019 | return err; | |
cf23422b | 1020 | |
d221938c | 1021 | cpu_maps_update_begin(); |
e761b772 MK |
1022 | |
1023 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1024 | err = -EBUSY; |
e761b772 MK |
1025 | goto out; |
1026 | } | |
1027 | ||
af1f4045 | 1028 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1029 | out: |
d221938c | 1030 | cpu_maps_update_done(); |
e3920fb4 RW |
1031 | return err; |
1032 | } | |
af1f4045 TG |
1033 | |
1034 | int cpu_up(unsigned int cpu) | |
1035 | { | |
1036 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1037 | } | |
a513f6ba | 1038 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1039 | |
f3de4be9 | 1040 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1041 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1042 | |
d391e552 | 1043 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1044 | { |
d391e552 | 1045 | int cpu, error = 0; |
e3920fb4 | 1046 | |
d221938c | 1047 | cpu_maps_update_begin(); |
d391e552 JM |
1048 | if (!cpu_online(primary)) |
1049 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
1050 | /* |
1051 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1052 | * with the userspace trying to use the CPU hotplug at the same time |
1053 | */ | |
e0b582ec | 1054 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1055 | |
84117da5 | 1056 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1057 | for_each_online_cpu(cpu) { |
d391e552 | 1058 | if (cpu == primary) |
e3920fb4 | 1059 | continue; |
bb3632c6 | 1060 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1061 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1062 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1063 | if (!error) |
e0b582ec | 1064 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1065 | else { |
84117da5 | 1066 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1067 | break; |
1068 | } | |
1069 | } | |
86886e55 | 1070 | |
89af7ba5 | 1071 | if (!error) |
e3920fb4 | 1072 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1073 | else |
84117da5 | 1074 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1075 | |
1076 | /* | |
1077 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1078 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1079 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1080 | */ | |
1081 | cpu_hotplug_disabled++; | |
1082 | ||
d221938c | 1083 | cpu_maps_update_done(); |
e3920fb4 RW |
1084 | return error; |
1085 | } | |
1086 | ||
d0af9eed SS |
1087 | void __weak arch_enable_nonboot_cpus_begin(void) |
1088 | { | |
1089 | } | |
1090 | ||
1091 | void __weak arch_enable_nonboot_cpus_end(void) | |
1092 | { | |
1093 | } | |
1094 | ||
71cf5aee | 1095 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1096 | { |
1097 | int cpu, error; | |
1098 | ||
1099 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1100 | cpu_maps_update_begin(); |
01b41159 | 1101 | __cpu_hotplug_enable(); |
e0b582ec | 1102 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1103 | goto out; |
e3920fb4 | 1104 | |
84117da5 | 1105 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1106 | |
1107 | arch_enable_nonboot_cpus_begin(); | |
1108 | ||
e0b582ec | 1109 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1110 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1111 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1112 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1113 | if (!error) { |
84117da5 | 1114 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1115 | continue; |
1116 | } | |
84117da5 | 1117 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1118 | } |
d0af9eed SS |
1119 | |
1120 | arch_enable_nonboot_cpus_end(); | |
1121 | ||
e0b582ec | 1122 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1123 | out: |
d221938c | 1124 | cpu_maps_update_done(); |
1da177e4 | 1125 | } |
e0b582ec | 1126 | |
d7268a31 | 1127 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1128 | { |
1129 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1130 | return -ENOMEM; | |
1131 | return 0; | |
1132 | } | |
1133 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1134 | |
79cfbdfa SB |
1135 | /* |
1136 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1137 | * ensure that the state of the system with respect to the tasks being frozen | |
1138 | * or not, as reported by the notification, remains unchanged *throughout the | |
1139 | * duration* of the execution of the callbacks. | |
1140 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1141 | * | |
1142 | * This synchronization is implemented by mutually excluding regular CPU | |
1143 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1144 | * Hibernate notifications. | |
1145 | */ | |
1146 | static int | |
1147 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1148 | unsigned long action, void *ptr) | |
1149 | { | |
1150 | switch (action) { | |
1151 | ||
1152 | case PM_SUSPEND_PREPARE: | |
1153 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1154 | cpu_hotplug_disable(); |
79cfbdfa SB |
1155 | break; |
1156 | ||
1157 | case PM_POST_SUSPEND: | |
1158 | case PM_POST_HIBERNATION: | |
16e53dbf | 1159 | cpu_hotplug_enable(); |
79cfbdfa SB |
1160 | break; |
1161 | ||
1162 | default: | |
1163 | return NOTIFY_DONE; | |
1164 | } | |
1165 | ||
1166 | return NOTIFY_OK; | |
1167 | } | |
1168 | ||
1169 | ||
d7268a31 | 1170 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1171 | { |
6e32d479 FY |
1172 | /* |
1173 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1174 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1175 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1176 | */ | |
79cfbdfa SB |
1177 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1178 | return 0; | |
1179 | } | |
1180 | core_initcall(cpu_hotplug_pm_sync_init); | |
1181 | ||
f3de4be9 | 1182 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1183 | |
8ce371f9 PZ |
1184 | int __boot_cpu_id; |
1185 | ||
68f4f1ec | 1186 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1187 | |
cff7d378 TG |
1188 | /* Boot processor state steps */ |
1189 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1190 | [CPUHP_OFFLINE] = { | |
1191 | .name = "offline", | |
3c1627e9 TG |
1192 | .startup.single = NULL, |
1193 | .teardown.single = NULL, | |
cff7d378 TG |
1194 | }, |
1195 | #ifdef CONFIG_SMP | |
1196 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1197 | .name = "threads:prepare", |
3c1627e9 TG |
1198 | .startup.single = smpboot_create_threads, |
1199 | .teardown.single = NULL, | |
757c989b | 1200 | .cant_stop = true, |
cff7d378 | 1201 | }, |
00e16c3d | 1202 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1203 | .name = "perf:prepare", |
1204 | .startup.single = perf_event_init_cpu, | |
1205 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1206 | }, |
7ee681b2 | 1207 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1208 | .name = "workqueue:prepare", |
1209 | .startup.single = workqueue_prepare_cpu, | |
1210 | .teardown.single = NULL, | |
7ee681b2 | 1211 | }, |
27590dc1 | 1212 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1213 | .name = "hrtimers:prepare", |
1214 | .startup.single = hrtimers_prepare_cpu, | |
1215 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1216 | }, |
31487f83 | 1217 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1218 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1219 | .startup.single = smpcfd_prepare_cpu, |
1220 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1221 | }, |
e6d4989a RW |
1222 | [CPUHP_RELAY_PREPARE] = { |
1223 | .name = "relay:prepare", | |
1224 | .startup.single = relay_prepare_cpu, | |
1225 | .teardown.single = NULL, | |
1226 | }, | |
6731d4f1 SAS |
1227 | [CPUHP_SLAB_PREPARE] = { |
1228 | .name = "slab:prepare", | |
1229 | .startup.single = slab_prepare_cpu, | |
1230 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1231 | }, |
4df83742 | 1232 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1233 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1234 | .startup.single = rcutree_prepare_cpu, |
1235 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1236 | }, |
4fae16df RC |
1237 | /* |
1238 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1239 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1240 | * otherwise a RCU stall occurs. | |
1241 | */ | |
1242 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1243 | .name = "timers:dead", |
1244 | .startup.single = NULL, | |
1245 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1246 | }, |
d10ef6f9 | 1247 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1248 | [CPUHP_BRINGUP_CPU] = { |
1249 | .name = "cpu:bringup", | |
3c1627e9 TG |
1250 | .startup.single = bringup_cpu, |
1251 | .teardown.single = NULL, | |
757c989b | 1252 | .cant_stop = true, |
4baa0afc | 1253 | }, |
31487f83 | 1254 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1255 | .name = "smpcfd:dying", |
3c1627e9 TG |
1256 | .startup.single = NULL, |
1257 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1258 | }, |
d10ef6f9 TG |
1259 | /* |
1260 | * Handled on controll processor until the plugged processor manages | |
1261 | * this itself. | |
1262 | */ | |
4baa0afc TG |
1263 | [CPUHP_TEARDOWN_CPU] = { |
1264 | .name = "cpu:teardown", | |
3c1627e9 TG |
1265 | .startup.single = NULL, |
1266 | .teardown.single = takedown_cpu, | |
757c989b | 1267 | .cant_stop = true, |
cff7d378 | 1268 | }, |
a7c73414 TG |
1269 | #else |
1270 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1271 | #endif |
cff7d378 TG |
1272 | }; |
1273 | ||
4baa0afc TG |
1274 | /* Application processor state steps */ |
1275 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1276 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1277 | /* Final state before CPU kills itself */ |
1278 | [CPUHP_AP_IDLE_DEAD] = { | |
1279 | .name = "idle:dead", | |
1280 | }, | |
1281 | /* | |
1282 | * Last state before CPU enters the idle loop to die. Transient state | |
1283 | * for synchronization. | |
1284 | */ | |
1285 | [CPUHP_AP_OFFLINE] = { | |
1286 | .name = "ap:offline", | |
1287 | .cant_stop = true, | |
1288 | }, | |
9cf7243d TG |
1289 | /* First state is scheduler control. Interrupts are disabled */ |
1290 | [CPUHP_AP_SCHED_STARTING] = { | |
1291 | .name = "sched:starting", | |
3c1627e9 TG |
1292 | .startup.single = sched_cpu_starting, |
1293 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1294 | }, |
4df83742 | 1295 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1296 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1297 | .startup.single = NULL, |
1298 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1299 | }, |
d10ef6f9 TG |
1300 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1301 | * state for synchronsization */ | |
1302 | [CPUHP_AP_ONLINE] = { | |
1303 | .name = "ap:online", | |
1304 | }, | |
1305 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1306 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1307 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1308 | .startup.single = smpboot_unpark_threads, |
1309 | .teardown.single = NULL, | |
1cf4f629 | 1310 | }, |
c5cb83bb TG |
1311 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1312 | .name = "irq/affinity:online", | |
1313 | .startup.single = irq_affinity_online_cpu, | |
1314 | .teardown.single = NULL, | |
1315 | }, | |
00e16c3d | 1316 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1317 | .name = "perf:online", |
1318 | .startup.single = perf_event_init_cpu, | |
1319 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1320 | }, |
7ee681b2 | 1321 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1322 | .name = "workqueue:online", |
1323 | .startup.single = workqueue_online_cpu, | |
1324 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1325 | }, |
4df83742 | 1326 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1327 | .name = "RCU/tree:online", |
3c1627e9 TG |
1328 | .startup.single = rcutree_online_cpu, |
1329 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1330 | }, |
4baa0afc | 1331 | #endif |
d10ef6f9 TG |
1332 | /* |
1333 | * The dynamically registered state space is here | |
1334 | */ | |
1335 | ||
aaddd7d1 TG |
1336 | #ifdef CONFIG_SMP |
1337 | /* Last state is scheduler control setting the cpu active */ | |
1338 | [CPUHP_AP_ACTIVE] = { | |
1339 | .name = "sched:active", | |
3c1627e9 TG |
1340 | .startup.single = sched_cpu_activate, |
1341 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1342 | }, |
1343 | #endif | |
1344 | ||
d10ef6f9 | 1345 | /* CPU is fully up and running. */ |
4baa0afc TG |
1346 | [CPUHP_ONLINE] = { |
1347 | .name = "online", | |
3c1627e9 TG |
1348 | .startup.single = NULL, |
1349 | .teardown.single = NULL, | |
4baa0afc TG |
1350 | }, |
1351 | }; | |
1352 | ||
5b7aa87e TG |
1353 | /* Sanity check for callbacks */ |
1354 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1355 | { | |
1356 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1357 | return -EINVAL; | |
1358 | return 0; | |
1359 | } | |
1360 | ||
dc280d93 TG |
1361 | /* |
1362 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1363 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1364 | * by having no name assigned. | |
1365 | */ | |
1366 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1367 | { | |
4205e478 TG |
1368 | enum cpuhp_state i, end; |
1369 | struct cpuhp_step *step; | |
dc280d93 | 1370 | |
4205e478 TG |
1371 | switch (state) { |
1372 | case CPUHP_AP_ONLINE_DYN: | |
1373 | step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; | |
1374 | end = CPUHP_AP_ONLINE_DYN_END; | |
1375 | break; | |
1376 | case CPUHP_BP_PREPARE_DYN: | |
1377 | step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; | |
1378 | end = CPUHP_BP_PREPARE_DYN_END; | |
1379 | break; | |
1380 | default: | |
1381 | return -EINVAL; | |
1382 | } | |
1383 | ||
1384 | for (i = state; i <= end; i++, step++) { | |
1385 | if (!step->name) | |
dc280d93 TG |
1386 | return i; |
1387 | } | |
1388 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1389 | return -ENOSPC; | |
1390 | } | |
1391 | ||
1392 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1393 | int (*startup)(unsigned int cpu), | |
1394 | int (*teardown)(unsigned int cpu), | |
1395 | bool multi_instance) | |
5b7aa87e TG |
1396 | { |
1397 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1398 | struct cpuhp_step *sp; | |
dc280d93 | 1399 | int ret = 0; |
5b7aa87e | 1400 | |
0c96b273 EB |
1401 | /* |
1402 | * If name is NULL, then the state gets removed. | |
1403 | * | |
1404 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
1405 | * the first allocation from these dynamic ranges, so the removal | |
1406 | * would trigger a new allocation and clear the wrong (already | |
1407 | * empty) state, leaving the callbacks of the to be cleared state | |
1408 | * dangling, which causes wreckage on the next hotplug operation. | |
1409 | */ | |
1410 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
1411 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
1412 | ret = cpuhp_reserve_state(state); |
1413 | if (ret < 0) | |
dc434e05 | 1414 | return ret; |
dc280d93 TG |
1415 | state = ret; |
1416 | } | |
5b7aa87e | 1417 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1418 | if (name && sp->name) |
1419 | return -EBUSY; | |
1420 | ||
3c1627e9 TG |
1421 | sp->startup.single = startup; |
1422 | sp->teardown.single = teardown; | |
5b7aa87e | 1423 | sp->name = name; |
cf392d10 TG |
1424 | sp->multi_instance = multi_instance; |
1425 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1426 | return ret; |
5b7aa87e TG |
1427 | } |
1428 | ||
1429 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1430 | { | |
3c1627e9 | 1431 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1432 | } |
1433 | ||
5b7aa87e TG |
1434 | /* |
1435 | * Call the startup/teardown function for a step either on the AP or | |
1436 | * on the current CPU. | |
1437 | */ | |
cf392d10 TG |
1438 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1439 | struct hlist_node *node) | |
5b7aa87e | 1440 | { |
a724632c | 1441 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1442 | int ret; |
1443 | ||
4dddfb5f PZ |
1444 | /* |
1445 | * If there's nothing to do, we done. | |
1446 | * Relies on the union for multi_instance. | |
1447 | */ | |
3c1627e9 TG |
1448 | if ((bringup && !sp->startup.single) || |
1449 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1450 | return 0; |
5b7aa87e TG |
1451 | /* |
1452 | * The non AP bound callbacks can fail on bringup. On teardown | |
1453 | * e.g. module removal we crash for now. | |
1454 | */ | |
1cf4f629 TG |
1455 | #ifdef CONFIG_SMP |
1456 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1457 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1458 | else |
96abb968 | 1459 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1460 | #else |
96abb968 | 1461 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1462 | #endif |
5b7aa87e TG |
1463 | BUG_ON(ret && !bringup); |
1464 | return ret; | |
1465 | } | |
1466 | ||
1467 | /* | |
1468 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1469 | * | |
1470 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1471 | */ | |
1472 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1473 | struct hlist_node *node) |
5b7aa87e TG |
1474 | { |
1475 | int cpu; | |
1476 | ||
5b7aa87e TG |
1477 | /* Roll back the already executed steps on the other cpus */ |
1478 | for_each_present_cpu(cpu) { | |
1479 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1480 | int cpustate = st->state; | |
1481 | ||
1482 | if (cpu >= failedcpu) | |
1483 | break; | |
1484 | ||
1485 | /* Did we invoke the startup call on that cpu ? */ | |
1486 | if (cpustate >= state) | |
cf392d10 | 1487 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1488 | } |
1489 | } | |
1490 | ||
9805c673 TG |
1491 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1492 | struct hlist_node *node, | |
1493 | bool invoke) | |
cf392d10 TG |
1494 | { |
1495 | struct cpuhp_step *sp; | |
1496 | int cpu; | |
1497 | int ret; | |
1498 | ||
9805c673 TG |
1499 | lockdep_assert_cpus_held(); |
1500 | ||
cf392d10 TG |
1501 | sp = cpuhp_get_step(state); |
1502 | if (sp->multi_instance == false) | |
1503 | return -EINVAL; | |
1504 | ||
dc434e05 | 1505 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1506 | |
3c1627e9 | 1507 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1508 | goto add_node; |
1509 | ||
1510 | /* | |
1511 | * Try to call the startup callback for each present cpu | |
1512 | * depending on the hotplug state of the cpu. | |
1513 | */ | |
1514 | for_each_present_cpu(cpu) { | |
1515 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1516 | int cpustate = st->state; | |
1517 | ||
1518 | if (cpustate < state) | |
1519 | continue; | |
1520 | ||
1521 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1522 | if (ret) { | |
3c1627e9 | 1523 | if (sp->teardown.multi) |
cf392d10 | 1524 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1525 | goto unlock; |
cf392d10 TG |
1526 | } |
1527 | } | |
1528 | add_node: | |
1529 | ret = 0; | |
cf392d10 | 1530 | hlist_add_head(node, &sp->list); |
dc434e05 | 1531 | unlock: |
cf392d10 | 1532 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1533 | return ret; |
1534 | } | |
1535 | ||
1536 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1537 | bool invoke) | |
1538 | { | |
1539 | int ret; | |
1540 | ||
1541 | cpus_read_lock(); | |
1542 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1543 | cpus_read_unlock(); |
cf392d10 TG |
1544 | return ret; |
1545 | } | |
1546 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1547 | ||
5b7aa87e | 1548 | /** |
71def423 | 1549 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1550 | * @state: The state to setup |
1551 | * @invoke: If true, the startup function is invoked for cpus where | |
1552 | * cpu state >= @state | |
1553 | * @startup: startup callback function | |
1554 | * @teardown: teardown callback function | |
1555 | * @multi_instance: State is set up for multiple instances which get | |
1556 | * added afterwards. | |
5b7aa87e | 1557 | * |
71def423 | 1558 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1559 | * Returns: |
1560 | * On success: | |
1561 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1562 | * 0 for all other states | |
1563 | * On failure: proper (negative) error code | |
5b7aa87e | 1564 | */ |
71def423 SAS |
1565 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1566 | const char *name, bool invoke, | |
1567 | int (*startup)(unsigned int cpu), | |
1568 | int (*teardown)(unsigned int cpu), | |
1569 | bool multi_instance) | |
5b7aa87e TG |
1570 | { |
1571 | int cpu, ret = 0; | |
b9d9d691 | 1572 | bool dynstate; |
5b7aa87e | 1573 | |
71def423 SAS |
1574 | lockdep_assert_cpus_held(); |
1575 | ||
5b7aa87e TG |
1576 | if (cpuhp_cb_check(state) || !name) |
1577 | return -EINVAL; | |
1578 | ||
dc434e05 | 1579 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1580 | |
dc280d93 TG |
1581 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1582 | multi_instance); | |
5b7aa87e | 1583 | |
b9d9d691 TG |
1584 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1585 | if (ret > 0 && dynstate) { | |
1586 | state = ret; | |
1587 | ret = 0; | |
1588 | } | |
1589 | ||
dc280d93 | 1590 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1591 | goto out; |
1592 | ||
1593 | /* | |
1594 | * Try to call the startup callback for each present cpu | |
1595 | * depending on the hotplug state of the cpu. | |
1596 | */ | |
1597 | for_each_present_cpu(cpu) { | |
1598 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1599 | int cpustate = st->state; | |
1600 | ||
1601 | if (cpustate < state) | |
1602 | continue; | |
1603 | ||
cf392d10 | 1604 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1605 | if (ret) { |
a724632c | 1606 | if (teardown) |
cf392d10 TG |
1607 | cpuhp_rollback_install(cpu, state, NULL); |
1608 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1609 | goto out; |
1610 | } | |
1611 | } | |
1612 | out: | |
dc434e05 | 1613 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1614 | /* |
1615 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1616 | * dynamically allocated state in case of success. | |
1617 | */ | |
b9d9d691 | 1618 | if (!ret && dynstate) |
5b7aa87e TG |
1619 | return state; |
1620 | return ret; | |
1621 | } | |
71def423 SAS |
1622 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1623 | ||
1624 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1625 | const char *name, bool invoke, | |
1626 | int (*startup)(unsigned int cpu), | |
1627 | int (*teardown)(unsigned int cpu), | |
1628 | bool multi_instance) | |
1629 | { | |
1630 | int ret; | |
1631 | ||
1632 | cpus_read_lock(); | |
1633 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1634 | teardown, multi_instance); | |
1635 | cpus_read_unlock(); | |
1636 | return ret; | |
1637 | } | |
5b7aa87e TG |
1638 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1639 | ||
cf392d10 TG |
1640 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1641 | struct hlist_node *node, bool invoke) | |
1642 | { | |
1643 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1644 | int cpu; | |
1645 | ||
1646 | BUG_ON(cpuhp_cb_check(state)); | |
1647 | ||
1648 | if (!sp->multi_instance) | |
1649 | return -EINVAL; | |
1650 | ||
8f553c49 | 1651 | cpus_read_lock(); |
dc434e05 SAS |
1652 | mutex_lock(&cpuhp_state_mutex); |
1653 | ||
cf392d10 TG |
1654 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1655 | goto remove; | |
1656 | /* | |
1657 | * Call the teardown callback for each present cpu depending | |
1658 | * on the hotplug state of the cpu. This function is not | |
1659 | * allowed to fail currently! | |
1660 | */ | |
1661 | for_each_present_cpu(cpu) { | |
1662 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1663 | int cpustate = st->state; | |
1664 | ||
1665 | if (cpustate >= state) | |
1666 | cpuhp_issue_call(cpu, state, false, node); | |
1667 | } | |
1668 | ||
1669 | remove: | |
cf392d10 TG |
1670 | hlist_del(node); |
1671 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1672 | cpus_read_unlock(); |
cf392d10 TG |
1673 | |
1674 | return 0; | |
1675 | } | |
1676 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1677 | |
5b7aa87e | 1678 | /** |
71def423 | 1679 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1680 | * @state: The state to remove |
1681 | * @invoke: If true, the teardown function is invoked for cpus where | |
1682 | * cpu state >= @state | |
1683 | * | |
71def423 | 1684 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1685 | * The teardown callback is currently not allowed to fail. Think |
1686 | * about module removal! | |
1687 | */ | |
71def423 | 1688 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1689 | { |
cf392d10 | 1690 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1691 | int cpu; |
1692 | ||
1693 | BUG_ON(cpuhp_cb_check(state)); | |
1694 | ||
71def423 | 1695 | lockdep_assert_cpus_held(); |
5b7aa87e | 1696 | |
dc434e05 | 1697 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1698 | if (sp->multi_instance) { |
1699 | WARN(!hlist_empty(&sp->list), | |
1700 | "Error: Removing state %d which has instances left.\n", | |
1701 | state); | |
1702 | goto remove; | |
1703 | } | |
1704 | ||
a724632c | 1705 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1706 | goto remove; |
1707 | ||
1708 | /* | |
1709 | * Call the teardown callback for each present cpu depending | |
1710 | * on the hotplug state of the cpu. This function is not | |
1711 | * allowed to fail currently! | |
1712 | */ | |
1713 | for_each_present_cpu(cpu) { | |
1714 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1715 | int cpustate = st->state; | |
1716 | ||
1717 | if (cpustate >= state) | |
cf392d10 | 1718 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1719 | } |
1720 | remove: | |
cf392d10 | 1721 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1722 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1723 | } |
1724 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1725 | ||
1726 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1727 | { | |
1728 | cpus_read_lock(); | |
1729 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1730 | cpus_read_unlock(); |
5b7aa87e TG |
1731 | } |
1732 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1733 | ||
98f8cdce TG |
1734 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1735 | static ssize_t show_cpuhp_state(struct device *dev, | |
1736 | struct device_attribute *attr, char *buf) | |
1737 | { | |
1738 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1739 | ||
1740 | return sprintf(buf, "%d\n", st->state); | |
1741 | } | |
1742 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1743 | ||
757c989b TG |
1744 | static ssize_t write_cpuhp_target(struct device *dev, |
1745 | struct device_attribute *attr, | |
1746 | const char *buf, size_t count) | |
1747 | { | |
1748 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1749 | struct cpuhp_step *sp; | |
1750 | int target, ret; | |
1751 | ||
1752 | ret = kstrtoint(buf, 10, &target); | |
1753 | if (ret) | |
1754 | return ret; | |
1755 | ||
1756 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1757 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1758 | return -EINVAL; | |
1759 | #else | |
1760 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1761 | return -EINVAL; | |
1762 | #endif | |
1763 | ||
1764 | ret = lock_device_hotplug_sysfs(); | |
1765 | if (ret) | |
1766 | return ret; | |
1767 | ||
1768 | mutex_lock(&cpuhp_state_mutex); | |
1769 | sp = cpuhp_get_step(target); | |
1770 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1771 | mutex_unlock(&cpuhp_state_mutex); | |
1772 | if (ret) | |
40da1b11 | 1773 | goto out; |
757c989b TG |
1774 | |
1775 | if (st->state < target) | |
1776 | ret = do_cpu_up(dev->id, target); | |
1777 | else | |
1778 | ret = do_cpu_down(dev->id, target); | |
40da1b11 | 1779 | out: |
757c989b TG |
1780 | unlock_device_hotplug(); |
1781 | return ret ? ret : count; | |
1782 | } | |
1783 | ||
98f8cdce TG |
1784 | static ssize_t show_cpuhp_target(struct device *dev, |
1785 | struct device_attribute *attr, char *buf) | |
1786 | { | |
1787 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1788 | ||
1789 | return sprintf(buf, "%d\n", st->target); | |
1790 | } | |
757c989b | 1791 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1792 | |
1793 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1794 | &dev_attr_state.attr, | |
1795 | &dev_attr_target.attr, | |
1796 | NULL | |
1797 | }; | |
1798 | ||
993647a2 | 1799 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
1800 | .attrs = cpuhp_cpu_attrs, |
1801 | .name = "hotplug", | |
1802 | NULL | |
1803 | }; | |
1804 | ||
1805 | static ssize_t show_cpuhp_states(struct device *dev, | |
1806 | struct device_attribute *attr, char *buf) | |
1807 | { | |
1808 | ssize_t cur, res = 0; | |
1809 | int i; | |
1810 | ||
1811 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1812 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1813 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1814 | ||
1815 | if (sp->name) { | |
1816 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1817 | buf += cur; | |
1818 | res += cur; | |
1819 | } | |
1820 | } | |
1821 | mutex_unlock(&cpuhp_state_mutex); | |
1822 | return res; | |
1823 | } | |
1824 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1825 | ||
1826 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1827 | &dev_attr_states.attr, | |
1828 | NULL | |
1829 | }; | |
1830 | ||
993647a2 | 1831 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
1832 | .attrs = cpuhp_cpu_root_attrs, |
1833 | .name = "hotplug", | |
1834 | NULL | |
1835 | }; | |
1836 | ||
1837 | static int __init cpuhp_sysfs_init(void) | |
1838 | { | |
1839 | int cpu, ret; | |
1840 | ||
1841 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1842 | &cpuhp_cpu_root_attr_group); | |
1843 | if (ret) | |
1844 | return ret; | |
1845 | ||
1846 | for_each_possible_cpu(cpu) { | |
1847 | struct device *dev = get_cpu_device(cpu); | |
1848 | ||
1849 | if (!dev) | |
1850 | continue; | |
1851 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1852 | if (ret) | |
1853 | return ret; | |
1854 | } | |
1855 | return 0; | |
1856 | } | |
1857 | device_initcall(cpuhp_sysfs_init); | |
1858 | #endif | |
1859 | ||
e56b3bc7 LT |
1860 | /* |
1861 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1862 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1863 | * | |
e0b582ec | 1864 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1865 | * mask value that has a single bit set only. |
1866 | */ | |
b8d317d1 | 1867 | |
e56b3bc7 | 1868 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1869 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1870 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1871 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1872 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1873 | |
e56b3bc7 LT |
1874 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1875 | ||
1876 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1877 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1878 | #if BITS_PER_LONG > 32 | |
1879 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1880 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1881 | #endif |
1882 | }; | |
e56b3bc7 | 1883 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1884 | |
1885 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1886 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1887 | |
1888 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1889 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1890 | = {CPU_BITS_ALL}; |
b3199c02 | 1891 | #else |
4b804c85 | 1892 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1893 | #endif |
4b804c85 | 1894 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1895 | |
4b804c85 RV |
1896 | struct cpumask __cpu_online_mask __read_mostly; |
1897 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1898 | |
4b804c85 RV |
1899 | struct cpumask __cpu_present_mask __read_mostly; |
1900 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1901 | |
4b804c85 RV |
1902 | struct cpumask __cpu_active_mask __read_mostly; |
1903 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1904 | |
3fa41520 RR |
1905 | void init_cpu_present(const struct cpumask *src) |
1906 | { | |
c4c54dd1 | 1907 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1908 | } |
1909 | ||
1910 | void init_cpu_possible(const struct cpumask *src) | |
1911 | { | |
c4c54dd1 | 1912 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1913 | } |
1914 | ||
1915 | void init_cpu_online(const struct cpumask *src) | |
1916 | { | |
c4c54dd1 | 1917 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1918 | } |
cff7d378 TG |
1919 | |
1920 | /* | |
1921 | * Activate the first processor. | |
1922 | */ | |
1923 | void __init boot_cpu_init(void) | |
1924 | { | |
1925 | int cpu = smp_processor_id(); | |
1926 | ||
1927 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1928 | set_cpu_online(cpu, true); | |
1929 | set_cpu_active(cpu, true); | |
1930 | set_cpu_present(cpu, true); | |
1931 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
1932 | |
1933 | #ifdef CONFIG_SMP | |
1934 | __boot_cpu_id = cpu; | |
1935 | #endif | |
cff7d378 TG |
1936 | } |
1937 | ||
1938 | /* | |
1939 | * Must be called _AFTER_ setting up the per_cpu areas | |
1940 | */ | |
1941 | void __init boot_cpu_state_init(void) | |
1942 | { | |
1943 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1944 | } |