]>
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 | 48 | * @result: Result of the operation |
5ebe7742 PZ |
49 | * @done_up: Signal completion to the issuer of the task for cpu-up |
50 | * @done_down: Signal completion to the issuer of the task for cpu-down | |
cff7d378 TG |
51 | */ |
52 | struct cpuhp_cpu_state { | |
53 | enum cpuhp_state state; | |
54 | enum cpuhp_state target; | |
4cb28ced TG |
55 | #ifdef CONFIG_SMP |
56 | struct task_struct *thread; | |
57 | bool should_run; | |
3b9d6da6 | 58 | bool rollback; |
a724632c TG |
59 | bool single; |
60 | bool bringup; | |
cf392d10 | 61 | struct hlist_node *node; |
4dddfb5f | 62 | struct hlist_node *last; |
4cb28ced | 63 | enum cpuhp_state cb_state; |
4cb28ced | 64 | int result; |
5ebe7742 PZ |
65 | struct completion done_up; |
66 | struct completion done_down; | |
4cb28ced | 67 | #endif |
cff7d378 TG |
68 | }; |
69 | ||
70 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
71 | ||
49dfe2a6 | 72 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
73 | static struct lockdep_map cpuhp_state_up_map = |
74 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
75 | static struct lockdep_map cpuhp_state_down_map = | |
76 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
77 | ||
78 | ||
79 | static void inline cpuhp_lock_acquire(bool bringup) | |
80 | { | |
81 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
82 | } | |
83 | ||
84 | static void inline cpuhp_lock_release(bool bringup) | |
85 | { | |
86 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
87 | } | |
88 | #else | |
89 | ||
90 | static void inline cpuhp_lock_acquire(bool bringup) { } | |
91 | static void inline cpuhp_lock_release(bool bringup) { } | |
92 | ||
49dfe2a6 TG |
93 | #endif |
94 | ||
cff7d378 TG |
95 | /** |
96 | * cpuhp_step - Hotplug state machine step | |
97 | * @name: Name of the step | |
98 | * @startup: Startup function of the step | |
99 | * @teardown: Teardown function of the step | |
100 | * @skip_onerr: Do not invoke the functions on error rollback | |
101 | * Will go away once the notifiers are gone | |
757c989b | 102 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
103 | */ |
104 | struct cpuhp_step { | |
cf392d10 TG |
105 | const char *name; |
106 | union { | |
3c1627e9 TG |
107 | int (*single)(unsigned int cpu); |
108 | int (*multi)(unsigned int cpu, | |
109 | struct hlist_node *node); | |
110 | } startup; | |
cf392d10 | 111 | union { |
3c1627e9 TG |
112 | int (*single)(unsigned int cpu); |
113 | int (*multi)(unsigned int cpu, | |
114 | struct hlist_node *node); | |
115 | } teardown; | |
cf392d10 TG |
116 | struct hlist_head list; |
117 | bool skip_onerr; | |
118 | bool cant_stop; | |
119 | bool multi_instance; | |
cff7d378 TG |
120 | }; |
121 | ||
98f8cdce | 122 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 123 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 124 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 | 125 | |
a724632c TG |
126 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
127 | { | |
128 | /* | |
129 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
130 | * purposes as that state is handled explicitly in cpu_down. | |
131 | */ | |
132 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
133 | } | |
134 | ||
135 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
136 | { | |
137 | struct cpuhp_step *sp; | |
138 | ||
139 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
140 | return sp + state; | |
141 | } | |
142 | ||
cff7d378 TG |
143 | /** |
144 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
145 | * @cpu: The cpu for which the callback should be invoked | |
96abb968 | 146 | * @state: The state to do callbacks for |
a724632c | 147 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
148 | * @node: For multi-instance, do a single entry callback for install/remove |
149 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 150 | * |
cf392d10 | 151 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 152 | */ |
a724632c | 153 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
154 | bool bringup, struct hlist_node *node, |
155 | struct hlist_node **lastp) | |
cff7d378 TG |
156 | { |
157 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 158 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
159 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
160 | int (*cb)(unsigned int cpu); | |
161 | int ret, cnt; | |
162 | ||
163 | if (!step->multi_instance) { | |
96abb968 | 164 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 165 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
166 | if (!cb) |
167 | return 0; | |
a724632c | 168 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 169 | ret = cb(cpu); |
a724632c | 170 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
171 | return ret; |
172 | } | |
3c1627e9 | 173 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
174 | if (!cbm) |
175 | return 0; | |
176 | ||
177 | /* Single invocation for instance add/remove */ | |
178 | if (node) { | |
96abb968 | 179 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
180 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
181 | ret = cbm(cpu, node); | |
182 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
183 | return ret; | |
184 | } | |
185 | ||
186 | /* State transition. Invoke on all instances */ | |
187 | cnt = 0; | |
188 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
189 | if (lastp && node == *lastp) |
190 | break; | |
191 | ||
cf392d10 TG |
192 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
193 | ret = cbm(cpu, node); | |
194 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
195 | if (ret) { |
196 | if (!lastp) | |
197 | goto err; | |
198 | ||
199 | *lastp = node; | |
200 | return ret; | |
201 | } | |
cf392d10 TG |
202 | cnt++; |
203 | } | |
96abb968 PZ |
204 | if (lastp) |
205 | *lastp = NULL; | |
cf392d10 TG |
206 | return 0; |
207 | err: | |
208 | /* Rollback the instances if one failed */ | |
3c1627e9 | 209 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
210 | if (!cbm) |
211 | return ret; | |
212 | ||
213 | hlist_for_each(node, &step->list) { | |
214 | if (!cnt--) | |
215 | break; | |
724a8688 PZ |
216 | |
217 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
218 | ret = cbm(cpu, node); | |
219 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
220 | /* | |
221 | * Rollback must not fail, | |
222 | */ | |
223 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
224 | } |
225 | return ret; | |
226 | } | |
227 | ||
98a79d6a | 228 | #ifdef CONFIG_SMP |
5ebe7742 PZ |
229 | static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup) |
230 | { | |
231 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
232 | wait_for_completion(done); | |
233 | } | |
234 | ||
235 | static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup) | |
236 | { | |
237 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
238 | complete(done); | |
239 | } | |
240 | ||
241 | /* | |
242 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
243 | */ | |
244 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
245 | { | |
246 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
247 | } | |
248 | ||
b3199c02 | 249 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 250 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
251 | bool cpuhp_tasks_frozen; |
252 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 253 | |
79a6cdeb | 254 | /* |
93ae4f97 SB |
255 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
256 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
257 | */ |
258 | void cpu_maps_update_begin(void) | |
259 | { | |
260 | mutex_lock(&cpu_add_remove_lock); | |
261 | } | |
262 | ||
263 | void cpu_maps_update_done(void) | |
264 | { | |
265 | mutex_unlock(&cpu_add_remove_lock); | |
266 | } | |
1da177e4 | 267 | |
fc8dffd3 TG |
268 | /* |
269 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
270 | * Should always be manipulated under cpu_add_remove_lock |
271 | */ | |
272 | static int cpu_hotplug_disabled; | |
273 | ||
79a6cdeb LJ |
274 | #ifdef CONFIG_HOTPLUG_CPU |
275 | ||
fc8dffd3 | 276 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 277 | |
8f553c49 | 278 | void cpus_read_lock(void) |
a9d9baa1 | 279 | { |
fc8dffd3 | 280 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 281 | } |
8f553c49 | 282 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 283 | |
8f553c49 | 284 | void cpus_read_unlock(void) |
a9d9baa1 | 285 | { |
fc8dffd3 | 286 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 287 | } |
8f553c49 | 288 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 289 | |
8f553c49 | 290 | void cpus_write_lock(void) |
d221938c | 291 | { |
fc8dffd3 | 292 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 293 | } |
87af9e7f | 294 | |
8f553c49 | 295 | void cpus_write_unlock(void) |
d221938c | 296 | { |
fc8dffd3 | 297 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
298 | } |
299 | ||
fc8dffd3 | 300 | void lockdep_assert_cpus_held(void) |
d221938c | 301 | { |
fc8dffd3 | 302 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 303 | } |
79a6cdeb | 304 | |
16e53dbf SB |
305 | /* |
306 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
307 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
308 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
309 | * hotplug path before performing hotplug operations. So acquiring that lock | |
310 | * guarantees mutual exclusion from any currently running hotplug operations. | |
311 | */ | |
312 | void cpu_hotplug_disable(void) | |
313 | { | |
314 | cpu_maps_update_begin(); | |
89af7ba5 | 315 | cpu_hotplug_disabled++; |
16e53dbf SB |
316 | cpu_maps_update_done(); |
317 | } | |
32145c46 | 318 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 319 | |
01b41159 LW |
320 | static void __cpu_hotplug_enable(void) |
321 | { | |
322 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
323 | return; | |
324 | cpu_hotplug_disabled--; | |
325 | } | |
326 | ||
16e53dbf SB |
327 | void cpu_hotplug_enable(void) |
328 | { | |
329 | cpu_maps_update_begin(); | |
01b41159 | 330 | __cpu_hotplug_enable(); |
16e53dbf SB |
331 | cpu_maps_update_done(); |
332 | } | |
32145c46 | 333 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 334 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 335 | |
4dddfb5f PZ |
336 | static inline enum cpuhp_state |
337 | cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
338 | { | |
339 | enum cpuhp_state prev_state = st->state; | |
340 | ||
341 | st->rollback = false; | |
342 | st->last = NULL; | |
343 | ||
344 | st->target = target; | |
345 | st->single = false; | |
346 | st->bringup = st->state < target; | |
347 | ||
348 | return prev_state; | |
349 | } | |
350 | ||
351 | static inline void | |
352 | cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state) | |
353 | { | |
354 | st->rollback = true; | |
355 | ||
356 | /* | |
357 | * If we have st->last we need to undo partial multi_instance of this | |
358 | * state first. Otherwise start undo at the previous state. | |
359 | */ | |
360 | if (!st->last) { | |
361 | if (st->bringup) | |
362 | st->state--; | |
363 | else | |
364 | st->state++; | |
365 | } | |
366 | ||
367 | st->target = prev_state; | |
368 | st->bringup = !st->bringup; | |
369 | } | |
370 | ||
371 | /* Regular hotplug invocation of the AP hotplug thread */ | |
372 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
373 | { | |
374 | if (!st->single && st->state == st->target) | |
375 | return; | |
376 | ||
377 | st->result = 0; | |
378 | /* | |
379 | * Make sure the above stores are visible before should_run becomes | |
380 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
381 | */ | |
382 | smp_mb(); | |
383 | st->should_run = true; | |
384 | wake_up_process(st->thread); | |
5ebe7742 | 385 | wait_for_ap_thread(st, st->bringup); |
4dddfb5f PZ |
386 | } |
387 | ||
388 | static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
389 | { | |
390 | enum cpuhp_state prev_state; | |
391 | int ret; | |
392 | ||
393 | prev_state = cpuhp_set_state(st, target); | |
394 | __cpuhp_kick_ap(st); | |
395 | if ((ret = st->result)) { | |
396 | cpuhp_reset_state(st, prev_state); | |
397 | __cpuhp_kick_ap(st); | |
398 | } | |
399 | ||
400 | return ret; | |
401 | } | |
9cd4f1a4 | 402 | |
8df3e07e TG |
403 | static int bringup_wait_for_ap(unsigned int cpu) |
404 | { | |
405 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
406 | ||
9cd4f1a4 | 407 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
5ebe7742 | 408 | wait_for_ap_thread(st, true); |
dea1d0f5 TG |
409 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
410 | return -ECANCELED; | |
9cd4f1a4 TG |
411 | |
412 | /* Unpark the stopper thread and the hotplug thread of the target cpu */ | |
413 | stop_machine_unpark(cpu); | |
414 | kthread_unpark(st->thread); | |
415 | ||
4dddfb5f PZ |
416 | if (st->target <= CPUHP_AP_ONLINE_IDLE) |
417 | return 0; | |
418 | ||
419 | return cpuhp_kick_ap(st, st->target); | |
8df3e07e TG |
420 | } |
421 | ||
ba997462 TG |
422 | static int bringup_cpu(unsigned int cpu) |
423 | { | |
424 | struct task_struct *idle = idle_thread_get(cpu); | |
425 | int ret; | |
426 | ||
aa877175 BO |
427 | /* |
428 | * Some architectures have to walk the irq descriptors to | |
429 | * setup the vector space for the cpu which comes online. | |
430 | * Prevent irq alloc/free across the bringup. | |
431 | */ | |
432 | irq_lock_sparse(); | |
433 | ||
ba997462 TG |
434 | /* Arch-specific enabling code. */ |
435 | ret = __cpu_up(cpu, idle); | |
aa877175 | 436 | irq_unlock_sparse(); |
530e9b76 | 437 | if (ret) |
ba997462 | 438 | return ret; |
9cd4f1a4 | 439 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
440 | } |
441 | ||
2e1a3483 TG |
442 | /* |
443 | * Hotplug state machine related functions | |
444 | */ | |
2e1a3483 | 445 | |
a724632c | 446 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 TG |
447 | { |
448 | for (st->state--; st->state > st->target; st->state--) { | |
a724632c | 449 | struct cpuhp_step *step = cpuhp_get_step(st->state); |
2e1a3483 TG |
450 | |
451 | if (!step->skip_onerr) | |
96abb968 | 452 | cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); |
2e1a3483 TG |
453 | } |
454 | } | |
455 | ||
456 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 457 | enum cpuhp_state target) |
2e1a3483 TG |
458 | { |
459 | enum cpuhp_state prev_state = st->state; | |
460 | int ret = 0; | |
461 | ||
462 | while (st->state < target) { | |
2e1a3483 | 463 | st->state++; |
96abb968 | 464 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
2e1a3483 TG |
465 | if (ret) { |
466 | st->target = prev_state; | |
a724632c | 467 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
468 | break; |
469 | } | |
470 | } | |
471 | return ret; | |
472 | } | |
473 | ||
4cb28ced TG |
474 | /* |
475 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
476 | */ | |
477 | static void cpuhp_create(unsigned int cpu) | |
478 | { | |
479 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
480 | ||
5ebe7742 PZ |
481 | init_completion(&st->done_up); |
482 | init_completion(&st->done_down); | |
4cb28ced TG |
483 | } |
484 | ||
485 | static int cpuhp_should_run(unsigned int cpu) | |
486 | { | |
487 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
488 | ||
489 | return st->should_run; | |
490 | } | |
491 | ||
4cb28ced TG |
492 | /* |
493 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
494 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
495 | * |
496 | * Each invocation of this function by the smpboot thread does a single AP | |
497 | * state callback. | |
498 | * | |
499 | * It has 3 modes of operation: | |
500 | * - single: runs st->cb_state | |
501 | * - up: runs ++st->state, while st->state < st->target | |
502 | * - down: runs st->state--, while st->state > st->target | |
503 | * | |
504 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
505 | */ |
506 | static void cpuhp_thread_fun(unsigned int cpu) | |
507 | { | |
508 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
509 | bool bringup = st->bringup; |
510 | enum cpuhp_state state; | |
4cb28ced TG |
511 | |
512 | /* | |
4dddfb5f PZ |
513 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
514 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
515 | */ |
516 | smp_mb(); | |
4cb28ced | 517 | |
4dddfb5f PZ |
518 | if (WARN_ON_ONCE(!st->should_run)) |
519 | return; | |
4cb28ced | 520 | |
5f4b55e1 | 521 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 522 | |
a724632c | 523 | if (st->single) { |
4dddfb5f PZ |
524 | state = st->cb_state; |
525 | st->should_run = false; | |
526 | } else { | |
527 | if (bringup) { | |
528 | st->state++; | |
529 | state = st->state; | |
530 | st->should_run = (st->state < st->target); | |
531 | WARN_ON_ONCE(st->state > st->target); | |
4cb28ced | 532 | } else { |
4dddfb5f PZ |
533 | state = st->state; |
534 | st->state--; | |
535 | st->should_run = (st->state > st->target); | |
536 | WARN_ON_ONCE(st->state < st->target); | |
4cb28ced | 537 | } |
4dddfb5f PZ |
538 | } |
539 | ||
540 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
541 | ||
542 | if (st->rollback) { | |
543 | struct cpuhp_step *step = cpuhp_get_step(state); | |
544 | if (step->skip_onerr) | |
545 | goto next; | |
546 | } | |
547 | ||
548 | if (cpuhp_is_atomic_state(state)) { | |
549 | local_irq_disable(); | |
550 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
551 | local_irq_enable(); | |
3b9d6da6 | 552 | |
4dddfb5f PZ |
553 | /* |
554 | * STARTING/DYING must not fail! | |
555 | */ | |
556 | WARN_ON_ONCE(st->result); | |
4cb28ced | 557 | } else { |
4dddfb5f PZ |
558 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
559 | } | |
560 | ||
561 | if (st->result) { | |
562 | /* | |
563 | * If we fail on a rollback, we're up a creek without no | |
564 | * paddle, no way forward, no way back. We loose, thanks for | |
565 | * playing. | |
566 | */ | |
567 | WARN_ON_ONCE(st->rollback); | |
568 | st->should_run = false; | |
4cb28ced | 569 | } |
4dddfb5f PZ |
570 | |
571 | next: | |
5f4b55e1 | 572 | cpuhp_lock_release(bringup); |
4dddfb5f PZ |
573 | |
574 | if (!st->should_run) | |
5ebe7742 | 575 | complete_ap_thread(st, bringup); |
4cb28ced TG |
576 | } |
577 | ||
578 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 579 | static int |
cf392d10 TG |
580 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
581 | struct hlist_node *node) | |
4cb28ced TG |
582 | { |
583 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 584 | int ret; |
4cb28ced TG |
585 | |
586 | if (!cpu_online(cpu)) | |
587 | return 0; | |
588 | ||
5f4b55e1 PZ |
589 | cpuhp_lock_acquire(false); |
590 | cpuhp_lock_release(false); | |
591 | ||
592 | cpuhp_lock_acquire(true); | |
593 | cpuhp_lock_release(true); | |
49dfe2a6 | 594 | |
6a4e2451 TG |
595 | /* |
596 | * If we are up and running, use the hotplug thread. For early calls | |
597 | * we invoke the thread function directly. | |
598 | */ | |
599 | if (!st->thread) | |
96abb968 | 600 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 601 | |
4dddfb5f PZ |
602 | st->rollback = false; |
603 | st->last = NULL; | |
604 | ||
605 | st->node = node; | |
606 | st->bringup = bringup; | |
4cb28ced | 607 | st->cb_state = state; |
a724632c | 608 | st->single = true; |
a724632c | 609 | |
4dddfb5f | 610 | __cpuhp_kick_ap(st); |
4cb28ced | 611 | |
4cb28ced | 612 | /* |
4dddfb5f | 613 | * If we failed and did a partial, do a rollback. |
4cb28ced | 614 | */ |
4dddfb5f PZ |
615 | if ((ret = st->result) && st->last) { |
616 | st->rollback = true; | |
617 | st->bringup = !bringup; | |
618 | ||
619 | __cpuhp_kick_ap(st); | |
620 | } | |
621 | ||
622 | return ret; | |
1cf4f629 TG |
623 | } |
624 | ||
625 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
626 | { | |
627 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
628 | enum cpuhp_state prev_state = st->state; |
629 | int ret; | |
1cf4f629 | 630 | |
5f4b55e1 PZ |
631 | cpuhp_lock_acquire(false); |
632 | cpuhp_lock_release(false); | |
633 | ||
634 | cpuhp_lock_acquire(true); | |
635 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
636 | |
637 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
638 | ret = cpuhp_kick_ap(st, st->target); | |
639 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); | |
640 | ||
641 | return ret; | |
4cb28ced TG |
642 | } |
643 | ||
644 | static struct smp_hotplug_thread cpuhp_threads = { | |
645 | .store = &cpuhp_state.thread, | |
646 | .create = &cpuhp_create, | |
647 | .thread_should_run = cpuhp_should_run, | |
648 | .thread_fn = cpuhp_thread_fun, | |
649 | .thread_comm = "cpuhp/%u", | |
650 | .selfparking = true, | |
651 | }; | |
652 | ||
653 | void __init cpuhp_threads_init(void) | |
654 | { | |
655 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
656 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
657 | } | |
658 | ||
777c6e0d | 659 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
660 | /** |
661 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
662 | * @cpu: a CPU id | |
663 | * | |
664 | * This function walks all processes, finds a valid mm struct for each one and | |
665 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
666 | * trivial, there are various non-obvious corner cases, which this function | |
667 | * tries to solve in a safe manner. | |
668 | * | |
669 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
670 | * be called only for an already offlined CPU. | |
671 | */ | |
cb79295e AV |
672 | void clear_tasks_mm_cpumask(int cpu) |
673 | { | |
674 | struct task_struct *p; | |
675 | ||
676 | /* | |
677 | * This function is called after the cpu is taken down and marked | |
678 | * offline, so its not like new tasks will ever get this cpu set in | |
679 | * their mm mask. -- Peter Zijlstra | |
680 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
681 | * full-fledged tasklist_lock. | |
682 | */ | |
e4cc2f87 | 683 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
684 | rcu_read_lock(); |
685 | for_each_process(p) { | |
686 | struct task_struct *t; | |
687 | ||
e4cc2f87 AV |
688 | /* |
689 | * Main thread might exit, but other threads may still have | |
690 | * a valid mm. Find one. | |
691 | */ | |
cb79295e AV |
692 | t = find_lock_task_mm(p); |
693 | if (!t) | |
694 | continue; | |
695 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
696 | task_unlock(t); | |
697 | } | |
698 | rcu_read_unlock(); | |
699 | } | |
700 | ||
1da177e4 | 701 | /* Take this CPU down. */ |
71cf5aee | 702 | static int take_cpu_down(void *_param) |
1da177e4 | 703 | { |
4baa0afc TG |
704 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
705 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 706 | int err, cpu = smp_processor_id(); |
724a8688 | 707 | int ret; |
1da177e4 | 708 | |
1da177e4 LT |
709 | /* Ensure this CPU doesn't handle any more interrupts. */ |
710 | err = __cpu_disable(); | |
711 | if (err < 0) | |
f3705136 | 712 | return err; |
1da177e4 | 713 | |
a724632c TG |
714 | /* |
715 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
716 | * do this step again. | |
717 | */ | |
718 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
719 | st->state--; | |
4baa0afc | 720 | /* Invoke the former CPU_DYING callbacks */ |
724a8688 PZ |
721 | for (; st->state > target; st->state--) { |
722 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
723 | /* | |
724 | * DYING must not fail! | |
725 | */ | |
726 | WARN_ON_ONCE(ret); | |
727 | } | |
4baa0afc | 728 | |
52c063d1 TG |
729 | /* Give up timekeeping duties */ |
730 | tick_handover_do_timer(); | |
14e568e7 | 731 | /* Park the stopper thread */ |
090e77c3 | 732 | stop_machine_park(cpu); |
f3705136 | 733 | return 0; |
1da177e4 LT |
734 | } |
735 | ||
98458172 | 736 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 737 | { |
e69aab13 | 738 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 739 | int err; |
1da177e4 | 740 | |
2a58c527 | 741 | /* Park the smpboot threads */ |
1cf4f629 | 742 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
2a58c527 | 743 | smpboot_park_threads(cpu); |
1cf4f629 | 744 | |
6acce3ef | 745 | /* |
a8994181 TG |
746 | * Prevent irq alloc/free while the dying cpu reorganizes the |
747 | * interrupt affinities. | |
6acce3ef | 748 | */ |
a8994181 | 749 | irq_lock_sparse(); |
6acce3ef | 750 | |
a8994181 TG |
751 | /* |
752 | * So now all preempt/rcu users must observe !cpu_active(). | |
753 | */ | |
210e2133 | 754 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 755 | if (err) { |
3b9d6da6 | 756 | /* CPU refused to die */ |
a8994181 | 757 | irq_unlock_sparse(); |
3b9d6da6 SAS |
758 | /* Unpark the hotplug thread so we can rollback there */ |
759 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 760 | return err; |
8fa1d7d3 | 761 | } |
04321587 | 762 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 763 | |
48c5ccae | 764 | /* |
ee1e714b | 765 | * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all |
48c5ccae PZ |
766 | * runnable tasks from the cpu, there's only the idle task left now |
767 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
768 | * |
769 | * Wait for the stop thread to go away. | |
48c5ccae | 770 | */ |
5ebe7742 | 771 | wait_for_ap_thread(st, false); |
e69aab13 | 772 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); |
1da177e4 | 773 | |
a8994181 TG |
774 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
775 | irq_unlock_sparse(); | |
776 | ||
345527b1 | 777 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
778 | /* This actually kills the CPU. */ |
779 | __cpu_die(cpu); | |
780 | ||
a49b116d | 781 | tick_cleanup_dead_cpu(cpu); |
a58163d8 | 782 | rcutree_migrate_callbacks(cpu); |
98458172 TG |
783 | return 0; |
784 | } | |
1da177e4 | 785 | |
71f87b2f TG |
786 | static void cpuhp_complete_idle_dead(void *arg) |
787 | { | |
788 | struct cpuhp_cpu_state *st = arg; | |
789 | ||
5ebe7742 | 790 | complete_ap_thread(st, false); |
71f87b2f TG |
791 | } |
792 | ||
e69aab13 TG |
793 | void cpuhp_report_idle_dead(void) |
794 | { | |
795 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
796 | ||
797 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 798 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
799 | st->state = CPUHP_AP_IDLE_DEAD; |
800 | /* | |
801 | * We cannot call complete after rcu_report_dead() so we delegate it | |
802 | * to an online cpu. | |
803 | */ | |
804 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
805 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
806 | } |
807 | ||
4dddfb5f PZ |
808 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
809 | { | |
810 | for (st->state++; st->state < st->target; st->state++) { | |
811 | struct cpuhp_step *step = cpuhp_get_step(st->state); | |
cff7d378 | 812 | |
4dddfb5f PZ |
813 | if (!step->skip_onerr) |
814 | cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); | |
815 | } | |
816 | } | |
817 | ||
818 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
819 | enum cpuhp_state target) | |
820 | { | |
821 | enum cpuhp_state prev_state = st->state; | |
822 | int ret = 0; | |
823 | ||
824 | for (; st->state > target; st->state--) { | |
825 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
826 | if (ret) { | |
827 | st->target = prev_state; | |
828 | undo_cpu_down(cpu, st); | |
829 | break; | |
830 | } | |
831 | } | |
832 | return ret; | |
833 | } | |
cff7d378 | 834 | |
98458172 | 835 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
836 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
837 | enum cpuhp_state target) | |
98458172 | 838 | { |
cff7d378 TG |
839 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
840 | int prev_state, ret = 0; | |
98458172 TG |
841 | |
842 | if (num_online_cpus() == 1) | |
843 | return -EBUSY; | |
844 | ||
757c989b | 845 | if (!cpu_present(cpu)) |
98458172 TG |
846 | return -EINVAL; |
847 | ||
8f553c49 | 848 | cpus_write_lock(); |
98458172 TG |
849 | |
850 | cpuhp_tasks_frozen = tasks_frozen; | |
851 | ||
4dddfb5f | 852 | prev_state = cpuhp_set_state(st, target); |
1cf4f629 TG |
853 | /* |
854 | * If the current CPU state is in the range of the AP hotplug thread, | |
855 | * then we need to kick the thread. | |
856 | */ | |
8df3e07e | 857 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 858 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
859 | ret = cpuhp_kick_ap_work(cpu); |
860 | /* | |
861 | * The AP side has done the error rollback already. Just | |
862 | * return the error code.. | |
863 | */ | |
864 | if (ret) | |
865 | goto out; | |
866 | ||
867 | /* | |
868 | * We might have stopped still in the range of the AP hotplug | |
869 | * thread. Nothing to do anymore. | |
870 | */ | |
8df3e07e | 871 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 872 | goto out; |
4dddfb5f PZ |
873 | |
874 | st->target = target; | |
1cf4f629 TG |
875 | } |
876 | /* | |
8df3e07e | 877 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
878 | * to do the further cleanups. |
879 | */ | |
a724632c | 880 | ret = cpuhp_down_callbacks(cpu, st, target); |
3b9d6da6 | 881 | if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
4dddfb5f PZ |
882 | cpuhp_reset_state(st, prev_state); |
883 | __cpuhp_kick_ap(st); | |
3b9d6da6 | 884 | } |
98458172 | 885 | |
1cf4f629 | 886 | out: |
8f553c49 | 887 | cpus_write_unlock(); |
cff7d378 | 888 | return ret; |
e3920fb4 RW |
889 | } |
890 | ||
af1f4045 | 891 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 892 | { |
9ea09af3 | 893 | int err; |
e3920fb4 | 894 | |
d221938c | 895 | cpu_maps_update_begin(); |
e761b772 MK |
896 | |
897 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 898 | err = -EBUSY; |
e761b772 MK |
899 | goto out; |
900 | } | |
901 | ||
af1f4045 | 902 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 903 | |
e761b772 | 904 | out: |
d221938c | 905 | cpu_maps_update_done(); |
1da177e4 LT |
906 | return err; |
907 | } | |
4dddfb5f | 908 | |
af1f4045 TG |
909 | int cpu_down(unsigned int cpu) |
910 | { | |
911 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
912 | } | |
b62b8ef9 | 913 | EXPORT_SYMBOL(cpu_down); |
4dddfb5f PZ |
914 | |
915 | #else | |
916 | #define takedown_cpu NULL | |
1da177e4 LT |
917 | #endif /*CONFIG_HOTPLUG_CPU*/ |
918 | ||
4baa0afc | 919 | /** |
ee1e714b | 920 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
921 | * @cpu: cpu that just started |
922 | * | |
4baa0afc TG |
923 | * It must be called by the arch code on the new cpu, before the new cpu |
924 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
925 | */ | |
926 | void notify_cpu_starting(unsigned int cpu) | |
927 | { | |
928 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
929 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
724a8688 | 930 | int ret; |
4baa0afc | 931 | |
0c6d4576 | 932 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
4baa0afc | 933 | while (st->state < target) { |
4baa0afc | 934 | st->state++; |
724a8688 PZ |
935 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
936 | /* | |
937 | * STARTING must not fail! | |
938 | */ | |
939 | WARN_ON_ONCE(ret); | |
4baa0afc TG |
940 | } |
941 | } | |
942 | ||
949338e3 | 943 | /* |
9cd4f1a4 TG |
944 | * Called from the idle task. Wake up the controlling task which brings the |
945 | * stopper and the hotplug thread of the upcoming CPU up and then delegates | |
946 | * the rest of the online bringup to the hotplug thread. | |
949338e3 | 947 | */ |
8df3e07e | 948 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 949 | { |
8df3e07e | 950 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
951 | |
952 | /* Happens for the boot cpu */ | |
953 | if (state != CPUHP_AP_ONLINE_IDLE) | |
954 | return; | |
955 | ||
956 | st->state = CPUHP_AP_ONLINE_IDLE; | |
5ebe7742 | 957 | complete_ap_thread(st, true); |
949338e3 TG |
958 | } |
959 | ||
e3920fb4 | 960 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 961 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 962 | { |
cff7d378 | 963 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 964 | struct task_struct *idle; |
2e1a3483 | 965 | int ret = 0; |
1da177e4 | 966 | |
8f553c49 | 967 | cpus_write_lock(); |
38498a67 | 968 | |
757c989b | 969 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
970 | ret = -EINVAL; |
971 | goto out; | |
972 | } | |
973 | ||
757c989b TG |
974 | /* |
975 | * The caller of do_cpu_up might have raced with another | |
976 | * caller. Ignore it for now. | |
977 | */ | |
978 | if (st->state >= target) | |
38498a67 | 979 | goto out; |
757c989b TG |
980 | |
981 | if (st->state == CPUHP_OFFLINE) { | |
982 | /* Let it fail before we try to bring the cpu up */ | |
983 | idle = idle_thread_get(cpu); | |
984 | if (IS_ERR(idle)) { | |
985 | ret = PTR_ERR(idle); | |
986 | goto out; | |
987 | } | |
3bb5d2ee | 988 | } |
38498a67 | 989 | |
ba997462 TG |
990 | cpuhp_tasks_frozen = tasks_frozen; |
991 | ||
4dddfb5f | 992 | cpuhp_set_state(st, target); |
1cf4f629 TG |
993 | /* |
994 | * If the current CPU state is in the range of the AP hotplug thread, | |
995 | * then we need to kick the thread once more. | |
996 | */ | |
8df3e07e | 997 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
998 | ret = cpuhp_kick_ap_work(cpu); |
999 | /* | |
1000 | * The AP side has done the error rollback already. Just | |
1001 | * return the error code.. | |
1002 | */ | |
1003 | if (ret) | |
1004 | goto out; | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1009 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1010 | * responsible for bringing it up to the target state. |
1011 | */ | |
8df3e07e | 1012 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1013 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1014 | out: |
8f553c49 | 1015 | cpus_write_unlock(); |
e3920fb4 RW |
1016 | return ret; |
1017 | } | |
1018 | ||
af1f4045 | 1019 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1020 | { |
1021 | int err = 0; | |
cf23422b | 1022 | |
e0b582ec | 1023 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1024 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1025 | cpu); | |
87d5e023 | 1026 | #if defined(CONFIG_IA64) |
84117da5 | 1027 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1028 | #endif |
1029 | return -EINVAL; | |
1030 | } | |
e3920fb4 | 1031 | |
01b0f197 TK |
1032 | err = try_online_node(cpu_to_node(cpu)); |
1033 | if (err) | |
1034 | return err; | |
cf23422b | 1035 | |
d221938c | 1036 | cpu_maps_update_begin(); |
e761b772 MK |
1037 | |
1038 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1039 | err = -EBUSY; |
e761b772 MK |
1040 | goto out; |
1041 | } | |
1042 | ||
af1f4045 | 1043 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1044 | out: |
d221938c | 1045 | cpu_maps_update_done(); |
e3920fb4 RW |
1046 | return err; |
1047 | } | |
af1f4045 TG |
1048 | |
1049 | int cpu_up(unsigned int cpu) | |
1050 | { | |
1051 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1052 | } | |
a513f6ba | 1053 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1054 | |
f3de4be9 | 1055 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1056 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1057 | |
d391e552 | 1058 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1059 | { |
d391e552 | 1060 | int cpu, error = 0; |
e3920fb4 | 1061 | |
d221938c | 1062 | cpu_maps_update_begin(); |
d391e552 JM |
1063 | if (!cpu_online(primary)) |
1064 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
1065 | /* |
1066 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1067 | * with the userspace trying to use the CPU hotplug at the same time |
1068 | */ | |
e0b582ec | 1069 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1070 | |
84117da5 | 1071 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1072 | for_each_online_cpu(cpu) { |
d391e552 | 1073 | if (cpu == primary) |
e3920fb4 | 1074 | continue; |
bb3632c6 | 1075 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1076 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1077 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1078 | if (!error) |
e0b582ec | 1079 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1080 | else { |
84117da5 | 1081 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1082 | break; |
1083 | } | |
1084 | } | |
86886e55 | 1085 | |
89af7ba5 | 1086 | if (!error) |
e3920fb4 | 1087 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1088 | else |
84117da5 | 1089 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1090 | |
1091 | /* | |
1092 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1093 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1094 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1095 | */ | |
1096 | cpu_hotplug_disabled++; | |
1097 | ||
d221938c | 1098 | cpu_maps_update_done(); |
e3920fb4 RW |
1099 | return error; |
1100 | } | |
1101 | ||
d0af9eed SS |
1102 | void __weak arch_enable_nonboot_cpus_begin(void) |
1103 | { | |
1104 | } | |
1105 | ||
1106 | void __weak arch_enable_nonboot_cpus_end(void) | |
1107 | { | |
1108 | } | |
1109 | ||
71cf5aee | 1110 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1111 | { |
1112 | int cpu, error; | |
1113 | ||
1114 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1115 | cpu_maps_update_begin(); |
01b41159 | 1116 | __cpu_hotplug_enable(); |
e0b582ec | 1117 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1118 | goto out; |
e3920fb4 | 1119 | |
84117da5 | 1120 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1121 | |
1122 | arch_enable_nonboot_cpus_begin(); | |
1123 | ||
e0b582ec | 1124 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1125 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1126 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1127 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1128 | if (!error) { |
84117da5 | 1129 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1130 | continue; |
1131 | } | |
84117da5 | 1132 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1133 | } |
d0af9eed SS |
1134 | |
1135 | arch_enable_nonboot_cpus_end(); | |
1136 | ||
e0b582ec | 1137 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1138 | out: |
d221938c | 1139 | cpu_maps_update_done(); |
1da177e4 | 1140 | } |
e0b582ec | 1141 | |
d7268a31 | 1142 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1143 | { |
1144 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1145 | return -ENOMEM; | |
1146 | return 0; | |
1147 | } | |
1148 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1149 | |
79cfbdfa SB |
1150 | /* |
1151 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1152 | * ensure that the state of the system with respect to the tasks being frozen | |
1153 | * or not, as reported by the notification, remains unchanged *throughout the | |
1154 | * duration* of the execution of the callbacks. | |
1155 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1156 | * | |
1157 | * This synchronization is implemented by mutually excluding regular CPU | |
1158 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1159 | * Hibernate notifications. | |
1160 | */ | |
1161 | static int | |
1162 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1163 | unsigned long action, void *ptr) | |
1164 | { | |
1165 | switch (action) { | |
1166 | ||
1167 | case PM_SUSPEND_PREPARE: | |
1168 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1169 | cpu_hotplug_disable(); |
79cfbdfa SB |
1170 | break; |
1171 | ||
1172 | case PM_POST_SUSPEND: | |
1173 | case PM_POST_HIBERNATION: | |
16e53dbf | 1174 | cpu_hotplug_enable(); |
79cfbdfa SB |
1175 | break; |
1176 | ||
1177 | default: | |
1178 | return NOTIFY_DONE; | |
1179 | } | |
1180 | ||
1181 | return NOTIFY_OK; | |
1182 | } | |
1183 | ||
1184 | ||
d7268a31 | 1185 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1186 | { |
6e32d479 FY |
1187 | /* |
1188 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1189 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1190 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1191 | */ | |
79cfbdfa SB |
1192 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1193 | return 0; | |
1194 | } | |
1195 | core_initcall(cpu_hotplug_pm_sync_init); | |
1196 | ||
f3de4be9 | 1197 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1198 | |
8ce371f9 PZ |
1199 | int __boot_cpu_id; |
1200 | ||
68f4f1ec | 1201 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1202 | |
cff7d378 TG |
1203 | /* Boot processor state steps */ |
1204 | static struct cpuhp_step cpuhp_bp_states[] = { | |
1205 | [CPUHP_OFFLINE] = { | |
1206 | .name = "offline", | |
3c1627e9 TG |
1207 | .startup.single = NULL, |
1208 | .teardown.single = NULL, | |
cff7d378 TG |
1209 | }, |
1210 | #ifdef CONFIG_SMP | |
1211 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1212 | .name = "threads:prepare", |
3c1627e9 TG |
1213 | .startup.single = smpboot_create_threads, |
1214 | .teardown.single = NULL, | |
757c989b | 1215 | .cant_stop = true, |
cff7d378 | 1216 | }, |
00e16c3d | 1217 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1218 | .name = "perf:prepare", |
1219 | .startup.single = perf_event_init_cpu, | |
1220 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1221 | }, |
7ee681b2 | 1222 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1223 | .name = "workqueue:prepare", |
1224 | .startup.single = workqueue_prepare_cpu, | |
1225 | .teardown.single = NULL, | |
7ee681b2 | 1226 | }, |
27590dc1 | 1227 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1228 | .name = "hrtimers:prepare", |
1229 | .startup.single = hrtimers_prepare_cpu, | |
1230 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1231 | }, |
31487f83 | 1232 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1233 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1234 | .startup.single = smpcfd_prepare_cpu, |
1235 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1236 | }, |
e6d4989a RW |
1237 | [CPUHP_RELAY_PREPARE] = { |
1238 | .name = "relay:prepare", | |
1239 | .startup.single = relay_prepare_cpu, | |
1240 | .teardown.single = NULL, | |
1241 | }, | |
6731d4f1 SAS |
1242 | [CPUHP_SLAB_PREPARE] = { |
1243 | .name = "slab:prepare", | |
1244 | .startup.single = slab_prepare_cpu, | |
1245 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1246 | }, |
4df83742 | 1247 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1248 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1249 | .startup.single = rcutree_prepare_cpu, |
1250 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1251 | }, |
4fae16df RC |
1252 | /* |
1253 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1254 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1255 | * otherwise a RCU stall occurs. | |
1256 | */ | |
1257 | [CPUHP_TIMERS_DEAD] = { | |
3c1627e9 TG |
1258 | .name = "timers:dead", |
1259 | .startup.single = NULL, | |
1260 | .teardown.single = timers_dead_cpu, | |
4fae16df | 1261 | }, |
d10ef6f9 | 1262 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1263 | [CPUHP_BRINGUP_CPU] = { |
1264 | .name = "cpu:bringup", | |
3c1627e9 TG |
1265 | .startup.single = bringup_cpu, |
1266 | .teardown.single = NULL, | |
757c989b | 1267 | .cant_stop = true, |
4baa0afc | 1268 | }, |
31487f83 | 1269 | [CPUHP_AP_SMPCFD_DYING] = { |
677f6646 | 1270 | .name = "smpcfd:dying", |
3c1627e9 TG |
1271 | .startup.single = NULL, |
1272 | .teardown.single = smpcfd_dying_cpu, | |
31487f83 | 1273 | }, |
d10ef6f9 TG |
1274 | /* |
1275 | * Handled on controll processor until the plugged processor manages | |
1276 | * this itself. | |
1277 | */ | |
4baa0afc TG |
1278 | [CPUHP_TEARDOWN_CPU] = { |
1279 | .name = "cpu:teardown", | |
3c1627e9 TG |
1280 | .startup.single = NULL, |
1281 | .teardown.single = takedown_cpu, | |
757c989b | 1282 | .cant_stop = true, |
cff7d378 | 1283 | }, |
a7c73414 TG |
1284 | #else |
1285 | [CPUHP_BRINGUP_CPU] = { }, | |
cff7d378 | 1286 | #endif |
cff7d378 TG |
1287 | }; |
1288 | ||
4baa0afc TG |
1289 | /* Application processor state steps */ |
1290 | static struct cpuhp_step cpuhp_ap_states[] = { | |
1291 | #ifdef CONFIG_SMP | |
d10ef6f9 TG |
1292 | /* Final state before CPU kills itself */ |
1293 | [CPUHP_AP_IDLE_DEAD] = { | |
1294 | .name = "idle:dead", | |
1295 | }, | |
1296 | /* | |
1297 | * Last state before CPU enters the idle loop to die. Transient state | |
1298 | * for synchronization. | |
1299 | */ | |
1300 | [CPUHP_AP_OFFLINE] = { | |
1301 | .name = "ap:offline", | |
1302 | .cant_stop = true, | |
1303 | }, | |
9cf7243d TG |
1304 | /* First state is scheduler control. Interrupts are disabled */ |
1305 | [CPUHP_AP_SCHED_STARTING] = { | |
1306 | .name = "sched:starting", | |
3c1627e9 TG |
1307 | .startup.single = sched_cpu_starting, |
1308 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1309 | }, |
4df83742 | 1310 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1311 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1312 | .startup.single = NULL, |
1313 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1314 | }, |
d10ef6f9 TG |
1315 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1316 | * state for synchronsization */ | |
1317 | [CPUHP_AP_ONLINE] = { | |
1318 | .name = "ap:online", | |
1319 | }, | |
1320 | /* Handle smpboot threads park/unpark */ | |
1cf4f629 | 1321 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1322 | .name = "smpboot/threads:online", |
3c1627e9 TG |
1323 | .startup.single = smpboot_unpark_threads, |
1324 | .teardown.single = NULL, | |
1cf4f629 | 1325 | }, |
c5cb83bb TG |
1326 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1327 | .name = "irq/affinity:online", | |
1328 | .startup.single = irq_affinity_online_cpu, | |
1329 | .teardown.single = NULL, | |
1330 | }, | |
00e16c3d | 1331 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1332 | .name = "perf:online", |
1333 | .startup.single = perf_event_init_cpu, | |
1334 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1335 | }, |
7ee681b2 | 1336 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1337 | .name = "workqueue:online", |
1338 | .startup.single = workqueue_online_cpu, | |
1339 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1340 | }, |
4df83742 | 1341 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1342 | .name = "RCU/tree:online", |
3c1627e9 TG |
1343 | .startup.single = rcutree_online_cpu, |
1344 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1345 | }, |
4baa0afc | 1346 | #endif |
d10ef6f9 TG |
1347 | /* |
1348 | * The dynamically registered state space is here | |
1349 | */ | |
1350 | ||
aaddd7d1 TG |
1351 | #ifdef CONFIG_SMP |
1352 | /* Last state is scheduler control setting the cpu active */ | |
1353 | [CPUHP_AP_ACTIVE] = { | |
1354 | .name = "sched:active", | |
3c1627e9 TG |
1355 | .startup.single = sched_cpu_activate, |
1356 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1357 | }, |
1358 | #endif | |
1359 | ||
d10ef6f9 | 1360 | /* CPU is fully up and running. */ |
4baa0afc TG |
1361 | [CPUHP_ONLINE] = { |
1362 | .name = "online", | |
3c1627e9 TG |
1363 | .startup.single = NULL, |
1364 | .teardown.single = NULL, | |
4baa0afc TG |
1365 | }, |
1366 | }; | |
1367 | ||
5b7aa87e TG |
1368 | /* Sanity check for callbacks */ |
1369 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1370 | { | |
1371 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1372 | return -EINVAL; | |
1373 | return 0; | |
1374 | } | |
1375 | ||
dc280d93 TG |
1376 | /* |
1377 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1378 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1379 | * by having no name assigned. | |
1380 | */ | |
1381 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1382 | { | |
4205e478 TG |
1383 | enum cpuhp_state i, end; |
1384 | struct cpuhp_step *step; | |
dc280d93 | 1385 | |
4205e478 TG |
1386 | switch (state) { |
1387 | case CPUHP_AP_ONLINE_DYN: | |
1388 | step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN; | |
1389 | end = CPUHP_AP_ONLINE_DYN_END; | |
1390 | break; | |
1391 | case CPUHP_BP_PREPARE_DYN: | |
1392 | step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN; | |
1393 | end = CPUHP_BP_PREPARE_DYN_END; | |
1394 | break; | |
1395 | default: | |
1396 | return -EINVAL; | |
1397 | } | |
1398 | ||
1399 | for (i = state; i <= end; i++, step++) { | |
1400 | if (!step->name) | |
dc280d93 TG |
1401 | return i; |
1402 | } | |
1403 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1404 | return -ENOSPC; | |
1405 | } | |
1406 | ||
1407 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1408 | int (*startup)(unsigned int cpu), | |
1409 | int (*teardown)(unsigned int cpu), | |
1410 | bool multi_instance) | |
5b7aa87e TG |
1411 | { |
1412 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1413 | struct cpuhp_step *sp; | |
dc280d93 | 1414 | int ret = 0; |
5b7aa87e | 1415 | |
0c96b273 EB |
1416 | /* |
1417 | * If name is NULL, then the state gets removed. | |
1418 | * | |
1419 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
1420 | * the first allocation from these dynamic ranges, so the removal | |
1421 | * would trigger a new allocation and clear the wrong (already | |
1422 | * empty) state, leaving the callbacks of the to be cleared state | |
1423 | * dangling, which causes wreckage on the next hotplug operation. | |
1424 | */ | |
1425 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
1426 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
1427 | ret = cpuhp_reserve_state(state); |
1428 | if (ret < 0) | |
dc434e05 | 1429 | return ret; |
dc280d93 TG |
1430 | state = ret; |
1431 | } | |
5b7aa87e | 1432 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1433 | if (name && sp->name) |
1434 | return -EBUSY; | |
1435 | ||
3c1627e9 TG |
1436 | sp->startup.single = startup; |
1437 | sp->teardown.single = teardown; | |
5b7aa87e | 1438 | sp->name = name; |
cf392d10 TG |
1439 | sp->multi_instance = multi_instance; |
1440 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1441 | return ret; |
5b7aa87e TG |
1442 | } |
1443 | ||
1444 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1445 | { | |
3c1627e9 | 1446 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1447 | } |
1448 | ||
5b7aa87e TG |
1449 | /* |
1450 | * Call the startup/teardown function for a step either on the AP or | |
1451 | * on the current CPU. | |
1452 | */ | |
cf392d10 TG |
1453 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1454 | struct hlist_node *node) | |
5b7aa87e | 1455 | { |
a724632c | 1456 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1457 | int ret; |
1458 | ||
4dddfb5f PZ |
1459 | /* |
1460 | * If there's nothing to do, we done. | |
1461 | * Relies on the union for multi_instance. | |
1462 | */ | |
3c1627e9 TG |
1463 | if ((bringup && !sp->startup.single) || |
1464 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1465 | return 0; |
5b7aa87e TG |
1466 | /* |
1467 | * The non AP bound callbacks can fail on bringup. On teardown | |
1468 | * e.g. module removal we crash for now. | |
1469 | */ | |
1cf4f629 TG |
1470 | #ifdef CONFIG_SMP |
1471 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1472 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1473 | else |
96abb968 | 1474 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1475 | #else |
96abb968 | 1476 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1477 | #endif |
5b7aa87e TG |
1478 | BUG_ON(ret && !bringup); |
1479 | return ret; | |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1484 | * | |
1485 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1486 | */ | |
1487 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1488 | struct hlist_node *node) |
5b7aa87e TG |
1489 | { |
1490 | int cpu; | |
1491 | ||
5b7aa87e TG |
1492 | /* Roll back the already executed steps on the other cpus */ |
1493 | for_each_present_cpu(cpu) { | |
1494 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1495 | int cpustate = st->state; | |
1496 | ||
1497 | if (cpu >= failedcpu) | |
1498 | break; | |
1499 | ||
1500 | /* Did we invoke the startup call on that cpu ? */ | |
1501 | if (cpustate >= state) | |
cf392d10 | 1502 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1503 | } |
1504 | } | |
1505 | ||
9805c673 TG |
1506 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1507 | struct hlist_node *node, | |
1508 | bool invoke) | |
cf392d10 TG |
1509 | { |
1510 | struct cpuhp_step *sp; | |
1511 | int cpu; | |
1512 | int ret; | |
1513 | ||
9805c673 TG |
1514 | lockdep_assert_cpus_held(); |
1515 | ||
cf392d10 TG |
1516 | sp = cpuhp_get_step(state); |
1517 | if (sp->multi_instance == false) | |
1518 | return -EINVAL; | |
1519 | ||
dc434e05 | 1520 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1521 | |
3c1627e9 | 1522 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1523 | goto add_node; |
1524 | ||
1525 | /* | |
1526 | * Try to call the startup callback for each present cpu | |
1527 | * depending on the hotplug state of the cpu. | |
1528 | */ | |
1529 | for_each_present_cpu(cpu) { | |
1530 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1531 | int cpustate = st->state; | |
1532 | ||
1533 | if (cpustate < state) | |
1534 | continue; | |
1535 | ||
1536 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1537 | if (ret) { | |
3c1627e9 | 1538 | if (sp->teardown.multi) |
cf392d10 | 1539 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1540 | goto unlock; |
cf392d10 TG |
1541 | } |
1542 | } | |
1543 | add_node: | |
1544 | ret = 0; | |
cf392d10 | 1545 | hlist_add_head(node, &sp->list); |
dc434e05 | 1546 | unlock: |
cf392d10 | 1547 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1548 | return ret; |
1549 | } | |
1550 | ||
1551 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1552 | bool invoke) | |
1553 | { | |
1554 | int ret; | |
1555 | ||
1556 | cpus_read_lock(); | |
1557 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1558 | cpus_read_unlock(); |
cf392d10 TG |
1559 | return ret; |
1560 | } | |
1561 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1562 | ||
5b7aa87e | 1563 | /** |
71def423 | 1564 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1565 | * @state: The state to setup |
1566 | * @invoke: If true, the startup function is invoked for cpus where | |
1567 | * cpu state >= @state | |
1568 | * @startup: startup callback function | |
1569 | * @teardown: teardown callback function | |
1570 | * @multi_instance: State is set up for multiple instances which get | |
1571 | * added afterwards. | |
5b7aa87e | 1572 | * |
71def423 | 1573 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1574 | * Returns: |
1575 | * On success: | |
1576 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1577 | * 0 for all other states | |
1578 | * On failure: proper (negative) error code | |
5b7aa87e | 1579 | */ |
71def423 SAS |
1580 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1581 | const char *name, bool invoke, | |
1582 | int (*startup)(unsigned int cpu), | |
1583 | int (*teardown)(unsigned int cpu), | |
1584 | bool multi_instance) | |
5b7aa87e TG |
1585 | { |
1586 | int cpu, ret = 0; | |
b9d9d691 | 1587 | bool dynstate; |
5b7aa87e | 1588 | |
71def423 SAS |
1589 | lockdep_assert_cpus_held(); |
1590 | ||
5b7aa87e TG |
1591 | if (cpuhp_cb_check(state) || !name) |
1592 | return -EINVAL; | |
1593 | ||
dc434e05 | 1594 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1595 | |
dc280d93 TG |
1596 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1597 | multi_instance); | |
5b7aa87e | 1598 | |
b9d9d691 TG |
1599 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1600 | if (ret > 0 && dynstate) { | |
1601 | state = ret; | |
1602 | ret = 0; | |
1603 | } | |
1604 | ||
dc280d93 | 1605 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1606 | goto out; |
1607 | ||
1608 | /* | |
1609 | * Try to call the startup callback for each present cpu | |
1610 | * depending on the hotplug state of the cpu. | |
1611 | */ | |
1612 | for_each_present_cpu(cpu) { | |
1613 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1614 | int cpustate = st->state; | |
1615 | ||
1616 | if (cpustate < state) | |
1617 | continue; | |
1618 | ||
cf392d10 | 1619 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1620 | if (ret) { |
a724632c | 1621 | if (teardown) |
cf392d10 TG |
1622 | cpuhp_rollback_install(cpu, state, NULL); |
1623 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1624 | goto out; |
1625 | } | |
1626 | } | |
1627 | out: | |
dc434e05 | 1628 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1629 | /* |
1630 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1631 | * dynamically allocated state in case of success. | |
1632 | */ | |
b9d9d691 | 1633 | if (!ret && dynstate) |
5b7aa87e TG |
1634 | return state; |
1635 | return ret; | |
1636 | } | |
71def423 SAS |
1637 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1638 | ||
1639 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1640 | const char *name, bool invoke, | |
1641 | int (*startup)(unsigned int cpu), | |
1642 | int (*teardown)(unsigned int cpu), | |
1643 | bool multi_instance) | |
1644 | { | |
1645 | int ret; | |
1646 | ||
1647 | cpus_read_lock(); | |
1648 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1649 | teardown, multi_instance); | |
1650 | cpus_read_unlock(); | |
1651 | return ret; | |
1652 | } | |
5b7aa87e TG |
1653 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1654 | ||
cf392d10 TG |
1655 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1656 | struct hlist_node *node, bool invoke) | |
1657 | { | |
1658 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1659 | int cpu; | |
1660 | ||
1661 | BUG_ON(cpuhp_cb_check(state)); | |
1662 | ||
1663 | if (!sp->multi_instance) | |
1664 | return -EINVAL; | |
1665 | ||
8f553c49 | 1666 | cpus_read_lock(); |
dc434e05 SAS |
1667 | mutex_lock(&cpuhp_state_mutex); |
1668 | ||
cf392d10 TG |
1669 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1670 | goto remove; | |
1671 | /* | |
1672 | * Call the teardown callback for each present cpu depending | |
1673 | * on the hotplug state of the cpu. This function is not | |
1674 | * allowed to fail currently! | |
1675 | */ | |
1676 | for_each_present_cpu(cpu) { | |
1677 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1678 | int cpustate = st->state; | |
1679 | ||
1680 | if (cpustate >= state) | |
1681 | cpuhp_issue_call(cpu, state, false, node); | |
1682 | } | |
1683 | ||
1684 | remove: | |
cf392d10 TG |
1685 | hlist_del(node); |
1686 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1687 | cpus_read_unlock(); |
cf392d10 TG |
1688 | |
1689 | return 0; | |
1690 | } | |
1691 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1692 | |
5b7aa87e | 1693 | /** |
71def423 | 1694 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1695 | * @state: The state to remove |
1696 | * @invoke: If true, the teardown function is invoked for cpus where | |
1697 | * cpu state >= @state | |
1698 | * | |
71def423 | 1699 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1700 | * The teardown callback is currently not allowed to fail. Think |
1701 | * about module removal! | |
1702 | */ | |
71def423 | 1703 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1704 | { |
cf392d10 | 1705 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1706 | int cpu; |
1707 | ||
1708 | BUG_ON(cpuhp_cb_check(state)); | |
1709 | ||
71def423 | 1710 | lockdep_assert_cpus_held(); |
5b7aa87e | 1711 | |
dc434e05 | 1712 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1713 | if (sp->multi_instance) { |
1714 | WARN(!hlist_empty(&sp->list), | |
1715 | "Error: Removing state %d which has instances left.\n", | |
1716 | state); | |
1717 | goto remove; | |
1718 | } | |
1719 | ||
a724632c | 1720 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1721 | goto remove; |
1722 | ||
1723 | /* | |
1724 | * Call the teardown callback for each present cpu depending | |
1725 | * on the hotplug state of the cpu. This function is not | |
1726 | * allowed to fail currently! | |
1727 | */ | |
1728 | for_each_present_cpu(cpu) { | |
1729 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1730 | int cpustate = st->state; | |
1731 | ||
1732 | if (cpustate >= state) | |
cf392d10 | 1733 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1734 | } |
1735 | remove: | |
cf392d10 | 1736 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1737 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1738 | } |
1739 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1740 | ||
1741 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1742 | { | |
1743 | cpus_read_lock(); | |
1744 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1745 | cpus_read_unlock(); |
5b7aa87e TG |
1746 | } |
1747 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1748 | ||
98f8cdce TG |
1749 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1750 | static ssize_t show_cpuhp_state(struct device *dev, | |
1751 | struct device_attribute *attr, char *buf) | |
1752 | { | |
1753 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1754 | ||
1755 | return sprintf(buf, "%d\n", st->state); | |
1756 | } | |
1757 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1758 | ||
757c989b TG |
1759 | static ssize_t write_cpuhp_target(struct device *dev, |
1760 | struct device_attribute *attr, | |
1761 | const char *buf, size_t count) | |
1762 | { | |
1763 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1764 | struct cpuhp_step *sp; | |
1765 | int target, ret; | |
1766 | ||
1767 | ret = kstrtoint(buf, 10, &target); | |
1768 | if (ret) | |
1769 | return ret; | |
1770 | ||
1771 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1772 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1773 | return -EINVAL; | |
1774 | #else | |
1775 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1776 | return -EINVAL; | |
1777 | #endif | |
1778 | ||
1779 | ret = lock_device_hotplug_sysfs(); | |
1780 | if (ret) | |
1781 | return ret; | |
1782 | ||
1783 | mutex_lock(&cpuhp_state_mutex); | |
1784 | sp = cpuhp_get_step(target); | |
1785 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1786 | mutex_unlock(&cpuhp_state_mutex); | |
1787 | if (ret) | |
40da1b11 | 1788 | goto out; |
757c989b TG |
1789 | |
1790 | if (st->state < target) | |
1791 | ret = do_cpu_up(dev->id, target); | |
1792 | else | |
1793 | ret = do_cpu_down(dev->id, target); | |
40da1b11 | 1794 | out: |
757c989b TG |
1795 | unlock_device_hotplug(); |
1796 | return ret ? ret : count; | |
1797 | } | |
1798 | ||
98f8cdce TG |
1799 | static ssize_t show_cpuhp_target(struct device *dev, |
1800 | struct device_attribute *attr, char *buf) | |
1801 | { | |
1802 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1803 | ||
1804 | return sprintf(buf, "%d\n", st->target); | |
1805 | } | |
757c989b | 1806 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1807 | |
1808 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1809 | &dev_attr_state.attr, | |
1810 | &dev_attr_target.attr, | |
1811 | NULL | |
1812 | }; | |
1813 | ||
993647a2 | 1814 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
1815 | .attrs = cpuhp_cpu_attrs, |
1816 | .name = "hotplug", | |
1817 | NULL | |
1818 | }; | |
1819 | ||
1820 | static ssize_t show_cpuhp_states(struct device *dev, | |
1821 | struct device_attribute *attr, char *buf) | |
1822 | { | |
1823 | ssize_t cur, res = 0; | |
1824 | int i; | |
1825 | ||
1826 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1827 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1828 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1829 | ||
1830 | if (sp->name) { | |
1831 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1832 | buf += cur; | |
1833 | res += cur; | |
1834 | } | |
1835 | } | |
1836 | mutex_unlock(&cpuhp_state_mutex); | |
1837 | return res; | |
1838 | } | |
1839 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1840 | ||
1841 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1842 | &dev_attr_states.attr, | |
1843 | NULL | |
1844 | }; | |
1845 | ||
993647a2 | 1846 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
1847 | .attrs = cpuhp_cpu_root_attrs, |
1848 | .name = "hotplug", | |
1849 | NULL | |
1850 | }; | |
1851 | ||
1852 | static int __init cpuhp_sysfs_init(void) | |
1853 | { | |
1854 | int cpu, ret; | |
1855 | ||
1856 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1857 | &cpuhp_cpu_root_attr_group); | |
1858 | if (ret) | |
1859 | return ret; | |
1860 | ||
1861 | for_each_possible_cpu(cpu) { | |
1862 | struct device *dev = get_cpu_device(cpu); | |
1863 | ||
1864 | if (!dev) | |
1865 | continue; | |
1866 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1867 | if (ret) | |
1868 | return ret; | |
1869 | } | |
1870 | return 0; | |
1871 | } | |
1872 | device_initcall(cpuhp_sysfs_init); | |
1873 | #endif | |
1874 | ||
e56b3bc7 LT |
1875 | /* |
1876 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1877 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1878 | * | |
e0b582ec | 1879 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1880 | * mask value that has a single bit set only. |
1881 | */ | |
b8d317d1 | 1882 | |
e56b3bc7 | 1883 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1884 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1885 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1886 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1887 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1888 | |
e56b3bc7 LT |
1889 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1890 | ||
1891 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1892 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1893 | #if BITS_PER_LONG > 32 | |
1894 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1895 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1896 | #endif |
1897 | }; | |
e56b3bc7 | 1898 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1899 | |
1900 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1901 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1902 | |
1903 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1904 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1905 | = {CPU_BITS_ALL}; |
b3199c02 | 1906 | #else |
4b804c85 | 1907 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1908 | #endif |
4b804c85 | 1909 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1910 | |
4b804c85 RV |
1911 | struct cpumask __cpu_online_mask __read_mostly; |
1912 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1913 | |
4b804c85 RV |
1914 | struct cpumask __cpu_present_mask __read_mostly; |
1915 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1916 | |
4b804c85 RV |
1917 | struct cpumask __cpu_active_mask __read_mostly; |
1918 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1919 | |
3fa41520 RR |
1920 | void init_cpu_present(const struct cpumask *src) |
1921 | { | |
c4c54dd1 | 1922 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1923 | } |
1924 | ||
1925 | void init_cpu_possible(const struct cpumask *src) | |
1926 | { | |
c4c54dd1 | 1927 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1928 | } |
1929 | ||
1930 | void init_cpu_online(const struct cpumask *src) | |
1931 | { | |
c4c54dd1 | 1932 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1933 | } |
cff7d378 TG |
1934 | |
1935 | /* | |
1936 | * Activate the first processor. | |
1937 | */ | |
1938 | void __init boot_cpu_init(void) | |
1939 | { | |
1940 | int cpu = smp_processor_id(); | |
1941 | ||
1942 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1943 | set_cpu_online(cpu, true); | |
1944 | set_cpu_active(cpu, true); | |
1945 | set_cpu_present(cpu, true); | |
1946 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
1947 | |
1948 | #ifdef CONFIG_SMP | |
1949 | __boot_cpu_id = cpu; | |
1950 | #endif | |
cff7d378 TG |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * Must be called _AFTER_ setting up the per_cpu areas | |
1955 | */ | |
1956 | void __init boot_cpu_state_init(void) | |
1957 | { | |
1958 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1959 | } |