]>
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> |
a74cfffb | 13 | #include <linux/sched/smt.h> |
1da177e4 LT |
14 | #include <linux/unistd.h> |
15 | #include <linux/cpu.h> | |
cb79295e AV |
16 | #include <linux/oom.h> |
17 | #include <linux/rcupdate.h> | |
9984de1a | 18 | #include <linux/export.h> |
e4cc2f87 | 19 | #include <linux/bug.h> |
1da177e4 LT |
20 | #include <linux/kthread.h> |
21 | #include <linux/stop_machine.h> | |
81615b62 | 22 | #include <linux/mutex.h> |
5a0e3ad6 | 23 | #include <linux/gfp.h> |
79cfbdfa | 24 | #include <linux/suspend.h> |
a19423b9 | 25 | #include <linux/lockdep.h> |
345527b1 | 26 | #include <linux/tick.h> |
a8994181 | 27 | #include <linux/irq.h> |
941154bd | 28 | #include <linux/nmi.h> |
4cb28ced | 29 | #include <linux/smpboot.h> |
e6d4989a | 30 | #include <linux/relay.h> |
6731d4f1 | 31 | #include <linux/slab.h> |
fc8dffd3 | 32 | #include <linux/percpu-rwsem.h> |
cff7d378 | 33 | |
bb3632c6 | 34 | #include <trace/events/power.h> |
cff7d378 TG |
35 | #define CREATE_TRACE_POINTS |
36 | #include <trace/events/cpuhp.h> | |
1da177e4 | 37 | |
38498a67 TG |
38 | #include "smpboot.h" |
39 | ||
cff7d378 TG |
40 | /** |
41 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
42 | * @state: The current cpu state | |
43 | * @target: The target state | |
4cb28ced TG |
44 | * @thread: Pointer to the hotplug thread |
45 | * @should_run: Thread should execute | |
3b9d6da6 | 46 | * @rollback: Perform a rollback |
a724632c TG |
47 | * @single: Single callback invocation |
48 | * @bringup: Single callback bringup or teardown selector | |
49 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced | 50 | * @result: Result of the operation |
5ebe7742 PZ |
51 | * @done_up: Signal completion to the issuer of the task for cpu-up |
52 | * @done_down: Signal completion to the issuer of the task for cpu-down | |
cff7d378 TG |
53 | */ |
54 | struct cpuhp_cpu_state { | |
55 | enum cpuhp_state state; | |
56 | enum cpuhp_state target; | |
1db49484 | 57 | enum cpuhp_state fail; |
4cb28ced TG |
58 | #ifdef CONFIG_SMP |
59 | struct task_struct *thread; | |
60 | bool should_run; | |
3b9d6da6 | 61 | bool rollback; |
a724632c TG |
62 | bool single; |
63 | bool bringup; | |
0cc3cd21 | 64 | bool booted_once; |
cf392d10 | 65 | struct hlist_node *node; |
4dddfb5f | 66 | struct hlist_node *last; |
4cb28ced | 67 | enum cpuhp_state cb_state; |
4cb28ced | 68 | int result; |
5ebe7742 PZ |
69 | struct completion done_up; |
70 | struct completion done_down; | |
4cb28ced | 71 | #endif |
cff7d378 TG |
72 | }; |
73 | ||
1db49484 PZ |
74 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = { |
75 | .fail = CPUHP_INVALID, | |
76 | }; | |
cff7d378 | 77 | |
49dfe2a6 | 78 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
79 | static struct lockdep_map cpuhp_state_up_map = |
80 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
81 | static struct lockdep_map cpuhp_state_down_map = | |
82 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
83 | ||
84 | ||
76dc6c09 | 85 | static inline void cpuhp_lock_acquire(bool bringup) |
5f4b55e1 PZ |
86 | { |
87 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
88 | } | |
89 | ||
76dc6c09 | 90 | static inline void cpuhp_lock_release(bool bringup) |
5f4b55e1 PZ |
91 | { |
92 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
93 | } | |
94 | #else | |
95 | ||
76dc6c09 MM |
96 | static inline void cpuhp_lock_acquire(bool bringup) { } |
97 | static inline void cpuhp_lock_release(bool bringup) { } | |
5f4b55e1 | 98 | |
49dfe2a6 TG |
99 | #endif |
100 | ||
cff7d378 TG |
101 | /** |
102 | * cpuhp_step - Hotplug state machine step | |
103 | * @name: Name of the step | |
104 | * @startup: Startup function of the step | |
105 | * @teardown: Teardown function of the step | |
757c989b | 106 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
107 | */ |
108 | struct cpuhp_step { | |
cf392d10 TG |
109 | const char *name; |
110 | union { | |
3c1627e9 TG |
111 | int (*single)(unsigned int cpu); |
112 | int (*multi)(unsigned int cpu, | |
113 | struct hlist_node *node); | |
114 | } startup; | |
cf392d10 | 115 | union { |
3c1627e9 TG |
116 | int (*single)(unsigned int cpu); |
117 | int (*multi)(unsigned int cpu, | |
118 | struct hlist_node *node); | |
119 | } teardown; | |
cf392d10 | 120 | struct hlist_head list; |
cf392d10 TG |
121 | bool cant_stop; |
122 | bool multi_instance; | |
cff7d378 TG |
123 | }; |
124 | ||
98f8cdce | 125 | static DEFINE_MUTEX(cpuhp_state_mutex); |
17a2f1ce | 126 | static struct cpuhp_step cpuhp_hp_states[]; |
cff7d378 | 127 | |
a724632c TG |
128 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) |
129 | { | |
17a2f1ce | 130 | return cpuhp_hp_states + state; |
a724632c TG |
131 | } |
132 | ||
cff7d378 TG |
133 | /** |
134 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
135 | * @cpu: The cpu for which the callback should be invoked | |
96abb968 | 136 | * @state: The state to do callbacks for |
a724632c | 137 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
138 | * @node: For multi-instance, do a single entry callback for install/remove |
139 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 140 | * |
cf392d10 | 141 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 142 | */ |
a724632c | 143 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
144 | bool bringup, struct hlist_node *node, |
145 | struct hlist_node **lastp) | |
cff7d378 TG |
146 | { |
147 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 148 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
149 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
150 | int (*cb)(unsigned int cpu); | |
151 | int ret, cnt; | |
152 | ||
1db49484 PZ |
153 | if (st->fail == state) { |
154 | st->fail = CPUHP_INVALID; | |
155 | ||
156 | if (!(bringup ? step->startup.single : step->teardown.single)) | |
157 | return 0; | |
158 | ||
159 | return -EAGAIN; | |
160 | } | |
161 | ||
cf392d10 | 162 | if (!step->multi_instance) { |
96abb968 | 163 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 164 | cb = bringup ? step->startup.single : step->teardown.single; |
cf392d10 TG |
165 | if (!cb) |
166 | return 0; | |
a724632c | 167 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 168 | ret = cb(cpu); |
a724632c | 169 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
170 | return ret; |
171 | } | |
3c1627e9 | 172 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
173 | if (!cbm) |
174 | return 0; | |
175 | ||
176 | /* Single invocation for instance add/remove */ | |
177 | if (node) { | |
96abb968 | 178 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
179 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
180 | ret = cbm(cpu, node); | |
181 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
182 | return ret; | |
183 | } | |
184 | ||
185 | /* State transition. Invoke on all instances */ | |
186 | cnt = 0; | |
187 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
188 | if (lastp && node == *lastp) |
189 | break; | |
190 | ||
cf392d10 TG |
191 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
192 | ret = cbm(cpu, node); | |
193 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
194 | if (ret) { |
195 | if (!lastp) | |
196 | goto err; | |
197 | ||
198 | *lastp = node; | |
199 | return ret; | |
200 | } | |
cf392d10 TG |
201 | cnt++; |
202 | } | |
96abb968 PZ |
203 | if (lastp) |
204 | *lastp = NULL; | |
cf392d10 TG |
205 | return 0; |
206 | err: | |
207 | /* Rollback the instances if one failed */ | |
3c1627e9 | 208 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
209 | if (!cbm) |
210 | return ret; | |
211 | ||
212 | hlist_for_each(node, &step->list) { | |
213 | if (!cnt--) | |
214 | break; | |
724a8688 PZ |
215 | |
216 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
217 | ret = cbm(cpu, node); | |
218 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
219 | /* | |
220 | * Rollback must not fail, | |
221 | */ | |
222 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
223 | } |
224 | return ret; | |
225 | } | |
226 | ||
98a79d6a | 227 | #ifdef CONFIG_SMP |
fcb3029a AB |
228 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
229 | { | |
230 | /* | |
231 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
232 | * purposes as that state is handled explicitly in cpu_down. | |
233 | */ | |
234 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
235 | } | |
236 | ||
5ebe7742 PZ |
237 | static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup) |
238 | { | |
239 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
240 | wait_for_completion(done); | |
241 | } | |
242 | ||
243 | static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup) | |
244 | { | |
245 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
246 | complete(done); | |
247 | } | |
248 | ||
249 | /* | |
250 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
251 | */ | |
252 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
253 | { | |
254 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
255 | } | |
256 | ||
b3199c02 | 257 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 258 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
259 | bool cpuhp_tasks_frozen; |
260 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 261 | |
79a6cdeb | 262 | /* |
93ae4f97 SB |
263 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
264 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
265 | */ |
266 | void cpu_maps_update_begin(void) | |
267 | { | |
268 | mutex_lock(&cpu_add_remove_lock); | |
269 | } | |
270 | ||
271 | void cpu_maps_update_done(void) | |
272 | { | |
273 | mutex_unlock(&cpu_add_remove_lock); | |
274 | } | |
1da177e4 | 275 | |
fc8dffd3 TG |
276 | /* |
277 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
278 | * Should always be manipulated under cpu_add_remove_lock |
279 | */ | |
280 | static int cpu_hotplug_disabled; | |
281 | ||
79a6cdeb LJ |
282 | #ifdef CONFIG_HOTPLUG_CPU |
283 | ||
fc8dffd3 | 284 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 285 | |
8f553c49 | 286 | void cpus_read_lock(void) |
a9d9baa1 | 287 | { |
fc8dffd3 | 288 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 289 | } |
8f553c49 | 290 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 291 | |
6f4ceee9 WL |
292 | int cpus_read_trylock(void) |
293 | { | |
294 | return percpu_down_read_trylock(&cpu_hotplug_lock); | |
295 | } | |
296 | EXPORT_SYMBOL_GPL(cpus_read_trylock); | |
297 | ||
8f553c49 | 298 | void cpus_read_unlock(void) |
a9d9baa1 | 299 | { |
fc8dffd3 | 300 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 301 | } |
8f553c49 | 302 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 303 | |
8f553c49 | 304 | void cpus_write_lock(void) |
d221938c | 305 | { |
fc8dffd3 | 306 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 307 | } |
87af9e7f | 308 | |
8f553c49 | 309 | void cpus_write_unlock(void) |
d221938c | 310 | { |
fc8dffd3 | 311 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
312 | } |
313 | ||
fc8dffd3 | 314 | void lockdep_assert_cpus_held(void) |
d221938c | 315 | { |
ce48c457 VS |
316 | /* |
317 | * We can't have hotplug operations before userspace starts running, | |
318 | * and some init codepaths will knowingly not take the hotplug lock. | |
319 | * This is all valid, so mute lockdep until it makes sense to report | |
320 | * unheld locks. | |
321 | */ | |
322 | if (system_state < SYSTEM_RUNNING) | |
323 | return; | |
324 | ||
fc8dffd3 | 325 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 326 | } |
79a6cdeb | 327 | |
cb92173d PZ |
328 | static void lockdep_acquire_cpus_lock(void) |
329 | { | |
330 | rwsem_acquire(&cpu_hotplug_lock.rw_sem.dep_map, 0, 0, _THIS_IP_); | |
331 | } | |
332 | ||
333 | static void lockdep_release_cpus_lock(void) | |
334 | { | |
335 | rwsem_release(&cpu_hotplug_lock.rw_sem.dep_map, 1, _THIS_IP_); | |
336 | } | |
337 | ||
16e53dbf SB |
338 | /* |
339 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
340 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
341 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
342 | * hotplug path before performing hotplug operations. So acquiring that lock | |
343 | * guarantees mutual exclusion from any currently running hotplug operations. | |
344 | */ | |
345 | void cpu_hotplug_disable(void) | |
346 | { | |
347 | cpu_maps_update_begin(); | |
89af7ba5 | 348 | cpu_hotplug_disabled++; |
16e53dbf SB |
349 | cpu_maps_update_done(); |
350 | } | |
32145c46 | 351 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 352 | |
01b41159 LW |
353 | static void __cpu_hotplug_enable(void) |
354 | { | |
355 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
356 | return; | |
357 | cpu_hotplug_disabled--; | |
358 | } | |
359 | ||
16e53dbf SB |
360 | void cpu_hotplug_enable(void) |
361 | { | |
362 | cpu_maps_update_begin(); | |
01b41159 | 363 | __cpu_hotplug_enable(); |
16e53dbf SB |
364 | cpu_maps_update_done(); |
365 | } | |
32145c46 | 366 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
cb92173d PZ |
367 | |
368 | #else | |
369 | ||
370 | static void lockdep_acquire_cpus_lock(void) | |
371 | { | |
372 | } | |
373 | ||
374 | static void lockdep_release_cpus_lock(void) | |
375 | { | |
376 | } | |
377 | ||
b9d10be7 | 378 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 379 | |
a74cfffb TG |
380 | /* |
381 | * Architectures that need SMT-specific errata handling during SMT hotplug | |
382 | * should override this. | |
383 | */ | |
384 | void __weak arch_smt_update(void) { } | |
385 | ||
0cc3cd21 TG |
386 | #ifdef CONFIG_HOTPLUG_SMT |
387 | enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED; | |
bc2d8d26 | 388 | |
8e1b706b | 389 | void __init cpu_smt_disable(bool force) |
0cc3cd21 | 390 | { |
8e1b706b JK |
391 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED || |
392 | cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
393 | return; | |
394 | ||
395 | if (force) { | |
0cc3cd21 TG |
396 | pr_info("SMT: Force disabled\n"); |
397 | cpu_smt_control = CPU_SMT_FORCE_DISABLED; | |
8e1b706b | 398 | } else { |
d0e7d144 | 399 | pr_info("SMT: disabled\n"); |
8e1b706b | 400 | cpu_smt_control = CPU_SMT_DISABLED; |
0cc3cd21 | 401 | } |
8e1b706b JK |
402 | } |
403 | ||
fee0aede TG |
404 | /* |
405 | * The decision whether SMT is supported can only be done after the full | |
b284909a | 406 | * CPU identification. Called from architecture code. |
bc2d8d26 TG |
407 | */ |
408 | void __init cpu_smt_check_topology(void) | |
409 | { | |
b284909a | 410 | if (!topology_smt_supported()) |
bc2d8d26 TG |
411 | cpu_smt_control = CPU_SMT_NOT_SUPPORTED; |
412 | } | |
413 | ||
8e1b706b JK |
414 | static int __init smt_cmdline_disable(char *str) |
415 | { | |
416 | cpu_smt_disable(str && !strcmp(str, "force")); | |
0cc3cd21 TG |
417 | return 0; |
418 | } | |
419 | early_param("nosmt", smt_cmdline_disable); | |
420 | ||
421 | static inline bool cpu_smt_allowed(unsigned int cpu) | |
422 | { | |
b284909a | 423 | if (cpu_smt_control == CPU_SMT_ENABLED) |
0cc3cd21 TG |
424 | return true; |
425 | ||
b284909a | 426 | if (topology_is_primary_thread(cpu)) |
0cc3cd21 TG |
427 | return true; |
428 | ||
429 | /* | |
430 | * On x86 it's required to boot all logical CPUs at least once so | |
431 | * that the init code can get a chance to set CR4.MCE on each | |
432 | * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any | |
433 | * core will shutdown the machine. | |
434 | */ | |
435 | return !per_cpu(cpuhp_state, cpu).booted_once; | |
436 | } | |
437 | #else | |
438 | static inline bool cpu_smt_allowed(unsigned int cpu) { return true; } | |
439 | #endif | |
440 | ||
4dddfb5f PZ |
441 | static inline enum cpuhp_state |
442 | cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
443 | { | |
444 | enum cpuhp_state prev_state = st->state; | |
445 | ||
446 | st->rollback = false; | |
447 | st->last = NULL; | |
448 | ||
449 | st->target = target; | |
450 | st->single = false; | |
451 | st->bringup = st->state < target; | |
452 | ||
453 | return prev_state; | |
454 | } | |
455 | ||
456 | static inline void | |
457 | cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state) | |
458 | { | |
459 | st->rollback = true; | |
460 | ||
461 | /* | |
462 | * If we have st->last we need to undo partial multi_instance of this | |
463 | * state first. Otherwise start undo at the previous state. | |
464 | */ | |
465 | if (!st->last) { | |
466 | if (st->bringup) | |
467 | st->state--; | |
468 | else | |
469 | st->state++; | |
470 | } | |
471 | ||
472 | st->target = prev_state; | |
473 | st->bringup = !st->bringup; | |
474 | } | |
475 | ||
476 | /* Regular hotplug invocation of the AP hotplug thread */ | |
477 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
478 | { | |
479 | if (!st->single && st->state == st->target) | |
480 | return; | |
481 | ||
482 | st->result = 0; | |
483 | /* | |
484 | * Make sure the above stores are visible before should_run becomes | |
485 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
486 | */ | |
487 | smp_mb(); | |
488 | st->should_run = true; | |
489 | wake_up_process(st->thread); | |
5ebe7742 | 490 | wait_for_ap_thread(st, st->bringup); |
4dddfb5f PZ |
491 | } |
492 | ||
493 | static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
494 | { | |
495 | enum cpuhp_state prev_state; | |
496 | int ret; | |
497 | ||
498 | prev_state = cpuhp_set_state(st, target); | |
499 | __cpuhp_kick_ap(st); | |
500 | if ((ret = st->result)) { | |
501 | cpuhp_reset_state(st, prev_state); | |
502 | __cpuhp_kick_ap(st); | |
503 | } | |
504 | ||
505 | return ret; | |
506 | } | |
9cd4f1a4 | 507 | |
8df3e07e TG |
508 | static int bringup_wait_for_ap(unsigned int cpu) |
509 | { | |
510 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
511 | ||
9cd4f1a4 | 512 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
5ebe7742 | 513 | wait_for_ap_thread(st, true); |
dea1d0f5 TG |
514 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
515 | return -ECANCELED; | |
9cd4f1a4 TG |
516 | |
517 | /* Unpark the stopper thread and the hotplug thread of the target cpu */ | |
518 | stop_machine_unpark(cpu); | |
519 | kthread_unpark(st->thread); | |
520 | ||
0cc3cd21 TG |
521 | /* |
522 | * SMT soft disabling on X86 requires to bring the CPU out of the | |
523 | * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The | |
524 | * CPU marked itself as booted_once in cpu_notify_starting() so the | |
525 | * cpu_smt_allowed() check will now return false if this is not the | |
526 | * primary sibling. | |
527 | */ | |
528 | if (!cpu_smt_allowed(cpu)) | |
529 | return -ECANCELED; | |
530 | ||
4dddfb5f PZ |
531 | if (st->target <= CPUHP_AP_ONLINE_IDLE) |
532 | return 0; | |
533 | ||
534 | return cpuhp_kick_ap(st, st->target); | |
8df3e07e TG |
535 | } |
536 | ||
ba997462 TG |
537 | static int bringup_cpu(unsigned int cpu) |
538 | { | |
539 | struct task_struct *idle = idle_thread_get(cpu); | |
540 | int ret; | |
541 | ||
aa877175 BO |
542 | /* |
543 | * Some architectures have to walk the irq descriptors to | |
544 | * setup the vector space for the cpu which comes online. | |
545 | * Prevent irq alloc/free across the bringup. | |
546 | */ | |
547 | irq_lock_sparse(); | |
548 | ||
ba997462 TG |
549 | /* Arch-specific enabling code. */ |
550 | ret = __cpu_up(cpu, idle); | |
aa877175 | 551 | irq_unlock_sparse(); |
530e9b76 | 552 | if (ret) |
ba997462 | 553 | return ret; |
9cd4f1a4 | 554 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
555 | } |
556 | ||
2e1a3483 TG |
557 | /* |
558 | * Hotplug state machine related functions | |
559 | */ | |
2e1a3483 | 560 | |
a724632c | 561 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st) |
2e1a3483 | 562 | { |
6fb86d97 MO |
563 | for (st->state--; st->state > st->target; st->state--) |
564 | cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
2e1a3483 TG |
565 | } |
566 | ||
567 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
a724632c | 568 | enum cpuhp_state target) |
2e1a3483 TG |
569 | { |
570 | enum cpuhp_state prev_state = st->state; | |
571 | int ret = 0; | |
572 | ||
573 | while (st->state < target) { | |
2e1a3483 | 574 | st->state++; |
96abb968 | 575 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
2e1a3483 TG |
576 | if (ret) { |
577 | st->target = prev_state; | |
a724632c | 578 | undo_cpu_up(cpu, st); |
2e1a3483 TG |
579 | break; |
580 | } | |
581 | } | |
582 | return ret; | |
583 | } | |
584 | ||
4cb28ced TG |
585 | /* |
586 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
587 | */ | |
588 | static void cpuhp_create(unsigned int cpu) | |
589 | { | |
590 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
591 | ||
5ebe7742 PZ |
592 | init_completion(&st->done_up); |
593 | init_completion(&st->done_down); | |
4cb28ced TG |
594 | } |
595 | ||
596 | static int cpuhp_should_run(unsigned int cpu) | |
597 | { | |
598 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
599 | ||
600 | return st->should_run; | |
601 | } | |
602 | ||
4cb28ced TG |
603 | /* |
604 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
605 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
606 | * |
607 | * Each invocation of this function by the smpboot thread does a single AP | |
608 | * state callback. | |
609 | * | |
610 | * It has 3 modes of operation: | |
611 | * - single: runs st->cb_state | |
612 | * - up: runs ++st->state, while st->state < st->target | |
613 | * - down: runs st->state--, while st->state > st->target | |
614 | * | |
615 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
616 | */ |
617 | static void cpuhp_thread_fun(unsigned int cpu) | |
618 | { | |
619 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
620 | bool bringup = st->bringup; |
621 | enum cpuhp_state state; | |
4cb28ced | 622 | |
f8b7530a NU |
623 | if (WARN_ON_ONCE(!st->should_run)) |
624 | return; | |
625 | ||
4cb28ced | 626 | /* |
4dddfb5f PZ |
627 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
628 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
629 | */ |
630 | smp_mb(); | |
4cb28ced | 631 | |
cb92173d PZ |
632 | /* |
633 | * The BP holds the hotplug lock, but we're now running on the AP, | |
634 | * ensure that anybody asserting the lock is held, will actually find | |
635 | * it so. | |
636 | */ | |
637 | lockdep_acquire_cpus_lock(); | |
5f4b55e1 | 638 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 639 | |
a724632c | 640 | if (st->single) { |
4dddfb5f PZ |
641 | state = st->cb_state; |
642 | st->should_run = false; | |
643 | } else { | |
644 | if (bringup) { | |
645 | st->state++; | |
646 | state = st->state; | |
647 | st->should_run = (st->state < st->target); | |
648 | WARN_ON_ONCE(st->state > st->target); | |
4cb28ced | 649 | } else { |
4dddfb5f PZ |
650 | state = st->state; |
651 | st->state--; | |
652 | st->should_run = (st->state > st->target); | |
653 | WARN_ON_ONCE(st->state < st->target); | |
4cb28ced | 654 | } |
4dddfb5f PZ |
655 | } |
656 | ||
657 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
658 | ||
4dddfb5f PZ |
659 | if (cpuhp_is_atomic_state(state)) { |
660 | local_irq_disable(); | |
661 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
662 | local_irq_enable(); | |
3b9d6da6 | 663 | |
4dddfb5f PZ |
664 | /* |
665 | * STARTING/DYING must not fail! | |
666 | */ | |
667 | WARN_ON_ONCE(st->result); | |
4cb28ced | 668 | } else { |
4dddfb5f PZ |
669 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
670 | } | |
671 | ||
672 | if (st->result) { | |
673 | /* | |
674 | * If we fail on a rollback, we're up a creek without no | |
675 | * paddle, no way forward, no way back. We loose, thanks for | |
676 | * playing. | |
677 | */ | |
678 | WARN_ON_ONCE(st->rollback); | |
679 | st->should_run = false; | |
4cb28ced | 680 | } |
4dddfb5f | 681 | |
5f4b55e1 | 682 | cpuhp_lock_release(bringup); |
cb92173d | 683 | lockdep_release_cpus_lock(); |
4dddfb5f PZ |
684 | |
685 | if (!st->should_run) | |
5ebe7742 | 686 | complete_ap_thread(st, bringup); |
4cb28ced TG |
687 | } |
688 | ||
689 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 690 | static int |
cf392d10 TG |
691 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
692 | struct hlist_node *node) | |
4cb28ced TG |
693 | { |
694 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 695 | int ret; |
4cb28ced TG |
696 | |
697 | if (!cpu_online(cpu)) | |
698 | return 0; | |
699 | ||
5f4b55e1 PZ |
700 | cpuhp_lock_acquire(false); |
701 | cpuhp_lock_release(false); | |
702 | ||
703 | cpuhp_lock_acquire(true); | |
704 | cpuhp_lock_release(true); | |
49dfe2a6 | 705 | |
6a4e2451 TG |
706 | /* |
707 | * If we are up and running, use the hotplug thread. For early calls | |
708 | * we invoke the thread function directly. | |
709 | */ | |
710 | if (!st->thread) | |
96abb968 | 711 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 712 | |
4dddfb5f PZ |
713 | st->rollback = false; |
714 | st->last = NULL; | |
715 | ||
716 | st->node = node; | |
717 | st->bringup = bringup; | |
4cb28ced | 718 | st->cb_state = state; |
a724632c | 719 | st->single = true; |
a724632c | 720 | |
4dddfb5f | 721 | __cpuhp_kick_ap(st); |
4cb28ced | 722 | |
4cb28ced | 723 | /* |
4dddfb5f | 724 | * If we failed and did a partial, do a rollback. |
4cb28ced | 725 | */ |
4dddfb5f PZ |
726 | if ((ret = st->result) && st->last) { |
727 | st->rollback = true; | |
728 | st->bringup = !bringup; | |
729 | ||
730 | __cpuhp_kick_ap(st); | |
731 | } | |
732 | ||
1f7c70d6 TG |
733 | /* |
734 | * Clean up the leftovers so the next hotplug operation wont use stale | |
735 | * data. | |
736 | */ | |
737 | st->node = st->last = NULL; | |
4dddfb5f | 738 | return ret; |
1cf4f629 TG |
739 | } |
740 | ||
741 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
742 | { | |
743 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
744 | enum cpuhp_state prev_state = st->state; |
745 | int ret; | |
1cf4f629 | 746 | |
5f4b55e1 PZ |
747 | cpuhp_lock_acquire(false); |
748 | cpuhp_lock_release(false); | |
749 | ||
750 | cpuhp_lock_acquire(true); | |
751 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
752 | |
753 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
754 | ret = cpuhp_kick_ap(st, st->target); | |
755 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); | |
756 | ||
757 | return ret; | |
4cb28ced TG |
758 | } |
759 | ||
760 | static struct smp_hotplug_thread cpuhp_threads = { | |
761 | .store = &cpuhp_state.thread, | |
762 | .create = &cpuhp_create, | |
763 | .thread_should_run = cpuhp_should_run, | |
764 | .thread_fn = cpuhp_thread_fun, | |
765 | .thread_comm = "cpuhp/%u", | |
766 | .selfparking = true, | |
767 | }; | |
768 | ||
769 | void __init cpuhp_threads_init(void) | |
770 | { | |
771 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
772 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
773 | } | |
774 | ||
777c6e0d | 775 | #ifdef CONFIG_HOTPLUG_CPU |
e4cc2f87 AV |
776 | /** |
777 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
778 | * @cpu: a CPU id | |
779 | * | |
780 | * This function walks all processes, finds a valid mm struct for each one and | |
781 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
782 | * trivial, there are various non-obvious corner cases, which this function | |
783 | * tries to solve in a safe manner. | |
784 | * | |
785 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
786 | * be called only for an already offlined CPU. | |
787 | */ | |
cb79295e AV |
788 | void clear_tasks_mm_cpumask(int cpu) |
789 | { | |
790 | struct task_struct *p; | |
791 | ||
792 | /* | |
793 | * This function is called after the cpu is taken down and marked | |
794 | * offline, so its not like new tasks will ever get this cpu set in | |
795 | * their mm mask. -- Peter Zijlstra | |
796 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
797 | * full-fledged tasklist_lock. | |
798 | */ | |
e4cc2f87 | 799 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
800 | rcu_read_lock(); |
801 | for_each_process(p) { | |
802 | struct task_struct *t; | |
803 | ||
e4cc2f87 AV |
804 | /* |
805 | * Main thread might exit, but other threads may still have | |
806 | * a valid mm. Find one. | |
807 | */ | |
cb79295e AV |
808 | t = find_lock_task_mm(p); |
809 | if (!t) | |
810 | continue; | |
811 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
812 | task_unlock(t); | |
813 | } | |
814 | rcu_read_unlock(); | |
815 | } | |
816 | ||
1da177e4 | 817 | /* Take this CPU down. */ |
71cf5aee | 818 | static int take_cpu_down(void *_param) |
1da177e4 | 819 | { |
4baa0afc TG |
820 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
821 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 822 | int err, cpu = smp_processor_id(); |
724a8688 | 823 | int ret; |
1da177e4 | 824 | |
1da177e4 LT |
825 | /* Ensure this CPU doesn't handle any more interrupts. */ |
826 | err = __cpu_disable(); | |
827 | if (err < 0) | |
f3705136 | 828 | return err; |
1da177e4 | 829 | |
a724632c TG |
830 | /* |
831 | * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not | |
832 | * do this step again. | |
833 | */ | |
834 | WARN_ON(st->state != CPUHP_TEARDOWN_CPU); | |
835 | st->state--; | |
4baa0afc | 836 | /* Invoke the former CPU_DYING callbacks */ |
724a8688 PZ |
837 | for (; st->state > target; st->state--) { |
838 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
839 | /* | |
840 | * DYING must not fail! | |
841 | */ | |
842 | WARN_ON_ONCE(ret); | |
843 | } | |
4baa0afc | 844 | |
52c063d1 TG |
845 | /* Give up timekeeping duties */ |
846 | tick_handover_do_timer(); | |
14e568e7 | 847 | /* Park the stopper thread */ |
090e77c3 | 848 | stop_machine_park(cpu); |
f3705136 | 849 | return 0; |
1da177e4 LT |
850 | } |
851 | ||
98458172 | 852 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 853 | { |
e69aab13 | 854 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 855 | int err; |
1da177e4 | 856 | |
2a58c527 | 857 | /* Park the smpboot threads */ |
1cf4f629 TG |
858 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
859 | ||
6acce3ef | 860 | /* |
a8994181 TG |
861 | * Prevent irq alloc/free while the dying cpu reorganizes the |
862 | * interrupt affinities. | |
6acce3ef | 863 | */ |
a8994181 | 864 | irq_lock_sparse(); |
6acce3ef | 865 | |
a8994181 TG |
866 | /* |
867 | * So now all preempt/rcu users must observe !cpu_active(). | |
868 | */ | |
210e2133 | 869 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 870 | if (err) { |
3b9d6da6 | 871 | /* CPU refused to die */ |
a8994181 | 872 | irq_unlock_sparse(); |
3b9d6da6 SAS |
873 | /* Unpark the hotplug thread so we can rollback there */ |
874 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 875 | return err; |
8fa1d7d3 | 876 | } |
04321587 | 877 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 878 | |
48c5ccae | 879 | /* |
5b1ead68 BJ |
880 | * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed |
881 | * all runnable tasks from the CPU, there's only the idle task left now | |
48c5ccae | 882 | * that the migration thread is done doing the stop_machine thing. |
51a96c77 PZ |
883 | * |
884 | * Wait for the stop thread to go away. | |
48c5ccae | 885 | */ |
5ebe7742 | 886 | wait_for_ap_thread(st, false); |
e69aab13 | 887 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); |
1da177e4 | 888 | |
a8994181 TG |
889 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
890 | irq_unlock_sparse(); | |
891 | ||
345527b1 | 892 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
893 | /* This actually kills the CPU. */ |
894 | __cpu_die(cpu); | |
895 | ||
a49b116d | 896 | tick_cleanup_dead_cpu(cpu); |
a58163d8 | 897 | rcutree_migrate_callbacks(cpu); |
98458172 TG |
898 | return 0; |
899 | } | |
1da177e4 | 900 | |
71f87b2f TG |
901 | static void cpuhp_complete_idle_dead(void *arg) |
902 | { | |
903 | struct cpuhp_cpu_state *st = arg; | |
904 | ||
5ebe7742 | 905 | complete_ap_thread(st, false); |
71f87b2f TG |
906 | } |
907 | ||
e69aab13 TG |
908 | void cpuhp_report_idle_dead(void) |
909 | { | |
910 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
911 | ||
912 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 913 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
914 | st->state = CPUHP_AP_IDLE_DEAD; |
915 | /* | |
916 | * We cannot call complete after rcu_report_dead() so we delegate it | |
917 | * to an online cpu. | |
918 | */ | |
919 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
920 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
921 | } |
922 | ||
4dddfb5f PZ |
923 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st) |
924 | { | |
6fb86d97 MO |
925 | for (st->state++; st->state < st->target; st->state++) |
926 | cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); | |
4dddfb5f PZ |
927 | } |
928 | ||
929 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
930 | enum cpuhp_state target) | |
931 | { | |
932 | enum cpuhp_state prev_state = st->state; | |
933 | int ret = 0; | |
934 | ||
935 | for (; st->state > target; st->state--) { | |
936 | ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL); | |
937 | if (ret) { | |
938 | st->target = prev_state; | |
69fa6eb7 TG |
939 | if (st->state < prev_state) |
940 | undo_cpu_down(cpu, st); | |
4dddfb5f PZ |
941 | break; |
942 | } | |
943 | } | |
944 | return ret; | |
945 | } | |
cff7d378 | 946 | |
98458172 | 947 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
948 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
949 | enum cpuhp_state target) | |
98458172 | 950 | { |
cff7d378 TG |
951 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
952 | int prev_state, ret = 0; | |
98458172 TG |
953 | |
954 | if (num_online_cpus() == 1) | |
955 | return -EBUSY; | |
956 | ||
757c989b | 957 | if (!cpu_present(cpu)) |
98458172 TG |
958 | return -EINVAL; |
959 | ||
8f553c49 | 960 | cpus_write_lock(); |
98458172 TG |
961 | |
962 | cpuhp_tasks_frozen = tasks_frozen; | |
963 | ||
4dddfb5f | 964 | prev_state = cpuhp_set_state(st, target); |
1cf4f629 TG |
965 | /* |
966 | * If the current CPU state is in the range of the AP hotplug thread, | |
967 | * then we need to kick the thread. | |
968 | */ | |
8df3e07e | 969 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 970 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
971 | ret = cpuhp_kick_ap_work(cpu); |
972 | /* | |
973 | * The AP side has done the error rollback already. Just | |
974 | * return the error code.. | |
975 | */ | |
976 | if (ret) | |
977 | goto out; | |
978 | ||
979 | /* | |
980 | * We might have stopped still in the range of the AP hotplug | |
981 | * thread. Nothing to do anymore. | |
982 | */ | |
8df3e07e | 983 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 984 | goto out; |
4dddfb5f PZ |
985 | |
986 | st->target = target; | |
1cf4f629 TG |
987 | } |
988 | /* | |
8df3e07e | 989 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
990 | * to do the further cleanups. |
991 | */ | |
a724632c | 992 | ret = cpuhp_down_callbacks(cpu, st, target); |
69fa6eb7 | 993 | if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) { |
4dddfb5f PZ |
994 | cpuhp_reset_state(st, prev_state); |
995 | __cpuhp_kick_ap(st); | |
3b9d6da6 | 996 | } |
98458172 | 997 | |
1cf4f629 | 998 | out: |
8f553c49 | 999 | cpus_write_unlock(); |
941154bd TG |
1000 | /* |
1001 | * Do post unplug cleanup. This is still protected against | |
1002 | * concurrent CPU hotplug via cpu_add_remove_lock. | |
1003 | */ | |
1004 | lockup_detector_cleanup(); | |
a74cfffb | 1005 | arch_smt_update(); |
cff7d378 | 1006 | return ret; |
e3920fb4 RW |
1007 | } |
1008 | ||
cc1fe215 TG |
1009 | static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target) |
1010 | { | |
1011 | if (cpu_hotplug_disabled) | |
1012 | return -EBUSY; | |
1013 | return _cpu_down(cpu, 0, target); | |
1014 | } | |
1015 | ||
af1f4045 | 1016 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 1017 | { |
9ea09af3 | 1018 | int err; |
e3920fb4 | 1019 | |
d221938c | 1020 | cpu_maps_update_begin(); |
cc1fe215 | 1021 | err = cpu_down_maps_locked(cpu, target); |
d221938c | 1022 | cpu_maps_update_done(); |
1da177e4 LT |
1023 | return err; |
1024 | } | |
4dddfb5f | 1025 | |
af1f4045 TG |
1026 | int cpu_down(unsigned int cpu) |
1027 | { | |
1028 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
1029 | } | |
b62b8ef9 | 1030 | EXPORT_SYMBOL(cpu_down); |
4dddfb5f PZ |
1031 | |
1032 | #else | |
1033 | #define takedown_cpu NULL | |
1da177e4 LT |
1034 | #endif /*CONFIG_HOTPLUG_CPU*/ |
1035 | ||
4baa0afc | 1036 | /** |
ee1e714b | 1037 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
1038 | * @cpu: cpu that just started |
1039 | * | |
4baa0afc TG |
1040 | * It must be called by the arch code on the new cpu, before the new cpu |
1041 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
1042 | */ | |
1043 | void notify_cpu_starting(unsigned int cpu) | |
1044 | { | |
1045 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1046 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
724a8688 | 1047 | int ret; |
4baa0afc | 1048 | |
0c6d4576 | 1049 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
0cc3cd21 | 1050 | st->booted_once = true; |
4baa0afc | 1051 | while (st->state < target) { |
4baa0afc | 1052 | st->state++; |
724a8688 PZ |
1053 | ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL); |
1054 | /* | |
1055 | * STARTING must not fail! | |
1056 | */ | |
1057 | WARN_ON_ONCE(ret); | |
4baa0afc TG |
1058 | } |
1059 | } | |
1060 | ||
949338e3 | 1061 | /* |
9cd4f1a4 TG |
1062 | * Called from the idle task. Wake up the controlling task which brings the |
1063 | * stopper and the hotplug thread of the upcoming CPU up and then delegates | |
1064 | * the rest of the online bringup to the hotplug thread. | |
949338e3 | 1065 | */ |
8df3e07e | 1066 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 1067 | { |
8df3e07e | 1068 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
1069 | |
1070 | /* Happens for the boot cpu */ | |
1071 | if (state != CPUHP_AP_ONLINE_IDLE) | |
1072 | return; | |
1073 | ||
1074 | st->state = CPUHP_AP_ONLINE_IDLE; | |
5ebe7742 | 1075 | complete_ap_thread(st, true); |
949338e3 TG |
1076 | } |
1077 | ||
e3920fb4 | 1078 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1079 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1080 | { |
cff7d378 | 1081 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1082 | struct task_struct *idle; |
2e1a3483 | 1083 | int ret = 0; |
1da177e4 | 1084 | |
8f553c49 | 1085 | cpus_write_lock(); |
38498a67 | 1086 | |
757c989b | 1087 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1088 | ret = -EINVAL; |
1089 | goto out; | |
1090 | } | |
1091 | ||
757c989b TG |
1092 | /* |
1093 | * The caller of do_cpu_up might have raced with another | |
1094 | * caller. Ignore it for now. | |
1095 | */ | |
1096 | if (st->state >= target) | |
38498a67 | 1097 | goto out; |
757c989b TG |
1098 | |
1099 | if (st->state == CPUHP_OFFLINE) { | |
1100 | /* Let it fail before we try to bring the cpu up */ | |
1101 | idle = idle_thread_get(cpu); | |
1102 | if (IS_ERR(idle)) { | |
1103 | ret = PTR_ERR(idle); | |
1104 | goto out; | |
1105 | } | |
3bb5d2ee | 1106 | } |
38498a67 | 1107 | |
ba997462 TG |
1108 | cpuhp_tasks_frozen = tasks_frozen; |
1109 | ||
4dddfb5f | 1110 | cpuhp_set_state(st, target); |
1cf4f629 TG |
1111 | /* |
1112 | * If the current CPU state is in the range of the AP hotplug thread, | |
1113 | * then we need to kick the thread once more. | |
1114 | */ | |
8df3e07e | 1115 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1116 | ret = cpuhp_kick_ap_work(cpu); |
1117 | /* | |
1118 | * The AP side has done the error rollback already. Just | |
1119 | * return the error code.. | |
1120 | */ | |
1121 | if (ret) | |
1122 | goto out; | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1127 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1128 | * responsible for bringing it up to the target state. |
1129 | */ | |
8df3e07e | 1130 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1131 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1132 | out: |
8f553c49 | 1133 | cpus_write_unlock(); |
a74cfffb | 1134 | arch_smt_update(); |
e3920fb4 RW |
1135 | return ret; |
1136 | } | |
1137 | ||
af1f4045 | 1138 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1139 | { |
1140 | int err = 0; | |
cf23422b | 1141 | |
e0b582ec | 1142 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1143 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1144 | cpu); | |
87d5e023 | 1145 | #if defined(CONFIG_IA64) |
84117da5 | 1146 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1147 | #endif |
1148 | return -EINVAL; | |
1149 | } | |
e3920fb4 | 1150 | |
01b0f197 TK |
1151 | err = try_online_node(cpu_to_node(cpu)); |
1152 | if (err) | |
1153 | return err; | |
cf23422b | 1154 | |
d221938c | 1155 | cpu_maps_update_begin(); |
e761b772 MK |
1156 | |
1157 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1158 | err = -EBUSY; |
e761b772 MK |
1159 | goto out; |
1160 | } | |
05736e4a TG |
1161 | if (!cpu_smt_allowed(cpu)) { |
1162 | err = -EPERM; | |
1163 | goto out; | |
1164 | } | |
e761b772 | 1165 | |
af1f4045 | 1166 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1167 | out: |
d221938c | 1168 | cpu_maps_update_done(); |
e3920fb4 RW |
1169 | return err; |
1170 | } | |
af1f4045 TG |
1171 | |
1172 | int cpu_up(unsigned int cpu) | |
1173 | { | |
1174 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
1175 | } | |
a513f6ba | 1176 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 1177 | |
f3de4be9 | 1178 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1179 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1180 | |
d391e552 | 1181 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1182 | { |
d391e552 | 1183 | int cpu, error = 0; |
e3920fb4 | 1184 | |
d221938c | 1185 | cpu_maps_update_begin(); |
d391e552 JM |
1186 | if (!cpu_online(primary)) |
1187 | primary = cpumask_first(cpu_online_mask); | |
9ee349ad XF |
1188 | /* |
1189 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1190 | * with the userspace trying to use the CPU hotplug at the same time |
1191 | */ | |
e0b582ec | 1192 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1193 | |
84117da5 | 1194 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1195 | for_each_online_cpu(cpu) { |
d391e552 | 1196 | if (cpu == primary) |
e3920fb4 | 1197 | continue; |
bb3632c6 | 1198 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1199 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1200 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1201 | if (!error) |
e0b582ec | 1202 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1203 | else { |
84117da5 | 1204 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1205 | break; |
1206 | } | |
1207 | } | |
86886e55 | 1208 | |
89af7ba5 | 1209 | if (!error) |
e3920fb4 | 1210 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1211 | else |
84117da5 | 1212 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1213 | |
1214 | /* | |
1215 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
1216 | * this even in case of failure as all disable_nonboot_cpus() users are | |
1217 | * supposed to do enable_nonboot_cpus() on the failure path. | |
1218 | */ | |
1219 | cpu_hotplug_disabled++; | |
1220 | ||
d221938c | 1221 | cpu_maps_update_done(); |
e3920fb4 RW |
1222 | return error; |
1223 | } | |
1224 | ||
d0af9eed SS |
1225 | void __weak arch_enable_nonboot_cpus_begin(void) |
1226 | { | |
1227 | } | |
1228 | ||
1229 | void __weak arch_enable_nonboot_cpus_end(void) | |
1230 | { | |
1231 | } | |
1232 | ||
71cf5aee | 1233 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
1234 | { |
1235 | int cpu, error; | |
1236 | ||
1237 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1238 | cpu_maps_update_begin(); |
01b41159 | 1239 | __cpu_hotplug_enable(); |
e0b582ec | 1240 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1241 | goto out; |
e3920fb4 | 1242 | |
84117da5 | 1243 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
1244 | |
1245 | arch_enable_nonboot_cpus_begin(); | |
1246 | ||
e0b582ec | 1247 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1248 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1249 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1250 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1251 | if (!error) { |
84117da5 | 1252 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1253 | continue; |
1254 | } | |
84117da5 | 1255 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1256 | } |
d0af9eed SS |
1257 | |
1258 | arch_enable_nonboot_cpus_end(); | |
1259 | ||
e0b582ec | 1260 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1261 | out: |
d221938c | 1262 | cpu_maps_update_done(); |
1da177e4 | 1263 | } |
e0b582ec | 1264 | |
d7268a31 | 1265 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1266 | { |
1267 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1268 | return -ENOMEM; | |
1269 | return 0; | |
1270 | } | |
1271 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1272 | |
79cfbdfa SB |
1273 | /* |
1274 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1275 | * ensure that the state of the system with respect to the tasks being frozen | |
1276 | * or not, as reported by the notification, remains unchanged *throughout the | |
1277 | * duration* of the execution of the callbacks. | |
1278 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1279 | * | |
1280 | * This synchronization is implemented by mutually excluding regular CPU | |
1281 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1282 | * Hibernate notifications. | |
1283 | */ | |
1284 | static int | |
1285 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1286 | unsigned long action, void *ptr) | |
1287 | { | |
1288 | switch (action) { | |
1289 | ||
1290 | case PM_SUSPEND_PREPARE: | |
1291 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1292 | cpu_hotplug_disable(); |
79cfbdfa SB |
1293 | break; |
1294 | ||
1295 | case PM_POST_SUSPEND: | |
1296 | case PM_POST_HIBERNATION: | |
16e53dbf | 1297 | cpu_hotplug_enable(); |
79cfbdfa SB |
1298 | break; |
1299 | ||
1300 | default: | |
1301 | return NOTIFY_DONE; | |
1302 | } | |
1303 | ||
1304 | return NOTIFY_OK; | |
1305 | } | |
1306 | ||
1307 | ||
d7268a31 | 1308 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1309 | { |
6e32d479 FY |
1310 | /* |
1311 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1312 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1313 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1314 | */ | |
79cfbdfa SB |
1315 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1316 | return 0; | |
1317 | } | |
1318 | core_initcall(cpu_hotplug_pm_sync_init); | |
1319 | ||
f3de4be9 | 1320 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1321 | |
8ce371f9 PZ |
1322 | int __boot_cpu_id; |
1323 | ||
68f4f1ec | 1324 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1325 | |
cff7d378 | 1326 | /* Boot processor state steps */ |
17a2f1ce | 1327 | static struct cpuhp_step cpuhp_hp_states[] = { |
cff7d378 TG |
1328 | [CPUHP_OFFLINE] = { |
1329 | .name = "offline", | |
3c1627e9 TG |
1330 | .startup.single = NULL, |
1331 | .teardown.single = NULL, | |
cff7d378 TG |
1332 | }, |
1333 | #ifdef CONFIG_SMP | |
1334 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1335 | .name = "threads:prepare", |
3c1627e9 TG |
1336 | .startup.single = smpboot_create_threads, |
1337 | .teardown.single = NULL, | |
757c989b | 1338 | .cant_stop = true, |
cff7d378 | 1339 | }, |
00e16c3d | 1340 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1341 | .name = "perf:prepare", |
1342 | .startup.single = perf_event_init_cpu, | |
1343 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1344 | }, |
7ee681b2 | 1345 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1346 | .name = "workqueue:prepare", |
1347 | .startup.single = workqueue_prepare_cpu, | |
1348 | .teardown.single = NULL, | |
7ee681b2 | 1349 | }, |
27590dc1 | 1350 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1351 | .name = "hrtimers:prepare", |
1352 | .startup.single = hrtimers_prepare_cpu, | |
1353 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1354 | }, |
31487f83 | 1355 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1356 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1357 | .startup.single = smpcfd_prepare_cpu, |
1358 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1359 | }, |
e6d4989a RW |
1360 | [CPUHP_RELAY_PREPARE] = { |
1361 | .name = "relay:prepare", | |
1362 | .startup.single = relay_prepare_cpu, | |
1363 | .teardown.single = NULL, | |
1364 | }, | |
6731d4f1 SAS |
1365 | [CPUHP_SLAB_PREPARE] = { |
1366 | .name = "slab:prepare", | |
1367 | .startup.single = slab_prepare_cpu, | |
1368 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1369 | }, |
4df83742 | 1370 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1371 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1372 | .startup.single = rcutree_prepare_cpu, |
1373 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1374 | }, |
4fae16df RC |
1375 | /* |
1376 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1377 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1378 | * otherwise a RCU stall occurs. | |
1379 | */ | |
26456f87 | 1380 | [CPUHP_TIMERS_PREPARE] = { |
d018031f | 1381 | .name = "timers:prepare", |
26456f87 | 1382 | .startup.single = timers_prepare_cpu, |
3c1627e9 | 1383 | .teardown.single = timers_dead_cpu, |
4fae16df | 1384 | }, |
d10ef6f9 | 1385 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1386 | [CPUHP_BRINGUP_CPU] = { |
1387 | .name = "cpu:bringup", | |
3c1627e9 TG |
1388 | .startup.single = bringup_cpu, |
1389 | .teardown.single = NULL, | |
757c989b | 1390 | .cant_stop = true, |
4baa0afc | 1391 | }, |
d10ef6f9 TG |
1392 | /* Final state before CPU kills itself */ |
1393 | [CPUHP_AP_IDLE_DEAD] = { | |
1394 | .name = "idle:dead", | |
1395 | }, | |
1396 | /* | |
1397 | * Last state before CPU enters the idle loop to die. Transient state | |
1398 | * for synchronization. | |
1399 | */ | |
1400 | [CPUHP_AP_OFFLINE] = { | |
1401 | .name = "ap:offline", | |
1402 | .cant_stop = true, | |
1403 | }, | |
9cf7243d TG |
1404 | /* First state is scheduler control. Interrupts are disabled */ |
1405 | [CPUHP_AP_SCHED_STARTING] = { | |
1406 | .name = "sched:starting", | |
3c1627e9 TG |
1407 | .startup.single = sched_cpu_starting, |
1408 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1409 | }, |
4df83742 | 1410 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1411 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1412 | .startup.single = NULL, |
1413 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1414 | }, |
46febd37 LJ |
1415 | [CPUHP_AP_SMPCFD_DYING] = { |
1416 | .name = "smpcfd:dying", | |
1417 | .startup.single = NULL, | |
1418 | .teardown.single = smpcfd_dying_cpu, | |
1419 | }, | |
d10ef6f9 TG |
1420 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1421 | * state for synchronsization */ | |
1422 | [CPUHP_AP_ONLINE] = { | |
1423 | .name = "ap:online", | |
1424 | }, | |
17a2f1ce LJ |
1425 | /* |
1426 | * Handled on controll processor until the plugged processor manages | |
1427 | * this itself. | |
1428 | */ | |
1429 | [CPUHP_TEARDOWN_CPU] = { | |
1430 | .name = "cpu:teardown", | |
1431 | .startup.single = NULL, | |
1432 | .teardown.single = takedown_cpu, | |
1433 | .cant_stop = true, | |
1434 | }, | |
d10ef6f9 | 1435 | /* Handle smpboot threads park/unpark */ |
1cf4f629 | 1436 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1437 | .name = "smpboot/threads:online", |
3c1627e9 | 1438 | .startup.single = smpboot_unpark_threads, |
c4de6569 | 1439 | .teardown.single = smpboot_park_threads, |
1cf4f629 | 1440 | }, |
c5cb83bb TG |
1441 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1442 | .name = "irq/affinity:online", | |
1443 | .startup.single = irq_affinity_online_cpu, | |
1444 | .teardown.single = NULL, | |
1445 | }, | |
00e16c3d | 1446 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1447 | .name = "perf:online", |
1448 | .startup.single = perf_event_init_cpu, | |
1449 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1450 | }, |
9cf57731 PZ |
1451 | [CPUHP_AP_WATCHDOG_ONLINE] = { |
1452 | .name = "lockup_detector:online", | |
1453 | .startup.single = lockup_detector_online_cpu, | |
1454 | .teardown.single = lockup_detector_offline_cpu, | |
1455 | }, | |
7ee681b2 | 1456 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1457 | .name = "workqueue:online", |
1458 | .startup.single = workqueue_online_cpu, | |
1459 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1460 | }, |
4df83742 | 1461 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1462 | .name = "RCU/tree:online", |
3c1627e9 TG |
1463 | .startup.single = rcutree_online_cpu, |
1464 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1465 | }, |
4baa0afc | 1466 | #endif |
d10ef6f9 TG |
1467 | /* |
1468 | * The dynamically registered state space is here | |
1469 | */ | |
1470 | ||
aaddd7d1 TG |
1471 | #ifdef CONFIG_SMP |
1472 | /* Last state is scheduler control setting the cpu active */ | |
1473 | [CPUHP_AP_ACTIVE] = { | |
1474 | .name = "sched:active", | |
3c1627e9 TG |
1475 | .startup.single = sched_cpu_activate, |
1476 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1477 | }, |
1478 | #endif | |
1479 | ||
d10ef6f9 | 1480 | /* CPU is fully up and running. */ |
4baa0afc TG |
1481 | [CPUHP_ONLINE] = { |
1482 | .name = "online", | |
3c1627e9 TG |
1483 | .startup.single = NULL, |
1484 | .teardown.single = NULL, | |
4baa0afc TG |
1485 | }, |
1486 | }; | |
1487 | ||
5b7aa87e TG |
1488 | /* Sanity check for callbacks */ |
1489 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1490 | { | |
1491 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1492 | return -EINVAL; | |
1493 | return 0; | |
1494 | } | |
1495 | ||
dc280d93 TG |
1496 | /* |
1497 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1498 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1499 | * by having no name assigned. | |
1500 | */ | |
1501 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1502 | { | |
4205e478 TG |
1503 | enum cpuhp_state i, end; |
1504 | struct cpuhp_step *step; | |
dc280d93 | 1505 | |
4205e478 TG |
1506 | switch (state) { |
1507 | case CPUHP_AP_ONLINE_DYN: | |
17a2f1ce | 1508 | step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN; |
4205e478 TG |
1509 | end = CPUHP_AP_ONLINE_DYN_END; |
1510 | break; | |
1511 | case CPUHP_BP_PREPARE_DYN: | |
17a2f1ce | 1512 | step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN; |
4205e478 TG |
1513 | end = CPUHP_BP_PREPARE_DYN_END; |
1514 | break; | |
1515 | default: | |
1516 | return -EINVAL; | |
1517 | } | |
1518 | ||
1519 | for (i = state; i <= end; i++, step++) { | |
1520 | if (!step->name) | |
dc280d93 TG |
1521 | return i; |
1522 | } | |
1523 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1524 | return -ENOSPC; | |
1525 | } | |
1526 | ||
1527 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1528 | int (*startup)(unsigned int cpu), | |
1529 | int (*teardown)(unsigned int cpu), | |
1530 | bool multi_instance) | |
5b7aa87e TG |
1531 | { |
1532 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1533 | struct cpuhp_step *sp; | |
dc280d93 | 1534 | int ret = 0; |
5b7aa87e | 1535 | |
0c96b273 EB |
1536 | /* |
1537 | * If name is NULL, then the state gets removed. | |
1538 | * | |
1539 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
1540 | * the first allocation from these dynamic ranges, so the removal | |
1541 | * would trigger a new allocation and clear the wrong (already | |
1542 | * empty) state, leaving the callbacks of the to be cleared state | |
1543 | * dangling, which causes wreckage on the next hotplug operation. | |
1544 | */ | |
1545 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
1546 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
1547 | ret = cpuhp_reserve_state(state); |
1548 | if (ret < 0) | |
dc434e05 | 1549 | return ret; |
dc280d93 TG |
1550 | state = ret; |
1551 | } | |
5b7aa87e | 1552 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1553 | if (name && sp->name) |
1554 | return -EBUSY; | |
1555 | ||
3c1627e9 TG |
1556 | sp->startup.single = startup; |
1557 | sp->teardown.single = teardown; | |
5b7aa87e | 1558 | sp->name = name; |
cf392d10 TG |
1559 | sp->multi_instance = multi_instance; |
1560 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1561 | return ret; |
5b7aa87e TG |
1562 | } |
1563 | ||
1564 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1565 | { | |
3c1627e9 | 1566 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1567 | } |
1568 | ||
5b7aa87e TG |
1569 | /* |
1570 | * Call the startup/teardown function for a step either on the AP or | |
1571 | * on the current CPU. | |
1572 | */ | |
cf392d10 TG |
1573 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1574 | struct hlist_node *node) | |
5b7aa87e | 1575 | { |
a724632c | 1576 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1577 | int ret; |
1578 | ||
4dddfb5f PZ |
1579 | /* |
1580 | * If there's nothing to do, we done. | |
1581 | * Relies on the union for multi_instance. | |
1582 | */ | |
3c1627e9 TG |
1583 | if ((bringup && !sp->startup.single) || |
1584 | (!bringup && !sp->teardown.single)) | |
5b7aa87e | 1585 | return 0; |
5b7aa87e TG |
1586 | /* |
1587 | * The non AP bound callbacks can fail on bringup. On teardown | |
1588 | * e.g. module removal we crash for now. | |
1589 | */ | |
1cf4f629 TG |
1590 | #ifdef CONFIG_SMP |
1591 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1592 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1593 | else |
96abb968 | 1594 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1595 | #else |
96abb968 | 1596 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1597 | #endif |
5b7aa87e TG |
1598 | BUG_ON(ret && !bringup); |
1599 | return ret; | |
1600 | } | |
1601 | ||
1602 | /* | |
1603 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1604 | * | |
1605 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1606 | */ | |
1607 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1608 | struct hlist_node *node) |
5b7aa87e TG |
1609 | { |
1610 | int cpu; | |
1611 | ||
5b7aa87e TG |
1612 | /* Roll back the already executed steps on the other cpus */ |
1613 | for_each_present_cpu(cpu) { | |
1614 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1615 | int cpustate = st->state; | |
1616 | ||
1617 | if (cpu >= failedcpu) | |
1618 | break; | |
1619 | ||
1620 | /* Did we invoke the startup call on that cpu ? */ | |
1621 | if (cpustate >= state) | |
cf392d10 | 1622 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1623 | } |
1624 | } | |
1625 | ||
9805c673 TG |
1626 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1627 | struct hlist_node *node, | |
1628 | bool invoke) | |
cf392d10 TG |
1629 | { |
1630 | struct cpuhp_step *sp; | |
1631 | int cpu; | |
1632 | int ret; | |
1633 | ||
9805c673 TG |
1634 | lockdep_assert_cpus_held(); |
1635 | ||
cf392d10 TG |
1636 | sp = cpuhp_get_step(state); |
1637 | if (sp->multi_instance == false) | |
1638 | return -EINVAL; | |
1639 | ||
dc434e05 | 1640 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1641 | |
3c1627e9 | 1642 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1643 | goto add_node; |
1644 | ||
1645 | /* | |
1646 | * Try to call the startup callback for each present cpu | |
1647 | * depending on the hotplug state of the cpu. | |
1648 | */ | |
1649 | for_each_present_cpu(cpu) { | |
1650 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1651 | int cpustate = st->state; | |
1652 | ||
1653 | if (cpustate < state) | |
1654 | continue; | |
1655 | ||
1656 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1657 | if (ret) { | |
3c1627e9 | 1658 | if (sp->teardown.multi) |
cf392d10 | 1659 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1660 | goto unlock; |
cf392d10 TG |
1661 | } |
1662 | } | |
1663 | add_node: | |
1664 | ret = 0; | |
cf392d10 | 1665 | hlist_add_head(node, &sp->list); |
dc434e05 | 1666 | unlock: |
cf392d10 | 1667 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1668 | return ret; |
1669 | } | |
1670 | ||
1671 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1672 | bool invoke) | |
1673 | { | |
1674 | int ret; | |
1675 | ||
1676 | cpus_read_lock(); | |
1677 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1678 | cpus_read_unlock(); |
cf392d10 TG |
1679 | return ret; |
1680 | } | |
1681 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1682 | ||
5b7aa87e | 1683 | /** |
71def423 | 1684 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1685 | * @state: The state to setup |
1686 | * @invoke: If true, the startup function is invoked for cpus where | |
1687 | * cpu state >= @state | |
1688 | * @startup: startup callback function | |
1689 | * @teardown: teardown callback function | |
1690 | * @multi_instance: State is set up for multiple instances which get | |
1691 | * added afterwards. | |
5b7aa87e | 1692 | * |
71def423 | 1693 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1694 | * Returns: |
1695 | * On success: | |
1696 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1697 | * 0 for all other states | |
1698 | * On failure: proper (negative) error code | |
5b7aa87e | 1699 | */ |
71def423 SAS |
1700 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1701 | const char *name, bool invoke, | |
1702 | int (*startup)(unsigned int cpu), | |
1703 | int (*teardown)(unsigned int cpu), | |
1704 | bool multi_instance) | |
5b7aa87e TG |
1705 | { |
1706 | int cpu, ret = 0; | |
b9d9d691 | 1707 | bool dynstate; |
5b7aa87e | 1708 | |
71def423 SAS |
1709 | lockdep_assert_cpus_held(); |
1710 | ||
5b7aa87e TG |
1711 | if (cpuhp_cb_check(state) || !name) |
1712 | return -EINVAL; | |
1713 | ||
dc434e05 | 1714 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1715 | |
dc280d93 TG |
1716 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1717 | multi_instance); | |
5b7aa87e | 1718 | |
b9d9d691 TG |
1719 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1720 | if (ret > 0 && dynstate) { | |
1721 | state = ret; | |
1722 | ret = 0; | |
1723 | } | |
1724 | ||
dc280d93 | 1725 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1726 | goto out; |
1727 | ||
1728 | /* | |
1729 | * Try to call the startup callback for each present cpu | |
1730 | * depending on the hotplug state of the cpu. | |
1731 | */ | |
1732 | for_each_present_cpu(cpu) { | |
1733 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1734 | int cpustate = st->state; | |
1735 | ||
1736 | if (cpustate < state) | |
1737 | continue; | |
1738 | ||
cf392d10 | 1739 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1740 | if (ret) { |
a724632c | 1741 | if (teardown) |
cf392d10 TG |
1742 | cpuhp_rollback_install(cpu, state, NULL); |
1743 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1744 | goto out; |
1745 | } | |
1746 | } | |
1747 | out: | |
dc434e05 | 1748 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1749 | /* |
1750 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1751 | * dynamically allocated state in case of success. | |
1752 | */ | |
b9d9d691 | 1753 | if (!ret && dynstate) |
5b7aa87e TG |
1754 | return state; |
1755 | return ret; | |
1756 | } | |
71def423 SAS |
1757 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
1758 | ||
1759 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1760 | const char *name, bool invoke, | |
1761 | int (*startup)(unsigned int cpu), | |
1762 | int (*teardown)(unsigned int cpu), | |
1763 | bool multi_instance) | |
1764 | { | |
1765 | int ret; | |
1766 | ||
1767 | cpus_read_lock(); | |
1768 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
1769 | teardown, multi_instance); | |
1770 | cpus_read_unlock(); | |
1771 | return ret; | |
1772 | } | |
5b7aa87e TG |
1773 | EXPORT_SYMBOL(__cpuhp_setup_state); |
1774 | ||
cf392d10 TG |
1775 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
1776 | struct hlist_node *node, bool invoke) | |
1777 | { | |
1778 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
1779 | int cpu; | |
1780 | ||
1781 | BUG_ON(cpuhp_cb_check(state)); | |
1782 | ||
1783 | if (!sp->multi_instance) | |
1784 | return -EINVAL; | |
1785 | ||
8f553c49 | 1786 | cpus_read_lock(); |
dc434e05 SAS |
1787 | mutex_lock(&cpuhp_state_mutex); |
1788 | ||
cf392d10 TG |
1789 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
1790 | goto remove; | |
1791 | /* | |
1792 | * Call the teardown callback for each present cpu depending | |
1793 | * on the hotplug state of the cpu. This function is not | |
1794 | * allowed to fail currently! | |
1795 | */ | |
1796 | for_each_present_cpu(cpu) { | |
1797 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1798 | int cpustate = st->state; | |
1799 | ||
1800 | if (cpustate >= state) | |
1801 | cpuhp_issue_call(cpu, state, false, node); | |
1802 | } | |
1803 | ||
1804 | remove: | |
cf392d10 TG |
1805 | hlist_del(node); |
1806 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 1807 | cpus_read_unlock(); |
cf392d10 TG |
1808 | |
1809 | return 0; | |
1810 | } | |
1811 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 1812 | |
5b7aa87e | 1813 | /** |
71def423 | 1814 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
1815 | * @state: The state to remove |
1816 | * @invoke: If true, the teardown function is invoked for cpus where | |
1817 | * cpu state >= @state | |
1818 | * | |
71def423 | 1819 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
1820 | * The teardown callback is currently not allowed to fail. Think |
1821 | * about module removal! | |
1822 | */ | |
71def423 | 1823 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 1824 | { |
cf392d10 | 1825 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1826 | int cpu; |
1827 | ||
1828 | BUG_ON(cpuhp_cb_check(state)); | |
1829 | ||
71def423 | 1830 | lockdep_assert_cpus_held(); |
5b7aa87e | 1831 | |
dc434e05 | 1832 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
1833 | if (sp->multi_instance) { |
1834 | WARN(!hlist_empty(&sp->list), | |
1835 | "Error: Removing state %d which has instances left.\n", | |
1836 | state); | |
1837 | goto remove; | |
1838 | } | |
1839 | ||
a724632c | 1840 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
1841 | goto remove; |
1842 | ||
1843 | /* | |
1844 | * Call the teardown callback for each present cpu depending | |
1845 | * on the hotplug state of the cpu. This function is not | |
1846 | * allowed to fail currently! | |
1847 | */ | |
1848 | for_each_present_cpu(cpu) { | |
1849 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1850 | int cpustate = st->state; | |
1851 | ||
1852 | if (cpustate >= state) | |
cf392d10 | 1853 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
1854 | } |
1855 | remove: | |
cf392d10 | 1856 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 1857 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
1858 | } |
1859 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
1860 | ||
1861 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1862 | { | |
1863 | cpus_read_lock(); | |
1864 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 1865 | cpus_read_unlock(); |
5b7aa87e TG |
1866 | } |
1867 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1868 | ||
98f8cdce TG |
1869 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1870 | static ssize_t show_cpuhp_state(struct device *dev, | |
1871 | struct device_attribute *attr, char *buf) | |
1872 | { | |
1873 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1874 | ||
1875 | return sprintf(buf, "%d\n", st->state); | |
1876 | } | |
1877 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1878 | ||
757c989b TG |
1879 | static ssize_t write_cpuhp_target(struct device *dev, |
1880 | struct device_attribute *attr, | |
1881 | const char *buf, size_t count) | |
1882 | { | |
1883 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1884 | struct cpuhp_step *sp; | |
1885 | int target, ret; | |
1886 | ||
1887 | ret = kstrtoint(buf, 10, &target); | |
1888 | if (ret) | |
1889 | return ret; | |
1890 | ||
1891 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1892 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1893 | return -EINVAL; | |
1894 | #else | |
1895 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1896 | return -EINVAL; | |
1897 | #endif | |
1898 | ||
1899 | ret = lock_device_hotplug_sysfs(); | |
1900 | if (ret) | |
1901 | return ret; | |
1902 | ||
1903 | mutex_lock(&cpuhp_state_mutex); | |
1904 | sp = cpuhp_get_step(target); | |
1905 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1906 | mutex_unlock(&cpuhp_state_mutex); | |
1907 | if (ret) | |
40da1b11 | 1908 | goto out; |
757c989b TG |
1909 | |
1910 | if (st->state < target) | |
1911 | ret = do_cpu_up(dev->id, target); | |
1912 | else | |
1913 | ret = do_cpu_down(dev->id, target); | |
40da1b11 | 1914 | out: |
757c989b TG |
1915 | unlock_device_hotplug(); |
1916 | return ret ? ret : count; | |
1917 | } | |
1918 | ||
98f8cdce TG |
1919 | static ssize_t show_cpuhp_target(struct device *dev, |
1920 | struct device_attribute *attr, char *buf) | |
1921 | { | |
1922 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1923 | ||
1924 | return sprintf(buf, "%d\n", st->target); | |
1925 | } | |
757c989b | 1926 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce | 1927 | |
1db49484 PZ |
1928 | |
1929 | static ssize_t write_cpuhp_fail(struct device *dev, | |
1930 | struct device_attribute *attr, | |
1931 | const char *buf, size_t count) | |
1932 | { | |
1933 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1934 | struct cpuhp_step *sp; | |
1935 | int fail, ret; | |
1936 | ||
1937 | ret = kstrtoint(buf, 10, &fail); | |
1938 | if (ret) | |
1939 | return ret; | |
1940 | ||
1941 | /* | |
1942 | * Cannot fail STARTING/DYING callbacks. | |
1943 | */ | |
1944 | if (cpuhp_is_atomic_state(fail)) | |
1945 | return -EINVAL; | |
1946 | ||
1947 | /* | |
1948 | * Cannot fail anything that doesn't have callbacks. | |
1949 | */ | |
1950 | mutex_lock(&cpuhp_state_mutex); | |
1951 | sp = cpuhp_get_step(fail); | |
1952 | if (!sp->startup.single && !sp->teardown.single) | |
1953 | ret = -EINVAL; | |
1954 | mutex_unlock(&cpuhp_state_mutex); | |
1955 | if (ret) | |
1956 | return ret; | |
1957 | ||
1958 | st->fail = fail; | |
1959 | ||
1960 | return count; | |
1961 | } | |
1962 | ||
1963 | static ssize_t show_cpuhp_fail(struct device *dev, | |
1964 | struct device_attribute *attr, char *buf) | |
1965 | { | |
1966 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1967 | ||
1968 | return sprintf(buf, "%d\n", st->fail); | |
1969 | } | |
1970 | ||
1971 | static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail); | |
1972 | ||
98f8cdce TG |
1973 | static struct attribute *cpuhp_cpu_attrs[] = { |
1974 | &dev_attr_state.attr, | |
1975 | &dev_attr_target.attr, | |
1db49484 | 1976 | &dev_attr_fail.attr, |
98f8cdce TG |
1977 | NULL |
1978 | }; | |
1979 | ||
993647a2 | 1980 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
1981 | .attrs = cpuhp_cpu_attrs, |
1982 | .name = "hotplug", | |
1983 | NULL | |
1984 | }; | |
1985 | ||
1986 | static ssize_t show_cpuhp_states(struct device *dev, | |
1987 | struct device_attribute *attr, char *buf) | |
1988 | { | |
1989 | ssize_t cur, res = 0; | |
1990 | int i; | |
1991 | ||
1992 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1993 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1994 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1995 | ||
1996 | if (sp->name) { | |
1997 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1998 | buf += cur; | |
1999 | res += cur; | |
2000 | } | |
2001 | } | |
2002 | mutex_unlock(&cpuhp_state_mutex); | |
2003 | return res; | |
2004 | } | |
2005 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
2006 | ||
2007 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
2008 | &dev_attr_states.attr, | |
2009 | NULL | |
2010 | }; | |
2011 | ||
993647a2 | 2012 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
2013 | .attrs = cpuhp_cpu_root_attrs, |
2014 | .name = "hotplug", | |
2015 | NULL | |
2016 | }; | |
2017 | ||
05736e4a TG |
2018 | #ifdef CONFIG_HOTPLUG_SMT |
2019 | ||
2020 | static const char *smt_states[] = { | |
2021 | [CPU_SMT_ENABLED] = "on", | |
2022 | [CPU_SMT_DISABLED] = "off", | |
2023 | [CPU_SMT_FORCE_DISABLED] = "forceoff", | |
2024 | [CPU_SMT_NOT_SUPPORTED] = "notsupported", | |
2025 | }; | |
2026 | ||
2027 | static ssize_t | |
2028 | show_smt_control(struct device *dev, struct device_attribute *attr, char *buf) | |
2029 | { | |
2030 | return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]); | |
2031 | } | |
2032 | ||
2033 | static void cpuhp_offline_cpu_device(unsigned int cpu) | |
2034 | { | |
2035 | struct device *dev = get_cpu_device(cpu); | |
2036 | ||
2037 | dev->offline = true; | |
2038 | /* Tell user space about the state change */ | |
2039 | kobject_uevent(&dev->kobj, KOBJ_OFFLINE); | |
2040 | } | |
2041 | ||
215af549 TG |
2042 | static void cpuhp_online_cpu_device(unsigned int cpu) |
2043 | { | |
2044 | struct device *dev = get_cpu_device(cpu); | |
2045 | ||
2046 | dev->offline = false; | |
2047 | /* Tell user space about the state change */ | |
2048 | kobject_uevent(&dev->kobj, KOBJ_ONLINE); | |
2049 | } | |
2050 | ||
05736e4a TG |
2051 | static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval) |
2052 | { | |
2053 | int cpu, ret = 0; | |
2054 | ||
2055 | cpu_maps_update_begin(); | |
2056 | for_each_online_cpu(cpu) { | |
2057 | if (topology_is_primary_thread(cpu)) | |
2058 | continue; | |
2059 | ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); | |
2060 | if (ret) | |
2061 | break; | |
2062 | /* | |
2063 | * As this needs to hold the cpu maps lock it's impossible | |
2064 | * to call device_offline() because that ends up calling | |
2065 | * cpu_down() which takes cpu maps lock. cpu maps lock | |
2066 | * needs to be held as this might race against in kernel | |
2067 | * abusers of the hotplug machinery (thermal management). | |
2068 | * | |
2069 | * So nothing would update device:offline state. That would | |
2070 | * leave the sysfs entry stale and prevent onlining after | |
2071 | * smt control has been changed to 'off' again. This is | |
2072 | * called under the sysfs hotplug lock, so it is properly | |
2073 | * serialized against the regular offline usage. | |
2074 | */ | |
2075 | cpuhp_offline_cpu_device(cpu); | |
2076 | } | |
34d66caf | 2077 | if (!ret) |
05736e4a TG |
2078 | cpu_smt_control = ctrlval; |
2079 | cpu_maps_update_done(); | |
2080 | return ret; | |
2081 | } | |
2082 | ||
215af549 | 2083 | static int cpuhp_smt_enable(void) |
05736e4a | 2084 | { |
215af549 TG |
2085 | int cpu, ret = 0; |
2086 | ||
05736e4a TG |
2087 | cpu_maps_update_begin(); |
2088 | cpu_smt_control = CPU_SMT_ENABLED; | |
215af549 TG |
2089 | for_each_present_cpu(cpu) { |
2090 | /* Skip online CPUs and CPUs on offline nodes */ | |
2091 | if (cpu_online(cpu) || !node_online(cpu_to_node(cpu))) | |
2092 | continue; | |
2093 | ret = _cpu_up(cpu, 0, CPUHP_ONLINE); | |
2094 | if (ret) | |
2095 | break; | |
2096 | /* See comment in cpuhp_smt_disable() */ | |
2097 | cpuhp_online_cpu_device(cpu); | |
2098 | } | |
05736e4a | 2099 | cpu_maps_update_done(); |
215af549 | 2100 | return ret; |
05736e4a TG |
2101 | } |
2102 | ||
2103 | static ssize_t | |
2104 | store_smt_control(struct device *dev, struct device_attribute *attr, | |
2105 | const char *buf, size_t count) | |
2106 | { | |
2107 | int ctrlval, ret; | |
2108 | ||
2109 | if (sysfs_streq(buf, "on")) | |
2110 | ctrlval = CPU_SMT_ENABLED; | |
2111 | else if (sysfs_streq(buf, "off")) | |
2112 | ctrlval = CPU_SMT_DISABLED; | |
2113 | else if (sysfs_streq(buf, "forceoff")) | |
2114 | ctrlval = CPU_SMT_FORCE_DISABLED; | |
2115 | else | |
2116 | return -EINVAL; | |
2117 | ||
2118 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED) | |
2119 | return -EPERM; | |
2120 | ||
2121 | if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
2122 | return -ENODEV; | |
2123 | ||
2124 | ret = lock_device_hotplug_sysfs(); | |
2125 | if (ret) | |
2126 | return ret; | |
2127 | ||
2128 | if (ctrlval != cpu_smt_control) { | |
2129 | switch (ctrlval) { | |
2130 | case CPU_SMT_ENABLED: | |
215af549 | 2131 | ret = cpuhp_smt_enable(); |
05736e4a TG |
2132 | break; |
2133 | case CPU_SMT_DISABLED: | |
2134 | case CPU_SMT_FORCE_DISABLED: | |
2135 | ret = cpuhp_smt_disable(ctrlval); | |
2136 | break; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | unlock_device_hotplug(); | |
2141 | return ret ? ret : count; | |
2142 | } | |
2143 | static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control); | |
2144 | ||
2145 | static ssize_t | |
2146 | show_smt_active(struct device *dev, struct device_attribute *attr, char *buf) | |
2147 | { | |
2148 | bool active = topology_max_smt_threads() > 1; | |
2149 | ||
2150 | return snprintf(buf, PAGE_SIZE - 2, "%d\n", active); | |
2151 | } | |
2152 | static DEVICE_ATTR(active, 0444, show_smt_active, NULL); | |
2153 | ||
2154 | static struct attribute *cpuhp_smt_attrs[] = { | |
2155 | &dev_attr_control.attr, | |
2156 | &dev_attr_active.attr, | |
2157 | NULL | |
2158 | }; | |
2159 | ||
2160 | static const struct attribute_group cpuhp_smt_attr_group = { | |
2161 | .attrs = cpuhp_smt_attrs, | |
2162 | .name = "smt", | |
2163 | NULL | |
2164 | }; | |
2165 | ||
2166 | static int __init cpu_smt_state_init(void) | |
2167 | { | |
05736e4a TG |
2168 | return sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2169 | &cpuhp_smt_attr_group); | |
2170 | } | |
2171 | ||
2172 | #else | |
2173 | static inline int cpu_smt_state_init(void) { return 0; } | |
2174 | #endif | |
2175 | ||
98f8cdce TG |
2176 | static int __init cpuhp_sysfs_init(void) |
2177 | { | |
2178 | int cpu, ret; | |
2179 | ||
05736e4a TG |
2180 | ret = cpu_smt_state_init(); |
2181 | if (ret) | |
2182 | return ret; | |
2183 | ||
98f8cdce TG |
2184 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2185 | &cpuhp_cpu_root_attr_group); | |
2186 | if (ret) | |
2187 | return ret; | |
2188 | ||
2189 | for_each_possible_cpu(cpu) { | |
2190 | struct device *dev = get_cpu_device(cpu); | |
2191 | ||
2192 | if (!dev) | |
2193 | continue; | |
2194 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
2195 | if (ret) | |
2196 | return ret; | |
2197 | } | |
2198 | return 0; | |
2199 | } | |
2200 | device_initcall(cpuhp_sysfs_init); | |
2201 | #endif | |
2202 | ||
e56b3bc7 LT |
2203 | /* |
2204 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
2205 | * represents all NR_CPUS bits binary values of 1<<nr. | |
2206 | * | |
e0b582ec | 2207 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
2208 | * mask value that has a single bit set only. |
2209 | */ | |
b8d317d1 | 2210 | |
e56b3bc7 | 2211 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 2212 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
2213 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
2214 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
2215 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 2216 | |
e56b3bc7 LT |
2217 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
2218 | ||
2219 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
2220 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
2221 | #if BITS_PER_LONG > 32 | |
2222 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
2223 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
2224 | #endif |
2225 | }; | |
e56b3bc7 | 2226 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
2227 | |
2228 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
2229 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
2230 | |
2231 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 2232 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 2233 | = {CPU_BITS_ALL}; |
b3199c02 | 2234 | #else |
4b804c85 | 2235 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 2236 | #endif |
4b804c85 | 2237 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 2238 | |
4b804c85 RV |
2239 | struct cpumask __cpu_online_mask __read_mostly; |
2240 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 2241 | |
4b804c85 RV |
2242 | struct cpumask __cpu_present_mask __read_mostly; |
2243 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 2244 | |
4b804c85 RV |
2245 | struct cpumask __cpu_active_mask __read_mostly; |
2246 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 2247 | |
3fa41520 RR |
2248 | void init_cpu_present(const struct cpumask *src) |
2249 | { | |
c4c54dd1 | 2250 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
2251 | } |
2252 | ||
2253 | void init_cpu_possible(const struct cpumask *src) | |
2254 | { | |
c4c54dd1 | 2255 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
2256 | } |
2257 | ||
2258 | void init_cpu_online(const struct cpumask *src) | |
2259 | { | |
c4c54dd1 | 2260 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 2261 | } |
cff7d378 TG |
2262 | |
2263 | /* | |
2264 | * Activate the first processor. | |
2265 | */ | |
2266 | void __init boot_cpu_init(void) | |
2267 | { | |
2268 | int cpu = smp_processor_id(); | |
2269 | ||
2270 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
2271 | set_cpu_online(cpu, true); | |
2272 | set_cpu_active(cpu, true); | |
2273 | set_cpu_present(cpu, true); | |
2274 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
2275 | |
2276 | #ifdef CONFIG_SMP | |
2277 | __boot_cpu_id = cpu; | |
2278 | #endif | |
cff7d378 TG |
2279 | } |
2280 | ||
2281 | /* | |
2282 | * Must be called _AFTER_ setting up the per_cpu areas | |
2283 | */ | |
b5b1404d | 2284 | void __init boot_cpu_hotplug_init(void) |
cff7d378 | 2285 | { |
269777aa | 2286 | #ifdef CONFIG_SMP |
0cc3cd21 | 2287 | this_cpu_write(cpuhp_state.booted_once, true); |
269777aa | 2288 | #endif |
0cc3cd21 | 2289 | this_cpu_write(cpuhp_state.state, CPUHP_ONLINE); |
cff7d378 | 2290 | } |