]>
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 | */ | |
bf2c59fc | 6 | #include <linux/sched/mm.h> |
1da177e4 LT |
7 | #include <linux/proc_fs.h> |
8 | #include <linux/smp.h> | |
9 | #include <linux/init.h> | |
10 | #include <linux/notifier.h> | |
3f07c014 | 11 | #include <linux/sched/signal.h> |
ef8bd77f | 12 | #include <linux/sched/hotplug.h> |
9ca12ac0 | 13 | #include <linux/sched/isolation.h> |
29930025 | 14 | #include <linux/sched/task.h> |
a74cfffb | 15 | #include <linux/sched/smt.h> |
1da177e4 LT |
16 | #include <linux/unistd.h> |
17 | #include <linux/cpu.h> | |
cb79295e AV |
18 | #include <linux/oom.h> |
19 | #include <linux/rcupdate.h> | |
9984de1a | 20 | #include <linux/export.h> |
e4cc2f87 | 21 | #include <linux/bug.h> |
1da177e4 LT |
22 | #include <linux/kthread.h> |
23 | #include <linux/stop_machine.h> | |
81615b62 | 24 | #include <linux/mutex.h> |
5a0e3ad6 | 25 | #include <linux/gfp.h> |
79cfbdfa | 26 | #include <linux/suspend.h> |
a19423b9 | 27 | #include <linux/lockdep.h> |
345527b1 | 28 | #include <linux/tick.h> |
a8994181 | 29 | #include <linux/irq.h> |
941154bd | 30 | #include <linux/nmi.h> |
4cb28ced | 31 | #include <linux/smpboot.h> |
e6d4989a | 32 | #include <linux/relay.h> |
6731d4f1 | 33 | #include <linux/slab.h> |
fc8dffd3 | 34 | #include <linux/percpu-rwsem.h> |
cff7d378 | 35 | |
bb3632c6 | 36 | #include <trace/events/power.h> |
cff7d378 TG |
37 | #define CREATE_TRACE_POINTS |
38 | #include <trace/events/cpuhp.h> | |
1da177e4 | 39 | |
38498a67 TG |
40 | #include "smpboot.h" |
41 | ||
cff7d378 TG |
42 | /** |
43 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
44 | * @state: The current cpu state | |
45 | * @target: The target state | |
4cb28ced TG |
46 | * @thread: Pointer to the hotplug thread |
47 | * @should_run: Thread should execute | |
3b9d6da6 | 48 | * @rollback: Perform a rollback |
a724632c TG |
49 | * @single: Single callback invocation |
50 | * @bringup: Single callback bringup or teardown selector | |
51 | * @cb_state: The state for a single callback (install/uninstall) | |
4cb28ced | 52 | * @result: Result of the operation |
5ebe7742 PZ |
53 | * @done_up: Signal completion to the issuer of the task for cpu-up |
54 | * @done_down: Signal completion to the issuer of the task for cpu-down | |
cff7d378 TG |
55 | */ |
56 | struct cpuhp_cpu_state { | |
57 | enum cpuhp_state state; | |
58 | enum cpuhp_state target; | |
1db49484 | 59 | enum cpuhp_state fail; |
4cb28ced TG |
60 | #ifdef CONFIG_SMP |
61 | struct task_struct *thread; | |
62 | bool should_run; | |
3b9d6da6 | 63 | bool rollback; |
a724632c TG |
64 | bool single; |
65 | bool bringup; | |
2ea46c6f | 66 | int cpu; |
cf392d10 | 67 | struct hlist_node *node; |
4dddfb5f | 68 | struct hlist_node *last; |
4cb28ced | 69 | enum cpuhp_state cb_state; |
4cb28ced | 70 | int result; |
5ebe7742 PZ |
71 | struct completion done_up; |
72 | struct completion done_down; | |
4cb28ced | 73 | #endif |
cff7d378 TG |
74 | }; |
75 | ||
1db49484 PZ |
76 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = { |
77 | .fail = CPUHP_INVALID, | |
78 | }; | |
cff7d378 | 79 | |
e797bda3 TG |
80 | #ifdef CONFIG_SMP |
81 | cpumask_t cpus_booted_once_mask; | |
82 | #endif | |
83 | ||
49dfe2a6 | 84 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
85 | static struct lockdep_map cpuhp_state_up_map = |
86 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
87 | static struct lockdep_map cpuhp_state_down_map = | |
88 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
89 | ||
90 | ||
76dc6c09 | 91 | static inline void cpuhp_lock_acquire(bool bringup) |
5f4b55e1 PZ |
92 | { |
93 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
94 | } | |
95 | ||
76dc6c09 | 96 | static inline void cpuhp_lock_release(bool bringup) |
5f4b55e1 PZ |
97 | { |
98 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
99 | } | |
100 | #else | |
101 | ||
76dc6c09 MM |
102 | static inline void cpuhp_lock_acquire(bool bringup) { } |
103 | static inline void cpuhp_lock_release(bool bringup) { } | |
5f4b55e1 | 104 | |
49dfe2a6 TG |
105 | #endif |
106 | ||
cff7d378 TG |
107 | /** |
108 | * cpuhp_step - Hotplug state machine step | |
109 | * @name: Name of the step | |
110 | * @startup: Startup function of the step | |
111 | * @teardown: Teardown function of the step | |
757c989b | 112 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
113 | */ |
114 | struct cpuhp_step { | |
cf392d10 TG |
115 | const char *name; |
116 | union { | |
3c1627e9 TG |
117 | int (*single)(unsigned int cpu); |
118 | int (*multi)(unsigned int cpu, | |
119 | struct hlist_node *node); | |
120 | } startup; | |
cf392d10 | 121 | union { |
3c1627e9 TG |
122 | int (*single)(unsigned int cpu); |
123 | int (*multi)(unsigned int cpu, | |
124 | struct hlist_node *node); | |
125 | } teardown; | |
cf392d10 | 126 | struct hlist_head list; |
cf392d10 TG |
127 | bool cant_stop; |
128 | bool multi_instance; | |
cff7d378 TG |
129 | }; |
130 | ||
98f8cdce | 131 | static DEFINE_MUTEX(cpuhp_state_mutex); |
17a2f1ce | 132 | static struct cpuhp_step cpuhp_hp_states[]; |
cff7d378 | 133 | |
a724632c TG |
134 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) |
135 | { | |
17a2f1ce | 136 | return cpuhp_hp_states + state; |
a724632c TG |
137 | } |
138 | ||
453e4108 VD |
139 | static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step) |
140 | { | |
141 | return bringup ? !step->startup.single : !step->teardown.single; | |
142 | } | |
143 | ||
cff7d378 TG |
144 | /** |
145 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
146 | * @cpu: The cpu for which the callback should be invoked | |
96abb968 | 147 | * @state: The state to do callbacks for |
a724632c | 148 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
149 | * @node: For multi-instance, do a single entry callback for install/remove |
150 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 151 | * |
cf392d10 | 152 | * Called from cpu hotplug and from the state register machinery. |
cff7d378 | 153 | */ |
a724632c | 154 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
155 | bool bringup, struct hlist_node *node, |
156 | struct hlist_node **lastp) | |
cff7d378 TG |
157 | { |
158 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 159 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
160 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
161 | int (*cb)(unsigned int cpu); | |
162 | int ret, cnt; | |
163 | ||
1db49484 PZ |
164 | if (st->fail == state) { |
165 | st->fail = CPUHP_INVALID; | |
1db49484 PZ |
166 | return -EAGAIN; |
167 | } | |
168 | ||
453e4108 VD |
169 | if (cpuhp_step_empty(bringup, step)) { |
170 | WARN_ON_ONCE(1); | |
171 | return 0; | |
172 | } | |
173 | ||
cf392d10 | 174 | if (!step->multi_instance) { |
96abb968 | 175 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 176 | cb = bringup ? step->startup.single : step->teardown.single; |
453e4108 | 177 | |
a724632c | 178 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 179 | ret = cb(cpu); |
a724632c | 180 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
181 | return ret; |
182 | } | |
3c1627e9 | 183 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
184 | |
185 | /* Single invocation for instance add/remove */ | |
186 | if (node) { | |
96abb968 | 187 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
188 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
189 | ret = cbm(cpu, node); | |
190 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
191 | return ret; | |
192 | } | |
193 | ||
194 | /* State transition. Invoke on all instances */ | |
195 | cnt = 0; | |
196 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
197 | if (lastp && node == *lastp) |
198 | break; | |
199 | ||
cf392d10 TG |
200 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
201 | ret = cbm(cpu, node); | |
202 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
203 | if (ret) { |
204 | if (!lastp) | |
205 | goto err; | |
206 | ||
207 | *lastp = node; | |
208 | return ret; | |
209 | } | |
cf392d10 TG |
210 | cnt++; |
211 | } | |
96abb968 PZ |
212 | if (lastp) |
213 | *lastp = NULL; | |
cf392d10 TG |
214 | return 0; |
215 | err: | |
216 | /* Rollback the instances if one failed */ | |
3c1627e9 | 217 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
218 | if (!cbm) |
219 | return ret; | |
220 | ||
221 | hlist_for_each(node, &step->list) { | |
222 | if (!cnt--) | |
223 | break; | |
724a8688 PZ |
224 | |
225 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
226 | ret = cbm(cpu, node); | |
227 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
228 | /* | |
229 | * Rollback must not fail, | |
230 | */ | |
231 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
232 | } |
233 | return ret; | |
234 | } | |
235 | ||
98a79d6a | 236 | #ifdef CONFIG_SMP |
fcb3029a AB |
237 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
238 | { | |
239 | /* | |
240 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
241 | * purposes as that state is handled explicitly in cpu_down. | |
242 | */ | |
243 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
244 | } | |
245 | ||
5ebe7742 PZ |
246 | static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup) |
247 | { | |
248 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
249 | wait_for_completion(done); | |
250 | } | |
251 | ||
252 | static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup) | |
253 | { | |
254 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
255 | complete(done); | |
256 | } | |
257 | ||
258 | /* | |
259 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
260 | */ | |
261 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
262 | { | |
263 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
264 | } | |
265 | ||
b3199c02 | 266 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 267 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
268 | bool cpuhp_tasks_frozen; |
269 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 270 | |
79a6cdeb | 271 | /* |
93ae4f97 SB |
272 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
273 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
274 | */ |
275 | void cpu_maps_update_begin(void) | |
276 | { | |
277 | mutex_lock(&cpu_add_remove_lock); | |
278 | } | |
279 | ||
280 | void cpu_maps_update_done(void) | |
281 | { | |
282 | mutex_unlock(&cpu_add_remove_lock); | |
283 | } | |
1da177e4 | 284 | |
fc8dffd3 TG |
285 | /* |
286 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
287 | * Should always be manipulated under cpu_add_remove_lock |
288 | */ | |
289 | static int cpu_hotplug_disabled; | |
290 | ||
79a6cdeb LJ |
291 | #ifdef CONFIG_HOTPLUG_CPU |
292 | ||
fc8dffd3 | 293 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 294 | |
8f553c49 | 295 | void cpus_read_lock(void) |
a9d9baa1 | 296 | { |
fc8dffd3 | 297 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 298 | } |
8f553c49 | 299 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 300 | |
6f4ceee9 WL |
301 | int cpus_read_trylock(void) |
302 | { | |
303 | return percpu_down_read_trylock(&cpu_hotplug_lock); | |
304 | } | |
305 | EXPORT_SYMBOL_GPL(cpus_read_trylock); | |
306 | ||
8f553c49 | 307 | void cpus_read_unlock(void) |
a9d9baa1 | 308 | { |
fc8dffd3 | 309 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 310 | } |
8f553c49 | 311 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 312 | |
8f553c49 | 313 | void cpus_write_lock(void) |
d221938c | 314 | { |
fc8dffd3 | 315 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 316 | } |
87af9e7f | 317 | |
8f553c49 | 318 | void cpus_write_unlock(void) |
d221938c | 319 | { |
fc8dffd3 | 320 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
321 | } |
322 | ||
fc8dffd3 | 323 | void lockdep_assert_cpus_held(void) |
d221938c | 324 | { |
ce48c457 VS |
325 | /* |
326 | * We can't have hotplug operations before userspace starts running, | |
327 | * and some init codepaths will knowingly not take the hotplug lock. | |
328 | * This is all valid, so mute lockdep until it makes sense to report | |
329 | * unheld locks. | |
330 | */ | |
331 | if (system_state < SYSTEM_RUNNING) | |
332 | return; | |
333 | ||
fc8dffd3 | 334 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 335 | } |
79a6cdeb | 336 | |
43759fe5 FW |
337 | #ifdef CONFIG_LOCKDEP |
338 | int lockdep_is_cpus_held(void) | |
339 | { | |
340 | return percpu_rwsem_is_held(&cpu_hotplug_lock); | |
341 | } | |
342 | #endif | |
343 | ||
cb92173d PZ |
344 | static void lockdep_acquire_cpus_lock(void) |
345 | { | |
1751060e | 346 | rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_); |
cb92173d PZ |
347 | } |
348 | ||
349 | static void lockdep_release_cpus_lock(void) | |
350 | { | |
1751060e | 351 | rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_); |
cb92173d PZ |
352 | } |
353 | ||
16e53dbf SB |
354 | /* |
355 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
356 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
357 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
358 | * hotplug path before performing hotplug operations. So acquiring that lock | |
359 | * guarantees mutual exclusion from any currently running hotplug operations. | |
360 | */ | |
361 | void cpu_hotplug_disable(void) | |
362 | { | |
363 | cpu_maps_update_begin(); | |
89af7ba5 | 364 | cpu_hotplug_disabled++; |
16e53dbf SB |
365 | cpu_maps_update_done(); |
366 | } | |
32145c46 | 367 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 368 | |
01b41159 LW |
369 | static void __cpu_hotplug_enable(void) |
370 | { | |
371 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
372 | return; | |
373 | cpu_hotplug_disabled--; | |
374 | } | |
375 | ||
16e53dbf SB |
376 | void cpu_hotplug_enable(void) |
377 | { | |
378 | cpu_maps_update_begin(); | |
01b41159 | 379 | __cpu_hotplug_enable(); |
16e53dbf SB |
380 | cpu_maps_update_done(); |
381 | } | |
32145c46 | 382 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
cb92173d PZ |
383 | |
384 | #else | |
385 | ||
386 | static void lockdep_acquire_cpus_lock(void) | |
387 | { | |
388 | } | |
389 | ||
390 | static void lockdep_release_cpus_lock(void) | |
391 | { | |
392 | } | |
393 | ||
b9d10be7 | 394 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 395 | |
a74cfffb TG |
396 | /* |
397 | * Architectures that need SMT-specific errata handling during SMT hotplug | |
398 | * should override this. | |
399 | */ | |
400 | void __weak arch_smt_update(void) { } | |
401 | ||
0cc3cd21 TG |
402 | #ifdef CONFIG_HOTPLUG_SMT |
403 | enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED; | |
bc2d8d26 | 404 | |
8e1b706b | 405 | void __init cpu_smt_disable(bool force) |
0cc3cd21 | 406 | { |
e1572f1d | 407 | if (!cpu_smt_possible()) |
8e1b706b JK |
408 | return; |
409 | ||
410 | if (force) { | |
0cc3cd21 TG |
411 | pr_info("SMT: Force disabled\n"); |
412 | cpu_smt_control = CPU_SMT_FORCE_DISABLED; | |
8e1b706b | 413 | } else { |
d0e7d144 | 414 | pr_info("SMT: disabled\n"); |
8e1b706b | 415 | cpu_smt_control = CPU_SMT_DISABLED; |
0cc3cd21 | 416 | } |
8e1b706b JK |
417 | } |
418 | ||
fee0aede TG |
419 | /* |
420 | * The decision whether SMT is supported can only be done after the full | |
b284909a | 421 | * CPU identification. Called from architecture code. |
bc2d8d26 TG |
422 | */ |
423 | void __init cpu_smt_check_topology(void) | |
424 | { | |
b284909a | 425 | if (!topology_smt_supported()) |
bc2d8d26 TG |
426 | cpu_smt_control = CPU_SMT_NOT_SUPPORTED; |
427 | } | |
428 | ||
8e1b706b JK |
429 | static int __init smt_cmdline_disable(char *str) |
430 | { | |
431 | cpu_smt_disable(str && !strcmp(str, "force")); | |
0cc3cd21 TG |
432 | return 0; |
433 | } | |
434 | early_param("nosmt", smt_cmdline_disable); | |
435 | ||
436 | static inline bool cpu_smt_allowed(unsigned int cpu) | |
437 | { | |
b284909a | 438 | if (cpu_smt_control == CPU_SMT_ENABLED) |
0cc3cd21 TG |
439 | return true; |
440 | ||
b284909a | 441 | if (topology_is_primary_thread(cpu)) |
0cc3cd21 TG |
442 | return true; |
443 | ||
444 | /* | |
445 | * On x86 it's required to boot all logical CPUs at least once so | |
446 | * that the init code can get a chance to set CR4.MCE on each | |
182e073f | 447 | * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any |
0cc3cd21 TG |
448 | * core will shutdown the machine. |
449 | */ | |
e797bda3 | 450 | return !cpumask_test_cpu(cpu, &cpus_booted_once_mask); |
0cc3cd21 | 451 | } |
e1572f1d VK |
452 | |
453 | /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */ | |
454 | bool cpu_smt_possible(void) | |
455 | { | |
456 | return cpu_smt_control != CPU_SMT_FORCE_DISABLED && | |
457 | cpu_smt_control != CPU_SMT_NOT_SUPPORTED; | |
458 | } | |
459 | EXPORT_SYMBOL_GPL(cpu_smt_possible); | |
0cc3cd21 TG |
460 | #else |
461 | static inline bool cpu_smt_allowed(unsigned int cpu) { return true; } | |
462 | #endif | |
463 | ||
4dddfb5f PZ |
464 | static inline enum cpuhp_state |
465 | cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
466 | { | |
467 | enum cpuhp_state prev_state = st->state; | |
2ea46c6f | 468 | bool bringup = st->state < target; |
4dddfb5f PZ |
469 | |
470 | st->rollback = false; | |
471 | st->last = NULL; | |
472 | ||
473 | st->target = target; | |
474 | st->single = false; | |
2ea46c6f PZ |
475 | st->bringup = bringup; |
476 | if (cpu_dying(st->cpu) != !bringup) | |
477 | set_cpu_dying(st->cpu, !bringup); | |
4dddfb5f PZ |
478 | |
479 | return prev_state; | |
480 | } | |
481 | ||
482 | static inline void | |
483 | cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state) | |
484 | { | |
2ea46c6f PZ |
485 | bool bringup = !st->bringup; |
486 | ||
453e4108 VD |
487 | st->target = prev_state; |
488 | ||
489 | /* | |
490 | * Already rolling back. No need invert the bringup value or to change | |
491 | * the current state. | |
492 | */ | |
493 | if (st->rollback) | |
494 | return; | |
495 | ||
4dddfb5f PZ |
496 | st->rollback = true; |
497 | ||
498 | /* | |
499 | * If we have st->last we need to undo partial multi_instance of this | |
500 | * state first. Otherwise start undo at the previous state. | |
501 | */ | |
502 | if (!st->last) { | |
503 | if (st->bringup) | |
504 | st->state--; | |
505 | else | |
506 | st->state++; | |
507 | } | |
508 | ||
2ea46c6f PZ |
509 | st->bringup = bringup; |
510 | if (cpu_dying(st->cpu) != !bringup) | |
511 | set_cpu_dying(st->cpu, !bringup); | |
4dddfb5f PZ |
512 | } |
513 | ||
514 | /* Regular hotplug invocation of the AP hotplug thread */ | |
515 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
516 | { | |
517 | if (!st->single && st->state == st->target) | |
518 | return; | |
519 | ||
520 | st->result = 0; | |
521 | /* | |
522 | * Make sure the above stores are visible before should_run becomes | |
523 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
524 | */ | |
525 | smp_mb(); | |
526 | st->should_run = true; | |
527 | wake_up_process(st->thread); | |
5ebe7742 | 528 | wait_for_ap_thread(st, st->bringup); |
4dddfb5f PZ |
529 | } |
530 | ||
531 | static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target) | |
532 | { | |
533 | enum cpuhp_state prev_state; | |
534 | int ret; | |
535 | ||
536 | prev_state = cpuhp_set_state(st, target); | |
537 | __cpuhp_kick_ap(st); | |
538 | if ((ret = st->result)) { | |
539 | cpuhp_reset_state(st, prev_state); | |
540 | __cpuhp_kick_ap(st); | |
541 | } | |
542 | ||
543 | return ret; | |
544 | } | |
9cd4f1a4 | 545 | |
8df3e07e TG |
546 | static int bringup_wait_for_ap(unsigned int cpu) |
547 | { | |
548 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
549 | ||
9cd4f1a4 | 550 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
5ebe7742 | 551 | wait_for_ap_thread(st, true); |
dea1d0f5 TG |
552 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
553 | return -ECANCELED; | |
9cd4f1a4 | 554 | |
45178ac0 | 555 | /* Unpark the hotplug thread of the target cpu */ |
9cd4f1a4 TG |
556 | kthread_unpark(st->thread); |
557 | ||
0cc3cd21 TG |
558 | /* |
559 | * SMT soft disabling on X86 requires to bring the CPU out of the | |
560 | * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The | |
f5602011 | 561 | * CPU marked itself as booted_once in notify_cpu_starting() so the |
0cc3cd21 TG |
562 | * cpu_smt_allowed() check will now return false if this is not the |
563 | * primary sibling. | |
564 | */ | |
565 | if (!cpu_smt_allowed(cpu)) | |
566 | return -ECANCELED; | |
567 | ||
4dddfb5f PZ |
568 | if (st->target <= CPUHP_AP_ONLINE_IDLE) |
569 | return 0; | |
570 | ||
571 | return cpuhp_kick_ap(st, st->target); | |
8df3e07e TG |
572 | } |
573 | ||
ba997462 TG |
574 | static int bringup_cpu(unsigned int cpu) |
575 | { | |
576 | struct task_struct *idle = idle_thread_get(cpu); | |
577 | int ret; | |
578 | ||
aa877175 BO |
579 | /* |
580 | * Some architectures have to walk the irq descriptors to | |
581 | * setup the vector space for the cpu which comes online. | |
582 | * Prevent irq alloc/free across the bringup. | |
583 | */ | |
584 | irq_lock_sparse(); | |
585 | ||
ba997462 TG |
586 | /* Arch-specific enabling code. */ |
587 | ret = __cpu_up(cpu, idle); | |
aa877175 | 588 | irq_unlock_sparse(); |
530e9b76 | 589 | if (ret) |
ba997462 | 590 | return ret; |
9cd4f1a4 | 591 | return bringup_wait_for_ap(cpu); |
ba997462 TG |
592 | } |
593 | ||
bf2c59fc PZ |
594 | static int finish_cpu(unsigned int cpu) |
595 | { | |
596 | struct task_struct *idle = idle_thread_get(cpu); | |
597 | struct mm_struct *mm = idle->active_mm; | |
598 | ||
599 | /* | |
600 | * idle_task_exit() will have switched to &init_mm, now | |
601 | * clean up any remaining active_mm state. | |
602 | */ | |
603 | if (mm != &init_mm) | |
604 | idle->active_mm = &init_mm; | |
605 | mmdrop(mm); | |
606 | return 0; | |
607 | } | |
608 | ||
2e1a3483 TG |
609 | /* |
610 | * Hotplug state machine related functions | |
611 | */ | |
2e1a3483 | 612 | |
453e4108 VD |
613 | /* |
614 | * Get the next state to run. Empty ones will be skipped. Returns true if a | |
615 | * state must be run. | |
616 | * | |
617 | * st->state will be modified ahead of time, to match state_to_run, as if it | |
618 | * has already ran. | |
619 | */ | |
620 | static bool cpuhp_next_state(bool bringup, | |
621 | enum cpuhp_state *state_to_run, | |
622 | struct cpuhp_cpu_state *st, | |
623 | enum cpuhp_state target) | |
2e1a3483 | 624 | { |
453e4108 VD |
625 | do { |
626 | if (bringup) { | |
627 | if (st->state >= target) | |
628 | return false; | |
629 | ||
630 | *state_to_run = ++st->state; | |
631 | } else { | |
632 | if (st->state <= target) | |
633 | return false; | |
634 | ||
635 | *state_to_run = st->state--; | |
636 | } | |
637 | ||
638 | if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run))) | |
639 | break; | |
640 | } while (true); | |
641 | ||
642 | return true; | |
643 | } | |
644 | ||
645 | static int cpuhp_invoke_callback_range(bool bringup, | |
646 | unsigned int cpu, | |
647 | struct cpuhp_cpu_state *st, | |
648 | enum cpuhp_state target) | |
649 | { | |
650 | enum cpuhp_state state; | |
651 | int err = 0; | |
652 | ||
653 | while (cpuhp_next_state(bringup, &state, st, target)) { | |
654 | err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL); | |
655 | if (err) | |
656 | break; | |
657 | } | |
658 | ||
659 | return err; | |
2e1a3483 TG |
660 | } |
661 | ||
206b9235 TG |
662 | static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st) |
663 | { | |
664 | if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) | |
665 | return true; | |
666 | /* | |
667 | * When CPU hotplug is disabled, then taking the CPU down is not | |
668 | * possible because takedown_cpu() and the architecture and | |
669 | * subsystem specific mechanisms are not available. So the CPU | |
670 | * which would be completely unplugged again needs to stay around | |
671 | * in the current state. | |
672 | */ | |
673 | return st->state <= CPUHP_BRINGUP_CPU; | |
674 | } | |
675 | ||
2e1a3483 | 676 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, |
a724632c | 677 | enum cpuhp_state target) |
2e1a3483 TG |
678 | { |
679 | enum cpuhp_state prev_state = st->state; | |
680 | int ret = 0; | |
681 | ||
453e4108 VD |
682 | ret = cpuhp_invoke_callback_range(true, cpu, st, target); |
683 | if (ret) { | |
684 | cpuhp_reset_state(st, prev_state); | |
685 | if (can_rollback_cpu(st)) | |
686 | WARN_ON(cpuhp_invoke_callback_range(false, cpu, st, | |
687 | prev_state)); | |
2e1a3483 TG |
688 | } |
689 | return ret; | |
690 | } | |
691 | ||
4cb28ced TG |
692 | /* |
693 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
694 | */ | |
695 | static void cpuhp_create(unsigned int cpu) | |
696 | { | |
697 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
698 | ||
5ebe7742 PZ |
699 | init_completion(&st->done_up); |
700 | init_completion(&st->done_down); | |
2ea46c6f | 701 | st->cpu = cpu; |
4cb28ced TG |
702 | } |
703 | ||
704 | static int cpuhp_should_run(unsigned int cpu) | |
705 | { | |
706 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
707 | ||
708 | return st->should_run; | |
709 | } | |
710 | ||
4cb28ced TG |
711 | /* |
712 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
713 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
714 | * |
715 | * Each invocation of this function by the smpboot thread does a single AP | |
716 | * state callback. | |
717 | * | |
718 | * It has 3 modes of operation: | |
719 | * - single: runs st->cb_state | |
720 | * - up: runs ++st->state, while st->state < st->target | |
721 | * - down: runs st->state--, while st->state > st->target | |
722 | * | |
723 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
724 | */ |
725 | static void cpuhp_thread_fun(unsigned int cpu) | |
726 | { | |
727 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
728 | bool bringup = st->bringup; |
729 | enum cpuhp_state state; | |
4cb28ced | 730 | |
f8b7530a NU |
731 | if (WARN_ON_ONCE(!st->should_run)) |
732 | return; | |
733 | ||
4cb28ced | 734 | /* |
4dddfb5f PZ |
735 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
736 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
737 | */ |
738 | smp_mb(); | |
4cb28ced | 739 | |
cb92173d PZ |
740 | /* |
741 | * The BP holds the hotplug lock, but we're now running on the AP, | |
742 | * ensure that anybody asserting the lock is held, will actually find | |
743 | * it so. | |
744 | */ | |
745 | lockdep_acquire_cpus_lock(); | |
5f4b55e1 | 746 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 747 | |
a724632c | 748 | if (st->single) { |
4dddfb5f PZ |
749 | state = st->cb_state; |
750 | st->should_run = false; | |
751 | } else { | |
453e4108 VD |
752 | st->should_run = cpuhp_next_state(bringup, &state, st, st->target); |
753 | if (!st->should_run) | |
754 | goto end; | |
4dddfb5f PZ |
755 | } |
756 | ||
757 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
758 | ||
4dddfb5f PZ |
759 | if (cpuhp_is_atomic_state(state)) { |
760 | local_irq_disable(); | |
761 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
762 | local_irq_enable(); | |
3b9d6da6 | 763 | |
4dddfb5f PZ |
764 | /* |
765 | * STARTING/DYING must not fail! | |
766 | */ | |
767 | WARN_ON_ONCE(st->result); | |
4cb28ced | 768 | } else { |
4dddfb5f PZ |
769 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
770 | } | |
771 | ||
772 | if (st->result) { | |
773 | /* | |
774 | * If we fail on a rollback, we're up a creek without no | |
775 | * paddle, no way forward, no way back. We loose, thanks for | |
776 | * playing. | |
777 | */ | |
778 | WARN_ON_ONCE(st->rollback); | |
779 | st->should_run = false; | |
4cb28ced | 780 | } |
4dddfb5f | 781 | |
453e4108 | 782 | end: |
5f4b55e1 | 783 | cpuhp_lock_release(bringup); |
cb92173d | 784 | lockdep_release_cpus_lock(); |
4dddfb5f PZ |
785 | |
786 | if (!st->should_run) | |
5ebe7742 | 787 | complete_ap_thread(st, bringup); |
4cb28ced TG |
788 | } |
789 | ||
790 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 791 | static int |
cf392d10 TG |
792 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
793 | struct hlist_node *node) | |
4cb28ced TG |
794 | { |
795 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 796 | int ret; |
4cb28ced TG |
797 | |
798 | if (!cpu_online(cpu)) | |
799 | return 0; | |
800 | ||
5f4b55e1 PZ |
801 | cpuhp_lock_acquire(false); |
802 | cpuhp_lock_release(false); | |
803 | ||
804 | cpuhp_lock_acquire(true); | |
805 | cpuhp_lock_release(true); | |
49dfe2a6 | 806 | |
6a4e2451 TG |
807 | /* |
808 | * If we are up and running, use the hotplug thread. For early calls | |
809 | * we invoke the thread function directly. | |
810 | */ | |
811 | if (!st->thread) | |
96abb968 | 812 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 813 | |
4dddfb5f PZ |
814 | st->rollback = false; |
815 | st->last = NULL; | |
816 | ||
817 | st->node = node; | |
818 | st->bringup = bringup; | |
4cb28ced | 819 | st->cb_state = state; |
a724632c | 820 | st->single = true; |
a724632c | 821 | |
4dddfb5f | 822 | __cpuhp_kick_ap(st); |
4cb28ced | 823 | |
4cb28ced | 824 | /* |
4dddfb5f | 825 | * If we failed and did a partial, do a rollback. |
4cb28ced | 826 | */ |
4dddfb5f PZ |
827 | if ((ret = st->result) && st->last) { |
828 | st->rollback = true; | |
829 | st->bringup = !bringup; | |
830 | ||
831 | __cpuhp_kick_ap(st); | |
832 | } | |
833 | ||
1f7c70d6 TG |
834 | /* |
835 | * Clean up the leftovers so the next hotplug operation wont use stale | |
836 | * data. | |
837 | */ | |
838 | st->node = st->last = NULL; | |
4dddfb5f | 839 | return ret; |
1cf4f629 TG |
840 | } |
841 | ||
842 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
843 | { | |
844 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
845 | enum cpuhp_state prev_state = st->state; |
846 | int ret; | |
1cf4f629 | 847 | |
5f4b55e1 PZ |
848 | cpuhp_lock_acquire(false); |
849 | cpuhp_lock_release(false); | |
850 | ||
851 | cpuhp_lock_acquire(true); | |
852 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
853 | |
854 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
855 | ret = cpuhp_kick_ap(st, st->target); | |
856 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); | |
857 | ||
858 | return ret; | |
4cb28ced TG |
859 | } |
860 | ||
861 | static struct smp_hotplug_thread cpuhp_threads = { | |
862 | .store = &cpuhp_state.thread, | |
863 | .create = &cpuhp_create, | |
864 | .thread_should_run = cpuhp_should_run, | |
865 | .thread_fn = cpuhp_thread_fun, | |
866 | .thread_comm = "cpuhp/%u", | |
867 | .selfparking = true, | |
868 | }; | |
869 | ||
870 | void __init cpuhp_threads_init(void) | |
871 | { | |
872 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); | |
873 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
874 | } | |
875 | ||
777c6e0d | 876 | #ifdef CONFIG_HOTPLUG_CPU |
8ff00399 NP |
877 | #ifndef arch_clear_mm_cpumask_cpu |
878 | #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm)) | |
879 | #endif | |
880 | ||
e4cc2f87 AV |
881 | /** |
882 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
883 | * @cpu: a CPU id | |
884 | * | |
885 | * This function walks all processes, finds a valid mm struct for each one and | |
886 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
887 | * trivial, there are various non-obvious corner cases, which this function | |
888 | * tries to solve in a safe manner. | |
889 | * | |
890 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
891 | * be called only for an already offlined CPU. | |
892 | */ | |
cb79295e AV |
893 | void clear_tasks_mm_cpumask(int cpu) |
894 | { | |
895 | struct task_struct *p; | |
896 | ||
897 | /* | |
898 | * This function is called after the cpu is taken down and marked | |
899 | * offline, so its not like new tasks will ever get this cpu set in | |
900 | * their mm mask. -- Peter Zijlstra | |
901 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
902 | * full-fledged tasklist_lock. | |
903 | */ | |
e4cc2f87 | 904 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
905 | rcu_read_lock(); |
906 | for_each_process(p) { | |
907 | struct task_struct *t; | |
908 | ||
e4cc2f87 AV |
909 | /* |
910 | * Main thread might exit, but other threads may still have | |
911 | * a valid mm. Find one. | |
912 | */ | |
cb79295e AV |
913 | t = find_lock_task_mm(p); |
914 | if (!t) | |
915 | continue; | |
8ff00399 | 916 | arch_clear_mm_cpumask_cpu(cpu, t->mm); |
cb79295e AV |
917 | task_unlock(t); |
918 | } | |
919 | rcu_read_unlock(); | |
920 | } | |
921 | ||
1da177e4 | 922 | /* Take this CPU down. */ |
71cf5aee | 923 | static int take_cpu_down(void *_param) |
1da177e4 | 924 | { |
4baa0afc TG |
925 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
926 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 927 | int err, cpu = smp_processor_id(); |
724a8688 | 928 | int ret; |
1da177e4 | 929 | |
1da177e4 LT |
930 | /* Ensure this CPU doesn't handle any more interrupts. */ |
931 | err = __cpu_disable(); | |
932 | if (err < 0) | |
f3705136 | 933 | return err; |
1da177e4 | 934 | |
a724632c | 935 | /* |
453e4108 VD |
936 | * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going |
937 | * down, that the current state is CPUHP_TEARDOWN_CPU - 1. | |
a724632c | 938 | */ |
453e4108 VD |
939 | WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1)); |
940 | ||
4baa0afc | 941 | /* Invoke the former CPU_DYING callbacks */ |
453e4108 VD |
942 | ret = cpuhp_invoke_callback_range(false, cpu, st, target); |
943 | ||
944 | /* | |
945 | * DYING must not fail! | |
946 | */ | |
947 | WARN_ON_ONCE(ret); | |
4baa0afc | 948 | |
52c063d1 TG |
949 | /* Give up timekeeping duties */ |
950 | tick_handover_do_timer(); | |
1b72d432 TG |
951 | /* Remove CPU from timer broadcasting */ |
952 | tick_offline_cpu(cpu); | |
14e568e7 | 953 | /* Park the stopper thread */ |
090e77c3 | 954 | stop_machine_park(cpu); |
f3705136 | 955 | return 0; |
1da177e4 LT |
956 | } |
957 | ||
98458172 | 958 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 959 | { |
e69aab13 | 960 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 961 | int err; |
1da177e4 | 962 | |
2a58c527 | 963 | /* Park the smpboot threads */ |
1cf4f629 TG |
964 | kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread); |
965 | ||
6acce3ef | 966 | /* |
a8994181 TG |
967 | * Prevent irq alloc/free while the dying cpu reorganizes the |
968 | * interrupt affinities. | |
6acce3ef | 969 | */ |
a8994181 | 970 | irq_lock_sparse(); |
6acce3ef | 971 | |
a8994181 TG |
972 | /* |
973 | * So now all preempt/rcu users must observe !cpu_active(). | |
974 | */ | |
210e2133 | 975 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 976 | if (err) { |
3b9d6da6 | 977 | /* CPU refused to die */ |
a8994181 | 978 | irq_unlock_sparse(); |
3b9d6da6 SAS |
979 | /* Unpark the hotplug thread so we can rollback there */ |
980 | kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread); | |
98458172 | 981 | return err; |
8fa1d7d3 | 982 | } |
04321587 | 983 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 984 | |
48c5ccae | 985 | /* |
5b1ead68 BJ |
986 | * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed |
987 | * all runnable tasks from the CPU, there's only the idle task left now | |
48c5ccae | 988 | * that the migration thread is done doing the stop_machine thing. |
51a96c77 PZ |
989 | * |
990 | * Wait for the stop thread to go away. | |
48c5ccae | 991 | */ |
5ebe7742 | 992 | wait_for_ap_thread(st, false); |
e69aab13 | 993 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); |
1da177e4 | 994 | |
a8994181 TG |
995 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
996 | irq_unlock_sparse(); | |
997 | ||
345527b1 | 998 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
999 | /* This actually kills the CPU. */ |
1000 | __cpu_die(cpu); | |
1001 | ||
a49b116d | 1002 | tick_cleanup_dead_cpu(cpu); |
a58163d8 | 1003 | rcutree_migrate_callbacks(cpu); |
98458172 TG |
1004 | return 0; |
1005 | } | |
1da177e4 | 1006 | |
71f87b2f TG |
1007 | static void cpuhp_complete_idle_dead(void *arg) |
1008 | { | |
1009 | struct cpuhp_cpu_state *st = arg; | |
1010 | ||
5ebe7742 | 1011 | complete_ap_thread(st, false); |
71f87b2f TG |
1012 | } |
1013 | ||
e69aab13 TG |
1014 | void cpuhp_report_idle_dead(void) |
1015 | { | |
1016 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
1017 | ||
1018 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
27d50c7e | 1019 | rcu_report_dead(smp_processor_id()); |
71f87b2f TG |
1020 | st->state = CPUHP_AP_IDLE_DEAD; |
1021 | /* | |
1022 | * We cannot call complete after rcu_report_dead() so we delegate it | |
1023 | * to an online cpu. | |
1024 | */ | |
1025 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
1026 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
1027 | } |
1028 | ||
4dddfb5f PZ |
1029 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, |
1030 | enum cpuhp_state target) | |
1031 | { | |
1032 | enum cpuhp_state prev_state = st->state; | |
1033 | int ret = 0; | |
1034 | ||
453e4108 VD |
1035 | ret = cpuhp_invoke_callback_range(false, cpu, st, target); |
1036 | if (ret) { | |
1037 | ||
1038 | cpuhp_reset_state(st, prev_state); | |
1039 | ||
1040 | if (st->state < prev_state) | |
1041 | WARN_ON(cpuhp_invoke_callback_range(true, cpu, st, | |
1042 | prev_state)); | |
4dddfb5f | 1043 | } |
453e4108 | 1044 | |
4dddfb5f PZ |
1045 | return ret; |
1046 | } | |
cff7d378 | 1047 | |
98458172 | 1048 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
1049 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
1050 | enum cpuhp_state target) | |
98458172 | 1051 | { |
cff7d378 TG |
1052 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
1053 | int prev_state, ret = 0; | |
98458172 TG |
1054 | |
1055 | if (num_online_cpus() == 1) | |
1056 | return -EBUSY; | |
1057 | ||
757c989b | 1058 | if (!cpu_present(cpu)) |
98458172 TG |
1059 | return -EINVAL; |
1060 | ||
8f553c49 | 1061 | cpus_write_lock(); |
98458172 TG |
1062 | |
1063 | cpuhp_tasks_frozen = tasks_frozen; | |
1064 | ||
4dddfb5f | 1065 | prev_state = cpuhp_set_state(st, target); |
1cf4f629 TG |
1066 | /* |
1067 | * If the current CPU state is in the range of the AP hotplug thread, | |
1068 | * then we need to kick the thread. | |
1069 | */ | |
8df3e07e | 1070 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 1071 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
1072 | ret = cpuhp_kick_ap_work(cpu); |
1073 | /* | |
1074 | * The AP side has done the error rollback already. Just | |
1075 | * return the error code.. | |
1076 | */ | |
1077 | if (ret) | |
1078 | goto out; | |
1079 | ||
1080 | /* | |
1081 | * We might have stopped still in the range of the AP hotplug | |
1082 | * thread. Nothing to do anymore. | |
1083 | */ | |
8df3e07e | 1084 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 1085 | goto out; |
4dddfb5f PZ |
1086 | |
1087 | st->target = target; | |
1cf4f629 TG |
1088 | } |
1089 | /* | |
8df3e07e | 1090 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
1091 | * to do the further cleanups. |
1092 | */ | |
a724632c | 1093 | ret = cpuhp_down_callbacks(cpu, st, target); |
62f25069 VD |
1094 | if (ret && st->state < prev_state) { |
1095 | if (st->state == CPUHP_TEARDOWN_CPU) { | |
1096 | cpuhp_reset_state(st, prev_state); | |
1097 | __cpuhp_kick_ap(st); | |
1098 | } else { | |
1099 | WARN(1, "DEAD callback error for CPU%d", cpu); | |
1100 | } | |
3b9d6da6 | 1101 | } |
98458172 | 1102 | |
1cf4f629 | 1103 | out: |
8f553c49 | 1104 | cpus_write_unlock(); |
941154bd TG |
1105 | /* |
1106 | * Do post unplug cleanup. This is still protected against | |
1107 | * concurrent CPU hotplug via cpu_add_remove_lock. | |
1108 | */ | |
1109 | lockup_detector_cleanup(); | |
a74cfffb | 1110 | arch_smt_update(); |
cff7d378 | 1111 | return ret; |
e3920fb4 RW |
1112 | } |
1113 | ||
cc1fe215 TG |
1114 | static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target) |
1115 | { | |
1116 | if (cpu_hotplug_disabled) | |
1117 | return -EBUSY; | |
1118 | return _cpu_down(cpu, 0, target); | |
1119 | } | |
1120 | ||
33c3736e | 1121 | static int cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 1122 | { |
9ea09af3 | 1123 | int err; |
e3920fb4 | 1124 | |
d221938c | 1125 | cpu_maps_update_begin(); |
cc1fe215 | 1126 | err = cpu_down_maps_locked(cpu, target); |
d221938c | 1127 | cpu_maps_update_done(); |
1da177e4 LT |
1128 | return err; |
1129 | } | |
4dddfb5f | 1130 | |
33c3736e QY |
1131 | /** |
1132 | * cpu_device_down - Bring down a cpu device | |
1133 | * @dev: Pointer to the cpu device to offline | |
1134 | * | |
1135 | * This function is meant to be used by device core cpu subsystem only. | |
1136 | * | |
1137 | * Other subsystems should use remove_cpu() instead. | |
1138 | */ | |
1139 | int cpu_device_down(struct device *dev) | |
af1f4045 | 1140 | { |
33c3736e | 1141 | return cpu_down(dev->id, CPUHP_OFFLINE); |
af1f4045 | 1142 | } |
4dddfb5f | 1143 | |
93ef1429 QY |
1144 | int remove_cpu(unsigned int cpu) |
1145 | { | |
1146 | int ret; | |
1147 | ||
1148 | lock_device_hotplug(); | |
1149 | ret = device_offline(get_cpu_device(cpu)); | |
1150 | unlock_device_hotplug(); | |
1151 | ||
1152 | return ret; | |
1153 | } | |
1154 | EXPORT_SYMBOL_GPL(remove_cpu); | |
1155 | ||
0441a559 QY |
1156 | void smp_shutdown_nonboot_cpus(unsigned int primary_cpu) |
1157 | { | |
1158 | unsigned int cpu; | |
1159 | int error; | |
1160 | ||
1161 | cpu_maps_update_begin(); | |
1162 | ||
1163 | /* | |
1164 | * Make certain the cpu I'm about to reboot on is online. | |
1165 | * | |
1166 | * This is inline to what migrate_to_reboot_cpu() already do. | |
1167 | */ | |
1168 | if (!cpu_online(primary_cpu)) | |
1169 | primary_cpu = cpumask_first(cpu_online_mask); | |
1170 | ||
1171 | for_each_online_cpu(cpu) { | |
1172 | if (cpu == primary_cpu) | |
1173 | continue; | |
1174 | ||
1175 | error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); | |
1176 | if (error) { | |
1177 | pr_err("Failed to offline CPU%d - error=%d", | |
1178 | cpu, error); | |
1179 | break; | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | /* | |
1184 | * Ensure all but the reboot CPU are offline. | |
1185 | */ | |
1186 | BUG_ON(num_online_cpus() > 1); | |
1187 | ||
1188 | /* | |
1189 | * Make sure the CPUs won't be enabled by someone else after this | |
1190 | * point. Kexec will reboot to a new kernel shortly resetting | |
1191 | * everything along the way. | |
1192 | */ | |
1193 | cpu_hotplug_disabled++; | |
1194 | ||
1195 | cpu_maps_update_done(); | |
af1f4045 | 1196 | } |
4dddfb5f PZ |
1197 | |
1198 | #else | |
1199 | #define takedown_cpu NULL | |
1da177e4 LT |
1200 | #endif /*CONFIG_HOTPLUG_CPU*/ |
1201 | ||
4baa0afc | 1202 | /** |
ee1e714b | 1203 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
1204 | * @cpu: cpu that just started |
1205 | * | |
4baa0afc TG |
1206 | * It must be called by the arch code on the new cpu, before the new cpu |
1207 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
1208 | */ | |
1209 | void notify_cpu_starting(unsigned int cpu) | |
1210 | { | |
1211 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1212 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
724a8688 | 1213 | int ret; |
4baa0afc | 1214 | |
0c6d4576 | 1215 | rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
e797bda3 | 1216 | cpumask_set_cpu(cpu, &cpus_booted_once_mask); |
453e4108 VD |
1217 | ret = cpuhp_invoke_callback_range(true, cpu, st, target); |
1218 | ||
1219 | /* | |
1220 | * STARTING must not fail! | |
1221 | */ | |
1222 | WARN_ON_ONCE(ret); | |
4baa0afc TG |
1223 | } |
1224 | ||
949338e3 | 1225 | /* |
9cd4f1a4 | 1226 | * Called from the idle task. Wake up the controlling task which brings the |
45178ac0 PZ |
1227 | * hotplug thread of the upcoming CPU up and then delegates the rest of the |
1228 | * online bringup to the hotplug thread. | |
949338e3 | 1229 | */ |
8df3e07e | 1230 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 1231 | { |
8df3e07e | 1232 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
1233 | |
1234 | /* Happens for the boot cpu */ | |
1235 | if (state != CPUHP_AP_ONLINE_IDLE) | |
1236 | return; | |
1237 | ||
45178ac0 PZ |
1238 | /* |
1239 | * Unpart the stopper thread before we start the idle loop (and start | |
1240 | * scheduling); this ensures the stopper task is always available. | |
1241 | */ | |
1242 | stop_machine_unpark(smp_processor_id()); | |
1243 | ||
8df3e07e | 1244 | st->state = CPUHP_AP_ONLINE_IDLE; |
5ebe7742 | 1245 | complete_ap_thread(st, true); |
949338e3 TG |
1246 | } |
1247 | ||
e3920fb4 | 1248 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1249 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1250 | { |
cff7d378 | 1251 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1252 | struct task_struct *idle; |
2e1a3483 | 1253 | int ret = 0; |
1da177e4 | 1254 | |
8f553c49 | 1255 | cpus_write_lock(); |
38498a67 | 1256 | |
757c989b | 1257 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1258 | ret = -EINVAL; |
1259 | goto out; | |
1260 | } | |
1261 | ||
757c989b | 1262 | /* |
33c3736e QY |
1263 | * The caller of cpu_up() might have raced with another |
1264 | * caller. Nothing to do. | |
757c989b TG |
1265 | */ |
1266 | if (st->state >= target) | |
38498a67 | 1267 | goto out; |
757c989b TG |
1268 | |
1269 | if (st->state == CPUHP_OFFLINE) { | |
1270 | /* Let it fail before we try to bring the cpu up */ | |
1271 | idle = idle_thread_get(cpu); | |
1272 | if (IS_ERR(idle)) { | |
1273 | ret = PTR_ERR(idle); | |
1274 | goto out; | |
1275 | } | |
3bb5d2ee | 1276 | } |
38498a67 | 1277 | |
ba997462 TG |
1278 | cpuhp_tasks_frozen = tasks_frozen; |
1279 | ||
4dddfb5f | 1280 | cpuhp_set_state(st, target); |
1cf4f629 TG |
1281 | /* |
1282 | * If the current CPU state is in the range of the AP hotplug thread, | |
1283 | * then we need to kick the thread once more. | |
1284 | */ | |
8df3e07e | 1285 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1286 | ret = cpuhp_kick_ap_work(cpu); |
1287 | /* | |
1288 | * The AP side has done the error rollback already. Just | |
1289 | * return the error code.. | |
1290 | */ | |
1291 | if (ret) | |
1292 | goto out; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1297 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1298 | * responsible for bringing it up to the target state. |
1299 | */ | |
8df3e07e | 1300 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1301 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1302 | out: |
8f553c49 | 1303 | cpus_write_unlock(); |
a74cfffb | 1304 | arch_smt_update(); |
e3920fb4 RW |
1305 | return ret; |
1306 | } | |
1307 | ||
33c3736e | 1308 | static int cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1309 | { |
1310 | int err = 0; | |
cf23422b | 1311 | |
e0b582ec | 1312 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1313 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1314 | cpu); | |
87d5e023 | 1315 | #if defined(CONFIG_IA64) |
84117da5 | 1316 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
1317 | #endif |
1318 | return -EINVAL; | |
1319 | } | |
e3920fb4 | 1320 | |
01b0f197 TK |
1321 | err = try_online_node(cpu_to_node(cpu)); |
1322 | if (err) | |
1323 | return err; | |
cf23422b | 1324 | |
d221938c | 1325 | cpu_maps_update_begin(); |
e761b772 MK |
1326 | |
1327 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1328 | err = -EBUSY; |
e761b772 MK |
1329 | goto out; |
1330 | } | |
05736e4a TG |
1331 | if (!cpu_smt_allowed(cpu)) { |
1332 | err = -EPERM; | |
1333 | goto out; | |
1334 | } | |
e761b772 | 1335 | |
af1f4045 | 1336 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1337 | out: |
d221938c | 1338 | cpu_maps_update_done(); |
e3920fb4 RW |
1339 | return err; |
1340 | } | |
af1f4045 | 1341 | |
33c3736e QY |
1342 | /** |
1343 | * cpu_device_up - Bring up a cpu device | |
1344 | * @dev: Pointer to the cpu device to online | |
1345 | * | |
1346 | * This function is meant to be used by device core cpu subsystem only. | |
1347 | * | |
1348 | * Other subsystems should use add_cpu() instead. | |
1349 | */ | |
1350 | int cpu_device_up(struct device *dev) | |
af1f4045 | 1351 | { |
33c3736e | 1352 | return cpu_up(dev->id, CPUHP_ONLINE); |
af1f4045 | 1353 | } |
e3920fb4 | 1354 | |
93ef1429 QY |
1355 | int add_cpu(unsigned int cpu) |
1356 | { | |
1357 | int ret; | |
1358 | ||
1359 | lock_device_hotplug(); | |
1360 | ret = device_online(get_cpu_device(cpu)); | |
1361 | unlock_device_hotplug(); | |
1362 | ||
1363 | return ret; | |
1364 | } | |
1365 | EXPORT_SYMBOL_GPL(add_cpu); | |
1366 | ||
d720f986 QY |
1367 | /** |
1368 | * bringup_hibernate_cpu - Bring up the CPU that we hibernated on | |
1369 | * @sleep_cpu: The cpu we hibernated on and should be brought up. | |
1370 | * | |
1371 | * On some architectures like arm64, we can hibernate on any CPU, but on | |
1372 | * wake up the CPU we hibernated on might be offline as a side effect of | |
1373 | * using maxcpus= for example. | |
1374 | */ | |
1375 | int bringup_hibernate_cpu(unsigned int sleep_cpu) | |
af1f4045 | 1376 | { |
d720f986 QY |
1377 | int ret; |
1378 | ||
1379 | if (!cpu_online(sleep_cpu)) { | |
1380 | pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n"); | |
33c3736e | 1381 | ret = cpu_up(sleep_cpu, CPUHP_ONLINE); |
d720f986 QY |
1382 | if (ret) { |
1383 | pr_err("Failed to bring hibernate-CPU up!\n"); | |
1384 | return ret; | |
1385 | } | |
1386 | } | |
1387 | return 0; | |
1388 | } | |
1389 | ||
b99a2659 QY |
1390 | void bringup_nonboot_cpus(unsigned int setup_max_cpus) |
1391 | { | |
1392 | unsigned int cpu; | |
1393 | ||
1394 | for_each_present_cpu(cpu) { | |
1395 | if (num_online_cpus() >= setup_max_cpus) | |
1396 | break; | |
1397 | if (!cpu_online(cpu)) | |
33c3736e | 1398 | cpu_up(cpu, CPUHP_ONLINE); |
b99a2659 | 1399 | } |
af1f4045 | 1400 | } |
e3920fb4 | 1401 | |
f3de4be9 | 1402 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1403 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1404 | |
fb7fb84a | 1405 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1406 | { |
d391e552 | 1407 | int cpu, error = 0; |
e3920fb4 | 1408 | |
d221938c | 1409 | cpu_maps_update_begin(); |
9ca12ac0 | 1410 | if (primary == -1) { |
d391e552 | 1411 | primary = cpumask_first(cpu_online_mask); |
9ca12ac0 NP |
1412 | if (!housekeeping_cpu(primary, HK_FLAG_TIMER)) |
1413 | primary = housekeeping_any_cpu(HK_FLAG_TIMER); | |
1414 | } else { | |
1415 | if (!cpu_online(primary)) | |
1416 | primary = cpumask_first(cpu_online_mask); | |
1417 | } | |
1418 | ||
9ee349ad XF |
1419 | /* |
1420 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1421 | * with the userspace trying to use the CPU hotplug at the same time |
1422 | */ | |
e0b582ec | 1423 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1424 | |
84117da5 | 1425 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1426 | for_each_online_cpu(cpu) { |
d391e552 | 1427 | if (cpu == primary) |
e3920fb4 | 1428 | continue; |
a66d955e | 1429 | |
fb7fb84a | 1430 | if (pm_wakeup_pending()) { |
a66d955e PK |
1431 | pr_info("Wakeup pending. Abort CPU freeze\n"); |
1432 | error = -EBUSY; | |
1433 | break; | |
1434 | } | |
1435 | ||
bb3632c6 | 1436 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1437 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1438 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1439 | if (!error) |
e0b582ec | 1440 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1441 | else { |
84117da5 | 1442 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1443 | break; |
1444 | } | |
1445 | } | |
86886e55 | 1446 | |
89af7ba5 | 1447 | if (!error) |
e3920fb4 | 1448 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1449 | else |
84117da5 | 1450 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1451 | |
1452 | /* | |
1453 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
56555855 QY |
1454 | * this even in case of failure as all freeze_secondary_cpus() users are |
1455 | * supposed to do thaw_secondary_cpus() on the failure path. | |
89af7ba5 VK |
1456 | */ |
1457 | cpu_hotplug_disabled++; | |
1458 | ||
d221938c | 1459 | cpu_maps_update_done(); |
e3920fb4 RW |
1460 | return error; |
1461 | } | |
1462 | ||
56555855 | 1463 | void __weak arch_thaw_secondary_cpus_begin(void) |
d0af9eed SS |
1464 | { |
1465 | } | |
1466 | ||
56555855 | 1467 | void __weak arch_thaw_secondary_cpus_end(void) |
d0af9eed SS |
1468 | { |
1469 | } | |
1470 | ||
56555855 | 1471 | void thaw_secondary_cpus(void) |
e3920fb4 RW |
1472 | { |
1473 | int cpu, error; | |
1474 | ||
1475 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1476 | cpu_maps_update_begin(); |
01b41159 | 1477 | __cpu_hotplug_enable(); |
e0b582ec | 1478 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1479 | goto out; |
e3920fb4 | 1480 | |
84117da5 | 1481 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed | 1482 | |
56555855 | 1483 | arch_thaw_secondary_cpus_begin(); |
d0af9eed | 1484 | |
e0b582ec | 1485 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1486 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1487 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1488 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1489 | if (!error) { |
84117da5 | 1490 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1491 | continue; |
1492 | } | |
84117da5 | 1493 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1494 | } |
d0af9eed | 1495 | |
56555855 | 1496 | arch_thaw_secondary_cpus_end(); |
d0af9eed | 1497 | |
e0b582ec | 1498 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1499 | out: |
d221938c | 1500 | cpu_maps_update_done(); |
1da177e4 | 1501 | } |
e0b582ec | 1502 | |
d7268a31 | 1503 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1504 | { |
1505 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1506 | return -ENOMEM; | |
1507 | return 0; | |
1508 | } | |
1509 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1510 | |
79cfbdfa SB |
1511 | /* |
1512 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1513 | * ensure that the state of the system with respect to the tasks being frozen | |
1514 | * or not, as reported by the notification, remains unchanged *throughout the | |
1515 | * duration* of the execution of the callbacks. | |
1516 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1517 | * | |
1518 | * This synchronization is implemented by mutually excluding regular CPU | |
1519 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1520 | * Hibernate notifications. | |
1521 | */ | |
1522 | static int | |
1523 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1524 | unsigned long action, void *ptr) | |
1525 | { | |
1526 | switch (action) { | |
1527 | ||
1528 | case PM_SUSPEND_PREPARE: | |
1529 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 1530 | cpu_hotplug_disable(); |
79cfbdfa SB |
1531 | break; |
1532 | ||
1533 | case PM_POST_SUSPEND: | |
1534 | case PM_POST_HIBERNATION: | |
16e53dbf | 1535 | cpu_hotplug_enable(); |
79cfbdfa SB |
1536 | break; |
1537 | ||
1538 | default: | |
1539 | return NOTIFY_DONE; | |
1540 | } | |
1541 | ||
1542 | return NOTIFY_OK; | |
1543 | } | |
1544 | ||
1545 | ||
d7268a31 | 1546 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 1547 | { |
6e32d479 FY |
1548 | /* |
1549 | * cpu_hotplug_pm_callback has higher priority than x86 | |
1550 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
1551 | * to disable cpu hotplug to avoid cpu hotplug race. | |
1552 | */ | |
79cfbdfa SB |
1553 | pm_notifier(cpu_hotplug_pm_callback, 0); |
1554 | return 0; | |
1555 | } | |
1556 | core_initcall(cpu_hotplug_pm_sync_init); | |
1557 | ||
f3de4be9 | 1558 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 1559 | |
8ce371f9 PZ |
1560 | int __boot_cpu_id; |
1561 | ||
68f4f1ec | 1562 | #endif /* CONFIG_SMP */ |
b8d317d1 | 1563 | |
cff7d378 | 1564 | /* Boot processor state steps */ |
17a2f1ce | 1565 | static struct cpuhp_step cpuhp_hp_states[] = { |
cff7d378 TG |
1566 | [CPUHP_OFFLINE] = { |
1567 | .name = "offline", | |
3c1627e9 TG |
1568 | .startup.single = NULL, |
1569 | .teardown.single = NULL, | |
cff7d378 TG |
1570 | }, |
1571 | #ifdef CONFIG_SMP | |
1572 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 1573 | .name = "threads:prepare", |
3c1627e9 TG |
1574 | .startup.single = smpboot_create_threads, |
1575 | .teardown.single = NULL, | |
757c989b | 1576 | .cant_stop = true, |
cff7d378 | 1577 | }, |
00e16c3d | 1578 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
1579 | .name = "perf:prepare", |
1580 | .startup.single = perf_event_init_cpu, | |
1581 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1582 | }, |
7ee681b2 | 1583 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
1584 | .name = "workqueue:prepare", |
1585 | .startup.single = workqueue_prepare_cpu, | |
1586 | .teardown.single = NULL, | |
7ee681b2 | 1587 | }, |
27590dc1 | 1588 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
1589 | .name = "hrtimers:prepare", |
1590 | .startup.single = hrtimers_prepare_cpu, | |
1591 | .teardown.single = hrtimers_dead_cpu, | |
27590dc1 | 1592 | }, |
31487f83 | 1593 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 1594 | .name = "smpcfd:prepare", |
3c1627e9 TG |
1595 | .startup.single = smpcfd_prepare_cpu, |
1596 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 1597 | }, |
e6d4989a RW |
1598 | [CPUHP_RELAY_PREPARE] = { |
1599 | .name = "relay:prepare", | |
1600 | .startup.single = relay_prepare_cpu, | |
1601 | .teardown.single = NULL, | |
1602 | }, | |
6731d4f1 SAS |
1603 | [CPUHP_SLAB_PREPARE] = { |
1604 | .name = "slab:prepare", | |
1605 | .startup.single = slab_prepare_cpu, | |
1606 | .teardown.single = slab_dead_cpu, | |
31487f83 | 1607 | }, |
4df83742 | 1608 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 1609 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
1610 | .startup.single = rcutree_prepare_cpu, |
1611 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 1612 | }, |
4fae16df RC |
1613 | /* |
1614 | * On the tear-down path, timers_dead_cpu() must be invoked | |
1615 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
1616 | * otherwise a RCU stall occurs. | |
1617 | */ | |
26456f87 | 1618 | [CPUHP_TIMERS_PREPARE] = { |
d018031f | 1619 | .name = "timers:prepare", |
26456f87 | 1620 | .startup.single = timers_prepare_cpu, |
3c1627e9 | 1621 | .teardown.single = timers_dead_cpu, |
4fae16df | 1622 | }, |
d10ef6f9 | 1623 | /* Kicks the plugged cpu into life */ |
cff7d378 TG |
1624 | [CPUHP_BRINGUP_CPU] = { |
1625 | .name = "cpu:bringup", | |
3c1627e9 | 1626 | .startup.single = bringup_cpu, |
bf2c59fc | 1627 | .teardown.single = finish_cpu, |
757c989b | 1628 | .cant_stop = true, |
4baa0afc | 1629 | }, |
d10ef6f9 TG |
1630 | /* Final state before CPU kills itself */ |
1631 | [CPUHP_AP_IDLE_DEAD] = { | |
1632 | .name = "idle:dead", | |
1633 | }, | |
1634 | /* | |
1635 | * Last state before CPU enters the idle loop to die. Transient state | |
1636 | * for synchronization. | |
1637 | */ | |
1638 | [CPUHP_AP_OFFLINE] = { | |
1639 | .name = "ap:offline", | |
1640 | .cant_stop = true, | |
1641 | }, | |
9cf7243d TG |
1642 | /* First state is scheduler control. Interrupts are disabled */ |
1643 | [CPUHP_AP_SCHED_STARTING] = { | |
1644 | .name = "sched:starting", | |
3c1627e9 TG |
1645 | .startup.single = sched_cpu_starting, |
1646 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 1647 | }, |
4df83742 | 1648 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 1649 | .name = "RCU/tree:dying", |
3c1627e9 TG |
1650 | .startup.single = NULL, |
1651 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 1652 | }, |
46febd37 LJ |
1653 | [CPUHP_AP_SMPCFD_DYING] = { |
1654 | .name = "smpcfd:dying", | |
1655 | .startup.single = NULL, | |
1656 | .teardown.single = smpcfd_dying_cpu, | |
1657 | }, | |
d10ef6f9 TG |
1658 | /* Entry state on starting. Interrupts enabled from here on. Transient |
1659 | * state for synchronsization */ | |
1660 | [CPUHP_AP_ONLINE] = { | |
1661 | .name = "ap:online", | |
1662 | }, | |
17a2f1ce | 1663 | /* |
1cf12e08 | 1664 | * Handled on control processor until the plugged processor manages |
17a2f1ce LJ |
1665 | * this itself. |
1666 | */ | |
1667 | [CPUHP_TEARDOWN_CPU] = { | |
1668 | .name = "cpu:teardown", | |
1669 | .startup.single = NULL, | |
1670 | .teardown.single = takedown_cpu, | |
1671 | .cant_stop = true, | |
1672 | }, | |
1cf12e08 TG |
1673 | |
1674 | [CPUHP_AP_SCHED_WAIT_EMPTY] = { | |
1675 | .name = "sched:waitempty", | |
1676 | .startup.single = NULL, | |
1677 | .teardown.single = sched_cpu_wait_empty, | |
1678 | }, | |
1679 | ||
d10ef6f9 | 1680 | /* Handle smpboot threads park/unpark */ |
1cf4f629 | 1681 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 1682 | .name = "smpboot/threads:online", |
3c1627e9 | 1683 | .startup.single = smpboot_unpark_threads, |
c4de6569 | 1684 | .teardown.single = smpboot_park_threads, |
1cf4f629 | 1685 | }, |
c5cb83bb TG |
1686 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
1687 | .name = "irq/affinity:online", | |
1688 | .startup.single = irq_affinity_online_cpu, | |
1689 | .teardown.single = NULL, | |
1690 | }, | |
00e16c3d | 1691 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
1692 | .name = "perf:online", |
1693 | .startup.single = perf_event_init_cpu, | |
1694 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 1695 | }, |
9cf57731 PZ |
1696 | [CPUHP_AP_WATCHDOG_ONLINE] = { |
1697 | .name = "lockup_detector:online", | |
1698 | .startup.single = lockup_detector_online_cpu, | |
1699 | .teardown.single = lockup_detector_offline_cpu, | |
1700 | }, | |
7ee681b2 | 1701 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
1702 | .name = "workqueue:online", |
1703 | .startup.single = workqueue_online_cpu, | |
1704 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 1705 | }, |
4df83742 | 1706 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 1707 | .name = "RCU/tree:online", |
3c1627e9 TG |
1708 | .startup.single = rcutree_online_cpu, |
1709 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 1710 | }, |
4baa0afc | 1711 | #endif |
d10ef6f9 TG |
1712 | /* |
1713 | * The dynamically registered state space is here | |
1714 | */ | |
1715 | ||
aaddd7d1 TG |
1716 | #ifdef CONFIG_SMP |
1717 | /* Last state is scheduler control setting the cpu active */ | |
1718 | [CPUHP_AP_ACTIVE] = { | |
1719 | .name = "sched:active", | |
3c1627e9 TG |
1720 | .startup.single = sched_cpu_activate, |
1721 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
1722 | }, |
1723 | #endif | |
1724 | ||
d10ef6f9 | 1725 | /* CPU is fully up and running. */ |
4baa0afc TG |
1726 | [CPUHP_ONLINE] = { |
1727 | .name = "online", | |
3c1627e9 TG |
1728 | .startup.single = NULL, |
1729 | .teardown.single = NULL, | |
4baa0afc TG |
1730 | }, |
1731 | }; | |
1732 | ||
5b7aa87e TG |
1733 | /* Sanity check for callbacks */ |
1734 | static int cpuhp_cb_check(enum cpuhp_state state) | |
1735 | { | |
1736 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
1737 | return -EINVAL; | |
1738 | return 0; | |
1739 | } | |
1740 | ||
dc280d93 TG |
1741 | /* |
1742 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1743 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1744 | * by having no name assigned. | |
1745 | */ | |
1746 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1747 | { | |
4205e478 TG |
1748 | enum cpuhp_state i, end; |
1749 | struct cpuhp_step *step; | |
dc280d93 | 1750 | |
4205e478 TG |
1751 | switch (state) { |
1752 | case CPUHP_AP_ONLINE_DYN: | |
17a2f1ce | 1753 | step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN; |
4205e478 TG |
1754 | end = CPUHP_AP_ONLINE_DYN_END; |
1755 | break; | |
1756 | case CPUHP_BP_PREPARE_DYN: | |
17a2f1ce | 1757 | step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN; |
4205e478 TG |
1758 | end = CPUHP_BP_PREPARE_DYN_END; |
1759 | break; | |
1760 | default: | |
1761 | return -EINVAL; | |
1762 | } | |
1763 | ||
1764 | for (i = state; i <= end; i++, step++) { | |
1765 | if (!step->name) | |
dc280d93 TG |
1766 | return i; |
1767 | } | |
1768 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1769 | return -ENOSPC; | |
1770 | } | |
1771 | ||
1772 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
1773 | int (*startup)(unsigned int cpu), | |
1774 | int (*teardown)(unsigned int cpu), | |
1775 | bool multi_instance) | |
5b7aa87e TG |
1776 | { |
1777 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1778 | struct cpuhp_step *sp; | |
dc280d93 | 1779 | int ret = 0; |
5b7aa87e | 1780 | |
0c96b273 EB |
1781 | /* |
1782 | * If name is NULL, then the state gets removed. | |
1783 | * | |
1784 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
1785 | * the first allocation from these dynamic ranges, so the removal | |
1786 | * would trigger a new allocation and clear the wrong (already | |
1787 | * empty) state, leaving the callbacks of the to be cleared state | |
1788 | * dangling, which causes wreckage on the next hotplug operation. | |
1789 | */ | |
1790 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
1791 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
1792 | ret = cpuhp_reserve_state(state); |
1793 | if (ret < 0) | |
dc434e05 | 1794 | return ret; |
dc280d93 TG |
1795 | state = ret; |
1796 | } | |
5b7aa87e | 1797 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
1798 | if (name && sp->name) |
1799 | return -EBUSY; | |
1800 | ||
3c1627e9 TG |
1801 | sp->startup.single = startup; |
1802 | sp->teardown.single = teardown; | |
5b7aa87e | 1803 | sp->name = name; |
cf392d10 TG |
1804 | sp->multi_instance = multi_instance; |
1805 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 1806 | return ret; |
5b7aa87e TG |
1807 | } |
1808 | ||
1809 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1810 | { | |
3c1627e9 | 1811 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
1812 | } |
1813 | ||
5b7aa87e TG |
1814 | /* |
1815 | * Call the startup/teardown function for a step either on the AP or | |
1816 | * on the current CPU. | |
1817 | */ | |
cf392d10 TG |
1818 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
1819 | struct hlist_node *node) | |
5b7aa87e | 1820 | { |
a724632c | 1821 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
1822 | int ret; |
1823 | ||
4dddfb5f PZ |
1824 | /* |
1825 | * If there's nothing to do, we done. | |
1826 | * Relies on the union for multi_instance. | |
1827 | */ | |
453e4108 | 1828 | if (cpuhp_step_empty(bringup, sp)) |
5b7aa87e | 1829 | return 0; |
5b7aa87e TG |
1830 | /* |
1831 | * The non AP bound callbacks can fail on bringup. On teardown | |
1832 | * e.g. module removal we crash for now. | |
1833 | */ | |
1cf4f629 TG |
1834 | #ifdef CONFIG_SMP |
1835 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 1836 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 1837 | else |
96abb968 | 1838 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1839 | #else |
96abb968 | 1840 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 1841 | #endif |
5b7aa87e TG |
1842 | BUG_ON(ret && !bringup); |
1843 | return ret; | |
1844 | } | |
1845 | ||
1846 | /* | |
1847 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1848 | * | |
1849 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1850 | */ | |
1851 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 1852 | struct hlist_node *node) |
5b7aa87e TG |
1853 | { |
1854 | int cpu; | |
1855 | ||
5b7aa87e TG |
1856 | /* Roll back the already executed steps on the other cpus */ |
1857 | for_each_present_cpu(cpu) { | |
1858 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1859 | int cpustate = st->state; | |
1860 | ||
1861 | if (cpu >= failedcpu) | |
1862 | break; | |
1863 | ||
1864 | /* Did we invoke the startup call on that cpu ? */ | |
1865 | if (cpustate >= state) | |
cf392d10 | 1866 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
1867 | } |
1868 | } | |
1869 | ||
9805c673 TG |
1870 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
1871 | struct hlist_node *node, | |
1872 | bool invoke) | |
cf392d10 TG |
1873 | { |
1874 | struct cpuhp_step *sp; | |
1875 | int cpu; | |
1876 | int ret; | |
1877 | ||
9805c673 TG |
1878 | lockdep_assert_cpus_held(); |
1879 | ||
cf392d10 TG |
1880 | sp = cpuhp_get_step(state); |
1881 | if (sp->multi_instance == false) | |
1882 | return -EINVAL; | |
1883 | ||
dc434e05 | 1884 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 1885 | |
3c1627e9 | 1886 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
1887 | goto add_node; |
1888 | ||
1889 | /* | |
1890 | * Try to call the startup callback for each present cpu | |
1891 | * depending on the hotplug state of the cpu. | |
1892 | */ | |
1893 | for_each_present_cpu(cpu) { | |
1894 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1895 | int cpustate = st->state; | |
1896 | ||
1897 | if (cpustate < state) | |
1898 | continue; | |
1899 | ||
1900 | ret = cpuhp_issue_call(cpu, state, true, node); | |
1901 | if (ret) { | |
3c1627e9 | 1902 | if (sp->teardown.multi) |
cf392d10 | 1903 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 1904 | goto unlock; |
cf392d10 TG |
1905 | } |
1906 | } | |
1907 | add_node: | |
1908 | ret = 0; | |
cf392d10 | 1909 | hlist_add_head(node, &sp->list); |
dc434e05 | 1910 | unlock: |
cf392d10 | 1911 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
1912 | return ret; |
1913 | } | |
1914 | ||
1915 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
1916 | bool invoke) | |
1917 | { | |
1918 | int ret; | |
1919 | ||
1920 | cpus_read_lock(); | |
1921 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 1922 | cpus_read_unlock(); |
cf392d10 TG |
1923 | return ret; |
1924 | } | |
1925 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
1926 | ||
5b7aa87e | 1927 | /** |
71def423 | 1928 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 TG |
1929 | * @state: The state to setup |
1930 | * @invoke: If true, the startup function is invoked for cpus where | |
1931 | * cpu state >= @state | |
1932 | * @startup: startup callback function | |
1933 | * @teardown: teardown callback function | |
1934 | * @multi_instance: State is set up for multiple instances which get | |
1935 | * added afterwards. | |
5b7aa87e | 1936 | * |
71def423 | 1937 | * The caller needs to hold cpus read locked while calling this function. |
512f0980 BO |
1938 | * Returns: |
1939 | * On success: | |
1940 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN | |
1941 | * 0 for all other states | |
1942 | * On failure: proper (negative) error code | |
5b7aa87e | 1943 | */ |
71def423 SAS |
1944 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
1945 | const char *name, bool invoke, | |
1946 | int (*startup)(unsigned int cpu), | |
1947 | int (*teardown)(unsigned int cpu), | |
1948 | bool multi_instance) | |
5b7aa87e TG |
1949 | { |
1950 | int cpu, ret = 0; | |
b9d9d691 | 1951 | bool dynstate; |
5b7aa87e | 1952 | |
71def423 SAS |
1953 | lockdep_assert_cpus_held(); |
1954 | ||
5b7aa87e TG |
1955 | if (cpuhp_cb_check(state) || !name) |
1956 | return -EINVAL; | |
1957 | ||
dc434e05 | 1958 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 1959 | |
dc280d93 TG |
1960 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
1961 | multi_instance); | |
5b7aa87e | 1962 | |
b9d9d691 TG |
1963 | dynstate = state == CPUHP_AP_ONLINE_DYN; |
1964 | if (ret > 0 && dynstate) { | |
1965 | state = ret; | |
1966 | ret = 0; | |
1967 | } | |
1968 | ||
dc280d93 | 1969 | if (ret || !invoke || !startup) |
5b7aa87e TG |
1970 | goto out; |
1971 | ||
1972 | /* | |
1973 | * Try to call the startup callback for each present cpu | |
1974 | * depending on the hotplug state of the cpu. | |
1975 | */ | |
1976 | for_each_present_cpu(cpu) { | |
1977 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1978 | int cpustate = st->state; | |
1979 | ||
1980 | if (cpustate < state) | |
1981 | continue; | |
1982 | ||
cf392d10 | 1983 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 1984 | if (ret) { |
a724632c | 1985 | if (teardown) |
cf392d10 TG |
1986 | cpuhp_rollback_install(cpu, state, NULL); |
1987 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
1988 | goto out; |
1989 | } | |
1990 | } | |
1991 | out: | |
dc434e05 | 1992 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 TG |
1993 | /* |
1994 | * If the requested state is CPUHP_AP_ONLINE_DYN, return the | |
1995 | * dynamically allocated state in case of success. | |
1996 | */ | |
b9d9d691 | 1997 | if (!ret && dynstate) |
5b7aa87e TG |
1998 | return state; |
1999 | return ret; | |
2000 | } | |
71def423 SAS |
2001 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
2002 | ||
2003 | int __cpuhp_setup_state(enum cpuhp_state state, | |
2004 | const char *name, bool invoke, | |
2005 | int (*startup)(unsigned int cpu), | |
2006 | int (*teardown)(unsigned int cpu), | |
2007 | bool multi_instance) | |
2008 | { | |
2009 | int ret; | |
2010 | ||
2011 | cpus_read_lock(); | |
2012 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
2013 | teardown, multi_instance); | |
2014 | cpus_read_unlock(); | |
2015 | return ret; | |
2016 | } | |
5b7aa87e TG |
2017 | EXPORT_SYMBOL(__cpuhp_setup_state); |
2018 | ||
cf392d10 TG |
2019 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
2020 | struct hlist_node *node, bool invoke) | |
2021 | { | |
2022 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
2023 | int cpu; | |
2024 | ||
2025 | BUG_ON(cpuhp_cb_check(state)); | |
2026 | ||
2027 | if (!sp->multi_instance) | |
2028 | return -EINVAL; | |
2029 | ||
8f553c49 | 2030 | cpus_read_lock(); |
dc434e05 SAS |
2031 | mutex_lock(&cpuhp_state_mutex); |
2032 | ||
cf392d10 TG |
2033 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
2034 | goto remove; | |
2035 | /* | |
2036 | * Call the teardown callback for each present cpu depending | |
2037 | * on the hotplug state of the cpu. This function is not | |
2038 | * allowed to fail currently! | |
2039 | */ | |
2040 | for_each_present_cpu(cpu) { | |
2041 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2042 | int cpustate = st->state; | |
2043 | ||
2044 | if (cpustate >= state) | |
2045 | cpuhp_issue_call(cpu, state, false, node); | |
2046 | } | |
2047 | ||
2048 | remove: | |
cf392d10 TG |
2049 | hlist_del(node); |
2050 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 2051 | cpus_read_unlock(); |
cf392d10 TG |
2052 | |
2053 | return 0; | |
2054 | } | |
2055 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 2056 | |
5b7aa87e | 2057 | /** |
71def423 | 2058 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
2059 | * @state: The state to remove |
2060 | * @invoke: If true, the teardown function is invoked for cpus where | |
2061 | * cpu state >= @state | |
2062 | * | |
71def423 | 2063 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
2064 | * The teardown callback is currently not allowed to fail. Think |
2065 | * about module removal! | |
2066 | */ | |
71def423 | 2067 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 2068 | { |
cf392d10 | 2069 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
2070 | int cpu; |
2071 | ||
2072 | BUG_ON(cpuhp_cb_check(state)); | |
2073 | ||
71def423 | 2074 | lockdep_assert_cpus_held(); |
5b7aa87e | 2075 | |
dc434e05 | 2076 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
2077 | if (sp->multi_instance) { |
2078 | WARN(!hlist_empty(&sp->list), | |
2079 | "Error: Removing state %d which has instances left.\n", | |
2080 | state); | |
2081 | goto remove; | |
2082 | } | |
2083 | ||
a724632c | 2084 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
2085 | goto remove; |
2086 | ||
2087 | /* | |
2088 | * Call the teardown callback for each present cpu depending | |
2089 | * on the hotplug state of the cpu. This function is not | |
2090 | * allowed to fail currently! | |
2091 | */ | |
2092 | for_each_present_cpu(cpu) { | |
2093 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2094 | int cpustate = st->state; | |
2095 | ||
2096 | if (cpustate >= state) | |
cf392d10 | 2097 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
2098 | } |
2099 | remove: | |
cf392d10 | 2100 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 2101 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
2102 | } |
2103 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
2104 | ||
2105 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
2106 | { | |
2107 | cpus_read_lock(); | |
2108 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 2109 | cpus_read_unlock(); |
5b7aa87e TG |
2110 | } |
2111 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
2112 | ||
dc8d37ed AB |
2113 | #ifdef CONFIG_HOTPLUG_SMT |
2114 | static void cpuhp_offline_cpu_device(unsigned int cpu) | |
2115 | { | |
2116 | struct device *dev = get_cpu_device(cpu); | |
2117 | ||
2118 | dev->offline = true; | |
2119 | /* Tell user space about the state change */ | |
2120 | kobject_uevent(&dev->kobj, KOBJ_OFFLINE); | |
2121 | } | |
2122 | ||
2123 | static void cpuhp_online_cpu_device(unsigned int cpu) | |
2124 | { | |
2125 | struct device *dev = get_cpu_device(cpu); | |
2126 | ||
2127 | dev->offline = false; | |
2128 | /* Tell user space about the state change */ | |
2129 | kobject_uevent(&dev->kobj, KOBJ_ONLINE); | |
2130 | } | |
2131 | ||
2132 | int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval) | |
2133 | { | |
2134 | int cpu, ret = 0; | |
2135 | ||
2136 | cpu_maps_update_begin(); | |
2137 | for_each_online_cpu(cpu) { | |
2138 | if (topology_is_primary_thread(cpu)) | |
2139 | continue; | |
2140 | ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); | |
2141 | if (ret) | |
2142 | break; | |
2143 | /* | |
2144 | * As this needs to hold the cpu maps lock it's impossible | |
2145 | * to call device_offline() because that ends up calling | |
2146 | * cpu_down() which takes cpu maps lock. cpu maps lock | |
2147 | * needs to be held as this might race against in kernel | |
2148 | * abusers of the hotplug machinery (thermal management). | |
2149 | * | |
2150 | * So nothing would update device:offline state. That would | |
2151 | * leave the sysfs entry stale and prevent onlining after | |
2152 | * smt control has been changed to 'off' again. This is | |
2153 | * called under the sysfs hotplug lock, so it is properly | |
2154 | * serialized against the regular offline usage. | |
2155 | */ | |
2156 | cpuhp_offline_cpu_device(cpu); | |
2157 | } | |
2158 | if (!ret) | |
2159 | cpu_smt_control = ctrlval; | |
2160 | cpu_maps_update_done(); | |
2161 | return ret; | |
2162 | } | |
2163 | ||
2164 | int cpuhp_smt_enable(void) | |
2165 | { | |
2166 | int cpu, ret = 0; | |
2167 | ||
2168 | cpu_maps_update_begin(); | |
2169 | cpu_smt_control = CPU_SMT_ENABLED; | |
2170 | for_each_present_cpu(cpu) { | |
2171 | /* Skip online CPUs and CPUs on offline nodes */ | |
2172 | if (cpu_online(cpu) || !node_online(cpu_to_node(cpu))) | |
2173 | continue; | |
2174 | ret = _cpu_up(cpu, 0, CPUHP_ONLINE); | |
2175 | if (ret) | |
2176 | break; | |
2177 | /* See comment in cpuhp_smt_disable() */ | |
2178 | cpuhp_online_cpu_device(cpu); | |
2179 | } | |
2180 | cpu_maps_update_done(); | |
2181 | return ret; | |
2182 | } | |
2183 | #endif | |
2184 | ||
98f8cdce TG |
2185 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
2186 | static ssize_t show_cpuhp_state(struct device *dev, | |
2187 | struct device_attribute *attr, char *buf) | |
2188 | { | |
2189 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2190 | ||
2191 | return sprintf(buf, "%d\n", st->state); | |
2192 | } | |
2193 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
2194 | ||
757c989b TG |
2195 | static ssize_t write_cpuhp_target(struct device *dev, |
2196 | struct device_attribute *attr, | |
2197 | const char *buf, size_t count) | |
2198 | { | |
2199 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2200 | struct cpuhp_step *sp; | |
2201 | int target, ret; | |
2202 | ||
2203 | ret = kstrtoint(buf, 10, &target); | |
2204 | if (ret) | |
2205 | return ret; | |
2206 | ||
2207 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
2208 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
2209 | return -EINVAL; | |
2210 | #else | |
2211 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
2212 | return -EINVAL; | |
2213 | #endif | |
2214 | ||
2215 | ret = lock_device_hotplug_sysfs(); | |
2216 | if (ret) | |
2217 | return ret; | |
2218 | ||
2219 | mutex_lock(&cpuhp_state_mutex); | |
2220 | sp = cpuhp_get_step(target); | |
2221 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
2222 | mutex_unlock(&cpuhp_state_mutex); | |
2223 | if (ret) | |
40da1b11 | 2224 | goto out; |
757c989b TG |
2225 | |
2226 | if (st->state < target) | |
33c3736e | 2227 | ret = cpu_up(dev->id, target); |
757c989b | 2228 | else |
33c3736e | 2229 | ret = cpu_down(dev->id, target); |
40da1b11 | 2230 | out: |
757c989b TG |
2231 | unlock_device_hotplug(); |
2232 | return ret ? ret : count; | |
2233 | } | |
2234 | ||
98f8cdce TG |
2235 | static ssize_t show_cpuhp_target(struct device *dev, |
2236 | struct device_attribute *attr, char *buf) | |
2237 | { | |
2238 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2239 | ||
2240 | return sprintf(buf, "%d\n", st->target); | |
2241 | } | |
757c989b | 2242 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce | 2243 | |
1db49484 PZ |
2244 | |
2245 | static ssize_t write_cpuhp_fail(struct device *dev, | |
2246 | struct device_attribute *attr, | |
2247 | const char *buf, size_t count) | |
2248 | { | |
2249 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2250 | struct cpuhp_step *sp; | |
2251 | int fail, ret; | |
2252 | ||
2253 | ret = kstrtoint(buf, 10, &fail); | |
2254 | if (ret) | |
2255 | return ret; | |
2256 | ||
3ae70c25 VD |
2257 | if (fail == CPUHP_INVALID) { |
2258 | st->fail = fail; | |
2259 | return count; | |
2260 | } | |
2261 | ||
33d4a5a7 ET |
2262 | if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE) |
2263 | return -EINVAL; | |
2264 | ||
1db49484 PZ |
2265 | /* |
2266 | * Cannot fail STARTING/DYING callbacks. | |
2267 | */ | |
2268 | if (cpuhp_is_atomic_state(fail)) | |
2269 | return -EINVAL; | |
2270 | ||
62f25069 VD |
2271 | /* |
2272 | * DEAD callbacks cannot fail... | |
2273 | * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter | |
2274 | * triggering STARTING callbacks, a failure in this state would | |
2275 | * hinder rollback. | |
2276 | */ | |
2277 | if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU) | |
2278 | return -EINVAL; | |
2279 | ||
1db49484 PZ |
2280 | /* |
2281 | * Cannot fail anything that doesn't have callbacks. | |
2282 | */ | |
2283 | mutex_lock(&cpuhp_state_mutex); | |
2284 | sp = cpuhp_get_step(fail); | |
2285 | if (!sp->startup.single && !sp->teardown.single) | |
2286 | ret = -EINVAL; | |
2287 | mutex_unlock(&cpuhp_state_mutex); | |
2288 | if (ret) | |
2289 | return ret; | |
2290 | ||
2291 | st->fail = fail; | |
2292 | ||
2293 | return count; | |
2294 | } | |
2295 | ||
2296 | static ssize_t show_cpuhp_fail(struct device *dev, | |
2297 | struct device_attribute *attr, char *buf) | |
2298 | { | |
2299 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2300 | ||
2301 | return sprintf(buf, "%d\n", st->fail); | |
2302 | } | |
2303 | ||
2304 | static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail); | |
2305 | ||
98f8cdce TG |
2306 | static struct attribute *cpuhp_cpu_attrs[] = { |
2307 | &dev_attr_state.attr, | |
2308 | &dev_attr_target.attr, | |
1db49484 | 2309 | &dev_attr_fail.attr, |
98f8cdce TG |
2310 | NULL |
2311 | }; | |
2312 | ||
993647a2 | 2313 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
2314 | .attrs = cpuhp_cpu_attrs, |
2315 | .name = "hotplug", | |
2316 | NULL | |
2317 | }; | |
2318 | ||
2319 | static ssize_t show_cpuhp_states(struct device *dev, | |
2320 | struct device_attribute *attr, char *buf) | |
2321 | { | |
2322 | ssize_t cur, res = 0; | |
2323 | int i; | |
2324 | ||
2325 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 2326 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
2327 | struct cpuhp_step *sp = cpuhp_get_step(i); |
2328 | ||
2329 | if (sp->name) { | |
2330 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
2331 | buf += cur; | |
2332 | res += cur; | |
2333 | } | |
2334 | } | |
2335 | mutex_unlock(&cpuhp_state_mutex); | |
2336 | return res; | |
2337 | } | |
2338 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
2339 | ||
2340 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
2341 | &dev_attr_states.attr, | |
2342 | NULL | |
2343 | }; | |
2344 | ||
993647a2 | 2345 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
2346 | .attrs = cpuhp_cpu_root_attrs, |
2347 | .name = "hotplug", | |
2348 | NULL | |
2349 | }; | |
2350 | ||
05736e4a TG |
2351 | #ifdef CONFIG_HOTPLUG_SMT |
2352 | ||
05736e4a | 2353 | static ssize_t |
de7b77e5 JP |
2354 | __store_smt_control(struct device *dev, struct device_attribute *attr, |
2355 | const char *buf, size_t count) | |
05736e4a TG |
2356 | { |
2357 | int ctrlval, ret; | |
2358 | ||
2359 | if (sysfs_streq(buf, "on")) | |
2360 | ctrlval = CPU_SMT_ENABLED; | |
2361 | else if (sysfs_streq(buf, "off")) | |
2362 | ctrlval = CPU_SMT_DISABLED; | |
2363 | else if (sysfs_streq(buf, "forceoff")) | |
2364 | ctrlval = CPU_SMT_FORCE_DISABLED; | |
2365 | else | |
2366 | return -EINVAL; | |
2367 | ||
2368 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED) | |
2369 | return -EPERM; | |
2370 | ||
2371 | if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
2372 | return -ENODEV; | |
2373 | ||
2374 | ret = lock_device_hotplug_sysfs(); | |
2375 | if (ret) | |
2376 | return ret; | |
2377 | ||
2378 | if (ctrlval != cpu_smt_control) { | |
2379 | switch (ctrlval) { | |
2380 | case CPU_SMT_ENABLED: | |
215af549 | 2381 | ret = cpuhp_smt_enable(); |
05736e4a TG |
2382 | break; |
2383 | case CPU_SMT_DISABLED: | |
2384 | case CPU_SMT_FORCE_DISABLED: | |
2385 | ret = cpuhp_smt_disable(ctrlval); | |
2386 | break; | |
2387 | } | |
2388 | } | |
2389 | ||
2390 | unlock_device_hotplug(); | |
2391 | return ret ? ret : count; | |
2392 | } | |
de7b77e5 JP |
2393 | |
2394 | #else /* !CONFIG_HOTPLUG_SMT */ | |
2395 | static ssize_t | |
2396 | __store_smt_control(struct device *dev, struct device_attribute *attr, | |
2397 | const char *buf, size_t count) | |
2398 | { | |
2399 | return -ENODEV; | |
2400 | } | |
2401 | #endif /* CONFIG_HOTPLUG_SMT */ | |
2402 | ||
2403 | static const char *smt_states[] = { | |
2404 | [CPU_SMT_ENABLED] = "on", | |
2405 | [CPU_SMT_DISABLED] = "off", | |
2406 | [CPU_SMT_FORCE_DISABLED] = "forceoff", | |
2407 | [CPU_SMT_NOT_SUPPORTED] = "notsupported", | |
2408 | [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented", | |
2409 | }; | |
2410 | ||
2411 | static ssize_t | |
2412 | show_smt_control(struct device *dev, struct device_attribute *attr, char *buf) | |
2413 | { | |
2414 | const char *state = smt_states[cpu_smt_control]; | |
2415 | ||
2416 | return snprintf(buf, PAGE_SIZE - 2, "%s\n", state); | |
2417 | } | |
2418 | ||
2419 | static ssize_t | |
2420 | store_smt_control(struct device *dev, struct device_attribute *attr, | |
2421 | const char *buf, size_t count) | |
2422 | { | |
2423 | return __store_smt_control(dev, attr, buf, count); | |
2424 | } | |
05736e4a TG |
2425 | static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control); |
2426 | ||
2427 | static ssize_t | |
2428 | show_smt_active(struct device *dev, struct device_attribute *attr, char *buf) | |
2429 | { | |
de7b77e5 | 2430 | return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active()); |
05736e4a TG |
2431 | } |
2432 | static DEVICE_ATTR(active, 0444, show_smt_active, NULL); | |
2433 | ||
2434 | static struct attribute *cpuhp_smt_attrs[] = { | |
2435 | &dev_attr_control.attr, | |
2436 | &dev_attr_active.attr, | |
2437 | NULL | |
2438 | }; | |
2439 | ||
2440 | static const struct attribute_group cpuhp_smt_attr_group = { | |
2441 | .attrs = cpuhp_smt_attrs, | |
2442 | .name = "smt", | |
2443 | NULL | |
2444 | }; | |
2445 | ||
de7b77e5 | 2446 | static int __init cpu_smt_sysfs_init(void) |
05736e4a | 2447 | { |
05736e4a TG |
2448 | return sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2449 | &cpuhp_smt_attr_group); | |
2450 | } | |
2451 | ||
98f8cdce TG |
2452 | static int __init cpuhp_sysfs_init(void) |
2453 | { | |
2454 | int cpu, ret; | |
2455 | ||
de7b77e5 | 2456 | ret = cpu_smt_sysfs_init(); |
05736e4a TG |
2457 | if (ret) |
2458 | return ret; | |
2459 | ||
98f8cdce TG |
2460 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, |
2461 | &cpuhp_cpu_root_attr_group); | |
2462 | if (ret) | |
2463 | return ret; | |
2464 | ||
2465 | for_each_possible_cpu(cpu) { | |
2466 | struct device *dev = get_cpu_device(cpu); | |
2467 | ||
2468 | if (!dev) | |
2469 | continue; | |
2470 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
2471 | if (ret) | |
2472 | return ret; | |
2473 | } | |
2474 | return 0; | |
2475 | } | |
2476 | device_initcall(cpuhp_sysfs_init); | |
de7b77e5 | 2477 | #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */ |
98f8cdce | 2478 | |
e56b3bc7 LT |
2479 | /* |
2480 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
2481 | * represents all NR_CPUS bits binary values of 1<<nr. | |
2482 | * | |
e0b582ec | 2483 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
2484 | * mask value that has a single bit set only. |
2485 | */ | |
b8d317d1 | 2486 | |
e56b3bc7 | 2487 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 2488 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
2489 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
2490 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
2491 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 2492 | |
e56b3bc7 LT |
2493 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
2494 | ||
2495 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
2496 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
2497 | #if BITS_PER_LONG > 32 | |
2498 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
2499 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
2500 | #endif |
2501 | }; | |
e56b3bc7 | 2502 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
2503 | |
2504 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
2505 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
2506 | |
2507 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 2508 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 2509 | = {CPU_BITS_ALL}; |
b3199c02 | 2510 | #else |
4b804c85 | 2511 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 2512 | #endif |
4b804c85 | 2513 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 2514 | |
4b804c85 RV |
2515 | struct cpumask __cpu_online_mask __read_mostly; |
2516 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 2517 | |
4b804c85 RV |
2518 | struct cpumask __cpu_present_mask __read_mostly; |
2519 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 2520 | |
4b804c85 RV |
2521 | struct cpumask __cpu_active_mask __read_mostly; |
2522 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 2523 | |
e40f74c5 PZ |
2524 | struct cpumask __cpu_dying_mask __read_mostly; |
2525 | EXPORT_SYMBOL(__cpu_dying_mask); | |
2526 | ||
0c09ab96 TG |
2527 | atomic_t __num_online_cpus __read_mostly; |
2528 | EXPORT_SYMBOL(__num_online_cpus); | |
2529 | ||
3fa41520 RR |
2530 | void init_cpu_present(const struct cpumask *src) |
2531 | { | |
c4c54dd1 | 2532 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
2533 | } |
2534 | ||
2535 | void init_cpu_possible(const struct cpumask *src) | |
2536 | { | |
c4c54dd1 | 2537 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
2538 | } |
2539 | ||
2540 | void init_cpu_online(const struct cpumask *src) | |
2541 | { | |
c4c54dd1 | 2542 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 2543 | } |
cff7d378 | 2544 | |
0c09ab96 TG |
2545 | void set_cpu_online(unsigned int cpu, bool online) |
2546 | { | |
2547 | /* | |
2548 | * atomic_inc/dec() is required to handle the horrid abuse of this | |
2549 | * function by the reboot and kexec code which invoke it from | |
2550 | * IPI/NMI broadcasts when shutting down CPUs. Invocation from | |
2551 | * regular CPU hotplug is properly serialized. | |
2552 | * | |
2553 | * Note, that the fact that __num_online_cpus is of type atomic_t | |
2554 | * does not protect readers which are not serialized against | |
2555 | * concurrent hotplug operations. | |
2556 | */ | |
2557 | if (online) { | |
2558 | if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask)) | |
2559 | atomic_inc(&__num_online_cpus); | |
2560 | } else { | |
2561 | if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask)) | |
2562 | atomic_dec(&__num_online_cpus); | |
2563 | } | |
2564 | } | |
2565 | ||
cff7d378 TG |
2566 | /* |
2567 | * Activate the first processor. | |
2568 | */ | |
2569 | void __init boot_cpu_init(void) | |
2570 | { | |
2571 | int cpu = smp_processor_id(); | |
2572 | ||
2573 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
2574 | set_cpu_online(cpu, true); | |
2575 | set_cpu_active(cpu, true); | |
2576 | set_cpu_present(cpu, true); | |
2577 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
2578 | |
2579 | #ifdef CONFIG_SMP | |
2580 | __boot_cpu_id = cpu; | |
2581 | #endif | |
cff7d378 TG |
2582 | } |
2583 | ||
2584 | /* | |
2585 | * Must be called _AFTER_ setting up the per_cpu areas | |
2586 | */ | |
b5b1404d | 2587 | void __init boot_cpu_hotplug_init(void) |
cff7d378 | 2588 | { |
269777aa | 2589 | #ifdef CONFIG_SMP |
e797bda3 | 2590 | cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask); |
269777aa | 2591 | #endif |
0cc3cd21 | 2592 | this_cpu_write(cpuhp_state.state, CPUHP_ONLINE); |
cff7d378 | 2593 | } |
98af8452 | 2594 | |
731dc9df TH |
2595 | /* |
2596 | * These are used for a global "mitigations=" cmdline option for toggling | |
2597 | * optional CPU mitigations. | |
2598 | */ | |
2599 | enum cpu_mitigations { | |
2600 | CPU_MITIGATIONS_OFF, | |
2601 | CPU_MITIGATIONS_AUTO, | |
2602 | CPU_MITIGATIONS_AUTO_NOSMT, | |
2603 | }; | |
2604 | ||
2605 | static enum cpu_mitigations cpu_mitigations __ro_after_init = | |
2606 | CPU_MITIGATIONS_AUTO; | |
98af8452 JP |
2607 | |
2608 | static int __init mitigations_parse_cmdline(char *arg) | |
2609 | { | |
2610 | if (!strcmp(arg, "off")) | |
2611 | cpu_mitigations = CPU_MITIGATIONS_OFF; | |
2612 | else if (!strcmp(arg, "auto")) | |
2613 | cpu_mitigations = CPU_MITIGATIONS_AUTO; | |
2614 | else if (!strcmp(arg, "auto,nosmt")) | |
2615 | cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT; | |
1bf72720 GU |
2616 | else |
2617 | pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n", | |
2618 | arg); | |
98af8452 JP |
2619 | |
2620 | return 0; | |
2621 | } | |
2622 | early_param("mitigations", mitigations_parse_cmdline); | |
731dc9df TH |
2623 | |
2624 | /* mitigations=off */ | |
2625 | bool cpu_mitigations_off(void) | |
2626 | { | |
2627 | return cpu_mitigations == CPU_MITIGATIONS_OFF; | |
2628 | } | |
2629 | EXPORT_SYMBOL_GPL(cpu_mitigations_off); | |
2630 | ||
2631 | /* mitigations=auto,nosmt */ | |
2632 | bool cpu_mitigations_auto_nosmt(void) | |
2633 | { | |
2634 | return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT; | |
2635 | } | |
2636 | EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt); |