]>
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> | |
6f062123 | 20 | #include <linux/delay.h> |
9984de1a | 21 | #include <linux/export.h> |
e4cc2f87 | 22 | #include <linux/bug.h> |
1da177e4 LT |
23 | #include <linux/kthread.h> |
24 | #include <linux/stop_machine.h> | |
81615b62 | 25 | #include <linux/mutex.h> |
5a0e3ad6 | 26 | #include <linux/gfp.h> |
79cfbdfa | 27 | #include <linux/suspend.h> |
a19423b9 | 28 | #include <linux/lockdep.h> |
345527b1 | 29 | #include <linux/tick.h> |
a8994181 | 30 | #include <linux/irq.h> |
941154bd | 31 | #include <linux/nmi.h> |
4cb28ced | 32 | #include <linux/smpboot.h> |
e6d4989a | 33 | #include <linux/relay.h> |
6731d4f1 | 34 | #include <linux/slab.h> |
dce1ca05 | 35 | #include <linux/scs.h> |
fc8dffd3 | 36 | #include <linux/percpu-rwsem.h> |
b22afcdf | 37 | #include <linux/cpuset.h> |
3191dd5a | 38 | #include <linux/random.h> |
bae1a962 | 39 | #include <linux/cc_platform.h> |
cff7d378 | 40 | |
bb3632c6 | 41 | #include <trace/events/power.h> |
cff7d378 TG |
42 | #define CREATE_TRACE_POINTS |
43 | #include <trace/events/cpuhp.h> | |
1da177e4 | 44 | |
38498a67 TG |
45 | #include "smpboot.h" |
46 | ||
cff7d378 | 47 | /** |
11bc021d | 48 | * struct cpuhp_cpu_state - Per cpu hotplug state storage |
cff7d378 TG |
49 | * @state: The current cpu state |
50 | * @target: The target state | |
11bc021d | 51 | * @fail: Current CPU hotplug callback state |
4cb28ced TG |
52 | * @thread: Pointer to the hotplug thread |
53 | * @should_run: Thread should execute | |
3b9d6da6 | 54 | * @rollback: Perform a rollback |
a724632c TG |
55 | * @single: Single callback invocation |
56 | * @bringup: Single callback bringup or teardown selector | |
11bc021d RD |
57 | * @node: Remote CPU node; for multi-instance, do a |
58 | * single entry callback for install/remove | |
59 | * @last: For multi-instance rollback, remember how far we got | |
a724632c | 60 | * @cb_state: The state for a single callback (install/uninstall) |
4cb28ced | 61 | * @result: Result of the operation |
6f062123 | 62 | * @ap_sync_state: State for AP synchronization |
5ebe7742 PZ |
63 | * @done_up: Signal completion to the issuer of the task for cpu-up |
64 | * @done_down: Signal completion to the issuer of the task for cpu-down | |
cff7d378 TG |
65 | */ |
66 | struct cpuhp_cpu_state { | |
67 | enum cpuhp_state state; | |
68 | enum cpuhp_state target; | |
1db49484 | 69 | enum cpuhp_state fail; |
4cb28ced TG |
70 | #ifdef CONFIG_SMP |
71 | struct task_struct *thread; | |
72 | bool should_run; | |
3b9d6da6 | 73 | bool rollback; |
a724632c TG |
74 | bool single; |
75 | bool bringup; | |
cf392d10 | 76 | struct hlist_node *node; |
4dddfb5f | 77 | struct hlist_node *last; |
4cb28ced | 78 | enum cpuhp_state cb_state; |
4cb28ced | 79 | int result; |
6f062123 | 80 | atomic_t ap_sync_state; |
5ebe7742 PZ |
81 | struct completion done_up; |
82 | struct completion done_down; | |
4cb28ced | 83 | #endif |
cff7d378 TG |
84 | }; |
85 | ||
1db49484 PZ |
86 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = { |
87 | .fail = CPUHP_INVALID, | |
88 | }; | |
cff7d378 | 89 | |
e797bda3 TG |
90 | #ifdef CONFIG_SMP |
91 | cpumask_t cpus_booted_once_mask; | |
92 | #endif | |
93 | ||
49dfe2a6 | 94 | #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP) |
5f4b55e1 PZ |
95 | static struct lockdep_map cpuhp_state_up_map = |
96 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map); | |
97 | static struct lockdep_map cpuhp_state_down_map = | |
98 | STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map); | |
99 | ||
100 | ||
76dc6c09 | 101 | static inline void cpuhp_lock_acquire(bool bringup) |
5f4b55e1 PZ |
102 | { |
103 | lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
104 | } | |
105 | ||
76dc6c09 | 106 | static inline void cpuhp_lock_release(bool bringup) |
5f4b55e1 PZ |
107 | { |
108 | lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map); | |
109 | } | |
110 | #else | |
111 | ||
76dc6c09 MM |
112 | static inline void cpuhp_lock_acquire(bool bringup) { } |
113 | static inline void cpuhp_lock_release(bool bringup) { } | |
5f4b55e1 | 114 | |
49dfe2a6 TG |
115 | #endif |
116 | ||
cff7d378 | 117 | /** |
11bc021d | 118 | * struct cpuhp_step - Hotplug state machine step |
cff7d378 TG |
119 | * @name: Name of the step |
120 | * @startup: Startup function of the step | |
121 | * @teardown: Teardown function of the step | |
757c989b | 122 | * @cant_stop: Bringup/teardown can't be stopped at this step |
11bc021d | 123 | * @multi_instance: State has multiple instances which get added afterwards |
cff7d378 TG |
124 | */ |
125 | struct cpuhp_step { | |
cf392d10 TG |
126 | const char *name; |
127 | union { | |
3c1627e9 TG |
128 | int (*single)(unsigned int cpu); |
129 | int (*multi)(unsigned int cpu, | |
130 | struct hlist_node *node); | |
131 | } startup; | |
cf392d10 | 132 | union { |
3c1627e9 TG |
133 | int (*single)(unsigned int cpu); |
134 | int (*multi)(unsigned int cpu, | |
135 | struct hlist_node *node); | |
136 | } teardown; | |
11bc021d | 137 | /* private: */ |
cf392d10 | 138 | struct hlist_head list; |
11bc021d | 139 | /* public: */ |
cf392d10 TG |
140 | bool cant_stop; |
141 | bool multi_instance; | |
cff7d378 TG |
142 | }; |
143 | ||
98f8cdce | 144 | static DEFINE_MUTEX(cpuhp_state_mutex); |
17a2f1ce | 145 | static struct cpuhp_step cpuhp_hp_states[]; |
cff7d378 | 146 | |
a724632c TG |
147 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) |
148 | { | |
17a2f1ce | 149 | return cpuhp_hp_states + state; |
a724632c TG |
150 | } |
151 | ||
453e4108 VD |
152 | static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step) |
153 | { | |
154 | return bringup ? !step->startup.single : !step->teardown.single; | |
155 | } | |
156 | ||
cff7d378 | 157 | /** |
11bc021d | 158 | * cpuhp_invoke_callback - Invoke the callbacks for a given state |
cff7d378 | 159 | * @cpu: The cpu for which the callback should be invoked |
96abb968 | 160 | * @state: The state to do callbacks for |
a724632c | 161 | * @bringup: True if the bringup callback should be invoked |
96abb968 PZ |
162 | * @node: For multi-instance, do a single entry callback for install/remove |
163 | * @lastp: For multi-instance rollback, remember how far we got | |
cff7d378 | 164 | * |
cf392d10 | 165 | * Called from cpu hotplug and from the state register machinery. |
11bc021d RD |
166 | * |
167 | * Return: %0 on success or a negative errno code | |
cff7d378 | 168 | */ |
a724632c | 169 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state, |
96abb968 PZ |
170 | bool bringup, struct hlist_node *node, |
171 | struct hlist_node **lastp) | |
cff7d378 TG |
172 | { |
173 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
a724632c | 174 | struct cpuhp_step *step = cpuhp_get_step(state); |
cf392d10 TG |
175 | int (*cbm)(unsigned int cpu, struct hlist_node *node); |
176 | int (*cb)(unsigned int cpu); | |
177 | int ret, cnt; | |
178 | ||
1db49484 PZ |
179 | if (st->fail == state) { |
180 | st->fail = CPUHP_INVALID; | |
1db49484 PZ |
181 | return -EAGAIN; |
182 | } | |
183 | ||
453e4108 VD |
184 | if (cpuhp_step_empty(bringup, step)) { |
185 | WARN_ON_ONCE(1); | |
186 | return 0; | |
187 | } | |
188 | ||
cf392d10 | 189 | if (!step->multi_instance) { |
96abb968 | 190 | WARN_ON_ONCE(lastp && *lastp); |
3c1627e9 | 191 | cb = bringup ? step->startup.single : step->teardown.single; |
453e4108 | 192 | |
a724632c | 193 | trace_cpuhp_enter(cpu, st->target, state, cb); |
cff7d378 | 194 | ret = cb(cpu); |
a724632c | 195 | trace_cpuhp_exit(cpu, st->state, state, ret); |
cf392d10 TG |
196 | return ret; |
197 | } | |
3c1627e9 | 198 | cbm = bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
199 | |
200 | /* Single invocation for instance add/remove */ | |
201 | if (node) { | |
96abb968 | 202 | WARN_ON_ONCE(lastp && *lastp); |
cf392d10 TG |
203 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
204 | ret = cbm(cpu, node); | |
205 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
206 | return ret; | |
207 | } | |
208 | ||
209 | /* State transition. Invoke on all instances */ | |
210 | cnt = 0; | |
211 | hlist_for_each(node, &step->list) { | |
96abb968 PZ |
212 | if (lastp && node == *lastp) |
213 | break; | |
214 | ||
cf392d10 TG |
215 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); |
216 | ret = cbm(cpu, node); | |
217 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
96abb968 PZ |
218 | if (ret) { |
219 | if (!lastp) | |
220 | goto err; | |
221 | ||
222 | *lastp = node; | |
223 | return ret; | |
224 | } | |
cf392d10 TG |
225 | cnt++; |
226 | } | |
96abb968 PZ |
227 | if (lastp) |
228 | *lastp = NULL; | |
cf392d10 TG |
229 | return 0; |
230 | err: | |
231 | /* Rollback the instances if one failed */ | |
3c1627e9 | 232 | cbm = !bringup ? step->startup.multi : step->teardown.multi; |
cf392d10 TG |
233 | if (!cbm) |
234 | return ret; | |
235 | ||
236 | hlist_for_each(node, &step->list) { | |
237 | if (!cnt--) | |
238 | break; | |
724a8688 PZ |
239 | |
240 | trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node); | |
241 | ret = cbm(cpu, node); | |
242 | trace_cpuhp_exit(cpu, st->state, state, ret); | |
243 | /* | |
244 | * Rollback must not fail, | |
245 | */ | |
246 | WARN_ON_ONCE(ret); | |
cff7d378 TG |
247 | } |
248 | return ret; | |
249 | } | |
250 | ||
98a79d6a | 251 | #ifdef CONFIG_SMP |
fcb3029a AB |
252 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
253 | { | |
254 | /* | |
255 | * The extra check for CPUHP_TEARDOWN_CPU is only for documentation | |
256 | * purposes as that state is handled explicitly in cpu_down. | |
257 | */ | |
258 | return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU; | |
259 | } | |
260 | ||
5ebe7742 PZ |
261 | static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup) |
262 | { | |
263 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
264 | wait_for_completion(done); | |
265 | } | |
266 | ||
267 | static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup) | |
268 | { | |
269 | struct completion *done = bringup ? &st->done_up : &st->done_down; | |
270 | complete(done); | |
271 | } | |
272 | ||
273 | /* | |
274 | * The former STARTING/DYING states, ran with IRQs disabled and must not fail. | |
275 | */ | |
276 | static bool cpuhp_is_atomic_state(enum cpuhp_state state) | |
277 | { | |
278 | return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE; | |
279 | } | |
280 | ||
6f062123 TG |
281 | /* Synchronization state management */ |
282 | enum cpuhp_sync_state { | |
283 | SYNC_STATE_DEAD, | |
284 | SYNC_STATE_KICKED, | |
285 | SYNC_STATE_SHOULD_DIE, | |
286 | SYNC_STATE_ALIVE, | |
287 | SYNC_STATE_SHOULD_ONLINE, | |
288 | SYNC_STATE_ONLINE, | |
289 | }; | |
290 | ||
291 | #ifdef CONFIG_HOTPLUG_CORE_SYNC | |
292 | /** | |
293 | * cpuhp_ap_update_sync_state - Update synchronization state during bringup/teardown | |
294 | * @state: The synchronization state to set | |
295 | * | |
296 | * No synchronization point. Just update of the synchronization state, but implies | |
297 | * a full barrier so that the AP changes are visible before the control CPU proceeds. | |
298 | */ | |
299 | static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state) | |
300 | { | |
301 | atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state); | |
302 | ||
303 | (void)atomic_xchg(st, state); | |
304 | } | |
305 | ||
306 | void __weak arch_cpuhp_sync_state_poll(void) { cpu_relax(); } | |
307 | ||
308 | static bool cpuhp_wait_for_sync_state(unsigned int cpu, enum cpuhp_sync_state state, | |
309 | enum cpuhp_sync_state next_state) | |
310 | { | |
311 | atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu); | |
312 | ktime_t now, end, start = ktime_get(); | |
313 | int sync; | |
314 | ||
315 | end = start + 10ULL * NSEC_PER_SEC; | |
316 | ||
317 | sync = atomic_read(st); | |
318 | while (1) { | |
319 | if (sync == state) { | |
320 | if (!atomic_try_cmpxchg(st, &sync, next_state)) | |
321 | continue; | |
322 | return true; | |
323 | } | |
324 | ||
325 | now = ktime_get(); | |
326 | if (now > end) { | |
327 | /* Timeout. Leave the state unchanged */ | |
328 | return false; | |
329 | } else if (now - start < NSEC_PER_MSEC) { | |
330 | /* Poll for one millisecond */ | |
331 | arch_cpuhp_sync_state_poll(); | |
332 | } else { | |
333 | usleep_range_state(USEC_PER_MSEC, 2 * USEC_PER_MSEC, TASK_UNINTERRUPTIBLE); | |
334 | } | |
335 | sync = atomic_read(st); | |
336 | } | |
337 | return true; | |
338 | } | |
339 | #else /* CONFIG_HOTPLUG_CORE_SYNC */ | |
340 | static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state) { } | |
341 | #endif /* !CONFIG_HOTPLUG_CORE_SYNC */ | |
342 | ||
343 | #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD | |
344 | /** | |
345 | * cpuhp_ap_report_dead - Update synchronization state to DEAD | |
346 | * | |
347 | * No synchronization point. Just update of the synchronization state. | |
348 | */ | |
349 | void cpuhp_ap_report_dead(void) | |
350 | { | |
351 | cpuhp_ap_update_sync_state(SYNC_STATE_DEAD); | |
352 | } | |
353 | ||
354 | void __weak arch_cpuhp_cleanup_dead_cpu(unsigned int cpu) { } | |
355 | ||
356 | /* | |
357 | * Late CPU shutdown synchronization point. Cannot use cpuhp_state::done_down | |
358 | * because the AP cannot issue complete() at this stage. | |
359 | */ | |
360 | static void cpuhp_bp_sync_dead(unsigned int cpu) | |
361 | { | |
362 | atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu); | |
363 | int sync = atomic_read(st); | |
364 | ||
365 | do { | |
366 | /* CPU can have reported dead already. Don't overwrite that! */ | |
367 | if (sync == SYNC_STATE_DEAD) | |
368 | break; | |
369 | } while (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_SHOULD_DIE)); | |
370 | ||
371 | if (cpuhp_wait_for_sync_state(cpu, SYNC_STATE_DEAD, SYNC_STATE_DEAD)) { | |
372 | /* CPU reached dead state. Invoke the cleanup function */ | |
373 | arch_cpuhp_cleanup_dead_cpu(cpu); | |
374 | return; | |
375 | } | |
376 | ||
377 | /* No further action possible. Emit message and give up. */ | |
378 | pr_err("CPU%u failed to report dead state\n", cpu); | |
379 | } | |
380 | #else /* CONFIG_HOTPLUG_CORE_SYNC_DEAD */ | |
381 | static inline void cpuhp_bp_sync_dead(unsigned int cpu) { } | |
382 | #endif /* !CONFIG_HOTPLUG_CORE_SYNC_DEAD */ | |
383 | ||
384 | #ifdef CONFIG_HOTPLUG_CORE_SYNC_FULL | |
385 | /** | |
386 | * cpuhp_ap_sync_alive - Synchronize AP with the control CPU once it is alive | |
387 | * | |
388 | * Updates the AP synchronization state to SYNC_STATE_ALIVE and waits | |
389 | * for the BP to release it. | |
390 | */ | |
391 | void cpuhp_ap_sync_alive(void) | |
392 | { | |
393 | atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state); | |
394 | ||
395 | cpuhp_ap_update_sync_state(SYNC_STATE_ALIVE); | |
396 | ||
397 | /* Wait for the control CPU to release it. */ | |
398 | while (atomic_read(st) != SYNC_STATE_SHOULD_ONLINE) | |
399 | cpu_relax(); | |
400 | } | |
401 | ||
402 | static bool cpuhp_can_boot_ap(unsigned int cpu) | |
403 | { | |
404 | atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu); | |
405 | int sync = atomic_read(st); | |
406 | ||
407 | again: | |
408 | switch (sync) { | |
409 | case SYNC_STATE_DEAD: | |
410 | /* CPU is properly dead */ | |
411 | break; | |
412 | case SYNC_STATE_KICKED: | |
413 | /* CPU did not come up in previous attempt */ | |
414 | break; | |
415 | case SYNC_STATE_ALIVE: | |
416 | /* CPU is stuck cpuhp_ap_sync_alive(). */ | |
417 | break; | |
418 | default: | |
419 | /* CPU failed to report online or dead and is in limbo state. */ | |
420 | return false; | |
421 | } | |
422 | ||
423 | /* Prepare for booting */ | |
424 | if (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_KICKED)) | |
425 | goto again; | |
426 | ||
427 | return true; | |
428 | } | |
429 | ||
430 | void __weak arch_cpuhp_cleanup_kick_cpu(unsigned int cpu) { } | |
431 | ||
432 | /* | |
433 | * Early CPU bringup synchronization point. Cannot use cpuhp_state::done_up | |
434 | * because the AP cannot issue complete() so early in the bringup. | |
435 | */ | |
436 | static int cpuhp_bp_sync_alive(unsigned int cpu) | |
437 | { | |
438 | int ret = 0; | |
439 | ||
440 | if (!IS_ENABLED(CONFIG_HOTPLUG_CORE_SYNC_FULL)) | |
441 | return 0; | |
442 | ||
443 | if (!cpuhp_wait_for_sync_state(cpu, SYNC_STATE_ALIVE, SYNC_STATE_SHOULD_ONLINE)) { | |
444 | pr_err("CPU%u failed to report alive state\n", cpu); | |
445 | ret = -EIO; | |
446 | } | |
447 | ||
448 | /* Let the architecture cleanup the kick alive mechanics. */ | |
449 | arch_cpuhp_cleanup_kick_cpu(cpu); | |
450 | return ret; | |
451 | } | |
452 | #else /* CONFIG_HOTPLUG_CORE_SYNC_FULL */ | |
453 | static inline int cpuhp_bp_sync_alive(unsigned int cpu) { return 0; } | |
454 | static inline bool cpuhp_can_boot_ap(unsigned int cpu) { return true; } | |
455 | #endif /* !CONFIG_HOTPLUG_CORE_SYNC_FULL */ | |
456 | ||
b3199c02 | 457 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 458 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
459 | bool cpuhp_tasks_frozen; |
460 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 461 | |
79a6cdeb | 462 | /* |
93ae4f97 SB |
463 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
464 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
79a6cdeb LJ |
465 | */ |
466 | void cpu_maps_update_begin(void) | |
467 | { | |
468 | mutex_lock(&cpu_add_remove_lock); | |
469 | } | |
470 | ||
471 | void cpu_maps_update_done(void) | |
472 | { | |
473 | mutex_unlock(&cpu_add_remove_lock); | |
474 | } | |
1da177e4 | 475 | |
fc8dffd3 TG |
476 | /* |
477 | * If set, cpu_up and cpu_down will return -EBUSY and do nothing. | |
e3920fb4 RW |
478 | * Should always be manipulated under cpu_add_remove_lock |
479 | */ | |
480 | static int cpu_hotplug_disabled; | |
481 | ||
79a6cdeb LJ |
482 | #ifdef CONFIG_HOTPLUG_CPU |
483 | ||
fc8dffd3 | 484 | DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock); |
a19423b9 | 485 | |
8f553c49 | 486 | void cpus_read_lock(void) |
a9d9baa1 | 487 | { |
fc8dffd3 | 488 | percpu_down_read(&cpu_hotplug_lock); |
a9d9baa1 | 489 | } |
8f553c49 | 490 | EXPORT_SYMBOL_GPL(cpus_read_lock); |
90d45d17 | 491 | |
6f4ceee9 WL |
492 | int cpus_read_trylock(void) |
493 | { | |
494 | return percpu_down_read_trylock(&cpu_hotplug_lock); | |
495 | } | |
496 | EXPORT_SYMBOL_GPL(cpus_read_trylock); | |
497 | ||
8f553c49 | 498 | void cpus_read_unlock(void) |
a9d9baa1 | 499 | { |
fc8dffd3 | 500 | percpu_up_read(&cpu_hotplug_lock); |
a9d9baa1 | 501 | } |
8f553c49 | 502 | EXPORT_SYMBOL_GPL(cpus_read_unlock); |
a9d9baa1 | 503 | |
8f553c49 | 504 | void cpus_write_lock(void) |
d221938c | 505 | { |
fc8dffd3 | 506 | percpu_down_write(&cpu_hotplug_lock); |
d221938c | 507 | } |
87af9e7f | 508 | |
8f553c49 | 509 | void cpus_write_unlock(void) |
d221938c | 510 | { |
fc8dffd3 | 511 | percpu_up_write(&cpu_hotplug_lock); |
d221938c GS |
512 | } |
513 | ||
fc8dffd3 | 514 | void lockdep_assert_cpus_held(void) |
d221938c | 515 | { |
ce48c457 VS |
516 | /* |
517 | * We can't have hotplug operations before userspace starts running, | |
518 | * and some init codepaths will knowingly not take the hotplug lock. | |
519 | * This is all valid, so mute lockdep until it makes sense to report | |
520 | * unheld locks. | |
521 | */ | |
522 | if (system_state < SYSTEM_RUNNING) | |
523 | return; | |
524 | ||
fc8dffd3 | 525 | percpu_rwsem_assert_held(&cpu_hotplug_lock); |
d221938c | 526 | } |
79a6cdeb | 527 | |
43759fe5 FW |
528 | #ifdef CONFIG_LOCKDEP |
529 | int lockdep_is_cpus_held(void) | |
530 | { | |
531 | return percpu_rwsem_is_held(&cpu_hotplug_lock); | |
532 | } | |
533 | #endif | |
534 | ||
cb92173d PZ |
535 | static void lockdep_acquire_cpus_lock(void) |
536 | { | |
1751060e | 537 | rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_); |
cb92173d PZ |
538 | } |
539 | ||
540 | static void lockdep_release_cpus_lock(void) | |
541 | { | |
1751060e | 542 | rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_); |
cb92173d PZ |
543 | } |
544 | ||
16e53dbf SB |
545 | /* |
546 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
547 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
548 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
549 | * hotplug path before performing hotplug operations. So acquiring that lock | |
550 | * guarantees mutual exclusion from any currently running hotplug operations. | |
551 | */ | |
552 | void cpu_hotplug_disable(void) | |
553 | { | |
554 | cpu_maps_update_begin(); | |
89af7ba5 | 555 | cpu_hotplug_disabled++; |
16e53dbf SB |
556 | cpu_maps_update_done(); |
557 | } | |
32145c46 | 558 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf | 559 | |
01b41159 LW |
560 | static void __cpu_hotplug_enable(void) |
561 | { | |
562 | if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n")) | |
563 | return; | |
564 | cpu_hotplug_disabled--; | |
565 | } | |
566 | ||
16e53dbf SB |
567 | void cpu_hotplug_enable(void) |
568 | { | |
569 | cpu_maps_update_begin(); | |
01b41159 | 570 | __cpu_hotplug_enable(); |
16e53dbf SB |
571 | cpu_maps_update_done(); |
572 | } | |
32145c46 | 573 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
cb92173d PZ |
574 | |
575 | #else | |
576 | ||
577 | static void lockdep_acquire_cpus_lock(void) | |
578 | { | |
579 | } | |
580 | ||
581 | static void lockdep_release_cpus_lock(void) | |
582 | { | |
583 | } | |
584 | ||
b9d10be7 | 585 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 586 | |
a74cfffb TG |
587 | /* |
588 | * Architectures that need SMT-specific errata handling during SMT hotplug | |
589 | * should override this. | |
590 | */ | |
591 | void __weak arch_smt_update(void) { } | |
592 | ||
0cc3cd21 | 593 | #ifdef CONFIG_HOTPLUG_SMT |
3f916919 | 594 | |
0cc3cd21 | 595 | enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED; |
447ae4ac ME |
596 | static unsigned int cpu_smt_max_threads __ro_after_init; |
597 | unsigned int cpu_smt_num_threads __read_mostly = UINT_MAX; | |
bc2d8d26 | 598 | |
8e1b706b | 599 | void __init cpu_smt_disable(bool force) |
0cc3cd21 | 600 | { |
e1572f1d | 601 | if (!cpu_smt_possible()) |
8e1b706b JK |
602 | return; |
603 | ||
604 | if (force) { | |
0cc3cd21 TG |
605 | pr_info("SMT: Force disabled\n"); |
606 | cpu_smt_control = CPU_SMT_FORCE_DISABLED; | |
8e1b706b | 607 | } else { |
d0e7d144 | 608 | pr_info("SMT: disabled\n"); |
8e1b706b | 609 | cpu_smt_control = CPU_SMT_DISABLED; |
0cc3cd21 | 610 | } |
447ae4ac | 611 | cpu_smt_num_threads = 1; |
8e1b706b JK |
612 | } |
613 | ||
fee0aede TG |
614 | /* |
615 | * The decision whether SMT is supported can only be done after the full | |
b284909a | 616 | * CPU identification. Called from architecture code. |
bc2d8d26 | 617 | */ |
447ae4ac ME |
618 | void __init cpu_smt_set_num_threads(unsigned int num_threads, |
619 | unsigned int max_threads) | |
bc2d8d26 | 620 | { |
447ae4ac ME |
621 | WARN_ON(!num_threads || (num_threads > max_threads)); |
622 | ||
91b4a7db | 623 | if (max_threads == 1) |
bc2d8d26 | 624 | cpu_smt_control = CPU_SMT_NOT_SUPPORTED; |
447ae4ac ME |
625 | |
626 | cpu_smt_max_threads = max_threads; | |
627 | ||
628 | /* | |
629 | * If SMT has been disabled via the kernel command line or SMT is | |
630 | * not supported, set cpu_smt_num_threads to 1 for consistency. | |
631 | * If enabled, take the architecture requested number of threads | |
632 | * to bring up into account. | |
633 | */ | |
634 | if (cpu_smt_control != CPU_SMT_ENABLED) | |
635 | cpu_smt_num_threads = 1; | |
636 | else if (num_threads < cpu_smt_num_threads) | |
637 | cpu_smt_num_threads = num_threads; | |
bc2d8d26 TG |
638 | } |
639 | ||
8e1b706b JK |
640 | static int __init smt_cmdline_disable(char *str) |
641 | { | |
642 | cpu_smt_disable(str && !strcmp(str, "force")); | |
0cc3cd21 TG |
643 | return 0; |
644 | } | |
645 | early_param("nosmt", smt_cmdline_disable); | |
646 | ||
38253464 ME |
647 | /* |
648 | * For Archicture supporting partial SMT states check if the thread is allowed. | |
649 | * Otherwise this has already been checked through cpu_smt_max_threads when | |
650 | * setting the SMT level. | |
651 | */ | |
652 | static inline bool cpu_smt_thread_allowed(unsigned int cpu) | |
653 | { | |
654 | #ifdef CONFIG_SMT_NUM_THREADS_DYNAMIC | |
655 | return topology_smt_thread_allowed(cpu); | |
656 | #else | |
657 | return true; | |
658 | #endif | |
659 | } | |
660 | ||
d91bdd96 | 661 | static inline bool cpu_bootable(unsigned int cpu) |
0cc3cd21 | 662 | { |
38253464 | 663 | if (cpu_smt_control == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu)) |
0cc3cd21 TG |
664 | return true; |
665 | ||
d91bdd96 TG |
666 | /* All CPUs are bootable if controls are not configured */ |
667 | if (cpu_smt_control == CPU_SMT_NOT_IMPLEMENTED) | |
668 | return true; | |
669 | ||
670 | /* All CPUs are bootable if CPU is not SMT capable */ | |
671 | if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
672 | return true; | |
673 | ||
b284909a | 674 | if (topology_is_primary_thread(cpu)) |
0cc3cd21 TG |
675 | return true; |
676 | ||
677 | /* | |
678 | * On x86 it's required to boot all logical CPUs at least once so | |
679 | * that the init code can get a chance to set CR4.MCE on each | |
182e073f | 680 | * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any |
0cc3cd21 TG |
681 | * core will shutdown the machine. |
682 | */ | |
e797bda3 | 683 | return !cpumask_test_cpu(cpu, &cpus_booted_once_mask); |
0cc3cd21 | 684 | } |
e1572f1d | 685 | |
52b38b7a | 686 | /* Returns true if SMT is supported and not forcefully (irreversibly) disabled */ |
e1572f1d VK |
687 | bool cpu_smt_possible(void) |
688 | { | |
689 | return cpu_smt_control != CPU_SMT_FORCE_DISABLED && | |
690 | cpu_smt_control != CPU_SMT_NOT_SUPPORTED; | |
691 | } | |
692 | EXPORT_SYMBOL_GPL(cpu_smt_possible); | |
18415f33 | 693 | |
0cc3cd21 | 694 | #else |
d91bdd96 | 695 | static inline bool cpu_bootable(unsigned int cpu) { return true; } |
0cc3cd21 TG |
696 | #endif |
697 | ||
4dddfb5f | 698 | static inline enum cpuhp_state |
b7ba6d8d | 699 | cpuhp_set_state(int cpu, struct cpuhp_cpu_state *st, enum cpuhp_state target) |
4dddfb5f PZ |
700 | { |
701 | enum cpuhp_state prev_state = st->state; | |
2ea46c6f | 702 | bool bringup = st->state < target; |
4dddfb5f PZ |
703 | |
704 | st->rollback = false; | |
705 | st->last = NULL; | |
706 | ||
707 | st->target = target; | |
708 | st->single = false; | |
2ea46c6f | 709 | st->bringup = bringup; |
b7ba6d8d SP |
710 | if (cpu_dying(cpu) != !bringup) |
711 | set_cpu_dying(cpu, !bringup); | |
4dddfb5f PZ |
712 | |
713 | return prev_state; | |
714 | } | |
715 | ||
716 | static inline void | |
b7ba6d8d SP |
717 | cpuhp_reset_state(int cpu, struct cpuhp_cpu_state *st, |
718 | enum cpuhp_state prev_state) | |
4dddfb5f | 719 | { |
2ea46c6f PZ |
720 | bool bringup = !st->bringup; |
721 | ||
453e4108 VD |
722 | st->target = prev_state; |
723 | ||
724 | /* | |
725 | * Already rolling back. No need invert the bringup value or to change | |
726 | * the current state. | |
727 | */ | |
728 | if (st->rollback) | |
729 | return; | |
730 | ||
4dddfb5f PZ |
731 | st->rollback = true; |
732 | ||
733 | /* | |
734 | * If we have st->last we need to undo partial multi_instance of this | |
735 | * state first. Otherwise start undo at the previous state. | |
736 | */ | |
737 | if (!st->last) { | |
738 | if (st->bringup) | |
739 | st->state--; | |
740 | else | |
741 | st->state++; | |
742 | } | |
743 | ||
2ea46c6f | 744 | st->bringup = bringup; |
b7ba6d8d SP |
745 | if (cpu_dying(cpu) != !bringup) |
746 | set_cpu_dying(cpu, !bringup); | |
4dddfb5f PZ |
747 | } |
748 | ||
749 | /* Regular hotplug invocation of the AP hotplug thread */ | |
750 | static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st) | |
751 | { | |
752 | if (!st->single && st->state == st->target) | |
753 | return; | |
754 | ||
755 | st->result = 0; | |
756 | /* | |
757 | * Make sure the above stores are visible before should_run becomes | |
758 | * true. Paired with the mb() above in cpuhp_thread_fun() | |
759 | */ | |
760 | smp_mb(); | |
761 | st->should_run = true; | |
762 | wake_up_process(st->thread); | |
5ebe7742 | 763 | wait_for_ap_thread(st, st->bringup); |
4dddfb5f PZ |
764 | } |
765 | ||
b7ba6d8d SP |
766 | static int cpuhp_kick_ap(int cpu, struct cpuhp_cpu_state *st, |
767 | enum cpuhp_state target) | |
4dddfb5f PZ |
768 | { |
769 | enum cpuhp_state prev_state; | |
770 | int ret; | |
771 | ||
b7ba6d8d | 772 | prev_state = cpuhp_set_state(cpu, st, target); |
4dddfb5f PZ |
773 | __cpuhp_kick_ap(st); |
774 | if ((ret = st->result)) { | |
b7ba6d8d | 775 | cpuhp_reset_state(cpu, st, prev_state); |
4dddfb5f PZ |
776 | __cpuhp_kick_ap(st); |
777 | } | |
778 | ||
779 | return ret; | |
780 | } | |
9cd4f1a4 | 781 | |
22b612e2 | 782 | static int bringup_wait_for_ap_online(unsigned int cpu) |
8df3e07e TG |
783 | { |
784 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
785 | ||
9cd4f1a4 | 786 | /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */ |
5ebe7742 | 787 | wait_for_ap_thread(st, true); |
dea1d0f5 TG |
788 | if (WARN_ON_ONCE((!cpu_online(cpu)))) |
789 | return -ECANCELED; | |
9cd4f1a4 | 790 | |
45178ac0 | 791 | /* Unpark the hotplug thread of the target cpu */ |
9cd4f1a4 TG |
792 | kthread_unpark(st->thread); |
793 | ||
0cc3cd21 TG |
794 | /* |
795 | * SMT soft disabling on X86 requires to bring the CPU out of the | |
796 | * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The | |
f5602011 | 797 | * CPU marked itself as booted_once in notify_cpu_starting() so the |
d91bdd96 | 798 | * cpu_bootable() check will now return false if this is not the |
0cc3cd21 TG |
799 | * primary sibling. |
800 | */ | |
d91bdd96 | 801 | if (!cpu_bootable(cpu)) |
0cc3cd21 | 802 | return -ECANCELED; |
22b612e2 | 803 | return 0; |
8df3e07e TG |
804 | } |
805 | ||
a631be92 TG |
806 | #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP |
807 | static int cpuhp_kick_ap_alive(unsigned int cpu) | |
808 | { | |
809 | if (!cpuhp_can_boot_ap(cpu)) | |
810 | return -EAGAIN; | |
811 | ||
812 | return arch_cpuhp_kick_ap_alive(cpu, idle_thread_get(cpu)); | |
813 | } | |
814 | ||
815 | static int cpuhp_bringup_ap(unsigned int cpu) | |
816 | { | |
817 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
818 | int ret; | |
819 | ||
820 | /* | |
821 | * Some architectures have to walk the irq descriptors to | |
822 | * setup the vector space for the cpu which comes online. | |
823 | * Prevent irq alloc/free across the bringup. | |
824 | */ | |
825 | irq_lock_sparse(); | |
826 | ||
827 | ret = cpuhp_bp_sync_alive(cpu); | |
828 | if (ret) | |
829 | goto out_unlock; | |
830 | ||
831 | ret = bringup_wait_for_ap_online(cpu); | |
832 | if (ret) | |
833 | goto out_unlock; | |
834 | ||
835 | irq_unlock_sparse(); | |
836 | ||
837 | if (st->target <= CPUHP_AP_ONLINE_IDLE) | |
838 | return 0; | |
839 | ||
840 | return cpuhp_kick_ap(cpu, st, st->target); | |
841 | ||
842 | out_unlock: | |
843 | irq_unlock_sparse(); | |
844 | return ret; | |
845 | } | |
846 | #else | |
ba997462 TG |
847 | static int bringup_cpu(unsigned int cpu) |
848 | { | |
22b612e2 | 849 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
ba997462 TG |
850 | struct task_struct *idle = idle_thread_get(cpu); |
851 | int ret; | |
852 | ||
6f062123 TG |
853 | if (!cpuhp_can_boot_ap(cpu)) |
854 | return -EAGAIN; | |
855 | ||
aa877175 BO |
856 | /* |
857 | * Some architectures have to walk the irq descriptors to | |
858 | * setup the vector space for the cpu which comes online. | |
22b612e2 TG |
859 | * |
860 | * Prevent irq alloc/free across the bringup by acquiring the | |
861 | * sparse irq lock. Hold it until the upcoming CPU completes the | |
862 | * startup in cpuhp_online_idle() which allows to avoid | |
863 | * intermediate synchronization points in the architecture code. | |
aa877175 BO |
864 | */ |
865 | irq_lock_sparse(); | |
866 | ||
ba997462 | 867 | ret = __cpu_up(cpu, idle); |
530e9b76 | 868 | if (ret) |
22b612e2 TG |
869 | goto out_unlock; |
870 | ||
6f062123 TG |
871 | ret = cpuhp_bp_sync_alive(cpu); |
872 | if (ret) | |
873 | goto out_unlock; | |
874 | ||
22b612e2 TG |
875 | ret = bringup_wait_for_ap_online(cpu); |
876 | if (ret) | |
877 | goto out_unlock; | |
878 | ||
879 | irq_unlock_sparse(); | |
880 | ||
881 | if (st->target <= CPUHP_AP_ONLINE_IDLE) | |
882 | return 0; | |
883 | ||
884 | return cpuhp_kick_ap(cpu, st, st->target); | |
885 | ||
886 | out_unlock: | |
887 | irq_unlock_sparse(); | |
888 | return ret; | |
ba997462 | 889 | } |
a631be92 | 890 | #endif |
ba997462 | 891 | |
bf2c59fc PZ |
892 | static int finish_cpu(unsigned int cpu) |
893 | { | |
894 | struct task_struct *idle = idle_thread_get(cpu); | |
895 | struct mm_struct *mm = idle->active_mm; | |
896 | ||
897 | /* | |
898 | * idle_task_exit() will have switched to &init_mm, now | |
899 | * clean up any remaining active_mm state. | |
900 | */ | |
901 | if (mm != &init_mm) | |
902 | idle->active_mm = &init_mm; | |
aa464ba9 | 903 | mmdrop_lazy_tlb(mm); |
bf2c59fc PZ |
904 | return 0; |
905 | } | |
906 | ||
2e1a3483 TG |
907 | /* |
908 | * Hotplug state machine related functions | |
909 | */ | |
2e1a3483 | 910 | |
453e4108 VD |
911 | /* |
912 | * Get the next state to run. Empty ones will be skipped. Returns true if a | |
913 | * state must be run. | |
914 | * | |
915 | * st->state will be modified ahead of time, to match state_to_run, as if it | |
916 | * has already ran. | |
917 | */ | |
918 | static bool cpuhp_next_state(bool bringup, | |
919 | enum cpuhp_state *state_to_run, | |
920 | struct cpuhp_cpu_state *st, | |
921 | enum cpuhp_state target) | |
2e1a3483 | 922 | { |
453e4108 VD |
923 | do { |
924 | if (bringup) { | |
925 | if (st->state >= target) | |
926 | return false; | |
927 | ||
928 | *state_to_run = ++st->state; | |
929 | } else { | |
930 | if (st->state <= target) | |
931 | return false; | |
932 | ||
933 | *state_to_run = st->state--; | |
934 | } | |
935 | ||
936 | if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run))) | |
937 | break; | |
938 | } while (true); | |
939 | ||
940 | return true; | |
941 | } | |
942 | ||
6f855b39 VD |
943 | static int __cpuhp_invoke_callback_range(bool bringup, |
944 | unsigned int cpu, | |
945 | struct cpuhp_cpu_state *st, | |
946 | enum cpuhp_state target, | |
947 | bool nofail) | |
453e4108 VD |
948 | { |
949 | enum cpuhp_state state; | |
6f855b39 | 950 | int ret = 0; |
453e4108 VD |
951 | |
952 | while (cpuhp_next_state(bringup, &state, st, target)) { | |
6f855b39 VD |
953 | int err; |
954 | ||
453e4108 | 955 | err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL); |
6f855b39 VD |
956 | if (!err) |
957 | continue; | |
958 | ||
959 | if (nofail) { | |
960 | pr_warn("CPU %u %s state %s (%d) failed (%d)\n", | |
961 | cpu, bringup ? "UP" : "DOWN", | |
962 | cpuhp_get_step(st->state)->name, | |
963 | st->state, err); | |
964 | ret = -1; | |
965 | } else { | |
966 | ret = err; | |
453e4108 | 967 | break; |
6f855b39 | 968 | } |
453e4108 VD |
969 | } |
970 | ||
6f855b39 VD |
971 | return ret; |
972 | } | |
973 | ||
974 | static inline int cpuhp_invoke_callback_range(bool bringup, | |
975 | unsigned int cpu, | |
976 | struct cpuhp_cpu_state *st, | |
977 | enum cpuhp_state target) | |
978 | { | |
979 | return __cpuhp_invoke_callback_range(bringup, cpu, st, target, false); | |
980 | } | |
981 | ||
982 | static inline void cpuhp_invoke_callback_range_nofail(bool bringup, | |
983 | unsigned int cpu, | |
984 | struct cpuhp_cpu_state *st, | |
985 | enum cpuhp_state target) | |
986 | { | |
987 | __cpuhp_invoke_callback_range(bringup, cpu, st, target, true); | |
2e1a3483 TG |
988 | } |
989 | ||
206b9235 TG |
990 | static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st) |
991 | { | |
992 | if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) | |
993 | return true; | |
994 | /* | |
995 | * When CPU hotplug is disabled, then taking the CPU down is not | |
996 | * possible because takedown_cpu() and the architecture and | |
997 | * subsystem specific mechanisms are not available. So the CPU | |
998 | * which would be completely unplugged again needs to stay around | |
999 | * in the current state. | |
1000 | */ | |
1001 | return st->state <= CPUHP_BRINGUP_CPU; | |
1002 | } | |
1003 | ||
2e1a3483 | 1004 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, |
a724632c | 1005 | enum cpuhp_state target) |
2e1a3483 TG |
1006 | { |
1007 | enum cpuhp_state prev_state = st->state; | |
1008 | int ret = 0; | |
1009 | ||
453e4108 VD |
1010 | ret = cpuhp_invoke_callback_range(true, cpu, st, target); |
1011 | if (ret) { | |
ebca71a8 DZ |
1012 | pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n", |
1013 | ret, cpu, cpuhp_get_step(st->state)->name, | |
1014 | st->state); | |
1015 | ||
b7ba6d8d | 1016 | cpuhp_reset_state(cpu, st, prev_state); |
453e4108 VD |
1017 | if (can_rollback_cpu(st)) |
1018 | WARN_ON(cpuhp_invoke_callback_range(false, cpu, st, | |
1019 | prev_state)); | |
2e1a3483 TG |
1020 | } |
1021 | return ret; | |
1022 | } | |
1023 | ||
4cb28ced TG |
1024 | /* |
1025 | * The cpu hotplug threads manage the bringup and teardown of the cpus | |
1026 | */ | |
4cb28ced TG |
1027 | static int cpuhp_should_run(unsigned int cpu) |
1028 | { | |
1029 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
1030 | ||
1031 | return st->should_run; | |
1032 | } | |
1033 | ||
4cb28ced TG |
1034 | /* |
1035 | * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke | |
1036 | * callbacks when a state gets [un]installed at runtime. | |
4dddfb5f PZ |
1037 | * |
1038 | * Each invocation of this function by the smpboot thread does a single AP | |
1039 | * state callback. | |
1040 | * | |
1041 | * It has 3 modes of operation: | |
1042 | * - single: runs st->cb_state | |
1043 | * - up: runs ++st->state, while st->state < st->target | |
1044 | * - down: runs st->state--, while st->state > st->target | |
1045 | * | |
1046 | * When complete or on error, should_run is cleared and the completion is fired. | |
4cb28ced TG |
1047 | */ |
1048 | static void cpuhp_thread_fun(unsigned int cpu) | |
1049 | { | |
1050 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
4dddfb5f PZ |
1051 | bool bringup = st->bringup; |
1052 | enum cpuhp_state state; | |
4cb28ced | 1053 | |
f8b7530a NU |
1054 | if (WARN_ON_ONCE(!st->should_run)) |
1055 | return; | |
1056 | ||
4cb28ced | 1057 | /* |
4dddfb5f PZ |
1058 | * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures |
1059 | * that if we see ->should_run we also see the rest of the state. | |
4cb28ced TG |
1060 | */ |
1061 | smp_mb(); | |
4cb28ced | 1062 | |
cb92173d PZ |
1063 | /* |
1064 | * The BP holds the hotplug lock, but we're now running on the AP, | |
1065 | * ensure that anybody asserting the lock is held, will actually find | |
1066 | * it so. | |
1067 | */ | |
1068 | lockdep_acquire_cpus_lock(); | |
5f4b55e1 | 1069 | cpuhp_lock_acquire(bringup); |
4dddfb5f | 1070 | |
a724632c | 1071 | if (st->single) { |
4dddfb5f PZ |
1072 | state = st->cb_state; |
1073 | st->should_run = false; | |
1074 | } else { | |
453e4108 VD |
1075 | st->should_run = cpuhp_next_state(bringup, &state, st, st->target); |
1076 | if (!st->should_run) | |
1077 | goto end; | |
4dddfb5f PZ |
1078 | } |
1079 | ||
1080 | WARN_ON_ONCE(!cpuhp_is_ap_state(state)); | |
1081 | ||
4dddfb5f PZ |
1082 | if (cpuhp_is_atomic_state(state)) { |
1083 | local_irq_disable(); | |
1084 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); | |
1085 | local_irq_enable(); | |
3b9d6da6 | 1086 | |
4dddfb5f PZ |
1087 | /* |
1088 | * STARTING/DYING must not fail! | |
1089 | */ | |
1090 | WARN_ON_ONCE(st->result); | |
4cb28ced | 1091 | } else { |
4dddfb5f PZ |
1092 | st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last); |
1093 | } | |
1094 | ||
1095 | if (st->result) { | |
1096 | /* | |
1097 | * If we fail on a rollback, we're up a creek without no | |
1098 | * paddle, no way forward, no way back. We loose, thanks for | |
1099 | * playing. | |
1100 | */ | |
1101 | WARN_ON_ONCE(st->rollback); | |
1102 | st->should_run = false; | |
4cb28ced | 1103 | } |
4dddfb5f | 1104 | |
453e4108 | 1105 | end: |
5f4b55e1 | 1106 | cpuhp_lock_release(bringup); |
cb92173d | 1107 | lockdep_release_cpus_lock(); |
4dddfb5f PZ |
1108 | |
1109 | if (!st->should_run) | |
5ebe7742 | 1110 | complete_ap_thread(st, bringup); |
4cb28ced TG |
1111 | } |
1112 | ||
1113 | /* Invoke a single callback on a remote cpu */ | |
a724632c | 1114 | static int |
cf392d10 TG |
1115 | cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup, |
1116 | struct hlist_node *node) | |
4cb28ced TG |
1117 | { |
1118 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f | 1119 | int ret; |
4cb28ced TG |
1120 | |
1121 | if (!cpu_online(cpu)) | |
1122 | return 0; | |
1123 | ||
5f4b55e1 PZ |
1124 | cpuhp_lock_acquire(false); |
1125 | cpuhp_lock_release(false); | |
1126 | ||
1127 | cpuhp_lock_acquire(true); | |
1128 | cpuhp_lock_release(true); | |
49dfe2a6 | 1129 | |
6a4e2451 TG |
1130 | /* |
1131 | * If we are up and running, use the hotplug thread. For early calls | |
1132 | * we invoke the thread function directly. | |
1133 | */ | |
1134 | if (!st->thread) | |
96abb968 | 1135 | return cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
6a4e2451 | 1136 | |
4dddfb5f PZ |
1137 | st->rollback = false; |
1138 | st->last = NULL; | |
1139 | ||
1140 | st->node = node; | |
1141 | st->bringup = bringup; | |
4cb28ced | 1142 | st->cb_state = state; |
a724632c | 1143 | st->single = true; |
a724632c | 1144 | |
4dddfb5f | 1145 | __cpuhp_kick_ap(st); |
4cb28ced | 1146 | |
4cb28ced | 1147 | /* |
4dddfb5f | 1148 | * If we failed and did a partial, do a rollback. |
4cb28ced | 1149 | */ |
4dddfb5f PZ |
1150 | if ((ret = st->result) && st->last) { |
1151 | st->rollback = true; | |
1152 | st->bringup = !bringup; | |
1153 | ||
1154 | __cpuhp_kick_ap(st); | |
1155 | } | |
1156 | ||
1f7c70d6 TG |
1157 | /* |
1158 | * Clean up the leftovers so the next hotplug operation wont use stale | |
1159 | * data. | |
1160 | */ | |
1161 | st->node = st->last = NULL; | |
4dddfb5f | 1162 | return ret; |
1cf4f629 TG |
1163 | } |
1164 | ||
1165 | static int cpuhp_kick_ap_work(unsigned int cpu) | |
1166 | { | |
1167 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
4dddfb5f PZ |
1168 | enum cpuhp_state prev_state = st->state; |
1169 | int ret; | |
1cf4f629 | 1170 | |
5f4b55e1 PZ |
1171 | cpuhp_lock_acquire(false); |
1172 | cpuhp_lock_release(false); | |
1173 | ||
1174 | cpuhp_lock_acquire(true); | |
1175 | cpuhp_lock_release(true); | |
4dddfb5f PZ |
1176 | |
1177 | trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work); | |
b7ba6d8d | 1178 | ret = cpuhp_kick_ap(cpu, st, st->target); |
4dddfb5f PZ |
1179 | trace_cpuhp_exit(cpu, st->state, prev_state, ret); |
1180 | ||
1181 | return ret; | |
4cb28ced TG |
1182 | } |
1183 | ||
1184 | static struct smp_hotplug_thread cpuhp_threads = { | |
1185 | .store = &cpuhp_state.thread, | |
4cb28ced TG |
1186 | .thread_should_run = cpuhp_should_run, |
1187 | .thread_fn = cpuhp_thread_fun, | |
1188 | .thread_comm = "cpuhp/%u", | |
1189 | .selfparking = true, | |
1190 | }; | |
1191 | ||
d308077e SP |
1192 | static __init void cpuhp_init_state(void) |
1193 | { | |
1194 | struct cpuhp_cpu_state *st; | |
1195 | int cpu; | |
1196 | ||
1197 | for_each_possible_cpu(cpu) { | |
1198 | st = per_cpu_ptr(&cpuhp_state, cpu); | |
1199 | init_completion(&st->done_up); | |
1200 | init_completion(&st->done_down); | |
1201 | } | |
1202 | } | |
1203 | ||
4cb28ced TG |
1204 | void __init cpuhp_threads_init(void) |
1205 | { | |
d308077e | 1206 | cpuhp_init_state(); |
4cb28ced TG |
1207 | BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads)); |
1208 | kthread_unpark(this_cpu_read(cpuhp_state.thread)); | |
1209 | } | |
1210 | ||
777c6e0d | 1211 | #ifdef CONFIG_HOTPLUG_CPU |
8ff00399 NP |
1212 | #ifndef arch_clear_mm_cpumask_cpu |
1213 | #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm)) | |
1214 | #endif | |
1215 | ||
e4cc2f87 AV |
1216 | /** |
1217 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
1218 | * @cpu: a CPU id | |
1219 | * | |
1220 | * This function walks all processes, finds a valid mm struct for each one and | |
1221 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
1222 | * trivial, there are various non-obvious corner cases, which this function | |
1223 | * tries to solve in a safe manner. | |
1224 | * | |
1225 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
1226 | * be called only for an already offlined CPU. | |
1227 | */ | |
cb79295e AV |
1228 | void clear_tasks_mm_cpumask(int cpu) |
1229 | { | |
1230 | struct task_struct *p; | |
1231 | ||
1232 | /* | |
1233 | * This function is called after the cpu is taken down and marked | |
1234 | * offline, so its not like new tasks will ever get this cpu set in | |
1235 | * their mm mask. -- Peter Zijlstra | |
1236 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
1237 | * full-fledged tasklist_lock. | |
1238 | */ | |
e4cc2f87 | 1239 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
1240 | rcu_read_lock(); |
1241 | for_each_process(p) { | |
1242 | struct task_struct *t; | |
1243 | ||
e4cc2f87 AV |
1244 | /* |
1245 | * Main thread might exit, but other threads may still have | |
1246 | * a valid mm. Find one. | |
1247 | */ | |
cb79295e AV |
1248 | t = find_lock_task_mm(p); |
1249 | if (!t) | |
1250 | continue; | |
8ff00399 | 1251 | arch_clear_mm_cpumask_cpu(cpu, t->mm); |
cb79295e AV |
1252 | task_unlock(t); |
1253 | } | |
1254 | rcu_read_unlock(); | |
1255 | } | |
1256 | ||
1da177e4 | 1257 | /* Take this CPU down. */ |
71cf5aee | 1258 | static int take_cpu_down(void *_param) |
1da177e4 | 1259 | { |
4baa0afc TG |
1260 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
1261 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 1262 | int err, cpu = smp_processor_id(); |
1da177e4 | 1263 | |
1da177e4 LT |
1264 | /* Ensure this CPU doesn't handle any more interrupts. */ |
1265 | err = __cpu_disable(); | |
1266 | if (err < 0) | |
f3705136 | 1267 | return err; |
1da177e4 | 1268 | |
a724632c | 1269 | /* |
453e4108 VD |
1270 | * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going |
1271 | * down, that the current state is CPUHP_TEARDOWN_CPU - 1. | |
a724632c | 1272 | */ |
453e4108 VD |
1273 | WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1)); |
1274 | ||
453e4108 | 1275 | /* |
6f855b39 | 1276 | * Invoke the former CPU_DYING callbacks. DYING must not fail! |
453e4108 | 1277 | */ |
6f855b39 | 1278 | cpuhp_invoke_callback_range_nofail(false, cpu, st, target); |
4baa0afc | 1279 | |
14e568e7 | 1280 | /* Park the stopper thread */ |
090e77c3 | 1281 | stop_machine_park(cpu); |
f3705136 | 1282 | return 0; |
1da177e4 LT |
1283 | } |
1284 | ||
98458172 | 1285 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 1286 | { |
e69aab13 | 1287 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
98458172 | 1288 | int err; |
1da177e4 | 1289 | |
2a58c527 | 1290 | /* Park the smpboot threads */ |
13070833 | 1291 | kthread_park(st->thread); |
1cf4f629 | 1292 | |
6acce3ef | 1293 | /* |
a8994181 TG |
1294 | * Prevent irq alloc/free while the dying cpu reorganizes the |
1295 | * interrupt affinities. | |
6acce3ef | 1296 | */ |
a8994181 | 1297 | irq_lock_sparse(); |
6acce3ef | 1298 | |
a8994181 TG |
1299 | /* |
1300 | * So now all preempt/rcu users must observe !cpu_active(). | |
1301 | */ | |
210e2133 | 1302 | err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 1303 | if (err) { |
3b9d6da6 | 1304 | /* CPU refused to die */ |
a8994181 | 1305 | irq_unlock_sparse(); |
3b9d6da6 | 1306 | /* Unpark the hotplug thread so we can rollback there */ |
13070833 | 1307 | kthread_unpark(st->thread); |
98458172 | 1308 | return err; |
8fa1d7d3 | 1309 | } |
04321587 | 1310 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 1311 | |
48c5ccae | 1312 | /* |
5b1ead68 BJ |
1313 | * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed |
1314 | * all runnable tasks from the CPU, there's only the idle task left now | |
48c5ccae | 1315 | * that the migration thread is done doing the stop_machine thing. |
51a96c77 PZ |
1316 | * |
1317 | * Wait for the stop thread to go away. | |
48c5ccae | 1318 | */ |
5ebe7742 | 1319 | wait_for_ap_thread(st, false); |
e69aab13 | 1320 | BUG_ON(st->state != CPUHP_AP_IDLE_DEAD); |
1da177e4 | 1321 | |
a8994181 TG |
1322 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
1323 | irq_unlock_sparse(); | |
1324 | ||
345527b1 | 1325 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
1326 | /* This actually kills the CPU. */ |
1327 | __cpu_die(cpu); | |
1328 | ||
6f062123 TG |
1329 | cpuhp_bp_sync_dead(cpu); |
1330 | ||
a49b116d | 1331 | tick_cleanup_dead_cpu(cpu); |
a28ab03b FW |
1332 | |
1333 | /* | |
1334 | * Callbacks must be re-integrated right away to the RCU state machine. | |
1335 | * Otherwise an RCU callback could block a further teardown function | |
1336 | * waiting for its completion. | |
1337 | */ | |
a58163d8 | 1338 | rcutree_migrate_callbacks(cpu); |
a28ab03b | 1339 | |
98458172 TG |
1340 | return 0; |
1341 | } | |
1da177e4 | 1342 | |
71f87b2f TG |
1343 | static void cpuhp_complete_idle_dead(void *arg) |
1344 | { | |
1345 | struct cpuhp_cpu_state *st = arg; | |
1346 | ||
5ebe7742 | 1347 | complete_ap_thread(st, false); |
71f87b2f TG |
1348 | } |
1349 | ||
e69aab13 TG |
1350 | void cpuhp_report_idle_dead(void) |
1351 | { | |
1352 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); | |
1353 | ||
1354 | BUG_ON(st->state != CPUHP_AP_OFFLINE); | |
500f8f9b | 1355 | tick_assert_timekeeping_handover(); |
448e9f34 | 1356 | rcutree_report_cpu_dead(); |
71f87b2f TG |
1357 | st->state = CPUHP_AP_IDLE_DEAD; |
1358 | /* | |
448e9f34 | 1359 | * We cannot call complete after rcutree_report_cpu_dead() so we delegate it |
71f87b2f TG |
1360 | * to an online cpu. |
1361 | */ | |
1362 | smp_call_function_single(cpumask_first(cpu_online_mask), | |
1363 | cpuhp_complete_idle_dead, st, 0); | |
e69aab13 TG |
1364 | } |
1365 | ||
4dddfb5f PZ |
1366 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, |
1367 | enum cpuhp_state target) | |
1368 | { | |
1369 | enum cpuhp_state prev_state = st->state; | |
1370 | int ret = 0; | |
1371 | ||
453e4108 VD |
1372 | ret = cpuhp_invoke_callback_range(false, cpu, st, target); |
1373 | if (ret) { | |
ebca71a8 DZ |
1374 | pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n", |
1375 | ret, cpu, cpuhp_get_step(st->state)->name, | |
1376 | st->state); | |
453e4108 | 1377 | |
b7ba6d8d | 1378 | cpuhp_reset_state(cpu, st, prev_state); |
453e4108 VD |
1379 | |
1380 | if (st->state < prev_state) | |
1381 | WARN_ON(cpuhp_invoke_callback_range(true, cpu, st, | |
1382 | prev_state)); | |
4dddfb5f | 1383 | } |
453e4108 | 1384 | |
4dddfb5f PZ |
1385 | return ret; |
1386 | } | |
cff7d378 | 1387 | |
98458172 | 1388 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
1389 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
1390 | enum cpuhp_state target) | |
98458172 | 1391 | { |
cff7d378 TG |
1392 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
1393 | int prev_state, ret = 0; | |
98458172 TG |
1394 | |
1395 | if (num_online_cpus() == 1) | |
1396 | return -EBUSY; | |
1397 | ||
757c989b | 1398 | if (!cpu_present(cpu)) |
98458172 TG |
1399 | return -EINVAL; |
1400 | ||
8f553c49 | 1401 | cpus_write_lock(); |
98458172 TG |
1402 | |
1403 | cpuhp_tasks_frozen = tasks_frozen; | |
1404 | ||
b7ba6d8d | 1405 | prev_state = cpuhp_set_state(cpu, st, target); |
1cf4f629 TG |
1406 | /* |
1407 | * If the current CPU state is in the range of the AP hotplug thread, | |
1408 | * then we need to kick the thread. | |
1409 | */ | |
8df3e07e | 1410 | if (st->state > CPUHP_TEARDOWN_CPU) { |
4dddfb5f | 1411 | st->target = max((int)target, CPUHP_TEARDOWN_CPU); |
1cf4f629 TG |
1412 | ret = cpuhp_kick_ap_work(cpu); |
1413 | /* | |
1414 | * The AP side has done the error rollback already. Just | |
1415 | * return the error code.. | |
1416 | */ | |
1417 | if (ret) | |
1418 | goto out; | |
1419 | ||
1420 | /* | |
1421 | * We might have stopped still in the range of the AP hotplug | |
1422 | * thread. Nothing to do anymore. | |
1423 | */ | |
8df3e07e | 1424 | if (st->state > CPUHP_TEARDOWN_CPU) |
1cf4f629 | 1425 | goto out; |
4dddfb5f PZ |
1426 | |
1427 | st->target = target; | |
1cf4f629 TG |
1428 | } |
1429 | /* | |
8df3e07e | 1430 | * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need |
1cf4f629 TG |
1431 | * to do the further cleanups. |
1432 | */ | |
a724632c | 1433 | ret = cpuhp_down_callbacks(cpu, st, target); |
62f25069 VD |
1434 | if (ret && st->state < prev_state) { |
1435 | if (st->state == CPUHP_TEARDOWN_CPU) { | |
b7ba6d8d | 1436 | cpuhp_reset_state(cpu, st, prev_state); |
62f25069 VD |
1437 | __cpuhp_kick_ap(st); |
1438 | } else { | |
1439 | WARN(1, "DEAD callback error for CPU%d", cpu); | |
1440 | } | |
3b9d6da6 | 1441 | } |
98458172 | 1442 | |
1cf4f629 | 1443 | out: |
8f553c49 | 1444 | cpus_write_unlock(); |
941154bd TG |
1445 | /* |
1446 | * Do post unplug cleanup. This is still protected against | |
1447 | * concurrent CPU hotplug via cpu_add_remove_lock. | |
1448 | */ | |
1449 | lockup_detector_cleanup(); | |
a74cfffb | 1450 | arch_smt_update(); |
cff7d378 | 1451 | return ret; |
e3920fb4 RW |
1452 | } |
1453 | ||
2b8272ff TG |
1454 | struct cpu_down_work { |
1455 | unsigned int cpu; | |
1456 | enum cpuhp_state target; | |
1457 | }; | |
1458 | ||
1459 | static long __cpu_down_maps_locked(void *arg) | |
1460 | { | |
1461 | struct cpu_down_work *work = arg; | |
1462 | ||
1463 | return _cpu_down(work->cpu, 0, work->target); | |
1464 | } | |
1465 | ||
cc1fe215 TG |
1466 | static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target) |
1467 | { | |
2b8272ff TG |
1468 | struct cpu_down_work work = { .cpu = cpu, .target = target, }; |
1469 | ||
bae1a962 KS |
1470 | /* |
1471 | * If the platform does not support hotplug, report it explicitly to | |
1472 | * differentiate it from a transient offlining failure. | |
1473 | */ | |
1474 | if (cc_platform_has(CC_ATTR_HOTPLUG_DISABLED)) | |
1475 | return -EOPNOTSUPP; | |
cc1fe215 TG |
1476 | if (cpu_hotplug_disabled) |
1477 | return -EBUSY; | |
2b8272ff TG |
1478 | |
1479 | /* | |
1480 | * Ensure that the control task does not run on the to be offlined | |
1481 | * CPU to prevent a deadlock against cfs_b->period_timer. | |
38685e2a RX |
1482 | * Also keep at least one housekeeping cpu onlined to avoid generating |
1483 | * an empty sched_domain span. | |
2b8272ff | 1484 | */ |
38685e2a RX |
1485 | for_each_cpu_and(cpu, cpu_online_mask, housekeeping_cpumask(HK_TYPE_DOMAIN)) { |
1486 | if (cpu != work.cpu) | |
1487 | return work_on_cpu(cpu, __cpu_down_maps_locked, &work); | |
1488 | } | |
1489 | return -EBUSY; | |
cc1fe215 TG |
1490 | } |
1491 | ||
33c3736e | 1492 | static int cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 1493 | { |
9ea09af3 | 1494 | int err; |
e3920fb4 | 1495 | |
d221938c | 1496 | cpu_maps_update_begin(); |
cc1fe215 | 1497 | err = cpu_down_maps_locked(cpu, target); |
d221938c | 1498 | cpu_maps_update_done(); |
1da177e4 LT |
1499 | return err; |
1500 | } | |
4dddfb5f | 1501 | |
33c3736e QY |
1502 | /** |
1503 | * cpu_device_down - Bring down a cpu device | |
1504 | * @dev: Pointer to the cpu device to offline | |
1505 | * | |
1506 | * This function is meant to be used by device core cpu subsystem only. | |
1507 | * | |
1508 | * Other subsystems should use remove_cpu() instead. | |
11bc021d RD |
1509 | * |
1510 | * Return: %0 on success or a negative errno code | |
33c3736e QY |
1511 | */ |
1512 | int cpu_device_down(struct device *dev) | |
af1f4045 | 1513 | { |
33c3736e | 1514 | return cpu_down(dev->id, CPUHP_OFFLINE); |
af1f4045 | 1515 | } |
4dddfb5f | 1516 | |
93ef1429 QY |
1517 | int remove_cpu(unsigned int cpu) |
1518 | { | |
1519 | int ret; | |
1520 | ||
1521 | lock_device_hotplug(); | |
1522 | ret = device_offline(get_cpu_device(cpu)); | |
1523 | unlock_device_hotplug(); | |
1524 | ||
1525 | return ret; | |
1526 | } | |
1527 | EXPORT_SYMBOL_GPL(remove_cpu); | |
1528 | ||
0441a559 QY |
1529 | void smp_shutdown_nonboot_cpus(unsigned int primary_cpu) |
1530 | { | |
1531 | unsigned int cpu; | |
1532 | int error; | |
1533 | ||
1534 | cpu_maps_update_begin(); | |
1535 | ||
1536 | /* | |
1537 | * Make certain the cpu I'm about to reboot on is online. | |
1538 | * | |
1539 | * This is inline to what migrate_to_reboot_cpu() already do. | |
1540 | */ | |
1541 | if (!cpu_online(primary_cpu)) | |
1542 | primary_cpu = cpumask_first(cpu_online_mask); | |
1543 | ||
1544 | for_each_online_cpu(cpu) { | |
1545 | if (cpu == primary_cpu) | |
1546 | continue; | |
1547 | ||
1548 | error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); | |
1549 | if (error) { | |
1550 | pr_err("Failed to offline CPU%d - error=%d", | |
1551 | cpu, error); | |
1552 | break; | |
1553 | } | |
1554 | } | |
1555 | ||
1556 | /* | |
1557 | * Ensure all but the reboot CPU are offline. | |
1558 | */ | |
1559 | BUG_ON(num_online_cpus() > 1); | |
1560 | ||
1561 | /* | |
1562 | * Make sure the CPUs won't be enabled by someone else after this | |
1563 | * point. Kexec will reboot to a new kernel shortly resetting | |
1564 | * everything along the way. | |
1565 | */ | |
1566 | cpu_hotplug_disabled++; | |
1567 | ||
1568 | cpu_maps_update_done(); | |
af1f4045 | 1569 | } |
4dddfb5f PZ |
1570 | |
1571 | #else | |
1572 | #define takedown_cpu NULL | |
1da177e4 LT |
1573 | #endif /*CONFIG_HOTPLUG_CPU*/ |
1574 | ||
4baa0afc | 1575 | /** |
ee1e714b | 1576 | * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU |
4baa0afc TG |
1577 | * @cpu: cpu that just started |
1578 | * | |
4baa0afc TG |
1579 | * It must be called by the arch code on the new cpu, before the new cpu |
1580 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
1581 | */ | |
1582 | void notify_cpu_starting(unsigned int cpu) | |
1583 | { | |
1584 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1585 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
1586 | ||
448e9f34 | 1587 | rcutree_report_cpu_starting(cpu); /* Enables RCU usage on this CPU. */ |
e797bda3 | 1588 | cpumask_set_cpu(cpu, &cpus_booted_once_mask); |
453e4108 VD |
1589 | |
1590 | /* | |
1591 | * STARTING must not fail! | |
1592 | */ | |
6f855b39 | 1593 | cpuhp_invoke_callback_range_nofail(true, cpu, st, target); |
4baa0afc TG |
1594 | } |
1595 | ||
949338e3 | 1596 | /* |
9cd4f1a4 | 1597 | * Called from the idle task. Wake up the controlling task which brings the |
45178ac0 PZ |
1598 | * hotplug thread of the upcoming CPU up and then delegates the rest of the |
1599 | * online bringup to the hotplug thread. | |
949338e3 | 1600 | */ |
8df3e07e | 1601 | void cpuhp_online_idle(enum cpuhp_state state) |
949338e3 | 1602 | { |
8df3e07e | 1603 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
8df3e07e TG |
1604 | |
1605 | /* Happens for the boot cpu */ | |
1606 | if (state != CPUHP_AP_ONLINE_IDLE) | |
1607 | return; | |
1608 | ||
6f062123 TG |
1609 | cpuhp_ap_update_sync_state(SYNC_STATE_ONLINE); |
1610 | ||
45178ac0 | 1611 | /* |
6f062123 | 1612 | * Unpark the stopper thread before we start the idle loop (and start |
45178ac0 PZ |
1613 | * scheduling); this ensures the stopper task is always available. |
1614 | */ | |
1615 | stop_machine_unpark(smp_processor_id()); | |
1616 | ||
8df3e07e | 1617 | st->state = CPUHP_AP_ONLINE_IDLE; |
5ebe7742 | 1618 | complete_ap_thread(st, true); |
949338e3 TG |
1619 | } |
1620 | ||
e3920fb4 | 1621 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 1622 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 1623 | { |
cff7d378 | 1624 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 1625 | struct task_struct *idle; |
2e1a3483 | 1626 | int ret = 0; |
1da177e4 | 1627 | |
8f553c49 | 1628 | cpus_write_lock(); |
38498a67 | 1629 | |
757c989b | 1630 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
1631 | ret = -EINVAL; |
1632 | goto out; | |
1633 | } | |
1634 | ||
757c989b | 1635 | /* |
33c3736e QY |
1636 | * The caller of cpu_up() might have raced with another |
1637 | * caller. Nothing to do. | |
757c989b TG |
1638 | */ |
1639 | if (st->state >= target) | |
38498a67 | 1640 | goto out; |
757c989b TG |
1641 | |
1642 | if (st->state == CPUHP_OFFLINE) { | |
1643 | /* Let it fail before we try to bring the cpu up */ | |
1644 | idle = idle_thread_get(cpu); | |
1645 | if (IS_ERR(idle)) { | |
1646 | ret = PTR_ERR(idle); | |
1647 | goto out; | |
1648 | } | |
6d712b9b DW |
1649 | |
1650 | /* | |
1651 | * Reset stale stack state from the last time this CPU was online. | |
1652 | */ | |
1653 | scs_task_reset(idle); | |
1654 | kasan_unpoison_task_stack(idle); | |
3bb5d2ee | 1655 | } |
38498a67 | 1656 | |
ba997462 TG |
1657 | cpuhp_tasks_frozen = tasks_frozen; |
1658 | ||
b7ba6d8d | 1659 | cpuhp_set_state(cpu, st, target); |
1cf4f629 TG |
1660 | /* |
1661 | * If the current CPU state is in the range of the AP hotplug thread, | |
1662 | * then we need to kick the thread once more. | |
1663 | */ | |
8df3e07e | 1664 | if (st->state > CPUHP_BRINGUP_CPU) { |
1cf4f629 TG |
1665 | ret = cpuhp_kick_ap_work(cpu); |
1666 | /* | |
1667 | * The AP side has done the error rollback already. Just | |
1668 | * return the error code.. | |
1669 | */ | |
1670 | if (ret) | |
1671 | goto out; | |
1672 | } | |
1673 | ||
1674 | /* | |
1675 | * Try to reach the target state. We max out on the BP at | |
8df3e07e | 1676 | * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is |
1cf4f629 TG |
1677 | * responsible for bringing it up to the target state. |
1678 | */ | |
8df3e07e | 1679 | target = min((int)target, CPUHP_BRINGUP_CPU); |
a724632c | 1680 | ret = cpuhp_up_callbacks(cpu, st, target); |
38498a67 | 1681 | out: |
8f553c49 | 1682 | cpus_write_unlock(); |
a74cfffb | 1683 | arch_smt_update(); |
e3920fb4 RW |
1684 | return ret; |
1685 | } | |
1686 | ||
33c3736e | 1687 | static int cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
1688 | { |
1689 | int err = 0; | |
cf23422b | 1690 | |
e0b582ec | 1691 | if (!cpu_possible(cpu)) { |
84117da5 FF |
1692 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
1693 | cpu); | |
73e753a5 KH |
1694 | return -EINVAL; |
1695 | } | |
e3920fb4 | 1696 | |
01b0f197 TK |
1697 | err = try_online_node(cpu_to_node(cpu)); |
1698 | if (err) | |
1699 | return err; | |
cf23422b | 1700 | |
d221938c | 1701 | cpu_maps_update_begin(); |
e761b772 MK |
1702 | |
1703 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 1704 | err = -EBUSY; |
e761b772 MK |
1705 | goto out; |
1706 | } | |
d91bdd96 | 1707 | if (!cpu_bootable(cpu)) { |
05736e4a TG |
1708 | err = -EPERM; |
1709 | goto out; | |
1710 | } | |
e761b772 | 1711 | |
af1f4045 | 1712 | err = _cpu_up(cpu, 0, target); |
e761b772 | 1713 | out: |
d221938c | 1714 | cpu_maps_update_done(); |
e3920fb4 RW |
1715 | return err; |
1716 | } | |
af1f4045 | 1717 | |
33c3736e QY |
1718 | /** |
1719 | * cpu_device_up - Bring up a cpu device | |
1720 | * @dev: Pointer to the cpu device to online | |
1721 | * | |
1722 | * This function is meant to be used by device core cpu subsystem only. | |
1723 | * | |
1724 | * Other subsystems should use add_cpu() instead. | |
11bc021d RD |
1725 | * |
1726 | * Return: %0 on success or a negative errno code | |
33c3736e QY |
1727 | */ |
1728 | int cpu_device_up(struct device *dev) | |
af1f4045 | 1729 | { |
33c3736e | 1730 | return cpu_up(dev->id, CPUHP_ONLINE); |
af1f4045 | 1731 | } |
e3920fb4 | 1732 | |
93ef1429 QY |
1733 | int add_cpu(unsigned int cpu) |
1734 | { | |
1735 | int ret; | |
1736 | ||
1737 | lock_device_hotplug(); | |
1738 | ret = device_online(get_cpu_device(cpu)); | |
1739 | unlock_device_hotplug(); | |
1740 | ||
1741 | return ret; | |
1742 | } | |
1743 | EXPORT_SYMBOL_GPL(add_cpu); | |
1744 | ||
d720f986 QY |
1745 | /** |
1746 | * bringup_hibernate_cpu - Bring up the CPU that we hibernated on | |
1747 | * @sleep_cpu: The cpu we hibernated on and should be brought up. | |
1748 | * | |
1749 | * On some architectures like arm64, we can hibernate on any CPU, but on | |
1750 | * wake up the CPU we hibernated on might be offline as a side effect of | |
1751 | * using maxcpus= for example. | |
11bc021d RD |
1752 | * |
1753 | * Return: %0 on success or a negative errno code | |
d720f986 QY |
1754 | */ |
1755 | int bringup_hibernate_cpu(unsigned int sleep_cpu) | |
af1f4045 | 1756 | { |
d720f986 QY |
1757 | int ret; |
1758 | ||
1759 | if (!cpu_online(sleep_cpu)) { | |
1760 | pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n"); | |
33c3736e | 1761 | ret = cpu_up(sleep_cpu, CPUHP_ONLINE); |
d720f986 QY |
1762 | if (ret) { |
1763 | pr_err("Failed to bring hibernate-CPU up!\n"); | |
1764 | return ret; | |
1765 | } | |
1766 | } | |
1767 | return 0; | |
1768 | } | |
1769 | ||
18415f33 TG |
1770 | static void __init cpuhp_bringup_mask(const struct cpumask *mask, unsigned int ncpus, |
1771 | enum cpuhp_state target) | |
b99a2659 QY |
1772 | { |
1773 | unsigned int cpu; | |
1774 | ||
18415f33 TG |
1775 | for_each_cpu(cpu, mask) { |
1776 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1777 | ||
18415f33 TG |
1778 | if (cpu_up(cpu, target) && can_rollback_cpu(st)) { |
1779 | /* | |
1780 | * If this failed then cpu_up() might have only | |
1781 | * rolled back to CPUHP_BP_KICK_AP for the final | |
1782 | * online. Clean it up. NOOP if already rolled back. | |
1783 | */ | |
1784 | WARN_ON(cpuhp_invoke_callback_range(false, cpu, st, CPUHP_OFFLINE)); | |
1785 | } | |
06c6796e TG |
1786 | |
1787 | if (!--ncpus) | |
1788 | break; | |
b99a2659 | 1789 | } |
af1f4045 | 1790 | } |
e3920fb4 | 1791 | |
18415f33 TG |
1792 | #ifdef CONFIG_HOTPLUG_PARALLEL |
1793 | static bool __cpuhp_parallel_bringup __ro_after_init = true; | |
1794 | ||
1795 | static int __init parallel_bringup_parse_param(char *arg) | |
1796 | { | |
1797 | return kstrtobool(arg, &__cpuhp_parallel_bringup); | |
1798 | } | |
1799 | early_param("cpuhp.parallel", parallel_bringup_parse_param); | |
1800 | ||
7a4dcb4a LD |
1801 | static inline bool cpuhp_smt_aware(void) |
1802 | { | |
91b4a7db | 1803 | return cpu_smt_max_threads > 1; |
7a4dcb4a LD |
1804 | } |
1805 | ||
1806 | static inline const struct cpumask *cpuhp_get_primary_thread_mask(void) | |
1807 | { | |
1808 | return cpu_primary_thread_mask; | |
1809 | } | |
1810 | ||
18415f33 TG |
1811 | /* |
1812 | * On architectures which have enabled parallel bringup this invokes all BP | |
1813 | * prepare states for each of the to be onlined APs first. The last state | |
1814 | * sends the startup IPI to the APs. The APs proceed through the low level | |
1815 | * bringup code in parallel and then wait for the control CPU to release | |
1816 | * them one by one for the final onlining procedure. | |
1817 | * | |
1818 | * This avoids waiting for each AP to respond to the startup IPI in | |
1819 | * CPUHP_BRINGUP_CPU. | |
1820 | */ | |
1821 | static bool __init cpuhp_bringup_cpus_parallel(unsigned int ncpus) | |
1822 | { | |
1823 | const struct cpumask *mask = cpu_present_mask; | |
1824 | ||
1825 | if (__cpuhp_parallel_bringup) | |
1826 | __cpuhp_parallel_bringup = arch_cpuhp_init_parallel_bringup(); | |
1827 | if (!__cpuhp_parallel_bringup) | |
1828 | return false; | |
1829 | ||
1830 | if (cpuhp_smt_aware()) { | |
1831 | const struct cpumask *pmask = cpuhp_get_primary_thread_mask(); | |
1832 | static struct cpumask tmp_mask __initdata; | |
1833 | ||
1834 | /* | |
1835 | * X86 requires to prevent that SMT siblings stopped while | |
1836 | * the primary thread does a microcode update for various | |
1837 | * reasons. Bring the primary threads up first. | |
1838 | */ | |
1839 | cpumask_and(&tmp_mask, mask, pmask); | |
1840 | cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_BP_KICK_AP); | |
1841 | cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_ONLINE); | |
1842 | /* Account for the online CPUs */ | |
1843 | ncpus -= num_online_cpus(); | |
1844 | if (!ncpus) | |
1845 | return true; | |
1846 | /* Create the mask for secondary CPUs */ | |
1847 | cpumask_andnot(&tmp_mask, mask, pmask); | |
1848 | mask = &tmp_mask; | |
1849 | } | |
1850 | ||
1851 | /* Bring the not-yet started CPUs up */ | |
1852 | cpuhp_bringup_mask(mask, ncpus, CPUHP_BP_KICK_AP); | |
1853 | cpuhp_bringup_mask(mask, ncpus, CPUHP_ONLINE); | |
1854 | return true; | |
1855 | } | |
1856 | #else | |
1857 | static inline bool cpuhp_bringup_cpus_parallel(unsigned int ncpus) { return false; } | |
1858 | #endif /* CONFIG_HOTPLUG_PARALLEL */ | |
1859 | ||
4c8a4985 | 1860 | void __init bringup_nonboot_cpus(unsigned int max_cpus) |
18415f33 | 1861 | { |
6ef8eb51 HC |
1862 | if (!max_cpus) |
1863 | return; | |
1864 | ||
18415f33 | 1865 | /* Try parallel bringup optimization if enabled */ |
4c8a4985 | 1866 | if (cpuhp_bringup_cpus_parallel(max_cpus)) |
18415f33 TG |
1867 | return; |
1868 | ||
1869 | /* Full per CPU serialized bringup */ | |
4c8a4985 | 1870 | cpuhp_bringup_mask(cpu_present_mask, max_cpus, CPUHP_ONLINE); |
18415f33 TG |
1871 | } |
1872 | ||
f3de4be9 | 1873 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 1874 | static cpumask_var_t frozen_cpus; |
e3920fb4 | 1875 | |
fb7fb84a | 1876 | int freeze_secondary_cpus(int primary) |
e3920fb4 | 1877 | { |
d391e552 | 1878 | int cpu, error = 0; |
e3920fb4 | 1879 | |
d221938c | 1880 | cpu_maps_update_begin(); |
9ca12ac0 | 1881 | if (primary == -1) { |
d391e552 | 1882 | primary = cpumask_first(cpu_online_mask); |
04d4e665 FW |
1883 | if (!housekeeping_cpu(primary, HK_TYPE_TIMER)) |
1884 | primary = housekeeping_any_cpu(HK_TYPE_TIMER); | |
9ca12ac0 NP |
1885 | } else { |
1886 | if (!cpu_online(primary)) | |
1887 | primary = cpumask_first(cpu_online_mask); | |
1888 | } | |
1889 | ||
9ee349ad XF |
1890 | /* |
1891 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
1892 | * with the userspace trying to use the CPU hotplug at the same time |
1893 | */ | |
e0b582ec | 1894 | cpumask_clear(frozen_cpus); |
6ad4c188 | 1895 | |
84117da5 | 1896 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 | 1897 | for_each_online_cpu(cpu) { |
d391e552 | 1898 | if (cpu == primary) |
e3920fb4 | 1899 | continue; |
a66d955e | 1900 | |
fb7fb84a | 1901 | if (pm_wakeup_pending()) { |
a66d955e PK |
1902 | pr_info("Wakeup pending. Abort CPU freeze\n"); |
1903 | error = -EBUSY; | |
1904 | break; | |
1905 | } | |
1906 | ||
bb3632c6 | 1907 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 1908 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 1909 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 1910 | if (!error) |
e0b582ec | 1911 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 1912 | else { |
84117da5 | 1913 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
1914 | break; |
1915 | } | |
1916 | } | |
86886e55 | 1917 | |
89af7ba5 | 1918 | if (!error) |
e3920fb4 | 1919 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 1920 | else |
84117da5 | 1921 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
1922 | |
1923 | /* | |
1924 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
56555855 QY |
1925 | * this even in case of failure as all freeze_secondary_cpus() users are |
1926 | * supposed to do thaw_secondary_cpus() on the failure path. | |
89af7ba5 VK |
1927 | */ |
1928 | cpu_hotplug_disabled++; | |
1929 | ||
d221938c | 1930 | cpu_maps_update_done(); |
e3920fb4 RW |
1931 | return error; |
1932 | } | |
1933 | ||
56555855 | 1934 | void __weak arch_thaw_secondary_cpus_begin(void) |
d0af9eed SS |
1935 | { |
1936 | } | |
1937 | ||
56555855 | 1938 | void __weak arch_thaw_secondary_cpus_end(void) |
d0af9eed SS |
1939 | { |
1940 | } | |
1941 | ||
56555855 | 1942 | void thaw_secondary_cpus(void) |
e3920fb4 RW |
1943 | { |
1944 | int cpu, error; | |
1945 | ||
1946 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 1947 | cpu_maps_update_begin(); |
01b41159 | 1948 | __cpu_hotplug_enable(); |
e0b582ec | 1949 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 1950 | goto out; |
e3920fb4 | 1951 | |
84117da5 | 1952 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed | 1953 | |
56555855 | 1954 | arch_thaw_secondary_cpus_begin(); |
d0af9eed | 1955 | |
e0b582ec | 1956 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 1957 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 1958 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 1959 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 1960 | if (!error) { |
84117da5 | 1961 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
1962 | continue; |
1963 | } | |
84117da5 | 1964 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 1965 | } |
d0af9eed | 1966 | |
56555855 | 1967 | arch_thaw_secondary_cpus_end(); |
d0af9eed | 1968 | |
e0b582ec | 1969 | cpumask_clear(frozen_cpus); |
1d64b9cb | 1970 | out: |
d221938c | 1971 | cpu_maps_update_done(); |
1da177e4 | 1972 | } |
e0b582ec | 1973 | |
d7268a31 | 1974 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
1975 | { |
1976 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
1977 | return -ENOMEM; | |
1978 | return 0; | |
1979 | } | |
1980 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 1981 | |
79cfbdfa SB |
1982 | /* |
1983 | * When callbacks for CPU hotplug notifications are being executed, we must | |
1984 | * ensure that the state of the system with respect to the tasks being frozen | |
1985 | * or not, as reported by the notification, remains unchanged *throughout the | |
1986 | * duration* of the execution of the callbacks. | |
1987 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
1988 | * | |
1989 | * This synchronization is implemented by mutually excluding regular CPU | |
1990 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
1991 | * Hibernate notifications. | |
1992 | */ | |
1993 | static int | |
1994 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
1995 | unsigned long action, void *ptr) | |
1996 | { | |
1997 | switch (action) { | |
1998 | ||
1999 | case PM_SUSPEND_PREPARE: | |
2000 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 2001 | cpu_hotplug_disable(); |
79cfbdfa SB |
2002 | break; |
2003 | ||
2004 | case PM_POST_SUSPEND: | |
2005 | case PM_POST_HIBERNATION: | |
16e53dbf | 2006 | cpu_hotplug_enable(); |
79cfbdfa SB |
2007 | break; |
2008 | ||
2009 | default: | |
2010 | return NOTIFY_DONE; | |
2011 | } | |
2012 | ||
2013 | return NOTIFY_OK; | |
2014 | } | |
2015 | ||
2016 | ||
d7268a31 | 2017 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 2018 | { |
6e32d479 FY |
2019 | /* |
2020 | * cpu_hotplug_pm_callback has higher priority than x86 | |
2021 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
2022 | * to disable cpu hotplug to avoid cpu hotplug race. | |
2023 | */ | |
79cfbdfa SB |
2024 | pm_notifier(cpu_hotplug_pm_callback, 0); |
2025 | return 0; | |
2026 | } | |
2027 | core_initcall(cpu_hotplug_pm_sync_init); | |
2028 | ||
f3de4be9 | 2029 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec | 2030 | |
8ce371f9 PZ |
2031 | int __boot_cpu_id; |
2032 | ||
68f4f1ec | 2033 | #endif /* CONFIG_SMP */ |
b8d317d1 | 2034 | |
cff7d378 | 2035 | /* Boot processor state steps */ |
17a2f1ce | 2036 | static struct cpuhp_step cpuhp_hp_states[] = { |
cff7d378 TG |
2037 | [CPUHP_OFFLINE] = { |
2038 | .name = "offline", | |
3c1627e9 TG |
2039 | .startup.single = NULL, |
2040 | .teardown.single = NULL, | |
cff7d378 TG |
2041 | }, |
2042 | #ifdef CONFIG_SMP | |
2043 | [CPUHP_CREATE_THREADS]= { | |
677f6646 | 2044 | .name = "threads:prepare", |
3c1627e9 TG |
2045 | .startup.single = smpboot_create_threads, |
2046 | .teardown.single = NULL, | |
757c989b | 2047 | .cant_stop = true, |
cff7d378 | 2048 | }, |
00e16c3d | 2049 | [CPUHP_PERF_PREPARE] = { |
3c1627e9 TG |
2050 | .name = "perf:prepare", |
2051 | .startup.single = perf_event_init_cpu, | |
2052 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 2053 | }, |
3191dd5a JD |
2054 | [CPUHP_RANDOM_PREPARE] = { |
2055 | .name = "random:prepare", | |
2056 | .startup.single = random_prepare_cpu, | |
2057 | .teardown.single = NULL, | |
2058 | }, | |
7ee681b2 | 2059 | [CPUHP_WORKQUEUE_PREP] = { |
3c1627e9 TG |
2060 | .name = "workqueue:prepare", |
2061 | .startup.single = workqueue_prepare_cpu, | |
2062 | .teardown.single = NULL, | |
7ee681b2 | 2063 | }, |
27590dc1 | 2064 | [CPUHP_HRTIMERS_PREPARE] = { |
3c1627e9 TG |
2065 | .name = "hrtimers:prepare", |
2066 | .startup.single = hrtimers_prepare_cpu, | |
5c0930cc | 2067 | .teardown.single = NULL, |
27590dc1 | 2068 | }, |
31487f83 | 2069 | [CPUHP_SMPCFD_PREPARE] = { |
677f6646 | 2070 | .name = "smpcfd:prepare", |
3c1627e9 TG |
2071 | .startup.single = smpcfd_prepare_cpu, |
2072 | .teardown.single = smpcfd_dead_cpu, | |
31487f83 | 2073 | }, |
e6d4989a RW |
2074 | [CPUHP_RELAY_PREPARE] = { |
2075 | .name = "relay:prepare", | |
2076 | .startup.single = relay_prepare_cpu, | |
2077 | .teardown.single = NULL, | |
2078 | }, | |
4df83742 | 2079 | [CPUHP_RCUTREE_PREP] = { |
677f6646 | 2080 | .name = "RCU/tree:prepare", |
3c1627e9 TG |
2081 | .startup.single = rcutree_prepare_cpu, |
2082 | .teardown.single = rcutree_dead_cpu, | |
4df83742 | 2083 | }, |
4fae16df RC |
2084 | /* |
2085 | * On the tear-down path, timers_dead_cpu() must be invoked | |
2086 | * before blk_mq_queue_reinit_notify() from notify_dead(), | |
2087 | * otherwise a RCU stall occurs. | |
2088 | */ | |
26456f87 | 2089 | [CPUHP_TIMERS_PREPARE] = { |
d018031f | 2090 | .name = "timers:prepare", |
26456f87 | 2091 | .startup.single = timers_prepare_cpu, |
3c1627e9 | 2092 | .teardown.single = timers_dead_cpu, |
4fae16df | 2093 | }, |
a631be92 TG |
2094 | |
2095 | #ifdef CONFIG_HOTPLUG_SPLIT_STARTUP | |
2096 | /* | |
2097 | * Kicks the AP alive. AP will wait in cpuhp_ap_sync_alive() until | |
2098 | * the next step will release it. | |
2099 | */ | |
2100 | [CPUHP_BP_KICK_AP] = { | |
2101 | .name = "cpu:kick_ap", | |
2102 | .startup.single = cpuhp_kick_ap_alive, | |
2103 | }, | |
2104 | ||
2105 | /* | |
2106 | * Waits for the AP to reach cpuhp_ap_sync_alive() and then | |
2107 | * releases it for the complete bringup. | |
2108 | */ | |
2109 | [CPUHP_BRINGUP_CPU] = { | |
2110 | .name = "cpu:bringup", | |
2111 | .startup.single = cpuhp_bringup_ap, | |
2112 | .teardown.single = finish_cpu, | |
2113 | .cant_stop = true, | |
2114 | }, | |
2115 | #else | |
2116 | /* | |
2117 | * All-in-one CPU bringup state which includes the kick alive. | |
2118 | */ | |
cff7d378 TG |
2119 | [CPUHP_BRINGUP_CPU] = { |
2120 | .name = "cpu:bringup", | |
3c1627e9 | 2121 | .startup.single = bringup_cpu, |
bf2c59fc | 2122 | .teardown.single = finish_cpu, |
757c989b | 2123 | .cant_stop = true, |
4baa0afc | 2124 | }, |
a631be92 | 2125 | #endif |
d10ef6f9 TG |
2126 | /* Final state before CPU kills itself */ |
2127 | [CPUHP_AP_IDLE_DEAD] = { | |
2128 | .name = "idle:dead", | |
2129 | }, | |
2130 | /* | |
2131 | * Last state before CPU enters the idle loop to die. Transient state | |
2132 | * for synchronization. | |
2133 | */ | |
2134 | [CPUHP_AP_OFFLINE] = { | |
2135 | .name = "ap:offline", | |
2136 | .cant_stop = true, | |
2137 | }, | |
9cf7243d TG |
2138 | /* First state is scheduler control. Interrupts are disabled */ |
2139 | [CPUHP_AP_SCHED_STARTING] = { | |
2140 | .name = "sched:starting", | |
3c1627e9 TG |
2141 | .startup.single = sched_cpu_starting, |
2142 | .teardown.single = sched_cpu_dying, | |
9cf7243d | 2143 | }, |
4df83742 | 2144 | [CPUHP_AP_RCUTREE_DYING] = { |
677f6646 | 2145 | .name = "RCU/tree:dying", |
3c1627e9 TG |
2146 | .startup.single = NULL, |
2147 | .teardown.single = rcutree_dying_cpu, | |
4baa0afc | 2148 | }, |
46febd37 LJ |
2149 | [CPUHP_AP_SMPCFD_DYING] = { |
2150 | .name = "smpcfd:dying", | |
2151 | .startup.single = NULL, | |
2152 | .teardown.single = smpcfd_dying_cpu, | |
2153 | }, | |
5c0930cc TG |
2154 | [CPUHP_AP_HRTIMERS_DYING] = { |
2155 | .name = "hrtimers:dying", | |
2156 | .startup.single = NULL, | |
2157 | .teardown.single = hrtimers_cpu_dying, | |
2158 | }, | |
3ad6eb06 FW |
2159 | [CPUHP_AP_TICK_DYING] = { |
2160 | .name = "tick:dying", | |
2161 | .startup.single = NULL, | |
2162 | .teardown.single = tick_cpu_dying, | |
2163 | }, | |
d10ef6f9 TG |
2164 | /* Entry state on starting. Interrupts enabled from here on. Transient |
2165 | * state for synchronsization */ | |
2166 | [CPUHP_AP_ONLINE] = { | |
2167 | .name = "ap:online", | |
2168 | }, | |
17a2f1ce | 2169 | /* |
1cf12e08 | 2170 | * Handled on control processor until the plugged processor manages |
17a2f1ce LJ |
2171 | * this itself. |
2172 | */ | |
2173 | [CPUHP_TEARDOWN_CPU] = { | |
2174 | .name = "cpu:teardown", | |
2175 | .startup.single = NULL, | |
2176 | .teardown.single = takedown_cpu, | |
2177 | .cant_stop = true, | |
2178 | }, | |
1cf12e08 TG |
2179 | |
2180 | [CPUHP_AP_SCHED_WAIT_EMPTY] = { | |
2181 | .name = "sched:waitempty", | |
2182 | .startup.single = NULL, | |
2183 | .teardown.single = sched_cpu_wait_empty, | |
2184 | }, | |
2185 | ||
d10ef6f9 | 2186 | /* Handle smpboot threads park/unpark */ |
1cf4f629 | 2187 | [CPUHP_AP_SMPBOOT_THREADS] = { |
677f6646 | 2188 | .name = "smpboot/threads:online", |
3c1627e9 | 2189 | .startup.single = smpboot_unpark_threads, |
c4de6569 | 2190 | .teardown.single = smpboot_park_threads, |
1cf4f629 | 2191 | }, |
c5cb83bb TG |
2192 | [CPUHP_AP_IRQ_AFFINITY_ONLINE] = { |
2193 | .name = "irq/affinity:online", | |
2194 | .startup.single = irq_affinity_online_cpu, | |
2195 | .teardown.single = NULL, | |
2196 | }, | |
00e16c3d | 2197 | [CPUHP_AP_PERF_ONLINE] = { |
3c1627e9 TG |
2198 | .name = "perf:online", |
2199 | .startup.single = perf_event_init_cpu, | |
2200 | .teardown.single = perf_event_exit_cpu, | |
00e16c3d | 2201 | }, |
9cf57731 PZ |
2202 | [CPUHP_AP_WATCHDOG_ONLINE] = { |
2203 | .name = "lockup_detector:online", | |
2204 | .startup.single = lockup_detector_online_cpu, | |
2205 | .teardown.single = lockup_detector_offline_cpu, | |
2206 | }, | |
7ee681b2 | 2207 | [CPUHP_AP_WORKQUEUE_ONLINE] = { |
3c1627e9 TG |
2208 | .name = "workqueue:online", |
2209 | .startup.single = workqueue_online_cpu, | |
2210 | .teardown.single = workqueue_offline_cpu, | |
7ee681b2 | 2211 | }, |
3191dd5a JD |
2212 | [CPUHP_AP_RANDOM_ONLINE] = { |
2213 | .name = "random:online", | |
2214 | .startup.single = random_online_cpu, | |
2215 | .teardown.single = NULL, | |
2216 | }, | |
4df83742 | 2217 | [CPUHP_AP_RCUTREE_ONLINE] = { |
677f6646 | 2218 | .name = "RCU/tree:online", |
3c1627e9 TG |
2219 | .startup.single = rcutree_online_cpu, |
2220 | .teardown.single = rcutree_offline_cpu, | |
4df83742 | 2221 | }, |
4baa0afc | 2222 | #endif |
d10ef6f9 TG |
2223 | /* |
2224 | * The dynamically registered state space is here | |
2225 | */ | |
2226 | ||
aaddd7d1 TG |
2227 | #ifdef CONFIG_SMP |
2228 | /* Last state is scheduler control setting the cpu active */ | |
2229 | [CPUHP_AP_ACTIVE] = { | |
2230 | .name = "sched:active", | |
3c1627e9 TG |
2231 | .startup.single = sched_cpu_activate, |
2232 | .teardown.single = sched_cpu_deactivate, | |
aaddd7d1 TG |
2233 | }, |
2234 | #endif | |
2235 | ||
d10ef6f9 | 2236 | /* CPU is fully up and running. */ |
4baa0afc TG |
2237 | [CPUHP_ONLINE] = { |
2238 | .name = "online", | |
3c1627e9 TG |
2239 | .startup.single = NULL, |
2240 | .teardown.single = NULL, | |
4baa0afc TG |
2241 | }, |
2242 | }; | |
2243 | ||
5b7aa87e TG |
2244 | /* Sanity check for callbacks */ |
2245 | static int cpuhp_cb_check(enum cpuhp_state state) | |
2246 | { | |
2247 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
2248 | return -EINVAL; | |
2249 | return 0; | |
2250 | } | |
2251 | ||
dc280d93 TG |
2252 | /* |
2253 | * Returns a free for dynamic slot assignment of the Online state. The states | |
2254 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
2255 | * by having no name assigned. | |
2256 | */ | |
2257 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
2258 | { | |
4205e478 TG |
2259 | enum cpuhp_state i, end; |
2260 | struct cpuhp_step *step; | |
dc280d93 | 2261 | |
4205e478 TG |
2262 | switch (state) { |
2263 | case CPUHP_AP_ONLINE_DYN: | |
17a2f1ce | 2264 | step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN; |
4205e478 TG |
2265 | end = CPUHP_AP_ONLINE_DYN_END; |
2266 | break; | |
2267 | case CPUHP_BP_PREPARE_DYN: | |
17a2f1ce | 2268 | step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN; |
4205e478 TG |
2269 | end = CPUHP_BP_PREPARE_DYN_END; |
2270 | break; | |
2271 | default: | |
2272 | return -EINVAL; | |
2273 | } | |
2274 | ||
2275 | for (i = state; i <= end; i++, step++) { | |
2276 | if (!step->name) | |
dc280d93 TG |
2277 | return i; |
2278 | } | |
2279 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
2280 | return -ENOSPC; | |
2281 | } | |
2282 | ||
2283 | static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name, | |
2284 | int (*startup)(unsigned int cpu), | |
2285 | int (*teardown)(unsigned int cpu), | |
2286 | bool multi_instance) | |
5b7aa87e TG |
2287 | { |
2288 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
2289 | struct cpuhp_step *sp; | |
dc280d93 | 2290 | int ret = 0; |
5b7aa87e | 2291 | |
0c96b273 EB |
2292 | /* |
2293 | * If name is NULL, then the state gets removed. | |
2294 | * | |
2295 | * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on | |
2296 | * the first allocation from these dynamic ranges, so the removal | |
2297 | * would trigger a new allocation and clear the wrong (already | |
2298 | * empty) state, leaving the callbacks of the to be cleared state | |
2299 | * dangling, which causes wreckage on the next hotplug operation. | |
2300 | */ | |
2301 | if (name && (state == CPUHP_AP_ONLINE_DYN || | |
2302 | state == CPUHP_BP_PREPARE_DYN)) { | |
dc280d93 TG |
2303 | ret = cpuhp_reserve_state(state); |
2304 | if (ret < 0) | |
dc434e05 | 2305 | return ret; |
dc280d93 TG |
2306 | state = ret; |
2307 | } | |
5b7aa87e | 2308 | sp = cpuhp_get_step(state); |
dc434e05 SAS |
2309 | if (name && sp->name) |
2310 | return -EBUSY; | |
2311 | ||
3c1627e9 TG |
2312 | sp->startup.single = startup; |
2313 | sp->teardown.single = teardown; | |
5b7aa87e | 2314 | sp->name = name; |
cf392d10 TG |
2315 | sp->multi_instance = multi_instance; |
2316 | INIT_HLIST_HEAD(&sp->list); | |
dc280d93 | 2317 | return ret; |
5b7aa87e TG |
2318 | } |
2319 | ||
2320 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
2321 | { | |
3c1627e9 | 2322 | return cpuhp_get_step(state)->teardown.single; |
5b7aa87e TG |
2323 | } |
2324 | ||
5b7aa87e TG |
2325 | /* |
2326 | * Call the startup/teardown function for a step either on the AP or | |
2327 | * on the current CPU. | |
2328 | */ | |
cf392d10 TG |
2329 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup, |
2330 | struct hlist_node *node) | |
5b7aa87e | 2331 | { |
a724632c | 2332 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
2333 | int ret; |
2334 | ||
4dddfb5f PZ |
2335 | /* |
2336 | * If there's nothing to do, we done. | |
2337 | * Relies on the union for multi_instance. | |
2338 | */ | |
453e4108 | 2339 | if (cpuhp_step_empty(bringup, sp)) |
5b7aa87e | 2340 | return 0; |
5b7aa87e TG |
2341 | /* |
2342 | * The non AP bound callbacks can fail on bringup. On teardown | |
2343 | * e.g. module removal we crash for now. | |
2344 | */ | |
1cf4f629 TG |
2345 | #ifdef CONFIG_SMP |
2346 | if (cpuhp_is_ap_state(state)) | |
cf392d10 | 2347 | ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node); |
1cf4f629 | 2348 | else |
96abb968 | 2349 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 2350 | #else |
96abb968 | 2351 | ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL); |
1cf4f629 | 2352 | #endif |
5b7aa87e TG |
2353 | BUG_ON(ret && !bringup); |
2354 | return ret; | |
2355 | } | |
2356 | ||
2357 | /* | |
2358 | * Called from __cpuhp_setup_state on a recoverable failure. | |
2359 | * | |
2360 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
2361 | */ | |
2362 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
cf392d10 | 2363 | struct hlist_node *node) |
5b7aa87e TG |
2364 | { |
2365 | int cpu; | |
2366 | ||
5b7aa87e TG |
2367 | /* Roll back the already executed steps on the other cpus */ |
2368 | for_each_present_cpu(cpu) { | |
2369 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2370 | int cpustate = st->state; | |
2371 | ||
2372 | if (cpu >= failedcpu) | |
2373 | break; | |
2374 | ||
2375 | /* Did we invoke the startup call on that cpu ? */ | |
2376 | if (cpustate >= state) | |
cf392d10 | 2377 | cpuhp_issue_call(cpu, state, false, node); |
5b7aa87e TG |
2378 | } |
2379 | } | |
2380 | ||
9805c673 TG |
2381 | int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state, |
2382 | struct hlist_node *node, | |
2383 | bool invoke) | |
cf392d10 TG |
2384 | { |
2385 | struct cpuhp_step *sp; | |
2386 | int cpu; | |
2387 | int ret; | |
2388 | ||
9805c673 TG |
2389 | lockdep_assert_cpus_held(); |
2390 | ||
cf392d10 TG |
2391 | sp = cpuhp_get_step(state); |
2392 | if (sp->multi_instance == false) | |
2393 | return -EINVAL; | |
2394 | ||
dc434e05 | 2395 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 | 2396 | |
3c1627e9 | 2397 | if (!invoke || !sp->startup.multi) |
cf392d10 TG |
2398 | goto add_node; |
2399 | ||
2400 | /* | |
2401 | * Try to call the startup callback for each present cpu | |
2402 | * depending on the hotplug state of the cpu. | |
2403 | */ | |
2404 | for_each_present_cpu(cpu) { | |
2405 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2406 | int cpustate = st->state; | |
2407 | ||
2408 | if (cpustate < state) | |
2409 | continue; | |
2410 | ||
2411 | ret = cpuhp_issue_call(cpu, state, true, node); | |
2412 | if (ret) { | |
3c1627e9 | 2413 | if (sp->teardown.multi) |
cf392d10 | 2414 | cpuhp_rollback_install(cpu, state, node); |
dc434e05 | 2415 | goto unlock; |
cf392d10 TG |
2416 | } |
2417 | } | |
2418 | add_node: | |
2419 | ret = 0; | |
cf392d10 | 2420 | hlist_add_head(node, &sp->list); |
dc434e05 | 2421 | unlock: |
cf392d10 | 2422 | mutex_unlock(&cpuhp_state_mutex); |
9805c673 TG |
2423 | return ret; |
2424 | } | |
2425 | ||
2426 | int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node, | |
2427 | bool invoke) | |
2428 | { | |
2429 | int ret; | |
2430 | ||
2431 | cpus_read_lock(); | |
2432 | ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke); | |
8f553c49 | 2433 | cpus_read_unlock(); |
cf392d10 TG |
2434 | return ret; |
2435 | } | |
2436 | EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance); | |
2437 | ||
5b7aa87e | 2438 | /** |
71def423 | 2439 | * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state |
dc280d93 | 2440 | * @state: The state to setup |
ed3cd1da | 2441 | * @name: Name of the step |
dc280d93 TG |
2442 | * @invoke: If true, the startup function is invoked for cpus where |
2443 | * cpu state >= @state | |
2444 | * @startup: startup callback function | |
2445 | * @teardown: teardown callback function | |
2446 | * @multi_instance: State is set up for multiple instances which get | |
2447 | * added afterwards. | |
5b7aa87e | 2448 | * |
71def423 | 2449 | * The caller needs to hold cpus read locked while calling this function. |
11bc021d | 2450 | * Return: |
512f0980 | 2451 | * On success: |
932d8476 | 2452 | * Positive state number if @state is CPUHP_AP_ONLINE_DYN or CPUHP_BP_PREPARE_DYN; |
512f0980 BO |
2453 | * 0 for all other states |
2454 | * On failure: proper (negative) error code | |
5b7aa87e | 2455 | */ |
71def423 SAS |
2456 | int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state, |
2457 | const char *name, bool invoke, | |
2458 | int (*startup)(unsigned int cpu), | |
2459 | int (*teardown)(unsigned int cpu), | |
2460 | bool multi_instance) | |
5b7aa87e TG |
2461 | { |
2462 | int cpu, ret = 0; | |
b9d9d691 | 2463 | bool dynstate; |
5b7aa87e | 2464 | |
71def423 SAS |
2465 | lockdep_assert_cpus_held(); |
2466 | ||
5b7aa87e TG |
2467 | if (cpuhp_cb_check(state) || !name) |
2468 | return -EINVAL; | |
2469 | ||
dc434e05 | 2470 | mutex_lock(&cpuhp_state_mutex); |
5b7aa87e | 2471 | |
dc280d93 TG |
2472 | ret = cpuhp_store_callbacks(state, name, startup, teardown, |
2473 | multi_instance); | |
5b7aa87e | 2474 | |
932d8476 | 2475 | dynstate = state == CPUHP_AP_ONLINE_DYN || state == CPUHP_BP_PREPARE_DYN; |
b9d9d691 TG |
2476 | if (ret > 0 && dynstate) { |
2477 | state = ret; | |
2478 | ret = 0; | |
2479 | } | |
2480 | ||
dc280d93 | 2481 | if (ret || !invoke || !startup) |
5b7aa87e TG |
2482 | goto out; |
2483 | ||
2484 | /* | |
2485 | * Try to call the startup callback for each present cpu | |
2486 | * depending on the hotplug state of the cpu. | |
2487 | */ | |
2488 | for_each_present_cpu(cpu) { | |
2489 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2490 | int cpustate = st->state; | |
2491 | ||
2492 | if (cpustate < state) | |
2493 | continue; | |
2494 | ||
cf392d10 | 2495 | ret = cpuhp_issue_call(cpu, state, true, NULL); |
5b7aa87e | 2496 | if (ret) { |
a724632c | 2497 | if (teardown) |
cf392d10 TG |
2498 | cpuhp_rollback_install(cpu, state, NULL); |
2499 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); | |
5b7aa87e TG |
2500 | goto out; |
2501 | } | |
2502 | } | |
2503 | out: | |
dc434e05 | 2504 | mutex_unlock(&cpuhp_state_mutex); |
dc280d93 | 2505 | /* |
932d8476 YW |
2506 | * If the requested state is CPUHP_AP_ONLINE_DYN or CPUHP_BP_PREPARE_DYN, |
2507 | * return the dynamically allocated state in case of success. | |
dc280d93 | 2508 | */ |
b9d9d691 | 2509 | if (!ret && dynstate) |
5b7aa87e TG |
2510 | return state; |
2511 | return ret; | |
2512 | } | |
71def423 SAS |
2513 | EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked); |
2514 | ||
2515 | int __cpuhp_setup_state(enum cpuhp_state state, | |
2516 | const char *name, bool invoke, | |
2517 | int (*startup)(unsigned int cpu), | |
2518 | int (*teardown)(unsigned int cpu), | |
2519 | bool multi_instance) | |
2520 | { | |
2521 | int ret; | |
2522 | ||
2523 | cpus_read_lock(); | |
2524 | ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup, | |
2525 | teardown, multi_instance); | |
2526 | cpus_read_unlock(); | |
2527 | return ret; | |
2528 | } | |
5b7aa87e TG |
2529 | EXPORT_SYMBOL(__cpuhp_setup_state); |
2530 | ||
cf392d10 TG |
2531 | int __cpuhp_state_remove_instance(enum cpuhp_state state, |
2532 | struct hlist_node *node, bool invoke) | |
2533 | { | |
2534 | struct cpuhp_step *sp = cpuhp_get_step(state); | |
2535 | int cpu; | |
2536 | ||
2537 | BUG_ON(cpuhp_cb_check(state)); | |
2538 | ||
2539 | if (!sp->multi_instance) | |
2540 | return -EINVAL; | |
2541 | ||
8f553c49 | 2542 | cpus_read_lock(); |
dc434e05 SAS |
2543 | mutex_lock(&cpuhp_state_mutex); |
2544 | ||
cf392d10 TG |
2545 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
2546 | goto remove; | |
2547 | /* | |
2548 | * Call the teardown callback for each present cpu depending | |
2549 | * on the hotplug state of the cpu. This function is not | |
2550 | * allowed to fail currently! | |
2551 | */ | |
2552 | for_each_present_cpu(cpu) { | |
2553 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2554 | int cpustate = st->state; | |
2555 | ||
2556 | if (cpustate >= state) | |
2557 | cpuhp_issue_call(cpu, state, false, node); | |
2558 | } | |
2559 | ||
2560 | remove: | |
cf392d10 TG |
2561 | hlist_del(node); |
2562 | mutex_unlock(&cpuhp_state_mutex); | |
8f553c49 | 2563 | cpus_read_unlock(); |
cf392d10 TG |
2564 | |
2565 | return 0; | |
2566 | } | |
2567 | EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance); | |
dc434e05 | 2568 | |
5b7aa87e | 2569 | /** |
71def423 | 2570 | * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state |
5b7aa87e TG |
2571 | * @state: The state to remove |
2572 | * @invoke: If true, the teardown function is invoked for cpus where | |
2573 | * cpu state >= @state | |
2574 | * | |
71def423 | 2575 | * The caller needs to hold cpus read locked while calling this function. |
5b7aa87e TG |
2576 | * The teardown callback is currently not allowed to fail. Think |
2577 | * about module removal! | |
2578 | */ | |
71def423 | 2579 | void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke) |
5b7aa87e | 2580 | { |
cf392d10 | 2581 | struct cpuhp_step *sp = cpuhp_get_step(state); |
5b7aa87e TG |
2582 | int cpu; |
2583 | ||
2584 | BUG_ON(cpuhp_cb_check(state)); | |
2585 | ||
71def423 | 2586 | lockdep_assert_cpus_held(); |
5b7aa87e | 2587 | |
dc434e05 | 2588 | mutex_lock(&cpuhp_state_mutex); |
cf392d10 TG |
2589 | if (sp->multi_instance) { |
2590 | WARN(!hlist_empty(&sp->list), | |
2591 | "Error: Removing state %d which has instances left.\n", | |
2592 | state); | |
2593 | goto remove; | |
2594 | } | |
2595 | ||
a724632c | 2596 | if (!invoke || !cpuhp_get_teardown_cb(state)) |
5b7aa87e TG |
2597 | goto remove; |
2598 | ||
2599 | /* | |
2600 | * Call the teardown callback for each present cpu depending | |
2601 | * on the hotplug state of the cpu. This function is not | |
2602 | * allowed to fail currently! | |
2603 | */ | |
2604 | for_each_present_cpu(cpu) { | |
2605 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
2606 | int cpustate = st->state; | |
2607 | ||
2608 | if (cpustate >= state) | |
cf392d10 | 2609 | cpuhp_issue_call(cpu, state, false, NULL); |
5b7aa87e TG |
2610 | } |
2611 | remove: | |
cf392d10 | 2612 | cpuhp_store_callbacks(state, NULL, NULL, NULL, false); |
dc434e05 | 2613 | mutex_unlock(&cpuhp_state_mutex); |
71def423 SAS |
2614 | } |
2615 | EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked); | |
2616 | ||
2617 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
2618 | { | |
2619 | cpus_read_lock(); | |
2620 | __cpuhp_remove_state_cpuslocked(state, invoke); | |
8f553c49 | 2621 | cpus_read_unlock(); |
5b7aa87e TG |
2622 | } |
2623 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
2624 | ||
dc8d37ed AB |
2625 | #ifdef CONFIG_HOTPLUG_SMT |
2626 | static void cpuhp_offline_cpu_device(unsigned int cpu) | |
2627 | { | |
2628 | struct device *dev = get_cpu_device(cpu); | |
2629 | ||
2630 | dev->offline = true; | |
2631 | /* Tell user space about the state change */ | |
2632 | kobject_uevent(&dev->kobj, KOBJ_OFFLINE); | |
2633 | } | |
2634 | ||
2635 | static void cpuhp_online_cpu_device(unsigned int cpu) | |
2636 | { | |
2637 | struct device *dev = get_cpu_device(cpu); | |
2638 | ||
2639 | dev->offline = false; | |
2640 | /* Tell user space about the state change */ | |
2641 | kobject_uevent(&dev->kobj, KOBJ_ONLINE); | |
2642 | } | |
2643 | ||
2644 | int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval) | |
2645 | { | |
2646 | int cpu, ret = 0; | |
2647 | ||
2648 | cpu_maps_update_begin(); | |
2649 | for_each_online_cpu(cpu) { | |
2650 | if (topology_is_primary_thread(cpu)) | |
2651 | continue; | |
38253464 ME |
2652 | /* |
2653 | * Disable can be called with CPU_SMT_ENABLED when changing | |
2654 | * from a higher to lower number of SMT threads per core. | |
2655 | */ | |
2656 | if (ctrlval == CPU_SMT_ENABLED && cpu_smt_thread_allowed(cpu)) | |
2657 | continue; | |
dc8d37ed AB |
2658 | ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE); |
2659 | if (ret) | |
2660 | break; | |
2661 | /* | |
2662 | * As this needs to hold the cpu maps lock it's impossible | |
2663 | * to call device_offline() because that ends up calling | |
2664 | * cpu_down() which takes cpu maps lock. cpu maps lock | |
2665 | * needs to be held as this might race against in kernel | |
2666 | * abusers of the hotplug machinery (thermal management). | |
2667 | * | |
2668 | * So nothing would update device:offline state. That would | |
2669 | * leave the sysfs entry stale and prevent onlining after | |
2670 | * smt control has been changed to 'off' again. This is | |
2671 | * called under the sysfs hotplug lock, so it is properly | |
2672 | * serialized against the regular offline usage. | |
2673 | */ | |
2674 | cpuhp_offline_cpu_device(cpu); | |
2675 | } | |
2676 | if (!ret) | |
2677 | cpu_smt_control = ctrlval; | |
2678 | cpu_maps_update_done(); | |
2679 | return ret; | |
2680 | } | |
2681 | ||
2682 | int cpuhp_smt_enable(void) | |
2683 | { | |
2684 | int cpu, ret = 0; | |
2685 | ||
2686 | cpu_maps_update_begin(); | |
2687 | cpu_smt_control = CPU_SMT_ENABLED; | |
2688 | for_each_present_cpu(cpu) { | |
2689 | /* Skip online CPUs and CPUs on offline nodes */ | |
2690 | if (cpu_online(cpu) || !node_online(cpu_to_node(cpu))) | |
2691 | continue; | |
38253464 ME |
2692 | if (!cpu_smt_thread_allowed(cpu)) |
2693 | continue; | |
dc8d37ed AB |
2694 | ret = _cpu_up(cpu, 0, CPUHP_ONLINE); |
2695 | if (ret) | |
2696 | break; | |
2697 | /* See comment in cpuhp_smt_disable() */ | |
2698 | cpuhp_online_cpu_device(cpu); | |
2699 | } | |
2700 | cpu_maps_update_done(); | |
2701 | return ret; | |
2702 | } | |
2703 | #endif | |
2704 | ||
98f8cdce | 2705 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1782dc87 Y |
2706 | static ssize_t state_show(struct device *dev, |
2707 | struct device_attribute *attr, char *buf) | |
98f8cdce TG |
2708 | { |
2709 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2710 | ||
2711 | return sprintf(buf, "%d\n", st->state); | |
2712 | } | |
1782dc87 | 2713 | static DEVICE_ATTR_RO(state); |
98f8cdce | 2714 | |
1782dc87 Y |
2715 | static ssize_t target_store(struct device *dev, struct device_attribute *attr, |
2716 | const char *buf, size_t count) | |
757c989b TG |
2717 | { |
2718 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2719 | struct cpuhp_step *sp; | |
2720 | int target, ret; | |
2721 | ||
2722 | ret = kstrtoint(buf, 10, &target); | |
2723 | if (ret) | |
2724 | return ret; | |
2725 | ||
2726 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
2727 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
2728 | return -EINVAL; | |
2729 | #else | |
2730 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
2731 | return -EINVAL; | |
2732 | #endif | |
2733 | ||
2734 | ret = lock_device_hotplug_sysfs(); | |
2735 | if (ret) | |
2736 | return ret; | |
2737 | ||
2738 | mutex_lock(&cpuhp_state_mutex); | |
2739 | sp = cpuhp_get_step(target); | |
2740 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
2741 | mutex_unlock(&cpuhp_state_mutex); | |
2742 | if (ret) | |
40da1b11 | 2743 | goto out; |
757c989b TG |
2744 | |
2745 | if (st->state < target) | |
33c3736e | 2746 | ret = cpu_up(dev->id, target); |
64ea6e44 | 2747 | else if (st->state > target) |
33c3736e | 2748 | ret = cpu_down(dev->id, target); |
64ea6e44 PA |
2749 | else if (WARN_ON(st->target != target)) |
2750 | st->target = target; | |
40da1b11 | 2751 | out: |
757c989b TG |
2752 | unlock_device_hotplug(); |
2753 | return ret ? ret : count; | |
2754 | } | |
2755 | ||
1782dc87 Y |
2756 | static ssize_t target_show(struct device *dev, |
2757 | struct device_attribute *attr, char *buf) | |
98f8cdce TG |
2758 | { |
2759 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2760 | ||
2761 | return sprintf(buf, "%d\n", st->target); | |
2762 | } | |
1782dc87 | 2763 | static DEVICE_ATTR_RW(target); |
1db49484 | 2764 | |
1782dc87 Y |
2765 | static ssize_t fail_store(struct device *dev, struct device_attribute *attr, |
2766 | const char *buf, size_t count) | |
1db49484 PZ |
2767 | { |
2768 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2769 | struct cpuhp_step *sp; | |
2770 | int fail, ret; | |
2771 | ||
2772 | ret = kstrtoint(buf, 10, &fail); | |
2773 | if (ret) | |
2774 | return ret; | |
2775 | ||
3ae70c25 VD |
2776 | if (fail == CPUHP_INVALID) { |
2777 | st->fail = fail; | |
2778 | return count; | |
2779 | } | |
2780 | ||
33d4a5a7 ET |
2781 | if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE) |
2782 | return -EINVAL; | |
2783 | ||
1db49484 PZ |
2784 | /* |
2785 | * Cannot fail STARTING/DYING callbacks. | |
2786 | */ | |
2787 | if (cpuhp_is_atomic_state(fail)) | |
2788 | return -EINVAL; | |
2789 | ||
62f25069 VD |
2790 | /* |
2791 | * DEAD callbacks cannot fail... | |
2792 | * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter | |
2793 | * triggering STARTING callbacks, a failure in this state would | |
2794 | * hinder rollback. | |
2795 | */ | |
2796 | if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU) | |
2797 | return -EINVAL; | |
2798 | ||
1db49484 PZ |
2799 | /* |
2800 | * Cannot fail anything that doesn't have callbacks. | |
2801 | */ | |
2802 | mutex_lock(&cpuhp_state_mutex); | |
2803 | sp = cpuhp_get_step(fail); | |
2804 | if (!sp->startup.single && !sp->teardown.single) | |
2805 | ret = -EINVAL; | |
2806 | mutex_unlock(&cpuhp_state_mutex); | |
2807 | if (ret) | |
2808 | return ret; | |
2809 | ||
2810 | st->fail = fail; | |
2811 | ||
2812 | return count; | |
2813 | } | |
2814 | ||
1782dc87 Y |
2815 | static ssize_t fail_show(struct device *dev, |
2816 | struct device_attribute *attr, char *buf) | |
1db49484 PZ |
2817 | { |
2818 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
2819 | ||
2820 | return sprintf(buf, "%d\n", st->fail); | |
2821 | } | |
2822 | ||
1782dc87 | 2823 | static DEVICE_ATTR_RW(fail); |
1db49484 | 2824 | |
98f8cdce TG |
2825 | static struct attribute *cpuhp_cpu_attrs[] = { |
2826 | &dev_attr_state.attr, | |
2827 | &dev_attr_target.attr, | |
1db49484 | 2828 | &dev_attr_fail.attr, |
98f8cdce TG |
2829 | NULL |
2830 | }; | |
2831 | ||
993647a2 | 2832 | static const struct attribute_group cpuhp_cpu_attr_group = { |
98f8cdce TG |
2833 | .attrs = cpuhp_cpu_attrs, |
2834 | .name = "hotplug", | |
2835 | NULL | |
2836 | }; | |
2837 | ||
1782dc87 | 2838 | static ssize_t states_show(struct device *dev, |
98f8cdce TG |
2839 | struct device_attribute *attr, char *buf) |
2840 | { | |
2841 | ssize_t cur, res = 0; | |
2842 | int i; | |
2843 | ||
2844 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 2845 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
2846 | struct cpuhp_step *sp = cpuhp_get_step(i); |
2847 | ||
2848 | if (sp->name) { | |
2849 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
2850 | buf += cur; | |
2851 | res += cur; | |
2852 | } | |
2853 | } | |
2854 | mutex_unlock(&cpuhp_state_mutex); | |
2855 | return res; | |
2856 | } | |
1782dc87 | 2857 | static DEVICE_ATTR_RO(states); |
98f8cdce TG |
2858 | |
2859 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
2860 | &dev_attr_states.attr, | |
2861 | NULL | |
2862 | }; | |
2863 | ||
993647a2 | 2864 | static const struct attribute_group cpuhp_cpu_root_attr_group = { |
98f8cdce TG |
2865 | .attrs = cpuhp_cpu_root_attrs, |
2866 | .name = "hotplug", | |
2867 | NULL | |
2868 | }; | |
2869 | ||
05736e4a TG |
2870 | #ifdef CONFIG_HOTPLUG_SMT |
2871 | ||
7f48405c ME |
2872 | static bool cpu_smt_num_threads_valid(unsigned int threads) |
2873 | { | |
2874 | if (IS_ENABLED(CONFIG_SMT_NUM_THREADS_DYNAMIC)) | |
2875 | return threads >= 1 && threads <= cpu_smt_max_threads; | |
2876 | return threads == 1 || threads == cpu_smt_max_threads; | |
2877 | } | |
2878 | ||
05736e4a | 2879 | static ssize_t |
de7b77e5 JP |
2880 | __store_smt_control(struct device *dev, struct device_attribute *attr, |
2881 | const char *buf, size_t count) | |
05736e4a | 2882 | { |
7f48405c ME |
2883 | int ctrlval, ret, num_threads, orig_threads; |
2884 | bool force_off; | |
05736e4a | 2885 | |
c53361ce ME |
2886 | if (cpu_smt_control == CPU_SMT_FORCE_DISABLED) |
2887 | return -EPERM; | |
2888 | ||
2889 | if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED) | |
2890 | return -ENODEV; | |
2891 | ||
7f48405c | 2892 | if (sysfs_streq(buf, "on")) { |
05736e4a | 2893 | ctrlval = CPU_SMT_ENABLED; |
7f48405c ME |
2894 | num_threads = cpu_smt_max_threads; |
2895 | } else if (sysfs_streq(buf, "off")) { | |
05736e4a | 2896 | ctrlval = CPU_SMT_DISABLED; |
7f48405c ME |
2897 | num_threads = 1; |
2898 | } else if (sysfs_streq(buf, "forceoff")) { | |
05736e4a | 2899 | ctrlval = CPU_SMT_FORCE_DISABLED; |
7f48405c ME |
2900 | num_threads = 1; |
2901 | } else if (kstrtoint(buf, 10, &num_threads) == 0) { | |
2902 | if (num_threads == 1) | |
2903 | ctrlval = CPU_SMT_DISABLED; | |
2904 | else if (cpu_smt_num_threads_valid(num_threads)) | |
2905 | ctrlval = CPU_SMT_ENABLED; | |
2906 | else | |
2907 | return -EINVAL; | |
2908 | } else { | |
05736e4a | 2909 | return -EINVAL; |
7f48405c | 2910 | } |
05736e4a | 2911 | |
05736e4a TG |
2912 | ret = lock_device_hotplug_sysfs(); |
2913 | if (ret) | |
2914 | return ret; | |
2915 | ||
7f48405c ME |
2916 | orig_threads = cpu_smt_num_threads; |
2917 | cpu_smt_num_threads = num_threads; | |
2918 | ||
2919 | force_off = ctrlval != cpu_smt_control && ctrlval == CPU_SMT_FORCE_DISABLED; | |
2920 | ||
2921 | if (num_threads > orig_threads) | |
2922 | ret = cpuhp_smt_enable(); | |
2923 | else if (num_threads < orig_threads || force_off) | |
2924 | ret = cpuhp_smt_disable(ctrlval); | |
05736e4a TG |
2925 | |
2926 | unlock_device_hotplug(); | |
2927 | return ret ? ret : count; | |
2928 | } | |
de7b77e5 JP |
2929 | |
2930 | #else /* !CONFIG_HOTPLUG_SMT */ | |
2931 | static ssize_t | |
2932 | __store_smt_control(struct device *dev, struct device_attribute *attr, | |
2933 | const char *buf, size_t count) | |
2934 | { | |
2935 | return -ENODEV; | |
2936 | } | |
2937 | #endif /* CONFIG_HOTPLUG_SMT */ | |
2938 | ||
2939 | static const char *smt_states[] = { | |
2940 | [CPU_SMT_ENABLED] = "on", | |
2941 | [CPU_SMT_DISABLED] = "off", | |
2942 | [CPU_SMT_FORCE_DISABLED] = "forceoff", | |
2943 | [CPU_SMT_NOT_SUPPORTED] = "notsupported", | |
2944 | [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented", | |
2945 | }; | |
2946 | ||
1782dc87 Y |
2947 | static ssize_t control_show(struct device *dev, |
2948 | struct device_attribute *attr, char *buf) | |
de7b77e5 JP |
2949 | { |
2950 | const char *state = smt_states[cpu_smt_control]; | |
2951 | ||
7f48405c ME |
2952 | #ifdef CONFIG_HOTPLUG_SMT |
2953 | /* | |
2954 | * If SMT is enabled but not all threads are enabled then show the | |
2955 | * number of threads. If all threads are enabled show "on". Otherwise | |
2956 | * show the state name. | |
2957 | */ | |
2958 | if (cpu_smt_control == CPU_SMT_ENABLED && | |
2959 | cpu_smt_num_threads != cpu_smt_max_threads) | |
2960 | return sysfs_emit(buf, "%d\n", cpu_smt_num_threads); | |
2961 | #endif | |
2962 | ||
effe6d27 | 2963 | return sysfs_emit(buf, "%s\n", state); |
de7b77e5 JP |
2964 | } |
2965 | ||
1782dc87 Y |
2966 | static ssize_t control_store(struct device *dev, struct device_attribute *attr, |
2967 | const char *buf, size_t count) | |
de7b77e5 JP |
2968 | { |
2969 | return __store_smt_control(dev, attr, buf, count); | |
2970 | } | |
1782dc87 | 2971 | static DEVICE_ATTR_RW(control); |
05736e4a | 2972 | |
1782dc87 Y |
2973 | static ssize_t active_show(struct device *dev, |
2974 | struct device_attribute *attr, char *buf) | |
05736e4a | 2975 | { |
effe6d27 | 2976 | return sysfs_emit(buf, "%d\n", sched_smt_active()); |
05736e4a | 2977 | } |
1782dc87 | 2978 | static DEVICE_ATTR_RO(active); |
05736e4a TG |
2979 | |
2980 | static struct attribute *cpuhp_smt_attrs[] = { | |
2981 | &dev_attr_control.attr, | |
2982 | &dev_attr_active.attr, | |
2983 | NULL | |
2984 | }; | |
2985 | ||
2986 | static const struct attribute_group cpuhp_smt_attr_group = { | |
2987 | .attrs = cpuhp_smt_attrs, | |
2988 | .name = "smt", | |
2989 | NULL | |
2990 | }; | |
2991 | ||
de7b77e5 | 2992 | static int __init cpu_smt_sysfs_init(void) |
05736e4a | 2993 | { |
db281d59 GKH |
2994 | struct device *dev_root; |
2995 | int ret = -ENODEV; | |
2996 | ||
2997 | dev_root = bus_get_dev_root(&cpu_subsys); | |
2998 | if (dev_root) { | |
2999 | ret = sysfs_create_group(&dev_root->kobj, &cpuhp_smt_attr_group); | |
3000 | put_device(dev_root); | |
3001 | } | |
3002 | return ret; | |
05736e4a TG |
3003 | } |
3004 | ||
98f8cdce TG |
3005 | static int __init cpuhp_sysfs_init(void) |
3006 | { | |
db281d59 | 3007 | struct device *dev_root; |
98f8cdce TG |
3008 | int cpu, ret; |
3009 | ||
de7b77e5 | 3010 | ret = cpu_smt_sysfs_init(); |
05736e4a TG |
3011 | if (ret) |
3012 | return ret; | |
3013 | ||
db281d59 GKH |
3014 | dev_root = bus_get_dev_root(&cpu_subsys); |
3015 | if (dev_root) { | |
3016 | ret = sysfs_create_group(&dev_root->kobj, &cpuhp_cpu_root_attr_group); | |
3017 | put_device(dev_root); | |
3018 | if (ret) | |
3019 | return ret; | |
3020 | } | |
98f8cdce TG |
3021 | |
3022 | for_each_possible_cpu(cpu) { | |
3023 | struct device *dev = get_cpu_device(cpu); | |
3024 | ||
3025 | if (!dev) | |
3026 | continue; | |
3027 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
3028 | if (ret) | |
3029 | return ret; | |
3030 | } | |
3031 | return 0; | |
3032 | } | |
3033 | device_initcall(cpuhp_sysfs_init); | |
de7b77e5 | 3034 | #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */ |
98f8cdce | 3035 | |
e56b3bc7 LT |
3036 | /* |
3037 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
3038 | * represents all NR_CPUS bits binary values of 1<<nr. | |
3039 | * | |
e0b582ec | 3040 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
3041 | * mask value that has a single bit set only. |
3042 | */ | |
b8d317d1 | 3043 | |
e56b3bc7 | 3044 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 3045 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
3046 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
3047 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
3048 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 3049 | |
e56b3bc7 LT |
3050 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
3051 | ||
3052 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
3053 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
3054 | #if BITS_PER_LONG > 32 | |
3055 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
3056 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
3057 | #endif |
3058 | }; | |
e56b3bc7 | 3059 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
3060 | |
3061 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
3062 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
3063 | |
3064 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
266e9578 | 3065 | struct cpumask __cpu_possible_mask __ro_after_init |
c4c54dd1 | 3066 | = {CPU_BITS_ALL}; |
b3199c02 | 3067 | #else |
da92df49 | 3068 | struct cpumask __cpu_possible_mask __ro_after_init; |
b3199c02 | 3069 | #endif |
4b804c85 | 3070 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 3071 | |
4b804c85 RV |
3072 | struct cpumask __cpu_online_mask __read_mostly; |
3073 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 3074 | |
4b804c85 RV |
3075 | struct cpumask __cpu_present_mask __read_mostly; |
3076 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 3077 | |
4b804c85 RV |
3078 | struct cpumask __cpu_active_mask __read_mostly; |
3079 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 3080 | |
e40f74c5 PZ |
3081 | struct cpumask __cpu_dying_mask __read_mostly; |
3082 | EXPORT_SYMBOL(__cpu_dying_mask); | |
3083 | ||
0c09ab96 TG |
3084 | atomic_t __num_online_cpus __read_mostly; |
3085 | EXPORT_SYMBOL(__num_online_cpus); | |
3086 | ||
3fa41520 RR |
3087 | void init_cpu_present(const struct cpumask *src) |
3088 | { | |
c4c54dd1 | 3089 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
3090 | } |
3091 | ||
3092 | void init_cpu_possible(const struct cpumask *src) | |
3093 | { | |
c4c54dd1 | 3094 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
3095 | } |
3096 | ||
3097 | void init_cpu_online(const struct cpumask *src) | |
3098 | { | |
c4c54dd1 | 3099 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 3100 | } |
cff7d378 | 3101 | |
0c09ab96 TG |
3102 | void set_cpu_online(unsigned int cpu, bool online) |
3103 | { | |
3104 | /* | |
3105 | * atomic_inc/dec() is required to handle the horrid abuse of this | |
3106 | * function by the reboot and kexec code which invoke it from | |
3107 | * IPI/NMI broadcasts when shutting down CPUs. Invocation from | |
3108 | * regular CPU hotplug is properly serialized. | |
3109 | * | |
3110 | * Note, that the fact that __num_online_cpus is of type atomic_t | |
3111 | * does not protect readers which are not serialized against | |
3112 | * concurrent hotplug operations. | |
3113 | */ | |
3114 | if (online) { | |
3115 | if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask)) | |
3116 | atomic_inc(&__num_online_cpus); | |
3117 | } else { | |
3118 | if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask)) | |
3119 | atomic_dec(&__num_online_cpus); | |
3120 | } | |
3121 | } | |
3122 | ||
cff7d378 TG |
3123 | /* |
3124 | * Activate the first processor. | |
3125 | */ | |
3126 | void __init boot_cpu_init(void) | |
3127 | { | |
3128 | int cpu = smp_processor_id(); | |
3129 | ||
3130 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
3131 | set_cpu_online(cpu, true); | |
3132 | set_cpu_active(cpu, true); | |
3133 | set_cpu_present(cpu, true); | |
3134 | set_cpu_possible(cpu, true); | |
8ce371f9 PZ |
3135 | |
3136 | #ifdef CONFIG_SMP | |
3137 | __boot_cpu_id = cpu; | |
3138 | #endif | |
cff7d378 TG |
3139 | } |
3140 | ||
3141 | /* | |
3142 | * Must be called _AFTER_ setting up the per_cpu areas | |
3143 | */ | |
b5b1404d | 3144 | void __init boot_cpu_hotplug_init(void) |
cff7d378 | 3145 | { |
269777aa | 3146 | #ifdef CONFIG_SMP |
e797bda3 | 3147 | cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask); |
6f062123 | 3148 | atomic_set(this_cpu_ptr(&cpuhp_state.ap_sync_state), SYNC_STATE_ONLINE); |
269777aa | 3149 | #endif |
0cc3cd21 | 3150 | this_cpu_write(cpuhp_state.state, CPUHP_ONLINE); |
d385febc | 3151 | this_cpu_write(cpuhp_state.target, CPUHP_ONLINE); |
cff7d378 | 3152 | } |
98af8452 | 3153 | |
ce0abef6 | 3154 | #ifdef CONFIG_CPU_MITIGATIONS |
731dc9df TH |
3155 | /* |
3156 | * These are used for a global "mitigations=" cmdline option for toggling | |
3157 | * optional CPU mitigations. | |
3158 | */ | |
3159 | enum cpu_mitigations { | |
3160 | CPU_MITIGATIONS_OFF, | |
3161 | CPU_MITIGATIONS_AUTO, | |
3162 | CPU_MITIGATIONS_AUTO_NOSMT, | |
3163 | }; | |
3164 | ||
ce0abef6 | 3165 | static enum cpu_mitigations cpu_mitigations __ro_after_init = CPU_MITIGATIONS_AUTO; |
98af8452 JP |
3166 | |
3167 | static int __init mitigations_parse_cmdline(char *arg) | |
3168 | { | |
3169 | if (!strcmp(arg, "off")) | |
3170 | cpu_mitigations = CPU_MITIGATIONS_OFF; | |
3171 | else if (!strcmp(arg, "auto")) | |
3172 | cpu_mitigations = CPU_MITIGATIONS_AUTO; | |
3173 | else if (!strcmp(arg, "auto,nosmt")) | |
3174 | cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT; | |
1bf72720 GU |
3175 | else |
3176 | pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n", | |
3177 | arg); | |
98af8452 JP |
3178 | |
3179 | return 0; | |
3180 | } | |
731dc9df TH |
3181 | |
3182 | /* mitigations=off */ | |
3183 | bool cpu_mitigations_off(void) | |
3184 | { | |
3185 | return cpu_mitigations == CPU_MITIGATIONS_OFF; | |
3186 | } | |
3187 | EXPORT_SYMBOL_GPL(cpu_mitigations_off); | |
3188 | ||
3189 | /* mitigations=auto,nosmt */ | |
3190 | bool cpu_mitigations_auto_nosmt(void) | |
3191 | { | |
3192 | return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT; | |
3193 | } | |
3194 | EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt); | |
ce0abef6 SC |
3195 | #else |
3196 | static int __init mitigations_parse_cmdline(char *arg) | |
3197 | { | |
3198 | pr_crit("Kernel compiled without mitigations, ignoring 'mitigations'; system may still be vulnerable\n"); | |
3199 | return 0; | |
3200 | } | |
3201 | #endif | |
3202 | early_param("mitigations", mitigations_parse_cmdline); |