Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* CPU control. |
2 | * (C) 2001, 2002, 2003, 2004 Rusty Russell | |
3 | * | |
4 | * This code is licenced under the GPL. | |
5 | */ | |
6 | #include <linux/proc_fs.h> | |
7 | #include <linux/smp.h> | |
8 | #include <linux/init.h> | |
9 | #include <linux/notifier.h> | |
10 | #include <linux/sched.h> | |
11 | #include <linux/unistd.h> | |
12 | #include <linux/cpu.h> | |
cb79295e AV |
13 | #include <linux/oom.h> |
14 | #include <linux/rcupdate.h> | |
9984de1a | 15 | #include <linux/export.h> |
e4cc2f87 | 16 | #include <linux/bug.h> |
1da177e4 LT |
17 | #include <linux/kthread.h> |
18 | #include <linux/stop_machine.h> | |
81615b62 | 19 | #include <linux/mutex.h> |
5a0e3ad6 | 20 | #include <linux/gfp.h> |
79cfbdfa | 21 | #include <linux/suspend.h> |
a19423b9 | 22 | #include <linux/lockdep.h> |
345527b1 | 23 | #include <linux/tick.h> |
a8994181 | 24 | #include <linux/irq.h> |
cff7d378 | 25 | |
bb3632c6 | 26 | #include <trace/events/power.h> |
cff7d378 TG |
27 | #define CREATE_TRACE_POINTS |
28 | #include <trace/events/cpuhp.h> | |
1da177e4 | 29 | |
38498a67 TG |
30 | #include "smpboot.h" |
31 | ||
cff7d378 TG |
32 | /** |
33 | * cpuhp_cpu_state - Per cpu hotplug state storage | |
34 | * @state: The current cpu state | |
35 | * @target: The target state | |
36 | */ | |
37 | struct cpuhp_cpu_state { | |
38 | enum cpuhp_state state; | |
39 | enum cpuhp_state target; | |
40 | }; | |
41 | ||
42 | static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state); | |
43 | ||
44 | /** | |
45 | * cpuhp_step - Hotplug state machine step | |
46 | * @name: Name of the step | |
47 | * @startup: Startup function of the step | |
48 | * @teardown: Teardown function of the step | |
49 | * @skip_onerr: Do not invoke the functions on error rollback | |
50 | * Will go away once the notifiers are gone | |
757c989b | 51 | * @cant_stop: Bringup/teardown can't be stopped at this step |
cff7d378 TG |
52 | */ |
53 | struct cpuhp_step { | |
54 | const char *name; | |
55 | int (*startup)(unsigned int cpu); | |
56 | int (*teardown)(unsigned int cpu); | |
57 | bool skip_onerr; | |
757c989b | 58 | bool cant_stop; |
cff7d378 TG |
59 | }; |
60 | ||
98f8cdce | 61 | static DEFINE_MUTEX(cpuhp_state_mutex); |
cff7d378 | 62 | static struct cpuhp_step cpuhp_bp_states[]; |
4baa0afc | 63 | static struct cpuhp_step cpuhp_ap_states[]; |
cff7d378 TG |
64 | |
65 | /** | |
66 | * cpuhp_invoke_callback _ Invoke the callbacks for a given state | |
67 | * @cpu: The cpu for which the callback should be invoked | |
68 | * @step: The step in the state machine | |
69 | * @cb: The callback function to invoke | |
70 | * | |
71 | * Called from cpu hotplug and from the state register machinery | |
72 | */ | |
73 | static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step, | |
74 | int (*cb)(unsigned int)) | |
75 | { | |
76 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
77 | int ret = 0; | |
78 | ||
79 | if (cb) { | |
80 | trace_cpuhp_enter(cpu, st->target, step, cb); | |
81 | ret = cb(cpu); | |
82 | trace_cpuhp_exit(cpu, st->state, step, ret); | |
83 | } | |
84 | return ret; | |
85 | } | |
86 | ||
98a79d6a | 87 | #ifdef CONFIG_SMP |
b3199c02 | 88 | /* Serializes the updates to cpu_online_mask, cpu_present_mask */ |
aa953877 | 89 | static DEFINE_MUTEX(cpu_add_remove_lock); |
090e77c3 TG |
90 | bool cpuhp_tasks_frozen; |
91 | EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen); | |
1da177e4 | 92 | |
79a6cdeb | 93 | /* |
93ae4f97 SB |
94 | * The following two APIs (cpu_maps_update_begin/done) must be used when |
95 | * attempting to serialize the updates to cpu_online_mask & cpu_present_mask. | |
96 | * The APIs cpu_notifier_register_begin/done() must be used to protect CPU | |
97 | * hotplug callback (un)registration performed using __register_cpu_notifier() | |
98 | * or __unregister_cpu_notifier(). | |
79a6cdeb LJ |
99 | */ |
100 | void cpu_maps_update_begin(void) | |
101 | { | |
102 | mutex_lock(&cpu_add_remove_lock); | |
103 | } | |
93ae4f97 | 104 | EXPORT_SYMBOL(cpu_notifier_register_begin); |
79a6cdeb LJ |
105 | |
106 | void cpu_maps_update_done(void) | |
107 | { | |
108 | mutex_unlock(&cpu_add_remove_lock); | |
109 | } | |
93ae4f97 | 110 | EXPORT_SYMBOL(cpu_notifier_register_done); |
79a6cdeb | 111 | |
5c113fbe | 112 | static RAW_NOTIFIER_HEAD(cpu_chain); |
1da177e4 | 113 | |
e3920fb4 RW |
114 | /* If set, cpu_up and cpu_down will return -EBUSY and do nothing. |
115 | * Should always be manipulated under cpu_add_remove_lock | |
116 | */ | |
117 | static int cpu_hotplug_disabled; | |
118 | ||
79a6cdeb LJ |
119 | #ifdef CONFIG_HOTPLUG_CPU |
120 | ||
d221938c GS |
121 | static struct { |
122 | struct task_struct *active_writer; | |
87af9e7f DH |
123 | /* wait queue to wake up the active_writer */ |
124 | wait_queue_head_t wq; | |
125 | /* verifies that no writer will get active while readers are active */ | |
126 | struct mutex lock; | |
d221938c GS |
127 | /* |
128 | * Also blocks the new readers during | |
129 | * an ongoing cpu hotplug operation. | |
130 | */ | |
87af9e7f | 131 | atomic_t refcount; |
a19423b9 GS |
132 | |
133 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
134 | struct lockdep_map dep_map; | |
135 | #endif | |
31950eb6 LT |
136 | } cpu_hotplug = { |
137 | .active_writer = NULL, | |
87af9e7f | 138 | .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq), |
31950eb6 | 139 | .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock), |
a19423b9 GS |
140 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
141 | .dep_map = {.name = "cpu_hotplug.lock" }, | |
142 | #endif | |
31950eb6 | 143 | }; |
d221938c | 144 | |
a19423b9 GS |
145 | /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */ |
146 | #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map) | |
dd56af42 PM |
147 | #define cpuhp_lock_acquire_tryread() \ |
148 | lock_map_acquire_tryread(&cpu_hotplug.dep_map) | |
a19423b9 GS |
149 | #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map) |
150 | #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map) | |
151 | ||
62db99f4 | 152 | |
86ef5c9a | 153 | void get_online_cpus(void) |
a9d9baa1 | 154 | { |
d221938c GS |
155 | might_sleep(); |
156 | if (cpu_hotplug.active_writer == current) | |
aa953877 | 157 | return; |
a19423b9 | 158 | cpuhp_lock_acquire_read(); |
d221938c | 159 | mutex_lock(&cpu_hotplug.lock); |
87af9e7f | 160 | atomic_inc(&cpu_hotplug.refcount); |
d221938c | 161 | mutex_unlock(&cpu_hotplug.lock); |
a9d9baa1 | 162 | } |
86ef5c9a | 163 | EXPORT_SYMBOL_GPL(get_online_cpus); |
90d45d17 | 164 | |
86ef5c9a | 165 | void put_online_cpus(void) |
a9d9baa1 | 166 | { |
87af9e7f DH |
167 | int refcount; |
168 | ||
d221938c | 169 | if (cpu_hotplug.active_writer == current) |
aa953877 | 170 | return; |
075663d1 | 171 | |
87af9e7f DH |
172 | refcount = atomic_dec_return(&cpu_hotplug.refcount); |
173 | if (WARN_ON(refcount < 0)) /* try to fix things up */ | |
174 | atomic_inc(&cpu_hotplug.refcount); | |
175 | ||
176 | if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq)) | |
177 | wake_up(&cpu_hotplug.wq); | |
075663d1 | 178 | |
a19423b9 | 179 | cpuhp_lock_release(); |
d221938c | 180 | |
a9d9baa1 | 181 | } |
86ef5c9a | 182 | EXPORT_SYMBOL_GPL(put_online_cpus); |
a9d9baa1 | 183 | |
d221938c GS |
184 | /* |
185 | * This ensures that the hotplug operation can begin only when the | |
186 | * refcount goes to zero. | |
187 | * | |
188 | * Note that during a cpu-hotplug operation, the new readers, if any, | |
189 | * will be blocked by the cpu_hotplug.lock | |
190 | * | |
d2ba7e2a ON |
191 | * Since cpu_hotplug_begin() is always called after invoking |
192 | * cpu_maps_update_begin(), we can be sure that only one writer is active. | |
d221938c GS |
193 | * |
194 | * Note that theoretically, there is a possibility of a livelock: | |
195 | * - Refcount goes to zero, last reader wakes up the sleeping | |
196 | * writer. | |
197 | * - Last reader unlocks the cpu_hotplug.lock. | |
198 | * - A new reader arrives at this moment, bumps up the refcount. | |
199 | * - The writer acquires the cpu_hotplug.lock finds the refcount | |
200 | * non zero and goes to sleep again. | |
201 | * | |
202 | * However, this is very difficult to achieve in practice since | |
86ef5c9a | 203 | * get_online_cpus() not an api which is called all that often. |
d221938c GS |
204 | * |
205 | */ | |
b9d10be7 | 206 | void cpu_hotplug_begin(void) |
d221938c | 207 | { |
87af9e7f | 208 | DEFINE_WAIT(wait); |
d2ba7e2a | 209 | |
87af9e7f | 210 | cpu_hotplug.active_writer = current; |
a19423b9 | 211 | cpuhp_lock_acquire(); |
87af9e7f | 212 | |
d2ba7e2a ON |
213 | for (;;) { |
214 | mutex_lock(&cpu_hotplug.lock); | |
87af9e7f DH |
215 | prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE); |
216 | if (likely(!atomic_read(&cpu_hotplug.refcount))) | |
217 | break; | |
d221938c GS |
218 | mutex_unlock(&cpu_hotplug.lock); |
219 | schedule(); | |
d221938c | 220 | } |
87af9e7f | 221 | finish_wait(&cpu_hotplug.wq, &wait); |
d221938c GS |
222 | } |
223 | ||
b9d10be7 | 224 | void cpu_hotplug_done(void) |
d221938c GS |
225 | { |
226 | cpu_hotplug.active_writer = NULL; | |
227 | mutex_unlock(&cpu_hotplug.lock); | |
a19423b9 | 228 | cpuhp_lock_release(); |
d221938c | 229 | } |
79a6cdeb | 230 | |
16e53dbf SB |
231 | /* |
232 | * Wait for currently running CPU hotplug operations to complete (if any) and | |
233 | * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects | |
234 | * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the | |
235 | * hotplug path before performing hotplug operations. So acquiring that lock | |
236 | * guarantees mutual exclusion from any currently running hotplug operations. | |
237 | */ | |
238 | void cpu_hotplug_disable(void) | |
239 | { | |
240 | cpu_maps_update_begin(); | |
89af7ba5 | 241 | cpu_hotplug_disabled++; |
16e53dbf SB |
242 | cpu_maps_update_done(); |
243 | } | |
32145c46 | 244 | EXPORT_SYMBOL_GPL(cpu_hotplug_disable); |
16e53dbf SB |
245 | |
246 | void cpu_hotplug_enable(void) | |
247 | { | |
248 | cpu_maps_update_begin(); | |
89af7ba5 | 249 | WARN_ON(--cpu_hotplug_disabled < 0); |
16e53dbf SB |
250 | cpu_maps_update_done(); |
251 | } | |
32145c46 | 252 | EXPORT_SYMBOL_GPL(cpu_hotplug_enable); |
b9d10be7 | 253 | #endif /* CONFIG_HOTPLUG_CPU */ |
79a6cdeb | 254 | |
1da177e4 | 255 | /* Need to know about CPUs going up/down? */ |
71cf5aee | 256 | int register_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 257 | { |
bd5349cf | 258 | int ret; |
d221938c | 259 | cpu_maps_update_begin(); |
bd5349cf | 260 | ret = raw_notifier_chain_register(&cpu_chain, nb); |
d221938c | 261 | cpu_maps_update_done(); |
bd5349cf | 262 | return ret; |
1da177e4 | 263 | } |
65edc68c | 264 | |
71cf5aee | 265 | int __register_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
266 | { |
267 | return raw_notifier_chain_register(&cpu_chain, nb); | |
268 | } | |
269 | ||
090e77c3 | 270 | static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call, |
e9fb7631 AM |
271 | int *nr_calls) |
272 | { | |
090e77c3 TG |
273 | unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0; |
274 | void *hcpu = (void *)(long)cpu; | |
275 | ||
e6bde73b AM |
276 | int ret; |
277 | ||
090e77c3 | 278 | ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call, |
e9fb7631 | 279 | nr_calls); |
e6bde73b AM |
280 | |
281 | return notifier_to_errno(ret); | |
e9fb7631 AM |
282 | } |
283 | ||
090e77c3 | 284 | static int cpu_notify(unsigned long val, unsigned int cpu) |
e9fb7631 | 285 | { |
090e77c3 | 286 | return __cpu_notify(val, cpu, -1, NULL); |
e9fb7631 AM |
287 | } |
288 | ||
ba997462 TG |
289 | /* Notifier wrappers for transitioning to state machine */ |
290 | static int notify_prepare(unsigned int cpu) | |
291 | { | |
292 | int nr_calls = 0; | |
293 | int ret; | |
294 | ||
295 | ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls); | |
296 | if (ret) { | |
297 | nr_calls--; | |
298 | printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n", | |
299 | __func__, cpu); | |
300 | __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL); | |
301 | } | |
302 | return ret; | |
303 | } | |
304 | ||
305 | static int notify_online(unsigned int cpu) | |
306 | { | |
307 | cpu_notify(CPU_ONLINE, cpu); | |
308 | return 0; | |
309 | } | |
310 | ||
4baa0afc TG |
311 | static int notify_starting(unsigned int cpu) |
312 | { | |
313 | cpu_notify(CPU_STARTING, cpu); | |
314 | return 0; | |
315 | } | |
316 | ||
ba997462 TG |
317 | static int bringup_cpu(unsigned int cpu) |
318 | { | |
319 | struct task_struct *idle = idle_thread_get(cpu); | |
320 | int ret; | |
321 | ||
322 | /* Arch-specific enabling code. */ | |
323 | ret = __cpu_up(cpu, idle); | |
324 | if (ret) { | |
325 | cpu_notify(CPU_UP_CANCELED, cpu); | |
326 | return ret; | |
327 | } | |
328 | BUG_ON(!cpu_online(cpu)); | |
329 | return 0; | |
330 | } | |
331 | ||
2e1a3483 TG |
332 | /* |
333 | * Hotplug state machine related functions | |
334 | */ | |
335 | static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st, | |
336 | struct cpuhp_step *steps) | |
337 | { | |
338 | for (st->state++; st->state < st->target; st->state++) { | |
339 | struct cpuhp_step *step = steps + st->state; | |
340 | ||
341 | if (!step->skip_onerr) | |
342 | cpuhp_invoke_callback(cpu, st->state, step->startup); | |
343 | } | |
344 | } | |
345 | ||
346 | static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
347 | struct cpuhp_step *steps, enum cpuhp_state target) | |
348 | { | |
349 | enum cpuhp_state prev_state = st->state; | |
350 | int ret = 0; | |
351 | ||
352 | for (; st->state > target; st->state--) { | |
353 | struct cpuhp_step *step = steps + st->state; | |
354 | ||
355 | ret = cpuhp_invoke_callback(cpu, st->state, step->teardown); | |
356 | if (ret) { | |
357 | st->target = prev_state; | |
358 | undo_cpu_down(cpu, st, steps); | |
359 | break; | |
360 | } | |
361 | } | |
362 | return ret; | |
363 | } | |
364 | ||
365 | static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st, | |
366 | struct cpuhp_step *steps) | |
367 | { | |
368 | for (st->state--; st->state > st->target; st->state--) { | |
369 | struct cpuhp_step *step = steps + st->state; | |
370 | ||
371 | if (!step->skip_onerr) | |
372 | cpuhp_invoke_callback(cpu, st->state, step->teardown); | |
373 | } | |
374 | } | |
375 | ||
376 | static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st, | |
377 | struct cpuhp_step *steps, enum cpuhp_state target) | |
378 | { | |
379 | enum cpuhp_state prev_state = st->state; | |
380 | int ret = 0; | |
381 | ||
382 | while (st->state < target) { | |
383 | struct cpuhp_step *step; | |
384 | ||
385 | st->state++; | |
386 | step = steps + st->state; | |
387 | ret = cpuhp_invoke_callback(cpu, st->state, step->startup); | |
388 | if (ret) { | |
389 | st->target = prev_state; | |
390 | undo_cpu_up(cpu, st, steps); | |
391 | break; | |
392 | } | |
393 | } | |
394 | return ret; | |
395 | } | |
396 | ||
00b9b0af | 397 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 398 | EXPORT_SYMBOL(register_cpu_notifier); |
93ae4f97 | 399 | EXPORT_SYMBOL(__register_cpu_notifier); |
71cf5aee | 400 | void unregister_cpu_notifier(struct notifier_block *nb) |
1da177e4 | 401 | { |
d221938c | 402 | cpu_maps_update_begin(); |
bd5349cf | 403 | raw_notifier_chain_unregister(&cpu_chain, nb); |
d221938c | 404 | cpu_maps_update_done(); |
1da177e4 LT |
405 | } |
406 | EXPORT_SYMBOL(unregister_cpu_notifier); | |
407 | ||
71cf5aee | 408 | void __unregister_cpu_notifier(struct notifier_block *nb) |
93ae4f97 SB |
409 | { |
410 | raw_notifier_chain_unregister(&cpu_chain, nb); | |
411 | } | |
412 | EXPORT_SYMBOL(__unregister_cpu_notifier); | |
413 | ||
e4cc2f87 AV |
414 | /** |
415 | * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU | |
416 | * @cpu: a CPU id | |
417 | * | |
418 | * This function walks all processes, finds a valid mm struct for each one and | |
419 | * then clears a corresponding bit in mm's cpumask. While this all sounds | |
420 | * trivial, there are various non-obvious corner cases, which this function | |
421 | * tries to solve in a safe manner. | |
422 | * | |
423 | * Also note that the function uses a somewhat relaxed locking scheme, so it may | |
424 | * be called only for an already offlined CPU. | |
425 | */ | |
cb79295e AV |
426 | void clear_tasks_mm_cpumask(int cpu) |
427 | { | |
428 | struct task_struct *p; | |
429 | ||
430 | /* | |
431 | * This function is called after the cpu is taken down and marked | |
432 | * offline, so its not like new tasks will ever get this cpu set in | |
433 | * their mm mask. -- Peter Zijlstra | |
434 | * Thus, we may use rcu_read_lock() here, instead of grabbing | |
435 | * full-fledged tasklist_lock. | |
436 | */ | |
e4cc2f87 | 437 | WARN_ON(cpu_online(cpu)); |
cb79295e AV |
438 | rcu_read_lock(); |
439 | for_each_process(p) { | |
440 | struct task_struct *t; | |
441 | ||
e4cc2f87 AV |
442 | /* |
443 | * Main thread might exit, but other threads may still have | |
444 | * a valid mm. Find one. | |
445 | */ | |
cb79295e AV |
446 | t = find_lock_task_mm(p); |
447 | if (!t) | |
448 | continue; | |
449 | cpumask_clear_cpu(cpu, mm_cpumask(t->mm)); | |
450 | task_unlock(t); | |
451 | } | |
452 | rcu_read_unlock(); | |
453 | } | |
454 | ||
b728ca06 | 455 | static inline void check_for_tasks(int dead_cpu) |
1da177e4 | 456 | { |
b728ca06 | 457 | struct task_struct *g, *p; |
1da177e4 | 458 | |
a75a6068 ON |
459 | read_lock(&tasklist_lock); |
460 | for_each_process_thread(g, p) { | |
b728ca06 KT |
461 | if (!p->on_rq) |
462 | continue; | |
463 | /* | |
464 | * We do the check with unlocked task_rq(p)->lock. | |
465 | * Order the reading to do not warn about a task, | |
466 | * which was running on this cpu in the past, and | |
467 | * it's just been woken on another cpu. | |
468 | */ | |
469 | rmb(); | |
470 | if (task_cpu(p) != dead_cpu) | |
471 | continue; | |
472 | ||
473 | pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n", | |
474 | p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags); | |
a75a6068 ON |
475 | } |
476 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
477 | } |
478 | ||
98458172 TG |
479 | static void cpu_notify_nofail(unsigned long val, unsigned int cpu) |
480 | { | |
481 | BUG_ON(cpu_notify(val, cpu)); | |
482 | } | |
483 | ||
484 | static int notify_down_prepare(unsigned int cpu) | |
485 | { | |
486 | int err, nr_calls = 0; | |
487 | ||
488 | err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls); | |
489 | if (err) { | |
490 | nr_calls--; | |
491 | __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL); | |
492 | pr_warn("%s: attempt to take down CPU %u failed\n", | |
493 | __func__, cpu); | |
494 | } | |
495 | return err; | |
496 | } | |
497 | ||
4baa0afc TG |
498 | static int notify_dying(unsigned int cpu) |
499 | { | |
500 | cpu_notify(CPU_DYING, cpu); | |
501 | return 0; | |
502 | } | |
503 | ||
1da177e4 | 504 | /* Take this CPU down. */ |
71cf5aee | 505 | static int take_cpu_down(void *_param) |
1da177e4 | 506 | { |
4baa0afc TG |
507 | struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state); |
508 | enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE); | |
090e77c3 | 509 | int err, cpu = smp_processor_id(); |
1da177e4 | 510 | |
1da177e4 LT |
511 | /* Ensure this CPU doesn't handle any more interrupts. */ |
512 | err = __cpu_disable(); | |
513 | if (err < 0) | |
f3705136 | 514 | return err; |
1da177e4 | 515 | |
4baa0afc TG |
516 | /* Invoke the former CPU_DYING callbacks */ |
517 | for (; st->state > target; st->state--) { | |
518 | struct cpuhp_step *step = cpuhp_ap_states + st->state; | |
519 | ||
520 | cpuhp_invoke_callback(cpu, st->state, step->teardown); | |
521 | } | |
52c063d1 TG |
522 | /* Give up timekeeping duties */ |
523 | tick_handover_do_timer(); | |
14e568e7 | 524 | /* Park the stopper thread */ |
090e77c3 | 525 | stop_machine_park(cpu); |
f3705136 | 526 | return 0; |
1da177e4 LT |
527 | } |
528 | ||
98458172 | 529 | static int takedown_cpu(unsigned int cpu) |
1da177e4 | 530 | { |
98458172 | 531 | int err; |
1da177e4 | 532 | |
6acce3ef PZ |
533 | /* |
534 | * By now we've cleared cpu_active_mask, wait for all preempt-disabled | |
535 | * and RCU users of this state to go away such that all new such users | |
536 | * will observe it. | |
537 | * | |
538 | * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might | |
779de6ce | 539 | * not imply sync_sched(), so wait for both. |
106dd5af M |
540 | * |
541 | * Do sync before park smpboot threads to take care the rcu boost case. | |
6acce3ef | 542 | */ |
779de6ce PM |
543 | if (IS_ENABLED(CONFIG_PREEMPT)) |
544 | synchronize_rcu_mult(call_rcu, call_rcu_sched); | |
545 | else | |
546 | synchronize_rcu(); | |
6acce3ef PZ |
547 | |
548 | /* | |
a8994181 TG |
549 | * Prevent irq alloc/free while the dying cpu reorganizes the |
550 | * interrupt affinities. | |
6acce3ef | 551 | */ |
a8994181 | 552 | irq_lock_sparse(); |
6acce3ef | 553 | |
a8994181 TG |
554 | /* |
555 | * So now all preempt/rcu users must observe !cpu_active(). | |
556 | */ | |
090e77c3 | 557 | err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu)); |
04321587 | 558 | if (err) { |
1da177e4 | 559 | /* CPU didn't die: tell everyone. Can't complain. */ |
090e77c3 | 560 | cpu_notify_nofail(CPU_DOWN_FAILED, cpu); |
a8994181 | 561 | irq_unlock_sparse(); |
98458172 | 562 | return err; |
8fa1d7d3 | 563 | } |
04321587 | 564 | BUG_ON(cpu_online(cpu)); |
1da177e4 | 565 | |
48c5ccae PZ |
566 | /* |
567 | * The migration_call() CPU_DYING callback will have removed all | |
568 | * runnable tasks from the cpu, there's only the idle task left now | |
569 | * that the migration thread is done doing the stop_machine thing. | |
51a96c77 PZ |
570 | * |
571 | * Wait for the stop thread to go away. | |
48c5ccae | 572 | */ |
528a25b0 | 573 | while (!per_cpu(cpu_dead_idle, cpu)) |
51a96c77 | 574 | cpu_relax(); |
528a25b0 PM |
575 | smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */ |
576 | per_cpu(cpu_dead_idle, cpu) = false; | |
1da177e4 | 577 | |
a8994181 TG |
578 | /* Interrupts are moved away from the dying cpu, reenable alloc/free */ |
579 | irq_unlock_sparse(); | |
580 | ||
345527b1 | 581 | hotplug_cpu__broadcast_tick_pull(cpu); |
1da177e4 LT |
582 | /* This actually kills the CPU. */ |
583 | __cpu_die(cpu); | |
584 | ||
a49b116d | 585 | tick_cleanup_dead_cpu(cpu); |
98458172 TG |
586 | return 0; |
587 | } | |
1da177e4 | 588 | |
98458172 TG |
589 | static int notify_dead(unsigned int cpu) |
590 | { | |
591 | cpu_notify_nofail(CPU_DEAD, cpu); | |
1da177e4 | 592 | check_for_tasks(cpu); |
98458172 TG |
593 | return 0; |
594 | } | |
595 | ||
cff7d378 TG |
596 | #else |
597 | #define notify_down_prepare NULL | |
598 | #define takedown_cpu NULL | |
599 | #define notify_dead NULL | |
4baa0afc | 600 | #define notify_dying NULL |
cff7d378 TG |
601 | #endif |
602 | ||
603 | #ifdef CONFIG_HOTPLUG_CPU | |
cff7d378 | 604 | |
98458172 | 605 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 TG |
606 | static int __ref _cpu_down(unsigned int cpu, int tasks_frozen, |
607 | enum cpuhp_state target) | |
98458172 | 608 | { |
cff7d378 TG |
609 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
610 | int prev_state, ret = 0; | |
611 | bool hasdied = false; | |
98458172 TG |
612 | |
613 | if (num_online_cpus() == 1) | |
614 | return -EBUSY; | |
615 | ||
757c989b | 616 | if (!cpu_present(cpu)) |
98458172 TG |
617 | return -EINVAL; |
618 | ||
619 | cpu_hotplug_begin(); | |
620 | ||
621 | cpuhp_tasks_frozen = tasks_frozen; | |
622 | ||
cff7d378 | 623 | prev_state = st->state; |
af1f4045 | 624 | st->target = target; |
2e1a3483 | 625 | ret = cpuhp_down_callbacks(cpu, st, cpuhp_bp_states, target); |
98458172 | 626 | |
cff7d378 | 627 | hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE; |
1da177e4 | 628 | |
d221938c | 629 | cpu_hotplug_done(); |
cff7d378 TG |
630 | /* This post dead nonsense must die */ |
631 | if (!ret && hasdied) | |
090e77c3 | 632 | cpu_notify_nofail(CPU_POST_DEAD, cpu); |
cff7d378 | 633 | return ret; |
e3920fb4 RW |
634 | } |
635 | ||
af1f4045 | 636 | static int do_cpu_down(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 | 637 | { |
9ea09af3 | 638 | int err; |
e3920fb4 | 639 | |
d221938c | 640 | cpu_maps_update_begin(); |
e761b772 MK |
641 | |
642 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 643 | err = -EBUSY; |
e761b772 MK |
644 | goto out; |
645 | } | |
646 | ||
af1f4045 | 647 | err = _cpu_down(cpu, 0, target); |
e3920fb4 | 648 | |
e761b772 | 649 | out: |
d221938c | 650 | cpu_maps_update_done(); |
1da177e4 LT |
651 | return err; |
652 | } | |
af1f4045 TG |
653 | int cpu_down(unsigned int cpu) |
654 | { | |
655 | return do_cpu_down(cpu, CPUHP_OFFLINE); | |
656 | } | |
b62b8ef9 | 657 | EXPORT_SYMBOL(cpu_down); |
1da177e4 LT |
658 | #endif /*CONFIG_HOTPLUG_CPU*/ |
659 | ||
4baa0afc TG |
660 | /** |
661 | * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers | |
662 | * @cpu: cpu that just started | |
663 | * | |
664 | * This function calls the cpu_chain notifiers with CPU_STARTING. | |
665 | * It must be called by the arch code on the new cpu, before the new cpu | |
666 | * enables interrupts and before the "boot" cpu returns from __cpu_up(). | |
667 | */ | |
668 | void notify_cpu_starting(unsigned int cpu) | |
669 | { | |
670 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
671 | enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE); | |
672 | ||
673 | while (st->state < target) { | |
674 | struct cpuhp_step *step; | |
675 | ||
676 | st->state++; | |
677 | step = cpuhp_ap_states + st->state; | |
678 | cpuhp_invoke_callback(cpu, st->state, step->startup); | |
679 | } | |
680 | } | |
681 | ||
949338e3 TG |
682 | /* |
683 | * Called from the idle task. We need to set active here, so we can kick off | |
684 | * the stopper thread. | |
685 | */ | |
686 | static int cpuhp_set_cpu_active(unsigned int cpu) | |
687 | { | |
688 | /* The cpu is marked online, set it active now */ | |
689 | set_cpu_active(cpu, true); | |
690 | /* Unpark the stopper thread */ | |
691 | stop_machine_unpark(cpu); | |
692 | return 0; | |
693 | } | |
694 | ||
e3920fb4 | 695 | /* Requires cpu_add_remove_lock to be held */ |
af1f4045 | 696 | static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target) |
1da177e4 | 697 | { |
cff7d378 | 698 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); |
3bb5d2ee | 699 | struct task_struct *idle; |
2e1a3483 | 700 | int ret = 0; |
1da177e4 | 701 | |
d221938c | 702 | cpu_hotplug_begin(); |
38498a67 | 703 | |
757c989b | 704 | if (!cpu_present(cpu)) { |
5e5041f3 YI |
705 | ret = -EINVAL; |
706 | goto out; | |
707 | } | |
708 | ||
757c989b TG |
709 | /* |
710 | * The caller of do_cpu_up might have raced with another | |
711 | * caller. Ignore it for now. | |
712 | */ | |
713 | if (st->state >= target) | |
38498a67 | 714 | goto out; |
757c989b TG |
715 | |
716 | if (st->state == CPUHP_OFFLINE) { | |
717 | /* Let it fail before we try to bring the cpu up */ | |
718 | idle = idle_thread_get(cpu); | |
719 | if (IS_ERR(idle)) { | |
720 | ret = PTR_ERR(idle); | |
721 | goto out; | |
722 | } | |
3bb5d2ee | 723 | } |
38498a67 | 724 | |
ba997462 TG |
725 | cpuhp_tasks_frozen = tasks_frozen; |
726 | ||
af1f4045 | 727 | st->target = target; |
2e1a3483 | 728 | ret = cpuhp_up_callbacks(cpu, st, cpuhp_bp_states, target); |
38498a67 | 729 | out: |
d221938c | 730 | cpu_hotplug_done(); |
e3920fb4 RW |
731 | return ret; |
732 | } | |
733 | ||
af1f4045 | 734 | static int do_cpu_up(unsigned int cpu, enum cpuhp_state target) |
e3920fb4 RW |
735 | { |
736 | int err = 0; | |
cf23422b | 737 | |
e0b582ec | 738 | if (!cpu_possible(cpu)) { |
84117da5 FF |
739 | pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n", |
740 | cpu); | |
87d5e023 | 741 | #if defined(CONFIG_IA64) |
84117da5 | 742 | pr_err("please check additional_cpus= boot parameter\n"); |
73e753a5 KH |
743 | #endif |
744 | return -EINVAL; | |
745 | } | |
e3920fb4 | 746 | |
01b0f197 TK |
747 | err = try_online_node(cpu_to_node(cpu)); |
748 | if (err) | |
749 | return err; | |
cf23422b | 750 | |
d221938c | 751 | cpu_maps_update_begin(); |
e761b772 MK |
752 | |
753 | if (cpu_hotplug_disabled) { | |
e3920fb4 | 754 | err = -EBUSY; |
e761b772 MK |
755 | goto out; |
756 | } | |
757 | ||
af1f4045 | 758 | err = _cpu_up(cpu, 0, target); |
e761b772 | 759 | out: |
d221938c | 760 | cpu_maps_update_done(); |
e3920fb4 RW |
761 | return err; |
762 | } | |
af1f4045 TG |
763 | |
764 | int cpu_up(unsigned int cpu) | |
765 | { | |
766 | return do_cpu_up(cpu, CPUHP_ONLINE); | |
767 | } | |
a513f6ba | 768 | EXPORT_SYMBOL_GPL(cpu_up); |
e3920fb4 | 769 | |
f3de4be9 | 770 | #ifdef CONFIG_PM_SLEEP_SMP |
e0b582ec | 771 | static cpumask_var_t frozen_cpus; |
e3920fb4 RW |
772 | |
773 | int disable_nonboot_cpus(void) | |
774 | { | |
e9a5f426 | 775 | int cpu, first_cpu, error = 0; |
e3920fb4 | 776 | |
d221938c | 777 | cpu_maps_update_begin(); |
e0b582ec | 778 | first_cpu = cpumask_first(cpu_online_mask); |
9ee349ad XF |
779 | /* |
780 | * We take down all of the non-boot CPUs in one shot to avoid races | |
e3920fb4 RW |
781 | * with the userspace trying to use the CPU hotplug at the same time |
782 | */ | |
e0b582ec | 783 | cpumask_clear(frozen_cpus); |
6ad4c188 | 784 | |
84117da5 | 785 | pr_info("Disabling non-boot CPUs ...\n"); |
e3920fb4 RW |
786 | for_each_online_cpu(cpu) { |
787 | if (cpu == first_cpu) | |
788 | continue; | |
bb3632c6 | 789 | trace_suspend_resume(TPS("CPU_OFF"), cpu, true); |
af1f4045 | 790 | error = _cpu_down(cpu, 1, CPUHP_OFFLINE); |
bb3632c6 | 791 | trace_suspend_resume(TPS("CPU_OFF"), cpu, false); |
feae3203 | 792 | if (!error) |
e0b582ec | 793 | cpumask_set_cpu(cpu, frozen_cpus); |
feae3203 | 794 | else { |
84117da5 | 795 | pr_err("Error taking CPU%d down: %d\n", cpu, error); |
e3920fb4 RW |
796 | break; |
797 | } | |
798 | } | |
86886e55 | 799 | |
89af7ba5 | 800 | if (!error) |
e3920fb4 | 801 | BUG_ON(num_online_cpus() > 1); |
89af7ba5 | 802 | else |
84117da5 | 803 | pr_err("Non-boot CPUs are not disabled\n"); |
89af7ba5 VK |
804 | |
805 | /* | |
806 | * Make sure the CPUs won't be enabled by someone else. We need to do | |
807 | * this even in case of failure as all disable_nonboot_cpus() users are | |
808 | * supposed to do enable_nonboot_cpus() on the failure path. | |
809 | */ | |
810 | cpu_hotplug_disabled++; | |
811 | ||
d221938c | 812 | cpu_maps_update_done(); |
e3920fb4 RW |
813 | return error; |
814 | } | |
815 | ||
d0af9eed SS |
816 | void __weak arch_enable_nonboot_cpus_begin(void) |
817 | { | |
818 | } | |
819 | ||
820 | void __weak arch_enable_nonboot_cpus_end(void) | |
821 | { | |
822 | } | |
823 | ||
71cf5aee | 824 | void enable_nonboot_cpus(void) |
e3920fb4 RW |
825 | { |
826 | int cpu, error; | |
827 | ||
828 | /* Allow everyone to use the CPU hotplug again */ | |
d221938c | 829 | cpu_maps_update_begin(); |
89af7ba5 | 830 | WARN_ON(--cpu_hotplug_disabled < 0); |
e0b582ec | 831 | if (cpumask_empty(frozen_cpus)) |
1d64b9cb | 832 | goto out; |
e3920fb4 | 833 | |
84117da5 | 834 | pr_info("Enabling non-boot CPUs ...\n"); |
d0af9eed SS |
835 | |
836 | arch_enable_nonboot_cpus_begin(); | |
837 | ||
e0b582ec | 838 | for_each_cpu(cpu, frozen_cpus) { |
bb3632c6 | 839 | trace_suspend_resume(TPS("CPU_ON"), cpu, true); |
af1f4045 | 840 | error = _cpu_up(cpu, 1, CPUHP_ONLINE); |
bb3632c6 | 841 | trace_suspend_resume(TPS("CPU_ON"), cpu, false); |
e3920fb4 | 842 | if (!error) { |
84117da5 | 843 | pr_info("CPU%d is up\n", cpu); |
e3920fb4 RW |
844 | continue; |
845 | } | |
84117da5 | 846 | pr_warn("Error taking CPU%d up: %d\n", cpu, error); |
e3920fb4 | 847 | } |
d0af9eed SS |
848 | |
849 | arch_enable_nonboot_cpus_end(); | |
850 | ||
e0b582ec | 851 | cpumask_clear(frozen_cpus); |
1d64b9cb | 852 | out: |
d221938c | 853 | cpu_maps_update_done(); |
1da177e4 | 854 | } |
e0b582ec | 855 | |
d7268a31 | 856 | static int __init alloc_frozen_cpus(void) |
e0b582ec RR |
857 | { |
858 | if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO)) | |
859 | return -ENOMEM; | |
860 | return 0; | |
861 | } | |
862 | core_initcall(alloc_frozen_cpus); | |
79cfbdfa | 863 | |
79cfbdfa SB |
864 | /* |
865 | * When callbacks for CPU hotplug notifications are being executed, we must | |
866 | * ensure that the state of the system with respect to the tasks being frozen | |
867 | * or not, as reported by the notification, remains unchanged *throughout the | |
868 | * duration* of the execution of the callbacks. | |
869 | * Hence we need to prevent the freezer from racing with regular CPU hotplug. | |
870 | * | |
871 | * This synchronization is implemented by mutually excluding regular CPU | |
872 | * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/ | |
873 | * Hibernate notifications. | |
874 | */ | |
875 | static int | |
876 | cpu_hotplug_pm_callback(struct notifier_block *nb, | |
877 | unsigned long action, void *ptr) | |
878 | { | |
879 | switch (action) { | |
880 | ||
881 | case PM_SUSPEND_PREPARE: | |
882 | case PM_HIBERNATION_PREPARE: | |
16e53dbf | 883 | cpu_hotplug_disable(); |
79cfbdfa SB |
884 | break; |
885 | ||
886 | case PM_POST_SUSPEND: | |
887 | case PM_POST_HIBERNATION: | |
16e53dbf | 888 | cpu_hotplug_enable(); |
79cfbdfa SB |
889 | break; |
890 | ||
891 | default: | |
892 | return NOTIFY_DONE; | |
893 | } | |
894 | ||
895 | return NOTIFY_OK; | |
896 | } | |
897 | ||
898 | ||
d7268a31 | 899 | static int __init cpu_hotplug_pm_sync_init(void) |
79cfbdfa | 900 | { |
6e32d479 FY |
901 | /* |
902 | * cpu_hotplug_pm_callback has higher priority than x86 | |
903 | * bsp_pm_callback which depends on cpu_hotplug_pm_callback | |
904 | * to disable cpu hotplug to avoid cpu hotplug race. | |
905 | */ | |
79cfbdfa SB |
906 | pm_notifier(cpu_hotplug_pm_callback, 0); |
907 | return 0; | |
908 | } | |
909 | core_initcall(cpu_hotplug_pm_sync_init); | |
910 | ||
f3de4be9 | 911 | #endif /* CONFIG_PM_SLEEP_SMP */ |
68f4f1ec MK |
912 | |
913 | #endif /* CONFIG_SMP */ | |
b8d317d1 | 914 | |
cff7d378 TG |
915 | /* Boot processor state steps */ |
916 | static struct cpuhp_step cpuhp_bp_states[] = { | |
917 | [CPUHP_OFFLINE] = { | |
918 | .name = "offline", | |
919 | .startup = NULL, | |
920 | .teardown = NULL, | |
921 | }, | |
922 | #ifdef CONFIG_SMP | |
923 | [CPUHP_CREATE_THREADS]= { | |
924 | .name = "threads:create", | |
925 | .startup = smpboot_create_threads, | |
926 | .teardown = NULL, | |
757c989b | 927 | .cant_stop = true, |
cff7d378 TG |
928 | }, |
929 | [CPUHP_NOTIFY_PREPARE] = { | |
930 | .name = "notify:prepare", | |
931 | .startup = notify_prepare, | |
932 | .teardown = notify_dead, | |
933 | .skip_onerr = true, | |
757c989b | 934 | .cant_stop = true, |
cff7d378 TG |
935 | }, |
936 | [CPUHP_BRINGUP_CPU] = { | |
937 | .name = "cpu:bringup", | |
938 | .startup = bringup_cpu, | |
4baa0afc | 939 | .teardown = NULL, |
757c989b | 940 | .cant_stop = true, |
4baa0afc TG |
941 | }, |
942 | [CPUHP_TEARDOWN_CPU] = { | |
943 | .name = "cpu:teardown", | |
944 | .startup = NULL, | |
cff7d378 | 945 | .teardown = takedown_cpu, |
757c989b | 946 | .cant_stop = true, |
cff7d378 | 947 | }, |
949338e3 TG |
948 | [CPUHP_CPU_SET_ACTIVE] = { |
949 | .name = "cpu:active", | |
950 | .startup = cpuhp_set_cpu_active, | |
951 | .teardown = NULL, | |
952 | }, | |
931ef163 TG |
953 | [CPUHP_SMPBOOT_THREADS] = { |
954 | .name = "smpboot:threads", | |
955 | .startup = smpboot_unpark_threads, | |
956 | .teardown = smpboot_park_threads, | |
957 | }, | |
cff7d378 TG |
958 | [CPUHP_NOTIFY_ONLINE] = { |
959 | .name = "notify:online", | |
960 | .startup = notify_online, | |
961 | .teardown = notify_down_prepare, | |
757c989b | 962 | .cant_stop = true, |
cff7d378 TG |
963 | }, |
964 | #endif | |
965 | [CPUHP_ONLINE] = { | |
966 | .name = "online", | |
967 | .startup = NULL, | |
968 | .teardown = NULL, | |
969 | }, | |
970 | }; | |
971 | ||
4baa0afc TG |
972 | /* Application processor state steps */ |
973 | static struct cpuhp_step cpuhp_ap_states[] = { | |
974 | #ifdef CONFIG_SMP | |
975 | [CPUHP_AP_NOTIFY_STARTING] = { | |
976 | .name = "notify:starting", | |
977 | .startup = notify_starting, | |
978 | .teardown = notify_dying, | |
979 | .skip_onerr = true, | |
757c989b | 980 | .cant_stop = true, |
4baa0afc TG |
981 | }, |
982 | #endif | |
983 | [CPUHP_ONLINE] = { | |
984 | .name = "online", | |
985 | .startup = NULL, | |
986 | .teardown = NULL, | |
987 | }, | |
988 | }; | |
989 | ||
5b7aa87e TG |
990 | /* Sanity check for callbacks */ |
991 | static int cpuhp_cb_check(enum cpuhp_state state) | |
992 | { | |
993 | if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE) | |
994 | return -EINVAL; | |
995 | return 0; | |
996 | } | |
997 | ||
98f8cdce TG |
998 | static bool cpuhp_is_ap_state(enum cpuhp_state state) |
999 | { | |
1000 | return (state > CPUHP_AP_OFFLINE && state < CPUHP_AP_ONLINE); | |
1001 | } | |
1002 | ||
1003 | static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state) | |
1004 | { | |
1005 | struct cpuhp_step *sp; | |
1006 | ||
1007 | sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states; | |
1008 | return sp + state; | |
1009 | } | |
1010 | ||
5b7aa87e TG |
1011 | static void cpuhp_store_callbacks(enum cpuhp_state state, |
1012 | const char *name, | |
1013 | int (*startup)(unsigned int cpu), | |
1014 | int (*teardown)(unsigned int cpu)) | |
1015 | { | |
1016 | /* (Un)Install the callbacks for further cpu hotplug operations */ | |
1017 | struct cpuhp_step *sp; | |
1018 | ||
1019 | mutex_lock(&cpuhp_state_mutex); | |
1020 | sp = cpuhp_get_step(state); | |
1021 | sp->startup = startup; | |
1022 | sp->teardown = teardown; | |
1023 | sp->name = name; | |
1024 | mutex_unlock(&cpuhp_state_mutex); | |
1025 | } | |
1026 | ||
1027 | static void *cpuhp_get_teardown_cb(enum cpuhp_state state) | |
1028 | { | |
1029 | return cpuhp_get_step(state)->teardown; | |
1030 | } | |
1031 | ||
1032 | /* Helper function to run callback on the target cpu */ | |
1033 | static void cpuhp_on_cpu_cb(void *__cb) | |
1034 | { | |
1035 | int (*cb)(unsigned int cpu) = __cb; | |
1036 | ||
1037 | BUG_ON(cb(smp_processor_id())); | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Call the startup/teardown function for a step either on the AP or | |
1042 | * on the current CPU. | |
1043 | */ | |
1044 | static int cpuhp_issue_call(int cpu, enum cpuhp_state state, | |
1045 | int (*cb)(unsigned int), bool bringup) | |
1046 | { | |
1047 | int ret; | |
1048 | ||
1049 | if (!cb) | |
1050 | return 0; | |
1051 | ||
1052 | /* | |
1053 | * This invokes the callback directly for now. In a later step we | |
1054 | * convert that to use cpuhp_invoke_callback(). | |
1055 | */ | |
1056 | if (cpuhp_is_ap_state(state)) { | |
1057 | /* | |
1058 | * Note, that a function called on the AP is not | |
1059 | * allowed to fail. | |
1060 | */ | |
1061 | if (cpu_online(cpu)) | |
1062 | smp_call_function_single(cpu, cpuhp_on_cpu_cb, cb, 1); | |
1063 | return 0; | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * The non AP bound callbacks can fail on bringup. On teardown | |
1068 | * e.g. module removal we crash for now. | |
1069 | */ | |
1070 | ret = cb(cpu); | |
1071 | BUG_ON(ret && !bringup); | |
1072 | return ret; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Called from __cpuhp_setup_state on a recoverable failure. | |
1077 | * | |
1078 | * Note: The teardown callbacks for rollback are not allowed to fail! | |
1079 | */ | |
1080 | static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state, | |
1081 | int (*teardown)(unsigned int cpu)) | |
1082 | { | |
1083 | int cpu; | |
1084 | ||
1085 | if (!teardown) | |
1086 | return; | |
1087 | ||
1088 | /* Roll back the already executed steps on the other cpus */ | |
1089 | for_each_present_cpu(cpu) { | |
1090 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1091 | int cpustate = st->state; | |
1092 | ||
1093 | if (cpu >= failedcpu) | |
1094 | break; | |
1095 | ||
1096 | /* Did we invoke the startup call on that cpu ? */ | |
1097 | if (cpustate >= state) | |
1098 | cpuhp_issue_call(cpu, state, teardown, false); | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * Returns a free for dynamic slot assignment of the Online state. The states | |
1104 | * are protected by the cpuhp_slot_states mutex and an empty slot is identified | |
1105 | * by having no name assigned. | |
1106 | */ | |
1107 | static int cpuhp_reserve_state(enum cpuhp_state state) | |
1108 | { | |
1109 | enum cpuhp_state i; | |
1110 | ||
1111 | mutex_lock(&cpuhp_state_mutex); | |
1112 | for (i = CPUHP_ONLINE_DYN; i <= CPUHP_ONLINE_DYN_END; i++) { | |
1113 | if (cpuhp_bp_states[i].name) | |
1114 | continue; | |
1115 | ||
1116 | cpuhp_bp_states[i].name = "Reserved"; | |
1117 | mutex_unlock(&cpuhp_state_mutex); | |
1118 | return i; | |
1119 | } | |
1120 | mutex_unlock(&cpuhp_state_mutex); | |
1121 | WARN(1, "No more dynamic states available for CPU hotplug\n"); | |
1122 | return -ENOSPC; | |
1123 | } | |
1124 | ||
1125 | /** | |
1126 | * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state | |
1127 | * @state: The state to setup | |
1128 | * @invoke: If true, the startup function is invoked for cpus where | |
1129 | * cpu state >= @state | |
1130 | * @startup: startup callback function | |
1131 | * @teardown: teardown callback function | |
1132 | * | |
1133 | * Returns 0 if successful, otherwise a proper error code | |
1134 | */ | |
1135 | int __cpuhp_setup_state(enum cpuhp_state state, | |
1136 | const char *name, bool invoke, | |
1137 | int (*startup)(unsigned int cpu), | |
1138 | int (*teardown)(unsigned int cpu)) | |
1139 | { | |
1140 | int cpu, ret = 0; | |
1141 | int dyn_state = 0; | |
1142 | ||
1143 | if (cpuhp_cb_check(state) || !name) | |
1144 | return -EINVAL; | |
1145 | ||
1146 | get_online_cpus(); | |
1147 | ||
1148 | /* currently assignments for the ONLINE state are possible */ | |
1149 | if (state == CPUHP_ONLINE_DYN) { | |
1150 | dyn_state = 1; | |
1151 | ret = cpuhp_reserve_state(state); | |
1152 | if (ret < 0) | |
1153 | goto out; | |
1154 | state = ret; | |
1155 | } | |
1156 | ||
1157 | cpuhp_store_callbacks(state, name, startup, teardown); | |
1158 | ||
1159 | if (!invoke || !startup) | |
1160 | goto out; | |
1161 | ||
1162 | /* | |
1163 | * Try to call the startup callback for each present cpu | |
1164 | * depending on the hotplug state of the cpu. | |
1165 | */ | |
1166 | for_each_present_cpu(cpu) { | |
1167 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1168 | int cpustate = st->state; | |
1169 | ||
1170 | if (cpustate < state) | |
1171 | continue; | |
1172 | ||
1173 | ret = cpuhp_issue_call(cpu, state, startup, true); | |
1174 | if (ret) { | |
1175 | cpuhp_rollback_install(cpu, state, teardown); | |
1176 | cpuhp_store_callbacks(state, NULL, NULL, NULL); | |
1177 | goto out; | |
1178 | } | |
1179 | } | |
1180 | out: | |
1181 | put_online_cpus(); | |
1182 | if (!ret && dyn_state) | |
1183 | return state; | |
1184 | return ret; | |
1185 | } | |
1186 | EXPORT_SYMBOL(__cpuhp_setup_state); | |
1187 | ||
1188 | /** | |
1189 | * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state | |
1190 | * @state: The state to remove | |
1191 | * @invoke: If true, the teardown function is invoked for cpus where | |
1192 | * cpu state >= @state | |
1193 | * | |
1194 | * The teardown callback is currently not allowed to fail. Think | |
1195 | * about module removal! | |
1196 | */ | |
1197 | void __cpuhp_remove_state(enum cpuhp_state state, bool invoke) | |
1198 | { | |
1199 | int (*teardown)(unsigned int cpu) = cpuhp_get_teardown_cb(state); | |
1200 | int cpu; | |
1201 | ||
1202 | BUG_ON(cpuhp_cb_check(state)); | |
1203 | ||
1204 | get_online_cpus(); | |
1205 | ||
1206 | if (!invoke || !teardown) | |
1207 | goto remove; | |
1208 | ||
1209 | /* | |
1210 | * Call the teardown callback for each present cpu depending | |
1211 | * on the hotplug state of the cpu. This function is not | |
1212 | * allowed to fail currently! | |
1213 | */ | |
1214 | for_each_present_cpu(cpu) { | |
1215 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu); | |
1216 | int cpustate = st->state; | |
1217 | ||
1218 | if (cpustate >= state) | |
1219 | cpuhp_issue_call(cpu, state, teardown, false); | |
1220 | } | |
1221 | remove: | |
1222 | cpuhp_store_callbacks(state, NULL, NULL, NULL); | |
1223 | put_online_cpus(); | |
1224 | } | |
1225 | EXPORT_SYMBOL(__cpuhp_remove_state); | |
1226 | ||
98f8cdce TG |
1227 | #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU) |
1228 | static ssize_t show_cpuhp_state(struct device *dev, | |
1229 | struct device_attribute *attr, char *buf) | |
1230 | { | |
1231 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1232 | ||
1233 | return sprintf(buf, "%d\n", st->state); | |
1234 | } | |
1235 | static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL); | |
1236 | ||
757c989b TG |
1237 | static ssize_t write_cpuhp_target(struct device *dev, |
1238 | struct device_attribute *attr, | |
1239 | const char *buf, size_t count) | |
1240 | { | |
1241 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1242 | struct cpuhp_step *sp; | |
1243 | int target, ret; | |
1244 | ||
1245 | ret = kstrtoint(buf, 10, &target); | |
1246 | if (ret) | |
1247 | return ret; | |
1248 | ||
1249 | #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL | |
1250 | if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE) | |
1251 | return -EINVAL; | |
1252 | #else | |
1253 | if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE) | |
1254 | return -EINVAL; | |
1255 | #endif | |
1256 | ||
1257 | ret = lock_device_hotplug_sysfs(); | |
1258 | if (ret) | |
1259 | return ret; | |
1260 | ||
1261 | mutex_lock(&cpuhp_state_mutex); | |
1262 | sp = cpuhp_get_step(target); | |
1263 | ret = !sp->name || sp->cant_stop ? -EINVAL : 0; | |
1264 | mutex_unlock(&cpuhp_state_mutex); | |
1265 | if (ret) | |
1266 | return ret; | |
1267 | ||
1268 | if (st->state < target) | |
1269 | ret = do_cpu_up(dev->id, target); | |
1270 | else | |
1271 | ret = do_cpu_down(dev->id, target); | |
1272 | ||
1273 | unlock_device_hotplug(); | |
1274 | return ret ? ret : count; | |
1275 | } | |
1276 | ||
98f8cdce TG |
1277 | static ssize_t show_cpuhp_target(struct device *dev, |
1278 | struct device_attribute *attr, char *buf) | |
1279 | { | |
1280 | struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id); | |
1281 | ||
1282 | return sprintf(buf, "%d\n", st->target); | |
1283 | } | |
757c989b | 1284 | static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target); |
98f8cdce TG |
1285 | |
1286 | static struct attribute *cpuhp_cpu_attrs[] = { | |
1287 | &dev_attr_state.attr, | |
1288 | &dev_attr_target.attr, | |
1289 | NULL | |
1290 | }; | |
1291 | ||
1292 | static struct attribute_group cpuhp_cpu_attr_group = { | |
1293 | .attrs = cpuhp_cpu_attrs, | |
1294 | .name = "hotplug", | |
1295 | NULL | |
1296 | }; | |
1297 | ||
1298 | static ssize_t show_cpuhp_states(struct device *dev, | |
1299 | struct device_attribute *attr, char *buf) | |
1300 | { | |
1301 | ssize_t cur, res = 0; | |
1302 | int i; | |
1303 | ||
1304 | mutex_lock(&cpuhp_state_mutex); | |
757c989b | 1305 | for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) { |
98f8cdce TG |
1306 | struct cpuhp_step *sp = cpuhp_get_step(i); |
1307 | ||
1308 | if (sp->name) { | |
1309 | cur = sprintf(buf, "%3d: %s\n", i, sp->name); | |
1310 | buf += cur; | |
1311 | res += cur; | |
1312 | } | |
1313 | } | |
1314 | mutex_unlock(&cpuhp_state_mutex); | |
1315 | return res; | |
1316 | } | |
1317 | static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL); | |
1318 | ||
1319 | static struct attribute *cpuhp_cpu_root_attrs[] = { | |
1320 | &dev_attr_states.attr, | |
1321 | NULL | |
1322 | }; | |
1323 | ||
1324 | static struct attribute_group cpuhp_cpu_root_attr_group = { | |
1325 | .attrs = cpuhp_cpu_root_attrs, | |
1326 | .name = "hotplug", | |
1327 | NULL | |
1328 | }; | |
1329 | ||
1330 | static int __init cpuhp_sysfs_init(void) | |
1331 | { | |
1332 | int cpu, ret; | |
1333 | ||
1334 | ret = sysfs_create_group(&cpu_subsys.dev_root->kobj, | |
1335 | &cpuhp_cpu_root_attr_group); | |
1336 | if (ret) | |
1337 | return ret; | |
1338 | ||
1339 | for_each_possible_cpu(cpu) { | |
1340 | struct device *dev = get_cpu_device(cpu); | |
1341 | ||
1342 | if (!dev) | |
1343 | continue; | |
1344 | ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group); | |
1345 | if (ret) | |
1346 | return ret; | |
1347 | } | |
1348 | return 0; | |
1349 | } | |
1350 | device_initcall(cpuhp_sysfs_init); | |
1351 | #endif | |
1352 | ||
e56b3bc7 LT |
1353 | /* |
1354 | * cpu_bit_bitmap[] is a special, "compressed" data structure that | |
1355 | * represents all NR_CPUS bits binary values of 1<<nr. | |
1356 | * | |
e0b582ec | 1357 | * It is used by cpumask_of() to get a constant address to a CPU |
e56b3bc7 LT |
1358 | * mask value that has a single bit set only. |
1359 | */ | |
b8d317d1 | 1360 | |
e56b3bc7 | 1361 | /* cpu_bit_bitmap[0] is empty - so we can back into it */ |
4d51985e | 1362 | #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x)) |
e56b3bc7 LT |
1363 | #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1) |
1364 | #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2) | |
1365 | #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4) | |
b8d317d1 | 1366 | |
e56b3bc7 LT |
1367 | const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = { |
1368 | ||
1369 | MASK_DECLARE_8(0), MASK_DECLARE_8(8), | |
1370 | MASK_DECLARE_8(16), MASK_DECLARE_8(24), | |
1371 | #if BITS_PER_LONG > 32 | |
1372 | MASK_DECLARE_8(32), MASK_DECLARE_8(40), | |
1373 | MASK_DECLARE_8(48), MASK_DECLARE_8(56), | |
b8d317d1 MT |
1374 | #endif |
1375 | }; | |
e56b3bc7 | 1376 | EXPORT_SYMBOL_GPL(cpu_bit_bitmap); |
2d3854a3 RR |
1377 | |
1378 | const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL; | |
1379 | EXPORT_SYMBOL(cpu_all_bits); | |
b3199c02 RR |
1380 | |
1381 | #ifdef CONFIG_INIT_ALL_POSSIBLE | |
4b804c85 | 1382 | struct cpumask __cpu_possible_mask __read_mostly |
c4c54dd1 | 1383 | = {CPU_BITS_ALL}; |
b3199c02 | 1384 | #else |
4b804c85 | 1385 | struct cpumask __cpu_possible_mask __read_mostly; |
b3199c02 | 1386 | #endif |
4b804c85 | 1387 | EXPORT_SYMBOL(__cpu_possible_mask); |
b3199c02 | 1388 | |
4b804c85 RV |
1389 | struct cpumask __cpu_online_mask __read_mostly; |
1390 | EXPORT_SYMBOL(__cpu_online_mask); | |
b3199c02 | 1391 | |
4b804c85 RV |
1392 | struct cpumask __cpu_present_mask __read_mostly; |
1393 | EXPORT_SYMBOL(__cpu_present_mask); | |
b3199c02 | 1394 | |
4b804c85 RV |
1395 | struct cpumask __cpu_active_mask __read_mostly; |
1396 | EXPORT_SYMBOL(__cpu_active_mask); | |
3fa41520 | 1397 | |
3fa41520 RR |
1398 | void init_cpu_present(const struct cpumask *src) |
1399 | { | |
c4c54dd1 | 1400 | cpumask_copy(&__cpu_present_mask, src); |
3fa41520 RR |
1401 | } |
1402 | ||
1403 | void init_cpu_possible(const struct cpumask *src) | |
1404 | { | |
c4c54dd1 | 1405 | cpumask_copy(&__cpu_possible_mask, src); |
3fa41520 RR |
1406 | } |
1407 | ||
1408 | void init_cpu_online(const struct cpumask *src) | |
1409 | { | |
c4c54dd1 | 1410 | cpumask_copy(&__cpu_online_mask, src); |
3fa41520 | 1411 | } |
cff7d378 TG |
1412 | |
1413 | /* | |
1414 | * Activate the first processor. | |
1415 | */ | |
1416 | void __init boot_cpu_init(void) | |
1417 | { | |
1418 | int cpu = smp_processor_id(); | |
1419 | ||
1420 | /* Mark the boot cpu "present", "online" etc for SMP and UP case */ | |
1421 | set_cpu_online(cpu, true); | |
1422 | set_cpu_active(cpu, true); | |
1423 | set_cpu_present(cpu, true); | |
1424 | set_cpu_possible(cpu, true); | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * Must be called _AFTER_ setting up the per_cpu areas | |
1429 | */ | |
1430 | void __init boot_cpu_state_init(void) | |
1431 | { | |
1432 | per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE; | |
1433 | } |