cpu/hotplug: Let upcoming cpu bring itself fully up
[linux-2.6-block.git] / kernel / cpu.c
CommitLineData
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>
4cb28ced 25#include <linux/smpboot.h>
cff7d378 26
bb3632c6 27#include <trace/events/power.h>
cff7d378
TG
28#define CREATE_TRACE_POINTS
29#include <trace/events/cpuhp.h>
1da177e4 30
38498a67
TG
31#include "smpboot.h"
32
cff7d378
TG
33/**
34 * cpuhp_cpu_state - Per cpu hotplug state storage
35 * @state: The current cpu state
36 * @target: The target state
4cb28ced
TG
37 * @thread: Pointer to the hotplug thread
38 * @should_run: Thread should execute
39 * @cb_stat: The state for a single callback (install/uninstall)
40 * @cb: Single callback function (install/uninstall)
41 * @result: Result of the operation
42 * @done: Signal completion to the issuer of the task
cff7d378
TG
43 */
44struct cpuhp_cpu_state {
45 enum cpuhp_state state;
46 enum cpuhp_state target;
4cb28ced
TG
47#ifdef CONFIG_SMP
48 struct task_struct *thread;
49 bool should_run;
50 enum cpuhp_state cb_state;
51 int (*cb)(unsigned int cpu);
52 int result;
53 struct completion done;
54#endif
cff7d378
TG
55};
56
57static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
58
59/**
60 * cpuhp_step - Hotplug state machine step
61 * @name: Name of the step
62 * @startup: Startup function of the step
63 * @teardown: Teardown function of the step
64 * @skip_onerr: Do not invoke the functions on error rollback
65 * Will go away once the notifiers are gone
757c989b 66 * @cant_stop: Bringup/teardown can't be stopped at this step
cff7d378
TG
67 */
68struct cpuhp_step {
69 const char *name;
70 int (*startup)(unsigned int cpu);
71 int (*teardown)(unsigned int cpu);
72 bool skip_onerr;
757c989b 73 bool cant_stop;
cff7d378
TG
74};
75
98f8cdce 76static DEFINE_MUTEX(cpuhp_state_mutex);
cff7d378 77static struct cpuhp_step cpuhp_bp_states[];
4baa0afc 78static struct cpuhp_step cpuhp_ap_states[];
cff7d378
TG
79
80/**
81 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
82 * @cpu: The cpu for which the callback should be invoked
83 * @step: The step in the state machine
84 * @cb: The callback function to invoke
85 *
86 * Called from cpu hotplug and from the state register machinery
87 */
88static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step,
89 int (*cb)(unsigned int))
90{
91 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
92 int ret = 0;
93
94 if (cb) {
95 trace_cpuhp_enter(cpu, st->target, step, cb);
96 ret = cb(cpu);
97 trace_cpuhp_exit(cpu, st->state, step, ret);
98 }
99 return ret;
100}
101
98a79d6a 102#ifdef CONFIG_SMP
b3199c02 103/* Serializes the updates to cpu_online_mask, cpu_present_mask */
aa953877 104static DEFINE_MUTEX(cpu_add_remove_lock);
090e77c3
TG
105bool cpuhp_tasks_frozen;
106EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
1da177e4 107
79a6cdeb 108/*
93ae4f97
SB
109 * The following two APIs (cpu_maps_update_begin/done) must be used when
110 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
111 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
112 * hotplug callback (un)registration performed using __register_cpu_notifier()
113 * or __unregister_cpu_notifier().
79a6cdeb
LJ
114 */
115void cpu_maps_update_begin(void)
116{
117 mutex_lock(&cpu_add_remove_lock);
118}
93ae4f97 119EXPORT_SYMBOL(cpu_notifier_register_begin);
79a6cdeb
LJ
120
121void cpu_maps_update_done(void)
122{
123 mutex_unlock(&cpu_add_remove_lock);
124}
93ae4f97 125EXPORT_SYMBOL(cpu_notifier_register_done);
79a6cdeb 126
5c113fbe 127static RAW_NOTIFIER_HEAD(cpu_chain);
1da177e4 128
e3920fb4
RW
129/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
130 * Should always be manipulated under cpu_add_remove_lock
131 */
132static int cpu_hotplug_disabled;
133
79a6cdeb
LJ
134#ifdef CONFIG_HOTPLUG_CPU
135
d221938c
GS
136static struct {
137 struct task_struct *active_writer;
87af9e7f
DH
138 /* wait queue to wake up the active_writer */
139 wait_queue_head_t wq;
140 /* verifies that no writer will get active while readers are active */
141 struct mutex lock;
d221938c
GS
142 /*
143 * Also blocks the new readers during
144 * an ongoing cpu hotplug operation.
145 */
87af9e7f 146 atomic_t refcount;
a19423b9
GS
147
148#ifdef CONFIG_DEBUG_LOCK_ALLOC
149 struct lockdep_map dep_map;
150#endif
31950eb6
LT
151} cpu_hotplug = {
152 .active_writer = NULL,
87af9e7f 153 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
31950eb6 154 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
a19423b9
GS
155#ifdef CONFIG_DEBUG_LOCK_ALLOC
156 .dep_map = {.name = "cpu_hotplug.lock" },
157#endif
31950eb6 158};
d221938c 159
a19423b9
GS
160/* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
161#define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
dd56af42
PM
162#define cpuhp_lock_acquire_tryread() \
163 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
a19423b9
GS
164#define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
165#define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
166
62db99f4 167
86ef5c9a 168void get_online_cpus(void)
a9d9baa1 169{
d221938c
GS
170 might_sleep();
171 if (cpu_hotplug.active_writer == current)
aa953877 172 return;
a19423b9 173 cpuhp_lock_acquire_read();
d221938c 174 mutex_lock(&cpu_hotplug.lock);
87af9e7f 175 atomic_inc(&cpu_hotplug.refcount);
d221938c 176 mutex_unlock(&cpu_hotplug.lock);
a9d9baa1 177}
86ef5c9a 178EXPORT_SYMBOL_GPL(get_online_cpus);
90d45d17 179
86ef5c9a 180void put_online_cpus(void)
a9d9baa1 181{
87af9e7f
DH
182 int refcount;
183
d221938c 184 if (cpu_hotplug.active_writer == current)
aa953877 185 return;
075663d1 186
87af9e7f
DH
187 refcount = atomic_dec_return(&cpu_hotplug.refcount);
188 if (WARN_ON(refcount < 0)) /* try to fix things up */
189 atomic_inc(&cpu_hotplug.refcount);
190
191 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
192 wake_up(&cpu_hotplug.wq);
075663d1 193
a19423b9 194 cpuhp_lock_release();
d221938c 195
a9d9baa1 196}
86ef5c9a 197EXPORT_SYMBOL_GPL(put_online_cpus);
a9d9baa1 198
d221938c
GS
199/*
200 * This ensures that the hotplug operation can begin only when the
201 * refcount goes to zero.
202 *
203 * Note that during a cpu-hotplug operation, the new readers, if any,
204 * will be blocked by the cpu_hotplug.lock
205 *
d2ba7e2a
ON
206 * Since cpu_hotplug_begin() is always called after invoking
207 * cpu_maps_update_begin(), we can be sure that only one writer is active.
d221938c
GS
208 *
209 * Note that theoretically, there is a possibility of a livelock:
210 * - Refcount goes to zero, last reader wakes up the sleeping
211 * writer.
212 * - Last reader unlocks the cpu_hotplug.lock.
213 * - A new reader arrives at this moment, bumps up the refcount.
214 * - The writer acquires the cpu_hotplug.lock finds the refcount
215 * non zero and goes to sleep again.
216 *
217 * However, this is very difficult to achieve in practice since
86ef5c9a 218 * get_online_cpus() not an api which is called all that often.
d221938c
GS
219 *
220 */
b9d10be7 221void cpu_hotplug_begin(void)
d221938c 222{
87af9e7f 223 DEFINE_WAIT(wait);
d2ba7e2a 224
87af9e7f 225 cpu_hotplug.active_writer = current;
a19423b9 226 cpuhp_lock_acquire();
87af9e7f 227
d2ba7e2a
ON
228 for (;;) {
229 mutex_lock(&cpu_hotplug.lock);
87af9e7f
DH
230 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
231 if (likely(!atomic_read(&cpu_hotplug.refcount)))
232 break;
d221938c
GS
233 mutex_unlock(&cpu_hotplug.lock);
234 schedule();
d221938c 235 }
87af9e7f 236 finish_wait(&cpu_hotplug.wq, &wait);
d221938c
GS
237}
238
b9d10be7 239void cpu_hotplug_done(void)
d221938c
GS
240{
241 cpu_hotplug.active_writer = NULL;
242 mutex_unlock(&cpu_hotplug.lock);
a19423b9 243 cpuhp_lock_release();
d221938c 244}
79a6cdeb 245
16e53dbf
SB
246/*
247 * Wait for currently running CPU hotplug operations to complete (if any) and
248 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
249 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
250 * hotplug path before performing hotplug operations. So acquiring that lock
251 * guarantees mutual exclusion from any currently running hotplug operations.
252 */
253void cpu_hotplug_disable(void)
254{
255 cpu_maps_update_begin();
89af7ba5 256 cpu_hotplug_disabled++;
16e53dbf
SB
257 cpu_maps_update_done();
258}
32145c46 259EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
16e53dbf
SB
260
261void cpu_hotplug_enable(void)
262{
263 cpu_maps_update_begin();
89af7ba5 264 WARN_ON(--cpu_hotplug_disabled < 0);
16e53dbf
SB
265 cpu_maps_update_done();
266}
32145c46 267EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
b9d10be7 268#endif /* CONFIG_HOTPLUG_CPU */
79a6cdeb 269
1da177e4 270/* Need to know about CPUs going up/down? */
71cf5aee 271int register_cpu_notifier(struct notifier_block *nb)
1da177e4 272{
bd5349cf 273 int ret;
d221938c 274 cpu_maps_update_begin();
bd5349cf 275 ret = raw_notifier_chain_register(&cpu_chain, nb);
d221938c 276 cpu_maps_update_done();
bd5349cf 277 return ret;
1da177e4 278}
65edc68c 279
71cf5aee 280int __register_cpu_notifier(struct notifier_block *nb)
93ae4f97
SB
281{
282 return raw_notifier_chain_register(&cpu_chain, nb);
283}
284
090e77c3 285static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
e9fb7631
AM
286 int *nr_calls)
287{
090e77c3
TG
288 unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
289 void *hcpu = (void *)(long)cpu;
290
e6bde73b
AM
291 int ret;
292
090e77c3 293 ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
e9fb7631 294 nr_calls);
e6bde73b
AM
295
296 return notifier_to_errno(ret);
e9fb7631
AM
297}
298
090e77c3 299static int cpu_notify(unsigned long val, unsigned int cpu)
e9fb7631 300{
090e77c3 301 return __cpu_notify(val, cpu, -1, NULL);
e9fb7631
AM
302}
303
ba997462
TG
304/* Notifier wrappers for transitioning to state machine */
305static int notify_prepare(unsigned int cpu)
306{
307 int nr_calls = 0;
308 int ret;
309
310 ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
311 if (ret) {
312 nr_calls--;
313 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
314 __func__, cpu);
315 __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
316 }
317 return ret;
318}
319
320static int notify_online(unsigned int cpu)
321{
322 cpu_notify(CPU_ONLINE, cpu);
323 return 0;
324}
325
4baa0afc
TG
326static int notify_starting(unsigned int cpu)
327{
328 cpu_notify(CPU_STARTING, cpu);
329 return 0;
330}
331
8df3e07e
TG
332static int bringup_wait_for_ap(unsigned int cpu)
333{
334 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
335
336 wait_for_completion(&st->done);
337 return st->result;
338}
339
ba997462
TG
340static int bringup_cpu(unsigned int cpu)
341{
342 struct task_struct *idle = idle_thread_get(cpu);
343 int ret;
344
345 /* Arch-specific enabling code. */
346 ret = __cpu_up(cpu, idle);
347 if (ret) {
348 cpu_notify(CPU_UP_CANCELED, cpu);
349 return ret;
350 }
8df3e07e 351 ret = bringup_wait_for_ap(cpu);
ba997462 352 BUG_ON(!cpu_online(cpu));
8df3e07e 353 return ret;
ba997462
TG
354}
355
2e1a3483
TG
356/*
357 * Hotplug state machine related functions
358 */
359static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st,
360 struct cpuhp_step *steps)
361{
362 for (st->state++; st->state < st->target; st->state++) {
363 struct cpuhp_step *step = steps + st->state;
364
365 if (!step->skip_onerr)
366 cpuhp_invoke_callback(cpu, st->state, step->startup);
367 }
368}
369
370static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
371 struct cpuhp_step *steps, enum cpuhp_state target)
372{
373 enum cpuhp_state prev_state = st->state;
374 int ret = 0;
375
376 for (; st->state > target; st->state--) {
377 struct cpuhp_step *step = steps + st->state;
378
379 ret = cpuhp_invoke_callback(cpu, st->state, step->teardown);
380 if (ret) {
381 st->target = prev_state;
382 undo_cpu_down(cpu, st, steps);
383 break;
384 }
385 }
386 return ret;
387}
388
389static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st,
390 struct cpuhp_step *steps)
391{
392 for (st->state--; st->state > st->target; st->state--) {
393 struct cpuhp_step *step = steps + st->state;
394
395 if (!step->skip_onerr)
396 cpuhp_invoke_callback(cpu, st->state, step->teardown);
397 }
398}
399
400static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
401 struct cpuhp_step *steps, enum cpuhp_state target)
402{
403 enum cpuhp_state prev_state = st->state;
404 int ret = 0;
405
406 while (st->state < target) {
407 struct cpuhp_step *step;
408
409 st->state++;
410 step = steps + st->state;
411 ret = cpuhp_invoke_callback(cpu, st->state, step->startup);
412 if (ret) {
413 st->target = prev_state;
414 undo_cpu_up(cpu, st, steps);
415 break;
416 }
417 }
418 return ret;
419}
420
4cb28ced
TG
421/*
422 * The cpu hotplug threads manage the bringup and teardown of the cpus
423 */
424static void cpuhp_create(unsigned int cpu)
425{
426 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
427
428 init_completion(&st->done);
429}
430
431static int cpuhp_should_run(unsigned int cpu)
432{
433 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
434
435 return st->should_run;
436}
437
438/* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
439static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
440{
1cf4f629 441 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
4cb28ced
TG
442
443 return cpuhp_down_callbacks(cpu, st, cpuhp_ap_states, target);
444}
445
446/* Execute the online startup callbacks. Used to be CPU_ONLINE */
447static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st)
448{
449 return cpuhp_up_callbacks(cpu, st, cpuhp_ap_states, st->target);
450}
451
452/*
453 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
454 * callbacks when a state gets [un]installed at runtime.
455 */
456static void cpuhp_thread_fun(unsigned int cpu)
457{
458 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
459 int ret = 0;
460
461 /*
462 * Paired with the mb() in cpuhp_kick_ap_work and
463 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
464 */
465 smp_mb();
466 if (!st->should_run)
467 return;
468
469 st->should_run = false;
470
471 /* Single callback invocation for [un]install ? */
472 if (st->cb) {
473 if (st->cb_state < CPUHP_AP_ONLINE) {
474 local_irq_disable();
475 ret = cpuhp_invoke_callback(cpu, st->cb_state, st->cb);
476 local_irq_enable();
477 } else {
478 ret = cpuhp_invoke_callback(cpu, st->cb_state, st->cb);
479 }
480 } else {
1cf4f629 481 /* Cannot happen .... */
8df3e07e 482 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
1cf4f629 483
4cb28ced
TG
484 /* Regular hotplug work */
485 if (st->state < st->target)
486 ret = cpuhp_ap_online(cpu, st);
487 else if (st->state > st->target)
488 ret = cpuhp_ap_offline(cpu, st);
489 }
490 st->result = ret;
491 complete(&st->done);
492}
493
494/* Invoke a single callback on a remote cpu */
495static int cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state,
496 int (*cb)(unsigned int))
497{
498 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
499
500 if (!cpu_online(cpu))
501 return 0;
502
503 st->cb_state = state;
504 st->cb = cb;
505 /*
506 * Make sure the above stores are visible before should_run becomes
507 * true. Paired with the mb() above in cpuhp_thread_fun()
508 */
509 smp_mb();
510 st->should_run = true;
511 wake_up_process(st->thread);
512 wait_for_completion(&st->done);
513 return st->result;
514}
515
516/* Regular hotplug invocation of the AP hotplug thread */
1cf4f629 517static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
4cb28ced 518{
4cb28ced
TG
519 st->result = 0;
520 st->cb = NULL;
521 /*
522 * Make sure the above stores are visible before should_run becomes
523 * true. Paired with the mb() above in cpuhp_thread_fun()
524 */
525 smp_mb();
526 st->should_run = true;
527 wake_up_process(st->thread);
1cf4f629
TG
528}
529
530static int cpuhp_kick_ap_work(unsigned int cpu)
531{
532 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
533 enum cpuhp_state state = st->state;
534
535 trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work);
536 __cpuhp_kick_ap_work(st);
4cb28ced
TG
537 wait_for_completion(&st->done);
538 trace_cpuhp_exit(cpu, st->state, state, st->result);
539 return st->result;
540}
541
542static struct smp_hotplug_thread cpuhp_threads = {
543 .store = &cpuhp_state.thread,
544 .create = &cpuhp_create,
545 .thread_should_run = cpuhp_should_run,
546 .thread_fn = cpuhp_thread_fun,
547 .thread_comm = "cpuhp/%u",
548 .selfparking = true,
549};
550
551void __init cpuhp_threads_init(void)
552{
553 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
554 kthread_unpark(this_cpu_read(cpuhp_state.thread));
555}
556
00b9b0af 557#ifdef CONFIG_HOTPLUG_CPU
1da177e4 558EXPORT_SYMBOL(register_cpu_notifier);
93ae4f97 559EXPORT_SYMBOL(__register_cpu_notifier);
71cf5aee 560void unregister_cpu_notifier(struct notifier_block *nb)
1da177e4 561{
d221938c 562 cpu_maps_update_begin();
bd5349cf 563 raw_notifier_chain_unregister(&cpu_chain, nb);
d221938c 564 cpu_maps_update_done();
1da177e4
LT
565}
566EXPORT_SYMBOL(unregister_cpu_notifier);
567
71cf5aee 568void __unregister_cpu_notifier(struct notifier_block *nb)
93ae4f97
SB
569{
570 raw_notifier_chain_unregister(&cpu_chain, nb);
571}
572EXPORT_SYMBOL(__unregister_cpu_notifier);
573
e4cc2f87
AV
574/**
575 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
576 * @cpu: a CPU id
577 *
578 * This function walks all processes, finds a valid mm struct for each one and
579 * then clears a corresponding bit in mm's cpumask. While this all sounds
580 * trivial, there are various non-obvious corner cases, which this function
581 * tries to solve in a safe manner.
582 *
583 * Also note that the function uses a somewhat relaxed locking scheme, so it may
584 * be called only for an already offlined CPU.
585 */
cb79295e
AV
586void clear_tasks_mm_cpumask(int cpu)
587{
588 struct task_struct *p;
589
590 /*
591 * This function is called after the cpu is taken down and marked
592 * offline, so its not like new tasks will ever get this cpu set in
593 * their mm mask. -- Peter Zijlstra
594 * Thus, we may use rcu_read_lock() here, instead of grabbing
595 * full-fledged tasklist_lock.
596 */
e4cc2f87 597 WARN_ON(cpu_online(cpu));
cb79295e
AV
598 rcu_read_lock();
599 for_each_process(p) {
600 struct task_struct *t;
601
e4cc2f87
AV
602 /*
603 * Main thread might exit, but other threads may still have
604 * a valid mm. Find one.
605 */
cb79295e
AV
606 t = find_lock_task_mm(p);
607 if (!t)
608 continue;
609 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
610 task_unlock(t);
611 }
612 rcu_read_unlock();
613}
614
b728ca06 615static inline void check_for_tasks(int dead_cpu)
1da177e4 616{
b728ca06 617 struct task_struct *g, *p;
1da177e4 618
a75a6068
ON
619 read_lock(&tasklist_lock);
620 for_each_process_thread(g, p) {
b728ca06
KT
621 if (!p->on_rq)
622 continue;
623 /*
624 * We do the check with unlocked task_rq(p)->lock.
625 * Order the reading to do not warn about a task,
626 * which was running on this cpu in the past, and
627 * it's just been woken on another cpu.
628 */
629 rmb();
630 if (task_cpu(p) != dead_cpu)
631 continue;
632
633 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
634 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
a75a6068
ON
635 }
636 read_unlock(&tasklist_lock);
1da177e4
LT
637}
638
98458172
TG
639static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
640{
641 BUG_ON(cpu_notify(val, cpu));
642}
643
644static int notify_down_prepare(unsigned int cpu)
645{
646 int err, nr_calls = 0;
647
648 err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
649 if (err) {
650 nr_calls--;
651 __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
652 pr_warn("%s: attempt to take down CPU %u failed\n",
653 __func__, cpu);
654 }
655 return err;
656}
657
4baa0afc
TG
658static int notify_dying(unsigned int cpu)
659{
660 cpu_notify(CPU_DYING, cpu);
661 return 0;
662}
663
1da177e4 664/* Take this CPU down. */
71cf5aee 665static int take_cpu_down(void *_param)
1da177e4 666{
4baa0afc
TG
667 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
668 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
090e77c3 669 int err, cpu = smp_processor_id();
1da177e4 670
1da177e4
LT
671 /* Ensure this CPU doesn't handle any more interrupts. */
672 err = __cpu_disable();
673 if (err < 0)
f3705136 674 return err;
1da177e4 675
4baa0afc
TG
676 /* Invoke the former CPU_DYING callbacks */
677 for (; st->state > target; st->state--) {
678 struct cpuhp_step *step = cpuhp_ap_states + st->state;
679
680 cpuhp_invoke_callback(cpu, st->state, step->teardown);
681 }
52c063d1
TG
682 /* Give up timekeeping duties */
683 tick_handover_do_timer();
14e568e7 684 /* Park the stopper thread */
090e77c3 685 stop_machine_park(cpu);
f3705136 686 return 0;
1da177e4
LT
687}
688
98458172 689static int takedown_cpu(unsigned int cpu)
1da177e4 690{
98458172 691 int err;
1da177e4 692
6acce3ef
PZ
693 /*
694 * By now we've cleared cpu_active_mask, wait for all preempt-disabled
695 * and RCU users of this state to go away such that all new such users
696 * will observe it.
697 *
698 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
779de6ce 699 * not imply sync_sched(), so wait for both.
106dd5af
M
700 *
701 * Do sync before park smpboot threads to take care the rcu boost case.
6acce3ef 702 */
779de6ce
PM
703 if (IS_ENABLED(CONFIG_PREEMPT))
704 synchronize_rcu_mult(call_rcu, call_rcu_sched);
705 else
706 synchronize_rcu();
6acce3ef 707
1cf4f629
TG
708 /* Park the hotplug thread */
709 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
710
6acce3ef 711 /*
a8994181
TG
712 * Prevent irq alloc/free while the dying cpu reorganizes the
713 * interrupt affinities.
6acce3ef 714 */
a8994181 715 irq_lock_sparse();
6acce3ef 716
a8994181
TG
717 /*
718 * So now all preempt/rcu users must observe !cpu_active().
719 */
090e77c3 720 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
04321587 721 if (err) {
1da177e4 722 /* CPU didn't die: tell everyone. Can't complain. */
090e77c3 723 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
a8994181 724 irq_unlock_sparse();
98458172 725 return err;
8fa1d7d3 726 }
04321587 727 BUG_ON(cpu_online(cpu));
1da177e4 728
48c5ccae
PZ
729 /*
730 * The migration_call() CPU_DYING callback will have removed all
731 * runnable tasks from the cpu, there's only the idle task left now
732 * that the migration thread is done doing the stop_machine thing.
51a96c77
PZ
733 *
734 * Wait for the stop thread to go away.
48c5ccae 735 */
528a25b0 736 while (!per_cpu(cpu_dead_idle, cpu))
51a96c77 737 cpu_relax();
528a25b0
PM
738 smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */
739 per_cpu(cpu_dead_idle, cpu) = false;
1da177e4 740
a8994181
TG
741 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
742 irq_unlock_sparse();
743
345527b1 744 hotplug_cpu__broadcast_tick_pull(cpu);
1da177e4
LT
745 /* This actually kills the CPU. */
746 __cpu_die(cpu);
747
a49b116d 748 tick_cleanup_dead_cpu(cpu);
98458172
TG
749 return 0;
750}
1da177e4 751
98458172
TG
752static int notify_dead(unsigned int cpu)
753{
754 cpu_notify_nofail(CPU_DEAD, cpu);
1da177e4 755 check_for_tasks(cpu);
98458172
TG
756 return 0;
757}
758
cff7d378
TG
759#else
760#define notify_down_prepare NULL
761#define takedown_cpu NULL
762#define notify_dead NULL
4baa0afc 763#define notify_dying NULL
cff7d378
TG
764#endif
765
766#ifdef CONFIG_HOTPLUG_CPU
cff7d378 767
98458172 768/* Requires cpu_add_remove_lock to be held */
af1f4045
TG
769static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
770 enum cpuhp_state target)
98458172 771{
cff7d378
TG
772 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
773 int prev_state, ret = 0;
774 bool hasdied = false;
98458172
TG
775
776 if (num_online_cpus() == 1)
777 return -EBUSY;
778
757c989b 779 if (!cpu_present(cpu))
98458172
TG
780 return -EINVAL;
781
782 cpu_hotplug_begin();
783
784 cpuhp_tasks_frozen = tasks_frozen;
785
cff7d378 786 prev_state = st->state;
af1f4045 787 st->target = target;
1cf4f629
TG
788 /*
789 * If the current CPU state is in the range of the AP hotplug thread,
790 * then we need to kick the thread.
791 */
8df3e07e 792 if (st->state > CPUHP_TEARDOWN_CPU) {
1cf4f629
TG
793 ret = cpuhp_kick_ap_work(cpu);
794 /*
795 * The AP side has done the error rollback already. Just
796 * return the error code..
797 */
798 if (ret)
799 goto out;
800
801 /*
802 * We might have stopped still in the range of the AP hotplug
803 * thread. Nothing to do anymore.
804 */
8df3e07e 805 if (st->state > CPUHP_TEARDOWN_CPU)
1cf4f629
TG
806 goto out;
807 }
808 /*
8df3e07e 809 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1cf4f629
TG
810 * to do the further cleanups.
811 */
2e1a3483 812 ret = cpuhp_down_callbacks(cpu, st, cpuhp_bp_states, target);
98458172 813
cff7d378 814 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
1cf4f629 815out:
d221938c 816 cpu_hotplug_done();
cff7d378
TG
817 /* This post dead nonsense must die */
818 if (!ret && hasdied)
090e77c3 819 cpu_notify_nofail(CPU_POST_DEAD, cpu);
cff7d378 820 return ret;
e3920fb4
RW
821}
822
af1f4045 823static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
e3920fb4 824{
9ea09af3 825 int err;
e3920fb4 826
d221938c 827 cpu_maps_update_begin();
e761b772
MK
828
829 if (cpu_hotplug_disabled) {
e3920fb4 830 err = -EBUSY;
e761b772
MK
831 goto out;
832 }
833
af1f4045 834 err = _cpu_down(cpu, 0, target);
e3920fb4 835
e761b772 836out:
d221938c 837 cpu_maps_update_done();
1da177e4
LT
838 return err;
839}
af1f4045
TG
840int cpu_down(unsigned int cpu)
841{
842 return do_cpu_down(cpu, CPUHP_OFFLINE);
843}
b62b8ef9 844EXPORT_SYMBOL(cpu_down);
1da177e4
LT
845#endif /*CONFIG_HOTPLUG_CPU*/
846
4baa0afc
TG
847/**
848 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
849 * @cpu: cpu that just started
850 *
851 * This function calls the cpu_chain notifiers with CPU_STARTING.
852 * It must be called by the arch code on the new cpu, before the new cpu
853 * enables interrupts and before the "boot" cpu returns from __cpu_up().
854 */
855void notify_cpu_starting(unsigned int cpu)
856{
857 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
858 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
859
860 while (st->state < target) {
861 struct cpuhp_step *step;
862
863 st->state++;
864 step = cpuhp_ap_states + st->state;
865 cpuhp_invoke_callback(cpu, st->state, step->startup);
866 }
867}
868
949338e3
TG
869/*
870 * Called from the idle task. We need to set active here, so we can kick off
8df3e07e
TG
871 * the stopper thread and unpark the smpboot threads. If the target state is
872 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
873 * cpu further.
949338e3 874 */
8df3e07e 875void cpuhp_online_idle(enum cpuhp_state state)
949338e3 876{
8df3e07e
TG
877 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
878 unsigned int cpu = smp_processor_id();
879
880 /* Happens for the boot cpu */
881 if (state != CPUHP_AP_ONLINE_IDLE)
882 return;
883
884 st->state = CPUHP_AP_ONLINE_IDLE;
1cf4f629 885
949338e3
TG
886 /* The cpu is marked online, set it active now */
887 set_cpu_active(cpu, true);
8df3e07e 888 /* Unpark the stopper thread and the hotplug thread of this cpu */
949338e3 889 stop_machine_unpark(cpu);
1cf4f629 890 kthread_unpark(st->thread);
8df3e07e
TG
891
892 /* Should we go further up ? */
893 if (st->target > CPUHP_AP_ONLINE_IDLE)
894 __cpuhp_kick_ap_work(st);
895 else
896 complete(&st->done);
949338e3
TG
897}
898
e3920fb4 899/* Requires cpu_add_remove_lock to be held */
af1f4045 900static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1da177e4 901{
cff7d378 902 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
3bb5d2ee 903 struct task_struct *idle;
2e1a3483 904 int ret = 0;
1da177e4 905
d221938c 906 cpu_hotplug_begin();
38498a67 907
757c989b 908 if (!cpu_present(cpu)) {
5e5041f3
YI
909 ret = -EINVAL;
910 goto out;
911 }
912
757c989b
TG
913 /*
914 * The caller of do_cpu_up might have raced with another
915 * caller. Ignore it for now.
916 */
917 if (st->state >= target)
38498a67 918 goto out;
757c989b
TG
919
920 if (st->state == CPUHP_OFFLINE) {
921 /* Let it fail before we try to bring the cpu up */
922 idle = idle_thread_get(cpu);
923 if (IS_ERR(idle)) {
924 ret = PTR_ERR(idle);
925 goto out;
926 }
3bb5d2ee 927 }
38498a67 928
ba997462
TG
929 cpuhp_tasks_frozen = tasks_frozen;
930
af1f4045 931 st->target = target;
1cf4f629
TG
932 /*
933 * If the current CPU state is in the range of the AP hotplug thread,
934 * then we need to kick the thread once more.
935 */
8df3e07e 936 if (st->state > CPUHP_BRINGUP_CPU) {
1cf4f629
TG
937 ret = cpuhp_kick_ap_work(cpu);
938 /*
939 * The AP side has done the error rollback already. Just
940 * return the error code..
941 */
942 if (ret)
943 goto out;
944 }
945
946 /*
947 * Try to reach the target state. We max out on the BP at
8df3e07e 948 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1cf4f629
TG
949 * responsible for bringing it up to the target state.
950 */
8df3e07e 951 target = min((int)target, CPUHP_BRINGUP_CPU);
2e1a3483 952 ret = cpuhp_up_callbacks(cpu, st, cpuhp_bp_states, target);
38498a67 953out:
d221938c 954 cpu_hotplug_done();
e3920fb4
RW
955 return ret;
956}
957
af1f4045 958static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
e3920fb4
RW
959{
960 int err = 0;
cf23422b 961
e0b582ec 962 if (!cpu_possible(cpu)) {
84117da5
FF
963 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
964 cpu);
87d5e023 965#if defined(CONFIG_IA64)
84117da5 966 pr_err("please check additional_cpus= boot parameter\n");
73e753a5
KH
967#endif
968 return -EINVAL;
969 }
e3920fb4 970
01b0f197
TK
971 err = try_online_node(cpu_to_node(cpu));
972 if (err)
973 return err;
cf23422b 974
d221938c 975 cpu_maps_update_begin();
e761b772
MK
976
977 if (cpu_hotplug_disabled) {
e3920fb4 978 err = -EBUSY;
e761b772
MK
979 goto out;
980 }
981
af1f4045 982 err = _cpu_up(cpu, 0, target);
e761b772 983out:
d221938c 984 cpu_maps_update_done();
e3920fb4
RW
985 return err;
986}
af1f4045
TG
987
988int cpu_up(unsigned int cpu)
989{
990 return do_cpu_up(cpu, CPUHP_ONLINE);
991}
a513f6ba 992EXPORT_SYMBOL_GPL(cpu_up);
e3920fb4 993
f3de4be9 994#ifdef CONFIG_PM_SLEEP_SMP
e0b582ec 995static cpumask_var_t frozen_cpus;
e3920fb4
RW
996
997int disable_nonboot_cpus(void)
998{
e9a5f426 999 int cpu, first_cpu, error = 0;
e3920fb4 1000
d221938c 1001 cpu_maps_update_begin();
e0b582ec 1002 first_cpu = cpumask_first(cpu_online_mask);
9ee349ad
XF
1003 /*
1004 * We take down all of the non-boot CPUs in one shot to avoid races
e3920fb4
RW
1005 * with the userspace trying to use the CPU hotplug at the same time
1006 */
e0b582ec 1007 cpumask_clear(frozen_cpus);
6ad4c188 1008
84117da5 1009 pr_info("Disabling non-boot CPUs ...\n");
e3920fb4
RW
1010 for_each_online_cpu(cpu) {
1011 if (cpu == first_cpu)
1012 continue;
bb3632c6 1013 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
af1f4045 1014 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
bb3632c6 1015 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
feae3203 1016 if (!error)
e0b582ec 1017 cpumask_set_cpu(cpu, frozen_cpus);
feae3203 1018 else {
84117da5 1019 pr_err("Error taking CPU%d down: %d\n", cpu, error);
e3920fb4
RW
1020 break;
1021 }
1022 }
86886e55 1023
89af7ba5 1024 if (!error)
e3920fb4 1025 BUG_ON(num_online_cpus() > 1);
89af7ba5 1026 else
84117da5 1027 pr_err("Non-boot CPUs are not disabled\n");
89af7ba5
VK
1028
1029 /*
1030 * Make sure the CPUs won't be enabled by someone else. We need to do
1031 * this even in case of failure as all disable_nonboot_cpus() users are
1032 * supposed to do enable_nonboot_cpus() on the failure path.
1033 */
1034 cpu_hotplug_disabled++;
1035
d221938c 1036 cpu_maps_update_done();
e3920fb4
RW
1037 return error;
1038}
1039
d0af9eed
SS
1040void __weak arch_enable_nonboot_cpus_begin(void)
1041{
1042}
1043
1044void __weak arch_enable_nonboot_cpus_end(void)
1045{
1046}
1047
71cf5aee 1048void enable_nonboot_cpus(void)
e3920fb4
RW
1049{
1050 int cpu, error;
1051
1052 /* Allow everyone to use the CPU hotplug again */
d221938c 1053 cpu_maps_update_begin();
89af7ba5 1054 WARN_ON(--cpu_hotplug_disabled < 0);
e0b582ec 1055 if (cpumask_empty(frozen_cpus))
1d64b9cb 1056 goto out;
e3920fb4 1057
84117da5 1058 pr_info("Enabling non-boot CPUs ...\n");
d0af9eed
SS
1059
1060 arch_enable_nonboot_cpus_begin();
1061
e0b582ec 1062 for_each_cpu(cpu, frozen_cpus) {
bb3632c6 1063 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
af1f4045 1064 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
bb3632c6 1065 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
e3920fb4 1066 if (!error) {
84117da5 1067 pr_info("CPU%d is up\n", cpu);
e3920fb4
RW
1068 continue;
1069 }
84117da5 1070 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
e3920fb4 1071 }
d0af9eed
SS
1072
1073 arch_enable_nonboot_cpus_end();
1074
e0b582ec 1075 cpumask_clear(frozen_cpus);
1d64b9cb 1076out:
d221938c 1077 cpu_maps_update_done();
1da177e4 1078}
e0b582ec 1079
d7268a31 1080static int __init alloc_frozen_cpus(void)
e0b582ec
RR
1081{
1082 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1083 return -ENOMEM;
1084 return 0;
1085}
1086core_initcall(alloc_frozen_cpus);
79cfbdfa 1087
79cfbdfa
SB
1088/*
1089 * When callbacks for CPU hotplug notifications are being executed, we must
1090 * ensure that the state of the system with respect to the tasks being frozen
1091 * or not, as reported by the notification, remains unchanged *throughout the
1092 * duration* of the execution of the callbacks.
1093 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1094 *
1095 * This synchronization is implemented by mutually excluding regular CPU
1096 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1097 * Hibernate notifications.
1098 */
1099static int
1100cpu_hotplug_pm_callback(struct notifier_block *nb,
1101 unsigned long action, void *ptr)
1102{
1103 switch (action) {
1104
1105 case PM_SUSPEND_PREPARE:
1106 case PM_HIBERNATION_PREPARE:
16e53dbf 1107 cpu_hotplug_disable();
79cfbdfa
SB
1108 break;
1109
1110 case PM_POST_SUSPEND:
1111 case PM_POST_HIBERNATION:
16e53dbf 1112 cpu_hotplug_enable();
79cfbdfa
SB
1113 break;
1114
1115 default:
1116 return NOTIFY_DONE;
1117 }
1118
1119 return NOTIFY_OK;
1120}
1121
1122
d7268a31 1123static int __init cpu_hotplug_pm_sync_init(void)
79cfbdfa 1124{
6e32d479
FY
1125 /*
1126 * cpu_hotplug_pm_callback has higher priority than x86
1127 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1128 * to disable cpu hotplug to avoid cpu hotplug race.
1129 */
79cfbdfa
SB
1130 pm_notifier(cpu_hotplug_pm_callback, 0);
1131 return 0;
1132}
1133core_initcall(cpu_hotplug_pm_sync_init);
1134
f3de4be9 1135#endif /* CONFIG_PM_SLEEP_SMP */
68f4f1ec
MK
1136
1137#endif /* CONFIG_SMP */
b8d317d1 1138
cff7d378
TG
1139/* Boot processor state steps */
1140static struct cpuhp_step cpuhp_bp_states[] = {
1141 [CPUHP_OFFLINE] = {
1142 .name = "offline",
1143 .startup = NULL,
1144 .teardown = NULL,
1145 },
1146#ifdef CONFIG_SMP
1147 [CPUHP_CREATE_THREADS]= {
1148 .name = "threads:create",
1149 .startup = smpboot_create_threads,
1150 .teardown = NULL,
757c989b 1151 .cant_stop = true,
cff7d378
TG
1152 },
1153 [CPUHP_NOTIFY_PREPARE] = {
1154 .name = "notify:prepare",
1155 .startup = notify_prepare,
1156 .teardown = notify_dead,
1157 .skip_onerr = true,
757c989b 1158 .cant_stop = true,
cff7d378
TG
1159 },
1160 [CPUHP_BRINGUP_CPU] = {
1161 .name = "cpu:bringup",
1162 .startup = bringup_cpu,
4baa0afc 1163 .teardown = NULL,
757c989b 1164 .cant_stop = true,
4baa0afc
TG
1165 },
1166 [CPUHP_TEARDOWN_CPU] = {
1167 .name = "cpu:teardown",
1168 .startup = NULL,
cff7d378 1169 .teardown = takedown_cpu,
757c989b 1170 .cant_stop = true,
cff7d378 1171 },
cff7d378 1172#endif
cff7d378
TG
1173};
1174
4baa0afc
TG
1175/* Application processor state steps */
1176static struct cpuhp_step cpuhp_ap_states[] = {
1177#ifdef CONFIG_SMP
1178 [CPUHP_AP_NOTIFY_STARTING] = {
1179 .name = "notify:starting",
1180 .startup = notify_starting,
1181 .teardown = notify_dying,
1182 .skip_onerr = true,
757c989b 1183 .cant_stop = true,
4baa0afc 1184 },
1cf4f629
TG
1185 [CPUHP_AP_SMPBOOT_THREADS] = {
1186 .name = "smpboot:threads",
1187 .startup = smpboot_unpark_threads,
1188 .teardown = smpboot_park_threads,
1189 },
1190 [CPUHP_AP_NOTIFY_ONLINE] = {
1191 .name = "notify:online",
1192 .startup = notify_online,
1193 .teardown = notify_down_prepare,
1194 },
4baa0afc
TG
1195#endif
1196 [CPUHP_ONLINE] = {
1197 .name = "online",
1198 .startup = NULL,
1199 .teardown = NULL,
1200 },
1201};
1202
5b7aa87e
TG
1203/* Sanity check for callbacks */
1204static int cpuhp_cb_check(enum cpuhp_state state)
1205{
1206 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1207 return -EINVAL;
1208 return 0;
1209}
1210
98f8cdce
TG
1211static bool cpuhp_is_ap_state(enum cpuhp_state state)
1212{
1cf4f629
TG
1213 if (state >= CPUHP_AP_OFFLINE && state <= CPUHP_AP_ONLINE)
1214 return true;
8df3e07e 1215 return state > CPUHP_BRINGUP_CPU;
98f8cdce
TG
1216}
1217
1218static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
1219{
1220 struct cpuhp_step *sp;
1221
1222 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
1223 return sp + state;
1224}
1225
5b7aa87e
TG
1226static void cpuhp_store_callbacks(enum cpuhp_state state,
1227 const char *name,
1228 int (*startup)(unsigned int cpu),
1229 int (*teardown)(unsigned int cpu))
1230{
1231 /* (Un)Install the callbacks for further cpu hotplug operations */
1232 struct cpuhp_step *sp;
1233
1234 mutex_lock(&cpuhp_state_mutex);
1235 sp = cpuhp_get_step(state);
1236 sp->startup = startup;
1237 sp->teardown = teardown;
1238 sp->name = name;
1239 mutex_unlock(&cpuhp_state_mutex);
1240}
1241
1242static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1243{
1244 return cpuhp_get_step(state)->teardown;
1245}
1246
5b7aa87e
TG
1247/*
1248 * Call the startup/teardown function for a step either on the AP or
1249 * on the current CPU.
1250 */
1251static int cpuhp_issue_call(int cpu, enum cpuhp_state state,
1252 int (*cb)(unsigned int), bool bringup)
1253{
1254 int ret;
1255
1256 if (!cb)
1257 return 0;
5b7aa87e
TG
1258 /*
1259 * The non AP bound callbacks can fail on bringup. On teardown
1260 * e.g. module removal we crash for now.
1261 */
1cf4f629
TG
1262#ifdef CONFIG_SMP
1263 if (cpuhp_is_ap_state(state))
1264 ret = cpuhp_invoke_ap_callback(cpu, state, cb);
1265 else
1266 ret = cpuhp_invoke_callback(cpu, state, cb);
1267#else
1268 ret = cpuhp_invoke_callback(cpu, state, cb);
1269#endif
5b7aa87e
TG
1270 BUG_ON(ret && !bringup);
1271 return ret;
1272}
1273
1274/*
1275 * Called from __cpuhp_setup_state on a recoverable failure.
1276 *
1277 * Note: The teardown callbacks for rollback are not allowed to fail!
1278 */
1279static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1280 int (*teardown)(unsigned int cpu))
1281{
1282 int cpu;
1283
1284 if (!teardown)
1285 return;
1286
1287 /* Roll back the already executed steps on the other cpus */
1288 for_each_present_cpu(cpu) {
1289 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1290 int cpustate = st->state;
1291
1292 if (cpu >= failedcpu)
1293 break;
1294
1295 /* Did we invoke the startup call on that cpu ? */
1296 if (cpustate >= state)
1297 cpuhp_issue_call(cpu, state, teardown, false);
1298 }
1299}
1300
1301/*
1302 * Returns a free for dynamic slot assignment of the Online state. The states
1303 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1304 * by having no name assigned.
1305 */
1306static int cpuhp_reserve_state(enum cpuhp_state state)
1307{
1308 enum cpuhp_state i;
1309
1310 mutex_lock(&cpuhp_state_mutex);
1cf4f629
TG
1311 for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1312 if (cpuhp_ap_states[i].name)
5b7aa87e
TG
1313 continue;
1314
1cf4f629 1315 cpuhp_ap_states[i].name = "Reserved";
5b7aa87e
TG
1316 mutex_unlock(&cpuhp_state_mutex);
1317 return i;
1318 }
1319 mutex_unlock(&cpuhp_state_mutex);
1320 WARN(1, "No more dynamic states available for CPU hotplug\n");
1321 return -ENOSPC;
1322}
1323
1324/**
1325 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1326 * @state: The state to setup
1327 * @invoke: If true, the startup function is invoked for cpus where
1328 * cpu state >= @state
1329 * @startup: startup callback function
1330 * @teardown: teardown callback function
1331 *
1332 * Returns 0 if successful, otherwise a proper error code
1333 */
1334int __cpuhp_setup_state(enum cpuhp_state state,
1335 const char *name, bool invoke,
1336 int (*startup)(unsigned int cpu),
1337 int (*teardown)(unsigned int cpu))
1338{
1339 int cpu, ret = 0;
1340 int dyn_state = 0;
1341
1342 if (cpuhp_cb_check(state) || !name)
1343 return -EINVAL;
1344
1345 get_online_cpus();
1346
1347 /* currently assignments for the ONLINE state are possible */
1cf4f629 1348 if (state == CPUHP_AP_ONLINE_DYN) {
5b7aa87e
TG
1349 dyn_state = 1;
1350 ret = cpuhp_reserve_state(state);
1351 if (ret < 0)
1352 goto out;
1353 state = ret;
1354 }
1355
1356 cpuhp_store_callbacks(state, name, startup, teardown);
1357
1358 if (!invoke || !startup)
1359 goto out;
1360
1361 /*
1362 * Try to call the startup callback for each present cpu
1363 * depending on the hotplug state of the cpu.
1364 */
1365 for_each_present_cpu(cpu) {
1366 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1367 int cpustate = st->state;
1368
1369 if (cpustate < state)
1370 continue;
1371
1372 ret = cpuhp_issue_call(cpu, state, startup, true);
1373 if (ret) {
1374 cpuhp_rollback_install(cpu, state, teardown);
1375 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1376 goto out;
1377 }
1378 }
1379out:
1380 put_online_cpus();
1381 if (!ret && dyn_state)
1382 return state;
1383 return ret;
1384}
1385EXPORT_SYMBOL(__cpuhp_setup_state);
1386
1387/**
1388 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1389 * @state: The state to remove
1390 * @invoke: If true, the teardown function is invoked for cpus where
1391 * cpu state >= @state
1392 *
1393 * The teardown callback is currently not allowed to fail. Think
1394 * about module removal!
1395 */
1396void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1397{
1398 int (*teardown)(unsigned int cpu) = cpuhp_get_teardown_cb(state);
1399 int cpu;
1400
1401 BUG_ON(cpuhp_cb_check(state));
1402
1403 get_online_cpus();
1404
1405 if (!invoke || !teardown)
1406 goto remove;
1407
1408 /*
1409 * Call the teardown callback for each present cpu depending
1410 * on the hotplug state of the cpu. This function is not
1411 * allowed to fail currently!
1412 */
1413 for_each_present_cpu(cpu) {
1414 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1415 int cpustate = st->state;
1416
1417 if (cpustate >= state)
1418 cpuhp_issue_call(cpu, state, teardown, false);
1419 }
1420remove:
1421 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1422 put_online_cpus();
1423}
1424EXPORT_SYMBOL(__cpuhp_remove_state);
1425
98f8cdce
TG
1426#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1427static ssize_t show_cpuhp_state(struct device *dev,
1428 struct device_attribute *attr, char *buf)
1429{
1430 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1431
1432 return sprintf(buf, "%d\n", st->state);
1433}
1434static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1435
757c989b
TG
1436static ssize_t write_cpuhp_target(struct device *dev,
1437 struct device_attribute *attr,
1438 const char *buf, size_t count)
1439{
1440 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1441 struct cpuhp_step *sp;
1442 int target, ret;
1443
1444 ret = kstrtoint(buf, 10, &target);
1445 if (ret)
1446 return ret;
1447
1448#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1449 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1450 return -EINVAL;
1451#else
1452 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1453 return -EINVAL;
1454#endif
1455
1456 ret = lock_device_hotplug_sysfs();
1457 if (ret)
1458 return ret;
1459
1460 mutex_lock(&cpuhp_state_mutex);
1461 sp = cpuhp_get_step(target);
1462 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1463 mutex_unlock(&cpuhp_state_mutex);
1464 if (ret)
1465 return ret;
1466
1467 if (st->state < target)
1468 ret = do_cpu_up(dev->id, target);
1469 else
1470 ret = do_cpu_down(dev->id, target);
1471
1472 unlock_device_hotplug();
1473 return ret ? ret : count;
1474}
1475
98f8cdce
TG
1476static ssize_t show_cpuhp_target(struct device *dev,
1477 struct device_attribute *attr, char *buf)
1478{
1479 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1480
1481 return sprintf(buf, "%d\n", st->target);
1482}
757c989b 1483static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
98f8cdce
TG
1484
1485static struct attribute *cpuhp_cpu_attrs[] = {
1486 &dev_attr_state.attr,
1487 &dev_attr_target.attr,
1488 NULL
1489};
1490
1491static struct attribute_group cpuhp_cpu_attr_group = {
1492 .attrs = cpuhp_cpu_attrs,
1493 .name = "hotplug",
1494 NULL
1495};
1496
1497static ssize_t show_cpuhp_states(struct device *dev,
1498 struct device_attribute *attr, char *buf)
1499{
1500 ssize_t cur, res = 0;
1501 int i;
1502
1503 mutex_lock(&cpuhp_state_mutex);
757c989b 1504 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
98f8cdce
TG
1505 struct cpuhp_step *sp = cpuhp_get_step(i);
1506
1507 if (sp->name) {
1508 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1509 buf += cur;
1510 res += cur;
1511 }
1512 }
1513 mutex_unlock(&cpuhp_state_mutex);
1514 return res;
1515}
1516static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1517
1518static struct attribute *cpuhp_cpu_root_attrs[] = {
1519 &dev_attr_states.attr,
1520 NULL
1521};
1522
1523static struct attribute_group cpuhp_cpu_root_attr_group = {
1524 .attrs = cpuhp_cpu_root_attrs,
1525 .name = "hotplug",
1526 NULL
1527};
1528
1529static int __init cpuhp_sysfs_init(void)
1530{
1531 int cpu, ret;
1532
1533 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1534 &cpuhp_cpu_root_attr_group);
1535 if (ret)
1536 return ret;
1537
1538 for_each_possible_cpu(cpu) {
1539 struct device *dev = get_cpu_device(cpu);
1540
1541 if (!dev)
1542 continue;
1543 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1544 if (ret)
1545 return ret;
1546 }
1547 return 0;
1548}
1549device_initcall(cpuhp_sysfs_init);
1550#endif
1551
e56b3bc7
LT
1552/*
1553 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1554 * represents all NR_CPUS bits binary values of 1<<nr.
1555 *
e0b582ec 1556 * It is used by cpumask_of() to get a constant address to a CPU
e56b3bc7
LT
1557 * mask value that has a single bit set only.
1558 */
b8d317d1 1559
e56b3bc7 1560/* cpu_bit_bitmap[0] is empty - so we can back into it */
4d51985e 1561#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
e56b3bc7
LT
1562#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1563#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1564#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
b8d317d1 1565
e56b3bc7
LT
1566const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1567
1568 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1569 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1570#if BITS_PER_LONG > 32
1571 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1572 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
b8d317d1
MT
1573#endif
1574};
e56b3bc7 1575EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2d3854a3
RR
1576
1577const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1578EXPORT_SYMBOL(cpu_all_bits);
b3199c02
RR
1579
1580#ifdef CONFIG_INIT_ALL_POSSIBLE
4b804c85 1581struct cpumask __cpu_possible_mask __read_mostly
c4c54dd1 1582 = {CPU_BITS_ALL};
b3199c02 1583#else
4b804c85 1584struct cpumask __cpu_possible_mask __read_mostly;
b3199c02 1585#endif
4b804c85 1586EXPORT_SYMBOL(__cpu_possible_mask);
b3199c02 1587
4b804c85
RV
1588struct cpumask __cpu_online_mask __read_mostly;
1589EXPORT_SYMBOL(__cpu_online_mask);
b3199c02 1590
4b804c85
RV
1591struct cpumask __cpu_present_mask __read_mostly;
1592EXPORT_SYMBOL(__cpu_present_mask);
b3199c02 1593
4b804c85
RV
1594struct cpumask __cpu_active_mask __read_mostly;
1595EXPORT_SYMBOL(__cpu_active_mask);
3fa41520 1596
3fa41520
RR
1597void init_cpu_present(const struct cpumask *src)
1598{
c4c54dd1 1599 cpumask_copy(&__cpu_present_mask, src);
3fa41520
RR
1600}
1601
1602void init_cpu_possible(const struct cpumask *src)
1603{
c4c54dd1 1604 cpumask_copy(&__cpu_possible_mask, src);
3fa41520
RR
1605}
1606
1607void init_cpu_online(const struct cpumask *src)
1608{
c4c54dd1 1609 cpumask_copy(&__cpu_online_mask, src);
3fa41520 1610}
cff7d378
TG
1611
1612/*
1613 * Activate the first processor.
1614 */
1615void __init boot_cpu_init(void)
1616{
1617 int cpu = smp_processor_id();
1618
1619 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1620 set_cpu_online(cpu, true);
1621 set_cpu_active(cpu, true);
1622 set_cpu_present(cpu, true);
1623 set_cpu_possible(cpu, true);
1624}
1625
1626/*
1627 * Must be called _AFTER_ setting up the per_cpu areas
1628 */
1629void __init boot_cpu_state_init(void)
1630{
1631 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;
1632}