]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/cpu.c
mac80211: Add support for beacon report radio measurement
[mirror_ubuntu-bionic-kernel.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
3b9d6da6 39 * @rollback: Perform a rollback
4cb28ced
TG
40 * @cb_stat: The state for a single callback (install/uninstall)
41 * @cb: Single callback function (install/uninstall)
42 * @result: Result of the operation
43 * @done: Signal completion to the issuer of the task
cff7d378
TG
44 */
45struct cpuhp_cpu_state {
46 enum cpuhp_state state;
47 enum cpuhp_state target;
4cb28ced
TG
48#ifdef CONFIG_SMP
49 struct task_struct *thread;
50 bool should_run;
3b9d6da6 51 bool rollback;
4cb28ced
TG
52 enum cpuhp_state cb_state;
53 int (*cb)(unsigned int cpu);
54 int result;
55 struct completion done;
56#endif
cff7d378
TG
57};
58
59static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
60
61/**
62 * cpuhp_step - Hotplug state machine step
63 * @name: Name of the step
64 * @startup: Startup function of the step
65 * @teardown: Teardown function of the step
66 * @skip_onerr: Do not invoke the functions on error rollback
67 * Will go away once the notifiers are gone
757c989b 68 * @cant_stop: Bringup/teardown can't be stopped at this step
cff7d378
TG
69 */
70struct cpuhp_step {
71 const char *name;
72 int (*startup)(unsigned int cpu);
73 int (*teardown)(unsigned int cpu);
74 bool skip_onerr;
757c989b 75 bool cant_stop;
cff7d378
TG
76};
77
98f8cdce 78static DEFINE_MUTEX(cpuhp_state_mutex);
cff7d378 79static struct cpuhp_step cpuhp_bp_states[];
4baa0afc 80static struct cpuhp_step cpuhp_ap_states[];
cff7d378
TG
81
82/**
83 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
84 * @cpu: The cpu for which the callback should be invoked
85 * @step: The step in the state machine
86 * @cb: The callback function to invoke
87 *
88 * Called from cpu hotplug and from the state register machinery
89 */
90static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step,
91 int (*cb)(unsigned int))
92{
93 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
94 int ret = 0;
95
96 if (cb) {
97 trace_cpuhp_enter(cpu, st->target, step, cb);
98 ret = cb(cpu);
99 trace_cpuhp_exit(cpu, st->state, step, ret);
100 }
101 return ret;
102}
103
98a79d6a 104#ifdef CONFIG_SMP
b3199c02 105/* Serializes the updates to cpu_online_mask, cpu_present_mask */
aa953877 106static DEFINE_MUTEX(cpu_add_remove_lock);
090e77c3
TG
107bool cpuhp_tasks_frozen;
108EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
1da177e4 109
79a6cdeb 110/*
93ae4f97
SB
111 * The following two APIs (cpu_maps_update_begin/done) must be used when
112 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
113 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
114 * hotplug callback (un)registration performed using __register_cpu_notifier()
115 * or __unregister_cpu_notifier().
79a6cdeb
LJ
116 */
117void cpu_maps_update_begin(void)
118{
119 mutex_lock(&cpu_add_remove_lock);
120}
93ae4f97 121EXPORT_SYMBOL(cpu_notifier_register_begin);
79a6cdeb
LJ
122
123void cpu_maps_update_done(void)
124{
125 mutex_unlock(&cpu_add_remove_lock);
126}
93ae4f97 127EXPORT_SYMBOL(cpu_notifier_register_done);
79a6cdeb 128
5c113fbe 129static RAW_NOTIFIER_HEAD(cpu_chain);
1da177e4 130
e3920fb4
RW
131/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
132 * Should always be manipulated under cpu_add_remove_lock
133 */
134static int cpu_hotplug_disabled;
135
79a6cdeb
LJ
136#ifdef CONFIG_HOTPLUG_CPU
137
d221938c
GS
138static struct {
139 struct task_struct *active_writer;
87af9e7f
DH
140 /* wait queue to wake up the active_writer */
141 wait_queue_head_t wq;
142 /* verifies that no writer will get active while readers are active */
143 struct mutex lock;
d221938c
GS
144 /*
145 * Also blocks the new readers during
146 * an ongoing cpu hotplug operation.
147 */
87af9e7f 148 atomic_t refcount;
a19423b9
GS
149
150#ifdef CONFIG_DEBUG_LOCK_ALLOC
151 struct lockdep_map dep_map;
152#endif
31950eb6
LT
153} cpu_hotplug = {
154 .active_writer = NULL,
87af9e7f 155 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
31950eb6 156 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
a19423b9
GS
157#ifdef CONFIG_DEBUG_LOCK_ALLOC
158 .dep_map = {.name = "cpu_hotplug.lock" },
159#endif
31950eb6 160};
d221938c 161
a19423b9
GS
162/* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
163#define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
dd56af42
PM
164#define cpuhp_lock_acquire_tryread() \
165 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
a19423b9
GS
166#define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
167#define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
168
62db99f4 169
86ef5c9a 170void get_online_cpus(void)
a9d9baa1 171{
d221938c
GS
172 might_sleep();
173 if (cpu_hotplug.active_writer == current)
aa953877 174 return;
a19423b9 175 cpuhp_lock_acquire_read();
d221938c 176 mutex_lock(&cpu_hotplug.lock);
87af9e7f 177 atomic_inc(&cpu_hotplug.refcount);
d221938c 178 mutex_unlock(&cpu_hotplug.lock);
a9d9baa1 179}
86ef5c9a 180EXPORT_SYMBOL_GPL(get_online_cpus);
90d45d17 181
86ef5c9a 182void put_online_cpus(void)
a9d9baa1 183{
87af9e7f
DH
184 int refcount;
185
d221938c 186 if (cpu_hotplug.active_writer == current)
aa953877 187 return;
075663d1 188
87af9e7f
DH
189 refcount = atomic_dec_return(&cpu_hotplug.refcount);
190 if (WARN_ON(refcount < 0)) /* try to fix things up */
191 atomic_inc(&cpu_hotplug.refcount);
192
193 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
194 wake_up(&cpu_hotplug.wq);
075663d1 195
a19423b9 196 cpuhp_lock_release();
d221938c 197
a9d9baa1 198}
86ef5c9a 199EXPORT_SYMBOL_GPL(put_online_cpus);
a9d9baa1 200
d221938c
GS
201/*
202 * This ensures that the hotplug operation can begin only when the
203 * refcount goes to zero.
204 *
205 * Note that during a cpu-hotplug operation, the new readers, if any,
206 * will be blocked by the cpu_hotplug.lock
207 *
d2ba7e2a
ON
208 * Since cpu_hotplug_begin() is always called after invoking
209 * cpu_maps_update_begin(), we can be sure that only one writer is active.
d221938c
GS
210 *
211 * Note that theoretically, there is a possibility of a livelock:
212 * - Refcount goes to zero, last reader wakes up the sleeping
213 * writer.
214 * - Last reader unlocks the cpu_hotplug.lock.
215 * - A new reader arrives at this moment, bumps up the refcount.
216 * - The writer acquires the cpu_hotplug.lock finds the refcount
217 * non zero and goes to sleep again.
218 *
219 * However, this is very difficult to achieve in practice since
86ef5c9a 220 * get_online_cpus() not an api which is called all that often.
d221938c
GS
221 *
222 */
b9d10be7 223void cpu_hotplug_begin(void)
d221938c 224{
87af9e7f 225 DEFINE_WAIT(wait);
d2ba7e2a 226
87af9e7f 227 cpu_hotplug.active_writer = current;
a19423b9 228 cpuhp_lock_acquire();
87af9e7f 229
d2ba7e2a
ON
230 for (;;) {
231 mutex_lock(&cpu_hotplug.lock);
87af9e7f
DH
232 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
233 if (likely(!atomic_read(&cpu_hotplug.refcount)))
234 break;
d221938c
GS
235 mutex_unlock(&cpu_hotplug.lock);
236 schedule();
d221938c 237 }
87af9e7f 238 finish_wait(&cpu_hotplug.wq, &wait);
d221938c
GS
239}
240
b9d10be7 241void cpu_hotplug_done(void)
d221938c
GS
242{
243 cpu_hotplug.active_writer = NULL;
244 mutex_unlock(&cpu_hotplug.lock);
a19423b9 245 cpuhp_lock_release();
d221938c 246}
79a6cdeb 247
16e53dbf
SB
248/*
249 * Wait for currently running CPU hotplug operations to complete (if any) and
250 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
251 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
252 * hotplug path before performing hotplug operations. So acquiring that lock
253 * guarantees mutual exclusion from any currently running hotplug operations.
254 */
255void cpu_hotplug_disable(void)
256{
257 cpu_maps_update_begin();
89af7ba5 258 cpu_hotplug_disabled++;
16e53dbf
SB
259 cpu_maps_update_done();
260}
32145c46 261EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
16e53dbf
SB
262
263void cpu_hotplug_enable(void)
264{
265 cpu_maps_update_begin();
89af7ba5 266 WARN_ON(--cpu_hotplug_disabled < 0);
16e53dbf
SB
267 cpu_maps_update_done();
268}
32145c46 269EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
b9d10be7 270#endif /* CONFIG_HOTPLUG_CPU */
79a6cdeb 271
1da177e4 272/* Need to know about CPUs going up/down? */
71cf5aee 273int register_cpu_notifier(struct notifier_block *nb)
1da177e4 274{
bd5349cf 275 int ret;
d221938c 276 cpu_maps_update_begin();
bd5349cf 277 ret = raw_notifier_chain_register(&cpu_chain, nb);
d221938c 278 cpu_maps_update_done();
bd5349cf 279 return ret;
1da177e4 280}
65edc68c 281
71cf5aee 282int __register_cpu_notifier(struct notifier_block *nb)
93ae4f97
SB
283{
284 return raw_notifier_chain_register(&cpu_chain, nb);
285}
286
090e77c3 287static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
e9fb7631
AM
288 int *nr_calls)
289{
090e77c3
TG
290 unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
291 void *hcpu = (void *)(long)cpu;
292
e6bde73b
AM
293 int ret;
294
090e77c3 295 ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
e9fb7631 296 nr_calls);
e6bde73b
AM
297
298 return notifier_to_errno(ret);
e9fb7631
AM
299}
300
090e77c3 301static int cpu_notify(unsigned long val, unsigned int cpu)
e9fb7631 302{
090e77c3 303 return __cpu_notify(val, cpu, -1, NULL);
e9fb7631
AM
304}
305
3b9d6da6
SAS
306static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
307{
308 BUG_ON(cpu_notify(val, cpu));
309}
310
ba997462
TG
311/* Notifier wrappers for transitioning to state machine */
312static int notify_prepare(unsigned int cpu)
313{
314 int nr_calls = 0;
315 int ret;
316
317 ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
318 if (ret) {
319 nr_calls--;
320 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
321 __func__, cpu);
322 __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
323 }
324 return ret;
325}
326
327static int notify_online(unsigned int cpu)
328{
329 cpu_notify(CPU_ONLINE, cpu);
330 return 0;
331}
332
4baa0afc
TG
333static int notify_starting(unsigned int cpu)
334{
335 cpu_notify(CPU_STARTING, cpu);
336 return 0;
337}
338
8df3e07e
TG
339static int bringup_wait_for_ap(unsigned int cpu)
340{
341 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
342
343 wait_for_completion(&st->done);
344 return st->result;
345}
346
ba997462
TG
347static int bringup_cpu(unsigned int cpu)
348{
349 struct task_struct *idle = idle_thread_get(cpu);
350 int ret;
351
352 /* Arch-specific enabling code. */
353 ret = __cpu_up(cpu, idle);
354 if (ret) {
355 cpu_notify(CPU_UP_CANCELED, cpu);
356 return ret;
357 }
8df3e07e 358 ret = bringup_wait_for_ap(cpu);
ba997462 359 BUG_ON(!cpu_online(cpu));
8df3e07e 360 return ret;
ba997462
TG
361}
362
2e1a3483
TG
363/*
364 * Hotplug state machine related functions
365 */
366static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st,
367 struct cpuhp_step *steps)
368{
369 for (st->state++; st->state < st->target; st->state++) {
370 struct cpuhp_step *step = steps + st->state;
371
372 if (!step->skip_onerr)
373 cpuhp_invoke_callback(cpu, st->state, step->startup);
374 }
375}
376
377static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
378 struct cpuhp_step *steps, enum cpuhp_state target)
379{
380 enum cpuhp_state prev_state = st->state;
381 int ret = 0;
382
383 for (; st->state > target; st->state--) {
384 struct cpuhp_step *step = steps + st->state;
385
386 ret = cpuhp_invoke_callback(cpu, st->state, step->teardown);
387 if (ret) {
388 st->target = prev_state;
389 undo_cpu_down(cpu, st, steps);
390 break;
391 }
392 }
393 return ret;
394}
395
396static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st,
397 struct cpuhp_step *steps)
398{
399 for (st->state--; st->state > st->target; st->state--) {
400 struct cpuhp_step *step = steps + st->state;
401
402 if (!step->skip_onerr)
403 cpuhp_invoke_callback(cpu, st->state, step->teardown);
404 }
405}
406
407static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
408 struct cpuhp_step *steps, enum cpuhp_state target)
409{
410 enum cpuhp_state prev_state = st->state;
411 int ret = 0;
412
413 while (st->state < target) {
414 struct cpuhp_step *step;
415
416 st->state++;
417 step = steps + st->state;
418 ret = cpuhp_invoke_callback(cpu, st->state, step->startup);
419 if (ret) {
420 st->target = prev_state;
421 undo_cpu_up(cpu, st, steps);
422 break;
423 }
424 }
425 return ret;
426}
427
4cb28ced
TG
428/*
429 * The cpu hotplug threads manage the bringup and teardown of the cpus
430 */
431static void cpuhp_create(unsigned int cpu)
432{
433 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
434
435 init_completion(&st->done);
436}
437
438static int cpuhp_should_run(unsigned int cpu)
439{
440 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
441
442 return st->should_run;
443}
444
445/* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
446static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
447{
1cf4f629 448 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
4cb28ced
TG
449
450 return cpuhp_down_callbacks(cpu, st, cpuhp_ap_states, target);
451}
452
453/* Execute the online startup callbacks. Used to be CPU_ONLINE */
454static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st)
455{
456 return cpuhp_up_callbacks(cpu, st, cpuhp_ap_states, st->target);
457}
458
459/*
460 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
461 * callbacks when a state gets [un]installed at runtime.
462 */
463static void cpuhp_thread_fun(unsigned int cpu)
464{
465 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
466 int ret = 0;
467
468 /*
469 * Paired with the mb() in cpuhp_kick_ap_work and
470 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
471 */
472 smp_mb();
473 if (!st->should_run)
474 return;
475
476 st->should_run = false;
477
478 /* Single callback invocation for [un]install ? */
479 if (st->cb) {
480 if (st->cb_state < CPUHP_AP_ONLINE) {
481 local_irq_disable();
482 ret = cpuhp_invoke_callback(cpu, st->cb_state, st->cb);
483 local_irq_enable();
484 } else {
485 ret = cpuhp_invoke_callback(cpu, st->cb_state, st->cb);
486 }
3b9d6da6
SAS
487 } else if (st->rollback) {
488 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
489
490 undo_cpu_down(cpu, st, cpuhp_ap_states);
491 /*
492 * This is a momentary workaround to keep the notifier users
493 * happy. Will go away once we got rid of the notifiers.
494 */
495 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
496 st->rollback = false;
4cb28ced 497 } else {
1cf4f629 498 /* Cannot happen .... */
8df3e07e 499 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
1cf4f629 500
4cb28ced
TG
501 /* Regular hotplug work */
502 if (st->state < st->target)
503 ret = cpuhp_ap_online(cpu, st);
504 else if (st->state > st->target)
505 ret = cpuhp_ap_offline(cpu, st);
506 }
507 st->result = ret;
508 complete(&st->done);
509}
510
511/* Invoke a single callback on a remote cpu */
512static int cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state,
513 int (*cb)(unsigned int))
514{
515 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
516
517 if (!cpu_online(cpu))
518 return 0;
519
520 st->cb_state = state;
521 st->cb = cb;
522 /*
523 * Make sure the above stores are visible before should_run becomes
524 * true. Paired with the mb() above in cpuhp_thread_fun()
525 */
526 smp_mb();
527 st->should_run = true;
528 wake_up_process(st->thread);
529 wait_for_completion(&st->done);
530 return st->result;
531}
532
533/* Regular hotplug invocation of the AP hotplug thread */
1cf4f629 534static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
4cb28ced 535{
4cb28ced
TG
536 st->result = 0;
537 st->cb = NULL;
538 /*
539 * Make sure the above stores are visible before should_run becomes
540 * true. Paired with the mb() above in cpuhp_thread_fun()
541 */
542 smp_mb();
543 st->should_run = true;
544 wake_up_process(st->thread);
1cf4f629
TG
545}
546
547static int cpuhp_kick_ap_work(unsigned int cpu)
548{
549 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
550 enum cpuhp_state state = st->state;
551
552 trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work);
553 __cpuhp_kick_ap_work(st);
4cb28ced
TG
554 wait_for_completion(&st->done);
555 trace_cpuhp_exit(cpu, st->state, state, st->result);
556 return st->result;
557}
558
559static struct smp_hotplug_thread cpuhp_threads = {
560 .store = &cpuhp_state.thread,
561 .create = &cpuhp_create,
562 .thread_should_run = cpuhp_should_run,
563 .thread_fn = cpuhp_thread_fun,
564 .thread_comm = "cpuhp/%u",
565 .selfparking = true,
566};
567
568void __init cpuhp_threads_init(void)
569{
570 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
571 kthread_unpark(this_cpu_read(cpuhp_state.thread));
572}
573
00b9b0af 574#ifdef CONFIG_HOTPLUG_CPU
1da177e4 575EXPORT_SYMBOL(register_cpu_notifier);
93ae4f97 576EXPORT_SYMBOL(__register_cpu_notifier);
71cf5aee 577void unregister_cpu_notifier(struct notifier_block *nb)
1da177e4 578{
d221938c 579 cpu_maps_update_begin();
bd5349cf 580 raw_notifier_chain_unregister(&cpu_chain, nb);
d221938c 581 cpu_maps_update_done();
1da177e4
LT
582}
583EXPORT_SYMBOL(unregister_cpu_notifier);
584
71cf5aee 585void __unregister_cpu_notifier(struct notifier_block *nb)
93ae4f97
SB
586{
587 raw_notifier_chain_unregister(&cpu_chain, nb);
588}
589EXPORT_SYMBOL(__unregister_cpu_notifier);
590
e4cc2f87
AV
591/**
592 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
593 * @cpu: a CPU id
594 *
595 * This function walks all processes, finds a valid mm struct for each one and
596 * then clears a corresponding bit in mm's cpumask. While this all sounds
597 * trivial, there are various non-obvious corner cases, which this function
598 * tries to solve in a safe manner.
599 *
600 * Also note that the function uses a somewhat relaxed locking scheme, so it may
601 * be called only for an already offlined CPU.
602 */
cb79295e
AV
603void clear_tasks_mm_cpumask(int cpu)
604{
605 struct task_struct *p;
606
607 /*
608 * This function is called after the cpu is taken down and marked
609 * offline, so its not like new tasks will ever get this cpu set in
610 * their mm mask. -- Peter Zijlstra
611 * Thus, we may use rcu_read_lock() here, instead of grabbing
612 * full-fledged tasklist_lock.
613 */
e4cc2f87 614 WARN_ON(cpu_online(cpu));
cb79295e
AV
615 rcu_read_lock();
616 for_each_process(p) {
617 struct task_struct *t;
618
e4cc2f87
AV
619 /*
620 * Main thread might exit, but other threads may still have
621 * a valid mm. Find one.
622 */
cb79295e
AV
623 t = find_lock_task_mm(p);
624 if (!t)
625 continue;
626 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
627 task_unlock(t);
628 }
629 rcu_read_unlock();
630}
631
b728ca06 632static inline void check_for_tasks(int dead_cpu)
1da177e4 633{
b728ca06 634 struct task_struct *g, *p;
1da177e4 635
a75a6068
ON
636 read_lock(&tasklist_lock);
637 for_each_process_thread(g, p) {
b728ca06
KT
638 if (!p->on_rq)
639 continue;
640 /*
641 * We do the check with unlocked task_rq(p)->lock.
642 * Order the reading to do not warn about a task,
643 * which was running on this cpu in the past, and
644 * it's just been woken on another cpu.
645 */
646 rmb();
647 if (task_cpu(p) != dead_cpu)
648 continue;
649
650 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
651 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
a75a6068
ON
652 }
653 read_unlock(&tasklist_lock);
1da177e4
LT
654}
655
98458172
TG
656static int notify_down_prepare(unsigned int cpu)
657{
658 int err, nr_calls = 0;
659
660 err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
661 if (err) {
662 nr_calls--;
663 __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
664 pr_warn("%s: attempt to take down CPU %u failed\n",
665 __func__, cpu);
666 }
667 return err;
668}
669
4baa0afc
TG
670static int notify_dying(unsigned int cpu)
671{
672 cpu_notify(CPU_DYING, cpu);
673 return 0;
674}
675
1da177e4 676/* Take this CPU down. */
71cf5aee 677static int take_cpu_down(void *_param)
1da177e4 678{
4baa0afc
TG
679 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
680 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
090e77c3 681 int err, cpu = smp_processor_id();
1da177e4 682
1da177e4
LT
683 /* Ensure this CPU doesn't handle any more interrupts. */
684 err = __cpu_disable();
685 if (err < 0)
f3705136 686 return err;
1da177e4 687
4baa0afc
TG
688 /* Invoke the former CPU_DYING callbacks */
689 for (; st->state > target; st->state--) {
690 struct cpuhp_step *step = cpuhp_ap_states + st->state;
691
692 cpuhp_invoke_callback(cpu, st->state, step->teardown);
693 }
52c063d1
TG
694 /* Give up timekeeping duties */
695 tick_handover_do_timer();
14e568e7 696 /* Park the stopper thread */
090e77c3 697 stop_machine_park(cpu);
f3705136 698 return 0;
1da177e4
LT
699}
700
98458172 701static int takedown_cpu(unsigned int cpu)
1da177e4 702{
e69aab13 703 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
98458172 704 int err;
1da177e4 705
2a58c527 706 /* Park the smpboot threads */
1cf4f629 707 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
2a58c527 708 smpboot_park_threads(cpu);
1cf4f629 709
6acce3ef 710 /*
a8994181
TG
711 * Prevent irq alloc/free while the dying cpu reorganizes the
712 * interrupt affinities.
6acce3ef 713 */
a8994181 714 irq_lock_sparse();
6acce3ef 715
a8994181
TG
716 /*
717 * So now all preempt/rcu users must observe !cpu_active().
718 */
090e77c3 719 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
04321587 720 if (err) {
3b9d6da6 721 /* CPU refused to die */
a8994181 722 irq_unlock_sparse();
3b9d6da6
SAS
723 /* Unpark the hotplug thread so we can rollback there */
724 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
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 */
e69aab13
TG
736 wait_for_completion(&st->done);
737 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1da177e4 738
a8994181
TG
739 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
740 irq_unlock_sparse();
741
345527b1 742 hotplug_cpu__broadcast_tick_pull(cpu);
1da177e4
LT
743 /* This actually kills the CPU. */
744 __cpu_die(cpu);
745
a49b116d 746 tick_cleanup_dead_cpu(cpu);
98458172
TG
747 return 0;
748}
1da177e4 749
98458172
TG
750static int notify_dead(unsigned int cpu)
751{
752 cpu_notify_nofail(CPU_DEAD, cpu);
1da177e4 753 check_for_tasks(cpu);
98458172
TG
754 return 0;
755}
756
71f87b2f
TG
757static void cpuhp_complete_idle_dead(void *arg)
758{
759 struct cpuhp_cpu_state *st = arg;
760
761 complete(&st->done);
762}
763
e69aab13
TG
764void cpuhp_report_idle_dead(void)
765{
766 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
767
768 BUG_ON(st->state != CPUHP_AP_OFFLINE);
27d50c7e 769 rcu_report_dead(smp_processor_id());
71f87b2f
TG
770 st->state = CPUHP_AP_IDLE_DEAD;
771 /*
772 * We cannot call complete after rcu_report_dead() so we delegate it
773 * to an online cpu.
774 */
775 smp_call_function_single(cpumask_first(cpu_online_mask),
776 cpuhp_complete_idle_dead, st, 0);
e69aab13
TG
777}
778
cff7d378
TG
779#else
780#define notify_down_prepare NULL
781#define takedown_cpu NULL
782#define notify_dead NULL
4baa0afc 783#define notify_dying NULL
cff7d378
TG
784#endif
785
786#ifdef CONFIG_HOTPLUG_CPU
cff7d378 787
98458172 788/* Requires cpu_add_remove_lock to be held */
af1f4045
TG
789static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
790 enum cpuhp_state target)
98458172 791{
cff7d378
TG
792 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
793 int prev_state, ret = 0;
794 bool hasdied = false;
98458172
TG
795
796 if (num_online_cpus() == 1)
797 return -EBUSY;
798
757c989b 799 if (!cpu_present(cpu))
98458172
TG
800 return -EINVAL;
801
802 cpu_hotplug_begin();
803
804 cpuhp_tasks_frozen = tasks_frozen;
805
cff7d378 806 prev_state = st->state;
af1f4045 807 st->target = target;
1cf4f629
TG
808 /*
809 * If the current CPU state is in the range of the AP hotplug thread,
810 * then we need to kick the thread.
811 */
8df3e07e 812 if (st->state > CPUHP_TEARDOWN_CPU) {
1cf4f629
TG
813 ret = cpuhp_kick_ap_work(cpu);
814 /*
815 * The AP side has done the error rollback already. Just
816 * return the error code..
817 */
818 if (ret)
819 goto out;
820
821 /*
822 * We might have stopped still in the range of the AP hotplug
823 * thread. Nothing to do anymore.
824 */
8df3e07e 825 if (st->state > CPUHP_TEARDOWN_CPU)
1cf4f629
TG
826 goto out;
827 }
828 /*
8df3e07e 829 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1cf4f629
TG
830 * to do the further cleanups.
831 */
2e1a3483 832 ret = cpuhp_down_callbacks(cpu, st, cpuhp_bp_states, target);
3b9d6da6
SAS
833 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
834 st->target = prev_state;
835 st->rollback = true;
836 cpuhp_kick_ap_work(cpu);
837 }
98458172 838
cff7d378 839 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
1cf4f629 840out:
d221938c 841 cpu_hotplug_done();
cff7d378
TG
842 /* This post dead nonsense must die */
843 if (!ret && hasdied)
090e77c3 844 cpu_notify_nofail(CPU_POST_DEAD, cpu);
cff7d378 845 return ret;
e3920fb4
RW
846}
847
af1f4045 848static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
e3920fb4 849{
9ea09af3 850 int err;
e3920fb4 851
d221938c 852 cpu_maps_update_begin();
e761b772
MK
853
854 if (cpu_hotplug_disabled) {
e3920fb4 855 err = -EBUSY;
e761b772
MK
856 goto out;
857 }
858
af1f4045 859 err = _cpu_down(cpu, 0, target);
e3920fb4 860
e761b772 861out:
d221938c 862 cpu_maps_update_done();
1da177e4
LT
863 return err;
864}
af1f4045
TG
865int cpu_down(unsigned int cpu)
866{
867 return do_cpu_down(cpu, CPUHP_OFFLINE);
868}
b62b8ef9 869EXPORT_SYMBOL(cpu_down);
1da177e4
LT
870#endif /*CONFIG_HOTPLUG_CPU*/
871
4baa0afc
TG
872/**
873 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
874 * @cpu: cpu that just started
875 *
876 * This function calls the cpu_chain notifiers with CPU_STARTING.
877 * It must be called by the arch code on the new cpu, before the new cpu
878 * enables interrupts and before the "boot" cpu returns from __cpu_up().
879 */
880void notify_cpu_starting(unsigned int cpu)
881{
882 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
883 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
884
885 while (st->state < target) {
886 struct cpuhp_step *step;
887
888 st->state++;
889 step = cpuhp_ap_states + st->state;
890 cpuhp_invoke_callback(cpu, st->state, step->startup);
891 }
892}
893
949338e3
TG
894/*
895 * Called from the idle task. We need to set active here, so we can kick off
8df3e07e
TG
896 * the stopper thread and unpark the smpboot threads. If the target state is
897 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
898 * cpu further.
949338e3 899 */
8df3e07e 900void cpuhp_online_idle(enum cpuhp_state state)
949338e3 901{
8df3e07e
TG
902 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
903 unsigned int cpu = smp_processor_id();
904
905 /* Happens for the boot cpu */
906 if (state != CPUHP_AP_ONLINE_IDLE)
907 return;
908
909 st->state = CPUHP_AP_ONLINE_IDLE;
1cf4f629 910
8df3e07e 911 /* Unpark the stopper thread and the hotplug thread of this cpu */
949338e3 912 stop_machine_unpark(cpu);
1cf4f629 913 kthread_unpark(st->thread);
8df3e07e
TG
914
915 /* Should we go further up ? */
916 if (st->target > CPUHP_AP_ONLINE_IDLE)
917 __cpuhp_kick_ap_work(st);
918 else
919 complete(&st->done);
949338e3
TG
920}
921
e3920fb4 922/* Requires cpu_add_remove_lock to be held */
af1f4045 923static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1da177e4 924{
cff7d378 925 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
3bb5d2ee 926 struct task_struct *idle;
2e1a3483 927 int ret = 0;
1da177e4 928
d221938c 929 cpu_hotplug_begin();
38498a67 930
757c989b 931 if (!cpu_present(cpu)) {
5e5041f3
YI
932 ret = -EINVAL;
933 goto out;
934 }
935
757c989b
TG
936 /*
937 * The caller of do_cpu_up might have raced with another
938 * caller. Ignore it for now.
939 */
940 if (st->state >= target)
38498a67 941 goto out;
757c989b
TG
942
943 if (st->state == CPUHP_OFFLINE) {
944 /* Let it fail before we try to bring the cpu up */
945 idle = idle_thread_get(cpu);
946 if (IS_ERR(idle)) {
947 ret = PTR_ERR(idle);
948 goto out;
949 }
3bb5d2ee 950 }
38498a67 951
ba997462
TG
952 cpuhp_tasks_frozen = tasks_frozen;
953
af1f4045 954 st->target = target;
1cf4f629
TG
955 /*
956 * If the current CPU state is in the range of the AP hotplug thread,
957 * then we need to kick the thread once more.
958 */
8df3e07e 959 if (st->state > CPUHP_BRINGUP_CPU) {
1cf4f629
TG
960 ret = cpuhp_kick_ap_work(cpu);
961 /*
962 * The AP side has done the error rollback already. Just
963 * return the error code..
964 */
965 if (ret)
966 goto out;
967 }
968
969 /*
970 * Try to reach the target state. We max out on the BP at
8df3e07e 971 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1cf4f629
TG
972 * responsible for bringing it up to the target state.
973 */
8df3e07e 974 target = min((int)target, CPUHP_BRINGUP_CPU);
2e1a3483 975 ret = cpuhp_up_callbacks(cpu, st, cpuhp_bp_states, target);
38498a67 976out:
d221938c 977 cpu_hotplug_done();
e3920fb4
RW
978 return ret;
979}
980
af1f4045 981static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
e3920fb4
RW
982{
983 int err = 0;
cf23422b 984
e0b582ec 985 if (!cpu_possible(cpu)) {
84117da5
FF
986 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
987 cpu);
87d5e023 988#if defined(CONFIG_IA64)
84117da5 989 pr_err("please check additional_cpus= boot parameter\n");
73e753a5
KH
990#endif
991 return -EINVAL;
992 }
e3920fb4 993
01b0f197
TK
994 err = try_online_node(cpu_to_node(cpu));
995 if (err)
996 return err;
cf23422b 997
d221938c 998 cpu_maps_update_begin();
e761b772
MK
999
1000 if (cpu_hotplug_disabled) {
e3920fb4 1001 err = -EBUSY;
e761b772
MK
1002 goto out;
1003 }
1004
af1f4045 1005 err = _cpu_up(cpu, 0, target);
e761b772 1006out:
d221938c 1007 cpu_maps_update_done();
e3920fb4
RW
1008 return err;
1009}
af1f4045
TG
1010
1011int cpu_up(unsigned int cpu)
1012{
1013 return do_cpu_up(cpu, CPUHP_ONLINE);
1014}
a513f6ba 1015EXPORT_SYMBOL_GPL(cpu_up);
e3920fb4 1016
f3de4be9 1017#ifdef CONFIG_PM_SLEEP_SMP
e0b582ec 1018static cpumask_var_t frozen_cpus;
e3920fb4
RW
1019
1020int disable_nonboot_cpus(void)
1021{
e9a5f426 1022 int cpu, first_cpu, error = 0;
e3920fb4 1023
d221938c 1024 cpu_maps_update_begin();
e0b582ec 1025 first_cpu = cpumask_first(cpu_online_mask);
9ee349ad
XF
1026 /*
1027 * We take down all of the non-boot CPUs in one shot to avoid races
e3920fb4
RW
1028 * with the userspace trying to use the CPU hotplug at the same time
1029 */
e0b582ec 1030 cpumask_clear(frozen_cpus);
6ad4c188 1031
84117da5 1032 pr_info("Disabling non-boot CPUs ...\n");
e3920fb4
RW
1033 for_each_online_cpu(cpu) {
1034 if (cpu == first_cpu)
1035 continue;
bb3632c6 1036 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
af1f4045 1037 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
bb3632c6 1038 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
feae3203 1039 if (!error)
e0b582ec 1040 cpumask_set_cpu(cpu, frozen_cpus);
feae3203 1041 else {
84117da5 1042 pr_err("Error taking CPU%d down: %d\n", cpu, error);
e3920fb4
RW
1043 break;
1044 }
1045 }
86886e55 1046
89af7ba5 1047 if (!error)
e3920fb4 1048 BUG_ON(num_online_cpus() > 1);
89af7ba5 1049 else
84117da5 1050 pr_err("Non-boot CPUs are not disabled\n");
89af7ba5
VK
1051
1052 /*
1053 * Make sure the CPUs won't be enabled by someone else. We need to do
1054 * this even in case of failure as all disable_nonboot_cpus() users are
1055 * supposed to do enable_nonboot_cpus() on the failure path.
1056 */
1057 cpu_hotplug_disabled++;
1058
d221938c 1059 cpu_maps_update_done();
e3920fb4
RW
1060 return error;
1061}
1062
d0af9eed
SS
1063void __weak arch_enable_nonboot_cpus_begin(void)
1064{
1065}
1066
1067void __weak arch_enable_nonboot_cpus_end(void)
1068{
1069}
1070
71cf5aee 1071void enable_nonboot_cpus(void)
e3920fb4
RW
1072{
1073 int cpu, error;
1074
1075 /* Allow everyone to use the CPU hotplug again */
d221938c 1076 cpu_maps_update_begin();
89af7ba5 1077 WARN_ON(--cpu_hotplug_disabled < 0);
e0b582ec 1078 if (cpumask_empty(frozen_cpus))
1d64b9cb 1079 goto out;
e3920fb4 1080
84117da5 1081 pr_info("Enabling non-boot CPUs ...\n");
d0af9eed
SS
1082
1083 arch_enable_nonboot_cpus_begin();
1084
e0b582ec 1085 for_each_cpu(cpu, frozen_cpus) {
bb3632c6 1086 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
af1f4045 1087 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
bb3632c6 1088 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
e3920fb4 1089 if (!error) {
84117da5 1090 pr_info("CPU%d is up\n", cpu);
e3920fb4
RW
1091 continue;
1092 }
84117da5 1093 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
e3920fb4 1094 }
d0af9eed
SS
1095
1096 arch_enable_nonboot_cpus_end();
1097
e0b582ec 1098 cpumask_clear(frozen_cpus);
1d64b9cb 1099out:
d221938c 1100 cpu_maps_update_done();
1da177e4 1101}
e0b582ec 1102
d7268a31 1103static int __init alloc_frozen_cpus(void)
e0b582ec
RR
1104{
1105 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1106 return -ENOMEM;
1107 return 0;
1108}
1109core_initcall(alloc_frozen_cpus);
79cfbdfa 1110
79cfbdfa
SB
1111/*
1112 * When callbacks for CPU hotplug notifications are being executed, we must
1113 * ensure that the state of the system with respect to the tasks being frozen
1114 * or not, as reported by the notification, remains unchanged *throughout the
1115 * duration* of the execution of the callbacks.
1116 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1117 *
1118 * This synchronization is implemented by mutually excluding regular CPU
1119 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1120 * Hibernate notifications.
1121 */
1122static int
1123cpu_hotplug_pm_callback(struct notifier_block *nb,
1124 unsigned long action, void *ptr)
1125{
1126 switch (action) {
1127
1128 case PM_SUSPEND_PREPARE:
1129 case PM_HIBERNATION_PREPARE:
16e53dbf 1130 cpu_hotplug_disable();
79cfbdfa
SB
1131 break;
1132
1133 case PM_POST_SUSPEND:
1134 case PM_POST_HIBERNATION:
16e53dbf 1135 cpu_hotplug_enable();
79cfbdfa
SB
1136 break;
1137
1138 default:
1139 return NOTIFY_DONE;
1140 }
1141
1142 return NOTIFY_OK;
1143}
1144
1145
d7268a31 1146static int __init cpu_hotplug_pm_sync_init(void)
79cfbdfa 1147{
6e32d479
FY
1148 /*
1149 * cpu_hotplug_pm_callback has higher priority than x86
1150 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1151 * to disable cpu hotplug to avoid cpu hotplug race.
1152 */
79cfbdfa
SB
1153 pm_notifier(cpu_hotplug_pm_callback, 0);
1154 return 0;
1155}
1156core_initcall(cpu_hotplug_pm_sync_init);
1157
f3de4be9 1158#endif /* CONFIG_PM_SLEEP_SMP */
68f4f1ec
MK
1159
1160#endif /* CONFIG_SMP */
b8d317d1 1161
cff7d378
TG
1162/* Boot processor state steps */
1163static struct cpuhp_step cpuhp_bp_states[] = {
1164 [CPUHP_OFFLINE] = {
1165 .name = "offline",
1166 .startup = NULL,
1167 .teardown = NULL,
1168 },
1169#ifdef CONFIG_SMP
1170 [CPUHP_CREATE_THREADS]= {
1171 .name = "threads:create",
1172 .startup = smpboot_create_threads,
1173 .teardown = NULL,
757c989b 1174 .cant_stop = true,
cff7d378 1175 },
d10ef6f9
TG
1176 /*
1177 * Preparatory and dead notifiers. Will be replaced once the notifiers
1178 * are converted to states.
1179 */
cff7d378
TG
1180 [CPUHP_NOTIFY_PREPARE] = {
1181 .name = "notify:prepare",
1182 .startup = notify_prepare,
1183 .teardown = notify_dead,
1184 .skip_onerr = true,
757c989b 1185 .cant_stop = true,
cff7d378 1186 },
d10ef6f9 1187 /* Kicks the plugged cpu into life */
cff7d378
TG
1188 [CPUHP_BRINGUP_CPU] = {
1189 .name = "cpu:bringup",
1190 .startup = bringup_cpu,
4baa0afc 1191 .teardown = NULL,
757c989b 1192 .cant_stop = true,
4baa0afc 1193 },
d10ef6f9
TG
1194 /*
1195 * Handled on controll processor until the plugged processor manages
1196 * this itself.
1197 */
4baa0afc
TG
1198 [CPUHP_TEARDOWN_CPU] = {
1199 .name = "cpu:teardown",
1200 .startup = NULL,
cff7d378 1201 .teardown = takedown_cpu,
757c989b 1202 .cant_stop = true,
cff7d378 1203 },
cff7d378 1204#endif
cff7d378
TG
1205};
1206
4baa0afc
TG
1207/* Application processor state steps */
1208static struct cpuhp_step cpuhp_ap_states[] = {
1209#ifdef CONFIG_SMP
d10ef6f9
TG
1210 /* Final state before CPU kills itself */
1211 [CPUHP_AP_IDLE_DEAD] = {
1212 .name = "idle:dead",
1213 },
1214 /*
1215 * Last state before CPU enters the idle loop to die. Transient state
1216 * for synchronization.
1217 */
1218 [CPUHP_AP_OFFLINE] = {
1219 .name = "ap:offline",
1220 .cant_stop = true,
1221 },
9cf7243d
TG
1222 /* First state is scheduler control. Interrupts are disabled */
1223 [CPUHP_AP_SCHED_STARTING] = {
1224 .name = "sched:starting",
1225 .startup = sched_cpu_starting,
f2785ddb 1226 .teardown = sched_cpu_dying,
9cf7243d 1227 },
d10ef6f9
TG
1228 /*
1229 * Low level startup/teardown notifiers. Run with interrupts
1230 * disabled. Will be removed once the notifiers are converted to
1231 * states.
1232 */
4baa0afc
TG
1233 [CPUHP_AP_NOTIFY_STARTING] = {
1234 .name = "notify:starting",
1235 .startup = notify_starting,
1236 .teardown = notify_dying,
1237 .skip_onerr = true,
757c989b 1238 .cant_stop = true,
4baa0afc 1239 },
d10ef6f9
TG
1240 /* Entry state on starting. Interrupts enabled from here on. Transient
1241 * state for synchronsization */
1242 [CPUHP_AP_ONLINE] = {
1243 .name = "ap:online",
1244 },
1245 /* Handle smpboot threads park/unpark */
1cf4f629
TG
1246 [CPUHP_AP_SMPBOOT_THREADS] = {
1247 .name = "smpboot:threads",
1248 .startup = smpboot_unpark_threads,
2a58c527 1249 .teardown = NULL,
1cf4f629 1250 },
d10ef6f9
TG
1251 /*
1252 * Online/down_prepare notifiers. Will be removed once the notifiers
1253 * are converted to states.
1254 */
1cf4f629
TG
1255 [CPUHP_AP_NOTIFY_ONLINE] = {
1256 .name = "notify:online",
1257 .startup = notify_online,
1258 .teardown = notify_down_prepare,
3b9d6da6 1259 .skip_onerr = true,
1cf4f629 1260 },
4baa0afc 1261#endif
d10ef6f9
TG
1262 /*
1263 * The dynamically registered state space is here
1264 */
1265
aaddd7d1
TG
1266#ifdef CONFIG_SMP
1267 /* Last state is scheduler control setting the cpu active */
1268 [CPUHP_AP_ACTIVE] = {
1269 .name = "sched:active",
1270 .startup = sched_cpu_activate,
1271 .teardown = sched_cpu_deactivate,
1272 },
1273#endif
1274
d10ef6f9 1275 /* CPU is fully up and running. */
4baa0afc
TG
1276 [CPUHP_ONLINE] = {
1277 .name = "online",
1278 .startup = NULL,
1279 .teardown = NULL,
1280 },
1281};
1282
5b7aa87e
TG
1283/* Sanity check for callbacks */
1284static int cpuhp_cb_check(enum cpuhp_state state)
1285{
1286 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1287 return -EINVAL;
1288 return 0;
1289}
1290
98f8cdce
TG
1291static bool cpuhp_is_ap_state(enum cpuhp_state state)
1292{
d10ef6f9
TG
1293 /*
1294 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
1295 * purposes as that state is handled explicitely in cpu_down.
1296 */
1297 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
98f8cdce
TG
1298}
1299
1300static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
1301{
1302 struct cpuhp_step *sp;
1303
1304 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
1305 return sp + state;
1306}
1307
5b7aa87e
TG
1308static void cpuhp_store_callbacks(enum cpuhp_state state,
1309 const char *name,
1310 int (*startup)(unsigned int cpu),
1311 int (*teardown)(unsigned int cpu))
1312{
1313 /* (Un)Install the callbacks for further cpu hotplug operations */
1314 struct cpuhp_step *sp;
1315
1316 mutex_lock(&cpuhp_state_mutex);
1317 sp = cpuhp_get_step(state);
1318 sp->startup = startup;
1319 sp->teardown = teardown;
1320 sp->name = name;
1321 mutex_unlock(&cpuhp_state_mutex);
1322}
1323
1324static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1325{
1326 return cpuhp_get_step(state)->teardown;
1327}
1328
5b7aa87e
TG
1329/*
1330 * Call the startup/teardown function for a step either on the AP or
1331 * on the current CPU.
1332 */
1333static int cpuhp_issue_call(int cpu, enum cpuhp_state state,
1334 int (*cb)(unsigned int), bool bringup)
1335{
1336 int ret;
1337
1338 if (!cb)
1339 return 0;
5b7aa87e
TG
1340 /*
1341 * The non AP bound callbacks can fail on bringup. On teardown
1342 * e.g. module removal we crash for now.
1343 */
1cf4f629
TG
1344#ifdef CONFIG_SMP
1345 if (cpuhp_is_ap_state(state))
1346 ret = cpuhp_invoke_ap_callback(cpu, state, cb);
1347 else
1348 ret = cpuhp_invoke_callback(cpu, state, cb);
1349#else
1350 ret = cpuhp_invoke_callback(cpu, state, cb);
1351#endif
5b7aa87e
TG
1352 BUG_ON(ret && !bringup);
1353 return ret;
1354}
1355
1356/*
1357 * Called from __cpuhp_setup_state on a recoverable failure.
1358 *
1359 * Note: The teardown callbacks for rollback are not allowed to fail!
1360 */
1361static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1362 int (*teardown)(unsigned int cpu))
1363{
1364 int cpu;
1365
1366 if (!teardown)
1367 return;
1368
1369 /* Roll back the already executed steps on the other cpus */
1370 for_each_present_cpu(cpu) {
1371 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1372 int cpustate = st->state;
1373
1374 if (cpu >= failedcpu)
1375 break;
1376
1377 /* Did we invoke the startup call on that cpu ? */
1378 if (cpustate >= state)
1379 cpuhp_issue_call(cpu, state, teardown, false);
1380 }
1381}
1382
1383/*
1384 * Returns a free for dynamic slot assignment of the Online state. The states
1385 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1386 * by having no name assigned.
1387 */
1388static int cpuhp_reserve_state(enum cpuhp_state state)
1389{
1390 enum cpuhp_state i;
1391
1392 mutex_lock(&cpuhp_state_mutex);
1cf4f629
TG
1393 for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1394 if (cpuhp_ap_states[i].name)
5b7aa87e
TG
1395 continue;
1396
1cf4f629 1397 cpuhp_ap_states[i].name = "Reserved";
5b7aa87e
TG
1398 mutex_unlock(&cpuhp_state_mutex);
1399 return i;
1400 }
1401 mutex_unlock(&cpuhp_state_mutex);
1402 WARN(1, "No more dynamic states available for CPU hotplug\n");
1403 return -ENOSPC;
1404}
1405
1406/**
1407 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1408 * @state: The state to setup
1409 * @invoke: If true, the startup function is invoked for cpus where
1410 * cpu state >= @state
1411 * @startup: startup callback function
1412 * @teardown: teardown callback function
1413 *
1414 * Returns 0 if successful, otherwise a proper error code
1415 */
1416int __cpuhp_setup_state(enum cpuhp_state state,
1417 const char *name, bool invoke,
1418 int (*startup)(unsigned int cpu),
1419 int (*teardown)(unsigned int cpu))
1420{
1421 int cpu, ret = 0;
1422 int dyn_state = 0;
1423
1424 if (cpuhp_cb_check(state) || !name)
1425 return -EINVAL;
1426
1427 get_online_cpus();
1428
1429 /* currently assignments for the ONLINE state are possible */
1cf4f629 1430 if (state == CPUHP_AP_ONLINE_DYN) {
5b7aa87e
TG
1431 dyn_state = 1;
1432 ret = cpuhp_reserve_state(state);
1433 if (ret < 0)
1434 goto out;
1435 state = ret;
1436 }
1437
1438 cpuhp_store_callbacks(state, name, startup, teardown);
1439
1440 if (!invoke || !startup)
1441 goto out;
1442
1443 /*
1444 * Try to call the startup callback for each present cpu
1445 * depending on the hotplug state of the cpu.
1446 */
1447 for_each_present_cpu(cpu) {
1448 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1449 int cpustate = st->state;
1450
1451 if (cpustate < state)
1452 continue;
1453
1454 ret = cpuhp_issue_call(cpu, state, startup, true);
1455 if (ret) {
1456 cpuhp_rollback_install(cpu, state, teardown);
1457 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1458 goto out;
1459 }
1460 }
1461out:
1462 put_online_cpus();
1463 if (!ret && dyn_state)
1464 return state;
1465 return ret;
1466}
1467EXPORT_SYMBOL(__cpuhp_setup_state);
1468
1469/**
1470 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1471 * @state: The state to remove
1472 * @invoke: If true, the teardown function is invoked for cpus where
1473 * cpu state >= @state
1474 *
1475 * The teardown callback is currently not allowed to fail. Think
1476 * about module removal!
1477 */
1478void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1479{
1480 int (*teardown)(unsigned int cpu) = cpuhp_get_teardown_cb(state);
1481 int cpu;
1482
1483 BUG_ON(cpuhp_cb_check(state));
1484
1485 get_online_cpus();
1486
1487 if (!invoke || !teardown)
1488 goto remove;
1489
1490 /*
1491 * Call the teardown callback for each present cpu depending
1492 * on the hotplug state of the cpu. This function is not
1493 * allowed to fail currently!
1494 */
1495 for_each_present_cpu(cpu) {
1496 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1497 int cpustate = st->state;
1498
1499 if (cpustate >= state)
1500 cpuhp_issue_call(cpu, state, teardown, false);
1501 }
1502remove:
1503 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1504 put_online_cpus();
1505}
1506EXPORT_SYMBOL(__cpuhp_remove_state);
1507
98f8cdce
TG
1508#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1509static ssize_t show_cpuhp_state(struct device *dev,
1510 struct device_attribute *attr, char *buf)
1511{
1512 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1513
1514 return sprintf(buf, "%d\n", st->state);
1515}
1516static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1517
757c989b
TG
1518static ssize_t write_cpuhp_target(struct device *dev,
1519 struct device_attribute *attr,
1520 const char *buf, size_t count)
1521{
1522 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1523 struct cpuhp_step *sp;
1524 int target, ret;
1525
1526 ret = kstrtoint(buf, 10, &target);
1527 if (ret)
1528 return ret;
1529
1530#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1531 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1532 return -EINVAL;
1533#else
1534 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1535 return -EINVAL;
1536#endif
1537
1538 ret = lock_device_hotplug_sysfs();
1539 if (ret)
1540 return ret;
1541
1542 mutex_lock(&cpuhp_state_mutex);
1543 sp = cpuhp_get_step(target);
1544 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1545 mutex_unlock(&cpuhp_state_mutex);
1546 if (ret)
1547 return ret;
1548
1549 if (st->state < target)
1550 ret = do_cpu_up(dev->id, target);
1551 else
1552 ret = do_cpu_down(dev->id, target);
1553
1554 unlock_device_hotplug();
1555 return ret ? ret : count;
1556}
1557
98f8cdce
TG
1558static ssize_t show_cpuhp_target(struct device *dev,
1559 struct device_attribute *attr, char *buf)
1560{
1561 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1562
1563 return sprintf(buf, "%d\n", st->target);
1564}
757c989b 1565static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
98f8cdce
TG
1566
1567static struct attribute *cpuhp_cpu_attrs[] = {
1568 &dev_attr_state.attr,
1569 &dev_attr_target.attr,
1570 NULL
1571};
1572
1573static struct attribute_group cpuhp_cpu_attr_group = {
1574 .attrs = cpuhp_cpu_attrs,
1575 .name = "hotplug",
1576 NULL
1577};
1578
1579static ssize_t show_cpuhp_states(struct device *dev,
1580 struct device_attribute *attr, char *buf)
1581{
1582 ssize_t cur, res = 0;
1583 int i;
1584
1585 mutex_lock(&cpuhp_state_mutex);
757c989b 1586 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
98f8cdce
TG
1587 struct cpuhp_step *sp = cpuhp_get_step(i);
1588
1589 if (sp->name) {
1590 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1591 buf += cur;
1592 res += cur;
1593 }
1594 }
1595 mutex_unlock(&cpuhp_state_mutex);
1596 return res;
1597}
1598static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1599
1600static struct attribute *cpuhp_cpu_root_attrs[] = {
1601 &dev_attr_states.attr,
1602 NULL
1603};
1604
1605static struct attribute_group cpuhp_cpu_root_attr_group = {
1606 .attrs = cpuhp_cpu_root_attrs,
1607 .name = "hotplug",
1608 NULL
1609};
1610
1611static int __init cpuhp_sysfs_init(void)
1612{
1613 int cpu, ret;
1614
1615 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1616 &cpuhp_cpu_root_attr_group);
1617 if (ret)
1618 return ret;
1619
1620 for_each_possible_cpu(cpu) {
1621 struct device *dev = get_cpu_device(cpu);
1622
1623 if (!dev)
1624 continue;
1625 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1626 if (ret)
1627 return ret;
1628 }
1629 return 0;
1630}
1631device_initcall(cpuhp_sysfs_init);
1632#endif
1633
e56b3bc7
LT
1634/*
1635 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1636 * represents all NR_CPUS bits binary values of 1<<nr.
1637 *
e0b582ec 1638 * It is used by cpumask_of() to get a constant address to a CPU
e56b3bc7
LT
1639 * mask value that has a single bit set only.
1640 */
b8d317d1 1641
e56b3bc7 1642/* cpu_bit_bitmap[0] is empty - so we can back into it */
4d51985e 1643#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
e56b3bc7
LT
1644#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1645#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1646#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
b8d317d1 1647
e56b3bc7
LT
1648const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1649
1650 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1651 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1652#if BITS_PER_LONG > 32
1653 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1654 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
b8d317d1
MT
1655#endif
1656};
e56b3bc7 1657EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2d3854a3
RR
1658
1659const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1660EXPORT_SYMBOL(cpu_all_bits);
b3199c02
RR
1661
1662#ifdef CONFIG_INIT_ALL_POSSIBLE
4b804c85 1663struct cpumask __cpu_possible_mask __read_mostly
c4c54dd1 1664 = {CPU_BITS_ALL};
b3199c02 1665#else
4b804c85 1666struct cpumask __cpu_possible_mask __read_mostly;
b3199c02 1667#endif
4b804c85 1668EXPORT_SYMBOL(__cpu_possible_mask);
b3199c02 1669
4b804c85
RV
1670struct cpumask __cpu_online_mask __read_mostly;
1671EXPORT_SYMBOL(__cpu_online_mask);
b3199c02 1672
4b804c85
RV
1673struct cpumask __cpu_present_mask __read_mostly;
1674EXPORT_SYMBOL(__cpu_present_mask);
b3199c02 1675
4b804c85
RV
1676struct cpumask __cpu_active_mask __read_mostly;
1677EXPORT_SYMBOL(__cpu_active_mask);
3fa41520 1678
3fa41520
RR
1679void init_cpu_present(const struct cpumask *src)
1680{
c4c54dd1 1681 cpumask_copy(&__cpu_present_mask, src);
3fa41520
RR
1682}
1683
1684void init_cpu_possible(const struct cpumask *src)
1685{
c4c54dd1 1686 cpumask_copy(&__cpu_possible_mask, src);
3fa41520
RR
1687}
1688
1689void init_cpu_online(const struct cpumask *src)
1690{
c4c54dd1 1691 cpumask_copy(&__cpu_online_mask, src);
3fa41520 1692}
cff7d378
TG
1693
1694/*
1695 * Activate the first processor.
1696 */
1697void __init boot_cpu_init(void)
1698{
1699 int cpu = smp_processor_id();
1700
1701 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1702 set_cpu_online(cpu, true);
1703 set_cpu_active(cpu, true);
1704 set_cpu_present(cpu, true);
1705 set_cpu_possible(cpu, true);
1706}
1707
1708/*
1709 * Must be called _AFTER_ setting up the per_cpu areas
1710 */
1711void __init boot_cpu_state_init(void)
1712{
1713 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;
1714}