blob: 93e9d89bb0ab80c9c71a8b9c4842ec425244f6c0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* 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>
Anton Vorontsovcb792952012-05-31 16:26:22 -070013#include <linux/oom.h>
14#include <linux/rcupdate.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040015#include <linux/export.h>
Anton Vorontsove4cc2f82012-05-31 16:26:26 -070016#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kthread.h>
18#include <linux/stop_machine.h>
Ingo Molnar81615b62006-06-26 00:24:32 -070019#include <linux/mutex.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/gfp.h>
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +010021#include <linux/suspend.h>
Gautham R. Shenoya19423b2014-03-11 02:04:03 +053022#include <linux/lockdep.h>
Preeti U Murthy345527b2015-03-30 14:59:19 +053023#include <linux/tick.h>
Thomas Gleixnera8994182015-07-05 17:12:30 +000024#include <linux/irq.h>
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000025#include <linux/smpboot.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000026
Todd E Brandtbb3632c2014-06-06 05:40:17 -070027#include <trace/events/power.h>
Thomas Gleixnercff7d372016-02-26 18:43:28 +000028#define CREATE_TRACE_POINTS
29#include <trace/events/cpuhp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Thomas Gleixner38498a62012-04-20 13:05:44 +000031#include "smpboot.h"
32
Thomas Gleixnercff7d372016-02-26 18:43:28 +000033/**
34 * cpuhp_cpu_state - Per cpu hotplug state storage
35 * @state: The current cpu state
36 * @target: The target state
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000037 * @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
Thomas Gleixnercff7d372016-02-26 18:43:28 +000043 */
44struct cpuhp_cpu_state {
45 enum cpuhp_state state;
46 enum cpuhp_state target;
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +000047#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
Thomas Gleixnercff7d372016-02-26 18:43:28 +000055};
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
Thomas Gleixner757c9892016-02-26 18:43:32 +000066 * @cant_stop: Bringup/teardown can't be stopped at this step
Thomas Gleixnercff7d372016-02-26 18:43:28 +000067 */
68struct cpuhp_step {
69 const char *name;
70 int (*startup)(unsigned int cpu);
71 int (*teardown)(unsigned int cpu);
72 bool skip_onerr;
Thomas Gleixner757c9892016-02-26 18:43:32 +000073 bool cant_stop;
Thomas Gleixnercff7d372016-02-26 18:43:28 +000074};
75
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +000076static DEFINE_MUTEX(cpuhp_state_mutex);
Thomas Gleixnercff7d372016-02-26 18:43:28 +000077static struct cpuhp_step cpuhp_bp_states[];
Thomas Gleixner4baa0af2016-02-26 18:43:29 +000078static struct cpuhp_step cpuhp_ap_states[];
Thomas Gleixnercff7d372016-02-26 18:43:28 +000079
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
Rusty Russell98a79d62008-12-13 21:19:41 +1030102#ifdef CONFIG_SMP
Rusty Russellb3199c02008-12-30 09:05:14 +1030103/* Serializes the updates to cpu_online_mask, cpu_present_mask */
Linus Torvaldsaa953872006-07-23 12:12:16 -0700104static DEFINE_MUTEX(cpu_add_remove_lock);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000105bool cpuhp_tasks_frozen;
106EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700108/*
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530109 * 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().
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700114 */
115void cpu_maps_update_begin(void)
116{
117 mutex_lock(&cpu_add_remove_lock);
118}
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530119EXPORT_SYMBOL(cpu_notifier_register_begin);
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700120
121void cpu_maps_update_done(void)
122{
123 mutex_unlock(&cpu_add_remove_lock);
124}
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530125EXPORT_SYMBOL(cpu_notifier_register_done);
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700126
Daniel J Blueman5c113fb2010-06-01 12:15:11 +0100127static RAW_NOTIFIER_HEAD(cpu_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700129/* 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
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700134#ifdef CONFIG_HOTPLUG_CPU
135
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100136static struct {
137 struct task_struct *active_writer;
David Hildenbrand87af9e72014-12-12 10:11:44 +0100138 /* 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;
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100142 /*
143 * Also blocks the new readers during
144 * an ongoing cpu hotplug operation.
145 */
David Hildenbrand87af9e72014-12-12 10:11:44 +0100146 atomic_t refcount;
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530147
148#ifdef CONFIG_DEBUG_LOCK_ALLOC
149 struct lockdep_map dep_map;
150#endif
Linus Torvalds31950eb2009-06-22 21:18:12 -0700151} cpu_hotplug = {
152 .active_writer = NULL,
David Hildenbrand87af9e72014-12-12 10:11:44 +0100153 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
Linus Torvalds31950eb2009-06-22 21:18:12 -0700154 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530155#ifdef CONFIG_DEBUG_LOCK_ALLOC
156 .dep_map = {.name = "cpu_hotplug.lock" },
157#endif
Linus Torvalds31950eb2009-06-22 21:18:12 -0700158};
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100159
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530160/* 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)
Paul E. McKenneydd56af42014-08-25 20:25:06 -0700162#define cpuhp_lock_acquire_tryread() \
163 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530164#define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
165#define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
166
Paul E. McKenney62db99f2014-10-22 14:51:49 -0700167
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100168void get_online_cpus(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800169{
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100170 might_sleep();
171 if (cpu_hotplug.active_writer == current)
Linus Torvaldsaa953872006-07-23 12:12:16 -0700172 return;
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530173 cpuhp_lock_acquire_read();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100174 mutex_lock(&cpu_hotplug.lock);
David Hildenbrand87af9e72014-12-12 10:11:44 +0100175 atomic_inc(&cpu_hotplug.refcount);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100176 mutex_unlock(&cpu_hotplug.lock);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800177}
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100178EXPORT_SYMBOL_GPL(get_online_cpus);
Ashok Raj90d45d12005-11-08 21:34:24 -0800179
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100180void put_online_cpus(void)
Ashok Raja9d9baa2005-11-28 13:43:46 -0800181{
David Hildenbrand87af9e72014-12-12 10:11:44 +0100182 int refcount;
183
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100184 if (cpu_hotplug.active_writer == current)
Linus Torvaldsaa953872006-07-23 12:12:16 -0700185 return;
Srivatsa S. Bhat075663d2012-10-08 16:28:20 -0700186
David Hildenbrand87af9e72014-12-12 10:11:44 +0100187 refcount = atomic_dec_return(&cpu_hotplug.refcount);
188 if (WARN_ON(refcount < 0)) /* try to fix things up */
189 atomic_inc(&cpu_hotplug.refcount);
Srivatsa S. Bhat075663d2012-10-08 16:28:20 -0700190
David Hildenbrand87af9e72014-12-12 10:11:44 +0100191 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
192 wake_up(&cpu_hotplug.wq);
193
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530194 cpuhp_lock_release();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100195
Ashok Raja9d9baa2005-11-28 13:43:46 -0800196}
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100197EXPORT_SYMBOL_GPL(put_online_cpus);
Ashok Raja9d9baa2005-11-28 13:43:46 -0800198
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100199/*
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 *
Oleg Nesterovd2ba7e22008-04-29 01:00:29 -0700206 * Since cpu_hotplug_begin() is always called after invoking
207 * cpu_maps_update_begin(), we can be sure that only one writer is active.
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100208 *
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
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100218 * get_online_cpus() not an api which is called all that often.
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100219 *
220 */
Toshi Kanib9d10be2013-08-12 09:45:53 -0600221void cpu_hotplug_begin(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100222{
David Hildenbrand87af9e72014-12-12 10:11:44 +0100223 DEFINE_WAIT(wait);
Oleg Nesterovd2ba7e22008-04-29 01:00:29 -0700224
David Hildenbrand87af9e72014-12-12 10:11:44 +0100225 cpu_hotplug.active_writer = current;
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530226 cpuhp_lock_acquire();
David Hildenbrand87af9e72014-12-12 10:11:44 +0100227
Oleg Nesterovd2ba7e22008-04-29 01:00:29 -0700228 for (;;) {
229 mutex_lock(&cpu_hotplug.lock);
David Hildenbrand87af9e72014-12-12 10:11:44 +0100230 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
231 if (likely(!atomic_read(&cpu_hotplug.refcount)))
232 break;
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100233 mutex_unlock(&cpu_hotplug.lock);
234 schedule();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100235 }
David Hildenbrand87af9e72014-12-12 10:11:44 +0100236 finish_wait(&cpu_hotplug.wq, &wait);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100237}
238
Toshi Kanib9d10be2013-08-12 09:45:53 -0600239void cpu_hotplug_done(void)
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100240{
241 cpu_hotplug.active_writer = NULL;
242 mutex_unlock(&cpu_hotplug.lock);
Gautham R. Shenoya19423b2014-03-11 02:04:03 +0530243 cpuhp_lock_release();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100244}
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700245
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700246/*
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();
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -0700256 cpu_hotplug_disabled++;
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700257 cpu_maps_update_done();
258}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700259EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700260
261void cpu_hotplug_enable(void)
262{
263 cpu_maps_update_begin();
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -0700264 WARN_ON(--cpu_hotplug_disabled < 0);
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -0700265 cpu_maps_update_done();
266}
Vitaly Kuznetsov32145c42015-08-05 00:52:47 -0700267EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
Toshi Kanib9d10be2013-08-12 09:45:53 -0600268#endif /* CONFIG_HOTPLUG_CPU */
Lai Jiangshan79a6cde2010-05-26 14:43:36 -0700269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270/* Need to know about CPUs going up/down? */
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200271int register_cpu_notifier(struct notifier_block *nb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Neil Brownbd5349c2006-10-17 00:10:35 -0700273 int ret;
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100274 cpu_maps_update_begin();
Neil Brownbd5349c2006-10-17 00:10:35 -0700275 ret = raw_notifier_chain_register(&cpu_chain, nb);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100276 cpu_maps_update_done();
Neil Brownbd5349c2006-10-17 00:10:35 -0700277 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
Chandra Seetharaman65edc682006-06-27 02:54:08 -0700279
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200280int __register_cpu_notifier(struct notifier_block *nb)
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530281{
282 return raw_notifier_chain_register(&cpu_chain, nb);
283}
284
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000285static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
Akinobu Mitae9fb7632010-05-26 14:43:28 -0700286 int *nr_calls)
287{
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000288 unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
289 void *hcpu = (void *)(long)cpu;
290
Akinobu Mitae6bde732010-05-26 14:43:29 -0700291 int ret;
292
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000293 ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
Akinobu Mitae9fb7632010-05-26 14:43:28 -0700294 nr_calls);
Akinobu Mitae6bde732010-05-26 14:43:29 -0700295
296 return notifier_to_errno(ret);
Akinobu Mitae9fb7632010-05-26 14:43:28 -0700297}
298
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000299static int cpu_notify(unsigned long val, unsigned int cpu)
Akinobu Mitae9fb7632010-05-26 14:43:28 -0700300{
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000301 return __cpu_notify(val, cpu, -1, NULL);
Akinobu Mitae9fb7632010-05-26 14:43:28 -0700302}
303
Thomas Gleixnerba997462016-02-26 18:43:24 +0000304/* 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
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000326static int notify_starting(unsigned int cpu)
327{
328 cpu_notify(CPU_STARTING, cpu);
329 return 0;
330}
331
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000332static 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
Thomas Gleixnerba997462016-02-26 18:43:24 +0000340static 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 }
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000351 ret = bringup_wait_for_ap(cpu);
Thomas Gleixnerba997462016-02-26 18:43:24 +0000352 BUG_ON(!cpu_online(cpu));
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000353 return ret;
Thomas Gleixnerba997462016-02-26 18:43:24 +0000354}
355
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000356/*
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
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000421/*
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{
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000441 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000442
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 {
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000481 /* Cannot happen .... */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000482 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000483
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000484 /* 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 */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000517static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000518{
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000519 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);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000528}
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);
Thomas Gleixner4cb28ce2016-02-26 18:43:38 +0000537 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
Linus Torvalds00b9b0a2010-05-27 10:32:08 -0700557#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558EXPORT_SYMBOL(register_cpu_notifier);
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530559EXPORT_SYMBOL(__register_cpu_notifier);
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200560void unregister_cpu_notifier(struct notifier_block *nb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561{
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100562 cpu_maps_update_begin();
Neil Brownbd5349c2006-10-17 00:10:35 -0700563 raw_notifier_chain_unregister(&cpu_chain, nb);
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100564 cpu_maps_update_done();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565}
566EXPORT_SYMBOL(unregister_cpu_notifier);
567
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200568void __unregister_cpu_notifier(struct notifier_block *nb)
Srivatsa S. Bhat93ae4f92014-03-11 02:04:14 +0530569{
570 raw_notifier_chain_unregister(&cpu_chain, nb);
571}
572EXPORT_SYMBOL(__unregister_cpu_notifier);
573
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700574/**
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 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700586void 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 */
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700597 WARN_ON(cpu_online(cpu));
Anton Vorontsovcb792952012-05-31 16:26:22 -0700598 rcu_read_lock();
599 for_each_process(p) {
600 struct task_struct *t;
601
Anton Vorontsove4cc2f82012-05-31 16:26:26 -0700602 /*
603 * Main thread might exit, but other threads may still have
604 * a valid mm. Find one.
605 */
Anton Vorontsovcb792952012-05-31 16:26:22 -0700606 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
Kirill Tkhaib728ca02014-06-25 12:19:55 +0400615static inline void check_for_tasks(int dead_cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Kirill Tkhaib728ca02014-06-25 12:19:55 +0400617 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Oleg Nesterova75a6062015-09-10 15:07:50 +0200619 read_lock(&tasklist_lock);
620 for_each_process_thread(g, p) {
Kirill Tkhaib728ca02014-06-25 12:19:55 +0400621 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);
Oleg Nesterova75a6062015-09-10 15:07:50 +0200635 }
636 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637}
638
Thomas Gleixner98458172016-02-26 18:43:25 +0000639static 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
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000658static int notify_dying(unsigned int cpu)
659{
660 cpu_notify(CPU_DYING, cpu);
661 return 0;
662}
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664/* Take this CPU down. */
Mathias Krause71cf5ae2015-07-19 20:06:22 +0200665static int take_cpu_down(void *_param)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000667 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
668 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000669 int err, cpu = smp_processor_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 /* Ensure this CPU doesn't handle any more interrupts. */
672 err = __cpu_disable();
673 if (err < 0)
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700674 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000676 /* 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 }
Thomas Gleixner52c063d2015-04-03 02:37:24 +0200682 /* Give up timekeeping duties */
683 tick_handover_do_timer();
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000684 /* Park the stopper thread */
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000685 stop_machine_park(cpu);
Zwane Mwaikambof3705132005-06-25 14:54:50 -0700686 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Thomas Gleixner98458172016-02-26 18:43:25 +0000689static int takedown_cpu(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000691 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000692 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200694 /*
695 * By now we've cleared cpu_active_mask, wait for all preempt-disabled
696 * and RCU users of this state to go away such that all new such users
697 * will observe it.
698 *
699 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
Paul E. McKenney779de6c2015-06-10 13:34:41 -0700700 * not imply sync_sched(), so wait for both.
Michael wang106dd5a2013-11-13 11:10:56 +0800701 *
702 * Do sync before park smpboot threads to take care the rcu boost case.
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200703 */
Paul E. McKenney779de6c2015-06-10 13:34:41 -0700704 if (IS_ENABLED(CONFIG_PREEMPT))
705 synchronize_rcu_mult(call_rcu, call_rcu_sched);
706 else
707 synchronize_rcu();
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200708
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000709 /* Park the hotplug thread */
710 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
711
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200712 /*
Thomas Gleixnera8994182015-07-05 17:12:30 +0000713 * Prevent irq alloc/free while the dying cpu reorganizes the
714 * interrupt affinities.
715 */
716 irq_lock_sparse();
717
718 /*
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200719 * So now all preempt/rcu users must observe !cpu_active().
720 */
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000721 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
Rusty Russell04321582008-07-28 12:16:29 -0500722 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 /* CPU didn't die: tell everyone. Can't complain. */
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000724 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
Thomas Gleixnera8994182015-07-05 17:12:30 +0000725 irq_unlock_sparse();
Thomas Gleixner98458172016-02-26 18:43:25 +0000726 return err;
Satoru Takeuchi8fa1d7d2006-10-28 10:38:57 -0700727 }
Rusty Russell04321582008-07-28 12:16:29 -0500728 BUG_ON(cpu_online(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Peter Zijlstra48c5cca2010-11-13 19:32:29 +0100730 /*
731 * The migration_call() CPU_DYING callback will have removed all
732 * runnable tasks from the cpu, there's only the idle task left now
733 * that the migration thread is done doing the stop_machine thing.
Peter Zijlstra51a96c72010-11-19 20:37:53 +0100734 *
735 * Wait for the stop thread to go away.
Peter Zijlstra48c5cca2010-11-13 19:32:29 +0100736 */
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000737 wait_for_completion(&st->done);
738 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Thomas Gleixnera8994182015-07-05 17:12:30 +0000740 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
741 irq_unlock_sparse();
742
Preeti U Murthy345527b2015-03-30 14:59:19 +0530743 hotplug_cpu__broadcast_tick_pull(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /* This actually kills the CPU. */
745 __cpu_die(cpu);
746
Thomas Gleixnera49b1162015-04-03 02:38:05 +0200747 tick_cleanup_dead_cpu(cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000748 return 0;
749}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Thomas Gleixner98458172016-02-26 18:43:25 +0000751static int notify_dead(unsigned int cpu)
752{
753 cpu_notify_nofail(CPU_DEAD, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 check_for_tasks(cpu);
Thomas Gleixner98458172016-02-26 18:43:25 +0000755 return 0;
756}
757
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100758static void cpuhp_complete_idle_dead(void *arg)
759{
760 struct cpuhp_cpu_state *st = arg;
761
762 complete(&st->done);
763}
764
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000765void cpuhp_report_idle_dead(void)
766{
767 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
768
769 BUG_ON(st->state != CPUHP_AP_OFFLINE);
Thomas Gleixner27d50c72016-02-26 18:43:44 +0000770 rcu_report_dead(smp_processor_id());
Thomas Gleixner71f87b22016-03-03 10:52:10 +0100771 st->state = CPUHP_AP_IDLE_DEAD;
772 /*
773 * We cannot call complete after rcu_report_dead() so we delegate it
774 * to an online cpu.
775 */
776 smp_call_function_single(cpumask_first(cpu_online_mask),
777 cpuhp_complete_idle_dead, st, 0);
Thomas Gleixnere69aab12016-02-26 18:43:43 +0000778}
779
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000780#else
781#define notify_down_prepare NULL
782#define takedown_cpu NULL
783#define notify_dead NULL
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000784#define notify_dying NULL
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000785#endif
786
787#ifdef CONFIG_HOTPLUG_CPU
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000788
Thomas Gleixner98458172016-02-26 18:43:25 +0000789/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000790static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
791 enum cpuhp_state target)
Thomas Gleixner98458172016-02-26 18:43:25 +0000792{
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000793 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
794 int prev_state, ret = 0;
795 bool hasdied = false;
Thomas Gleixner98458172016-02-26 18:43:25 +0000796
797 if (num_online_cpus() == 1)
798 return -EBUSY;
799
Thomas Gleixner757c9892016-02-26 18:43:32 +0000800 if (!cpu_present(cpu))
Thomas Gleixner98458172016-02-26 18:43:25 +0000801 return -EINVAL;
802
803 cpu_hotplug_begin();
804
805 cpuhp_tasks_frozen = tasks_frozen;
806
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000807 prev_state = st->state;
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000808 st->target = target;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000809 /*
810 * If the current CPU state is in the range of the AP hotplug thread,
811 * then we need to kick the thread.
812 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000813 if (st->state > CPUHP_TEARDOWN_CPU) {
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000814 ret = cpuhp_kick_ap_work(cpu);
815 /*
816 * The AP side has done the error rollback already. Just
817 * return the error code..
818 */
819 if (ret)
820 goto out;
821
822 /*
823 * We might have stopped still in the range of the AP hotplug
824 * thread. Nothing to do anymore.
825 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000826 if (st->state > CPUHP_TEARDOWN_CPU)
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000827 goto out;
828 }
829 /*
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000830 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000831 * to do the further cleanups.
832 */
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000833 ret = cpuhp_down_callbacks(cpu, st, cpuhp_bp_states, target);
Thomas Gleixner98458172016-02-26 18:43:25 +0000834
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000835 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000836out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100837 cpu_hotplug_done();
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000838 /* This post dead nonsense must die */
839 if (!ret && hasdied)
Thomas Gleixner090e77c2016-02-26 18:43:23 +0000840 cpu_notify_nofail(CPU_POST_DEAD, cpu);
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000841 return ret;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700842}
843
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000844static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700845{
Heiko Carstens9ea09af2008-12-22 12:36:30 +0100846 int err;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700847
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100848 cpu_maps_update_begin();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700849
Max Krasnyanskye761b772008-07-15 04:43:49 -0700850 if (cpu_hotplug_disabled) {
851 err = -EBUSY;
852 goto out;
853 }
854
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000855 err = _cpu_down(cpu, 0, target);
Max Krasnyanskye761b772008-07-15 04:43:49 -0700856
Max Krasnyanskye761b772008-07-15 04:43:49 -0700857out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100858 cpu_maps_update_done();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 return err;
860}
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000861int cpu_down(unsigned int cpu)
862{
863 return do_cpu_down(cpu, CPUHP_OFFLINE);
864}
Zhang Ruib62b8ef2008-04-29 02:35:56 -0400865EXPORT_SYMBOL(cpu_down);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866#endif /*CONFIG_HOTPLUG_CPU*/
867
Thomas Gleixner4baa0af2016-02-26 18:43:29 +0000868/**
869 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
870 * @cpu: cpu that just started
871 *
872 * This function calls the cpu_chain notifiers with CPU_STARTING.
873 * It must be called by the arch code on the new cpu, before the new cpu
874 * enables interrupts and before the "boot" cpu returns from __cpu_up().
875 */
876void notify_cpu_starting(unsigned int cpu)
877{
878 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
879 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
880
881 while (st->state < target) {
882 struct cpuhp_step *step;
883
884 st->state++;
885 step = cpuhp_ap_states + st->state;
886 cpuhp_invoke_callback(cpu, st->state, step->startup);
887 }
888}
889
Thomas Gleixner949338e2016-02-26 18:43:35 +0000890/*
891 * Called from the idle task. We need to set active here, so we can kick off
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000892 * the stopper thread and unpark the smpboot threads. If the target state is
893 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
894 * cpu further.
Thomas Gleixner949338e2016-02-26 18:43:35 +0000895 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000896void cpuhp_online_idle(enum cpuhp_state state)
Thomas Gleixner949338e2016-02-26 18:43:35 +0000897{
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000898 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
899 unsigned int cpu = smp_processor_id();
900
901 /* Happens for the boot cpu */
902 if (state != CPUHP_AP_ONLINE_IDLE)
903 return;
904
905 st->state = CPUHP_AP_ONLINE_IDLE;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000906
Thomas Gleixner949338e2016-02-26 18:43:35 +0000907 /* The cpu is marked online, set it active now */
908 set_cpu_active(cpu, true);
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000909 /* Unpark the stopper thread and the hotplug thread of this cpu */
Thomas Gleixner949338e2016-02-26 18:43:35 +0000910 stop_machine_unpark(cpu);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000911 kthread_unpark(st->thread);
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000912
913 /* Should we go further up ? */
914 if (st->target > CPUHP_AP_ONLINE_IDLE)
915 __cpuhp_kick_ap_work(st);
916 else
917 complete(&st->done);
Thomas Gleixner949338e2016-02-26 18:43:35 +0000918}
919
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700920/* Requires cpu_add_remove_lock to be held */
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000921static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922{
Thomas Gleixnercff7d372016-02-26 18:43:28 +0000923 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -0700924 struct task_struct *idle;
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000925 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100927 cpu_hotplug_begin();
Thomas Gleixner38498a62012-04-20 13:05:44 +0000928
Thomas Gleixner757c9892016-02-26 18:43:32 +0000929 if (!cpu_present(cpu)) {
Yasuaki Ishimatsu5e5041f2012-10-23 01:30:54 +0200930 ret = -EINVAL;
931 goto out;
932 }
933
Thomas Gleixner757c9892016-02-26 18:43:32 +0000934 /*
935 * The caller of do_cpu_up might have raced with another
936 * caller. Ignore it for now.
937 */
938 if (st->state >= target)
Thomas Gleixner38498a62012-04-20 13:05:44 +0000939 goto out;
Thomas Gleixner757c9892016-02-26 18:43:32 +0000940
941 if (st->state == CPUHP_OFFLINE) {
942 /* Let it fail before we try to bring the cpu up */
943 idle = idle_thread_get(cpu);
944 if (IS_ERR(idle)) {
945 ret = PTR_ERR(idle);
946 goto out;
947 }
Suresh Siddha3bb5d2e2012-04-20 17:08:50 -0700948 }
Thomas Gleixner38498a62012-04-20 13:05:44 +0000949
Thomas Gleixnerba997462016-02-26 18:43:24 +0000950 cpuhp_tasks_frozen = tasks_frozen;
951
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000952 st->target = target;
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000953 /*
954 * If the current CPU state is in the range of the AP hotplug thread,
955 * then we need to kick the thread once more.
956 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000957 if (st->state > CPUHP_BRINGUP_CPU) {
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000958 ret = cpuhp_kick_ap_work(cpu);
959 /*
960 * The AP side has done the error rollback already. Just
961 * return the error code..
962 */
963 if (ret)
964 goto out;
965 }
966
967 /*
968 * Try to reach the target state. We max out on the BP at
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000969 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
Thomas Gleixner1cf4f622016-02-26 18:43:39 +0000970 * responsible for bringing it up to the target state.
971 */
Thomas Gleixner8df3e072016-02-26 18:43:41 +0000972 target = min((int)target, CPUHP_BRINGUP_CPU);
Thomas Gleixner2e1a3482016-02-26 18:43:37 +0000973 ret = cpuhp_up_callbacks(cpu, st, cpuhp_bp_states, target);
Thomas Gleixner38498a62012-04-20 13:05:44 +0000974out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100975 cpu_hotplug_done();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return ret;
977}
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700978
Thomas Gleixneraf1f4042016-02-26 18:43:30 +0000979static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700980{
981 int err = 0;
minskey guocf234222010-05-24 14:32:41 -0700982
Rusty Russelle0b582e2009-01-01 10:12:28 +1030983 if (!cpu_possible(cpu)) {
Fabian Frederick84117da2014-06-04 16:11:17 -0700984 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
985 cpu);
Chen Gong87d5e022010-03-05 13:42:38 -0800986#if defined(CONFIG_IA64)
Fabian Frederick84117da2014-06-04 16:11:17 -0700987 pr_err("please check additional_cpus= boot parameter\n");
KAMEZAWA Hiroyuki73e753a2007-10-18 23:40:47 -0700988#endif
989 return -EINVAL;
990 }
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700991
Toshi Kani01b0f192013-11-12 15:07:25 -0800992 err = try_online_node(cpu_to_node(cpu));
993 if (err)
994 return err;
minskey guocf234222010-05-24 14:32:41 -0700995
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100996 cpu_maps_update_begin();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -0700997
Max Krasnyanskye761b772008-07-15 04:43:49 -0700998 if (cpu_hotplug_disabled) {
999 err = -EBUSY;
1000 goto out;
1001 }
1002
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001003 err = _cpu_up(cpu, 0, target);
Max Krasnyanskye761b772008-07-15 04:43:49 -07001004out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001005 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001006 return err;
1007}
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001008
1009int cpu_up(unsigned int cpu)
1010{
1011 return do_cpu_up(cpu, CPUHP_ONLINE);
1012}
Paul E. McKenneya513f6b2011-12-11 21:54:45 -08001013EXPORT_SYMBOL_GPL(cpu_up);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001014
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001015#ifdef CONFIG_PM_SLEEP_SMP
Rusty Russelle0b582e2009-01-01 10:12:28 +10301016static cpumask_var_t frozen_cpus;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001017
1018int disable_nonboot_cpus(void)
1019{
Rafael J. Wysockie9a5f422010-05-27 22:16:22 +02001020 int cpu, first_cpu, error = 0;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001021
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001022 cpu_maps_update_begin();
Rusty Russelle0b582e2009-01-01 10:12:28 +10301023 first_cpu = cpumask_first(cpu_online_mask);
Xiaotian Feng9ee349a2009-12-16 18:04:32 +01001024 /*
1025 * We take down all of the non-boot CPUs in one shot to avoid races
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001026 * with the userspace trying to use the CPU hotplug at the same time
1027 */
Rusty Russelle0b582e2009-01-01 10:12:28 +10301028 cpumask_clear(frozen_cpus);
Peter Zijlstra6ad4c182009-11-25 13:31:39 +01001029
Fabian Frederick84117da2014-06-04 16:11:17 -07001030 pr_info("Disabling non-boot CPUs ...\n");
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001031 for_each_online_cpu(cpu) {
1032 if (cpu == first_cpu)
1033 continue;
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001034 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001035 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001036 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
Mike Travisfeae3202009-11-17 18:22:13 -06001037 if (!error)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301038 cpumask_set_cpu(cpu, frozen_cpus);
Mike Travisfeae3202009-11-17 18:22:13 -06001039 else {
Fabian Frederick84117da2014-06-04 16:11:17 -07001040 pr_err("Error taking CPU%d down: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001041 break;
1042 }
1043 }
Joseph Cihula86886e52009-06-30 19:31:07 -07001044
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001045 if (!error)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001046 BUG_ON(num_online_cpus() > 1);
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001047 else
Fabian Frederick84117da2014-06-04 16:11:17 -07001048 pr_err("Non-boot CPUs are not disabled\n");
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001049
1050 /*
1051 * Make sure the CPUs won't be enabled by someone else. We need to do
1052 * this even in case of failure as all disable_nonboot_cpus() users are
1053 * supposed to do enable_nonboot_cpus() on the failure path.
1054 */
1055 cpu_hotplug_disabled++;
1056
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001057 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001058 return error;
1059}
1060
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001061void __weak arch_enable_nonboot_cpus_begin(void)
1062{
1063}
1064
1065void __weak arch_enable_nonboot_cpus_end(void)
1066{
1067}
1068
Mathias Krause71cf5ae2015-07-19 20:06:22 +02001069void enable_nonboot_cpus(void)
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001070{
1071 int cpu, error;
1072
1073 /* Allow everyone to use the CPU hotplug again */
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001074 cpu_maps_update_begin();
Vitaly Kuznetsov89af7ba2015-08-05 00:52:46 -07001075 WARN_ON(--cpu_hotplug_disabled < 0);
Rusty Russelle0b582e2009-01-01 10:12:28 +10301076 if (cpumask_empty(frozen_cpus))
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001077 goto out;
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001078
Fabian Frederick84117da2014-06-04 16:11:17 -07001079 pr_info("Enabling non-boot CPUs ...\n");
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001080
1081 arch_enable_nonboot_cpus_begin();
1082
Rusty Russelle0b582e2009-01-01 10:12:28 +10301083 for_each_cpu(cpu, frozen_cpus) {
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001084 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
Thomas Gleixneraf1f4042016-02-26 18:43:30 +00001085 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
Todd E Brandtbb3632c2014-06-06 05:40:17 -07001086 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001087 if (!error) {
Fabian Frederick84117da2014-06-04 16:11:17 -07001088 pr_info("CPU%d is up\n", cpu);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001089 continue;
1090 }
Fabian Frederick84117da2014-06-04 16:11:17 -07001091 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001092 }
Suresh Siddhad0af9ee2009-08-19 18:05:36 -07001093
1094 arch_enable_nonboot_cpus_end();
1095
Rusty Russelle0b582e2009-01-01 10:12:28 +10301096 cpumask_clear(frozen_cpus);
Rafael J. Wysocki1d64b9c2007-04-01 23:49:49 -07001097out:
Gautham R Shenoyd2219382008-01-25 21:08:01 +01001098 cpu_maps_update_done();
Rafael J. Wysockie3920fb2006-09-25 23:32:48 -07001099}
Rusty Russelle0b582e2009-01-01 10:12:28 +10301100
Fenghua Yud7268a32011-11-15 21:59:31 +01001101static int __init alloc_frozen_cpus(void)
Rusty Russelle0b582e2009-01-01 10:12:28 +10301102{
1103 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1104 return -ENOMEM;
1105 return 0;
1106}
1107core_initcall(alloc_frozen_cpus);
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001108
1109/*
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001110 * When callbacks for CPU hotplug notifications are being executed, we must
1111 * ensure that the state of the system with respect to the tasks being frozen
1112 * or not, as reported by the notification, remains unchanged *throughout the
1113 * duration* of the execution of the callbacks.
1114 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1115 *
1116 * This synchronization is implemented by mutually excluding regular CPU
1117 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1118 * Hibernate notifications.
1119 */
1120static int
1121cpu_hotplug_pm_callback(struct notifier_block *nb,
1122 unsigned long action, void *ptr)
1123{
1124 switch (action) {
1125
1126 case PM_SUSPEND_PREPARE:
1127 case PM_HIBERNATION_PREPARE:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001128 cpu_hotplug_disable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001129 break;
1130
1131 case PM_POST_SUSPEND:
1132 case PM_POST_HIBERNATION:
Srivatsa S. Bhat16e53db2013-06-12 14:04:36 -07001133 cpu_hotplug_enable();
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001134 break;
1135
1136 default:
1137 return NOTIFY_DONE;
1138 }
1139
1140 return NOTIFY_OK;
1141}
1142
1143
Fenghua Yud7268a32011-11-15 21:59:31 +01001144static int __init cpu_hotplug_pm_sync_init(void)
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001145{
Fenghua Yu6e32d472012-11-13 11:32:43 -08001146 /*
1147 * cpu_hotplug_pm_callback has higher priority than x86
1148 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1149 * to disable cpu hotplug to avoid cpu hotplug race.
1150 */
Srivatsa S. Bhat79cfbdf2011-11-03 00:59:25 +01001151 pm_notifier(cpu_hotplug_pm_callback, 0);
1152 return 0;
1153}
1154core_initcall(cpu_hotplug_pm_sync_init);
1155
Rafael J. Wysockif3de4be2007-08-30 23:56:29 -07001156#endif /* CONFIG_PM_SLEEP_SMP */
Max Krasnyansky68f4f1e2008-05-29 11:17:02 -07001157
1158#endif /* CONFIG_SMP */
Mike Travisb8d317d2008-07-24 18:21:29 -07001159
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001160/* Boot processor state steps */
1161static struct cpuhp_step cpuhp_bp_states[] = {
1162 [CPUHP_OFFLINE] = {
1163 .name = "offline",
1164 .startup = NULL,
1165 .teardown = NULL,
1166 },
1167#ifdef CONFIG_SMP
1168 [CPUHP_CREATE_THREADS]= {
1169 .name = "threads:create",
1170 .startup = smpboot_create_threads,
1171 .teardown = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001172 .cant_stop = true,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001173 },
1174 [CPUHP_NOTIFY_PREPARE] = {
1175 .name = "notify:prepare",
1176 .startup = notify_prepare,
1177 .teardown = notify_dead,
1178 .skip_onerr = true,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001179 .cant_stop = true,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001180 },
1181 [CPUHP_BRINGUP_CPU] = {
1182 .name = "cpu:bringup",
1183 .startup = bringup_cpu,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001184 .teardown = NULL,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001185 .cant_stop = true,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001186 },
1187 [CPUHP_TEARDOWN_CPU] = {
1188 .name = "cpu:teardown",
1189 .startup = NULL,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001190 .teardown = takedown_cpu,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001191 .cant_stop = true,
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001192 },
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001193#endif
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001194};
1195
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001196/* Application processor state steps */
1197static struct cpuhp_step cpuhp_ap_states[] = {
1198#ifdef CONFIG_SMP
1199 [CPUHP_AP_NOTIFY_STARTING] = {
1200 .name = "notify:starting",
1201 .startup = notify_starting,
1202 .teardown = notify_dying,
1203 .skip_onerr = true,
Thomas Gleixner757c9892016-02-26 18:43:32 +00001204 .cant_stop = true,
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001205 },
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001206 [CPUHP_AP_SMPBOOT_THREADS] = {
1207 .name = "smpboot:threads",
1208 .startup = smpboot_unpark_threads,
1209 .teardown = smpboot_park_threads,
1210 },
1211 [CPUHP_AP_NOTIFY_ONLINE] = {
1212 .name = "notify:online",
1213 .startup = notify_online,
1214 .teardown = notify_down_prepare,
1215 },
Thomas Gleixner4baa0af2016-02-26 18:43:29 +00001216#endif
1217 [CPUHP_ONLINE] = {
1218 .name = "online",
1219 .startup = NULL,
1220 .teardown = NULL,
1221 },
1222};
1223
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001224/* Sanity check for callbacks */
1225static int cpuhp_cb_check(enum cpuhp_state state)
1226{
1227 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1228 return -EINVAL;
1229 return 0;
1230}
1231
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001232static bool cpuhp_is_ap_state(enum cpuhp_state state)
1233{
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001234 if (state >= CPUHP_AP_OFFLINE && state <= CPUHP_AP_ONLINE)
1235 return true;
Thomas Gleixner8df3e072016-02-26 18:43:41 +00001236 return state > CPUHP_BRINGUP_CPU;
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001237}
1238
1239static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
1240{
1241 struct cpuhp_step *sp;
1242
1243 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
1244 return sp + state;
1245}
1246
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001247static void cpuhp_store_callbacks(enum cpuhp_state state,
1248 const char *name,
1249 int (*startup)(unsigned int cpu),
1250 int (*teardown)(unsigned int cpu))
1251{
1252 /* (Un)Install the callbacks for further cpu hotplug operations */
1253 struct cpuhp_step *sp;
1254
1255 mutex_lock(&cpuhp_state_mutex);
1256 sp = cpuhp_get_step(state);
1257 sp->startup = startup;
1258 sp->teardown = teardown;
1259 sp->name = name;
1260 mutex_unlock(&cpuhp_state_mutex);
1261}
1262
1263static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1264{
1265 return cpuhp_get_step(state)->teardown;
1266}
1267
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001268/*
1269 * Call the startup/teardown function for a step either on the AP or
1270 * on the current CPU.
1271 */
1272static int cpuhp_issue_call(int cpu, enum cpuhp_state state,
1273 int (*cb)(unsigned int), bool bringup)
1274{
1275 int ret;
1276
1277 if (!cb)
1278 return 0;
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001279 /*
1280 * The non AP bound callbacks can fail on bringup. On teardown
1281 * e.g. module removal we crash for now.
1282 */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001283#ifdef CONFIG_SMP
1284 if (cpuhp_is_ap_state(state))
1285 ret = cpuhp_invoke_ap_callback(cpu, state, cb);
1286 else
1287 ret = cpuhp_invoke_callback(cpu, state, cb);
1288#else
1289 ret = cpuhp_invoke_callback(cpu, state, cb);
1290#endif
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001291 BUG_ON(ret && !bringup);
1292 return ret;
1293}
1294
1295/*
1296 * Called from __cpuhp_setup_state on a recoverable failure.
1297 *
1298 * Note: The teardown callbacks for rollback are not allowed to fail!
1299 */
1300static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1301 int (*teardown)(unsigned int cpu))
1302{
1303 int cpu;
1304
1305 if (!teardown)
1306 return;
1307
1308 /* Roll back the already executed steps on the other cpus */
1309 for_each_present_cpu(cpu) {
1310 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1311 int cpustate = st->state;
1312
1313 if (cpu >= failedcpu)
1314 break;
1315
1316 /* Did we invoke the startup call on that cpu ? */
1317 if (cpustate >= state)
1318 cpuhp_issue_call(cpu, state, teardown, false);
1319 }
1320}
1321
1322/*
1323 * Returns a free for dynamic slot assignment of the Online state. The states
1324 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1325 * by having no name assigned.
1326 */
1327static int cpuhp_reserve_state(enum cpuhp_state state)
1328{
1329 enum cpuhp_state i;
1330
1331 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001332 for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1333 if (cpuhp_ap_states[i].name)
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001334 continue;
1335
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001336 cpuhp_ap_states[i].name = "Reserved";
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001337 mutex_unlock(&cpuhp_state_mutex);
1338 return i;
1339 }
1340 mutex_unlock(&cpuhp_state_mutex);
1341 WARN(1, "No more dynamic states available for CPU hotplug\n");
1342 return -ENOSPC;
1343}
1344
1345/**
1346 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1347 * @state: The state to setup
1348 * @invoke: If true, the startup function is invoked for cpus where
1349 * cpu state >= @state
1350 * @startup: startup callback function
1351 * @teardown: teardown callback function
1352 *
1353 * Returns 0 if successful, otherwise a proper error code
1354 */
1355int __cpuhp_setup_state(enum cpuhp_state state,
1356 const char *name, bool invoke,
1357 int (*startup)(unsigned int cpu),
1358 int (*teardown)(unsigned int cpu))
1359{
1360 int cpu, ret = 0;
1361 int dyn_state = 0;
1362
1363 if (cpuhp_cb_check(state) || !name)
1364 return -EINVAL;
1365
1366 get_online_cpus();
1367
1368 /* currently assignments for the ONLINE state are possible */
Thomas Gleixner1cf4f622016-02-26 18:43:39 +00001369 if (state == CPUHP_AP_ONLINE_DYN) {
Thomas Gleixner5b7aa872016-02-26 18:43:33 +00001370 dyn_state = 1;
1371 ret = cpuhp_reserve_state(state);
1372 if (ret < 0)
1373 goto out;
1374 state = ret;
1375 }
1376
1377 cpuhp_store_callbacks(state, name, startup, teardown);
1378
1379 if (!invoke || !startup)
1380 goto out;
1381
1382 /*
1383 * Try to call the startup callback for each present cpu
1384 * depending on the hotplug state of the cpu.
1385 */
1386 for_each_present_cpu(cpu) {
1387 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1388 int cpustate = st->state;
1389
1390 if (cpustate < state)
1391 continue;
1392
1393 ret = cpuhp_issue_call(cpu, state, startup, true);
1394 if (ret) {
1395 cpuhp_rollback_install(cpu, state, teardown);
1396 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1397 goto out;
1398 }
1399 }
1400out:
1401 put_online_cpus();
1402 if (!ret && dyn_state)
1403 return state;
1404 return ret;
1405}
1406EXPORT_SYMBOL(__cpuhp_setup_state);
1407
1408/**
1409 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1410 * @state: The state to remove
1411 * @invoke: If true, the teardown function is invoked for cpus where
1412 * cpu state >= @state
1413 *
1414 * The teardown callback is currently not allowed to fail. Think
1415 * about module removal!
1416 */
1417void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1418{
1419 int (*teardown)(unsigned int cpu) = cpuhp_get_teardown_cb(state);
1420 int cpu;
1421
1422 BUG_ON(cpuhp_cb_check(state));
1423
1424 get_online_cpus();
1425
1426 if (!invoke || !teardown)
1427 goto remove;
1428
1429 /*
1430 * Call the teardown callback for each present cpu depending
1431 * on the hotplug state of the cpu. This function is not
1432 * allowed to fail currently!
1433 */
1434 for_each_present_cpu(cpu) {
1435 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1436 int cpustate = st->state;
1437
1438 if (cpustate >= state)
1439 cpuhp_issue_call(cpu, state, teardown, false);
1440 }
1441remove:
1442 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1443 put_online_cpus();
1444}
1445EXPORT_SYMBOL(__cpuhp_remove_state);
1446
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001447#if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1448static ssize_t show_cpuhp_state(struct device *dev,
1449 struct device_attribute *attr, char *buf)
1450{
1451 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1452
1453 return sprintf(buf, "%d\n", st->state);
1454}
1455static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1456
Thomas Gleixner757c9892016-02-26 18:43:32 +00001457static ssize_t write_cpuhp_target(struct device *dev,
1458 struct device_attribute *attr,
1459 const char *buf, size_t count)
1460{
1461 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1462 struct cpuhp_step *sp;
1463 int target, ret;
1464
1465 ret = kstrtoint(buf, 10, &target);
1466 if (ret)
1467 return ret;
1468
1469#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1470 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1471 return -EINVAL;
1472#else
1473 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1474 return -EINVAL;
1475#endif
1476
1477 ret = lock_device_hotplug_sysfs();
1478 if (ret)
1479 return ret;
1480
1481 mutex_lock(&cpuhp_state_mutex);
1482 sp = cpuhp_get_step(target);
1483 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1484 mutex_unlock(&cpuhp_state_mutex);
1485 if (ret)
1486 return ret;
1487
1488 if (st->state < target)
1489 ret = do_cpu_up(dev->id, target);
1490 else
1491 ret = do_cpu_down(dev->id, target);
1492
1493 unlock_device_hotplug();
1494 return ret ? ret : count;
1495}
1496
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001497static ssize_t show_cpuhp_target(struct device *dev,
1498 struct device_attribute *attr, char *buf)
1499{
1500 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1501
1502 return sprintf(buf, "%d\n", st->target);
1503}
Thomas Gleixner757c9892016-02-26 18:43:32 +00001504static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001505
1506static struct attribute *cpuhp_cpu_attrs[] = {
1507 &dev_attr_state.attr,
1508 &dev_attr_target.attr,
1509 NULL
1510};
1511
1512static struct attribute_group cpuhp_cpu_attr_group = {
1513 .attrs = cpuhp_cpu_attrs,
1514 .name = "hotplug",
1515 NULL
1516};
1517
1518static ssize_t show_cpuhp_states(struct device *dev,
1519 struct device_attribute *attr, char *buf)
1520{
1521 ssize_t cur, res = 0;
1522 int i;
1523
1524 mutex_lock(&cpuhp_state_mutex);
Thomas Gleixner757c9892016-02-26 18:43:32 +00001525 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
Thomas Gleixner98f8cdc2016-02-26 18:43:31 +00001526 struct cpuhp_step *sp = cpuhp_get_step(i);
1527
1528 if (sp->name) {
1529 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1530 buf += cur;
1531 res += cur;
1532 }
1533 }
1534 mutex_unlock(&cpuhp_state_mutex);
1535 return res;
1536}
1537static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1538
1539static struct attribute *cpuhp_cpu_root_attrs[] = {
1540 &dev_attr_states.attr,
1541 NULL
1542};
1543
1544static struct attribute_group cpuhp_cpu_root_attr_group = {
1545 .attrs = cpuhp_cpu_root_attrs,
1546 .name = "hotplug",
1547 NULL
1548};
1549
1550static int __init cpuhp_sysfs_init(void)
1551{
1552 int cpu, ret;
1553
1554 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1555 &cpuhp_cpu_root_attr_group);
1556 if (ret)
1557 return ret;
1558
1559 for_each_possible_cpu(cpu) {
1560 struct device *dev = get_cpu_device(cpu);
1561
1562 if (!dev)
1563 continue;
1564 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1565 if (ret)
1566 return ret;
1567 }
1568 return 0;
1569}
1570device_initcall(cpuhp_sysfs_init);
1571#endif
1572
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001573/*
1574 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1575 * represents all NR_CPUS bits binary values of 1<<nr.
1576 *
Rusty Russelle0b582e2009-01-01 10:12:28 +10301577 * It is used by cpumask_of() to get a constant address to a CPU
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001578 * mask value that has a single bit set only.
1579 */
Mike Travisb8d317d2008-07-24 18:21:29 -07001580
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001581/* cpu_bit_bitmap[0] is empty - so we can back into it */
Michael Rodriguez4d519852011-03-22 16:34:07 -07001582#define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001583#define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1584#define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1585#define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
Mike Travisb8d317d2008-07-24 18:21:29 -07001586
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001587const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
Mike Travisb8d317d2008-07-24 18:21:29 -07001588
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001589 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1590 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1591#if BITS_PER_LONG > 32
1592 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1593 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
Mike Travisb8d317d2008-07-24 18:21:29 -07001594#endif
1595};
Linus Torvaldse56b3bc2008-07-28 11:32:33 -07001596EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
Rusty Russell2d3854a2008-11-05 13:39:10 +11001597
1598const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1599EXPORT_SYMBOL(cpu_all_bits);
Rusty Russellb3199c02008-12-30 09:05:14 +10301600
1601#ifdef CONFIG_INIT_ALL_POSSIBLE
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001602struct cpumask __cpu_possible_mask __read_mostly
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08001603 = {CPU_BITS_ALL};
Rusty Russellb3199c02008-12-30 09:05:14 +10301604#else
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001605struct cpumask __cpu_possible_mask __read_mostly;
Rusty Russellb3199c02008-12-30 09:05:14 +10301606#endif
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001607EXPORT_SYMBOL(__cpu_possible_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10301608
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001609struct cpumask __cpu_online_mask __read_mostly;
1610EXPORT_SYMBOL(__cpu_online_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10301611
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001612struct cpumask __cpu_present_mask __read_mostly;
1613EXPORT_SYMBOL(__cpu_present_mask);
Rusty Russellb3199c02008-12-30 09:05:14 +10301614
Rasmus Villemoes4b804c82016-01-20 15:00:19 -08001615struct cpumask __cpu_active_mask __read_mostly;
1616EXPORT_SYMBOL(__cpu_active_mask);
Rusty Russell3fa41522008-12-30 09:05:16 +10301617
Rusty Russell3fa41522008-12-30 09:05:16 +10301618void init_cpu_present(const struct cpumask *src)
1619{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08001620 cpumask_copy(&__cpu_present_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10301621}
1622
1623void init_cpu_possible(const struct cpumask *src)
1624{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08001625 cpumask_copy(&__cpu_possible_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10301626}
1627
1628void init_cpu_online(const struct cpumask *src)
1629{
Rasmus Villemoesc4c54dd2016-01-20 15:00:16 -08001630 cpumask_copy(&__cpu_online_mask, src);
Rusty Russell3fa41522008-12-30 09:05:16 +10301631}
Thomas Gleixnercff7d372016-02-26 18:43:28 +00001632
1633/*
1634 * Activate the first processor.
1635 */
1636void __init boot_cpu_init(void)
1637{
1638 int cpu = smp_processor_id();
1639
1640 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1641 set_cpu_online(cpu, true);
1642 set_cpu_active(cpu, true);
1643 set_cpu_present(cpu, true);
1644 set_cpu_possible(cpu, true);
1645}
1646
1647/*
1648 * Must be called _AFTER_ setting up the per_cpu areas
1649 */
1650void __init boot_cpu_state_init(void)
1651{
1652 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;
1653}