]> git.karo-electronics.de Git - linux-beck.git/blob - kernel/cpu.c
93e9d89bb0ab80c9c71a8b9c4842ec425244f6c0
[linux-beck.git] / kernel / cpu.c
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>
13 #include <linux/oom.h>
14 #include <linux/rcupdate.h>
15 #include <linux/export.h>
16 #include <linux/bug.h>
17 #include <linux/kthread.h>
18 #include <linux/stop_machine.h>
19 #include <linux/mutex.h>
20 #include <linux/gfp.h>
21 #include <linux/suspend.h>
22 #include <linux/lockdep.h>
23 #include <linux/tick.h>
24 #include <linux/irq.h>
25 #include <linux/smpboot.h>
26
27 #include <trace/events/power.h>
28 #define CREATE_TRACE_POINTS
29 #include <trace/events/cpuhp.h>
30
31 #include "smpboot.h"
32
33 /**
34  * cpuhp_cpu_state - Per cpu hotplug state storage
35  * @state:      The current cpu state
36  * @target:     The target state
37  * @thread:     Pointer to the hotplug thread
38  * @should_run: Thread should execute
39  * @cb_stat:    The state for a single callback (install/uninstall)
40  * @cb:         Single callback function (install/uninstall)
41  * @result:     Result of the operation
42  * @done:       Signal completion to the issuer of the task
43  */
44 struct cpuhp_cpu_state {
45         enum cpuhp_state        state;
46         enum cpuhp_state        target;
47 #ifdef CONFIG_SMP
48         struct task_struct      *thread;
49         bool                    should_run;
50         enum cpuhp_state        cb_state;
51         int                     (*cb)(unsigned int cpu);
52         int                     result;
53         struct completion       done;
54 #endif
55 };
56
57 static 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
66  * @cant_stop:  Bringup/teardown can't be stopped at this step
67  */
68 struct cpuhp_step {
69         const char      *name;
70         int             (*startup)(unsigned int cpu);
71         int             (*teardown)(unsigned int cpu);
72         bool            skip_onerr;
73         bool            cant_stop;
74 };
75
76 static DEFINE_MUTEX(cpuhp_state_mutex);
77 static struct cpuhp_step cpuhp_bp_states[];
78 static struct cpuhp_step cpuhp_ap_states[];
79
80 /**
81  * cpuhp_invoke_callback _ Invoke the callbacks for a given state
82  * @cpu:        The cpu for which the callback should be invoked
83  * @step:       The step in the state machine
84  * @cb:         The callback function to invoke
85  *
86  * Called from cpu hotplug and from the state register machinery
87  */
88 static 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
102 #ifdef CONFIG_SMP
103 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
104 static DEFINE_MUTEX(cpu_add_remove_lock);
105 bool cpuhp_tasks_frozen;
106 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
107
108 /*
109  * The following two APIs (cpu_maps_update_begin/done) must be used when
110  * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
111  * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
112  * hotplug callback (un)registration performed using __register_cpu_notifier()
113  * or __unregister_cpu_notifier().
114  */
115 void cpu_maps_update_begin(void)
116 {
117         mutex_lock(&cpu_add_remove_lock);
118 }
119 EXPORT_SYMBOL(cpu_notifier_register_begin);
120
121 void cpu_maps_update_done(void)
122 {
123         mutex_unlock(&cpu_add_remove_lock);
124 }
125 EXPORT_SYMBOL(cpu_notifier_register_done);
126
127 static RAW_NOTIFIER_HEAD(cpu_chain);
128
129 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
130  * Should always be manipulated under cpu_add_remove_lock
131  */
132 static int cpu_hotplug_disabled;
133
134 #ifdef CONFIG_HOTPLUG_CPU
135
136 static struct {
137         struct task_struct *active_writer;
138         /* wait queue to wake up the active_writer */
139         wait_queue_head_t wq;
140         /* verifies that no writer will get active while readers are active */
141         struct mutex lock;
142         /*
143          * Also blocks the new readers during
144          * an ongoing cpu hotplug operation.
145          */
146         atomic_t refcount;
147
148 #ifdef CONFIG_DEBUG_LOCK_ALLOC
149         struct lockdep_map dep_map;
150 #endif
151 } cpu_hotplug = {
152         .active_writer = NULL,
153         .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
154         .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
155 #ifdef CONFIG_DEBUG_LOCK_ALLOC
156         .dep_map = {.name = "cpu_hotplug.lock" },
157 #endif
158 };
159
160 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
161 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
162 #define cpuhp_lock_acquire_tryread() \
163                                   lock_map_acquire_tryread(&cpu_hotplug.dep_map)
164 #define cpuhp_lock_acquire()      lock_map_acquire(&cpu_hotplug.dep_map)
165 #define cpuhp_lock_release()      lock_map_release(&cpu_hotplug.dep_map)
166
167
168 void get_online_cpus(void)
169 {
170         might_sleep();
171         if (cpu_hotplug.active_writer == current)
172                 return;
173         cpuhp_lock_acquire_read();
174         mutex_lock(&cpu_hotplug.lock);
175         atomic_inc(&cpu_hotplug.refcount);
176         mutex_unlock(&cpu_hotplug.lock);
177 }
178 EXPORT_SYMBOL_GPL(get_online_cpus);
179
180 void put_online_cpus(void)
181 {
182         int refcount;
183
184         if (cpu_hotplug.active_writer == current)
185                 return;
186
187         refcount = atomic_dec_return(&cpu_hotplug.refcount);
188         if (WARN_ON(refcount < 0)) /* try to fix things up */
189                 atomic_inc(&cpu_hotplug.refcount);
190
191         if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
192                 wake_up(&cpu_hotplug.wq);
193
194         cpuhp_lock_release();
195
196 }
197 EXPORT_SYMBOL_GPL(put_online_cpus);
198
199 /*
200  * This ensures that the hotplug operation can begin only when the
201  * refcount goes to zero.
202  *
203  * Note that during a cpu-hotplug operation, the new readers, if any,
204  * will be blocked by the cpu_hotplug.lock
205  *
206  * Since cpu_hotplug_begin() is always called after invoking
207  * cpu_maps_update_begin(), we can be sure that only one writer is active.
208  *
209  * Note that theoretically, there is a possibility of a livelock:
210  * - Refcount goes to zero, last reader wakes up the sleeping
211  *   writer.
212  * - Last reader unlocks the cpu_hotplug.lock.
213  * - A new reader arrives at this moment, bumps up the refcount.
214  * - The writer acquires the cpu_hotplug.lock finds the refcount
215  *   non zero and goes to sleep again.
216  *
217  * However, this is very difficult to achieve in practice since
218  * get_online_cpus() not an api which is called all that often.
219  *
220  */
221 void cpu_hotplug_begin(void)
222 {
223         DEFINE_WAIT(wait);
224
225         cpu_hotplug.active_writer = current;
226         cpuhp_lock_acquire();
227
228         for (;;) {
229                 mutex_lock(&cpu_hotplug.lock);
230                 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
231                 if (likely(!atomic_read(&cpu_hotplug.refcount)))
232                                 break;
233                 mutex_unlock(&cpu_hotplug.lock);
234                 schedule();
235         }
236         finish_wait(&cpu_hotplug.wq, &wait);
237 }
238
239 void cpu_hotplug_done(void)
240 {
241         cpu_hotplug.active_writer = NULL;
242         mutex_unlock(&cpu_hotplug.lock);
243         cpuhp_lock_release();
244 }
245
246 /*
247  * Wait for currently running CPU hotplug operations to complete (if any) and
248  * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
249  * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
250  * hotplug path before performing hotplug operations. So acquiring that lock
251  * guarantees mutual exclusion from any currently running hotplug operations.
252  */
253 void cpu_hotplug_disable(void)
254 {
255         cpu_maps_update_begin();
256         cpu_hotplug_disabled++;
257         cpu_maps_update_done();
258 }
259 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
260
261 void cpu_hotplug_enable(void)
262 {
263         cpu_maps_update_begin();
264         WARN_ON(--cpu_hotplug_disabled < 0);
265         cpu_maps_update_done();
266 }
267 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
268 #endif  /* CONFIG_HOTPLUG_CPU */
269
270 /* Need to know about CPUs going up/down? */
271 int register_cpu_notifier(struct notifier_block *nb)
272 {
273         int ret;
274         cpu_maps_update_begin();
275         ret = raw_notifier_chain_register(&cpu_chain, nb);
276         cpu_maps_update_done();
277         return ret;
278 }
279
280 int __register_cpu_notifier(struct notifier_block *nb)
281 {
282         return raw_notifier_chain_register(&cpu_chain, nb);
283 }
284
285 static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
286                         int *nr_calls)
287 {
288         unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
289         void *hcpu = (void *)(long)cpu;
290
291         int ret;
292
293         ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
294                                         nr_calls);
295
296         return notifier_to_errno(ret);
297 }
298
299 static int cpu_notify(unsigned long val, unsigned int cpu)
300 {
301         return __cpu_notify(val, cpu, -1, NULL);
302 }
303
304 /* Notifier wrappers for transitioning to state machine */
305 static 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
320 static int notify_online(unsigned int cpu)
321 {
322         cpu_notify(CPU_ONLINE, cpu);
323         return 0;
324 }
325
326 static int notify_starting(unsigned int cpu)
327 {
328         cpu_notify(CPU_STARTING, cpu);
329         return 0;
330 }
331
332 static 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
340 static 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         }
351         ret = bringup_wait_for_ap(cpu);
352         BUG_ON(!cpu_online(cpu));
353         return ret;
354 }
355
356 /*
357  * Hotplug state machine related functions
358  */
359 static 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
370 static 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
389 static 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
400 static 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
421 /*
422  * The cpu hotplug threads manage the bringup and teardown of the cpus
423  */
424 static 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
431 static 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 */
439 static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
440 {
441         enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
442
443         return cpuhp_down_callbacks(cpu, st, cpuhp_ap_states, target);
444 }
445
446 /* Execute the online startup callbacks. Used to be CPU_ONLINE */
447 static 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  */
456 static 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 {
481                 /* Cannot happen .... */
482                 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
483
484                 /* Regular hotplug work */
485                 if (st->state < st->target)
486                         ret = cpuhp_ap_online(cpu, st);
487                 else if (st->state > st->target)
488                         ret = cpuhp_ap_offline(cpu, st);
489         }
490         st->result = ret;
491         complete(&st->done);
492 }
493
494 /* Invoke a single callback on a remote cpu */
495 static 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 */
517 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
518 {
519         st->result = 0;
520         st->cb = NULL;
521         /*
522          * Make sure the above stores are visible before should_run becomes
523          * true. Paired with the mb() above in cpuhp_thread_fun()
524          */
525         smp_mb();
526         st->should_run = true;
527         wake_up_process(st->thread);
528 }
529
530 static 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);
537         wait_for_completion(&st->done);
538         trace_cpuhp_exit(cpu, st->state, state, st->result);
539         return st->result;
540 }
541
542 static 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
551 void __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
557 #ifdef CONFIG_HOTPLUG_CPU
558 EXPORT_SYMBOL(register_cpu_notifier);
559 EXPORT_SYMBOL(__register_cpu_notifier);
560 void unregister_cpu_notifier(struct notifier_block *nb)
561 {
562         cpu_maps_update_begin();
563         raw_notifier_chain_unregister(&cpu_chain, nb);
564         cpu_maps_update_done();
565 }
566 EXPORT_SYMBOL(unregister_cpu_notifier);
567
568 void __unregister_cpu_notifier(struct notifier_block *nb)
569 {
570         raw_notifier_chain_unregister(&cpu_chain, nb);
571 }
572 EXPORT_SYMBOL(__unregister_cpu_notifier);
573
574 /**
575  * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
576  * @cpu: a CPU id
577  *
578  * This function walks all processes, finds a valid mm struct for each one and
579  * then clears a corresponding bit in mm's cpumask.  While this all sounds
580  * trivial, there are various non-obvious corner cases, which this function
581  * tries to solve in a safe manner.
582  *
583  * Also note that the function uses a somewhat relaxed locking scheme, so it may
584  * be called only for an already offlined CPU.
585  */
586 void 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          */
597         WARN_ON(cpu_online(cpu));
598         rcu_read_lock();
599         for_each_process(p) {
600                 struct task_struct *t;
601
602                 /*
603                  * Main thread might exit, but other threads may still have
604                  * a valid mm. Find one.
605                  */
606                 t = find_lock_task_mm(p);
607                 if (!t)
608                         continue;
609                 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
610                 task_unlock(t);
611         }
612         rcu_read_unlock();
613 }
614
615 static inline void check_for_tasks(int dead_cpu)
616 {
617         struct task_struct *g, *p;
618
619         read_lock(&tasklist_lock);
620         for_each_process_thread(g, p) {
621                 if (!p->on_rq)
622                         continue;
623                 /*
624                  * We do the check with unlocked task_rq(p)->lock.
625                  * Order the reading to do not warn about a task,
626                  * which was running on this cpu in the past, and
627                  * it's just been woken on another cpu.
628                  */
629                 rmb();
630                 if (task_cpu(p) != dead_cpu)
631                         continue;
632
633                 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
634                         p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
635         }
636         read_unlock(&tasklist_lock);
637 }
638
639 static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
640 {
641         BUG_ON(cpu_notify(val, cpu));
642 }
643
644 static 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
658 static int notify_dying(unsigned int cpu)
659 {
660         cpu_notify(CPU_DYING, cpu);
661         return 0;
662 }
663
664 /* Take this CPU down. */
665 static int take_cpu_down(void *_param)
666 {
667         struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
668         enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
669         int err, cpu = smp_processor_id();
670
671         /* Ensure this CPU doesn't handle any more interrupts. */
672         err = __cpu_disable();
673         if (err < 0)
674                 return err;
675
676         /* Invoke the former CPU_DYING callbacks */
677         for (; st->state > target; st->state--) {
678                 struct cpuhp_step *step = cpuhp_ap_states + st->state;
679
680                 cpuhp_invoke_callback(cpu, st->state, step->teardown);
681         }
682         /* Give up timekeeping duties */
683         tick_handover_do_timer();
684         /* Park the stopper thread */
685         stop_machine_park(cpu);
686         return 0;
687 }
688
689 static int takedown_cpu(unsigned int cpu)
690 {
691         struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
692         int err;
693
694         /*
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
700          * not imply sync_sched(), so wait for both.
701          *
702          * Do sync before park smpboot threads to take care the rcu boost case.
703          */
704         if (IS_ENABLED(CONFIG_PREEMPT))
705                 synchronize_rcu_mult(call_rcu, call_rcu_sched);
706         else
707                 synchronize_rcu();
708
709         /* Park the hotplug thread */
710         kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
711
712         /*
713          * Prevent irq alloc/free while the dying cpu reorganizes the
714          * interrupt affinities.
715          */
716         irq_lock_sparse();
717
718         /*
719          * So now all preempt/rcu users must observe !cpu_active().
720          */
721         err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
722         if (err) {
723                 /* CPU didn't die: tell everyone.  Can't complain. */
724                 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
725                 irq_unlock_sparse();
726                 return err;
727         }
728         BUG_ON(cpu_online(cpu));
729
730         /*
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.
734          *
735          * Wait for the stop thread to go away.
736          */
737         wait_for_completion(&st->done);
738         BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
739
740         /* Interrupts are moved away from the dying cpu, reenable alloc/free */
741         irq_unlock_sparse();
742
743         hotplug_cpu__broadcast_tick_pull(cpu);
744         /* This actually kills the CPU. */
745         __cpu_die(cpu);
746
747         tick_cleanup_dead_cpu(cpu);
748         return 0;
749 }
750
751 static int notify_dead(unsigned int cpu)
752 {
753         cpu_notify_nofail(CPU_DEAD, cpu);
754         check_for_tasks(cpu);
755         return 0;
756 }
757
758 static void cpuhp_complete_idle_dead(void *arg)
759 {
760         struct cpuhp_cpu_state *st = arg;
761
762         complete(&st->done);
763 }
764
765 void 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);
770         rcu_report_dead(smp_processor_id());
771         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);
778 }
779
780 #else
781 #define notify_down_prepare     NULL
782 #define takedown_cpu            NULL
783 #define notify_dead             NULL
784 #define notify_dying            NULL
785 #endif
786
787 #ifdef CONFIG_HOTPLUG_CPU
788
789 /* Requires cpu_add_remove_lock to be held */
790 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
791                            enum cpuhp_state target)
792 {
793         struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
794         int prev_state, ret = 0;
795         bool hasdied = false;
796
797         if (num_online_cpus() == 1)
798                 return -EBUSY;
799
800         if (!cpu_present(cpu))
801                 return -EINVAL;
802
803         cpu_hotplug_begin();
804
805         cpuhp_tasks_frozen = tasks_frozen;
806
807         prev_state = st->state;
808         st->target = target;
809         /*
810          * If the current CPU state is in the range of the AP hotplug thread,
811          * then we need to kick the thread.
812          */
813         if (st->state > CPUHP_TEARDOWN_CPU) {
814                 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                  */
826                 if (st->state > CPUHP_TEARDOWN_CPU)
827                         goto out;
828         }
829         /*
830          * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
831          * to do the further cleanups.
832          */
833         ret = cpuhp_down_callbacks(cpu, st, cpuhp_bp_states, target);
834
835         hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
836 out:
837         cpu_hotplug_done();
838         /* This post dead nonsense must die */
839         if (!ret && hasdied)
840                 cpu_notify_nofail(CPU_POST_DEAD, cpu);
841         return ret;
842 }
843
844 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
845 {
846         int err;
847
848         cpu_maps_update_begin();
849
850         if (cpu_hotplug_disabled) {
851                 err = -EBUSY;
852                 goto out;
853         }
854
855         err = _cpu_down(cpu, 0, target);
856
857 out:
858         cpu_maps_update_done();
859         return err;
860 }
861 int cpu_down(unsigned int cpu)
862 {
863         return do_cpu_down(cpu, CPUHP_OFFLINE);
864 }
865 EXPORT_SYMBOL(cpu_down);
866 #endif /*CONFIG_HOTPLUG_CPU*/
867
868 /**
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  */
876 void 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
890 /*
891  * Called from the idle task. We need to set active here, so we can kick off
892  * 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.
895  */
896 void cpuhp_online_idle(enum cpuhp_state state)
897 {
898         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;
906
907         /* The cpu is marked online, set it active now */
908         set_cpu_active(cpu, true);
909         /* Unpark the stopper thread and the hotplug thread of this cpu */
910         stop_machine_unpark(cpu);
911         kthread_unpark(st->thread);
912
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);
918 }
919
920 /* Requires cpu_add_remove_lock to be held */
921 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
922 {
923         struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
924         struct task_struct *idle;
925         int ret = 0;
926
927         cpu_hotplug_begin();
928
929         if (!cpu_present(cpu)) {
930                 ret = -EINVAL;
931                 goto out;
932         }
933
934         /*
935          * The caller of do_cpu_up might have raced with another
936          * caller. Ignore it for now.
937          */
938         if (st->state >= target)
939                 goto out;
940
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                 }
948         }
949
950         cpuhp_tasks_frozen = tasks_frozen;
951
952         st->target = target;
953         /*
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          */
957         if (st->state > CPUHP_BRINGUP_CPU) {
958                 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
969          * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
970          * responsible for bringing it up to the target state.
971          */
972         target = min((int)target, CPUHP_BRINGUP_CPU);
973         ret = cpuhp_up_callbacks(cpu, st, cpuhp_bp_states, target);
974 out:
975         cpu_hotplug_done();
976         return ret;
977 }
978
979 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
980 {
981         int err = 0;
982
983         if (!cpu_possible(cpu)) {
984                 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
985                        cpu);
986 #if defined(CONFIG_IA64)
987                 pr_err("please check additional_cpus= boot parameter\n");
988 #endif
989                 return -EINVAL;
990         }
991
992         err = try_online_node(cpu_to_node(cpu));
993         if (err)
994                 return err;
995
996         cpu_maps_update_begin();
997
998         if (cpu_hotplug_disabled) {
999                 err = -EBUSY;
1000                 goto out;
1001         }
1002
1003         err = _cpu_up(cpu, 0, target);
1004 out:
1005         cpu_maps_update_done();
1006         return err;
1007 }
1008
1009 int cpu_up(unsigned int cpu)
1010 {
1011         return do_cpu_up(cpu, CPUHP_ONLINE);
1012 }
1013 EXPORT_SYMBOL_GPL(cpu_up);
1014
1015 #ifdef CONFIG_PM_SLEEP_SMP
1016 static cpumask_var_t frozen_cpus;
1017
1018 int disable_nonboot_cpus(void)
1019 {
1020         int cpu, first_cpu, error = 0;
1021
1022         cpu_maps_update_begin();
1023         first_cpu = cpumask_first(cpu_online_mask);
1024         /*
1025          * We take down all of the non-boot CPUs in one shot to avoid races
1026          * with the userspace trying to use the CPU hotplug at the same time
1027          */
1028         cpumask_clear(frozen_cpus);
1029
1030         pr_info("Disabling non-boot CPUs ...\n");
1031         for_each_online_cpu(cpu) {
1032                 if (cpu == first_cpu)
1033                         continue;
1034                 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1035                 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1036                 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1037                 if (!error)
1038                         cpumask_set_cpu(cpu, frozen_cpus);
1039                 else {
1040                         pr_err("Error taking CPU%d down: %d\n", cpu, error);
1041                         break;
1042                 }
1043         }
1044
1045         if (!error)
1046                 BUG_ON(num_online_cpus() > 1);
1047         else
1048                 pr_err("Non-boot CPUs are not disabled\n");
1049
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
1057         cpu_maps_update_done();
1058         return error;
1059 }
1060
1061 void __weak arch_enable_nonboot_cpus_begin(void)
1062 {
1063 }
1064
1065 void __weak arch_enable_nonboot_cpus_end(void)
1066 {
1067 }
1068
1069 void enable_nonboot_cpus(void)
1070 {
1071         int cpu, error;
1072
1073         /* Allow everyone to use the CPU hotplug again */
1074         cpu_maps_update_begin();
1075         WARN_ON(--cpu_hotplug_disabled < 0);
1076         if (cpumask_empty(frozen_cpus))
1077                 goto out;
1078
1079         pr_info("Enabling non-boot CPUs ...\n");
1080
1081         arch_enable_nonboot_cpus_begin();
1082
1083         for_each_cpu(cpu, frozen_cpus) {
1084                 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1085                 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1086                 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1087                 if (!error) {
1088                         pr_info("CPU%d is up\n", cpu);
1089                         continue;
1090                 }
1091                 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1092         }
1093
1094         arch_enable_nonboot_cpus_end();
1095
1096         cpumask_clear(frozen_cpus);
1097 out:
1098         cpu_maps_update_done();
1099 }
1100
1101 static int __init alloc_frozen_cpus(void)
1102 {
1103         if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1104                 return -ENOMEM;
1105         return 0;
1106 }
1107 core_initcall(alloc_frozen_cpus);
1108
1109 /*
1110  * 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  */
1120 static int
1121 cpu_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:
1128                 cpu_hotplug_disable();
1129                 break;
1130
1131         case PM_POST_SUSPEND:
1132         case PM_POST_HIBERNATION:
1133                 cpu_hotplug_enable();
1134                 break;
1135
1136         default:
1137                 return NOTIFY_DONE;
1138         }
1139
1140         return NOTIFY_OK;
1141 }
1142
1143
1144 static int __init cpu_hotplug_pm_sync_init(void)
1145 {
1146         /*
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          */
1151         pm_notifier(cpu_hotplug_pm_callback, 0);
1152         return 0;
1153 }
1154 core_initcall(cpu_hotplug_pm_sync_init);
1155
1156 #endif /* CONFIG_PM_SLEEP_SMP */
1157
1158 #endif /* CONFIG_SMP */
1159
1160 /* Boot processor state steps */
1161 static 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,
1172                 .cant_stop              = true,
1173         },
1174         [CPUHP_NOTIFY_PREPARE] = {
1175                 .name                   = "notify:prepare",
1176                 .startup                = notify_prepare,
1177                 .teardown               = notify_dead,
1178                 .skip_onerr             = true,
1179                 .cant_stop              = true,
1180         },
1181         [CPUHP_BRINGUP_CPU] = {
1182                 .name                   = "cpu:bringup",
1183                 .startup                = bringup_cpu,
1184                 .teardown               = NULL,
1185                 .cant_stop              = true,
1186         },
1187         [CPUHP_TEARDOWN_CPU] = {
1188                 .name                   = "cpu:teardown",
1189                 .startup                = NULL,
1190                 .teardown               = takedown_cpu,
1191                 .cant_stop              = true,
1192         },
1193 #endif
1194 };
1195
1196 /* Application processor state steps */
1197 static 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,
1204                 .cant_stop              = true,
1205         },
1206         [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         },
1216 #endif
1217         [CPUHP_ONLINE] = {
1218                 .name                   = "online",
1219                 .startup                = NULL,
1220                 .teardown               = NULL,
1221         },
1222 };
1223
1224 /* Sanity check for callbacks */
1225 static 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
1232 static bool cpuhp_is_ap_state(enum cpuhp_state state)
1233 {
1234         if (state >= CPUHP_AP_OFFLINE && state <= CPUHP_AP_ONLINE)
1235                 return true;
1236         return state > CPUHP_BRINGUP_CPU;
1237 }
1238
1239 static 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
1247 static 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
1263 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1264 {
1265         return cpuhp_get_step(state)->teardown;
1266 }
1267
1268 /*
1269  * Call the startup/teardown function for a step either on the AP or
1270  * on the current CPU.
1271  */
1272 static 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;
1279         /*
1280          * The non AP bound callbacks can fail on bringup. On teardown
1281          * e.g. module removal we crash for now.
1282          */
1283 #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
1291         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  */
1300 static 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  */
1327 static int cpuhp_reserve_state(enum cpuhp_state state)
1328 {
1329         enum cpuhp_state i;
1330
1331         mutex_lock(&cpuhp_state_mutex);
1332         for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1333                 if (cpuhp_ap_states[i].name)
1334                         continue;
1335
1336                 cpuhp_ap_states[i].name = "Reserved";
1337                 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  */
1355 int __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 */
1369         if (state == CPUHP_AP_ONLINE_DYN) {
1370                 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         }
1400 out:
1401         put_online_cpus();
1402         if (!ret && dyn_state)
1403                 return state;
1404         return ret;
1405 }
1406 EXPORT_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  */
1417 void __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         }
1441 remove:
1442         cpuhp_store_callbacks(state, NULL, NULL, NULL);
1443         put_online_cpus();
1444 }
1445 EXPORT_SYMBOL(__cpuhp_remove_state);
1446
1447 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1448 static 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 }
1455 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1456
1457 static 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
1497 static 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 }
1504 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1505
1506 static struct attribute *cpuhp_cpu_attrs[] = {
1507         &dev_attr_state.attr,
1508         &dev_attr_target.attr,
1509         NULL
1510 };
1511
1512 static struct attribute_group cpuhp_cpu_attr_group = {
1513         .attrs = cpuhp_cpu_attrs,
1514         .name = "hotplug",
1515         NULL
1516 };
1517
1518 static 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);
1525         for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1526                 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 }
1537 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1538
1539 static struct attribute *cpuhp_cpu_root_attrs[] = {
1540         &dev_attr_states.attr,
1541         NULL
1542 };
1543
1544 static struct attribute_group cpuhp_cpu_root_attr_group = {
1545         .attrs = cpuhp_cpu_root_attrs,
1546         .name = "hotplug",
1547         NULL
1548 };
1549
1550 static 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 }
1570 device_initcall(cpuhp_sysfs_init);
1571 #endif
1572
1573 /*
1574  * cpu_bit_bitmap[] is a special, "compressed" data structure that
1575  * represents all NR_CPUS bits binary values of 1<<nr.
1576  *
1577  * It is used by cpumask_of() to get a constant address to a CPU
1578  * mask value that has a single bit set only.
1579  */
1580
1581 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1582 #define MASK_DECLARE_1(x)       [x+1][0] = (1UL << (x))
1583 #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)
1586
1587 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1588
1589         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),
1594 #endif
1595 };
1596 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
1597
1598 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1599 EXPORT_SYMBOL(cpu_all_bits);
1600
1601 #ifdef CONFIG_INIT_ALL_POSSIBLE
1602 struct cpumask __cpu_possible_mask __read_mostly
1603         = {CPU_BITS_ALL};
1604 #else
1605 struct cpumask __cpu_possible_mask __read_mostly;
1606 #endif
1607 EXPORT_SYMBOL(__cpu_possible_mask);
1608
1609 struct cpumask __cpu_online_mask __read_mostly;
1610 EXPORT_SYMBOL(__cpu_online_mask);
1611
1612 struct cpumask __cpu_present_mask __read_mostly;
1613 EXPORT_SYMBOL(__cpu_present_mask);
1614
1615 struct cpumask __cpu_active_mask __read_mostly;
1616 EXPORT_SYMBOL(__cpu_active_mask);
1617
1618 void init_cpu_present(const struct cpumask *src)
1619 {
1620         cpumask_copy(&__cpu_present_mask, src);
1621 }
1622
1623 void init_cpu_possible(const struct cpumask *src)
1624 {
1625         cpumask_copy(&__cpu_possible_mask, src);
1626 }
1627
1628 void init_cpu_online(const struct cpumask *src)
1629 {
1630         cpumask_copy(&__cpu_online_mask, src);
1631 }
1632
1633 /*
1634  * Activate the first processor.
1635  */
1636 void __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  */
1650 void __init boot_cpu_state_init(void)
1651 {
1652         per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;
1653 }