2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/proc_fs.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>
27 #include <trace/events/power.h>
28 #define CREATE_TRACE_POINTS
29 #include <trace/events/cpuhp.h>
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 * @rollback: Perform a rollback
40 * @single: Single callback invocation
41 * @bringup: Single callback bringup or teardown selector
42 * @cb_state: The state for a single callback (install/uninstall)
43 * @result: Result of the operation
44 * @done: Signal completion to the issuer of the task
46 struct cpuhp_cpu_state {
47 enum cpuhp_state state;
48 enum cpuhp_state target;
50 struct task_struct *thread;
55 struct hlist_node *node;
56 enum cpuhp_state cb_state;
58 struct completion done;
62 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
65 * cpuhp_step - Hotplug state machine step
66 * @name: Name of the step
67 * @startup: Startup function of the step
68 * @teardown: Teardown function of the step
69 * @skip_onerr: Do not invoke the functions on error rollback
70 * Will go away once the notifiers are gone
71 * @cant_stop: Bringup/teardown can't be stopped at this step
76 int (*single)(unsigned int cpu);
77 int (*multi)(unsigned int cpu,
78 struct hlist_node *node);
81 int (*single)(unsigned int cpu);
82 int (*multi)(unsigned int cpu,
83 struct hlist_node *node);
85 struct hlist_head list;
91 static DEFINE_MUTEX(cpuhp_state_mutex);
92 static struct cpuhp_step cpuhp_bp_states[];
93 static struct cpuhp_step cpuhp_ap_states[];
95 static bool cpuhp_is_ap_state(enum cpuhp_state state)
98 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
99 * purposes as that state is handled explicitly in cpu_down.
101 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
104 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
106 struct cpuhp_step *sp;
108 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
113 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
114 * @cpu: The cpu for which the callback should be invoked
115 * @step: The step in the state machine
116 * @bringup: True if the bringup callback should be invoked
118 * Called from cpu hotplug and from the state register machinery.
120 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
121 bool bringup, struct hlist_node *node)
123 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
124 struct cpuhp_step *step = cpuhp_get_step(state);
125 int (*cbm)(unsigned int cpu, struct hlist_node *node);
126 int (*cb)(unsigned int cpu);
129 if (!step->multi_instance) {
130 cb = bringup ? step->startup.single : step->teardown.single;
133 trace_cpuhp_enter(cpu, st->target, state, cb);
135 trace_cpuhp_exit(cpu, st->state, state, ret);
138 cbm = bringup ? step->startup.multi : step->teardown.multi;
142 /* Single invocation for instance add/remove */
144 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
145 ret = cbm(cpu, node);
146 trace_cpuhp_exit(cpu, st->state, state, ret);
150 /* State transition. Invoke on all instances */
152 hlist_for_each(node, &step->list) {
153 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
154 ret = cbm(cpu, node);
155 trace_cpuhp_exit(cpu, st->state, state, ret);
162 /* Rollback the instances if one failed */
163 cbm = !bringup ? step->startup.multi : step->teardown.multi;
167 hlist_for_each(node, &step->list) {
176 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
177 static DEFINE_MUTEX(cpu_add_remove_lock);
178 bool cpuhp_tasks_frozen;
179 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
182 * The following two APIs (cpu_maps_update_begin/done) must be used when
183 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
184 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
185 * hotplug callback (un)registration performed using __register_cpu_notifier()
186 * or __unregister_cpu_notifier().
188 void cpu_maps_update_begin(void)
190 mutex_lock(&cpu_add_remove_lock);
192 EXPORT_SYMBOL(cpu_notifier_register_begin);
194 void cpu_maps_update_done(void)
196 mutex_unlock(&cpu_add_remove_lock);
198 EXPORT_SYMBOL(cpu_notifier_register_done);
200 static RAW_NOTIFIER_HEAD(cpu_chain);
202 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
203 * Should always be manipulated under cpu_add_remove_lock
205 static int cpu_hotplug_disabled;
207 #ifdef CONFIG_HOTPLUG_CPU
210 struct task_struct *active_writer;
211 /* wait queue to wake up the active_writer */
212 wait_queue_head_t wq;
213 /* verifies that no writer will get active while readers are active */
216 * Also blocks the new readers during
217 * an ongoing cpu hotplug operation.
221 #ifdef CONFIG_DEBUG_LOCK_ALLOC
222 struct lockdep_map dep_map;
225 .active_writer = NULL,
226 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
227 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
228 #ifdef CONFIG_DEBUG_LOCK_ALLOC
229 .dep_map = {.name = "cpu_hotplug.lock" },
233 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
234 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
235 #define cpuhp_lock_acquire_tryread() \
236 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
237 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
238 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
241 void get_online_cpus(void)
244 if (cpu_hotplug.active_writer == current)
246 cpuhp_lock_acquire_read();
247 mutex_lock(&cpu_hotplug.lock);
248 atomic_inc(&cpu_hotplug.refcount);
249 mutex_unlock(&cpu_hotplug.lock);
251 EXPORT_SYMBOL_GPL(get_online_cpus);
253 void put_online_cpus(void)
257 if (cpu_hotplug.active_writer == current)
260 refcount = atomic_dec_return(&cpu_hotplug.refcount);
261 if (WARN_ON(refcount < 0)) /* try to fix things up */
262 atomic_inc(&cpu_hotplug.refcount);
264 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
265 wake_up(&cpu_hotplug.wq);
267 cpuhp_lock_release();
270 EXPORT_SYMBOL_GPL(put_online_cpus);
273 * This ensures that the hotplug operation can begin only when the
274 * refcount goes to zero.
276 * Note that during a cpu-hotplug operation, the new readers, if any,
277 * will be blocked by the cpu_hotplug.lock
279 * Since cpu_hotplug_begin() is always called after invoking
280 * cpu_maps_update_begin(), we can be sure that only one writer is active.
282 * Note that theoretically, there is a possibility of a livelock:
283 * - Refcount goes to zero, last reader wakes up the sleeping
285 * - Last reader unlocks the cpu_hotplug.lock.
286 * - A new reader arrives at this moment, bumps up the refcount.
287 * - The writer acquires the cpu_hotplug.lock finds the refcount
288 * non zero and goes to sleep again.
290 * However, this is very difficult to achieve in practice since
291 * get_online_cpus() not an api which is called all that often.
294 void cpu_hotplug_begin(void)
298 cpu_hotplug.active_writer = current;
299 cpuhp_lock_acquire();
302 mutex_lock(&cpu_hotplug.lock);
303 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
304 if (likely(!atomic_read(&cpu_hotplug.refcount)))
306 mutex_unlock(&cpu_hotplug.lock);
309 finish_wait(&cpu_hotplug.wq, &wait);
312 void cpu_hotplug_done(void)
314 cpu_hotplug.active_writer = NULL;
315 mutex_unlock(&cpu_hotplug.lock);
316 cpuhp_lock_release();
320 * Wait for currently running CPU hotplug operations to complete (if any) and
321 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
322 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
323 * hotplug path before performing hotplug operations. So acquiring that lock
324 * guarantees mutual exclusion from any currently running hotplug operations.
326 void cpu_hotplug_disable(void)
328 cpu_maps_update_begin();
329 cpu_hotplug_disabled++;
330 cpu_maps_update_done();
332 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
334 static void __cpu_hotplug_enable(void)
336 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
338 cpu_hotplug_disabled--;
341 void cpu_hotplug_enable(void)
343 cpu_maps_update_begin();
344 __cpu_hotplug_enable();
345 cpu_maps_update_done();
347 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
348 #endif /* CONFIG_HOTPLUG_CPU */
350 /* Need to know about CPUs going up/down? */
351 int register_cpu_notifier(struct notifier_block *nb)
354 cpu_maps_update_begin();
355 ret = raw_notifier_chain_register(&cpu_chain, nb);
356 cpu_maps_update_done();
360 int __register_cpu_notifier(struct notifier_block *nb)
362 return raw_notifier_chain_register(&cpu_chain, nb);
365 static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
368 unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
369 void *hcpu = (void *)(long)cpu;
373 ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
376 return notifier_to_errno(ret);
379 static int cpu_notify(unsigned long val, unsigned int cpu)
381 return __cpu_notify(val, cpu, -1, NULL);
384 static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
386 BUG_ON(cpu_notify(val, cpu));
389 /* Notifier wrappers for transitioning to state machine */
390 static int notify_prepare(unsigned int cpu)
395 ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
398 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
400 __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
405 static int notify_online(unsigned int cpu)
407 cpu_notify(CPU_ONLINE, cpu);
411 static int notify_starting(unsigned int cpu)
413 cpu_notify(CPU_STARTING, cpu);
417 static int bringup_wait_for_ap(unsigned int cpu)
419 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
421 wait_for_completion(&st->done);
425 static int bringup_cpu(unsigned int cpu)
427 struct task_struct *idle = idle_thread_get(cpu);
431 * Some architectures have to walk the irq descriptors to
432 * setup the vector space for the cpu which comes online.
433 * Prevent irq alloc/free across the bringup.
437 /* Arch-specific enabling code. */
438 ret = __cpu_up(cpu, idle);
441 cpu_notify(CPU_UP_CANCELED, cpu);
444 ret = bringup_wait_for_ap(cpu);
445 BUG_ON(!cpu_online(cpu));
450 * Hotplug state machine related functions
452 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
454 for (st->state++; st->state < st->target; st->state++) {
455 struct cpuhp_step *step = cpuhp_get_step(st->state);
457 if (!step->skip_onerr)
458 cpuhp_invoke_callback(cpu, st->state, true, NULL);
462 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
463 enum cpuhp_state target)
465 enum cpuhp_state prev_state = st->state;
468 for (; st->state > target; st->state--) {
469 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL);
471 st->target = prev_state;
472 undo_cpu_down(cpu, st);
479 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
481 for (st->state--; st->state > st->target; st->state--) {
482 struct cpuhp_step *step = cpuhp_get_step(st->state);
484 if (!step->skip_onerr)
485 cpuhp_invoke_callback(cpu, st->state, false, NULL);
489 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
490 enum cpuhp_state target)
492 enum cpuhp_state prev_state = st->state;
495 while (st->state < target) {
497 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL);
499 st->target = prev_state;
500 undo_cpu_up(cpu, st);
508 * The cpu hotplug threads manage the bringup and teardown of the cpus
510 static void cpuhp_create(unsigned int cpu)
512 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
514 init_completion(&st->done);
517 static int cpuhp_should_run(unsigned int cpu)
519 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
521 return st->should_run;
524 /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
525 static int cpuhp_ap_offline(unsigned int cpu, struct cpuhp_cpu_state *st)
527 enum cpuhp_state target = max((int)st->target, CPUHP_TEARDOWN_CPU);
529 return cpuhp_down_callbacks(cpu, st, target);
532 /* Execute the online startup callbacks. Used to be CPU_ONLINE */
533 static int cpuhp_ap_online(unsigned int cpu, struct cpuhp_cpu_state *st)
535 return cpuhp_up_callbacks(cpu, st, st->target);
539 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
540 * callbacks when a state gets [un]installed at runtime.
542 static void cpuhp_thread_fun(unsigned int cpu)
544 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
548 * Paired with the mb() in cpuhp_kick_ap_work and
549 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
555 st->should_run = false;
557 /* Single callback invocation for [un]install ? */
559 if (st->cb_state < CPUHP_AP_ONLINE) {
561 ret = cpuhp_invoke_callback(cpu, st->cb_state,
562 st->bringup, st->node);
565 ret = cpuhp_invoke_callback(cpu, st->cb_state,
566 st->bringup, st->node);
568 } else if (st->rollback) {
569 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
571 undo_cpu_down(cpu, st);
573 * This is a momentary workaround to keep the notifier users
574 * happy. Will go away once we got rid of the notifiers.
576 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
577 st->rollback = false;
579 /* Cannot happen .... */
580 BUG_ON(st->state < CPUHP_AP_ONLINE_IDLE);
582 /* Regular hotplug work */
583 if (st->state < st->target)
584 ret = cpuhp_ap_online(cpu, st);
585 else if (st->state > st->target)
586 ret = cpuhp_ap_offline(cpu, st);
592 /* Invoke a single callback on a remote cpu */
594 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
595 struct hlist_node *node)
597 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
599 if (!cpu_online(cpu))
603 * If we are up and running, use the hotplug thread. For early calls
604 * we invoke the thread function directly.
607 return cpuhp_invoke_callback(cpu, state, bringup, node);
609 st->cb_state = state;
611 st->bringup = bringup;
615 * Make sure the above stores are visible before should_run becomes
616 * true. Paired with the mb() above in cpuhp_thread_fun()
619 st->should_run = true;
620 wake_up_process(st->thread);
621 wait_for_completion(&st->done);
625 /* Regular hotplug invocation of the AP hotplug thread */
626 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state *st)
631 * Make sure the above stores are visible before should_run becomes
632 * true. Paired with the mb() above in cpuhp_thread_fun()
635 st->should_run = true;
636 wake_up_process(st->thread);
639 static int cpuhp_kick_ap_work(unsigned int cpu)
641 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
642 enum cpuhp_state state = st->state;
644 trace_cpuhp_enter(cpu, st->target, state, cpuhp_kick_ap_work);
645 __cpuhp_kick_ap_work(st);
646 wait_for_completion(&st->done);
647 trace_cpuhp_exit(cpu, st->state, state, st->result);
651 static struct smp_hotplug_thread cpuhp_threads = {
652 .store = &cpuhp_state.thread,
653 .create = &cpuhp_create,
654 .thread_should_run = cpuhp_should_run,
655 .thread_fn = cpuhp_thread_fun,
656 .thread_comm = "cpuhp/%u",
660 void __init cpuhp_threads_init(void)
662 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
663 kthread_unpark(this_cpu_read(cpuhp_state.thread));
666 #ifdef CONFIG_HOTPLUG_CPU
667 EXPORT_SYMBOL(register_cpu_notifier);
668 EXPORT_SYMBOL(__register_cpu_notifier);
669 void unregister_cpu_notifier(struct notifier_block *nb)
671 cpu_maps_update_begin();
672 raw_notifier_chain_unregister(&cpu_chain, nb);
673 cpu_maps_update_done();
675 EXPORT_SYMBOL(unregister_cpu_notifier);
677 void __unregister_cpu_notifier(struct notifier_block *nb)
679 raw_notifier_chain_unregister(&cpu_chain, nb);
681 EXPORT_SYMBOL(__unregister_cpu_notifier);
684 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
687 * This function walks all processes, finds a valid mm struct for each one and
688 * then clears a corresponding bit in mm's cpumask. While this all sounds
689 * trivial, there are various non-obvious corner cases, which this function
690 * tries to solve in a safe manner.
692 * Also note that the function uses a somewhat relaxed locking scheme, so it may
693 * be called only for an already offlined CPU.
695 void clear_tasks_mm_cpumask(int cpu)
697 struct task_struct *p;
700 * This function is called after the cpu is taken down and marked
701 * offline, so its not like new tasks will ever get this cpu set in
702 * their mm mask. -- Peter Zijlstra
703 * Thus, we may use rcu_read_lock() here, instead of grabbing
704 * full-fledged tasklist_lock.
706 WARN_ON(cpu_online(cpu));
708 for_each_process(p) {
709 struct task_struct *t;
712 * Main thread might exit, but other threads may still have
713 * a valid mm. Find one.
715 t = find_lock_task_mm(p);
718 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
724 static inline void check_for_tasks(int dead_cpu)
726 struct task_struct *g, *p;
728 read_lock(&tasklist_lock);
729 for_each_process_thread(g, p) {
733 * We do the check with unlocked task_rq(p)->lock.
734 * Order the reading to do not warn about a task,
735 * which was running on this cpu in the past, and
736 * it's just been woken on another cpu.
739 if (task_cpu(p) != dead_cpu)
742 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
743 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
745 read_unlock(&tasklist_lock);
748 static int notify_down_prepare(unsigned int cpu)
750 int err, nr_calls = 0;
752 err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
755 __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
756 pr_warn("%s: attempt to take down CPU %u failed\n",
762 static int notify_dying(unsigned int cpu)
764 cpu_notify(CPU_DYING, cpu);
768 /* Take this CPU down. */
769 static int take_cpu_down(void *_param)
771 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
772 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
773 int err, cpu = smp_processor_id();
775 /* Ensure this CPU doesn't handle any more interrupts. */
776 err = __cpu_disable();
781 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
782 * do this step again.
784 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
786 /* Invoke the former CPU_DYING callbacks */
787 for (; st->state > target; st->state--)
788 cpuhp_invoke_callback(cpu, st->state, false, NULL);
790 /* Give up timekeeping duties */
791 tick_handover_do_timer();
792 /* Park the stopper thread */
793 stop_machine_park(cpu);
797 static int takedown_cpu(unsigned int cpu)
799 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
802 /* Park the smpboot threads */
803 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
804 smpboot_park_threads(cpu);
807 * Prevent irq alloc/free while the dying cpu reorganizes the
808 * interrupt affinities.
813 * So now all preempt/rcu users must observe !cpu_active().
815 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
817 /* CPU refused to die */
819 /* Unpark the hotplug thread so we can rollback there */
820 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
823 BUG_ON(cpu_online(cpu));
826 * The migration_call() CPU_DYING callback will have removed all
827 * runnable tasks from the cpu, there's only the idle task left now
828 * that the migration thread is done doing the stop_machine thing.
830 * Wait for the stop thread to go away.
832 wait_for_completion(&st->done);
833 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
835 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
838 hotplug_cpu__broadcast_tick_pull(cpu);
839 /* This actually kills the CPU. */
842 tick_cleanup_dead_cpu(cpu);
846 static int notify_dead(unsigned int cpu)
848 cpu_notify_nofail(CPU_DEAD, cpu);
849 check_for_tasks(cpu);
853 static void cpuhp_complete_idle_dead(void *arg)
855 struct cpuhp_cpu_state *st = arg;
860 void cpuhp_report_idle_dead(void)
862 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
864 BUG_ON(st->state != CPUHP_AP_OFFLINE);
865 rcu_report_dead(smp_processor_id());
866 st->state = CPUHP_AP_IDLE_DEAD;
868 * We cannot call complete after rcu_report_dead() so we delegate it
871 smp_call_function_single(cpumask_first(cpu_online_mask),
872 cpuhp_complete_idle_dead, st, 0);
876 #define notify_down_prepare NULL
877 #define takedown_cpu NULL
878 #define notify_dead NULL
879 #define notify_dying NULL
882 #ifdef CONFIG_HOTPLUG_CPU
884 /* Requires cpu_add_remove_lock to be held */
885 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
886 enum cpuhp_state target)
888 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
889 int prev_state, ret = 0;
890 bool hasdied = false;
892 if (num_online_cpus() == 1)
895 if (!cpu_present(cpu))
900 cpuhp_tasks_frozen = tasks_frozen;
902 prev_state = st->state;
905 * If the current CPU state is in the range of the AP hotplug thread,
906 * then we need to kick the thread.
908 if (st->state > CPUHP_TEARDOWN_CPU) {
909 ret = cpuhp_kick_ap_work(cpu);
911 * The AP side has done the error rollback already. Just
912 * return the error code..
918 * We might have stopped still in the range of the AP hotplug
919 * thread. Nothing to do anymore.
921 if (st->state > CPUHP_TEARDOWN_CPU)
925 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
926 * to do the further cleanups.
928 ret = cpuhp_down_callbacks(cpu, st, target);
929 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
930 st->target = prev_state;
932 cpuhp_kick_ap_work(cpu);
935 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
938 /* This post dead nonsense must die */
940 cpu_notify_nofail(CPU_POST_DEAD, cpu);
944 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
948 cpu_maps_update_begin();
950 if (cpu_hotplug_disabled) {
955 err = _cpu_down(cpu, 0, target);
958 cpu_maps_update_done();
961 int cpu_down(unsigned int cpu)
963 return do_cpu_down(cpu, CPUHP_OFFLINE);
965 EXPORT_SYMBOL(cpu_down);
966 #endif /*CONFIG_HOTPLUG_CPU*/
969 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
970 * @cpu: cpu that just started
972 * This function calls the cpu_chain notifiers with CPU_STARTING.
973 * It must be called by the arch code on the new cpu, before the new cpu
974 * enables interrupts and before the "boot" cpu returns from __cpu_up().
976 void notify_cpu_starting(unsigned int cpu)
978 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
979 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
981 while (st->state < target) {
983 cpuhp_invoke_callback(cpu, st->state, true, NULL);
988 * Called from the idle task. We need to set active here, so we can kick off
989 * the stopper thread and unpark the smpboot threads. If the target state is
990 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
993 void cpuhp_online_idle(enum cpuhp_state state)
995 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
996 unsigned int cpu = smp_processor_id();
998 /* Happens for the boot cpu */
999 if (state != CPUHP_AP_ONLINE_IDLE)
1002 st->state = CPUHP_AP_ONLINE_IDLE;
1004 /* Unpark the stopper thread and the hotplug thread of this cpu */
1005 stop_machine_unpark(cpu);
1006 kthread_unpark(st->thread);
1008 /* Should we go further up ? */
1009 if (st->target > CPUHP_AP_ONLINE_IDLE)
1010 __cpuhp_kick_ap_work(st);
1012 complete(&st->done);
1015 /* Requires cpu_add_remove_lock to be held */
1016 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1018 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1019 struct task_struct *idle;
1022 cpu_hotplug_begin();
1024 if (!cpu_present(cpu)) {
1030 * The caller of do_cpu_up might have raced with another
1031 * caller. Ignore it for now.
1033 if (st->state >= target)
1036 if (st->state == CPUHP_OFFLINE) {
1037 /* Let it fail before we try to bring the cpu up */
1038 idle = idle_thread_get(cpu);
1040 ret = PTR_ERR(idle);
1045 cpuhp_tasks_frozen = tasks_frozen;
1047 st->target = target;
1049 * If the current CPU state is in the range of the AP hotplug thread,
1050 * then we need to kick the thread once more.
1052 if (st->state > CPUHP_BRINGUP_CPU) {
1053 ret = cpuhp_kick_ap_work(cpu);
1055 * The AP side has done the error rollback already. Just
1056 * return the error code..
1063 * Try to reach the target state. We max out on the BP at
1064 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1065 * responsible for bringing it up to the target state.
1067 target = min((int)target, CPUHP_BRINGUP_CPU);
1068 ret = cpuhp_up_callbacks(cpu, st, target);
1074 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1078 if (!cpu_possible(cpu)) {
1079 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1081 #if defined(CONFIG_IA64)
1082 pr_err("please check additional_cpus= boot parameter\n");
1087 err = try_online_node(cpu_to_node(cpu));
1091 cpu_maps_update_begin();
1093 if (cpu_hotplug_disabled) {
1098 err = _cpu_up(cpu, 0, target);
1100 cpu_maps_update_done();
1104 int cpu_up(unsigned int cpu)
1106 return do_cpu_up(cpu, CPUHP_ONLINE);
1108 EXPORT_SYMBOL_GPL(cpu_up);
1110 #ifdef CONFIG_PM_SLEEP_SMP
1111 static cpumask_var_t frozen_cpus;
1113 int disable_nonboot_cpus(void)
1115 int cpu, first_cpu, error = 0;
1117 cpu_maps_update_begin();
1118 first_cpu = cpumask_first(cpu_online_mask);
1120 * We take down all of the non-boot CPUs in one shot to avoid races
1121 * with the userspace trying to use the CPU hotplug at the same time
1123 cpumask_clear(frozen_cpus);
1125 pr_info("Disabling non-boot CPUs ...\n");
1126 for_each_online_cpu(cpu) {
1127 if (cpu == first_cpu)
1129 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1130 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1131 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1133 cpumask_set_cpu(cpu, frozen_cpus);
1135 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1141 BUG_ON(num_online_cpus() > 1);
1143 pr_err("Non-boot CPUs are not disabled\n");
1146 * Make sure the CPUs won't be enabled by someone else. We need to do
1147 * this even in case of failure as all disable_nonboot_cpus() users are
1148 * supposed to do enable_nonboot_cpus() on the failure path.
1150 cpu_hotplug_disabled++;
1152 cpu_maps_update_done();
1156 void __weak arch_enable_nonboot_cpus_begin(void)
1160 void __weak arch_enable_nonboot_cpus_end(void)
1164 void enable_nonboot_cpus(void)
1168 /* Allow everyone to use the CPU hotplug again */
1169 cpu_maps_update_begin();
1170 __cpu_hotplug_enable();
1171 if (cpumask_empty(frozen_cpus))
1174 pr_info("Enabling non-boot CPUs ...\n");
1176 arch_enable_nonboot_cpus_begin();
1178 for_each_cpu(cpu, frozen_cpus) {
1179 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1180 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1181 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1183 pr_info("CPU%d is up\n", cpu);
1186 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1189 arch_enable_nonboot_cpus_end();
1191 cpumask_clear(frozen_cpus);
1193 cpu_maps_update_done();
1196 static int __init alloc_frozen_cpus(void)
1198 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1202 core_initcall(alloc_frozen_cpus);
1205 * When callbacks for CPU hotplug notifications are being executed, we must
1206 * ensure that the state of the system with respect to the tasks being frozen
1207 * or not, as reported by the notification, remains unchanged *throughout the
1208 * duration* of the execution of the callbacks.
1209 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1211 * This synchronization is implemented by mutually excluding regular CPU
1212 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1213 * Hibernate notifications.
1216 cpu_hotplug_pm_callback(struct notifier_block *nb,
1217 unsigned long action, void *ptr)
1221 case PM_SUSPEND_PREPARE:
1222 case PM_HIBERNATION_PREPARE:
1223 cpu_hotplug_disable();
1226 case PM_POST_SUSPEND:
1227 case PM_POST_HIBERNATION:
1228 cpu_hotplug_enable();
1239 static int __init cpu_hotplug_pm_sync_init(void)
1242 * cpu_hotplug_pm_callback has higher priority than x86
1243 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1244 * to disable cpu hotplug to avoid cpu hotplug race.
1246 pm_notifier(cpu_hotplug_pm_callback, 0);
1249 core_initcall(cpu_hotplug_pm_sync_init);
1251 #endif /* CONFIG_PM_SLEEP_SMP */
1253 #endif /* CONFIG_SMP */
1255 /* Boot processor state steps */
1256 static struct cpuhp_step cpuhp_bp_states[] = {
1259 .startup.single = NULL,
1260 .teardown.single = NULL,
1263 [CPUHP_CREATE_THREADS]= {
1264 .name = "threads:create",
1265 .startup.single = smpboot_create_threads,
1266 .teardown.single = NULL,
1269 [CPUHP_PERF_PREPARE] = {
1270 .name = "perf:prepare",
1271 .startup.single = perf_event_init_cpu,
1272 .teardown.single = perf_event_exit_cpu,
1274 [CPUHP_WORKQUEUE_PREP] = {
1275 .name = "workqueue:prepare",
1276 .startup.single = workqueue_prepare_cpu,
1277 .teardown.single = NULL,
1279 [CPUHP_HRTIMERS_PREPARE] = {
1280 .name = "hrtimers:prepare",
1281 .startup.single = hrtimers_prepare_cpu,
1282 .teardown.single = hrtimers_dead_cpu,
1284 [CPUHP_SMPCFD_PREPARE] = {
1285 .name = "SMPCFD:prepare",
1286 .startup.single = smpcfd_prepare_cpu,
1287 .teardown.single = smpcfd_dead_cpu,
1289 [CPUHP_RCUTREE_PREP] = {
1290 .name = "RCU-tree:prepare",
1291 .startup.single = rcutree_prepare_cpu,
1292 .teardown.single = rcutree_dead_cpu,
1295 * Preparatory and dead notifiers. Will be replaced once the notifiers
1296 * are converted to states.
1298 [CPUHP_NOTIFY_PREPARE] = {
1299 .name = "notify:prepare",
1300 .startup.single = notify_prepare,
1301 .teardown.single = notify_dead,
1306 * On the tear-down path, timers_dead_cpu() must be invoked
1307 * before blk_mq_queue_reinit_notify() from notify_dead(),
1308 * otherwise a RCU stall occurs.
1310 [CPUHP_TIMERS_DEAD] = {
1311 .name = "timers:dead",
1312 .startup.single = NULL,
1313 .teardown.single = timers_dead_cpu,
1315 /* Kicks the plugged cpu into life */
1316 [CPUHP_BRINGUP_CPU] = {
1317 .name = "cpu:bringup",
1318 .startup.single = bringup_cpu,
1319 .teardown.single = NULL,
1322 [CPUHP_AP_SMPCFD_DYING] = {
1323 .name = "SMPCFD:dying",
1324 .startup.single = NULL,
1325 .teardown.single = smpcfd_dying_cpu,
1328 * Handled on controll processor until the plugged processor manages
1331 [CPUHP_TEARDOWN_CPU] = {
1332 .name = "cpu:teardown",
1333 .startup.single = NULL,
1334 .teardown.single = takedown_cpu,
1338 [CPUHP_BRINGUP_CPU] = { },
1342 /* Application processor state steps */
1343 static struct cpuhp_step cpuhp_ap_states[] = {
1345 /* Final state before CPU kills itself */
1346 [CPUHP_AP_IDLE_DEAD] = {
1347 .name = "idle:dead",
1350 * Last state before CPU enters the idle loop to die. Transient state
1351 * for synchronization.
1353 [CPUHP_AP_OFFLINE] = {
1354 .name = "ap:offline",
1357 /* First state is scheduler control. Interrupts are disabled */
1358 [CPUHP_AP_SCHED_STARTING] = {
1359 .name = "sched:starting",
1360 .startup.single = sched_cpu_starting,
1361 .teardown.single = sched_cpu_dying,
1363 [CPUHP_AP_RCUTREE_DYING] = {
1364 .name = "RCU-tree:dying",
1365 .startup.single = NULL,
1366 .teardown.single = rcutree_dying_cpu,
1369 * Low level startup.single/teardown notifiers. Run with interrupts
1370 * disabled. Will be removed once the notifiers are converted to
1373 [CPUHP_AP_NOTIFY_STARTING] = {
1374 .name = "notify:starting",
1375 .startup.single = notify_starting,
1376 .teardown.single = notify_dying,
1380 /* Entry state on starting. Interrupts enabled from here on. Transient
1381 * state for synchronsization */
1382 [CPUHP_AP_ONLINE] = {
1383 .name = "ap:online",
1385 /* Handle smpboot threads park/unpark */
1386 [CPUHP_AP_SMPBOOT_THREADS] = {
1387 .name = "smpboot:threads",
1388 .startup.single = smpboot_unpark_threads,
1389 .teardown.single = NULL,
1391 [CPUHP_AP_PERF_ONLINE] = {
1392 .name = "perf:online",
1393 .startup.single = perf_event_init_cpu,
1394 .teardown.single = perf_event_exit_cpu,
1396 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1397 .name = "workqueue:online",
1398 .startup.single = workqueue_online_cpu,
1399 .teardown.single = workqueue_offline_cpu,
1401 [CPUHP_AP_RCUTREE_ONLINE] = {
1402 .name = "RCU-tree:online",
1403 .startup.single = rcutree_online_cpu,
1404 .teardown.single = rcutree_offline_cpu,
1408 * Online/down_prepare notifiers. Will be removed once the notifiers
1409 * are converted to states.
1411 [CPUHP_AP_NOTIFY_ONLINE] = {
1412 .name = "notify:online",
1413 .startup.single = notify_online,
1414 .teardown.single = notify_down_prepare,
1419 * The dynamically registered state space is here
1423 /* Last state is scheduler control setting the cpu active */
1424 [CPUHP_AP_ACTIVE] = {
1425 .name = "sched:active",
1426 .startup.single = sched_cpu_activate,
1427 .teardown.single = sched_cpu_deactivate,
1431 /* CPU is fully up and running. */
1434 .startup.single = NULL,
1435 .teardown.single = NULL,
1439 /* Sanity check for callbacks */
1440 static int cpuhp_cb_check(enum cpuhp_state state)
1442 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1447 static void cpuhp_store_callbacks(enum cpuhp_state state,
1449 int (*startup)(unsigned int cpu),
1450 int (*teardown)(unsigned int cpu),
1451 bool multi_instance)
1453 /* (Un)Install the callbacks for further cpu hotplug operations */
1454 struct cpuhp_step *sp;
1456 mutex_lock(&cpuhp_state_mutex);
1457 sp = cpuhp_get_step(state);
1458 sp->startup.single = startup;
1459 sp->teardown.single = teardown;
1461 sp->multi_instance = multi_instance;
1462 INIT_HLIST_HEAD(&sp->list);
1463 mutex_unlock(&cpuhp_state_mutex);
1466 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1468 return cpuhp_get_step(state)->teardown.single;
1472 * Call the startup/teardown function for a step either on the AP or
1473 * on the current CPU.
1475 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1476 struct hlist_node *node)
1478 struct cpuhp_step *sp = cpuhp_get_step(state);
1481 if ((bringup && !sp->startup.single) ||
1482 (!bringup && !sp->teardown.single))
1485 * The non AP bound callbacks can fail on bringup. On teardown
1486 * e.g. module removal we crash for now.
1489 if (cpuhp_is_ap_state(state))
1490 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1492 ret = cpuhp_invoke_callback(cpu, state, bringup, node);
1494 ret = cpuhp_invoke_callback(cpu, state, bringup, node);
1496 BUG_ON(ret && !bringup);
1501 * Called from __cpuhp_setup_state on a recoverable failure.
1503 * Note: The teardown callbacks for rollback are not allowed to fail!
1505 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1506 struct hlist_node *node)
1510 /* Roll back the already executed steps on the other cpus */
1511 for_each_present_cpu(cpu) {
1512 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1513 int cpustate = st->state;
1515 if (cpu >= failedcpu)
1518 /* Did we invoke the startup call on that cpu ? */
1519 if (cpustate >= state)
1520 cpuhp_issue_call(cpu, state, false, node);
1525 * Returns a free for dynamic slot assignment of the Online state. The states
1526 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1527 * by having no name assigned.
1529 static int cpuhp_reserve_state(enum cpuhp_state state)
1533 mutex_lock(&cpuhp_state_mutex);
1534 for (i = CPUHP_AP_ONLINE_DYN; i <= CPUHP_AP_ONLINE_DYN_END; i++) {
1535 if (cpuhp_ap_states[i].name)
1538 cpuhp_ap_states[i].name = "Reserved";
1539 mutex_unlock(&cpuhp_state_mutex);
1542 mutex_unlock(&cpuhp_state_mutex);
1543 WARN(1, "No more dynamic states available for CPU hotplug\n");
1547 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1550 struct cpuhp_step *sp;
1554 sp = cpuhp_get_step(state);
1555 if (sp->multi_instance == false)
1560 if (!invoke || !sp->startup.multi)
1564 * Try to call the startup callback for each present cpu
1565 * depending on the hotplug state of the cpu.
1567 for_each_present_cpu(cpu) {
1568 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1569 int cpustate = st->state;
1571 if (cpustate < state)
1574 ret = cpuhp_issue_call(cpu, state, true, node);
1576 if (sp->teardown.multi)
1577 cpuhp_rollback_install(cpu, state, node);
1583 mutex_lock(&cpuhp_state_mutex);
1584 hlist_add_head(node, &sp->list);
1585 mutex_unlock(&cpuhp_state_mutex);
1591 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1594 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1595 * @state: The state to setup
1596 * @invoke: If true, the startup function is invoked for cpus where
1597 * cpu state >= @state
1598 * @startup: startup callback function
1599 * @teardown: teardown callback function
1601 * Returns 0 if successful, otherwise a proper error code
1603 int __cpuhp_setup_state(enum cpuhp_state state,
1604 const char *name, bool invoke,
1605 int (*startup)(unsigned int cpu),
1606 int (*teardown)(unsigned int cpu),
1607 bool multi_instance)
1612 if (cpuhp_cb_check(state) || !name)
1617 /* currently assignments for the ONLINE state are possible */
1618 if (state == CPUHP_AP_ONLINE_DYN) {
1620 ret = cpuhp_reserve_state(state);
1626 cpuhp_store_callbacks(state, name, startup, teardown, multi_instance);
1628 if (!invoke || !startup)
1632 * Try to call the startup callback for each present cpu
1633 * depending on the hotplug state of the cpu.
1635 for_each_present_cpu(cpu) {
1636 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1637 int cpustate = st->state;
1639 if (cpustate < state)
1642 ret = cpuhp_issue_call(cpu, state, true, NULL);
1645 cpuhp_rollback_install(cpu, state, NULL);
1646 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1652 if (!ret && dyn_state)
1656 EXPORT_SYMBOL(__cpuhp_setup_state);
1658 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1659 struct hlist_node *node, bool invoke)
1661 struct cpuhp_step *sp = cpuhp_get_step(state);
1664 BUG_ON(cpuhp_cb_check(state));
1666 if (!sp->multi_instance)
1670 if (!invoke || !cpuhp_get_teardown_cb(state))
1673 * Call the teardown callback for each present cpu depending
1674 * on the hotplug state of the cpu. This function is not
1675 * allowed to fail currently!
1677 for_each_present_cpu(cpu) {
1678 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1679 int cpustate = st->state;
1681 if (cpustate >= state)
1682 cpuhp_issue_call(cpu, state, false, node);
1686 mutex_lock(&cpuhp_state_mutex);
1688 mutex_unlock(&cpuhp_state_mutex);
1693 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1695 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1696 * @state: The state to remove
1697 * @invoke: If true, the teardown function is invoked for cpus where
1698 * cpu state >= @state
1700 * The teardown callback is currently not allowed to fail. Think
1701 * about module removal!
1703 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1705 struct cpuhp_step *sp = cpuhp_get_step(state);
1708 BUG_ON(cpuhp_cb_check(state));
1712 if (sp->multi_instance) {
1713 WARN(!hlist_empty(&sp->list),
1714 "Error: Removing state %d which has instances left.\n",
1719 if (!invoke || !cpuhp_get_teardown_cb(state))
1723 * Call the teardown callback for each present cpu depending
1724 * on the hotplug state of the cpu. This function is not
1725 * allowed to fail currently!
1727 for_each_present_cpu(cpu) {
1728 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1729 int cpustate = st->state;
1731 if (cpustate >= state)
1732 cpuhp_issue_call(cpu, state, false, NULL);
1735 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1738 EXPORT_SYMBOL(__cpuhp_remove_state);
1740 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1741 static ssize_t show_cpuhp_state(struct device *dev,
1742 struct device_attribute *attr, char *buf)
1744 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1746 return sprintf(buf, "%d\n", st->state);
1748 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1750 static ssize_t write_cpuhp_target(struct device *dev,
1751 struct device_attribute *attr,
1752 const char *buf, size_t count)
1754 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1755 struct cpuhp_step *sp;
1758 ret = kstrtoint(buf, 10, &target);
1762 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1763 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1766 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1770 ret = lock_device_hotplug_sysfs();
1774 mutex_lock(&cpuhp_state_mutex);
1775 sp = cpuhp_get_step(target);
1776 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1777 mutex_unlock(&cpuhp_state_mutex);
1781 if (st->state < target)
1782 ret = do_cpu_up(dev->id, target);
1784 ret = do_cpu_down(dev->id, target);
1786 unlock_device_hotplug();
1787 return ret ? ret : count;
1790 static ssize_t show_cpuhp_target(struct device *dev,
1791 struct device_attribute *attr, char *buf)
1793 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1795 return sprintf(buf, "%d\n", st->target);
1797 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1799 static struct attribute *cpuhp_cpu_attrs[] = {
1800 &dev_attr_state.attr,
1801 &dev_attr_target.attr,
1805 static struct attribute_group cpuhp_cpu_attr_group = {
1806 .attrs = cpuhp_cpu_attrs,
1811 static ssize_t show_cpuhp_states(struct device *dev,
1812 struct device_attribute *attr, char *buf)
1814 ssize_t cur, res = 0;
1817 mutex_lock(&cpuhp_state_mutex);
1818 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1819 struct cpuhp_step *sp = cpuhp_get_step(i);
1822 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1827 mutex_unlock(&cpuhp_state_mutex);
1830 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1832 static struct attribute *cpuhp_cpu_root_attrs[] = {
1833 &dev_attr_states.attr,
1837 static struct attribute_group cpuhp_cpu_root_attr_group = {
1838 .attrs = cpuhp_cpu_root_attrs,
1843 static int __init cpuhp_sysfs_init(void)
1847 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1848 &cpuhp_cpu_root_attr_group);
1852 for_each_possible_cpu(cpu) {
1853 struct device *dev = get_cpu_device(cpu);
1857 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1863 device_initcall(cpuhp_sysfs_init);
1867 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1868 * represents all NR_CPUS bits binary values of 1<<nr.
1870 * It is used by cpumask_of() to get a constant address to a CPU
1871 * mask value that has a single bit set only.
1874 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1875 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1876 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1877 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1878 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1880 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1882 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1883 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1884 #if BITS_PER_LONG > 32
1885 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1886 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1889 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
1891 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1892 EXPORT_SYMBOL(cpu_all_bits);
1894 #ifdef CONFIG_INIT_ALL_POSSIBLE
1895 struct cpumask __cpu_possible_mask __read_mostly
1898 struct cpumask __cpu_possible_mask __read_mostly;
1900 EXPORT_SYMBOL(__cpu_possible_mask);
1902 struct cpumask __cpu_online_mask __read_mostly;
1903 EXPORT_SYMBOL(__cpu_online_mask);
1905 struct cpumask __cpu_present_mask __read_mostly;
1906 EXPORT_SYMBOL(__cpu_present_mask);
1908 struct cpumask __cpu_active_mask __read_mostly;
1909 EXPORT_SYMBOL(__cpu_active_mask);
1911 void init_cpu_present(const struct cpumask *src)
1913 cpumask_copy(&__cpu_present_mask, src);
1916 void init_cpu_possible(const struct cpumask *src)
1918 cpumask_copy(&__cpu_possible_mask, src);
1921 void init_cpu_online(const struct cpumask *src)
1923 cpumask_copy(&__cpu_online_mask, src);
1927 * Activate the first processor.
1929 void __init boot_cpu_init(void)
1931 int cpu = smp_processor_id();
1933 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1934 set_cpu_online(cpu, true);
1935 set_cpu_active(cpu, true);
1936 set_cpu_present(cpu, true);
1937 set_cpu_possible(cpu, true);
1941 * Must be called _AFTER_ setting up the per_cpu areas
1943 void __init boot_cpu_state_init(void)
1945 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;