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>
26 #include <trace/events/power.h>
27 #define CREATE_TRACE_POINTS
28 #include <trace/events/cpuhp.h>
33 * cpuhp_cpu_state - Per cpu hotplug state storage
34 * @state: The current cpu state
35 * @target: The target state
37 struct cpuhp_cpu_state {
38 enum cpuhp_state state;
39 enum cpuhp_state target;
42 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
45 * cpuhp_step - Hotplug state machine step
46 * @name: Name of the step
47 * @startup: Startup function of the step
48 * @teardown: Teardown function of the step
49 * @skip_onerr: Do not invoke the functions on error rollback
50 * Will go away once the notifiers are gone
51 * @cant_stop: Bringup/teardown can't be stopped at this step
55 int (*startup)(unsigned int cpu);
56 int (*teardown)(unsigned int cpu);
61 static DEFINE_MUTEX(cpuhp_state_mutex);
62 static struct cpuhp_step cpuhp_bp_states[];
63 static struct cpuhp_step cpuhp_ap_states[];
66 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
67 * @cpu: The cpu for which the callback should be invoked
68 * @step: The step in the state machine
69 * @cb: The callback function to invoke
71 * Called from cpu hotplug and from the state register machinery
73 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step,
74 int (*cb)(unsigned int))
76 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
80 trace_cpuhp_enter(cpu, st->target, step, cb);
82 trace_cpuhp_exit(cpu, st->state, step, ret);
88 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
89 static DEFINE_MUTEX(cpu_add_remove_lock);
90 bool cpuhp_tasks_frozen;
91 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
94 * The following two APIs (cpu_maps_update_begin/done) must be used when
95 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
96 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
97 * hotplug callback (un)registration performed using __register_cpu_notifier()
98 * or __unregister_cpu_notifier().
100 void cpu_maps_update_begin(void)
102 mutex_lock(&cpu_add_remove_lock);
104 EXPORT_SYMBOL(cpu_notifier_register_begin);
106 void cpu_maps_update_done(void)
108 mutex_unlock(&cpu_add_remove_lock);
110 EXPORT_SYMBOL(cpu_notifier_register_done);
112 static RAW_NOTIFIER_HEAD(cpu_chain);
114 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
115 * Should always be manipulated under cpu_add_remove_lock
117 static int cpu_hotplug_disabled;
119 #ifdef CONFIG_HOTPLUG_CPU
122 struct task_struct *active_writer;
123 /* wait queue to wake up the active_writer */
124 wait_queue_head_t wq;
125 /* verifies that no writer will get active while readers are active */
128 * Also blocks the new readers during
129 * an ongoing cpu hotplug operation.
133 #ifdef CONFIG_DEBUG_LOCK_ALLOC
134 struct lockdep_map dep_map;
137 .active_writer = NULL,
138 .wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
139 .lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
141 .dep_map = {.name = "cpu_hotplug.lock" },
145 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
146 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
147 #define cpuhp_lock_acquire_tryread() \
148 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
149 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
150 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
153 void get_online_cpus(void)
156 if (cpu_hotplug.active_writer == current)
158 cpuhp_lock_acquire_read();
159 mutex_lock(&cpu_hotplug.lock);
160 atomic_inc(&cpu_hotplug.refcount);
161 mutex_unlock(&cpu_hotplug.lock);
163 EXPORT_SYMBOL_GPL(get_online_cpus);
165 void put_online_cpus(void)
169 if (cpu_hotplug.active_writer == current)
172 refcount = atomic_dec_return(&cpu_hotplug.refcount);
173 if (WARN_ON(refcount < 0)) /* try to fix things up */
174 atomic_inc(&cpu_hotplug.refcount);
176 if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
177 wake_up(&cpu_hotplug.wq);
179 cpuhp_lock_release();
182 EXPORT_SYMBOL_GPL(put_online_cpus);
185 * This ensures that the hotplug operation can begin only when the
186 * refcount goes to zero.
188 * Note that during a cpu-hotplug operation, the new readers, if any,
189 * will be blocked by the cpu_hotplug.lock
191 * Since cpu_hotplug_begin() is always called after invoking
192 * cpu_maps_update_begin(), we can be sure that only one writer is active.
194 * Note that theoretically, there is a possibility of a livelock:
195 * - Refcount goes to zero, last reader wakes up the sleeping
197 * - Last reader unlocks the cpu_hotplug.lock.
198 * - A new reader arrives at this moment, bumps up the refcount.
199 * - The writer acquires the cpu_hotplug.lock finds the refcount
200 * non zero and goes to sleep again.
202 * However, this is very difficult to achieve in practice since
203 * get_online_cpus() not an api which is called all that often.
206 void cpu_hotplug_begin(void)
210 cpu_hotplug.active_writer = current;
211 cpuhp_lock_acquire();
214 mutex_lock(&cpu_hotplug.lock);
215 prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
216 if (likely(!atomic_read(&cpu_hotplug.refcount)))
218 mutex_unlock(&cpu_hotplug.lock);
221 finish_wait(&cpu_hotplug.wq, &wait);
224 void cpu_hotplug_done(void)
226 cpu_hotplug.active_writer = NULL;
227 mutex_unlock(&cpu_hotplug.lock);
228 cpuhp_lock_release();
232 * Wait for currently running CPU hotplug operations to complete (if any) and
233 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
234 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
235 * hotplug path before performing hotplug operations. So acquiring that lock
236 * guarantees mutual exclusion from any currently running hotplug operations.
238 void cpu_hotplug_disable(void)
240 cpu_maps_update_begin();
241 cpu_hotplug_disabled++;
242 cpu_maps_update_done();
244 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
246 void cpu_hotplug_enable(void)
248 cpu_maps_update_begin();
249 WARN_ON(--cpu_hotplug_disabled < 0);
250 cpu_maps_update_done();
252 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
253 #endif /* CONFIG_HOTPLUG_CPU */
255 /* Need to know about CPUs going up/down? */
256 int register_cpu_notifier(struct notifier_block *nb)
259 cpu_maps_update_begin();
260 ret = raw_notifier_chain_register(&cpu_chain, nb);
261 cpu_maps_update_done();
265 int __register_cpu_notifier(struct notifier_block *nb)
267 return raw_notifier_chain_register(&cpu_chain, nb);
270 static int __cpu_notify(unsigned long val, unsigned int cpu, int nr_to_call,
273 unsigned long mod = cpuhp_tasks_frozen ? CPU_TASKS_FROZEN : 0;
274 void *hcpu = (void *)(long)cpu;
278 ret = __raw_notifier_call_chain(&cpu_chain, val | mod, hcpu, nr_to_call,
281 return notifier_to_errno(ret);
284 static int cpu_notify(unsigned long val, unsigned int cpu)
286 return __cpu_notify(val, cpu, -1, NULL);
289 /* Notifier wrappers for transitioning to state machine */
290 static int notify_prepare(unsigned int cpu)
295 ret = __cpu_notify(CPU_UP_PREPARE, cpu, -1, &nr_calls);
298 printk(KERN_WARNING "%s: attempt to bring up CPU %u failed\n",
300 __cpu_notify(CPU_UP_CANCELED, cpu, nr_calls, NULL);
305 static int notify_online(unsigned int cpu)
307 cpu_notify(CPU_ONLINE, cpu);
311 static int notify_starting(unsigned int cpu)
313 cpu_notify(CPU_STARTING, cpu);
317 static int bringup_cpu(unsigned int cpu)
319 struct task_struct *idle = idle_thread_get(cpu);
322 /* Arch-specific enabling code. */
323 ret = __cpu_up(cpu, idle);
325 cpu_notify(CPU_UP_CANCELED, cpu);
328 BUG_ON(!cpu_online(cpu));
332 #ifdef CONFIG_HOTPLUG_CPU
333 EXPORT_SYMBOL(register_cpu_notifier);
334 EXPORT_SYMBOL(__register_cpu_notifier);
336 void unregister_cpu_notifier(struct notifier_block *nb)
338 cpu_maps_update_begin();
339 raw_notifier_chain_unregister(&cpu_chain, nb);
340 cpu_maps_update_done();
342 EXPORT_SYMBOL(unregister_cpu_notifier);
344 void __unregister_cpu_notifier(struct notifier_block *nb)
346 raw_notifier_chain_unregister(&cpu_chain, nb);
348 EXPORT_SYMBOL(__unregister_cpu_notifier);
351 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
354 * This function walks all processes, finds a valid mm struct for each one and
355 * then clears a corresponding bit in mm's cpumask. While this all sounds
356 * trivial, there are various non-obvious corner cases, which this function
357 * tries to solve in a safe manner.
359 * Also note that the function uses a somewhat relaxed locking scheme, so it may
360 * be called only for an already offlined CPU.
362 void clear_tasks_mm_cpumask(int cpu)
364 struct task_struct *p;
367 * This function is called after the cpu is taken down and marked
368 * offline, so its not like new tasks will ever get this cpu set in
369 * their mm mask. -- Peter Zijlstra
370 * Thus, we may use rcu_read_lock() here, instead of grabbing
371 * full-fledged tasklist_lock.
373 WARN_ON(cpu_online(cpu));
375 for_each_process(p) {
376 struct task_struct *t;
379 * Main thread might exit, but other threads may still have
380 * a valid mm. Find one.
382 t = find_lock_task_mm(p);
385 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
391 static inline void check_for_tasks(int dead_cpu)
393 struct task_struct *g, *p;
395 read_lock(&tasklist_lock);
396 for_each_process_thread(g, p) {
400 * We do the check with unlocked task_rq(p)->lock.
401 * Order the reading to do not warn about a task,
402 * which was running on this cpu in the past, and
403 * it's just been woken on another cpu.
406 if (task_cpu(p) != dead_cpu)
409 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
410 p->comm, task_pid_nr(p), dead_cpu, p->state, p->flags);
412 read_unlock(&tasklist_lock);
415 static void cpu_notify_nofail(unsigned long val, unsigned int cpu)
417 BUG_ON(cpu_notify(val, cpu));
420 static int notify_down_prepare(unsigned int cpu)
422 int err, nr_calls = 0;
424 err = __cpu_notify(CPU_DOWN_PREPARE, cpu, -1, &nr_calls);
427 __cpu_notify(CPU_DOWN_FAILED, cpu, nr_calls, NULL);
428 pr_warn("%s: attempt to take down CPU %u failed\n",
434 static int notify_dying(unsigned int cpu)
436 cpu_notify(CPU_DYING, cpu);
440 /* Take this CPU down. */
441 static int take_cpu_down(void *_param)
443 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
444 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
445 int err, cpu = smp_processor_id();
447 /* Ensure this CPU doesn't handle any more interrupts. */
448 err = __cpu_disable();
452 /* Invoke the former CPU_DYING callbacks */
453 for (; st->state > target; st->state--) {
454 struct cpuhp_step *step = cpuhp_ap_states + st->state;
456 cpuhp_invoke_callback(cpu, st->state, step->teardown);
458 /* Give up timekeeping duties */
459 tick_handover_do_timer();
460 /* Park the stopper thread */
461 stop_machine_park(cpu);
465 static int takedown_cpu(unsigned int cpu)
470 * By now we've cleared cpu_active_mask, wait for all preempt-disabled
471 * and RCU users of this state to go away such that all new such users
474 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
475 * not imply sync_sched(), so wait for both.
477 * Do sync before park smpboot threads to take care the rcu boost case.
479 if (IS_ENABLED(CONFIG_PREEMPT))
480 synchronize_rcu_mult(call_rcu, call_rcu_sched);
485 * Prevent irq alloc/free while the dying cpu reorganizes the
486 * interrupt affinities.
491 * So now all preempt/rcu users must observe !cpu_active().
493 err = stop_machine(take_cpu_down, NULL, cpumask_of(cpu));
495 /* CPU didn't die: tell everyone. Can't complain. */
496 cpu_notify_nofail(CPU_DOWN_FAILED, cpu);
500 BUG_ON(cpu_online(cpu));
503 * The migration_call() CPU_DYING callback will have removed all
504 * runnable tasks from the cpu, there's only the idle task left now
505 * that the migration thread is done doing the stop_machine thing.
507 * Wait for the stop thread to go away.
509 while (!per_cpu(cpu_dead_idle, cpu))
511 smp_mb(); /* Read from cpu_dead_idle before __cpu_die(). */
512 per_cpu(cpu_dead_idle, cpu) = false;
514 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
517 hotplug_cpu__broadcast_tick_pull(cpu);
518 /* This actually kills the CPU. */
521 tick_cleanup_dead_cpu(cpu);
525 static int notify_dead(unsigned int cpu)
527 cpu_notify_nofail(CPU_DEAD, cpu);
528 check_for_tasks(cpu);
533 #define notify_down_prepare NULL
534 #define takedown_cpu NULL
535 #define notify_dead NULL
536 #define notify_dying NULL
539 #ifdef CONFIG_HOTPLUG_CPU
540 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
542 for (st->state++; st->state < st->target; st->state++) {
543 struct cpuhp_step *step = cpuhp_bp_states + st->state;
545 if (!step->skip_onerr)
546 cpuhp_invoke_callback(cpu, st->state, step->startup);
550 /* Requires cpu_add_remove_lock to be held */
551 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
552 enum cpuhp_state target)
554 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
555 int prev_state, ret = 0;
556 bool hasdied = false;
558 if (num_online_cpus() == 1)
561 if (!cpu_present(cpu))
566 cpuhp_tasks_frozen = tasks_frozen;
568 prev_state = st->state;
570 for (; st->state > st->target; st->state--) {
571 struct cpuhp_step *step = cpuhp_bp_states + st->state;
573 ret = cpuhp_invoke_callback(cpu, st->state, step->teardown);
575 st->target = prev_state;
576 undo_cpu_down(cpu, st);
580 hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
583 /* This post dead nonsense must die */
585 cpu_notify_nofail(CPU_POST_DEAD, cpu);
589 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
593 cpu_maps_update_begin();
595 if (cpu_hotplug_disabled) {
600 err = _cpu_down(cpu, 0, target);
603 cpu_maps_update_done();
606 int cpu_down(unsigned int cpu)
608 return do_cpu_down(cpu, CPUHP_OFFLINE);
610 EXPORT_SYMBOL(cpu_down);
611 #endif /*CONFIG_HOTPLUG_CPU*/
614 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
615 * @cpu: cpu that just started
617 * This function calls the cpu_chain notifiers with CPU_STARTING.
618 * It must be called by the arch code on the new cpu, before the new cpu
619 * enables interrupts and before the "boot" cpu returns from __cpu_up().
621 void notify_cpu_starting(unsigned int cpu)
623 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
624 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
626 while (st->state < target) {
627 struct cpuhp_step *step;
630 step = cpuhp_ap_states + st->state;
631 cpuhp_invoke_callback(cpu, st->state, step->startup);
636 * Called from the idle task. We need to set active here, so we can kick off
637 * the stopper thread.
639 static int cpuhp_set_cpu_active(unsigned int cpu)
641 /* The cpu is marked online, set it active now */
642 set_cpu_active(cpu, true);
643 /* Unpark the stopper thread */
644 stop_machine_unpark(cpu);
648 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
650 for (st->state--; st->state > st->target; st->state--) {
651 struct cpuhp_step *step = cpuhp_bp_states + st->state;
653 if (!step->skip_onerr)
654 cpuhp_invoke_callback(cpu, st->state, step->teardown);
658 /* Requires cpu_add_remove_lock to be held */
659 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
661 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
662 struct task_struct *idle;
663 int prev_state, ret = 0;
667 if (!cpu_present(cpu)) {
673 * The caller of do_cpu_up might have raced with another
674 * caller. Ignore it for now.
676 if (st->state >= target)
679 if (st->state == CPUHP_OFFLINE) {
680 /* Let it fail before we try to bring the cpu up */
681 idle = idle_thread_get(cpu);
688 cpuhp_tasks_frozen = tasks_frozen;
690 prev_state = st->state;
692 while (st->state < st->target) {
693 struct cpuhp_step *step;
696 step = cpuhp_bp_states + st->state;
697 ret = cpuhp_invoke_callback(cpu, st->state, step->startup);
699 st->target = prev_state;
700 undo_cpu_up(cpu, st);
709 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
713 if (!cpu_possible(cpu)) {
714 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
716 #if defined(CONFIG_IA64)
717 pr_err("please check additional_cpus= boot parameter\n");
722 err = try_online_node(cpu_to_node(cpu));
726 cpu_maps_update_begin();
728 if (cpu_hotplug_disabled) {
733 err = _cpu_up(cpu, 0, target);
735 cpu_maps_update_done();
739 int cpu_up(unsigned int cpu)
741 return do_cpu_up(cpu, CPUHP_ONLINE);
743 EXPORT_SYMBOL_GPL(cpu_up);
745 #ifdef CONFIG_PM_SLEEP_SMP
746 static cpumask_var_t frozen_cpus;
748 int disable_nonboot_cpus(void)
750 int cpu, first_cpu, error = 0;
752 cpu_maps_update_begin();
753 first_cpu = cpumask_first(cpu_online_mask);
755 * We take down all of the non-boot CPUs in one shot to avoid races
756 * with the userspace trying to use the CPU hotplug at the same time
758 cpumask_clear(frozen_cpus);
760 pr_info("Disabling non-boot CPUs ...\n");
761 for_each_online_cpu(cpu) {
762 if (cpu == first_cpu)
764 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
765 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
766 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
768 cpumask_set_cpu(cpu, frozen_cpus);
770 pr_err("Error taking CPU%d down: %d\n", cpu, error);
776 BUG_ON(num_online_cpus() > 1);
778 pr_err("Non-boot CPUs are not disabled\n");
781 * Make sure the CPUs won't be enabled by someone else. We need to do
782 * this even in case of failure as all disable_nonboot_cpus() users are
783 * supposed to do enable_nonboot_cpus() on the failure path.
785 cpu_hotplug_disabled++;
787 cpu_maps_update_done();
791 void __weak arch_enable_nonboot_cpus_begin(void)
795 void __weak arch_enable_nonboot_cpus_end(void)
799 void enable_nonboot_cpus(void)
803 /* Allow everyone to use the CPU hotplug again */
804 cpu_maps_update_begin();
805 WARN_ON(--cpu_hotplug_disabled < 0);
806 if (cpumask_empty(frozen_cpus))
809 pr_info("Enabling non-boot CPUs ...\n");
811 arch_enable_nonboot_cpus_begin();
813 for_each_cpu(cpu, frozen_cpus) {
814 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
815 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
816 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
818 pr_info("CPU%d is up\n", cpu);
821 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
824 arch_enable_nonboot_cpus_end();
826 cpumask_clear(frozen_cpus);
828 cpu_maps_update_done();
831 static int __init alloc_frozen_cpus(void)
833 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
837 core_initcall(alloc_frozen_cpus);
840 * When callbacks for CPU hotplug notifications are being executed, we must
841 * ensure that the state of the system with respect to the tasks being frozen
842 * or not, as reported by the notification, remains unchanged *throughout the
843 * duration* of the execution of the callbacks.
844 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
846 * This synchronization is implemented by mutually excluding regular CPU
847 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
848 * Hibernate notifications.
851 cpu_hotplug_pm_callback(struct notifier_block *nb,
852 unsigned long action, void *ptr)
856 case PM_SUSPEND_PREPARE:
857 case PM_HIBERNATION_PREPARE:
858 cpu_hotplug_disable();
861 case PM_POST_SUSPEND:
862 case PM_POST_HIBERNATION:
863 cpu_hotplug_enable();
874 static int __init cpu_hotplug_pm_sync_init(void)
877 * cpu_hotplug_pm_callback has higher priority than x86
878 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
879 * to disable cpu hotplug to avoid cpu hotplug race.
881 pm_notifier(cpu_hotplug_pm_callback, 0);
884 core_initcall(cpu_hotplug_pm_sync_init);
886 #endif /* CONFIG_PM_SLEEP_SMP */
888 #endif /* CONFIG_SMP */
890 /* Boot processor state steps */
891 static struct cpuhp_step cpuhp_bp_states[] = {
898 [CPUHP_CREATE_THREADS]= {
899 .name = "threads:create",
900 .startup = smpboot_create_threads,
904 [CPUHP_NOTIFY_PREPARE] = {
905 .name = "notify:prepare",
906 .startup = notify_prepare,
907 .teardown = notify_dead,
911 [CPUHP_BRINGUP_CPU] = {
912 .name = "cpu:bringup",
913 .startup = bringup_cpu,
917 [CPUHP_TEARDOWN_CPU] = {
918 .name = "cpu:teardown",
920 .teardown = takedown_cpu,
923 [CPUHP_CPU_SET_ACTIVE] = {
924 .name = "cpu:active",
925 .startup = cpuhp_set_cpu_active,
928 [CPUHP_SMPBOOT_THREADS] = {
929 .name = "smpboot:threads",
930 .startup = smpboot_unpark_threads,
931 .teardown = smpboot_park_threads,
933 [CPUHP_NOTIFY_ONLINE] = {
934 .name = "notify:online",
935 .startup = notify_online,
936 .teardown = notify_down_prepare,
947 /* Application processor state steps */
948 static struct cpuhp_step cpuhp_ap_states[] = {
950 [CPUHP_AP_NOTIFY_STARTING] = {
951 .name = "notify:starting",
952 .startup = notify_starting,
953 .teardown = notify_dying,
965 /* Sanity check for callbacks */
966 static int cpuhp_cb_check(enum cpuhp_state state)
968 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
973 static bool cpuhp_is_ap_state(enum cpuhp_state state)
975 return (state > CPUHP_AP_OFFLINE && state < CPUHP_AP_ONLINE);
978 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
980 struct cpuhp_step *sp;
982 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
986 static void cpuhp_store_callbacks(enum cpuhp_state state,
988 int (*startup)(unsigned int cpu),
989 int (*teardown)(unsigned int cpu))
991 /* (Un)Install the callbacks for further cpu hotplug operations */
992 struct cpuhp_step *sp;
994 mutex_lock(&cpuhp_state_mutex);
995 sp = cpuhp_get_step(state);
996 sp->startup = startup;
997 sp->teardown = teardown;
999 mutex_unlock(&cpuhp_state_mutex);
1002 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1004 return cpuhp_get_step(state)->teardown;
1007 /* Helper function to run callback on the target cpu */
1008 static void cpuhp_on_cpu_cb(void *__cb)
1010 int (*cb)(unsigned int cpu) = __cb;
1012 BUG_ON(cb(smp_processor_id()));
1016 * Call the startup/teardown function for a step either on the AP or
1017 * on the current CPU.
1019 static int cpuhp_issue_call(int cpu, enum cpuhp_state state,
1020 int (*cb)(unsigned int), bool bringup)
1028 * This invokes the callback directly for now. In a later step we
1029 * convert that to use cpuhp_invoke_callback().
1031 if (cpuhp_is_ap_state(state)) {
1033 * Note, that a function called on the AP is not
1036 if (cpu_online(cpu))
1037 smp_call_function_single(cpu, cpuhp_on_cpu_cb, cb, 1);
1042 * The non AP bound callbacks can fail on bringup. On teardown
1043 * e.g. module removal we crash for now.
1046 BUG_ON(ret && !bringup);
1051 * Called from __cpuhp_setup_state on a recoverable failure.
1053 * Note: The teardown callbacks for rollback are not allowed to fail!
1055 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1056 int (*teardown)(unsigned int cpu))
1063 /* Roll back the already executed steps on the other cpus */
1064 for_each_present_cpu(cpu) {
1065 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1066 int cpustate = st->state;
1068 if (cpu >= failedcpu)
1071 /* Did we invoke the startup call on that cpu ? */
1072 if (cpustate >= state)
1073 cpuhp_issue_call(cpu, state, teardown, false);
1078 * Returns a free for dynamic slot assignment of the Online state. The states
1079 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1080 * by having no name assigned.
1082 static int cpuhp_reserve_state(enum cpuhp_state state)
1086 mutex_lock(&cpuhp_state_mutex);
1087 for (i = CPUHP_ONLINE_DYN; i <= CPUHP_ONLINE_DYN_END; i++) {
1088 if (cpuhp_bp_states[i].name)
1091 cpuhp_bp_states[i].name = "Reserved";
1092 mutex_unlock(&cpuhp_state_mutex);
1095 mutex_unlock(&cpuhp_state_mutex);
1096 WARN(1, "No more dynamic states available for CPU hotplug\n");
1101 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1102 * @state: The state to setup
1103 * @invoke: If true, the startup function is invoked for cpus where
1104 * cpu state >= @state
1105 * @startup: startup callback function
1106 * @teardown: teardown callback function
1108 * Returns 0 if successful, otherwise a proper error code
1110 int __cpuhp_setup_state(enum cpuhp_state state,
1111 const char *name, bool invoke,
1112 int (*startup)(unsigned int cpu),
1113 int (*teardown)(unsigned int cpu))
1118 if (cpuhp_cb_check(state) || !name)
1123 /* currently assignments for the ONLINE state are possible */
1124 if (state == CPUHP_ONLINE_DYN) {
1126 ret = cpuhp_reserve_state(state);
1132 cpuhp_store_callbacks(state, name, startup, teardown);
1134 if (!invoke || !startup)
1138 * Try to call the startup callback for each present cpu
1139 * depending on the hotplug state of the cpu.
1141 for_each_present_cpu(cpu) {
1142 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1143 int cpustate = st->state;
1145 if (cpustate < state)
1148 ret = cpuhp_issue_call(cpu, state, startup, true);
1150 cpuhp_rollback_install(cpu, state, teardown);
1151 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1157 if (!ret && dyn_state)
1161 EXPORT_SYMBOL(__cpuhp_setup_state);
1164 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1165 * @state: The state to remove
1166 * @invoke: If true, the teardown function is invoked for cpus where
1167 * cpu state >= @state
1169 * The teardown callback is currently not allowed to fail. Think
1170 * about module removal!
1172 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1174 int (*teardown)(unsigned int cpu) = cpuhp_get_teardown_cb(state);
1177 BUG_ON(cpuhp_cb_check(state));
1181 if (!invoke || !teardown)
1185 * Call the teardown callback for each present cpu depending
1186 * on the hotplug state of the cpu. This function is not
1187 * allowed to fail currently!
1189 for_each_present_cpu(cpu) {
1190 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1191 int cpustate = st->state;
1193 if (cpustate >= state)
1194 cpuhp_issue_call(cpu, state, teardown, false);
1197 cpuhp_store_callbacks(state, NULL, NULL, NULL);
1200 EXPORT_SYMBOL(__cpuhp_remove_state);
1202 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1203 static ssize_t show_cpuhp_state(struct device *dev,
1204 struct device_attribute *attr, char *buf)
1206 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1208 return sprintf(buf, "%d\n", st->state);
1210 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1212 static ssize_t write_cpuhp_target(struct device *dev,
1213 struct device_attribute *attr,
1214 const char *buf, size_t count)
1216 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1217 struct cpuhp_step *sp;
1220 ret = kstrtoint(buf, 10, &target);
1224 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1225 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1228 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1232 ret = lock_device_hotplug_sysfs();
1236 mutex_lock(&cpuhp_state_mutex);
1237 sp = cpuhp_get_step(target);
1238 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1239 mutex_unlock(&cpuhp_state_mutex);
1243 if (st->state < target)
1244 ret = do_cpu_up(dev->id, target);
1246 ret = do_cpu_down(dev->id, target);
1248 unlock_device_hotplug();
1249 return ret ? ret : count;
1252 static ssize_t show_cpuhp_target(struct device *dev,
1253 struct device_attribute *attr, char *buf)
1255 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1257 return sprintf(buf, "%d\n", st->target);
1259 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1261 static struct attribute *cpuhp_cpu_attrs[] = {
1262 &dev_attr_state.attr,
1263 &dev_attr_target.attr,
1267 static struct attribute_group cpuhp_cpu_attr_group = {
1268 .attrs = cpuhp_cpu_attrs,
1273 static ssize_t show_cpuhp_states(struct device *dev,
1274 struct device_attribute *attr, char *buf)
1276 ssize_t cur, res = 0;
1279 mutex_lock(&cpuhp_state_mutex);
1280 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1281 struct cpuhp_step *sp = cpuhp_get_step(i);
1284 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1289 mutex_unlock(&cpuhp_state_mutex);
1292 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1294 static struct attribute *cpuhp_cpu_root_attrs[] = {
1295 &dev_attr_states.attr,
1299 static struct attribute_group cpuhp_cpu_root_attr_group = {
1300 .attrs = cpuhp_cpu_root_attrs,
1305 static int __init cpuhp_sysfs_init(void)
1309 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1310 &cpuhp_cpu_root_attr_group);
1314 for_each_possible_cpu(cpu) {
1315 struct device *dev = get_cpu_device(cpu);
1319 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1325 device_initcall(cpuhp_sysfs_init);
1329 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1330 * represents all NR_CPUS bits binary values of 1<<nr.
1332 * It is used by cpumask_of() to get a constant address to a CPU
1333 * mask value that has a single bit set only.
1336 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1337 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1338 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1339 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1340 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1342 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1344 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1345 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1346 #if BITS_PER_LONG > 32
1347 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1348 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1351 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
1353 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1354 EXPORT_SYMBOL(cpu_all_bits);
1356 #ifdef CONFIG_INIT_ALL_POSSIBLE
1357 struct cpumask __cpu_possible_mask __read_mostly
1360 struct cpumask __cpu_possible_mask __read_mostly;
1362 EXPORT_SYMBOL(__cpu_possible_mask);
1364 struct cpumask __cpu_online_mask __read_mostly;
1365 EXPORT_SYMBOL(__cpu_online_mask);
1367 struct cpumask __cpu_present_mask __read_mostly;
1368 EXPORT_SYMBOL(__cpu_present_mask);
1370 struct cpumask __cpu_active_mask __read_mostly;
1371 EXPORT_SYMBOL(__cpu_active_mask);
1373 void init_cpu_present(const struct cpumask *src)
1375 cpumask_copy(&__cpu_present_mask, src);
1378 void init_cpu_possible(const struct cpumask *src)
1380 cpumask_copy(&__cpu_possible_mask, src);
1383 void init_cpu_online(const struct cpumask *src)
1385 cpumask_copy(&__cpu_online_mask, src);
1389 * Activate the first processor.
1391 void __init boot_cpu_init(void)
1393 int cpu = smp_processor_id();
1395 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1396 set_cpu_online(cpu, true);
1397 set_cpu_active(cpu, true);
1398 set_cpu_present(cpu, true);
1399 set_cpu_possible(cpu, true);
1403 * Must be called _AFTER_ setting up the per_cpu areas
1405 void __init boot_cpu_state_init(void)
1407 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;