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>
26 #include <linux/relay.h>
27 #include <linux/slab.h>
29 #include <trace/events/power.h>
30 #define CREATE_TRACE_POINTS
31 #include <trace/events/cpuhp.h>
36 * cpuhp_cpu_state - Per cpu hotplug state storage
37 * @state: The current cpu state
38 * @target: The target state
39 * @thread: Pointer to the hotplug thread
40 * @should_run: Thread should execute
41 * @rollback: Perform a rollback
42 * @single: Single callback invocation
43 * @bringup: Single callback bringup or teardown selector
44 * @cb_state: The state for a single callback (install/uninstall)
45 * @result: Result of the operation
46 * @done: Signal completion to the issuer of the task
48 struct cpuhp_cpu_state
{
49 enum cpuhp_state state
;
50 enum cpuhp_state target
;
52 struct task_struct
*thread
;
57 struct hlist_node
*node
;
58 enum cpuhp_state cb_state
;
60 struct completion done
;
64 static DEFINE_PER_CPU(struct cpuhp_cpu_state
, cpuhp_state
);
67 * cpuhp_step - Hotplug state machine step
68 * @name: Name of the step
69 * @startup: Startup function of the step
70 * @teardown: Teardown function of the step
71 * @skip_onerr: Do not invoke the functions on error rollback
72 * Will go away once the notifiers are gone
73 * @cant_stop: Bringup/teardown can't be stopped at this step
78 int (*single
)(unsigned int cpu
);
79 int (*multi
)(unsigned int cpu
,
80 struct hlist_node
*node
);
83 int (*single
)(unsigned int cpu
);
84 int (*multi
)(unsigned int cpu
,
85 struct hlist_node
*node
);
87 struct hlist_head list
;
93 static DEFINE_MUTEX(cpuhp_state_mutex
);
94 static struct cpuhp_step cpuhp_bp_states
[];
95 static struct cpuhp_step cpuhp_ap_states
[];
97 static bool cpuhp_is_ap_state(enum cpuhp_state state
)
100 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
101 * purposes as that state is handled explicitly in cpu_down.
103 return state
> CPUHP_BRINGUP_CPU
&& state
!= CPUHP_TEARDOWN_CPU
;
106 static struct cpuhp_step
*cpuhp_get_step(enum cpuhp_state state
)
108 struct cpuhp_step
*sp
;
110 sp
= cpuhp_is_ap_state(state
) ? cpuhp_ap_states
: cpuhp_bp_states
;
115 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
116 * @cpu: The cpu for which the callback should be invoked
117 * @step: The step in the state machine
118 * @bringup: True if the bringup callback should be invoked
120 * Called from cpu hotplug and from the state register machinery.
122 static int cpuhp_invoke_callback(unsigned int cpu
, enum cpuhp_state state
,
123 bool bringup
, struct hlist_node
*node
)
125 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
126 struct cpuhp_step
*step
= cpuhp_get_step(state
);
127 int (*cbm
)(unsigned int cpu
, struct hlist_node
*node
);
128 int (*cb
)(unsigned int cpu
);
131 if (!step
->multi_instance
) {
132 cb
= bringup
? step
->startup
.single
: step
->teardown
.single
;
135 trace_cpuhp_enter(cpu
, st
->target
, state
, cb
);
137 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
140 cbm
= bringup
? step
->startup
.multi
: step
->teardown
.multi
;
144 /* Single invocation for instance add/remove */
146 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
147 ret
= cbm(cpu
, node
);
148 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
152 /* State transition. Invoke on all instances */
154 hlist_for_each(node
, &step
->list
) {
155 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
156 ret
= cbm(cpu
, node
);
157 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
164 /* Rollback the instances if one failed */
165 cbm
= !bringup
? step
->startup
.multi
: step
->teardown
.multi
;
169 hlist_for_each(node
, &step
->list
) {
178 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
179 static DEFINE_MUTEX(cpu_add_remove_lock
);
180 bool cpuhp_tasks_frozen
;
181 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen
);
184 * The following two APIs (cpu_maps_update_begin/done) must be used when
185 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
186 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
187 * hotplug callback (un)registration performed using __register_cpu_notifier()
188 * or __unregister_cpu_notifier().
190 void cpu_maps_update_begin(void)
192 mutex_lock(&cpu_add_remove_lock
);
194 EXPORT_SYMBOL(cpu_notifier_register_begin
);
196 void cpu_maps_update_done(void)
198 mutex_unlock(&cpu_add_remove_lock
);
200 EXPORT_SYMBOL(cpu_notifier_register_done
);
202 static RAW_NOTIFIER_HEAD(cpu_chain
);
204 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
205 * Should always be manipulated under cpu_add_remove_lock
207 static int cpu_hotplug_disabled
;
209 #ifdef CONFIG_HOTPLUG_CPU
212 struct task_struct
*active_writer
;
213 /* wait queue to wake up the active_writer */
214 wait_queue_head_t wq
;
215 /* verifies that no writer will get active while readers are active */
218 * Also blocks the new readers during
219 * an ongoing cpu hotplug operation.
223 #ifdef CONFIG_DEBUG_LOCK_ALLOC
224 struct lockdep_map dep_map
;
227 .active_writer
= NULL
,
228 .wq
= __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug
.wq
),
229 .lock
= __MUTEX_INITIALIZER(cpu_hotplug
.lock
),
230 #ifdef CONFIG_DEBUG_LOCK_ALLOC
231 .dep_map
= STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug
.dep_map
),
235 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
236 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
237 #define cpuhp_lock_acquire_tryread() \
238 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
239 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
240 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
243 void get_online_cpus(void)
246 if (cpu_hotplug
.active_writer
== current
)
248 cpuhp_lock_acquire_read();
249 mutex_lock(&cpu_hotplug
.lock
);
250 atomic_inc(&cpu_hotplug
.refcount
);
251 mutex_unlock(&cpu_hotplug
.lock
);
253 EXPORT_SYMBOL_GPL(get_online_cpus
);
255 void put_online_cpus(void)
259 if (cpu_hotplug
.active_writer
== current
)
262 refcount
= atomic_dec_return(&cpu_hotplug
.refcount
);
263 if (WARN_ON(refcount
< 0)) /* try to fix things up */
264 atomic_inc(&cpu_hotplug
.refcount
);
266 if (refcount
<= 0 && waitqueue_active(&cpu_hotplug
.wq
))
267 wake_up(&cpu_hotplug
.wq
);
269 cpuhp_lock_release();
272 EXPORT_SYMBOL_GPL(put_online_cpus
);
275 * This ensures that the hotplug operation can begin only when the
276 * refcount goes to zero.
278 * Note that during a cpu-hotplug operation, the new readers, if any,
279 * will be blocked by the cpu_hotplug.lock
281 * Since cpu_hotplug_begin() is always called after invoking
282 * cpu_maps_update_begin(), we can be sure that only one writer is active.
284 * Note that theoretically, there is a possibility of a livelock:
285 * - Refcount goes to zero, last reader wakes up the sleeping
287 * - Last reader unlocks the cpu_hotplug.lock.
288 * - A new reader arrives at this moment, bumps up the refcount.
289 * - The writer acquires the cpu_hotplug.lock finds the refcount
290 * non zero and goes to sleep again.
292 * However, this is very difficult to achieve in practice since
293 * get_online_cpus() not an api which is called all that often.
296 void cpu_hotplug_begin(void)
300 cpu_hotplug
.active_writer
= current
;
301 cpuhp_lock_acquire();
304 mutex_lock(&cpu_hotplug
.lock
);
305 prepare_to_wait(&cpu_hotplug
.wq
, &wait
, TASK_UNINTERRUPTIBLE
);
306 if (likely(!atomic_read(&cpu_hotplug
.refcount
)))
308 mutex_unlock(&cpu_hotplug
.lock
);
311 finish_wait(&cpu_hotplug
.wq
, &wait
);
314 void cpu_hotplug_done(void)
316 cpu_hotplug
.active_writer
= NULL
;
317 mutex_unlock(&cpu_hotplug
.lock
);
318 cpuhp_lock_release();
322 * Wait for currently running CPU hotplug operations to complete (if any) and
323 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
324 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
325 * hotplug path before performing hotplug operations. So acquiring that lock
326 * guarantees mutual exclusion from any currently running hotplug operations.
328 void cpu_hotplug_disable(void)
330 cpu_maps_update_begin();
331 cpu_hotplug_disabled
++;
332 cpu_maps_update_done();
334 EXPORT_SYMBOL_GPL(cpu_hotplug_disable
);
336 static void __cpu_hotplug_enable(void)
338 if (WARN_ONCE(!cpu_hotplug_disabled
, "Unbalanced cpu hotplug enable\n"))
340 cpu_hotplug_disabled
--;
343 void cpu_hotplug_enable(void)
345 cpu_maps_update_begin();
346 __cpu_hotplug_enable();
347 cpu_maps_update_done();
349 EXPORT_SYMBOL_GPL(cpu_hotplug_enable
);
350 #endif /* CONFIG_HOTPLUG_CPU */
352 /* Need to know about CPUs going up/down? */
353 int register_cpu_notifier(struct notifier_block
*nb
)
356 cpu_maps_update_begin();
357 ret
= raw_notifier_chain_register(&cpu_chain
, nb
);
358 cpu_maps_update_done();
362 int __register_cpu_notifier(struct notifier_block
*nb
)
364 return raw_notifier_chain_register(&cpu_chain
, nb
);
367 static int __cpu_notify(unsigned long val
, unsigned int cpu
, int nr_to_call
,
370 unsigned long mod
= cpuhp_tasks_frozen
? CPU_TASKS_FROZEN
: 0;
371 void *hcpu
= (void *)(long)cpu
;
375 ret
= __raw_notifier_call_chain(&cpu_chain
, val
| mod
, hcpu
, nr_to_call
,
378 return notifier_to_errno(ret
);
381 static int cpu_notify(unsigned long val
, unsigned int cpu
)
383 return __cpu_notify(val
, cpu
, -1, NULL
);
386 static void cpu_notify_nofail(unsigned long val
, unsigned int cpu
)
388 BUG_ON(cpu_notify(val
, cpu
));
391 /* Notifier wrappers for transitioning to state machine */
392 static int notify_prepare(unsigned int cpu
)
397 ret
= __cpu_notify(CPU_UP_PREPARE
, cpu
, -1, &nr_calls
);
400 printk(KERN_WARNING
"%s: attempt to bring up CPU %u failed\n",
402 __cpu_notify(CPU_UP_CANCELED
, cpu
, nr_calls
, NULL
);
407 static int notify_online(unsigned int cpu
)
409 cpu_notify(CPU_ONLINE
, cpu
);
413 static int bringup_wait_for_ap(unsigned int cpu
)
415 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
417 wait_for_completion(&st
->done
);
421 static int bringup_cpu(unsigned int cpu
)
423 struct task_struct
*idle
= idle_thread_get(cpu
);
427 * Some architectures have to walk the irq descriptors to
428 * setup the vector space for the cpu which comes online.
429 * Prevent irq alloc/free across the bringup.
433 /* Arch-specific enabling code. */
434 ret
= __cpu_up(cpu
, idle
);
437 cpu_notify(CPU_UP_CANCELED
, cpu
);
440 ret
= bringup_wait_for_ap(cpu
);
441 BUG_ON(!cpu_online(cpu
));
446 * Hotplug state machine related functions
448 static void undo_cpu_down(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
450 for (st
->state
++; st
->state
< st
->target
; st
->state
++) {
451 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
453 if (!step
->skip_onerr
)
454 cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
);
458 static int cpuhp_down_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
459 enum cpuhp_state target
)
461 enum cpuhp_state prev_state
= st
->state
;
464 for (; st
->state
> target
; st
->state
--) {
465 ret
= cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
);
467 st
->target
= prev_state
;
468 undo_cpu_down(cpu
, st
);
475 static void undo_cpu_up(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
477 for (st
->state
--; st
->state
> st
->target
; st
->state
--) {
478 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
480 if (!step
->skip_onerr
)
481 cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
);
485 static int cpuhp_up_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
486 enum cpuhp_state target
)
488 enum cpuhp_state prev_state
= st
->state
;
491 while (st
->state
< target
) {
493 ret
= cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
);
495 st
->target
= prev_state
;
496 undo_cpu_up(cpu
, st
);
504 * The cpu hotplug threads manage the bringup and teardown of the cpus
506 static void cpuhp_create(unsigned int cpu
)
508 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
510 init_completion(&st
->done
);
513 static int cpuhp_should_run(unsigned int cpu
)
515 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
517 return st
->should_run
;
520 /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
521 static int cpuhp_ap_offline(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
523 enum cpuhp_state target
= max((int)st
->target
, CPUHP_TEARDOWN_CPU
);
525 return cpuhp_down_callbacks(cpu
, st
, target
);
528 /* Execute the online startup callbacks. Used to be CPU_ONLINE */
529 static int cpuhp_ap_online(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
531 return cpuhp_up_callbacks(cpu
, st
, st
->target
);
535 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
536 * callbacks when a state gets [un]installed at runtime.
538 static void cpuhp_thread_fun(unsigned int cpu
)
540 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
544 * Paired with the mb() in cpuhp_kick_ap_work and
545 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
551 st
->should_run
= false;
553 /* Single callback invocation for [un]install ? */
555 if (st
->cb_state
< CPUHP_AP_ONLINE
) {
557 ret
= cpuhp_invoke_callback(cpu
, st
->cb_state
,
558 st
->bringup
, st
->node
);
561 ret
= cpuhp_invoke_callback(cpu
, st
->cb_state
,
562 st
->bringup
, st
->node
);
564 } else if (st
->rollback
) {
565 BUG_ON(st
->state
< CPUHP_AP_ONLINE_IDLE
);
567 undo_cpu_down(cpu
, st
);
569 * This is a momentary workaround to keep the notifier users
570 * happy. Will go away once we got rid of the notifiers.
572 cpu_notify_nofail(CPU_DOWN_FAILED
, cpu
);
573 st
->rollback
= false;
575 /* Cannot happen .... */
576 BUG_ON(st
->state
< CPUHP_AP_ONLINE_IDLE
);
578 /* Regular hotplug work */
579 if (st
->state
< st
->target
)
580 ret
= cpuhp_ap_online(cpu
, st
);
581 else if (st
->state
> st
->target
)
582 ret
= cpuhp_ap_offline(cpu
, st
);
588 /* Invoke a single callback on a remote cpu */
590 cpuhp_invoke_ap_callback(int cpu
, enum cpuhp_state state
, bool bringup
,
591 struct hlist_node
*node
)
593 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
595 if (!cpu_online(cpu
))
599 * If we are up and running, use the hotplug thread. For early calls
600 * we invoke the thread function directly.
603 return cpuhp_invoke_callback(cpu
, state
, bringup
, node
);
605 st
->cb_state
= state
;
607 st
->bringup
= bringup
;
611 * Make sure the above stores are visible before should_run becomes
612 * true. Paired with the mb() above in cpuhp_thread_fun()
615 st
->should_run
= true;
616 wake_up_process(st
->thread
);
617 wait_for_completion(&st
->done
);
621 /* Regular hotplug invocation of the AP hotplug thread */
622 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state
*st
)
627 * Make sure the above stores are visible before should_run becomes
628 * true. Paired with the mb() above in cpuhp_thread_fun()
631 st
->should_run
= true;
632 wake_up_process(st
->thread
);
635 static int cpuhp_kick_ap_work(unsigned int cpu
)
637 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
638 enum cpuhp_state state
= st
->state
;
640 trace_cpuhp_enter(cpu
, st
->target
, state
, cpuhp_kick_ap_work
);
641 __cpuhp_kick_ap_work(st
);
642 wait_for_completion(&st
->done
);
643 trace_cpuhp_exit(cpu
, st
->state
, state
, st
->result
);
647 static struct smp_hotplug_thread cpuhp_threads
= {
648 .store
= &cpuhp_state
.thread
,
649 .create
= &cpuhp_create
,
650 .thread_should_run
= cpuhp_should_run
,
651 .thread_fn
= cpuhp_thread_fun
,
652 .thread_comm
= "cpuhp/%u",
656 void __init
cpuhp_threads_init(void)
658 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads
));
659 kthread_unpark(this_cpu_read(cpuhp_state
.thread
));
662 #ifdef CONFIG_HOTPLUG_CPU
663 EXPORT_SYMBOL(register_cpu_notifier
);
664 EXPORT_SYMBOL(__register_cpu_notifier
);
665 void unregister_cpu_notifier(struct notifier_block
*nb
)
667 cpu_maps_update_begin();
668 raw_notifier_chain_unregister(&cpu_chain
, nb
);
669 cpu_maps_update_done();
671 EXPORT_SYMBOL(unregister_cpu_notifier
);
673 void __unregister_cpu_notifier(struct notifier_block
*nb
)
675 raw_notifier_chain_unregister(&cpu_chain
, nb
);
677 EXPORT_SYMBOL(__unregister_cpu_notifier
);
680 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
683 * This function walks all processes, finds a valid mm struct for each one and
684 * then clears a corresponding bit in mm's cpumask. While this all sounds
685 * trivial, there are various non-obvious corner cases, which this function
686 * tries to solve in a safe manner.
688 * Also note that the function uses a somewhat relaxed locking scheme, so it may
689 * be called only for an already offlined CPU.
691 void clear_tasks_mm_cpumask(int cpu
)
693 struct task_struct
*p
;
696 * This function is called after the cpu is taken down and marked
697 * offline, so its not like new tasks will ever get this cpu set in
698 * their mm mask. -- Peter Zijlstra
699 * Thus, we may use rcu_read_lock() here, instead of grabbing
700 * full-fledged tasklist_lock.
702 WARN_ON(cpu_online(cpu
));
704 for_each_process(p
) {
705 struct task_struct
*t
;
708 * Main thread might exit, but other threads may still have
709 * a valid mm. Find one.
711 t
= find_lock_task_mm(p
);
714 cpumask_clear_cpu(cpu
, mm_cpumask(t
->mm
));
720 static inline void check_for_tasks(int dead_cpu
)
722 struct task_struct
*g
, *p
;
724 read_lock(&tasklist_lock
);
725 for_each_process_thread(g
, p
) {
729 * We do the check with unlocked task_rq(p)->lock.
730 * Order the reading to do not warn about a task,
731 * which was running on this cpu in the past, and
732 * it's just been woken on another cpu.
735 if (task_cpu(p
) != dead_cpu
)
738 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
739 p
->comm
, task_pid_nr(p
), dead_cpu
, p
->state
, p
->flags
);
741 read_unlock(&tasklist_lock
);
744 static int notify_down_prepare(unsigned int cpu
)
746 int err
, nr_calls
= 0;
748 err
= __cpu_notify(CPU_DOWN_PREPARE
, cpu
, -1, &nr_calls
);
751 __cpu_notify(CPU_DOWN_FAILED
, cpu
, nr_calls
, NULL
);
752 pr_warn("%s: attempt to take down CPU %u failed\n",
758 /* Take this CPU down. */
759 static int take_cpu_down(void *_param
)
761 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
762 enum cpuhp_state target
= max((int)st
->target
, CPUHP_AP_OFFLINE
);
763 int err
, cpu
= smp_processor_id();
765 /* Ensure this CPU doesn't handle any more interrupts. */
766 err
= __cpu_disable();
771 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
772 * do this step again.
774 WARN_ON(st
->state
!= CPUHP_TEARDOWN_CPU
);
776 /* Invoke the former CPU_DYING callbacks */
777 for (; st
->state
> target
; st
->state
--)
778 cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
);
780 /* Give up timekeeping duties */
781 tick_handover_do_timer();
782 /* Park the stopper thread */
783 stop_machine_park(cpu
);
787 static int takedown_cpu(unsigned int cpu
)
789 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
792 /* Park the smpboot threads */
793 kthread_park(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
794 smpboot_park_threads(cpu
);
797 * Prevent irq alloc/free while the dying cpu reorganizes the
798 * interrupt affinities.
803 * So now all preempt/rcu users must observe !cpu_active().
805 err
= stop_machine(take_cpu_down
, NULL
, cpumask_of(cpu
));
807 /* CPU refused to die */
809 /* Unpark the hotplug thread so we can rollback there */
810 kthread_unpark(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
813 BUG_ON(cpu_online(cpu
));
816 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
817 * runnable tasks from the cpu, there's only the idle task left now
818 * that the migration thread is done doing the stop_machine thing.
820 * Wait for the stop thread to go away.
822 wait_for_completion(&st
->done
);
823 BUG_ON(st
->state
!= CPUHP_AP_IDLE_DEAD
);
825 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
828 hotplug_cpu__broadcast_tick_pull(cpu
);
829 /* This actually kills the CPU. */
832 tick_cleanup_dead_cpu(cpu
);
836 static int notify_dead(unsigned int cpu
)
838 cpu_notify_nofail(CPU_DEAD
, cpu
);
839 check_for_tasks(cpu
);
843 static void cpuhp_complete_idle_dead(void *arg
)
845 struct cpuhp_cpu_state
*st
= arg
;
850 void cpuhp_report_idle_dead(void)
852 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
854 BUG_ON(st
->state
!= CPUHP_AP_OFFLINE
);
855 rcu_report_dead(smp_processor_id());
856 st
->state
= CPUHP_AP_IDLE_DEAD
;
858 * We cannot call complete after rcu_report_dead() so we delegate it
861 smp_call_function_single(cpumask_first(cpu_online_mask
),
862 cpuhp_complete_idle_dead
, st
, 0);
866 #define notify_down_prepare NULL
867 #define takedown_cpu NULL
868 #define notify_dead NULL
871 #ifdef CONFIG_HOTPLUG_CPU
873 /* Requires cpu_add_remove_lock to be held */
874 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
,
875 enum cpuhp_state target
)
877 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
878 int prev_state
, ret
= 0;
879 bool hasdied
= false;
881 if (num_online_cpus() == 1)
884 if (!cpu_present(cpu
))
889 cpuhp_tasks_frozen
= tasks_frozen
;
891 prev_state
= st
->state
;
894 * If the current CPU state is in the range of the AP hotplug thread,
895 * then we need to kick the thread.
897 if (st
->state
> CPUHP_TEARDOWN_CPU
) {
898 ret
= cpuhp_kick_ap_work(cpu
);
900 * The AP side has done the error rollback already. Just
901 * return the error code..
907 * We might have stopped still in the range of the AP hotplug
908 * thread. Nothing to do anymore.
910 if (st
->state
> CPUHP_TEARDOWN_CPU
)
914 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
915 * to do the further cleanups.
917 ret
= cpuhp_down_callbacks(cpu
, st
, target
);
918 if (ret
&& st
->state
> CPUHP_TEARDOWN_CPU
&& st
->state
< prev_state
) {
919 st
->target
= prev_state
;
921 cpuhp_kick_ap_work(cpu
);
924 hasdied
= prev_state
!= st
->state
&& st
->state
== CPUHP_OFFLINE
;
927 /* This post dead nonsense must die */
929 cpu_notify_nofail(CPU_POST_DEAD
, cpu
);
933 static int do_cpu_down(unsigned int cpu
, enum cpuhp_state target
)
937 cpu_maps_update_begin();
939 if (cpu_hotplug_disabled
) {
944 err
= _cpu_down(cpu
, 0, target
);
947 cpu_maps_update_done();
950 int cpu_down(unsigned int cpu
)
952 return do_cpu_down(cpu
, CPUHP_OFFLINE
);
954 EXPORT_SYMBOL(cpu_down
);
955 #endif /*CONFIG_HOTPLUG_CPU*/
958 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
959 * @cpu: cpu that just started
961 * It must be called by the arch code on the new cpu, before the new cpu
962 * enables interrupts and before the "boot" cpu returns from __cpu_up().
964 void notify_cpu_starting(unsigned int cpu
)
966 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
967 enum cpuhp_state target
= min((int)st
->target
, CPUHP_AP_ONLINE
);
969 rcu_cpu_starting(cpu
); /* Enables RCU usage on this CPU. */
970 while (st
->state
< target
) {
972 cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
);
977 * Called from the idle task. We need to set active here, so we can kick off
978 * the stopper thread and unpark the smpboot threads. If the target state is
979 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
982 void cpuhp_online_idle(enum cpuhp_state state
)
984 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
985 unsigned int cpu
= smp_processor_id();
987 /* Happens for the boot cpu */
988 if (state
!= CPUHP_AP_ONLINE_IDLE
)
991 st
->state
= CPUHP_AP_ONLINE_IDLE
;
993 /* Unpark the stopper thread and the hotplug thread of this cpu */
994 stop_machine_unpark(cpu
);
995 kthread_unpark(st
->thread
);
997 /* Should we go further up ? */
998 if (st
->target
> CPUHP_AP_ONLINE_IDLE
)
999 __cpuhp_kick_ap_work(st
);
1001 complete(&st
->done
);
1004 /* Requires cpu_add_remove_lock to be held */
1005 static int _cpu_up(unsigned int cpu
, int tasks_frozen
, enum cpuhp_state target
)
1007 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1008 struct task_struct
*idle
;
1011 cpu_hotplug_begin();
1013 if (!cpu_present(cpu
)) {
1019 * The caller of do_cpu_up might have raced with another
1020 * caller. Ignore it for now.
1022 if (st
->state
>= target
)
1025 if (st
->state
== CPUHP_OFFLINE
) {
1026 /* Let it fail before we try to bring the cpu up */
1027 idle
= idle_thread_get(cpu
);
1029 ret
= PTR_ERR(idle
);
1034 cpuhp_tasks_frozen
= tasks_frozen
;
1036 st
->target
= target
;
1038 * If the current CPU state is in the range of the AP hotplug thread,
1039 * then we need to kick the thread once more.
1041 if (st
->state
> CPUHP_BRINGUP_CPU
) {
1042 ret
= cpuhp_kick_ap_work(cpu
);
1044 * The AP side has done the error rollback already. Just
1045 * return the error code..
1052 * Try to reach the target state. We max out on the BP at
1053 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1054 * responsible for bringing it up to the target state.
1056 target
= min((int)target
, CPUHP_BRINGUP_CPU
);
1057 ret
= cpuhp_up_callbacks(cpu
, st
, target
);
1063 static int do_cpu_up(unsigned int cpu
, enum cpuhp_state target
)
1067 if (!cpu_possible(cpu
)) {
1068 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1070 #if defined(CONFIG_IA64)
1071 pr_err("please check additional_cpus= boot parameter\n");
1076 err
= try_online_node(cpu_to_node(cpu
));
1080 cpu_maps_update_begin();
1082 if (cpu_hotplug_disabled
) {
1087 err
= _cpu_up(cpu
, 0, target
);
1089 cpu_maps_update_done();
1093 int cpu_up(unsigned int cpu
)
1095 return do_cpu_up(cpu
, CPUHP_ONLINE
);
1097 EXPORT_SYMBOL_GPL(cpu_up
);
1099 #ifdef CONFIG_PM_SLEEP_SMP
1100 static cpumask_var_t frozen_cpus
;
1102 int freeze_secondary_cpus(int primary
)
1106 cpu_maps_update_begin();
1107 if (!cpu_online(primary
))
1108 primary
= cpumask_first(cpu_online_mask
);
1110 * We take down all of the non-boot CPUs in one shot to avoid races
1111 * with the userspace trying to use the CPU hotplug at the same time
1113 cpumask_clear(frozen_cpus
);
1115 pr_info("Disabling non-boot CPUs ...\n");
1116 for_each_online_cpu(cpu
) {
1119 trace_suspend_resume(TPS("CPU_OFF"), cpu
, true);
1120 error
= _cpu_down(cpu
, 1, CPUHP_OFFLINE
);
1121 trace_suspend_resume(TPS("CPU_OFF"), cpu
, false);
1123 cpumask_set_cpu(cpu
, frozen_cpus
);
1125 pr_err("Error taking CPU%d down: %d\n", cpu
, error
);
1131 BUG_ON(num_online_cpus() > 1);
1133 pr_err("Non-boot CPUs are not disabled\n");
1136 * Make sure the CPUs won't be enabled by someone else. We need to do
1137 * this even in case of failure as all disable_nonboot_cpus() users are
1138 * supposed to do enable_nonboot_cpus() on the failure path.
1140 cpu_hotplug_disabled
++;
1142 cpu_maps_update_done();
1146 void __weak
arch_enable_nonboot_cpus_begin(void)
1150 void __weak
arch_enable_nonboot_cpus_end(void)
1154 void enable_nonboot_cpus(void)
1158 /* Allow everyone to use the CPU hotplug again */
1159 cpu_maps_update_begin();
1160 __cpu_hotplug_enable();
1161 if (cpumask_empty(frozen_cpus
))
1164 pr_info("Enabling non-boot CPUs ...\n");
1166 arch_enable_nonboot_cpus_begin();
1168 for_each_cpu(cpu
, frozen_cpus
) {
1169 trace_suspend_resume(TPS("CPU_ON"), cpu
, true);
1170 error
= _cpu_up(cpu
, 1, CPUHP_ONLINE
);
1171 trace_suspend_resume(TPS("CPU_ON"), cpu
, false);
1173 pr_info("CPU%d is up\n", cpu
);
1176 pr_warn("Error taking CPU%d up: %d\n", cpu
, error
);
1179 arch_enable_nonboot_cpus_end();
1181 cpumask_clear(frozen_cpus
);
1183 cpu_maps_update_done();
1186 static int __init
alloc_frozen_cpus(void)
1188 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
1192 core_initcall(alloc_frozen_cpus
);
1195 * When callbacks for CPU hotplug notifications are being executed, we must
1196 * ensure that the state of the system with respect to the tasks being frozen
1197 * or not, as reported by the notification, remains unchanged *throughout the
1198 * duration* of the execution of the callbacks.
1199 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1201 * This synchronization is implemented by mutually excluding regular CPU
1202 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1203 * Hibernate notifications.
1206 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
1207 unsigned long action
, void *ptr
)
1211 case PM_SUSPEND_PREPARE
:
1212 case PM_HIBERNATION_PREPARE
:
1213 cpu_hotplug_disable();
1216 case PM_POST_SUSPEND
:
1217 case PM_POST_HIBERNATION
:
1218 cpu_hotplug_enable();
1229 static int __init
cpu_hotplug_pm_sync_init(void)
1232 * cpu_hotplug_pm_callback has higher priority than x86
1233 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1234 * to disable cpu hotplug to avoid cpu hotplug race.
1236 pm_notifier(cpu_hotplug_pm_callback
, 0);
1239 core_initcall(cpu_hotplug_pm_sync_init
);
1241 #endif /* CONFIG_PM_SLEEP_SMP */
1243 #endif /* CONFIG_SMP */
1245 /* Boot processor state steps */
1246 static struct cpuhp_step cpuhp_bp_states
[] = {
1249 .startup
.single
= NULL
,
1250 .teardown
.single
= NULL
,
1253 [CPUHP_CREATE_THREADS
]= {
1254 .name
= "threads:prepare",
1255 .startup
.single
= smpboot_create_threads
,
1256 .teardown
.single
= NULL
,
1259 [CPUHP_PERF_PREPARE
] = {
1260 .name
= "perf:prepare",
1261 .startup
.single
= perf_event_init_cpu
,
1262 .teardown
.single
= perf_event_exit_cpu
,
1264 [CPUHP_WORKQUEUE_PREP
] = {
1265 .name
= "workqueue:prepare",
1266 .startup
.single
= workqueue_prepare_cpu
,
1267 .teardown
.single
= NULL
,
1269 [CPUHP_HRTIMERS_PREPARE
] = {
1270 .name
= "hrtimers:prepare",
1271 .startup
.single
= hrtimers_prepare_cpu
,
1272 .teardown
.single
= hrtimers_dead_cpu
,
1274 [CPUHP_SMPCFD_PREPARE
] = {
1275 .name
= "smpcfd:prepare",
1276 .startup
.single
= smpcfd_prepare_cpu
,
1277 .teardown
.single
= smpcfd_dead_cpu
,
1279 [CPUHP_RELAY_PREPARE
] = {
1280 .name
= "relay:prepare",
1281 .startup
.single
= relay_prepare_cpu
,
1282 .teardown
.single
= NULL
,
1284 [CPUHP_SLAB_PREPARE
] = {
1285 .name
= "slab:prepare",
1286 .startup
.single
= slab_prepare_cpu
,
1287 .teardown
.single
= slab_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
,
1368 /* Entry state on starting. Interrupts enabled from here on. Transient
1369 * state for synchronsization */
1370 [CPUHP_AP_ONLINE
] = {
1371 .name
= "ap:online",
1373 /* Handle smpboot threads park/unpark */
1374 [CPUHP_AP_SMPBOOT_THREADS
] = {
1375 .name
= "smpboot/threads:online",
1376 .startup
.single
= smpboot_unpark_threads
,
1377 .teardown
.single
= NULL
,
1379 [CPUHP_AP_PERF_ONLINE
] = {
1380 .name
= "perf:online",
1381 .startup
.single
= perf_event_init_cpu
,
1382 .teardown
.single
= perf_event_exit_cpu
,
1384 [CPUHP_AP_WORKQUEUE_ONLINE
] = {
1385 .name
= "workqueue:online",
1386 .startup
.single
= workqueue_online_cpu
,
1387 .teardown
.single
= workqueue_offline_cpu
,
1389 [CPUHP_AP_RCUTREE_ONLINE
] = {
1390 .name
= "RCU/tree:online",
1391 .startup
.single
= rcutree_online_cpu
,
1392 .teardown
.single
= rcutree_offline_cpu
,
1396 * Online/down_prepare notifiers. Will be removed once the notifiers
1397 * are converted to states.
1399 [CPUHP_AP_NOTIFY_ONLINE
] = {
1400 .name
= "notify:online",
1401 .startup
.single
= notify_online
,
1402 .teardown
.single
= notify_down_prepare
,
1407 * The dynamically registered state space is here
1411 /* Last state is scheduler control setting the cpu active */
1412 [CPUHP_AP_ACTIVE
] = {
1413 .name
= "sched:active",
1414 .startup
.single
= sched_cpu_activate
,
1415 .teardown
.single
= sched_cpu_deactivate
,
1419 /* CPU is fully up and running. */
1422 .startup
.single
= NULL
,
1423 .teardown
.single
= NULL
,
1427 /* Sanity check for callbacks */
1428 static int cpuhp_cb_check(enum cpuhp_state state
)
1430 if (state
<= CPUHP_OFFLINE
|| state
>= CPUHP_ONLINE
)
1435 static void cpuhp_store_callbacks(enum cpuhp_state state
,
1437 int (*startup
)(unsigned int cpu
),
1438 int (*teardown
)(unsigned int cpu
),
1439 bool multi_instance
)
1441 /* (Un)Install the callbacks for further cpu hotplug operations */
1442 struct cpuhp_step
*sp
;
1444 mutex_lock(&cpuhp_state_mutex
);
1445 sp
= cpuhp_get_step(state
);
1446 sp
->startup
.single
= startup
;
1447 sp
->teardown
.single
= teardown
;
1449 sp
->multi_instance
= multi_instance
;
1450 INIT_HLIST_HEAD(&sp
->list
);
1451 mutex_unlock(&cpuhp_state_mutex
);
1454 static void *cpuhp_get_teardown_cb(enum cpuhp_state state
)
1456 return cpuhp_get_step(state
)->teardown
.single
;
1460 * Call the startup/teardown function for a step either on the AP or
1461 * on the current CPU.
1463 static int cpuhp_issue_call(int cpu
, enum cpuhp_state state
, bool bringup
,
1464 struct hlist_node
*node
)
1466 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1469 if ((bringup
&& !sp
->startup
.single
) ||
1470 (!bringup
&& !sp
->teardown
.single
))
1473 * The non AP bound callbacks can fail on bringup. On teardown
1474 * e.g. module removal we crash for now.
1477 if (cpuhp_is_ap_state(state
))
1478 ret
= cpuhp_invoke_ap_callback(cpu
, state
, bringup
, node
);
1480 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
);
1482 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
);
1484 BUG_ON(ret
&& !bringup
);
1489 * Called from __cpuhp_setup_state on a recoverable failure.
1491 * Note: The teardown callbacks for rollback are not allowed to fail!
1493 static void cpuhp_rollback_install(int failedcpu
, enum cpuhp_state state
,
1494 struct hlist_node
*node
)
1498 /* Roll back the already executed steps on the other cpus */
1499 for_each_present_cpu(cpu
) {
1500 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1501 int cpustate
= st
->state
;
1503 if (cpu
>= failedcpu
)
1506 /* Did we invoke the startup call on that cpu ? */
1507 if (cpustate
>= state
)
1508 cpuhp_issue_call(cpu
, state
, false, node
);
1513 * Returns a free for dynamic slot assignment of the Online state. The states
1514 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1515 * by having no name assigned.
1517 static int cpuhp_reserve_state(enum cpuhp_state state
)
1521 mutex_lock(&cpuhp_state_mutex
);
1522 for (i
= CPUHP_AP_ONLINE_DYN
; i
<= CPUHP_AP_ONLINE_DYN_END
; i
++) {
1523 if (cpuhp_ap_states
[i
].name
)
1526 cpuhp_ap_states
[i
].name
= "Reserved";
1527 mutex_unlock(&cpuhp_state_mutex
);
1530 mutex_unlock(&cpuhp_state_mutex
);
1531 WARN(1, "No more dynamic states available for CPU hotplug\n");
1535 int __cpuhp_state_add_instance(enum cpuhp_state state
, struct hlist_node
*node
,
1538 struct cpuhp_step
*sp
;
1542 sp
= cpuhp_get_step(state
);
1543 if (sp
->multi_instance
== false)
1548 if (!invoke
|| !sp
->startup
.multi
)
1552 * Try to call the startup callback for each present cpu
1553 * depending on the hotplug state of the cpu.
1555 for_each_present_cpu(cpu
) {
1556 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1557 int cpustate
= st
->state
;
1559 if (cpustate
< state
)
1562 ret
= cpuhp_issue_call(cpu
, state
, true, node
);
1564 if (sp
->teardown
.multi
)
1565 cpuhp_rollback_install(cpu
, state
, node
);
1571 mutex_lock(&cpuhp_state_mutex
);
1572 hlist_add_head(node
, &sp
->list
);
1573 mutex_unlock(&cpuhp_state_mutex
);
1579 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance
);
1582 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1583 * @state: The state to setup
1584 * @invoke: If true, the startup function is invoked for cpus where
1585 * cpu state >= @state
1586 * @startup: startup callback function
1587 * @teardown: teardown callback function
1591 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1592 * 0 for all other states
1593 * On failure: proper (negative) error code
1595 int __cpuhp_setup_state(enum cpuhp_state state
,
1596 const char *name
, bool invoke
,
1597 int (*startup
)(unsigned int cpu
),
1598 int (*teardown
)(unsigned int cpu
),
1599 bool multi_instance
)
1604 if (cpuhp_cb_check(state
) || !name
)
1609 /* currently assignments for the ONLINE state are possible */
1610 if (state
== CPUHP_AP_ONLINE_DYN
) {
1612 ret
= cpuhp_reserve_state(state
);
1618 cpuhp_store_callbacks(state
, name
, startup
, teardown
, multi_instance
);
1620 if (!invoke
|| !startup
)
1624 * Try to call the startup callback for each present cpu
1625 * depending on the hotplug state of the cpu.
1627 for_each_present_cpu(cpu
) {
1628 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1629 int cpustate
= st
->state
;
1631 if (cpustate
< state
)
1634 ret
= cpuhp_issue_call(cpu
, state
, true, NULL
);
1637 cpuhp_rollback_install(cpu
, state
, NULL
);
1638 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1644 if (!ret
&& dyn_state
)
1648 EXPORT_SYMBOL(__cpuhp_setup_state
);
1650 int __cpuhp_state_remove_instance(enum cpuhp_state state
,
1651 struct hlist_node
*node
, bool invoke
)
1653 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1656 BUG_ON(cpuhp_cb_check(state
));
1658 if (!sp
->multi_instance
)
1662 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1665 * Call the teardown callback for each present cpu depending
1666 * on the hotplug state of the cpu. This function is not
1667 * allowed to fail currently!
1669 for_each_present_cpu(cpu
) {
1670 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1671 int cpustate
= st
->state
;
1673 if (cpustate
>= state
)
1674 cpuhp_issue_call(cpu
, state
, false, node
);
1678 mutex_lock(&cpuhp_state_mutex
);
1680 mutex_unlock(&cpuhp_state_mutex
);
1685 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance
);
1687 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1688 * @state: The state to remove
1689 * @invoke: If true, the teardown function is invoked for cpus where
1690 * cpu state >= @state
1692 * The teardown callback is currently not allowed to fail. Think
1693 * about module removal!
1695 void __cpuhp_remove_state(enum cpuhp_state state
, bool invoke
)
1697 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1700 BUG_ON(cpuhp_cb_check(state
));
1704 if (sp
->multi_instance
) {
1705 WARN(!hlist_empty(&sp
->list
),
1706 "Error: Removing state %d which has instances left.\n",
1711 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1715 * Call the teardown callback for each present cpu depending
1716 * on the hotplug state of the cpu. This function is not
1717 * allowed to fail currently!
1719 for_each_present_cpu(cpu
) {
1720 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1721 int cpustate
= st
->state
;
1723 if (cpustate
>= state
)
1724 cpuhp_issue_call(cpu
, state
, false, NULL
);
1727 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1730 EXPORT_SYMBOL(__cpuhp_remove_state
);
1732 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1733 static ssize_t
show_cpuhp_state(struct device
*dev
,
1734 struct device_attribute
*attr
, char *buf
)
1736 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1738 return sprintf(buf
, "%d\n", st
->state
);
1740 static DEVICE_ATTR(state
, 0444, show_cpuhp_state
, NULL
);
1742 static ssize_t
write_cpuhp_target(struct device
*dev
,
1743 struct device_attribute
*attr
,
1744 const char *buf
, size_t count
)
1746 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1747 struct cpuhp_step
*sp
;
1750 ret
= kstrtoint(buf
, 10, &target
);
1754 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1755 if (target
< CPUHP_OFFLINE
|| target
> CPUHP_ONLINE
)
1758 if (target
!= CPUHP_OFFLINE
&& target
!= CPUHP_ONLINE
)
1762 ret
= lock_device_hotplug_sysfs();
1766 mutex_lock(&cpuhp_state_mutex
);
1767 sp
= cpuhp_get_step(target
);
1768 ret
= !sp
->name
|| sp
->cant_stop
? -EINVAL
: 0;
1769 mutex_unlock(&cpuhp_state_mutex
);
1773 if (st
->state
< target
)
1774 ret
= do_cpu_up(dev
->id
, target
);
1776 ret
= do_cpu_down(dev
->id
, target
);
1778 unlock_device_hotplug();
1779 return ret
? ret
: count
;
1782 static ssize_t
show_cpuhp_target(struct device
*dev
,
1783 struct device_attribute
*attr
, char *buf
)
1785 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1787 return sprintf(buf
, "%d\n", st
->target
);
1789 static DEVICE_ATTR(target
, 0644, show_cpuhp_target
, write_cpuhp_target
);
1791 static struct attribute
*cpuhp_cpu_attrs
[] = {
1792 &dev_attr_state
.attr
,
1793 &dev_attr_target
.attr
,
1797 static struct attribute_group cpuhp_cpu_attr_group
= {
1798 .attrs
= cpuhp_cpu_attrs
,
1803 static ssize_t
show_cpuhp_states(struct device
*dev
,
1804 struct device_attribute
*attr
, char *buf
)
1806 ssize_t cur
, res
= 0;
1809 mutex_lock(&cpuhp_state_mutex
);
1810 for (i
= CPUHP_OFFLINE
; i
<= CPUHP_ONLINE
; i
++) {
1811 struct cpuhp_step
*sp
= cpuhp_get_step(i
);
1814 cur
= sprintf(buf
, "%3d: %s\n", i
, sp
->name
);
1819 mutex_unlock(&cpuhp_state_mutex
);
1822 static DEVICE_ATTR(states
, 0444, show_cpuhp_states
, NULL
);
1824 static struct attribute
*cpuhp_cpu_root_attrs
[] = {
1825 &dev_attr_states
.attr
,
1829 static struct attribute_group cpuhp_cpu_root_attr_group
= {
1830 .attrs
= cpuhp_cpu_root_attrs
,
1835 static int __init
cpuhp_sysfs_init(void)
1839 ret
= sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
1840 &cpuhp_cpu_root_attr_group
);
1844 for_each_possible_cpu(cpu
) {
1845 struct device
*dev
= get_cpu_device(cpu
);
1849 ret
= sysfs_create_group(&dev
->kobj
, &cpuhp_cpu_attr_group
);
1855 device_initcall(cpuhp_sysfs_init
);
1859 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1860 * represents all NR_CPUS bits binary values of 1<<nr.
1862 * It is used by cpumask_of() to get a constant address to a CPU
1863 * mask value that has a single bit set only.
1866 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1867 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1868 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1869 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1870 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1872 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
1874 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1875 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1876 #if BITS_PER_LONG > 32
1877 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1878 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1881 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
1883 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
1884 EXPORT_SYMBOL(cpu_all_bits
);
1886 #ifdef CONFIG_INIT_ALL_POSSIBLE
1887 struct cpumask __cpu_possible_mask __read_mostly
1890 struct cpumask __cpu_possible_mask __read_mostly
;
1892 EXPORT_SYMBOL(__cpu_possible_mask
);
1894 struct cpumask __cpu_online_mask __read_mostly
;
1895 EXPORT_SYMBOL(__cpu_online_mask
);
1897 struct cpumask __cpu_present_mask __read_mostly
;
1898 EXPORT_SYMBOL(__cpu_present_mask
);
1900 struct cpumask __cpu_active_mask __read_mostly
;
1901 EXPORT_SYMBOL(__cpu_active_mask
);
1903 void init_cpu_present(const struct cpumask
*src
)
1905 cpumask_copy(&__cpu_present_mask
, src
);
1908 void init_cpu_possible(const struct cpumask
*src
)
1910 cpumask_copy(&__cpu_possible_mask
, src
);
1913 void init_cpu_online(const struct cpumask
*src
)
1915 cpumask_copy(&__cpu_online_mask
, src
);
1919 * Activate the first processor.
1921 void __init
boot_cpu_init(void)
1923 int cpu
= smp_processor_id();
1925 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1926 set_cpu_online(cpu
, true);
1927 set_cpu_active(cpu
, true);
1928 set_cpu_present(cpu
, true);
1929 set_cpu_possible(cpu
, true);
1933 * Must be called _AFTER_ setting up the per_cpu areas
1935 void __init
boot_cpu_state_init(void)
1937 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->state
= CPUHP_ONLINE
;