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 * @cb_stat: The state for a single callback (install/uninstall)
40 * @cb: Single callback function (install/uninstall)
41 * @result: Result of the operation
42 * @done: Signal completion to the issuer of the task
44 struct cpuhp_cpu_state
{
45 enum cpuhp_state state
;
46 enum cpuhp_state target
;
48 struct task_struct
*thread
;
50 enum cpuhp_state cb_state
;
51 int (*cb
)(unsigned int cpu
);
53 struct completion done
;
57 static DEFINE_PER_CPU(struct cpuhp_cpu_state
, cpuhp_state
);
60 * cpuhp_step - Hotplug state machine step
61 * @name: Name of the step
62 * @startup: Startup function of the step
63 * @teardown: Teardown function of the step
64 * @skip_onerr: Do not invoke the functions on error rollback
65 * Will go away once the notifiers are gone
66 * @cant_stop: Bringup/teardown can't be stopped at this step
70 int (*startup
)(unsigned int cpu
);
71 int (*teardown
)(unsigned int cpu
);
76 static DEFINE_MUTEX(cpuhp_state_mutex
);
77 static struct cpuhp_step cpuhp_bp_states
[];
78 static struct cpuhp_step cpuhp_ap_states
[];
81 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
82 * @cpu: The cpu for which the callback should be invoked
83 * @step: The step in the state machine
84 * @cb: The callback function to invoke
86 * Called from cpu hotplug and from the state register machinery
88 static int cpuhp_invoke_callback(unsigned int cpu
, enum cpuhp_state step
,
89 int (*cb
)(unsigned int))
91 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
95 trace_cpuhp_enter(cpu
, st
->target
, step
, cb
);
97 trace_cpuhp_exit(cpu
, st
->state
, step
, ret
);
103 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
104 static DEFINE_MUTEX(cpu_add_remove_lock
);
105 bool cpuhp_tasks_frozen
;
106 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen
);
109 * The following two APIs (cpu_maps_update_begin/done) must be used when
110 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
111 * The APIs cpu_notifier_register_begin/done() must be used to protect CPU
112 * hotplug callback (un)registration performed using __register_cpu_notifier()
113 * or __unregister_cpu_notifier().
115 void cpu_maps_update_begin(void)
117 mutex_lock(&cpu_add_remove_lock
);
119 EXPORT_SYMBOL(cpu_notifier_register_begin
);
121 void cpu_maps_update_done(void)
123 mutex_unlock(&cpu_add_remove_lock
);
125 EXPORT_SYMBOL(cpu_notifier_register_done
);
127 static RAW_NOTIFIER_HEAD(cpu_chain
);
129 /* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
130 * Should always be manipulated under cpu_add_remove_lock
132 static int cpu_hotplug_disabled
;
134 #ifdef CONFIG_HOTPLUG_CPU
137 struct task_struct
*active_writer
;
138 /* wait queue to wake up the active_writer */
139 wait_queue_head_t wq
;
140 /* verifies that no writer will get active while readers are active */
143 * Also blocks the new readers during
144 * an ongoing cpu hotplug operation.
148 #ifdef CONFIG_DEBUG_LOCK_ALLOC
149 struct lockdep_map dep_map
;
152 .active_writer
= NULL
,
153 .wq
= __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug
.wq
),
154 .lock
= __MUTEX_INITIALIZER(cpu_hotplug
.lock
),
155 #ifdef CONFIG_DEBUG_LOCK_ALLOC
156 .dep_map
= {.name
= "cpu_hotplug.lock" },
160 /* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
161 #define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
162 #define cpuhp_lock_acquire_tryread() \
163 lock_map_acquire_tryread(&cpu_hotplug.dep_map)
164 #define cpuhp_lock_acquire() lock_map_acquire(&cpu_hotplug.dep_map)
165 #define cpuhp_lock_release() lock_map_release(&cpu_hotplug.dep_map)
168 void get_online_cpus(void)
171 if (cpu_hotplug
.active_writer
== current
)
173 cpuhp_lock_acquire_read();
174 mutex_lock(&cpu_hotplug
.lock
);
175 atomic_inc(&cpu_hotplug
.refcount
);
176 mutex_unlock(&cpu_hotplug
.lock
);
178 EXPORT_SYMBOL_GPL(get_online_cpus
);
180 void put_online_cpus(void)
184 if (cpu_hotplug
.active_writer
== current
)
187 refcount
= atomic_dec_return(&cpu_hotplug
.refcount
);
188 if (WARN_ON(refcount
< 0)) /* try to fix things up */
189 atomic_inc(&cpu_hotplug
.refcount
);
191 if (refcount
<= 0 && waitqueue_active(&cpu_hotplug
.wq
))
192 wake_up(&cpu_hotplug
.wq
);
194 cpuhp_lock_release();
197 EXPORT_SYMBOL_GPL(put_online_cpus
);
200 * This ensures that the hotplug operation can begin only when the
201 * refcount goes to zero.
203 * Note that during a cpu-hotplug operation, the new readers, if any,
204 * will be blocked by the cpu_hotplug.lock
206 * Since cpu_hotplug_begin() is always called after invoking
207 * cpu_maps_update_begin(), we can be sure that only one writer is active.
209 * Note that theoretically, there is a possibility of a livelock:
210 * - Refcount goes to zero, last reader wakes up the sleeping
212 * - Last reader unlocks the cpu_hotplug.lock.
213 * - A new reader arrives at this moment, bumps up the refcount.
214 * - The writer acquires the cpu_hotplug.lock finds the refcount
215 * non zero and goes to sleep again.
217 * However, this is very difficult to achieve in practice since
218 * get_online_cpus() not an api which is called all that often.
221 void cpu_hotplug_begin(void)
225 cpu_hotplug
.active_writer
= current
;
226 cpuhp_lock_acquire();
229 mutex_lock(&cpu_hotplug
.lock
);
230 prepare_to_wait(&cpu_hotplug
.wq
, &wait
, TASK_UNINTERRUPTIBLE
);
231 if (likely(!atomic_read(&cpu_hotplug
.refcount
)))
233 mutex_unlock(&cpu_hotplug
.lock
);
236 finish_wait(&cpu_hotplug
.wq
, &wait
);
239 void cpu_hotplug_done(void)
241 cpu_hotplug
.active_writer
= NULL
;
242 mutex_unlock(&cpu_hotplug
.lock
);
243 cpuhp_lock_release();
247 * Wait for currently running CPU hotplug operations to complete (if any) and
248 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
249 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
250 * hotplug path before performing hotplug operations. So acquiring that lock
251 * guarantees mutual exclusion from any currently running hotplug operations.
253 void cpu_hotplug_disable(void)
255 cpu_maps_update_begin();
256 cpu_hotplug_disabled
++;
257 cpu_maps_update_done();
259 EXPORT_SYMBOL_GPL(cpu_hotplug_disable
);
261 void cpu_hotplug_enable(void)
263 cpu_maps_update_begin();
264 WARN_ON(--cpu_hotplug_disabled
< 0);
265 cpu_maps_update_done();
267 EXPORT_SYMBOL_GPL(cpu_hotplug_enable
);
268 #endif /* CONFIG_HOTPLUG_CPU */
270 /* Need to know about CPUs going up/down? */
271 int register_cpu_notifier(struct notifier_block
*nb
)
274 cpu_maps_update_begin();
275 ret
= raw_notifier_chain_register(&cpu_chain
, nb
);
276 cpu_maps_update_done();
280 int __register_cpu_notifier(struct notifier_block
*nb
)
282 return raw_notifier_chain_register(&cpu_chain
, nb
);
285 static int __cpu_notify(unsigned long val
, unsigned int cpu
, int nr_to_call
,
288 unsigned long mod
= cpuhp_tasks_frozen
? CPU_TASKS_FROZEN
: 0;
289 void *hcpu
= (void *)(long)cpu
;
293 ret
= __raw_notifier_call_chain(&cpu_chain
, val
| mod
, hcpu
, nr_to_call
,
296 return notifier_to_errno(ret
);
299 static int cpu_notify(unsigned long val
, unsigned int cpu
)
301 return __cpu_notify(val
, cpu
, -1, NULL
);
304 /* Notifier wrappers for transitioning to state machine */
305 static int notify_prepare(unsigned int cpu
)
310 ret
= __cpu_notify(CPU_UP_PREPARE
, cpu
, -1, &nr_calls
);
313 printk(KERN_WARNING
"%s: attempt to bring up CPU %u failed\n",
315 __cpu_notify(CPU_UP_CANCELED
, cpu
, nr_calls
, NULL
);
320 static int notify_online(unsigned int cpu
)
322 cpu_notify(CPU_ONLINE
, cpu
);
326 static int notify_starting(unsigned int cpu
)
328 cpu_notify(CPU_STARTING
, cpu
);
332 static int bringup_wait_for_ap(unsigned int cpu
)
334 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
336 wait_for_completion(&st
->done
);
340 static int bringup_cpu(unsigned int cpu
)
342 struct task_struct
*idle
= idle_thread_get(cpu
);
345 /* Arch-specific enabling code. */
346 ret
= __cpu_up(cpu
, idle
);
348 cpu_notify(CPU_UP_CANCELED
, cpu
);
351 ret
= bringup_wait_for_ap(cpu
);
352 BUG_ON(!cpu_online(cpu
));
357 * Hotplug state machine related functions
359 static void undo_cpu_down(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
360 struct cpuhp_step
*steps
)
362 for (st
->state
++; st
->state
< st
->target
; st
->state
++) {
363 struct cpuhp_step
*step
= steps
+ st
->state
;
365 if (!step
->skip_onerr
)
366 cpuhp_invoke_callback(cpu
, st
->state
, step
->startup
);
370 static int cpuhp_down_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
371 struct cpuhp_step
*steps
, enum cpuhp_state target
)
373 enum cpuhp_state prev_state
= st
->state
;
376 for (; st
->state
> target
; st
->state
--) {
377 struct cpuhp_step
*step
= steps
+ st
->state
;
379 ret
= cpuhp_invoke_callback(cpu
, st
->state
, step
->teardown
);
381 st
->target
= prev_state
;
382 undo_cpu_down(cpu
, st
, steps
);
389 static void undo_cpu_up(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
390 struct cpuhp_step
*steps
)
392 for (st
->state
--; st
->state
> st
->target
; st
->state
--) {
393 struct cpuhp_step
*step
= steps
+ st
->state
;
395 if (!step
->skip_onerr
)
396 cpuhp_invoke_callback(cpu
, st
->state
, step
->teardown
);
400 static int cpuhp_up_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
401 struct cpuhp_step
*steps
, enum cpuhp_state target
)
403 enum cpuhp_state prev_state
= st
->state
;
406 while (st
->state
< target
) {
407 struct cpuhp_step
*step
;
410 step
= steps
+ st
->state
;
411 ret
= cpuhp_invoke_callback(cpu
, st
->state
, step
->startup
);
413 st
->target
= prev_state
;
414 undo_cpu_up(cpu
, st
, steps
);
422 * The cpu hotplug threads manage the bringup and teardown of the cpus
424 static void cpuhp_create(unsigned int cpu
)
426 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
428 init_completion(&st
->done
);
431 static int cpuhp_should_run(unsigned int cpu
)
433 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
435 return st
->should_run
;
438 /* Execute the teardown callbacks. Used to be CPU_DOWN_PREPARE */
439 static int cpuhp_ap_offline(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
441 enum cpuhp_state target
= max((int)st
->target
, CPUHP_TEARDOWN_CPU
);
443 return cpuhp_down_callbacks(cpu
, st
, cpuhp_ap_states
, target
);
446 /* Execute the online startup callbacks. Used to be CPU_ONLINE */
447 static int cpuhp_ap_online(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
449 return cpuhp_up_callbacks(cpu
, st
, cpuhp_ap_states
, st
->target
);
453 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
454 * callbacks when a state gets [un]installed at runtime.
456 static void cpuhp_thread_fun(unsigned int cpu
)
458 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
462 * Paired with the mb() in cpuhp_kick_ap_work and
463 * cpuhp_invoke_ap_callback, so the work set is consistent visible.
469 st
->should_run
= false;
471 /* Single callback invocation for [un]install ? */
473 if (st
->cb_state
< CPUHP_AP_ONLINE
) {
475 ret
= cpuhp_invoke_callback(cpu
, st
->cb_state
, st
->cb
);
478 ret
= cpuhp_invoke_callback(cpu
, st
->cb_state
, st
->cb
);
481 /* Cannot happen .... */
482 BUG_ON(st
->state
< CPUHP_AP_ONLINE_IDLE
);
484 /* Regular hotplug work */
485 if (st
->state
< st
->target
)
486 ret
= cpuhp_ap_online(cpu
, st
);
487 else if (st
->state
> st
->target
)
488 ret
= cpuhp_ap_offline(cpu
, st
);
494 /* Invoke a single callback on a remote cpu */
495 static int cpuhp_invoke_ap_callback(int cpu
, enum cpuhp_state state
,
496 int (*cb
)(unsigned int))
498 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
500 if (!cpu_online(cpu
))
503 st
->cb_state
= state
;
506 * Make sure the above stores are visible before should_run becomes
507 * true. Paired with the mb() above in cpuhp_thread_fun()
510 st
->should_run
= true;
511 wake_up_process(st
->thread
);
512 wait_for_completion(&st
->done
);
516 /* Regular hotplug invocation of the AP hotplug thread */
517 static void __cpuhp_kick_ap_work(struct cpuhp_cpu_state
*st
)
522 * Make sure the above stores are visible before should_run becomes
523 * true. Paired with the mb() above in cpuhp_thread_fun()
526 st
->should_run
= true;
527 wake_up_process(st
->thread
);
530 static int cpuhp_kick_ap_work(unsigned int cpu
)
532 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
533 enum cpuhp_state state
= st
->state
;
535 trace_cpuhp_enter(cpu
, st
->target
, state
, cpuhp_kick_ap_work
);
536 __cpuhp_kick_ap_work(st
);
537 wait_for_completion(&st
->done
);
538 trace_cpuhp_exit(cpu
, st
->state
, state
, st
->result
);
542 static struct smp_hotplug_thread cpuhp_threads
= {
543 .store
= &cpuhp_state
.thread
,
544 .create
= &cpuhp_create
,
545 .thread_should_run
= cpuhp_should_run
,
546 .thread_fn
= cpuhp_thread_fun
,
547 .thread_comm
= "cpuhp/%u",
551 void __init
cpuhp_threads_init(void)
553 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads
));
554 kthread_unpark(this_cpu_read(cpuhp_state
.thread
));
557 #ifdef CONFIG_HOTPLUG_CPU
558 EXPORT_SYMBOL(register_cpu_notifier
);
559 EXPORT_SYMBOL(__register_cpu_notifier
);
560 void unregister_cpu_notifier(struct notifier_block
*nb
)
562 cpu_maps_update_begin();
563 raw_notifier_chain_unregister(&cpu_chain
, nb
);
564 cpu_maps_update_done();
566 EXPORT_SYMBOL(unregister_cpu_notifier
);
568 void __unregister_cpu_notifier(struct notifier_block
*nb
)
570 raw_notifier_chain_unregister(&cpu_chain
, nb
);
572 EXPORT_SYMBOL(__unregister_cpu_notifier
);
575 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
578 * This function walks all processes, finds a valid mm struct for each one and
579 * then clears a corresponding bit in mm's cpumask. While this all sounds
580 * trivial, there are various non-obvious corner cases, which this function
581 * tries to solve in a safe manner.
583 * Also note that the function uses a somewhat relaxed locking scheme, so it may
584 * be called only for an already offlined CPU.
586 void clear_tasks_mm_cpumask(int cpu
)
588 struct task_struct
*p
;
591 * This function is called after the cpu is taken down and marked
592 * offline, so its not like new tasks will ever get this cpu set in
593 * their mm mask. -- Peter Zijlstra
594 * Thus, we may use rcu_read_lock() here, instead of grabbing
595 * full-fledged tasklist_lock.
597 WARN_ON(cpu_online(cpu
));
599 for_each_process(p
) {
600 struct task_struct
*t
;
603 * Main thread might exit, but other threads may still have
604 * a valid mm. Find one.
606 t
= find_lock_task_mm(p
);
609 cpumask_clear_cpu(cpu
, mm_cpumask(t
->mm
));
615 static inline void check_for_tasks(int dead_cpu
)
617 struct task_struct
*g
, *p
;
619 read_lock(&tasklist_lock
);
620 for_each_process_thread(g
, p
) {
624 * We do the check with unlocked task_rq(p)->lock.
625 * Order the reading to do not warn about a task,
626 * which was running on this cpu in the past, and
627 * it's just been woken on another cpu.
630 if (task_cpu(p
) != dead_cpu
)
633 pr_warn("Task %s (pid=%d) is on cpu %d (state=%ld, flags=%x)\n",
634 p
->comm
, task_pid_nr(p
), dead_cpu
, p
->state
, p
->flags
);
636 read_unlock(&tasklist_lock
);
639 static void cpu_notify_nofail(unsigned long val
, unsigned int cpu
)
641 BUG_ON(cpu_notify(val
, cpu
));
644 static int notify_down_prepare(unsigned int cpu
)
646 int err
, nr_calls
= 0;
648 err
= __cpu_notify(CPU_DOWN_PREPARE
, cpu
, -1, &nr_calls
);
651 __cpu_notify(CPU_DOWN_FAILED
, cpu
, nr_calls
, NULL
);
652 pr_warn("%s: attempt to take down CPU %u failed\n",
658 static int notify_dying(unsigned int cpu
)
660 cpu_notify(CPU_DYING
, cpu
);
664 /* Take this CPU down. */
665 static int take_cpu_down(void *_param
)
667 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
668 enum cpuhp_state target
= max((int)st
->target
, CPUHP_AP_OFFLINE
);
669 int err
, cpu
= smp_processor_id();
671 /* Ensure this CPU doesn't handle any more interrupts. */
672 err
= __cpu_disable();
676 /* Invoke the former CPU_DYING callbacks */
677 for (; st
->state
> target
; st
->state
--) {
678 struct cpuhp_step
*step
= cpuhp_ap_states
+ st
->state
;
680 cpuhp_invoke_callback(cpu
, st
->state
, step
->teardown
);
682 /* Give up timekeeping duties */
683 tick_handover_do_timer();
684 /* Park the stopper thread */
685 stop_machine_park(cpu
);
689 static int takedown_cpu(unsigned int cpu
)
691 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
695 * By now we've cleared cpu_active_mask, wait for all preempt-disabled
696 * and RCU users of this state to go away such that all new such users
699 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
700 * not imply sync_sched(), so wait for both.
702 * Do sync before park smpboot threads to take care the rcu boost case.
704 if (IS_ENABLED(CONFIG_PREEMPT
))
705 synchronize_rcu_mult(call_rcu
, call_rcu_sched
);
709 /* Park the hotplug thread */
710 kthread_park(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
713 * Prevent irq alloc/free while the dying cpu reorganizes the
714 * interrupt affinities.
719 * So now all preempt/rcu users must observe !cpu_active().
721 err
= stop_machine(take_cpu_down
, NULL
, cpumask_of(cpu
));
723 /* CPU didn't die: tell everyone. Can't complain. */
724 cpu_notify_nofail(CPU_DOWN_FAILED
, cpu
);
728 BUG_ON(cpu_online(cpu
));
731 * The migration_call() CPU_DYING callback will have removed all
732 * runnable tasks from the cpu, there's only the idle task left now
733 * that the migration thread is done doing the stop_machine thing.
735 * Wait for the stop thread to go away.
737 wait_for_completion(&st
->done
);
738 BUG_ON(st
->state
!= CPUHP_AP_IDLE_DEAD
);
740 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
743 hotplug_cpu__broadcast_tick_pull(cpu
);
744 /* This actually kills the CPU. */
747 tick_cleanup_dead_cpu(cpu
);
751 static int notify_dead(unsigned int cpu
)
753 cpu_notify_nofail(CPU_DEAD
, cpu
);
754 check_for_tasks(cpu
);
758 void cpuhp_report_idle_dead(void)
760 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
762 BUG_ON(st
->state
!= CPUHP_AP_OFFLINE
);
763 st
->state
= CPUHP_AP_IDLE_DEAD
;
765 rcu_report_dead(smp_processor_id());
769 #define notify_down_prepare NULL
770 #define takedown_cpu NULL
771 #define notify_dead NULL
772 #define notify_dying NULL
775 #ifdef CONFIG_HOTPLUG_CPU
777 /* Requires cpu_add_remove_lock to be held */
778 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
,
779 enum cpuhp_state target
)
781 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
782 int prev_state
, ret
= 0;
783 bool hasdied
= false;
785 if (num_online_cpus() == 1)
788 if (!cpu_present(cpu
))
793 cpuhp_tasks_frozen
= tasks_frozen
;
795 prev_state
= st
->state
;
798 * If the current CPU state is in the range of the AP hotplug thread,
799 * then we need to kick the thread.
801 if (st
->state
> CPUHP_TEARDOWN_CPU
) {
802 ret
= cpuhp_kick_ap_work(cpu
);
804 * The AP side has done the error rollback already. Just
805 * return the error code..
811 * We might have stopped still in the range of the AP hotplug
812 * thread. Nothing to do anymore.
814 if (st
->state
> CPUHP_TEARDOWN_CPU
)
818 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
819 * to do the further cleanups.
821 ret
= cpuhp_down_callbacks(cpu
, st
, cpuhp_bp_states
, target
);
823 hasdied
= prev_state
!= st
->state
&& st
->state
== CPUHP_OFFLINE
;
826 /* This post dead nonsense must die */
828 cpu_notify_nofail(CPU_POST_DEAD
, cpu
);
832 static int do_cpu_down(unsigned int cpu
, enum cpuhp_state target
)
836 cpu_maps_update_begin();
838 if (cpu_hotplug_disabled
) {
843 err
= _cpu_down(cpu
, 0, target
);
846 cpu_maps_update_done();
849 int cpu_down(unsigned int cpu
)
851 return do_cpu_down(cpu
, CPUHP_OFFLINE
);
853 EXPORT_SYMBOL(cpu_down
);
854 #endif /*CONFIG_HOTPLUG_CPU*/
857 * notify_cpu_starting(cpu) - call the CPU_STARTING notifiers
858 * @cpu: cpu that just started
860 * This function calls the cpu_chain notifiers with CPU_STARTING.
861 * It must be called by the arch code on the new cpu, before the new cpu
862 * enables interrupts and before the "boot" cpu returns from __cpu_up().
864 void notify_cpu_starting(unsigned int cpu
)
866 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
867 enum cpuhp_state target
= min((int)st
->target
, CPUHP_AP_ONLINE
);
869 while (st
->state
< target
) {
870 struct cpuhp_step
*step
;
873 step
= cpuhp_ap_states
+ st
->state
;
874 cpuhp_invoke_callback(cpu
, st
->state
, step
->startup
);
879 * Called from the idle task. We need to set active here, so we can kick off
880 * the stopper thread and unpark the smpboot threads. If the target state is
881 * beyond CPUHP_AP_ONLINE_IDLE we kick cpuhp thread and let it bring up the
884 void cpuhp_online_idle(enum cpuhp_state state
)
886 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
887 unsigned int cpu
= smp_processor_id();
889 /* Happens for the boot cpu */
890 if (state
!= CPUHP_AP_ONLINE_IDLE
)
893 st
->state
= CPUHP_AP_ONLINE_IDLE
;
895 /* The cpu is marked online, set it active now */
896 set_cpu_active(cpu
, true);
897 /* Unpark the stopper thread and the hotplug thread of this cpu */
898 stop_machine_unpark(cpu
);
899 kthread_unpark(st
->thread
);
901 /* Should we go further up ? */
902 if (st
->target
> CPUHP_AP_ONLINE_IDLE
)
903 __cpuhp_kick_ap_work(st
);
908 /* Requires cpu_add_remove_lock to be held */
909 static int _cpu_up(unsigned int cpu
, int tasks_frozen
, enum cpuhp_state target
)
911 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
912 struct task_struct
*idle
;
917 if (!cpu_present(cpu
)) {
923 * The caller of do_cpu_up might have raced with another
924 * caller. Ignore it for now.
926 if (st
->state
>= target
)
929 if (st
->state
== CPUHP_OFFLINE
) {
930 /* Let it fail before we try to bring the cpu up */
931 idle
= idle_thread_get(cpu
);
938 cpuhp_tasks_frozen
= tasks_frozen
;
942 * If the current CPU state is in the range of the AP hotplug thread,
943 * then we need to kick the thread once more.
945 if (st
->state
> CPUHP_BRINGUP_CPU
) {
946 ret
= cpuhp_kick_ap_work(cpu
);
948 * The AP side has done the error rollback already. Just
949 * return the error code..
956 * Try to reach the target state. We max out on the BP at
957 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
958 * responsible for bringing it up to the target state.
960 target
= min((int)target
, CPUHP_BRINGUP_CPU
);
961 ret
= cpuhp_up_callbacks(cpu
, st
, cpuhp_bp_states
, target
);
967 static int do_cpu_up(unsigned int cpu
, enum cpuhp_state target
)
971 if (!cpu_possible(cpu
)) {
972 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
974 #if defined(CONFIG_IA64)
975 pr_err("please check additional_cpus= boot parameter\n");
980 err
= try_online_node(cpu_to_node(cpu
));
984 cpu_maps_update_begin();
986 if (cpu_hotplug_disabled
) {
991 err
= _cpu_up(cpu
, 0, target
);
993 cpu_maps_update_done();
997 int cpu_up(unsigned int cpu
)
999 return do_cpu_up(cpu
, CPUHP_ONLINE
);
1001 EXPORT_SYMBOL_GPL(cpu_up
);
1003 #ifdef CONFIG_PM_SLEEP_SMP
1004 static cpumask_var_t frozen_cpus
;
1006 int disable_nonboot_cpus(void)
1008 int cpu
, first_cpu
, error
= 0;
1010 cpu_maps_update_begin();
1011 first_cpu
= cpumask_first(cpu_online_mask
);
1013 * We take down all of the non-boot CPUs in one shot to avoid races
1014 * with the userspace trying to use the CPU hotplug at the same time
1016 cpumask_clear(frozen_cpus
);
1018 pr_info("Disabling non-boot CPUs ...\n");
1019 for_each_online_cpu(cpu
) {
1020 if (cpu
== first_cpu
)
1022 trace_suspend_resume(TPS("CPU_OFF"), cpu
, true);
1023 error
= _cpu_down(cpu
, 1, CPUHP_OFFLINE
);
1024 trace_suspend_resume(TPS("CPU_OFF"), cpu
, false);
1026 cpumask_set_cpu(cpu
, frozen_cpus
);
1028 pr_err("Error taking CPU%d down: %d\n", cpu
, error
);
1034 BUG_ON(num_online_cpus() > 1);
1036 pr_err("Non-boot CPUs are not disabled\n");
1039 * Make sure the CPUs won't be enabled by someone else. We need to do
1040 * this even in case of failure as all disable_nonboot_cpus() users are
1041 * supposed to do enable_nonboot_cpus() on the failure path.
1043 cpu_hotplug_disabled
++;
1045 cpu_maps_update_done();
1049 void __weak
arch_enable_nonboot_cpus_begin(void)
1053 void __weak
arch_enable_nonboot_cpus_end(void)
1057 void enable_nonboot_cpus(void)
1061 /* Allow everyone to use the CPU hotplug again */
1062 cpu_maps_update_begin();
1063 WARN_ON(--cpu_hotplug_disabled
< 0);
1064 if (cpumask_empty(frozen_cpus
))
1067 pr_info("Enabling non-boot CPUs ...\n");
1069 arch_enable_nonboot_cpus_begin();
1071 for_each_cpu(cpu
, frozen_cpus
) {
1072 trace_suspend_resume(TPS("CPU_ON"), cpu
, true);
1073 error
= _cpu_up(cpu
, 1, CPUHP_ONLINE
);
1074 trace_suspend_resume(TPS("CPU_ON"), cpu
, false);
1076 pr_info("CPU%d is up\n", cpu
);
1079 pr_warn("Error taking CPU%d up: %d\n", cpu
, error
);
1082 arch_enable_nonboot_cpus_end();
1084 cpumask_clear(frozen_cpus
);
1086 cpu_maps_update_done();
1089 static int __init
alloc_frozen_cpus(void)
1091 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
1095 core_initcall(alloc_frozen_cpus
);
1098 * When callbacks for CPU hotplug notifications are being executed, we must
1099 * ensure that the state of the system with respect to the tasks being frozen
1100 * or not, as reported by the notification, remains unchanged *throughout the
1101 * duration* of the execution of the callbacks.
1102 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1104 * This synchronization is implemented by mutually excluding regular CPU
1105 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1106 * Hibernate notifications.
1109 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
1110 unsigned long action
, void *ptr
)
1114 case PM_SUSPEND_PREPARE
:
1115 case PM_HIBERNATION_PREPARE
:
1116 cpu_hotplug_disable();
1119 case PM_POST_SUSPEND
:
1120 case PM_POST_HIBERNATION
:
1121 cpu_hotplug_enable();
1132 static int __init
cpu_hotplug_pm_sync_init(void)
1135 * cpu_hotplug_pm_callback has higher priority than x86
1136 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1137 * to disable cpu hotplug to avoid cpu hotplug race.
1139 pm_notifier(cpu_hotplug_pm_callback
, 0);
1142 core_initcall(cpu_hotplug_pm_sync_init
);
1144 #endif /* CONFIG_PM_SLEEP_SMP */
1146 #endif /* CONFIG_SMP */
1148 /* Boot processor state steps */
1149 static struct cpuhp_step cpuhp_bp_states
[] = {
1156 [CPUHP_CREATE_THREADS
]= {
1157 .name
= "threads:create",
1158 .startup
= smpboot_create_threads
,
1162 [CPUHP_NOTIFY_PREPARE
] = {
1163 .name
= "notify:prepare",
1164 .startup
= notify_prepare
,
1165 .teardown
= notify_dead
,
1169 [CPUHP_BRINGUP_CPU
] = {
1170 .name
= "cpu:bringup",
1171 .startup
= bringup_cpu
,
1175 [CPUHP_TEARDOWN_CPU
] = {
1176 .name
= "cpu:teardown",
1178 .teardown
= takedown_cpu
,
1184 /* Application processor state steps */
1185 static struct cpuhp_step cpuhp_ap_states
[] = {
1187 [CPUHP_AP_NOTIFY_STARTING
] = {
1188 .name
= "notify:starting",
1189 .startup
= notify_starting
,
1190 .teardown
= notify_dying
,
1194 [CPUHP_AP_SMPBOOT_THREADS
] = {
1195 .name
= "smpboot:threads",
1196 .startup
= smpboot_unpark_threads
,
1197 .teardown
= smpboot_park_threads
,
1199 [CPUHP_AP_NOTIFY_ONLINE
] = {
1200 .name
= "notify:online",
1201 .startup
= notify_online
,
1202 .teardown
= notify_down_prepare
,
1212 /* Sanity check for callbacks */
1213 static int cpuhp_cb_check(enum cpuhp_state state
)
1215 if (state
<= CPUHP_OFFLINE
|| state
>= CPUHP_ONLINE
)
1220 static bool cpuhp_is_ap_state(enum cpuhp_state state
)
1222 if (state
>= CPUHP_AP_OFFLINE
&& state
<= CPUHP_AP_ONLINE
)
1224 return state
> CPUHP_BRINGUP_CPU
;
1227 static struct cpuhp_step
*cpuhp_get_step(enum cpuhp_state state
)
1229 struct cpuhp_step
*sp
;
1231 sp
= cpuhp_is_ap_state(state
) ? cpuhp_ap_states
: cpuhp_bp_states
;
1235 static void cpuhp_store_callbacks(enum cpuhp_state state
,
1237 int (*startup
)(unsigned int cpu
),
1238 int (*teardown
)(unsigned int cpu
))
1240 /* (Un)Install the callbacks for further cpu hotplug operations */
1241 struct cpuhp_step
*sp
;
1243 mutex_lock(&cpuhp_state_mutex
);
1244 sp
= cpuhp_get_step(state
);
1245 sp
->startup
= startup
;
1246 sp
->teardown
= teardown
;
1248 mutex_unlock(&cpuhp_state_mutex
);
1251 static void *cpuhp_get_teardown_cb(enum cpuhp_state state
)
1253 return cpuhp_get_step(state
)->teardown
;
1257 * Call the startup/teardown function for a step either on the AP or
1258 * on the current CPU.
1260 static int cpuhp_issue_call(int cpu
, enum cpuhp_state state
,
1261 int (*cb
)(unsigned int), bool bringup
)
1268 * The non AP bound callbacks can fail on bringup. On teardown
1269 * e.g. module removal we crash for now.
1272 if (cpuhp_is_ap_state(state
))
1273 ret
= cpuhp_invoke_ap_callback(cpu
, state
, cb
);
1275 ret
= cpuhp_invoke_callback(cpu
, state
, cb
);
1277 ret
= cpuhp_invoke_callback(cpu
, state
, cb
);
1279 BUG_ON(ret
&& !bringup
);
1284 * Called from __cpuhp_setup_state on a recoverable failure.
1286 * Note: The teardown callbacks for rollback are not allowed to fail!
1288 static void cpuhp_rollback_install(int failedcpu
, enum cpuhp_state state
,
1289 int (*teardown
)(unsigned int cpu
))
1296 /* Roll back the already executed steps on the other cpus */
1297 for_each_present_cpu(cpu
) {
1298 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1299 int cpustate
= st
->state
;
1301 if (cpu
>= failedcpu
)
1304 /* Did we invoke the startup call on that cpu ? */
1305 if (cpustate
>= state
)
1306 cpuhp_issue_call(cpu
, state
, teardown
, false);
1311 * Returns a free for dynamic slot assignment of the Online state. The states
1312 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1313 * by having no name assigned.
1315 static int cpuhp_reserve_state(enum cpuhp_state state
)
1319 mutex_lock(&cpuhp_state_mutex
);
1320 for (i
= CPUHP_AP_ONLINE_DYN
; i
<= CPUHP_AP_ONLINE_DYN_END
; i
++) {
1321 if (cpuhp_ap_states
[i
].name
)
1324 cpuhp_ap_states
[i
].name
= "Reserved";
1325 mutex_unlock(&cpuhp_state_mutex
);
1328 mutex_unlock(&cpuhp_state_mutex
);
1329 WARN(1, "No more dynamic states available for CPU hotplug\n");
1334 * __cpuhp_setup_state - Setup the callbacks for an hotplug machine state
1335 * @state: The state to setup
1336 * @invoke: If true, the startup function is invoked for cpus where
1337 * cpu state >= @state
1338 * @startup: startup callback function
1339 * @teardown: teardown callback function
1341 * Returns 0 if successful, otherwise a proper error code
1343 int __cpuhp_setup_state(enum cpuhp_state state
,
1344 const char *name
, bool invoke
,
1345 int (*startup
)(unsigned int cpu
),
1346 int (*teardown
)(unsigned int cpu
))
1351 if (cpuhp_cb_check(state
) || !name
)
1356 /* currently assignments for the ONLINE state are possible */
1357 if (state
== CPUHP_AP_ONLINE_DYN
) {
1359 ret
= cpuhp_reserve_state(state
);
1365 cpuhp_store_callbacks(state
, name
, startup
, teardown
);
1367 if (!invoke
|| !startup
)
1371 * Try to call the startup callback for each present cpu
1372 * depending on the hotplug state of the cpu.
1374 for_each_present_cpu(cpu
) {
1375 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1376 int cpustate
= st
->state
;
1378 if (cpustate
< state
)
1381 ret
= cpuhp_issue_call(cpu
, state
, startup
, true);
1383 cpuhp_rollback_install(cpu
, state
, teardown
);
1384 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
);
1390 if (!ret
&& dyn_state
)
1394 EXPORT_SYMBOL(__cpuhp_setup_state
);
1397 * __cpuhp_remove_state - Remove the callbacks for an hotplug machine state
1398 * @state: The state to remove
1399 * @invoke: If true, the teardown function is invoked for cpus where
1400 * cpu state >= @state
1402 * The teardown callback is currently not allowed to fail. Think
1403 * about module removal!
1405 void __cpuhp_remove_state(enum cpuhp_state state
, bool invoke
)
1407 int (*teardown
)(unsigned int cpu
) = cpuhp_get_teardown_cb(state
);
1410 BUG_ON(cpuhp_cb_check(state
));
1414 if (!invoke
|| !teardown
)
1418 * Call the teardown callback for each present cpu depending
1419 * on the hotplug state of the cpu. This function is not
1420 * allowed to fail currently!
1422 for_each_present_cpu(cpu
) {
1423 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1424 int cpustate
= st
->state
;
1426 if (cpustate
>= state
)
1427 cpuhp_issue_call(cpu
, state
, teardown
, false);
1430 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
);
1433 EXPORT_SYMBOL(__cpuhp_remove_state
);
1435 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1436 static ssize_t
show_cpuhp_state(struct device
*dev
,
1437 struct device_attribute
*attr
, char *buf
)
1439 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1441 return sprintf(buf
, "%d\n", st
->state
);
1443 static DEVICE_ATTR(state
, 0444, show_cpuhp_state
, NULL
);
1445 static ssize_t
write_cpuhp_target(struct device
*dev
,
1446 struct device_attribute
*attr
,
1447 const char *buf
, size_t count
)
1449 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1450 struct cpuhp_step
*sp
;
1453 ret
= kstrtoint(buf
, 10, &target
);
1457 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1458 if (target
< CPUHP_OFFLINE
|| target
> CPUHP_ONLINE
)
1461 if (target
!= CPUHP_OFFLINE
&& target
!= CPUHP_ONLINE
)
1465 ret
= lock_device_hotplug_sysfs();
1469 mutex_lock(&cpuhp_state_mutex
);
1470 sp
= cpuhp_get_step(target
);
1471 ret
= !sp
->name
|| sp
->cant_stop
? -EINVAL
: 0;
1472 mutex_unlock(&cpuhp_state_mutex
);
1476 if (st
->state
< target
)
1477 ret
= do_cpu_up(dev
->id
, target
);
1479 ret
= do_cpu_down(dev
->id
, target
);
1481 unlock_device_hotplug();
1482 return ret
? ret
: count
;
1485 static ssize_t
show_cpuhp_target(struct device
*dev
,
1486 struct device_attribute
*attr
, char *buf
)
1488 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1490 return sprintf(buf
, "%d\n", st
->target
);
1492 static DEVICE_ATTR(target
, 0644, show_cpuhp_target
, write_cpuhp_target
);
1494 static struct attribute
*cpuhp_cpu_attrs
[] = {
1495 &dev_attr_state
.attr
,
1496 &dev_attr_target
.attr
,
1500 static struct attribute_group cpuhp_cpu_attr_group
= {
1501 .attrs
= cpuhp_cpu_attrs
,
1506 static ssize_t
show_cpuhp_states(struct device
*dev
,
1507 struct device_attribute
*attr
, char *buf
)
1509 ssize_t cur
, res
= 0;
1512 mutex_lock(&cpuhp_state_mutex
);
1513 for (i
= CPUHP_OFFLINE
; i
<= CPUHP_ONLINE
; i
++) {
1514 struct cpuhp_step
*sp
= cpuhp_get_step(i
);
1517 cur
= sprintf(buf
, "%3d: %s\n", i
, sp
->name
);
1522 mutex_unlock(&cpuhp_state_mutex
);
1525 static DEVICE_ATTR(states
, 0444, show_cpuhp_states
, NULL
);
1527 static struct attribute
*cpuhp_cpu_root_attrs
[] = {
1528 &dev_attr_states
.attr
,
1532 static struct attribute_group cpuhp_cpu_root_attr_group
= {
1533 .attrs
= cpuhp_cpu_root_attrs
,
1538 static int __init
cpuhp_sysfs_init(void)
1542 ret
= sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
1543 &cpuhp_cpu_root_attr_group
);
1547 for_each_possible_cpu(cpu
) {
1548 struct device
*dev
= get_cpu_device(cpu
);
1552 ret
= sysfs_create_group(&dev
->kobj
, &cpuhp_cpu_attr_group
);
1558 device_initcall(cpuhp_sysfs_init
);
1562 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1563 * represents all NR_CPUS bits binary values of 1<<nr.
1565 * It is used by cpumask_of() to get a constant address to a CPU
1566 * mask value that has a single bit set only.
1569 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1570 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1571 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1572 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1573 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1575 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
1577 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1578 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1579 #if BITS_PER_LONG > 32
1580 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1581 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1584 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
1586 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
1587 EXPORT_SYMBOL(cpu_all_bits
);
1589 #ifdef CONFIG_INIT_ALL_POSSIBLE
1590 struct cpumask __cpu_possible_mask __read_mostly
1593 struct cpumask __cpu_possible_mask __read_mostly
;
1595 EXPORT_SYMBOL(__cpu_possible_mask
);
1597 struct cpumask __cpu_online_mask __read_mostly
;
1598 EXPORT_SYMBOL(__cpu_online_mask
);
1600 struct cpumask __cpu_present_mask __read_mostly
;
1601 EXPORT_SYMBOL(__cpu_present_mask
);
1603 struct cpumask __cpu_active_mask __read_mostly
;
1604 EXPORT_SYMBOL(__cpu_active_mask
);
1606 void init_cpu_present(const struct cpumask
*src
)
1608 cpumask_copy(&__cpu_present_mask
, src
);
1611 void init_cpu_possible(const struct cpumask
*src
)
1613 cpumask_copy(&__cpu_possible_mask
, src
);
1616 void init_cpu_online(const struct cpumask
*src
)
1618 cpumask_copy(&__cpu_online_mask
, src
);
1622 * Activate the first processor.
1624 void __init
boot_cpu_init(void)
1626 int cpu
= smp_processor_id();
1628 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
1629 set_cpu_online(cpu
, true);
1630 set_cpu_active(cpu
, true);
1631 set_cpu_present(cpu
, true);
1632 set_cpu_possible(cpu
, true);
1636 * Must be called _AFTER_ setting up the per_cpu areas
1638 void __init
boot_cpu_state_init(void)
1640 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->state
= CPUHP_ONLINE
;