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/signal.h>
11 #include <linux/sched/hotplug.h>
12 #include <linux/sched/task.h>
13 #include <linux/unistd.h>
14 #include <linux/cpu.h>
15 #include <linux/oom.h>
16 #include <linux/rcupdate.h>
17 #include <linux/export.h>
18 #include <linux/bug.h>
19 #include <linux/kthread.h>
20 #include <linux/stop_machine.h>
21 #include <linux/mutex.h>
22 #include <linux/gfp.h>
23 #include <linux/suspend.h>
24 #include <linux/lockdep.h>
25 #include <linux/tick.h>
26 #include <linux/irq.h>
27 #include <linux/nmi.h>
28 #include <linux/smpboot.h>
29 #include <linux/relay.h>
30 #include <linux/slab.h>
31 #include <linux/percpu-rwsem.h>
33 #include <trace/events/power.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/cpuhp.h>
40 * cpuhp_cpu_state - Per cpu hotplug state storage
41 * @state: The current cpu state
42 * @target: The target state
43 * @thread: Pointer to the hotplug thread
44 * @should_run: Thread should execute
45 * @rollback: Perform a rollback
46 * @single: Single callback invocation
47 * @bringup: Single callback bringup or teardown selector
48 * @cb_state: The state for a single callback (install/uninstall)
49 * @result: Result of the operation
50 * @done_up: Signal completion to the issuer of the task for cpu-up
51 * @done_down: Signal completion to the issuer of the task for cpu-down
53 struct cpuhp_cpu_state {
54 enum cpuhp_state state;
55 enum cpuhp_state target;
56 enum cpuhp_state fail;
58 struct task_struct *thread;
63 struct hlist_node *node;
64 struct hlist_node *last;
65 enum cpuhp_state cb_state;
67 struct completion done_up;
68 struct completion done_down;
72 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
73 .fail = CPUHP_INVALID,
76 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
77 static struct lockdep_map cpuhp_state_up_map =
78 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
79 static struct lockdep_map cpuhp_state_down_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
83 static inline void cpuhp_lock_acquire(bool bringup)
85 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
88 static inline void cpuhp_lock_release(bool bringup)
90 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
94 static inline void cpuhp_lock_acquire(bool bringup) { }
95 static inline void cpuhp_lock_release(bool bringup) { }
100 * cpuhp_step - Hotplug state machine step
101 * @name: Name of the step
102 * @startup: Startup function of the step
103 * @teardown: Teardown function of the step
104 * @skip_onerr: Do not invoke the functions on error rollback
105 * Will go away once the notifiers are gone
106 * @cant_stop: Bringup/teardown can't be stopped at this step
111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
120 struct hlist_head list;
126 static DEFINE_MUTEX(cpuhp_state_mutex);
127 static struct cpuhp_step cpuhp_hp_states[];
129 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
131 return cpuhp_hp_states + state;
135 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
136 * @cpu: The cpu for which the callback should be invoked
137 * @state: The state to do callbacks for
138 * @bringup: True if the bringup callback should be invoked
139 * @node: For multi-instance, do a single entry callback for install/remove
140 * @lastp: For multi-instance rollback, remember how far we got
142 * Called from cpu hotplug and from the state register machinery.
144 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
145 bool bringup, struct hlist_node *node,
146 struct hlist_node **lastp)
148 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
149 struct cpuhp_step *step = cpuhp_get_step(state);
150 int (*cbm)(unsigned int cpu, struct hlist_node *node);
151 int (*cb)(unsigned int cpu);
154 if (st->fail == state) {
155 st->fail = CPUHP_INVALID;
157 if (!(bringup ? step->startup.single : step->teardown.single))
163 if (!step->multi_instance) {
164 WARN_ON_ONCE(lastp && *lastp);
165 cb = bringup ? step->startup.single : step->teardown.single;
168 trace_cpuhp_enter(cpu, st->target, state, cb);
170 trace_cpuhp_exit(cpu, st->state, state, ret);
173 cbm = bringup ? step->startup.multi : step->teardown.multi;
177 /* Single invocation for instance add/remove */
179 WARN_ON_ONCE(lastp && *lastp);
180 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
181 ret = cbm(cpu, node);
182 trace_cpuhp_exit(cpu, st->state, state, ret);
186 /* State transition. Invoke on all instances */
188 hlist_for_each(node, &step->list) {
189 if (lastp && node == *lastp)
192 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
193 ret = cbm(cpu, node);
194 trace_cpuhp_exit(cpu, st->state, state, ret);
208 /* Rollback the instances if one failed */
209 cbm = !bringup ? step->startup.multi : step->teardown.multi;
213 hlist_for_each(node, &step->list) {
217 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
218 ret = cbm(cpu, node);
219 trace_cpuhp_exit(cpu, st->state, state, ret);
221 * Rollback must not fail,
229 static bool cpuhp_is_ap_state(enum cpuhp_state state)
232 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
233 * purposes as that state is handled explicitly in cpu_down.
235 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
238 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
240 struct completion *done = bringup ? &st->done_up : &st->done_down;
241 wait_for_completion(done);
244 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
246 struct completion *done = bringup ? &st->done_up : &st->done_down;
251 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
253 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
255 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
258 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
259 static DEFINE_MUTEX(cpu_add_remove_lock);
260 bool cpuhp_tasks_frozen;
261 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
264 * The following two APIs (cpu_maps_update_begin/done) must be used when
265 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
267 void cpu_maps_update_begin(void)
269 mutex_lock(&cpu_add_remove_lock);
272 void cpu_maps_update_done(void)
274 mutex_unlock(&cpu_add_remove_lock);
278 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
279 * Should always be manipulated under cpu_add_remove_lock
281 static int cpu_hotplug_disabled;
283 #ifdef CONFIG_HOTPLUG_CPU
285 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
287 void cpus_read_lock(void)
289 percpu_down_read(&cpu_hotplug_lock);
291 EXPORT_SYMBOL_GPL(cpus_read_lock);
293 int cpus_read_trylock(void)
295 return percpu_down_read_trylock(&cpu_hotplug_lock);
297 EXPORT_SYMBOL_GPL(cpus_read_trylock);
299 void cpus_read_unlock(void)
301 percpu_up_read(&cpu_hotplug_lock);
303 EXPORT_SYMBOL_GPL(cpus_read_unlock);
305 void cpus_write_lock(void)
307 percpu_down_write(&cpu_hotplug_lock);
310 void cpus_write_unlock(void)
312 percpu_up_write(&cpu_hotplug_lock);
315 void lockdep_assert_cpus_held(void)
317 percpu_rwsem_assert_held(&cpu_hotplug_lock);
321 * Wait for currently running CPU hotplug operations to complete (if any) and
322 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
323 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
324 * hotplug path before performing hotplug operations. So acquiring that lock
325 * guarantees mutual exclusion from any currently running hotplug operations.
327 void cpu_hotplug_disable(void)
329 cpu_maps_update_begin();
330 cpu_hotplug_disabled++;
331 cpu_maps_update_done();
333 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
335 static void __cpu_hotplug_enable(void)
337 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
339 cpu_hotplug_disabled--;
342 void cpu_hotplug_enable(void)
344 cpu_maps_update_begin();
345 __cpu_hotplug_enable();
346 cpu_maps_update_done();
348 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
349 #endif /* CONFIG_HOTPLUG_CPU */
351 static inline enum cpuhp_state
352 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
354 enum cpuhp_state prev_state = st->state;
356 st->rollback = false;
361 st->bringup = st->state < target;
367 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
372 * If we have st->last we need to undo partial multi_instance of this
373 * state first. Otherwise start undo at the previous state.
382 st->target = prev_state;
383 st->bringup = !st->bringup;
386 /* Regular hotplug invocation of the AP hotplug thread */
387 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
389 if (!st->single && st->state == st->target)
394 * Make sure the above stores are visible before should_run becomes
395 * true. Paired with the mb() above in cpuhp_thread_fun()
398 st->should_run = true;
399 wake_up_process(st->thread);
400 wait_for_ap_thread(st, st->bringup);
403 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
405 enum cpuhp_state prev_state;
408 prev_state = cpuhp_set_state(st, target);
410 if ((ret = st->result)) {
411 cpuhp_reset_state(st, prev_state);
418 static int bringup_wait_for_ap(unsigned int cpu)
420 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
422 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
423 wait_for_ap_thread(st, true);
424 if (WARN_ON_ONCE((!cpu_online(cpu))))
427 /* Unpark the stopper thread and the hotplug thread of the target cpu */
428 stop_machine_unpark(cpu);
429 kthread_unpark(st->thread);
431 if (st->target <= CPUHP_AP_ONLINE_IDLE)
434 return cpuhp_kick_ap(st, st->target);
437 static int bringup_cpu(unsigned int cpu)
439 struct task_struct *idle = idle_thread_get(cpu);
443 * Some architectures have to walk the irq descriptors to
444 * setup the vector space for the cpu which comes online.
445 * Prevent irq alloc/free across the bringup.
449 /* Arch-specific enabling code. */
450 ret = __cpu_up(cpu, idle);
454 return bringup_wait_for_ap(cpu);
458 * Hotplug state machine related functions
461 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
463 for (st->state--; st->state > st->target; st->state--) {
464 struct cpuhp_step *step = cpuhp_get_step(st->state);
466 if (!step->skip_onerr)
467 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
471 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
472 enum cpuhp_state target)
474 enum cpuhp_state prev_state = st->state;
477 while (st->state < target) {
479 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
481 st->target = prev_state;
482 undo_cpu_up(cpu, st);
490 * The cpu hotplug threads manage the bringup and teardown of the cpus
492 static void cpuhp_create(unsigned int cpu)
494 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
496 init_completion(&st->done_up);
497 init_completion(&st->done_down);
500 static int cpuhp_should_run(unsigned int cpu)
502 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
504 return st->should_run;
508 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
509 * callbacks when a state gets [un]installed at runtime.
511 * Each invocation of this function by the smpboot thread does a single AP
514 * It has 3 modes of operation:
515 * - single: runs st->cb_state
516 * - up: runs ++st->state, while st->state < st->target
517 * - down: runs st->state--, while st->state > st->target
519 * When complete or on error, should_run is cleared and the completion is fired.
521 static void cpuhp_thread_fun(unsigned int cpu)
523 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
524 bool bringup = st->bringup;
525 enum cpuhp_state state;
528 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
529 * that if we see ->should_run we also see the rest of the state.
533 if (WARN_ON_ONCE(!st->should_run))
536 cpuhp_lock_acquire(bringup);
539 state = st->cb_state;
540 st->should_run = false;
545 st->should_run = (st->state < st->target);
546 WARN_ON_ONCE(st->state > st->target);
550 st->should_run = (st->state > st->target);
551 WARN_ON_ONCE(st->state < st->target);
555 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
558 struct cpuhp_step *step = cpuhp_get_step(state);
559 if (step->skip_onerr)
563 if (cpuhp_is_atomic_state(state)) {
565 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
569 * STARTING/DYING must not fail!
571 WARN_ON_ONCE(st->result);
573 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
578 * If we fail on a rollback, we're up a creek without no
579 * paddle, no way forward, no way back. We loose, thanks for
582 WARN_ON_ONCE(st->rollback);
583 st->should_run = false;
587 cpuhp_lock_release(bringup);
590 complete_ap_thread(st, bringup);
593 /* Invoke a single callback on a remote cpu */
595 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
596 struct hlist_node *node)
598 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
601 if (!cpu_online(cpu))
604 cpuhp_lock_acquire(false);
605 cpuhp_lock_release(false);
607 cpuhp_lock_acquire(true);
608 cpuhp_lock_release(true);
611 * If we are up and running, use the hotplug thread. For early calls
612 * we invoke the thread function directly.
615 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
617 st->rollback = false;
621 st->bringup = bringup;
622 st->cb_state = state;
628 * If we failed and did a partial, do a rollback.
630 if ((ret = st->result) && st->last) {
632 st->bringup = !bringup;
638 * Clean up the leftovers so the next hotplug operation wont use stale
641 st->node = st->last = NULL;
645 static int cpuhp_kick_ap_work(unsigned int cpu)
647 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
648 enum cpuhp_state prev_state = st->state;
651 cpuhp_lock_acquire(false);
652 cpuhp_lock_release(false);
654 cpuhp_lock_acquire(true);
655 cpuhp_lock_release(true);
657 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
658 ret = cpuhp_kick_ap(st, st->target);
659 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
664 static struct smp_hotplug_thread cpuhp_threads = {
665 .store = &cpuhp_state.thread,
666 .create = &cpuhp_create,
667 .thread_should_run = cpuhp_should_run,
668 .thread_fn = cpuhp_thread_fun,
669 .thread_comm = "cpuhp/%u",
673 void __init cpuhp_threads_init(void)
675 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
676 kthread_unpark(this_cpu_read(cpuhp_state.thread));
679 #ifdef CONFIG_HOTPLUG_CPU
681 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
684 * This function walks all processes, finds a valid mm struct for each one and
685 * then clears a corresponding bit in mm's cpumask. While this all sounds
686 * trivial, there are various non-obvious corner cases, which this function
687 * tries to solve in a safe manner.
689 * Also note that the function uses a somewhat relaxed locking scheme, so it may
690 * be called only for an already offlined CPU.
692 void clear_tasks_mm_cpumask(int cpu)
694 struct task_struct *p;
697 * This function is called after the cpu is taken down and marked
698 * offline, so its not like new tasks will ever get this cpu set in
699 * their mm mask. -- Peter Zijlstra
700 * Thus, we may use rcu_read_lock() here, instead of grabbing
701 * full-fledged tasklist_lock.
703 WARN_ON(cpu_online(cpu));
705 for_each_process(p) {
706 struct task_struct *t;
709 * Main thread might exit, but other threads may still have
710 * a valid mm. Find one.
712 t = find_lock_task_mm(p);
715 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
721 /* Take this CPU down. */
722 static int take_cpu_down(void *_param)
724 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
725 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
726 int err, cpu = smp_processor_id();
729 /* Ensure this CPU doesn't handle any more interrupts. */
730 err = __cpu_disable();
735 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
736 * do this step again.
738 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
740 /* Invoke the former CPU_DYING callbacks */
741 for (; st->state > target; st->state--) {
742 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
744 * DYING must not fail!
749 /* Give up timekeeping duties */
750 tick_handover_do_timer();
751 /* Park the stopper thread */
752 stop_machine_park(cpu);
756 static int takedown_cpu(unsigned int cpu)
758 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
761 /* Park the smpboot threads */
762 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
763 smpboot_park_threads(cpu);
766 * Prevent irq alloc/free while the dying cpu reorganizes the
767 * interrupt affinities.
772 * So now all preempt/rcu users must observe !cpu_active().
774 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
776 /* CPU refused to die */
778 /* Unpark the hotplug thread so we can rollback there */
779 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
782 BUG_ON(cpu_online(cpu));
785 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
786 * all runnable tasks from the CPU, there's only the idle task left now
787 * that the migration thread is done doing the stop_machine thing.
789 * Wait for the stop thread to go away.
791 wait_for_ap_thread(st, false);
792 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
794 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
797 hotplug_cpu__broadcast_tick_pull(cpu);
798 /* This actually kills the CPU. */
801 tick_cleanup_dead_cpu(cpu);
802 rcutree_migrate_callbacks(cpu);
806 static void cpuhp_complete_idle_dead(void *arg)
808 struct cpuhp_cpu_state *st = arg;
810 complete_ap_thread(st, false);
813 void cpuhp_report_idle_dead(void)
815 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
817 BUG_ON(st->state != CPUHP_AP_OFFLINE);
818 rcu_report_dead(smp_processor_id());
819 st->state = CPUHP_AP_IDLE_DEAD;
821 * We cannot call complete after rcu_report_dead() so we delegate it
824 smp_call_function_single(cpumask_first(cpu_online_mask),
825 cpuhp_complete_idle_dead, st, 0);
828 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
830 for (st->state++; st->state < st->target; st->state++) {
831 struct cpuhp_step *step = cpuhp_get_step(st->state);
833 if (!step->skip_onerr)
834 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
838 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
839 enum cpuhp_state target)
841 enum cpuhp_state prev_state = st->state;
844 for (; st->state > target; st->state--) {
845 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
847 st->target = prev_state;
848 undo_cpu_down(cpu, st);
855 /* Requires cpu_add_remove_lock to be held */
856 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
857 enum cpuhp_state target)
859 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
860 int prev_state, ret = 0;
862 if (num_online_cpus() == 1)
865 if (!cpu_present(cpu))
870 cpuhp_tasks_frozen = tasks_frozen;
872 prev_state = cpuhp_set_state(st, target);
874 * If the current CPU state is in the range of the AP hotplug thread,
875 * then we need to kick the thread.
877 if (st->state > CPUHP_TEARDOWN_CPU) {
878 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
879 ret = cpuhp_kick_ap_work(cpu);
881 * The AP side has done the error rollback already. Just
882 * return the error code..
888 * We might have stopped still in the range of the AP hotplug
889 * thread. Nothing to do anymore.
891 if (st->state > CPUHP_TEARDOWN_CPU)
897 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
898 * to do the further cleanups.
900 ret = cpuhp_down_callbacks(cpu, st, target);
901 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
902 cpuhp_reset_state(st, prev_state);
909 * Do post unplug cleanup. This is still protected against
910 * concurrent CPU hotplug via cpu_add_remove_lock.
912 lockup_detector_cleanup();
916 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
920 cpu_maps_update_begin();
922 if (cpu_hotplug_disabled) {
927 err = _cpu_down(cpu, 0, target);
930 cpu_maps_update_done();
934 int cpu_down(unsigned int cpu)
936 return do_cpu_down(cpu, CPUHP_OFFLINE);
938 EXPORT_SYMBOL(cpu_down);
941 #define takedown_cpu NULL
942 #endif /*CONFIG_HOTPLUG_CPU*/
945 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
946 * @cpu: cpu that just started
948 * It must be called by the arch code on the new cpu, before the new cpu
949 * enables interrupts and before the "boot" cpu returns from __cpu_up().
951 void notify_cpu_starting(unsigned int cpu)
953 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
954 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
957 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
958 while (st->state < target) {
960 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
962 * STARTING must not fail!
969 * Called from the idle task. Wake up the controlling task which brings the
970 * stopper and the hotplug thread of the upcoming CPU up and then delegates
971 * the rest of the online bringup to the hotplug thread.
973 void cpuhp_online_idle(enum cpuhp_state state)
975 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
977 /* Happens for the boot cpu */
978 if (state != CPUHP_AP_ONLINE_IDLE)
981 st->state = CPUHP_AP_ONLINE_IDLE;
982 complete_ap_thread(st, true);
985 /* Requires cpu_add_remove_lock to be held */
986 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
988 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
989 struct task_struct *idle;
994 if (!cpu_present(cpu)) {
1000 * The caller of do_cpu_up might have raced with another
1001 * caller. Ignore it for now.
1003 if (st->state >= target)
1006 if (st->state == CPUHP_OFFLINE) {
1007 /* Let it fail before we try to bring the cpu up */
1008 idle = idle_thread_get(cpu);
1010 ret = PTR_ERR(idle);
1015 cpuhp_tasks_frozen = tasks_frozen;
1017 cpuhp_set_state(st, target);
1019 * If the current CPU state is in the range of the AP hotplug thread,
1020 * then we need to kick the thread once more.
1022 if (st->state > CPUHP_BRINGUP_CPU) {
1023 ret = cpuhp_kick_ap_work(cpu);
1025 * The AP side has done the error rollback already. Just
1026 * return the error code..
1033 * Try to reach the target state. We max out on the BP at
1034 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1035 * responsible for bringing it up to the target state.
1037 target = min((int)target, CPUHP_BRINGUP_CPU);
1038 ret = cpuhp_up_callbacks(cpu, st, target);
1040 cpus_write_unlock();
1044 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1048 if (!cpu_possible(cpu)) {
1049 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1051 #if defined(CONFIG_IA64)
1052 pr_err("please check additional_cpus= boot parameter\n");
1057 err = try_online_node(cpu_to_node(cpu));
1061 cpu_maps_update_begin();
1063 if (cpu_hotplug_disabled) {
1068 err = _cpu_up(cpu, 0, target);
1070 cpu_maps_update_done();
1074 int cpu_up(unsigned int cpu)
1076 return do_cpu_up(cpu, CPUHP_ONLINE);
1078 EXPORT_SYMBOL_GPL(cpu_up);
1080 #ifdef CONFIG_PM_SLEEP_SMP
1081 static cpumask_var_t frozen_cpus;
1083 int freeze_secondary_cpus(int primary)
1087 cpu_maps_update_begin();
1088 if (!cpu_online(primary))
1089 primary = cpumask_first(cpu_online_mask);
1091 * We take down all of the non-boot CPUs in one shot to avoid races
1092 * with the userspace trying to use the CPU hotplug at the same time
1094 cpumask_clear(frozen_cpus);
1096 pr_info("Disabling non-boot CPUs ...\n");
1097 for_each_online_cpu(cpu) {
1100 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1101 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1102 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1104 cpumask_set_cpu(cpu, frozen_cpus);
1106 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1112 BUG_ON(num_online_cpus() > 1);
1114 pr_err("Non-boot CPUs are not disabled\n");
1117 * Make sure the CPUs won't be enabled by someone else. We need to do
1118 * this even in case of failure as all disable_nonboot_cpus() users are
1119 * supposed to do enable_nonboot_cpus() on the failure path.
1121 cpu_hotplug_disabled++;
1123 cpu_maps_update_done();
1127 void __weak arch_enable_nonboot_cpus_begin(void)
1131 void __weak arch_enable_nonboot_cpus_end(void)
1135 void enable_nonboot_cpus(void)
1139 /* Allow everyone to use the CPU hotplug again */
1140 cpu_maps_update_begin();
1141 __cpu_hotplug_enable();
1142 if (cpumask_empty(frozen_cpus))
1145 pr_info("Enabling non-boot CPUs ...\n");
1147 arch_enable_nonboot_cpus_begin();
1149 for_each_cpu(cpu, frozen_cpus) {
1150 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1151 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1152 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1154 pr_info("CPU%d is up\n", cpu);
1157 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1160 arch_enable_nonboot_cpus_end();
1162 cpumask_clear(frozen_cpus);
1164 cpu_maps_update_done();
1167 static int __init alloc_frozen_cpus(void)
1169 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1173 core_initcall(alloc_frozen_cpus);
1176 * When callbacks for CPU hotplug notifications are being executed, we must
1177 * ensure that the state of the system with respect to the tasks being frozen
1178 * or not, as reported by the notification, remains unchanged *throughout the
1179 * duration* of the execution of the callbacks.
1180 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1182 * This synchronization is implemented by mutually excluding regular CPU
1183 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1184 * Hibernate notifications.
1187 cpu_hotplug_pm_callback(struct notifier_block *nb,
1188 unsigned long action, void *ptr)
1192 case PM_SUSPEND_PREPARE:
1193 case PM_HIBERNATION_PREPARE:
1194 cpu_hotplug_disable();
1197 case PM_POST_SUSPEND:
1198 case PM_POST_HIBERNATION:
1199 cpu_hotplug_enable();
1210 static int __init cpu_hotplug_pm_sync_init(void)
1213 * cpu_hotplug_pm_callback has higher priority than x86
1214 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1215 * to disable cpu hotplug to avoid cpu hotplug race.
1217 pm_notifier(cpu_hotplug_pm_callback, 0);
1220 core_initcall(cpu_hotplug_pm_sync_init);
1222 #endif /* CONFIG_PM_SLEEP_SMP */
1226 #endif /* CONFIG_SMP */
1228 /* Boot processor state steps */
1229 static struct cpuhp_step cpuhp_hp_states[] = {
1232 .startup.single = NULL,
1233 .teardown.single = NULL,
1236 [CPUHP_CREATE_THREADS]= {
1237 .name = "threads:prepare",
1238 .startup.single = smpboot_create_threads,
1239 .teardown.single = NULL,
1242 [CPUHP_PERF_PREPARE] = {
1243 .name = "perf:prepare",
1244 .startup.single = perf_event_init_cpu,
1245 .teardown.single = perf_event_exit_cpu,
1247 [CPUHP_WORKQUEUE_PREP] = {
1248 .name = "workqueue:prepare",
1249 .startup.single = workqueue_prepare_cpu,
1250 .teardown.single = NULL,
1252 [CPUHP_HRTIMERS_PREPARE] = {
1253 .name = "hrtimers:prepare",
1254 .startup.single = hrtimers_prepare_cpu,
1255 .teardown.single = hrtimers_dead_cpu,
1257 [CPUHP_SMPCFD_PREPARE] = {
1258 .name = "smpcfd:prepare",
1259 .startup.single = smpcfd_prepare_cpu,
1260 .teardown.single = smpcfd_dead_cpu,
1262 [CPUHP_RELAY_PREPARE] = {
1263 .name = "relay:prepare",
1264 .startup.single = relay_prepare_cpu,
1265 .teardown.single = NULL,
1267 [CPUHP_SLAB_PREPARE] = {
1268 .name = "slab:prepare",
1269 .startup.single = slab_prepare_cpu,
1270 .teardown.single = slab_dead_cpu,
1272 [CPUHP_RCUTREE_PREP] = {
1273 .name = "RCU/tree:prepare",
1274 .startup.single = rcutree_prepare_cpu,
1275 .teardown.single = rcutree_dead_cpu,
1278 * On the tear-down path, timers_dead_cpu() must be invoked
1279 * before blk_mq_queue_reinit_notify() from notify_dead(),
1280 * otherwise a RCU stall occurs.
1282 [CPUHP_TIMERS_PREPARE] = {
1283 .name = "timers:dead",
1284 .startup.single = timers_prepare_cpu,
1285 .teardown.single = timers_dead_cpu,
1287 /* Kicks the plugged cpu into life */
1288 [CPUHP_BRINGUP_CPU] = {
1289 .name = "cpu:bringup",
1290 .startup.single = bringup_cpu,
1291 .teardown.single = NULL,
1294 /* Final state before CPU kills itself */
1295 [CPUHP_AP_IDLE_DEAD] = {
1296 .name = "idle:dead",
1299 * Last state before CPU enters the idle loop to die. Transient state
1300 * for synchronization.
1302 [CPUHP_AP_OFFLINE] = {
1303 .name = "ap:offline",
1306 /* First state is scheduler control. Interrupts are disabled */
1307 [CPUHP_AP_SCHED_STARTING] = {
1308 .name = "sched:starting",
1309 .startup.single = sched_cpu_starting,
1310 .teardown.single = sched_cpu_dying,
1312 [CPUHP_AP_RCUTREE_DYING] = {
1313 .name = "RCU/tree:dying",
1314 .startup.single = NULL,
1315 .teardown.single = rcutree_dying_cpu,
1317 [CPUHP_AP_SMPCFD_DYING] = {
1318 .name = "smpcfd:dying",
1319 .startup.single = NULL,
1320 .teardown.single = smpcfd_dying_cpu,
1322 /* Entry state on starting. Interrupts enabled from here on. Transient
1323 * state for synchronsization */
1324 [CPUHP_AP_ONLINE] = {
1325 .name = "ap:online",
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,
1337 /* Handle smpboot threads park/unpark */
1338 [CPUHP_AP_SMPBOOT_THREADS] = {
1339 .name = "smpboot/threads:online",
1340 .startup.single = smpboot_unpark_threads,
1341 .teardown.single = NULL,
1343 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1344 .name = "irq/affinity:online",
1345 .startup.single = irq_affinity_online_cpu,
1346 .teardown.single = NULL,
1348 [CPUHP_AP_PERF_ONLINE] = {
1349 .name = "perf:online",
1350 .startup.single = perf_event_init_cpu,
1351 .teardown.single = perf_event_exit_cpu,
1353 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1354 .name = "workqueue:online",
1355 .startup.single = workqueue_online_cpu,
1356 .teardown.single = workqueue_offline_cpu,
1358 [CPUHP_AP_RCUTREE_ONLINE] = {
1359 .name = "RCU/tree:online",
1360 .startup.single = rcutree_online_cpu,
1361 .teardown.single = rcutree_offline_cpu,
1365 * The dynamically registered state space is here
1369 /* Last state is scheduler control setting the cpu active */
1370 [CPUHP_AP_ACTIVE] = {
1371 .name = "sched:active",
1372 .startup.single = sched_cpu_activate,
1373 .teardown.single = sched_cpu_deactivate,
1377 /* CPU is fully up and running. */
1380 .startup.single = NULL,
1381 .teardown.single = NULL,
1385 /* Sanity check for callbacks */
1386 static int cpuhp_cb_check(enum cpuhp_state state)
1388 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1394 * Returns a free for dynamic slot assignment of the Online state. The states
1395 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1396 * by having no name assigned.
1398 static int cpuhp_reserve_state(enum cpuhp_state state)
1400 enum cpuhp_state i, end;
1401 struct cpuhp_step *step;
1404 case CPUHP_AP_ONLINE_DYN:
1405 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1406 end = CPUHP_AP_ONLINE_DYN_END;
1408 case CPUHP_BP_PREPARE_DYN:
1409 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1410 end = CPUHP_BP_PREPARE_DYN_END;
1416 for (i = state; i <= end; i++, step++) {
1420 WARN(1, "No more dynamic states available for CPU hotplug\n");
1424 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1425 int (*startup)(unsigned int cpu),
1426 int (*teardown)(unsigned int cpu),
1427 bool multi_instance)
1429 /* (Un)Install the callbacks for further cpu hotplug operations */
1430 struct cpuhp_step *sp;
1434 * If name is NULL, then the state gets removed.
1436 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1437 * the first allocation from these dynamic ranges, so the removal
1438 * would trigger a new allocation and clear the wrong (already
1439 * empty) state, leaving the callbacks of the to be cleared state
1440 * dangling, which causes wreckage on the next hotplug operation.
1442 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1443 state == CPUHP_BP_PREPARE_DYN)) {
1444 ret = cpuhp_reserve_state(state);
1449 sp = cpuhp_get_step(state);
1450 if (name && sp->name)
1453 sp->startup.single = startup;
1454 sp->teardown.single = teardown;
1456 sp->multi_instance = multi_instance;
1457 INIT_HLIST_HEAD(&sp->list);
1461 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1463 return cpuhp_get_step(state)->teardown.single;
1467 * Call the startup/teardown function for a step either on the AP or
1468 * on the current CPU.
1470 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1471 struct hlist_node *node)
1473 struct cpuhp_step *sp = cpuhp_get_step(state);
1477 * If there's nothing to do, we done.
1478 * Relies on the union for multi_instance.
1480 if ((bringup && !sp->startup.single) ||
1481 (!bringup && !sp->teardown.single))
1484 * The non AP bound callbacks can fail on bringup. On teardown
1485 * e.g. module removal we crash for now.
1488 if (cpuhp_is_ap_state(state))
1489 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1491 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1493 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1495 BUG_ON(ret && !bringup);
1500 * Called from __cpuhp_setup_state on a recoverable failure.
1502 * Note: The teardown callbacks for rollback are not allowed to fail!
1504 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1505 struct hlist_node *node)
1509 /* Roll back the already executed steps on the other cpus */
1510 for_each_present_cpu(cpu) {
1511 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1512 int cpustate = st->state;
1514 if (cpu >= failedcpu)
1517 /* Did we invoke the startup call on that cpu ? */
1518 if (cpustate >= state)
1519 cpuhp_issue_call(cpu, state, false, node);
1523 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1524 struct hlist_node *node,
1527 struct cpuhp_step *sp;
1531 lockdep_assert_cpus_held();
1533 sp = cpuhp_get_step(state);
1534 if (sp->multi_instance == false)
1537 mutex_lock(&cpuhp_state_mutex);
1539 if (!invoke || !sp->startup.multi)
1543 * Try to call the startup callback for each present cpu
1544 * depending on the hotplug state of the cpu.
1546 for_each_present_cpu(cpu) {
1547 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1548 int cpustate = st->state;
1550 if (cpustate < state)
1553 ret = cpuhp_issue_call(cpu, state, true, node);
1555 if (sp->teardown.multi)
1556 cpuhp_rollback_install(cpu, state, node);
1562 hlist_add_head(node, &sp->list);
1564 mutex_unlock(&cpuhp_state_mutex);
1568 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1574 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1578 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1581 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1582 * @state: The state to setup
1583 * @invoke: If true, the startup function is invoked for cpus where
1584 * cpu state >= @state
1585 * @startup: startup callback function
1586 * @teardown: teardown callback function
1587 * @multi_instance: State is set up for multiple instances which get
1590 * The caller needs to hold cpus read locked while calling this function.
1593 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1594 * 0 for all other states
1595 * On failure: proper (negative) error code
1597 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1598 const char *name, bool invoke,
1599 int (*startup)(unsigned int cpu),
1600 int (*teardown)(unsigned int cpu),
1601 bool multi_instance)
1606 lockdep_assert_cpus_held();
1608 if (cpuhp_cb_check(state) || !name)
1611 mutex_lock(&cpuhp_state_mutex);
1613 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1616 dynstate = state == CPUHP_AP_ONLINE_DYN;
1617 if (ret > 0 && dynstate) {
1622 if (ret || !invoke || !startup)
1626 * Try to call the startup callback for each present cpu
1627 * depending on the hotplug state of the cpu.
1629 for_each_present_cpu(cpu) {
1630 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1631 int cpustate = st->state;
1633 if (cpustate < state)
1636 ret = cpuhp_issue_call(cpu, state, true, NULL);
1639 cpuhp_rollback_install(cpu, state, NULL);
1640 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1645 mutex_unlock(&cpuhp_state_mutex);
1647 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1648 * dynamically allocated state in case of success.
1650 if (!ret && dynstate)
1654 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1656 int __cpuhp_setup_state(enum cpuhp_state state,
1657 const char *name, bool invoke,
1658 int (*startup)(unsigned int cpu),
1659 int (*teardown)(unsigned int cpu),
1660 bool multi_instance)
1665 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1666 teardown, multi_instance);
1670 EXPORT_SYMBOL(__cpuhp_setup_state);
1672 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1673 struct hlist_node *node, bool invoke)
1675 struct cpuhp_step *sp = cpuhp_get_step(state);
1678 BUG_ON(cpuhp_cb_check(state));
1680 if (!sp->multi_instance)
1684 mutex_lock(&cpuhp_state_mutex);
1686 if (!invoke || !cpuhp_get_teardown_cb(state))
1689 * Call the teardown callback for each present cpu depending
1690 * on the hotplug state of the cpu. This function is not
1691 * allowed to fail currently!
1693 for_each_present_cpu(cpu) {
1694 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1695 int cpustate = st->state;
1697 if (cpustate >= state)
1698 cpuhp_issue_call(cpu, state, false, node);
1703 mutex_unlock(&cpuhp_state_mutex);
1708 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1711 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1712 * @state: The state to remove
1713 * @invoke: If true, the teardown function is invoked for cpus where
1714 * cpu state >= @state
1716 * The caller needs to hold cpus read locked while calling this function.
1717 * The teardown callback is currently not allowed to fail. Think
1718 * about module removal!
1720 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1722 struct cpuhp_step *sp = cpuhp_get_step(state);
1725 BUG_ON(cpuhp_cb_check(state));
1727 lockdep_assert_cpus_held();
1729 mutex_lock(&cpuhp_state_mutex);
1730 if (sp->multi_instance) {
1731 WARN(!hlist_empty(&sp->list),
1732 "Error: Removing state %d which has instances left.\n",
1737 if (!invoke || !cpuhp_get_teardown_cb(state))
1741 * Call the teardown callback for each present cpu depending
1742 * on the hotplug state of the cpu. This function is not
1743 * allowed to fail currently!
1745 for_each_present_cpu(cpu) {
1746 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1747 int cpustate = st->state;
1749 if (cpustate >= state)
1750 cpuhp_issue_call(cpu, state, false, NULL);
1753 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1754 mutex_unlock(&cpuhp_state_mutex);
1756 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1758 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1761 __cpuhp_remove_state_cpuslocked(state, invoke);
1764 EXPORT_SYMBOL(__cpuhp_remove_state);
1766 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1767 static ssize_t show_cpuhp_state(struct device *dev,
1768 struct device_attribute *attr, char *buf)
1770 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1772 return sprintf(buf, "%d\n", st->state);
1774 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1776 static ssize_t write_cpuhp_target(struct device *dev,
1777 struct device_attribute *attr,
1778 const char *buf, size_t count)
1780 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1781 struct cpuhp_step *sp;
1784 ret = kstrtoint(buf, 10, &target);
1788 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1789 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1792 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1796 ret = lock_device_hotplug_sysfs();
1800 mutex_lock(&cpuhp_state_mutex);
1801 sp = cpuhp_get_step(target);
1802 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1803 mutex_unlock(&cpuhp_state_mutex);
1807 if (st->state < target)
1808 ret = do_cpu_up(dev->id, target);
1810 ret = do_cpu_down(dev->id, target);
1812 unlock_device_hotplug();
1813 return ret ? ret : count;
1816 static ssize_t show_cpuhp_target(struct device *dev,
1817 struct device_attribute *attr, char *buf)
1819 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1821 return sprintf(buf, "%d\n", st->target);
1823 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1826 static ssize_t write_cpuhp_fail(struct device *dev,
1827 struct device_attribute *attr,
1828 const char *buf, size_t count)
1830 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1831 struct cpuhp_step *sp;
1834 ret = kstrtoint(buf, 10, &fail);
1839 * Cannot fail STARTING/DYING callbacks.
1841 if (cpuhp_is_atomic_state(fail))
1845 * Cannot fail anything that doesn't have callbacks.
1847 mutex_lock(&cpuhp_state_mutex);
1848 sp = cpuhp_get_step(fail);
1849 if (!sp->startup.single && !sp->teardown.single)
1851 mutex_unlock(&cpuhp_state_mutex);
1860 static ssize_t show_cpuhp_fail(struct device *dev,
1861 struct device_attribute *attr, char *buf)
1863 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1865 return sprintf(buf, "%d\n", st->fail);
1868 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1870 static struct attribute *cpuhp_cpu_attrs[] = {
1871 &dev_attr_state.attr,
1872 &dev_attr_target.attr,
1873 &dev_attr_fail.attr,
1877 static const struct attribute_group cpuhp_cpu_attr_group = {
1878 .attrs = cpuhp_cpu_attrs,
1883 static ssize_t show_cpuhp_states(struct device *dev,
1884 struct device_attribute *attr, char *buf)
1886 ssize_t cur, res = 0;
1889 mutex_lock(&cpuhp_state_mutex);
1890 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1891 struct cpuhp_step *sp = cpuhp_get_step(i);
1894 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1899 mutex_unlock(&cpuhp_state_mutex);
1902 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1904 static struct attribute *cpuhp_cpu_root_attrs[] = {
1905 &dev_attr_states.attr,
1909 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1910 .attrs = cpuhp_cpu_root_attrs,
1915 static int __init cpuhp_sysfs_init(void)
1919 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
1920 &cpuhp_cpu_root_attr_group);
1924 for_each_possible_cpu(cpu) {
1925 struct device *dev = get_cpu_device(cpu);
1929 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
1935 device_initcall(cpuhp_sysfs_init);
1939 * cpu_bit_bitmap[] is a special, "compressed" data structure that
1940 * represents all NR_CPUS bits binary values of 1<<nr.
1942 * It is used by cpumask_of() to get a constant address to a CPU
1943 * mask value that has a single bit set only.
1946 /* cpu_bit_bitmap[0] is empty - so we can back into it */
1947 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
1948 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
1949 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
1950 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
1952 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
1954 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
1955 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
1956 #if BITS_PER_LONG > 32
1957 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
1958 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
1961 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
1963 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
1964 EXPORT_SYMBOL(cpu_all_bits);
1966 #ifdef CONFIG_INIT_ALL_POSSIBLE
1967 struct cpumask __cpu_possible_mask __read_mostly
1970 struct cpumask __cpu_possible_mask __read_mostly;
1972 EXPORT_SYMBOL(__cpu_possible_mask);
1974 struct cpumask __cpu_online_mask __read_mostly;
1975 EXPORT_SYMBOL(__cpu_online_mask);
1977 struct cpumask __cpu_present_mask __read_mostly;
1978 EXPORT_SYMBOL(__cpu_present_mask);
1980 struct cpumask __cpu_active_mask __read_mostly;
1981 EXPORT_SYMBOL(__cpu_active_mask);
1983 void init_cpu_present(const struct cpumask *src)
1985 cpumask_copy(&__cpu_present_mask, src);
1988 void init_cpu_possible(const struct cpumask *src)
1990 cpumask_copy(&__cpu_possible_mask, src);
1993 void init_cpu_online(const struct cpumask *src)
1995 cpumask_copy(&__cpu_online_mask, src);
1999 * Activate the first processor.
2001 void __init boot_cpu_init(void)
2003 int cpu = smp_processor_id();
2005 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2006 set_cpu_online(cpu, true);
2007 set_cpu_active(cpu, true);
2008 set_cpu_present(cpu, true);
2009 set_cpu_possible(cpu, true);
2012 __boot_cpu_id = cpu;
2017 * Must be called _AFTER_ setting up the per_cpu areas
2019 void __init boot_cpu_state_init(void)
2021 per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;