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;
64 struct hlist_node *node;
65 struct hlist_node *last;
66 enum cpuhp_state cb_state;
68 struct completion done_up;
69 struct completion done_down;
73 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
74 .fail = CPUHP_INVALID,
77 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
78 static struct lockdep_map cpuhp_state_up_map =
79 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
80 static struct lockdep_map cpuhp_state_down_map =
81 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
84 static void inline cpuhp_lock_acquire(bool bringup)
86 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
89 static void inline cpuhp_lock_release(bool bringup)
91 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
95 static void inline cpuhp_lock_acquire(bool bringup) { }
96 static void inline cpuhp_lock_release(bool bringup) { }
101 * cpuhp_step - Hotplug state machine step
102 * @name: Name of the step
103 * @startup: Startup function of the step
104 * @teardown: Teardown function of the step
105 * @skip_onerr: Do not invoke the functions on error rollback
106 * Will go away once the notifiers are gone
107 * @cant_stop: Bringup/teardown can't be stopped at this step
112 int (*single)(unsigned int cpu);
113 int (*multi)(unsigned int cpu,
114 struct hlist_node *node);
117 int (*single)(unsigned int cpu);
118 int (*multi)(unsigned int cpu,
119 struct hlist_node *node);
121 struct hlist_head list;
127 static DEFINE_MUTEX(cpuhp_state_mutex);
128 static struct cpuhp_step cpuhp_bp_states[];
129 static struct cpuhp_step cpuhp_ap_states[];
131 static bool cpuhp_is_ap_state(enum cpuhp_state state)
134 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
135 * purposes as that state is handled explicitly in cpu_down.
137 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
140 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
142 struct cpuhp_step *sp;
144 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
149 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
150 * @cpu: The cpu for which the callback should be invoked
151 * @state: The state to do callbacks for
152 * @bringup: True if the bringup callback should be invoked
153 * @node: For multi-instance, do a single entry callback for install/remove
154 * @lastp: For multi-instance rollback, remember how far we got
156 * Called from cpu hotplug and from the state register machinery.
158 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
159 bool bringup, struct hlist_node *node,
160 struct hlist_node **lastp)
162 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
163 struct cpuhp_step *step = cpuhp_get_step(state);
164 int (*cbm)(unsigned int cpu, struct hlist_node *node);
165 int (*cb)(unsigned int cpu);
168 if (st->fail == state) {
169 st->fail = CPUHP_INVALID;
171 if (!(bringup ? step->startup.single : step->teardown.single))
177 if (!step->multi_instance) {
178 WARN_ON_ONCE(lastp && *lastp);
179 cb = bringup ? step->startup.single : step->teardown.single;
182 trace_cpuhp_enter(cpu, st->target, state, cb);
184 trace_cpuhp_exit(cpu, st->state, state, ret);
187 cbm = bringup ? step->startup.multi : step->teardown.multi;
191 /* Single invocation for instance add/remove */
193 WARN_ON_ONCE(lastp && *lastp);
194 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
195 ret = cbm(cpu, node);
196 trace_cpuhp_exit(cpu, st->state, state, ret);
200 /* State transition. Invoke on all instances */
202 hlist_for_each(node, &step->list) {
203 if (lastp && node == *lastp)
206 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
207 ret = cbm(cpu, node);
208 trace_cpuhp_exit(cpu, st->state, state, ret);
222 /* Rollback the instances if one failed */
223 cbm = !bringup ? step->startup.multi : step->teardown.multi;
227 hlist_for_each(node, &step->list) {
231 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
232 ret = cbm(cpu, node);
233 trace_cpuhp_exit(cpu, st->state, state, ret);
235 * Rollback must not fail,
243 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
246 wait_for_completion(done);
249 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
251 struct completion *done = bringup ? &st->done_up : &st->done_down;
256 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
258 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
260 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
263 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
264 static DEFINE_MUTEX(cpu_add_remove_lock);
265 bool cpuhp_tasks_frozen;
266 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
269 * The following two APIs (cpu_maps_update_begin/done) must be used when
270 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
272 void cpu_maps_update_begin(void)
274 mutex_lock(&cpu_add_remove_lock);
277 void cpu_maps_update_done(void)
279 mutex_unlock(&cpu_add_remove_lock);
283 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
284 * Should always be manipulated under cpu_add_remove_lock
286 static int cpu_hotplug_disabled;
288 #ifdef CONFIG_HOTPLUG_CPU
290 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
292 void cpus_read_lock(void)
294 percpu_down_read(&cpu_hotplug_lock);
296 EXPORT_SYMBOL_GPL(cpus_read_lock);
298 void cpus_read_unlock(void)
300 percpu_up_read(&cpu_hotplug_lock);
302 EXPORT_SYMBOL_GPL(cpus_read_unlock);
304 void cpus_write_lock(void)
306 percpu_down_write(&cpu_hotplug_lock);
309 void cpus_write_unlock(void)
311 percpu_up_write(&cpu_hotplug_lock);
314 void lockdep_assert_cpus_held(void)
316 percpu_rwsem_assert_held(&cpu_hotplug_lock);
320 * Wait for currently running CPU hotplug operations to complete (if any) and
321 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
322 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
323 * hotplug path before performing hotplug operations. So acquiring that lock
324 * guarantees mutual exclusion from any currently running hotplug operations.
326 void cpu_hotplug_disable(void)
328 cpu_maps_update_begin();
329 cpu_hotplug_disabled++;
330 cpu_maps_update_done();
332 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
334 static void __cpu_hotplug_enable(void)
336 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
338 cpu_hotplug_disabled--;
341 void cpu_hotplug_enable(void)
343 cpu_maps_update_begin();
344 __cpu_hotplug_enable();
345 cpu_maps_update_done();
347 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
348 #endif /* CONFIG_HOTPLUG_CPU */
350 #ifdef CONFIG_HOTPLUG_SMT
351 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
352 EXPORT_SYMBOL_GPL(cpu_smt_control);
354 static int __init smt_cmdline_disable(char *str)
356 cpu_smt_control = CPU_SMT_DISABLED;
357 if (str && !strcmp(str, "force")) {
358 pr_info("SMT: Force disabled\n");
359 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
363 early_param("nosmt", smt_cmdline_disable);
365 static inline bool cpu_smt_allowed(unsigned int cpu)
367 if (cpu_smt_control == CPU_SMT_ENABLED)
370 if (topology_is_primary_thread(cpu))
374 * On x86 it's required to boot all logical CPUs at least once so
375 * that the init code can get a chance to set CR4.MCE on each
376 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
377 * core will shutdown the machine.
379 return !per_cpu(cpuhp_state, cpu).booted_once;
382 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
385 static inline enum cpuhp_state
386 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
388 enum cpuhp_state prev_state = st->state;
390 st->rollback = false;
395 st->bringup = st->state < target;
401 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
406 * If we have st->last we need to undo partial multi_instance of this
407 * state first. Otherwise start undo at the previous state.
416 st->target = prev_state;
417 st->bringup = !st->bringup;
420 /* Regular hotplug invocation of the AP hotplug thread */
421 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
423 if (!st->single && st->state == st->target)
428 * Make sure the above stores are visible before should_run becomes
429 * true. Paired with the mb() above in cpuhp_thread_fun()
432 st->should_run = true;
433 wake_up_process(st->thread);
434 wait_for_ap_thread(st, st->bringup);
437 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
439 enum cpuhp_state prev_state;
442 prev_state = cpuhp_set_state(st, target);
444 if ((ret = st->result)) {
445 cpuhp_reset_state(st, prev_state);
452 static int bringup_wait_for_ap(unsigned int cpu)
454 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
456 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
457 wait_for_ap_thread(st, true);
458 if (WARN_ON_ONCE((!cpu_online(cpu))))
461 /* Unpark the stopper thread and the hotplug thread of the target cpu */
462 stop_machine_unpark(cpu);
463 kthread_unpark(st->thread);
466 * SMT soft disabling on X86 requires to bring the CPU out of the
467 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
468 * CPU marked itself as booted_once in cpu_notify_starting() so the
469 * cpu_smt_allowed() check will now return false if this is not the
472 if (!cpu_smt_allowed(cpu))
475 if (st->target <= CPUHP_AP_ONLINE_IDLE)
478 return cpuhp_kick_ap(st, st->target);
481 static int bringup_cpu(unsigned int cpu)
483 struct task_struct *idle = idle_thread_get(cpu);
487 * Some architectures have to walk the irq descriptors to
488 * setup the vector space for the cpu which comes online.
489 * Prevent irq alloc/free across the bringup.
493 /* Arch-specific enabling code. */
494 ret = __cpu_up(cpu, idle);
498 return bringup_wait_for_ap(cpu);
502 * Hotplug state machine related functions
505 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
507 for (st->state--; st->state > st->target; st->state--) {
508 struct cpuhp_step *step = cpuhp_get_step(st->state);
510 if (!step->skip_onerr)
511 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
515 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
516 enum cpuhp_state target)
518 enum cpuhp_state prev_state = st->state;
521 while (st->state < target) {
523 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
525 st->target = prev_state;
526 undo_cpu_up(cpu, st);
534 * The cpu hotplug threads manage the bringup and teardown of the cpus
536 static void cpuhp_create(unsigned int cpu)
538 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
540 init_completion(&st->done_up);
541 init_completion(&st->done_down);
544 static int cpuhp_should_run(unsigned int cpu)
546 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
548 return st->should_run;
552 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
553 * callbacks when a state gets [un]installed at runtime.
555 * Each invocation of this function by the smpboot thread does a single AP
558 * It has 3 modes of operation:
559 * - single: runs st->cb_state
560 * - up: runs ++st->state, while st->state < st->target
561 * - down: runs st->state--, while st->state > st->target
563 * When complete or on error, should_run is cleared and the completion is fired.
565 static void cpuhp_thread_fun(unsigned int cpu)
567 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
568 bool bringup = st->bringup;
569 enum cpuhp_state state;
572 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
573 * that if we see ->should_run we also see the rest of the state.
577 if (WARN_ON_ONCE(!st->should_run))
580 cpuhp_lock_acquire(bringup);
583 state = st->cb_state;
584 st->should_run = false;
589 st->should_run = (st->state < st->target);
590 WARN_ON_ONCE(st->state > st->target);
594 st->should_run = (st->state > st->target);
595 WARN_ON_ONCE(st->state < st->target);
599 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
602 struct cpuhp_step *step = cpuhp_get_step(state);
603 if (step->skip_onerr)
607 if (cpuhp_is_atomic_state(state)) {
609 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
613 * STARTING/DYING must not fail!
615 WARN_ON_ONCE(st->result);
617 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
622 * If we fail on a rollback, we're up a creek without no
623 * paddle, no way forward, no way back. We loose, thanks for
626 WARN_ON_ONCE(st->rollback);
627 st->should_run = false;
631 cpuhp_lock_release(bringup);
634 complete_ap_thread(st, bringup);
637 /* Invoke a single callback on a remote cpu */
639 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
640 struct hlist_node *node)
642 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
645 if (!cpu_online(cpu))
648 cpuhp_lock_acquire(false);
649 cpuhp_lock_release(false);
651 cpuhp_lock_acquire(true);
652 cpuhp_lock_release(true);
655 * If we are up and running, use the hotplug thread. For early calls
656 * we invoke the thread function directly.
659 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
661 st->rollback = false;
665 st->bringup = bringup;
666 st->cb_state = state;
672 * If we failed and did a partial, do a rollback.
674 if ((ret = st->result) && st->last) {
676 st->bringup = !bringup;
682 * Clean up the leftovers so the next hotplug operation wont use stale
685 st->node = st->last = NULL;
689 static int cpuhp_kick_ap_work(unsigned int cpu)
691 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
692 enum cpuhp_state prev_state = st->state;
695 cpuhp_lock_acquire(false);
696 cpuhp_lock_release(false);
698 cpuhp_lock_acquire(true);
699 cpuhp_lock_release(true);
701 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
702 ret = cpuhp_kick_ap(st, st->target);
703 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
708 static struct smp_hotplug_thread cpuhp_threads = {
709 .store = &cpuhp_state.thread,
710 .create = &cpuhp_create,
711 .thread_should_run = cpuhp_should_run,
712 .thread_fn = cpuhp_thread_fun,
713 .thread_comm = "cpuhp/%u",
717 void __init cpuhp_threads_init(void)
719 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
720 kthread_unpark(this_cpu_read(cpuhp_state.thread));
723 #ifdef CONFIG_HOTPLUG_CPU
725 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
728 * This function walks all processes, finds a valid mm struct for each one and
729 * then clears a corresponding bit in mm's cpumask. While this all sounds
730 * trivial, there are various non-obvious corner cases, which this function
731 * tries to solve in a safe manner.
733 * Also note that the function uses a somewhat relaxed locking scheme, so it may
734 * be called only for an already offlined CPU.
736 void clear_tasks_mm_cpumask(int cpu)
738 struct task_struct *p;
741 * This function is called after the cpu is taken down and marked
742 * offline, so its not like new tasks will ever get this cpu set in
743 * their mm mask. -- Peter Zijlstra
744 * Thus, we may use rcu_read_lock() here, instead of grabbing
745 * full-fledged tasklist_lock.
747 WARN_ON(cpu_online(cpu));
749 for_each_process(p) {
750 struct task_struct *t;
753 * Main thread might exit, but other threads may still have
754 * a valid mm. Find one.
756 t = find_lock_task_mm(p);
759 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
765 /* Take this CPU down. */
766 static int take_cpu_down(void *_param)
768 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
769 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
770 int err, cpu = smp_processor_id();
773 /* Ensure this CPU doesn't handle any more interrupts. */
774 err = __cpu_disable();
779 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
780 * do this step again.
782 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
784 /* Invoke the former CPU_DYING callbacks */
785 for (; st->state > target; st->state--) {
786 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
788 * DYING must not fail!
793 /* Give up timekeeping duties */
794 tick_handover_do_timer();
795 /* Park the stopper thread */
796 stop_machine_park(cpu);
800 static int takedown_cpu(unsigned int cpu)
802 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
805 /* Park the smpboot threads */
806 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
809 * Prevent irq alloc/free while the dying cpu reorganizes the
810 * interrupt affinities.
815 * So now all preempt/rcu users must observe !cpu_active().
817 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
819 /* CPU refused to die */
821 /* Unpark the hotplug thread so we can rollback there */
822 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
825 BUG_ON(cpu_online(cpu));
828 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
829 * runnable tasks from the cpu, there's only the idle task left now
830 * that the migration thread is done doing the stop_machine thing.
832 * Wait for the stop thread to go away.
834 wait_for_ap_thread(st, false);
835 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
837 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
840 hotplug_cpu__broadcast_tick_pull(cpu);
841 /* This actually kills the CPU. */
844 tick_cleanup_dead_cpu(cpu);
845 rcutree_migrate_callbacks(cpu);
849 static void cpuhp_complete_idle_dead(void *arg)
851 struct cpuhp_cpu_state *st = arg;
853 complete_ap_thread(st, false);
856 void cpuhp_report_idle_dead(void)
858 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
860 BUG_ON(st->state != CPUHP_AP_OFFLINE);
861 rcu_report_dead(smp_processor_id());
862 st->state = CPUHP_AP_IDLE_DEAD;
864 * We cannot call complete after rcu_report_dead() so we delegate it
867 smp_call_function_single(cpumask_first(cpu_online_mask),
868 cpuhp_complete_idle_dead, st, 0);
871 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
873 for (st->state++; st->state < st->target; st->state++) {
874 struct cpuhp_step *step = cpuhp_get_step(st->state);
876 if (!step->skip_onerr)
877 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
881 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
882 enum cpuhp_state target)
884 enum cpuhp_state prev_state = st->state;
887 for (; st->state > target; st->state--) {
888 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
890 st->target = prev_state;
891 undo_cpu_down(cpu, st);
898 /* Requires cpu_add_remove_lock to be held */
899 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
900 enum cpuhp_state target)
902 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
903 int prev_state, ret = 0;
905 if (num_online_cpus() == 1)
908 if (!cpu_present(cpu))
913 cpuhp_tasks_frozen = tasks_frozen;
915 prev_state = cpuhp_set_state(st, target);
917 * If the current CPU state is in the range of the AP hotplug thread,
918 * then we need to kick the thread.
920 if (st->state > CPUHP_TEARDOWN_CPU) {
921 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
922 ret = cpuhp_kick_ap_work(cpu);
924 * The AP side has done the error rollback already. Just
925 * return the error code..
931 * We might have stopped still in the range of the AP hotplug
932 * thread. Nothing to do anymore.
934 if (st->state > CPUHP_TEARDOWN_CPU)
940 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
941 * to do the further cleanups.
943 ret = cpuhp_down_callbacks(cpu, st, target);
944 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
945 cpuhp_reset_state(st, prev_state);
952 * Do post unplug cleanup. This is still protected against
953 * concurrent CPU hotplug via cpu_add_remove_lock.
955 lockup_detector_cleanup();
959 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
961 if (cpu_hotplug_disabled)
963 return _cpu_down(cpu, 0, target);
966 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
970 cpu_maps_update_begin();
971 err = cpu_down_maps_locked(cpu, target);
972 cpu_maps_update_done();
976 int cpu_down(unsigned int cpu)
978 return do_cpu_down(cpu, CPUHP_OFFLINE);
980 EXPORT_SYMBOL(cpu_down);
983 #define takedown_cpu NULL
984 #endif /*CONFIG_HOTPLUG_CPU*/
987 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
988 * @cpu: cpu that just started
990 * It must be called by the arch code on the new cpu, before the new cpu
991 * enables interrupts and before the "boot" cpu returns from __cpu_up().
993 void notify_cpu_starting(unsigned int cpu)
995 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
996 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
999 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1000 st->booted_once = true;
1001 while (st->state < target) {
1003 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1005 * STARTING must not fail!
1012 * Called from the idle task. Wake up the controlling task which brings the
1013 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1014 * the rest of the online bringup to the hotplug thread.
1016 void cpuhp_online_idle(enum cpuhp_state state)
1018 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1020 /* Happens for the boot cpu */
1021 if (state != CPUHP_AP_ONLINE_IDLE)
1024 st->state = CPUHP_AP_ONLINE_IDLE;
1025 complete_ap_thread(st, true);
1028 /* Requires cpu_add_remove_lock to be held */
1029 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1031 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1032 struct task_struct *idle;
1037 if (!cpu_present(cpu)) {
1043 * The caller of do_cpu_up might have raced with another
1044 * caller. Ignore it for now.
1046 if (st->state >= target)
1049 if (st->state == CPUHP_OFFLINE) {
1050 /* Let it fail before we try to bring the cpu up */
1051 idle = idle_thread_get(cpu);
1053 ret = PTR_ERR(idle);
1058 cpuhp_tasks_frozen = tasks_frozen;
1060 cpuhp_set_state(st, target);
1062 * If the current CPU state is in the range of the AP hotplug thread,
1063 * then we need to kick the thread once more.
1065 if (st->state > CPUHP_BRINGUP_CPU) {
1066 ret = cpuhp_kick_ap_work(cpu);
1068 * The AP side has done the error rollback already. Just
1069 * return the error code..
1076 * Try to reach the target state. We max out on the BP at
1077 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1078 * responsible for bringing it up to the target state.
1080 target = min((int)target, CPUHP_BRINGUP_CPU);
1081 ret = cpuhp_up_callbacks(cpu, st, target);
1083 cpus_write_unlock();
1087 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1091 if (!cpu_possible(cpu)) {
1092 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1094 #if defined(CONFIG_IA64)
1095 pr_err("please check additional_cpus= boot parameter\n");
1100 err = try_online_node(cpu_to_node(cpu));
1104 cpu_maps_update_begin();
1106 if (cpu_hotplug_disabled) {
1110 if (!cpu_smt_allowed(cpu)) {
1115 err = _cpu_up(cpu, 0, target);
1117 cpu_maps_update_done();
1121 int cpu_up(unsigned int cpu)
1123 return do_cpu_up(cpu, CPUHP_ONLINE);
1125 EXPORT_SYMBOL_GPL(cpu_up);
1127 #ifdef CONFIG_PM_SLEEP_SMP
1128 static cpumask_var_t frozen_cpus;
1130 int freeze_secondary_cpus(int primary)
1134 cpu_maps_update_begin();
1135 if (!cpu_online(primary))
1136 primary = cpumask_first(cpu_online_mask);
1138 * We take down all of the non-boot CPUs in one shot to avoid races
1139 * with the userspace trying to use the CPU hotplug at the same time
1141 cpumask_clear(frozen_cpus);
1143 pr_info("Disabling non-boot CPUs ...\n");
1144 for_each_online_cpu(cpu) {
1147 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1148 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1149 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1151 cpumask_set_cpu(cpu, frozen_cpus);
1153 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1159 BUG_ON(num_online_cpus() > 1);
1161 pr_err("Non-boot CPUs are not disabled\n");
1164 * Make sure the CPUs won't be enabled by someone else. We need to do
1165 * this even in case of failure as all disable_nonboot_cpus() users are
1166 * supposed to do enable_nonboot_cpus() on the failure path.
1168 cpu_hotplug_disabled++;
1170 cpu_maps_update_done();
1174 void __weak arch_enable_nonboot_cpus_begin(void)
1178 void __weak arch_enable_nonboot_cpus_end(void)
1182 void enable_nonboot_cpus(void)
1186 /* Allow everyone to use the CPU hotplug again */
1187 cpu_maps_update_begin();
1188 __cpu_hotplug_enable();
1189 if (cpumask_empty(frozen_cpus))
1192 pr_info("Enabling non-boot CPUs ...\n");
1194 arch_enable_nonboot_cpus_begin();
1196 for_each_cpu(cpu, frozen_cpus) {
1197 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1198 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1199 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1201 pr_info("CPU%d is up\n", cpu);
1204 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1207 arch_enable_nonboot_cpus_end();
1209 cpumask_clear(frozen_cpus);
1211 cpu_maps_update_done();
1214 static int __init alloc_frozen_cpus(void)
1216 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1220 core_initcall(alloc_frozen_cpus);
1223 * When callbacks for CPU hotplug notifications are being executed, we must
1224 * ensure that the state of the system with respect to the tasks being frozen
1225 * or not, as reported by the notification, remains unchanged *throughout the
1226 * duration* of the execution of the callbacks.
1227 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1229 * This synchronization is implemented by mutually excluding regular CPU
1230 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1231 * Hibernate notifications.
1234 cpu_hotplug_pm_callback(struct notifier_block *nb,
1235 unsigned long action, void *ptr)
1239 case PM_SUSPEND_PREPARE:
1240 case PM_HIBERNATION_PREPARE:
1241 cpu_hotplug_disable();
1244 case PM_POST_SUSPEND:
1245 case PM_POST_HIBERNATION:
1246 cpu_hotplug_enable();
1257 static int __init cpu_hotplug_pm_sync_init(void)
1260 * cpu_hotplug_pm_callback has higher priority than x86
1261 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1262 * to disable cpu hotplug to avoid cpu hotplug race.
1264 pm_notifier(cpu_hotplug_pm_callback, 0);
1267 core_initcall(cpu_hotplug_pm_sync_init);
1269 #endif /* CONFIG_PM_SLEEP_SMP */
1273 #endif /* CONFIG_SMP */
1275 /* Boot processor state steps */
1276 static struct cpuhp_step cpuhp_bp_states[] = {
1279 .startup.single = NULL,
1280 .teardown.single = NULL,
1283 [CPUHP_CREATE_THREADS]= {
1284 .name = "threads:prepare",
1285 .startup.single = smpboot_create_threads,
1286 .teardown.single = NULL,
1289 [CPUHP_PERF_PREPARE] = {
1290 .name = "perf:prepare",
1291 .startup.single = perf_event_init_cpu,
1292 .teardown.single = perf_event_exit_cpu,
1294 [CPUHP_WORKQUEUE_PREP] = {
1295 .name = "workqueue:prepare",
1296 .startup.single = workqueue_prepare_cpu,
1297 .teardown.single = NULL,
1299 [CPUHP_HRTIMERS_PREPARE] = {
1300 .name = "hrtimers:prepare",
1301 .startup.single = hrtimers_prepare_cpu,
1302 .teardown.single = hrtimers_dead_cpu,
1304 [CPUHP_SMPCFD_PREPARE] = {
1305 .name = "smpcfd:prepare",
1306 .startup.single = smpcfd_prepare_cpu,
1307 .teardown.single = smpcfd_dead_cpu,
1309 [CPUHP_RELAY_PREPARE] = {
1310 .name = "relay:prepare",
1311 .startup.single = relay_prepare_cpu,
1312 .teardown.single = NULL,
1314 [CPUHP_SLAB_PREPARE] = {
1315 .name = "slab:prepare",
1316 .startup.single = slab_prepare_cpu,
1317 .teardown.single = slab_dead_cpu,
1319 [CPUHP_RCUTREE_PREP] = {
1320 .name = "RCU/tree:prepare",
1321 .startup.single = rcutree_prepare_cpu,
1322 .teardown.single = rcutree_dead_cpu,
1325 * On the tear-down path, timers_dead_cpu() must be invoked
1326 * before blk_mq_queue_reinit_notify() from notify_dead(),
1327 * otherwise a RCU stall occurs.
1329 [CPUHP_TIMERS_PREPARE] = {
1330 .name = "timers:dead",
1331 .startup.single = timers_prepare_cpu,
1332 .teardown.single = timers_dead_cpu,
1334 /* Kicks the plugged cpu into life */
1335 [CPUHP_BRINGUP_CPU] = {
1336 .name = "cpu:bringup",
1337 .startup.single = bringup_cpu,
1338 .teardown.single = NULL,
1342 * Handled on controll processor until the plugged processor manages
1345 [CPUHP_TEARDOWN_CPU] = {
1346 .name = "cpu:teardown",
1347 .startup.single = NULL,
1348 .teardown.single = takedown_cpu,
1352 [CPUHP_BRINGUP_CPU] = { },
1356 /* Application processor state steps */
1357 static struct cpuhp_step cpuhp_ap_states[] = {
1359 /* Final state before CPU kills itself */
1360 [CPUHP_AP_IDLE_DEAD] = {
1361 .name = "idle:dead",
1364 * Last state before CPU enters the idle loop to die. Transient state
1365 * for synchronization.
1367 [CPUHP_AP_OFFLINE] = {
1368 .name = "ap:offline",
1371 /* First state is scheduler control. Interrupts are disabled */
1372 [CPUHP_AP_SCHED_STARTING] = {
1373 .name = "sched:starting",
1374 .startup.single = sched_cpu_starting,
1375 .teardown.single = sched_cpu_dying,
1377 [CPUHP_AP_RCUTREE_DYING] = {
1378 .name = "RCU/tree:dying",
1379 .startup.single = NULL,
1380 .teardown.single = rcutree_dying_cpu,
1382 [CPUHP_AP_SMPCFD_DYING] = {
1383 .name = "smpcfd:dying",
1384 .startup.single = NULL,
1385 .teardown.single = smpcfd_dying_cpu,
1387 /* Entry state on starting. Interrupts enabled from here on. Transient
1388 * state for synchronsization */
1389 [CPUHP_AP_ONLINE] = {
1390 .name = "ap:online",
1392 /* Handle smpboot threads park/unpark */
1393 [CPUHP_AP_SMPBOOT_THREADS] = {
1394 .name = "smpboot/threads:online",
1395 .startup.single = smpboot_unpark_threads,
1396 .teardown.single = smpboot_park_threads,
1398 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1399 .name = "irq/affinity:online",
1400 .startup.single = irq_affinity_online_cpu,
1401 .teardown.single = NULL,
1403 [CPUHP_AP_PERF_ONLINE] = {
1404 .name = "perf:online",
1405 .startup.single = perf_event_init_cpu,
1406 .teardown.single = perf_event_exit_cpu,
1408 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1409 .name = "workqueue:online",
1410 .startup.single = workqueue_online_cpu,
1411 .teardown.single = workqueue_offline_cpu,
1413 [CPUHP_AP_RCUTREE_ONLINE] = {
1414 .name = "RCU/tree:online",
1415 .startup.single = rcutree_online_cpu,
1416 .teardown.single = rcutree_offline_cpu,
1420 * The dynamically registered state space is here
1424 /* Last state is scheduler control setting the cpu active */
1425 [CPUHP_AP_ACTIVE] = {
1426 .name = "sched:active",
1427 .startup.single = sched_cpu_activate,
1428 .teardown.single = sched_cpu_deactivate,
1432 /* CPU is fully up and running. */
1435 .startup.single = NULL,
1436 .teardown.single = NULL,
1440 /* Sanity check for callbacks */
1441 static int cpuhp_cb_check(enum cpuhp_state state)
1443 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1449 * Returns a free for dynamic slot assignment of the Online state. The states
1450 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1451 * by having no name assigned.
1453 static int cpuhp_reserve_state(enum cpuhp_state state)
1455 enum cpuhp_state i, end;
1456 struct cpuhp_step *step;
1459 case CPUHP_AP_ONLINE_DYN:
1460 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN;
1461 end = CPUHP_AP_ONLINE_DYN_END;
1463 case CPUHP_BP_PREPARE_DYN:
1464 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN;
1465 end = CPUHP_BP_PREPARE_DYN_END;
1471 for (i = state; i <= end; i++, step++) {
1475 WARN(1, "No more dynamic states available for CPU hotplug\n");
1479 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1480 int (*startup)(unsigned int cpu),
1481 int (*teardown)(unsigned int cpu),
1482 bool multi_instance)
1484 /* (Un)Install the callbacks for further cpu hotplug operations */
1485 struct cpuhp_step *sp;
1489 * If name is NULL, then the state gets removed.
1491 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1492 * the first allocation from these dynamic ranges, so the removal
1493 * would trigger a new allocation and clear the wrong (already
1494 * empty) state, leaving the callbacks of the to be cleared state
1495 * dangling, which causes wreckage on the next hotplug operation.
1497 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1498 state == CPUHP_BP_PREPARE_DYN)) {
1499 ret = cpuhp_reserve_state(state);
1504 sp = cpuhp_get_step(state);
1505 if (name && sp->name)
1508 sp->startup.single = startup;
1509 sp->teardown.single = teardown;
1511 sp->multi_instance = multi_instance;
1512 INIT_HLIST_HEAD(&sp->list);
1516 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1518 return cpuhp_get_step(state)->teardown.single;
1522 * Call the startup/teardown function for a step either on the AP or
1523 * on the current CPU.
1525 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1526 struct hlist_node *node)
1528 struct cpuhp_step *sp = cpuhp_get_step(state);
1532 * If there's nothing to do, we done.
1533 * Relies on the union for multi_instance.
1535 if ((bringup && !sp->startup.single) ||
1536 (!bringup && !sp->teardown.single))
1539 * The non AP bound callbacks can fail on bringup. On teardown
1540 * e.g. module removal we crash for now.
1543 if (cpuhp_is_ap_state(state))
1544 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1546 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1548 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1550 BUG_ON(ret && !bringup);
1555 * Called from __cpuhp_setup_state on a recoverable failure.
1557 * Note: The teardown callbacks for rollback are not allowed to fail!
1559 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1560 struct hlist_node *node)
1564 /* Roll back the already executed steps on the other cpus */
1565 for_each_present_cpu(cpu) {
1566 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1567 int cpustate = st->state;
1569 if (cpu >= failedcpu)
1572 /* Did we invoke the startup call on that cpu ? */
1573 if (cpustate >= state)
1574 cpuhp_issue_call(cpu, state, false, node);
1578 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1579 struct hlist_node *node,
1582 struct cpuhp_step *sp;
1586 lockdep_assert_cpus_held();
1588 sp = cpuhp_get_step(state);
1589 if (sp->multi_instance == false)
1592 mutex_lock(&cpuhp_state_mutex);
1594 if (!invoke || !sp->startup.multi)
1598 * Try to call the startup callback for each present cpu
1599 * depending on the hotplug state of the cpu.
1601 for_each_present_cpu(cpu) {
1602 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1603 int cpustate = st->state;
1605 if (cpustate < state)
1608 ret = cpuhp_issue_call(cpu, state, true, node);
1610 if (sp->teardown.multi)
1611 cpuhp_rollback_install(cpu, state, node);
1617 hlist_add_head(node, &sp->list);
1619 mutex_unlock(&cpuhp_state_mutex);
1623 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1629 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1633 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1636 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1637 * @state: The state to setup
1638 * @invoke: If true, the startup function is invoked for cpus where
1639 * cpu state >= @state
1640 * @startup: startup callback function
1641 * @teardown: teardown callback function
1642 * @multi_instance: State is set up for multiple instances which get
1645 * The caller needs to hold cpus read locked while calling this function.
1648 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1649 * 0 for all other states
1650 * On failure: proper (negative) error code
1652 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1653 const char *name, bool invoke,
1654 int (*startup)(unsigned int cpu),
1655 int (*teardown)(unsigned int cpu),
1656 bool multi_instance)
1661 lockdep_assert_cpus_held();
1663 if (cpuhp_cb_check(state) || !name)
1666 mutex_lock(&cpuhp_state_mutex);
1668 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1671 dynstate = state == CPUHP_AP_ONLINE_DYN;
1672 if (ret > 0 && dynstate) {
1677 if (ret || !invoke || !startup)
1681 * Try to call the startup callback for each present cpu
1682 * depending on the hotplug state of the cpu.
1684 for_each_present_cpu(cpu) {
1685 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1686 int cpustate = st->state;
1688 if (cpustate < state)
1691 ret = cpuhp_issue_call(cpu, state, true, NULL);
1694 cpuhp_rollback_install(cpu, state, NULL);
1695 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1700 mutex_unlock(&cpuhp_state_mutex);
1702 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1703 * dynamically allocated state in case of success.
1705 if (!ret && dynstate)
1709 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1711 int __cpuhp_setup_state(enum cpuhp_state state,
1712 const char *name, bool invoke,
1713 int (*startup)(unsigned int cpu),
1714 int (*teardown)(unsigned int cpu),
1715 bool multi_instance)
1720 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1721 teardown, multi_instance);
1725 EXPORT_SYMBOL(__cpuhp_setup_state);
1727 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1728 struct hlist_node *node, bool invoke)
1730 struct cpuhp_step *sp = cpuhp_get_step(state);
1733 BUG_ON(cpuhp_cb_check(state));
1735 if (!sp->multi_instance)
1739 mutex_lock(&cpuhp_state_mutex);
1741 if (!invoke || !cpuhp_get_teardown_cb(state))
1744 * Call the teardown callback for each present cpu depending
1745 * on the hotplug state of the cpu. This function is not
1746 * allowed to fail currently!
1748 for_each_present_cpu(cpu) {
1749 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1750 int cpustate = st->state;
1752 if (cpustate >= state)
1753 cpuhp_issue_call(cpu, state, false, node);
1758 mutex_unlock(&cpuhp_state_mutex);
1763 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1766 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1767 * @state: The state to remove
1768 * @invoke: If true, the teardown function is invoked for cpus where
1769 * cpu state >= @state
1771 * The caller needs to hold cpus read locked while calling this function.
1772 * The teardown callback is currently not allowed to fail. Think
1773 * about module removal!
1775 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1777 struct cpuhp_step *sp = cpuhp_get_step(state);
1780 BUG_ON(cpuhp_cb_check(state));
1782 lockdep_assert_cpus_held();
1784 mutex_lock(&cpuhp_state_mutex);
1785 if (sp->multi_instance) {
1786 WARN(!hlist_empty(&sp->list),
1787 "Error: Removing state %d which has instances left.\n",
1792 if (!invoke || !cpuhp_get_teardown_cb(state))
1796 * Call the teardown callback for each present cpu depending
1797 * on the hotplug state of the cpu. This function is not
1798 * allowed to fail currently!
1800 for_each_present_cpu(cpu) {
1801 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1802 int cpustate = st->state;
1804 if (cpustate >= state)
1805 cpuhp_issue_call(cpu, state, false, NULL);
1808 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1809 mutex_unlock(&cpuhp_state_mutex);
1811 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1813 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1816 __cpuhp_remove_state_cpuslocked(state, invoke);
1819 EXPORT_SYMBOL(__cpuhp_remove_state);
1821 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1822 static ssize_t show_cpuhp_state(struct device *dev,
1823 struct device_attribute *attr, char *buf)
1825 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1827 return sprintf(buf, "%d\n", st->state);
1829 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1831 static ssize_t write_cpuhp_target(struct device *dev,
1832 struct device_attribute *attr,
1833 const char *buf, size_t count)
1835 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1836 struct cpuhp_step *sp;
1839 ret = kstrtoint(buf, 10, &target);
1843 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1844 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1847 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1851 ret = lock_device_hotplug_sysfs();
1855 mutex_lock(&cpuhp_state_mutex);
1856 sp = cpuhp_get_step(target);
1857 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1858 mutex_unlock(&cpuhp_state_mutex);
1862 if (st->state < target)
1863 ret = do_cpu_up(dev->id, target);
1865 ret = do_cpu_down(dev->id, target);
1867 unlock_device_hotplug();
1868 return ret ? ret : count;
1871 static ssize_t show_cpuhp_target(struct device *dev,
1872 struct device_attribute *attr, char *buf)
1874 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1876 return sprintf(buf, "%d\n", st->target);
1878 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1881 static ssize_t write_cpuhp_fail(struct device *dev,
1882 struct device_attribute *attr,
1883 const char *buf, size_t count)
1885 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1886 struct cpuhp_step *sp;
1889 ret = kstrtoint(buf, 10, &fail);
1894 * Cannot fail STARTING/DYING callbacks.
1896 if (cpuhp_is_atomic_state(fail))
1900 * Cannot fail anything that doesn't have callbacks.
1902 mutex_lock(&cpuhp_state_mutex);
1903 sp = cpuhp_get_step(fail);
1904 if (!sp->startup.single && !sp->teardown.single)
1906 mutex_unlock(&cpuhp_state_mutex);
1915 static ssize_t show_cpuhp_fail(struct device *dev,
1916 struct device_attribute *attr, char *buf)
1918 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1920 return sprintf(buf, "%d\n", st->fail);
1923 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1925 static struct attribute *cpuhp_cpu_attrs[] = {
1926 &dev_attr_state.attr,
1927 &dev_attr_target.attr,
1928 &dev_attr_fail.attr,
1932 static const struct attribute_group cpuhp_cpu_attr_group = {
1933 .attrs = cpuhp_cpu_attrs,
1938 static ssize_t show_cpuhp_states(struct device *dev,
1939 struct device_attribute *attr, char *buf)
1941 ssize_t cur, res = 0;
1944 mutex_lock(&cpuhp_state_mutex);
1945 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1946 struct cpuhp_step *sp = cpuhp_get_step(i);
1949 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1954 mutex_unlock(&cpuhp_state_mutex);
1957 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1959 static struct attribute *cpuhp_cpu_root_attrs[] = {
1960 &dev_attr_states.attr,
1964 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1965 .attrs = cpuhp_cpu_root_attrs,
1970 #ifdef CONFIG_HOTPLUG_SMT
1972 static const char *smt_states[] = {
1973 [CPU_SMT_ENABLED] = "on",
1974 [CPU_SMT_DISABLED] = "off",
1975 [CPU_SMT_FORCE_DISABLED] = "forceoff",
1976 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
1980 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
1982 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
1985 static void cpuhp_offline_cpu_device(unsigned int cpu)
1987 struct device *dev = get_cpu_device(cpu);
1989 dev->offline = true;
1990 /* Tell user space about the state change */
1991 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
1994 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
1998 cpu_maps_update_begin();
1999 for_each_online_cpu(cpu) {
2000 if (topology_is_primary_thread(cpu))
2002 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2006 * As this needs to hold the cpu maps lock it's impossible
2007 * to call device_offline() because that ends up calling
2008 * cpu_down() which takes cpu maps lock. cpu maps lock
2009 * needs to be held as this might race against in kernel
2010 * abusers of the hotplug machinery (thermal management).
2012 * So nothing would update device:offline state. That would
2013 * leave the sysfs entry stale and prevent onlining after
2014 * smt control has been changed to 'off' again. This is
2015 * called under the sysfs hotplug lock, so it is properly
2016 * serialized against the regular offline usage.
2018 cpuhp_offline_cpu_device(cpu);
2021 cpu_smt_control = ctrlval;
2022 cpu_maps_update_done();
2026 static void cpuhp_smt_enable(void)
2028 cpu_maps_update_begin();
2029 cpu_smt_control = CPU_SMT_ENABLED;
2030 cpu_maps_update_done();
2034 store_smt_control(struct device *dev, struct device_attribute *attr,
2035 const char *buf, size_t count)
2039 if (sysfs_streq(buf, "on"))
2040 ctrlval = CPU_SMT_ENABLED;
2041 else if (sysfs_streq(buf, "off"))
2042 ctrlval = CPU_SMT_DISABLED;
2043 else if (sysfs_streq(buf, "forceoff"))
2044 ctrlval = CPU_SMT_FORCE_DISABLED;
2048 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2051 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2054 ret = lock_device_hotplug_sysfs();
2058 if (ctrlval != cpu_smt_control) {
2060 case CPU_SMT_ENABLED:
2063 case CPU_SMT_DISABLED:
2064 case CPU_SMT_FORCE_DISABLED:
2065 ret = cpuhp_smt_disable(ctrlval);
2070 unlock_device_hotplug();
2071 return ret ? ret : count;
2073 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2076 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2078 bool active = topology_max_smt_threads() > 1;
2080 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2082 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2084 static struct attribute *cpuhp_smt_attrs[] = {
2085 &dev_attr_control.attr,
2086 &dev_attr_active.attr,
2090 static const struct attribute_group cpuhp_smt_attr_group = {
2091 .attrs = cpuhp_smt_attrs,
2096 static int __init cpu_smt_state_init(void)
2098 if (!topology_smt_supported())
2099 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
2101 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2102 &cpuhp_smt_attr_group);
2106 static inline int cpu_smt_state_init(void) { return 0; }
2109 static int __init cpuhp_sysfs_init(void)
2113 ret = cpu_smt_state_init();
2117 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2118 &cpuhp_cpu_root_attr_group);
2122 for_each_possible_cpu(cpu) {
2123 struct device *dev = get_cpu_device(cpu);
2127 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2133 device_initcall(cpuhp_sysfs_init);
2137 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2138 * represents all NR_CPUS bits binary values of 1<<nr.
2140 * It is used by cpumask_of() to get a constant address to a CPU
2141 * mask value that has a single bit set only.
2144 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2145 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2146 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2147 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2148 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2150 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2152 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2153 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2154 #if BITS_PER_LONG > 32
2155 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2156 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2159 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2161 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2162 EXPORT_SYMBOL(cpu_all_bits);
2164 #ifdef CONFIG_INIT_ALL_POSSIBLE
2165 struct cpumask __cpu_possible_mask __read_mostly
2168 struct cpumask __cpu_possible_mask __read_mostly;
2170 EXPORT_SYMBOL(__cpu_possible_mask);
2172 struct cpumask __cpu_online_mask __read_mostly;
2173 EXPORT_SYMBOL(__cpu_online_mask);
2175 struct cpumask __cpu_present_mask __read_mostly;
2176 EXPORT_SYMBOL(__cpu_present_mask);
2178 struct cpumask __cpu_active_mask __read_mostly;
2179 EXPORT_SYMBOL(__cpu_active_mask);
2181 void init_cpu_present(const struct cpumask *src)
2183 cpumask_copy(&__cpu_present_mask, src);
2186 void init_cpu_possible(const struct cpumask *src)
2188 cpumask_copy(&__cpu_possible_mask, src);
2191 void init_cpu_online(const struct cpumask *src)
2193 cpumask_copy(&__cpu_online_mask, src);
2197 * Activate the first processor.
2199 void __init boot_cpu_init(void)
2201 int cpu = smp_processor_id();
2203 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2204 set_cpu_online(cpu, true);
2205 set_cpu_active(cpu, true);
2206 set_cpu_present(cpu, true);
2207 set_cpu_possible(cpu, true);
2210 __boot_cpu_id = cpu;
2215 * Must be called _AFTER_ setting up the per_cpu areas
2217 void __init boot_cpu_hotplug_init(void)
2219 this_cpu_write(cpuhp_state.booted_once, true);
2220 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);