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 void __init cpu_smt_disable(bool force)
356 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
357 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
361 pr_info("SMT: Force disabled\n");
362 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
364 cpu_smt_control = CPU_SMT_DISABLED;
368 static int __init smt_cmdline_disable(char *str)
370 cpu_smt_disable(str && !strcmp(str, "force"));
373 early_param("nosmt", smt_cmdline_disable);
375 static inline bool cpu_smt_allowed(unsigned int cpu)
377 if (cpu_smt_control == CPU_SMT_ENABLED)
380 if (topology_is_primary_thread(cpu))
384 * On x86 it's required to boot all logical CPUs at least once so
385 * that the init code can get a chance to set CR4.MCE on each
386 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
387 * core will shutdown the machine.
389 return !per_cpu(cpuhp_state, cpu).booted_once;
392 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
395 static inline enum cpuhp_state
396 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
398 enum cpuhp_state prev_state = st->state;
400 st->rollback = false;
405 st->bringup = st->state < target;
411 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
416 * If we have st->last we need to undo partial multi_instance of this
417 * state first. Otherwise start undo at the previous state.
426 st->target = prev_state;
427 st->bringup = !st->bringup;
430 /* Regular hotplug invocation of the AP hotplug thread */
431 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
433 if (!st->single && st->state == st->target)
438 * Make sure the above stores are visible before should_run becomes
439 * true. Paired with the mb() above in cpuhp_thread_fun()
442 st->should_run = true;
443 wake_up_process(st->thread);
444 wait_for_ap_thread(st, st->bringup);
447 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
449 enum cpuhp_state prev_state;
452 prev_state = cpuhp_set_state(st, target);
454 if ((ret = st->result)) {
455 cpuhp_reset_state(st, prev_state);
462 static int bringup_wait_for_ap(unsigned int cpu)
464 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
466 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
467 wait_for_ap_thread(st, true);
468 if (WARN_ON_ONCE((!cpu_online(cpu))))
471 /* Unpark the stopper thread and the hotplug thread of the target cpu */
472 stop_machine_unpark(cpu);
473 kthread_unpark(st->thread);
476 * SMT soft disabling on X86 requires to bring the CPU out of the
477 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
478 * CPU marked itself as booted_once in cpu_notify_starting() so the
479 * cpu_smt_allowed() check will now return false if this is not the
482 if (!cpu_smt_allowed(cpu))
485 if (st->target <= CPUHP_AP_ONLINE_IDLE)
488 return cpuhp_kick_ap(st, st->target);
491 static int bringup_cpu(unsigned int cpu)
493 struct task_struct *idle = idle_thread_get(cpu);
497 * Some architectures have to walk the irq descriptors to
498 * setup the vector space for the cpu which comes online.
499 * Prevent irq alloc/free across the bringup.
503 /* Arch-specific enabling code. */
504 ret = __cpu_up(cpu, idle);
508 return bringup_wait_for_ap(cpu);
512 * Hotplug state machine related functions
515 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
517 for (st->state--; st->state > st->target; st->state--) {
518 struct cpuhp_step *step = cpuhp_get_step(st->state);
520 if (!step->skip_onerr)
521 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
525 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
526 enum cpuhp_state target)
528 enum cpuhp_state prev_state = st->state;
531 while (st->state < target) {
533 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
535 st->target = prev_state;
536 undo_cpu_up(cpu, st);
544 * The cpu hotplug threads manage the bringup and teardown of the cpus
546 static void cpuhp_create(unsigned int cpu)
548 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
550 init_completion(&st->done_up);
551 init_completion(&st->done_down);
554 static int cpuhp_should_run(unsigned int cpu)
556 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
558 return st->should_run;
562 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
563 * callbacks when a state gets [un]installed at runtime.
565 * Each invocation of this function by the smpboot thread does a single AP
568 * It has 3 modes of operation:
569 * - single: runs st->cb_state
570 * - up: runs ++st->state, while st->state < st->target
571 * - down: runs st->state--, while st->state > st->target
573 * When complete or on error, should_run is cleared and the completion is fired.
575 static void cpuhp_thread_fun(unsigned int cpu)
577 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
578 bool bringup = st->bringup;
579 enum cpuhp_state state;
582 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
583 * that if we see ->should_run we also see the rest of the state.
587 if (WARN_ON_ONCE(!st->should_run))
590 cpuhp_lock_acquire(bringup);
593 state = st->cb_state;
594 st->should_run = false;
599 st->should_run = (st->state < st->target);
600 WARN_ON_ONCE(st->state > st->target);
604 st->should_run = (st->state > st->target);
605 WARN_ON_ONCE(st->state < st->target);
609 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
612 struct cpuhp_step *step = cpuhp_get_step(state);
613 if (step->skip_onerr)
617 if (cpuhp_is_atomic_state(state)) {
619 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
623 * STARTING/DYING must not fail!
625 WARN_ON_ONCE(st->result);
627 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
632 * If we fail on a rollback, we're up a creek without no
633 * paddle, no way forward, no way back. We loose, thanks for
636 WARN_ON_ONCE(st->rollback);
637 st->should_run = false;
641 cpuhp_lock_release(bringup);
644 complete_ap_thread(st, bringup);
647 /* Invoke a single callback on a remote cpu */
649 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
650 struct hlist_node *node)
652 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
655 if (!cpu_online(cpu))
658 cpuhp_lock_acquire(false);
659 cpuhp_lock_release(false);
661 cpuhp_lock_acquire(true);
662 cpuhp_lock_release(true);
665 * If we are up and running, use the hotplug thread. For early calls
666 * we invoke the thread function directly.
669 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
671 st->rollback = false;
675 st->bringup = bringup;
676 st->cb_state = state;
682 * If we failed and did a partial, do a rollback.
684 if ((ret = st->result) && st->last) {
686 st->bringup = !bringup;
692 * Clean up the leftovers so the next hotplug operation wont use stale
695 st->node = st->last = NULL;
699 static int cpuhp_kick_ap_work(unsigned int cpu)
701 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
702 enum cpuhp_state prev_state = st->state;
705 cpuhp_lock_acquire(false);
706 cpuhp_lock_release(false);
708 cpuhp_lock_acquire(true);
709 cpuhp_lock_release(true);
711 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
712 ret = cpuhp_kick_ap(st, st->target);
713 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
718 static struct smp_hotplug_thread cpuhp_threads = {
719 .store = &cpuhp_state.thread,
720 .create = &cpuhp_create,
721 .thread_should_run = cpuhp_should_run,
722 .thread_fn = cpuhp_thread_fun,
723 .thread_comm = "cpuhp/%u",
727 void __init cpuhp_threads_init(void)
729 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
730 kthread_unpark(this_cpu_read(cpuhp_state.thread));
733 #ifdef CONFIG_HOTPLUG_CPU
735 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
738 * This function walks all processes, finds a valid mm struct for each one and
739 * then clears a corresponding bit in mm's cpumask. While this all sounds
740 * trivial, there are various non-obvious corner cases, which this function
741 * tries to solve in a safe manner.
743 * Also note that the function uses a somewhat relaxed locking scheme, so it may
744 * be called only for an already offlined CPU.
746 void clear_tasks_mm_cpumask(int cpu)
748 struct task_struct *p;
751 * This function is called after the cpu is taken down and marked
752 * offline, so its not like new tasks will ever get this cpu set in
753 * their mm mask. -- Peter Zijlstra
754 * Thus, we may use rcu_read_lock() here, instead of grabbing
755 * full-fledged tasklist_lock.
757 WARN_ON(cpu_online(cpu));
759 for_each_process(p) {
760 struct task_struct *t;
763 * Main thread might exit, but other threads may still have
764 * a valid mm. Find one.
766 t = find_lock_task_mm(p);
769 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
775 /* Take this CPU down. */
776 static int take_cpu_down(void *_param)
778 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
779 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
780 int err, cpu = smp_processor_id();
783 /* Ensure this CPU doesn't handle any more interrupts. */
784 err = __cpu_disable();
789 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
790 * do this step again.
792 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
794 /* Invoke the former CPU_DYING callbacks */
795 for (; st->state > target; st->state--) {
796 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
798 * DYING must not fail!
803 /* Give up timekeeping duties */
804 tick_handover_do_timer();
805 /* Park the stopper thread */
806 stop_machine_park(cpu);
810 static int takedown_cpu(unsigned int cpu)
812 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
815 /* Park the smpboot threads */
816 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
819 * Prevent irq alloc/free while the dying cpu reorganizes the
820 * interrupt affinities.
825 * So now all preempt/rcu users must observe !cpu_active().
827 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
829 /* CPU refused to die */
831 /* Unpark the hotplug thread so we can rollback there */
832 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
835 BUG_ON(cpu_online(cpu));
838 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
839 * runnable tasks from the cpu, there's only the idle task left now
840 * that the migration thread is done doing the stop_machine thing.
842 * Wait for the stop thread to go away.
844 wait_for_ap_thread(st, false);
845 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
847 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
850 hotplug_cpu__broadcast_tick_pull(cpu);
851 /* This actually kills the CPU. */
854 tick_cleanup_dead_cpu(cpu);
855 rcutree_migrate_callbacks(cpu);
859 static void cpuhp_complete_idle_dead(void *arg)
861 struct cpuhp_cpu_state *st = arg;
863 complete_ap_thread(st, false);
866 void cpuhp_report_idle_dead(void)
868 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
870 BUG_ON(st->state != CPUHP_AP_OFFLINE);
871 rcu_report_dead(smp_processor_id());
872 st->state = CPUHP_AP_IDLE_DEAD;
874 * We cannot call complete after rcu_report_dead() so we delegate it
877 smp_call_function_single(cpumask_first(cpu_online_mask),
878 cpuhp_complete_idle_dead, st, 0);
881 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
883 for (st->state++; st->state < st->target; st->state++) {
884 struct cpuhp_step *step = cpuhp_get_step(st->state);
886 if (!step->skip_onerr)
887 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
891 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
892 enum cpuhp_state target)
894 enum cpuhp_state prev_state = st->state;
897 for (; st->state > target; st->state--) {
898 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
900 st->target = prev_state;
901 undo_cpu_down(cpu, st);
908 /* Requires cpu_add_remove_lock to be held */
909 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
910 enum cpuhp_state target)
912 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
913 int prev_state, ret = 0;
915 if (num_online_cpus() == 1)
918 if (!cpu_present(cpu))
923 cpuhp_tasks_frozen = tasks_frozen;
925 prev_state = cpuhp_set_state(st, target);
927 * If the current CPU state is in the range of the AP hotplug thread,
928 * then we need to kick the thread.
930 if (st->state > CPUHP_TEARDOWN_CPU) {
931 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
932 ret = cpuhp_kick_ap_work(cpu);
934 * The AP side has done the error rollback already. Just
935 * return the error code..
941 * We might have stopped still in the range of the AP hotplug
942 * thread. Nothing to do anymore.
944 if (st->state > CPUHP_TEARDOWN_CPU)
950 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
951 * to do the further cleanups.
953 ret = cpuhp_down_callbacks(cpu, st, target);
954 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
955 cpuhp_reset_state(st, prev_state);
962 * Do post unplug cleanup. This is still protected against
963 * concurrent CPU hotplug via cpu_add_remove_lock.
965 lockup_detector_cleanup();
969 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
971 if (cpu_hotplug_disabled)
973 return _cpu_down(cpu, 0, target);
976 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
980 cpu_maps_update_begin();
981 err = cpu_down_maps_locked(cpu, target);
982 cpu_maps_update_done();
986 int cpu_down(unsigned int cpu)
988 return do_cpu_down(cpu, CPUHP_OFFLINE);
990 EXPORT_SYMBOL(cpu_down);
993 #define takedown_cpu NULL
994 #endif /*CONFIG_HOTPLUG_CPU*/
997 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
998 * @cpu: cpu that just started
1000 * It must be called by the arch code on the new cpu, before the new cpu
1001 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1003 void notify_cpu_starting(unsigned int cpu)
1005 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1006 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1009 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1010 st->booted_once = true;
1011 while (st->state < target) {
1013 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1015 * STARTING must not fail!
1022 * Called from the idle task. Wake up the controlling task which brings the
1023 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1024 * the rest of the online bringup to the hotplug thread.
1026 void cpuhp_online_idle(enum cpuhp_state state)
1028 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1030 /* Happens for the boot cpu */
1031 if (state != CPUHP_AP_ONLINE_IDLE)
1034 st->state = CPUHP_AP_ONLINE_IDLE;
1035 complete_ap_thread(st, true);
1038 /* Requires cpu_add_remove_lock to be held */
1039 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1041 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1042 struct task_struct *idle;
1047 if (!cpu_present(cpu)) {
1053 * The caller of do_cpu_up might have raced with another
1054 * caller. Ignore it for now.
1056 if (st->state >= target)
1059 if (st->state == CPUHP_OFFLINE) {
1060 /* Let it fail before we try to bring the cpu up */
1061 idle = idle_thread_get(cpu);
1063 ret = PTR_ERR(idle);
1068 cpuhp_tasks_frozen = tasks_frozen;
1070 cpuhp_set_state(st, target);
1072 * If the current CPU state is in the range of the AP hotplug thread,
1073 * then we need to kick the thread once more.
1075 if (st->state > CPUHP_BRINGUP_CPU) {
1076 ret = cpuhp_kick_ap_work(cpu);
1078 * The AP side has done the error rollback already. Just
1079 * return the error code..
1086 * Try to reach the target state. We max out on the BP at
1087 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1088 * responsible for bringing it up to the target state.
1090 target = min((int)target, CPUHP_BRINGUP_CPU);
1091 ret = cpuhp_up_callbacks(cpu, st, target);
1093 cpus_write_unlock();
1097 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1101 if (!cpu_possible(cpu)) {
1102 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1104 #if defined(CONFIG_IA64)
1105 pr_err("please check additional_cpus= boot parameter\n");
1110 err = try_online_node(cpu_to_node(cpu));
1114 cpu_maps_update_begin();
1116 if (cpu_hotplug_disabled) {
1120 if (!cpu_smt_allowed(cpu)) {
1125 err = _cpu_up(cpu, 0, target);
1127 cpu_maps_update_done();
1131 int cpu_up(unsigned int cpu)
1133 return do_cpu_up(cpu, CPUHP_ONLINE);
1135 EXPORT_SYMBOL_GPL(cpu_up);
1137 #ifdef CONFIG_PM_SLEEP_SMP
1138 static cpumask_var_t frozen_cpus;
1140 int freeze_secondary_cpus(int primary)
1144 cpu_maps_update_begin();
1145 if (!cpu_online(primary))
1146 primary = cpumask_first(cpu_online_mask);
1148 * We take down all of the non-boot CPUs in one shot to avoid races
1149 * with the userspace trying to use the CPU hotplug at the same time
1151 cpumask_clear(frozen_cpus);
1153 pr_info("Disabling non-boot CPUs ...\n");
1154 for_each_online_cpu(cpu) {
1157 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1158 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1159 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1161 cpumask_set_cpu(cpu, frozen_cpus);
1163 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1169 BUG_ON(num_online_cpus() > 1);
1171 pr_err("Non-boot CPUs are not disabled\n");
1174 * Make sure the CPUs won't be enabled by someone else. We need to do
1175 * this even in case of failure as all disable_nonboot_cpus() users are
1176 * supposed to do enable_nonboot_cpus() on the failure path.
1178 cpu_hotplug_disabled++;
1180 cpu_maps_update_done();
1184 void __weak arch_enable_nonboot_cpus_begin(void)
1188 void __weak arch_enable_nonboot_cpus_end(void)
1192 void enable_nonboot_cpus(void)
1196 /* Allow everyone to use the CPU hotplug again */
1197 cpu_maps_update_begin();
1198 __cpu_hotplug_enable();
1199 if (cpumask_empty(frozen_cpus))
1202 pr_info("Enabling non-boot CPUs ...\n");
1204 arch_enable_nonboot_cpus_begin();
1206 for_each_cpu(cpu, frozen_cpus) {
1207 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1208 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1209 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1211 pr_info("CPU%d is up\n", cpu);
1214 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1217 arch_enable_nonboot_cpus_end();
1219 cpumask_clear(frozen_cpus);
1221 cpu_maps_update_done();
1224 static int __init alloc_frozen_cpus(void)
1226 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1230 core_initcall(alloc_frozen_cpus);
1233 * When callbacks for CPU hotplug notifications are being executed, we must
1234 * ensure that the state of the system with respect to the tasks being frozen
1235 * or not, as reported by the notification, remains unchanged *throughout the
1236 * duration* of the execution of the callbacks.
1237 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1239 * This synchronization is implemented by mutually excluding regular CPU
1240 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1241 * Hibernate notifications.
1244 cpu_hotplug_pm_callback(struct notifier_block *nb,
1245 unsigned long action, void *ptr)
1249 case PM_SUSPEND_PREPARE:
1250 case PM_HIBERNATION_PREPARE:
1251 cpu_hotplug_disable();
1254 case PM_POST_SUSPEND:
1255 case PM_POST_HIBERNATION:
1256 cpu_hotplug_enable();
1267 static int __init cpu_hotplug_pm_sync_init(void)
1270 * cpu_hotplug_pm_callback has higher priority than x86
1271 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1272 * to disable cpu hotplug to avoid cpu hotplug race.
1274 pm_notifier(cpu_hotplug_pm_callback, 0);
1277 core_initcall(cpu_hotplug_pm_sync_init);
1279 #endif /* CONFIG_PM_SLEEP_SMP */
1283 #endif /* CONFIG_SMP */
1285 /* Boot processor state steps */
1286 static struct cpuhp_step cpuhp_bp_states[] = {
1289 .startup.single = NULL,
1290 .teardown.single = NULL,
1293 [CPUHP_CREATE_THREADS]= {
1294 .name = "threads:prepare",
1295 .startup.single = smpboot_create_threads,
1296 .teardown.single = NULL,
1299 [CPUHP_PERF_PREPARE] = {
1300 .name = "perf:prepare",
1301 .startup.single = perf_event_init_cpu,
1302 .teardown.single = perf_event_exit_cpu,
1304 [CPUHP_WORKQUEUE_PREP] = {
1305 .name = "workqueue:prepare",
1306 .startup.single = workqueue_prepare_cpu,
1307 .teardown.single = NULL,
1309 [CPUHP_HRTIMERS_PREPARE] = {
1310 .name = "hrtimers:prepare",
1311 .startup.single = hrtimers_prepare_cpu,
1312 .teardown.single = hrtimers_dead_cpu,
1314 [CPUHP_SMPCFD_PREPARE] = {
1315 .name = "smpcfd:prepare",
1316 .startup.single = smpcfd_prepare_cpu,
1317 .teardown.single = smpcfd_dead_cpu,
1319 [CPUHP_RELAY_PREPARE] = {
1320 .name = "relay:prepare",
1321 .startup.single = relay_prepare_cpu,
1322 .teardown.single = NULL,
1324 [CPUHP_SLAB_PREPARE] = {
1325 .name = "slab:prepare",
1326 .startup.single = slab_prepare_cpu,
1327 .teardown.single = slab_dead_cpu,
1329 [CPUHP_RCUTREE_PREP] = {
1330 .name = "RCU/tree:prepare",
1331 .startup.single = rcutree_prepare_cpu,
1332 .teardown.single = rcutree_dead_cpu,
1335 * On the tear-down path, timers_dead_cpu() must be invoked
1336 * before blk_mq_queue_reinit_notify() from notify_dead(),
1337 * otherwise a RCU stall occurs.
1339 [CPUHP_TIMERS_PREPARE] = {
1340 .name = "timers:dead",
1341 .startup.single = timers_prepare_cpu,
1342 .teardown.single = timers_dead_cpu,
1344 /* Kicks the plugged cpu into life */
1345 [CPUHP_BRINGUP_CPU] = {
1346 .name = "cpu:bringup",
1347 .startup.single = bringup_cpu,
1348 .teardown.single = NULL,
1352 * Handled on controll processor until the plugged processor manages
1355 [CPUHP_TEARDOWN_CPU] = {
1356 .name = "cpu:teardown",
1357 .startup.single = NULL,
1358 .teardown.single = takedown_cpu,
1362 [CPUHP_BRINGUP_CPU] = { },
1366 /* Application processor state steps */
1367 static struct cpuhp_step cpuhp_ap_states[] = {
1369 /* Final state before CPU kills itself */
1370 [CPUHP_AP_IDLE_DEAD] = {
1371 .name = "idle:dead",
1374 * Last state before CPU enters the idle loop to die. Transient state
1375 * for synchronization.
1377 [CPUHP_AP_OFFLINE] = {
1378 .name = "ap:offline",
1381 /* First state is scheduler control. Interrupts are disabled */
1382 [CPUHP_AP_SCHED_STARTING] = {
1383 .name = "sched:starting",
1384 .startup.single = sched_cpu_starting,
1385 .teardown.single = sched_cpu_dying,
1387 [CPUHP_AP_RCUTREE_DYING] = {
1388 .name = "RCU/tree:dying",
1389 .startup.single = NULL,
1390 .teardown.single = rcutree_dying_cpu,
1392 [CPUHP_AP_SMPCFD_DYING] = {
1393 .name = "smpcfd:dying",
1394 .startup.single = NULL,
1395 .teardown.single = smpcfd_dying_cpu,
1397 /* Entry state on starting. Interrupts enabled from here on. Transient
1398 * state for synchronsization */
1399 [CPUHP_AP_ONLINE] = {
1400 .name = "ap:online",
1402 /* Handle smpboot threads park/unpark */
1403 [CPUHP_AP_SMPBOOT_THREADS] = {
1404 .name = "smpboot/threads:online",
1405 .startup.single = smpboot_unpark_threads,
1406 .teardown.single = smpboot_park_threads,
1408 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1409 .name = "irq/affinity:online",
1410 .startup.single = irq_affinity_online_cpu,
1411 .teardown.single = NULL,
1413 [CPUHP_AP_PERF_ONLINE] = {
1414 .name = "perf:online",
1415 .startup.single = perf_event_init_cpu,
1416 .teardown.single = perf_event_exit_cpu,
1418 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1419 .name = "workqueue:online",
1420 .startup.single = workqueue_online_cpu,
1421 .teardown.single = workqueue_offline_cpu,
1423 [CPUHP_AP_RCUTREE_ONLINE] = {
1424 .name = "RCU/tree:online",
1425 .startup.single = rcutree_online_cpu,
1426 .teardown.single = rcutree_offline_cpu,
1430 * The dynamically registered state space is here
1434 /* Last state is scheduler control setting the cpu active */
1435 [CPUHP_AP_ACTIVE] = {
1436 .name = "sched:active",
1437 .startup.single = sched_cpu_activate,
1438 .teardown.single = sched_cpu_deactivate,
1442 /* CPU is fully up and running. */
1445 .startup.single = NULL,
1446 .teardown.single = NULL,
1450 /* Sanity check for callbacks */
1451 static int cpuhp_cb_check(enum cpuhp_state state)
1453 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1459 * Returns a free for dynamic slot assignment of the Online state. The states
1460 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1461 * by having no name assigned.
1463 static int cpuhp_reserve_state(enum cpuhp_state state)
1465 enum cpuhp_state i, end;
1466 struct cpuhp_step *step;
1469 case CPUHP_AP_ONLINE_DYN:
1470 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN;
1471 end = CPUHP_AP_ONLINE_DYN_END;
1473 case CPUHP_BP_PREPARE_DYN:
1474 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN;
1475 end = CPUHP_BP_PREPARE_DYN_END;
1481 for (i = state; i <= end; i++, step++) {
1485 WARN(1, "No more dynamic states available for CPU hotplug\n");
1489 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1490 int (*startup)(unsigned int cpu),
1491 int (*teardown)(unsigned int cpu),
1492 bool multi_instance)
1494 /* (Un)Install the callbacks for further cpu hotplug operations */
1495 struct cpuhp_step *sp;
1499 * If name is NULL, then the state gets removed.
1501 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1502 * the first allocation from these dynamic ranges, so the removal
1503 * would trigger a new allocation and clear the wrong (already
1504 * empty) state, leaving the callbacks of the to be cleared state
1505 * dangling, which causes wreckage on the next hotplug operation.
1507 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1508 state == CPUHP_BP_PREPARE_DYN)) {
1509 ret = cpuhp_reserve_state(state);
1514 sp = cpuhp_get_step(state);
1515 if (name && sp->name)
1518 sp->startup.single = startup;
1519 sp->teardown.single = teardown;
1521 sp->multi_instance = multi_instance;
1522 INIT_HLIST_HEAD(&sp->list);
1526 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1528 return cpuhp_get_step(state)->teardown.single;
1532 * Call the startup/teardown function for a step either on the AP or
1533 * on the current CPU.
1535 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1536 struct hlist_node *node)
1538 struct cpuhp_step *sp = cpuhp_get_step(state);
1542 * If there's nothing to do, we done.
1543 * Relies on the union for multi_instance.
1545 if ((bringup && !sp->startup.single) ||
1546 (!bringup && !sp->teardown.single))
1549 * The non AP bound callbacks can fail on bringup. On teardown
1550 * e.g. module removal we crash for now.
1553 if (cpuhp_is_ap_state(state))
1554 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1556 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1558 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1560 BUG_ON(ret && !bringup);
1565 * Called from __cpuhp_setup_state on a recoverable failure.
1567 * Note: The teardown callbacks for rollback are not allowed to fail!
1569 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1570 struct hlist_node *node)
1574 /* Roll back the already executed steps on the other cpus */
1575 for_each_present_cpu(cpu) {
1576 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1577 int cpustate = st->state;
1579 if (cpu >= failedcpu)
1582 /* Did we invoke the startup call on that cpu ? */
1583 if (cpustate >= state)
1584 cpuhp_issue_call(cpu, state, false, node);
1588 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1589 struct hlist_node *node,
1592 struct cpuhp_step *sp;
1596 lockdep_assert_cpus_held();
1598 sp = cpuhp_get_step(state);
1599 if (sp->multi_instance == false)
1602 mutex_lock(&cpuhp_state_mutex);
1604 if (!invoke || !sp->startup.multi)
1608 * Try to call the startup callback for each present cpu
1609 * depending on the hotplug state of the cpu.
1611 for_each_present_cpu(cpu) {
1612 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1613 int cpustate = st->state;
1615 if (cpustate < state)
1618 ret = cpuhp_issue_call(cpu, state, true, node);
1620 if (sp->teardown.multi)
1621 cpuhp_rollback_install(cpu, state, node);
1627 hlist_add_head(node, &sp->list);
1629 mutex_unlock(&cpuhp_state_mutex);
1633 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1639 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1643 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1646 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1647 * @state: The state to setup
1648 * @invoke: If true, the startup function is invoked for cpus where
1649 * cpu state >= @state
1650 * @startup: startup callback function
1651 * @teardown: teardown callback function
1652 * @multi_instance: State is set up for multiple instances which get
1655 * The caller needs to hold cpus read locked while calling this function.
1658 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1659 * 0 for all other states
1660 * On failure: proper (negative) error code
1662 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1663 const char *name, bool invoke,
1664 int (*startup)(unsigned int cpu),
1665 int (*teardown)(unsigned int cpu),
1666 bool multi_instance)
1671 lockdep_assert_cpus_held();
1673 if (cpuhp_cb_check(state) || !name)
1676 mutex_lock(&cpuhp_state_mutex);
1678 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1681 dynstate = state == CPUHP_AP_ONLINE_DYN;
1682 if (ret > 0 && dynstate) {
1687 if (ret || !invoke || !startup)
1691 * Try to call the startup callback for each present cpu
1692 * depending on the hotplug state of the cpu.
1694 for_each_present_cpu(cpu) {
1695 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1696 int cpustate = st->state;
1698 if (cpustate < state)
1701 ret = cpuhp_issue_call(cpu, state, true, NULL);
1704 cpuhp_rollback_install(cpu, state, NULL);
1705 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1710 mutex_unlock(&cpuhp_state_mutex);
1712 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1713 * dynamically allocated state in case of success.
1715 if (!ret && dynstate)
1719 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1721 int __cpuhp_setup_state(enum cpuhp_state state,
1722 const char *name, bool invoke,
1723 int (*startup)(unsigned int cpu),
1724 int (*teardown)(unsigned int cpu),
1725 bool multi_instance)
1730 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1731 teardown, multi_instance);
1735 EXPORT_SYMBOL(__cpuhp_setup_state);
1737 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1738 struct hlist_node *node, bool invoke)
1740 struct cpuhp_step *sp = cpuhp_get_step(state);
1743 BUG_ON(cpuhp_cb_check(state));
1745 if (!sp->multi_instance)
1749 mutex_lock(&cpuhp_state_mutex);
1751 if (!invoke || !cpuhp_get_teardown_cb(state))
1754 * Call the teardown callback for each present cpu depending
1755 * on the hotplug state of the cpu. This function is not
1756 * allowed to fail currently!
1758 for_each_present_cpu(cpu) {
1759 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1760 int cpustate = st->state;
1762 if (cpustate >= state)
1763 cpuhp_issue_call(cpu, state, false, node);
1768 mutex_unlock(&cpuhp_state_mutex);
1773 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1776 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1777 * @state: The state to remove
1778 * @invoke: If true, the teardown function is invoked for cpus where
1779 * cpu state >= @state
1781 * The caller needs to hold cpus read locked while calling this function.
1782 * The teardown callback is currently not allowed to fail. Think
1783 * about module removal!
1785 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1787 struct cpuhp_step *sp = cpuhp_get_step(state);
1790 BUG_ON(cpuhp_cb_check(state));
1792 lockdep_assert_cpus_held();
1794 mutex_lock(&cpuhp_state_mutex);
1795 if (sp->multi_instance) {
1796 WARN(!hlist_empty(&sp->list),
1797 "Error: Removing state %d which has instances left.\n",
1802 if (!invoke || !cpuhp_get_teardown_cb(state))
1806 * Call the teardown callback for each present cpu depending
1807 * on the hotplug state of the cpu. This function is not
1808 * allowed to fail currently!
1810 for_each_present_cpu(cpu) {
1811 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1812 int cpustate = st->state;
1814 if (cpustate >= state)
1815 cpuhp_issue_call(cpu, state, false, NULL);
1818 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1819 mutex_unlock(&cpuhp_state_mutex);
1821 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1823 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1826 __cpuhp_remove_state_cpuslocked(state, invoke);
1829 EXPORT_SYMBOL(__cpuhp_remove_state);
1831 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1832 static ssize_t show_cpuhp_state(struct device *dev,
1833 struct device_attribute *attr, char *buf)
1835 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1837 return sprintf(buf, "%d\n", st->state);
1839 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1841 static ssize_t write_cpuhp_target(struct device *dev,
1842 struct device_attribute *attr,
1843 const char *buf, size_t count)
1845 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1846 struct cpuhp_step *sp;
1849 ret = kstrtoint(buf, 10, &target);
1853 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1854 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1857 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1861 ret = lock_device_hotplug_sysfs();
1865 mutex_lock(&cpuhp_state_mutex);
1866 sp = cpuhp_get_step(target);
1867 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1868 mutex_unlock(&cpuhp_state_mutex);
1872 if (st->state < target)
1873 ret = do_cpu_up(dev->id, target);
1875 ret = do_cpu_down(dev->id, target);
1877 unlock_device_hotplug();
1878 return ret ? ret : count;
1881 static ssize_t show_cpuhp_target(struct device *dev,
1882 struct device_attribute *attr, char *buf)
1884 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1886 return sprintf(buf, "%d\n", st->target);
1888 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1891 static ssize_t write_cpuhp_fail(struct device *dev,
1892 struct device_attribute *attr,
1893 const char *buf, size_t count)
1895 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1896 struct cpuhp_step *sp;
1899 ret = kstrtoint(buf, 10, &fail);
1904 * Cannot fail STARTING/DYING callbacks.
1906 if (cpuhp_is_atomic_state(fail))
1910 * Cannot fail anything that doesn't have callbacks.
1912 mutex_lock(&cpuhp_state_mutex);
1913 sp = cpuhp_get_step(fail);
1914 if (!sp->startup.single && !sp->teardown.single)
1916 mutex_unlock(&cpuhp_state_mutex);
1925 static ssize_t show_cpuhp_fail(struct device *dev,
1926 struct device_attribute *attr, char *buf)
1928 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1930 return sprintf(buf, "%d\n", st->fail);
1933 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1935 static struct attribute *cpuhp_cpu_attrs[] = {
1936 &dev_attr_state.attr,
1937 &dev_attr_target.attr,
1938 &dev_attr_fail.attr,
1942 static const struct attribute_group cpuhp_cpu_attr_group = {
1943 .attrs = cpuhp_cpu_attrs,
1948 static ssize_t show_cpuhp_states(struct device *dev,
1949 struct device_attribute *attr, char *buf)
1951 ssize_t cur, res = 0;
1954 mutex_lock(&cpuhp_state_mutex);
1955 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1956 struct cpuhp_step *sp = cpuhp_get_step(i);
1959 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1964 mutex_unlock(&cpuhp_state_mutex);
1967 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1969 static struct attribute *cpuhp_cpu_root_attrs[] = {
1970 &dev_attr_states.attr,
1974 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1975 .attrs = cpuhp_cpu_root_attrs,
1980 #ifdef CONFIG_HOTPLUG_SMT
1982 static const char *smt_states[] = {
1983 [CPU_SMT_ENABLED] = "on",
1984 [CPU_SMT_DISABLED] = "off",
1985 [CPU_SMT_FORCE_DISABLED] = "forceoff",
1986 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
1990 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
1992 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
1995 static void cpuhp_offline_cpu_device(unsigned int cpu)
1997 struct device *dev = get_cpu_device(cpu);
1999 dev->offline = true;
2000 /* Tell user space about the state change */
2001 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2004 static void cpuhp_online_cpu_device(unsigned int cpu)
2006 struct device *dev = get_cpu_device(cpu);
2008 dev->offline = false;
2009 /* Tell user space about the state change */
2010 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2013 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2017 cpu_maps_update_begin();
2018 for_each_online_cpu(cpu) {
2019 if (topology_is_primary_thread(cpu))
2021 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2025 * As this needs to hold the cpu maps lock it's impossible
2026 * to call device_offline() because that ends up calling
2027 * cpu_down() which takes cpu maps lock. cpu maps lock
2028 * needs to be held as this might race against in kernel
2029 * abusers of the hotplug machinery (thermal management).
2031 * So nothing would update device:offline state. That would
2032 * leave the sysfs entry stale and prevent onlining after
2033 * smt control has been changed to 'off' again. This is
2034 * called under the sysfs hotplug lock, so it is properly
2035 * serialized against the regular offline usage.
2037 cpuhp_offline_cpu_device(cpu);
2040 cpu_smt_control = ctrlval;
2041 cpu_maps_update_done();
2045 static int cpuhp_smt_enable(void)
2049 cpu_maps_update_begin();
2050 cpu_smt_control = CPU_SMT_ENABLED;
2051 for_each_present_cpu(cpu) {
2052 /* Skip online CPUs and CPUs on offline nodes */
2053 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2055 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2058 /* See comment in cpuhp_smt_disable() */
2059 cpuhp_online_cpu_device(cpu);
2061 cpu_maps_update_done();
2066 store_smt_control(struct device *dev, struct device_attribute *attr,
2067 const char *buf, size_t count)
2071 if (sysfs_streq(buf, "on"))
2072 ctrlval = CPU_SMT_ENABLED;
2073 else if (sysfs_streq(buf, "off"))
2074 ctrlval = CPU_SMT_DISABLED;
2075 else if (sysfs_streq(buf, "forceoff"))
2076 ctrlval = CPU_SMT_FORCE_DISABLED;
2080 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2083 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2086 ret = lock_device_hotplug_sysfs();
2090 if (ctrlval != cpu_smt_control) {
2092 case CPU_SMT_ENABLED:
2093 ret = cpuhp_smt_enable();
2095 case CPU_SMT_DISABLED:
2096 case CPU_SMT_FORCE_DISABLED:
2097 ret = cpuhp_smt_disable(ctrlval);
2102 unlock_device_hotplug();
2103 return ret ? ret : count;
2105 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2108 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2110 bool active = topology_max_smt_threads() > 1;
2112 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2114 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2116 static struct attribute *cpuhp_smt_attrs[] = {
2117 &dev_attr_control.attr,
2118 &dev_attr_active.attr,
2122 static const struct attribute_group cpuhp_smt_attr_group = {
2123 .attrs = cpuhp_smt_attrs,
2128 static int __init cpu_smt_state_init(void)
2130 if (!topology_smt_supported())
2131 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
2133 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2134 &cpuhp_smt_attr_group);
2138 static inline int cpu_smt_state_init(void) { return 0; }
2141 static int __init cpuhp_sysfs_init(void)
2145 ret = cpu_smt_state_init();
2149 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2150 &cpuhp_cpu_root_attr_group);
2154 for_each_possible_cpu(cpu) {
2155 struct device *dev = get_cpu_device(cpu);
2159 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2165 device_initcall(cpuhp_sysfs_init);
2169 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2170 * represents all NR_CPUS bits binary values of 1<<nr.
2172 * It is used by cpumask_of() to get a constant address to a CPU
2173 * mask value that has a single bit set only.
2176 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2177 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2178 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2179 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2180 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2182 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2184 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2185 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2186 #if BITS_PER_LONG > 32
2187 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2188 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2191 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2193 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2194 EXPORT_SYMBOL(cpu_all_bits);
2196 #ifdef CONFIG_INIT_ALL_POSSIBLE
2197 struct cpumask __cpu_possible_mask __read_mostly
2200 struct cpumask __cpu_possible_mask __read_mostly;
2202 EXPORT_SYMBOL(__cpu_possible_mask);
2204 struct cpumask __cpu_online_mask __read_mostly;
2205 EXPORT_SYMBOL(__cpu_online_mask);
2207 struct cpumask __cpu_present_mask __read_mostly;
2208 EXPORT_SYMBOL(__cpu_present_mask);
2210 struct cpumask __cpu_active_mask __read_mostly;
2211 EXPORT_SYMBOL(__cpu_active_mask);
2213 void init_cpu_present(const struct cpumask *src)
2215 cpumask_copy(&__cpu_present_mask, src);
2218 void init_cpu_possible(const struct cpumask *src)
2220 cpumask_copy(&__cpu_possible_mask, src);
2223 void init_cpu_online(const struct cpumask *src)
2225 cpumask_copy(&__cpu_online_mask, src);
2229 * Activate the first processor.
2231 void __init boot_cpu_init(void)
2233 int cpu = smp_processor_id();
2235 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2236 set_cpu_online(cpu, true);
2237 set_cpu_active(cpu, true);
2238 set_cpu_present(cpu, true);
2239 set_cpu_possible(cpu, true);
2242 __boot_cpu_id = cpu;
2247 * Must be called _AFTER_ setting up the per_cpu areas
2249 void __init boot_cpu_hotplug_init(void)
2251 this_cpu_write(cpuhp_state.booted_once, true);
2252 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);