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/sched/smt.h>
14 #include <linux/unistd.h>
15 #include <linux/cpu.h>
16 #include <linux/oom.h>
17 #include <linux/rcupdate.h>
18 #include <linux/export.h>
19 #include <linux/bug.h>
20 #include <linux/kthread.h>
21 #include <linux/stop_machine.h>
22 #include <linux/mutex.h>
23 #include <linux/gfp.h>
24 #include <linux/suspend.h>
25 #include <linux/lockdep.h>
26 #include <linux/tick.h>
27 #include <linux/irq.h>
28 #include <linux/nmi.h>
29 #include <linux/smpboot.h>
30 #include <linux/relay.h>
31 #include <linux/slab.h>
32 #include <linux/percpu-rwsem.h>
34 #include <trace/events/power.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/cpuhp.h>
41 * cpuhp_cpu_state - Per cpu hotplug state storage
42 * @state: The current cpu state
43 * @target: The target state
44 * @thread: Pointer to the hotplug thread
45 * @should_run: Thread should execute
46 * @rollback: Perform a rollback
47 * @single: Single callback invocation
48 * @bringup: Single callback bringup or teardown selector
49 * @cb_state: The state for a single callback (install/uninstall)
50 * @result: Result of the operation
51 * @done_up: Signal completion to the issuer of the task for cpu-up
52 * @done_down: Signal completion to the issuer of the task for cpu-down
54 struct cpuhp_cpu_state {
55 enum cpuhp_state state;
56 enum cpuhp_state target;
57 enum cpuhp_state fail;
59 struct task_struct *thread;
65 struct hlist_node *node;
66 struct hlist_node *last;
67 enum cpuhp_state cb_state;
69 struct completion done_up;
70 struct completion done_down;
74 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
75 .fail = CPUHP_INVALID,
78 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
79 static struct lockdep_map cpuhp_state_up_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
81 static struct lockdep_map cpuhp_state_down_map =
82 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
85 static inline void cpuhp_lock_acquire(bool bringup)
87 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
90 static inline void cpuhp_lock_release(bool bringup)
92 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
96 static inline void cpuhp_lock_acquire(bool bringup) { }
97 static inline void cpuhp_lock_release(bool bringup) { }
102 * cpuhp_step - Hotplug state machine step
103 * @name: Name of the step
104 * @startup: Startup function of the step
105 * @teardown: Teardown function of the step
106 * @cant_stop: Bringup/teardown can't be stopped at this step
111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
120 struct hlist_head list;
125 static DEFINE_MUTEX(cpuhp_state_mutex);
126 static struct cpuhp_step cpuhp_hp_states[];
128 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
130 return cpuhp_hp_states + state;
134 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
135 * @cpu: The cpu for which the callback should be invoked
136 * @state: The state to do callbacks for
137 * @bringup: True if the bringup callback should be invoked
138 * @node: For multi-instance, do a single entry callback for install/remove
139 * @lastp: For multi-instance rollback, remember how far we got
141 * Called from cpu hotplug and from the state register machinery.
143 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
144 bool bringup, struct hlist_node *node,
145 struct hlist_node **lastp)
147 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
148 struct cpuhp_step *step = cpuhp_get_step(state);
149 int (*cbm)(unsigned int cpu, struct hlist_node *node);
150 int (*cb)(unsigned int cpu);
153 if (st->fail == state) {
154 st->fail = CPUHP_INVALID;
156 if (!(bringup ? step->startup.single : step->teardown.single))
162 if (!step->multi_instance) {
163 WARN_ON_ONCE(lastp && *lastp);
164 cb = bringup ? step->startup.single : step->teardown.single;
167 trace_cpuhp_enter(cpu, st->target, state, cb);
169 trace_cpuhp_exit(cpu, st->state, state, ret);
172 cbm = bringup ? step->startup.multi : step->teardown.multi;
176 /* Single invocation for instance add/remove */
178 WARN_ON_ONCE(lastp && *lastp);
179 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
180 ret = cbm(cpu, node);
181 trace_cpuhp_exit(cpu, st->state, state, ret);
185 /* State transition. Invoke on all instances */
187 hlist_for_each(node, &step->list) {
188 if (lastp && node == *lastp)
191 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
192 ret = cbm(cpu, node);
193 trace_cpuhp_exit(cpu, st->state, state, ret);
207 /* Rollback the instances if one failed */
208 cbm = !bringup ? step->startup.multi : step->teardown.multi;
212 hlist_for_each(node, &step->list) {
216 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
217 ret = cbm(cpu, node);
218 trace_cpuhp_exit(cpu, st->state, state, ret);
220 * Rollback must not fail,
228 static bool cpuhp_is_ap_state(enum cpuhp_state state)
231 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
232 * purposes as that state is handled explicitly in cpu_down.
234 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
237 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
239 struct completion *done = bringup ? &st->done_up : &st->done_down;
240 wait_for_completion(done);
243 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
250 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
252 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
254 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
257 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
258 static DEFINE_MUTEX(cpu_add_remove_lock);
259 bool cpuhp_tasks_frozen;
260 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
263 * The following two APIs (cpu_maps_update_begin/done) must be used when
264 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
266 void cpu_maps_update_begin(void)
268 mutex_lock(&cpu_add_remove_lock);
271 void cpu_maps_update_done(void)
273 mutex_unlock(&cpu_add_remove_lock);
277 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
278 * Should always be manipulated under cpu_add_remove_lock
280 static int cpu_hotplug_disabled;
282 #ifdef CONFIG_HOTPLUG_CPU
284 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
286 void cpus_read_lock(void)
288 percpu_down_read(&cpu_hotplug_lock);
290 EXPORT_SYMBOL_GPL(cpus_read_lock);
292 int cpus_read_trylock(void)
294 return percpu_down_read_trylock(&cpu_hotplug_lock);
296 EXPORT_SYMBOL_GPL(cpus_read_trylock);
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)
317 * We can't have hotplug operations before userspace starts running,
318 * and some init codepaths will knowingly not take the hotplug lock.
319 * This is all valid, so mute lockdep until it makes sense to report
322 if (system_state < SYSTEM_RUNNING)
325 percpu_rwsem_assert_held(&cpu_hotplug_lock);
328 static void lockdep_acquire_cpus_lock(void)
330 rwsem_acquire(&cpu_hotplug_lock.rw_sem.dep_map, 0, 0, _THIS_IP_);
333 static void lockdep_release_cpus_lock(void)
335 rwsem_release(&cpu_hotplug_lock.rw_sem.dep_map, 1, _THIS_IP_);
339 * Wait for currently running CPU hotplug operations to complete (if any) and
340 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
341 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
342 * hotplug path before performing hotplug operations. So acquiring that lock
343 * guarantees mutual exclusion from any currently running hotplug operations.
345 void cpu_hotplug_disable(void)
347 cpu_maps_update_begin();
348 cpu_hotplug_disabled++;
349 cpu_maps_update_done();
351 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
353 static void __cpu_hotplug_enable(void)
355 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
357 cpu_hotplug_disabled--;
360 void cpu_hotplug_enable(void)
362 cpu_maps_update_begin();
363 __cpu_hotplug_enable();
364 cpu_maps_update_done();
366 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
370 static void lockdep_acquire_cpus_lock(void)
374 static void lockdep_release_cpus_lock(void)
378 #endif /* CONFIG_HOTPLUG_CPU */
381 * Architectures that need SMT-specific errata handling during SMT hotplug
382 * should override this.
384 void __weak arch_smt_update(void) { }
386 #ifdef CONFIG_HOTPLUG_SMT
387 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
389 void __init cpu_smt_disable(bool force)
391 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
392 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
396 pr_info("SMT: Force disabled\n");
397 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
399 pr_info("SMT: disabled\n");
400 cpu_smt_control = CPU_SMT_DISABLED;
405 * The decision whether SMT is supported can only be done after the full
406 * CPU identification. Called from architecture code.
408 void __init cpu_smt_check_topology(void)
410 if (!topology_smt_supported())
411 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
414 static int __init smt_cmdline_disable(char *str)
416 cpu_smt_disable(str && !strcmp(str, "force"));
419 early_param("nosmt", smt_cmdline_disable);
421 static inline bool cpu_smt_allowed(unsigned int cpu)
423 if (cpu_smt_control == CPU_SMT_ENABLED)
426 if (topology_is_primary_thread(cpu))
430 * On x86 it's required to boot all logical CPUs at least once so
431 * that the init code can get a chance to set CR4.MCE on each
432 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
433 * core will shutdown the machine.
435 return !per_cpu(cpuhp_state, cpu).booted_once;
438 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
441 static inline enum cpuhp_state
442 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
444 enum cpuhp_state prev_state = st->state;
446 st->rollback = false;
451 st->bringup = st->state < target;
457 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
462 * If we have st->last we need to undo partial multi_instance of this
463 * state first. Otherwise start undo at the previous state.
472 st->target = prev_state;
473 st->bringup = !st->bringup;
476 /* Regular hotplug invocation of the AP hotplug thread */
477 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
479 if (!st->single && st->state == st->target)
484 * Make sure the above stores are visible before should_run becomes
485 * true. Paired with the mb() above in cpuhp_thread_fun()
488 st->should_run = true;
489 wake_up_process(st->thread);
490 wait_for_ap_thread(st, st->bringup);
493 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
495 enum cpuhp_state prev_state;
498 prev_state = cpuhp_set_state(st, target);
500 if ((ret = st->result)) {
501 cpuhp_reset_state(st, prev_state);
508 static int bringup_wait_for_ap(unsigned int cpu)
510 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
512 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
513 wait_for_ap_thread(st, true);
514 if (WARN_ON_ONCE((!cpu_online(cpu))))
517 /* Unpark the stopper thread and the hotplug thread of the target cpu */
518 stop_machine_unpark(cpu);
519 kthread_unpark(st->thread);
522 * SMT soft disabling on X86 requires to bring the CPU out of the
523 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
524 * CPU marked itself as booted_once in cpu_notify_starting() so the
525 * cpu_smt_allowed() check will now return false if this is not the
528 if (!cpu_smt_allowed(cpu))
531 if (st->target <= CPUHP_AP_ONLINE_IDLE)
534 return cpuhp_kick_ap(st, st->target);
537 static int bringup_cpu(unsigned int cpu)
539 struct task_struct *idle = idle_thread_get(cpu);
543 * Some architectures have to walk the irq descriptors to
544 * setup the vector space for the cpu which comes online.
545 * Prevent irq alloc/free across the bringup.
549 /* Arch-specific enabling code. */
550 ret = __cpu_up(cpu, idle);
554 return bringup_wait_for_ap(cpu);
558 * Hotplug state machine related functions
561 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
563 for (st->state--; st->state > st->target; st->state--)
564 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
567 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
568 enum cpuhp_state target)
570 enum cpuhp_state prev_state = st->state;
573 while (st->state < target) {
575 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
577 st->target = prev_state;
578 undo_cpu_up(cpu, st);
586 * The cpu hotplug threads manage the bringup and teardown of the cpus
588 static void cpuhp_create(unsigned int cpu)
590 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
592 init_completion(&st->done_up);
593 init_completion(&st->done_down);
596 static int cpuhp_should_run(unsigned int cpu)
598 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
600 return st->should_run;
604 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
605 * callbacks when a state gets [un]installed at runtime.
607 * Each invocation of this function by the smpboot thread does a single AP
610 * It has 3 modes of operation:
611 * - single: runs st->cb_state
612 * - up: runs ++st->state, while st->state < st->target
613 * - down: runs st->state--, while st->state > st->target
615 * When complete or on error, should_run is cleared and the completion is fired.
617 static void cpuhp_thread_fun(unsigned int cpu)
619 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
620 bool bringup = st->bringup;
621 enum cpuhp_state state;
623 if (WARN_ON_ONCE(!st->should_run))
627 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
628 * that if we see ->should_run we also see the rest of the state.
633 * The BP holds the hotplug lock, but we're now running on the AP,
634 * ensure that anybody asserting the lock is held, will actually find
637 lockdep_acquire_cpus_lock();
638 cpuhp_lock_acquire(bringup);
641 state = st->cb_state;
642 st->should_run = false;
647 st->should_run = (st->state < st->target);
648 WARN_ON_ONCE(st->state > st->target);
652 st->should_run = (st->state > st->target);
653 WARN_ON_ONCE(st->state < st->target);
657 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
659 if (cpuhp_is_atomic_state(state)) {
661 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
665 * STARTING/DYING must not fail!
667 WARN_ON_ONCE(st->result);
669 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
674 * If we fail on a rollback, we're up a creek without no
675 * paddle, no way forward, no way back. We loose, thanks for
678 WARN_ON_ONCE(st->rollback);
679 st->should_run = false;
682 cpuhp_lock_release(bringup);
683 lockdep_release_cpus_lock();
686 complete_ap_thread(st, bringup);
689 /* Invoke a single callback on a remote cpu */
691 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
692 struct hlist_node *node)
694 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
697 if (!cpu_online(cpu))
700 cpuhp_lock_acquire(false);
701 cpuhp_lock_release(false);
703 cpuhp_lock_acquire(true);
704 cpuhp_lock_release(true);
707 * If we are up and running, use the hotplug thread. For early calls
708 * we invoke the thread function directly.
711 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
713 st->rollback = false;
717 st->bringup = bringup;
718 st->cb_state = state;
724 * If we failed and did a partial, do a rollback.
726 if ((ret = st->result) && st->last) {
728 st->bringup = !bringup;
734 * Clean up the leftovers so the next hotplug operation wont use stale
737 st->node = st->last = NULL;
741 static int cpuhp_kick_ap_work(unsigned int cpu)
743 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
744 enum cpuhp_state prev_state = st->state;
747 cpuhp_lock_acquire(false);
748 cpuhp_lock_release(false);
750 cpuhp_lock_acquire(true);
751 cpuhp_lock_release(true);
753 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
754 ret = cpuhp_kick_ap(st, st->target);
755 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
760 static struct smp_hotplug_thread cpuhp_threads = {
761 .store = &cpuhp_state.thread,
762 .create = &cpuhp_create,
763 .thread_should_run = cpuhp_should_run,
764 .thread_fn = cpuhp_thread_fun,
765 .thread_comm = "cpuhp/%u",
769 void __init cpuhp_threads_init(void)
771 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
772 kthread_unpark(this_cpu_read(cpuhp_state.thread));
775 #ifdef CONFIG_HOTPLUG_CPU
777 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
780 * This function walks all processes, finds a valid mm struct for each one and
781 * then clears a corresponding bit in mm's cpumask. While this all sounds
782 * trivial, there are various non-obvious corner cases, which this function
783 * tries to solve in a safe manner.
785 * Also note that the function uses a somewhat relaxed locking scheme, so it may
786 * be called only for an already offlined CPU.
788 void clear_tasks_mm_cpumask(int cpu)
790 struct task_struct *p;
793 * This function is called after the cpu is taken down and marked
794 * offline, so its not like new tasks will ever get this cpu set in
795 * their mm mask. -- Peter Zijlstra
796 * Thus, we may use rcu_read_lock() here, instead of grabbing
797 * full-fledged tasklist_lock.
799 WARN_ON(cpu_online(cpu));
801 for_each_process(p) {
802 struct task_struct *t;
805 * Main thread might exit, but other threads may still have
806 * a valid mm. Find one.
808 t = find_lock_task_mm(p);
811 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
817 /* Take this CPU down. */
818 static int take_cpu_down(void *_param)
820 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
821 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
822 int err, cpu = smp_processor_id();
825 /* Ensure this CPU doesn't handle any more interrupts. */
826 err = __cpu_disable();
831 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
832 * do this step again.
834 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
836 /* Invoke the former CPU_DYING callbacks */
837 for (; st->state > target; st->state--) {
838 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
840 * DYING must not fail!
845 /* Give up timekeeping duties */
846 tick_handover_do_timer();
847 /* Park the stopper thread */
848 stop_machine_park(cpu);
852 static int takedown_cpu(unsigned int cpu)
854 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
857 /* Park the smpboot threads */
858 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
861 * Prevent irq alloc/free while the dying cpu reorganizes the
862 * interrupt affinities.
867 * So now all preempt/rcu users must observe !cpu_active().
869 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
871 /* CPU refused to die */
873 /* Unpark the hotplug thread so we can rollback there */
874 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
877 BUG_ON(cpu_online(cpu));
880 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
881 * all runnable tasks from the CPU, there's only the idle task left now
882 * that the migration thread is done doing the stop_machine thing.
884 * Wait for the stop thread to go away.
886 wait_for_ap_thread(st, false);
887 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
889 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
892 hotplug_cpu__broadcast_tick_pull(cpu);
893 /* This actually kills the CPU. */
896 tick_cleanup_dead_cpu(cpu);
897 rcutree_migrate_callbacks(cpu);
901 static void cpuhp_complete_idle_dead(void *arg)
903 struct cpuhp_cpu_state *st = arg;
905 complete_ap_thread(st, false);
908 void cpuhp_report_idle_dead(void)
910 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
912 BUG_ON(st->state != CPUHP_AP_OFFLINE);
913 rcu_report_dead(smp_processor_id());
914 st->state = CPUHP_AP_IDLE_DEAD;
916 * We cannot call complete after rcu_report_dead() so we delegate it
919 smp_call_function_single(cpumask_first(cpu_online_mask),
920 cpuhp_complete_idle_dead, st, 0);
923 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
925 for (st->state++; st->state < st->target; st->state++)
926 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
929 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
930 enum cpuhp_state target)
932 enum cpuhp_state prev_state = st->state;
935 for (; st->state > target; st->state--) {
936 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
938 st->target = prev_state;
939 if (st->state < prev_state)
940 undo_cpu_down(cpu, st);
947 /* Requires cpu_add_remove_lock to be held */
948 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
949 enum cpuhp_state target)
951 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
952 int prev_state, ret = 0;
954 if (num_online_cpus() == 1)
957 if (!cpu_present(cpu))
962 cpuhp_tasks_frozen = tasks_frozen;
964 prev_state = cpuhp_set_state(st, target);
966 * If the current CPU state is in the range of the AP hotplug thread,
967 * then we need to kick the thread.
969 if (st->state > CPUHP_TEARDOWN_CPU) {
970 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
971 ret = cpuhp_kick_ap_work(cpu);
973 * The AP side has done the error rollback already. Just
974 * return the error code..
980 * We might have stopped still in the range of the AP hotplug
981 * thread. Nothing to do anymore.
983 if (st->state > CPUHP_TEARDOWN_CPU)
989 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
990 * to do the further cleanups.
992 ret = cpuhp_down_callbacks(cpu, st, target);
993 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
994 cpuhp_reset_state(st, prev_state);
1001 * Do post unplug cleanup. This is still protected against
1002 * concurrent CPU hotplug via cpu_add_remove_lock.
1004 lockup_detector_cleanup();
1009 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1011 if (cpu_hotplug_disabled)
1013 return _cpu_down(cpu, 0, target);
1016 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
1020 cpu_maps_update_begin();
1021 err = cpu_down_maps_locked(cpu, target);
1022 cpu_maps_update_done();
1026 int cpu_down(unsigned int cpu)
1028 return do_cpu_down(cpu, CPUHP_OFFLINE);
1030 EXPORT_SYMBOL(cpu_down);
1033 #define takedown_cpu NULL
1034 #endif /*CONFIG_HOTPLUG_CPU*/
1037 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1038 * @cpu: cpu that just started
1040 * It must be called by the arch code on the new cpu, before the new cpu
1041 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1043 void notify_cpu_starting(unsigned int cpu)
1045 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1046 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1049 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1050 st->booted_once = true;
1051 while (st->state < target) {
1053 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1055 * STARTING must not fail!
1062 * Called from the idle task. Wake up the controlling task which brings the
1063 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1064 * the rest of the online bringup to the hotplug thread.
1066 void cpuhp_online_idle(enum cpuhp_state state)
1068 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1070 /* Happens for the boot cpu */
1071 if (state != CPUHP_AP_ONLINE_IDLE)
1074 st->state = CPUHP_AP_ONLINE_IDLE;
1075 complete_ap_thread(st, true);
1078 /* Requires cpu_add_remove_lock to be held */
1079 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1081 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1082 struct task_struct *idle;
1087 if (!cpu_present(cpu)) {
1093 * The caller of do_cpu_up might have raced with another
1094 * caller. Ignore it for now.
1096 if (st->state >= target)
1099 if (st->state == CPUHP_OFFLINE) {
1100 /* Let it fail before we try to bring the cpu up */
1101 idle = idle_thread_get(cpu);
1103 ret = PTR_ERR(idle);
1108 cpuhp_tasks_frozen = tasks_frozen;
1110 cpuhp_set_state(st, target);
1112 * If the current CPU state is in the range of the AP hotplug thread,
1113 * then we need to kick the thread once more.
1115 if (st->state > CPUHP_BRINGUP_CPU) {
1116 ret = cpuhp_kick_ap_work(cpu);
1118 * The AP side has done the error rollback already. Just
1119 * return the error code..
1126 * Try to reach the target state. We max out on the BP at
1127 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1128 * responsible for bringing it up to the target state.
1130 target = min((int)target, CPUHP_BRINGUP_CPU);
1131 ret = cpuhp_up_callbacks(cpu, st, target);
1133 cpus_write_unlock();
1138 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1142 if (!cpu_possible(cpu)) {
1143 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1145 #if defined(CONFIG_IA64)
1146 pr_err("please check additional_cpus= boot parameter\n");
1151 err = try_online_node(cpu_to_node(cpu));
1155 cpu_maps_update_begin();
1157 if (cpu_hotplug_disabled) {
1161 if (!cpu_smt_allowed(cpu)) {
1166 err = _cpu_up(cpu, 0, target);
1168 cpu_maps_update_done();
1172 int cpu_up(unsigned int cpu)
1174 return do_cpu_up(cpu, CPUHP_ONLINE);
1176 EXPORT_SYMBOL_GPL(cpu_up);
1178 #ifdef CONFIG_PM_SLEEP_SMP
1179 static cpumask_var_t frozen_cpus;
1181 int freeze_secondary_cpus(int primary)
1185 cpu_maps_update_begin();
1186 if (!cpu_online(primary))
1187 primary = cpumask_first(cpu_online_mask);
1189 * We take down all of the non-boot CPUs in one shot to avoid races
1190 * with the userspace trying to use the CPU hotplug at the same time
1192 cpumask_clear(frozen_cpus);
1194 pr_info("Disabling non-boot CPUs ...\n");
1195 for_each_online_cpu(cpu) {
1198 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1199 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1200 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1202 cpumask_set_cpu(cpu, frozen_cpus);
1204 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1210 BUG_ON(num_online_cpus() > 1);
1212 pr_err("Non-boot CPUs are not disabled\n");
1215 * Make sure the CPUs won't be enabled by someone else. We need to do
1216 * this even in case of failure as all disable_nonboot_cpus() users are
1217 * supposed to do enable_nonboot_cpus() on the failure path.
1219 cpu_hotplug_disabled++;
1221 cpu_maps_update_done();
1225 void __weak arch_enable_nonboot_cpus_begin(void)
1229 void __weak arch_enable_nonboot_cpus_end(void)
1233 void enable_nonboot_cpus(void)
1237 /* Allow everyone to use the CPU hotplug again */
1238 cpu_maps_update_begin();
1239 __cpu_hotplug_enable();
1240 if (cpumask_empty(frozen_cpus))
1243 pr_info("Enabling non-boot CPUs ...\n");
1245 arch_enable_nonboot_cpus_begin();
1247 for_each_cpu(cpu, frozen_cpus) {
1248 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1249 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1250 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1252 pr_info("CPU%d is up\n", cpu);
1255 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1258 arch_enable_nonboot_cpus_end();
1260 cpumask_clear(frozen_cpus);
1262 cpu_maps_update_done();
1265 static int __init alloc_frozen_cpus(void)
1267 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1271 core_initcall(alloc_frozen_cpus);
1274 * When callbacks for CPU hotplug notifications are being executed, we must
1275 * ensure that the state of the system with respect to the tasks being frozen
1276 * or not, as reported by the notification, remains unchanged *throughout the
1277 * duration* of the execution of the callbacks.
1278 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1280 * This synchronization is implemented by mutually excluding regular CPU
1281 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1282 * Hibernate notifications.
1285 cpu_hotplug_pm_callback(struct notifier_block *nb,
1286 unsigned long action, void *ptr)
1290 case PM_SUSPEND_PREPARE:
1291 case PM_HIBERNATION_PREPARE:
1292 cpu_hotplug_disable();
1295 case PM_POST_SUSPEND:
1296 case PM_POST_HIBERNATION:
1297 cpu_hotplug_enable();
1308 static int __init cpu_hotplug_pm_sync_init(void)
1311 * cpu_hotplug_pm_callback has higher priority than x86
1312 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1313 * to disable cpu hotplug to avoid cpu hotplug race.
1315 pm_notifier(cpu_hotplug_pm_callback, 0);
1318 core_initcall(cpu_hotplug_pm_sync_init);
1320 #endif /* CONFIG_PM_SLEEP_SMP */
1324 #endif /* CONFIG_SMP */
1326 /* Boot processor state steps */
1327 static struct cpuhp_step cpuhp_hp_states[] = {
1330 .startup.single = NULL,
1331 .teardown.single = NULL,
1334 [CPUHP_CREATE_THREADS]= {
1335 .name = "threads:prepare",
1336 .startup.single = smpboot_create_threads,
1337 .teardown.single = NULL,
1340 [CPUHP_PERF_PREPARE] = {
1341 .name = "perf:prepare",
1342 .startup.single = perf_event_init_cpu,
1343 .teardown.single = perf_event_exit_cpu,
1345 [CPUHP_WORKQUEUE_PREP] = {
1346 .name = "workqueue:prepare",
1347 .startup.single = workqueue_prepare_cpu,
1348 .teardown.single = NULL,
1350 [CPUHP_HRTIMERS_PREPARE] = {
1351 .name = "hrtimers:prepare",
1352 .startup.single = hrtimers_prepare_cpu,
1353 .teardown.single = hrtimers_dead_cpu,
1355 [CPUHP_SMPCFD_PREPARE] = {
1356 .name = "smpcfd:prepare",
1357 .startup.single = smpcfd_prepare_cpu,
1358 .teardown.single = smpcfd_dead_cpu,
1360 [CPUHP_RELAY_PREPARE] = {
1361 .name = "relay:prepare",
1362 .startup.single = relay_prepare_cpu,
1363 .teardown.single = NULL,
1365 [CPUHP_SLAB_PREPARE] = {
1366 .name = "slab:prepare",
1367 .startup.single = slab_prepare_cpu,
1368 .teardown.single = slab_dead_cpu,
1370 [CPUHP_RCUTREE_PREP] = {
1371 .name = "RCU/tree:prepare",
1372 .startup.single = rcutree_prepare_cpu,
1373 .teardown.single = rcutree_dead_cpu,
1376 * On the tear-down path, timers_dead_cpu() must be invoked
1377 * before blk_mq_queue_reinit_notify() from notify_dead(),
1378 * otherwise a RCU stall occurs.
1380 [CPUHP_TIMERS_PREPARE] = {
1381 .name = "timers:prepare",
1382 .startup.single = timers_prepare_cpu,
1383 .teardown.single = timers_dead_cpu,
1385 /* Kicks the plugged cpu into life */
1386 [CPUHP_BRINGUP_CPU] = {
1387 .name = "cpu:bringup",
1388 .startup.single = bringup_cpu,
1389 .teardown.single = NULL,
1392 /* Final state before CPU kills itself */
1393 [CPUHP_AP_IDLE_DEAD] = {
1394 .name = "idle:dead",
1397 * Last state before CPU enters the idle loop to die. Transient state
1398 * for synchronization.
1400 [CPUHP_AP_OFFLINE] = {
1401 .name = "ap:offline",
1404 /* First state is scheduler control. Interrupts are disabled */
1405 [CPUHP_AP_SCHED_STARTING] = {
1406 .name = "sched:starting",
1407 .startup.single = sched_cpu_starting,
1408 .teardown.single = sched_cpu_dying,
1410 [CPUHP_AP_RCUTREE_DYING] = {
1411 .name = "RCU/tree:dying",
1412 .startup.single = NULL,
1413 .teardown.single = rcutree_dying_cpu,
1415 [CPUHP_AP_SMPCFD_DYING] = {
1416 .name = "smpcfd:dying",
1417 .startup.single = NULL,
1418 .teardown.single = smpcfd_dying_cpu,
1420 /* Entry state on starting. Interrupts enabled from here on. Transient
1421 * state for synchronsization */
1422 [CPUHP_AP_ONLINE] = {
1423 .name = "ap:online",
1426 * Handled on controll processor until the plugged processor manages
1429 [CPUHP_TEARDOWN_CPU] = {
1430 .name = "cpu:teardown",
1431 .startup.single = NULL,
1432 .teardown.single = takedown_cpu,
1435 /* Handle smpboot threads park/unpark */
1436 [CPUHP_AP_SMPBOOT_THREADS] = {
1437 .name = "smpboot/threads:online",
1438 .startup.single = smpboot_unpark_threads,
1439 .teardown.single = smpboot_park_threads,
1441 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1442 .name = "irq/affinity:online",
1443 .startup.single = irq_affinity_online_cpu,
1444 .teardown.single = NULL,
1446 [CPUHP_AP_PERF_ONLINE] = {
1447 .name = "perf:online",
1448 .startup.single = perf_event_init_cpu,
1449 .teardown.single = perf_event_exit_cpu,
1451 [CPUHP_AP_WATCHDOG_ONLINE] = {
1452 .name = "lockup_detector:online",
1453 .startup.single = lockup_detector_online_cpu,
1454 .teardown.single = lockup_detector_offline_cpu,
1456 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1457 .name = "workqueue:online",
1458 .startup.single = workqueue_online_cpu,
1459 .teardown.single = workqueue_offline_cpu,
1461 [CPUHP_AP_RCUTREE_ONLINE] = {
1462 .name = "RCU/tree:online",
1463 .startup.single = rcutree_online_cpu,
1464 .teardown.single = rcutree_offline_cpu,
1468 * The dynamically registered state space is here
1472 /* Last state is scheduler control setting the cpu active */
1473 [CPUHP_AP_ACTIVE] = {
1474 .name = "sched:active",
1475 .startup.single = sched_cpu_activate,
1476 .teardown.single = sched_cpu_deactivate,
1480 /* CPU is fully up and running. */
1483 .startup.single = NULL,
1484 .teardown.single = NULL,
1488 /* Sanity check for callbacks */
1489 static int cpuhp_cb_check(enum cpuhp_state state)
1491 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1497 * Returns a free for dynamic slot assignment of the Online state. The states
1498 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1499 * by having no name assigned.
1501 static int cpuhp_reserve_state(enum cpuhp_state state)
1503 enum cpuhp_state i, end;
1504 struct cpuhp_step *step;
1507 case CPUHP_AP_ONLINE_DYN:
1508 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1509 end = CPUHP_AP_ONLINE_DYN_END;
1511 case CPUHP_BP_PREPARE_DYN:
1512 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1513 end = CPUHP_BP_PREPARE_DYN_END;
1519 for (i = state; i <= end; i++, step++) {
1523 WARN(1, "No more dynamic states available for CPU hotplug\n");
1527 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1528 int (*startup)(unsigned int cpu),
1529 int (*teardown)(unsigned int cpu),
1530 bool multi_instance)
1532 /* (Un)Install the callbacks for further cpu hotplug operations */
1533 struct cpuhp_step *sp;
1537 * If name is NULL, then the state gets removed.
1539 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1540 * the first allocation from these dynamic ranges, so the removal
1541 * would trigger a new allocation and clear the wrong (already
1542 * empty) state, leaving the callbacks of the to be cleared state
1543 * dangling, which causes wreckage on the next hotplug operation.
1545 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1546 state == CPUHP_BP_PREPARE_DYN)) {
1547 ret = cpuhp_reserve_state(state);
1552 sp = cpuhp_get_step(state);
1553 if (name && sp->name)
1556 sp->startup.single = startup;
1557 sp->teardown.single = teardown;
1559 sp->multi_instance = multi_instance;
1560 INIT_HLIST_HEAD(&sp->list);
1564 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1566 return cpuhp_get_step(state)->teardown.single;
1570 * Call the startup/teardown function for a step either on the AP or
1571 * on the current CPU.
1573 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1574 struct hlist_node *node)
1576 struct cpuhp_step *sp = cpuhp_get_step(state);
1580 * If there's nothing to do, we done.
1581 * Relies on the union for multi_instance.
1583 if ((bringup && !sp->startup.single) ||
1584 (!bringup && !sp->teardown.single))
1587 * The non AP bound callbacks can fail on bringup. On teardown
1588 * e.g. module removal we crash for now.
1591 if (cpuhp_is_ap_state(state))
1592 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1594 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1596 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1598 BUG_ON(ret && !bringup);
1603 * Called from __cpuhp_setup_state on a recoverable failure.
1605 * Note: The teardown callbacks for rollback are not allowed to fail!
1607 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1608 struct hlist_node *node)
1612 /* Roll back the already executed steps on the other cpus */
1613 for_each_present_cpu(cpu) {
1614 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1615 int cpustate = st->state;
1617 if (cpu >= failedcpu)
1620 /* Did we invoke the startup call on that cpu ? */
1621 if (cpustate >= state)
1622 cpuhp_issue_call(cpu, state, false, node);
1626 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1627 struct hlist_node *node,
1630 struct cpuhp_step *sp;
1634 lockdep_assert_cpus_held();
1636 sp = cpuhp_get_step(state);
1637 if (sp->multi_instance == false)
1640 mutex_lock(&cpuhp_state_mutex);
1642 if (!invoke || !sp->startup.multi)
1646 * Try to call the startup callback for each present cpu
1647 * depending on the hotplug state of the cpu.
1649 for_each_present_cpu(cpu) {
1650 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1651 int cpustate = st->state;
1653 if (cpustate < state)
1656 ret = cpuhp_issue_call(cpu, state, true, node);
1658 if (sp->teardown.multi)
1659 cpuhp_rollback_install(cpu, state, node);
1665 hlist_add_head(node, &sp->list);
1667 mutex_unlock(&cpuhp_state_mutex);
1671 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1677 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1681 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1684 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1685 * @state: The state to setup
1686 * @invoke: If true, the startup function is invoked for cpus where
1687 * cpu state >= @state
1688 * @startup: startup callback function
1689 * @teardown: teardown callback function
1690 * @multi_instance: State is set up for multiple instances which get
1693 * The caller needs to hold cpus read locked while calling this function.
1696 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1697 * 0 for all other states
1698 * On failure: proper (negative) error code
1700 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1701 const char *name, bool invoke,
1702 int (*startup)(unsigned int cpu),
1703 int (*teardown)(unsigned int cpu),
1704 bool multi_instance)
1709 lockdep_assert_cpus_held();
1711 if (cpuhp_cb_check(state) || !name)
1714 mutex_lock(&cpuhp_state_mutex);
1716 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1719 dynstate = state == CPUHP_AP_ONLINE_DYN;
1720 if (ret > 0 && dynstate) {
1725 if (ret || !invoke || !startup)
1729 * Try to call the startup callback for each present cpu
1730 * depending on the hotplug state of the cpu.
1732 for_each_present_cpu(cpu) {
1733 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1734 int cpustate = st->state;
1736 if (cpustate < state)
1739 ret = cpuhp_issue_call(cpu, state, true, NULL);
1742 cpuhp_rollback_install(cpu, state, NULL);
1743 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1748 mutex_unlock(&cpuhp_state_mutex);
1750 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1751 * dynamically allocated state in case of success.
1753 if (!ret && dynstate)
1757 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1759 int __cpuhp_setup_state(enum cpuhp_state state,
1760 const char *name, bool invoke,
1761 int (*startup)(unsigned int cpu),
1762 int (*teardown)(unsigned int cpu),
1763 bool multi_instance)
1768 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1769 teardown, multi_instance);
1773 EXPORT_SYMBOL(__cpuhp_setup_state);
1775 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1776 struct hlist_node *node, bool invoke)
1778 struct cpuhp_step *sp = cpuhp_get_step(state);
1781 BUG_ON(cpuhp_cb_check(state));
1783 if (!sp->multi_instance)
1787 mutex_lock(&cpuhp_state_mutex);
1789 if (!invoke || !cpuhp_get_teardown_cb(state))
1792 * Call the teardown callback for each present cpu depending
1793 * on the hotplug state of the cpu. This function is not
1794 * allowed to fail currently!
1796 for_each_present_cpu(cpu) {
1797 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1798 int cpustate = st->state;
1800 if (cpustate >= state)
1801 cpuhp_issue_call(cpu, state, false, node);
1806 mutex_unlock(&cpuhp_state_mutex);
1811 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1814 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1815 * @state: The state to remove
1816 * @invoke: If true, the teardown function is invoked for cpus where
1817 * cpu state >= @state
1819 * The caller needs to hold cpus read locked while calling this function.
1820 * The teardown callback is currently not allowed to fail. Think
1821 * about module removal!
1823 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1825 struct cpuhp_step *sp = cpuhp_get_step(state);
1828 BUG_ON(cpuhp_cb_check(state));
1830 lockdep_assert_cpus_held();
1832 mutex_lock(&cpuhp_state_mutex);
1833 if (sp->multi_instance) {
1834 WARN(!hlist_empty(&sp->list),
1835 "Error: Removing state %d which has instances left.\n",
1840 if (!invoke || !cpuhp_get_teardown_cb(state))
1844 * Call the teardown callback for each present cpu depending
1845 * on the hotplug state of the cpu. This function is not
1846 * allowed to fail currently!
1848 for_each_present_cpu(cpu) {
1849 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1850 int cpustate = st->state;
1852 if (cpustate >= state)
1853 cpuhp_issue_call(cpu, state, false, NULL);
1856 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1857 mutex_unlock(&cpuhp_state_mutex);
1859 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1861 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1864 __cpuhp_remove_state_cpuslocked(state, invoke);
1867 EXPORT_SYMBOL(__cpuhp_remove_state);
1869 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1870 static ssize_t show_cpuhp_state(struct device *dev,
1871 struct device_attribute *attr, char *buf)
1873 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1875 return sprintf(buf, "%d\n", st->state);
1877 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1879 static ssize_t write_cpuhp_target(struct device *dev,
1880 struct device_attribute *attr,
1881 const char *buf, size_t count)
1883 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1884 struct cpuhp_step *sp;
1887 ret = kstrtoint(buf, 10, &target);
1891 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1892 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1895 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1899 ret = lock_device_hotplug_sysfs();
1903 mutex_lock(&cpuhp_state_mutex);
1904 sp = cpuhp_get_step(target);
1905 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1906 mutex_unlock(&cpuhp_state_mutex);
1910 if (st->state < target)
1911 ret = do_cpu_up(dev->id, target);
1913 ret = do_cpu_down(dev->id, target);
1915 unlock_device_hotplug();
1916 return ret ? ret : count;
1919 static ssize_t show_cpuhp_target(struct device *dev,
1920 struct device_attribute *attr, char *buf)
1922 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1924 return sprintf(buf, "%d\n", st->target);
1926 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1929 static ssize_t write_cpuhp_fail(struct device *dev,
1930 struct device_attribute *attr,
1931 const char *buf, size_t count)
1933 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1934 struct cpuhp_step *sp;
1937 ret = kstrtoint(buf, 10, &fail);
1942 * Cannot fail STARTING/DYING callbacks.
1944 if (cpuhp_is_atomic_state(fail))
1948 * Cannot fail anything that doesn't have callbacks.
1950 mutex_lock(&cpuhp_state_mutex);
1951 sp = cpuhp_get_step(fail);
1952 if (!sp->startup.single && !sp->teardown.single)
1954 mutex_unlock(&cpuhp_state_mutex);
1963 static ssize_t show_cpuhp_fail(struct device *dev,
1964 struct device_attribute *attr, char *buf)
1966 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1968 return sprintf(buf, "%d\n", st->fail);
1971 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1973 static struct attribute *cpuhp_cpu_attrs[] = {
1974 &dev_attr_state.attr,
1975 &dev_attr_target.attr,
1976 &dev_attr_fail.attr,
1980 static const struct attribute_group cpuhp_cpu_attr_group = {
1981 .attrs = cpuhp_cpu_attrs,
1986 static ssize_t show_cpuhp_states(struct device *dev,
1987 struct device_attribute *attr, char *buf)
1989 ssize_t cur, res = 0;
1992 mutex_lock(&cpuhp_state_mutex);
1993 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1994 struct cpuhp_step *sp = cpuhp_get_step(i);
1997 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2002 mutex_unlock(&cpuhp_state_mutex);
2005 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2007 static struct attribute *cpuhp_cpu_root_attrs[] = {
2008 &dev_attr_states.attr,
2012 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2013 .attrs = cpuhp_cpu_root_attrs,
2018 #ifdef CONFIG_HOTPLUG_SMT
2020 static const char *smt_states[] = {
2021 [CPU_SMT_ENABLED] = "on",
2022 [CPU_SMT_DISABLED] = "off",
2023 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2024 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2028 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2030 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2033 static void cpuhp_offline_cpu_device(unsigned int cpu)
2035 struct device *dev = get_cpu_device(cpu);
2037 dev->offline = true;
2038 /* Tell user space about the state change */
2039 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2042 static void cpuhp_online_cpu_device(unsigned int cpu)
2044 struct device *dev = get_cpu_device(cpu);
2046 dev->offline = false;
2047 /* Tell user space about the state change */
2048 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2051 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2055 cpu_maps_update_begin();
2056 for_each_online_cpu(cpu) {
2057 if (topology_is_primary_thread(cpu))
2059 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2063 * As this needs to hold the cpu maps lock it's impossible
2064 * to call device_offline() because that ends up calling
2065 * cpu_down() which takes cpu maps lock. cpu maps lock
2066 * needs to be held as this might race against in kernel
2067 * abusers of the hotplug machinery (thermal management).
2069 * So nothing would update device:offline state. That would
2070 * leave the sysfs entry stale and prevent onlining after
2071 * smt control has been changed to 'off' again. This is
2072 * called under the sysfs hotplug lock, so it is properly
2073 * serialized against the regular offline usage.
2075 cpuhp_offline_cpu_device(cpu);
2078 cpu_smt_control = ctrlval;
2079 cpu_maps_update_done();
2083 static int cpuhp_smt_enable(void)
2087 cpu_maps_update_begin();
2088 cpu_smt_control = CPU_SMT_ENABLED;
2089 for_each_present_cpu(cpu) {
2090 /* Skip online CPUs and CPUs on offline nodes */
2091 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2093 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2096 /* See comment in cpuhp_smt_disable() */
2097 cpuhp_online_cpu_device(cpu);
2099 cpu_maps_update_done();
2104 store_smt_control(struct device *dev, struct device_attribute *attr,
2105 const char *buf, size_t count)
2109 if (sysfs_streq(buf, "on"))
2110 ctrlval = CPU_SMT_ENABLED;
2111 else if (sysfs_streq(buf, "off"))
2112 ctrlval = CPU_SMT_DISABLED;
2113 else if (sysfs_streq(buf, "forceoff"))
2114 ctrlval = CPU_SMT_FORCE_DISABLED;
2118 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2121 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2124 ret = lock_device_hotplug_sysfs();
2128 if (ctrlval != cpu_smt_control) {
2130 case CPU_SMT_ENABLED:
2131 ret = cpuhp_smt_enable();
2133 case CPU_SMT_DISABLED:
2134 case CPU_SMT_FORCE_DISABLED:
2135 ret = cpuhp_smt_disable(ctrlval);
2140 unlock_device_hotplug();
2141 return ret ? ret : count;
2143 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2146 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2148 bool active = topology_max_smt_threads() > 1;
2150 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2152 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2154 static struct attribute *cpuhp_smt_attrs[] = {
2155 &dev_attr_control.attr,
2156 &dev_attr_active.attr,
2160 static const struct attribute_group cpuhp_smt_attr_group = {
2161 .attrs = cpuhp_smt_attrs,
2166 static int __init cpu_smt_state_init(void)
2168 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2169 &cpuhp_smt_attr_group);
2173 static inline int cpu_smt_state_init(void) { return 0; }
2176 static int __init cpuhp_sysfs_init(void)
2180 ret = cpu_smt_state_init();
2184 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2185 &cpuhp_cpu_root_attr_group);
2189 for_each_possible_cpu(cpu) {
2190 struct device *dev = get_cpu_device(cpu);
2194 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2200 device_initcall(cpuhp_sysfs_init);
2204 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2205 * represents all NR_CPUS bits binary values of 1<<nr.
2207 * It is used by cpumask_of() to get a constant address to a CPU
2208 * mask value that has a single bit set only.
2211 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2212 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2213 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2214 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2215 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2217 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2219 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2220 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2221 #if BITS_PER_LONG > 32
2222 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2223 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2226 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2228 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2229 EXPORT_SYMBOL(cpu_all_bits);
2231 #ifdef CONFIG_INIT_ALL_POSSIBLE
2232 struct cpumask __cpu_possible_mask __read_mostly
2235 struct cpumask __cpu_possible_mask __read_mostly;
2237 EXPORT_SYMBOL(__cpu_possible_mask);
2239 struct cpumask __cpu_online_mask __read_mostly;
2240 EXPORT_SYMBOL(__cpu_online_mask);
2242 struct cpumask __cpu_present_mask __read_mostly;
2243 EXPORT_SYMBOL(__cpu_present_mask);
2245 struct cpumask __cpu_active_mask __read_mostly;
2246 EXPORT_SYMBOL(__cpu_active_mask);
2248 void init_cpu_present(const struct cpumask *src)
2250 cpumask_copy(&__cpu_present_mask, src);
2253 void init_cpu_possible(const struct cpumask *src)
2255 cpumask_copy(&__cpu_possible_mask, src);
2258 void init_cpu_online(const struct cpumask *src)
2260 cpumask_copy(&__cpu_online_mask, src);
2264 * Activate the first processor.
2266 void __init boot_cpu_init(void)
2268 int cpu = smp_processor_id();
2270 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2271 set_cpu_online(cpu, true);
2272 set_cpu_active(cpu, true);
2273 set_cpu_present(cpu, true);
2274 set_cpu_possible(cpu, true);
2277 __boot_cpu_id = cpu;
2282 * Must be called _AFTER_ setting up the per_cpu areas
2284 void __init boot_cpu_hotplug_init(void)
2287 this_cpu_write(cpuhp_state.booted_once, true);
2289 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);