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;
353 static int __init smt_cmdline_disable(char *str)
355 cpu_smt_control = CPU_SMT_DISABLED;
356 if (str && !strcmp(str, "force")) {
357 pr_info("SMT: Force disabled\n");
358 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
362 early_param("nosmt", smt_cmdline_disable);
364 static inline bool cpu_smt_allowed(unsigned int cpu)
366 if (cpu_smt_control == CPU_SMT_ENABLED)
369 if (topology_is_primary_thread(cpu))
373 * On x86 it's required to boot all logical CPUs at least once so
374 * that the init code can get a chance to set CR4.MCE on each
375 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
376 * core will shutdown the machine.
378 return !per_cpu(cpuhp_state, cpu).booted_once;
381 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
384 static inline enum cpuhp_state
385 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
387 enum cpuhp_state prev_state = st->state;
389 st->rollback = false;
394 st->bringup = st->state < target;
400 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
405 * If we have st->last we need to undo partial multi_instance of this
406 * state first. Otherwise start undo at the previous state.
415 st->target = prev_state;
416 st->bringup = !st->bringup;
419 /* Regular hotplug invocation of the AP hotplug thread */
420 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
422 if (!st->single && st->state == st->target)
427 * Make sure the above stores are visible before should_run becomes
428 * true. Paired with the mb() above in cpuhp_thread_fun()
431 st->should_run = true;
432 wake_up_process(st->thread);
433 wait_for_ap_thread(st, st->bringup);
436 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
438 enum cpuhp_state prev_state;
441 prev_state = cpuhp_set_state(st, target);
443 if ((ret = st->result)) {
444 cpuhp_reset_state(st, prev_state);
451 static int bringup_wait_for_ap(unsigned int cpu)
453 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
455 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
456 wait_for_ap_thread(st, true);
457 if (WARN_ON_ONCE((!cpu_online(cpu))))
460 /* Unpark the stopper thread and the hotplug thread of the target cpu */
461 stop_machine_unpark(cpu);
462 kthread_unpark(st->thread);
465 * SMT soft disabling on X86 requires to bring the CPU out of the
466 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
467 * CPU marked itself as booted_once in cpu_notify_starting() so the
468 * cpu_smt_allowed() check will now return false if this is not the
471 if (!cpu_smt_allowed(cpu))
474 if (st->target <= CPUHP_AP_ONLINE_IDLE)
477 return cpuhp_kick_ap(st, st->target);
480 static int bringup_cpu(unsigned int cpu)
482 struct task_struct *idle = idle_thread_get(cpu);
486 * Some architectures have to walk the irq descriptors to
487 * setup the vector space for the cpu which comes online.
488 * Prevent irq alloc/free across the bringup.
492 /* Arch-specific enabling code. */
493 ret = __cpu_up(cpu, idle);
497 return bringup_wait_for_ap(cpu);
501 * Hotplug state machine related functions
504 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
506 for (st->state--; st->state > st->target; st->state--) {
507 struct cpuhp_step *step = cpuhp_get_step(st->state);
509 if (!step->skip_onerr)
510 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
514 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
515 enum cpuhp_state target)
517 enum cpuhp_state prev_state = st->state;
520 while (st->state < target) {
522 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
524 st->target = prev_state;
525 undo_cpu_up(cpu, st);
533 * The cpu hotplug threads manage the bringup and teardown of the cpus
535 static void cpuhp_create(unsigned int cpu)
537 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
539 init_completion(&st->done_up);
540 init_completion(&st->done_down);
543 static int cpuhp_should_run(unsigned int cpu)
545 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
547 return st->should_run;
551 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
552 * callbacks when a state gets [un]installed at runtime.
554 * Each invocation of this function by the smpboot thread does a single AP
557 * It has 3 modes of operation:
558 * - single: runs st->cb_state
559 * - up: runs ++st->state, while st->state < st->target
560 * - down: runs st->state--, while st->state > st->target
562 * When complete or on error, should_run is cleared and the completion is fired.
564 static void cpuhp_thread_fun(unsigned int cpu)
566 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
567 bool bringup = st->bringup;
568 enum cpuhp_state state;
571 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
572 * that if we see ->should_run we also see the rest of the state.
576 if (WARN_ON_ONCE(!st->should_run))
579 cpuhp_lock_acquire(bringup);
582 state = st->cb_state;
583 st->should_run = false;
588 st->should_run = (st->state < st->target);
589 WARN_ON_ONCE(st->state > st->target);
593 st->should_run = (st->state > st->target);
594 WARN_ON_ONCE(st->state < st->target);
598 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
601 struct cpuhp_step *step = cpuhp_get_step(state);
602 if (step->skip_onerr)
606 if (cpuhp_is_atomic_state(state)) {
608 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
612 * STARTING/DYING must not fail!
614 WARN_ON_ONCE(st->result);
616 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
621 * If we fail on a rollback, we're up a creek without no
622 * paddle, no way forward, no way back. We loose, thanks for
625 WARN_ON_ONCE(st->rollback);
626 st->should_run = false;
630 cpuhp_lock_release(bringup);
633 complete_ap_thread(st, bringup);
636 /* Invoke a single callback on a remote cpu */
638 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
639 struct hlist_node *node)
641 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
644 if (!cpu_online(cpu))
647 cpuhp_lock_acquire(false);
648 cpuhp_lock_release(false);
650 cpuhp_lock_acquire(true);
651 cpuhp_lock_release(true);
654 * If we are up and running, use the hotplug thread. For early calls
655 * we invoke the thread function directly.
658 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
660 st->rollback = false;
664 st->bringup = bringup;
665 st->cb_state = state;
671 * If we failed and did a partial, do a rollback.
673 if ((ret = st->result) && st->last) {
675 st->bringup = !bringup;
681 * Clean up the leftovers so the next hotplug operation wont use stale
684 st->node = st->last = NULL;
688 static int cpuhp_kick_ap_work(unsigned int cpu)
690 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
691 enum cpuhp_state prev_state = st->state;
694 cpuhp_lock_acquire(false);
695 cpuhp_lock_release(false);
697 cpuhp_lock_acquire(true);
698 cpuhp_lock_release(true);
700 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
701 ret = cpuhp_kick_ap(st, st->target);
702 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
707 static struct smp_hotplug_thread cpuhp_threads = {
708 .store = &cpuhp_state.thread,
709 .create = &cpuhp_create,
710 .thread_should_run = cpuhp_should_run,
711 .thread_fn = cpuhp_thread_fun,
712 .thread_comm = "cpuhp/%u",
716 void __init cpuhp_threads_init(void)
718 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
719 kthread_unpark(this_cpu_read(cpuhp_state.thread));
722 #ifdef CONFIG_HOTPLUG_CPU
724 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
727 * This function walks all processes, finds a valid mm struct for each one and
728 * then clears a corresponding bit in mm's cpumask. While this all sounds
729 * trivial, there are various non-obvious corner cases, which this function
730 * tries to solve in a safe manner.
732 * Also note that the function uses a somewhat relaxed locking scheme, so it may
733 * be called only for an already offlined CPU.
735 void clear_tasks_mm_cpumask(int cpu)
737 struct task_struct *p;
740 * This function is called after the cpu is taken down and marked
741 * offline, so its not like new tasks will ever get this cpu set in
742 * their mm mask. -- Peter Zijlstra
743 * Thus, we may use rcu_read_lock() here, instead of grabbing
744 * full-fledged tasklist_lock.
746 WARN_ON(cpu_online(cpu));
748 for_each_process(p) {
749 struct task_struct *t;
752 * Main thread might exit, but other threads may still have
753 * a valid mm. Find one.
755 t = find_lock_task_mm(p);
758 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
764 /* Take this CPU down. */
765 static int take_cpu_down(void *_param)
767 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
768 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
769 int err, cpu = smp_processor_id();
772 /* Ensure this CPU doesn't handle any more interrupts. */
773 err = __cpu_disable();
778 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
779 * do this step again.
781 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
783 /* Invoke the former CPU_DYING callbacks */
784 for (; st->state > target; st->state--) {
785 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
787 * DYING must not fail!
792 /* Give up timekeeping duties */
793 tick_handover_do_timer();
794 /* Park the stopper thread */
795 stop_machine_park(cpu);
799 static int takedown_cpu(unsigned int cpu)
801 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
804 /* Park the smpboot threads */
805 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
808 * Prevent irq alloc/free while the dying cpu reorganizes the
809 * interrupt affinities.
814 * So now all preempt/rcu users must observe !cpu_active().
816 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
818 /* CPU refused to die */
820 /* Unpark the hotplug thread so we can rollback there */
821 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
824 BUG_ON(cpu_online(cpu));
827 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
828 * runnable tasks from the cpu, there's only the idle task left now
829 * that the migration thread is done doing the stop_machine thing.
831 * Wait for the stop thread to go away.
833 wait_for_ap_thread(st, false);
834 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
836 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
839 hotplug_cpu__broadcast_tick_pull(cpu);
840 /* This actually kills the CPU. */
843 tick_cleanup_dead_cpu(cpu);
844 rcutree_migrate_callbacks(cpu);
848 static void cpuhp_complete_idle_dead(void *arg)
850 struct cpuhp_cpu_state *st = arg;
852 complete_ap_thread(st, false);
855 void cpuhp_report_idle_dead(void)
857 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
859 BUG_ON(st->state != CPUHP_AP_OFFLINE);
860 rcu_report_dead(smp_processor_id());
861 st->state = CPUHP_AP_IDLE_DEAD;
863 * We cannot call complete after rcu_report_dead() so we delegate it
866 smp_call_function_single(cpumask_first(cpu_online_mask),
867 cpuhp_complete_idle_dead, st, 0);
870 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
872 for (st->state++; st->state < st->target; st->state++) {
873 struct cpuhp_step *step = cpuhp_get_step(st->state);
875 if (!step->skip_onerr)
876 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
880 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
881 enum cpuhp_state target)
883 enum cpuhp_state prev_state = st->state;
886 for (; st->state > target; st->state--) {
887 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
889 st->target = prev_state;
890 undo_cpu_down(cpu, st);
897 /* Requires cpu_add_remove_lock to be held */
898 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
899 enum cpuhp_state target)
901 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
902 int prev_state, ret = 0;
904 if (num_online_cpus() == 1)
907 if (!cpu_present(cpu))
912 cpuhp_tasks_frozen = tasks_frozen;
914 prev_state = cpuhp_set_state(st, target);
916 * If the current CPU state is in the range of the AP hotplug thread,
917 * then we need to kick the thread.
919 if (st->state > CPUHP_TEARDOWN_CPU) {
920 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
921 ret = cpuhp_kick_ap_work(cpu);
923 * The AP side has done the error rollback already. Just
924 * return the error code..
930 * We might have stopped still in the range of the AP hotplug
931 * thread. Nothing to do anymore.
933 if (st->state > CPUHP_TEARDOWN_CPU)
939 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
940 * to do the further cleanups.
942 ret = cpuhp_down_callbacks(cpu, st, target);
943 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
944 cpuhp_reset_state(st, prev_state);
951 * Do post unplug cleanup. This is still protected against
952 * concurrent CPU hotplug via cpu_add_remove_lock.
954 lockup_detector_cleanup();
958 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
960 if (cpu_hotplug_disabled)
962 return _cpu_down(cpu, 0, target);
965 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
969 cpu_maps_update_begin();
970 err = cpu_down_maps_locked(cpu, target);
971 cpu_maps_update_done();
975 int cpu_down(unsigned int cpu)
977 return do_cpu_down(cpu, CPUHP_OFFLINE);
979 EXPORT_SYMBOL(cpu_down);
982 #define takedown_cpu NULL
983 #endif /*CONFIG_HOTPLUG_CPU*/
986 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
987 * @cpu: cpu that just started
989 * It must be called by the arch code on the new cpu, before the new cpu
990 * enables interrupts and before the "boot" cpu returns from __cpu_up().
992 void notify_cpu_starting(unsigned int cpu)
994 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
995 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
998 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
999 st->booted_once = true;
1000 while (st->state < target) {
1002 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1004 * STARTING must not fail!
1011 * Called from the idle task. Wake up the controlling task which brings the
1012 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1013 * the rest of the online bringup to the hotplug thread.
1015 void cpuhp_online_idle(enum cpuhp_state state)
1017 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1019 /* Happens for the boot cpu */
1020 if (state != CPUHP_AP_ONLINE_IDLE)
1023 st->state = CPUHP_AP_ONLINE_IDLE;
1024 complete_ap_thread(st, true);
1027 /* Requires cpu_add_remove_lock to be held */
1028 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1030 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1031 struct task_struct *idle;
1036 if (!cpu_present(cpu)) {
1042 * The caller of do_cpu_up might have raced with another
1043 * caller. Ignore it for now.
1045 if (st->state >= target)
1048 if (st->state == CPUHP_OFFLINE) {
1049 /* Let it fail before we try to bring the cpu up */
1050 idle = idle_thread_get(cpu);
1052 ret = PTR_ERR(idle);
1057 cpuhp_tasks_frozen = tasks_frozen;
1059 cpuhp_set_state(st, target);
1061 * If the current CPU state is in the range of the AP hotplug thread,
1062 * then we need to kick the thread once more.
1064 if (st->state > CPUHP_BRINGUP_CPU) {
1065 ret = cpuhp_kick_ap_work(cpu);
1067 * The AP side has done the error rollback already. Just
1068 * return the error code..
1075 * Try to reach the target state. We max out on the BP at
1076 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1077 * responsible for bringing it up to the target state.
1079 target = min((int)target, CPUHP_BRINGUP_CPU);
1080 ret = cpuhp_up_callbacks(cpu, st, target);
1082 cpus_write_unlock();
1086 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1090 if (!cpu_possible(cpu)) {
1091 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1093 #if defined(CONFIG_IA64)
1094 pr_err("please check additional_cpus= boot parameter\n");
1099 err = try_online_node(cpu_to_node(cpu));
1103 cpu_maps_update_begin();
1105 if (cpu_hotplug_disabled) {
1109 if (!cpu_smt_allowed(cpu)) {
1114 err = _cpu_up(cpu, 0, target);
1116 cpu_maps_update_done();
1120 int cpu_up(unsigned int cpu)
1122 return do_cpu_up(cpu, CPUHP_ONLINE);
1124 EXPORT_SYMBOL_GPL(cpu_up);
1126 #ifdef CONFIG_PM_SLEEP_SMP
1127 static cpumask_var_t frozen_cpus;
1129 int freeze_secondary_cpus(int primary)
1133 cpu_maps_update_begin();
1134 if (!cpu_online(primary))
1135 primary = cpumask_first(cpu_online_mask);
1137 * We take down all of the non-boot CPUs in one shot to avoid races
1138 * with the userspace trying to use the CPU hotplug at the same time
1140 cpumask_clear(frozen_cpus);
1142 pr_info("Disabling non-boot CPUs ...\n");
1143 for_each_online_cpu(cpu) {
1146 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1147 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1148 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1150 cpumask_set_cpu(cpu, frozen_cpus);
1152 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1158 BUG_ON(num_online_cpus() > 1);
1160 pr_err("Non-boot CPUs are not disabled\n");
1163 * Make sure the CPUs won't be enabled by someone else. We need to do
1164 * this even in case of failure as all disable_nonboot_cpus() users are
1165 * supposed to do enable_nonboot_cpus() on the failure path.
1167 cpu_hotplug_disabled++;
1169 cpu_maps_update_done();
1173 void __weak arch_enable_nonboot_cpus_begin(void)
1177 void __weak arch_enable_nonboot_cpus_end(void)
1181 void enable_nonboot_cpus(void)
1185 /* Allow everyone to use the CPU hotplug again */
1186 cpu_maps_update_begin();
1187 __cpu_hotplug_enable();
1188 if (cpumask_empty(frozen_cpus))
1191 pr_info("Enabling non-boot CPUs ...\n");
1193 arch_enable_nonboot_cpus_begin();
1195 for_each_cpu(cpu, frozen_cpus) {
1196 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1197 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1198 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1200 pr_info("CPU%d is up\n", cpu);
1203 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1206 arch_enable_nonboot_cpus_end();
1208 cpumask_clear(frozen_cpus);
1210 cpu_maps_update_done();
1213 static int __init alloc_frozen_cpus(void)
1215 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1219 core_initcall(alloc_frozen_cpus);
1222 * When callbacks for CPU hotplug notifications are being executed, we must
1223 * ensure that the state of the system with respect to the tasks being frozen
1224 * or not, as reported by the notification, remains unchanged *throughout the
1225 * duration* of the execution of the callbacks.
1226 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1228 * This synchronization is implemented by mutually excluding regular CPU
1229 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1230 * Hibernate notifications.
1233 cpu_hotplug_pm_callback(struct notifier_block *nb,
1234 unsigned long action, void *ptr)
1238 case PM_SUSPEND_PREPARE:
1239 case PM_HIBERNATION_PREPARE:
1240 cpu_hotplug_disable();
1243 case PM_POST_SUSPEND:
1244 case PM_POST_HIBERNATION:
1245 cpu_hotplug_enable();
1256 static int __init cpu_hotplug_pm_sync_init(void)
1259 * cpu_hotplug_pm_callback has higher priority than x86
1260 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1261 * to disable cpu hotplug to avoid cpu hotplug race.
1263 pm_notifier(cpu_hotplug_pm_callback, 0);
1266 core_initcall(cpu_hotplug_pm_sync_init);
1268 #endif /* CONFIG_PM_SLEEP_SMP */
1272 #endif /* CONFIG_SMP */
1274 /* Boot processor state steps */
1275 static struct cpuhp_step cpuhp_bp_states[] = {
1278 .startup.single = NULL,
1279 .teardown.single = NULL,
1282 [CPUHP_CREATE_THREADS]= {
1283 .name = "threads:prepare",
1284 .startup.single = smpboot_create_threads,
1285 .teardown.single = NULL,
1288 [CPUHP_PERF_PREPARE] = {
1289 .name = "perf:prepare",
1290 .startup.single = perf_event_init_cpu,
1291 .teardown.single = perf_event_exit_cpu,
1293 [CPUHP_WORKQUEUE_PREP] = {
1294 .name = "workqueue:prepare",
1295 .startup.single = workqueue_prepare_cpu,
1296 .teardown.single = NULL,
1298 [CPUHP_HRTIMERS_PREPARE] = {
1299 .name = "hrtimers:prepare",
1300 .startup.single = hrtimers_prepare_cpu,
1301 .teardown.single = hrtimers_dead_cpu,
1303 [CPUHP_SMPCFD_PREPARE] = {
1304 .name = "smpcfd:prepare",
1305 .startup.single = smpcfd_prepare_cpu,
1306 .teardown.single = smpcfd_dead_cpu,
1308 [CPUHP_RELAY_PREPARE] = {
1309 .name = "relay:prepare",
1310 .startup.single = relay_prepare_cpu,
1311 .teardown.single = NULL,
1313 [CPUHP_SLAB_PREPARE] = {
1314 .name = "slab:prepare",
1315 .startup.single = slab_prepare_cpu,
1316 .teardown.single = slab_dead_cpu,
1318 [CPUHP_RCUTREE_PREP] = {
1319 .name = "RCU/tree:prepare",
1320 .startup.single = rcutree_prepare_cpu,
1321 .teardown.single = rcutree_dead_cpu,
1324 * On the tear-down path, timers_dead_cpu() must be invoked
1325 * before blk_mq_queue_reinit_notify() from notify_dead(),
1326 * otherwise a RCU stall occurs.
1328 [CPUHP_TIMERS_PREPARE] = {
1329 .name = "timers:dead",
1330 .startup.single = timers_prepare_cpu,
1331 .teardown.single = timers_dead_cpu,
1333 /* Kicks the plugged cpu into life */
1334 [CPUHP_BRINGUP_CPU] = {
1335 .name = "cpu:bringup",
1336 .startup.single = bringup_cpu,
1337 .teardown.single = NULL,
1341 * Handled on controll processor until the plugged processor manages
1344 [CPUHP_TEARDOWN_CPU] = {
1345 .name = "cpu:teardown",
1346 .startup.single = NULL,
1347 .teardown.single = takedown_cpu,
1351 [CPUHP_BRINGUP_CPU] = { },
1355 /* Application processor state steps */
1356 static struct cpuhp_step cpuhp_ap_states[] = {
1358 /* Final state before CPU kills itself */
1359 [CPUHP_AP_IDLE_DEAD] = {
1360 .name = "idle:dead",
1363 * Last state before CPU enters the idle loop to die. Transient state
1364 * for synchronization.
1366 [CPUHP_AP_OFFLINE] = {
1367 .name = "ap:offline",
1370 /* First state is scheduler control. Interrupts are disabled */
1371 [CPUHP_AP_SCHED_STARTING] = {
1372 .name = "sched:starting",
1373 .startup.single = sched_cpu_starting,
1374 .teardown.single = sched_cpu_dying,
1376 [CPUHP_AP_RCUTREE_DYING] = {
1377 .name = "RCU/tree:dying",
1378 .startup.single = NULL,
1379 .teardown.single = rcutree_dying_cpu,
1381 [CPUHP_AP_SMPCFD_DYING] = {
1382 .name = "smpcfd:dying",
1383 .startup.single = NULL,
1384 .teardown.single = smpcfd_dying_cpu,
1386 /* Entry state on starting. Interrupts enabled from here on. Transient
1387 * state for synchronsization */
1388 [CPUHP_AP_ONLINE] = {
1389 .name = "ap:online",
1391 /* Handle smpboot threads park/unpark */
1392 [CPUHP_AP_SMPBOOT_THREADS] = {
1393 .name = "smpboot/threads:online",
1394 .startup.single = smpboot_unpark_threads,
1395 .teardown.single = smpboot_park_threads,
1397 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1398 .name = "irq/affinity:online",
1399 .startup.single = irq_affinity_online_cpu,
1400 .teardown.single = NULL,
1402 [CPUHP_AP_PERF_ONLINE] = {
1403 .name = "perf:online",
1404 .startup.single = perf_event_init_cpu,
1405 .teardown.single = perf_event_exit_cpu,
1407 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1408 .name = "workqueue:online",
1409 .startup.single = workqueue_online_cpu,
1410 .teardown.single = workqueue_offline_cpu,
1412 [CPUHP_AP_RCUTREE_ONLINE] = {
1413 .name = "RCU/tree:online",
1414 .startup.single = rcutree_online_cpu,
1415 .teardown.single = rcutree_offline_cpu,
1419 * The dynamically registered state space is here
1423 /* Last state is scheduler control setting the cpu active */
1424 [CPUHP_AP_ACTIVE] = {
1425 .name = "sched:active",
1426 .startup.single = sched_cpu_activate,
1427 .teardown.single = sched_cpu_deactivate,
1431 /* CPU is fully up and running. */
1434 .startup.single = NULL,
1435 .teardown.single = NULL,
1439 /* Sanity check for callbacks */
1440 static int cpuhp_cb_check(enum cpuhp_state state)
1442 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1448 * Returns a free for dynamic slot assignment of the Online state. The states
1449 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1450 * by having no name assigned.
1452 static int cpuhp_reserve_state(enum cpuhp_state state)
1454 enum cpuhp_state i, end;
1455 struct cpuhp_step *step;
1458 case CPUHP_AP_ONLINE_DYN:
1459 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN;
1460 end = CPUHP_AP_ONLINE_DYN_END;
1462 case CPUHP_BP_PREPARE_DYN:
1463 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN;
1464 end = CPUHP_BP_PREPARE_DYN_END;
1470 for (i = state; i <= end; i++, step++) {
1474 WARN(1, "No more dynamic states available for CPU hotplug\n");
1478 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1479 int (*startup)(unsigned int cpu),
1480 int (*teardown)(unsigned int cpu),
1481 bool multi_instance)
1483 /* (Un)Install the callbacks for further cpu hotplug operations */
1484 struct cpuhp_step *sp;
1488 * If name is NULL, then the state gets removed.
1490 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1491 * the first allocation from these dynamic ranges, so the removal
1492 * would trigger a new allocation and clear the wrong (already
1493 * empty) state, leaving the callbacks of the to be cleared state
1494 * dangling, which causes wreckage on the next hotplug operation.
1496 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1497 state == CPUHP_BP_PREPARE_DYN)) {
1498 ret = cpuhp_reserve_state(state);
1503 sp = cpuhp_get_step(state);
1504 if (name && sp->name)
1507 sp->startup.single = startup;
1508 sp->teardown.single = teardown;
1510 sp->multi_instance = multi_instance;
1511 INIT_HLIST_HEAD(&sp->list);
1515 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1517 return cpuhp_get_step(state)->teardown.single;
1521 * Call the startup/teardown function for a step either on the AP or
1522 * on the current CPU.
1524 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1525 struct hlist_node *node)
1527 struct cpuhp_step *sp = cpuhp_get_step(state);
1531 * If there's nothing to do, we done.
1532 * Relies on the union for multi_instance.
1534 if ((bringup && !sp->startup.single) ||
1535 (!bringup && !sp->teardown.single))
1538 * The non AP bound callbacks can fail on bringup. On teardown
1539 * e.g. module removal we crash for now.
1542 if (cpuhp_is_ap_state(state))
1543 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1545 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1547 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1549 BUG_ON(ret && !bringup);
1554 * Called from __cpuhp_setup_state on a recoverable failure.
1556 * Note: The teardown callbacks for rollback are not allowed to fail!
1558 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1559 struct hlist_node *node)
1563 /* Roll back the already executed steps on the other cpus */
1564 for_each_present_cpu(cpu) {
1565 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1566 int cpustate = st->state;
1568 if (cpu >= failedcpu)
1571 /* Did we invoke the startup call on that cpu ? */
1572 if (cpustate >= state)
1573 cpuhp_issue_call(cpu, state, false, node);
1577 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1578 struct hlist_node *node,
1581 struct cpuhp_step *sp;
1585 lockdep_assert_cpus_held();
1587 sp = cpuhp_get_step(state);
1588 if (sp->multi_instance == false)
1591 mutex_lock(&cpuhp_state_mutex);
1593 if (!invoke || !sp->startup.multi)
1597 * Try to call the startup callback for each present cpu
1598 * depending on the hotplug state of the cpu.
1600 for_each_present_cpu(cpu) {
1601 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1602 int cpustate = st->state;
1604 if (cpustate < state)
1607 ret = cpuhp_issue_call(cpu, state, true, node);
1609 if (sp->teardown.multi)
1610 cpuhp_rollback_install(cpu, state, node);
1616 hlist_add_head(node, &sp->list);
1618 mutex_unlock(&cpuhp_state_mutex);
1622 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1628 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1632 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1635 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1636 * @state: The state to setup
1637 * @invoke: If true, the startup function is invoked for cpus where
1638 * cpu state >= @state
1639 * @startup: startup callback function
1640 * @teardown: teardown callback function
1641 * @multi_instance: State is set up for multiple instances which get
1644 * The caller needs to hold cpus read locked while calling this function.
1647 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1648 * 0 for all other states
1649 * On failure: proper (negative) error code
1651 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1652 const char *name, bool invoke,
1653 int (*startup)(unsigned int cpu),
1654 int (*teardown)(unsigned int cpu),
1655 bool multi_instance)
1660 lockdep_assert_cpus_held();
1662 if (cpuhp_cb_check(state) || !name)
1665 mutex_lock(&cpuhp_state_mutex);
1667 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1670 dynstate = state == CPUHP_AP_ONLINE_DYN;
1671 if (ret > 0 && dynstate) {
1676 if (ret || !invoke || !startup)
1680 * Try to call the startup callback for each present cpu
1681 * depending on the hotplug state of the cpu.
1683 for_each_present_cpu(cpu) {
1684 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1685 int cpustate = st->state;
1687 if (cpustate < state)
1690 ret = cpuhp_issue_call(cpu, state, true, NULL);
1693 cpuhp_rollback_install(cpu, state, NULL);
1694 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1699 mutex_unlock(&cpuhp_state_mutex);
1701 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1702 * dynamically allocated state in case of success.
1704 if (!ret && dynstate)
1708 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1710 int __cpuhp_setup_state(enum cpuhp_state state,
1711 const char *name, bool invoke,
1712 int (*startup)(unsigned int cpu),
1713 int (*teardown)(unsigned int cpu),
1714 bool multi_instance)
1719 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1720 teardown, multi_instance);
1724 EXPORT_SYMBOL(__cpuhp_setup_state);
1726 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1727 struct hlist_node *node, bool invoke)
1729 struct cpuhp_step *sp = cpuhp_get_step(state);
1732 BUG_ON(cpuhp_cb_check(state));
1734 if (!sp->multi_instance)
1738 mutex_lock(&cpuhp_state_mutex);
1740 if (!invoke || !cpuhp_get_teardown_cb(state))
1743 * Call the teardown callback for each present cpu depending
1744 * on the hotplug state of the cpu. This function is not
1745 * allowed to fail currently!
1747 for_each_present_cpu(cpu) {
1748 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1749 int cpustate = st->state;
1751 if (cpustate >= state)
1752 cpuhp_issue_call(cpu, state, false, node);
1757 mutex_unlock(&cpuhp_state_mutex);
1762 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1765 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1766 * @state: The state to remove
1767 * @invoke: If true, the teardown function is invoked for cpus where
1768 * cpu state >= @state
1770 * The caller needs to hold cpus read locked while calling this function.
1771 * The teardown callback is currently not allowed to fail. Think
1772 * about module removal!
1774 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1776 struct cpuhp_step *sp = cpuhp_get_step(state);
1779 BUG_ON(cpuhp_cb_check(state));
1781 lockdep_assert_cpus_held();
1783 mutex_lock(&cpuhp_state_mutex);
1784 if (sp->multi_instance) {
1785 WARN(!hlist_empty(&sp->list),
1786 "Error: Removing state %d which has instances left.\n",
1791 if (!invoke || !cpuhp_get_teardown_cb(state))
1795 * Call the teardown callback for each present cpu depending
1796 * on the hotplug state of the cpu. This function is not
1797 * allowed to fail currently!
1799 for_each_present_cpu(cpu) {
1800 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1801 int cpustate = st->state;
1803 if (cpustate >= state)
1804 cpuhp_issue_call(cpu, state, false, NULL);
1807 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1808 mutex_unlock(&cpuhp_state_mutex);
1810 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1812 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1815 __cpuhp_remove_state_cpuslocked(state, invoke);
1818 EXPORT_SYMBOL(__cpuhp_remove_state);
1820 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1821 static ssize_t show_cpuhp_state(struct device *dev,
1822 struct device_attribute *attr, char *buf)
1824 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1826 return sprintf(buf, "%d\n", st->state);
1828 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1830 static ssize_t write_cpuhp_target(struct device *dev,
1831 struct device_attribute *attr,
1832 const char *buf, size_t count)
1834 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1835 struct cpuhp_step *sp;
1838 ret = kstrtoint(buf, 10, &target);
1842 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1843 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1846 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1850 ret = lock_device_hotplug_sysfs();
1854 mutex_lock(&cpuhp_state_mutex);
1855 sp = cpuhp_get_step(target);
1856 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1857 mutex_unlock(&cpuhp_state_mutex);
1861 if (st->state < target)
1862 ret = do_cpu_up(dev->id, target);
1864 ret = do_cpu_down(dev->id, target);
1866 unlock_device_hotplug();
1867 return ret ? ret : count;
1870 static ssize_t show_cpuhp_target(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->target);
1877 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1880 static ssize_t write_cpuhp_fail(struct device *dev,
1881 struct device_attribute *attr,
1882 const char *buf, size_t count)
1884 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1885 struct cpuhp_step *sp;
1888 ret = kstrtoint(buf, 10, &fail);
1893 * Cannot fail STARTING/DYING callbacks.
1895 if (cpuhp_is_atomic_state(fail))
1899 * Cannot fail anything that doesn't have callbacks.
1901 mutex_lock(&cpuhp_state_mutex);
1902 sp = cpuhp_get_step(fail);
1903 if (!sp->startup.single && !sp->teardown.single)
1905 mutex_unlock(&cpuhp_state_mutex);
1914 static ssize_t show_cpuhp_fail(struct device *dev,
1915 struct device_attribute *attr, char *buf)
1917 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1919 return sprintf(buf, "%d\n", st->fail);
1922 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1924 static struct attribute *cpuhp_cpu_attrs[] = {
1925 &dev_attr_state.attr,
1926 &dev_attr_target.attr,
1927 &dev_attr_fail.attr,
1931 static const struct attribute_group cpuhp_cpu_attr_group = {
1932 .attrs = cpuhp_cpu_attrs,
1937 static ssize_t show_cpuhp_states(struct device *dev,
1938 struct device_attribute *attr, char *buf)
1940 ssize_t cur, res = 0;
1943 mutex_lock(&cpuhp_state_mutex);
1944 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1945 struct cpuhp_step *sp = cpuhp_get_step(i);
1948 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1953 mutex_unlock(&cpuhp_state_mutex);
1956 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1958 static struct attribute *cpuhp_cpu_root_attrs[] = {
1959 &dev_attr_states.attr,
1963 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1964 .attrs = cpuhp_cpu_root_attrs,
1969 #ifdef CONFIG_HOTPLUG_SMT
1971 static const char *smt_states[] = {
1972 [CPU_SMT_ENABLED] = "on",
1973 [CPU_SMT_DISABLED] = "off",
1974 [CPU_SMT_FORCE_DISABLED] = "forceoff",
1975 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
1979 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
1981 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
1984 static void cpuhp_offline_cpu_device(unsigned int cpu)
1986 struct device *dev = get_cpu_device(cpu);
1988 dev->offline = true;
1989 /* Tell user space about the state change */
1990 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
1993 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
1997 cpu_maps_update_begin();
1998 for_each_online_cpu(cpu) {
1999 if (topology_is_primary_thread(cpu))
2001 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2005 * As this needs to hold the cpu maps lock it's impossible
2006 * to call device_offline() because that ends up calling
2007 * cpu_down() which takes cpu maps lock. cpu maps lock
2008 * needs to be held as this might race against in kernel
2009 * abusers of the hotplug machinery (thermal management).
2011 * So nothing would update device:offline state. That would
2012 * leave the sysfs entry stale and prevent onlining after
2013 * smt control has been changed to 'off' again. This is
2014 * called under the sysfs hotplug lock, so it is properly
2015 * serialized against the regular offline usage.
2017 cpuhp_offline_cpu_device(cpu);
2020 cpu_smt_control = ctrlval;
2021 cpu_maps_update_done();
2025 static void cpuhp_smt_enable(void)
2027 cpu_maps_update_begin();
2028 cpu_smt_control = CPU_SMT_ENABLED;
2029 cpu_maps_update_done();
2033 store_smt_control(struct device *dev, struct device_attribute *attr,
2034 const char *buf, size_t count)
2038 if (sysfs_streq(buf, "on"))
2039 ctrlval = CPU_SMT_ENABLED;
2040 else if (sysfs_streq(buf, "off"))
2041 ctrlval = CPU_SMT_DISABLED;
2042 else if (sysfs_streq(buf, "forceoff"))
2043 ctrlval = CPU_SMT_FORCE_DISABLED;
2047 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2050 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2053 ret = lock_device_hotplug_sysfs();
2057 if (ctrlval != cpu_smt_control) {
2059 case CPU_SMT_ENABLED:
2062 case CPU_SMT_DISABLED:
2063 case CPU_SMT_FORCE_DISABLED:
2064 ret = cpuhp_smt_disable(ctrlval);
2069 unlock_device_hotplug();
2070 return ret ? ret : count;
2072 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2075 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2077 bool active = topology_max_smt_threads() > 1;
2079 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2081 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2083 static struct attribute *cpuhp_smt_attrs[] = {
2084 &dev_attr_control.attr,
2085 &dev_attr_active.attr,
2089 static const struct attribute_group cpuhp_smt_attr_group = {
2090 .attrs = cpuhp_smt_attrs,
2095 static int __init cpu_smt_state_init(void)
2097 if (!topology_smt_supported())
2098 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
2100 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2101 &cpuhp_smt_attr_group);
2105 static inline int cpu_smt_state_init(void) { return 0; }
2108 static int __init cpuhp_sysfs_init(void)
2112 ret = cpu_smt_state_init();
2116 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2117 &cpuhp_cpu_root_attr_group);
2121 for_each_possible_cpu(cpu) {
2122 struct device *dev = get_cpu_device(cpu);
2126 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2132 device_initcall(cpuhp_sysfs_init);
2136 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2137 * represents all NR_CPUS bits binary values of 1<<nr.
2139 * It is used by cpumask_of() to get a constant address to a CPU
2140 * mask value that has a single bit set only.
2143 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2144 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2145 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2146 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2147 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2149 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2151 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2152 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2153 #if BITS_PER_LONG > 32
2154 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2155 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2158 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2160 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2161 EXPORT_SYMBOL(cpu_all_bits);
2163 #ifdef CONFIG_INIT_ALL_POSSIBLE
2164 struct cpumask __cpu_possible_mask __read_mostly
2167 struct cpumask __cpu_possible_mask __read_mostly;
2169 EXPORT_SYMBOL(__cpu_possible_mask);
2171 struct cpumask __cpu_online_mask __read_mostly;
2172 EXPORT_SYMBOL(__cpu_online_mask);
2174 struct cpumask __cpu_present_mask __read_mostly;
2175 EXPORT_SYMBOL(__cpu_present_mask);
2177 struct cpumask __cpu_active_mask __read_mostly;
2178 EXPORT_SYMBOL(__cpu_active_mask);
2180 void init_cpu_present(const struct cpumask *src)
2182 cpumask_copy(&__cpu_present_mask, src);
2185 void init_cpu_possible(const struct cpumask *src)
2187 cpumask_copy(&__cpu_possible_mask, src);
2190 void init_cpu_online(const struct cpumask *src)
2192 cpumask_copy(&__cpu_online_mask, src);
2196 * Activate the first processor.
2198 void __init boot_cpu_init(void)
2200 int cpu = smp_processor_id();
2202 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2203 set_cpu_online(cpu, true);
2204 set_cpu_active(cpu, true);
2205 set_cpu_present(cpu, true);
2206 set_cpu_possible(cpu, true);
2209 __boot_cpu_id = cpu;
2214 * Must be called _AFTER_ setting up the per_cpu areas
2216 void __init boot_cpu_hotplug_init(void)
2218 this_cpu_write(cpuhp_state.booted_once, true);
2219 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);