2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/proc_fs.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/hotplug.h>
12 #include <linux/sched/task.h>
13 #include <linux/unistd.h>
14 #include <linux/cpu.h>
15 #include <linux/oom.h>
16 #include <linux/rcupdate.h>
17 #include <linux/export.h>
18 #include <linux/bug.h>
19 #include <linux/kthread.h>
20 #include <linux/stop_machine.h>
21 #include <linux/mutex.h>
22 #include <linux/gfp.h>
23 #include <linux/suspend.h>
24 #include <linux/lockdep.h>
25 #include <linux/tick.h>
26 #include <linux/irq.h>
27 #include <linux/nmi.h>
28 #include <linux/smpboot.h>
29 #include <linux/relay.h>
30 #include <linux/slab.h>
31 #include <linux/percpu-rwsem.h>
33 #include <trace/events/power.h>
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/cpuhp.h>
40 * cpuhp_cpu_state - Per cpu hotplug state storage
41 * @state: The current cpu state
42 * @target: The target state
43 * @thread: Pointer to the hotplug thread
44 * @should_run: Thread should execute
45 * @rollback: Perform a rollback
46 * @single: Single callback invocation
47 * @bringup: Single callback bringup or teardown selector
48 * @cb_state: The state for a single callback (install/uninstall)
49 * @result: Result of the operation
50 * @done_up: Signal completion to the issuer of the task for cpu-up
51 * @done_down: Signal completion to the issuer of the task for cpu-down
53 struct cpuhp_cpu_state {
54 enum cpuhp_state state;
55 enum cpuhp_state target;
56 enum cpuhp_state fail;
58 struct task_struct *thread;
64 struct hlist_node *node;
65 struct hlist_node *last;
66 enum cpuhp_state cb_state;
68 struct completion done_up;
69 struct completion done_down;
73 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
74 .fail = CPUHP_INVALID,
77 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
78 static struct lockdep_map cpuhp_state_up_map =
79 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
80 static struct lockdep_map cpuhp_state_down_map =
81 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
84 static void inline cpuhp_lock_acquire(bool bringup)
86 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
89 static void inline cpuhp_lock_release(bool bringup)
91 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
95 static void inline cpuhp_lock_acquire(bool bringup) { }
96 static void inline cpuhp_lock_release(bool bringup) { }
101 * cpuhp_step - Hotplug state machine step
102 * @name: Name of the step
103 * @startup: Startup function of the step
104 * @teardown: Teardown function of the step
105 * @skip_onerr: Do not invoke the functions on error rollback
106 * Will go away once the notifiers are gone
107 * @cant_stop: Bringup/teardown can't be stopped at this step
112 int (*single)(unsigned int cpu);
113 int (*multi)(unsigned int cpu,
114 struct hlist_node *node);
117 int (*single)(unsigned int cpu);
118 int (*multi)(unsigned int cpu,
119 struct hlist_node *node);
121 struct hlist_head list;
127 static DEFINE_MUTEX(cpuhp_state_mutex);
128 static struct cpuhp_step cpuhp_bp_states[];
129 static struct cpuhp_step cpuhp_ap_states[];
131 static bool cpuhp_is_ap_state(enum cpuhp_state state)
134 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
135 * purposes as that state is handled explicitly in cpu_down.
137 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
140 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
142 struct cpuhp_step *sp;
144 sp = cpuhp_is_ap_state(state) ? cpuhp_ap_states : cpuhp_bp_states;
149 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
150 * @cpu: The cpu for which the callback should be invoked
151 * @state: The state to do callbacks for
152 * @bringup: True if the bringup callback should be invoked
153 * @node: For multi-instance, do a single entry callback for install/remove
154 * @lastp: For multi-instance rollback, remember how far we got
156 * Called from cpu hotplug and from the state register machinery.
158 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
159 bool bringup, struct hlist_node *node,
160 struct hlist_node **lastp)
162 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
163 struct cpuhp_step *step = cpuhp_get_step(state);
164 int (*cbm)(unsigned int cpu, struct hlist_node *node);
165 int (*cb)(unsigned int cpu);
168 if (st->fail == state) {
169 st->fail = CPUHP_INVALID;
171 if (!(bringup ? step->startup.single : step->teardown.single))
177 if (!step->multi_instance) {
178 WARN_ON_ONCE(lastp && *lastp);
179 cb = bringup ? step->startup.single : step->teardown.single;
182 trace_cpuhp_enter(cpu, st->target, state, cb);
184 trace_cpuhp_exit(cpu, st->state, state, ret);
187 cbm = bringup ? step->startup.multi : step->teardown.multi;
191 /* Single invocation for instance add/remove */
193 WARN_ON_ONCE(lastp && *lastp);
194 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
195 ret = cbm(cpu, node);
196 trace_cpuhp_exit(cpu, st->state, state, ret);
200 /* State transition. Invoke on all instances */
202 hlist_for_each(node, &step->list) {
203 if (lastp && node == *lastp)
206 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
207 ret = cbm(cpu, node);
208 trace_cpuhp_exit(cpu, st->state, state, ret);
222 /* Rollback the instances if one failed */
223 cbm = !bringup ? step->startup.multi : step->teardown.multi;
227 hlist_for_each(node, &step->list) {
231 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
232 ret = cbm(cpu, node);
233 trace_cpuhp_exit(cpu, st->state, state, ret);
235 * Rollback must not fail,
243 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
246 wait_for_completion(done);
249 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
251 struct completion *done = bringup ? &st->done_up : &st->done_down;
256 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
258 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
260 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
263 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
264 static DEFINE_MUTEX(cpu_add_remove_lock);
265 bool cpuhp_tasks_frozen;
266 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
269 * The following two APIs (cpu_maps_update_begin/done) must be used when
270 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
272 void cpu_maps_update_begin(void)
274 mutex_lock(&cpu_add_remove_lock);
277 void cpu_maps_update_done(void)
279 mutex_unlock(&cpu_add_remove_lock);
283 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
284 * Should always be manipulated under cpu_add_remove_lock
286 static int cpu_hotplug_disabled;
288 #ifdef CONFIG_HOTPLUG_CPU
290 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
292 void cpus_read_lock(void)
294 percpu_down_read(&cpu_hotplug_lock);
296 EXPORT_SYMBOL_GPL(cpus_read_lock);
298 void cpus_read_unlock(void)
300 percpu_up_read(&cpu_hotplug_lock);
302 EXPORT_SYMBOL_GPL(cpus_read_unlock);
304 void cpus_write_lock(void)
306 percpu_down_write(&cpu_hotplug_lock);
309 void cpus_write_unlock(void)
311 percpu_up_write(&cpu_hotplug_lock);
314 void lockdep_assert_cpus_held(void)
316 percpu_rwsem_assert_held(&cpu_hotplug_lock);
320 * Wait for currently running CPU hotplug operations to complete (if any) and
321 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
322 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
323 * hotplug path before performing hotplug operations. So acquiring that lock
324 * guarantees mutual exclusion from any currently running hotplug operations.
326 void cpu_hotplug_disable(void)
328 cpu_maps_update_begin();
329 cpu_hotplug_disabled++;
330 cpu_maps_update_done();
332 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
334 static void __cpu_hotplug_enable(void)
336 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
338 cpu_hotplug_disabled--;
341 void cpu_hotplug_enable(void)
343 cpu_maps_update_begin();
344 __cpu_hotplug_enable();
345 cpu_maps_update_done();
347 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
348 #endif /* CONFIG_HOTPLUG_CPU */
350 #ifdef CONFIG_HOTPLUG_SMT
351 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
352 EXPORT_SYMBOL_GPL(cpu_smt_control);
354 void __init cpu_smt_disable(bool force)
356 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
357 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
361 pr_info("SMT: Force disabled\n");
362 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
364 cpu_smt_control = CPU_SMT_DISABLED;
369 * The decision whether SMT is supported can only be done after the full
370 * CPU identification. Called from architecture code.
372 void __init cpu_smt_check_topology(void)
374 if (!topology_smt_supported())
375 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
378 static int __init smt_cmdline_disable(char *str)
380 cpu_smt_disable(str && !strcmp(str, "force"));
383 early_param("nosmt", smt_cmdline_disable);
385 static inline bool cpu_smt_allowed(unsigned int cpu)
387 if (cpu_smt_control == CPU_SMT_ENABLED)
390 if (topology_is_primary_thread(cpu))
394 * On x86 it's required to boot all logical CPUs at least once so
395 * that the init code can get a chance to set CR4.MCE on each
396 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
397 * core will shutdown the machine.
399 return !per_cpu(cpuhp_state, cpu).booted_once;
402 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
405 static inline enum cpuhp_state
406 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
408 enum cpuhp_state prev_state = st->state;
410 st->rollback = false;
415 st->bringup = st->state < target;
421 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
426 * If we have st->last we need to undo partial multi_instance of this
427 * state first. Otherwise start undo at the previous state.
436 st->target = prev_state;
437 st->bringup = !st->bringup;
440 /* Regular hotplug invocation of the AP hotplug thread */
441 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
443 if (!st->single && st->state == st->target)
448 * Make sure the above stores are visible before should_run becomes
449 * true. Paired with the mb() above in cpuhp_thread_fun()
452 st->should_run = true;
453 wake_up_process(st->thread);
454 wait_for_ap_thread(st, st->bringup);
457 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
459 enum cpuhp_state prev_state;
462 prev_state = cpuhp_set_state(st, target);
464 if ((ret = st->result)) {
465 cpuhp_reset_state(st, prev_state);
472 static int bringup_wait_for_ap(unsigned int cpu)
474 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
476 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
477 wait_for_ap_thread(st, true);
478 if (WARN_ON_ONCE((!cpu_online(cpu))))
481 /* Unpark the stopper thread and the hotplug thread of the target cpu */
482 stop_machine_unpark(cpu);
483 kthread_unpark(st->thread);
486 * SMT soft disabling on X86 requires to bring the CPU out of the
487 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
488 * CPU marked itself as booted_once in cpu_notify_starting() so the
489 * cpu_smt_allowed() check will now return false if this is not the
492 if (!cpu_smt_allowed(cpu))
495 if (st->target <= CPUHP_AP_ONLINE_IDLE)
498 return cpuhp_kick_ap(st, st->target);
501 static int bringup_cpu(unsigned int cpu)
503 struct task_struct *idle = idle_thread_get(cpu);
507 * Some architectures have to walk the irq descriptors to
508 * setup the vector space for the cpu which comes online.
509 * Prevent irq alloc/free across the bringup.
513 /* Arch-specific enabling code. */
514 ret = __cpu_up(cpu, idle);
518 return bringup_wait_for_ap(cpu);
522 * Hotplug state machine related functions
525 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
527 for (st->state--; st->state > st->target; st->state--) {
528 struct cpuhp_step *step = cpuhp_get_step(st->state);
530 if (!step->skip_onerr)
531 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
535 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
536 enum cpuhp_state target)
538 enum cpuhp_state prev_state = st->state;
541 while (st->state < target) {
543 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
545 st->target = prev_state;
546 undo_cpu_up(cpu, st);
554 * The cpu hotplug threads manage the bringup and teardown of the cpus
556 static void cpuhp_create(unsigned int cpu)
558 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
560 init_completion(&st->done_up);
561 init_completion(&st->done_down);
564 static int cpuhp_should_run(unsigned int cpu)
566 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
568 return st->should_run;
572 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
573 * callbacks when a state gets [un]installed at runtime.
575 * Each invocation of this function by the smpboot thread does a single AP
578 * It has 3 modes of operation:
579 * - single: runs st->cb_state
580 * - up: runs ++st->state, while st->state < st->target
581 * - down: runs st->state--, while st->state > st->target
583 * When complete or on error, should_run is cleared and the completion is fired.
585 static void cpuhp_thread_fun(unsigned int cpu)
587 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
588 bool bringup = st->bringup;
589 enum cpuhp_state state;
592 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
593 * that if we see ->should_run we also see the rest of the state.
597 if (WARN_ON_ONCE(!st->should_run))
600 cpuhp_lock_acquire(bringup);
603 state = st->cb_state;
604 st->should_run = false;
609 st->should_run = (st->state < st->target);
610 WARN_ON_ONCE(st->state > st->target);
614 st->should_run = (st->state > st->target);
615 WARN_ON_ONCE(st->state < st->target);
619 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
622 struct cpuhp_step *step = cpuhp_get_step(state);
623 if (step->skip_onerr)
627 if (cpuhp_is_atomic_state(state)) {
629 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
633 * STARTING/DYING must not fail!
635 WARN_ON_ONCE(st->result);
637 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
642 * If we fail on a rollback, we're up a creek without no
643 * paddle, no way forward, no way back. We loose, thanks for
646 WARN_ON_ONCE(st->rollback);
647 st->should_run = false;
651 cpuhp_lock_release(bringup);
654 complete_ap_thread(st, bringup);
657 /* Invoke a single callback on a remote cpu */
659 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
660 struct hlist_node *node)
662 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
665 if (!cpu_online(cpu))
668 cpuhp_lock_acquire(false);
669 cpuhp_lock_release(false);
671 cpuhp_lock_acquire(true);
672 cpuhp_lock_release(true);
675 * If we are up and running, use the hotplug thread. For early calls
676 * we invoke the thread function directly.
679 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
681 st->rollback = false;
685 st->bringup = bringup;
686 st->cb_state = state;
692 * If we failed and did a partial, do a rollback.
694 if ((ret = st->result) && st->last) {
696 st->bringup = !bringup;
702 * Clean up the leftovers so the next hotplug operation wont use stale
705 st->node = st->last = NULL;
709 static int cpuhp_kick_ap_work(unsigned int cpu)
711 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
712 enum cpuhp_state prev_state = st->state;
715 cpuhp_lock_acquire(false);
716 cpuhp_lock_release(false);
718 cpuhp_lock_acquire(true);
719 cpuhp_lock_release(true);
721 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
722 ret = cpuhp_kick_ap(st, st->target);
723 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
728 static struct smp_hotplug_thread cpuhp_threads = {
729 .store = &cpuhp_state.thread,
730 .create = &cpuhp_create,
731 .thread_should_run = cpuhp_should_run,
732 .thread_fn = cpuhp_thread_fun,
733 .thread_comm = "cpuhp/%u",
737 void __init cpuhp_threads_init(void)
739 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
740 kthread_unpark(this_cpu_read(cpuhp_state.thread));
743 #ifdef CONFIG_HOTPLUG_CPU
745 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
748 * This function walks all processes, finds a valid mm struct for each one and
749 * then clears a corresponding bit in mm's cpumask. While this all sounds
750 * trivial, there are various non-obvious corner cases, which this function
751 * tries to solve in a safe manner.
753 * Also note that the function uses a somewhat relaxed locking scheme, so it may
754 * be called only for an already offlined CPU.
756 void clear_tasks_mm_cpumask(int cpu)
758 struct task_struct *p;
761 * This function is called after the cpu is taken down and marked
762 * offline, so its not like new tasks will ever get this cpu set in
763 * their mm mask. -- Peter Zijlstra
764 * Thus, we may use rcu_read_lock() here, instead of grabbing
765 * full-fledged tasklist_lock.
767 WARN_ON(cpu_online(cpu));
769 for_each_process(p) {
770 struct task_struct *t;
773 * Main thread might exit, but other threads may still have
774 * a valid mm. Find one.
776 t = find_lock_task_mm(p);
779 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
785 /* Take this CPU down. */
786 static int take_cpu_down(void *_param)
788 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
789 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
790 int err, cpu = smp_processor_id();
793 /* Ensure this CPU doesn't handle any more interrupts. */
794 err = __cpu_disable();
799 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
800 * do this step again.
802 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
804 /* Invoke the former CPU_DYING callbacks */
805 for (; st->state > target; st->state--) {
806 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
808 * DYING must not fail!
813 /* Give up timekeeping duties */
814 tick_handover_do_timer();
815 /* Park the stopper thread */
816 stop_machine_park(cpu);
820 static int takedown_cpu(unsigned int cpu)
822 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
825 /* Park the smpboot threads */
826 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
829 * Prevent irq alloc/free while the dying cpu reorganizes the
830 * interrupt affinities.
835 * So now all preempt/rcu users must observe !cpu_active().
837 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
839 /* CPU refused to die */
841 /* Unpark the hotplug thread so we can rollback there */
842 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
845 BUG_ON(cpu_online(cpu));
848 * The CPUHP_AP_SCHED_MIGRATE_DYING callback will have removed all
849 * runnable tasks from the cpu, there's only the idle task left now
850 * that the migration thread is done doing the stop_machine thing.
852 * Wait for the stop thread to go away.
854 wait_for_ap_thread(st, false);
855 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
857 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
860 hotplug_cpu__broadcast_tick_pull(cpu);
861 /* This actually kills the CPU. */
864 tick_cleanup_dead_cpu(cpu);
865 rcutree_migrate_callbacks(cpu);
869 static void cpuhp_complete_idle_dead(void *arg)
871 struct cpuhp_cpu_state *st = arg;
873 complete_ap_thread(st, false);
876 void cpuhp_report_idle_dead(void)
878 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
880 BUG_ON(st->state != CPUHP_AP_OFFLINE);
881 rcu_report_dead(smp_processor_id());
882 st->state = CPUHP_AP_IDLE_DEAD;
884 * We cannot call complete after rcu_report_dead() so we delegate it
887 smp_call_function_single(cpumask_first(cpu_online_mask),
888 cpuhp_complete_idle_dead, st, 0);
891 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
893 for (st->state++; st->state < st->target; st->state++) {
894 struct cpuhp_step *step = cpuhp_get_step(st->state);
896 if (!step->skip_onerr)
897 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
901 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
902 enum cpuhp_state target)
904 enum cpuhp_state prev_state = st->state;
907 for (; st->state > target; st->state--) {
908 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
910 st->target = prev_state;
911 undo_cpu_down(cpu, st);
918 /* Requires cpu_add_remove_lock to be held */
919 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
920 enum cpuhp_state target)
922 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
923 int prev_state, ret = 0;
925 if (num_online_cpus() == 1)
928 if (!cpu_present(cpu))
933 cpuhp_tasks_frozen = tasks_frozen;
935 prev_state = cpuhp_set_state(st, target);
937 * If the current CPU state is in the range of the AP hotplug thread,
938 * then we need to kick the thread.
940 if (st->state > CPUHP_TEARDOWN_CPU) {
941 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
942 ret = cpuhp_kick_ap_work(cpu);
944 * The AP side has done the error rollback already. Just
945 * return the error code..
951 * We might have stopped still in the range of the AP hotplug
952 * thread. Nothing to do anymore.
954 if (st->state > CPUHP_TEARDOWN_CPU)
960 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
961 * to do the further cleanups.
963 ret = cpuhp_down_callbacks(cpu, st, target);
964 if (ret && st->state > CPUHP_TEARDOWN_CPU && st->state < prev_state) {
965 cpuhp_reset_state(st, prev_state);
972 * Do post unplug cleanup. This is still protected against
973 * concurrent CPU hotplug via cpu_add_remove_lock.
975 lockup_detector_cleanup();
979 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
981 if (cpu_hotplug_disabled)
983 return _cpu_down(cpu, 0, target);
986 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
990 cpu_maps_update_begin();
991 err = cpu_down_maps_locked(cpu, target);
992 cpu_maps_update_done();
996 int cpu_down(unsigned int cpu)
998 return do_cpu_down(cpu, CPUHP_OFFLINE);
1000 EXPORT_SYMBOL(cpu_down);
1003 #define takedown_cpu NULL
1004 #endif /*CONFIG_HOTPLUG_CPU*/
1007 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1008 * @cpu: cpu that just started
1010 * It must be called by the arch code on the new cpu, before the new cpu
1011 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1013 void notify_cpu_starting(unsigned int cpu)
1015 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1016 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1019 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1020 st->booted_once = true;
1021 while (st->state < target) {
1023 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1025 * STARTING must not fail!
1032 * Called from the idle task. Wake up the controlling task which brings the
1033 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1034 * the rest of the online bringup to the hotplug thread.
1036 void cpuhp_online_idle(enum cpuhp_state state)
1038 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1040 /* Happens for the boot cpu */
1041 if (state != CPUHP_AP_ONLINE_IDLE)
1044 st->state = CPUHP_AP_ONLINE_IDLE;
1045 complete_ap_thread(st, true);
1048 /* Requires cpu_add_remove_lock to be held */
1049 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1051 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1052 struct task_struct *idle;
1057 if (!cpu_present(cpu)) {
1063 * The caller of do_cpu_up might have raced with another
1064 * caller. Ignore it for now.
1066 if (st->state >= target)
1069 if (st->state == CPUHP_OFFLINE) {
1070 /* Let it fail before we try to bring the cpu up */
1071 idle = idle_thread_get(cpu);
1073 ret = PTR_ERR(idle);
1078 cpuhp_tasks_frozen = tasks_frozen;
1080 cpuhp_set_state(st, target);
1082 * If the current CPU state is in the range of the AP hotplug thread,
1083 * then we need to kick the thread once more.
1085 if (st->state > CPUHP_BRINGUP_CPU) {
1086 ret = cpuhp_kick_ap_work(cpu);
1088 * The AP side has done the error rollback already. Just
1089 * return the error code..
1096 * Try to reach the target state. We max out on the BP at
1097 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1098 * responsible for bringing it up to the target state.
1100 target = min((int)target, CPUHP_BRINGUP_CPU);
1101 ret = cpuhp_up_callbacks(cpu, st, target);
1103 cpus_write_unlock();
1107 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1111 if (!cpu_possible(cpu)) {
1112 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1114 #if defined(CONFIG_IA64)
1115 pr_err("please check additional_cpus= boot parameter\n");
1120 err = try_online_node(cpu_to_node(cpu));
1124 cpu_maps_update_begin();
1126 if (cpu_hotplug_disabled) {
1130 if (!cpu_smt_allowed(cpu)) {
1135 err = _cpu_up(cpu, 0, target);
1137 cpu_maps_update_done();
1141 int cpu_up(unsigned int cpu)
1143 return do_cpu_up(cpu, CPUHP_ONLINE);
1145 EXPORT_SYMBOL_GPL(cpu_up);
1147 #ifdef CONFIG_PM_SLEEP_SMP
1148 static cpumask_var_t frozen_cpus;
1150 int freeze_secondary_cpus(int primary)
1154 cpu_maps_update_begin();
1155 if (!cpu_online(primary))
1156 primary = cpumask_first(cpu_online_mask);
1158 * We take down all of the non-boot CPUs in one shot to avoid races
1159 * with the userspace trying to use the CPU hotplug at the same time
1161 cpumask_clear(frozen_cpus);
1163 pr_info("Disabling non-boot CPUs ...\n");
1164 for_each_online_cpu(cpu) {
1167 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1168 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1169 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1171 cpumask_set_cpu(cpu, frozen_cpus);
1173 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1179 BUG_ON(num_online_cpus() > 1);
1181 pr_err("Non-boot CPUs are not disabled\n");
1184 * Make sure the CPUs won't be enabled by someone else. We need to do
1185 * this even in case of failure as all disable_nonboot_cpus() users are
1186 * supposed to do enable_nonboot_cpus() on the failure path.
1188 cpu_hotplug_disabled++;
1190 cpu_maps_update_done();
1194 void __weak arch_enable_nonboot_cpus_begin(void)
1198 void __weak arch_enable_nonboot_cpus_end(void)
1202 void enable_nonboot_cpus(void)
1206 /* Allow everyone to use the CPU hotplug again */
1207 cpu_maps_update_begin();
1208 __cpu_hotplug_enable();
1209 if (cpumask_empty(frozen_cpus))
1212 pr_info("Enabling non-boot CPUs ...\n");
1214 arch_enable_nonboot_cpus_begin();
1216 for_each_cpu(cpu, frozen_cpus) {
1217 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1218 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1219 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1221 pr_info("CPU%d is up\n", cpu);
1224 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1227 arch_enable_nonboot_cpus_end();
1229 cpumask_clear(frozen_cpus);
1231 cpu_maps_update_done();
1234 static int __init alloc_frozen_cpus(void)
1236 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1240 core_initcall(alloc_frozen_cpus);
1243 * When callbacks for CPU hotplug notifications are being executed, we must
1244 * ensure that the state of the system with respect to the tasks being frozen
1245 * or not, as reported by the notification, remains unchanged *throughout the
1246 * duration* of the execution of the callbacks.
1247 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1249 * This synchronization is implemented by mutually excluding regular CPU
1250 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1251 * Hibernate notifications.
1254 cpu_hotplug_pm_callback(struct notifier_block *nb,
1255 unsigned long action, void *ptr)
1259 case PM_SUSPEND_PREPARE:
1260 case PM_HIBERNATION_PREPARE:
1261 cpu_hotplug_disable();
1264 case PM_POST_SUSPEND:
1265 case PM_POST_HIBERNATION:
1266 cpu_hotplug_enable();
1277 static int __init cpu_hotplug_pm_sync_init(void)
1280 * cpu_hotplug_pm_callback has higher priority than x86
1281 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1282 * to disable cpu hotplug to avoid cpu hotplug race.
1284 pm_notifier(cpu_hotplug_pm_callback, 0);
1287 core_initcall(cpu_hotplug_pm_sync_init);
1289 #endif /* CONFIG_PM_SLEEP_SMP */
1293 #endif /* CONFIG_SMP */
1295 /* Boot processor state steps */
1296 static struct cpuhp_step cpuhp_bp_states[] = {
1299 .startup.single = NULL,
1300 .teardown.single = NULL,
1303 [CPUHP_CREATE_THREADS]= {
1304 .name = "threads:prepare",
1305 .startup.single = smpboot_create_threads,
1306 .teardown.single = NULL,
1309 [CPUHP_PERF_PREPARE] = {
1310 .name = "perf:prepare",
1311 .startup.single = perf_event_init_cpu,
1312 .teardown.single = perf_event_exit_cpu,
1314 [CPUHP_WORKQUEUE_PREP] = {
1315 .name = "workqueue:prepare",
1316 .startup.single = workqueue_prepare_cpu,
1317 .teardown.single = NULL,
1319 [CPUHP_HRTIMERS_PREPARE] = {
1320 .name = "hrtimers:prepare",
1321 .startup.single = hrtimers_prepare_cpu,
1322 .teardown.single = hrtimers_dead_cpu,
1324 [CPUHP_SMPCFD_PREPARE] = {
1325 .name = "smpcfd:prepare",
1326 .startup.single = smpcfd_prepare_cpu,
1327 .teardown.single = smpcfd_dead_cpu,
1329 [CPUHP_RELAY_PREPARE] = {
1330 .name = "relay:prepare",
1331 .startup.single = relay_prepare_cpu,
1332 .teardown.single = NULL,
1334 [CPUHP_SLAB_PREPARE] = {
1335 .name = "slab:prepare",
1336 .startup.single = slab_prepare_cpu,
1337 .teardown.single = slab_dead_cpu,
1339 [CPUHP_RCUTREE_PREP] = {
1340 .name = "RCU/tree:prepare",
1341 .startup.single = rcutree_prepare_cpu,
1342 .teardown.single = rcutree_dead_cpu,
1345 * On the tear-down path, timers_dead_cpu() must be invoked
1346 * before blk_mq_queue_reinit_notify() from notify_dead(),
1347 * otherwise a RCU stall occurs.
1349 [CPUHP_TIMERS_PREPARE] = {
1350 .name = "timers:dead",
1351 .startup.single = timers_prepare_cpu,
1352 .teardown.single = timers_dead_cpu,
1354 /* Kicks the plugged cpu into life */
1355 [CPUHP_BRINGUP_CPU] = {
1356 .name = "cpu:bringup",
1357 .startup.single = bringup_cpu,
1358 .teardown.single = NULL,
1362 * Handled on controll processor until the plugged processor manages
1365 [CPUHP_TEARDOWN_CPU] = {
1366 .name = "cpu:teardown",
1367 .startup.single = NULL,
1368 .teardown.single = takedown_cpu,
1372 [CPUHP_BRINGUP_CPU] = { },
1376 /* Application processor state steps */
1377 static struct cpuhp_step cpuhp_ap_states[] = {
1379 /* Final state before CPU kills itself */
1380 [CPUHP_AP_IDLE_DEAD] = {
1381 .name = "idle:dead",
1384 * Last state before CPU enters the idle loop to die. Transient state
1385 * for synchronization.
1387 [CPUHP_AP_OFFLINE] = {
1388 .name = "ap:offline",
1391 /* First state is scheduler control. Interrupts are disabled */
1392 [CPUHP_AP_SCHED_STARTING] = {
1393 .name = "sched:starting",
1394 .startup.single = sched_cpu_starting,
1395 .teardown.single = sched_cpu_dying,
1397 [CPUHP_AP_RCUTREE_DYING] = {
1398 .name = "RCU/tree:dying",
1399 .startup.single = NULL,
1400 .teardown.single = rcutree_dying_cpu,
1402 [CPUHP_AP_SMPCFD_DYING] = {
1403 .name = "smpcfd:dying",
1404 .startup.single = NULL,
1405 .teardown.single = smpcfd_dying_cpu,
1407 /* Entry state on starting. Interrupts enabled from here on. Transient
1408 * state for synchronsization */
1409 [CPUHP_AP_ONLINE] = {
1410 .name = "ap:online",
1412 /* Handle smpboot threads park/unpark */
1413 [CPUHP_AP_SMPBOOT_THREADS] = {
1414 .name = "smpboot/threads:online",
1415 .startup.single = smpboot_unpark_threads,
1416 .teardown.single = smpboot_park_threads,
1418 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1419 .name = "irq/affinity:online",
1420 .startup.single = irq_affinity_online_cpu,
1421 .teardown.single = NULL,
1423 [CPUHP_AP_PERF_ONLINE] = {
1424 .name = "perf:online",
1425 .startup.single = perf_event_init_cpu,
1426 .teardown.single = perf_event_exit_cpu,
1428 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1429 .name = "workqueue:online",
1430 .startup.single = workqueue_online_cpu,
1431 .teardown.single = workqueue_offline_cpu,
1433 [CPUHP_AP_RCUTREE_ONLINE] = {
1434 .name = "RCU/tree:online",
1435 .startup.single = rcutree_online_cpu,
1436 .teardown.single = rcutree_offline_cpu,
1440 * The dynamically registered state space is here
1444 /* Last state is scheduler control setting the cpu active */
1445 [CPUHP_AP_ACTIVE] = {
1446 .name = "sched:active",
1447 .startup.single = sched_cpu_activate,
1448 .teardown.single = sched_cpu_deactivate,
1452 /* CPU is fully up and running. */
1455 .startup.single = NULL,
1456 .teardown.single = NULL,
1460 /* Sanity check for callbacks */
1461 static int cpuhp_cb_check(enum cpuhp_state state)
1463 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1469 * Returns a free for dynamic slot assignment of the Online state. The states
1470 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1471 * by having no name assigned.
1473 static int cpuhp_reserve_state(enum cpuhp_state state)
1475 enum cpuhp_state i, end;
1476 struct cpuhp_step *step;
1479 case CPUHP_AP_ONLINE_DYN:
1480 step = cpuhp_ap_states + CPUHP_AP_ONLINE_DYN;
1481 end = CPUHP_AP_ONLINE_DYN_END;
1483 case CPUHP_BP_PREPARE_DYN:
1484 step = cpuhp_bp_states + CPUHP_BP_PREPARE_DYN;
1485 end = CPUHP_BP_PREPARE_DYN_END;
1491 for (i = state; i <= end; i++, step++) {
1495 WARN(1, "No more dynamic states available for CPU hotplug\n");
1499 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1500 int (*startup)(unsigned int cpu),
1501 int (*teardown)(unsigned int cpu),
1502 bool multi_instance)
1504 /* (Un)Install the callbacks for further cpu hotplug operations */
1505 struct cpuhp_step *sp;
1509 * If name is NULL, then the state gets removed.
1511 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1512 * the first allocation from these dynamic ranges, so the removal
1513 * would trigger a new allocation and clear the wrong (already
1514 * empty) state, leaving the callbacks of the to be cleared state
1515 * dangling, which causes wreckage on the next hotplug operation.
1517 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1518 state == CPUHP_BP_PREPARE_DYN)) {
1519 ret = cpuhp_reserve_state(state);
1524 sp = cpuhp_get_step(state);
1525 if (name && sp->name)
1528 sp->startup.single = startup;
1529 sp->teardown.single = teardown;
1531 sp->multi_instance = multi_instance;
1532 INIT_HLIST_HEAD(&sp->list);
1536 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1538 return cpuhp_get_step(state)->teardown.single;
1542 * Call the startup/teardown function for a step either on the AP or
1543 * on the current CPU.
1545 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1546 struct hlist_node *node)
1548 struct cpuhp_step *sp = cpuhp_get_step(state);
1552 * If there's nothing to do, we done.
1553 * Relies on the union for multi_instance.
1555 if ((bringup && !sp->startup.single) ||
1556 (!bringup && !sp->teardown.single))
1559 * The non AP bound callbacks can fail on bringup. On teardown
1560 * e.g. module removal we crash for now.
1563 if (cpuhp_is_ap_state(state))
1564 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1566 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1568 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1570 BUG_ON(ret && !bringup);
1575 * Called from __cpuhp_setup_state on a recoverable failure.
1577 * Note: The teardown callbacks for rollback are not allowed to fail!
1579 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1580 struct hlist_node *node)
1584 /* Roll back the already executed steps on the other cpus */
1585 for_each_present_cpu(cpu) {
1586 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1587 int cpustate = st->state;
1589 if (cpu >= failedcpu)
1592 /* Did we invoke the startup call on that cpu ? */
1593 if (cpustate >= state)
1594 cpuhp_issue_call(cpu, state, false, node);
1598 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1599 struct hlist_node *node,
1602 struct cpuhp_step *sp;
1606 lockdep_assert_cpus_held();
1608 sp = cpuhp_get_step(state);
1609 if (sp->multi_instance == false)
1612 mutex_lock(&cpuhp_state_mutex);
1614 if (!invoke || !sp->startup.multi)
1618 * Try to call the startup callback for each present cpu
1619 * depending on the hotplug state of the cpu.
1621 for_each_present_cpu(cpu) {
1622 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1623 int cpustate = st->state;
1625 if (cpustate < state)
1628 ret = cpuhp_issue_call(cpu, state, true, node);
1630 if (sp->teardown.multi)
1631 cpuhp_rollback_install(cpu, state, node);
1637 hlist_add_head(node, &sp->list);
1639 mutex_unlock(&cpuhp_state_mutex);
1643 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1649 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1653 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1656 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1657 * @state: The state to setup
1658 * @invoke: If true, the startup function is invoked for cpus where
1659 * cpu state >= @state
1660 * @startup: startup callback function
1661 * @teardown: teardown callback function
1662 * @multi_instance: State is set up for multiple instances which get
1665 * The caller needs to hold cpus read locked while calling this function.
1668 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1669 * 0 for all other states
1670 * On failure: proper (negative) error code
1672 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1673 const char *name, bool invoke,
1674 int (*startup)(unsigned int cpu),
1675 int (*teardown)(unsigned int cpu),
1676 bool multi_instance)
1681 lockdep_assert_cpus_held();
1683 if (cpuhp_cb_check(state) || !name)
1686 mutex_lock(&cpuhp_state_mutex);
1688 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1691 dynstate = state == CPUHP_AP_ONLINE_DYN;
1692 if (ret > 0 && dynstate) {
1697 if (ret || !invoke || !startup)
1701 * Try to call the startup callback for each present cpu
1702 * depending on the hotplug state of the cpu.
1704 for_each_present_cpu(cpu) {
1705 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1706 int cpustate = st->state;
1708 if (cpustate < state)
1711 ret = cpuhp_issue_call(cpu, state, true, NULL);
1714 cpuhp_rollback_install(cpu, state, NULL);
1715 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1720 mutex_unlock(&cpuhp_state_mutex);
1722 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1723 * dynamically allocated state in case of success.
1725 if (!ret && dynstate)
1729 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1731 int __cpuhp_setup_state(enum cpuhp_state state,
1732 const char *name, bool invoke,
1733 int (*startup)(unsigned int cpu),
1734 int (*teardown)(unsigned int cpu),
1735 bool multi_instance)
1740 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1741 teardown, multi_instance);
1745 EXPORT_SYMBOL(__cpuhp_setup_state);
1747 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1748 struct hlist_node *node, bool invoke)
1750 struct cpuhp_step *sp = cpuhp_get_step(state);
1753 BUG_ON(cpuhp_cb_check(state));
1755 if (!sp->multi_instance)
1759 mutex_lock(&cpuhp_state_mutex);
1761 if (!invoke || !cpuhp_get_teardown_cb(state))
1764 * Call the teardown callback for each present cpu depending
1765 * on the hotplug state of the cpu. This function is not
1766 * allowed to fail currently!
1768 for_each_present_cpu(cpu) {
1769 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1770 int cpustate = st->state;
1772 if (cpustate >= state)
1773 cpuhp_issue_call(cpu, state, false, node);
1778 mutex_unlock(&cpuhp_state_mutex);
1783 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1786 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1787 * @state: The state to remove
1788 * @invoke: If true, the teardown function is invoked for cpus where
1789 * cpu state >= @state
1791 * The caller needs to hold cpus read locked while calling this function.
1792 * The teardown callback is currently not allowed to fail. Think
1793 * about module removal!
1795 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1797 struct cpuhp_step *sp = cpuhp_get_step(state);
1800 BUG_ON(cpuhp_cb_check(state));
1802 lockdep_assert_cpus_held();
1804 mutex_lock(&cpuhp_state_mutex);
1805 if (sp->multi_instance) {
1806 WARN(!hlist_empty(&sp->list),
1807 "Error: Removing state %d which has instances left.\n",
1812 if (!invoke || !cpuhp_get_teardown_cb(state))
1816 * Call the teardown callback for each present cpu depending
1817 * on the hotplug state of the cpu. This function is not
1818 * allowed to fail currently!
1820 for_each_present_cpu(cpu) {
1821 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1822 int cpustate = st->state;
1824 if (cpustate >= state)
1825 cpuhp_issue_call(cpu, state, false, NULL);
1828 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1829 mutex_unlock(&cpuhp_state_mutex);
1831 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1833 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1836 __cpuhp_remove_state_cpuslocked(state, invoke);
1839 EXPORT_SYMBOL(__cpuhp_remove_state);
1841 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1842 static ssize_t show_cpuhp_state(struct device *dev,
1843 struct device_attribute *attr, char *buf)
1845 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1847 return sprintf(buf, "%d\n", st->state);
1849 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1851 static ssize_t write_cpuhp_target(struct device *dev,
1852 struct device_attribute *attr,
1853 const char *buf, size_t count)
1855 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1856 struct cpuhp_step *sp;
1859 ret = kstrtoint(buf, 10, &target);
1863 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1864 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1867 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1871 ret = lock_device_hotplug_sysfs();
1875 mutex_lock(&cpuhp_state_mutex);
1876 sp = cpuhp_get_step(target);
1877 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1878 mutex_unlock(&cpuhp_state_mutex);
1882 if (st->state < target)
1883 ret = do_cpu_up(dev->id, target);
1885 ret = do_cpu_down(dev->id, target);
1887 unlock_device_hotplug();
1888 return ret ? ret : count;
1891 static ssize_t show_cpuhp_target(struct device *dev,
1892 struct device_attribute *attr, char *buf)
1894 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1896 return sprintf(buf, "%d\n", st->target);
1898 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1901 static ssize_t write_cpuhp_fail(struct device *dev,
1902 struct device_attribute *attr,
1903 const char *buf, size_t count)
1905 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1906 struct cpuhp_step *sp;
1909 ret = kstrtoint(buf, 10, &fail);
1914 * Cannot fail STARTING/DYING callbacks.
1916 if (cpuhp_is_atomic_state(fail))
1920 * Cannot fail anything that doesn't have callbacks.
1922 mutex_lock(&cpuhp_state_mutex);
1923 sp = cpuhp_get_step(fail);
1924 if (!sp->startup.single && !sp->teardown.single)
1926 mutex_unlock(&cpuhp_state_mutex);
1935 static ssize_t show_cpuhp_fail(struct device *dev,
1936 struct device_attribute *attr, char *buf)
1938 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1940 return sprintf(buf, "%d\n", st->fail);
1943 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1945 static struct attribute *cpuhp_cpu_attrs[] = {
1946 &dev_attr_state.attr,
1947 &dev_attr_target.attr,
1948 &dev_attr_fail.attr,
1952 static const struct attribute_group cpuhp_cpu_attr_group = {
1953 .attrs = cpuhp_cpu_attrs,
1958 static ssize_t show_cpuhp_states(struct device *dev,
1959 struct device_attribute *attr, char *buf)
1961 ssize_t cur, res = 0;
1964 mutex_lock(&cpuhp_state_mutex);
1965 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1966 struct cpuhp_step *sp = cpuhp_get_step(i);
1969 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1974 mutex_unlock(&cpuhp_state_mutex);
1977 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1979 static struct attribute *cpuhp_cpu_root_attrs[] = {
1980 &dev_attr_states.attr,
1984 static const struct attribute_group cpuhp_cpu_root_attr_group = {
1985 .attrs = cpuhp_cpu_root_attrs,
1990 #ifdef CONFIG_HOTPLUG_SMT
1992 static const char *smt_states[] = {
1993 [CPU_SMT_ENABLED] = "on",
1994 [CPU_SMT_DISABLED] = "off",
1995 [CPU_SMT_FORCE_DISABLED] = "forceoff",
1996 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2000 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2002 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2005 static void cpuhp_offline_cpu_device(unsigned int cpu)
2007 struct device *dev = get_cpu_device(cpu);
2009 dev->offline = true;
2010 /* Tell user space about the state change */
2011 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2014 static void cpuhp_online_cpu_device(unsigned int cpu)
2016 struct device *dev = get_cpu_device(cpu);
2018 dev->offline = false;
2019 /* Tell user space about the state change */
2020 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2023 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2027 cpu_maps_update_begin();
2028 for_each_online_cpu(cpu) {
2029 if (topology_is_primary_thread(cpu))
2031 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2035 * As this needs to hold the cpu maps lock it's impossible
2036 * to call device_offline() because that ends up calling
2037 * cpu_down() which takes cpu maps lock. cpu maps lock
2038 * needs to be held as this might race against in kernel
2039 * abusers of the hotplug machinery (thermal management).
2041 * So nothing would update device:offline state. That would
2042 * leave the sysfs entry stale and prevent onlining after
2043 * smt control has been changed to 'off' again. This is
2044 * called under the sysfs hotplug lock, so it is properly
2045 * serialized against the regular offline usage.
2047 cpuhp_offline_cpu_device(cpu);
2050 cpu_smt_control = ctrlval;
2051 cpu_maps_update_done();
2055 static int cpuhp_smt_enable(void)
2059 cpu_maps_update_begin();
2060 cpu_smt_control = CPU_SMT_ENABLED;
2061 for_each_present_cpu(cpu) {
2062 /* Skip online CPUs and CPUs on offline nodes */
2063 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2065 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2068 /* See comment in cpuhp_smt_disable() */
2069 cpuhp_online_cpu_device(cpu);
2071 cpu_maps_update_done();
2076 store_smt_control(struct device *dev, struct device_attribute *attr,
2077 const char *buf, size_t count)
2081 if (sysfs_streq(buf, "on"))
2082 ctrlval = CPU_SMT_ENABLED;
2083 else if (sysfs_streq(buf, "off"))
2084 ctrlval = CPU_SMT_DISABLED;
2085 else if (sysfs_streq(buf, "forceoff"))
2086 ctrlval = CPU_SMT_FORCE_DISABLED;
2090 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2093 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2096 ret = lock_device_hotplug_sysfs();
2100 if (ctrlval != cpu_smt_control) {
2102 case CPU_SMT_ENABLED:
2103 ret = cpuhp_smt_enable();
2105 case CPU_SMT_DISABLED:
2106 case CPU_SMT_FORCE_DISABLED:
2107 ret = cpuhp_smt_disable(ctrlval);
2112 unlock_device_hotplug();
2113 return ret ? ret : count;
2115 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2118 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2120 bool active = topology_max_smt_threads() > 1;
2122 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2124 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2126 static struct attribute *cpuhp_smt_attrs[] = {
2127 &dev_attr_control.attr,
2128 &dev_attr_active.attr,
2132 static const struct attribute_group cpuhp_smt_attr_group = {
2133 .attrs = cpuhp_smt_attrs,
2138 static int __init cpu_smt_state_init(void)
2141 * If SMT was disabled by BIOS, detect it here, after the CPUs have
2142 * been brought online. This ensures the smt/l1tf sysfs entries are
2143 * consistent with reality. Note this may overwrite cpu_smt_control's
2146 if (topology_max_smt_threads() == 1)
2147 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
2149 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2150 &cpuhp_smt_attr_group);
2154 static inline int cpu_smt_state_init(void) { return 0; }
2157 static int __init cpuhp_sysfs_init(void)
2161 ret = cpu_smt_state_init();
2165 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2166 &cpuhp_cpu_root_attr_group);
2170 for_each_possible_cpu(cpu) {
2171 struct device *dev = get_cpu_device(cpu);
2175 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2181 device_initcall(cpuhp_sysfs_init);
2185 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2186 * represents all NR_CPUS bits binary values of 1<<nr.
2188 * It is used by cpumask_of() to get a constant address to a CPU
2189 * mask value that has a single bit set only.
2192 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2193 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2194 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2195 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2196 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2198 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2200 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2201 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2202 #if BITS_PER_LONG > 32
2203 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2204 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2207 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2209 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2210 EXPORT_SYMBOL(cpu_all_bits);
2212 #ifdef CONFIG_INIT_ALL_POSSIBLE
2213 struct cpumask __cpu_possible_mask __read_mostly
2216 struct cpumask __cpu_possible_mask __read_mostly;
2218 EXPORT_SYMBOL(__cpu_possible_mask);
2220 struct cpumask __cpu_online_mask __read_mostly;
2221 EXPORT_SYMBOL(__cpu_online_mask);
2223 struct cpumask __cpu_present_mask __read_mostly;
2224 EXPORT_SYMBOL(__cpu_present_mask);
2226 struct cpumask __cpu_active_mask __read_mostly;
2227 EXPORT_SYMBOL(__cpu_active_mask);
2229 void init_cpu_present(const struct cpumask *src)
2231 cpumask_copy(&__cpu_present_mask, src);
2234 void init_cpu_possible(const struct cpumask *src)
2236 cpumask_copy(&__cpu_possible_mask, src);
2239 void init_cpu_online(const struct cpumask *src)
2241 cpumask_copy(&__cpu_online_mask, src);
2245 * Activate the first processor.
2247 void __init boot_cpu_init(void)
2249 int cpu = smp_processor_id();
2251 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2252 set_cpu_online(cpu, true);
2253 set_cpu_active(cpu, true);
2254 set_cpu_present(cpu, true);
2255 set_cpu_possible(cpu, true);
2258 __boot_cpu_id = cpu;
2263 * Must be called _AFTER_ setting up the per_cpu areas
2265 void __init boot_cpu_hotplug_init(void)
2267 this_cpu_write(cpuhp_state.booted_once, true);
2268 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);