2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/proc_fs.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/hotplug.h>
12 #include <linux/sched/task.h>
13 #include <linux/sched/smt.h>
14 #include <linux/unistd.h>
15 #include <linux/cpu.h>
16 #include <linux/oom.h>
17 #include <linux/rcupdate.h>
18 #include <linux/export.h>
19 #include <linux/bug.h>
20 #include <linux/kthread.h>
21 #include <linux/stop_machine.h>
22 #include <linux/mutex.h>
23 #include <linux/gfp.h>
24 #include <linux/suspend.h>
25 #include <linux/lockdep.h>
26 #include <linux/tick.h>
27 #include <linux/irq.h>
28 #include <linux/nmi.h>
29 #include <linux/smpboot.h>
30 #include <linux/relay.h>
31 #include <linux/slab.h>
32 #include <linux/percpu-rwsem.h>
34 #include <trace/events/power.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/cpuhp.h>
41 * cpuhp_cpu_state - Per cpu hotplug state storage
42 * @state: The current cpu state
43 * @target: The target state
44 * @thread: Pointer to the hotplug thread
45 * @should_run: Thread should execute
46 * @rollback: Perform a rollback
47 * @single: Single callback invocation
48 * @bringup: Single callback bringup or teardown selector
49 * @cb_state: The state for a single callback (install/uninstall)
50 * @result: Result of the operation
51 * @done_up: Signal completion to the issuer of the task for cpu-up
52 * @done_down: Signal completion to the issuer of the task for cpu-down
54 struct cpuhp_cpu_state {
55 enum cpuhp_state state;
56 enum cpuhp_state target;
57 enum cpuhp_state fail;
59 struct task_struct *thread;
65 struct hlist_node *node;
66 struct hlist_node *last;
67 enum cpuhp_state cb_state;
69 struct completion done_up;
70 struct completion done_down;
74 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
75 .fail = CPUHP_INVALID,
78 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
79 static struct lockdep_map cpuhp_state_up_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
81 static struct lockdep_map cpuhp_state_down_map =
82 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
85 static inline void cpuhp_lock_acquire(bool bringup)
87 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
90 static inline void cpuhp_lock_release(bool bringup)
92 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
96 static inline void cpuhp_lock_acquire(bool bringup) { }
97 static inline void cpuhp_lock_release(bool bringup) { }
102 * cpuhp_step - Hotplug state machine step
103 * @name: Name of the step
104 * @startup: Startup function of the step
105 * @teardown: Teardown function of the step
106 * @cant_stop: Bringup/teardown can't be stopped at this step
111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
120 struct hlist_head list;
125 static DEFINE_MUTEX(cpuhp_state_mutex);
126 static struct cpuhp_step cpuhp_hp_states[];
128 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
130 return cpuhp_hp_states + state;
134 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
135 * @cpu: The cpu for which the callback should be invoked
136 * @state: The state to do callbacks for
137 * @bringup: True if the bringup callback should be invoked
138 * @node: For multi-instance, do a single entry callback for install/remove
139 * @lastp: For multi-instance rollback, remember how far we got
141 * Called from cpu hotplug and from the state register machinery.
143 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
144 bool bringup, struct hlist_node *node,
145 struct hlist_node **lastp)
147 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
148 struct cpuhp_step *step = cpuhp_get_step(state);
149 int (*cbm)(unsigned int cpu, struct hlist_node *node);
150 int (*cb)(unsigned int cpu);
153 if (st->fail == state) {
154 st->fail = CPUHP_INVALID;
156 if (!(bringup ? step->startup.single : step->teardown.single))
162 if (!step->multi_instance) {
163 WARN_ON_ONCE(lastp && *lastp);
164 cb = bringup ? step->startup.single : step->teardown.single;
167 trace_cpuhp_enter(cpu, st->target, state, cb);
169 trace_cpuhp_exit(cpu, st->state, state, ret);
172 cbm = bringup ? step->startup.multi : step->teardown.multi;
176 /* Single invocation for instance add/remove */
178 WARN_ON_ONCE(lastp && *lastp);
179 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
180 ret = cbm(cpu, node);
181 trace_cpuhp_exit(cpu, st->state, state, ret);
185 /* State transition. Invoke on all instances */
187 hlist_for_each(node, &step->list) {
188 if (lastp && node == *lastp)
191 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
192 ret = cbm(cpu, node);
193 trace_cpuhp_exit(cpu, st->state, state, ret);
207 /* Rollback the instances if one failed */
208 cbm = !bringup ? step->startup.multi : step->teardown.multi;
212 hlist_for_each(node, &step->list) {
216 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
217 ret = cbm(cpu, node);
218 trace_cpuhp_exit(cpu, st->state, state, ret);
220 * Rollback must not fail,
228 static bool cpuhp_is_ap_state(enum cpuhp_state state)
231 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
232 * purposes as that state is handled explicitly in cpu_down.
234 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
237 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
239 struct completion *done = bringup ? &st->done_up : &st->done_down;
240 wait_for_completion(done);
243 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
250 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
252 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
254 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
257 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
258 static DEFINE_MUTEX(cpu_add_remove_lock);
259 bool cpuhp_tasks_frozen;
260 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
263 * The following two APIs (cpu_maps_update_begin/done) must be used when
264 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
266 void cpu_maps_update_begin(void)
268 mutex_lock(&cpu_add_remove_lock);
271 void cpu_maps_update_done(void)
273 mutex_unlock(&cpu_add_remove_lock);
277 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
278 * Should always be manipulated under cpu_add_remove_lock
280 static int cpu_hotplug_disabled;
282 #ifdef CONFIG_HOTPLUG_CPU
284 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
286 void cpus_read_lock(void)
288 percpu_down_read(&cpu_hotplug_lock);
290 EXPORT_SYMBOL_GPL(cpus_read_lock);
292 int cpus_read_trylock(void)
294 return percpu_down_read_trylock(&cpu_hotplug_lock);
296 EXPORT_SYMBOL_GPL(cpus_read_trylock);
298 void cpus_read_unlock(void)
300 percpu_up_read(&cpu_hotplug_lock);
302 EXPORT_SYMBOL_GPL(cpus_read_unlock);
304 void cpus_write_lock(void)
306 percpu_down_write(&cpu_hotplug_lock);
309 void cpus_write_unlock(void)
311 percpu_up_write(&cpu_hotplug_lock);
314 void lockdep_assert_cpus_held(void)
316 percpu_rwsem_assert_held(&cpu_hotplug_lock);
319 static void lockdep_acquire_cpus_lock(void)
321 rwsem_acquire(&cpu_hotplug_lock.rw_sem.dep_map, 0, 0, _THIS_IP_);
324 static void lockdep_release_cpus_lock(void)
326 rwsem_release(&cpu_hotplug_lock.rw_sem.dep_map, 1, _THIS_IP_);
330 * Wait for currently running CPU hotplug operations to complete (if any) and
331 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
332 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
333 * hotplug path before performing hotplug operations. So acquiring that lock
334 * guarantees mutual exclusion from any currently running hotplug operations.
336 void cpu_hotplug_disable(void)
338 cpu_maps_update_begin();
339 cpu_hotplug_disabled++;
340 cpu_maps_update_done();
342 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
344 static void __cpu_hotplug_enable(void)
346 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
348 cpu_hotplug_disabled--;
351 void cpu_hotplug_enable(void)
353 cpu_maps_update_begin();
354 __cpu_hotplug_enable();
355 cpu_maps_update_done();
357 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
361 static void lockdep_acquire_cpus_lock(void)
365 static void lockdep_release_cpus_lock(void)
369 #endif /* CONFIG_HOTPLUG_CPU */
372 * Architectures that need SMT-specific errata handling during SMT hotplug
373 * should override this.
375 void __weak arch_smt_update(void) { }
377 #ifdef CONFIG_HOTPLUG_SMT
378 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
380 void __init cpu_smt_disable(bool force)
382 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
383 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
387 pr_info("SMT: Force disabled\n");
388 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
390 pr_info("SMT: disabled\n");
391 cpu_smt_control = CPU_SMT_DISABLED;
396 * The decision whether SMT is supported can only be done after the full
397 * CPU identification. Called from architecture code.
399 void __init cpu_smt_check_topology(void)
401 if (!topology_smt_supported())
402 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
405 static int __init smt_cmdline_disable(char *str)
407 cpu_smt_disable(str && !strcmp(str, "force"));
410 early_param("nosmt", smt_cmdline_disable);
412 static inline bool cpu_smt_allowed(unsigned int cpu)
414 if (cpu_smt_control == CPU_SMT_ENABLED)
417 if (topology_is_primary_thread(cpu))
421 * On x86 it's required to boot all logical CPUs at least once so
422 * that the init code can get a chance to set CR4.MCE on each
423 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
424 * core will shutdown the machine.
426 return !per_cpu(cpuhp_state, cpu).booted_once;
429 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
432 static inline enum cpuhp_state
433 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
435 enum cpuhp_state prev_state = st->state;
437 st->rollback = false;
442 st->bringup = st->state < target;
448 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
453 * If we have st->last we need to undo partial multi_instance of this
454 * state first. Otherwise start undo at the previous state.
463 st->target = prev_state;
464 st->bringup = !st->bringup;
467 /* Regular hotplug invocation of the AP hotplug thread */
468 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
470 if (!st->single && st->state == st->target)
475 * Make sure the above stores are visible before should_run becomes
476 * true. Paired with the mb() above in cpuhp_thread_fun()
479 st->should_run = true;
480 wake_up_process(st->thread);
481 wait_for_ap_thread(st, st->bringup);
484 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
486 enum cpuhp_state prev_state;
489 prev_state = cpuhp_set_state(st, target);
491 if ((ret = st->result)) {
492 cpuhp_reset_state(st, prev_state);
499 static int bringup_wait_for_ap(unsigned int cpu)
501 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
503 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
504 wait_for_ap_thread(st, true);
505 if (WARN_ON_ONCE((!cpu_online(cpu))))
508 /* Unpark the stopper thread and the hotplug thread of the target cpu */
509 stop_machine_unpark(cpu);
510 kthread_unpark(st->thread);
513 * SMT soft disabling on X86 requires to bring the CPU out of the
514 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
515 * CPU marked itself as booted_once in cpu_notify_starting() so the
516 * cpu_smt_allowed() check will now return false if this is not the
519 if (!cpu_smt_allowed(cpu))
522 if (st->target <= CPUHP_AP_ONLINE_IDLE)
525 return cpuhp_kick_ap(st, st->target);
528 static int bringup_cpu(unsigned int cpu)
530 struct task_struct *idle = idle_thread_get(cpu);
534 * Some architectures have to walk the irq descriptors to
535 * setup the vector space for the cpu which comes online.
536 * Prevent irq alloc/free across the bringup.
540 /* Arch-specific enabling code. */
541 ret = __cpu_up(cpu, idle);
545 return bringup_wait_for_ap(cpu);
549 * Hotplug state machine related functions
552 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
554 for (st->state--; st->state > st->target; st->state--)
555 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
558 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
559 enum cpuhp_state target)
561 enum cpuhp_state prev_state = st->state;
564 while (st->state < target) {
566 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
568 st->target = prev_state;
569 undo_cpu_up(cpu, st);
577 * The cpu hotplug threads manage the bringup and teardown of the cpus
579 static void cpuhp_create(unsigned int cpu)
581 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
583 init_completion(&st->done_up);
584 init_completion(&st->done_down);
587 static int cpuhp_should_run(unsigned int cpu)
589 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
591 return st->should_run;
595 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
596 * callbacks when a state gets [un]installed at runtime.
598 * Each invocation of this function by the smpboot thread does a single AP
601 * It has 3 modes of operation:
602 * - single: runs st->cb_state
603 * - up: runs ++st->state, while st->state < st->target
604 * - down: runs st->state--, while st->state > st->target
606 * When complete or on error, should_run is cleared and the completion is fired.
608 static void cpuhp_thread_fun(unsigned int cpu)
610 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
611 bool bringup = st->bringup;
612 enum cpuhp_state state;
614 if (WARN_ON_ONCE(!st->should_run))
618 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
619 * that if we see ->should_run we also see the rest of the state.
624 * The BP holds the hotplug lock, but we're now running on the AP,
625 * ensure that anybody asserting the lock is held, will actually find
628 lockdep_acquire_cpus_lock();
629 cpuhp_lock_acquire(bringup);
632 state = st->cb_state;
633 st->should_run = false;
638 st->should_run = (st->state < st->target);
639 WARN_ON_ONCE(st->state > st->target);
643 st->should_run = (st->state > st->target);
644 WARN_ON_ONCE(st->state < st->target);
648 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
650 if (cpuhp_is_atomic_state(state)) {
652 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
656 * STARTING/DYING must not fail!
658 WARN_ON_ONCE(st->result);
660 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
665 * If we fail on a rollback, we're up a creek without no
666 * paddle, no way forward, no way back. We loose, thanks for
669 WARN_ON_ONCE(st->rollback);
670 st->should_run = false;
673 cpuhp_lock_release(bringup);
674 lockdep_release_cpus_lock();
677 complete_ap_thread(st, bringup);
680 /* Invoke a single callback on a remote cpu */
682 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
683 struct hlist_node *node)
685 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
688 if (!cpu_online(cpu))
691 cpuhp_lock_acquire(false);
692 cpuhp_lock_release(false);
694 cpuhp_lock_acquire(true);
695 cpuhp_lock_release(true);
698 * If we are up and running, use the hotplug thread. For early calls
699 * we invoke the thread function directly.
702 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
704 st->rollback = false;
708 st->bringup = bringup;
709 st->cb_state = state;
715 * If we failed and did a partial, do a rollback.
717 if ((ret = st->result) && st->last) {
719 st->bringup = !bringup;
725 * Clean up the leftovers so the next hotplug operation wont use stale
728 st->node = st->last = NULL;
732 static int cpuhp_kick_ap_work(unsigned int cpu)
734 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
735 enum cpuhp_state prev_state = st->state;
738 cpuhp_lock_acquire(false);
739 cpuhp_lock_release(false);
741 cpuhp_lock_acquire(true);
742 cpuhp_lock_release(true);
744 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
745 ret = cpuhp_kick_ap(st, st->target);
746 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
751 static struct smp_hotplug_thread cpuhp_threads = {
752 .store = &cpuhp_state.thread,
753 .create = &cpuhp_create,
754 .thread_should_run = cpuhp_should_run,
755 .thread_fn = cpuhp_thread_fun,
756 .thread_comm = "cpuhp/%u",
760 void __init cpuhp_threads_init(void)
762 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
763 kthread_unpark(this_cpu_read(cpuhp_state.thread));
766 #ifdef CONFIG_HOTPLUG_CPU
768 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
771 * This function walks all processes, finds a valid mm struct for each one and
772 * then clears a corresponding bit in mm's cpumask. While this all sounds
773 * trivial, there are various non-obvious corner cases, which this function
774 * tries to solve in a safe manner.
776 * Also note that the function uses a somewhat relaxed locking scheme, so it may
777 * be called only for an already offlined CPU.
779 void clear_tasks_mm_cpumask(int cpu)
781 struct task_struct *p;
784 * This function is called after the cpu is taken down and marked
785 * offline, so its not like new tasks will ever get this cpu set in
786 * their mm mask. -- Peter Zijlstra
787 * Thus, we may use rcu_read_lock() here, instead of grabbing
788 * full-fledged tasklist_lock.
790 WARN_ON(cpu_online(cpu));
792 for_each_process(p) {
793 struct task_struct *t;
796 * Main thread might exit, but other threads may still have
797 * a valid mm. Find one.
799 t = find_lock_task_mm(p);
802 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
808 /* Take this CPU down. */
809 static int take_cpu_down(void *_param)
811 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
812 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
813 int err, cpu = smp_processor_id();
816 /* Ensure this CPU doesn't handle any more interrupts. */
817 err = __cpu_disable();
822 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
823 * do this step again.
825 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
827 /* Invoke the former CPU_DYING callbacks */
828 for (; st->state > target; st->state--) {
829 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
831 * DYING must not fail!
836 /* Give up timekeeping duties */
837 tick_handover_do_timer();
838 /* Park the stopper thread */
839 stop_machine_park(cpu);
843 static int takedown_cpu(unsigned int cpu)
845 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
848 /* Park the smpboot threads */
849 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
852 * Prevent irq alloc/free while the dying cpu reorganizes the
853 * interrupt affinities.
858 * So now all preempt/rcu users must observe !cpu_active().
860 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
862 /* CPU refused to die */
864 /* Unpark the hotplug thread so we can rollback there */
865 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
868 BUG_ON(cpu_online(cpu));
871 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
872 * all runnable tasks from the CPU, there's only the idle task left now
873 * that the migration thread is done doing the stop_machine thing.
875 * Wait for the stop thread to go away.
877 wait_for_ap_thread(st, false);
878 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
880 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
883 hotplug_cpu__broadcast_tick_pull(cpu);
884 /* This actually kills the CPU. */
887 tick_cleanup_dead_cpu(cpu);
888 rcutree_migrate_callbacks(cpu);
892 static void cpuhp_complete_idle_dead(void *arg)
894 struct cpuhp_cpu_state *st = arg;
896 complete_ap_thread(st, false);
899 void cpuhp_report_idle_dead(void)
901 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
903 BUG_ON(st->state != CPUHP_AP_OFFLINE);
904 rcu_report_dead(smp_processor_id());
905 st->state = CPUHP_AP_IDLE_DEAD;
907 * We cannot call complete after rcu_report_dead() so we delegate it
910 smp_call_function_single(cpumask_first(cpu_online_mask),
911 cpuhp_complete_idle_dead, st, 0);
914 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
916 for (st->state++; st->state < st->target; st->state++)
917 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
920 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
921 enum cpuhp_state target)
923 enum cpuhp_state prev_state = st->state;
926 for (; st->state > target; st->state--) {
927 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
929 st->target = prev_state;
930 if (st->state < prev_state)
931 undo_cpu_down(cpu, st);
938 /* Requires cpu_add_remove_lock to be held */
939 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
940 enum cpuhp_state target)
942 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
943 int prev_state, ret = 0;
945 if (num_online_cpus() == 1)
948 if (!cpu_present(cpu))
953 cpuhp_tasks_frozen = tasks_frozen;
955 prev_state = cpuhp_set_state(st, target);
957 * If the current CPU state is in the range of the AP hotplug thread,
958 * then we need to kick the thread.
960 if (st->state > CPUHP_TEARDOWN_CPU) {
961 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
962 ret = cpuhp_kick_ap_work(cpu);
964 * The AP side has done the error rollback already. Just
965 * return the error code..
971 * We might have stopped still in the range of the AP hotplug
972 * thread. Nothing to do anymore.
974 if (st->state > CPUHP_TEARDOWN_CPU)
980 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
981 * to do the further cleanups.
983 ret = cpuhp_down_callbacks(cpu, st, target);
984 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
985 cpuhp_reset_state(st, prev_state);
992 * Do post unplug cleanup. This is still protected against
993 * concurrent CPU hotplug via cpu_add_remove_lock.
995 lockup_detector_cleanup();
1000 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1002 if (cpu_hotplug_disabled)
1004 return _cpu_down(cpu, 0, target);
1007 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
1011 cpu_maps_update_begin();
1012 err = cpu_down_maps_locked(cpu, target);
1013 cpu_maps_update_done();
1017 int cpu_down(unsigned int cpu)
1019 return do_cpu_down(cpu, CPUHP_OFFLINE);
1021 EXPORT_SYMBOL(cpu_down);
1024 #define takedown_cpu NULL
1025 #endif /*CONFIG_HOTPLUG_CPU*/
1028 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1029 * @cpu: cpu that just started
1031 * It must be called by the arch code on the new cpu, before the new cpu
1032 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1034 void notify_cpu_starting(unsigned int cpu)
1036 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1037 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1040 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1041 st->booted_once = true;
1042 while (st->state < target) {
1044 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1046 * STARTING must not fail!
1053 * Called from the idle task. Wake up the controlling task which brings the
1054 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1055 * the rest of the online bringup to the hotplug thread.
1057 void cpuhp_online_idle(enum cpuhp_state state)
1059 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1061 /* Happens for the boot cpu */
1062 if (state != CPUHP_AP_ONLINE_IDLE)
1065 st->state = CPUHP_AP_ONLINE_IDLE;
1066 complete_ap_thread(st, true);
1069 /* Requires cpu_add_remove_lock to be held */
1070 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1072 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1073 struct task_struct *idle;
1078 if (!cpu_present(cpu)) {
1084 * The caller of do_cpu_up might have raced with another
1085 * caller. Ignore it for now.
1087 if (st->state >= target)
1090 if (st->state == CPUHP_OFFLINE) {
1091 /* Let it fail before we try to bring the cpu up */
1092 idle = idle_thread_get(cpu);
1094 ret = PTR_ERR(idle);
1099 cpuhp_tasks_frozen = tasks_frozen;
1101 cpuhp_set_state(st, target);
1103 * If the current CPU state is in the range of the AP hotplug thread,
1104 * then we need to kick the thread once more.
1106 if (st->state > CPUHP_BRINGUP_CPU) {
1107 ret = cpuhp_kick_ap_work(cpu);
1109 * The AP side has done the error rollback already. Just
1110 * return the error code..
1117 * Try to reach the target state. We max out on the BP at
1118 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1119 * responsible for bringing it up to the target state.
1121 target = min((int)target, CPUHP_BRINGUP_CPU);
1122 ret = cpuhp_up_callbacks(cpu, st, target);
1124 cpus_write_unlock();
1129 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1133 if (!cpu_possible(cpu)) {
1134 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1136 #if defined(CONFIG_IA64)
1137 pr_err("please check additional_cpus= boot parameter\n");
1142 err = try_online_node(cpu_to_node(cpu));
1146 cpu_maps_update_begin();
1148 if (cpu_hotplug_disabled) {
1152 if (!cpu_smt_allowed(cpu)) {
1157 err = _cpu_up(cpu, 0, target);
1159 cpu_maps_update_done();
1163 int cpu_up(unsigned int cpu)
1165 return do_cpu_up(cpu, CPUHP_ONLINE);
1167 EXPORT_SYMBOL_GPL(cpu_up);
1169 #ifdef CONFIG_PM_SLEEP_SMP
1170 static cpumask_var_t frozen_cpus;
1172 int freeze_secondary_cpus(int primary)
1176 cpu_maps_update_begin();
1177 if (!cpu_online(primary))
1178 primary = cpumask_first(cpu_online_mask);
1180 * We take down all of the non-boot CPUs in one shot to avoid races
1181 * with the userspace trying to use the CPU hotplug at the same time
1183 cpumask_clear(frozen_cpus);
1185 pr_info("Disabling non-boot CPUs ...\n");
1186 for_each_online_cpu(cpu) {
1189 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1190 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1191 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1193 cpumask_set_cpu(cpu, frozen_cpus);
1195 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1201 BUG_ON(num_online_cpus() > 1);
1203 pr_err("Non-boot CPUs are not disabled\n");
1206 * Make sure the CPUs won't be enabled by someone else. We need to do
1207 * this even in case of failure as all disable_nonboot_cpus() users are
1208 * supposed to do enable_nonboot_cpus() on the failure path.
1210 cpu_hotplug_disabled++;
1212 cpu_maps_update_done();
1216 void __weak arch_enable_nonboot_cpus_begin(void)
1220 void __weak arch_enable_nonboot_cpus_end(void)
1224 void enable_nonboot_cpus(void)
1228 /* Allow everyone to use the CPU hotplug again */
1229 cpu_maps_update_begin();
1230 __cpu_hotplug_enable();
1231 if (cpumask_empty(frozen_cpus))
1234 pr_info("Enabling non-boot CPUs ...\n");
1236 arch_enable_nonboot_cpus_begin();
1238 for_each_cpu(cpu, frozen_cpus) {
1239 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1240 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1241 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1243 pr_info("CPU%d is up\n", cpu);
1246 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1249 arch_enable_nonboot_cpus_end();
1251 cpumask_clear(frozen_cpus);
1253 cpu_maps_update_done();
1256 static int __init alloc_frozen_cpus(void)
1258 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1262 core_initcall(alloc_frozen_cpus);
1265 * When callbacks for CPU hotplug notifications are being executed, we must
1266 * ensure that the state of the system with respect to the tasks being frozen
1267 * or not, as reported by the notification, remains unchanged *throughout the
1268 * duration* of the execution of the callbacks.
1269 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1271 * This synchronization is implemented by mutually excluding regular CPU
1272 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1273 * Hibernate notifications.
1276 cpu_hotplug_pm_callback(struct notifier_block *nb,
1277 unsigned long action, void *ptr)
1281 case PM_SUSPEND_PREPARE:
1282 case PM_HIBERNATION_PREPARE:
1283 cpu_hotplug_disable();
1286 case PM_POST_SUSPEND:
1287 case PM_POST_HIBERNATION:
1288 cpu_hotplug_enable();
1299 static int __init cpu_hotplug_pm_sync_init(void)
1302 * cpu_hotplug_pm_callback has higher priority than x86
1303 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1304 * to disable cpu hotplug to avoid cpu hotplug race.
1306 pm_notifier(cpu_hotplug_pm_callback, 0);
1309 core_initcall(cpu_hotplug_pm_sync_init);
1311 #endif /* CONFIG_PM_SLEEP_SMP */
1315 #endif /* CONFIG_SMP */
1317 /* Boot processor state steps */
1318 static struct cpuhp_step cpuhp_hp_states[] = {
1321 .startup.single = NULL,
1322 .teardown.single = NULL,
1325 [CPUHP_CREATE_THREADS]= {
1326 .name = "threads:prepare",
1327 .startup.single = smpboot_create_threads,
1328 .teardown.single = NULL,
1331 [CPUHP_PERF_PREPARE] = {
1332 .name = "perf:prepare",
1333 .startup.single = perf_event_init_cpu,
1334 .teardown.single = perf_event_exit_cpu,
1336 [CPUHP_WORKQUEUE_PREP] = {
1337 .name = "workqueue:prepare",
1338 .startup.single = workqueue_prepare_cpu,
1339 .teardown.single = NULL,
1341 [CPUHP_HRTIMERS_PREPARE] = {
1342 .name = "hrtimers:prepare",
1343 .startup.single = hrtimers_prepare_cpu,
1344 .teardown.single = hrtimers_dead_cpu,
1346 [CPUHP_SMPCFD_PREPARE] = {
1347 .name = "smpcfd:prepare",
1348 .startup.single = smpcfd_prepare_cpu,
1349 .teardown.single = smpcfd_dead_cpu,
1351 [CPUHP_RELAY_PREPARE] = {
1352 .name = "relay:prepare",
1353 .startup.single = relay_prepare_cpu,
1354 .teardown.single = NULL,
1356 [CPUHP_SLAB_PREPARE] = {
1357 .name = "slab:prepare",
1358 .startup.single = slab_prepare_cpu,
1359 .teardown.single = slab_dead_cpu,
1361 [CPUHP_RCUTREE_PREP] = {
1362 .name = "RCU/tree:prepare",
1363 .startup.single = rcutree_prepare_cpu,
1364 .teardown.single = rcutree_dead_cpu,
1367 * On the tear-down path, timers_dead_cpu() must be invoked
1368 * before blk_mq_queue_reinit_notify() from notify_dead(),
1369 * otherwise a RCU stall occurs.
1371 [CPUHP_TIMERS_PREPARE] = {
1372 .name = "timers:prepare",
1373 .startup.single = timers_prepare_cpu,
1374 .teardown.single = timers_dead_cpu,
1376 /* Kicks the plugged cpu into life */
1377 [CPUHP_BRINGUP_CPU] = {
1378 .name = "cpu:bringup",
1379 .startup.single = bringup_cpu,
1380 .teardown.single = NULL,
1383 /* Final state before CPU kills itself */
1384 [CPUHP_AP_IDLE_DEAD] = {
1385 .name = "idle:dead",
1388 * Last state before CPU enters the idle loop to die. Transient state
1389 * for synchronization.
1391 [CPUHP_AP_OFFLINE] = {
1392 .name = "ap:offline",
1395 /* First state is scheduler control. Interrupts are disabled */
1396 [CPUHP_AP_SCHED_STARTING] = {
1397 .name = "sched:starting",
1398 .startup.single = sched_cpu_starting,
1399 .teardown.single = sched_cpu_dying,
1401 [CPUHP_AP_RCUTREE_DYING] = {
1402 .name = "RCU/tree:dying",
1403 .startup.single = NULL,
1404 .teardown.single = rcutree_dying_cpu,
1406 [CPUHP_AP_SMPCFD_DYING] = {
1407 .name = "smpcfd:dying",
1408 .startup.single = NULL,
1409 .teardown.single = smpcfd_dying_cpu,
1411 /* Entry state on starting. Interrupts enabled from here on. Transient
1412 * state for synchronsization */
1413 [CPUHP_AP_ONLINE] = {
1414 .name = "ap:online",
1417 * Handled on controll processor until the plugged processor manages
1420 [CPUHP_TEARDOWN_CPU] = {
1421 .name = "cpu:teardown",
1422 .startup.single = NULL,
1423 .teardown.single = takedown_cpu,
1426 /* Handle smpboot threads park/unpark */
1427 [CPUHP_AP_SMPBOOT_THREADS] = {
1428 .name = "smpboot/threads:online",
1429 .startup.single = smpboot_unpark_threads,
1430 .teardown.single = smpboot_park_threads,
1432 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1433 .name = "irq/affinity:online",
1434 .startup.single = irq_affinity_online_cpu,
1435 .teardown.single = NULL,
1437 [CPUHP_AP_PERF_ONLINE] = {
1438 .name = "perf:online",
1439 .startup.single = perf_event_init_cpu,
1440 .teardown.single = perf_event_exit_cpu,
1442 [CPUHP_AP_WATCHDOG_ONLINE] = {
1443 .name = "lockup_detector:online",
1444 .startup.single = lockup_detector_online_cpu,
1445 .teardown.single = lockup_detector_offline_cpu,
1447 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1448 .name = "workqueue:online",
1449 .startup.single = workqueue_online_cpu,
1450 .teardown.single = workqueue_offline_cpu,
1452 [CPUHP_AP_RCUTREE_ONLINE] = {
1453 .name = "RCU/tree:online",
1454 .startup.single = rcutree_online_cpu,
1455 .teardown.single = rcutree_offline_cpu,
1459 * The dynamically registered state space is here
1463 /* Last state is scheduler control setting the cpu active */
1464 [CPUHP_AP_ACTIVE] = {
1465 .name = "sched:active",
1466 .startup.single = sched_cpu_activate,
1467 .teardown.single = sched_cpu_deactivate,
1471 /* CPU is fully up and running. */
1474 .startup.single = NULL,
1475 .teardown.single = NULL,
1479 /* Sanity check for callbacks */
1480 static int cpuhp_cb_check(enum cpuhp_state state)
1482 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1488 * Returns a free for dynamic slot assignment of the Online state. The states
1489 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1490 * by having no name assigned.
1492 static int cpuhp_reserve_state(enum cpuhp_state state)
1494 enum cpuhp_state i, end;
1495 struct cpuhp_step *step;
1498 case CPUHP_AP_ONLINE_DYN:
1499 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1500 end = CPUHP_AP_ONLINE_DYN_END;
1502 case CPUHP_BP_PREPARE_DYN:
1503 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1504 end = CPUHP_BP_PREPARE_DYN_END;
1510 for (i = state; i <= end; i++, step++) {
1514 WARN(1, "No more dynamic states available for CPU hotplug\n");
1518 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1519 int (*startup)(unsigned int cpu),
1520 int (*teardown)(unsigned int cpu),
1521 bool multi_instance)
1523 /* (Un)Install the callbacks for further cpu hotplug operations */
1524 struct cpuhp_step *sp;
1528 * If name is NULL, then the state gets removed.
1530 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1531 * the first allocation from these dynamic ranges, so the removal
1532 * would trigger a new allocation and clear the wrong (already
1533 * empty) state, leaving the callbacks of the to be cleared state
1534 * dangling, which causes wreckage on the next hotplug operation.
1536 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1537 state == CPUHP_BP_PREPARE_DYN)) {
1538 ret = cpuhp_reserve_state(state);
1543 sp = cpuhp_get_step(state);
1544 if (name && sp->name)
1547 sp->startup.single = startup;
1548 sp->teardown.single = teardown;
1550 sp->multi_instance = multi_instance;
1551 INIT_HLIST_HEAD(&sp->list);
1555 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1557 return cpuhp_get_step(state)->teardown.single;
1561 * Call the startup/teardown function for a step either on the AP or
1562 * on the current CPU.
1564 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1565 struct hlist_node *node)
1567 struct cpuhp_step *sp = cpuhp_get_step(state);
1571 * If there's nothing to do, we done.
1572 * Relies on the union for multi_instance.
1574 if ((bringup && !sp->startup.single) ||
1575 (!bringup && !sp->teardown.single))
1578 * The non AP bound callbacks can fail on bringup. On teardown
1579 * e.g. module removal we crash for now.
1582 if (cpuhp_is_ap_state(state))
1583 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1585 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1587 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1589 BUG_ON(ret && !bringup);
1594 * Called from __cpuhp_setup_state on a recoverable failure.
1596 * Note: The teardown callbacks for rollback are not allowed to fail!
1598 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1599 struct hlist_node *node)
1603 /* Roll back the already executed steps on the other cpus */
1604 for_each_present_cpu(cpu) {
1605 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1606 int cpustate = st->state;
1608 if (cpu >= failedcpu)
1611 /* Did we invoke the startup call on that cpu ? */
1612 if (cpustate >= state)
1613 cpuhp_issue_call(cpu, state, false, node);
1617 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1618 struct hlist_node *node,
1621 struct cpuhp_step *sp;
1625 lockdep_assert_cpus_held();
1627 sp = cpuhp_get_step(state);
1628 if (sp->multi_instance == false)
1631 mutex_lock(&cpuhp_state_mutex);
1633 if (!invoke || !sp->startup.multi)
1637 * Try to call the startup callback for each present cpu
1638 * depending on the hotplug state of the cpu.
1640 for_each_present_cpu(cpu) {
1641 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1642 int cpustate = st->state;
1644 if (cpustate < state)
1647 ret = cpuhp_issue_call(cpu, state, true, node);
1649 if (sp->teardown.multi)
1650 cpuhp_rollback_install(cpu, state, node);
1656 hlist_add_head(node, &sp->list);
1658 mutex_unlock(&cpuhp_state_mutex);
1662 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1668 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1672 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1675 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1676 * @state: The state to setup
1677 * @invoke: If true, the startup function is invoked for cpus where
1678 * cpu state >= @state
1679 * @startup: startup callback function
1680 * @teardown: teardown callback function
1681 * @multi_instance: State is set up for multiple instances which get
1684 * The caller needs to hold cpus read locked while calling this function.
1687 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1688 * 0 for all other states
1689 * On failure: proper (negative) error code
1691 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1692 const char *name, bool invoke,
1693 int (*startup)(unsigned int cpu),
1694 int (*teardown)(unsigned int cpu),
1695 bool multi_instance)
1700 lockdep_assert_cpus_held();
1702 if (cpuhp_cb_check(state) || !name)
1705 mutex_lock(&cpuhp_state_mutex);
1707 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1710 dynstate = state == CPUHP_AP_ONLINE_DYN;
1711 if (ret > 0 && dynstate) {
1716 if (ret || !invoke || !startup)
1720 * Try to call the startup callback for each present cpu
1721 * depending on the hotplug state of the cpu.
1723 for_each_present_cpu(cpu) {
1724 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1725 int cpustate = st->state;
1727 if (cpustate < state)
1730 ret = cpuhp_issue_call(cpu, state, true, NULL);
1733 cpuhp_rollback_install(cpu, state, NULL);
1734 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1739 mutex_unlock(&cpuhp_state_mutex);
1741 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1742 * dynamically allocated state in case of success.
1744 if (!ret && dynstate)
1748 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1750 int __cpuhp_setup_state(enum cpuhp_state state,
1751 const char *name, bool invoke,
1752 int (*startup)(unsigned int cpu),
1753 int (*teardown)(unsigned int cpu),
1754 bool multi_instance)
1759 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1760 teardown, multi_instance);
1764 EXPORT_SYMBOL(__cpuhp_setup_state);
1766 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1767 struct hlist_node *node, bool invoke)
1769 struct cpuhp_step *sp = cpuhp_get_step(state);
1772 BUG_ON(cpuhp_cb_check(state));
1774 if (!sp->multi_instance)
1778 mutex_lock(&cpuhp_state_mutex);
1780 if (!invoke || !cpuhp_get_teardown_cb(state))
1783 * Call the teardown callback for each present cpu depending
1784 * on the hotplug state of the cpu. This function is not
1785 * allowed to fail currently!
1787 for_each_present_cpu(cpu) {
1788 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1789 int cpustate = st->state;
1791 if (cpustate >= state)
1792 cpuhp_issue_call(cpu, state, false, node);
1797 mutex_unlock(&cpuhp_state_mutex);
1802 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1805 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1806 * @state: The state to remove
1807 * @invoke: If true, the teardown function is invoked for cpus where
1808 * cpu state >= @state
1810 * The caller needs to hold cpus read locked while calling this function.
1811 * The teardown callback is currently not allowed to fail. Think
1812 * about module removal!
1814 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1816 struct cpuhp_step *sp = cpuhp_get_step(state);
1819 BUG_ON(cpuhp_cb_check(state));
1821 lockdep_assert_cpus_held();
1823 mutex_lock(&cpuhp_state_mutex);
1824 if (sp->multi_instance) {
1825 WARN(!hlist_empty(&sp->list),
1826 "Error: Removing state %d which has instances left.\n",
1831 if (!invoke || !cpuhp_get_teardown_cb(state))
1835 * Call the teardown callback for each present cpu depending
1836 * on the hotplug state of the cpu. This function is not
1837 * allowed to fail currently!
1839 for_each_present_cpu(cpu) {
1840 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1841 int cpustate = st->state;
1843 if (cpustate >= state)
1844 cpuhp_issue_call(cpu, state, false, NULL);
1847 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1848 mutex_unlock(&cpuhp_state_mutex);
1850 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1852 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1855 __cpuhp_remove_state_cpuslocked(state, invoke);
1858 EXPORT_SYMBOL(__cpuhp_remove_state);
1860 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1861 static ssize_t show_cpuhp_state(struct device *dev,
1862 struct device_attribute *attr, char *buf)
1864 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1866 return sprintf(buf, "%d\n", st->state);
1868 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1870 static ssize_t write_cpuhp_target(struct device *dev,
1871 struct device_attribute *attr,
1872 const char *buf, size_t count)
1874 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1875 struct cpuhp_step *sp;
1878 ret = kstrtoint(buf, 10, &target);
1882 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1883 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1886 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1890 ret = lock_device_hotplug_sysfs();
1894 mutex_lock(&cpuhp_state_mutex);
1895 sp = cpuhp_get_step(target);
1896 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1897 mutex_unlock(&cpuhp_state_mutex);
1901 if (st->state < target)
1902 ret = do_cpu_up(dev->id, target);
1904 ret = do_cpu_down(dev->id, target);
1906 unlock_device_hotplug();
1907 return ret ? ret : count;
1910 static ssize_t show_cpuhp_target(struct device *dev,
1911 struct device_attribute *attr, char *buf)
1913 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1915 return sprintf(buf, "%d\n", st->target);
1917 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1920 static ssize_t write_cpuhp_fail(struct device *dev,
1921 struct device_attribute *attr,
1922 const char *buf, size_t count)
1924 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1925 struct cpuhp_step *sp;
1928 ret = kstrtoint(buf, 10, &fail);
1933 * Cannot fail STARTING/DYING callbacks.
1935 if (cpuhp_is_atomic_state(fail))
1939 * Cannot fail anything that doesn't have callbacks.
1941 mutex_lock(&cpuhp_state_mutex);
1942 sp = cpuhp_get_step(fail);
1943 if (!sp->startup.single && !sp->teardown.single)
1945 mutex_unlock(&cpuhp_state_mutex);
1954 static ssize_t show_cpuhp_fail(struct device *dev,
1955 struct device_attribute *attr, char *buf)
1957 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1959 return sprintf(buf, "%d\n", st->fail);
1962 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1964 static struct attribute *cpuhp_cpu_attrs[] = {
1965 &dev_attr_state.attr,
1966 &dev_attr_target.attr,
1967 &dev_attr_fail.attr,
1971 static const struct attribute_group cpuhp_cpu_attr_group = {
1972 .attrs = cpuhp_cpu_attrs,
1977 static ssize_t show_cpuhp_states(struct device *dev,
1978 struct device_attribute *attr, char *buf)
1980 ssize_t cur, res = 0;
1983 mutex_lock(&cpuhp_state_mutex);
1984 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1985 struct cpuhp_step *sp = cpuhp_get_step(i);
1988 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1993 mutex_unlock(&cpuhp_state_mutex);
1996 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1998 static struct attribute *cpuhp_cpu_root_attrs[] = {
1999 &dev_attr_states.attr,
2003 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2004 .attrs = cpuhp_cpu_root_attrs,
2009 #ifdef CONFIG_HOTPLUG_SMT
2011 static const char *smt_states[] = {
2012 [CPU_SMT_ENABLED] = "on",
2013 [CPU_SMT_DISABLED] = "off",
2014 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2015 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2019 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2021 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2024 static void cpuhp_offline_cpu_device(unsigned int cpu)
2026 struct device *dev = get_cpu_device(cpu);
2028 dev->offline = true;
2029 /* Tell user space about the state change */
2030 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2033 static void cpuhp_online_cpu_device(unsigned int cpu)
2035 struct device *dev = get_cpu_device(cpu);
2037 dev->offline = false;
2038 /* Tell user space about the state change */
2039 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2042 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2046 cpu_maps_update_begin();
2047 for_each_online_cpu(cpu) {
2048 if (topology_is_primary_thread(cpu))
2050 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2054 * As this needs to hold the cpu maps lock it's impossible
2055 * to call device_offline() because that ends up calling
2056 * cpu_down() which takes cpu maps lock. cpu maps lock
2057 * needs to be held as this might race against in kernel
2058 * abusers of the hotplug machinery (thermal management).
2060 * So nothing would update device:offline state. That would
2061 * leave the sysfs entry stale and prevent onlining after
2062 * smt control has been changed to 'off' again. This is
2063 * called under the sysfs hotplug lock, so it is properly
2064 * serialized against the regular offline usage.
2066 cpuhp_offline_cpu_device(cpu);
2069 cpu_smt_control = ctrlval;
2070 cpu_maps_update_done();
2074 static int cpuhp_smt_enable(void)
2078 cpu_maps_update_begin();
2079 cpu_smt_control = CPU_SMT_ENABLED;
2080 for_each_present_cpu(cpu) {
2081 /* Skip online CPUs and CPUs on offline nodes */
2082 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2084 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2087 /* See comment in cpuhp_smt_disable() */
2088 cpuhp_online_cpu_device(cpu);
2090 cpu_maps_update_done();
2095 store_smt_control(struct device *dev, struct device_attribute *attr,
2096 const char *buf, size_t count)
2100 if (sysfs_streq(buf, "on"))
2101 ctrlval = CPU_SMT_ENABLED;
2102 else if (sysfs_streq(buf, "off"))
2103 ctrlval = CPU_SMT_DISABLED;
2104 else if (sysfs_streq(buf, "forceoff"))
2105 ctrlval = CPU_SMT_FORCE_DISABLED;
2109 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2112 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2115 ret = lock_device_hotplug_sysfs();
2119 if (ctrlval != cpu_smt_control) {
2121 case CPU_SMT_ENABLED:
2122 ret = cpuhp_smt_enable();
2124 case CPU_SMT_DISABLED:
2125 case CPU_SMT_FORCE_DISABLED:
2126 ret = cpuhp_smt_disable(ctrlval);
2131 unlock_device_hotplug();
2132 return ret ? ret : count;
2134 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2137 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2139 bool active = topology_max_smt_threads() > 1;
2141 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2143 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2145 static struct attribute *cpuhp_smt_attrs[] = {
2146 &dev_attr_control.attr,
2147 &dev_attr_active.attr,
2151 static const struct attribute_group cpuhp_smt_attr_group = {
2152 .attrs = cpuhp_smt_attrs,
2157 static int __init cpu_smt_state_init(void)
2159 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2160 &cpuhp_smt_attr_group);
2164 static inline int cpu_smt_state_init(void) { return 0; }
2167 static int __init cpuhp_sysfs_init(void)
2171 ret = cpu_smt_state_init();
2175 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2176 &cpuhp_cpu_root_attr_group);
2180 for_each_possible_cpu(cpu) {
2181 struct device *dev = get_cpu_device(cpu);
2185 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2191 device_initcall(cpuhp_sysfs_init);
2195 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2196 * represents all NR_CPUS bits binary values of 1<<nr.
2198 * It is used by cpumask_of() to get a constant address to a CPU
2199 * mask value that has a single bit set only.
2202 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2203 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2204 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2205 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2206 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2208 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2210 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2211 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2212 #if BITS_PER_LONG > 32
2213 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2214 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2217 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2219 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2220 EXPORT_SYMBOL(cpu_all_bits);
2222 #ifdef CONFIG_INIT_ALL_POSSIBLE
2223 struct cpumask __cpu_possible_mask __read_mostly
2226 struct cpumask __cpu_possible_mask __read_mostly;
2228 EXPORT_SYMBOL(__cpu_possible_mask);
2230 struct cpumask __cpu_online_mask __read_mostly;
2231 EXPORT_SYMBOL(__cpu_online_mask);
2233 struct cpumask __cpu_present_mask __read_mostly;
2234 EXPORT_SYMBOL(__cpu_present_mask);
2236 struct cpumask __cpu_active_mask __read_mostly;
2237 EXPORT_SYMBOL(__cpu_active_mask);
2239 void init_cpu_present(const struct cpumask *src)
2241 cpumask_copy(&__cpu_present_mask, src);
2244 void init_cpu_possible(const struct cpumask *src)
2246 cpumask_copy(&__cpu_possible_mask, src);
2249 void init_cpu_online(const struct cpumask *src)
2251 cpumask_copy(&__cpu_online_mask, src);
2255 * Activate the first processor.
2257 void __init boot_cpu_init(void)
2259 int cpu = smp_processor_id();
2261 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2262 set_cpu_online(cpu, true);
2263 set_cpu_active(cpu, true);
2264 set_cpu_present(cpu, true);
2265 set_cpu_possible(cpu, true);
2268 __boot_cpu_id = cpu;
2273 * Must be called _AFTER_ setting up the per_cpu areas
2275 void __init boot_cpu_hotplug_init(void)
2278 this_cpu_write(cpuhp_state.booted_once, true);
2280 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);