2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/sched/mm.h>
7 #include <linux/proc_fs.h>
9 #include <linux/init.h>
10 #include <linux/notifier.h>
11 #include <linux/sched/signal.h>
12 #include <linux/sched/hotplug.h>
13 #include <linux/sched/isolation.h>
14 #include <linux/sched/task.h>
15 #include <linux/sched/smt.h>
16 #include <linux/unistd.h>
17 #include <linux/cpu.h>
18 #include <linux/oom.h>
19 #include <linux/rcupdate.h>
20 #include <linux/export.h>
21 #include <linux/bug.h>
22 #include <linux/kthread.h>
23 #include <linux/stop_machine.h>
24 #include <linux/mutex.h>
25 #include <linux/gfp.h>
26 #include <linux/suspend.h>
27 #include <linux/lockdep.h>
28 #include <linux/tick.h>
29 #include <linux/irq.h>
30 #include <linux/nmi.h>
31 #include <linux/smpboot.h>
32 #include <linux/relay.h>
33 #include <linux/slab.h>
34 #include <linux/scs.h>
35 #include <linux/percpu-rwsem.h>
36 #include <linux/cpuset.h>
38 #include <trace/events/power.h>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/cpuhp.h>
45 * cpuhp_cpu_state - Per cpu hotplug state storage
46 * @state: The current cpu state
47 * @target: The target state
48 * @thread: Pointer to the hotplug thread
49 * @should_run: Thread should execute
50 * @rollback: Perform a rollback
51 * @single: Single callback invocation
52 * @bringup: Single callback bringup or teardown selector
53 * @cb_state: The state for a single callback (install/uninstall)
54 * @result: Result of the operation
55 * @done_up: Signal completion to the issuer of the task for cpu-up
56 * @done_down: Signal completion to the issuer of the task for cpu-down
58 struct cpuhp_cpu_state {
59 enum cpuhp_state state;
60 enum cpuhp_state target;
61 enum cpuhp_state fail;
63 struct task_struct *thread;
68 struct hlist_node *node;
69 struct hlist_node *last;
70 enum cpuhp_state cb_state;
72 struct completion done_up;
73 struct completion done_down;
77 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
78 .fail = CPUHP_INVALID,
82 cpumask_t cpus_booted_once_mask;
85 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
86 static struct lockdep_map cpuhp_state_up_map =
87 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
88 static struct lockdep_map cpuhp_state_down_map =
89 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
92 static inline void cpuhp_lock_acquire(bool bringup)
94 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
97 static inline void cpuhp_lock_release(bool bringup)
99 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
103 static inline void cpuhp_lock_acquire(bool bringup) { }
104 static inline void cpuhp_lock_release(bool bringup) { }
109 * cpuhp_step - Hotplug state machine step
110 * @name: Name of the step
111 * @startup: Startup function of the step
112 * @teardown: Teardown function of the step
113 * @cant_stop: Bringup/teardown can't be stopped at this step
118 int (*single)(unsigned int cpu);
119 int (*multi)(unsigned int cpu,
120 struct hlist_node *node);
123 int (*single)(unsigned int cpu);
124 int (*multi)(unsigned int cpu,
125 struct hlist_node *node);
127 struct hlist_head list;
132 static DEFINE_MUTEX(cpuhp_state_mutex);
133 static struct cpuhp_step cpuhp_hp_states[];
135 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
137 return cpuhp_hp_states + state;
141 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
142 * @cpu: The cpu for which the callback should be invoked
143 * @state: The state to do callbacks for
144 * @bringup: True if the bringup callback should be invoked
145 * @node: For multi-instance, do a single entry callback for install/remove
146 * @lastp: For multi-instance rollback, remember how far we got
148 * Called from cpu hotplug and from the state register machinery.
150 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
151 bool bringup, struct hlist_node *node,
152 struct hlist_node **lastp)
154 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
155 struct cpuhp_step *step = cpuhp_get_step(state);
156 int (*cbm)(unsigned int cpu, struct hlist_node *node);
157 int (*cb)(unsigned int cpu);
160 if (st->fail == state) {
161 st->fail = CPUHP_INVALID;
163 if (!(bringup ? step->startup.single : step->teardown.single))
169 if (!step->multi_instance) {
170 WARN_ON_ONCE(lastp && *lastp);
171 cb = bringup ? step->startup.single : step->teardown.single;
174 trace_cpuhp_enter(cpu, st->target, state, cb);
176 trace_cpuhp_exit(cpu, st->state, state, ret);
179 cbm = bringup ? step->startup.multi : step->teardown.multi;
183 /* Single invocation for instance add/remove */
185 WARN_ON_ONCE(lastp && *lastp);
186 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
187 ret = cbm(cpu, node);
188 trace_cpuhp_exit(cpu, st->state, state, ret);
192 /* State transition. Invoke on all instances */
194 hlist_for_each(node, &step->list) {
195 if (lastp && node == *lastp)
198 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
199 ret = cbm(cpu, node);
200 trace_cpuhp_exit(cpu, st->state, state, ret);
214 /* Rollback the instances if one failed */
215 cbm = !bringup ? step->startup.multi : step->teardown.multi;
219 hlist_for_each(node, &step->list) {
223 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
224 ret = cbm(cpu, node);
225 trace_cpuhp_exit(cpu, st->state, state, ret);
227 * Rollback must not fail,
235 static bool cpuhp_is_ap_state(enum cpuhp_state state)
238 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
239 * purposes as that state is handled explicitly in cpu_down.
241 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
244 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
246 struct completion *done = bringup ? &st->done_up : &st->done_down;
247 wait_for_completion(done);
250 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
252 struct completion *done = bringup ? &st->done_up : &st->done_down;
257 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
259 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
261 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
264 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
265 static DEFINE_MUTEX(cpu_add_remove_lock);
266 bool cpuhp_tasks_frozen;
267 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
270 * The following two APIs (cpu_maps_update_begin/done) must be used when
271 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
273 void cpu_maps_update_begin(void)
275 mutex_lock(&cpu_add_remove_lock);
278 void cpu_maps_update_done(void)
280 mutex_unlock(&cpu_add_remove_lock);
284 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
285 * Should always be manipulated under cpu_add_remove_lock
287 static int cpu_hotplug_disabled;
289 #ifdef CONFIG_HOTPLUG_CPU
291 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
293 void cpus_read_lock(void)
295 percpu_down_read(&cpu_hotplug_lock);
297 EXPORT_SYMBOL_GPL(cpus_read_lock);
299 int cpus_read_trylock(void)
301 return percpu_down_read_trylock(&cpu_hotplug_lock);
303 EXPORT_SYMBOL_GPL(cpus_read_trylock);
305 void cpus_read_unlock(void)
307 percpu_up_read(&cpu_hotplug_lock);
309 EXPORT_SYMBOL_GPL(cpus_read_unlock);
311 void cpus_write_lock(void)
313 percpu_down_write(&cpu_hotplug_lock);
316 void cpus_write_unlock(void)
318 percpu_up_write(&cpu_hotplug_lock);
321 void lockdep_assert_cpus_held(void)
324 * We can't have hotplug operations before userspace starts running,
325 * and some init codepaths will knowingly not take the hotplug lock.
326 * This is all valid, so mute lockdep until it makes sense to report
329 if (system_state < SYSTEM_RUNNING)
332 percpu_rwsem_assert_held(&cpu_hotplug_lock);
335 static void lockdep_acquire_cpus_lock(void)
337 rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
340 static void lockdep_release_cpus_lock(void)
342 rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
346 * Wait for currently running CPU hotplug operations to complete (if any) and
347 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
348 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
349 * hotplug path before performing hotplug operations. So acquiring that lock
350 * guarantees mutual exclusion from any currently running hotplug operations.
352 void cpu_hotplug_disable(void)
354 cpu_maps_update_begin();
355 cpu_hotplug_disabled++;
356 cpu_maps_update_done();
358 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
360 static void __cpu_hotplug_enable(void)
362 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
364 cpu_hotplug_disabled--;
367 void cpu_hotplug_enable(void)
369 cpu_maps_update_begin();
370 __cpu_hotplug_enable();
371 cpu_maps_update_done();
373 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
377 static void lockdep_acquire_cpus_lock(void)
381 static void lockdep_release_cpus_lock(void)
385 #endif /* CONFIG_HOTPLUG_CPU */
388 * Architectures that need SMT-specific errata handling during SMT hotplug
389 * should override this.
391 void __weak arch_smt_update(void) { }
393 #ifdef CONFIG_HOTPLUG_SMT
394 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
396 void __init cpu_smt_disable(bool force)
398 if (!cpu_smt_possible())
402 pr_info("SMT: Force disabled\n");
403 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
405 pr_info("SMT: disabled\n");
406 cpu_smt_control = CPU_SMT_DISABLED;
411 * The decision whether SMT is supported can only be done after the full
412 * CPU identification. Called from architecture code.
414 void __init cpu_smt_check_topology(void)
416 if (!topology_smt_supported())
417 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
420 static int __init smt_cmdline_disable(char *str)
422 cpu_smt_disable(str && !strcmp(str, "force"));
425 early_param("nosmt", smt_cmdline_disable);
427 static inline bool cpu_smt_allowed(unsigned int cpu)
429 if (cpu_smt_control == CPU_SMT_ENABLED)
432 if (topology_is_primary_thread(cpu))
436 * On x86 it's required to boot all logical CPUs at least once so
437 * that the init code can get a chance to set CR4.MCE on each
438 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
439 * core will shutdown the machine.
441 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
444 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
445 bool cpu_smt_possible(void)
447 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
448 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
450 EXPORT_SYMBOL_GPL(cpu_smt_possible);
452 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
455 static inline enum cpuhp_state
456 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
458 enum cpuhp_state prev_state = st->state;
460 st->rollback = false;
465 st->bringup = st->state < target;
471 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
476 * If we have st->last we need to undo partial multi_instance of this
477 * state first. Otherwise start undo at the previous state.
486 st->target = prev_state;
487 st->bringup = !st->bringup;
490 /* Regular hotplug invocation of the AP hotplug thread */
491 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
493 if (!st->single && st->state == st->target)
498 * Make sure the above stores are visible before should_run becomes
499 * true. Paired with the mb() above in cpuhp_thread_fun()
502 st->should_run = true;
503 wake_up_process(st->thread);
504 wait_for_ap_thread(st, st->bringup);
507 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
509 enum cpuhp_state prev_state;
512 prev_state = cpuhp_set_state(st, target);
514 if ((ret = st->result)) {
515 cpuhp_reset_state(st, prev_state);
522 static int bringup_wait_for_ap(unsigned int cpu)
524 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
526 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
527 wait_for_ap_thread(st, true);
528 if (WARN_ON_ONCE((!cpu_online(cpu))))
531 /* Unpark the hotplug thread of the target cpu */
532 kthread_unpark(st->thread);
535 * SMT soft disabling on X86 requires to bring the CPU out of the
536 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
537 * CPU marked itself as booted_once in notify_cpu_starting() so the
538 * cpu_smt_allowed() check will now return false if this is not the
541 if (!cpu_smt_allowed(cpu))
544 if (st->target <= CPUHP_AP_ONLINE_IDLE)
547 return cpuhp_kick_ap(st, st->target);
550 static int bringup_cpu(unsigned int cpu)
552 struct task_struct *idle = idle_thread_get(cpu);
556 * Reset stale stack state from the last time this CPU was online.
558 scs_task_reset(idle);
559 kasan_unpoison_task_stack(idle);
562 * Some architectures have to walk the irq descriptors to
563 * setup the vector space for the cpu which comes online.
564 * Prevent irq alloc/free across the bringup.
568 /* Arch-specific enabling code. */
569 ret = __cpu_up(cpu, idle);
573 return bringup_wait_for_ap(cpu);
576 static int finish_cpu(unsigned int cpu)
578 struct task_struct *idle = idle_thread_get(cpu);
579 struct mm_struct *mm = idle->active_mm;
582 * idle_task_exit() will have switched to &init_mm, now
583 * clean up any remaining active_mm state.
586 idle->active_mm = &init_mm;
592 * Hotplug state machine related functions
595 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
597 for (st->state--; st->state > st->target; st->state--)
598 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
601 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
603 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
606 * When CPU hotplug is disabled, then taking the CPU down is not
607 * possible because takedown_cpu() and the architecture and
608 * subsystem specific mechanisms are not available. So the CPU
609 * which would be completely unplugged again needs to stay around
610 * in the current state.
612 return st->state <= CPUHP_BRINGUP_CPU;
615 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
616 enum cpuhp_state target)
618 enum cpuhp_state prev_state = st->state;
621 while (st->state < target) {
623 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
625 if (can_rollback_cpu(st)) {
626 st->target = prev_state;
627 undo_cpu_up(cpu, st);
636 * The cpu hotplug threads manage the bringup and teardown of the cpus
638 static void cpuhp_create(unsigned int cpu)
640 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
642 init_completion(&st->done_up);
643 init_completion(&st->done_down);
646 static int cpuhp_should_run(unsigned int cpu)
648 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
650 return st->should_run;
654 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
655 * callbacks when a state gets [un]installed at runtime.
657 * Each invocation of this function by the smpboot thread does a single AP
660 * It has 3 modes of operation:
661 * - single: runs st->cb_state
662 * - up: runs ++st->state, while st->state < st->target
663 * - down: runs st->state--, while st->state > st->target
665 * When complete or on error, should_run is cleared and the completion is fired.
667 static void cpuhp_thread_fun(unsigned int cpu)
669 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
670 bool bringup = st->bringup;
671 enum cpuhp_state state;
673 if (WARN_ON_ONCE(!st->should_run))
677 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
678 * that if we see ->should_run we also see the rest of the state.
683 * The BP holds the hotplug lock, but we're now running on the AP,
684 * ensure that anybody asserting the lock is held, will actually find
687 lockdep_acquire_cpus_lock();
688 cpuhp_lock_acquire(bringup);
691 state = st->cb_state;
692 st->should_run = false;
697 st->should_run = (st->state < st->target);
698 WARN_ON_ONCE(st->state > st->target);
702 st->should_run = (st->state > st->target);
703 WARN_ON_ONCE(st->state < st->target);
707 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
709 if (cpuhp_is_atomic_state(state)) {
711 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
715 * STARTING/DYING must not fail!
717 WARN_ON_ONCE(st->result);
719 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
724 * If we fail on a rollback, we're up a creek without no
725 * paddle, no way forward, no way back. We loose, thanks for
728 WARN_ON_ONCE(st->rollback);
729 st->should_run = false;
732 cpuhp_lock_release(bringup);
733 lockdep_release_cpus_lock();
736 complete_ap_thread(st, bringup);
739 /* Invoke a single callback on a remote cpu */
741 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
742 struct hlist_node *node)
744 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
747 if (!cpu_online(cpu))
750 cpuhp_lock_acquire(false);
751 cpuhp_lock_release(false);
753 cpuhp_lock_acquire(true);
754 cpuhp_lock_release(true);
757 * If we are up and running, use the hotplug thread. For early calls
758 * we invoke the thread function directly.
761 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
763 st->rollback = false;
767 st->bringup = bringup;
768 st->cb_state = state;
774 * If we failed and did a partial, do a rollback.
776 if ((ret = st->result) && st->last) {
778 st->bringup = !bringup;
784 * Clean up the leftovers so the next hotplug operation wont use stale
787 st->node = st->last = NULL;
791 static int cpuhp_kick_ap_work(unsigned int cpu)
793 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
794 enum cpuhp_state prev_state = st->state;
797 cpuhp_lock_acquire(false);
798 cpuhp_lock_release(false);
800 cpuhp_lock_acquire(true);
801 cpuhp_lock_release(true);
803 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
804 ret = cpuhp_kick_ap(st, st->target);
805 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
810 static struct smp_hotplug_thread cpuhp_threads = {
811 .store = &cpuhp_state.thread,
812 .create = &cpuhp_create,
813 .thread_should_run = cpuhp_should_run,
814 .thread_fn = cpuhp_thread_fun,
815 .thread_comm = "cpuhp/%u",
819 void __init cpuhp_threads_init(void)
821 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
822 kthread_unpark(this_cpu_read(cpuhp_state.thread));
827 * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
830 * The operation is still serialized against concurrent CPU hotplug via
831 * cpu_add_remove_lock, i.e. CPU map protection. But it is _not_
832 * serialized against other hotplug related activity like adding or
833 * removing of state callbacks and state instances, which invoke either the
834 * startup or the teardown callback of the affected state.
836 * This is required for subsystems which are unfixable vs. CPU hotplug and
837 * evade lock inversion problems by scheduling work which has to be
838 * completed _before_ cpu_up()/_cpu_down() returns.
840 * Don't even think about adding anything to this for any new code or even
841 * drivers. It's only purpose is to keep existing lock order trainwrecks
844 * For cpu_down() there might be valid reasons to finish cleanups which are
845 * not required to be done under cpu_hotplug_lock, but that's a different
846 * story and would be not invoked via this.
848 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
851 * cpusets delegate hotplug operations to a worker to "solve" the
852 * lock order problems. Wait for the worker, but only if tasks are
853 * _not_ frozen (suspend, hibernate) as that would wait forever.
855 * The wait is required because otherwise the hotplug operation
856 * returns with inconsistent state, which could even be observed in
857 * user space when a new CPU is brought up. The CPU plug uevent
858 * would be delivered and user space reacting on it would fail to
859 * move tasks to the newly plugged CPU up to the point where the
860 * work has finished because up to that point the newly plugged CPU
861 * is not assignable in cpusets/cgroups. On unplug that's not
862 * necessarily a visible issue, but it is still inconsistent state,
863 * which is the real problem which needs to be "fixed". This can't
864 * prevent the transient state between scheduling the work and
865 * returning from waiting for it.
868 cpuset_wait_for_hotplug();
871 #ifdef CONFIG_HOTPLUG_CPU
872 #ifndef arch_clear_mm_cpumask_cpu
873 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
877 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
880 * This function walks all processes, finds a valid mm struct for each one and
881 * then clears a corresponding bit in mm's cpumask. While this all sounds
882 * trivial, there are various non-obvious corner cases, which this function
883 * tries to solve in a safe manner.
885 * Also note that the function uses a somewhat relaxed locking scheme, so it may
886 * be called only for an already offlined CPU.
888 void clear_tasks_mm_cpumask(int cpu)
890 struct task_struct *p;
893 * This function is called after the cpu is taken down and marked
894 * offline, so its not like new tasks will ever get this cpu set in
895 * their mm mask. -- Peter Zijlstra
896 * Thus, we may use rcu_read_lock() here, instead of grabbing
897 * full-fledged tasklist_lock.
899 WARN_ON(cpu_online(cpu));
901 for_each_process(p) {
902 struct task_struct *t;
905 * Main thread might exit, but other threads may still have
906 * a valid mm. Find one.
908 t = find_lock_task_mm(p);
911 arch_clear_mm_cpumask_cpu(cpu, t->mm);
917 /* Take this CPU down. */
918 static int take_cpu_down(void *_param)
920 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
921 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
922 int err, cpu = smp_processor_id();
925 /* Ensure this CPU doesn't handle any more interrupts. */
926 err = __cpu_disable();
931 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
932 * do this step again.
934 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
936 /* Invoke the former CPU_DYING callbacks */
937 for (; st->state > target; st->state--) {
938 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
940 * DYING must not fail!
945 /* Give up timekeeping duties */
946 tick_handover_do_timer();
947 /* Remove CPU from timer broadcasting */
948 tick_offline_cpu(cpu);
949 /* Park the stopper thread */
950 stop_machine_park(cpu);
954 static int takedown_cpu(unsigned int cpu)
956 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
959 /* Park the smpboot threads */
960 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
963 * Prevent irq alloc/free while the dying cpu reorganizes the
964 * interrupt affinities.
969 * So now all preempt/rcu users must observe !cpu_active().
971 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
973 /* CPU refused to die */
975 /* Unpark the hotplug thread so we can rollback there */
976 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
979 BUG_ON(cpu_online(cpu));
982 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
983 * all runnable tasks from the CPU, there's only the idle task left now
984 * that the migration thread is done doing the stop_machine thing.
986 * Wait for the stop thread to go away.
988 wait_for_ap_thread(st, false);
989 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
991 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
994 hotplug_cpu__broadcast_tick_pull(cpu);
995 /* This actually kills the CPU. */
998 tick_cleanup_dead_cpu(cpu);
999 rcutree_migrate_callbacks(cpu);
1003 static void cpuhp_complete_idle_dead(void *arg)
1005 struct cpuhp_cpu_state *st = arg;
1007 complete_ap_thread(st, false);
1010 void cpuhp_report_idle_dead(void)
1012 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1014 BUG_ON(st->state != CPUHP_AP_OFFLINE);
1015 rcu_report_dead(smp_processor_id());
1016 st->state = CPUHP_AP_IDLE_DEAD;
1018 * We cannot call complete after rcu_report_dead() so we delegate it
1021 smp_call_function_single(cpumask_first(cpu_online_mask),
1022 cpuhp_complete_idle_dead, st, 0);
1025 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
1027 for (st->state++; st->state < st->target; st->state++)
1028 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1031 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1032 enum cpuhp_state target)
1034 enum cpuhp_state prev_state = st->state;
1037 for (; st->state > target; st->state--) {
1038 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
1040 st->target = prev_state;
1041 if (st->state < prev_state)
1042 undo_cpu_down(cpu, st);
1049 /* Requires cpu_add_remove_lock to be held */
1050 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1051 enum cpuhp_state target)
1053 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1054 int prev_state, ret = 0;
1056 if (num_online_cpus() == 1)
1059 if (!cpu_present(cpu))
1064 cpuhp_tasks_frozen = tasks_frozen;
1066 prev_state = cpuhp_set_state(st, target);
1068 * If the current CPU state is in the range of the AP hotplug thread,
1069 * then we need to kick the thread.
1071 if (st->state > CPUHP_TEARDOWN_CPU) {
1072 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1073 ret = cpuhp_kick_ap_work(cpu);
1075 * The AP side has done the error rollback already. Just
1076 * return the error code..
1082 * We might have stopped still in the range of the AP hotplug
1083 * thread. Nothing to do anymore.
1085 if (st->state > CPUHP_TEARDOWN_CPU)
1088 st->target = target;
1091 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1092 * to do the further cleanups.
1094 ret = cpuhp_down_callbacks(cpu, st, target);
1095 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
1096 cpuhp_reset_state(st, prev_state);
1097 __cpuhp_kick_ap(st);
1101 cpus_write_unlock();
1103 * Do post unplug cleanup. This is still protected against
1104 * concurrent CPU hotplug via cpu_add_remove_lock.
1106 lockup_detector_cleanup();
1108 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1112 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1114 if (cpu_hotplug_disabled)
1116 return _cpu_down(cpu, 0, target);
1119 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1123 cpu_maps_update_begin();
1124 err = cpu_down_maps_locked(cpu, target);
1125 cpu_maps_update_done();
1130 * cpu_device_down - Bring down a cpu device
1131 * @dev: Pointer to the cpu device to offline
1133 * This function is meant to be used by device core cpu subsystem only.
1135 * Other subsystems should use remove_cpu() instead.
1137 int cpu_device_down(struct device *dev)
1139 return cpu_down(dev->id, CPUHP_OFFLINE);
1142 int remove_cpu(unsigned int cpu)
1146 lock_device_hotplug();
1147 ret = device_offline(get_cpu_device(cpu));
1148 unlock_device_hotplug();
1152 EXPORT_SYMBOL_GPL(remove_cpu);
1154 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1159 cpu_maps_update_begin();
1162 * Make certain the cpu I'm about to reboot on is online.
1164 * This is inline to what migrate_to_reboot_cpu() already do.
1166 if (!cpu_online(primary_cpu))
1167 primary_cpu = cpumask_first(cpu_online_mask);
1169 for_each_online_cpu(cpu) {
1170 if (cpu == primary_cpu)
1173 error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1175 pr_err("Failed to offline CPU%d - error=%d",
1182 * Ensure all but the reboot CPU are offline.
1184 BUG_ON(num_online_cpus() > 1);
1187 * Make sure the CPUs won't be enabled by someone else after this
1188 * point. Kexec will reboot to a new kernel shortly resetting
1189 * everything along the way.
1191 cpu_hotplug_disabled++;
1193 cpu_maps_update_done();
1197 #define takedown_cpu NULL
1198 #endif /*CONFIG_HOTPLUG_CPU*/
1201 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1202 * @cpu: cpu that just started
1204 * It must be called by the arch code on the new cpu, before the new cpu
1205 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1207 void notify_cpu_starting(unsigned int cpu)
1209 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1210 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1213 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1214 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1215 while (st->state < target) {
1217 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1219 * STARTING must not fail!
1226 * Called from the idle task. Wake up the controlling task which brings the
1227 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1228 * online bringup to the hotplug thread.
1230 void cpuhp_online_idle(enum cpuhp_state state)
1232 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1234 /* Happens for the boot cpu */
1235 if (state != CPUHP_AP_ONLINE_IDLE)
1239 * Unpart the stopper thread before we start the idle loop (and start
1240 * scheduling); this ensures the stopper task is always available.
1242 stop_machine_unpark(smp_processor_id());
1244 st->state = CPUHP_AP_ONLINE_IDLE;
1245 complete_ap_thread(st, true);
1248 /* Requires cpu_add_remove_lock to be held */
1249 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1251 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1252 struct task_struct *idle;
1257 if (!cpu_present(cpu)) {
1263 * The caller of cpu_up() might have raced with another
1264 * caller. Nothing to do.
1266 if (st->state >= target)
1269 if (st->state == CPUHP_OFFLINE) {
1270 /* Let it fail before we try to bring the cpu up */
1271 idle = idle_thread_get(cpu);
1273 ret = PTR_ERR(idle);
1278 cpuhp_tasks_frozen = tasks_frozen;
1280 cpuhp_set_state(st, target);
1282 * If the current CPU state is in the range of the AP hotplug thread,
1283 * then we need to kick the thread once more.
1285 if (st->state > CPUHP_BRINGUP_CPU) {
1286 ret = cpuhp_kick_ap_work(cpu);
1288 * The AP side has done the error rollback already. Just
1289 * return the error code..
1296 * Try to reach the target state. We max out on the BP at
1297 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1298 * responsible for bringing it up to the target state.
1300 target = min((int)target, CPUHP_BRINGUP_CPU);
1301 ret = cpuhp_up_callbacks(cpu, st, target);
1303 cpus_write_unlock();
1305 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1309 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1313 if (!cpu_possible(cpu)) {
1314 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1316 #if defined(CONFIG_IA64)
1317 pr_err("please check additional_cpus= boot parameter\n");
1322 err = try_online_node(cpu_to_node(cpu));
1326 cpu_maps_update_begin();
1328 if (cpu_hotplug_disabled) {
1332 if (!cpu_smt_allowed(cpu)) {
1337 err = _cpu_up(cpu, 0, target);
1339 cpu_maps_update_done();
1344 * cpu_device_up - Bring up a cpu device
1345 * @dev: Pointer to the cpu device to online
1347 * This function is meant to be used by device core cpu subsystem only.
1349 * Other subsystems should use add_cpu() instead.
1351 int cpu_device_up(struct device *dev)
1353 return cpu_up(dev->id, CPUHP_ONLINE);
1356 int add_cpu(unsigned int cpu)
1360 lock_device_hotplug();
1361 ret = device_online(get_cpu_device(cpu));
1362 unlock_device_hotplug();
1366 EXPORT_SYMBOL_GPL(add_cpu);
1369 * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1370 * @sleep_cpu: The cpu we hibernated on and should be brought up.
1372 * On some architectures like arm64, we can hibernate on any CPU, but on
1373 * wake up the CPU we hibernated on might be offline as a side effect of
1374 * using maxcpus= for example.
1376 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1380 if (!cpu_online(sleep_cpu)) {
1381 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1382 ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1384 pr_err("Failed to bring hibernate-CPU up!\n");
1391 void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1395 for_each_present_cpu(cpu) {
1396 if (num_online_cpus() >= setup_max_cpus)
1398 if (!cpu_online(cpu))
1399 cpu_up(cpu, CPUHP_ONLINE);
1403 #ifdef CONFIG_PM_SLEEP_SMP
1404 static cpumask_var_t frozen_cpus;
1406 int freeze_secondary_cpus(int primary)
1410 cpu_maps_update_begin();
1411 if (primary == -1) {
1412 primary = cpumask_first(cpu_online_mask);
1413 if (!housekeeping_cpu(primary, HK_FLAG_TIMER))
1414 primary = housekeeping_any_cpu(HK_FLAG_TIMER);
1416 if (!cpu_online(primary))
1417 primary = cpumask_first(cpu_online_mask);
1421 * We take down all of the non-boot CPUs in one shot to avoid races
1422 * with the userspace trying to use the CPU hotplug at the same time
1424 cpumask_clear(frozen_cpus);
1426 pr_info("Disabling non-boot CPUs ...\n");
1427 for_each_online_cpu(cpu) {
1431 if (pm_wakeup_pending()) {
1432 pr_info("Wakeup pending. Abort CPU freeze\n");
1437 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1438 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1439 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1441 cpumask_set_cpu(cpu, frozen_cpus);
1443 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1449 BUG_ON(num_online_cpus() > 1);
1451 pr_err("Non-boot CPUs are not disabled\n");
1454 * Make sure the CPUs won't be enabled by someone else. We need to do
1455 * this even in case of failure as all freeze_secondary_cpus() users are
1456 * supposed to do thaw_secondary_cpus() on the failure path.
1458 cpu_hotplug_disabled++;
1460 cpu_maps_update_done();
1464 void __weak arch_thaw_secondary_cpus_begin(void)
1468 void __weak arch_thaw_secondary_cpus_end(void)
1472 void thaw_secondary_cpus(void)
1476 /* Allow everyone to use the CPU hotplug again */
1477 cpu_maps_update_begin();
1478 __cpu_hotplug_enable();
1479 if (cpumask_empty(frozen_cpus))
1482 pr_info("Enabling non-boot CPUs ...\n");
1484 arch_thaw_secondary_cpus_begin();
1486 for_each_cpu(cpu, frozen_cpus) {
1487 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1488 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1489 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1491 pr_info("CPU%d is up\n", cpu);
1494 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1497 arch_thaw_secondary_cpus_end();
1499 cpumask_clear(frozen_cpus);
1501 cpu_maps_update_done();
1504 static int __init alloc_frozen_cpus(void)
1506 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1510 core_initcall(alloc_frozen_cpus);
1513 * When callbacks for CPU hotplug notifications are being executed, we must
1514 * ensure that the state of the system with respect to the tasks being frozen
1515 * or not, as reported by the notification, remains unchanged *throughout the
1516 * duration* of the execution of the callbacks.
1517 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1519 * This synchronization is implemented by mutually excluding regular CPU
1520 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1521 * Hibernate notifications.
1524 cpu_hotplug_pm_callback(struct notifier_block *nb,
1525 unsigned long action, void *ptr)
1529 case PM_SUSPEND_PREPARE:
1530 case PM_HIBERNATION_PREPARE:
1531 cpu_hotplug_disable();
1534 case PM_POST_SUSPEND:
1535 case PM_POST_HIBERNATION:
1536 cpu_hotplug_enable();
1547 static int __init cpu_hotplug_pm_sync_init(void)
1550 * cpu_hotplug_pm_callback has higher priority than x86
1551 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1552 * to disable cpu hotplug to avoid cpu hotplug race.
1554 pm_notifier(cpu_hotplug_pm_callback, 0);
1557 core_initcall(cpu_hotplug_pm_sync_init);
1559 #endif /* CONFIG_PM_SLEEP_SMP */
1563 #endif /* CONFIG_SMP */
1565 /* Boot processor state steps */
1566 static struct cpuhp_step cpuhp_hp_states[] = {
1569 .startup.single = NULL,
1570 .teardown.single = NULL,
1573 [CPUHP_CREATE_THREADS]= {
1574 .name = "threads:prepare",
1575 .startup.single = smpboot_create_threads,
1576 .teardown.single = NULL,
1579 [CPUHP_PERF_PREPARE] = {
1580 .name = "perf:prepare",
1581 .startup.single = perf_event_init_cpu,
1582 .teardown.single = perf_event_exit_cpu,
1584 [CPUHP_WORKQUEUE_PREP] = {
1585 .name = "workqueue:prepare",
1586 .startup.single = workqueue_prepare_cpu,
1587 .teardown.single = NULL,
1589 [CPUHP_HRTIMERS_PREPARE] = {
1590 .name = "hrtimers:prepare",
1591 .startup.single = hrtimers_prepare_cpu,
1592 .teardown.single = hrtimers_dead_cpu,
1594 [CPUHP_SMPCFD_PREPARE] = {
1595 .name = "smpcfd:prepare",
1596 .startup.single = smpcfd_prepare_cpu,
1597 .teardown.single = smpcfd_dead_cpu,
1599 [CPUHP_RELAY_PREPARE] = {
1600 .name = "relay:prepare",
1601 .startup.single = relay_prepare_cpu,
1602 .teardown.single = NULL,
1604 [CPUHP_SLAB_PREPARE] = {
1605 .name = "slab:prepare",
1606 .startup.single = slab_prepare_cpu,
1607 .teardown.single = slab_dead_cpu,
1609 [CPUHP_RCUTREE_PREP] = {
1610 .name = "RCU/tree:prepare",
1611 .startup.single = rcutree_prepare_cpu,
1612 .teardown.single = rcutree_dead_cpu,
1615 * On the tear-down path, timers_dead_cpu() must be invoked
1616 * before blk_mq_queue_reinit_notify() from notify_dead(),
1617 * otherwise a RCU stall occurs.
1619 [CPUHP_TIMERS_PREPARE] = {
1620 .name = "timers:prepare",
1621 .startup.single = timers_prepare_cpu,
1622 .teardown.single = timers_dead_cpu,
1624 /* Kicks the plugged cpu into life */
1625 [CPUHP_BRINGUP_CPU] = {
1626 .name = "cpu:bringup",
1627 .startup.single = bringup_cpu,
1628 .teardown.single = finish_cpu,
1631 /* Final state before CPU kills itself */
1632 [CPUHP_AP_IDLE_DEAD] = {
1633 .name = "idle:dead",
1636 * Last state before CPU enters the idle loop to die. Transient state
1637 * for synchronization.
1639 [CPUHP_AP_OFFLINE] = {
1640 .name = "ap:offline",
1643 /* First state is scheduler control. Interrupts are disabled */
1644 [CPUHP_AP_SCHED_STARTING] = {
1645 .name = "sched:starting",
1646 .startup.single = sched_cpu_starting,
1647 .teardown.single = sched_cpu_dying,
1649 [CPUHP_AP_RCUTREE_DYING] = {
1650 .name = "RCU/tree:dying",
1651 .startup.single = NULL,
1652 .teardown.single = rcutree_dying_cpu,
1654 [CPUHP_AP_SMPCFD_DYING] = {
1655 .name = "smpcfd:dying",
1656 .startup.single = NULL,
1657 .teardown.single = smpcfd_dying_cpu,
1659 /* Entry state on starting. Interrupts enabled from here on. Transient
1660 * state for synchronsization */
1661 [CPUHP_AP_ONLINE] = {
1662 .name = "ap:online",
1665 * Handled on controll processor until the plugged processor manages
1668 [CPUHP_TEARDOWN_CPU] = {
1669 .name = "cpu:teardown",
1670 .startup.single = NULL,
1671 .teardown.single = takedown_cpu,
1674 /* Handle smpboot threads park/unpark */
1675 [CPUHP_AP_SMPBOOT_THREADS] = {
1676 .name = "smpboot/threads:online",
1677 .startup.single = smpboot_unpark_threads,
1678 .teardown.single = smpboot_park_threads,
1680 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1681 .name = "irq/affinity:online",
1682 .startup.single = irq_affinity_online_cpu,
1683 .teardown.single = NULL,
1685 [CPUHP_AP_PERF_ONLINE] = {
1686 .name = "perf:online",
1687 .startup.single = perf_event_init_cpu,
1688 .teardown.single = perf_event_exit_cpu,
1690 [CPUHP_AP_WATCHDOG_ONLINE] = {
1691 .name = "lockup_detector:online",
1692 .startup.single = lockup_detector_online_cpu,
1693 .teardown.single = lockup_detector_offline_cpu,
1695 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1696 .name = "workqueue:online",
1697 .startup.single = workqueue_online_cpu,
1698 .teardown.single = workqueue_offline_cpu,
1700 [CPUHP_AP_RCUTREE_ONLINE] = {
1701 .name = "RCU/tree:online",
1702 .startup.single = rcutree_online_cpu,
1703 .teardown.single = rcutree_offline_cpu,
1707 * The dynamically registered state space is here
1711 /* Last state is scheduler control setting the cpu active */
1712 [CPUHP_AP_ACTIVE] = {
1713 .name = "sched:active",
1714 .startup.single = sched_cpu_activate,
1715 .teardown.single = sched_cpu_deactivate,
1719 /* CPU is fully up and running. */
1722 .startup.single = NULL,
1723 .teardown.single = NULL,
1727 /* Sanity check for callbacks */
1728 static int cpuhp_cb_check(enum cpuhp_state state)
1730 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1736 * Returns a free for dynamic slot assignment of the Online state. The states
1737 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1738 * by having no name assigned.
1740 static int cpuhp_reserve_state(enum cpuhp_state state)
1742 enum cpuhp_state i, end;
1743 struct cpuhp_step *step;
1746 case CPUHP_AP_ONLINE_DYN:
1747 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1748 end = CPUHP_AP_ONLINE_DYN_END;
1750 case CPUHP_BP_PREPARE_DYN:
1751 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1752 end = CPUHP_BP_PREPARE_DYN_END;
1758 for (i = state; i <= end; i++, step++) {
1762 WARN(1, "No more dynamic states available for CPU hotplug\n");
1766 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1767 int (*startup)(unsigned int cpu),
1768 int (*teardown)(unsigned int cpu),
1769 bool multi_instance)
1771 /* (Un)Install the callbacks for further cpu hotplug operations */
1772 struct cpuhp_step *sp;
1776 * If name is NULL, then the state gets removed.
1778 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1779 * the first allocation from these dynamic ranges, so the removal
1780 * would trigger a new allocation and clear the wrong (already
1781 * empty) state, leaving the callbacks of the to be cleared state
1782 * dangling, which causes wreckage on the next hotplug operation.
1784 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1785 state == CPUHP_BP_PREPARE_DYN)) {
1786 ret = cpuhp_reserve_state(state);
1791 sp = cpuhp_get_step(state);
1792 if (name && sp->name)
1795 sp->startup.single = startup;
1796 sp->teardown.single = teardown;
1798 sp->multi_instance = multi_instance;
1799 INIT_HLIST_HEAD(&sp->list);
1803 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1805 return cpuhp_get_step(state)->teardown.single;
1809 * Call the startup/teardown function for a step either on the AP or
1810 * on the current CPU.
1812 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1813 struct hlist_node *node)
1815 struct cpuhp_step *sp = cpuhp_get_step(state);
1819 * If there's nothing to do, we done.
1820 * Relies on the union for multi_instance.
1822 if ((bringup && !sp->startup.single) ||
1823 (!bringup && !sp->teardown.single))
1826 * The non AP bound callbacks can fail on bringup. On teardown
1827 * e.g. module removal we crash for now.
1830 if (cpuhp_is_ap_state(state))
1831 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1833 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1835 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1837 BUG_ON(ret && !bringup);
1842 * Called from __cpuhp_setup_state on a recoverable failure.
1844 * Note: The teardown callbacks for rollback are not allowed to fail!
1846 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1847 struct hlist_node *node)
1851 /* Roll back the already executed steps on the other cpus */
1852 for_each_present_cpu(cpu) {
1853 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1854 int cpustate = st->state;
1856 if (cpu >= failedcpu)
1859 /* Did we invoke the startup call on that cpu ? */
1860 if (cpustate >= state)
1861 cpuhp_issue_call(cpu, state, false, node);
1865 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1866 struct hlist_node *node,
1869 struct cpuhp_step *sp;
1873 lockdep_assert_cpus_held();
1875 sp = cpuhp_get_step(state);
1876 if (sp->multi_instance == false)
1879 mutex_lock(&cpuhp_state_mutex);
1881 if (!invoke || !sp->startup.multi)
1885 * Try to call the startup callback for each present cpu
1886 * depending on the hotplug state of the cpu.
1888 for_each_present_cpu(cpu) {
1889 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1890 int cpustate = st->state;
1892 if (cpustate < state)
1895 ret = cpuhp_issue_call(cpu, state, true, node);
1897 if (sp->teardown.multi)
1898 cpuhp_rollback_install(cpu, state, node);
1904 hlist_add_head(node, &sp->list);
1906 mutex_unlock(&cpuhp_state_mutex);
1910 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1916 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1920 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1923 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1924 * @state: The state to setup
1925 * @invoke: If true, the startup function is invoked for cpus where
1926 * cpu state >= @state
1927 * @startup: startup callback function
1928 * @teardown: teardown callback function
1929 * @multi_instance: State is set up for multiple instances which get
1932 * The caller needs to hold cpus read locked while calling this function.
1935 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1936 * 0 for all other states
1937 * On failure: proper (negative) error code
1939 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1940 const char *name, bool invoke,
1941 int (*startup)(unsigned int cpu),
1942 int (*teardown)(unsigned int cpu),
1943 bool multi_instance)
1948 lockdep_assert_cpus_held();
1950 if (cpuhp_cb_check(state) || !name)
1953 mutex_lock(&cpuhp_state_mutex);
1955 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1958 dynstate = state == CPUHP_AP_ONLINE_DYN;
1959 if (ret > 0 && dynstate) {
1964 if (ret || !invoke || !startup)
1968 * Try to call the startup callback for each present cpu
1969 * depending on the hotplug state of the cpu.
1971 for_each_present_cpu(cpu) {
1972 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1973 int cpustate = st->state;
1975 if (cpustate < state)
1978 ret = cpuhp_issue_call(cpu, state, true, NULL);
1981 cpuhp_rollback_install(cpu, state, NULL);
1982 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1987 mutex_unlock(&cpuhp_state_mutex);
1989 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1990 * dynamically allocated state in case of success.
1992 if (!ret && dynstate)
1996 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1998 int __cpuhp_setup_state(enum cpuhp_state state,
1999 const char *name, bool invoke,
2000 int (*startup)(unsigned int cpu),
2001 int (*teardown)(unsigned int cpu),
2002 bool multi_instance)
2007 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2008 teardown, multi_instance);
2012 EXPORT_SYMBOL(__cpuhp_setup_state);
2014 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2015 struct hlist_node *node, bool invoke)
2017 struct cpuhp_step *sp = cpuhp_get_step(state);
2020 BUG_ON(cpuhp_cb_check(state));
2022 if (!sp->multi_instance)
2026 mutex_lock(&cpuhp_state_mutex);
2028 if (!invoke || !cpuhp_get_teardown_cb(state))
2031 * Call the teardown callback for each present cpu depending
2032 * on the hotplug state of the cpu. This function is not
2033 * allowed to fail currently!
2035 for_each_present_cpu(cpu) {
2036 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2037 int cpustate = st->state;
2039 if (cpustate >= state)
2040 cpuhp_issue_call(cpu, state, false, node);
2045 mutex_unlock(&cpuhp_state_mutex);
2050 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2053 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2054 * @state: The state to remove
2055 * @invoke: If true, the teardown function is invoked for cpus where
2056 * cpu state >= @state
2058 * The caller needs to hold cpus read locked while calling this function.
2059 * The teardown callback is currently not allowed to fail. Think
2060 * about module removal!
2062 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2064 struct cpuhp_step *sp = cpuhp_get_step(state);
2067 BUG_ON(cpuhp_cb_check(state));
2069 lockdep_assert_cpus_held();
2071 mutex_lock(&cpuhp_state_mutex);
2072 if (sp->multi_instance) {
2073 WARN(!hlist_empty(&sp->list),
2074 "Error: Removing state %d which has instances left.\n",
2079 if (!invoke || !cpuhp_get_teardown_cb(state))
2083 * Call the teardown callback for each present cpu depending
2084 * on the hotplug state of the cpu. This function is not
2085 * allowed to fail currently!
2087 for_each_present_cpu(cpu) {
2088 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2089 int cpustate = st->state;
2091 if (cpustate >= state)
2092 cpuhp_issue_call(cpu, state, false, NULL);
2095 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2096 mutex_unlock(&cpuhp_state_mutex);
2098 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2100 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2103 __cpuhp_remove_state_cpuslocked(state, invoke);
2106 EXPORT_SYMBOL(__cpuhp_remove_state);
2108 #ifdef CONFIG_HOTPLUG_SMT
2109 static void cpuhp_offline_cpu_device(unsigned int cpu)
2111 struct device *dev = get_cpu_device(cpu);
2113 dev->offline = true;
2114 /* Tell user space about the state change */
2115 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2118 static void cpuhp_online_cpu_device(unsigned int cpu)
2120 struct device *dev = get_cpu_device(cpu);
2122 dev->offline = false;
2123 /* Tell user space about the state change */
2124 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2127 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2131 cpu_maps_update_begin();
2132 for_each_online_cpu(cpu) {
2133 if (topology_is_primary_thread(cpu))
2135 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2139 * As this needs to hold the cpu maps lock it's impossible
2140 * to call device_offline() because that ends up calling
2141 * cpu_down() which takes cpu maps lock. cpu maps lock
2142 * needs to be held as this might race against in kernel
2143 * abusers of the hotplug machinery (thermal management).
2145 * So nothing would update device:offline state. That would
2146 * leave the sysfs entry stale and prevent onlining after
2147 * smt control has been changed to 'off' again. This is
2148 * called under the sysfs hotplug lock, so it is properly
2149 * serialized against the regular offline usage.
2151 cpuhp_offline_cpu_device(cpu);
2154 cpu_smt_control = ctrlval;
2155 cpu_maps_update_done();
2159 int cpuhp_smt_enable(void)
2163 cpu_maps_update_begin();
2164 cpu_smt_control = CPU_SMT_ENABLED;
2165 for_each_present_cpu(cpu) {
2166 /* Skip online CPUs and CPUs on offline nodes */
2167 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2169 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2172 /* See comment in cpuhp_smt_disable() */
2173 cpuhp_online_cpu_device(cpu);
2175 cpu_maps_update_done();
2180 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2181 static ssize_t show_cpuhp_state(struct device *dev,
2182 struct device_attribute *attr, char *buf)
2184 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2186 return sprintf(buf, "%d\n", st->state);
2188 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
2190 static ssize_t write_cpuhp_target(struct device *dev,
2191 struct device_attribute *attr,
2192 const char *buf, size_t count)
2194 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2195 struct cpuhp_step *sp;
2198 ret = kstrtoint(buf, 10, &target);
2202 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2203 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2206 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2210 ret = lock_device_hotplug_sysfs();
2214 mutex_lock(&cpuhp_state_mutex);
2215 sp = cpuhp_get_step(target);
2216 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2217 mutex_unlock(&cpuhp_state_mutex);
2221 if (st->state < target)
2222 ret = cpu_up(dev->id, target);
2224 ret = cpu_down(dev->id, target);
2226 unlock_device_hotplug();
2227 return ret ? ret : count;
2230 static ssize_t show_cpuhp_target(struct device *dev,
2231 struct device_attribute *attr, char *buf)
2233 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2235 return sprintf(buf, "%d\n", st->target);
2237 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
2240 static ssize_t write_cpuhp_fail(struct device *dev,
2241 struct device_attribute *attr,
2242 const char *buf, size_t count)
2244 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2245 struct cpuhp_step *sp;
2248 ret = kstrtoint(buf, 10, &fail);
2252 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2256 * Cannot fail STARTING/DYING callbacks.
2258 if (cpuhp_is_atomic_state(fail))
2262 * Cannot fail anything that doesn't have callbacks.
2264 mutex_lock(&cpuhp_state_mutex);
2265 sp = cpuhp_get_step(fail);
2266 if (!sp->startup.single && !sp->teardown.single)
2268 mutex_unlock(&cpuhp_state_mutex);
2277 static ssize_t show_cpuhp_fail(struct device *dev,
2278 struct device_attribute *attr, char *buf)
2280 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2282 return sprintf(buf, "%d\n", st->fail);
2285 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
2287 static struct attribute *cpuhp_cpu_attrs[] = {
2288 &dev_attr_state.attr,
2289 &dev_attr_target.attr,
2290 &dev_attr_fail.attr,
2294 static const struct attribute_group cpuhp_cpu_attr_group = {
2295 .attrs = cpuhp_cpu_attrs,
2300 static ssize_t show_cpuhp_states(struct device *dev,
2301 struct device_attribute *attr, char *buf)
2303 ssize_t cur, res = 0;
2306 mutex_lock(&cpuhp_state_mutex);
2307 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2308 struct cpuhp_step *sp = cpuhp_get_step(i);
2311 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2316 mutex_unlock(&cpuhp_state_mutex);
2319 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2321 static struct attribute *cpuhp_cpu_root_attrs[] = {
2322 &dev_attr_states.attr,
2326 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2327 .attrs = cpuhp_cpu_root_attrs,
2332 #ifdef CONFIG_HOTPLUG_SMT
2335 __store_smt_control(struct device *dev, struct device_attribute *attr,
2336 const char *buf, size_t count)
2340 if (sysfs_streq(buf, "on"))
2341 ctrlval = CPU_SMT_ENABLED;
2342 else if (sysfs_streq(buf, "off"))
2343 ctrlval = CPU_SMT_DISABLED;
2344 else if (sysfs_streq(buf, "forceoff"))
2345 ctrlval = CPU_SMT_FORCE_DISABLED;
2349 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2352 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2355 ret = lock_device_hotplug_sysfs();
2359 if (ctrlval != cpu_smt_control) {
2361 case CPU_SMT_ENABLED:
2362 ret = cpuhp_smt_enable();
2364 case CPU_SMT_DISABLED:
2365 case CPU_SMT_FORCE_DISABLED:
2366 ret = cpuhp_smt_disable(ctrlval);
2371 unlock_device_hotplug();
2372 return ret ? ret : count;
2375 #else /* !CONFIG_HOTPLUG_SMT */
2377 __store_smt_control(struct device *dev, struct device_attribute *attr,
2378 const char *buf, size_t count)
2382 #endif /* CONFIG_HOTPLUG_SMT */
2384 static const char *smt_states[] = {
2385 [CPU_SMT_ENABLED] = "on",
2386 [CPU_SMT_DISABLED] = "off",
2387 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2388 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2389 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2393 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2395 const char *state = smt_states[cpu_smt_control];
2397 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2401 store_smt_control(struct device *dev, struct device_attribute *attr,
2402 const char *buf, size_t count)
2404 return __store_smt_control(dev, attr, buf, count);
2406 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2409 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2411 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2413 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2415 static struct attribute *cpuhp_smt_attrs[] = {
2416 &dev_attr_control.attr,
2417 &dev_attr_active.attr,
2421 static const struct attribute_group cpuhp_smt_attr_group = {
2422 .attrs = cpuhp_smt_attrs,
2427 static int __init cpu_smt_sysfs_init(void)
2429 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2430 &cpuhp_smt_attr_group);
2433 static int __init cpuhp_sysfs_init(void)
2437 ret = cpu_smt_sysfs_init();
2441 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2442 &cpuhp_cpu_root_attr_group);
2446 for_each_possible_cpu(cpu) {
2447 struct device *dev = get_cpu_device(cpu);
2451 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2457 device_initcall(cpuhp_sysfs_init);
2458 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2461 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2462 * represents all NR_CPUS bits binary values of 1<<nr.
2464 * It is used by cpumask_of() to get a constant address to a CPU
2465 * mask value that has a single bit set only.
2468 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2469 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2470 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2471 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2472 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2474 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2476 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2477 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2478 #if BITS_PER_LONG > 32
2479 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2480 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2483 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2485 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2486 EXPORT_SYMBOL(cpu_all_bits);
2488 #ifdef CONFIG_INIT_ALL_POSSIBLE
2489 struct cpumask __cpu_possible_mask __read_mostly
2492 struct cpumask __cpu_possible_mask __read_mostly;
2494 EXPORT_SYMBOL(__cpu_possible_mask);
2496 struct cpumask __cpu_online_mask __read_mostly;
2497 EXPORT_SYMBOL(__cpu_online_mask);
2499 struct cpumask __cpu_present_mask __read_mostly;
2500 EXPORT_SYMBOL(__cpu_present_mask);
2502 struct cpumask __cpu_active_mask __read_mostly;
2503 EXPORT_SYMBOL(__cpu_active_mask);
2505 atomic_t __num_online_cpus __read_mostly;
2506 EXPORT_SYMBOL(__num_online_cpus);
2508 void init_cpu_present(const struct cpumask *src)
2510 cpumask_copy(&__cpu_present_mask, src);
2513 void init_cpu_possible(const struct cpumask *src)
2515 cpumask_copy(&__cpu_possible_mask, src);
2518 void init_cpu_online(const struct cpumask *src)
2520 cpumask_copy(&__cpu_online_mask, src);
2523 void set_cpu_online(unsigned int cpu, bool online)
2526 * atomic_inc/dec() is required to handle the horrid abuse of this
2527 * function by the reboot and kexec code which invoke it from
2528 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2529 * regular CPU hotplug is properly serialized.
2531 * Note, that the fact that __num_online_cpus is of type atomic_t
2532 * does not protect readers which are not serialized against
2533 * concurrent hotplug operations.
2536 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2537 atomic_inc(&__num_online_cpus);
2539 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2540 atomic_dec(&__num_online_cpus);
2545 * Activate the first processor.
2547 void __init boot_cpu_init(void)
2549 int cpu = smp_processor_id();
2551 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2552 set_cpu_online(cpu, true);
2553 set_cpu_active(cpu, true);
2554 set_cpu_present(cpu, true);
2555 set_cpu_possible(cpu, true);
2558 __boot_cpu_id = cpu;
2563 * Must be called _AFTER_ setting up the per_cpu areas
2565 void __init boot_cpu_hotplug_init(void)
2568 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2570 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2574 * These are used for a global "mitigations=" cmdline option for toggling
2575 * optional CPU mitigations.
2577 enum cpu_mitigations {
2578 CPU_MITIGATIONS_OFF,
2579 CPU_MITIGATIONS_AUTO,
2580 CPU_MITIGATIONS_AUTO_NOSMT,
2583 static enum cpu_mitigations cpu_mitigations __ro_after_init =
2584 CPU_MITIGATIONS_AUTO;
2586 static int __init mitigations_parse_cmdline(char *arg)
2588 if (!strcmp(arg, "off"))
2589 cpu_mitigations = CPU_MITIGATIONS_OFF;
2590 else if (!strcmp(arg, "auto"))
2591 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2592 else if (!strcmp(arg, "auto,nosmt"))
2593 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2595 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2600 early_param("mitigations", mitigations_parse_cmdline);
2602 /* mitigations=off */
2603 bool cpu_mitigations_off(void)
2605 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2607 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2609 /* mitigations=auto,nosmt */
2610 bool cpu_mitigations_auto_nosmt(void)
2612 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2614 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);