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>
37 #include <linux/random.h>
39 #include <trace/events/power.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/cpuhp.h>
46 * struct cpuhp_cpu_state - Per cpu hotplug state storage
47 * @state: The current cpu state
48 * @target: The target state
49 * @fail: Current CPU hotplug callback state
50 * @thread: Pointer to the hotplug thread
51 * @should_run: Thread should execute
52 * @rollback: Perform a rollback
53 * @single: Single callback invocation
54 * @bringup: Single callback bringup or teardown selector
56 * @node: Remote CPU node; for multi-instance, do a
57 * single entry callback for install/remove
58 * @last: For multi-instance rollback, remember how far we got
59 * @cb_state: The state for a single callback (install/uninstall)
60 * @result: Result of the operation
61 * @done_up: Signal completion to the issuer of the task for cpu-up
62 * @done_down: Signal completion to the issuer of the task for cpu-down
64 struct cpuhp_cpu_state {
65 enum cpuhp_state state;
66 enum cpuhp_state target;
67 enum cpuhp_state fail;
69 struct task_struct *thread;
74 struct hlist_node *node;
75 struct hlist_node *last;
76 enum cpuhp_state cb_state;
78 struct completion done_up;
79 struct completion done_down;
83 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
84 .fail = CPUHP_INVALID,
88 cpumask_t cpus_booted_once_mask;
91 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
92 static struct lockdep_map cpuhp_state_up_map =
93 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
94 static struct lockdep_map cpuhp_state_down_map =
95 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
98 static inline void cpuhp_lock_acquire(bool bringup)
100 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
103 static inline void cpuhp_lock_release(bool bringup)
105 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
109 static inline void cpuhp_lock_acquire(bool bringup) { }
110 static inline void cpuhp_lock_release(bool bringup) { }
115 * struct cpuhp_step - Hotplug state machine step
116 * @name: Name of the step
117 * @startup: Startup function of the step
118 * @teardown: Teardown function of the step
119 * @cant_stop: Bringup/teardown can't be stopped at this step
120 * @multi_instance: State has multiple instances which get added afterwards
125 int (*single)(unsigned int cpu);
126 int (*multi)(unsigned int cpu,
127 struct hlist_node *node);
130 int (*single)(unsigned int cpu);
131 int (*multi)(unsigned int cpu,
132 struct hlist_node *node);
135 struct hlist_head list;
141 static DEFINE_MUTEX(cpuhp_state_mutex);
142 static struct cpuhp_step cpuhp_hp_states[];
144 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
146 return cpuhp_hp_states + state;
149 static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step)
151 return bringup ? !step->startup.single : !step->teardown.single;
155 * cpuhp_invoke_callback - Invoke the callbacks for a given state
156 * @cpu: The cpu for which the callback should be invoked
157 * @state: The state to do callbacks for
158 * @bringup: True if the bringup callback should be invoked
159 * @node: For multi-instance, do a single entry callback for install/remove
160 * @lastp: For multi-instance rollback, remember how far we got
162 * Called from cpu hotplug and from the state register machinery.
164 * Return: %0 on success or a negative errno code
166 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
167 bool bringup, struct hlist_node *node,
168 struct hlist_node **lastp)
170 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
171 struct cpuhp_step *step = cpuhp_get_step(state);
172 int (*cbm)(unsigned int cpu, struct hlist_node *node);
173 int (*cb)(unsigned int cpu);
176 if (st->fail == state) {
177 st->fail = CPUHP_INVALID;
181 if (cpuhp_step_empty(bringup, step)) {
186 if (!step->multi_instance) {
187 WARN_ON_ONCE(lastp && *lastp);
188 cb = bringup ? step->startup.single : step->teardown.single;
190 trace_cpuhp_enter(cpu, st->target, state, cb);
192 trace_cpuhp_exit(cpu, st->state, state, ret);
195 cbm = bringup ? step->startup.multi : step->teardown.multi;
197 /* Single invocation for instance add/remove */
199 WARN_ON_ONCE(lastp && *lastp);
200 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
201 ret = cbm(cpu, node);
202 trace_cpuhp_exit(cpu, st->state, state, ret);
206 /* State transition. Invoke on all instances */
208 hlist_for_each(node, &step->list) {
209 if (lastp && node == *lastp)
212 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
213 ret = cbm(cpu, node);
214 trace_cpuhp_exit(cpu, st->state, state, ret);
228 /* Rollback the instances if one failed */
229 cbm = !bringup ? step->startup.multi : step->teardown.multi;
233 hlist_for_each(node, &step->list) {
237 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
238 ret = cbm(cpu, node);
239 trace_cpuhp_exit(cpu, st->state, state, ret);
241 * Rollback must not fail,
249 static bool cpuhp_is_ap_state(enum cpuhp_state state)
252 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
253 * purposes as that state is handled explicitly in cpu_down.
255 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
258 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
260 struct completion *done = bringup ? &st->done_up : &st->done_down;
261 wait_for_completion(done);
264 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
266 struct completion *done = bringup ? &st->done_up : &st->done_down;
271 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
273 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
275 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
278 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
279 static DEFINE_MUTEX(cpu_add_remove_lock);
280 bool cpuhp_tasks_frozen;
281 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
284 * The following two APIs (cpu_maps_update_begin/done) must be used when
285 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
287 void cpu_maps_update_begin(void)
289 mutex_lock(&cpu_add_remove_lock);
292 void cpu_maps_update_done(void)
294 mutex_unlock(&cpu_add_remove_lock);
298 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
299 * Should always be manipulated under cpu_add_remove_lock
301 static int cpu_hotplug_disabled;
303 #ifdef CONFIG_HOTPLUG_CPU
305 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
307 void cpus_read_lock(void)
309 percpu_down_read(&cpu_hotplug_lock);
311 EXPORT_SYMBOL_GPL(cpus_read_lock);
313 int cpus_read_trylock(void)
315 return percpu_down_read_trylock(&cpu_hotplug_lock);
317 EXPORT_SYMBOL_GPL(cpus_read_trylock);
319 void cpus_read_unlock(void)
321 percpu_up_read(&cpu_hotplug_lock);
323 EXPORT_SYMBOL_GPL(cpus_read_unlock);
325 void cpus_write_lock(void)
327 percpu_down_write(&cpu_hotplug_lock);
330 void cpus_write_unlock(void)
332 percpu_up_write(&cpu_hotplug_lock);
335 void lockdep_assert_cpus_held(void)
338 * We can't have hotplug operations before userspace starts running,
339 * and some init codepaths will knowingly not take the hotplug lock.
340 * This is all valid, so mute lockdep until it makes sense to report
343 if (system_state < SYSTEM_RUNNING)
346 percpu_rwsem_assert_held(&cpu_hotplug_lock);
349 #ifdef CONFIG_LOCKDEP
350 int lockdep_is_cpus_held(void)
352 return percpu_rwsem_is_held(&cpu_hotplug_lock);
356 static void lockdep_acquire_cpus_lock(void)
358 rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
361 static void lockdep_release_cpus_lock(void)
363 rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
367 * Wait for currently running CPU hotplug operations to complete (if any) and
368 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
369 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
370 * hotplug path before performing hotplug operations. So acquiring that lock
371 * guarantees mutual exclusion from any currently running hotplug operations.
373 void cpu_hotplug_disable(void)
375 cpu_maps_update_begin();
376 cpu_hotplug_disabled++;
377 cpu_maps_update_done();
379 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
381 static void __cpu_hotplug_enable(void)
383 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
385 cpu_hotplug_disabled--;
388 void cpu_hotplug_enable(void)
390 cpu_maps_update_begin();
391 __cpu_hotplug_enable();
392 cpu_maps_update_done();
394 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
398 static void lockdep_acquire_cpus_lock(void)
402 static void lockdep_release_cpus_lock(void)
406 #endif /* CONFIG_HOTPLUG_CPU */
409 * Architectures that need SMT-specific errata handling during SMT hotplug
410 * should override this.
412 void __weak arch_smt_update(void) { }
414 #ifdef CONFIG_HOTPLUG_SMT
415 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
417 void __init cpu_smt_disable(bool force)
419 if (!cpu_smt_possible())
423 pr_info("SMT: Force disabled\n");
424 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
426 pr_info("SMT: disabled\n");
427 cpu_smt_control = CPU_SMT_DISABLED;
432 * The decision whether SMT is supported can only be done after the full
433 * CPU identification. Called from architecture code.
435 void __init cpu_smt_check_topology(void)
437 if (!topology_smt_supported())
438 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
441 static int __init smt_cmdline_disable(char *str)
443 cpu_smt_disable(str && !strcmp(str, "force"));
446 early_param("nosmt", smt_cmdline_disable);
448 static inline bool cpu_smt_allowed(unsigned int cpu)
450 if (cpu_smt_control == CPU_SMT_ENABLED)
453 if (topology_is_primary_thread(cpu))
457 * On x86 it's required to boot all logical CPUs at least once so
458 * that the init code can get a chance to set CR4.MCE on each
459 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
460 * core will shutdown the machine.
462 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
465 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
466 bool cpu_smt_possible(void)
468 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
469 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
471 EXPORT_SYMBOL_GPL(cpu_smt_possible);
473 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
476 static inline enum cpuhp_state
477 cpuhp_set_state(int cpu, struct cpuhp_cpu_state *st, enum cpuhp_state target)
479 enum cpuhp_state prev_state = st->state;
480 bool bringup = st->state < target;
482 st->rollback = false;
487 st->bringup = bringup;
488 if (cpu_dying(cpu) != !bringup)
489 set_cpu_dying(cpu, !bringup);
495 cpuhp_reset_state(int cpu, struct cpuhp_cpu_state *st,
496 enum cpuhp_state prev_state)
498 bool bringup = !st->bringup;
500 st->target = prev_state;
503 * Already rolling back. No need invert the bringup value or to change
512 * If we have st->last we need to undo partial multi_instance of this
513 * state first. Otherwise start undo at the previous state.
522 st->bringup = bringup;
523 if (cpu_dying(cpu) != !bringup)
524 set_cpu_dying(cpu, !bringup);
527 /* Regular hotplug invocation of the AP hotplug thread */
528 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
530 if (!st->single && st->state == st->target)
535 * Make sure the above stores are visible before should_run becomes
536 * true. Paired with the mb() above in cpuhp_thread_fun()
539 st->should_run = true;
540 wake_up_process(st->thread);
541 wait_for_ap_thread(st, st->bringup);
544 static int cpuhp_kick_ap(int cpu, struct cpuhp_cpu_state *st,
545 enum cpuhp_state target)
547 enum cpuhp_state prev_state;
550 prev_state = cpuhp_set_state(cpu, st, target);
552 if ((ret = st->result)) {
553 cpuhp_reset_state(cpu, st, prev_state);
560 static int bringup_wait_for_ap(unsigned int cpu)
562 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
564 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
565 wait_for_ap_thread(st, true);
566 if (WARN_ON_ONCE((!cpu_online(cpu))))
569 /* Unpark the hotplug thread of the target cpu */
570 kthread_unpark(st->thread);
573 * SMT soft disabling on X86 requires to bring the CPU out of the
574 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
575 * CPU marked itself as booted_once in notify_cpu_starting() so the
576 * cpu_smt_allowed() check will now return false if this is not the
579 if (!cpu_smt_allowed(cpu))
582 if (st->target <= CPUHP_AP_ONLINE_IDLE)
585 return cpuhp_kick_ap(cpu, st, st->target);
588 static int bringup_cpu(unsigned int cpu)
590 struct task_struct *idle = idle_thread_get(cpu);
594 * Reset stale stack state from the last time this CPU was online.
596 scs_task_reset(idle);
597 kasan_unpoison_task_stack(idle);
600 * Some architectures have to walk the irq descriptors to
601 * setup the vector space for the cpu which comes online.
602 * Prevent irq alloc/free across the bringup.
606 /* Arch-specific enabling code. */
607 ret = __cpu_up(cpu, idle);
611 return bringup_wait_for_ap(cpu);
614 static int finish_cpu(unsigned int cpu)
616 struct task_struct *idle = idle_thread_get(cpu);
617 struct mm_struct *mm = idle->active_mm;
620 * idle_task_exit() will have switched to &init_mm, now
621 * clean up any remaining active_mm state.
624 idle->active_mm = &init_mm;
630 * Hotplug state machine related functions
634 * Get the next state to run. Empty ones will be skipped. Returns true if a
637 * st->state will be modified ahead of time, to match state_to_run, as if it
640 static bool cpuhp_next_state(bool bringup,
641 enum cpuhp_state *state_to_run,
642 struct cpuhp_cpu_state *st,
643 enum cpuhp_state target)
647 if (st->state >= target)
650 *state_to_run = ++st->state;
652 if (st->state <= target)
655 *state_to_run = st->state--;
658 if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run)))
665 static int __cpuhp_invoke_callback_range(bool bringup,
667 struct cpuhp_cpu_state *st,
668 enum cpuhp_state target,
671 enum cpuhp_state state;
674 while (cpuhp_next_state(bringup, &state, st, target)) {
677 err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL);
682 pr_warn("CPU %u %s state %s (%d) failed (%d)\n",
683 cpu, bringup ? "UP" : "DOWN",
684 cpuhp_get_step(st->state)->name,
696 static inline int cpuhp_invoke_callback_range(bool bringup,
698 struct cpuhp_cpu_state *st,
699 enum cpuhp_state target)
701 return __cpuhp_invoke_callback_range(bringup, cpu, st, target, false);
704 static inline void cpuhp_invoke_callback_range_nofail(bool bringup,
706 struct cpuhp_cpu_state *st,
707 enum cpuhp_state target)
709 __cpuhp_invoke_callback_range(bringup, cpu, st, target, true);
712 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
714 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
717 * When CPU hotplug is disabled, then taking the CPU down is not
718 * possible because takedown_cpu() and the architecture and
719 * subsystem specific mechanisms are not available. So the CPU
720 * which would be completely unplugged again needs to stay around
721 * in the current state.
723 return st->state <= CPUHP_BRINGUP_CPU;
726 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
727 enum cpuhp_state target)
729 enum cpuhp_state prev_state = st->state;
732 ret = cpuhp_invoke_callback_range(true, cpu, st, target);
734 pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n",
735 ret, cpu, cpuhp_get_step(st->state)->name,
738 cpuhp_reset_state(cpu, st, prev_state);
739 if (can_rollback_cpu(st))
740 WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
747 * The cpu hotplug threads manage the bringup and teardown of the cpus
749 static void cpuhp_create(unsigned int cpu)
751 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
753 init_completion(&st->done_up);
754 init_completion(&st->done_down);
757 static int cpuhp_should_run(unsigned int cpu)
759 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
761 return st->should_run;
765 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
766 * callbacks when a state gets [un]installed at runtime.
768 * Each invocation of this function by the smpboot thread does a single AP
771 * It has 3 modes of operation:
772 * - single: runs st->cb_state
773 * - up: runs ++st->state, while st->state < st->target
774 * - down: runs st->state--, while st->state > st->target
776 * When complete or on error, should_run is cleared and the completion is fired.
778 static void cpuhp_thread_fun(unsigned int cpu)
780 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
781 bool bringup = st->bringup;
782 enum cpuhp_state state;
784 if (WARN_ON_ONCE(!st->should_run))
788 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
789 * that if we see ->should_run we also see the rest of the state.
794 * The BP holds the hotplug lock, but we're now running on the AP,
795 * ensure that anybody asserting the lock is held, will actually find
798 lockdep_acquire_cpus_lock();
799 cpuhp_lock_acquire(bringup);
802 state = st->cb_state;
803 st->should_run = false;
805 st->should_run = cpuhp_next_state(bringup, &state, st, st->target);
810 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
812 if (cpuhp_is_atomic_state(state)) {
814 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
818 * STARTING/DYING must not fail!
820 WARN_ON_ONCE(st->result);
822 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
827 * If we fail on a rollback, we're up a creek without no
828 * paddle, no way forward, no way back. We loose, thanks for
831 WARN_ON_ONCE(st->rollback);
832 st->should_run = false;
836 cpuhp_lock_release(bringup);
837 lockdep_release_cpus_lock();
840 complete_ap_thread(st, bringup);
843 /* Invoke a single callback on a remote cpu */
845 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
846 struct hlist_node *node)
848 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
851 if (!cpu_online(cpu))
854 cpuhp_lock_acquire(false);
855 cpuhp_lock_release(false);
857 cpuhp_lock_acquire(true);
858 cpuhp_lock_release(true);
861 * If we are up and running, use the hotplug thread. For early calls
862 * we invoke the thread function directly.
865 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
867 st->rollback = false;
871 st->bringup = bringup;
872 st->cb_state = state;
878 * If we failed and did a partial, do a rollback.
880 if ((ret = st->result) && st->last) {
882 st->bringup = !bringup;
888 * Clean up the leftovers so the next hotplug operation wont use stale
891 st->node = st->last = NULL;
895 static int cpuhp_kick_ap_work(unsigned int cpu)
897 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
898 enum cpuhp_state prev_state = st->state;
901 cpuhp_lock_acquire(false);
902 cpuhp_lock_release(false);
904 cpuhp_lock_acquire(true);
905 cpuhp_lock_release(true);
907 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
908 ret = cpuhp_kick_ap(cpu, st, st->target);
909 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
914 static struct smp_hotplug_thread cpuhp_threads = {
915 .store = &cpuhp_state.thread,
916 .create = &cpuhp_create,
917 .thread_should_run = cpuhp_should_run,
918 .thread_fn = cpuhp_thread_fun,
919 .thread_comm = "cpuhp/%u",
923 void __init cpuhp_threads_init(void)
925 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
926 kthread_unpark(this_cpu_read(cpuhp_state.thread));
931 * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
934 * The operation is still serialized against concurrent CPU hotplug via
935 * cpu_add_remove_lock, i.e. CPU map protection. But it is _not_
936 * serialized against other hotplug related activity like adding or
937 * removing of state callbacks and state instances, which invoke either the
938 * startup or the teardown callback of the affected state.
940 * This is required for subsystems which are unfixable vs. CPU hotplug and
941 * evade lock inversion problems by scheduling work which has to be
942 * completed _before_ cpu_up()/_cpu_down() returns.
944 * Don't even think about adding anything to this for any new code or even
945 * drivers. It's only purpose is to keep existing lock order trainwrecks
948 * For cpu_down() there might be valid reasons to finish cleanups which are
949 * not required to be done under cpu_hotplug_lock, but that's a different
950 * story and would be not invoked via this.
952 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
955 * cpusets delegate hotplug operations to a worker to "solve" the
956 * lock order problems. Wait for the worker, but only if tasks are
957 * _not_ frozen (suspend, hibernate) as that would wait forever.
959 * The wait is required because otherwise the hotplug operation
960 * returns with inconsistent state, which could even be observed in
961 * user space when a new CPU is brought up. The CPU plug uevent
962 * would be delivered and user space reacting on it would fail to
963 * move tasks to the newly plugged CPU up to the point where the
964 * work has finished because up to that point the newly plugged CPU
965 * is not assignable in cpusets/cgroups. On unplug that's not
966 * necessarily a visible issue, but it is still inconsistent state,
967 * which is the real problem which needs to be "fixed". This can't
968 * prevent the transient state between scheduling the work and
969 * returning from waiting for it.
972 cpuset_wait_for_hotplug();
975 #ifdef CONFIG_HOTPLUG_CPU
976 #ifndef arch_clear_mm_cpumask_cpu
977 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
981 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
984 * This function walks all processes, finds a valid mm struct for each one and
985 * then clears a corresponding bit in mm's cpumask. While this all sounds
986 * trivial, there are various non-obvious corner cases, which this function
987 * tries to solve in a safe manner.
989 * Also note that the function uses a somewhat relaxed locking scheme, so it may
990 * be called only for an already offlined CPU.
992 void clear_tasks_mm_cpumask(int cpu)
994 struct task_struct *p;
997 * This function is called after the cpu is taken down and marked
998 * offline, so its not like new tasks will ever get this cpu set in
999 * their mm mask. -- Peter Zijlstra
1000 * Thus, we may use rcu_read_lock() here, instead of grabbing
1001 * full-fledged tasklist_lock.
1003 WARN_ON(cpu_online(cpu));
1005 for_each_process(p) {
1006 struct task_struct *t;
1009 * Main thread might exit, but other threads may still have
1010 * a valid mm. Find one.
1012 t = find_lock_task_mm(p);
1015 arch_clear_mm_cpumask_cpu(cpu, t->mm);
1021 /* Take this CPU down. */
1022 static int take_cpu_down(void *_param)
1024 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1025 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
1026 int err, cpu = smp_processor_id();
1028 /* Ensure this CPU doesn't handle any more interrupts. */
1029 err = __cpu_disable();
1034 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
1035 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
1037 WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1));
1040 * Invoke the former CPU_DYING callbacks. DYING must not fail!
1042 cpuhp_invoke_callback_range_nofail(false, cpu, st, target);
1044 /* Give up timekeeping duties */
1045 tick_handover_do_timer();
1046 /* Remove CPU from timer broadcasting */
1047 tick_offline_cpu(cpu);
1048 /* Park the stopper thread */
1049 stop_machine_park(cpu);
1053 static int takedown_cpu(unsigned int cpu)
1055 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1058 /* Park the smpboot threads */
1059 kthread_park(st->thread);
1062 * Prevent irq alloc/free while the dying cpu reorganizes the
1063 * interrupt affinities.
1068 * So now all preempt/rcu users must observe !cpu_active().
1070 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
1072 /* CPU refused to die */
1073 irq_unlock_sparse();
1074 /* Unpark the hotplug thread so we can rollback there */
1075 kthread_unpark(st->thread);
1078 BUG_ON(cpu_online(cpu));
1081 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1082 * all runnable tasks from the CPU, there's only the idle task left now
1083 * that the migration thread is done doing the stop_machine thing.
1085 * Wait for the stop thread to go away.
1087 wait_for_ap_thread(st, false);
1088 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1090 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
1091 irq_unlock_sparse();
1093 hotplug_cpu__broadcast_tick_pull(cpu);
1094 /* This actually kills the CPU. */
1097 tick_cleanup_dead_cpu(cpu);
1098 rcutree_migrate_callbacks(cpu);
1102 static void cpuhp_complete_idle_dead(void *arg)
1104 struct cpuhp_cpu_state *st = arg;
1106 complete_ap_thread(st, false);
1109 void cpuhp_report_idle_dead(void)
1111 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1113 BUG_ON(st->state != CPUHP_AP_OFFLINE);
1114 rcu_report_dead(smp_processor_id());
1115 st->state = CPUHP_AP_IDLE_DEAD;
1117 * We cannot call complete after rcu_report_dead() so we delegate it
1120 smp_call_function_single(cpumask_first(cpu_online_mask),
1121 cpuhp_complete_idle_dead, st, 0);
1124 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1125 enum cpuhp_state target)
1127 enum cpuhp_state prev_state = st->state;
1130 ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1132 pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n",
1133 ret, cpu, cpuhp_get_step(st->state)->name,
1136 cpuhp_reset_state(cpu, st, prev_state);
1138 if (st->state < prev_state)
1139 WARN_ON(cpuhp_invoke_callback_range(true, cpu, st,
1146 /* Requires cpu_add_remove_lock to be held */
1147 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1148 enum cpuhp_state target)
1150 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1151 int prev_state, ret = 0;
1153 if (num_online_cpus() == 1)
1156 if (!cpu_present(cpu))
1161 cpuhp_tasks_frozen = tasks_frozen;
1163 prev_state = cpuhp_set_state(cpu, st, target);
1165 * If the current CPU state is in the range of the AP hotplug thread,
1166 * then we need to kick the thread.
1168 if (st->state > CPUHP_TEARDOWN_CPU) {
1169 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1170 ret = cpuhp_kick_ap_work(cpu);
1172 * The AP side has done the error rollback already. Just
1173 * return the error code..
1179 * We might have stopped still in the range of the AP hotplug
1180 * thread. Nothing to do anymore.
1182 if (st->state > CPUHP_TEARDOWN_CPU)
1185 st->target = target;
1188 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1189 * to do the further cleanups.
1191 ret = cpuhp_down_callbacks(cpu, st, target);
1192 if (ret && st->state < prev_state) {
1193 if (st->state == CPUHP_TEARDOWN_CPU) {
1194 cpuhp_reset_state(cpu, st, prev_state);
1195 __cpuhp_kick_ap(st);
1197 WARN(1, "DEAD callback error for CPU%d", cpu);
1202 cpus_write_unlock();
1204 * Do post unplug cleanup. This is still protected against
1205 * concurrent CPU hotplug via cpu_add_remove_lock.
1207 lockup_detector_cleanup();
1209 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1213 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1215 if (cpu_hotplug_disabled)
1217 return _cpu_down(cpu, 0, target);
1220 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1224 cpu_maps_update_begin();
1225 err = cpu_down_maps_locked(cpu, target);
1226 cpu_maps_update_done();
1231 * cpu_device_down - Bring down a cpu device
1232 * @dev: Pointer to the cpu device to offline
1234 * This function is meant to be used by device core cpu subsystem only.
1236 * Other subsystems should use remove_cpu() instead.
1238 * Return: %0 on success or a negative errno code
1240 int cpu_device_down(struct device *dev)
1242 return cpu_down(dev->id, CPUHP_OFFLINE);
1245 int remove_cpu(unsigned int cpu)
1249 lock_device_hotplug();
1250 ret = device_offline(get_cpu_device(cpu));
1251 unlock_device_hotplug();
1255 EXPORT_SYMBOL_GPL(remove_cpu);
1257 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1262 cpu_maps_update_begin();
1265 * Make certain the cpu I'm about to reboot on is online.
1267 * This is inline to what migrate_to_reboot_cpu() already do.
1269 if (!cpu_online(primary_cpu))
1270 primary_cpu = cpumask_first(cpu_online_mask);
1272 for_each_online_cpu(cpu) {
1273 if (cpu == primary_cpu)
1276 error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1278 pr_err("Failed to offline CPU%d - error=%d",
1285 * Ensure all but the reboot CPU are offline.
1287 BUG_ON(num_online_cpus() > 1);
1290 * Make sure the CPUs won't be enabled by someone else after this
1291 * point. Kexec will reboot to a new kernel shortly resetting
1292 * everything along the way.
1294 cpu_hotplug_disabled++;
1296 cpu_maps_update_done();
1300 #define takedown_cpu NULL
1301 #endif /*CONFIG_HOTPLUG_CPU*/
1304 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1305 * @cpu: cpu that just started
1307 * It must be called by the arch code on the new cpu, before the new cpu
1308 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1310 void notify_cpu_starting(unsigned int cpu)
1312 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1313 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1315 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1316 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1319 * STARTING must not fail!
1321 cpuhp_invoke_callback_range_nofail(true, cpu, st, target);
1325 * Called from the idle task. Wake up the controlling task which brings the
1326 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1327 * online bringup to the hotplug thread.
1329 void cpuhp_online_idle(enum cpuhp_state state)
1331 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1333 /* Happens for the boot cpu */
1334 if (state != CPUHP_AP_ONLINE_IDLE)
1338 * Unpart the stopper thread before we start the idle loop (and start
1339 * scheduling); this ensures the stopper task is always available.
1341 stop_machine_unpark(smp_processor_id());
1343 st->state = CPUHP_AP_ONLINE_IDLE;
1344 complete_ap_thread(st, true);
1347 /* Requires cpu_add_remove_lock to be held */
1348 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1350 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1351 struct task_struct *idle;
1356 if (!cpu_present(cpu)) {
1362 * The caller of cpu_up() might have raced with another
1363 * caller. Nothing to do.
1365 if (st->state >= target)
1368 if (st->state == CPUHP_OFFLINE) {
1369 /* Let it fail before we try to bring the cpu up */
1370 idle = idle_thread_get(cpu);
1372 ret = PTR_ERR(idle);
1377 cpuhp_tasks_frozen = tasks_frozen;
1379 cpuhp_set_state(cpu, st, target);
1381 * If the current CPU state is in the range of the AP hotplug thread,
1382 * then we need to kick the thread once more.
1384 if (st->state > CPUHP_BRINGUP_CPU) {
1385 ret = cpuhp_kick_ap_work(cpu);
1387 * The AP side has done the error rollback already. Just
1388 * return the error code..
1395 * Try to reach the target state. We max out on the BP at
1396 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1397 * responsible for bringing it up to the target state.
1399 target = min((int)target, CPUHP_BRINGUP_CPU);
1400 ret = cpuhp_up_callbacks(cpu, st, target);
1402 cpus_write_unlock();
1404 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1408 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1412 if (!cpu_possible(cpu)) {
1413 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1415 #if defined(CONFIG_IA64)
1416 pr_err("please check additional_cpus= boot parameter\n");
1421 err = try_online_node(cpu_to_node(cpu));
1425 cpu_maps_update_begin();
1427 if (cpu_hotplug_disabled) {
1431 if (!cpu_smt_allowed(cpu)) {
1436 err = _cpu_up(cpu, 0, target);
1438 cpu_maps_update_done();
1443 * cpu_device_up - Bring up a cpu device
1444 * @dev: Pointer to the cpu device to online
1446 * This function is meant to be used by device core cpu subsystem only.
1448 * Other subsystems should use add_cpu() instead.
1450 * Return: %0 on success or a negative errno code
1452 int cpu_device_up(struct device *dev)
1454 return cpu_up(dev->id, CPUHP_ONLINE);
1457 int add_cpu(unsigned int cpu)
1461 lock_device_hotplug();
1462 ret = device_online(get_cpu_device(cpu));
1463 unlock_device_hotplug();
1467 EXPORT_SYMBOL_GPL(add_cpu);
1470 * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1471 * @sleep_cpu: The cpu we hibernated on and should be brought up.
1473 * On some architectures like arm64, we can hibernate on any CPU, but on
1474 * wake up the CPU we hibernated on might be offline as a side effect of
1475 * using maxcpus= for example.
1477 * Return: %0 on success or a negative errno code
1479 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1483 if (!cpu_online(sleep_cpu)) {
1484 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1485 ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1487 pr_err("Failed to bring hibernate-CPU up!\n");
1494 void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1498 for_each_present_cpu(cpu) {
1499 if (num_online_cpus() >= setup_max_cpus)
1501 if (!cpu_online(cpu))
1502 cpu_up(cpu, CPUHP_ONLINE);
1506 #ifdef CONFIG_PM_SLEEP_SMP
1507 static cpumask_var_t frozen_cpus;
1509 int freeze_secondary_cpus(int primary)
1513 cpu_maps_update_begin();
1514 if (primary == -1) {
1515 primary = cpumask_first(cpu_online_mask);
1516 if (!housekeeping_cpu(primary, HK_FLAG_TIMER))
1517 primary = housekeeping_any_cpu(HK_FLAG_TIMER);
1519 if (!cpu_online(primary))
1520 primary = cpumask_first(cpu_online_mask);
1524 * We take down all of the non-boot CPUs in one shot to avoid races
1525 * with the userspace trying to use the CPU hotplug at the same time
1527 cpumask_clear(frozen_cpus);
1529 pr_info("Disabling non-boot CPUs ...\n");
1530 for_each_online_cpu(cpu) {
1534 if (pm_wakeup_pending()) {
1535 pr_info("Wakeup pending. Abort CPU freeze\n");
1540 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1541 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1542 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1544 cpumask_set_cpu(cpu, frozen_cpus);
1546 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1552 BUG_ON(num_online_cpus() > 1);
1554 pr_err("Non-boot CPUs are not disabled\n");
1557 * Make sure the CPUs won't be enabled by someone else. We need to do
1558 * this even in case of failure as all freeze_secondary_cpus() users are
1559 * supposed to do thaw_secondary_cpus() on the failure path.
1561 cpu_hotplug_disabled++;
1563 cpu_maps_update_done();
1567 void __weak arch_thaw_secondary_cpus_begin(void)
1571 void __weak arch_thaw_secondary_cpus_end(void)
1575 void thaw_secondary_cpus(void)
1579 /* Allow everyone to use the CPU hotplug again */
1580 cpu_maps_update_begin();
1581 __cpu_hotplug_enable();
1582 if (cpumask_empty(frozen_cpus))
1585 pr_info("Enabling non-boot CPUs ...\n");
1587 arch_thaw_secondary_cpus_begin();
1589 for_each_cpu(cpu, frozen_cpus) {
1590 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1591 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1592 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1594 pr_info("CPU%d is up\n", cpu);
1597 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1600 arch_thaw_secondary_cpus_end();
1602 cpumask_clear(frozen_cpus);
1604 cpu_maps_update_done();
1607 static int __init alloc_frozen_cpus(void)
1609 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1613 core_initcall(alloc_frozen_cpus);
1616 * When callbacks for CPU hotplug notifications are being executed, we must
1617 * ensure that the state of the system with respect to the tasks being frozen
1618 * or not, as reported by the notification, remains unchanged *throughout the
1619 * duration* of the execution of the callbacks.
1620 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1622 * This synchronization is implemented by mutually excluding regular CPU
1623 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1624 * Hibernate notifications.
1627 cpu_hotplug_pm_callback(struct notifier_block *nb,
1628 unsigned long action, void *ptr)
1632 case PM_SUSPEND_PREPARE:
1633 case PM_HIBERNATION_PREPARE:
1634 cpu_hotplug_disable();
1637 case PM_POST_SUSPEND:
1638 case PM_POST_HIBERNATION:
1639 cpu_hotplug_enable();
1650 static int __init cpu_hotplug_pm_sync_init(void)
1653 * cpu_hotplug_pm_callback has higher priority than x86
1654 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1655 * to disable cpu hotplug to avoid cpu hotplug race.
1657 pm_notifier(cpu_hotplug_pm_callback, 0);
1660 core_initcall(cpu_hotplug_pm_sync_init);
1662 #endif /* CONFIG_PM_SLEEP_SMP */
1666 #endif /* CONFIG_SMP */
1668 /* Boot processor state steps */
1669 static struct cpuhp_step cpuhp_hp_states[] = {
1672 .startup.single = NULL,
1673 .teardown.single = NULL,
1676 [CPUHP_CREATE_THREADS]= {
1677 .name = "threads:prepare",
1678 .startup.single = smpboot_create_threads,
1679 .teardown.single = NULL,
1682 [CPUHP_PERF_PREPARE] = {
1683 .name = "perf:prepare",
1684 .startup.single = perf_event_init_cpu,
1685 .teardown.single = perf_event_exit_cpu,
1687 [CPUHP_RANDOM_PREPARE] = {
1688 .name = "random:prepare",
1689 .startup.single = random_prepare_cpu,
1690 .teardown.single = NULL,
1692 [CPUHP_WORKQUEUE_PREP] = {
1693 .name = "workqueue:prepare",
1694 .startup.single = workqueue_prepare_cpu,
1695 .teardown.single = NULL,
1697 [CPUHP_HRTIMERS_PREPARE] = {
1698 .name = "hrtimers:prepare",
1699 .startup.single = hrtimers_prepare_cpu,
1700 .teardown.single = hrtimers_dead_cpu,
1702 [CPUHP_SMPCFD_PREPARE] = {
1703 .name = "smpcfd:prepare",
1704 .startup.single = smpcfd_prepare_cpu,
1705 .teardown.single = smpcfd_dead_cpu,
1707 [CPUHP_RELAY_PREPARE] = {
1708 .name = "relay:prepare",
1709 .startup.single = relay_prepare_cpu,
1710 .teardown.single = NULL,
1712 [CPUHP_SLAB_PREPARE] = {
1713 .name = "slab:prepare",
1714 .startup.single = slab_prepare_cpu,
1715 .teardown.single = slab_dead_cpu,
1717 [CPUHP_RCUTREE_PREP] = {
1718 .name = "RCU/tree:prepare",
1719 .startup.single = rcutree_prepare_cpu,
1720 .teardown.single = rcutree_dead_cpu,
1723 * On the tear-down path, timers_dead_cpu() must be invoked
1724 * before blk_mq_queue_reinit_notify() from notify_dead(),
1725 * otherwise a RCU stall occurs.
1727 [CPUHP_TIMERS_PREPARE] = {
1728 .name = "timers:prepare",
1729 .startup.single = timers_prepare_cpu,
1730 .teardown.single = timers_dead_cpu,
1732 /* Kicks the plugged cpu into life */
1733 [CPUHP_BRINGUP_CPU] = {
1734 .name = "cpu:bringup",
1735 .startup.single = bringup_cpu,
1736 .teardown.single = finish_cpu,
1739 /* Final state before CPU kills itself */
1740 [CPUHP_AP_IDLE_DEAD] = {
1741 .name = "idle:dead",
1744 * Last state before CPU enters the idle loop to die. Transient state
1745 * for synchronization.
1747 [CPUHP_AP_OFFLINE] = {
1748 .name = "ap:offline",
1751 /* First state is scheduler control. Interrupts are disabled */
1752 [CPUHP_AP_SCHED_STARTING] = {
1753 .name = "sched:starting",
1754 .startup.single = sched_cpu_starting,
1755 .teardown.single = sched_cpu_dying,
1757 [CPUHP_AP_RCUTREE_DYING] = {
1758 .name = "RCU/tree:dying",
1759 .startup.single = NULL,
1760 .teardown.single = rcutree_dying_cpu,
1762 [CPUHP_AP_SMPCFD_DYING] = {
1763 .name = "smpcfd:dying",
1764 .startup.single = NULL,
1765 .teardown.single = smpcfd_dying_cpu,
1767 /* Entry state on starting. Interrupts enabled from here on. Transient
1768 * state for synchronsization */
1769 [CPUHP_AP_ONLINE] = {
1770 .name = "ap:online",
1773 * Handled on control processor until the plugged processor manages
1776 [CPUHP_TEARDOWN_CPU] = {
1777 .name = "cpu:teardown",
1778 .startup.single = NULL,
1779 .teardown.single = takedown_cpu,
1783 [CPUHP_AP_SCHED_WAIT_EMPTY] = {
1784 .name = "sched:waitempty",
1785 .startup.single = NULL,
1786 .teardown.single = sched_cpu_wait_empty,
1789 /* Handle smpboot threads park/unpark */
1790 [CPUHP_AP_SMPBOOT_THREADS] = {
1791 .name = "smpboot/threads:online",
1792 .startup.single = smpboot_unpark_threads,
1793 .teardown.single = smpboot_park_threads,
1795 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1796 .name = "irq/affinity:online",
1797 .startup.single = irq_affinity_online_cpu,
1798 .teardown.single = NULL,
1800 [CPUHP_AP_PERF_ONLINE] = {
1801 .name = "perf:online",
1802 .startup.single = perf_event_init_cpu,
1803 .teardown.single = perf_event_exit_cpu,
1805 [CPUHP_AP_WATCHDOG_ONLINE] = {
1806 .name = "lockup_detector:online",
1807 .startup.single = lockup_detector_online_cpu,
1808 .teardown.single = lockup_detector_offline_cpu,
1810 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1811 .name = "workqueue:online",
1812 .startup.single = workqueue_online_cpu,
1813 .teardown.single = workqueue_offline_cpu,
1815 [CPUHP_AP_RANDOM_ONLINE] = {
1816 .name = "random:online",
1817 .startup.single = random_online_cpu,
1818 .teardown.single = NULL,
1820 [CPUHP_AP_RCUTREE_ONLINE] = {
1821 .name = "RCU/tree:online",
1822 .startup.single = rcutree_online_cpu,
1823 .teardown.single = rcutree_offline_cpu,
1827 * The dynamically registered state space is here
1831 /* Last state is scheduler control setting the cpu active */
1832 [CPUHP_AP_ACTIVE] = {
1833 .name = "sched:active",
1834 .startup.single = sched_cpu_activate,
1835 .teardown.single = sched_cpu_deactivate,
1839 /* CPU is fully up and running. */
1842 .startup.single = NULL,
1843 .teardown.single = NULL,
1847 /* Sanity check for callbacks */
1848 static int cpuhp_cb_check(enum cpuhp_state state)
1850 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1856 * Returns a free for dynamic slot assignment of the Online state. The states
1857 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1858 * by having no name assigned.
1860 static int cpuhp_reserve_state(enum cpuhp_state state)
1862 enum cpuhp_state i, end;
1863 struct cpuhp_step *step;
1866 case CPUHP_AP_ONLINE_DYN:
1867 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1868 end = CPUHP_AP_ONLINE_DYN_END;
1870 case CPUHP_BP_PREPARE_DYN:
1871 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1872 end = CPUHP_BP_PREPARE_DYN_END;
1878 for (i = state; i <= end; i++, step++) {
1882 WARN(1, "No more dynamic states available for CPU hotplug\n");
1886 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1887 int (*startup)(unsigned int cpu),
1888 int (*teardown)(unsigned int cpu),
1889 bool multi_instance)
1891 /* (Un)Install the callbacks for further cpu hotplug operations */
1892 struct cpuhp_step *sp;
1896 * If name is NULL, then the state gets removed.
1898 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1899 * the first allocation from these dynamic ranges, so the removal
1900 * would trigger a new allocation and clear the wrong (already
1901 * empty) state, leaving the callbacks of the to be cleared state
1902 * dangling, which causes wreckage on the next hotplug operation.
1904 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1905 state == CPUHP_BP_PREPARE_DYN)) {
1906 ret = cpuhp_reserve_state(state);
1911 sp = cpuhp_get_step(state);
1912 if (name && sp->name)
1915 sp->startup.single = startup;
1916 sp->teardown.single = teardown;
1918 sp->multi_instance = multi_instance;
1919 INIT_HLIST_HEAD(&sp->list);
1923 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1925 return cpuhp_get_step(state)->teardown.single;
1929 * Call the startup/teardown function for a step either on the AP or
1930 * on the current CPU.
1932 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1933 struct hlist_node *node)
1935 struct cpuhp_step *sp = cpuhp_get_step(state);
1939 * If there's nothing to do, we done.
1940 * Relies on the union for multi_instance.
1942 if (cpuhp_step_empty(bringup, sp))
1945 * The non AP bound callbacks can fail on bringup. On teardown
1946 * e.g. module removal we crash for now.
1949 if (cpuhp_is_ap_state(state))
1950 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1952 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1954 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1956 BUG_ON(ret && !bringup);
1961 * Called from __cpuhp_setup_state on a recoverable failure.
1963 * Note: The teardown callbacks for rollback are not allowed to fail!
1965 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1966 struct hlist_node *node)
1970 /* Roll back the already executed steps on the other cpus */
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 (cpu >= failedcpu)
1978 /* Did we invoke the startup call on that cpu ? */
1979 if (cpustate >= state)
1980 cpuhp_issue_call(cpu, state, false, node);
1984 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1985 struct hlist_node *node,
1988 struct cpuhp_step *sp;
1992 lockdep_assert_cpus_held();
1994 sp = cpuhp_get_step(state);
1995 if (sp->multi_instance == false)
1998 mutex_lock(&cpuhp_state_mutex);
2000 if (!invoke || !sp->startup.multi)
2004 * Try to call the startup callback for each present cpu
2005 * depending on the hotplug state of the cpu.
2007 for_each_present_cpu(cpu) {
2008 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2009 int cpustate = st->state;
2011 if (cpustate < state)
2014 ret = cpuhp_issue_call(cpu, state, true, node);
2016 if (sp->teardown.multi)
2017 cpuhp_rollback_install(cpu, state, node);
2023 hlist_add_head(node, &sp->list);
2025 mutex_unlock(&cpuhp_state_mutex);
2029 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
2035 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
2039 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
2042 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
2043 * @state: The state to setup
2044 * @name: Name of the step
2045 * @invoke: If true, the startup function is invoked for cpus where
2046 * cpu state >= @state
2047 * @startup: startup callback function
2048 * @teardown: teardown callback function
2049 * @multi_instance: State is set up for multiple instances which get
2052 * The caller needs to hold cpus read locked while calling this function.
2055 * Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2056 * 0 for all other states
2057 * On failure: proper (negative) error code
2059 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
2060 const char *name, bool invoke,
2061 int (*startup)(unsigned int cpu),
2062 int (*teardown)(unsigned int cpu),
2063 bool multi_instance)
2068 lockdep_assert_cpus_held();
2070 if (cpuhp_cb_check(state) || !name)
2073 mutex_lock(&cpuhp_state_mutex);
2075 ret = cpuhp_store_callbacks(state, name, startup, teardown,
2078 dynstate = state == CPUHP_AP_ONLINE_DYN;
2079 if (ret > 0 && dynstate) {
2084 if (ret || !invoke || !startup)
2088 * Try to call the startup callback for each present cpu
2089 * depending on the hotplug state of the cpu.
2091 for_each_present_cpu(cpu) {
2092 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2093 int cpustate = st->state;
2095 if (cpustate < state)
2098 ret = cpuhp_issue_call(cpu, state, true, NULL);
2101 cpuhp_rollback_install(cpu, state, NULL);
2102 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2107 mutex_unlock(&cpuhp_state_mutex);
2109 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2110 * dynamically allocated state in case of success.
2112 if (!ret && dynstate)
2116 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
2118 int __cpuhp_setup_state(enum cpuhp_state state,
2119 const char *name, bool invoke,
2120 int (*startup)(unsigned int cpu),
2121 int (*teardown)(unsigned int cpu),
2122 bool multi_instance)
2127 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2128 teardown, multi_instance);
2132 EXPORT_SYMBOL(__cpuhp_setup_state);
2134 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2135 struct hlist_node *node, bool invoke)
2137 struct cpuhp_step *sp = cpuhp_get_step(state);
2140 BUG_ON(cpuhp_cb_check(state));
2142 if (!sp->multi_instance)
2146 mutex_lock(&cpuhp_state_mutex);
2148 if (!invoke || !cpuhp_get_teardown_cb(state))
2151 * Call the teardown callback for each present cpu depending
2152 * on the hotplug state of the cpu. This function is not
2153 * allowed to fail currently!
2155 for_each_present_cpu(cpu) {
2156 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2157 int cpustate = st->state;
2159 if (cpustate >= state)
2160 cpuhp_issue_call(cpu, state, false, node);
2165 mutex_unlock(&cpuhp_state_mutex);
2170 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2173 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2174 * @state: The state to remove
2175 * @invoke: If true, the teardown function is invoked for cpus where
2176 * cpu state >= @state
2178 * The caller needs to hold cpus read locked while calling this function.
2179 * The teardown callback is currently not allowed to fail. Think
2180 * about module removal!
2182 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2184 struct cpuhp_step *sp = cpuhp_get_step(state);
2187 BUG_ON(cpuhp_cb_check(state));
2189 lockdep_assert_cpus_held();
2191 mutex_lock(&cpuhp_state_mutex);
2192 if (sp->multi_instance) {
2193 WARN(!hlist_empty(&sp->list),
2194 "Error: Removing state %d which has instances left.\n",
2199 if (!invoke || !cpuhp_get_teardown_cb(state))
2203 * Call the teardown callback for each present cpu depending
2204 * on the hotplug state of the cpu. This function is not
2205 * allowed to fail currently!
2207 for_each_present_cpu(cpu) {
2208 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2209 int cpustate = st->state;
2211 if (cpustate >= state)
2212 cpuhp_issue_call(cpu, state, false, NULL);
2215 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2216 mutex_unlock(&cpuhp_state_mutex);
2218 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2220 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2223 __cpuhp_remove_state_cpuslocked(state, invoke);
2226 EXPORT_SYMBOL(__cpuhp_remove_state);
2228 #ifdef CONFIG_HOTPLUG_SMT
2229 static void cpuhp_offline_cpu_device(unsigned int cpu)
2231 struct device *dev = get_cpu_device(cpu);
2233 dev->offline = true;
2234 /* Tell user space about the state change */
2235 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2238 static void cpuhp_online_cpu_device(unsigned int cpu)
2240 struct device *dev = get_cpu_device(cpu);
2242 dev->offline = false;
2243 /* Tell user space about the state change */
2244 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2247 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2251 cpu_maps_update_begin();
2252 for_each_online_cpu(cpu) {
2253 if (topology_is_primary_thread(cpu))
2255 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2259 * As this needs to hold the cpu maps lock it's impossible
2260 * to call device_offline() because that ends up calling
2261 * cpu_down() which takes cpu maps lock. cpu maps lock
2262 * needs to be held as this might race against in kernel
2263 * abusers of the hotplug machinery (thermal management).
2265 * So nothing would update device:offline state. That would
2266 * leave the sysfs entry stale and prevent onlining after
2267 * smt control has been changed to 'off' again. This is
2268 * called under the sysfs hotplug lock, so it is properly
2269 * serialized against the regular offline usage.
2271 cpuhp_offline_cpu_device(cpu);
2274 cpu_smt_control = ctrlval;
2275 cpu_maps_update_done();
2279 int cpuhp_smt_enable(void)
2283 cpu_maps_update_begin();
2284 cpu_smt_control = CPU_SMT_ENABLED;
2285 for_each_present_cpu(cpu) {
2286 /* Skip online CPUs and CPUs on offline nodes */
2287 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2289 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2292 /* See comment in cpuhp_smt_disable() */
2293 cpuhp_online_cpu_device(cpu);
2295 cpu_maps_update_done();
2300 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2301 static ssize_t state_show(struct device *dev,
2302 struct device_attribute *attr, char *buf)
2304 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2306 return sprintf(buf, "%d\n", st->state);
2308 static DEVICE_ATTR_RO(state);
2310 static ssize_t target_store(struct device *dev, struct device_attribute *attr,
2311 const char *buf, size_t count)
2313 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2314 struct cpuhp_step *sp;
2317 ret = kstrtoint(buf, 10, &target);
2321 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2322 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2325 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2329 ret = lock_device_hotplug_sysfs();
2333 mutex_lock(&cpuhp_state_mutex);
2334 sp = cpuhp_get_step(target);
2335 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2336 mutex_unlock(&cpuhp_state_mutex);
2340 if (st->state < target)
2341 ret = cpu_up(dev->id, target);
2342 else if (st->state > target)
2343 ret = cpu_down(dev->id, target);
2344 else if (WARN_ON(st->target != target))
2345 st->target = target;
2347 unlock_device_hotplug();
2348 return ret ? ret : count;
2351 static ssize_t target_show(struct device *dev,
2352 struct device_attribute *attr, char *buf)
2354 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2356 return sprintf(buf, "%d\n", st->target);
2358 static DEVICE_ATTR_RW(target);
2360 static ssize_t fail_store(struct device *dev, struct device_attribute *attr,
2361 const char *buf, size_t count)
2363 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2364 struct cpuhp_step *sp;
2367 ret = kstrtoint(buf, 10, &fail);
2371 if (fail == CPUHP_INVALID) {
2376 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2380 * Cannot fail STARTING/DYING callbacks.
2382 if (cpuhp_is_atomic_state(fail))
2386 * DEAD callbacks cannot fail...
2387 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2388 * triggering STARTING callbacks, a failure in this state would
2391 if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU)
2395 * Cannot fail anything that doesn't have callbacks.
2397 mutex_lock(&cpuhp_state_mutex);
2398 sp = cpuhp_get_step(fail);
2399 if (!sp->startup.single && !sp->teardown.single)
2401 mutex_unlock(&cpuhp_state_mutex);
2410 static ssize_t fail_show(struct device *dev,
2411 struct device_attribute *attr, char *buf)
2413 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2415 return sprintf(buf, "%d\n", st->fail);
2418 static DEVICE_ATTR_RW(fail);
2420 static struct attribute *cpuhp_cpu_attrs[] = {
2421 &dev_attr_state.attr,
2422 &dev_attr_target.attr,
2423 &dev_attr_fail.attr,
2427 static const struct attribute_group cpuhp_cpu_attr_group = {
2428 .attrs = cpuhp_cpu_attrs,
2433 static ssize_t states_show(struct device *dev,
2434 struct device_attribute *attr, char *buf)
2436 ssize_t cur, res = 0;
2439 mutex_lock(&cpuhp_state_mutex);
2440 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2441 struct cpuhp_step *sp = cpuhp_get_step(i);
2444 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2449 mutex_unlock(&cpuhp_state_mutex);
2452 static DEVICE_ATTR_RO(states);
2454 static struct attribute *cpuhp_cpu_root_attrs[] = {
2455 &dev_attr_states.attr,
2459 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2460 .attrs = cpuhp_cpu_root_attrs,
2465 #ifdef CONFIG_HOTPLUG_SMT
2468 __store_smt_control(struct device *dev, struct device_attribute *attr,
2469 const char *buf, size_t count)
2473 if (sysfs_streq(buf, "on"))
2474 ctrlval = CPU_SMT_ENABLED;
2475 else if (sysfs_streq(buf, "off"))
2476 ctrlval = CPU_SMT_DISABLED;
2477 else if (sysfs_streq(buf, "forceoff"))
2478 ctrlval = CPU_SMT_FORCE_DISABLED;
2482 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2485 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2488 ret = lock_device_hotplug_sysfs();
2492 if (ctrlval != cpu_smt_control) {
2494 case CPU_SMT_ENABLED:
2495 ret = cpuhp_smt_enable();
2497 case CPU_SMT_DISABLED:
2498 case CPU_SMT_FORCE_DISABLED:
2499 ret = cpuhp_smt_disable(ctrlval);
2504 unlock_device_hotplug();
2505 return ret ? ret : count;
2508 #else /* !CONFIG_HOTPLUG_SMT */
2510 __store_smt_control(struct device *dev, struct device_attribute *attr,
2511 const char *buf, size_t count)
2515 #endif /* CONFIG_HOTPLUG_SMT */
2517 static const char *smt_states[] = {
2518 [CPU_SMT_ENABLED] = "on",
2519 [CPU_SMT_DISABLED] = "off",
2520 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2521 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2522 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2525 static ssize_t control_show(struct device *dev,
2526 struct device_attribute *attr, char *buf)
2528 const char *state = smt_states[cpu_smt_control];
2530 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2533 static ssize_t control_store(struct device *dev, struct device_attribute *attr,
2534 const char *buf, size_t count)
2536 return __store_smt_control(dev, attr, buf, count);
2538 static DEVICE_ATTR_RW(control);
2540 static ssize_t active_show(struct device *dev,
2541 struct device_attribute *attr, char *buf)
2543 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2545 static DEVICE_ATTR_RO(active);
2547 static struct attribute *cpuhp_smt_attrs[] = {
2548 &dev_attr_control.attr,
2549 &dev_attr_active.attr,
2553 static const struct attribute_group cpuhp_smt_attr_group = {
2554 .attrs = cpuhp_smt_attrs,
2559 static int __init cpu_smt_sysfs_init(void)
2561 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2562 &cpuhp_smt_attr_group);
2565 static int __init cpuhp_sysfs_init(void)
2569 ret = cpu_smt_sysfs_init();
2573 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2574 &cpuhp_cpu_root_attr_group);
2578 for_each_possible_cpu(cpu) {
2579 struct device *dev = get_cpu_device(cpu);
2583 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2589 device_initcall(cpuhp_sysfs_init);
2590 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2593 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2594 * represents all NR_CPUS bits binary values of 1<<nr.
2596 * It is used by cpumask_of() to get a constant address to a CPU
2597 * mask value that has a single bit set only.
2600 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2601 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2602 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2603 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2604 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2606 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2608 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2609 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2610 #if BITS_PER_LONG > 32
2611 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2612 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2615 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2617 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2618 EXPORT_SYMBOL(cpu_all_bits);
2620 #ifdef CONFIG_INIT_ALL_POSSIBLE
2621 struct cpumask __cpu_possible_mask __read_mostly
2624 struct cpumask __cpu_possible_mask __read_mostly;
2626 EXPORT_SYMBOL(__cpu_possible_mask);
2628 struct cpumask __cpu_online_mask __read_mostly;
2629 EXPORT_SYMBOL(__cpu_online_mask);
2631 struct cpumask __cpu_present_mask __read_mostly;
2632 EXPORT_SYMBOL(__cpu_present_mask);
2634 struct cpumask __cpu_active_mask __read_mostly;
2635 EXPORT_SYMBOL(__cpu_active_mask);
2637 struct cpumask __cpu_dying_mask __read_mostly;
2638 EXPORT_SYMBOL(__cpu_dying_mask);
2640 atomic_t __num_online_cpus __read_mostly;
2641 EXPORT_SYMBOL(__num_online_cpus);
2643 void init_cpu_present(const struct cpumask *src)
2645 cpumask_copy(&__cpu_present_mask, src);
2648 void init_cpu_possible(const struct cpumask *src)
2650 cpumask_copy(&__cpu_possible_mask, src);
2653 void init_cpu_online(const struct cpumask *src)
2655 cpumask_copy(&__cpu_online_mask, src);
2658 void set_cpu_online(unsigned int cpu, bool online)
2661 * atomic_inc/dec() is required to handle the horrid abuse of this
2662 * function by the reboot and kexec code which invoke it from
2663 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2664 * regular CPU hotplug is properly serialized.
2666 * Note, that the fact that __num_online_cpus is of type atomic_t
2667 * does not protect readers which are not serialized against
2668 * concurrent hotplug operations.
2671 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2672 atomic_inc(&__num_online_cpus);
2674 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2675 atomic_dec(&__num_online_cpus);
2680 * Activate the first processor.
2682 void __init boot_cpu_init(void)
2684 int cpu = smp_processor_id();
2686 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2687 set_cpu_online(cpu, true);
2688 set_cpu_active(cpu, true);
2689 set_cpu_present(cpu, true);
2690 set_cpu_possible(cpu, true);
2693 __boot_cpu_id = cpu;
2698 * Must be called _AFTER_ setting up the per_cpu areas
2700 void __init boot_cpu_hotplug_init(void)
2703 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2705 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2709 * These are used for a global "mitigations=" cmdline option for toggling
2710 * optional CPU mitigations.
2712 enum cpu_mitigations {
2713 CPU_MITIGATIONS_OFF,
2714 CPU_MITIGATIONS_AUTO,
2715 CPU_MITIGATIONS_AUTO_NOSMT,
2718 static enum cpu_mitigations cpu_mitigations __ro_after_init =
2719 CPU_MITIGATIONS_AUTO;
2721 static int __init mitigations_parse_cmdline(char *arg)
2723 if (!strcmp(arg, "off"))
2724 cpu_mitigations = CPU_MITIGATIONS_OFF;
2725 else if (!strcmp(arg, "auto"))
2726 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2727 else if (!strcmp(arg, "auto,nosmt"))
2728 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2730 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2735 early_param("mitigations", mitigations_parse_cmdline);
2737 /* mitigations=off */
2738 bool cpu_mitigations_off(void)
2740 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2742 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2744 /* mitigations=auto,nosmt */
2745 bool cpu_mitigations_auto_nosmt(void)
2747 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2749 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);