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/percpu-rwsem.h>
35 #include <linux/cpuset.h>
37 #include <trace/events/power.h>
38 #define CREATE_TRACE_POINTS
39 #include <trace/events/cpuhp.h>
44 * struct cpuhp_cpu_state - Per cpu hotplug state storage
45 * @state: The current cpu state
46 * @target: The target state
47 * @fail: Current CPU hotplug callback 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
54 * @node: Remote CPU node; for multi-instance, do a
55 * single entry callback for install/remove
56 * @last: For multi-instance rollback, remember how far we got
57 * @cb_state: The state for a single callback (install/uninstall)
58 * @result: Result of the operation
59 * @done_up: Signal completion to the issuer of the task for cpu-up
60 * @done_down: Signal completion to the issuer of the task for cpu-down
62 struct cpuhp_cpu_state {
63 enum cpuhp_state state;
64 enum cpuhp_state target;
65 enum cpuhp_state fail;
67 struct task_struct *thread;
73 struct hlist_node *node;
74 struct hlist_node *last;
75 enum cpuhp_state cb_state;
77 struct completion done_up;
78 struct completion done_down;
82 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
83 .fail = CPUHP_INVALID,
87 cpumask_t cpus_booted_once_mask;
90 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
91 static struct lockdep_map cpuhp_state_up_map =
92 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
93 static struct lockdep_map cpuhp_state_down_map =
94 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
97 static inline void cpuhp_lock_acquire(bool bringup)
99 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
102 static inline void cpuhp_lock_release(bool bringup)
104 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
108 static inline void cpuhp_lock_acquire(bool bringup) { }
109 static inline void cpuhp_lock_release(bool bringup) { }
114 * struct cpuhp_step - Hotplug state machine step
115 * @name: Name of the step
116 * @startup: Startup function of the step
117 * @teardown: Teardown function of the step
118 * @cant_stop: Bringup/teardown can't be stopped at this step
119 * @multi_instance: State has multiple instances which get added afterwards
124 int (*single)(unsigned int cpu);
125 int (*multi)(unsigned int cpu,
126 struct hlist_node *node);
129 int (*single)(unsigned int cpu);
130 int (*multi)(unsigned int cpu,
131 struct hlist_node *node);
134 struct hlist_head list;
140 static DEFINE_MUTEX(cpuhp_state_mutex);
141 static struct cpuhp_step cpuhp_hp_states[];
143 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
145 return cpuhp_hp_states + state;
148 static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step)
150 return bringup ? !step->startup.single : !step->teardown.single;
154 * cpuhp_invoke_callback - Invoke the callbacks for a given state
155 * @cpu: The cpu for which the callback should be invoked
156 * @state: The state to do callbacks for
157 * @bringup: True if the bringup callback should be invoked
158 * @node: For multi-instance, do a single entry callback for install/remove
159 * @lastp: For multi-instance rollback, remember how far we got
161 * Called from cpu hotplug and from the state register machinery.
163 * Return: %0 on success or a negative errno code
165 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
166 bool bringup, struct hlist_node *node,
167 struct hlist_node **lastp)
169 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
170 struct cpuhp_step *step = cpuhp_get_step(state);
171 int (*cbm)(unsigned int cpu, struct hlist_node *node);
172 int (*cb)(unsigned int cpu);
175 if (st->fail == state) {
176 st->fail = CPUHP_INVALID;
180 if (cpuhp_step_empty(bringup, step)) {
185 if (!step->multi_instance) {
186 WARN_ON_ONCE(lastp && *lastp);
187 cb = bringup ? step->startup.single : step->teardown.single;
189 trace_cpuhp_enter(cpu, st->target, state, cb);
191 trace_cpuhp_exit(cpu, st->state, state, ret);
194 cbm = bringup ? step->startup.multi : step->teardown.multi;
196 /* Single invocation for instance add/remove */
198 WARN_ON_ONCE(lastp && *lastp);
199 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
200 ret = cbm(cpu, node);
201 trace_cpuhp_exit(cpu, st->state, state, ret);
205 /* State transition. Invoke on all instances */
207 hlist_for_each(node, &step->list) {
208 if (lastp && node == *lastp)
211 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
212 ret = cbm(cpu, node);
213 trace_cpuhp_exit(cpu, st->state, state, ret);
227 /* Rollback the instances if one failed */
228 cbm = !bringup ? step->startup.multi : step->teardown.multi;
232 hlist_for_each(node, &step->list) {
236 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
237 ret = cbm(cpu, node);
238 trace_cpuhp_exit(cpu, st->state, state, ret);
240 * Rollback must not fail,
248 static bool cpuhp_is_ap_state(enum cpuhp_state state)
251 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
252 * purposes as that state is handled explicitly in cpu_down.
254 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
257 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
259 struct completion *done = bringup ? &st->done_up : &st->done_down;
260 wait_for_completion(done);
263 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
265 struct completion *done = bringup ? &st->done_up : &st->done_down;
270 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
272 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
274 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
277 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
278 static DEFINE_MUTEX(cpu_add_remove_lock);
279 bool cpuhp_tasks_frozen;
280 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
283 * The following two APIs (cpu_maps_update_begin/done) must be used when
284 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
286 void cpu_maps_update_begin(void)
288 mutex_lock(&cpu_add_remove_lock);
291 void cpu_maps_update_done(void)
293 mutex_unlock(&cpu_add_remove_lock);
297 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
298 * Should always be manipulated under cpu_add_remove_lock
300 static int cpu_hotplug_disabled;
302 #ifdef CONFIG_HOTPLUG_CPU
304 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
306 void cpus_read_lock(void)
308 percpu_down_read(&cpu_hotplug_lock);
310 EXPORT_SYMBOL_GPL(cpus_read_lock);
312 int cpus_read_trylock(void)
314 return percpu_down_read_trylock(&cpu_hotplug_lock);
316 EXPORT_SYMBOL_GPL(cpus_read_trylock);
318 void cpus_read_unlock(void)
320 percpu_up_read(&cpu_hotplug_lock);
322 EXPORT_SYMBOL_GPL(cpus_read_unlock);
324 void cpus_write_lock(void)
326 percpu_down_write(&cpu_hotplug_lock);
329 void cpus_write_unlock(void)
331 percpu_up_write(&cpu_hotplug_lock);
334 void lockdep_assert_cpus_held(void)
337 * We can't have hotplug operations before userspace starts running,
338 * and some init codepaths will knowingly not take the hotplug lock.
339 * This is all valid, so mute lockdep until it makes sense to report
342 if (system_state < SYSTEM_RUNNING)
345 percpu_rwsem_assert_held(&cpu_hotplug_lock);
348 #ifdef CONFIG_LOCKDEP
349 int lockdep_is_cpus_held(void)
351 return percpu_rwsem_is_held(&cpu_hotplug_lock);
355 static void lockdep_acquire_cpus_lock(void)
357 rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
360 static void lockdep_release_cpus_lock(void)
362 rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
366 * Wait for currently running CPU hotplug operations to complete (if any) and
367 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
368 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
369 * hotplug path before performing hotplug operations. So acquiring that lock
370 * guarantees mutual exclusion from any currently running hotplug operations.
372 void cpu_hotplug_disable(void)
374 cpu_maps_update_begin();
375 cpu_hotplug_disabled++;
376 cpu_maps_update_done();
378 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
380 static void __cpu_hotplug_enable(void)
382 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
384 cpu_hotplug_disabled--;
387 void cpu_hotplug_enable(void)
389 cpu_maps_update_begin();
390 __cpu_hotplug_enable();
391 cpu_maps_update_done();
393 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
397 static void lockdep_acquire_cpus_lock(void)
401 static void lockdep_release_cpus_lock(void)
405 #endif /* CONFIG_HOTPLUG_CPU */
408 * Architectures that need SMT-specific errata handling during SMT hotplug
409 * should override this.
411 void __weak arch_smt_update(void) { }
413 #ifdef CONFIG_HOTPLUG_SMT
414 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
416 void __init cpu_smt_disable(bool force)
418 if (!cpu_smt_possible())
422 pr_info("SMT: Force disabled\n");
423 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
425 pr_info("SMT: disabled\n");
426 cpu_smt_control = CPU_SMT_DISABLED;
431 * The decision whether SMT is supported can only be done after the full
432 * CPU identification. Called from architecture code.
434 void __init cpu_smt_check_topology(void)
436 if (!topology_smt_supported())
437 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
440 static int __init smt_cmdline_disable(char *str)
442 cpu_smt_disable(str && !strcmp(str, "force"));
445 early_param("nosmt", smt_cmdline_disable);
447 static inline bool cpu_smt_allowed(unsigned int cpu)
449 if (cpu_smt_control == CPU_SMT_ENABLED)
452 if (topology_is_primary_thread(cpu))
456 * On x86 it's required to boot all logical CPUs at least once so
457 * that the init code can get a chance to set CR4.MCE on each
458 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
459 * core will shutdown the machine.
461 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
464 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
465 bool cpu_smt_possible(void)
467 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
468 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
470 EXPORT_SYMBOL_GPL(cpu_smt_possible);
472 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
475 static inline enum cpuhp_state
476 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
478 enum cpuhp_state prev_state = st->state;
479 bool bringup = st->state < target;
481 st->rollback = false;
486 st->bringup = bringup;
487 if (cpu_dying(st->cpu) != !bringup)
488 set_cpu_dying(st->cpu, !bringup);
494 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
496 bool bringup = !st->bringup;
498 st->target = prev_state;
501 * Already rolling back. No need invert the bringup value or to change
510 * If we have st->last we need to undo partial multi_instance of this
511 * state first. Otherwise start undo at the previous state.
520 st->bringup = bringup;
521 if (cpu_dying(st->cpu) != !bringup)
522 set_cpu_dying(st->cpu, !bringup);
525 /* Regular hotplug invocation of the AP hotplug thread */
526 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
528 if (!st->single && st->state == st->target)
533 * Make sure the above stores are visible before should_run becomes
534 * true. Paired with the mb() above in cpuhp_thread_fun()
537 st->should_run = true;
538 wake_up_process(st->thread);
539 wait_for_ap_thread(st, st->bringup);
542 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
544 enum cpuhp_state prev_state;
547 prev_state = cpuhp_set_state(st, target);
549 if ((ret = st->result)) {
550 cpuhp_reset_state(st, prev_state);
557 static int bringup_wait_for_ap(unsigned int cpu)
559 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
561 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
562 wait_for_ap_thread(st, true);
563 if (WARN_ON_ONCE((!cpu_online(cpu))))
566 /* Unpark the hotplug thread of the target cpu */
567 kthread_unpark(st->thread);
570 * SMT soft disabling on X86 requires to bring the CPU out of the
571 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
572 * CPU marked itself as booted_once in notify_cpu_starting() so the
573 * cpu_smt_allowed() check will now return false if this is not the
576 if (!cpu_smt_allowed(cpu))
579 if (st->target <= CPUHP_AP_ONLINE_IDLE)
582 return cpuhp_kick_ap(st, st->target);
585 static int bringup_cpu(unsigned int cpu)
587 struct task_struct *idle = idle_thread_get(cpu);
591 * Some architectures have to walk the irq descriptors to
592 * setup the vector space for the cpu which comes online.
593 * Prevent irq alloc/free across the bringup.
597 /* Arch-specific enabling code. */
598 ret = __cpu_up(cpu, idle);
602 return bringup_wait_for_ap(cpu);
605 static int finish_cpu(unsigned int cpu)
607 struct task_struct *idle = idle_thread_get(cpu);
608 struct mm_struct *mm = idle->active_mm;
611 * idle_task_exit() will have switched to &init_mm, now
612 * clean up any remaining active_mm state.
615 idle->active_mm = &init_mm;
621 * Hotplug state machine related functions
625 * Get the next state to run. Empty ones will be skipped. Returns true if a
628 * st->state will be modified ahead of time, to match state_to_run, as if it
631 static bool cpuhp_next_state(bool bringup,
632 enum cpuhp_state *state_to_run,
633 struct cpuhp_cpu_state *st,
634 enum cpuhp_state target)
638 if (st->state >= target)
641 *state_to_run = ++st->state;
643 if (st->state <= target)
646 *state_to_run = st->state--;
649 if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run)))
656 static int cpuhp_invoke_callback_range(bool bringup,
658 struct cpuhp_cpu_state *st,
659 enum cpuhp_state target)
661 enum cpuhp_state state;
664 while (cpuhp_next_state(bringup, &state, st, target)) {
665 err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL);
673 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
675 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
678 * When CPU hotplug is disabled, then taking the CPU down is not
679 * possible because takedown_cpu() and the architecture and
680 * subsystem specific mechanisms are not available. So the CPU
681 * which would be completely unplugged again needs to stay around
682 * in the current state.
684 return st->state <= CPUHP_BRINGUP_CPU;
687 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
688 enum cpuhp_state target)
690 enum cpuhp_state prev_state = st->state;
693 ret = cpuhp_invoke_callback_range(true, cpu, st, target);
695 cpuhp_reset_state(st, prev_state);
696 if (can_rollback_cpu(st))
697 WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
704 * The cpu hotplug threads manage the bringup and teardown of the cpus
706 static void cpuhp_create(unsigned int cpu)
708 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
710 init_completion(&st->done_up);
711 init_completion(&st->done_down);
715 static int cpuhp_should_run(unsigned int cpu)
717 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
719 return st->should_run;
723 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
724 * callbacks when a state gets [un]installed at runtime.
726 * Each invocation of this function by the smpboot thread does a single AP
729 * It has 3 modes of operation:
730 * - single: runs st->cb_state
731 * - up: runs ++st->state, while st->state < st->target
732 * - down: runs st->state--, while st->state > st->target
734 * When complete or on error, should_run is cleared and the completion is fired.
736 static void cpuhp_thread_fun(unsigned int cpu)
738 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
739 bool bringup = st->bringup;
740 enum cpuhp_state state;
742 if (WARN_ON_ONCE(!st->should_run))
746 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
747 * that if we see ->should_run we also see the rest of the state.
752 * The BP holds the hotplug lock, but we're now running on the AP,
753 * ensure that anybody asserting the lock is held, will actually find
756 lockdep_acquire_cpus_lock();
757 cpuhp_lock_acquire(bringup);
760 state = st->cb_state;
761 st->should_run = false;
763 st->should_run = cpuhp_next_state(bringup, &state, st, st->target);
768 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
770 if (cpuhp_is_atomic_state(state)) {
772 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
776 * STARTING/DYING must not fail!
778 WARN_ON_ONCE(st->result);
780 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
785 * If we fail on a rollback, we're up a creek without no
786 * paddle, no way forward, no way back. We loose, thanks for
789 WARN_ON_ONCE(st->rollback);
790 st->should_run = false;
794 cpuhp_lock_release(bringup);
795 lockdep_release_cpus_lock();
798 complete_ap_thread(st, bringup);
801 /* Invoke a single callback on a remote cpu */
803 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
804 struct hlist_node *node)
806 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
809 if (!cpu_online(cpu))
812 cpuhp_lock_acquire(false);
813 cpuhp_lock_release(false);
815 cpuhp_lock_acquire(true);
816 cpuhp_lock_release(true);
819 * If we are up and running, use the hotplug thread. For early calls
820 * we invoke the thread function directly.
823 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
825 st->rollback = false;
829 st->bringup = bringup;
830 st->cb_state = state;
836 * If we failed and did a partial, do a rollback.
838 if ((ret = st->result) && st->last) {
840 st->bringup = !bringup;
846 * Clean up the leftovers so the next hotplug operation wont use stale
849 st->node = st->last = NULL;
853 static int cpuhp_kick_ap_work(unsigned int cpu)
855 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
856 enum cpuhp_state prev_state = st->state;
859 cpuhp_lock_acquire(false);
860 cpuhp_lock_release(false);
862 cpuhp_lock_acquire(true);
863 cpuhp_lock_release(true);
865 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
866 ret = cpuhp_kick_ap(st, st->target);
867 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
872 static struct smp_hotplug_thread cpuhp_threads = {
873 .store = &cpuhp_state.thread,
874 .create = &cpuhp_create,
875 .thread_should_run = cpuhp_should_run,
876 .thread_fn = cpuhp_thread_fun,
877 .thread_comm = "cpuhp/%u",
881 void __init cpuhp_threads_init(void)
883 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
884 kthread_unpark(this_cpu_read(cpuhp_state.thread));
889 * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
892 * The operation is still serialized against concurrent CPU hotplug via
893 * cpu_add_remove_lock, i.e. CPU map protection. But it is _not_
894 * serialized against other hotplug related activity like adding or
895 * removing of state callbacks and state instances, which invoke either the
896 * startup or the teardown callback of the affected state.
898 * This is required for subsystems which are unfixable vs. CPU hotplug and
899 * evade lock inversion problems by scheduling work which has to be
900 * completed _before_ cpu_up()/_cpu_down() returns.
902 * Don't even think about adding anything to this for any new code or even
903 * drivers. It's only purpose is to keep existing lock order trainwrecks
906 * For cpu_down() there might be valid reasons to finish cleanups which are
907 * not required to be done under cpu_hotplug_lock, but that's a different
908 * story and would be not invoked via this.
910 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
913 * cpusets delegate hotplug operations to a worker to "solve" the
914 * lock order problems. Wait for the worker, but only if tasks are
915 * _not_ frozen (suspend, hibernate) as that would wait forever.
917 * The wait is required because otherwise the hotplug operation
918 * returns with inconsistent state, which could even be observed in
919 * user space when a new CPU is brought up. The CPU plug uevent
920 * would be delivered and user space reacting on it would fail to
921 * move tasks to the newly plugged CPU up to the point where the
922 * work has finished because up to that point the newly plugged CPU
923 * is not assignable in cpusets/cgroups. On unplug that's not
924 * necessarily a visible issue, but it is still inconsistent state,
925 * which is the real problem which needs to be "fixed". This can't
926 * prevent the transient state between scheduling the work and
927 * returning from waiting for it.
930 cpuset_wait_for_hotplug();
933 #ifdef CONFIG_HOTPLUG_CPU
934 #ifndef arch_clear_mm_cpumask_cpu
935 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
939 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
942 * This function walks all processes, finds a valid mm struct for each one and
943 * then clears a corresponding bit in mm's cpumask. While this all sounds
944 * trivial, there are various non-obvious corner cases, which this function
945 * tries to solve in a safe manner.
947 * Also note that the function uses a somewhat relaxed locking scheme, so it may
948 * be called only for an already offlined CPU.
950 void clear_tasks_mm_cpumask(int cpu)
952 struct task_struct *p;
955 * This function is called after the cpu is taken down and marked
956 * offline, so its not like new tasks will ever get this cpu set in
957 * their mm mask. -- Peter Zijlstra
958 * Thus, we may use rcu_read_lock() here, instead of grabbing
959 * full-fledged tasklist_lock.
961 WARN_ON(cpu_online(cpu));
963 for_each_process(p) {
964 struct task_struct *t;
967 * Main thread might exit, but other threads may still have
968 * a valid mm. Find one.
970 t = find_lock_task_mm(p);
973 arch_clear_mm_cpumask_cpu(cpu, t->mm);
979 /* Take this CPU down. */
980 static int take_cpu_down(void *_param)
982 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
983 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
984 int err, cpu = smp_processor_id();
987 /* Ensure this CPU doesn't handle any more interrupts. */
988 err = __cpu_disable();
993 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
994 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
996 WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1));
998 /* Invoke the former CPU_DYING callbacks */
999 ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1002 * DYING must not fail!
1006 /* Give up timekeeping duties */
1007 tick_handover_do_timer();
1008 /* Remove CPU from timer broadcasting */
1009 tick_offline_cpu(cpu);
1010 /* Park the stopper thread */
1011 stop_machine_park(cpu);
1015 static int takedown_cpu(unsigned int cpu)
1017 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1020 /* Park the smpboot threads */
1021 kthread_park(st->thread);
1024 * Prevent irq alloc/free while the dying cpu reorganizes the
1025 * interrupt affinities.
1030 * So now all preempt/rcu users must observe !cpu_active().
1032 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
1034 /* CPU refused to die */
1035 irq_unlock_sparse();
1036 /* Unpark the hotplug thread so we can rollback there */
1037 kthread_unpark(st->thread);
1040 BUG_ON(cpu_online(cpu));
1043 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1044 * all runnable tasks from the CPU, there's only the idle task left now
1045 * that the migration thread is done doing the stop_machine thing.
1047 * Wait for the stop thread to go away.
1049 wait_for_ap_thread(st, false);
1050 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1052 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
1053 irq_unlock_sparse();
1055 hotplug_cpu__broadcast_tick_pull(cpu);
1056 /* This actually kills the CPU. */
1059 tick_cleanup_dead_cpu(cpu);
1060 rcutree_migrate_callbacks(cpu);
1064 static void cpuhp_complete_idle_dead(void *arg)
1066 struct cpuhp_cpu_state *st = arg;
1068 complete_ap_thread(st, false);
1071 void cpuhp_report_idle_dead(void)
1073 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1075 BUG_ON(st->state != CPUHP_AP_OFFLINE);
1076 rcu_report_dead(smp_processor_id());
1077 st->state = CPUHP_AP_IDLE_DEAD;
1079 * We cannot call complete after rcu_report_dead() so we delegate it
1082 smp_call_function_single(cpumask_first(cpu_online_mask),
1083 cpuhp_complete_idle_dead, st, 0);
1086 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1087 enum cpuhp_state target)
1089 enum cpuhp_state prev_state = st->state;
1092 ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1095 cpuhp_reset_state(st, prev_state);
1097 if (st->state < prev_state)
1098 WARN_ON(cpuhp_invoke_callback_range(true, cpu, st,
1105 /* Requires cpu_add_remove_lock to be held */
1106 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1107 enum cpuhp_state target)
1109 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1110 int prev_state, ret = 0;
1112 if (num_online_cpus() == 1)
1115 if (!cpu_present(cpu))
1120 cpuhp_tasks_frozen = tasks_frozen;
1122 prev_state = cpuhp_set_state(st, target);
1124 * If the current CPU state is in the range of the AP hotplug thread,
1125 * then we need to kick the thread.
1127 if (st->state > CPUHP_TEARDOWN_CPU) {
1128 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1129 ret = cpuhp_kick_ap_work(cpu);
1131 * The AP side has done the error rollback already. Just
1132 * return the error code..
1138 * We might have stopped still in the range of the AP hotplug
1139 * thread. Nothing to do anymore.
1141 if (st->state > CPUHP_TEARDOWN_CPU)
1144 st->target = target;
1147 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1148 * to do the further cleanups.
1150 ret = cpuhp_down_callbacks(cpu, st, target);
1151 if (ret && st->state < prev_state) {
1152 if (st->state == CPUHP_TEARDOWN_CPU) {
1153 cpuhp_reset_state(st, prev_state);
1154 __cpuhp_kick_ap(st);
1156 WARN(1, "DEAD callback error for CPU%d", cpu);
1161 cpus_write_unlock();
1163 * Do post unplug cleanup. This is still protected against
1164 * concurrent CPU hotplug via cpu_add_remove_lock.
1166 lockup_detector_cleanup();
1168 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1172 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1174 if (cpu_hotplug_disabled)
1176 return _cpu_down(cpu, 0, target);
1179 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1183 cpu_maps_update_begin();
1184 err = cpu_down_maps_locked(cpu, target);
1185 cpu_maps_update_done();
1190 * cpu_device_down - Bring down a cpu device
1191 * @dev: Pointer to the cpu device to offline
1193 * This function is meant to be used by device core cpu subsystem only.
1195 * Other subsystems should use remove_cpu() instead.
1197 * Return: %0 on success or a negative errno code
1199 int cpu_device_down(struct device *dev)
1201 return cpu_down(dev->id, CPUHP_OFFLINE);
1204 int remove_cpu(unsigned int cpu)
1208 lock_device_hotplug();
1209 ret = device_offline(get_cpu_device(cpu));
1210 unlock_device_hotplug();
1214 EXPORT_SYMBOL_GPL(remove_cpu);
1216 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1221 cpu_maps_update_begin();
1224 * Make certain the cpu I'm about to reboot on is online.
1226 * This is inline to what migrate_to_reboot_cpu() already do.
1228 if (!cpu_online(primary_cpu))
1229 primary_cpu = cpumask_first(cpu_online_mask);
1231 for_each_online_cpu(cpu) {
1232 if (cpu == primary_cpu)
1235 error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1237 pr_err("Failed to offline CPU%d - error=%d",
1244 * Ensure all but the reboot CPU are offline.
1246 BUG_ON(num_online_cpus() > 1);
1249 * Make sure the CPUs won't be enabled by someone else after this
1250 * point. Kexec will reboot to a new kernel shortly resetting
1251 * everything along the way.
1253 cpu_hotplug_disabled++;
1255 cpu_maps_update_done();
1259 #define takedown_cpu NULL
1260 #endif /*CONFIG_HOTPLUG_CPU*/
1263 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1264 * @cpu: cpu that just started
1266 * It must be called by the arch code on the new cpu, before the new cpu
1267 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1269 void notify_cpu_starting(unsigned int cpu)
1271 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1272 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1275 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1276 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1277 ret = cpuhp_invoke_callback_range(true, cpu, st, target);
1280 * STARTING must not fail!
1286 * Called from the idle task. Wake up the controlling task which brings the
1287 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1288 * online bringup to the hotplug thread.
1290 void cpuhp_online_idle(enum cpuhp_state state)
1292 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1294 /* Happens for the boot cpu */
1295 if (state != CPUHP_AP_ONLINE_IDLE)
1299 * Unpart the stopper thread before we start the idle loop (and start
1300 * scheduling); this ensures the stopper task is always available.
1302 stop_machine_unpark(smp_processor_id());
1304 st->state = CPUHP_AP_ONLINE_IDLE;
1305 complete_ap_thread(st, true);
1308 /* Requires cpu_add_remove_lock to be held */
1309 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1311 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1312 struct task_struct *idle;
1317 if (!cpu_present(cpu)) {
1323 * The caller of cpu_up() might have raced with another
1324 * caller. Nothing to do.
1326 if (st->state >= target)
1329 if (st->state == CPUHP_OFFLINE) {
1330 /* Let it fail before we try to bring the cpu up */
1331 idle = idle_thread_get(cpu);
1333 ret = PTR_ERR(idle);
1338 cpuhp_tasks_frozen = tasks_frozen;
1340 cpuhp_set_state(st, target);
1342 * If the current CPU state is in the range of the AP hotplug thread,
1343 * then we need to kick the thread once more.
1345 if (st->state > CPUHP_BRINGUP_CPU) {
1346 ret = cpuhp_kick_ap_work(cpu);
1348 * The AP side has done the error rollback already. Just
1349 * return the error code..
1356 * Try to reach the target state. We max out on the BP at
1357 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1358 * responsible for bringing it up to the target state.
1360 target = min((int)target, CPUHP_BRINGUP_CPU);
1361 ret = cpuhp_up_callbacks(cpu, st, target);
1363 cpus_write_unlock();
1365 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1369 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1373 if (!cpu_possible(cpu)) {
1374 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1376 #if defined(CONFIG_IA64)
1377 pr_err("please check additional_cpus= boot parameter\n");
1382 err = try_online_node(cpu_to_node(cpu));
1386 cpu_maps_update_begin();
1388 if (cpu_hotplug_disabled) {
1392 if (!cpu_smt_allowed(cpu)) {
1397 err = _cpu_up(cpu, 0, target);
1399 cpu_maps_update_done();
1404 * cpu_device_up - Bring up a cpu device
1405 * @dev: Pointer to the cpu device to online
1407 * This function is meant to be used by device core cpu subsystem only.
1409 * Other subsystems should use add_cpu() instead.
1411 * Return: %0 on success or a negative errno code
1413 int cpu_device_up(struct device *dev)
1415 return cpu_up(dev->id, CPUHP_ONLINE);
1418 int add_cpu(unsigned int cpu)
1422 lock_device_hotplug();
1423 ret = device_online(get_cpu_device(cpu));
1424 unlock_device_hotplug();
1428 EXPORT_SYMBOL_GPL(add_cpu);
1431 * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1432 * @sleep_cpu: The cpu we hibernated on and should be brought up.
1434 * On some architectures like arm64, we can hibernate on any CPU, but on
1435 * wake up the CPU we hibernated on might be offline as a side effect of
1436 * using maxcpus= for example.
1438 * Return: %0 on success or a negative errno code
1440 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1444 if (!cpu_online(sleep_cpu)) {
1445 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1446 ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1448 pr_err("Failed to bring hibernate-CPU up!\n");
1455 void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1459 for_each_present_cpu(cpu) {
1460 if (num_online_cpus() >= setup_max_cpus)
1462 if (!cpu_online(cpu))
1463 cpu_up(cpu, CPUHP_ONLINE);
1467 #ifdef CONFIG_PM_SLEEP_SMP
1468 static cpumask_var_t frozen_cpus;
1470 int freeze_secondary_cpus(int primary)
1474 cpu_maps_update_begin();
1475 if (primary == -1) {
1476 primary = cpumask_first(cpu_online_mask);
1477 if (!housekeeping_cpu(primary, HK_FLAG_TIMER))
1478 primary = housekeeping_any_cpu(HK_FLAG_TIMER);
1480 if (!cpu_online(primary))
1481 primary = cpumask_first(cpu_online_mask);
1485 * We take down all of the non-boot CPUs in one shot to avoid races
1486 * with the userspace trying to use the CPU hotplug at the same time
1488 cpumask_clear(frozen_cpus);
1490 pr_info("Disabling non-boot CPUs ...\n");
1491 for_each_online_cpu(cpu) {
1495 if (pm_wakeup_pending()) {
1496 pr_info("Wakeup pending. Abort CPU freeze\n");
1501 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1502 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1503 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1505 cpumask_set_cpu(cpu, frozen_cpus);
1507 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1513 BUG_ON(num_online_cpus() > 1);
1515 pr_err("Non-boot CPUs are not disabled\n");
1518 * Make sure the CPUs won't be enabled by someone else. We need to do
1519 * this even in case of failure as all freeze_secondary_cpus() users are
1520 * supposed to do thaw_secondary_cpus() on the failure path.
1522 cpu_hotplug_disabled++;
1524 cpu_maps_update_done();
1528 void __weak arch_thaw_secondary_cpus_begin(void)
1532 void __weak arch_thaw_secondary_cpus_end(void)
1536 void thaw_secondary_cpus(void)
1540 /* Allow everyone to use the CPU hotplug again */
1541 cpu_maps_update_begin();
1542 __cpu_hotplug_enable();
1543 if (cpumask_empty(frozen_cpus))
1546 pr_info("Enabling non-boot CPUs ...\n");
1548 arch_thaw_secondary_cpus_begin();
1550 for_each_cpu(cpu, frozen_cpus) {
1551 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1552 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1553 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1555 pr_info("CPU%d is up\n", cpu);
1558 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1561 arch_thaw_secondary_cpus_end();
1563 cpumask_clear(frozen_cpus);
1565 cpu_maps_update_done();
1568 static int __init alloc_frozen_cpus(void)
1570 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1574 core_initcall(alloc_frozen_cpus);
1577 * When callbacks for CPU hotplug notifications are being executed, we must
1578 * ensure that the state of the system with respect to the tasks being frozen
1579 * or not, as reported by the notification, remains unchanged *throughout the
1580 * duration* of the execution of the callbacks.
1581 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1583 * This synchronization is implemented by mutually excluding regular CPU
1584 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1585 * Hibernate notifications.
1588 cpu_hotplug_pm_callback(struct notifier_block *nb,
1589 unsigned long action, void *ptr)
1593 case PM_SUSPEND_PREPARE:
1594 case PM_HIBERNATION_PREPARE:
1595 cpu_hotplug_disable();
1598 case PM_POST_SUSPEND:
1599 case PM_POST_HIBERNATION:
1600 cpu_hotplug_enable();
1611 static int __init cpu_hotplug_pm_sync_init(void)
1614 * cpu_hotplug_pm_callback has higher priority than x86
1615 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1616 * to disable cpu hotplug to avoid cpu hotplug race.
1618 pm_notifier(cpu_hotplug_pm_callback, 0);
1621 core_initcall(cpu_hotplug_pm_sync_init);
1623 #endif /* CONFIG_PM_SLEEP_SMP */
1627 #endif /* CONFIG_SMP */
1629 /* Boot processor state steps */
1630 static struct cpuhp_step cpuhp_hp_states[] = {
1633 .startup.single = NULL,
1634 .teardown.single = NULL,
1637 [CPUHP_CREATE_THREADS]= {
1638 .name = "threads:prepare",
1639 .startup.single = smpboot_create_threads,
1640 .teardown.single = NULL,
1643 [CPUHP_PERF_PREPARE] = {
1644 .name = "perf:prepare",
1645 .startup.single = perf_event_init_cpu,
1646 .teardown.single = perf_event_exit_cpu,
1648 [CPUHP_WORKQUEUE_PREP] = {
1649 .name = "workqueue:prepare",
1650 .startup.single = workqueue_prepare_cpu,
1651 .teardown.single = NULL,
1653 [CPUHP_HRTIMERS_PREPARE] = {
1654 .name = "hrtimers:prepare",
1655 .startup.single = hrtimers_prepare_cpu,
1656 .teardown.single = hrtimers_dead_cpu,
1658 [CPUHP_SMPCFD_PREPARE] = {
1659 .name = "smpcfd:prepare",
1660 .startup.single = smpcfd_prepare_cpu,
1661 .teardown.single = smpcfd_dead_cpu,
1663 [CPUHP_RELAY_PREPARE] = {
1664 .name = "relay:prepare",
1665 .startup.single = relay_prepare_cpu,
1666 .teardown.single = NULL,
1668 [CPUHP_SLAB_PREPARE] = {
1669 .name = "slab:prepare",
1670 .startup.single = slab_prepare_cpu,
1671 .teardown.single = slab_dead_cpu,
1673 [CPUHP_RCUTREE_PREP] = {
1674 .name = "RCU/tree:prepare",
1675 .startup.single = rcutree_prepare_cpu,
1676 .teardown.single = rcutree_dead_cpu,
1679 * On the tear-down path, timers_dead_cpu() must be invoked
1680 * before blk_mq_queue_reinit_notify() from notify_dead(),
1681 * otherwise a RCU stall occurs.
1683 [CPUHP_TIMERS_PREPARE] = {
1684 .name = "timers:prepare",
1685 .startup.single = timers_prepare_cpu,
1686 .teardown.single = timers_dead_cpu,
1688 /* Kicks the plugged cpu into life */
1689 [CPUHP_BRINGUP_CPU] = {
1690 .name = "cpu:bringup",
1691 .startup.single = bringup_cpu,
1692 .teardown.single = finish_cpu,
1695 /* Final state before CPU kills itself */
1696 [CPUHP_AP_IDLE_DEAD] = {
1697 .name = "idle:dead",
1700 * Last state before CPU enters the idle loop to die. Transient state
1701 * for synchronization.
1703 [CPUHP_AP_OFFLINE] = {
1704 .name = "ap:offline",
1707 /* First state is scheduler control. Interrupts are disabled */
1708 [CPUHP_AP_SCHED_STARTING] = {
1709 .name = "sched:starting",
1710 .startup.single = sched_cpu_starting,
1711 .teardown.single = sched_cpu_dying,
1713 [CPUHP_AP_RCUTREE_DYING] = {
1714 .name = "RCU/tree:dying",
1715 .startup.single = NULL,
1716 .teardown.single = rcutree_dying_cpu,
1718 [CPUHP_AP_SMPCFD_DYING] = {
1719 .name = "smpcfd:dying",
1720 .startup.single = NULL,
1721 .teardown.single = smpcfd_dying_cpu,
1723 /* Entry state on starting. Interrupts enabled from here on. Transient
1724 * state for synchronsization */
1725 [CPUHP_AP_ONLINE] = {
1726 .name = "ap:online",
1729 * Handled on control processor until the plugged processor manages
1732 [CPUHP_TEARDOWN_CPU] = {
1733 .name = "cpu:teardown",
1734 .startup.single = NULL,
1735 .teardown.single = takedown_cpu,
1739 [CPUHP_AP_SCHED_WAIT_EMPTY] = {
1740 .name = "sched:waitempty",
1741 .startup.single = NULL,
1742 .teardown.single = sched_cpu_wait_empty,
1745 /* Handle smpboot threads park/unpark */
1746 [CPUHP_AP_SMPBOOT_THREADS] = {
1747 .name = "smpboot/threads:online",
1748 .startup.single = smpboot_unpark_threads,
1749 .teardown.single = smpboot_park_threads,
1751 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1752 .name = "irq/affinity:online",
1753 .startup.single = irq_affinity_online_cpu,
1754 .teardown.single = NULL,
1756 [CPUHP_AP_PERF_ONLINE] = {
1757 .name = "perf:online",
1758 .startup.single = perf_event_init_cpu,
1759 .teardown.single = perf_event_exit_cpu,
1761 [CPUHP_AP_WATCHDOG_ONLINE] = {
1762 .name = "lockup_detector:online",
1763 .startup.single = lockup_detector_online_cpu,
1764 .teardown.single = lockup_detector_offline_cpu,
1766 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1767 .name = "workqueue:online",
1768 .startup.single = workqueue_online_cpu,
1769 .teardown.single = workqueue_offline_cpu,
1771 [CPUHP_AP_RCUTREE_ONLINE] = {
1772 .name = "RCU/tree:online",
1773 .startup.single = rcutree_online_cpu,
1774 .teardown.single = rcutree_offline_cpu,
1778 * The dynamically registered state space is here
1782 /* Last state is scheduler control setting the cpu active */
1783 [CPUHP_AP_ACTIVE] = {
1784 .name = "sched:active",
1785 .startup.single = sched_cpu_activate,
1786 .teardown.single = sched_cpu_deactivate,
1790 /* CPU is fully up and running. */
1793 .startup.single = NULL,
1794 .teardown.single = NULL,
1798 /* Sanity check for callbacks */
1799 static int cpuhp_cb_check(enum cpuhp_state state)
1801 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1807 * Returns a free for dynamic slot assignment of the Online state. The states
1808 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1809 * by having no name assigned.
1811 static int cpuhp_reserve_state(enum cpuhp_state state)
1813 enum cpuhp_state i, end;
1814 struct cpuhp_step *step;
1817 case CPUHP_AP_ONLINE_DYN:
1818 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1819 end = CPUHP_AP_ONLINE_DYN_END;
1821 case CPUHP_BP_PREPARE_DYN:
1822 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1823 end = CPUHP_BP_PREPARE_DYN_END;
1829 for (i = state; i <= end; i++, step++) {
1833 WARN(1, "No more dynamic states available for CPU hotplug\n");
1837 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1838 int (*startup)(unsigned int cpu),
1839 int (*teardown)(unsigned int cpu),
1840 bool multi_instance)
1842 /* (Un)Install the callbacks for further cpu hotplug operations */
1843 struct cpuhp_step *sp;
1847 * If name is NULL, then the state gets removed.
1849 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1850 * the first allocation from these dynamic ranges, so the removal
1851 * would trigger a new allocation and clear the wrong (already
1852 * empty) state, leaving the callbacks of the to be cleared state
1853 * dangling, which causes wreckage on the next hotplug operation.
1855 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1856 state == CPUHP_BP_PREPARE_DYN)) {
1857 ret = cpuhp_reserve_state(state);
1862 sp = cpuhp_get_step(state);
1863 if (name && sp->name)
1866 sp->startup.single = startup;
1867 sp->teardown.single = teardown;
1869 sp->multi_instance = multi_instance;
1870 INIT_HLIST_HEAD(&sp->list);
1874 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1876 return cpuhp_get_step(state)->teardown.single;
1880 * Call the startup/teardown function for a step either on the AP or
1881 * on the current CPU.
1883 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1884 struct hlist_node *node)
1886 struct cpuhp_step *sp = cpuhp_get_step(state);
1890 * If there's nothing to do, we done.
1891 * Relies on the union for multi_instance.
1893 if (cpuhp_step_empty(bringup, sp))
1896 * The non AP bound callbacks can fail on bringup. On teardown
1897 * e.g. module removal we crash for now.
1900 if (cpuhp_is_ap_state(state))
1901 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1903 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1905 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1907 BUG_ON(ret && !bringup);
1912 * Called from __cpuhp_setup_state on a recoverable failure.
1914 * Note: The teardown callbacks for rollback are not allowed to fail!
1916 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1917 struct hlist_node *node)
1921 /* Roll back the already executed steps on the other cpus */
1922 for_each_present_cpu(cpu) {
1923 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1924 int cpustate = st->state;
1926 if (cpu >= failedcpu)
1929 /* Did we invoke the startup call on that cpu ? */
1930 if (cpustate >= state)
1931 cpuhp_issue_call(cpu, state, false, node);
1935 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1936 struct hlist_node *node,
1939 struct cpuhp_step *sp;
1943 lockdep_assert_cpus_held();
1945 sp = cpuhp_get_step(state);
1946 if (sp->multi_instance == false)
1949 mutex_lock(&cpuhp_state_mutex);
1951 if (!invoke || !sp->startup.multi)
1955 * Try to call the startup callback for each present cpu
1956 * depending on the hotplug state of the cpu.
1958 for_each_present_cpu(cpu) {
1959 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1960 int cpustate = st->state;
1962 if (cpustate < state)
1965 ret = cpuhp_issue_call(cpu, state, true, node);
1967 if (sp->teardown.multi)
1968 cpuhp_rollback_install(cpu, state, node);
1974 hlist_add_head(node, &sp->list);
1976 mutex_unlock(&cpuhp_state_mutex);
1980 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1986 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1990 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1993 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1994 * @state: The state to setup
1995 * @name: Name of the step
1996 * @invoke: If true, the startup function is invoked for cpus where
1997 * cpu state >= @state
1998 * @startup: startup callback function
1999 * @teardown: teardown callback function
2000 * @multi_instance: State is set up for multiple instances which get
2003 * The caller needs to hold cpus read locked while calling this function.
2006 * Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2007 * 0 for all other states
2008 * On failure: proper (negative) error code
2010 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
2011 const char *name, bool invoke,
2012 int (*startup)(unsigned int cpu),
2013 int (*teardown)(unsigned int cpu),
2014 bool multi_instance)
2019 lockdep_assert_cpus_held();
2021 if (cpuhp_cb_check(state) || !name)
2024 mutex_lock(&cpuhp_state_mutex);
2026 ret = cpuhp_store_callbacks(state, name, startup, teardown,
2029 dynstate = state == CPUHP_AP_ONLINE_DYN;
2030 if (ret > 0 && dynstate) {
2035 if (ret || !invoke || !startup)
2039 * Try to call the startup callback for each present cpu
2040 * depending on the hotplug state of the cpu.
2042 for_each_present_cpu(cpu) {
2043 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2044 int cpustate = st->state;
2046 if (cpustate < state)
2049 ret = cpuhp_issue_call(cpu, state, true, NULL);
2052 cpuhp_rollback_install(cpu, state, NULL);
2053 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2058 mutex_unlock(&cpuhp_state_mutex);
2060 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2061 * dynamically allocated state in case of success.
2063 if (!ret && dynstate)
2067 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
2069 int __cpuhp_setup_state(enum cpuhp_state state,
2070 const char *name, bool invoke,
2071 int (*startup)(unsigned int cpu),
2072 int (*teardown)(unsigned int cpu),
2073 bool multi_instance)
2078 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2079 teardown, multi_instance);
2083 EXPORT_SYMBOL(__cpuhp_setup_state);
2085 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2086 struct hlist_node *node, bool invoke)
2088 struct cpuhp_step *sp = cpuhp_get_step(state);
2091 BUG_ON(cpuhp_cb_check(state));
2093 if (!sp->multi_instance)
2097 mutex_lock(&cpuhp_state_mutex);
2099 if (!invoke || !cpuhp_get_teardown_cb(state))
2102 * Call the teardown callback for each present cpu depending
2103 * on the hotplug state of the cpu. This function is not
2104 * allowed to fail currently!
2106 for_each_present_cpu(cpu) {
2107 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2108 int cpustate = st->state;
2110 if (cpustate >= state)
2111 cpuhp_issue_call(cpu, state, false, node);
2116 mutex_unlock(&cpuhp_state_mutex);
2121 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2124 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2125 * @state: The state to remove
2126 * @invoke: If true, the teardown function is invoked for cpus where
2127 * cpu state >= @state
2129 * The caller needs to hold cpus read locked while calling this function.
2130 * The teardown callback is currently not allowed to fail. Think
2131 * about module removal!
2133 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2135 struct cpuhp_step *sp = cpuhp_get_step(state);
2138 BUG_ON(cpuhp_cb_check(state));
2140 lockdep_assert_cpus_held();
2142 mutex_lock(&cpuhp_state_mutex);
2143 if (sp->multi_instance) {
2144 WARN(!hlist_empty(&sp->list),
2145 "Error: Removing state %d which has instances left.\n",
2150 if (!invoke || !cpuhp_get_teardown_cb(state))
2154 * Call the teardown callback for each present cpu depending
2155 * on the hotplug state of the cpu. This function is not
2156 * allowed to fail currently!
2158 for_each_present_cpu(cpu) {
2159 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2160 int cpustate = st->state;
2162 if (cpustate >= state)
2163 cpuhp_issue_call(cpu, state, false, NULL);
2166 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2167 mutex_unlock(&cpuhp_state_mutex);
2169 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2171 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2174 __cpuhp_remove_state_cpuslocked(state, invoke);
2177 EXPORT_SYMBOL(__cpuhp_remove_state);
2179 #ifdef CONFIG_HOTPLUG_SMT
2180 static void cpuhp_offline_cpu_device(unsigned int cpu)
2182 struct device *dev = get_cpu_device(cpu);
2184 dev->offline = true;
2185 /* Tell user space about the state change */
2186 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2189 static void cpuhp_online_cpu_device(unsigned int cpu)
2191 struct device *dev = get_cpu_device(cpu);
2193 dev->offline = false;
2194 /* Tell user space about the state change */
2195 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2198 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2202 cpu_maps_update_begin();
2203 for_each_online_cpu(cpu) {
2204 if (topology_is_primary_thread(cpu))
2206 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2210 * As this needs to hold the cpu maps lock it's impossible
2211 * to call device_offline() because that ends up calling
2212 * cpu_down() which takes cpu maps lock. cpu maps lock
2213 * needs to be held as this might race against in kernel
2214 * abusers of the hotplug machinery (thermal management).
2216 * So nothing would update device:offline state. That would
2217 * leave the sysfs entry stale and prevent onlining after
2218 * smt control has been changed to 'off' again. This is
2219 * called under the sysfs hotplug lock, so it is properly
2220 * serialized against the regular offline usage.
2222 cpuhp_offline_cpu_device(cpu);
2225 cpu_smt_control = ctrlval;
2226 cpu_maps_update_done();
2230 int cpuhp_smt_enable(void)
2234 cpu_maps_update_begin();
2235 cpu_smt_control = CPU_SMT_ENABLED;
2236 for_each_present_cpu(cpu) {
2237 /* Skip online CPUs and CPUs on offline nodes */
2238 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2240 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2243 /* See comment in cpuhp_smt_disable() */
2244 cpuhp_online_cpu_device(cpu);
2246 cpu_maps_update_done();
2251 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2252 static ssize_t show_cpuhp_state(struct device *dev,
2253 struct device_attribute *attr, char *buf)
2255 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2257 return sprintf(buf, "%d\n", st->state);
2259 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
2261 static ssize_t write_cpuhp_target(struct device *dev,
2262 struct device_attribute *attr,
2263 const char *buf, size_t count)
2265 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2266 struct cpuhp_step *sp;
2269 ret = kstrtoint(buf, 10, &target);
2273 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2274 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2277 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2281 ret = lock_device_hotplug_sysfs();
2285 mutex_lock(&cpuhp_state_mutex);
2286 sp = cpuhp_get_step(target);
2287 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2288 mutex_unlock(&cpuhp_state_mutex);
2292 if (st->state < target)
2293 ret = cpu_up(dev->id, target);
2295 ret = cpu_down(dev->id, target);
2297 unlock_device_hotplug();
2298 return ret ? ret : count;
2301 static ssize_t show_cpuhp_target(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->target);
2308 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
2311 static ssize_t write_cpuhp_fail(struct device *dev,
2312 struct device_attribute *attr,
2313 const char *buf, size_t count)
2315 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2316 struct cpuhp_step *sp;
2319 ret = kstrtoint(buf, 10, &fail);
2323 if (fail == CPUHP_INVALID) {
2328 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2332 * Cannot fail STARTING/DYING callbacks.
2334 if (cpuhp_is_atomic_state(fail))
2338 * DEAD callbacks cannot fail...
2339 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2340 * triggering STARTING callbacks, a failure in this state would
2343 if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU)
2347 * Cannot fail anything that doesn't have callbacks.
2349 mutex_lock(&cpuhp_state_mutex);
2350 sp = cpuhp_get_step(fail);
2351 if (!sp->startup.single && !sp->teardown.single)
2353 mutex_unlock(&cpuhp_state_mutex);
2362 static ssize_t show_cpuhp_fail(struct device *dev,
2363 struct device_attribute *attr, char *buf)
2365 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2367 return sprintf(buf, "%d\n", st->fail);
2370 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
2372 static struct attribute *cpuhp_cpu_attrs[] = {
2373 &dev_attr_state.attr,
2374 &dev_attr_target.attr,
2375 &dev_attr_fail.attr,
2379 static const struct attribute_group cpuhp_cpu_attr_group = {
2380 .attrs = cpuhp_cpu_attrs,
2385 static ssize_t show_cpuhp_states(struct device *dev,
2386 struct device_attribute *attr, char *buf)
2388 ssize_t cur, res = 0;
2391 mutex_lock(&cpuhp_state_mutex);
2392 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2393 struct cpuhp_step *sp = cpuhp_get_step(i);
2396 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2401 mutex_unlock(&cpuhp_state_mutex);
2404 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
2406 static struct attribute *cpuhp_cpu_root_attrs[] = {
2407 &dev_attr_states.attr,
2411 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2412 .attrs = cpuhp_cpu_root_attrs,
2417 #ifdef CONFIG_HOTPLUG_SMT
2420 __store_smt_control(struct device *dev, struct device_attribute *attr,
2421 const char *buf, size_t count)
2425 if (sysfs_streq(buf, "on"))
2426 ctrlval = CPU_SMT_ENABLED;
2427 else if (sysfs_streq(buf, "off"))
2428 ctrlval = CPU_SMT_DISABLED;
2429 else if (sysfs_streq(buf, "forceoff"))
2430 ctrlval = CPU_SMT_FORCE_DISABLED;
2434 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2437 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2440 ret = lock_device_hotplug_sysfs();
2444 if (ctrlval != cpu_smt_control) {
2446 case CPU_SMT_ENABLED:
2447 ret = cpuhp_smt_enable();
2449 case CPU_SMT_DISABLED:
2450 case CPU_SMT_FORCE_DISABLED:
2451 ret = cpuhp_smt_disable(ctrlval);
2456 unlock_device_hotplug();
2457 return ret ? ret : count;
2460 #else /* !CONFIG_HOTPLUG_SMT */
2462 __store_smt_control(struct device *dev, struct device_attribute *attr,
2463 const char *buf, size_t count)
2467 #endif /* CONFIG_HOTPLUG_SMT */
2469 static const char *smt_states[] = {
2470 [CPU_SMT_ENABLED] = "on",
2471 [CPU_SMT_DISABLED] = "off",
2472 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2473 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2474 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2478 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2480 const char *state = smt_states[cpu_smt_control];
2482 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2486 store_smt_control(struct device *dev, struct device_attribute *attr,
2487 const char *buf, size_t count)
2489 return __store_smt_control(dev, attr, buf, count);
2491 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2494 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2496 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2498 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2500 static struct attribute *cpuhp_smt_attrs[] = {
2501 &dev_attr_control.attr,
2502 &dev_attr_active.attr,
2506 static const struct attribute_group cpuhp_smt_attr_group = {
2507 .attrs = cpuhp_smt_attrs,
2512 static int __init cpu_smt_sysfs_init(void)
2514 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2515 &cpuhp_smt_attr_group);
2518 static int __init cpuhp_sysfs_init(void)
2522 ret = cpu_smt_sysfs_init();
2526 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2527 &cpuhp_cpu_root_attr_group);
2531 for_each_possible_cpu(cpu) {
2532 struct device *dev = get_cpu_device(cpu);
2536 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2542 device_initcall(cpuhp_sysfs_init);
2543 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2546 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2547 * represents all NR_CPUS bits binary values of 1<<nr.
2549 * It is used by cpumask_of() to get a constant address to a CPU
2550 * mask value that has a single bit set only.
2553 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2554 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2555 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2556 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2557 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2559 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2561 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2562 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2563 #if BITS_PER_LONG > 32
2564 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2565 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2568 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2570 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2571 EXPORT_SYMBOL(cpu_all_bits);
2573 #ifdef CONFIG_INIT_ALL_POSSIBLE
2574 struct cpumask __cpu_possible_mask __read_mostly
2577 struct cpumask __cpu_possible_mask __read_mostly;
2579 EXPORT_SYMBOL(__cpu_possible_mask);
2581 struct cpumask __cpu_online_mask __read_mostly;
2582 EXPORT_SYMBOL(__cpu_online_mask);
2584 struct cpumask __cpu_present_mask __read_mostly;
2585 EXPORT_SYMBOL(__cpu_present_mask);
2587 struct cpumask __cpu_active_mask __read_mostly;
2588 EXPORT_SYMBOL(__cpu_active_mask);
2590 struct cpumask __cpu_dying_mask __read_mostly;
2591 EXPORT_SYMBOL(__cpu_dying_mask);
2593 atomic_t __num_online_cpus __read_mostly;
2594 EXPORT_SYMBOL(__num_online_cpus);
2596 void init_cpu_present(const struct cpumask *src)
2598 cpumask_copy(&__cpu_present_mask, src);
2601 void init_cpu_possible(const struct cpumask *src)
2603 cpumask_copy(&__cpu_possible_mask, src);
2606 void init_cpu_online(const struct cpumask *src)
2608 cpumask_copy(&__cpu_online_mask, src);
2611 void set_cpu_online(unsigned int cpu, bool online)
2614 * atomic_inc/dec() is required to handle the horrid abuse of this
2615 * function by the reboot and kexec code which invoke it from
2616 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2617 * regular CPU hotplug is properly serialized.
2619 * Note, that the fact that __num_online_cpus is of type atomic_t
2620 * does not protect readers which are not serialized against
2621 * concurrent hotplug operations.
2624 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2625 atomic_inc(&__num_online_cpus);
2627 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2628 atomic_dec(&__num_online_cpus);
2633 * Activate the first processor.
2635 void __init boot_cpu_init(void)
2637 int cpu = smp_processor_id();
2639 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2640 set_cpu_online(cpu, true);
2641 set_cpu_active(cpu, true);
2642 set_cpu_present(cpu, true);
2643 set_cpu_possible(cpu, true);
2646 __boot_cpu_id = cpu;
2651 * Must be called _AFTER_ setting up the per_cpu areas
2653 void __init boot_cpu_hotplug_init(void)
2656 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2658 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2662 * These are used for a global "mitigations=" cmdline option for toggling
2663 * optional CPU mitigations.
2665 enum cpu_mitigations {
2666 CPU_MITIGATIONS_OFF,
2667 CPU_MITIGATIONS_AUTO,
2668 CPU_MITIGATIONS_AUTO_NOSMT,
2671 static enum cpu_mitigations cpu_mitigations __ro_after_init =
2672 CPU_MITIGATIONS_AUTO;
2674 static int __init mitigations_parse_cmdline(char *arg)
2676 if (!strcmp(arg, "off"))
2677 cpu_mitigations = CPU_MITIGATIONS_OFF;
2678 else if (!strcmp(arg, "auto"))
2679 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2680 else if (!strcmp(arg, "auto,nosmt"))
2681 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2683 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2688 early_param("mitigations", mitigations_parse_cmdline);
2690 /* mitigations=off */
2691 bool cpu_mitigations_off(void)
2693 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2695 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2697 /* mitigations=auto,nosmt */
2698 bool cpu_mitigations_auto_nosmt(void)
2700 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2702 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);