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>
38 #include <linux/cc_platform.h>
40 #include <trace/events/power.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/cpuhp.h>
47 * struct cpuhp_cpu_state - Per cpu hotplug state storage
48 * @state: The current cpu state
49 * @target: The target state
50 * @fail: Current CPU hotplug callback state
51 * @thread: Pointer to the hotplug thread
52 * @should_run: Thread should execute
53 * @rollback: Perform a rollback
54 * @single: Single callback invocation
55 * @bringup: Single callback bringup or teardown selector
57 * @node: Remote CPU node; for multi-instance, do a
58 * single entry callback for install/remove
59 * @last: For multi-instance rollback, remember how far we got
60 * @cb_state: The state for a single callback (install/uninstall)
61 * @result: Result of the operation
62 * @done_up: Signal completion to the issuer of the task for cpu-up
63 * @done_down: Signal completion to the issuer of the task for cpu-down
65 struct cpuhp_cpu_state {
66 enum cpuhp_state state;
67 enum cpuhp_state target;
68 enum cpuhp_state fail;
70 struct task_struct *thread;
75 struct hlist_node *node;
76 struct hlist_node *last;
77 enum cpuhp_state cb_state;
79 struct completion done_up;
80 struct completion done_down;
84 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
85 .fail = CPUHP_INVALID,
89 cpumask_t cpus_booted_once_mask;
92 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
93 static struct lockdep_map cpuhp_state_up_map =
94 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
95 static struct lockdep_map cpuhp_state_down_map =
96 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
99 static inline void cpuhp_lock_acquire(bool bringup)
101 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
104 static inline void cpuhp_lock_release(bool bringup)
106 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
110 static inline void cpuhp_lock_acquire(bool bringup) { }
111 static inline void cpuhp_lock_release(bool bringup) { }
116 * struct cpuhp_step - Hotplug state machine step
117 * @name: Name of the step
118 * @startup: Startup function of the step
119 * @teardown: Teardown function of the step
120 * @cant_stop: Bringup/teardown can't be stopped at this step
121 * @multi_instance: State has multiple instances which get added afterwards
126 int (*single)(unsigned int cpu);
127 int (*multi)(unsigned int cpu,
128 struct hlist_node *node);
131 int (*single)(unsigned int cpu);
132 int (*multi)(unsigned int cpu,
133 struct hlist_node *node);
136 struct hlist_head list;
142 static DEFINE_MUTEX(cpuhp_state_mutex);
143 static struct cpuhp_step cpuhp_hp_states[];
145 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
147 return cpuhp_hp_states + state;
150 static bool cpuhp_step_empty(bool bringup, struct cpuhp_step *step)
152 return bringup ? !step->startup.single : !step->teardown.single;
156 * cpuhp_invoke_callback - Invoke the callbacks for a given state
157 * @cpu: The cpu for which the callback should be invoked
158 * @state: The state to do callbacks for
159 * @bringup: True if the bringup callback should be invoked
160 * @node: For multi-instance, do a single entry callback for install/remove
161 * @lastp: For multi-instance rollback, remember how far we got
163 * Called from cpu hotplug and from the state register machinery.
165 * Return: %0 on success or a negative errno code
167 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
168 bool bringup, struct hlist_node *node,
169 struct hlist_node **lastp)
171 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
172 struct cpuhp_step *step = cpuhp_get_step(state);
173 int (*cbm)(unsigned int cpu, struct hlist_node *node);
174 int (*cb)(unsigned int cpu);
177 if (st->fail == state) {
178 st->fail = CPUHP_INVALID;
182 if (cpuhp_step_empty(bringup, step)) {
187 if (!step->multi_instance) {
188 WARN_ON_ONCE(lastp && *lastp);
189 cb = bringup ? step->startup.single : step->teardown.single;
191 trace_cpuhp_enter(cpu, st->target, state, cb);
193 trace_cpuhp_exit(cpu, st->state, state, ret);
196 cbm = bringup ? step->startup.multi : step->teardown.multi;
198 /* Single invocation for instance add/remove */
200 WARN_ON_ONCE(lastp && *lastp);
201 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
202 ret = cbm(cpu, node);
203 trace_cpuhp_exit(cpu, st->state, state, ret);
207 /* State transition. Invoke on all instances */
209 hlist_for_each(node, &step->list) {
210 if (lastp && node == *lastp)
213 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
214 ret = cbm(cpu, node);
215 trace_cpuhp_exit(cpu, st->state, state, ret);
229 /* Rollback the instances if one failed */
230 cbm = !bringup ? step->startup.multi : step->teardown.multi;
234 hlist_for_each(node, &step->list) {
238 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
239 ret = cbm(cpu, node);
240 trace_cpuhp_exit(cpu, st->state, state, ret);
242 * Rollback must not fail,
250 static bool cpuhp_is_ap_state(enum cpuhp_state state)
253 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
254 * purposes as that state is handled explicitly in cpu_down.
256 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
259 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
261 struct completion *done = bringup ? &st->done_up : &st->done_down;
262 wait_for_completion(done);
265 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
267 struct completion *done = bringup ? &st->done_up : &st->done_down;
272 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
274 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
276 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
279 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
280 static DEFINE_MUTEX(cpu_add_remove_lock);
281 bool cpuhp_tasks_frozen;
282 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
285 * The following two APIs (cpu_maps_update_begin/done) must be used when
286 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
288 void cpu_maps_update_begin(void)
290 mutex_lock(&cpu_add_remove_lock);
293 void cpu_maps_update_done(void)
295 mutex_unlock(&cpu_add_remove_lock);
299 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
300 * Should always be manipulated under cpu_add_remove_lock
302 static int cpu_hotplug_disabled;
304 #ifdef CONFIG_HOTPLUG_CPU
306 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
308 void cpus_read_lock(void)
310 percpu_down_read(&cpu_hotplug_lock);
312 EXPORT_SYMBOL_GPL(cpus_read_lock);
314 int cpus_read_trylock(void)
316 return percpu_down_read_trylock(&cpu_hotplug_lock);
318 EXPORT_SYMBOL_GPL(cpus_read_trylock);
320 void cpus_read_unlock(void)
322 percpu_up_read(&cpu_hotplug_lock);
324 EXPORT_SYMBOL_GPL(cpus_read_unlock);
326 void cpus_write_lock(void)
328 percpu_down_write(&cpu_hotplug_lock);
331 void cpus_write_unlock(void)
333 percpu_up_write(&cpu_hotplug_lock);
336 void lockdep_assert_cpus_held(void)
339 * We can't have hotplug operations before userspace starts running,
340 * and some init codepaths will knowingly not take the hotplug lock.
341 * This is all valid, so mute lockdep until it makes sense to report
344 if (system_state < SYSTEM_RUNNING)
347 percpu_rwsem_assert_held(&cpu_hotplug_lock);
350 #ifdef CONFIG_LOCKDEP
351 int lockdep_is_cpus_held(void)
353 return percpu_rwsem_is_held(&cpu_hotplug_lock);
357 static void lockdep_acquire_cpus_lock(void)
359 rwsem_acquire(&cpu_hotplug_lock.dep_map, 0, 0, _THIS_IP_);
362 static void lockdep_release_cpus_lock(void)
364 rwsem_release(&cpu_hotplug_lock.dep_map, _THIS_IP_);
368 * Wait for currently running CPU hotplug operations to complete (if any) and
369 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
370 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
371 * hotplug path before performing hotplug operations. So acquiring that lock
372 * guarantees mutual exclusion from any currently running hotplug operations.
374 void cpu_hotplug_disable(void)
376 cpu_maps_update_begin();
377 cpu_hotplug_disabled++;
378 cpu_maps_update_done();
380 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
382 static void __cpu_hotplug_enable(void)
384 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
386 cpu_hotplug_disabled--;
389 void cpu_hotplug_enable(void)
391 cpu_maps_update_begin();
392 __cpu_hotplug_enable();
393 cpu_maps_update_done();
395 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
399 static void lockdep_acquire_cpus_lock(void)
403 static void lockdep_release_cpus_lock(void)
407 #endif /* CONFIG_HOTPLUG_CPU */
410 * Architectures that need SMT-specific errata handling during SMT hotplug
411 * should override this.
413 void __weak arch_smt_update(void) { }
415 #ifdef CONFIG_HOTPLUG_SMT
416 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
418 void __init cpu_smt_disable(bool force)
420 if (!cpu_smt_possible())
424 pr_info("SMT: Force disabled\n");
425 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
427 pr_info("SMT: disabled\n");
428 cpu_smt_control = CPU_SMT_DISABLED;
433 * The decision whether SMT is supported can only be done after the full
434 * CPU identification. Called from architecture code.
436 void __init cpu_smt_check_topology(void)
438 if (!topology_smt_supported())
439 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
442 static int __init smt_cmdline_disable(char *str)
444 cpu_smt_disable(str && !strcmp(str, "force"));
447 early_param("nosmt", smt_cmdline_disable);
449 static inline bool cpu_smt_allowed(unsigned int cpu)
451 if (cpu_smt_control == CPU_SMT_ENABLED)
454 if (topology_is_primary_thread(cpu))
458 * On x86 it's required to boot all logical CPUs at least once so
459 * that the init code can get a chance to set CR4.MCE on each
460 * CPU. Otherwise, a broadcasted MCE observing CR4.MCE=0b on any
461 * core will shutdown the machine.
463 return !cpumask_test_cpu(cpu, &cpus_booted_once_mask);
466 /* Returns true if SMT is not supported of forcefully (irreversibly) disabled */
467 bool cpu_smt_possible(void)
469 return cpu_smt_control != CPU_SMT_FORCE_DISABLED &&
470 cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
472 EXPORT_SYMBOL_GPL(cpu_smt_possible);
474 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
477 static inline enum cpuhp_state
478 cpuhp_set_state(int cpu, struct cpuhp_cpu_state *st, enum cpuhp_state target)
480 enum cpuhp_state prev_state = st->state;
481 bool bringup = st->state < target;
483 st->rollback = false;
488 st->bringup = bringup;
489 if (cpu_dying(cpu) != !bringup)
490 set_cpu_dying(cpu, !bringup);
496 cpuhp_reset_state(int cpu, struct cpuhp_cpu_state *st,
497 enum cpuhp_state prev_state)
499 bool bringup = !st->bringup;
501 st->target = prev_state;
504 * Already rolling back. No need invert the bringup value or to change
513 * If we have st->last we need to undo partial multi_instance of this
514 * state first. Otherwise start undo at the previous state.
523 st->bringup = bringup;
524 if (cpu_dying(cpu) != !bringup)
525 set_cpu_dying(cpu, !bringup);
528 /* Regular hotplug invocation of the AP hotplug thread */
529 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
531 if (!st->single && st->state == st->target)
536 * Make sure the above stores are visible before should_run becomes
537 * true. Paired with the mb() above in cpuhp_thread_fun()
540 st->should_run = true;
541 wake_up_process(st->thread);
542 wait_for_ap_thread(st, st->bringup);
545 static int cpuhp_kick_ap(int cpu, struct cpuhp_cpu_state *st,
546 enum cpuhp_state target)
548 enum cpuhp_state prev_state;
551 prev_state = cpuhp_set_state(cpu, st, target);
553 if ((ret = st->result)) {
554 cpuhp_reset_state(cpu, st, prev_state);
561 static int bringup_wait_for_ap(unsigned int cpu)
563 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
565 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
566 wait_for_ap_thread(st, true);
567 if (WARN_ON_ONCE((!cpu_online(cpu))))
570 /* Unpark the hotplug thread of the target cpu */
571 kthread_unpark(st->thread);
574 * SMT soft disabling on X86 requires to bring the CPU out of the
575 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
576 * CPU marked itself as booted_once in notify_cpu_starting() so the
577 * cpu_smt_allowed() check will now return false if this is not the
580 if (!cpu_smt_allowed(cpu))
583 if (st->target <= CPUHP_AP_ONLINE_IDLE)
586 return cpuhp_kick_ap(cpu, st, st->target);
589 static int bringup_cpu(unsigned int cpu)
591 struct task_struct *idle = idle_thread_get(cpu);
595 * Reset stale stack state from the last time this CPU was online.
597 scs_task_reset(idle);
598 kasan_unpoison_task_stack(idle);
601 * Some architectures have to walk the irq descriptors to
602 * setup the vector space for the cpu which comes online.
603 * Prevent irq alloc/free across the bringup.
607 /* Arch-specific enabling code. */
608 ret = __cpu_up(cpu, idle);
612 return bringup_wait_for_ap(cpu);
615 static int finish_cpu(unsigned int cpu)
617 struct task_struct *idle = idle_thread_get(cpu);
618 struct mm_struct *mm = idle->active_mm;
621 * idle_task_exit() will have switched to &init_mm, now
622 * clean up any remaining active_mm state.
625 idle->active_mm = &init_mm;
631 * Hotplug state machine related functions
635 * Get the next state to run. Empty ones will be skipped. Returns true if a
638 * st->state will be modified ahead of time, to match state_to_run, as if it
641 static bool cpuhp_next_state(bool bringup,
642 enum cpuhp_state *state_to_run,
643 struct cpuhp_cpu_state *st,
644 enum cpuhp_state target)
648 if (st->state >= target)
651 *state_to_run = ++st->state;
653 if (st->state <= target)
656 *state_to_run = st->state--;
659 if (!cpuhp_step_empty(bringup, cpuhp_get_step(*state_to_run)))
666 static int __cpuhp_invoke_callback_range(bool bringup,
668 struct cpuhp_cpu_state *st,
669 enum cpuhp_state target,
672 enum cpuhp_state state;
675 while (cpuhp_next_state(bringup, &state, st, target)) {
678 err = cpuhp_invoke_callback(cpu, state, bringup, NULL, NULL);
683 pr_warn("CPU %u %s state %s (%d) failed (%d)\n",
684 cpu, bringup ? "UP" : "DOWN",
685 cpuhp_get_step(st->state)->name,
697 static inline int cpuhp_invoke_callback_range(bool bringup,
699 struct cpuhp_cpu_state *st,
700 enum cpuhp_state target)
702 return __cpuhp_invoke_callback_range(bringup, cpu, st, target, false);
705 static inline void cpuhp_invoke_callback_range_nofail(bool bringup,
707 struct cpuhp_cpu_state *st,
708 enum cpuhp_state target)
710 __cpuhp_invoke_callback_range(bringup, cpu, st, target, true);
713 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
715 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
718 * When CPU hotplug is disabled, then taking the CPU down is not
719 * possible because takedown_cpu() and the architecture and
720 * subsystem specific mechanisms are not available. So the CPU
721 * which would be completely unplugged again needs to stay around
722 * in the current state.
724 return st->state <= CPUHP_BRINGUP_CPU;
727 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
728 enum cpuhp_state target)
730 enum cpuhp_state prev_state = st->state;
733 ret = cpuhp_invoke_callback_range(true, cpu, st, target);
735 pr_debug("CPU UP failed (%d) CPU %u state %s (%d)\n",
736 ret, cpu, cpuhp_get_step(st->state)->name,
739 cpuhp_reset_state(cpu, st, prev_state);
740 if (can_rollback_cpu(st))
741 WARN_ON(cpuhp_invoke_callback_range(false, cpu, st,
748 * The cpu hotplug threads manage the bringup and teardown of the cpus
750 static int cpuhp_should_run(unsigned int cpu)
752 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
754 return st->should_run;
758 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
759 * callbacks when a state gets [un]installed at runtime.
761 * Each invocation of this function by the smpboot thread does a single AP
764 * It has 3 modes of operation:
765 * - single: runs st->cb_state
766 * - up: runs ++st->state, while st->state < st->target
767 * - down: runs st->state--, while st->state > st->target
769 * When complete or on error, should_run is cleared and the completion is fired.
771 static void cpuhp_thread_fun(unsigned int cpu)
773 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
774 bool bringup = st->bringup;
775 enum cpuhp_state state;
777 if (WARN_ON_ONCE(!st->should_run))
781 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
782 * that if we see ->should_run we also see the rest of the state.
787 * The BP holds the hotplug lock, but we're now running on the AP,
788 * ensure that anybody asserting the lock is held, will actually find
791 lockdep_acquire_cpus_lock();
792 cpuhp_lock_acquire(bringup);
795 state = st->cb_state;
796 st->should_run = false;
798 st->should_run = cpuhp_next_state(bringup, &state, st, st->target);
803 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
805 if (cpuhp_is_atomic_state(state)) {
807 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
811 * STARTING/DYING must not fail!
813 WARN_ON_ONCE(st->result);
815 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
820 * If we fail on a rollback, we're up a creek without no
821 * paddle, no way forward, no way back. We loose, thanks for
824 WARN_ON_ONCE(st->rollback);
825 st->should_run = false;
829 cpuhp_lock_release(bringup);
830 lockdep_release_cpus_lock();
833 complete_ap_thread(st, bringup);
836 /* Invoke a single callback on a remote cpu */
838 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
839 struct hlist_node *node)
841 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
844 if (!cpu_online(cpu))
847 cpuhp_lock_acquire(false);
848 cpuhp_lock_release(false);
850 cpuhp_lock_acquire(true);
851 cpuhp_lock_release(true);
854 * If we are up and running, use the hotplug thread. For early calls
855 * we invoke the thread function directly.
858 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
860 st->rollback = false;
864 st->bringup = bringup;
865 st->cb_state = state;
871 * If we failed and did a partial, do a rollback.
873 if ((ret = st->result) && st->last) {
875 st->bringup = !bringup;
881 * Clean up the leftovers so the next hotplug operation wont use stale
884 st->node = st->last = NULL;
888 static int cpuhp_kick_ap_work(unsigned int cpu)
890 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
891 enum cpuhp_state prev_state = st->state;
894 cpuhp_lock_acquire(false);
895 cpuhp_lock_release(false);
897 cpuhp_lock_acquire(true);
898 cpuhp_lock_release(true);
900 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
901 ret = cpuhp_kick_ap(cpu, st, st->target);
902 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
907 static struct smp_hotplug_thread cpuhp_threads = {
908 .store = &cpuhp_state.thread,
909 .thread_should_run = cpuhp_should_run,
910 .thread_fn = cpuhp_thread_fun,
911 .thread_comm = "cpuhp/%u",
915 static __init void cpuhp_init_state(void)
917 struct cpuhp_cpu_state *st;
920 for_each_possible_cpu(cpu) {
921 st = per_cpu_ptr(&cpuhp_state, cpu);
922 init_completion(&st->done_up);
923 init_completion(&st->done_down);
927 void __init cpuhp_threads_init(void)
930 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
931 kthread_unpark(this_cpu_read(cpuhp_state.thread));
936 * Serialize hotplug trainwrecks outside of the cpu_hotplug_lock
939 * The operation is still serialized against concurrent CPU hotplug via
940 * cpu_add_remove_lock, i.e. CPU map protection. But it is _not_
941 * serialized against other hotplug related activity like adding or
942 * removing of state callbacks and state instances, which invoke either the
943 * startup or the teardown callback of the affected state.
945 * This is required for subsystems which are unfixable vs. CPU hotplug and
946 * evade lock inversion problems by scheduling work which has to be
947 * completed _before_ cpu_up()/_cpu_down() returns.
949 * Don't even think about adding anything to this for any new code or even
950 * drivers. It's only purpose is to keep existing lock order trainwrecks
953 * For cpu_down() there might be valid reasons to finish cleanups which are
954 * not required to be done under cpu_hotplug_lock, but that's a different
955 * story and would be not invoked via this.
957 static void cpu_up_down_serialize_trainwrecks(bool tasks_frozen)
960 * cpusets delegate hotplug operations to a worker to "solve" the
961 * lock order problems. Wait for the worker, but only if tasks are
962 * _not_ frozen (suspend, hibernate) as that would wait forever.
964 * The wait is required because otherwise the hotplug operation
965 * returns with inconsistent state, which could even be observed in
966 * user space when a new CPU is brought up. The CPU plug uevent
967 * would be delivered and user space reacting on it would fail to
968 * move tasks to the newly plugged CPU up to the point where the
969 * work has finished because up to that point the newly plugged CPU
970 * is not assignable in cpusets/cgroups. On unplug that's not
971 * necessarily a visible issue, but it is still inconsistent state,
972 * which is the real problem which needs to be "fixed". This can't
973 * prevent the transient state between scheduling the work and
974 * returning from waiting for it.
977 cpuset_wait_for_hotplug();
980 #ifdef CONFIG_HOTPLUG_CPU
981 #ifndef arch_clear_mm_cpumask_cpu
982 #define arch_clear_mm_cpumask_cpu(cpu, mm) cpumask_clear_cpu(cpu, mm_cpumask(mm))
986 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
989 * This function walks all processes, finds a valid mm struct for each one and
990 * then clears a corresponding bit in mm's cpumask. While this all sounds
991 * trivial, there are various non-obvious corner cases, which this function
992 * tries to solve in a safe manner.
994 * Also note that the function uses a somewhat relaxed locking scheme, so it may
995 * be called only for an already offlined CPU.
997 void clear_tasks_mm_cpumask(int cpu)
999 struct task_struct *p;
1002 * This function is called after the cpu is taken down and marked
1003 * offline, so its not like new tasks will ever get this cpu set in
1004 * their mm mask. -- Peter Zijlstra
1005 * Thus, we may use rcu_read_lock() here, instead of grabbing
1006 * full-fledged tasklist_lock.
1008 WARN_ON(cpu_online(cpu));
1010 for_each_process(p) {
1011 struct task_struct *t;
1014 * Main thread might exit, but other threads may still have
1015 * a valid mm. Find one.
1017 t = find_lock_task_mm(p);
1020 arch_clear_mm_cpumask_cpu(cpu, t->mm);
1026 /* Take this CPU down. */
1027 static int take_cpu_down(void *_param)
1029 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1030 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
1031 int err, cpu = smp_processor_id();
1033 /* Ensure this CPU doesn't handle any more interrupts. */
1034 err = __cpu_disable();
1039 * Must be called from CPUHP_TEARDOWN_CPU, which means, as we are going
1040 * down, that the current state is CPUHP_TEARDOWN_CPU - 1.
1042 WARN_ON(st->state != (CPUHP_TEARDOWN_CPU - 1));
1045 * Invoke the former CPU_DYING callbacks. DYING must not fail!
1047 cpuhp_invoke_callback_range_nofail(false, cpu, st, target);
1049 /* Give up timekeeping duties */
1050 tick_handover_do_timer();
1051 /* Remove CPU from timer broadcasting */
1052 tick_offline_cpu(cpu);
1053 /* Park the stopper thread */
1054 stop_machine_park(cpu);
1058 static int takedown_cpu(unsigned int cpu)
1060 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1063 /* Park the smpboot threads */
1064 kthread_park(st->thread);
1067 * Prevent irq alloc/free while the dying cpu reorganizes the
1068 * interrupt affinities.
1073 * So now all preempt/rcu users must observe !cpu_active().
1075 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
1077 /* CPU refused to die */
1078 irq_unlock_sparse();
1079 /* Unpark the hotplug thread so we can rollback there */
1080 kthread_unpark(st->thread);
1083 BUG_ON(cpu_online(cpu));
1086 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
1087 * all runnable tasks from the CPU, there's only the idle task left now
1088 * that the migration thread is done doing the stop_machine thing.
1090 * Wait for the stop thread to go away.
1092 wait_for_ap_thread(st, false);
1093 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
1095 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
1096 irq_unlock_sparse();
1098 hotplug_cpu__broadcast_tick_pull(cpu);
1099 /* This actually kills the CPU. */
1102 tick_cleanup_dead_cpu(cpu);
1103 rcutree_migrate_callbacks(cpu);
1107 static void cpuhp_complete_idle_dead(void *arg)
1109 struct cpuhp_cpu_state *st = arg;
1111 complete_ap_thread(st, false);
1114 void cpuhp_report_idle_dead(void)
1116 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1118 BUG_ON(st->state != CPUHP_AP_OFFLINE);
1119 rcu_report_dead(smp_processor_id());
1120 st->state = CPUHP_AP_IDLE_DEAD;
1122 * We cannot call complete after rcu_report_dead() so we delegate it
1125 smp_call_function_single(cpumask_first(cpu_online_mask),
1126 cpuhp_complete_idle_dead, st, 0);
1129 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
1130 enum cpuhp_state target)
1132 enum cpuhp_state prev_state = st->state;
1135 ret = cpuhp_invoke_callback_range(false, cpu, st, target);
1137 pr_debug("CPU DOWN failed (%d) CPU %u state %s (%d)\n",
1138 ret, cpu, cpuhp_get_step(st->state)->name,
1141 cpuhp_reset_state(cpu, st, prev_state);
1143 if (st->state < prev_state)
1144 WARN_ON(cpuhp_invoke_callback_range(true, cpu, st,
1151 /* Requires cpu_add_remove_lock to be held */
1152 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
1153 enum cpuhp_state target)
1155 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1156 int prev_state, ret = 0;
1158 if (num_online_cpus() == 1)
1161 if (!cpu_present(cpu))
1166 cpuhp_tasks_frozen = tasks_frozen;
1168 prev_state = cpuhp_set_state(cpu, st, target);
1170 * If the current CPU state is in the range of the AP hotplug thread,
1171 * then we need to kick the thread.
1173 if (st->state > CPUHP_TEARDOWN_CPU) {
1174 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
1175 ret = cpuhp_kick_ap_work(cpu);
1177 * The AP side has done the error rollback already. Just
1178 * return the error code..
1184 * We might have stopped still in the range of the AP hotplug
1185 * thread. Nothing to do anymore.
1187 if (st->state > CPUHP_TEARDOWN_CPU)
1190 st->target = target;
1193 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
1194 * to do the further cleanups.
1196 ret = cpuhp_down_callbacks(cpu, st, target);
1197 if (ret && st->state < prev_state) {
1198 if (st->state == CPUHP_TEARDOWN_CPU) {
1199 cpuhp_reset_state(cpu, st, prev_state);
1200 __cpuhp_kick_ap(st);
1202 WARN(1, "DEAD callback error for CPU%d", cpu);
1207 cpus_write_unlock();
1209 * Do post unplug cleanup. This is still protected against
1210 * concurrent CPU hotplug via cpu_add_remove_lock.
1212 lockup_detector_cleanup();
1214 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1218 struct cpu_down_work {
1220 enum cpuhp_state target;
1223 static long __cpu_down_maps_locked(void *arg)
1225 struct cpu_down_work *work = arg;
1227 return _cpu_down(work->cpu, 0, work->target);
1230 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
1232 struct cpu_down_work work = { .cpu = cpu, .target = target, };
1235 * If the platform does not support hotplug, report it explicitly to
1236 * differentiate it from a transient offlining failure.
1238 if (cc_platform_has(CC_ATTR_HOTPLUG_DISABLED))
1240 if (cpu_hotplug_disabled)
1244 * Ensure that the control task does not run on the to be offlined
1245 * CPU to prevent a deadlock against cfs_b->period_timer.
1246 * Also keep at least one housekeeping cpu onlined to avoid generating
1247 * an empty sched_domain span.
1249 for_each_cpu_and(cpu, cpu_online_mask, housekeeping_cpumask(HK_TYPE_DOMAIN)) {
1250 if (cpu != work.cpu)
1251 return work_on_cpu(cpu, __cpu_down_maps_locked, &work);
1256 static int cpu_down(unsigned int cpu, enum cpuhp_state target)
1260 cpu_maps_update_begin();
1261 err = cpu_down_maps_locked(cpu, target);
1262 cpu_maps_update_done();
1267 * cpu_device_down - Bring down a cpu device
1268 * @dev: Pointer to the cpu device to offline
1270 * This function is meant to be used by device core cpu subsystem only.
1272 * Other subsystems should use remove_cpu() instead.
1274 * Return: %0 on success or a negative errno code
1276 int cpu_device_down(struct device *dev)
1278 return cpu_down(dev->id, CPUHP_OFFLINE);
1281 int remove_cpu(unsigned int cpu)
1285 lock_device_hotplug();
1286 ret = device_offline(get_cpu_device(cpu));
1287 unlock_device_hotplug();
1291 EXPORT_SYMBOL_GPL(remove_cpu);
1293 void smp_shutdown_nonboot_cpus(unsigned int primary_cpu)
1298 cpu_maps_update_begin();
1301 * Make certain the cpu I'm about to reboot on is online.
1303 * This is inline to what migrate_to_reboot_cpu() already do.
1305 if (!cpu_online(primary_cpu))
1306 primary_cpu = cpumask_first(cpu_online_mask);
1308 for_each_online_cpu(cpu) {
1309 if (cpu == primary_cpu)
1312 error = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
1314 pr_err("Failed to offline CPU%d - error=%d",
1321 * Ensure all but the reboot CPU are offline.
1323 BUG_ON(num_online_cpus() > 1);
1326 * Make sure the CPUs won't be enabled by someone else after this
1327 * point. Kexec will reboot to a new kernel shortly resetting
1328 * everything along the way.
1330 cpu_hotplug_disabled++;
1332 cpu_maps_update_done();
1336 #define takedown_cpu NULL
1337 #endif /*CONFIG_HOTPLUG_CPU*/
1340 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1341 * @cpu: cpu that just started
1343 * It must be called by the arch code on the new cpu, before the new cpu
1344 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1346 void notify_cpu_starting(unsigned int cpu)
1348 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1349 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1351 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1352 cpumask_set_cpu(cpu, &cpus_booted_once_mask);
1355 * STARTING must not fail!
1357 cpuhp_invoke_callback_range_nofail(true, cpu, st, target);
1361 * Called from the idle task. Wake up the controlling task which brings the
1362 * hotplug thread of the upcoming CPU up and then delegates the rest of the
1363 * online bringup to the hotplug thread.
1365 void cpuhp_online_idle(enum cpuhp_state state)
1367 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1369 /* Happens for the boot cpu */
1370 if (state != CPUHP_AP_ONLINE_IDLE)
1374 * Unpart the stopper thread before we start the idle loop (and start
1375 * scheduling); this ensures the stopper task is always available.
1377 stop_machine_unpark(smp_processor_id());
1379 st->state = CPUHP_AP_ONLINE_IDLE;
1380 complete_ap_thread(st, true);
1383 /* Requires cpu_add_remove_lock to be held */
1384 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1386 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1387 struct task_struct *idle;
1392 if (!cpu_present(cpu)) {
1398 * The caller of cpu_up() might have raced with another
1399 * caller. Nothing to do.
1401 if (st->state >= target)
1404 if (st->state == CPUHP_OFFLINE) {
1405 /* Let it fail before we try to bring the cpu up */
1406 idle = idle_thread_get(cpu);
1408 ret = PTR_ERR(idle);
1413 cpuhp_tasks_frozen = tasks_frozen;
1415 cpuhp_set_state(cpu, st, target);
1417 * If the current CPU state is in the range of the AP hotplug thread,
1418 * then we need to kick the thread once more.
1420 if (st->state > CPUHP_BRINGUP_CPU) {
1421 ret = cpuhp_kick_ap_work(cpu);
1423 * The AP side has done the error rollback already. Just
1424 * return the error code..
1431 * Try to reach the target state. We max out on the BP at
1432 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1433 * responsible for bringing it up to the target state.
1435 target = min((int)target, CPUHP_BRINGUP_CPU);
1436 ret = cpuhp_up_callbacks(cpu, st, target);
1438 cpus_write_unlock();
1440 cpu_up_down_serialize_trainwrecks(tasks_frozen);
1444 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
1448 if (!cpu_possible(cpu)) {
1449 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1451 #if defined(CONFIG_IA64)
1452 pr_err("please check additional_cpus= boot parameter\n");
1457 err = try_online_node(cpu_to_node(cpu));
1461 cpu_maps_update_begin();
1463 if (cpu_hotplug_disabled) {
1467 if (!cpu_smt_allowed(cpu)) {
1472 err = _cpu_up(cpu, 0, target);
1474 cpu_maps_update_done();
1479 * cpu_device_up - Bring up a cpu device
1480 * @dev: Pointer to the cpu device to online
1482 * This function is meant to be used by device core cpu subsystem only.
1484 * Other subsystems should use add_cpu() instead.
1486 * Return: %0 on success or a negative errno code
1488 int cpu_device_up(struct device *dev)
1490 return cpu_up(dev->id, CPUHP_ONLINE);
1493 int add_cpu(unsigned int cpu)
1497 lock_device_hotplug();
1498 ret = device_online(get_cpu_device(cpu));
1499 unlock_device_hotplug();
1503 EXPORT_SYMBOL_GPL(add_cpu);
1506 * bringup_hibernate_cpu - Bring up the CPU that we hibernated on
1507 * @sleep_cpu: The cpu we hibernated on and should be brought up.
1509 * On some architectures like arm64, we can hibernate on any CPU, but on
1510 * wake up the CPU we hibernated on might be offline as a side effect of
1511 * using maxcpus= for example.
1513 * Return: %0 on success or a negative errno code
1515 int bringup_hibernate_cpu(unsigned int sleep_cpu)
1519 if (!cpu_online(sleep_cpu)) {
1520 pr_info("Hibernated on a CPU that is offline! Bringing CPU up.\n");
1521 ret = cpu_up(sleep_cpu, CPUHP_ONLINE);
1523 pr_err("Failed to bring hibernate-CPU up!\n");
1530 void bringup_nonboot_cpus(unsigned int setup_max_cpus)
1534 for_each_present_cpu(cpu) {
1535 if (num_online_cpus() >= setup_max_cpus)
1537 if (!cpu_online(cpu))
1538 cpu_up(cpu, CPUHP_ONLINE);
1542 #ifdef CONFIG_PM_SLEEP_SMP
1543 static cpumask_var_t frozen_cpus;
1545 int freeze_secondary_cpus(int primary)
1549 cpu_maps_update_begin();
1550 if (primary == -1) {
1551 primary = cpumask_first(cpu_online_mask);
1552 if (!housekeeping_cpu(primary, HK_TYPE_TIMER))
1553 primary = housekeeping_any_cpu(HK_TYPE_TIMER);
1555 if (!cpu_online(primary))
1556 primary = cpumask_first(cpu_online_mask);
1560 * We take down all of the non-boot CPUs in one shot to avoid races
1561 * with the userspace trying to use the CPU hotplug at the same time
1563 cpumask_clear(frozen_cpus);
1565 pr_info("Disabling non-boot CPUs ...\n");
1566 for_each_online_cpu(cpu) {
1570 if (pm_wakeup_pending()) {
1571 pr_info("Wakeup pending. Abort CPU freeze\n");
1576 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1577 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1578 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1580 cpumask_set_cpu(cpu, frozen_cpus);
1582 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1588 BUG_ON(num_online_cpus() > 1);
1590 pr_err("Non-boot CPUs are not disabled\n");
1593 * Make sure the CPUs won't be enabled by someone else. We need to do
1594 * this even in case of failure as all freeze_secondary_cpus() users are
1595 * supposed to do thaw_secondary_cpus() on the failure path.
1597 cpu_hotplug_disabled++;
1599 cpu_maps_update_done();
1603 void __weak arch_thaw_secondary_cpus_begin(void)
1607 void __weak arch_thaw_secondary_cpus_end(void)
1611 void thaw_secondary_cpus(void)
1615 /* Allow everyone to use the CPU hotplug again */
1616 cpu_maps_update_begin();
1617 __cpu_hotplug_enable();
1618 if (cpumask_empty(frozen_cpus))
1621 pr_info("Enabling non-boot CPUs ...\n");
1623 arch_thaw_secondary_cpus_begin();
1625 for_each_cpu(cpu, frozen_cpus) {
1626 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1627 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1628 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1630 pr_info("CPU%d is up\n", cpu);
1633 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1636 arch_thaw_secondary_cpus_end();
1638 cpumask_clear(frozen_cpus);
1640 cpu_maps_update_done();
1643 static int __init alloc_frozen_cpus(void)
1645 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1649 core_initcall(alloc_frozen_cpus);
1652 * When callbacks for CPU hotplug notifications are being executed, we must
1653 * ensure that the state of the system with respect to the tasks being frozen
1654 * or not, as reported by the notification, remains unchanged *throughout the
1655 * duration* of the execution of the callbacks.
1656 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1658 * This synchronization is implemented by mutually excluding regular CPU
1659 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1660 * Hibernate notifications.
1663 cpu_hotplug_pm_callback(struct notifier_block *nb,
1664 unsigned long action, void *ptr)
1668 case PM_SUSPEND_PREPARE:
1669 case PM_HIBERNATION_PREPARE:
1670 cpu_hotplug_disable();
1673 case PM_POST_SUSPEND:
1674 case PM_POST_HIBERNATION:
1675 cpu_hotplug_enable();
1686 static int __init cpu_hotplug_pm_sync_init(void)
1689 * cpu_hotplug_pm_callback has higher priority than x86
1690 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1691 * to disable cpu hotplug to avoid cpu hotplug race.
1693 pm_notifier(cpu_hotplug_pm_callback, 0);
1696 core_initcall(cpu_hotplug_pm_sync_init);
1698 #endif /* CONFIG_PM_SLEEP_SMP */
1702 #endif /* CONFIG_SMP */
1704 /* Boot processor state steps */
1705 static struct cpuhp_step cpuhp_hp_states[] = {
1708 .startup.single = NULL,
1709 .teardown.single = NULL,
1712 [CPUHP_CREATE_THREADS]= {
1713 .name = "threads:prepare",
1714 .startup.single = smpboot_create_threads,
1715 .teardown.single = NULL,
1718 [CPUHP_PERF_PREPARE] = {
1719 .name = "perf:prepare",
1720 .startup.single = perf_event_init_cpu,
1721 .teardown.single = perf_event_exit_cpu,
1723 [CPUHP_RANDOM_PREPARE] = {
1724 .name = "random:prepare",
1725 .startup.single = random_prepare_cpu,
1726 .teardown.single = NULL,
1728 [CPUHP_WORKQUEUE_PREP] = {
1729 .name = "workqueue:prepare",
1730 .startup.single = workqueue_prepare_cpu,
1731 .teardown.single = NULL,
1733 [CPUHP_HRTIMERS_PREPARE] = {
1734 .name = "hrtimers:prepare",
1735 .startup.single = hrtimers_prepare_cpu,
1736 .teardown.single = NULL,
1738 [CPUHP_SMPCFD_PREPARE] = {
1739 .name = "smpcfd:prepare",
1740 .startup.single = smpcfd_prepare_cpu,
1741 .teardown.single = smpcfd_dead_cpu,
1743 [CPUHP_RELAY_PREPARE] = {
1744 .name = "relay:prepare",
1745 .startup.single = relay_prepare_cpu,
1746 .teardown.single = NULL,
1748 [CPUHP_SLAB_PREPARE] = {
1749 .name = "slab:prepare",
1750 .startup.single = slab_prepare_cpu,
1751 .teardown.single = slab_dead_cpu,
1753 [CPUHP_RCUTREE_PREP] = {
1754 .name = "RCU/tree:prepare",
1755 .startup.single = rcutree_prepare_cpu,
1756 .teardown.single = rcutree_dead_cpu,
1759 * On the tear-down path, timers_dead_cpu() must be invoked
1760 * before blk_mq_queue_reinit_notify() from notify_dead(),
1761 * otherwise a RCU stall occurs.
1763 [CPUHP_TIMERS_PREPARE] = {
1764 .name = "timers:prepare",
1765 .startup.single = timers_prepare_cpu,
1766 .teardown.single = timers_dead_cpu,
1768 /* Kicks the plugged cpu into life */
1769 [CPUHP_BRINGUP_CPU] = {
1770 .name = "cpu:bringup",
1771 .startup.single = bringup_cpu,
1772 .teardown.single = finish_cpu,
1775 /* Final state before CPU kills itself */
1776 [CPUHP_AP_IDLE_DEAD] = {
1777 .name = "idle:dead",
1780 * Last state before CPU enters the idle loop to die. Transient state
1781 * for synchronization.
1783 [CPUHP_AP_OFFLINE] = {
1784 .name = "ap:offline",
1787 /* First state is scheduler control. Interrupts are disabled */
1788 [CPUHP_AP_SCHED_STARTING] = {
1789 .name = "sched:starting",
1790 .startup.single = sched_cpu_starting,
1791 .teardown.single = sched_cpu_dying,
1793 [CPUHP_AP_RCUTREE_DYING] = {
1794 .name = "RCU/tree:dying",
1795 .startup.single = NULL,
1796 .teardown.single = rcutree_dying_cpu,
1798 [CPUHP_AP_SMPCFD_DYING] = {
1799 .name = "smpcfd:dying",
1800 .startup.single = NULL,
1801 .teardown.single = smpcfd_dying_cpu,
1803 [CPUHP_AP_HRTIMERS_DYING] = {
1804 .name = "hrtimers:dying",
1805 .startup.single = NULL,
1806 .teardown.single = hrtimers_cpu_dying,
1809 /* Entry state on starting. Interrupts enabled from here on. Transient
1810 * state for synchronsization */
1811 [CPUHP_AP_ONLINE] = {
1812 .name = "ap:online",
1815 * Handled on control processor until the plugged processor manages
1818 [CPUHP_TEARDOWN_CPU] = {
1819 .name = "cpu:teardown",
1820 .startup.single = NULL,
1821 .teardown.single = takedown_cpu,
1825 [CPUHP_AP_SCHED_WAIT_EMPTY] = {
1826 .name = "sched:waitempty",
1827 .startup.single = NULL,
1828 .teardown.single = sched_cpu_wait_empty,
1831 /* Handle smpboot threads park/unpark */
1832 [CPUHP_AP_SMPBOOT_THREADS] = {
1833 .name = "smpboot/threads:online",
1834 .startup.single = smpboot_unpark_threads,
1835 .teardown.single = smpboot_park_threads,
1837 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1838 .name = "irq/affinity:online",
1839 .startup.single = irq_affinity_online_cpu,
1840 .teardown.single = NULL,
1842 [CPUHP_AP_PERF_ONLINE] = {
1843 .name = "perf:online",
1844 .startup.single = perf_event_init_cpu,
1845 .teardown.single = perf_event_exit_cpu,
1847 [CPUHP_AP_WATCHDOG_ONLINE] = {
1848 .name = "lockup_detector:online",
1849 .startup.single = lockup_detector_online_cpu,
1850 .teardown.single = lockup_detector_offline_cpu,
1852 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1853 .name = "workqueue:online",
1854 .startup.single = workqueue_online_cpu,
1855 .teardown.single = workqueue_offline_cpu,
1857 [CPUHP_AP_RANDOM_ONLINE] = {
1858 .name = "random:online",
1859 .startup.single = random_online_cpu,
1860 .teardown.single = NULL,
1862 [CPUHP_AP_RCUTREE_ONLINE] = {
1863 .name = "RCU/tree:online",
1864 .startup.single = rcutree_online_cpu,
1865 .teardown.single = rcutree_offline_cpu,
1869 * The dynamically registered state space is here
1873 /* Last state is scheduler control setting the cpu active */
1874 [CPUHP_AP_ACTIVE] = {
1875 .name = "sched:active",
1876 .startup.single = sched_cpu_activate,
1877 .teardown.single = sched_cpu_deactivate,
1881 /* CPU is fully up and running. */
1884 .startup.single = NULL,
1885 .teardown.single = NULL,
1889 /* Sanity check for callbacks */
1890 static int cpuhp_cb_check(enum cpuhp_state state)
1892 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1898 * Returns a free for dynamic slot assignment of the Online state. The states
1899 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1900 * by having no name assigned.
1902 static int cpuhp_reserve_state(enum cpuhp_state state)
1904 enum cpuhp_state i, end;
1905 struct cpuhp_step *step;
1908 case CPUHP_AP_ONLINE_DYN:
1909 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1910 end = CPUHP_AP_ONLINE_DYN_END;
1912 case CPUHP_BP_PREPARE_DYN:
1913 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1914 end = CPUHP_BP_PREPARE_DYN_END;
1920 for (i = state; i <= end; i++, step++) {
1924 WARN(1, "No more dynamic states available for CPU hotplug\n");
1928 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1929 int (*startup)(unsigned int cpu),
1930 int (*teardown)(unsigned int cpu),
1931 bool multi_instance)
1933 /* (Un)Install the callbacks for further cpu hotplug operations */
1934 struct cpuhp_step *sp;
1938 * If name is NULL, then the state gets removed.
1940 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1941 * the first allocation from these dynamic ranges, so the removal
1942 * would trigger a new allocation and clear the wrong (already
1943 * empty) state, leaving the callbacks of the to be cleared state
1944 * dangling, which causes wreckage on the next hotplug operation.
1946 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1947 state == CPUHP_BP_PREPARE_DYN)) {
1948 ret = cpuhp_reserve_state(state);
1953 sp = cpuhp_get_step(state);
1954 if (name && sp->name)
1957 sp->startup.single = startup;
1958 sp->teardown.single = teardown;
1960 sp->multi_instance = multi_instance;
1961 INIT_HLIST_HEAD(&sp->list);
1965 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1967 return cpuhp_get_step(state)->teardown.single;
1971 * Call the startup/teardown function for a step either on the AP or
1972 * on the current CPU.
1974 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1975 struct hlist_node *node)
1977 struct cpuhp_step *sp = cpuhp_get_step(state);
1981 * If there's nothing to do, we done.
1982 * Relies on the union for multi_instance.
1984 if (cpuhp_step_empty(bringup, sp))
1987 * The non AP bound callbacks can fail on bringup. On teardown
1988 * e.g. module removal we crash for now.
1991 if (cpuhp_is_ap_state(state))
1992 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1994 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1996 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1998 BUG_ON(ret && !bringup);
2003 * Called from __cpuhp_setup_state on a recoverable failure.
2005 * Note: The teardown callbacks for rollback are not allowed to fail!
2007 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
2008 struct hlist_node *node)
2012 /* Roll back the already executed steps on the other cpus */
2013 for_each_present_cpu(cpu) {
2014 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2015 int cpustate = st->state;
2017 if (cpu >= failedcpu)
2020 /* Did we invoke the startup call on that cpu ? */
2021 if (cpustate >= state)
2022 cpuhp_issue_call(cpu, state, false, node);
2026 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
2027 struct hlist_node *node,
2030 struct cpuhp_step *sp;
2034 lockdep_assert_cpus_held();
2036 sp = cpuhp_get_step(state);
2037 if (sp->multi_instance == false)
2040 mutex_lock(&cpuhp_state_mutex);
2042 if (!invoke || !sp->startup.multi)
2046 * Try to call the startup callback for each present cpu
2047 * depending on the hotplug state of the cpu.
2049 for_each_present_cpu(cpu) {
2050 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2051 int cpustate = st->state;
2053 if (cpustate < state)
2056 ret = cpuhp_issue_call(cpu, state, true, node);
2058 if (sp->teardown.multi)
2059 cpuhp_rollback_install(cpu, state, node);
2065 hlist_add_head(node, &sp->list);
2067 mutex_unlock(&cpuhp_state_mutex);
2071 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
2077 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
2081 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
2084 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
2085 * @state: The state to setup
2086 * @name: Name of the step
2087 * @invoke: If true, the startup function is invoked for cpus where
2088 * cpu state >= @state
2089 * @startup: startup callback function
2090 * @teardown: teardown callback function
2091 * @multi_instance: State is set up for multiple instances which get
2094 * The caller needs to hold cpus read locked while calling this function.
2097 * Positive state number if @state is CPUHP_AP_ONLINE_DYN;
2098 * 0 for all other states
2099 * On failure: proper (negative) error code
2101 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
2102 const char *name, bool invoke,
2103 int (*startup)(unsigned int cpu),
2104 int (*teardown)(unsigned int cpu),
2105 bool multi_instance)
2110 lockdep_assert_cpus_held();
2112 if (cpuhp_cb_check(state) || !name)
2115 mutex_lock(&cpuhp_state_mutex);
2117 ret = cpuhp_store_callbacks(state, name, startup, teardown,
2120 dynstate = state == CPUHP_AP_ONLINE_DYN;
2121 if (ret > 0 && dynstate) {
2126 if (ret || !invoke || !startup)
2130 * Try to call the startup callback for each present cpu
2131 * depending on the hotplug state of the cpu.
2133 for_each_present_cpu(cpu) {
2134 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2135 int cpustate = st->state;
2137 if (cpustate < state)
2140 ret = cpuhp_issue_call(cpu, state, true, NULL);
2143 cpuhp_rollback_install(cpu, state, NULL);
2144 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2149 mutex_unlock(&cpuhp_state_mutex);
2151 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
2152 * dynamically allocated state in case of success.
2154 if (!ret && dynstate)
2158 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
2160 int __cpuhp_setup_state(enum cpuhp_state state,
2161 const char *name, bool invoke,
2162 int (*startup)(unsigned int cpu),
2163 int (*teardown)(unsigned int cpu),
2164 bool multi_instance)
2169 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
2170 teardown, multi_instance);
2174 EXPORT_SYMBOL(__cpuhp_setup_state);
2176 int __cpuhp_state_remove_instance(enum cpuhp_state state,
2177 struct hlist_node *node, bool invoke)
2179 struct cpuhp_step *sp = cpuhp_get_step(state);
2182 BUG_ON(cpuhp_cb_check(state));
2184 if (!sp->multi_instance)
2188 mutex_lock(&cpuhp_state_mutex);
2190 if (!invoke || !cpuhp_get_teardown_cb(state))
2193 * Call the teardown callback for each present cpu depending
2194 * on the hotplug state of the cpu. This function is not
2195 * allowed to fail currently!
2197 for_each_present_cpu(cpu) {
2198 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2199 int cpustate = st->state;
2201 if (cpustate >= state)
2202 cpuhp_issue_call(cpu, state, false, node);
2207 mutex_unlock(&cpuhp_state_mutex);
2212 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
2215 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
2216 * @state: The state to remove
2217 * @invoke: If true, the teardown function is invoked for cpus where
2218 * cpu state >= @state
2220 * The caller needs to hold cpus read locked while calling this function.
2221 * The teardown callback is currently not allowed to fail. Think
2222 * about module removal!
2224 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
2226 struct cpuhp_step *sp = cpuhp_get_step(state);
2229 BUG_ON(cpuhp_cb_check(state));
2231 lockdep_assert_cpus_held();
2233 mutex_lock(&cpuhp_state_mutex);
2234 if (sp->multi_instance) {
2235 WARN(!hlist_empty(&sp->list),
2236 "Error: Removing state %d which has instances left.\n",
2241 if (!invoke || !cpuhp_get_teardown_cb(state))
2245 * Call the teardown callback for each present cpu depending
2246 * on the hotplug state of the cpu. This function is not
2247 * allowed to fail currently!
2249 for_each_present_cpu(cpu) {
2250 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
2251 int cpustate = st->state;
2253 if (cpustate >= state)
2254 cpuhp_issue_call(cpu, state, false, NULL);
2257 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
2258 mutex_unlock(&cpuhp_state_mutex);
2260 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
2262 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
2265 __cpuhp_remove_state_cpuslocked(state, invoke);
2268 EXPORT_SYMBOL(__cpuhp_remove_state);
2270 #ifdef CONFIG_HOTPLUG_SMT
2271 static void cpuhp_offline_cpu_device(unsigned int cpu)
2273 struct device *dev = get_cpu_device(cpu);
2275 dev->offline = true;
2276 /* Tell user space about the state change */
2277 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2280 static void cpuhp_online_cpu_device(unsigned int cpu)
2282 struct device *dev = get_cpu_device(cpu);
2284 dev->offline = false;
2285 /* Tell user space about the state change */
2286 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2289 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2293 cpu_maps_update_begin();
2294 for_each_online_cpu(cpu) {
2295 if (topology_is_primary_thread(cpu))
2297 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2301 * As this needs to hold the cpu maps lock it's impossible
2302 * to call device_offline() because that ends up calling
2303 * cpu_down() which takes cpu maps lock. cpu maps lock
2304 * needs to be held as this might race against in kernel
2305 * abusers of the hotplug machinery (thermal management).
2307 * So nothing would update device:offline state. That would
2308 * leave the sysfs entry stale and prevent onlining after
2309 * smt control has been changed to 'off' again. This is
2310 * called under the sysfs hotplug lock, so it is properly
2311 * serialized against the regular offline usage.
2313 cpuhp_offline_cpu_device(cpu);
2316 cpu_smt_control = ctrlval;
2317 cpu_maps_update_done();
2321 int cpuhp_smt_enable(void)
2325 cpu_maps_update_begin();
2326 cpu_smt_control = CPU_SMT_ENABLED;
2327 for_each_present_cpu(cpu) {
2328 /* Skip online CPUs and CPUs on offline nodes */
2329 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2331 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2334 /* See comment in cpuhp_smt_disable() */
2335 cpuhp_online_cpu_device(cpu);
2337 cpu_maps_update_done();
2342 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
2343 static ssize_t state_show(struct device *dev,
2344 struct device_attribute *attr, char *buf)
2346 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2348 return sprintf(buf, "%d\n", st->state);
2350 static DEVICE_ATTR_RO(state);
2352 static ssize_t target_store(struct device *dev, struct device_attribute *attr,
2353 const char *buf, size_t count)
2355 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2356 struct cpuhp_step *sp;
2359 ret = kstrtoint(buf, 10, &target);
2363 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
2364 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
2367 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
2371 ret = lock_device_hotplug_sysfs();
2375 mutex_lock(&cpuhp_state_mutex);
2376 sp = cpuhp_get_step(target);
2377 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
2378 mutex_unlock(&cpuhp_state_mutex);
2382 if (st->state < target)
2383 ret = cpu_up(dev->id, target);
2384 else if (st->state > target)
2385 ret = cpu_down(dev->id, target);
2386 else if (WARN_ON(st->target != target))
2387 st->target = target;
2389 unlock_device_hotplug();
2390 return ret ? ret : count;
2393 static ssize_t target_show(struct device *dev,
2394 struct device_attribute *attr, char *buf)
2396 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2398 return sprintf(buf, "%d\n", st->target);
2400 static DEVICE_ATTR_RW(target);
2402 static ssize_t fail_store(struct device *dev, struct device_attribute *attr,
2403 const char *buf, size_t count)
2405 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2406 struct cpuhp_step *sp;
2409 ret = kstrtoint(buf, 10, &fail);
2413 if (fail == CPUHP_INVALID) {
2418 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
2422 * Cannot fail STARTING/DYING callbacks.
2424 if (cpuhp_is_atomic_state(fail))
2428 * DEAD callbacks cannot fail...
2429 * ... neither can CPUHP_BRINGUP_CPU during hotunplug. The latter
2430 * triggering STARTING callbacks, a failure in this state would
2433 if (fail <= CPUHP_BRINGUP_CPU && st->state > CPUHP_BRINGUP_CPU)
2437 * Cannot fail anything that doesn't have callbacks.
2439 mutex_lock(&cpuhp_state_mutex);
2440 sp = cpuhp_get_step(fail);
2441 if (!sp->startup.single && !sp->teardown.single)
2443 mutex_unlock(&cpuhp_state_mutex);
2452 static ssize_t fail_show(struct device *dev,
2453 struct device_attribute *attr, char *buf)
2455 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
2457 return sprintf(buf, "%d\n", st->fail);
2460 static DEVICE_ATTR_RW(fail);
2462 static struct attribute *cpuhp_cpu_attrs[] = {
2463 &dev_attr_state.attr,
2464 &dev_attr_target.attr,
2465 &dev_attr_fail.attr,
2469 static const struct attribute_group cpuhp_cpu_attr_group = {
2470 .attrs = cpuhp_cpu_attrs,
2475 static ssize_t states_show(struct device *dev,
2476 struct device_attribute *attr, char *buf)
2478 ssize_t cur, res = 0;
2481 mutex_lock(&cpuhp_state_mutex);
2482 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
2483 struct cpuhp_step *sp = cpuhp_get_step(i);
2486 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
2491 mutex_unlock(&cpuhp_state_mutex);
2494 static DEVICE_ATTR_RO(states);
2496 static struct attribute *cpuhp_cpu_root_attrs[] = {
2497 &dev_attr_states.attr,
2501 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2502 .attrs = cpuhp_cpu_root_attrs,
2507 #ifdef CONFIG_HOTPLUG_SMT
2510 __store_smt_control(struct device *dev, struct device_attribute *attr,
2511 const char *buf, size_t count)
2515 if (sysfs_streq(buf, "on"))
2516 ctrlval = CPU_SMT_ENABLED;
2517 else if (sysfs_streq(buf, "off"))
2518 ctrlval = CPU_SMT_DISABLED;
2519 else if (sysfs_streq(buf, "forceoff"))
2520 ctrlval = CPU_SMT_FORCE_DISABLED;
2524 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2527 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2530 ret = lock_device_hotplug_sysfs();
2534 if (ctrlval != cpu_smt_control) {
2536 case CPU_SMT_ENABLED:
2537 ret = cpuhp_smt_enable();
2539 case CPU_SMT_DISABLED:
2540 case CPU_SMT_FORCE_DISABLED:
2541 ret = cpuhp_smt_disable(ctrlval);
2546 unlock_device_hotplug();
2547 return ret ? ret : count;
2550 #else /* !CONFIG_HOTPLUG_SMT */
2552 __store_smt_control(struct device *dev, struct device_attribute *attr,
2553 const char *buf, size_t count)
2557 #endif /* CONFIG_HOTPLUG_SMT */
2559 static const char *smt_states[] = {
2560 [CPU_SMT_ENABLED] = "on",
2561 [CPU_SMT_DISABLED] = "off",
2562 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2563 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2564 [CPU_SMT_NOT_IMPLEMENTED] = "notimplemented",
2567 static ssize_t control_show(struct device *dev,
2568 struct device_attribute *attr, char *buf)
2570 const char *state = smt_states[cpu_smt_control];
2572 return snprintf(buf, PAGE_SIZE - 2, "%s\n", state);
2575 static ssize_t control_store(struct device *dev, struct device_attribute *attr,
2576 const char *buf, size_t count)
2578 return __store_smt_control(dev, attr, buf, count);
2580 static DEVICE_ATTR_RW(control);
2582 static ssize_t active_show(struct device *dev,
2583 struct device_attribute *attr, char *buf)
2585 return snprintf(buf, PAGE_SIZE - 2, "%d\n", sched_smt_active());
2587 static DEVICE_ATTR_RO(active);
2589 static struct attribute *cpuhp_smt_attrs[] = {
2590 &dev_attr_control.attr,
2591 &dev_attr_active.attr,
2595 static const struct attribute_group cpuhp_smt_attr_group = {
2596 .attrs = cpuhp_smt_attrs,
2601 static int __init cpu_smt_sysfs_init(void)
2603 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2604 &cpuhp_smt_attr_group);
2607 static int __init cpuhp_sysfs_init(void)
2611 ret = cpu_smt_sysfs_init();
2615 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2616 &cpuhp_cpu_root_attr_group);
2620 for_each_possible_cpu(cpu) {
2621 struct device *dev = get_cpu_device(cpu);
2625 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2631 device_initcall(cpuhp_sysfs_init);
2632 #endif /* CONFIG_SYSFS && CONFIG_HOTPLUG_CPU */
2635 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2636 * represents all NR_CPUS bits binary values of 1<<nr.
2638 * It is used by cpumask_of() to get a constant address to a CPU
2639 * mask value that has a single bit set only.
2642 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2643 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2644 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2645 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2646 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2648 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2650 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2651 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2652 #if BITS_PER_LONG > 32
2653 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2654 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2657 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2659 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2660 EXPORT_SYMBOL(cpu_all_bits);
2662 #ifdef CONFIG_INIT_ALL_POSSIBLE
2663 struct cpumask __cpu_possible_mask __read_mostly
2666 struct cpumask __cpu_possible_mask __read_mostly;
2668 EXPORT_SYMBOL(__cpu_possible_mask);
2670 struct cpumask __cpu_online_mask __read_mostly;
2671 EXPORT_SYMBOL(__cpu_online_mask);
2673 struct cpumask __cpu_present_mask __read_mostly;
2674 EXPORT_SYMBOL(__cpu_present_mask);
2676 struct cpumask __cpu_active_mask __read_mostly;
2677 EXPORT_SYMBOL(__cpu_active_mask);
2679 struct cpumask __cpu_dying_mask __read_mostly;
2680 EXPORT_SYMBOL(__cpu_dying_mask);
2682 atomic_t __num_online_cpus __read_mostly;
2683 EXPORT_SYMBOL(__num_online_cpus);
2685 void init_cpu_present(const struct cpumask *src)
2687 cpumask_copy(&__cpu_present_mask, src);
2690 void init_cpu_possible(const struct cpumask *src)
2692 cpumask_copy(&__cpu_possible_mask, src);
2695 void init_cpu_online(const struct cpumask *src)
2697 cpumask_copy(&__cpu_online_mask, src);
2700 void set_cpu_online(unsigned int cpu, bool online)
2703 * atomic_inc/dec() is required to handle the horrid abuse of this
2704 * function by the reboot and kexec code which invoke it from
2705 * IPI/NMI broadcasts when shutting down CPUs. Invocation from
2706 * regular CPU hotplug is properly serialized.
2708 * Note, that the fact that __num_online_cpus is of type atomic_t
2709 * does not protect readers which are not serialized against
2710 * concurrent hotplug operations.
2713 if (!cpumask_test_and_set_cpu(cpu, &__cpu_online_mask))
2714 atomic_inc(&__num_online_cpus);
2716 if (cpumask_test_and_clear_cpu(cpu, &__cpu_online_mask))
2717 atomic_dec(&__num_online_cpus);
2722 * Activate the first processor.
2724 void __init boot_cpu_init(void)
2726 int cpu = smp_processor_id();
2728 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2729 set_cpu_online(cpu, true);
2730 set_cpu_active(cpu, true);
2731 set_cpu_present(cpu, true);
2732 set_cpu_possible(cpu, true);
2735 __boot_cpu_id = cpu;
2740 * Must be called _AFTER_ setting up the per_cpu areas
2742 void __init boot_cpu_hotplug_init(void)
2745 cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
2747 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2751 * These are used for a global "mitigations=" cmdline option for toggling
2752 * optional CPU mitigations.
2754 enum cpu_mitigations {
2755 CPU_MITIGATIONS_OFF,
2756 CPU_MITIGATIONS_AUTO,
2757 CPU_MITIGATIONS_AUTO_NOSMT,
2760 static enum cpu_mitigations cpu_mitigations __ro_after_init =
2761 CPU_MITIGATIONS_AUTO;
2763 static int __init mitigations_parse_cmdline(char *arg)
2765 if (!strcmp(arg, "off"))
2766 cpu_mitigations = CPU_MITIGATIONS_OFF;
2767 else if (!strcmp(arg, "auto"))
2768 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2769 else if (!strcmp(arg, "auto,nosmt"))
2770 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2772 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2777 early_param("mitigations", mitigations_parse_cmdline);
2779 /* mitigations=off */
2780 bool cpu_mitigations_off(void)
2782 return cpu_mitigations == CPU_MITIGATIONS_OFF;
2784 EXPORT_SYMBOL_GPL(cpu_mitigations_off);
2786 /* mitigations=auto,nosmt */
2787 bool cpu_mitigations_auto_nosmt(void)
2789 return cpu_mitigations == CPU_MITIGATIONS_AUTO_NOSMT;
2791 EXPORT_SYMBOL_GPL(cpu_mitigations_auto_nosmt);