2 * (C) 2001, 2002, 2003, 2004 Rusty Russell
4 * This code is licenced under the GPL.
6 #include <linux/proc_fs.h>
8 #include <linux/init.h>
9 #include <linux/notifier.h>
10 #include <linux/sched/signal.h>
11 #include <linux/sched/hotplug.h>
12 #include <linux/sched/task.h>
13 #include <linux/sched/smt.h>
14 #include <linux/unistd.h>
15 #include <linux/cpu.h>
16 #include <linux/oom.h>
17 #include <linux/rcupdate.h>
18 #include <linux/export.h>
19 #include <linux/bug.h>
20 #include <linux/kthread.h>
21 #include <linux/stop_machine.h>
22 #include <linux/mutex.h>
23 #include <linux/gfp.h>
24 #include <linux/suspend.h>
25 #include <linux/lockdep.h>
26 #include <linux/tick.h>
27 #include <linux/irq.h>
28 #include <linux/nmi.h>
29 #include <linux/smpboot.h>
30 #include <linux/relay.h>
31 #include <linux/slab.h>
32 #include <linux/percpu-rwsem.h>
34 #include <trace/events/power.h>
35 #define CREATE_TRACE_POINTS
36 #include <trace/events/cpuhp.h>
41 * cpuhp_cpu_state - Per cpu hotplug state storage
42 * @state: The current cpu state
43 * @target: The target state
44 * @thread: Pointer to the hotplug thread
45 * @should_run: Thread should execute
46 * @rollback: Perform a rollback
47 * @single: Single callback invocation
48 * @bringup: Single callback bringup or teardown selector
49 * @cb_state: The state for a single callback (install/uninstall)
50 * @result: Result of the operation
51 * @done_up: Signal completion to the issuer of the task for cpu-up
52 * @done_down: Signal completion to the issuer of the task for cpu-down
54 struct cpuhp_cpu_state {
55 enum cpuhp_state state;
56 enum cpuhp_state target;
57 enum cpuhp_state fail;
59 struct task_struct *thread;
65 struct hlist_node *node;
66 struct hlist_node *last;
67 enum cpuhp_state cb_state;
69 struct completion done_up;
70 struct completion done_down;
74 static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state) = {
75 .fail = CPUHP_INVALID,
78 #if defined(CONFIG_LOCKDEP) && defined(CONFIG_SMP)
79 static struct lockdep_map cpuhp_state_up_map =
80 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-up", &cpuhp_state_up_map);
81 static struct lockdep_map cpuhp_state_down_map =
82 STATIC_LOCKDEP_MAP_INIT("cpuhp_state-down", &cpuhp_state_down_map);
85 static inline void cpuhp_lock_acquire(bool bringup)
87 lock_map_acquire(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
90 static inline void cpuhp_lock_release(bool bringup)
92 lock_map_release(bringup ? &cpuhp_state_up_map : &cpuhp_state_down_map);
96 static inline void cpuhp_lock_acquire(bool bringup) { }
97 static inline void cpuhp_lock_release(bool bringup) { }
102 * cpuhp_step - Hotplug state machine step
103 * @name: Name of the step
104 * @startup: Startup function of the step
105 * @teardown: Teardown function of the step
106 * @cant_stop: Bringup/teardown can't be stopped at this step
111 int (*single)(unsigned int cpu);
112 int (*multi)(unsigned int cpu,
113 struct hlist_node *node);
116 int (*single)(unsigned int cpu);
117 int (*multi)(unsigned int cpu,
118 struct hlist_node *node);
120 struct hlist_head list;
125 static DEFINE_MUTEX(cpuhp_state_mutex);
126 static struct cpuhp_step cpuhp_hp_states[];
128 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
130 return cpuhp_hp_states + state;
134 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
135 * @cpu: The cpu for which the callback should be invoked
136 * @state: The state to do callbacks for
137 * @bringup: True if the bringup callback should be invoked
138 * @node: For multi-instance, do a single entry callback for install/remove
139 * @lastp: For multi-instance rollback, remember how far we got
141 * Called from cpu hotplug and from the state register machinery.
143 static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state state,
144 bool bringup, struct hlist_node *node,
145 struct hlist_node **lastp)
147 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
148 struct cpuhp_step *step = cpuhp_get_step(state);
149 int (*cbm)(unsigned int cpu, struct hlist_node *node);
150 int (*cb)(unsigned int cpu);
153 if (st->fail == state) {
154 st->fail = CPUHP_INVALID;
156 if (!(bringup ? step->startup.single : step->teardown.single))
162 if (!step->multi_instance) {
163 WARN_ON_ONCE(lastp && *lastp);
164 cb = bringup ? step->startup.single : step->teardown.single;
167 trace_cpuhp_enter(cpu, st->target, state, cb);
169 trace_cpuhp_exit(cpu, st->state, state, ret);
172 cbm = bringup ? step->startup.multi : step->teardown.multi;
176 /* Single invocation for instance add/remove */
178 WARN_ON_ONCE(lastp && *lastp);
179 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
180 ret = cbm(cpu, node);
181 trace_cpuhp_exit(cpu, st->state, state, ret);
185 /* State transition. Invoke on all instances */
187 hlist_for_each(node, &step->list) {
188 if (lastp && node == *lastp)
191 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
192 ret = cbm(cpu, node);
193 trace_cpuhp_exit(cpu, st->state, state, ret);
207 /* Rollback the instances if one failed */
208 cbm = !bringup ? step->startup.multi : step->teardown.multi;
212 hlist_for_each(node, &step->list) {
216 trace_cpuhp_multi_enter(cpu, st->target, state, cbm, node);
217 ret = cbm(cpu, node);
218 trace_cpuhp_exit(cpu, st->state, state, ret);
220 * Rollback must not fail,
228 static bool cpuhp_is_ap_state(enum cpuhp_state state)
231 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
232 * purposes as that state is handled explicitly in cpu_down.
234 return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
237 static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
239 struct completion *done = bringup ? &st->done_up : &st->done_down;
240 wait_for_completion(done);
243 static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
245 struct completion *done = bringup ? &st->done_up : &st->done_down;
250 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
252 static bool cpuhp_is_atomic_state(enum cpuhp_state state)
254 return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
257 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
258 static DEFINE_MUTEX(cpu_add_remove_lock);
259 bool cpuhp_tasks_frozen;
260 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen);
263 * The following two APIs (cpu_maps_update_begin/done) must be used when
264 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
266 void cpu_maps_update_begin(void)
268 mutex_lock(&cpu_add_remove_lock);
271 void cpu_maps_update_done(void)
273 mutex_unlock(&cpu_add_remove_lock);
277 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
278 * Should always be manipulated under cpu_add_remove_lock
280 static int cpu_hotplug_disabled;
282 #ifdef CONFIG_HOTPLUG_CPU
284 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
286 void cpus_read_lock(void)
288 percpu_down_read(&cpu_hotplug_lock);
290 EXPORT_SYMBOL_GPL(cpus_read_lock);
292 int cpus_read_trylock(void)
294 return percpu_down_read_trylock(&cpu_hotplug_lock);
296 EXPORT_SYMBOL_GPL(cpus_read_trylock);
298 void cpus_read_unlock(void)
300 percpu_up_read(&cpu_hotplug_lock);
302 EXPORT_SYMBOL_GPL(cpus_read_unlock);
304 void cpus_write_lock(void)
306 percpu_down_write(&cpu_hotplug_lock);
309 void cpus_write_unlock(void)
311 percpu_up_write(&cpu_hotplug_lock);
314 void lockdep_assert_cpus_held(void)
317 * We can't have hotplug operations before userspace starts running,
318 * and some init codepaths will knowingly not take the hotplug lock.
319 * This is all valid, so mute lockdep until it makes sense to report
322 if (system_state < SYSTEM_RUNNING)
325 percpu_rwsem_assert_held(&cpu_hotplug_lock);
329 * Wait for currently running CPU hotplug operations to complete (if any) and
330 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
331 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
332 * hotplug path before performing hotplug operations. So acquiring that lock
333 * guarantees mutual exclusion from any currently running hotplug operations.
335 void cpu_hotplug_disable(void)
337 cpu_maps_update_begin();
338 cpu_hotplug_disabled++;
339 cpu_maps_update_done();
341 EXPORT_SYMBOL_GPL(cpu_hotplug_disable);
343 static void __cpu_hotplug_enable(void)
345 if (WARN_ONCE(!cpu_hotplug_disabled, "Unbalanced cpu hotplug enable\n"))
347 cpu_hotplug_disabled--;
350 void cpu_hotplug_enable(void)
352 cpu_maps_update_begin();
353 __cpu_hotplug_enable();
354 cpu_maps_update_done();
356 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
357 #endif /* CONFIG_HOTPLUG_CPU */
360 * Architectures that need SMT-specific errata handling during SMT hotplug
361 * should override this.
363 void __weak arch_smt_update(void) { }
365 #ifdef CONFIG_HOTPLUG_SMT
366 enum cpuhp_smt_control cpu_smt_control __read_mostly = CPU_SMT_ENABLED;
368 void __init cpu_smt_disable(bool force)
370 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
371 cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
375 pr_info("SMT: Force disabled\n");
376 cpu_smt_control = CPU_SMT_FORCE_DISABLED;
378 cpu_smt_control = CPU_SMT_DISABLED;
383 * The decision whether SMT is supported can only be done after the full
384 * CPU identification. Called from architecture code.
386 void __init cpu_smt_check_topology(void)
388 if (!topology_smt_supported())
389 cpu_smt_control = CPU_SMT_NOT_SUPPORTED;
392 static int __init smt_cmdline_disable(char *str)
394 cpu_smt_disable(str && !strcmp(str, "force"));
397 early_param("nosmt", smt_cmdline_disable);
399 static inline bool cpu_smt_allowed(unsigned int cpu)
401 if (cpu_smt_control == CPU_SMT_ENABLED)
404 if (topology_is_primary_thread(cpu))
408 * On x86 it's required to boot all logical CPUs at least once so
409 * that the init code can get a chance to set CR4.MCE on each
410 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
411 * core will shutdown the machine.
413 return !per_cpu(cpuhp_state, cpu).booted_once;
416 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
419 static inline enum cpuhp_state
420 cpuhp_set_state(struct cpuhp_cpu_state *st, enum cpuhp_state target)
422 enum cpuhp_state prev_state = st->state;
424 st->rollback = false;
429 st->bringup = st->state < target;
435 cpuhp_reset_state(struct cpuhp_cpu_state *st, enum cpuhp_state prev_state)
440 * If we have st->last we need to undo partial multi_instance of this
441 * state first. Otherwise start undo at the previous state.
450 st->target = prev_state;
451 st->bringup = !st->bringup;
454 /* Regular hotplug invocation of the AP hotplug thread */
455 static void __cpuhp_kick_ap(struct cpuhp_cpu_state *st)
457 if (!st->single && st->state == st->target)
462 * Make sure the above stores are visible before should_run becomes
463 * true. Paired with the mb() above in cpuhp_thread_fun()
466 st->should_run = true;
467 wake_up_process(st->thread);
468 wait_for_ap_thread(st, st->bringup);
471 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
473 enum cpuhp_state prev_state;
476 prev_state = cpuhp_set_state(st, target);
478 if ((ret = st->result)) {
479 cpuhp_reset_state(st, prev_state);
486 static int bringup_wait_for_ap(unsigned int cpu)
488 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
490 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
491 wait_for_ap_thread(st, true);
492 if (WARN_ON_ONCE((!cpu_online(cpu))))
495 /* Unpark the stopper thread and the hotplug thread of the target cpu */
496 stop_machine_unpark(cpu);
497 kthread_unpark(st->thread);
500 * SMT soft disabling on X86 requires to bring the CPU out of the
501 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
502 * CPU marked itself as booted_once in cpu_notify_starting() so the
503 * cpu_smt_allowed() check will now return false if this is not the
506 if (!cpu_smt_allowed(cpu))
509 if (st->target <= CPUHP_AP_ONLINE_IDLE)
512 return cpuhp_kick_ap(st, st->target);
515 static int bringup_cpu(unsigned int cpu)
517 struct task_struct *idle = idle_thread_get(cpu);
521 * Some architectures have to walk the irq descriptors to
522 * setup the vector space for the cpu which comes online.
523 * Prevent irq alloc/free across the bringup.
527 /* Arch-specific enabling code. */
528 ret = __cpu_up(cpu, idle);
532 return bringup_wait_for_ap(cpu);
536 * Hotplug state machine related functions
539 static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
541 for (st->state--; st->state > st->target; st->state--)
542 cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
545 static inline bool can_rollback_cpu(struct cpuhp_cpu_state *st)
547 if (IS_ENABLED(CONFIG_HOTPLUG_CPU))
550 * When CPU hotplug is disabled, then taking the CPU down is not
551 * possible because takedown_cpu() and the architecture and
552 * subsystem specific mechanisms are not available. So the CPU
553 * which would be completely unplugged again needs to stay around
554 * in the current state.
556 return st->state <= CPUHP_BRINGUP_CPU;
559 static int cpuhp_up_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
560 enum cpuhp_state target)
562 enum cpuhp_state prev_state = st->state;
565 while (st->state < target) {
567 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
569 if (can_rollback_cpu(st)) {
570 st->target = prev_state;
571 undo_cpu_up(cpu, st);
580 * The cpu hotplug threads manage the bringup and teardown of the cpus
582 static void cpuhp_create(unsigned int cpu)
584 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
586 init_completion(&st->done_up);
587 init_completion(&st->done_down);
590 static int cpuhp_should_run(unsigned int cpu)
592 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
594 return st->should_run;
598 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
599 * callbacks when a state gets [un]installed at runtime.
601 * Each invocation of this function by the smpboot thread does a single AP
604 * It has 3 modes of operation:
605 * - single: runs st->cb_state
606 * - up: runs ++st->state, while st->state < st->target
607 * - down: runs st->state--, while st->state > st->target
609 * When complete or on error, should_run is cleared and the completion is fired.
611 static void cpuhp_thread_fun(unsigned int cpu)
613 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
614 bool bringup = st->bringup;
615 enum cpuhp_state state;
617 if (WARN_ON_ONCE(!st->should_run))
621 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
622 * that if we see ->should_run we also see the rest of the state.
626 cpuhp_lock_acquire(bringup);
629 state = st->cb_state;
630 st->should_run = false;
635 st->should_run = (st->state < st->target);
636 WARN_ON_ONCE(st->state > st->target);
640 st->should_run = (st->state > st->target);
641 WARN_ON_ONCE(st->state < st->target);
645 WARN_ON_ONCE(!cpuhp_is_ap_state(state));
647 if (cpuhp_is_atomic_state(state)) {
649 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
653 * STARTING/DYING must not fail!
655 WARN_ON_ONCE(st->result);
657 st->result = cpuhp_invoke_callback(cpu, state, bringup, st->node, &st->last);
662 * If we fail on a rollback, we're up a creek without no
663 * paddle, no way forward, no way back. We loose, thanks for
666 WARN_ON_ONCE(st->rollback);
667 st->should_run = false;
670 cpuhp_lock_release(bringup);
673 complete_ap_thread(st, bringup);
676 /* Invoke a single callback on a remote cpu */
678 cpuhp_invoke_ap_callback(int cpu, enum cpuhp_state state, bool bringup,
679 struct hlist_node *node)
681 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
684 if (!cpu_online(cpu))
687 cpuhp_lock_acquire(false);
688 cpuhp_lock_release(false);
690 cpuhp_lock_acquire(true);
691 cpuhp_lock_release(true);
694 * If we are up and running, use the hotplug thread. For early calls
695 * we invoke the thread function directly.
698 return cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
700 st->rollback = false;
704 st->bringup = bringup;
705 st->cb_state = state;
711 * If we failed and did a partial, do a rollback.
713 if ((ret = st->result) && st->last) {
715 st->bringup = !bringup;
721 * Clean up the leftovers so the next hotplug operation wont use stale
724 st->node = st->last = NULL;
728 static int cpuhp_kick_ap_work(unsigned int cpu)
730 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
731 enum cpuhp_state prev_state = st->state;
734 cpuhp_lock_acquire(false);
735 cpuhp_lock_release(false);
737 cpuhp_lock_acquire(true);
738 cpuhp_lock_release(true);
740 trace_cpuhp_enter(cpu, st->target, prev_state, cpuhp_kick_ap_work);
741 ret = cpuhp_kick_ap(st, st->target);
742 trace_cpuhp_exit(cpu, st->state, prev_state, ret);
747 static struct smp_hotplug_thread cpuhp_threads = {
748 .store = &cpuhp_state.thread,
749 .create = &cpuhp_create,
750 .thread_should_run = cpuhp_should_run,
751 .thread_fn = cpuhp_thread_fun,
752 .thread_comm = "cpuhp/%u",
756 void __init cpuhp_threads_init(void)
758 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads));
759 kthread_unpark(this_cpu_read(cpuhp_state.thread));
762 #ifdef CONFIG_HOTPLUG_CPU
764 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
767 * This function walks all processes, finds a valid mm struct for each one and
768 * then clears a corresponding bit in mm's cpumask. While this all sounds
769 * trivial, there are various non-obvious corner cases, which this function
770 * tries to solve in a safe manner.
772 * Also note that the function uses a somewhat relaxed locking scheme, so it may
773 * be called only for an already offlined CPU.
775 void clear_tasks_mm_cpumask(int cpu)
777 struct task_struct *p;
780 * This function is called after the cpu is taken down and marked
781 * offline, so its not like new tasks will ever get this cpu set in
782 * their mm mask. -- Peter Zijlstra
783 * Thus, we may use rcu_read_lock() here, instead of grabbing
784 * full-fledged tasklist_lock.
786 WARN_ON(cpu_online(cpu));
788 for_each_process(p) {
789 struct task_struct *t;
792 * Main thread might exit, but other threads may still have
793 * a valid mm. Find one.
795 t = find_lock_task_mm(p);
798 cpumask_clear_cpu(cpu, mm_cpumask(t->mm));
804 /* Take this CPU down. */
805 static int take_cpu_down(void *_param)
807 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
808 enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
809 int err, cpu = smp_processor_id();
812 /* Ensure this CPU doesn't handle any more interrupts. */
813 err = __cpu_disable();
818 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
819 * do this step again.
821 WARN_ON(st->state != CPUHP_TEARDOWN_CPU);
823 /* Invoke the former CPU_DYING callbacks */
824 for (; st->state > target; st->state--) {
825 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
827 * DYING must not fail!
832 /* Give up timekeeping duties */
833 tick_handover_do_timer();
834 /* Park the stopper thread */
835 stop_machine_park(cpu);
839 static int takedown_cpu(unsigned int cpu)
841 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
844 /* Park the smpboot threads */
845 kthread_park(per_cpu_ptr(&cpuhp_state, cpu)->thread);
848 * Prevent irq alloc/free while the dying cpu reorganizes the
849 * interrupt affinities.
854 * So now all preempt/rcu users must observe !cpu_active().
856 err = stop_machine_cpuslocked(take_cpu_down, NULL, cpumask_of(cpu));
858 /* CPU refused to die */
860 /* Unpark the hotplug thread so we can rollback there */
861 kthread_unpark(per_cpu_ptr(&cpuhp_state, cpu)->thread);
864 BUG_ON(cpu_online(cpu));
867 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
868 * all runnable tasks from the CPU, there's only the idle task left now
869 * that the migration thread is done doing the stop_machine thing.
871 * Wait for the stop thread to go away.
873 wait_for_ap_thread(st, false);
874 BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
876 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
879 hotplug_cpu__broadcast_tick_pull(cpu);
880 /* This actually kills the CPU. */
883 tick_cleanup_dead_cpu(cpu);
884 rcutree_migrate_callbacks(cpu);
888 static void cpuhp_complete_idle_dead(void *arg)
890 struct cpuhp_cpu_state *st = arg;
892 complete_ap_thread(st, false);
895 void cpuhp_report_idle_dead(void)
897 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
899 BUG_ON(st->state != CPUHP_AP_OFFLINE);
900 rcu_report_dead(smp_processor_id());
901 st->state = CPUHP_AP_IDLE_DEAD;
903 * We cannot call complete after rcu_report_dead() so we delegate it
906 smp_call_function_single(cpumask_first(cpu_online_mask),
907 cpuhp_complete_idle_dead, st, 0);
910 static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
912 for (st->state++; st->state < st->target; st->state++)
913 cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
916 static int cpuhp_down_callbacks(unsigned int cpu, struct cpuhp_cpu_state *st,
917 enum cpuhp_state target)
919 enum cpuhp_state prev_state = st->state;
922 for (; st->state > target; st->state--) {
923 ret = cpuhp_invoke_callback(cpu, st->state, false, NULL, NULL);
925 st->target = prev_state;
926 if (st->state < prev_state)
927 undo_cpu_down(cpu, st);
934 /* Requires cpu_add_remove_lock to be held */
935 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
936 enum cpuhp_state target)
938 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
939 int prev_state, ret = 0;
941 if (num_online_cpus() == 1)
944 if (!cpu_present(cpu))
949 cpuhp_tasks_frozen = tasks_frozen;
951 prev_state = cpuhp_set_state(st, target);
953 * If the current CPU state is in the range of the AP hotplug thread,
954 * then we need to kick the thread.
956 if (st->state > CPUHP_TEARDOWN_CPU) {
957 st->target = max((int)target, CPUHP_TEARDOWN_CPU);
958 ret = cpuhp_kick_ap_work(cpu);
960 * The AP side has done the error rollback already. Just
961 * return the error code..
967 * We might have stopped still in the range of the AP hotplug
968 * thread. Nothing to do anymore.
970 if (st->state > CPUHP_TEARDOWN_CPU)
976 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
977 * to do the further cleanups.
979 ret = cpuhp_down_callbacks(cpu, st, target);
980 if (ret && st->state == CPUHP_TEARDOWN_CPU && st->state < prev_state) {
981 cpuhp_reset_state(st, prev_state);
988 * Do post unplug cleanup. This is still protected against
989 * concurrent CPU hotplug via cpu_add_remove_lock.
991 lockup_detector_cleanup();
996 static int cpu_down_maps_locked(unsigned int cpu, enum cpuhp_state target)
998 if (cpu_hotplug_disabled)
1000 return _cpu_down(cpu, 0, target);
1003 static int do_cpu_down(unsigned int cpu, enum cpuhp_state target)
1007 cpu_maps_update_begin();
1008 err = cpu_down_maps_locked(cpu, target);
1009 cpu_maps_update_done();
1013 int cpu_down(unsigned int cpu)
1015 return do_cpu_down(cpu, CPUHP_OFFLINE);
1017 EXPORT_SYMBOL(cpu_down);
1020 #define takedown_cpu NULL
1021 #endif /*CONFIG_HOTPLUG_CPU*/
1024 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1025 * @cpu: cpu that just started
1027 * It must be called by the arch code on the new cpu, before the new cpu
1028 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1030 void notify_cpu_starting(unsigned int cpu)
1032 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1033 enum cpuhp_state target = min((int)st->target, CPUHP_AP_ONLINE);
1036 rcu_cpu_starting(cpu); /* Enables RCU usage on this CPU. */
1037 st->booted_once = true;
1038 while (st->state < target) {
1040 ret = cpuhp_invoke_callback(cpu, st->state, true, NULL, NULL);
1042 * STARTING must not fail!
1049 * Called from the idle task. Wake up the controlling task which brings the
1050 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1051 * the rest of the online bringup to the hotplug thread.
1053 void cpuhp_online_idle(enum cpuhp_state state)
1055 struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
1057 /* Happens for the boot cpu */
1058 if (state != CPUHP_AP_ONLINE_IDLE)
1061 st->state = CPUHP_AP_ONLINE_IDLE;
1062 complete_ap_thread(st, true);
1065 /* Requires cpu_add_remove_lock to be held */
1066 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
1068 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1069 struct task_struct *idle;
1074 if (!cpu_present(cpu)) {
1080 * The caller of do_cpu_up might have raced with another
1081 * caller. Ignore it for now.
1083 if (st->state >= target)
1086 if (st->state == CPUHP_OFFLINE) {
1087 /* Let it fail before we try to bring the cpu up */
1088 idle = idle_thread_get(cpu);
1090 ret = PTR_ERR(idle);
1095 cpuhp_tasks_frozen = tasks_frozen;
1097 cpuhp_set_state(st, target);
1099 * If the current CPU state is in the range of the AP hotplug thread,
1100 * then we need to kick the thread once more.
1102 if (st->state > CPUHP_BRINGUP_CPU) {
1103 ret = cpuhp_kick_ap_work(cpu);
1105 * The AP side has done the error rollback already. Just
1106 * return the error code..
1113 * Try to reach the target state. We max out on the BP at
1114 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1115 * responsible for bringing it up to the target state.
1117 target = min((int)target, CPUHP_BRINGUP_CPU);
1118 ret = cpuhp_up_callbacks(cpu, st, target);
1120 cpus_write_unlock();
1125 static int do_cpu_up(unsigned int cpu, enum cpuhp_state target)
1129 if (!cpu_possible(cpu)) {
1130 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1132 #if defined(CONFIG_IA64)
1133 pr_err("please check additional_cpus= boot parameter\n");
1138 err = try_online_node(cpu_to_node(cpu));
1142 cpu_maps_update_begin();
1144 if (cpu_hotplug_disabled) {
1148 if (!cpu_smt_allowed(cpu)) {
1153 err = _cpu_up(cpu, 0, target);
1155 cpu_maps_update_done();
1159 int cpu_up(unsigned int cpu)
1161 return do_cpu_up(cpu, CPUHP_ONLINE);
1163 EXPORT_SYMBOL_GPL(cpu_up);
1165 #ifdef CONFIG_PM_SLEEP_SMP
1166 static cpumask_var_t frozen_cpus;
1168 int freeze_secondary_cpus(int primary)
1172 cpu_maps_update_begin();
1173 if (!cpu_online(primary))
1174 primary = cpumask_first(cpu_online_mask);
1176 * We take down all of the non-boot CPUs in one shot to avoid races
1177 * with the userspace trying to use the CPU hotplug at the same time
1179 cpumask_clear(frozen_cpus);
1181 pr_info("Disabling non-boot CPUs ...\n");
1182 for_each_online_cpu(cpu) {
1185 trace_suspend_resume(TPS("CPU_OFF"), cpu, true);
1186 error = _cpu_down(cpu, 1, CPUHP_OFFLINE);
1187 trace_suspend_resume(TPS("CPU_OFF"), cpu, false);
1189 cpumask_set_cpu(cpu, frozen_cpus);
1191 pr_err("Error taking CPU%d down: %d\n", cpu, error);
1197 BUG_ON(num_online_cpus() > 1);
1199 pr_err("Non-boot CPUs are not disabled\n");
1202 * Make sure the CPUs won't be enabled by someone else. We need to do
1203 * this even in case of failure as all disable_nonboot_cpus() users are
1204 * supposed to do enable_nonboot_cpus() on the failure path.
1206 cpu_hotplug_disabled++;
1208 cpu_maps_update_done();
1212 void __weak arch_enable_nonboot_cpus_begin(void)
1216 void __weak arch_enable_nonboot_cpus_end(void)
1220 void enable_nonboot_cpus(void)
1224 /* Allow everyone to use the CPU hotplug again */
1225 cpu_maps_update_begin();
1226 __cpu_hotplug_enable();
1227 if (cpumask_empty(frozen_cpus))
1230 pr_info("Enabling non-boot CPUs ...\n");
1232 arch_enable_nonboot_cpus_begin();
1234 for_each_cpu(cpu, frozen_cpus) {
1235 trace_suspend_resume(TPS("CPU_ON"), cpu, true);
1236 error = _cpu_up(cpu, 1, CPUHP_ONLINE);
1237 trace_suspend_resume(TPS("CPU_ON"), cpu, false);
1239 pr_info("CPU%d is up\n", cpu);
1242 pr_warn("Error taking CPU%d up: %d\n", cpu, error);
1245 arch_enable_nonboot_cpus_end();
1247 cpumask_clear(frozen_cpus);
1249 cpu_maps_update_done();
1252 static int __init alloc_frozen_cpus(void)
1254 if (!alloc_cpumask_var(&frozen_cpus, GFP_KERNEL|__GFP_ZERO))
1258 core_initcall(alloc_frozen_cpus);
1261 * When callbacks for CPU hotplug notifications are being executed, we must
1262 * ensure that the state of the system with respect to the tasks being frozen
1263 * or not, as reported by the notification, remains unchanged *throughout the
1264 * duration* of the execution of the callbacks.
1265 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1267 * This synchronization is implemented by mutually excluding regular CPU
1268 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1269 * Hibernate notifications.
1272 cpu_hotplug_pm_callback(struct notifier_block *nb,
1273 unsigned long action, void *ptr)
1277 case PM_SUSPEND_PREPARE:
1278 case PM_HIBERNATION_PREPARE:
1279 cpu_hotplug_disable();
1282 case PM_POST_SUSPEND:
1283 case PM_POST_HIBERNATION:
1284 cpu_hotplug_enable();
1295 static int __init cpu_hotplug_pm_sync_init(void)
1298 * cpu_hotplug_pm_callback has higher priority than x86
1299 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1300 * to disable cpu hotplug to avoid cpu hotplug race.
1302 pm_notifier(cpu_hotplug_pm_callback, 0);
1305 core_initcall(cpu_hotplug_pm_sync_init);
1307 #endif /* CONFIG_PM_SLEEP_SMP */
1311 #endif /* CONFIG_SMP */
1313 /* Boot processor state steps */
1314 static struct cpuhp_step cpuhp_hp_states[] = {
1317 .startup.single = NULL,
1318 .teardown.single = NULL,
1321 [CPUHP_CREATE_THREADS]= {
1322 .name = "threads:prepare",
1323 .startup.single = smpboot_create_threads,
1324 .teardown.single = NULL,
1327 [CPUHP_PERF_PREPARE] = {
1328 .name = "perf:prepare",
1329 .startup.single = perf_event_init_cpu,
1330 .teardown.single = perf_event_exit_cpu,
1332 [CPUHP_WORKQUEUE_PREP] = {
1333 .name = "workqueue:prepare",
1334 .startup.single = workqueue_prepare_cpu,
1335 .teardown.single = NULL,
1337 [CPUHP_HRTIMERS_PREPARE] = {
1338 .name = "hrtimers:prepare",
1339 .startup.single = hrtimers_prepare_cpu,
1340 .teardown.single = hrtimers_dead_cpu,
1342 [CPUHP_SMPCFD_PREPARE] = {
1343 .name = "smpcfd:prepare",
1344 .startup.single = smpcfd_prepare_cpu,
1345 .teardown.single = smpcfd_dead_cpu,
1347 [CPUHP_RELAY_PREPARE] = {
1348 .name = "relay:prepare",
1349 .startup.single = relay_prepare_cpu,
1350 .teardown.single = NULL,
1352 [CPUHP_SLAB_PREPARE] = {
1353 .name = "slab:prepare",
1354 .startup.single = slab_prepare_cpu,
1355 .teardown.single = slab_dead_cpu,
1357 [CPUHP_RCUTREE_PREP] = {
1358 .name = "RCU/tree:prepare",
1359 .startup.single = rcutree_prepare_cpu,
1360 .teardown.single = rcutree_dead_cpu,
1363 * On the tear-down path, timers_dead_cpu() must be invoked
1364 * before blk_mq_queue_reinit_notify() from notify_dead(),
1365 * otherwise a RCU stall occurs.
1367 [CPUHP_TIMERS_PREPARE] = {
1368 .name = "timers:prepare",
1369 .startup.single = timers_prepare_cpu,
1370 .teardown.single = timers_dead_cpu,
1372 /* Kicks the plugged cpu into life */
1373 [CPUHP_BRINGUP_CPU] = {
1374 .name = "cpu:bringup",
1375 .startup.single = bringup_cpu,
1376 .teardown.single = NULL,
1379 /* Final state before CPU kills itself */
1380 [CPUHP_AP_IDLE_DEAD] = {
1381 .name = "idle:dead",
1384 * Last state before CPU enters the idle loop to die. Transient state
1385 * for synchronization.
1387 [CPUHP_AP_OFFLINE] = {
1388 .name = "ap:offline",
1391 /* First state is scheduler control. Interrupts are disabled */
1392 [CPUHP_AP_SCHED_STARTING] = {
1393 .name = "sched:starting",
1394 .startup.single = sched_cpu_starting,
1395 .teardown.single = sched_cpu_dying,
1397 [CPUHP_AP_RCUTREE_DYING] = {
1398 .name = "RCU/tree:dying",
1399 .startup.single = NULL,
1400 .teardown.single = rcutree_dying_cpu,
1402 [CPUHP_AP_SMPCFD_DYING] = {
1403 .name = "smpcfd:dying",
1404 .startup.single = NULL,
1405 .teardown.single = smpcfd_dying_cpu,
1407 /* Entry state on starting. Interrupts enabled from here on. Transient
1408 * state for synchronsization */
1409 [CPUHP_AP_ONLINE] = {
1410 .name = "ap:online",
1413 * Handled on controll processor until the plugged processor manages
1416 [CPUHP_TEARDOWN_CPU] = {
1417 .name = "cpu:teardown",
1418 .startup.single = NULL,
1419 .teardown.single = takedown_cpu,
1422 /* Handle smpboot threads park/unpark */
1423 [CPUHP_AP_SMPBOOT_THREADS] = {
1424 .name = "smpboot/threads:online",
1425 .startup.single = smpboot_unpark_threads,
1426 .teardown.single = smpboot_park_threads,
1428 [CPUHP_AP_IRQ_AFFINITY_ONLINE] = {
1429 .name = "irq/affinity:online",
1430 .startup.single = irq_affinity_online_cpu,
1431 .teardown.single = NULL,
1433 [CPUHP_AP_PERF_ONLINE] = {
1434 .name = "perf:online",
1435 .startup.single = perf_event_init_cpu,
1436 .teardown.single = perf_event_exit_cpu,
1438 [CPUHP_AP_WATCHDOG_ONLINE] = {
1439 .name = "lockup_detector:online",
1440 .startup.single = lockup_detector_online_cpu,
1441 .teardown.single = lockup_detector_offline_cpu,
1443 [CPUHP_AP_WORKQUEUE_ONLINE] = {
1444 .name = "workqueue:online",
1445 .startup.single = workqueue_online_cpu,
1446 .teardown.single = workqueue_offline_cpu,
1448 [CPUHP_AP_RCUTREE_ONLINE] = {
1449 .name = "RCU/tree:online",
1450 .startup.single = rcutree_online_cpu,
1451 .teardown.single = rcutree_offline_cpu,
1455 * The dynamically registered state space is here
1459 /* Last state is scheduler control setting the cpu active */
1460 [CPUHP_AP_ACTIVE] = {
1461 .name = "sched:active",
1462 .startup.single = sched_cpu_activate,
1463 .teardown.single = sched_cpu_deactivate,
1467 /* CPU is fully up and running. */
1470 .startup.single = NULL,
1471 .teardown.single = NULL,
1475 /* Sanity check for callbacks */
1476 static int cpuhp_cb_check(enum cpuhp_state state)
1478 if (state <= CPUHP_OFFLINE || state >= CPUHP_ONLINE)
1484 * Returns a free for dynamic slot assignment of the Online state. The states
1485 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1486 * by having no name assigned.
1488 static int cpuhp_reserve_state(enum cpuhp_state state)
1490 enum cpuhp_state i, end;
1491 struct cpuhp_step *step;
1494 case CPUHP_AP_ONLINE_DYN:
1495 step = cpuhp_hp_states + CPUHP_AP_ONLINE_DYN;
1496 end = CPUHP_AP_ONLINE_DYN_END;
1498 case CPUHP_BP_PREPARE_DYN:
1499 step = cpuhp_hp_states + CPUHP_BP_PREPARE_DYN;
1500 end = CPUHP_BP_PREPARE_DYN_END;
1506 for (i = state; i <= end; i++, step++) {
1510 WARN(1, "No more dynamic states available for CPU hotplug\n");
1514 static int cpuhp_store_callbacks(enum cpuhp_state state, const char *name,
1515 int (*startup)(unsigned int cpu),
1516 int (*teardown)(unsigned int cpu),
1517 bool multi_instance)
1519 /* (Un)Install the callbacks for further cpu hotplug operations */
1520 struct cpuhp_step *sp;
1524 * If name is NULL, then the state gets removed.
1526 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1527 * the first allocation from these dynamic ranges, so the removal
1528 * would trigger a new allocation and clear the wrong (already
1529 * empty) state, leaving the callbacks of the to be cleared state
1530 * dangling, which causes wreckage on the next hotplug operation.
1532 if (name && (state == CPUHP_AP_ONLINE_DYN ||
1533 state == CPUHP_BP_PREPARE_DYN)) {
1534 ret = cpuhp_reserve_state(state);
1539 sp = cpuhp_get_step(state);
1540 if (name && sp->name)
1543 sp->startup.single = startup;
1544 sp->teardown.single = teardown;
1546 sp->multi_instance = multi_instance;
1547 INIT_HLIST_HEAD(&sp->list);
1551 static void *cpuhp_get_teardown_cb(enum cpuhp_state state)
1553 return cpuhp_get_step(state)->teardown.single;
1557 * Call the startup/teardown function for a step either on the AP or
1558 * on the current CPU.
1560 static int cpuhp_issue_call(int cpu, enum cpuhp_state state, bool bringup,
1561 struct hlist_node *node)
1563 struct cpuhp_step *sp = cpuhp_get_step(state);
1567 * If there's nothing to do, we done.
1568 * Relies on the union for multi_instance.
1570 if ((bringup && !sp->startup.single) ||
1571 (!bringup && !sp->teardown.single))
1574 * The non AP bound callbacks can fail on bringup. On teardown
1575 * e.g. module removal we crash for now.
1578 if (cpuhp_is_ap_state(state))
1579 ret = cpuhp_invoke_ap_callback(cpu, state, bringup, node);
1581 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1583 ret = cpuhp_invoke_callback(cpu, state, bringup, node, NULL);
1585 BUG_ON(ret && !bringup);
1590 * Called from __cpuhp_setup_state on a recoverable failure.
1592 * Note: The teardown callbacks for rollback are not allowed to fail!
1594 static void cpuhp_rollback_install(int failedcpu, enum cpuhp_state state,
1595 struct hlist_node *node)
1599 /* Roll back the already executed steps on the other cpus */
1600 for_each_present_cpu(cpu) {
1601 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1602 int cpustate = st->state;
1604 if (cpu >= failedcpu)
1607 /* Did we invoke the startup call on that cpu ? */
1608 if (cpustate >= state)
1609 cpuhp_issue_call(cpu, state, false, node);
1613 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state,
1614 struct hlist_node *node,
1617 struct cpuhp_step *sp;
1621 lockdep_assert_cpus_held();
1623 sp = cpuhp_get_step(state);
1624 if (sp->multi_instance == false)
1627 mutex_lock(&cpuhp_state_mutex);
1629 if (!invoke || !sp->startup.multi)
1633 * Try to call the startup callback for each present cpu
1634 * depending on the hotplug state of the cpu.
1636 for_each_present_cpu(cpu) {
1637 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1638 int cpustate = st->state;
1640 if (cpustate < state)
1643 ret = cpuhp_issue_call(cpu, state, true, node);
1645 if (sp->teardown.multi)
1646 cpuhp_rollback_install(cpu, state, node);
1652 hlist_add_head(node, &sp->list);
1654 mutex_unlock(&cpuhp_state_mutex);
1658 int __cpuhp_state_add_instance(enum cpuhp_state state, struct hlist_node *node,
1664 ret = __cpuhp_state_add_instance_cpuslocked(state, node, invoke);
1668 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance);
1671 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1672 * @state: The state to setup
1673 * @invoke: If true, the startup function is invoked for cpus where
1674 * cpu state >= @state
1675 * @startup: startup callback function
1676 * @teardown: teardown callback function
1677 * @multi_instance: State is set up for multiple instances which get
1680 * The caller needs to hold cpus read locked while calling this function.
1683 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1684 * 0 for all other states
1685 * On failure: proper (negative) error code
1687 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state,
1688 const char *name, bool invoke,
1689 int (*startup)(unsigned int cpu),
1690 int (*teardown)(unsigned int cpu),
1691 bool multi_instance)
1696 lockdep_assert_cpus_held();
1698 if (cpuhp_cb_check(state) || !name)
1701 mutex_lock(&cpuhp_state_mutex);
1703 ret = cpuhp_store_callbacks(state, name, startup, teardown,
1706 dynstate = state == CPUHP_AP_ONLINE_DYN;
1707 if (ret > 0 && dynstate) {
1712 if (ret || !invoke || !startup)
1716 * Try to call the startup callback for each present cpu
1717 * depending on the hotplug state of the cpu.
1719 for_each_present_cpu(cpu) {
1720 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1721 int cpustate = st->state;
1723 if (cpustate < state)
1726 ret = cpuhp_issue_call(cpu, state, true, NULL);
1729 cpuhp_rollback_install(cpu, state, NULL);
1730 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1735 mutex_unlock(&cpuhp_state_mutex);
1737 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1738 * dynamically allocated state in case of success.
1740 if (!ret && dynstate)
1744 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked);
1746 int __cpuhp_setup_state(enum cpuhp_state state,
1747 const char *name, bool invoke,
1748 int (*startup)(unsigned int cpu),
1749 int (*teardown)(unsigned int cpu),
1750 bool multi_instance)
1755 ret = __cpuhp_setup_state_cpuslocked(state, name, invoke, startup,
1756 teardown, multi_instance);
1760 EXPORT_SYMBOL(__cpuhp_setup_state);
1762 int __cpuhp_state_remove_instance(enum cpuhp_state state,
1763 struct hlist_node *node, bool invoke)
1765 struct cpuhp_step *sp = cpuhp_get_step(state);
1768 BUG_ON(cpuhp_cb_check(state));
1770 if (!sp->multi_instance)
1774 mutex_lock(&cpuhp_state_mutex);
1776 if (!invoke || !cpuhp_get_teardown_cb(state))
1779 * Call the teardown callback for each present cpu depending
1780 * on the hotplug state of the cpu. This function is not
1781 * allowed to fail currently!
1783 for_each_present_cpu(cpu) {
1784 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1785 int cpustate = st->state;
1787 if (cpustate >= state)
1788 cpuhp_issue_call(cpu, state, false, node);
1793 mutex_unlock(&cpuhp_state_mutex);
1798 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance);
1801 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1802 * @state: The state to remove
1803 * @invoke: If true, the teardown function is invoked for cpus where
1804 * cpu state >= @state
1806 * The caller needs to hold cpus read locked while calling this function.
1807 * The teardown callback is currently not allowed to fail. Think
1808 * about module removal!
1810 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state, bool invoke)
1812 struct cpuhp_step *sp = cpuhp_get_step(state);
1815 BUG_ON(cpuhp_cb_check(state));
1817 lockdep_assert_cpus_held();
1819 mutex_lock(&cpuhp_state_mutex);
1820 if (sp->multi_instance) {
1821 WARN(!hlist_empty(&sp->list),
1822 "Error: Removing state %d which has instances left.\n",
1827 if (!invoke || !cpuhp_get_teardown_cb(state))
1831 * Call the teardown callback for each present cpu depending
1832 * on the hotplug state of the cpu. This function is not
1833 * allowed to fail currently!
1835 for_each_present_cpu(cpu) {
1836 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
1837 int cpustate = st->state;
1839 if (cpustate >= state)
1840 cpuhp_issue_call(cpu, state, false, NULL);
1843 cpuhp_store_callbacks(state, NULL, NULL, NULL, false);
1844 mutex_unlock(&cpuhp_state_mutex);
1846 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked);
1848 void __cpuhp_remove_state(enum cpuhp_state state, bool invoke)
1851 __cpuhp_remove_state_cpuslocked(state, invoke);
1854 EXPORT_SYMBOL(__cpuhp_remove_state);
1856 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1857 static ssize_t show_cpuhp_state(struct device *dev,
1858 struct device_attribute *attr, char *buf)
1860 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1862 return sprintf(buf, "%d\n", st->state);
1864 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
1866 static ssize_t write_cpuhp_target(struct device *dev,
1867 struct device_attribute *attr,
1868 const char *buf, size_t count)
1870 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1871 struct cpuhp_step *sp;
1874 ret = kstrtoint(buf, 10, &target);
1878 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1879 if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
1882 if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
1886 ret = lock_device_hotplug_sysfs();
1890 mutex_lock(&cpuhp_state_mutex);
1891 sp = cpuhp_get_step(target);
1892 ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
1893 mutex_unlock(&cpuhp_state_mutex);
1897 if (st->state < target)
1898 ret = do_cpu_up(dev->id, target);
1900 ret = do_cpu_down(dev->id, target);
1902 unlock_device_hotplug();
1903 return ret ? ret : count;
1906 static ssize_t show_cpuhp_target(struct device *dev,
1907 struct device_attribute *attr, char *buf)
1909 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1911 return sprintf(buf, "%d\n", st->target);
1913 static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
1916 static ssize_t write_cpuhp_fail(struct device *dev,
1917 struct device_attribute *attr,
1918 const char *buf, size_t count)
1920 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1921 struct cpuhp_step *sp;
1924 ret = kstrtoint(buf, 10, &fail);
1928 if (fail < CPUHP_OFFLINE || fail > CPUHP_ONLINE)
1932 * Cannot fail STARTING/DYING callbacks.
1934 if (cpuhp_is_atomic_state(fail))
1938 * Cannot fail anything that doesn't have callbacks.
1940 mutex_lock(&cpuhp_state_mutex);
1941 sp = cpuhp_get_step(fail);
1942 if (!sp->startup.single && !sp->teardown.single)
1944 mutex_unlock(&cpuhp_state_mutex);
1953 static ssize_t show_cpuhp_fail(struct device *dev,
1954 struct device_attribute *attr, char *buf)
1956 struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
1958 return sprintf(buf, "%d\n", st->fail);
1961 static DEVICE_ATTR(fail, 0644, show_cpuhp_fail, write_cpuhp_fail);
1963 static struct attribute *cpuhp_cpu_attrs[] = {
1964 &dev_attr_state.attr,
1965 &dev_attr_target.attr,
1966 &dev_attr_fail.attr,
1970 static const struct attribute_group cpuhp_cpu_attr_group = {
1971 .attrs = cpuhp_cpu_attrs,
1976 static ssize_t show_cpuhp_states(struct device *dev,
1977 struct device_attribute *attr, char *buf)
1979 ssize_t cur, res = 0;
1982 mutex_lock(&cpuhp_state_mutex);
1983 for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
1984 struct cpuhp_step *sp = cpuhp_get_step(i);
1987 cur = sprintf(buf, "%3d: %s\n", i, sp->name);
1992 mutex_unlock(&cpuhp_state_mutex);
1995 static DEVICE_ATTR(states, 0444, show_cpuhp_states, NULL);
1997 static struct attribute *cpuhp_cpu_root_attrs[] = {
1998 &dev_attr_states.attr,
2002 static const struct attribute_group cpuhp_cpu_root_attr_group = {
2003 .attrs = cpuhp_cpu_root_attrs,
2008 #ifdef CONFIG_HOTPLUG_SMT
2010 static const char *smt_states[] = {
2011 [CPU_SMT_ENABLED] = "on",
2012 [CPU_SMT_DISABLED] = "off",
2013 [CPU_SMT_FORCE_DISABLED] = "forceoff",
2014 [CPU_SMT_NOT_SUPPORTED] = "notsupported",
2018 show_smt_control(struct device *dev, struct device_attribute *attr, char *buf)
2020 return snprintf(buf, PAGE_SIZE - 2, "%s\n", smt_states[cpu_smt_control]);
2023 static void cpuhp_offline_cpu_device(unsigned int cpu)
2025 struct device *dev = get_cpu_device(cpu);
2027 dev->offline = true;
2028 /* Tell user space about the state change */
2029 kobject_uevent(&dev->kobj, KOBJ_OFFLINE);
2032 static void cpuhp_online_cpu_device(unsigned int cpu)
2034 struct device *dev = get_cpu_device(cpu);
2036 dev->offline = false;
2037 /* Tell user space about the state change */
2038 kobject_uevent(&dev->kobj, KOBJ_ONLINE);
2041 int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval)
2045 cpu_maps_update_begin();
2046 for_each_online_cpu(cpu) {
2047 if (topology_is_primary_thread(cpu))
2049 ret = cpu_down_maps_locked(cpu, CPUHP_OFFLINE);
2053 * As this needs to hold the cpu maps lock it's impossible
2054 * to call device_offline() because that ends up calling
2055 * cpu_down() which takes cpu maps lock. cpu maps lock
2056 * needs to be held as this might race against in kernel
2057 * abusers of the hotplug machinery (thermal management).
2059 * So nothing would update device:offline state. That would
2060 * leave the sysfs entry stale and prevent onlining after
2061 * smt control has been changed to 'off' again. This is
2062 * called under the sysfs hotplug lock, so it is properly
2063 * serialized against the regular offline usage.
2065 cpuhp_offline_cpu_device(cpu);
2068 cpu_smt_control = ctrlval;
2071 cpu_maps_update_done();
2075 int cpuhp_smt_enable(void)
2079 cpu_maps_update_begin();
2080 cpu_smt_control = CPU_SMT_ENABLED;
2082 for_each_present_cpu(cpu) {
2083 /* Skip online CPUs and CPUs on offline nodes */
2084 if (cpu_online(cpu) || !node_online(cpu_to_node(cpu)))
2086 ret = _cpu_up(cpu, 0, CPUHP_ONLINE);
2089 /* See comment in cpuhp_smt_disable() */
2090 cpuhp_online_cpu_device(cpu);
2092 cpu_maps_update_done();
2097 store_smt_control(struct device *dev, struct device_attribute *attr,
2098 const char *buf, size_t count)
2102 if (sysfs_streq(buf, "on"))
2103 ctrlval = CPU_SMT_ENABLED;
2104 else if (sysfs_streq(buf, "off"))
2105 ctrlval = CPU_SMT_DISABLED;
2106 else if (sysfs_streq(buf, "forceoff"))
2107 ctrlval = CPU_SMT_FORCE_DISABLED;
2111 if (cpu_smt_control == CPU_SMT_FORCE_DISABLED)
2114 if (cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
2117 ret = lock_device_hotplug_sysfs();
2121 if (ctrlval != cpu_smt_control) {
2123 case CPU_SMT_ENABLED:
2124 ret = cpuhp_smt_enable();
2126 case CPU_SMT_DISABLED:
2127 case CPU_SMT_FORCE_DISABLED:
2128 ret = cpuhp_smt_disable(ctrlval);
2133 unlock_device_hotplug();
2134 return ret ? ret : count;
2136 static DEVICE_ATTR(control, 0644, show_smt_control, store_smt_control);
2139 show_smt_active(struct device *dev, struct device_attribute *attr, char *buf)
2141 bool active = topology_max_smt_threads() > 1;
2143 return snprintf(buf, PAGE_SIZE - 2, "%d\n", active);
2145 static DEVICE_ATTR(active, 0444, show_smt_active, NULL);
2147 static struct attribute *cpuhp_smt_attrs[] = {
2148 &dev_attr_control.attr,
2149 &dev_attr_active.attr,
2153 static const struct attribute_group cpuhp_smt_attr_group = {
2154 .attrs = cpuhp_smt_attrs,
2159 static int __init cpu_smt_state_init(void)
2161 return sysfs_create_group(&cpu_subsys.dev_root->kobj,
2162 &cpuhp_smt_attr_group);
2166 static inline int cpu_smt_state_init(void) { return 0; }
2169 static int __init cpuhp_sysfs_init(void)
2173 ret = cpu_smt_state_init();
2177 ret = sysfs_create_group(&cpu_subsys.dev_root->kobj,
2178 &cpuhp_cpu_root_attr_group);
2182 for_each_possible_cpu(cpu) {
2183 struct device *dev = get_cpu_device(cpu);
2187 ret = sysfs_create_group(&dev->kobj, &cpuhp_cpu_attr_group);
2193 device_initcall(cpuhp_sysfs_init);
2197 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2198 * represents all NR_CPUS bits binary values of 1<<nr.
2200 * It is used by cpumask_of() to get a constant address to a CPU
2201 * mask value that has a single bit set only.
2204 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2205 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2206 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2207 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2208 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2210 const unsigned long cpu_bit_bitmap[BITS_PER_LONG+1][BITS_TO_LONGS(NR_CPUS)] = {
2212 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2213 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2214 #if BITS_PER_LONG > 32
2215 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2216 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2219 EXPORT_SYMBOL_GPL(cpu_bit_bitmap);
2221 const DECLARE_BITMAP(cpu_all_bits, NR_CPUS) = CPU_BITS_ALL;
2222 EXPORT_SYMBOL(cpu_all_bits);
2224 #ifdef CONFIG_INIT_ALL_POSSIBLE
2225 struct cpumask __cpu_possible_mask __read_mostly
2228 struct cpumask __cpu_possible_mask __read_mostly;
2230 EXPORT_SYMBOL(__cpu_possible_mask);
2232 struct cpumask __cpu_online_mask __read_mostly;
2233 EXPORT_SYMBOL(__cpu_online_mask);
2235 struct cpumask __cpu_present_mask __read_mostly;
2236 EXPORT_SYMBOL(__cpu_present_mask);
2238 struct cpumask __cpu_active_mask __read_mostly;
2239 EXPORT_SYMBOL(__cpu_active_mask);
2241 void init_cpu_present(const struct cpumask *src)
2243 cpumask_copy(&__cpu_present_mask, src);
2246 void init_cpu_possible(const struct cpumask *src)
2248 cpumask_copy(&__cpu_possible_mask, src);
2251 void init_cpu_online(const struct cpumask *src)
2253 cpumask_copy(&__cpu_online_mask, src);
2257 * Activate the first processor.
2259 void __init boot_cpu_init(void)
2261 int cpu = smp_processor_id();
2263 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2264 set_cpu_online(cpu, true);
2265 set_cpu_active(cpu, true);
2266 set_cpu_present(cpu, true);
2267 set_cpu_possible(cpu, true);
2270 __boot_cpu_id = cpu;
2275 * Must be called _AFTER_ setting up the per_cpu areas
2277 void __init boot_cpu_hotplug_init(void)
2280 this_cpu_write(cpuhp_state.booted_once, true);
2282 this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
2285 enum cpu_mitigations cpu_mitigations __ro_after_init = CPU_MITIGATIONS_AUTO;
2287 static int __init mitigations_parse_cmdline(char *arg)
2289 if (!strcmp(arg, "off"))
2290 cpu_mitigations = CPU_MITIGATIONS_OFF;
2291 else if (!strcmp(arg, "auto"))
2292 cpu_mitigations = CPU_MITIGATIONS_AUTO;
2293 else if (!strcmp(arg, "auto,nosmt"))
2294 cpu_mitigations = CPU_MITIGATIONS_AUTO_NOSMT;
2296 pr_crit("Unsupported mitigations=%s, system may still be vulnerable\n",
2301 early_param("mitigations", mitigations_parse_cmdline);