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 * @skip_onerr: Do not invoke the functions on error rollback
107 * Will go away once the notifiers are gone
108 * @cant_stop: Bringup/teardown can't be stopped at this step
113 int (*single
)(unsigned int cpu
);
114 int (*multi
)(unsigned int cpu
,
115 struct hlist_node
*node
);
118 int (*single
)(unsigned int cpu
);
119 int (*multi
)(unsigned int cpu
,
120 struct hlist_node
*node
);
122 struct hlist_head list
;
128 static DEFINE_MUTEX(cpuhp_state_mutex
);
129 static struct cpuhp_step cpuhp_bp_states
[];
130 static struct cpuhp_step cpuhp_ap_states
[];
132 static bool cpuhp_is_ap_state(enum cpuhp_state state
)
135 * The extra check for CPUHP_TEARDOWN_CPU is only for documentation
136 * purposes as that state is handled explicitly in cpu_down.
138 return state
> CPUHP_BRINGUP_CPU
&& state
!= CPUHP_TEARDOWN_CPU
;
141 static struct cpuhp_step
*cpuhp_get_step(enum cpuhp_state state
)
143 struct cpuhp_step
*sp
;
145 sp
= cpuhp_is_ap_state(state
) ? cpuhp_ap_states
: cpuhp_bp_states
;
150 * cpuhp_invoke_callback _ Invoke the callbacks for a given state
151 * @cpu: The cpu for which the callback should be invoked
152 * @state: The state to do callbacks for
153 * @bringup: True if the bringup callback should be invoked
154 * @node: For multi-instance, do a single entry callback for install/remove
155 * @lastp: For multi-instance rollback, remember how far we got
157 * Called from cpu hotplug and from the state register machinery.
159 static int cpuhp_invoke_callback(unsigned int cpu
, enum cpuhp_state state
,
160 bool bringup
, struct hlist_node
*node
,
161 struct hlist_node
**lastp
)
163 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
164 struct cpuhp_step
*step
= cpuhp_get_step(state
);
165 int (*cbm
)(unsigned int cpu
, struct hlist_node
*node
);
166 int (*cb
)(unsigned int cpu
);
169 if (st
->fail
== state
) {
170 st
->fail
= CPUHP_INVALID
;
172 if (!(bringup
? step
->startup
.single
: step
->teardown
.single
))
178 if (!step
->multi_instance
) {
179 WARN_ON_ONCE(lastp
&& *lastp
);
180 cb
= bringup
? step
->startup
.single
: step
->teardown
.single
;
183 trace_cpuhp_enter(cpu
, st
->target
, state
, cb
);
185 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
188 cbm
= bringup
? step
->startup
.multi
: step
->teardown
.multi
;
192 /* Single invocation for instance add/remove */
194 WARN_ON_ONCE(lastp
&& *lastp
);
195 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
196 ret
= cbm(cpu
, node
);
197 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
201 /* State transition. Invoke on all instances */
203 hlist_for_each(node
, &step
->list
) {
204 if (lastp
&& node
== *lastp
)
207 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
208 ret
= cbm(cpu
, node
);
209 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
223 /* Rollback the instances if one failed */
224 cbm
= !bringup
? step
->startup
.multi
: step
->teardown
.multi
;
228 hlist_for_each(node
, &step
->list
) {
232 trace_cpuhp_multi_enter(cpu
, st
->target
, state
, cbm
, node
);
233 ret
= cbm(cpu
, node
);
234 trace_cpuhp_exit(cpu
, st
->state
, state
, ret
);
236 * Rollback must not fail,
244 static inline void wait_for_ap_thread(struct cpuhp_cpu_state
*st
, bool bringup
)
246 struct completion
*done
= bringup
? &st
->done_up
: &st
->done_down
;
247 wait_for_completion(done
);
250 static inline void complete_ap_thread(struct cpuhp_cpu_state
*st
, bool bringup
)
252 struct completion
*done
= bringup
? &st
->done_up
: &st
->done_down
;
257 * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
259 static bool cpuhp_is_atomic_state(enum cpuhp_state state
)
261 return CPUHP_AP_IDLE_DEAD
<= state
&& state
< CPUHP_AP_ONLINE
;
264 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
265 static DEFINE_MUTEX(cpu_add_remove_lock
);
266 bool cpuhp_tasks_frozen
;
267 EXPORT_SYMBOL_GPL(cpuhp_tasks_frozen
);
270 * The following two APIs (cpu_maps_update_begin/done) must be used when
271 * attempting to serialize the updates to cpu_online_mask & cpu_present_mask.
273 void cpu_maps_update_begin(void)
275 mutex_lock(&cpu_add_remove_lock
);
278 void cpu_maps_update_done(void)
280 mutex_unlock(&cpu_add_remove_lock
);
284 * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
285 * Should always be manipulated under cpu_add_remove_lock
287 static int cpu_hotplug_disabled
;
289 #ifdef CONFIG_HOTPLUG_CPU
291 DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock
);
293 void cpus_read_lock(void)
295 percpu_down_read(&cpu_hotplug_lock
);
297 EXPORT_SYMBOL_GPL(cpus_read_lock
);
299 void cpus_read_unlock(void)
301 percpu_up_read(&cpu_hotplug_lock
);
303 EXPORT_SYMBOL_GPL(cpus_read_unlock
);
305 void cpus_write_lock(void)
307 percpu_down_write(&cpu_hotplug_lock
);
310 void cpus_write_unlock(void)
312 percpu_up_write(&cpu_hotplug_lock
);
315 void lockdep_assert_cpus_held(void)
317 percpu_rwsem_assert_held(&cpu_hotplug_lock
);
321 * Wait for currently running CPU hotplug operations to complete (if any) and
322 * disable future CPU hotplug (from sysfs). The 'cpu_add_remove_lock' protects
323 * the 'cpu_hotplug_disabled' flag. The same lock is also acquired by the
324 * hotplug path before performing hotplug operations. So acquiring that lock
325 * guarantees mutual exclusion from any currently running hotplug operations.
327 void cpu_hotplug_disable(void)
329 cpu_maps_update_begin();
330 cpu_hotplug_disabled
++;
331 cpu_maps_update_done();
333 EXPORT_SYMBOL_GPL(cpu_hotplug_disable
);
335 static void __cpu_hotplug_enable(void)
337 if (WARN_ONCE(!cpu_hotplug_disabled
, "Unbalanced cpu hotplug enable\n"))
339 cpu_hotplug_disabled
--;
342 void cpu_hotplug_enable(void)
344 cpu_maps_update_begin();
345 __cpu_hotplug_enable();
346 cpu_maps_update_done();
348 EXPORT_SYMBOL_GPL(cpu_hotplug_enable
);
349 #endif /* CONFIG_HOTPLUG_CPU */
352 * Architectures that need SMT-specific errata handling during SMT hotplug
353 * should override this.
355 void __weak
arch_smt_update(void) { }
357 #ifdef CONFIG_HOTPLUG_SMT
358 enum cpuhp_smt_control cpu_smt_control __read_mostly
= CPU_SMT_ENABLED
;
360 void __init
cpu_smt_disable(bool force
)
362 if (cpu_smt_control
== CPU_SMT_FORCE_DISABLED
||
363 cpu_smt_control
== CPU_SMT_NOT_SUPPORTED
)
367 pr_info("SMT: Force disabled\n");
368 cpu_smt_control
= CPU_SMT_FORCE_DISABLED
;
370 cpu_smt_control
= CPU_SMT_DISABLED
;
375 * The decision whether SMT is supported can only be done after the full
376 * CPU identification. Called from architecture code.
378 void __init
cpu_smt_check_topology(void)
380 if (!topology_smt_supported())
381 cpu_smt_control
= CPU_SMT_NOT_SUPPORTED
;
384 static int __init
smt_cmdline_disable(char *str
)
386 cpu_smt_disable(str
&& !strcmp(str
, "force"));
389 early_param("nosmt", smt_cmdline_disable
);
391 static inline bool cpu_smt_allowed(unsigned int cpu
)
393 if (cpu_smt_control
== CPU_SMT_ENABLED
)
396 if (topology_is_primary_thread(cpu
))
400 * On x86 it's required to boot all logical CPUs at least once so
401 * that the init code can get a chance to set CR4.MCE on each
402 * CPU. Otherwise, a broadacasted MCE observing CR4.MCE=0b on any
403 * core will shutdown the machine.
405 return !per_cpu(cpuhp_state
, cpu
).booted_once
;
408 static inline bool cpu_smt_allowed(unsigned int cpu
) { return true; }
411 static inline enum cpuhp_state
412 cpuhp_set_state(struct cpuhp_cpu_state
*st
, enum cpuhp_state target
)
414 enum cpuhp_state prev_state
= st
->state
;
416 st
->rollback
= false;
421 st
->bringup
= st
->state
< target
;
427 cpuhp_reset_state(struct cpuhp_cpu_state
*st
, enum cpuhp_state prev_state
)
432 * If we have st->last we need to undo partial multi_instance of this
433 * state first. Otherwise start undo at the previous state.
442 st
->target
= prev_state
;
443 st
->bringup
= !st
->bringup
;
446 /* Regular hotplug invocation of the AP hotplug thread */
447 static void __cpuhp_kick_ap(struct cpuhp_cpu_state
*st
)
449 if (!st
->single
&& st
->state
== st
->target
)
454 * Make sure the above stores are visible before should_run becomes
455 * true. Paired with the mb() above in cpuhp_thread_fun()
458 st
->should_run
= true;
459 wake_up_process(st
->thread
);
460 wait_for_ap_thread(st
, st
->bringup
);
463 static int cpuhp_kick_ap(struct cpuhp_cpu_state
*st
, enum cpuhp_state target
)
465 enum cpuhp_state prev_state
;
468 prev_state
= cpuhp_set_state(st
, target
);
470 if ((ret
= st
->result
)) {
471 cpuhp_reset_state(st
, prev_state
);
478 static int bringup_wait_for_ap(unsigned int cpu
)
480 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
482 /* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
483 wait_for_ap_thread(st
, true);
484 if (WARN_ON_ONCE((!cpu_online(cpu
))))
487 /* Unpark the stopper thread and the hotplug thread of the target cpu */
488 stop_machine_unpark(cpu
);
489 kthread_unpark(st
->thread
);
492 * SMT soft disabling on X86 requires to bring the CPU out of the
493 * BIOS 'wait for SIPI' state in order to set the CR4.MCE bit. The
494 * CPU marked itself as booted_once in cpu_notify_starting() so the
495 * cpu_smt_allowed() check will now return false if this is not the
498 if (!cpu_smt_allowed(cpu
))
501 if (st
->target
<= CPUHP_AP_ONLINE_IDLE
)
504 return cpuhp_kick_ap(st
, st
->target
);
507 static int bringup_cpu(unsigned int cpu
)
509 struct task_struct
*idle
= idle_thread_get(cpu
);
513 * Some architectures have to walk the irq descriptors to
514 * setup the vector space for the cpu which comes online.
515 * Prevent irq alloc/free across the bringup.
519 /* Arch-specific enabling code. */
520 ret
= __cpu_up(cpu
, idle
);
524 return bringup_wait_for_ap(cpu
);
528 * Hotplug state machine related functions
531 static void undo_cpu_up(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
533 for (st
->state
--; st
->state
> st
->target
; st
->state
--) {
534 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
536 if (!step
->skip_onerr
)
537 cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
541 static int cpuhp_up_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
542 enum cpuhp_state target
)
544 enum cpuhp_state prev_state
= st
->state
;
547 while (st
->state
< target
) {
549 ret
= cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
551 st
->target
= prev_state
;
552 undo_cpu_up(cpu
, st
);
560 * The cpu hotplug threads manage the bringup and teardown of the cpus
562 static void cpuhp_create(unsigned int cpu
)
564 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
566 init_completion(&st
->done_up
);
567 init_completion(&st
->done_down
);
570 static int cpuhp_should_run(unsigned int cpu
)
572 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
574 return st
->should_run
;
578 * Execute teardown/startup callbacks on the plugged cpu. Also used to invoke
579 * callbacks when a state gets [un]installed at runtime.
581 * Each invocation of this function by the smpboot thread does a single AP
584 * It has 3 modes of operation:
585 * - single: runs st->cb_state
586 * - up: runs ++st->state, while st->state < st->target
587 * - down: runs st->state--, while st->state > st->target
589 * When complete or on error, should_run is cleared and the completion is fired.
591 static void cpuhp_thread_fun(unsigned int cpu
)
593 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
594 bool bringup
= st
->bringup
;
595 enum cpuhp_state state
;
597 if (WARN_ON_ONCE(!st
->should_run
))
601 * ACQUIRE for the cpuhp_should_run() load of ->should_run. Ensures
602 * that if we see ->should_run we also see the rest of the state.
606 cpuhp_lock_acquire(bringup
);
609 state
= st
->cb_state
;
610 st
->should_run
= false;
615 st
->should_run
= (st
->state
< st
->target
);
616 WARN_ON_ONCE(st
->state
> st
->target
);
620 st
->should_run
= (st
->state
> st
->target
);
621 WARN_ON_ONCE(st
->state
< st
->target
);
625 WARN_ON_ONCE(!cpuhp_is_ap_state(state
));
628 struct cpuhp_step
*step
= cpuhp_get_step(state
);
629 if (step
->skip_onerr
)
633 if (cpuhp_is_atomic_state(state
)) {
635 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
639 * STARTING/DYING must not fail!
641 WARN_ON_ONCE(st
->result
);
643 st
->result
= cpuhp_invoke_callback(cpu
, state
, bringup
, st
->node
, &st
->last
);
648 * If we fail on a rollback, we're up a creek without no
649 * paddle, no way forward, no way back. We loose, thanks for
652 WARN_ON_ONCE(st
->rollback
);
653 st
->should_run
= false;
657 cpuhp_lock_release(bringup
);
660 complete_ap_thread(st
, bringup
);
663 /* Invoke a single callback on a remote cpu */
665 cpuhp_invoke_ap_callback(int cpu
, enum cpuhp_state state
, bool bringup
,
666 struct hlist_node
*node
)
668 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
671 if (!cpu_online(cpu
))
674 cpuhp_lock_acquire(false);
675 cpuhp_lock_release(false);
677 cpuhp_lock_acquire(true);
678 cpuhp_lock_release(true);
681 * If we are up and running, use the hotplug thread. For early calls
682 * we invoke the thread function directly.
685 return cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
687 st
->rollback
= false;
691 st
->bringup
= bringup
;
692 st
->cb_state
= state
;
698 * If we failed and did a partial, do a rollback.
700 if ((ret
= st
->result
) && st
->last
) {
702 st
->bringup
= !bringup
;
708 * Clean up the leftovers so the next hotplug operation wont use stale
711 st
->node
= st
->last
= NULL
;
715 static int cpuhp_kick_ap_work(unsigned int cpu
)
717 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
718 enum cpuhp_state prev_state
= st
->state
;
721 cpuhp_lock_acquire(false);
722 cpuhp_lock_release(false);
724 cpuhp_lock_acquire(true);
725 cpuhp_lock_release(true);
727 trace_cpuhp_enter(cpu
, st
->target
, prev_state
, cpuhp_kick_ap_work
);
728 ret
= cpuhp_kick_ap(st
, st
->target
);
729 trace_cpuhp_exit(cpu
, st
->state
, prev_state
, ret
);
734 static struct smp_hotplug_thread cpuhp_threads
= {
735 .store
= &cpuhp_state
.thread
,
736 .create
= &cpuhp_create
,
737 .thread_should_run
= cpuhp_should_run
,
738 .thread_fn
= cpuhp_thread_fun
,
739 .thread_comm
= "cpuhp/%u",
743 void __init
cpuhp_threads_init(void)
745 BUG_ON(smpboot_register_percpu_thread(&cpuhp_threads
));
746 kthread_unpark(this_cpu_read(cpuhp_state
.thread
));
749 #ifdef CONFIG_HOTPLUG_CPU
751 * clear_tasks_mm_cpumask - Safely clear tasks' mm_cpumask for a CPU
754 * This function walks all processes, finds a valid mm struct for each one and
755 * then clears a corresponding bit in mm's cpumask. While this all sounds
756 * trivial, there are various non-obvious corner cases, which this function
757 * tries to solve in a safe manner.
759 * Also note that the function uses a somewhat relaxed locking scheme, so it may
760 * be called only for an already offlined CPU.
762 void clear_tasks_mm_cpumask(int cpu
)
764 struct task_struct
*p
;
767 * This function is called after the cpu is taken down and marked
768 * offline, so its not like new tasks will ever get this cpu set in
769 * their mm mask. -- Peter Zijlstra
770 * Thus, we may use rcu_read_lock() here, instead of grabbing
771 * full-fledged tasklist_lock.
773 WARN_ON(cpu_online(cpu
));
775 for_each_process(p
) {
776 struct task_struct
*t
;
779 * Main thread might exit, but other threads may still have
780 * a valid mm. Find one.
782 t
= find_lock_task_mm(p
);
785 cpumask_clear_cpu(cpu
, mm_cpumask(t
->mm
));
791 /* Take this CPU down. */
792 static int take_cpu_down(void *_param
)
794 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
795 enum cpuhp_state target
= max((int)st
->target
, CPUHP_AP_OFFLINE
);
796 int err
, cpu
= smp_processor_id();
799 /* Ensure this CPU doesn't handle any more interrupts. */
800 err
= __cpu_disable();
805 * We get here while we are in CPUHP_TEARDOWN_CPU state and we must not
806 * do this step again.
808 WARN_ON(st
->state
!= CPUHP_TEARDOWN_CPU
);
810 /* Invoke the former CPU_DYING callbacks */
811 for (; st
->state
> target
; st
->state
--) {
812 ret
= cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
814 * DYING must not fail!
819 /* Give up timekeeping duties */
820 tick_handover_do_timer();
821 /* Park the stopper thread */
822 stop_machine_park(cpu
);
826 static int takedown_cpu(unsigned int cpu
)
828 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
831 /* Park the smpboot threads */
832 kthread_park(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
835 * Prevent irq alloc/free while the dying cpu reorganizes the
836 * interrupt affinities.
841 * So now all preempt/rcu users must observe !cpu_active().
843 err
= stop_machine_cpuslocked(take_cpu_down
, NULL
, cpumask_of(cpu
));
845 /* CPU refused to die */
847 /* Unpark the hotplug thread so we can rollback there */
848 kthread_unpark(per_cpu_ptr(&cpuhp_state
, cpu
)->thread
);
851 BUG_ON(cpu_online(cpu
));
854 * The teardown callback for CPUHP_AP_SCHED_STARTING will have removed
855 * all runnable tasks from the CPU, there's only the idle task left now
856 * that the migration thread is done doing the stop_machine thing.
858 * Wait for the stop thread to go away.
860 wait_for_ap_thread(st
, false);
861 BUG_ON(st
->state
!= CPUHP_AP_IDLE_DEAD
);
863 /* Interrupts are moved away from the dying cpu, reenable alloc/free */
866 hotplug_cpu__broadcast_tick_pull(cpu
);
867 /* This actually kills the CPU. */
870 tick_cleanup_dead_cpu(cpu
);
871 rcutree_migrate_callbacks(cpu
);
875 static void cpuhp_complete_idle_dead(void *arg
)
877 struct cpuhp_cpu_state
*st
= arg
;
879 complete_ap_thread(st
, false);
882 void cpuhp_report_idle_dead(void)
884 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
886 BUG_ON(st
->state
!= CPUHP_AP_OFFLINE
);
887 rcu_report_dead(smp_processor_id());
888 st
->state
= CPUHP_AP_IDLE_DEAD
;
890 * We cannot call complete after rcu_report_dead() so we delegate it
893 smp_call_function_single(cpumask_first(cpu_online_mask
),
894 cpuhp_complete_idle_dead
, st
, 0);
897 static void undo_cpu_down(unsigned int cpu
, struct cpuhp_cpu_state
*st
)
899 for (st
->state
++; st
->state
< st
->target
; st
->state
++) {
900 struct cpuhp_step
*step
= cpuhp_get_step(st
->state
);
902 if (!step
->skip_onerr
)
903 cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
907 static int cpuhp_down_callbacks(unsigned int cpu
, struct cpuhp_cpu_state
*st
,
908 enum cpuhp_state target
)
910 enum cpuhp_state prev_state
= st
->state
;
913 for (; st
->state
> target
; st
->state
--) {
914 ret
= cpuhp_invoke_callback(cpu
, st
->state
, false, NULL
, NULL
);
916 st
->target
= prev_state
;
917 if (st
->state
< prev_state
)
918 undo_cpu_down(cpu
, st
);
925 /* Requires cpu_add_remove_lock to be held */
926 static int __ref
_cpu_down(unsigned int cpu
, int tasks_frozen
,
927 enum cpuhp_state target
)
929 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
930 int prev_state
, ret
= 0;
932 if (num_online_cpus() == 1)
935 if (!cpu_present(cpu
))
940 cpuhp_tasks_frozen
= tasks_frozen
;
942 prev_state
= cpuhp_set_state(st
, target
);
944 * If the current CPU state is in the range of the AP hotplug thread,
945 * then we need to kick the thread.
947 if (st
->state
> CPUHP_TEARDOWN_CPU
) {
948 st
->target
= max((int)target
, CPUHP_TEARDOWN_CPU
);
949 ret
= cpuhp_kick_ap_work(cpu
);
951 * The AP side has done the error rollback already. Just
952 * return the error code..
958 * We might have stopped still in the range of the AP hotplug
959 * thread. Nothing to do anymore.
961 if (st
->state
> CPUHP_TEARDOWN_CPU
)
967 * The AP brought itself down to CPUHP_TEARDOWN_CPU. So we need
968 * to do the further cleanups.
970 ret
= cpuhp_down_callbacks(cpu
, st
, target
);
971 if (ret
&& st
->state
== CPUHP_TEARDOWN_CPU
&& st
->state
< prev_state
) {
972 cpuhp_reset_state(st
, prev_state
);
979 * Do post unplug cleanup. This is still protected against
980 * concurrent CPU hotplug via cpu_add_remove_lock.
982 lockup_detector_cleanup();
987 static int cpu_down_maps_locked(unsigned int cpu
, enum cpuhp_state target
)
989 if (cpu_hotplug_disabled
)
991 return _cpu_down(cpu
, 0, target
);
994 static int do_cpu_down(unsigned int cpu
, enum cpuhp_state target
)
998 cpu_maps_update_begin();
999 err
= cpu_down_maps_locked(cpu
, target
);
1000 cpu_maps_update_done();
1004 int cpu_down(unsigned int cpu
)
1006 return do_cpu_down(cpu
, CPUHP_OFFLINE
);
1008 EXPORT_SYMBOL(cpu_down
);
1011 #define takedown_cpu NULL
1012 #endif /*CONFIG_HOTPLUG_CPU*/
1015 * notify_cpu_starting(cpu) - Invoke the callbacks on the starting CPU
1016 * @cpu: cpu that just started
1018 * It must be called by the arch code on the new cpu, before the new cpu
1019 * enables interrupts and before the "boot" cpu returns from __cpu_up().
1021 void notify_cpu_starting(unsigned int cpu
)
1023 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1024 enum cpuhp_state target
= min((int)st
->target
, CPUHP_AP_ONLINE
);
1027 rcu_cpu_starting(cpu
); /* Enables RCU usage on this CPU. */
1028 st
->booted_once
= true;
1029 while (st
->state
< target
) {
1031 ret
= cpuhp_invoke_callback(cpu
, st
->state
, true, NULL
, NULL
);
1033 * STARTING must not fail!
1040 * Called from the idle task. Wake up the controlling task which brings the
1041 * stopper and the hotplug thread of the upcoming CPU up and then delegates
1042 * the rest of the online bringup to the hotplug thread.
1044 void cpuhp_online_idle(enum cpuhp_state state
)
1046 struct cpuhp_cpu_state
*st
= this_cpu_ptr(&cpuhp_state
);
1048 /* Happens for the boot cpu */
1049 if (state
!= CPUHP_AP_ONLINE_IDLE
)
1052 st
->state
= CPUHP_AP_ONLINE_IDLE
;
1053 complete_ap_thread(st
, true);
1056 /* Requires cpu_add_remove_lock to be held */
1057 static int _cpu_up(unsigned int cpu
, int tasks_frozen
, enum cpuhp_state target
)
1059 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1060 struct task_struct
*idle
;
1065 if (!cpu_present(cpu
)) {
1071 * The caller of do_cpu_up might have raced with another
1072 * caller. Ignore it for now.
1074 if (st
->state
>= target
)
1077 if (st
->state
== CPUHP_OFFLINE
) {
1078 /* Let it fail before we try to bring the cpu up */
1079 idle
= idle_thread_get(cpu
);
1081 ret
= PTR_ERR(idle
);
1086 cpuhp_tasks_frozen
= tasks_frozen
;
1088 cpuhp_set_state(st
, target
);
1090 * If the current CPU state is in the range of the AP hotplug thread,
1091 * then we need to kick the thread once more.
1093 if (st
->state
> CPUHP_BRINGUP_CPU
) {
1094 ret
= cpuhp_kick_ap_work(cpu
);
1096 * The AP side has done the error rollback already. Just
1097 * return the error code..
1104 * Try to reach the target state. We max out on the BP at
1105 * CPUHP_BRINGUP_CPU. After that the AP hotplug thread is
1106 * responsible for bringing it up to the target state.
1108 target
= min((int)target
, CPUHP_BRINGUP_CPU
);
1109 ret
= cpuhp_up_callbacks(cpu
, st
, target
);
1111 cpus_write_unlock();
1116 static int do_cpu_up(unsigned int cpu
, enum cpuhp_state target
)
1120 if (!cpu_possible(cpu
)) {
1121 pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
1123 #if defined(CONFIG_IA64)
1124 pr_err("please check additional_cpus= boot parameter\n");
1129 err
= try_online_node(cpu_to_node(cpu
));
1133 cpu_maps_update_begin();
1135 if (cpu_hotplug_disabled
) {
1139 if (!cpu_smt_allowed(cpu
)) {
1144 err
= _cpu_up(cpu
, 0, target
);
1146 cpu_maps_update_done();
1150 int cpu_up(unsigned int cpu
)
1152 return do_cpu_up(cpu
, CPUHP_ONLINE
);
1154 EXPORT_SYMBOL_GPL(cpu_up
);
1156 #ifdef CONFIG_PM_SLEEP_SMP
1157 static cpumask_var_t frozen_cpus
;
1159 int freeze_secondary_cpus(int primary
)
1163 cpu_maps_update_begin();
1164 if (!cpu_online(primary
))
1165 primary
= cpumask_first(cpu_online_mask
);
1167 * We take down all of the non-boot CPUs in one shot to avoid races
1168 * with the userspace trying to use the CPU hotplug at the same time
1170 cpumask_clear(frozen_cpus
);
1172 pr_info("Disabling non-boot CPUs ...\n");
1173 for_each_online_cpu(cpu
) {
1176 trace_suspend_resume(TPS("CPU_OFF"), cpu
, true);
1177 error
= _cpu_down(cpu
, 1, CPUHP_OFFLINE
);
1178 trace_suspend_resume(TPS("CPU_OFF"), cpu
, false);
1180 cpumask_set_cpu(cpu
, frozen_cpus
);
1182 pr_err("Error taking CPU%d down: %d\n", cpu
, error
);
1188 BUG_ON(num_online_cpus() > 1);
1190 pr_err("Non-boot CPUs are not disabled\n");
1193 * Make sure the CPUs won't be enabled by someone else. We need to do
1194 * this even in case of failure as all disable_nonboot_cpus() users are
1195 * supposed to do enable_nonboot_cpus() on the failure path.
1197 cpu_hotplug_disabled
++;
1199 cpu_maps_update_done();
1203 void __weak
arch_enable_nonboot_cpus_begin(void)
1207 void __weak
arch_enable_nonboot_cpus_end(void)
1211 void enable_nonboot_cpus(void)
1215 /* Allow everyone to use the CPU hotplug again */
1216 cpu_maps_update_begin();
1217 __cpu_hotplug_enable();
1218 if (cpumask_empty(frozen_cpus
))
1221 pr_info("Enabling non-boot CPUs ...\n");
1223 arch_enable_nonboot_cpus_begin();
1225 for_each_cpu(cpu
, frozen_cpus
) {
1226 trace_suspend_resume(TPS("CPU_ON"), cpu
, true);
1227 error
= _cpu_up(cpu
, 1, CPUHP_ONLINE
);
1228 trace_suspend_resume(TPS("CPU_ON"), cpu
, false);
1230 pr_info("CPU%d is up\n", cpu
);
1233 pr_warn("Error taking CPU%d up: %d\n", cpu
, error
);
1236 arch_enable_nonboot_cpus_end();
1238 cpumask_clear(frozen_cpus
);
1240 cpu_maps_update_done();
1243 static int __init
alloc_frozen_cpus(void)
1245 if (!alloc_cpumask_var(&frozen_cpus
, GFP_KERNEL
|__GFP_ZERO
))
1249 core_initcall(alloc_frozen_cpus
);
1252 * When callbacks for CPU hotplug notifications are being executed, we must
1253 * ensure that the state of the system with respect to the tasks being frozen
1254 * or not, as reported by the notification, remains unchanged *throughout the
1255 * duration* of the execution of the callbacks.
1256 * Hence we need to prevent the freezer from racing with regular CPU hotplug.
1258 * This synchronization is implemented by mutually excluding regular CPU
1259 * hotplug and Suspend/Hibernate call paths by hooking onto the Suspend/
1260 * Hibernate notifications.
1263 cpu_hotplug_pm_callback(struct notifier_block
*nb
,
1264 unsigned long action
, void *ptr
)
1268 case PM_SUSPEND_PREPARE
:
1269 case PM_HIBERNATION_PREPARE
:
1270 cpu_hotplug_disable();
1273 case PM_POST_SUSPEND
:
1274 case PM_POST_HIBERNATION
:
1275 cpu_hotplug_enable();
1286 static int __init
cpu_hotplug_pm_sync_init(void)
1289 * cpu_hotplug_pm_callback has higher priority than x86
1290 * bsp_pm_callback which depends on cpu_hotplug_pm_callback
1291 * to disable cpu hotplug to avoid cpu hotplug race.
1293 pm_notifier(cpu_hotplug_pm_callback
, 0);
1296 core_initcall(cpu_hotplug_pm_sync_init
);
1298 #endif /* CONFIG_PM_SLEEP_SMP */
1302 #endif /* CONFIG_SMP */
1304 /* Boot processor state steps */
1305 static struct cpuhp_step cpuhp_bp_states
[] = {
1308 .startup
.single
= NULL
,
1309 .teardown
.single
= NULL
,
1312 [CPUHP_CREATE_THREADS
]= {
1313 .name
= "threads:prepare",
1314 .startup
.single
= smpboot_create_threads
,
1315 .teardown
.single
= NULL
,
1318 [CPUHP_PERF_PREPARE
] = {
1319 .name
= "perf:prepare",
1320 .startup
.single
= perf_event_init_cpu
,
1321 .teardown
.single
= perf_event_exit_cpu
,
1323 [CPUHP_WORKQUEUE_PREP
] = {
1324 .name
= "workqueue:prepare",
1325 .startup
.single
= workqueue_prepare_cpu
,
1326 .teardown
.single
= NULL
,
1328 [CPUHP_HRTIMERS_PREPARE
] = {
1329 .name
= "hrtimers:prepare",
1330 .startup
.single
= hrtimers_prepare_cpu
,
1331 .teardown
.single
= hrtimers_dead_cpu
,
1333 [CPUHP_SMPCFD_PREPARE
] = {
1334 .name
= "smpcfd:prepare",
1335 .startup
.single
= smpcfd_prepare_cpu
,
1336 .teardown
.single
= smpcfd_dead_cpu
,
1338 [CPUHP_RELAY_PREPARE
] = {
1339 .name
= "relay:prepare",
1340 .startup
.single
= relay_prepare_cpu
,
1341 .teardown
.single
= NULL
,
1343 [CPUHP_SLAB_PREPARE
] = {
1344 .name
= "slab:prepare",
1345 .startup
.single
= slab_prepare_cpu
,
1346 .teardown
.single
= slab_dead_cpu
,
1348 [CPUHP_RCUTREE_PREP
] = {
1349 .name
= "RCU/tree:prepare",
1350 .startup
.single
= rcutree_prepare_cpu
,
1351 .teardown
.single
= rcutree_dead_cpu
,
1354 * On the tear-down path, timers_dead_cpu() must be invoked
1355 * before blk_mq_queue_reinit_notify() from notify_dead(),
1356 * otherwise a RCU stall occurs.
1358 [CPUHP_TIMERS_PREPARE
] = {
1359 .name
= "timers:dead",
1360 .startup
.single
= timers_prepare_cpu
,
1361 .teardown
.single
= timers_dead_cpu
,
1363 /* Kicks the plugged cpu into life */
1364 [CPUHP_BRINGUP_CPU
] = {
1365 .name
= "cpu:bringup",
1366 .startup
.single
= bringup_cpu
,
1367 .teardown
.single
= NULL
,
1371 * Handled on controll processor until the plugged processor manages
1374 [CPUHP_TEARDOWN_CPU
] = {
1375 .name
= "cpu:teardown",
1376 .startup
.single
= NULL
,
1377 .teardown
.single
= takedown_cpu
,
1381 [CPUHP_BRINGUP_CPU
] = { },
1385 /* Application processor state steps */
1386 static struct cpuhp_step cpuhp_ap_states
[] = {
1388 /* Final state before CPU kills itself */
1389 [CPUHP_AP_IDLE_DEAD
] = {
1390 .name
= "idle:dead",
1393 * Last state before CPU enters the idle loop to die. Transient state
1394 * for synchronization.
1396 [CPUHP_AP_OFFLINE
] = {
1397 .name
= "ap:offline",
1400 /* First state is scheduler control. Interrupts are disabled */
1401 [CPUHP_AP_SCHED_STARTING
] = {
1402 .name
= "sched:starting",
1403 .startup
.single
= sched_cpu_starting
,
1404 .teardown
.single
= sched_cpu_dying
,
1406 [CPUHP_AP_RCUTREE_DYING
] = {
1407 .name
= "RCU/tree:dying",
1408 .startup
.single
= NULL
,
1409 .teardown
.single
= rcutree_dying_cpu
,
1411 [CPUHP_AP_SMPCFD_DYING
] = {
1412 .name
= "smpcfd:dying",
1413 .startup
.single
= NULL
,
1414 .teardown
.single
= smpcfd_dying_cpu
,
1416 /* Entry state on starting. Interrupts enabled from here on. Transient
1417 * state for synchronsization */
1418 [CPUHP_AP_ONLINE
] = {
1419 .name
= "ap:online",
1421 /* Handle smpboot threads park/unpark */
1422 [CPUHP_AP_SMPBOOT_THREADS
] = {
1423 .name
= "smpboot/threads:online",
1424 .startup
.single
= smpboot_unpark_threads
,
1425 .teardown
.single
= smpboot_park_threads
,
1427 [CPUHP_AP_IRQ_AFFINITY_ONLINE
] = {
1428 .name
= "irq/affinity:online",
1429 .startup
.single
= irq_affinity_online_cpu
,
1430 .teardown
.single
= NULL
,
1432 [CPUHP_AP_PERF_ONLINE
] = {
1433 .name
= "perf:online",
1434 .startup
.single
= perf_event_init_cpu
,
1435 .teardown
.single
= perf_event_exit_cpu
,
1437 [CPUHP_AP_WORKQUEUE_ONLINE
] = {
1438 .name
= "workqueue:online",
1439 .startup
.single
= workqueue_online_cpu
,
1440 .teardown
.single
= workqueue_offline_cpu
,
1442 [CPUHP_AP_RCUTREE_ONLINE
] = {
1443 .name
= "RCU/tree:online",
1444 .startup
.single
= rcutree_online_cpu
,
1445 .teardown
.single
= rcutree_offline_cpu
,
1449 * The dynamically registered state space is here
1453 /* Last state is scheduler control setting the cpu active */
1454 [CPUHP_AP_ACTIVE
] = {
1455 .name
= "sched:active",
1456 .startup
.single
= sched_cpu_activate
,
1457 .teardown
.single
= sched_cpu_deactivate
,
1461 /* CPU is fully up and running. */
1464 .startup
.single
= NULL
,
1465 .teardown
.single
= NULL
,
1469 /* Sanity check for callbacks */
1470 static int cpuhp_cb_check(enum cpuhp_state state
)
1472 if (state
<= CPUHP_OFFLINE
|| state
>= CPUHP_ONLINE
)
1478 * Returns a free for dynamic slot assignment of the Online state. The states
1479 * are protected by the cpuhp_slot_states mutex and an empty slot is identified
1480 * by having no name assigned.
1482 static int cpuhp_reserve_state(enum cpuhp_state state
)
1484 enum cpuhp_state i
, end
;
1485 struct cpuhp_step
*step
;
1488 case CPUHP_AP_ONLINE_DYN
:
1489 step
= cpuhp_ap_states
+ CPUHP_AP_ONLINE_DYN
;
1490 end
= CPUHP_AP_ONLINE_DYN_END
;
1492 case CPUHP_BP_PREPARE_DYN
:
1493 step
= cpuhp_bp_states
+ CPUHP_BP_PREPARE_DYN
;
1494 end
= CPUHP_BP_PREPARE_DYN_END
;
1500 for (i
= state
; i
<= end
; i
++, step
++) {
1504 WARN(1, "No more dynamic states available for CPU hotplug\n");
1508 static int cpuhp_store_callbacks(enum cpuhp_state state
, const char *name
,
1509 int (*startup
)(unsigned int cpu
),
1510 int (*teardown
)(unsigned int cpu
),
1511 bool multi_instance
)
1513 /* (Un)Install the callbacks for further cpu hotplug operations */
1514 struct cpuhp_step
*sp
;
1518 * If name is NULL, then the state gets removed.
1520 * CPUHP_AP_ONLINE_DYN and CPUHP_BP_PREPARE_DYN are handed out on
1521 * the first allocation from these dynamic ranges, so the removal
1522 * would trigger a new allocation and clear the wrong (already
1523 * empty) state, leaving the callbacks of the to be cleared state
1524 * dangling, which causes wreckage on the next hotplug operation.
1526 if (name
&& (state
== CPUHP_AP_ONLINE_DYN
||
1527 state
== CPUHP_BP_PREPARE_DYN
)) {
1528 ret
= cpuhp_reserve_state(state
);
1533 sp
= cpuhp_get_step(state
);
1534 if (name
&& sp
->name
)
1537 sp
->startup
.single
= startup
;
1538 sp
->teardown
.single
= teardown
;
1540 sp
->multi_instance
= multi_instance
;
1541 INIT_HLIST_HEAD(&sp
->list
);
1545 static void *cpuhp_get_teardown_cb(enum cpuhp_state state
)
1547 return cpuhp_get_step(state
)->teardown
.single
;
1551 * Call the startup/teardown function for a step either on the AP or
1552 * on the current CPU.
1554 static int cpuhp_issue_call(int cpu
, enum cpuhp_state state
, bool bringup
,
1555 struct hlist_node
*node
)
1557 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1561 * If there's nothing to do, we done.
1562 * Relies on the union for multi_instance.
1564 if ((bringup
&& !sp
->startup
.single
) ||
1565 (!bringup
&& !sp
->teardown
.single
))
1568 * The non AP bound callbacks can fail on bringup. On teardown
1569 * e.g. module removal we crash for now.
1572 if (cpuhp_is_ap_state(state
))
1573 ret
= cpuhp_invoke_ap_callback(cpu
, state
, bringup
, node
);
1575 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1577 ret
= cpuhp_invoke_callback(cpu
, state
, bringup
, node
, NULL
);
1579 BUG_ON(ret
&& !bringup
);
1584 * Called from __cpuhp_setup_state on a recoverable failure.
1586 * Note: The teardown callbacks for rollback are not allowed to fail!
1588 static void cpuhp_rollback_install(int failedcpu
, enum cpuhp_state state
,
1589 struct hlist_node
*node
)
1593 /* Roll back the already executed steps on the other cpus */
1594 for_each_present_cpu(cpu
) {
1595 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1596 int cpustate
= st
->state
;
1598 if (cpu
>= failedcpu
)
1601 /* Did we invoke the startup call on that cpu ? */
1602 if (cpustate
>= state
)
1603 cpuhp_issue_call(cpu
, state
, false, node
);
1607 int __cpuhp_state_add_instance_cpuslocked(enum cpuhp_state state
,
1608 struct hlist_node
*node
,
1611 struct cpuhp_step
*sp
;
1615 lockdep_assert_cpus_held();
1617 sp
= cpuhp_get_step(state
);
1618 if (sp
->multi_instance
== false)
1621 mutex_lock(&cpuhp_state_mutex
);
1623 if (!invoke
|| !sp
->startup
.multi
)
1627 * Try to call the startup callback for each present cpu
1628 * depending on the hotplug state of the cpu.
1630 for_each_present_cpu(cpu
) {
1631 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1632 int cpustate
= st
->state
;
1634 if (cpustate
< state
)
1637 ret
= cpuhp_issue_call(cpu
, state
, true, node
);
1639 if (sp
->teardown
.multi
)
1640 cpuhp_rollback_install(cpu
, state
, node
);
1646 hlist_add_head(node
, &sp
->list
);
1648 mutex_unlock(&cpuhp_state_mutex
);
1652 int __cpuhp_state_add_instance(enum cpuhp_state state
, struct hlist_node
*node
,
1658 ret
= __cpuhp_state_add_instance_cpuslocked(state
, node
, invoke
);
1662 EXPORT_SYMBOL_GPL(__cpuhp_state_add_instance
);
1665 * __cpuhp_setup_state_cpuslocked - Setup the callbacks for an hotplug machine state
1666 * @state: The state to setup
1667 * @invoke: If true, the startup function is invoked for cpus where
1668 * cpu state >= @state
1669 * @startup: startup callback function
1670 * @teardown: teardown callback function
1671 * @multi_instance: State is set up for multiple instances which get
1674 * The caller needs to hold cpus read locked while calling this function.
1677 * Positive state number if @state is CPUHP_AP_ONLINE_DYN
1678 * 0 for all other states
1679 * On failure: proper (negative) error code
1681 int __cpuhp_setup_state_cpuslocked(enum cpuhp_state state
,
1682 const char *name
, bool invoke
,
1683 int (*startup
)(unsigned int cpu
),
1684 int (*teardown
)(unsigned int cpu
),
1685 bool multi_instance
)
1690 lockdep_assert_cpus_held();
1692 if (cpuhp_cb_check(state
) || !name
)
1695 mutex_lock(&cpuhp_state_mutex
);
1697 ret
= cpuhp_store_callbacks(state
, name
, startup
, teardown
,
1700 dynstate
= state
== CPUHP_AP_ONLINE_DYN
;
1701 if (ret
> 0 && dynstate
) {
1706 if (ret
|| !invoke
|| !startup
)
1710 * Try to call the startup callback for each present cpu
1711 * depending on the hotplug state of the cpu.
1713 for_each_present_cpu(cpu
) {
1714 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1715 int cpustate
= st
->state
;
1717 if (cpustate
< state
)
1720 ret
= cpuhp_issue_call(cpu
, state
, true, NULL
);
1723 cpuhp_rollback_install(cpu
, state
, NULL
);
1724 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1729 mutex_unlock(&cpuhp_state_mutex
);
1731 * If the requested state is CPUHP_AP_ONLINE_DYN, return the
1732 * dynamically allocated state in case of success.
1734 if (!ret
&& dynstate
)
1738 EXPORT_SYMBOL(__cpuhp_setup_state_cpuslocked
);
1740 int __cpuhp_setup_state(enum cpuhp_state state
,
1741 const char *name
, bool invoke
,
1742 int (*startup
)(unsigned int cpu
),
1743 int (*teardown
)(unsigned int cpu
),
1744 bool multi_instance
)
1749 ret
= __cpuhp_setup_state_cpuslocked(state
, name
, invoke
, startup
,
1750 teardown
, multi_instance
);
1754 EXPORT_SYMBOL(__cpuhp_setup_state
);
1756 int __cpuhp_state_remove_instance(enum cpuhp_state state
,
1757 struct hlist_node
*node
, bool invoke
)
1759 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1762 BUG_ON(cpuhp_cb_check(state
));
1764 if (!sp
->multi_instance
)
1768 mutex_lock(&cpuhp_state_mutex
);
1770 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1773 * Call the teardown callback for each present cpu depending
1774 * on the hotplug state of the cpu. This function is not
1775 * allowed to fail currently!
1777 for_each_present_cpu(cpu
) {
1778 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1779 int cpustate
= st
->state
;
1781 if (cpustate
>= state
)
1782 cpuhp_issue_call(cpu
, state
, false, node
);
1787 mutex_unlock(&cpuhp_state_mutex
);
1792 EXPORT_SYMBOL_GPL(__cpuhp_state_remove_instance
);
1795 * __cpuhp_remove_state_cpuslocked - Remove the callbacks for an hotplug machine state
1796 * @state: The state to remove
1797 * @invoke: If true, the teardown function is invoked for cpus where
1798 * cpu state >= @state
1800 * The caller needs to hold cpus read locked while calling this function.
1801 * The teardown callback is currently not allowed to fail. Think
1802 * about module removal!
1804 void __cpuhp_remove_state_cpuslocked(enum cpuhp_state state
, bool invoke
)
1806 struct cpuhp_step
*sp
= cpuhp_get_step(state
);
1809 BUG_ON(cpuhp_cb_check(state
));
1811 lockdep_assert_cpus_held();
1813 mutex_lock(&cpuhp_state_mutex
);
1814 if (sp
->multi_instance
) {
1815 WARN(!hlist_empty(&sp
->list
),
1816 "Error: Removing state %d which has instances left.\n",
1821 if (!invoke
|| !cpuhp_get_teardown_cb(state
))
1825 * Call the teardown callback for each present cpu depending
1826 * on the hotplug state of the cpu. This function is not
1827 * allowed to fail currently!
1829 for_each_present_cpu(cpu
) {
1830 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, cpu
);
1831 int cpustate
= st
->state
;
1833 if (cpustate
>= state
)
1834 cpuhp_issue_call(cpu
, state
, false, NULL
);
1837 cpuhp_store_callbacks(state
, NULL
, NULL
, NULL
, false);
1838 mutex_unlock(&cpuhp_state_mutex
);
1840 EXPORT_SYMBOL(__cpuhp_remove_state_cpuslocked
);
1842 void __cpuhp_remove_state(enum cpuhp_state state
, bool invoke
)
1845 __cpuhp_remove_state_cpuslocked(state
, invoke
);
1848 EXPORT_SYMBOL(__cpuhp_remove_state
);
1850 #if defined(CONFIG_SYSFS) && defined(CONFIG_HOTPLUG_CPU)
1851 static ssize_t
show_cpuhp_state(struct device
*dev
,
1852 struct device_attribute
*attr
, char *buf
)
1854 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1856 return sprintf(buf
, "%d\n", st
->state
);
1858 static DEVICE_ATTR(state
, 0444, show_cpuhp_state
, NULL
);
1860 static ssize_t
write_cpuhp_target(struct device
*dev
,
1861 struct device_attribute
*attr
,
1862 const char *buf
, size_t count
)
1864 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1865 struct cpuhp_step
*sp
;
1868 ret
= kstrtoint(buf
, 10, &target
);
1872 #ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
1873 if (target
< CPUHP_OFFLINE
|| target
> CPUHP_ONLINE
)
1876 if (target
!= CPUHP_OFFLINE
&& target
!= CPUHP_ONLINE
)
1880 ret
= lock_device_hotplug_sysfs();
1884 mutex_lock(&cpuhp_state_mutex
);
1885 sp
= cpuhp_get_step(target
);
1886 ret
= !sp
->name
|| sp
->cant_stop
? -EINVAL
: 0;
1887 mutex_unlock(&cpuhp_state_mutex
);
1891 if (st
->state
< target
)
1892 ret
= do_cpu_up(dev
->id
, target
);
1894 ret
= do_cpu_down(dev
->id
, target
);
1896 unlock_device_hotplug();
1897 return ret
? ret
: count
;
1900 static ssize_t
show_cpuhp_target(struct device
*dev
,
1901 struct device_attribute
*attr
, char *buf
)
1903 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1905 return sprintf(buf
, "%d\n", st
->target
);
1907 static DEVICE_ATTR(target
, 0644, show_cpuhp_target
, write_cpuhp_target
);
1910 static ssize_t
write_cpuhp_fail(struct device
*dev
,
1911 struct device_attribute
*attr
,
1912 const char *buf
, size_t count
)
1914 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1915 struct cpuhp_step
*sp
;
1918 ret
= kstrtoint(buf
, 10, &fail
);
1923 * Cannot fail STARTING/DYING callbacks.
1925 if (cpuhp_is_atomic_state(fail
))
1929 * Cannot fail anything that doesn't have callbacks.
1931 mutex_lock(&cpuhp_state_mutex
);
1932 sp
= cpuhp_get_step(fail
);
1933 if (!sp
->startup
.single
&& !sp
->teardown
.single
)
1935 mutex_unlock(&cpuhp_state_mutex
);
1944 static ssize_t
show_cpuhp_fail(struct device
*dev
,
1945 struct device_attribute
*attr
, char *buf
)
1947 struct cpuhp_cpu_state
*st
= per_cpu_ptr(&cpuhp_state
, dev
->id
);
1949 return sprintf(buf
, "%d\n", st
->fail
);
1952 static DEVICE_ATTR(fail
, 0644, show_cpuhp_fail
, write_cpuhp_fail
);
1954 static struct attribute
*cpuhp_cpu_attrs
[] = {
1955 &dev_attr_state
.attr
,
1956 &dev_attr_target
.attr
,
1957 &dev_attr_fail
.attr
,
1961 static const struct attribute_group cpuhp_cpu_attr_group
= {
1962 .attrs
= cpuhp_cpu_attrs
,
1967 static ssize_t
show_cpuhp_states(struct device
*dev
,
1968 struct device_attribute
*attr
, char *buf
)
1970 ssize_t cur
, res
= 0;
1973 mutex_lock(&cpuhp_state_mutex
);
1974 for (i
= CPUHP_OFFLINE
; i
<= CPUHP_ONLINE
; i
++) {
1975 struct cpuhp_step
*sp
= cpuhp_get_step(i
);
1978 cur
= sprintf(buf
, "%3d: %s\n", i
, sp
->name
);
1983 mutex_unlock(&cpuhp_state_mutex
);
1986 static DEVICE_ATTR(states
, 0444, show_cpuhp_states
, NULL
);
1988 static struct attribute
*cpuhp_cpu_root_attrs
[] = {
1989 &dev_attr_states
.attr
,
1993 static const struct attribute_group cpuhp_cpu_root_attr_group
= {
1994 .attrs
= cpuhp_cpu_root_attrs
,
1999 #ifdef CONFIG_HOTPLUG_SMT
2001 static const char *smt_states
[] = {
2002 [CPU_SMT_ENABLED
] = "on",
2003 [CPU_SMT_DISABLED
] = "off",
2004 [CPU_SMT_FORCE_DISABLED
] = "forceoff",
2005 [CPU_SMT_NOT_SUPPORTED
] = "notsupported",
2009 show_smt_control(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2011 return snprintf(buf
, PAGE_SIZE
- 2, "%s\n", smt_states
[cpu_smt_control
]);
2014 static void cpuhp_offline_cpu_device(unsigned int cpu
)
2016 struct device
*dev
= get_cpu_device(cpu
);
2018 dev
->offline
= true;
2019 /* Tell user space about the state change */
2020 kobject_uevent(&dev
->kobj
, KOBJ_OFFLINE
);
2023 static void cpuhp_online_cpu_device(unsigned int cpu
)
2025 struct device
*dev
= get_cpu_device(cpu
);
2027 dev
->offline
= false;
2028 /* Tell user space about the state change */
2029 kobject_uevent(&dev
->kobj
, KOBJ_ONLINE
);
2032 static int cpuhp_smt_disable(enum cpuhp_smt_control ctrlval
)
2036 cpu_maps_update_begin();
2037 for_each_online_cpu(cpu
) {
2038 if (topology_is_primary_thread(cpu
))
2040 ret
= cpu_down_maps_locked(cpu
, CPUHP_OFFLINE
);
2044 * As this needs to hold the cpu maps lock it's impossible
2045 * to call device_offline() because that ends up calling
2046 * cpu_down() which takes cpu maps lock. cpu maps lock
2047 * needs to be held as this might race against in kernel
2048 * abusers of the hotplug machinery (thermal management).
2050 * So nothing would update device:offline state. That would
2051 * leave the sysfs entry stale and prevent onlining after
2052 * smt control has been changed to 'off' again. This is
2053 * called under the sysfs hotplug lock, so it is properly
2054 * serialized against the regular offline usage.
2056 cpuhp_offline_cpu_device(cpu
);
2059 cpu_smt_control
= ctrlval
;
2062 cpu_maps_update_done();
2066 static int cpuhp_smt_enable(void)
2070 cpu_maps_update_begin();
2071 cpu_smt_control
= CPU_SMT_ENABLED
;
2073 for_each_present_cpu(cpu
) {
2074 /* Skip online CPUs and CPUs on offline nodes */
2075 if (cpu_online(cpu
) || !node_online(cpu_to_node(cpu
)))
2077 ret
= _cpu_up(cpu
, 0, CPUHP_ONLINE
);
2080 /* See comment in cpuhp_smt_disable() */
2081 cpuhp_online_cpu_device(cpu
);
2083 cpu_maps_update_done();
2088 store_smt_control(struct device
*dev
, struct device_attribute
*attr
,
2089 const char *buf
, size_t count
)
2093 if (sysfs_streq(buf
, "on"))
2094 ctrlval
= CPU_SMT_ENABLED
;
2095 else if (sysfs_streq(buf
, "off"))
2096 ctrlval
= CPU_SMT_DISABLED
;
2097 else if (sysfs_streq(buf
, "forceoff"))
2098 ctrlval
= CPU_SMT_FORCE_DISABLED
;
2102 if (cpu_smt_control
== CPU_SMT_FORCE_DISABLED
)
2105 if (cpu_smt_control
== CPU_SMT_NOT_SUPPORTED
)
2108 ret
= lock_device_hotplug_sysfs();
2112 if (ctrlval
!= cpu_smt_control
) {
2114 case CPU_SMT_ENABLED
:
2115 ret
= cpuhp_smt_enable();
2117 case CPU_SMT_DISABLED
:
2118 case CPU_SMT_FORCE_DISABLED
:
2119 ret
= cpuhp_smt_disable(ctrlval
);
2124 unlock_device_hotplug();
2125 return ret
? ret
: count
;
2127 static DEVICE_ATTR(control
, 0644, show_smt_control
, store_smt_control
);
2130 show_smt_active(struct device
*dev
, struct device_attribute
*attr
, char *buf
)
2132 bool active
= topology_max_smt_threads() > 1;
2134 return snprintf(buf
, PAGE_SIZE
- 2, "%d\n", active
);
2136 static DEVICE_ATTR(active
, 0444, show_smt_active
, NULL
);
2138 static struct attribute
*cpuhp_smt_attrs
[] = {
2139 &dev_attr_control
.attr
,
2140 &dev_attr_active
.attr
,
2144 static const struct attribute_group cpuhp_smt_attr_group
= {
2145 .attrs
= cpuhp_smt_attrs
,
2150 static int __init
cpu_smt_state_init(void)
2152 return sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2153 &cpuhp_smt_attr_group
);
2157 static inline int cpu_smt_state_init(void) { return 0; }
2160 static int __init
cpuhp_sysfs_init(void)
2164 ret
= cpu_smt_state_init();
2168 ret
= sysfs_create_group(&cpu_subsys
.dev_root
->kobj
,
2169 &cpuhp_cpu_root_attr_group
);
2173 for_each_possible_cpu(cpu
) {
2174 struct device
*dev
= get_cpu_device(cpu
);
2178 ret
= sysfs_create_group(&dev
->kobj
, &cpuhp_cpu_attr_group
);
2184 device_initcall(cpuhp_sysfs_init
);
2188 * cpu_bit_bitmap[] is a special, "compressed" data structure that
2189 * represents all NR_CPUS bits binary values of 1<<nr.
2191 * It is used by cpumask_of() to get a constant address to a CPU
2192 * mask value that has a single bit set only.
2195 /* cpu_bit_bitmap[0] is empty - so we can back into it */
2196 #define MASK_DECLARE_1(x) [x+1][0] = (1UL << (x))
2197 #define MASK_DECLARE_2(x) MASK_DECLARE_1(x), MASK_DECLARE_1(x+1)
2198 #define MASK_DECLARE_4(x) MASK_DECLARE_2(x), MASK_DECLARE_2(x+2)
2199 #define MASK_DECLARE_8(x) MASK_DECLARE_4(x), MASK_DECLARE_4(x+4)
2201 const unsigned long cpu_bit_bitmap
[BITS_PER_LONG
+1][BITS_TO_LONGS(NR_CPUS
)] = {
2203 MASK_DECLARE_8(0), MASK_DECLARE_8(8),
2204 MASK_DECLARE_8(16), MASK_DECLARE_8(24),
2205 #if BITS_PER_LONG > 32
2206 MASK_DECLARE_8(32), MASK_DECLARE_8(40),
2207 MASK_DECLARE_8(48), MASK_DECLARE_8(56),
2210 EXPORT_SYMBOL_GPL(cpu_bit_bitmap
);
2212 const DECLARE_BITMAP(cpu_all_bits
, NR_CPUS
) = CPU_BITS_ALL
;
2213 EXPORT_SYMBOL(cpu_all_bits
);
2215 #ifdef CONFIG_INIT_ALL_POSSIBLE
2216 struct cpumask __cpu_possible_mask __read_mostly
2219 struct cpumask __cpu_possible_mask __read_mostly
;
2221 EXPORT_SYMBOL(__cpu_possible_mask
);
2223 struct cpumask __cpu_online_mask __read_mostly
;
2224 EXPORT_SYMBOL(__cpu_online_mask
);
2226 struct cpumask __cpu_present_mask __read_mostly
;
2227 EXPORT_SYMBOL(__cpu_present_mask
);
2229 struct cpumask __cpu_active_mask __read_mostly
;
2230 EXPORT_SYMBOL(__cpu_active_mask
);
2232 void init_cpu_present(const struct cpumask
*src
)
2234 cpumask_copy(&__cpu_present_mask
, src
);
2237 void init_cpu_possible(const struct cpumask
*src
)
2239 cpumask_copy(&__cpu_possible_mask
, src
);
2242 void init_cpu_online(const struct cpumask
*src
)
2244 cpumask_copy(&__cpu_online_mask
, src
);
2248 * Activate the first processor.
2250 void __init
boot_cpu_init(void)
2252 int cpu
= smp_processor_id();
2254 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
2255 set_cpu_online(cpu
, true);
2256 set_cpu_active(cpu
, true);
2257 set_cpu_present(cpu
, true);
2258 set_cpu_possible(cpu
, true);
2261 __boot_cpu_id
= cpu
;
2266 * Must be called _AFTER_ setting up the per_cpu areas
2268 void __init
boot_cpu_hotplug_init(void)
2271 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->booted_once
= true;
2273 per_cpu_ptr(&cpuhp_state
, smp_processor_id())->state
= CPUHP_ONLINE
;
2276 enum cpu_mitigations cpu_mitigations __ro_after_init
= CPU_MITIGATIONS_AUTO
;
2278 static int __init
mitigations_parse_cmdline(char *arg
)
2280 if (!strcmp(arg
, "off"))
2281 cpu_mitigations
= CPU_MITIGATIONS_OFF
;
2282 else if (!strcmp(arg
, "auto"))
2283 cpu_mitigations
= CPU_MITIGATIONS_AUTO
;
2284 else if (!strcmp(arg
, "auto,nosmt"))
2285 cpu_mitigations
= CPU_MITIGATIONS_AUTO_NOSMT
;
2289 early_param("mitigations", mitigations_parse_cmdline
);